Whamcloud - gitweb
LU-1855 build: fix 'out-of-bounds access' errors
[fs/lustre-release.git] / lustre / ptlrpc / sec.c
index a92a5c4..c4aa034 100644 (file)
@@ -1,28 +1,43 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- * Copyright (C) 2004-2007 Cluster File Systems, Inc.
- *   Author: Eric Mei <ericm@clusterfs.com>
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
  *
- *   Lustre is free software; you can redistribute it and/or
- *   modify it under the terms of version 2 of the GNU General Public
- *   License as published by the Free Software Foundation.
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  *
- *   Lustre is distributed in the hope that it will be useful,
- *   but WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   GNU General Public License for more details.
+ * lustre/ptlrpc/sec.c
  *
- *   You should have received a copy of the GNU General Public License
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * Author: Eric Mei <ericm@clusterfs.com>
  */
 
-#ifndef EXPORT_SYMTAB
-#define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_SEC
 
 #include <libcfs/libcfs.h>
@@ -48,7 +63,7 @@
  * policy registers                            *
  ***********************************************/
 
-static rwlock_t policy_lock = RW_LOCK_UNLOCKED;
+static cfs_rwlock_t policy_lock;
 static struct ptlrpc_sec_policy *policies[SPTLRPC_POLICY_MAX] = {
         NULL,
 };
@@ -64,13 +79,13 @@ int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy)
         if (number >= SPTLRPC_POLICY_MAX)
                 return -EINVAL;
 
-        write_lock(&policy_lock);
+        cfs_write_lock(&policy_lock);
         if (unlikely(policies[number])) {
-                write_unlock(&policy_lock);
+                cfs_write_unlock(&policy_lock);
                 return -EALREADY;
         }
         policies[number] = policy;
-        write_unlock(&policy_lock);
+        cfs_write_unlock(&policy_lock);
 
         CDEBUG(D_SEC, "%s: registered\n", policy->sp_name);
         return 0;
@@ -83,16 +98,16 @@ int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
 
         LASSERT(number < SPTLRPC_POLICY_MAX);
 
-        write_lock(&policy_lock);
+        cfs_write_lock(&policy_lock);
         if (unlikely(policies[number] == NULL)) {
-                write_unlock(&policy_lock);
+                cfs_write_unlock(&policy_lock);
                 CERROR("%s: already unregistered\n", policy->sp_name);
                 return -EINVAL;
         }
 
         LASSERT(policies[number] == policy);
         policies[number] = NULL;
-        write_unlock(&policy_lock);
+        cfs_write_unlock(&policy_lock);
 
         CDEBUG(D_SEC, "%s: unregistered\n", policy->sp_name);
         return 0;
@@ -100,46 +115,48 @@ int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy)
 EXPORT_SYMBOL(sptlrpc_unregister_policy);
 
 static
-struct ptlrpc_sec_policy * sptlrpc_rpcflavor2policy(__u16 flavor)
+struct ptlrpc_sec_policy * sptlrpc_wireflavor2policy(__u32 flavor)
 {
-        static DECLARE_MUTEX(load_mutex);
-        static atomic_t           loaded = ATOMIC_INIT(0);
+        static CFS_DEFINE_MUTEX(load_mutex);
+        static cfs_atomic_t       loaded = CFS_ATOMIC_INIT(0);
         struct ptlrpc_sec_policy *policy;
-        __u16                     number = RPC_FLVR_POLICY(flavor), flag = 0;
+        __u16                     number = SPTLRPC_FLVR_POLICY(flavor);
+        __u16                     flag = 0;
 
         if (number >= SPTLRPC_POLICY_MAX)
                 return NULL;
 
         while (1) {
-                read_lock(&policy_lock);
+                cfs_read_lock(&policy_lock);
                 policy = policies[number];
-                if (policy && !try_module_get(policy->sp_owner))
+                if (policy && !cfs_try_module_get(policy->sp_owner))
                         policy = NULL;
                 if (policy == NULL)
-                        flag = atomic_read(&loaded);
-                read_unlock(&policy_lock);
+                        flag = cfs_atomic_read(&loaded);
+                cfs_read_unlock(&policy_lock);
 
                 if (policy != NULL || flag != 0 ||
                     number != SPTLRPC_POLICY_GSS)
                         break;
 
                 /* try to load gss module, once */
-                mutex_down(&load_mutex);
-                if (atomic_read(&loaded) == 0) {
-                        if (request_module("ptlrpc_gss") == 0)
-                                CWARN("module ptlrpc_gss loaded on demand\n");
+                cfs_mutex_lock(&load_mutex);
+                if (cfs_atomic_read(&loaded) == 0) {
+                        if (cfs_request_module("ptlrpc_gss") == 0)
+                                CDEBUG(D_SEC,
+                                       "module ptlrpc_gss loaded on demand\n");
                         else
                                 CERROR("Unable to load module ptlrpc_gss\n");
 
-                        atomic_set(&loaded, 1);
+                        cfs_atomic_set(&loaded, 1);
                 }
-                mutex_up(&load_mutex);
+                cfs_mutex_unlock(&load_mutex);
         }
 
         return policy;
 }
 
-__u16 sptlrpc_name2rpcflavor(const char *name)
+__u32 sptlrpc_name2flavor_base(const char *name)
 {
         if (!strcmp(name, "null"))
                 return SPTLRPC_FLVR_NULL;
@@ -156,51 +173,85 @@ __u16 sptlrpc_name2rpcflavor(const char *name)
 
         return SPTLRPC_FLVR_INVALID;
 }
-EXPORT_SYMBOL(sptlrpc_name2rpcflavor);
+EXPORT_SYMBOL(sptlrpc_name2flavor_base);
 
-const char *sptlrpc_rpcflavor2name(__u16 flavor)
+const char *sptlrpc_flavor2name_base(__u32 flvr)
 {
-        switch (flavor) {
-        case SPTLRPC_FLVR_NULL:
+        __u32   base = SPTLRPC_FLVR_BASE(flvr);
+
+        if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL))
                 return "null";
-        case SPTLRPC_FLVR_PLAIN:
+        else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_PLAIN))
                 return "plain";
-        case SPTLRPC_FLVR_KRB5N:
+        else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5N))
                 return "krb5n";
-        case SPTLRPC_FLVR_KRB5A:
+        else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5A))
                 return "krb5a";
-        case SPTLRPC_FLVR_KRB5I:
+        else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5I))
                 return "krb5i";
-        case SPTLRPC_FLVR_KRB5P:
+        else if (base == SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_KRB5P))
                 return "krb5p";
-        default:
-                CERROR("invalid rpc flavor 0x%x(p%u,s%u,v%u)\n", flavor,
-                       RPC_FLVR_POLICY(flavor), RPC_FLVR_MECH(flavor),
-                       RPC_FLVR_SVC(flavor));
-        }
-        return "unknown";
+
+        CERROR("invalid wire flavor 0x%x\n", flvr);
+        return "invalid";
 }
-EXPORT_SYMBOL(sptlrpc_rpcflavor2name);
+EXPORT_SYMBOL(sptlrpc_flavor2name_base);
 
-int sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
+char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
+                               char *buf, int bufsize)
 {
-        char           *bulk;
-
-        if (sf->sf_bulk_ciph != BULK_CIPH_ALG_NULL)
-                bulk = "bulkp";
-        else if (sf->sf_bulk_hash != BULK_HASH_ALG_NULL)
-                bulk = "bulki";
+        if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN)
+                snprintf(buf, bufsize, "hash:%s",
+                         sptlrpc_get_hash_name(sf->u_bulk.hash.hash_alg));
         else
-                bulk = "bulkn";
+                snprintf(buf, bufsize, "%s",
+                         sptlrpc_flavor2name_base(sf->sf_rpc));
 
-        snprintf(buf, bufsize, "%s-%s:%s/%s",
-                 sptlrpc_rpcflavor2name(sf->sf_rpc), bulk,
-                 sptlrpc_get_hash_name(sf->sf_bulk_hash),
-                 sptlrpc_get_ciph_name(sf->sf_bulk_ciph));
-        return 0;
+        buf[bufsize - 1] = '\0';
+        return buf;
+}
+EXPORT_SYMBOL(sptlrpc_flavor2name_bulk);
+
+char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize)
+{
+        snprintf(buf, bufsize, "%s", sptlrpc_flavor2name_base(sf->sf_rpc));
+
+        /*
+         * currently we don't support customized bulk specification for
+         * flavors other than plain
+         */
+        if (SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_PLAIN) {
+                char bspec[16];
+
+                bspec[0] = '-';
+                sptlrpc_flavor2name_bulk(sf, &bspec[1], sizeof(bspec) - 1);
+                strncat(buf, bspec, bufsize);
+        }
+
+        buf[bufsize - 1] = '\0';
+        return buf;
 }
 EXPORT_SYMBOL(sptlrpc_flavor2name);
 
+char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize)
+{
+       buf[0] = '\0';
+
+       if (flags & PTLRPC_SEC_FL_REVERSE)
+               strlcat(buf, "reverse,", bufsize);
+       if (flags & PTLRPC_SEC_FL_ROOTONLY)
+               strlcat(buf, "rootonly,", bufsize);
+       if (flags & PTLRPC_SEC_FL_UDESC)
+               strlcat(buf, "udesc,", bufsize);
+       if (flags & PTLRPC_SEC_FL_BULK)
+               strlcat(buf, "bulk,", bufsize);
+       if (buf[0] == '\0')
+               strlcat(buf, "-,", bufsize);
+
+       return buf;
+}
+EXPORT_SYMBOL(sptlrpc_secflags2str);
+
 /**************************************************
  * client context APIs                            *
  **************************************************/
@@ -223,8 +274,8 @@ struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
                         remove_dead = 0;
                 }
         } else {
-                vcred.vc_uid = cfs_current()->uid;
-                vcred.vc_gid = cfs_current()->gid;
+                vcred.vc_uid = cfs_curproc_uid();
+                vcred.vc_gid = cfs_curproc_gid();
         }
 
         return sec->ps_policy->sp_cops->lookup_ctx(sec, &vcred,
@@ -233,8 +284,7 @@ struct ptlrpc_cli_ctx *get_my_ctx(struct ptlrpc_sec *sec)
 
 struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx)
 {
-        LASSERT(atomic_read(&ctx->cc_refcount) > 0);
-        atomic_inc(&ctx->cc_refcount);
+        cfs_atomic_inc(&ctx->cc_refcount);
         return ctx;
 }
 EXPORT_SYMBOL(sptlrpc_cli_ctx_get);
@@ -244,18 +294,19 @@ void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync)
         struct ptlrpc_sec *sec = ctx->cc_sec;
 
         LASSERT(sec);
-        LASSERT(atomic_read(&ctx->cc_refcount));
+        LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 
-        if (!atomic_dec_and_test(&ctx->cc_refcount))
+        if (!cfs_atomic_dec_and_test(&ctx->cc_refcount))
                 return;
 
         sec->ps_policy->sp_cops->release_ctx(sec, ctx, sync);
 }
 EXPORT_SYMBOL(sptlrpc_cli_ctx_put);
 
-/*
- * expire the context immediately.
- * the caller must hold at least 1 ref on the ctx.
+/**
+ * Expire the client context immediately.
+ *
+ * \pre Caller must hold at least 1 reference on the \a ctx.
  */
 void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 {
@@ -264,16 +315,21 @@ void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
 }
 EXPORT_SYMBOL(sptlrpc_cli_ctx_expire);
 
+/**
+ * To wake up the threads who are waiting for this client context. Called
+ * after some status change happened on \a ctx.
+ */
 void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx)
 {
         struct ptlrpc_request *req, *next;
 
-        spin_lock(&ctx->cc_lock);
-        list_for_each_entry_safe(req, next, &ctx->cc_req_list, rq_ctx_chain) {
-                list_del_init(&req->rq_ctx_chain);
-                ptlrpc_wake_client_req(req);
+        cfs_spin_lock(&ctx->cc_lock);
+        cfs_list_for_each_entry_safe(req, next, &ctx->cc_req_list,
+                                     rq_ctx_chain) {
+                cfs_list_del_init(&req->rq_ctx_chain);
+                ptlrpc_client_wake_req(req);
         }
-        spin_unlock(&ctx->cc_lock);
+        cfs_spin_unlock(&ctx->cc_lock);
 }
 EXPORT_SYMBOL(sptlrpc_cli_ctx_wakeup);
 
@@ -287,17 +343,17 @@ int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize)
         return ctx->cc_ops->display(ctx, buf, bufsize);
 }
 
-static int sptlrpc_import_sec_check_expire(struct obd_import *imp)
+static int import_sec_check_expire(struct obd_import *imp)
 {
         int     adapt = 0;
 
-        spin_lock(&imp->imp_lock);
+        cfs_spin_lock(&imp->imp_lock);
         if (imp->imp_sec_expire &&
             imp->imp_sec_expire < cfs_time_current_sec()) {
                 adapt = 1;
                 imp->imp_sec_expire = 0;
         }
-        spin_unlock(&imp->imp_lock);
+        cfs_spin_unlock(&imp->imp_lock);
 
         if (!adapt)
                 return 0;
@@ -306,34 +362,54 @@ static int sptlrpc_import_sec_check_expire(struct obd_import *imp)
         return sptlrpc_import_sec_adapt(imp, NULL, 0);
 }
 
-int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
+static int import_sec_validate_get(struct obd_import *imp,
+                                   struct ptlrpc_sec **sec)
 {
-        struct obd_import *imp = req->rq_import;
-        struct ptlrpc_sec *sec;
-        int                rc;
-        ENTRY;
-
-        LASSERT(!req->rq_cli_ctx);
-        LASSERT(imp);
+        int     rc;
 
         if (unlikely(imp->imp_sec_expire)) {
-                rc = sptlrpc_import_sec_check_expire(imp);
+                rc = import_sec_check_expire(imp);
                 if (rc)
-                        RETURN(rc);
+                        return rc;
         }
 
-        sec = sptlrpc_import_sec_ref(imp);
-        if (sec == NULL) {
-                CERROR("import %p (%s) with no ptlrpc_sec\n",
+        *sec = sptlrpc_import_sec_ref(imp);
+        if (*sec == NULL) {
+                CERROR("import %p (%s) with no sec\n",
                        imp, ptlrpc_import_state_name(imp->imp_state));
-                RETURN(-EACCES);
+                return -EACCES;
         }
 
-        if (unlikely(sec->ps_dying)) {
+        if (unlikely((*sec)->ps_dying)) {
                 CERROR("attempt to use dying sec %p\n", sec);
+                sptlrpc_sec_put(*sec);
                 return -EACCES;
         }
 
+        return 0;
+}
+
+/**
+ * Given a \a req, find or allocate a appropriate context for it.
+ * \pre req->rq_cli_ctx == NULL.
+ *
+ * \retval 0 succeed, and req->rq_cli_ctx is set.
+ * \retval -ev error number, and req->rq_cli_ctx == NULL.
+ */
+int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
+{
+        struct obd_import *imp = req->rq_import;
+        struct ptlrpc_sec *sec;
+        int                rc;
+        ENTRY;
+
+        LASSERT(!req->rq_cli_ctx);
+        LASSERT(imp);
+
+        rc = import_sec_validate_get(imp, &sec);
+        if (rc)
+                RETURN(rc);
+
         req->rq_cli_ctx = get_my_ctx(sec);
 
         sptlrpc_sec_put(sec);
@@ -346,9 +422,14 @@ int sptlrpc_req_get_ctx(struct ptlrpc_request *req)
         RETURN(0);
 }
 
-/*
- * if @sync == 0, this function should return quickly without sleep;
- * otherwise might trigger ctx destroying rpc to server.
+/**
+ * Drop the context for \a req.
+ * \pre req->rq_cli_ctx != NULL.
+ * \post req->rq_cli_ctx == NULL.
+ *
+ * If \a sync == 0, this function should return quickly without sleep;
+ * otherwise it might trigger and wait for the whole process of sending
+ * an context-destroying rpc to server.
  */
 void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
 {
@@ -360,10 +441,10 @@ void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync)
         /* request might be asked to release earlier while still
          * in the context waiting list.
          */
-        if (!list_empty(&req->rq_ctx_chain)) {
-                spin_lock(&req->rq_cli_ctx->cc_lock);
-                list_del_init(&req->rq_ctx_chain);
-                spin_unlock(&req->rq_cli_ctx->cc_lock);
+        if (!cfs_list_empty(&req->rq_ctx_chain)) {
+                cfs_spin_lock(&req->rq_cli_ctx->cc_lock);
+                cfs_list_del_init(&req->rq_ctx_chain);
+                cfs_spin_unlock(&req->rq_cli_ctx->cc_lock);
         }
 
         sptlrpc_cli_ctx_put(req->rq_cli_ctx, sync);
@@ -377,31 +458,32 @@ int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
                            struct ptlrpc_cli_ctx *newctx)
 {
         struct sptlrpc_flavor   old_flvr;
-        char                   *reqmsg;
+        char                   *reqmsg = NULL; /* to workaround old gcc */
         int                     reqmsg_size;
-        int                     rc;
-
-        if (likely(oldctx->cc_sec == newctx->cc_sec))
-                return 0;
+        int                     rc = 0;
 
         LASSERT(req->rq_reqmsg);
         LASSERT(req->rq_reqlen);
         LASSERT(req->rq_replen);
 
-        CWARN("req %p: switch ctx %p -> %p, switch sec %p(%s) -> %p(%s)\n",
-              req, oldctx, newctx,
-              oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
-              newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
+        CDEBUG(D_SEC, "req %p: switch ctx %p(%u->%s) -> %p(%u->%s), "
+               "switch sec %p(%s) -> %p(%s)\n", req,
+               oldctx, oldctx->cc_vcred.vc_uid, sec2target_str(oldctx->cc_sec),
+               newctx, newctx->cc_vcred.vc_uid, sec2target_str(newctx->cc_sec),
+               oldctx->cc_sec, oldctx->cc_sec->ps_policy->sp_name,
+               newctx->cc_sec, newctx->cc_sec->ps_policy->sp_name);
 
         /* save flavor */
         old_flvr = req->rq_flvr;
 
         /* save request message */
         reqmsg_size = req->rq_reqlen;
-        OBD_ALLOC(reqmsg, reqmsg_size);
-        if (reqmsg == NULL)
-                return -ENOMEM;
-        memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
+        if (reqmsg_size != 0) {
+                OBD_ALLOC_LARGE(reqmsg, reqmsg_size);
+                if (reqmsg == NULL)
+                        return -ENOMEM;
+                memcpy(reqmsg, req->rq_reqmsg, reqmsg_size);
+        }
 
         /* release old req/rep buf */
         req->rq_cli_ctx = oldctx;
@@ -414,24 +496,29 @@ int sptlrpc_req_ctx_switch(struct ptlrpc_request *req,
 
         /* alloc new request buffer
          * we don't need to alloc reply buffer here, leave it to the
-         * rest procedure of ptlrpc
-         */
-        rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
-        if (!rc) {
-                LASSERT(req->rq_reqmsg);
-                memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
-        } else {
-                CWARN("failed to alloc reqbuf: %d\n", rc);
-                req->rq_flvr = old_flvr;
-        }
+         * rest procedure of ptlrpc */
+        if (reqmsg_size != 0) {
+                rc = sptlrpc_cli_alloc_reqbuf(req, reqmsg_size);
+                if (!rc) {
+                        LASSERT(req->rq_reqmsg);
+                        memcpy(req->rq_reqmsg, reqmsg, reqmsg_size);
+                } else {
+                        CWARN("failed to alloc reqbuf: %d\n", rc);
+                        req->rq_flvr = old_flvr;
+                }
 
-        OBD_FREE(reqmsg, reqmsg_size);
+                OBD_FREE_LARGE(reqmsg, reqmsg_size);
+        }
         return rc;
 }
 
-/*
- * request must have a context. in any case of failure, restore the
- * restore the old one. a request must have a ctx.
+/**
+ * If current context of \a req is dead somehow, e.g. we just switched flavor
+ * thus marked original contexts dead, we'll find a new context for it. if
+ * no switch is needed, \a req will end up with the same context.
+ *
+ * \note a request must have a context, to keep other parts of code happy.
+ * In any case of failure during the switching, we must restore the old one.
  */
 int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
 {
@@ -441,7 +528,6 @@ int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
         ENTRY;
 
         LASSERT(oldctx);
-        LASSERT(test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags));
 
         sptlrpc_cli_ctx_get(oldctx);
         sptlrpc_req_put_ctx(req, 0);
@@ -458,15 +544,22 @@ int sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req)
         newctx = req->rq_cli_ctx;
         LASSERT(newctx);
 
-        if (unlikely(newctx == oldctx)) {
+        if (unlikely(newctx == oldctx && 
+                     cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &oldctx->cc_flags))) {
                 /*
-                 * still get the old ctx, usually means system busy
+                 * still get the old dead ctx, usually means system too busy
                  */
-                CWARN("ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
-                      newctx, newctx->cc_flags);
+                CDEBUG(D_SEC,
+                       "ctx (%p, fl %lx) doesn't switch, relax a little bit\n",
+                       newctx, newctx->cc_flags);
 
-                schedule_timeout(HZ);
+                cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
+                                                   CFS_HZ);
         } else {
+                /*
+                 * it's possible newctx == oldctx if we're switching
+                 * subflavor with the same sec.
+                 */
                 rc = sptlrpc_req_ctx_switch(req, oldctx, newctx);
                 if (rc) {
                         /* restore old ctx */
@@ -500,7 +593,7 @@ int ctx_refresh_timeout(void *data)
         /* conn_cnt is needed in expire_one_request */
         lustre_msg_set_conn_cnt(req->rq_reqmsg, req->rq_import->imp_conn_cnt);
 
-        rc = ptlrpc_expire_one_request(req);
+        rc = ptlrpc_expire_one_request(req, 1);
         /* if we started recovery, we should mark this ctx dead; otherwise
          * in case of lgssd died nobody would retire this ctx, following
          * connecting will still find the same ctx thus cause deadlock.
@@ -517,56 +610,74 @@ void ctx_refresh_interrupt(void *data)
 {
         struct ptlrpc_request *req = data;
 
-        spin_lock(&req->rq_lock);
+        cfs_spin_lock(&req->rq_lock);
         req->rq_intr = 1;
-        spin_unlock(&req->rq_lock);
+        cfs_spin_unlock(&req->rq_lock);
 }
 
 static
 void req_off_ctx_list(struct ptlrpc_request *req, struct ptlrpc_cli_ctx *ctx)
 {
-        spin_lock(&ctx->cc_lock);
-        if (!list_empty(&req->rq_ctx_chain))
-                list_del_init(&req->rq_ctx_chain);
-        spin_unlock(&ctx->cc_lock);
+        cfs_spin_lock(&ctx->cc_lock);
+        if (!cfs_list_empty(&req->rq_ctx_chain))
+                cfs_list_del_init(&req->rq_ctx_chain);
+        cfs_spin_unlock(&ctx->cc_lock);
 }
 
-/*
- * the status of context could be subject to be changed by other threads at any
- * time. we allow this race. but once we return with 0, the caller will
- * suppose it's uptodated and keep using it until the owning rpc is done.
+/**
+ * To refresh the context of \req, if it's not up-to-date.
+ * \param timeout
+ * - < 0: don't wait
+ * - = 0: wait until success or fatal error occur
+ * - > 0: timeout value (in seconds)
  *
- * @timeout:
- *    < 0  - don't wait
- *    = 0  - wait until success or fatal error occur
- *    > 0  - timeout value
+ * The status of the context could be subject to be changed by other threads
+ * at any time. We allow this race, but once we return with 0, the caller will
+ * suppose it's uptodated and keep using it until the owning rpc is done.
  *
- * return 0 only if the context is uptodated.
+ * \retval 0 only if the context is uptodated.
+ * \retval -ev error number.
  */
 int sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout)
 {
         struct ptlrpc_cli_ctx  *ctx = req->rq_cli_ctx;
+        struct ptlrpc_sec      *sec;
         struct l_wait_info      lwi;
         int                     rc;
         ENTRY;
 
         LASSERT(ctx);
 
+        if (req->rq_ctx_init || req->rq_ctx_fini)
+                RETURN(0);
+
         /*
          * during the process a request's context might change type even
-         * (e.g. from gss ctx to plain ctx), so each loop we need to re-check
+         * (e.g. from gss ctx to null ctx), so each loop we need to re-check
          * everything
          */
 again:
-        /* skip special ctxs */
-        if (cli_ctx_is_eternal(ctx) || req->rq_ctx_init || req->rq_ctx_fini)
+        rc = import_sec_validate_get(req->rq_import, &sec);
+        if (rc)
+                RETURN(rc);
+
+        if (sec->ps_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
+                CDEBUG(D_SEC, "req %p: flavor has changed %x -> %x\n",
+                      req, req->rq_flvr.sf_rpc, sec->ps_flvr.sf_rpc);
+                req_off_ctx_list(req, ctx);
+                sptlrpc_req_replace_dead_ctx(req);
+                ctx = req->rq_cli_ctx;
+        }
+        sptlrpc_sec_put(sec);
+
+        if (cli_ctx_is_eternal(ctx))
                 RETURN(0);
 
-        if (test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags)) {
+        if (unlikely(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags))) {
                 LASSERT(ctx->cc_ops->refresh);
                 ctx->cc_ops->refresh(ctx);
         }
-        LASSERT(test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
+        LASSERT(cfs_test_bit(PTLRPC_CTX_NEW_BIT, &ctx->cc_flags) == 0);
 
         LASSERT(ctx->cc_ops->validate);
         if (ctx->cc_ops->validate(ctx) == 0) {
@@ -574,97 +685,108 @@ again:
                 RETURN(0);
         }
 
-        if (unlikely(test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
+        if (unlikely(cfs_test_bit(PTLRPC_CTX_ERROR_BIT, &ctx->cc_flags))) {
+                cfs_spin_lock(&req->rq_lock);
                 req->rq_err = 1;
+                cfs_spin_unlock(&req->rq_lock);
                 req_off_ctx_list(req, ctx);
                 RETURN(-EPERM);
         }
 
-        /* This is subtle. For resent message we have to keep original
-         * context to survive following situation:
-         *  1. the request sent to server
-         *  2. recovery was kick start
-         *  3. recovery finished, the request marked as resent
-         *  4. resend the request
-         *  5. old reply from server received (because xid is the same)
-         *  6. verify reply (has to be success)
-         *  7. new reply from server received, lnet drop it
+        /*
+         * There's a subtle issue for resending RPCs, suppose following
+         * situation:
+         *  1. the request was sent to server.
+         *  2. recovery was kicked start, after finished the request was
+         *     marked as resent.
+         *  3. resend the request.
+         *  4. old reply from server received, we accept and verify the reply.
+         *     this has to be success, otherwise the error will be aware
+         *     by application.
+         *  5. new reply from server received, dropped by LNet.
          *
-         * Note we can't simply change xid for resent request because
-         * server reply on it for reply reconstruction.
+         * Note the xid of old & new request is the same. We can't simply
+         * change xid for the resent request because the server replies on
+         * it for reply reconstruction.
          *
          * Commonly the original context should be uptodate because we
-         * have a expiry nice time; And server will keep their half part
-         * context because we at least hold a ref of old context which
-         * prevent the context detroy RPC be sent. So server still can
-         * accept the request and finish RPC. Two cases:
-         *  1. If server side context has been trimed, a NO_CONTEXT will
+         * have a expiry nice time; server will keep its context because
+         * we at least hold a ref of old context which prevent context
+         * destroying RPC being sent. So server still can accept the request
+         * and finish the RPC. But if that's not the case:
+         *  1. If server side context has been trimmed, a NO_CONTEXT will
          *     be returned, gss_cli_ctx_verify/unseal will switch to new
          *     context by force.
          *  2. Current context never be refreshed, then we are fine: we
          *     never really send request with old context before.
          */
-        if (test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
+        if (cfs_test_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags) &&
             unlikely(req->rq_reqmsg) &&
             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
                 req_off_ctx_list(req, ctx);
                 RETURN(0);
         }
 
-        if (unlikely(test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
+        if (unlikely(cfs_test_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags))) {
+                req_off_ctx_list(req, ctx);
+                /*
+                 * don't switch ctx if import was deactivated
+                 */
+                if (req->rq_import->imp_deactive) {
+                        cfs_spin_lock(&req->rq_lock);
+                        req->rq_err = 1;
+                        cfs_spin_unlock(&req->rq_lock);
+                        RETURN(-EINTR);
+                }
+
                 rc = sptlrpc_req_replace_dead_ctx(req);
                 if (rc) {
                         LASSERT(ctx == req->rq_cli_ctx);
                         CERROR("req %p: failed to replace dead ctx %p: %d\n",
                                 req, ctx, rc);
+                        cfs_spin_lock(&req->rq_lock);
                         req->rq_err = 1;
-                        LASSERT(list_empty(&req->rq_ctx_chain));
+                        cfs_spin_unlock(&req->rq_lock);
                         RETURN(rc);
                 }
 
-                CWARN("req %p: replace dead ctx %p => ctx %p (%u->%s)\n",
-                      req, ctx, req->rq_cli_ctx,
-                      req->rq_cli_ctx->cc_vcred.vc_uid,
-                      sec2target_str(req->rq_cli_ctx->cc_sec));
-
                 ctx = req->rq_cli_ctx;
-                LASSERT(list_empty(&req->rq_ctx_chain));
-
                 goto again;
         }
 
-        /* Now we're sure this context is during upcall, add myself into
+        /*
+         * Now we're sure this context is during upcall, add myself into
          * waiting list
          */
-        spin_lock(&ctx->cc_lock);
-        if (list_empty(&req->rq_ctx_chain))
-                list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
-        spin_unlock(&ctx->cc_lock);
+        cfs_spin_lock(&ctx->cc_lock);
+        if (cfs_list_empty(&req->rq_ctx_chain))
+                cfs_list_add(&req->rq_ctx_chain, &ctx->cc_req_list);
+        cfs_spin_unlock(&ctx->cc_lock);
 
-        if (timeout < 0) {
+        if (timeout < 0)
                 RETURN(-EWOULDBLOCK);
-        }
 
         /* Clear any flags that may be present from previous sends */
         LASSERT(req->rq_receiving_reply == 0);
-        spin_lock(&req->rq_lock);
+        cfs_spin_lock(&req->rq_lock);
         req->rq_err = 0;
         req->rq_timedout = 0;
         req->rq_resend = 0;
         req->rq_restart = 0;
-        spin_unlock(&req->rq_lock);
+        cfs_spin_unlock(&req->rq_lock);
 
-        lwi = LWI_TIMEOUT_INTR(timeout * HZ, ctx_refresh_timeout,
+        lwi = LWI_TIMEOUT_INTR(timeout * CFS_HZ, ctx_refresh_timeout,
                                ctx_refresh_interrupt, req);
         rc = l_wait_event(req->rq_reply_waitq, ctx_check_refresh(ctx), &lwi);
 
-        /* following cases we could be here:
+        /*
+         * following cases could lead us here:
          * - successfully refreshed;
-         * - interruptted;
+         * - interrupted;
          * - timedout, and we don't want recover from the failure;
          * - timedout, and waked up upon recovery finished;
          * - someone else mark this ctx dead by force;
-         * - someone invalidate the req and call wake_client_req(),
+         * - someone invalidate the req and call ptlrpc_client_wake_req(),
          *   e.g. ptlrpc_abort_inflight();
          */
         if (!cli_ctx_is_refreshed(ctx)) {
@@ -678,8 +800,10 @@ again:
         goto again;
 }
 
-/*
- * Note this could be called in two situations:
+/**
+ * Initialize flavor settings for \a req, according to \a opcode.
+ *
+ * \note this could be called in two situations:
  * - new request from ptlrpc_pre_req(), with proper @opcode
  * - old request which changed ctx in the middle, with @opcode == 0
  */
@@ -695,9 +819,13 @@ void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
         /* special security flags accoding to opcode */
         switch (opcode) {
         case OST_READ:
+        case MDS_READPAGE:
+        case MGS_CONFIG_READ:
+       case OBD_IDX_READ:
                 req->rq_bulk_read = 1;
                 break;
         case OST_WRITE:
+        case MDS_WRITEPAGE:
                 req->rq_bulk_write = 1;
                 break;
         case SEC_CTX_INIT:
@@ -719,16 +847,16 @@ void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 
         sec = req->rq_cli_ctx->cc_sec;
 
-        spin_lock(&sec->ps_lock);
+        cfs_spin_lock(&sec->ps_lock);
         req->rq_flvr = sec->ps_flvr;
-        spin_unlock(&sec->ps_lock);
+        cfs_spin_unlock(&sec->ps_lock);
 
         /* force SVC_NULL for context initiation rpc, SVC_INTG for context
          * destruction rpc */
         if (unlikely(req->rq_ctx_init))
-                rpc_flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
+                flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_NULL);
         else if (unlikely(req->rq_ctx_fini))
-                rpc_flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
+                flvr_set_svc(&req->rq_flvr.sf_rpc, SPTLRPC_SVC_INTG);
 
         /* user descriptor flag, null security can't do it anyway */
         if ((sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC) &&
@@ -737,14 +865,13 @@ void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode)
 
         /* bulk security flag */
         if ((req->rq_bulk_read || req->rq_bulk_write) &&
-            (req->rq_flvr.sf_bulk_ciph != BULK_CIPH_ALG_NULL ||
-             req->rq_flvr.sf_bulk_hash != BULK_HASH_ALG_NULL))
+            sptlrpc_flavor_has_bulk(&req->rq_flvr))
                 req->rq_pack_bulk = 1;
 }
 
 void sptlrpc_request_out_callback(struct ptlrpc_request *req)
 {
-        if (RPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
+        if (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) != SPTLRPC_SVC_PRIV)
                 return;
 
         LASSERT(req->rq_clrbuf);
@@ -756,10 +883,10 @@ void sptlrpc_request_out_callback(struct ptlrpc_request *req)
         req->rq_reqbuf_len = 0;
 }
 
-/*
- * check whether current user have valid context for an import or not.
- * might repeatedly try in case of non-fatal errors.
- * return 0 on success, < 0 on failure
+/**
+ * Given an import \a imp, check whether current user has a valid context
+ * or not. We may create a new context and try to refresh it, and try
+ * repeatedly try in case of non-fatal errors. Return 0 means success.
  */
 int sptlrpc_import_check_ctx(struct obd_import *imp)
 {
@@ -769,14 +896,14 @@ int sptlrpc_import_check_ctx(struct obd_import *imp)
         int rc;
         ENTRY;
 
-        might_sleep();
+        cfs_might_sleep();
 
         sec = sptlrpc_import_sec_ref(imp);
         ctx = get_my_ctx(sec);
         sptlrpc_sec_put(sec);
 
         if (!ctx)
-                RETURN(1);
+                RETURN(-ENOMEM);
 
         if (cli_ctx_is_eternal(ctx) ||
             ctx->cc_ops->validate(ctx) == 0) {
@@ -784,25 +911,37 @@ int sptlrpc_import_check_ctx(struct obd_import *imp)
                 RETURN(0);
         }
 
+        if (cli_ctx_is_error(ctx)) {
+                sptlrpc_cli_ctx_put(ctx, 1);
+                RETURN(-EACCES);
+        }
+
         OBD_ALLOC_PTR(req);
         if (!req)
                 RETURN(-ENOMEM);
 
-        spin_lock_init(&req->rq_lock);
-        atomic_set(&req->rq_refcount, 10000);
+        cfs_spin_lock_init(&req->rq_lock);
+        cfs_atomic_set(&req->rq_refcount, 10000);
         CFS_INIT_LIST_HEAD(&req->rq_ctx_chain);
-        init_waitqueue_head(&req->rq_reply_waitq);
+        cfs_waitq_init(&req->rq_reply_waitq);
+        cfs_waitq_init(&req->rq_set_waitq);
         req->rq_import = imp;
+        req->rq_flvr = sec->ps_flvr;
         req->rq_cli_ctx = ctx;
 
         rc = sptlrpc_req_refresh_ctx(req, 0);
-        LASSERT(list_empty(&req->rq_ctx_chain));
+        LASSERT(cfs_list_empty(&req->rq_ctx_chain));
         sptlrpc_cli_ctx_put(req->rq_cli_ctx, 1);
         OBD_FREE_PTR(req);
 
         RETURN(rc);
 }
 
+/**
+ * Used by ptlrpc client, to perform the pre-defined security transformation
+ * upon the request message of \a req. After this function called,
+ * req->rq_reqmsg is still accessible as clear text.
+ */
 int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
 {
         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
@@ -822,7 +961,7 @@ int sptlrpc_cli_wrap_request(struct ptlrpc_request *req)
                         RETURN(rc);
         }
 
-        switch (RPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
+        switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
         case SPTLRPC_SVC_NULL:
         case SPTLRPC_SVC_AUTH:
         case SPTLRPC_SVC_INTG:
@@ -850,7 +989,6 @@ static int do_cli_unwrap_reply(struct ptlrpc_request *req)
 {
         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
         int                    rc;
-        __u16                  rpc_flvr;
         ENTRY;
 
         LASSERT(ctx);
@@ -859,33 +997,34 @@ static int do_cli_unwrap_reply(struct ptlrpc_request *req)
         LASSERT(req->rq_repdata);
         LASSERT(req->rq_repmsg == NULL);
 
+        req->rq_rep_swab_mask = 0;
+
+        rc = __lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len);
+        switch (rc) {
+        case 1:
+                lustre_set_rep_swabbed(req, MSG_PTLRPC_HEADER_OFF);
+        case 0:
+                break;
+        default:
+                CERROR("failed unpack reply: x"LPU64"\n", req->rq_xid);
+                RETURN(-EPROTO);
+        }
+
         if (req->rq_repdata_len < sizeof(struct lustre_msg)) {
                 CERROR("replied data length %d too small\n",
                        req->rq_repdata_len);
                 RETURN(-EPROTO);
         }
 
-        /* v2 message, check request/reply policy match */
-        rpc_flvr = WIRE_FLVR_RPC(req->rq_repdata->lm_secflvr);
-
-        if (req->rq_repdata->lm_magic == LUSTRE_MSG_MAGIC_V2_SWABBED)
-                __swab16s(&rpc_flvr);
-
-        if (RPC_FLVR_POLICY(rpc_flvr) !=
-            RPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
-                CERROR("request policy was %u while reply with %u\n",
-                       RPC_FLVR_POLICY(req->rq_flvr.sf_rpc),
-                       RPC_FLVR_POLICY(rpc_flvr));
+        if (SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr) !=
+            SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc)) {
+                CERROR("reply policy %u doesn't match request policy %u\n",
+                       SPTLRPC_FLVR_POLICY(req->rq_repdata->lm_secflvr),
+                       SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc));
                 RETURN(-EPROTO);
         }
 
-        /* do nothing if it's null policy; otherwise unpack the
-         * wrapper message */
-        if (RPC_FLVR_POLICY(rpc_flvr) != SPTLRPC_POLICY_NULL &&
-            lustre_unpack_msg(req->rq_repdata, req->rq_repdata_len))
-                RETURN(-EPROTO);
-
-        switch (RPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
+        switch (SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc)) {
         case SPTLRPC_SVC_NULL:
         case SPTLRPC_SVC_AUTH:
         case SPTLRPC_SVC_INTG:
@@ -899,14 +1038,21 @@ static int do_cli_unwrap_reply(struct ptlrpc_request *req)
         default:
                 LBUG();
         }
-
         LASSERT(rc || req->rq_repmsg || req->rq_resend);
+
+        if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL &&
+            !req->rq_ctx_init)
+                req->rq_rep_swab_mask = 0;
         RETURN(rc);
 }
 
-/*
- * upon this be called, the reply buffer should have been un-posted,
- * so nothing is going to change.
+/**
+ * Used by ptlrpc client, to perform security transformation upon the reply
+ * message of \a req. After return successfully, req->rq_repmsg points to
+ * the reply message in clear text.
+ *
+ * \pre the reply buffer should have been un-posted from LNet, so nothing is
+ * going to change.
  */
 int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
 {
@@ -915,7 +1061,8 @@ int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
         LASSERT(req->rq_repmsg == NULL);
         LASSERT(req->rq_reply_off + req->rq_nob_received <= req->rq_repbuf_len);
 
-        if (req->rq_reply_off == 0) {
+        if (req->rq_reply_off == 0 &&
+            (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)) {
                 CERROR("real reply with offset 0\n");
                 return -EPROTO;
         }
@@ -932,114 +1079,120 @@ int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req)
         return do_cli_unwrap_reply(req);
 }
 
-/*
- * Upon called, the receive buffer might be still posted, so the reply data
- * might be changed at any time, no matter we're holding rq_lock or not. we
- * expect the rq_reply_off be 0, rq_nob_received is the early reply size.
- *
- * we allocate a separate buffer to hold early reply data, pointed by
- * rq_repdata, rq_repdata_len is the early reply size, and round up to power2
- * is the actual buffer size.
- *
- * caller _must_ call sptlrpc_cli_finish_early_reply() after this, before
- * process another early reply or real reply, to restore ptlrpc_request
- * to normal status.
+/**
+ * Used by ptlrpc client, to perform security transformation upon the early
+ * reply message of \a req. We expect the rq_reply_off is 0, and
+ * rq_nob_received is the early reply size.
+ * 
+ * Because the receive buffer might be still posted, the reply data might be
+ * changed at any time, no matter we're holding rq_lock or not. For this reason
+ * we allocate a separate ptlrpc_request and reply buffer for early reply
+ * processing.
+ * 
+ * \retval 0 success, \a req_ret is filled with a duplicated ptlrpc_request.
+ * Later the caller must call sptlrpc_cli_finish_early_reply() on the returned
+ * \a *req_ret to release it.
+ * \retval -ev error number, and \a req_ret will not be set.
  */
-int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req)
+int sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
+                                   struct ptlrpc_request **req_ret)
 {
-        struct lustre_msg      *early_buf;
+        struct ptlrpc_request  *early_req;
+        char                   *early_buf;
         int                     early_bufsz, early_size;
         int                     rc;
         ENTRY;
 
-        LASSERT(req->rq_repbuf);
-        LASSERT(req->rq_repdata == NULL);
-        LASSERT(req->rq_repmsg == NULL);
+        OBD_ALLOC_PTR(early_req);
+        if (early_req == NULL)
+                RETURN(-ENOMEM);
 
         early_size = req->rq_nob_received;
-        if (early_size < sizeof(struct lustre_msg)) {
-                CERROR("early reply length %d too small\n", early_size);
-                RETURN(-EPROTO);
-        }
-
         early_bufsz = size_roundup_power2(early_size);
-        OBD_ALLOC(early_buf, early_bufsz);
+        OBD_ALLOC_LARGE(early_buf, early_bufsz);
         if (early_buf == NULL)
-                RETURN(-ENOMEM);
+                GOTO(err_req, rc = -ENOMEM);
 
-        /* copy data out, do it inside spinlock */
-        spin_lock(&req->rq_lock);
+        /* sanity checkings and copy data out, do it inside spinlock */
+        cfs_spin_lock(&req->rq_lock);
 
         if (req->rq_replied) {
-                spin_unlock(&req->rq_lock);
-                GOTO(err_free, rc = -EALREADY);
+                cfs_spin_unlock(&req->rq_lock);
+                GOTO(err_buf, rc = -EALREADY);
         }
 
+        LASSERT(req->rq_repbuf);
+        LASSERT(req->rq_repdata == NULL);
+        LASSERT(req->rq_repmsg == NULL);
+
         if (req->rq_reply_off != 0) {
                 CERROR("early reply with offset %u\n", req->rq_reply_off);
-                GOTO(err_free, rc = -EPROTO);
+                cfs_spin_unlock(&req->rq_lock);
+                GOTO(err_buf, rc = -EPROTO);
         }
 
         if (req->rq_nob_received != early_size) {
                 /* even another early arrived the size should be the same */
-                CWARN("data size has changed from %u to %u\n",
-                      early_size, req->rq_nob_received);
-                spin_unlock(&req->rq_lock);
-                GOTO(err_free, rc = -EINVAL);
+                CERROR("data size has changed from %u to %u\n",
+                       early_size, req->rq_nob_received);
+                cfs_spin_unlock(&req->rq_lock);
+                GOTO(err_buf, rc = -EINVAL);
         }
 
         if (req->rq_nob_received < sizeof(struct lustre_msg)) {
                 CERROR("early reply length %d too small\n",
                        req->rq_nob_received);
-                spin_unlock(&req->rq_lock);
-                GOTO(err_free, rc = -EALREADY);
+                cfs_spin_unlock(&req->rq_lock);
+                GOTO(err_buf, rc = -EALREADY);
         }
 
         memcpy(early_buf, req->rq_repbuf, early_size);
-        spin_unlock(&req->rq_lock);
-
-        req->rq_repdata = early_buf;
-        req->rq_repdata_len = early_size;
-
-        rc = do_cli_unwrap_reply(req);
-
-        /* treate resend as an error case. in fact server should never ask
-         * resend via early reply. */
-        if (req->rq_resend) {
-                req->rq_resend = 0;
-                rc = -EPROTO;
-        }
-
+        cfs_spin_unlock(&req->rq_lock);
+
+        cfs_spin_lock_init(&early_req->rq_lock);
+        early_req->rq_cli_ctx = sptlrpc_cli_ctx_get(req->rq_cli_ctx);
+        early_req->rq_flvr = req->rq_flvr;
+        early_req->rq_repbuf = early_buf;
+        early_req->rq_repbuf_len = early_bufsz;
+        early_req->rq_repdata = (struct lustre_msg *) early_buf;
+        early_req->rq_repdata_len = early_size;
+        early_req->rq_early = 1;
+        early_req->rq_reqmsg = req->rq_reqmsg;
+
+        rc = do_cli_unwrap_reply(early_req);
         if (rc) {
-                LASSERT(req->rq_repmsg == NULL);
-                req->rq_repdata = NULL;
-                req->rq_repdata_len = 0;
-                GOTO(err_free, rc);
+                DEBUG_REQ(D_ADAPTTO, early_req,
+                          "error %d unwrap early reply", rc);
+                GOTO(err_ctx, rc);
         }
 
-        LASSERT(req->rq_repmsg);
+        LASSERT(early_req->rq_repmsg);
+        *req_ret = early_req;
         RETURN(0);
 
-err_free:
-        OBD_FREE(early_buf, early_bufsz);
+err_ctx:
+        sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
+err_buf:
+        OBD_FREE_LARGE(early_buf, early_bufsz);
+err_req:
+        OBD_FREE_PTR(early_req);
         RETURN(rc);
 }
 
-int sptlrpc_cli_finish_early_reply(struct ptlrpc_request *req)
+/**
+ * Used by ptlrpc client, to release a processed early reply \a early_req.
+ *
+ * \pre \a early_req was obtained from calling sptlrpc_cli_unwrap_early_reply().
+ */
+void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req)
 {
-        int     early_bufsz;
-
-        LASSERT(req->rq_repdata);
-        LASSERT(req->rq_repdata_len);
-        LASSERT(req->rq_repmsg);
+        LASSERT(early_req->rq_repbuf);
+        LASSERT(early_req->rq_repdata);
+        LASSERT(early_req->rq_repmsg);
 
-        early_bufsz = size_roundup_power2(req->rq_repdata_len);
-        OBD_FREE(req->rq_repdata, early_bufsz);
-
-        req->rq_repdata = NULL;
-        req->rq_repdata_len = 0;
-        req->rq_repmsg = NULL;
-        return 0;
+        sptlrpc_cli_ctx_put(early_req->rq_cli_ctx, 1);
+        OBD_FREE_LARGE(early_req->rq_repbuf, early_req->rq_repbuf_len);
+        OBD_FREE_PTR(early_req);
 }
 
 /**************************************************
@@ -1049,11 +1202,11 @@ int sptlrpc_cli_finish_early_reply(struct ptlrpc_request *req)
 /*
  * "fixed" sec (e.g. null) use sec_id < 0
  */
-static atomic_t sptlrpc_sec_id = ATOMIC_INIT(1);
+static cfs_atomic_t sptlrpc_sec_id = CFS_ATOMIC_INIT(1);
 
 int sptlrpc_get_next_secid(void)
 {
-        return atomic_inc_return(&sptlrpc_sec_id);
+        return cfs_atomic_inc_return(&sptlrpc_sec_id);
 }
 EXPORT_SYMBOL(sptlrpc_get_next_secid);
 
@@ -1076,8 +1229,8 @@ static void sec_cop_destroy_sec(struct ptlrpc_sec *sec)
 {
         struct ptlrpc_sec_policy *policy = sec->ps_policy;
 
-        LASSERT(atomic_read(&sec->ps_refcount) == 0);
-        LASSERT(atomic_read(&sec->ps_nctx) == 0);
+        LASSERT_ATOMIC_ZERO(&sec->ps_refcount);
+        LASSERT_ATOMIC_ZERO(&sec->ps_nctx);
         LASSERT(policy->sp_cops->destroy_sec);
 
         CDEBUG(D_SEC, "%s@%p: being destroied\n", sec->ps_policy->sp_name, sec);
@@ -1094,7 +1247,7 @@ EXPORT_SYMBOL(sptlrpc_sec_destroy);
 
 static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
 {
-        LASSERT(atomic_read(&sec->ps_refcount) > 0);
+        LASSERT_ATOMIC_POS(&sec->ps_refcount);
 
         if (sec->ps_policy->sp_cops->kill_sec) {
                 sec->ps_policy->sp_cops->kill_sec(sec);
@@ -1105,10 +1258,8 @@ static void sptlrpc_sec_kill(struct ptlrpc_sec *sec)
 
 struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec)
 {
-        if (sec) {
-                LASSERT(atomic_read(&sec->ps_refcount) > 0);
-                atomic_inc(&sec->ps_refcount);
-        }
+        if (sec)
+                cfs_atomic_inc(&sec->ps_refcount);
 
         return sec;
 }
@@ -1117,11 +1268,9 @@ EXPORT_SYMBOL(sptlrpc_sec_get);
 void sptlrpc_sec_put(struct ptlrpc_sec *sec)
 {
         if (sec) {
-                LASSERT(atomic_read(&sec->ps_refcount) > 0);
-
-                if (atomic_dec_and_test(&sec->ps_refcount)) {
-                        LASSERT(atomic_read(&sec->ps_nctx) == 0);
+                LASSERT_ATOMIC_POS(&sec->ps_refcount);
 
+                if (cfs_atomic_dec_and_test(&sec->ps_refcount)) {
                         sptlrpc_gc_del_sec(sec);
                         sec_cop_destroy_sec(sec);
                 }
@@ -1130,7 +1279,7 @@ void sptlrpc_sec_put(struct ptlrpc_sec *sec)
 EXPORT_SYMBOL(sptlrpc_sec_put);
 
 /*
- * it's policy module responsible for taking refrence of import
+ * policy module is responsible for taking refrence of import
  */
 static
 struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
@@ -1140,6 +1289,7 @@ struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
 {
         struct ptlrpc_sec_policy *policy;
         struct ptlrpc_sec        *sec;
+        char                      str[32];
         ENTRY;
 
         if (svc_ctx) {
@@ -1148,7 +1298,7 @@ struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
                 CDEBUG(D_SEC, "%s %s: reverse sec using flavor %s\n",
                        imp->imp_obd->obd_type->typ_name,
                        imp->imp_obd->obd_name,
-                       sptlrpc_rpcflavor2name(sf->sf_rpc));
+                       sptlrpc_flavor2name(sf, str, sizeof(str)));
 
                 policy = sptlrpc_policy_get(svc_ctx->sc_policy);
                 sf->sf_flags |= PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
@@ -1158,9 +1308,9 @@ struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
                 CDEBUG(D_SEC, "%s %s: select security flavor %s\n",
                        imp->imp_obd->obd_type->typ_name,
                        imp->imp_obd->obd_name,
-                       sptlrpc_rpcflavor2name(sf->sf_rpc));
+                       sptlrpc_flavor2name(sf, str, sizeof(str)));
 
-                policy = sptlrpc_rpcflavor2policy(sf->sf_rpc);
+                policy = sptlrpc_wireflavor2policy(sf->sf_rpc);
                 if (!policy) {
                         CERROR("invalid flavor 0x%x\n", sf->sf_rpc);
                         RETURN(NULL);
@@ -1169,7 +1319,7 @@ struct ptlrpc_sec * sptlrpc_sec_create(struct obd_import *imp,
 
         sec = policy->sp_cops->create_sec(imp, svc_ctx, sf);
         if (sec) {
-                atomic_inc(&sec->ps_refcount);
+                cfs_atomic_inc(&sec->ps_refcount);
 
                 sec->ps_part = sp;
 
@@ -1186,9 +1336,9 @@ struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp)
 {
         struct ptlrpc_sec *sec;
 
-        spin_lock(&imp->imp_lock);
+        cfs_spin_lock(&imp->imp_lock);
         sec = sptlrpc_sec_get(imp->imp_sec);
-        spin_unlock(&imp->imp_lock);
+        cfs_spin_unlock(&imp->imp_lock);
 
         return sec;
 }
@@ -1199,12 +1349,12 @@ static void sptlrpc_import_sec_install(struct obd_import *imp,
 {
         struct ptlrpc_sec *old_sec;
 
-        LASSERT(atomic_read(&sec->ps_refcount) > 0);
+        LASSERT_ATOMIC_POS(&sec->ps_refcount);
 
-        spin_lock(&imp->imp_lock);
+        cfs_spin_lock(&imp->imp_lock);
         old_sec = imp->imp_sec;
         imp->imp_sec = sec;
-        spin_unlock(&imp->imp_lock);
+        cfs_spin_unlock(&imp->imp_lock);
 
         if (old_sec) {
                 sptlrpc_sec_kill(old_sec);
@@ -1214,138 +1364,133 @@ static void sptlrpc_import_sec_install(struct obd_import *imp,
         }
 }
 
+static inline
+int flavor_equal(struct sptlrpc_flavor *sf1, struct sptlrpc_flavor *sf2)
+{
+        return (memcmp(sf1, sf2, sizeof(*sf1)) == 0);
+}
+
+static inline
+void flavor_copy(struct sptlrpc_flavor *dst, struct sptlrpc_flavor *src)
+{
+        *dst = *src;
+}
+
 static void sptlrpc_import_sec_adapt_inplace(struct obd_import *imp,
                                              struct ptlrpc_sec *sec,
                                              struct sptlrpc_flavor *sf)
 {
-        if (sf->sf_bulk_ciph != sec->ps_flvr.sf_bulk_ciph ||
-            sf->sf_bulk_hash != sec->ps_flvr.sf_bulk_hash) {
-                CWARN("imp %p (%s->%s): changing bulk flavor %s/%s -> %s/%s\n",
-                      imp, imp->imp_obd->obd_name,
-                      obd_uuid2str(&imp->imp_connection->c_remote_uuid),
-                      sptlrpc_get_ciph_name(sec->ps_flvr.sf_bulk_ciph),
-                      sptlrpc_get_hash_name(sec->ps_flvr.sf_bulk_hash),
-                      sptlrpc_get_ciph_name(sf->sf_bulk_ciph),
-                      sptlrpc_get_hash_name(sf->sf_bulk_hash));
-
-                spin_lock(&sec->ps_lock);
-                sec->ps_flvr.sf_bulk_ciph = sf->sf_bulk_ciph;
-                sec->ps_flvr.sf_bulk_hash = sf->sf_bulk_hash;
-                spin_unlock(&sec->ps_lock);
-        }
+        char    str1[32], str2[32];
 
-        if (!equi(sf->sf_flags & PTLRPC_SEC_FL_UDESC,
-                  sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_UDESC)) {
-                CWARN("imp %p (%s->%s): %s shipping user descriptor\n",
-                      imp, imp->imp_obd->obd_name,
-                      obd_uuid2str(&imp->imp_connection->c_remote_uuid),
-                      (sf->sf_flags & PTLRPC_SEC_FL_UDESC) ? "start" : "stop");
+        if (sec->ps_flvr.sf_flags != sf->sf_flags)
+                CDEBUG(D_SEC, "changing sec flags: %s -> %s\n",
+                       sptlrpc_secflags2str(sec->ps_flvr.sf_flags,
+                                            str1, sizeof(str1)),
+                       sptlrpc_secflags2str(sf->sf_flags,
+                                            str2, sizeof(str2)));
 
-                spin_lock(&sec->ps_lock);
-                sec->ps_flvr.sf_flags &= ~PTLRPC_SEC_FL_UDESC;
-                sec->ps_flvr.sf_flags |= sf->sf_flags & PTLRPC_SEC_FL_UDESC;
-                spin_unlock(&sec->ps_lock);
-        }
+        cfs_spin_lock(&sec->ps_lock);
+        flavor_copy(&sec->ps_flvr, sf);
+        cfs_spin_unlock(&sec->ps_lock);
 }
 
-/*
- * for normal import, @svc_ctx should be NULL and @rpc_flavor is ignored;
- * for reverse import, @svc_ctx and @rpc_flavor is from incoming request.
+/**
+ * To get an appropriate ptlrpc_sec for the \a imp, according to the current
+ * configuration. Upon called, imp->imp_sec may or may not be NULL.
+ *
+ *  - regular import: \a svc_ctx should be NULL and \a flvr is ignored;
+ *  - reverse import: \a svc_ctx and \a flvr are obtained from incoming request.
  */
 int sptlrpc_import_sec_adapt(struct obd_import *imp,
                              struct ptlrpc_svc_ctx *svc_ctx,
-                             __u16 rpc_flavor)
+                             struct sptlrpc_flavor *flvr)
 {
         struct ptlrpc_connection   *conn;
         struct sptlrpc_flavor       sf;
         struct ptlrpc_sec          *sec, *newsec;
         enum lustre_sec_part        sp;
-        int                         rc;
+        char                        str[24];
+        int                         rc = 0;
+        ENTRY;
+
+        cfs_might_sleep();
 
         if (imp == NULL)
-                return 0;
+                RETURN(0);
 
         conn = imp->imp_connection;
 
         if (svc_ctx == NULL) {
-                /* normal import, determine flavor from rule set */
-                sptlrpc_rule_set_choose(&imp->imp_obd->u.cli.cl_sptlrpc_rset,
-                                        LUSTRE_SP_ANY, conn->c_self, &sf);
-
-                sp = imp->imp_obd->u.cli.cl_sec_part;
+                struct client_obd *cliobd = &imp->imp_obd->u.cli;
+                /*
+                 * normal import, determine flavor from rule set, except
+                 * for mgc the flavor is predetermined.
+                 */
+                if (cliobd->cl_sp_me == LUSTRE_SP_MGC)
+                        sf = cliobd->cl_flvr_mgc;
+                else 
+                        sptlrpc_conf_choose_flavor(cliobd->cl_sp_me,
+                                                   cliobd->cl_sp_to,
+                                                   &cliobd->cl_target_uuid,
+                                                   conn->c_self, &sf);
+
+                sp = imp->imp_obd->u.cli.cl_sp_me;
         } else {
                 /* reverse import, determine flavor from incoming reqeust */
-                sf.sf_rpc = rpc_flavor;
-                sf.sf_bulk_ciph = BULK_CIPH_ALG_NULL;
-                sf.sf_bulk_hash = BULK_HASH_ALG_NULL;
-                sf.sf_flags = PTLRPC_SEC_FL_REVERSE | PTLRPC_SEC_FL_ROOTONLY;
+                sf = *flvr;
+
+                if (sf.sf_rpc != SPTLRPC_FLVR_NULL)
+                        sf.sf_flags = PTLRPC_SEC_FL_REVERSE |
+                                      PTLRPC_SEC_FL_ROOTONLY;
 
                 sp = sptlrpc_target_sec_part(imp->imp_obd);
         }
 
         sec = sptlrpc_import_sec_ref(imp);
         if (sec) {
-                if (svc_ctx == NULL) {
-                        /* normal import, only check rpc flavor, if just bulk
-                         * flavor or flags changed, we can handle it on the fly
-                         * without switching sec. */
-                        if (sf.sf_rpc == sec->ps_flvr.sf_rpc) {
-                                sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
-
-                                rc = 0;
-                                goto out;
-                        }
-                } else {
-                        /* reverse import, do not compare bulk flavor */
-                        if (sf.sf_rpc == sec->ps_flvr.sf_rpc) {
-                                rc = 0;
-                                goto out;
-                        }
-                }
+                char    str2[24];
 
-                CWARN("%simport %p (%s%s%s): changing flavor "
-                      "(%s, %s/%s) -> (%s, %s/%s)\n",
-                      svc_ctx ? "reverse " : "",
-                      imp, imp->imp_obd->obd_name,
-                      svc_ctx == NULL ? "->" : "<-",
-                      obd_uuid2str(&conn->c_remote_uuid),
-                      sptlrpc_rpcflavor2name(sec->ps_flvr.sf_rpc),
-                      sptlrpc_get_hash_name(sec->ps_flvr.sf_bulk_hash),
-                      sptlrpc_get_ciph_name(sec->ps_flvr.sf_bulk_ciph),
-                      sptlrpc_rpcflavor2name(sf.sf_rpc),
-                      sptlrpc_get_hash_name(sf.sf_bulk_hash),
-                      sptlrpc_get_ciph_name(sf.sf_bulk_ciph));
-        } else {
-                CWARN("%simport %p (%s%s%s) netid %x: "
-                      "select initial flavor (%s, %s/%s)\n",
-                      svc_ctx == NULL ? "" : "reverse ",
-                      imp, imp->imp_obd->obd_name,
-                      svc_ctx == NULL ? "->" : "<-",
-                      obd_uuid2str(&conn->c_remote_uuid),
-                      LNET_NIDNET(conn->c_self),
-                      sptlrpc_rpcflavor2name(sf.sf_rpc),
-                      sptlrpc_get_hash_name(sf.sf_bulk_hash),
-                      sptlrpc_get_ciph_name(sf.sf_bulk_ciph));
+                if (flavor_equal(&sf, &sec->ps_flvr))
+                        GOTO(out, rc);
+
+                CDEBUG(D_SEC, "import %s->%s: changing flavor %s -> %s\n",
+                       imp->imp_obd->obd_name,
+                       obd_uuid2str(&conn->c_remote_uuid),
+                       sptlrpc_flavor2name(&sec->ps_flvr, str, sizeof(str)),
+                       sptlrpc_flavor2name(&sf, str2, sizeof(str2)));
+
+                if (SPTLRPC_FLVR_POLICY(sf.sf_rpc) ==
+                    SPTLRPC_FLVR_POLICY(sec->ps_flvr.sf_rpc) &&
+                    SPTLRPC_FLVR_MECH(sf.sf_rpc) ==
+                    SPTLRPC_FLVR_MECH(sec->ps_flvr.sf_rpc)) {
+                        sptlrpc_import_sec_adapt_inplace(imp, sec, &sf);
+                        GOTO(out, rc);
+                }
+        } else if (SPTLRPC_FLVR_BASE(sf.sf_rpc) !=
+                   SPTLRPC_FLVR_BASE(SPTLRPC_FLVR_NULL)) {
+                CDEBUG(D_SEC, "import %s->%s netid %x: select flavor %s\n",
+                       imp->imp_obd->obd_name,
+                       obd_uuid2str(&conn->c_remote_uuid),
+                       LNET_NIDNET(conn->c_self),
+                       sptlrpc_flavor2name(&sf, str, sizeof(str)));
         }
 
-        mutex_down(&imp->imp_sec_mutex);
+        cfs_mutex_lock(&imp->imp_sec_mutex);
 
         newsec = sptlrpc_sec_create(imp, svc_ctx, &sf, sp);
         if (newsec) {
                 sptlrpc_import_sec_install(imp, newsec);
-                rc = 0;
         } else {
-                CERROR("%simport %p (%s): failed to create new sec\n",
-                       svc_ctx == NULL ? "" : "reverse ",
-                       imp, obd_uuid2str(&conn->c_remote_uuid));
+                CERROR("import %s->%s: failed to create new sec\n",
+                       imp->imp_obd->obd_name,
+                       obd_uuid2str(&conn->c_remote_uuid));
                 rc = -EPERM;
         }
 
-        mutex_up(&imp->imp_sec_mutex);
-
+        cfs_mutex_unlock(&imp->imp_sec_mutex);
 out:
         sptlrpc_sec_put(sec);
-        return 0;
+        RETURN(rc);
 }
 
 void sptlrpc_import_sec_put(struct obd_import *imp)
@@ -1374,12 +1519,6 @@ static void import_flush_ctx_common(struct obd_import *imp,
         sptlrpc_sec_put(sec);
 }
 
-void sptlrpc_import_inval_all_ctx(struct obd_import *imp)
-{
-        /* use grace == 0 */
-        import_flush_ctx_common(imp, -1, 0, 1);
-}
-
 void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
 {
         /* it's important to use grace mode, see explain in
@@ -1389,7 +1528,7 @@ void sptlrpc_import_flush_root_ctx(struct obd_import *imp)
 
 void sptlrpc_import_flush_my_ctx(struct obd_import *imp)
 {
-        import_flush_ctx_common(imp, cfs_current()->uid, 1, 1);
+        import_flush_ctx_common(imp, cfs_curproc_uid(), 1, 1);
 }
 EXPORT_SYMBOL(sptlrpc_import_flush_my_ctx);
 
@@ -1399,9 +1538,9 @@ void sptlrpc_import_flush_all_ctx(struct obd_import *imp)
 }
 EXPORT_SYMBOL(sptlrpc_import_flush_all_ctx);
 
-/*
- * when complete successfully, req->rq_reqmsg should point to the
- * right place.
+/**
+ * Used by ptlrpc client to allocate request buffer of \a req. Upon return
+ * successfully, req->rq_reqmsg points to a buffer with size \a msgsize.
  */
 int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
 {
@@ -1410,10 +1549,10 @@ int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
         int rc;
 
         LASSERT(ctx);
-        LASSERT(atomic_read(&ctx->cc_refcount));
         LASSERT(ctx->cc_sec);
         LASSERT(ctx->cc_sec->ps_policy);
         LASSERT(req->rq_reqmsg == NULL);
+        LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 
         policy = ctx->cc_sec->ps_policy;
         rc = policy->sp_cops->alloc_reqbuf(ctx->cc_sec, req, msgsize);
@@ -1429,21 +1568,26 @@ int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize)
         return rc;
 }
 
+/**
+ * Used by ptlrpc client to free request buffer of \a req. After this
+ * req->rq_reqmsg is set to NULL and should not be accessed anymore.
+ */
 void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req)
 {
         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
         struct ptlrpc_sec_policy *policy;
 
         LASSERT(ctx);
-        LASSERT(atomic_read(&ctx->cc_refcount));
         LASSERT(ctx->cc_sec);
         LASSERT(ctx->cc_sec->ps_policy);
+        LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 
         if (req->rq_reqbuf == NULL && req->rq_clrbuf == NULL)
                 return;
 
         policy = ctx->cc_sec->ps_policy;
         policy->sp_cops->free_reqbuf(ctx->cc_sec, req);
+        req->rq_reqmsg = NULL;
 }
 
 /*
@@ -1490,13 +1634,14 @@ void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
 }
 EXPORT_SYMBOL(_sptlrpc_enlarge_msg_inplace);
 
-/*
- * enlarge @segment of upper message req->rq_reqmsg to @newsize, all data
- * will be preserved after enlargement. this must be called after rq_reqmsg has
- * been intialized at least.
+/**
+ * Used by ptlrpc client to enlarge the \a segment of request message pointed
+ * by req->rq_reqmsg to size \a newsize, all previously filled-in data will be
+ * preserved after the enlargement. this must be called after original request
+ * buffer being allocated.
  *
- * caller's attention: upon return, rq_reqmsg and rq_reqlen might have
- * been changed.
+ * \note after this be called, rq_reqmsg and rq_reqlen might have been changed,
+ * so caller should refresh its local pointers if needed.
  */
 int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
                                int segment, int newsize)
@@ -1519,6 +1664,11 @@ int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
 }
 EXPORT_SYMBOL(sptlrpc_cli_enlarge_reqbuf);
 
+/**
+ * Used by ptlrpc client to allocate reply buffer of \a req.
+ *
+ * \note After this, req->rq_repmsg is still not accessible.
+ */
 int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
 {
         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
@@ -1526,7 +1676,6 @@ int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
         ENTRY;
 
         LASSERT(ctx);
-        LASSERT(atomic_read(&ctx->cc_refcount));
         LASSERT(ctx->cc_sec);
         LASSERT(ctx->cc_sec->ps_policy);
 
@@ -1537,6 +1686,10 @@ int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize)
         RETURN(policy->sp_cops->alloc_repbuf(ctx->cc_sec, req, msgsize));
 }
 
+/**
+ * Used by ptlrpc client to free reply buffer of \a req. After this
+ * req->rq_repmsg is set to NULL and should not be accessed anymore.
+ */
 void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
 {
         struct ptlrpc_cli_ctx *ctx = req->rq_cli_ctx;
@@ -1544,9 +1697,9 @@ void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
         ENTRY;
 
         LASSERT(ctx);
-        LASSERT(atomic_read(&ctx->cc_refcount));
         LASSERT(ctx->cc_sec);
         LASSERT(ctx->cc_sec->ps_policy);
+        LASSERT_ATOMIC_POS(&ctx->cc_refcount);
 
         if (req->rq_repbuf == NULL)
                 return;
@@ -1554,6 +1707,7 @@ void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req)
 
         policy = ctx->cc_sec->ps_policy;
         policy->sp_cops->free_repbuf(ctx->cc_sec, req);
+        req->rq_repmsg = NULL;
         EXIT;
 }
 
@@ -1586,12 +1740,13 @@ static int flavor_allowed(struct sptlrpc_flavor *exp,
 {
         struct sptlrpc_flavor *flvr = &req->rq_flvr;
 
-        if (exp->sf_rpc == flvr->sf_rpc)
+        if (exp->sf_rpc == SPTLRPC_FLVR_ANY || exp->sf_rpc == flvr->sf_rpc)
                 return 1;
 
         if ((req->rq_ctx_init || req->rq_ctx_fini) &&
-            RPC_FLVR_POLICY(exp->sf_rpc) == RPC_FLVR_POLICY(flvr->sf_rpc) &&
-            RPC_FLVR_MECH(exp->sf_rpc) == RPC_FLVR_MECH(flvr->sf_rpc))
+            SPTLRPC_FLVR_POLICY(exp->sf_rpc) ==
+            SPTLRPC_FLVR_POLICY(flvr->sf_rpc) &&
+            SPTLRPC_FLVR_MECH(exp->sf_rpc) == SPTLRPC_FLVR_MECH(flvr->sf_rpc))
                 return 1;
 
         return 0;
@@ -1599,6 +1754,11 @@ static int flavor_allowed(struct sptlrpc_flavor *exp,
 
 #define EXP_FLVR_UPDATE_EXPIRE      (OBD_TIMEOUT_DEFAULT + 10)
 
+/**
+ * Given an export \a exp, check whether the flavor of incoming \a req
+ * is allowed by the export \a exp. Main logic is about taking care of
+ * changing configurations. Return 0 means success.
+ */
 int sptlrpc_target_export_check(struct obd_export *exp,
                                 struct ptlrpc_request *req)
 {
@@ -1616,7 +1776,7 @@ int sptlrpc_target_export_check(struct obd_export *exp,
         if (req->rq_ctx_fini)
                 return 0;
 
-        spin_lock(&exp->exp_lock);
+        cfs_spin_lock(&exp->exp_lock);
 
         /* if flavor just changed (exp->exp_flvr_changed != 0), we wait for
          * the first req with the new flavor, then treat it as current flavor,
@@ -1643,20 +1803,22 @@ int sptlrpc_target_export_check(struct obd_export *exp,
 
                 /* if it's gss, we only interested in root ctx init */
                 if (req->rq_auth_gss &&
-                    !(req->rq_ctx_init && (req->rq_auth_usr_root ||
-                                           req->rq_auth_usr_mdt))) {
-                        spin_unlock(&exp->exp_lock);
-                        CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d)\n",
+                    !(req->rq_ctx_init &&
+                      (req->rq_auth_usr_root || req->rq_auth_usr_mdt ||
+                       req->rq_auth_usr_ost))) {
+                        cfs_spin_unlock(&exp->exp_lock);
+                        CDEBUG(D_SEC, "is good but not root(%d:%d:%d:%d:%d)\n",
                                req->rq_auth_gss, req->rq_ctx_init,
-                               req->rq_auth_usr_root, req->rq_auth_usr_mdt);
+                               req->rq_auth_usr_root, req->rq_auth_usr_mdt,
+                               req->rq_auth_usr_ost);
                         return 0;
                 }
 
                 exp->exp_flvr_adapt = 0;
-                spin_unlock(&exp->exp_lock);
+                cfs_spin_unlock(&exp->exp_lock);
 
                 return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
-                                                req->rq_svc_ctx, flavor.sf_rpc);
+                                                req->rq_svc_ctx, &flavor);
         }
 
         /* if it equals to the current flavor, we accept it, but need to
@@ -1665,8 +1827,9 @@ int sptlrpc_target_export_check(struct obd_export *exp,
                 /* most cases should return here, we only interested in
                  * gss root ctx init */
                 if (!req->rq_auth_gss || !req->rq_ctx_init ||
-                    (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt)) {
-                        spin_unlock(&exp->exp_lock);
+                    (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
+                     !req->rq_auth_usr_ost)) {
+                        cfs_spin_unlock(&exp->exp_lock);
                         return 0;
                 }
 
@@ -1675,7 +1838,7 @@ int sptlrpc_target_export_check(struct obd_export *exp,
                  * shortly, and let _this_ rpc pass through */
                 if (exp->exp_flvr_changed) {
                         LASSERT(exp->exp_flvr_adapt);
-                        spin_unlock(&exp->exp_lock);
+                        cfs_spin_unlock(&exp->exp_lock);
                         return 0;
                 }
 
@@ -1686,17 +1849,17 @@ int sptlrpc_target_export_check(struct obd_export *exp,
                                exp->exp_flvr_old[0].sf_rpc,
                                exp->exp_flvr_old[1].sf_rpc);
                         flavor = exp->exp_flvr;
-                        spin_unlock(&exp->exp_lock);
+                        cfs_spin_unlock(&exp->exp_lock);
 
                         return sptlrpc_import_sec_adapt(exp->exp_imp_reverse,
                                                         req->rq_svc_ctx,
-                                                        flavor.sf_rpc);
+                                                        &flavor);
                 } else {
                         CDEBUG(D_SEC, "exp %p (%x|%x|%x): is current flavor, "
                                "install rvs ctx\n", exp, exp->exp_flvr.sf_rpc,
                                exp->exp_flvr_old[0].sf_rpc,
                                exp->exp_flvr_old[1].sf_rpc);
-                        spin_unlock(&exp->exp_lock);
+                        cfs_spin_unlock(&exp->exp_lock);
 
                         return sptlrpc_svc_install_rvs_ctx(exp->exp_imp_reverse,
                                                            req->rq_svc_ctx);
@@ -1713,7 +1876,7 @@ int sptlrpc_target_export_check(struct obd_export *exp,
                                        exp->exp_flvr_old[1].sf_rpc,
                                        exp->exp_flvr_expire[0] -
                                                 cfs_time_current_sec());
-                                spin_unlock(&exp->exp_lock);
+                                cfs_spin_unlock(&exp->exp_lock);
                                 return 0;
                         }
                 } else {
@@ -1738,7 +1901,7 @@ int sptlrpc_target_export_check(struct obd_export *exp,
                                        exp->exp_flvr_old[1].sf_rpc,
                                        exp->exp_flvr_expire[1] -
                                                 cfs_time_current_sec());
-                                spin_unlock(&exp->exp_lock);
+                                cfs_spin_unlock(&exp->exp_lock);
                                 return 0;
                         }
                 } else {
@@ -1755,13 +1918,26 @@ int sptlrpc_target_export_check(struct obd_export *exp,
                        exp->exp_flvr_old[1].sf_rpc);
         }
 
-        spin_unlock(&exp->exp_lock);
+        cfs_spin_unlock(&exp->exp_lock);
 
-        CWARN("req %p: (%u|%u|%u|%u|%u) with unauthorized flavor %x\n",
+        CWARN("exp %p(%s): req %p (%u|%u|%u|%u|%u|%u) with "
+              "unauthorized flavor %x, expect %x|%x(%+ld)|%x(%+ld)\n",
+              exp, exp->exp_obd->obd_name,
               req, req->rq_auth_gss, req->rq_ctx_init, req->rq_ctx_fini,
-              req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_flvr.sf_rpc);
+              req->rq_auth_usr_root, req->rq_auth_usr_mdt, req->rq_auth_usr_ost,
+              req->rq_flvr.sf_rpc,
+              exp->exp_flvr.sf_rpc,
+              exp->exp_flvr_old[0].sf_rpc,
+              exp->exp_flvr_expire[0] ?
+              (unsigned long) (exp->exp_flvr_expire[0] -
+                               cfs_time_current_sec()) : 0,
+              exp->exp_flvr_old[1].sf_rpc,
+              exp->exp_flvr_expire[1] ?
+              (unsigned long) (exp->exp_flvr_expire[1] -
+                               cfs_time_current_sec()) : 0);
         return -EACCES;
 }
+EXPORT_SYMBOL(sptlrpc_target_export_check);
 
 void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
                                       struct sptlrpc_rule_set *rset)
@@ -1771,114 +1947,123 @@ void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
 
         LASSERT(obd);
 
-        spin_lock(&obd->obd_dev_lock);
+        cfs_spin_lock(&obd->obd_dev_lock);
 
-        list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
+        cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
                 if (exp->exp_connection == NULL)
                         continue;
 
                 /* note if this export had just been updated flavor
                  * (exp_flvr_changed == 1), this will override the
                  * previous one. */
-                spin_lock(&exp->exp_lock);
-                sptlrpc_rule_set_choose(rset, exp->exp_sp_peer,
-                                        exp->exp_connection->c_peer.nid,
-                                        &new_flvr);
+                cfs_spin_lock(&exp->exp_lock);
+                sptlrpc_target_choose_flavor(rset, exp->exp_sp_peer,
+                                             exp->exp_connection->c_peer.nid,
+                                             &new_flvr);
                 if (exp->exp_flvr_changed ||
-                    memcmp(&new_flvr, &exp->exp_flvr, sizeof(new_flvr))) {
+                    !flavor_equal(&new_flvr, &exp->exp_flvr)) {
                         exp->exp_flvr_old[1] = new_flvr;
                         exp->exp_flvr_expire[1] = 0;
                         exp->exp_flvr_changed = 1;
                         exp->exp_flvr_adapt = 1;
+
                         CDEBUG(D_SEC, "exp %p (%s): updated flavor %x->%x\n",
                                exp, sptlrpc_part2name(exp->exp_sp_peer),
                                exp->exp_flvr.sf_rpc,
                                exp->exp_flvr_old[1].sf_rpc);
                 }
-                spin_unlock(&exp->exp_lock);
+                cfs_spin_unlock(&exp->exp_lock);
         }
 
-        spin_unlock(&obd->obd_dev_lock);
+        cfs_spin_unlock(&obd->obd_dev_lock);
 }
 EXPORT_SYMBOL(sptlrpc_target_update_exp_flavor);
 
 static int sptlrpc_svc_check_from(struct ptlrpc_request *req, int svc_rc)
 {
-        if (svc_rc == SECSVC_DROP)
-                return SECSVC_DROP;
+        /* peer's claim is unreliable unless gss is being used */
+        if (!req->rq_auth_gss || svc_rc == SECSVC_DROP)
+                return svc_rc;
 
         switch (req->rq_sp_from) {
         case LUSTRE_SP_CLI:
+                if (req->rq_auth_usr_mdt || req->rq_auth_usr_ost) {
+                        DEBUG_REQ(D_ERROR, req, "faked source CLI");
+                        svc_rc = SECSVC_DROP;
+                }
+                break;
         case LUSTRE_SP_MDT:
+                if (!req->rq_auth_usr_mdt) {
+                        DEBUG_REQ(D_ERROR, req, "faked source MDT");
+                        svc_rc = SECSVC_DROP;
+                }
+                break;
         case LUSTRE_SP_OST:
+                if (!req->rq_auth_usr_ost) {
+                        DEBUG_REQ(D_ERROR, req, "faked source OST");
+                        svc_rc = SECSVC_DROP;
+                }
+                break;
         case LUSTRE_SP_MGS:
-        case LUSTRE_SP_ANY:
+        case LUSTRE_SP_MGC:
+                if (!req->rq_auth_usr_root && !req->rq_auth_usr_mdt &&
+                    !req->rq_auth_usr_ost) {
+                        DEBUG_REQ(D_ERROR, req, "faked source MGC/MGS");
+                        svc_rc = SECSVC_DROP;
+                }
                 break;
+        case LUSTRE_SP_ANY:
         default:
                 DEBUG_REQ(D_ERROR, req, "invalid source %u", req->rq_sp_from);
-                return SECSVC_DROP;
-        }
-
-        if (!req->rq_auth_gss)
-                return svc_rc;
-
-        if (unlikely(req->rq_sp_from == LUSTRE_SP_ANY)) {
-                CERROR("not specific part\n");
-                return SECSVC_DROP;
-        }
-
-        /* from MDT, must be authenticated as MDT */
-        if (unlikely(req->rq_sp_from == LUSTRE_SP_MDT &&
-                     !req->rq_auth_usr_mdt)) {
-                DEBUG_REQ(D_ERROR, req, "fake source MDT");
-                return SECSVC_DROP;
-        }
-
-        /* from OST, must be callback to MDT and CLI, the reverse sec
-         * was from mdt/root keytab, so it should be MDT or root FIXME */
-        if (unlikely(req->rq_sp_from == LUSTRE_SP_OST &&
-                     !req->rq_auth_usr_mdt && !req->rq_auth_usr_root)) {
-                DEBUG_REQ(D_ERROR, req, "fake source OST");
-                return SECSVC_DROP;
+                svc_rc = SECSVC_DROP;
         }
 
         return svc_rc;
 }
 
+/**
+ * Used by ptlrpc server, to perform transformation upon request message of
+ * incoming \a req. This must be the first thing to do with a incoming
+ * request in ptlrpc layer.
+ *
+ * \retval SECSVC_OK success, and req->rq_reqmsg point to request message in
+ * clear text, size is req->rq_reqlen; also req->rq_svc_ctx is set.
+ * \retval SECSVC_COMPLETE success, the request has been fully processed, and
+ * reply message has been prepared.
+ * \retval SECSVC_DROP failed, this request should be dropped.
+ */
 int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
 {
         struct ptlrpc_sec_policy *policy;
-        struct lustre_msg *msg = req->rq_reqbuf;
-        int rc;
+        struct lustre_msg        *msg = req->rq_reqbuf;
+        int                       rc;
         ENTRY;
 
         LASSERT(msg);
         LASSERT(req->rq_reqmsg == NULL);
         LASSERT(req->rq_repmsg == NULL);
+        LASSERT(req->rq_svc_ctx == NULL);
 
-        req->rq_sp_from = LUSTRE_SP_ANY;
-        req->rq_auth_uid = INVALID_UID;
-        req->rq_auth_mapped_uid = INVALID_UID;
+        req->rq_req_swab_mask = 0;
 
-        if (req->rq_reqdata_len < sizeof(struct lustre_msg)) {
-                CERROR("request size %d too small\n", req->rq_reqdata_len);
+        rc = __lustre_unpack_msg(msg, req->rq_reqdata_len);
+        switch (rc) {
+        case 1:
+                lustre_set_req_swabbed(req, MSG_PTLRPC_HEADER_OFF);
+        case 0:
+                break;
+        default:
+                CERROR("error unpacking request from %s x"LPU64"\n",
+                       libcfs_id2str(req->rq_peer), req->rq_xid);
                 RETURN(SECSVC_DROP);
         }
 
-        /*
-         * v2 message.
-         */
-        if (msg->lm_magic == LUSTRE_MSG_MAGIC_V2)
-                req->rq_flvr.sf_rpc = WIRE_FLVR_RPC(msg->lm_secflvr);
-        else
-                req->rq_flvr.sf_rpc = WIRE_FLVR_RPC(__swab32(msg->lm_secflvr));
-
-        /* unpack the wrapper message if the policy is not null */
-        if ((RPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL) &&
-             lustre_unpack_msg(msg, req->rq_reqdata_len))
-                RETURN(SECSVC_DROP);
+        req->rq_flvr.sf_rpc = WIRE_FLVR(msg->lm_secflvr);
+        req->rq_sp_from = LUSTRE_SP_ANY;
+        req->rq_auth_uid = INVALID_UID;
+        req->rq_auth_mapped_uid = INVALID_UID;
 
-        policy = sptlrpc_rpcflavor2policy(req->rq_flvr.sf_rpc);
+        policy = sptlrpc_wireflavor2policy(req->rq_flvr.sf_rpc);
         if (!policy) {
                 CERROR("unsupported rpc flavor %x\n", req->rq_flvr.sf_rpc);
                 RETURN(SECSVC_DROP);
@@ -1886,29 +2071,28 @@ int sptlrpc_svc_unwrap_request(struct ptlrpc_request *req)
 
         LASSERT(policy->sp_sops->accept);
         rc = policy->sp_sops->accept(req);
-
-        LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
         sptlrpc_policy_put(policy);
+        LASSERT(req->rq_reqmsg || rc != SECSVC_OK);
+        LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
+
+        /*
+         * if it's not null flavor (which means embedded packing msg),
+         * reset the swab mask for the comming inner msg unpacking.
+         */
+        if (SPTLRPC_FLVR_POLICY(req->rq_flvr.sf_rpc) != SPTLRPC_POLICY_NULL)
+                req->rq_req_swab_mask = 0;
 
         /* sanity check for the request source */
         rc = sptlrpc_svc_check_from(req, rc);
-
-        /* FIXME move to proper place */
-        if (rc == SECSVC_OK) {
-                __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
-
-                if (opc == OST_WRITE)
-                        req->rq_bulk_write = 1;
-                else if (opc == OST_READ)
-                        req->rq_bulk_read = 1;
-        }
-
-        LASSERT(req->rq_svc_ctx || rc == SECSVC_DROP);
         RETURN(rc);
 }
 
-int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req,
-                         int msglen)
+/**
+ * Used by ptlrpc server, to allocate reply buffer for \a req. If succeed,
+ * req->rq_reply_state is set, and req->rq_reply_state->rs_msg point to
+ * a buffer of \a msglen size.
+ */
+int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
 {
         struct ptlrpc_sec_policy *policy;
         struct ptlrpc_reply_state *rs;
@@ -1924,7 +2108,7 @@ int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req,
         rc = policy->sp_sops->alloc_rs(req, msglen);
         if (unlikely(rc == -ENOMEM)) {
                 /* failed alloc, try emergency pool */
-                rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_service);
+               rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_svcpt);
                 if (rs == NULL)
                         RETURN(-ENOMEM);
 
@@ -1942,6 +2126,12 @@ int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req,
         RETURN(rc);
 }
 
+/**
+ * Used by ptlrpc server, to perform transformation upon reply message.
+ *
+ * \post req->rq_reply_off is set to approriate server-controlled reply offset.
+ * \post req->rq_repmsg and req->rq_reply_state->rs_msg becomes inaccessible.
+ */
 int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
 {
         struct ptlrpc_sec_policy *policy;
@@ -1960,6 +2150,9 @@ int sptlrpc_svc_wrap_reply(struct ptlrpc_request *req)
         RETURN(rc);
 }
 
+/**
+ * Used by ptlrpc server, to free reply_state.
+ */
 void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs)
 {
         struct ptlrpc_sec_policy *policy;
@@ -1984,11 +2177,8 @@ void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req)
 {
         struct ptlrpc_svc_ctx *ctx = req->rq_svc_ctx;
 
-        if (ctx == NULL)
-                return;
-
-        LASSERT(atomic_read(&ctx->sc_refcount) > 0);
-        atomic_inc(&ctx->sc_refcount);
+        if (ctx != NULL)
+                cfs_atomic_inc(&ctx->sc_refcount);
 }
 
 void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
@@ -1998,8 +2188,8 @@ void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req)
         if (ctx == NULL)
                 return;
 
-        LASSERT(atomic_read(&ctx->sc_refcount) > 0);
-        if (atomic_dec_and_test(&ctx->sc_refcount)) {
+        LASSERT_ATOMIC_POS(&ctx->sc_refcount);
+        if (cfs_atomic_dec_and_test(&ctx->sc_refcount)) {
                 if (ctx->sc_policy->sp_sops->free_ctx)
                         ctx->sc_policy->sp_sops->free_ctx(ctx);
         }
@@ -2013,7 +2203,7 @@ void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req)
         if (ctx == NULL)
                 return;
 
-        LASSERT(atomic_read(&ctx->sc_refcount) > 0);
+        LASSERT_ATOMIC_POS(&ctx->sc_refcount);
         if (ctx->sc_policy->sp_sops->invalidate_ctx)
                 ctx->sc_policy->sp_sops->invalidate_ctx(ctx);
 }
@@ -2023,16 +2213,20 @@ EXPORT_SYMBOL(sptlrpc_svc_ctx_invalidate);
  * bulk security                        *
  ****************************************/
 
+/**
+ * Perform transformation upon bulk data pointed by \a desc. This is called
+ * before transforming the request message.
+ */
 int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
                           struct ptlrpc_bulk_desc *desc)
 {
         struct ptlrpc_cli_ctx *ctx;
 
+        LASSERT(req->rq_bulk_read || req->rq_bulk_write);
+
         if (!req->rq_pack_bulk)
                 return 0;
 
-        LASSERT(req->rq_bulk_read || req->rq_bulk_write);
-
         ctx = req->rq_cli_ctx;
         if (ctx->cc_ops->wrap_bulk)
                 return ctx->cc_ops->wrap_bulk(ctx, req, desc);
@@ -2040,92 +2234,82 @@ int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
 }
 EXPORT_SYMBOL(sptlrpc_cli_wrap_bulk);
 
-static
-void pga_to_bulk_desc(int nob, obd_count pg_count, struct brw_page **pga,
-                      struct ptlrpc_bulk_desc *desc)
-{
-        int i;
-
-        LASSERT(pga);
-        LASSERT(*pga);
-
-        for (i = 0; i < pg_count && nob > 0; i++) {
-#ifdef __KERNEL__
-                desc->bd_iov[i].kiov_page = pga[i]->pg;
-                desc->bd_iov[i].kiov_len = pga[i]->count > nob ?
-                                           nob : pga[i]->count;
-                desc->bd_iov[i].kiov_offset = pga[i]->off & ~CFS_PAGE_MASK;
-#else
-#warning FIXME for liblustre!
-                desc->bd_iov[i].iov_base = pga[i]->pg->addr;
-                desc->bd_iov[i].iov_len = pga[i]->count > nob ?
-                                           nob : pga[i]->count;
-#endif
-
-                desc->bd_iov_count++;
-                nob -= pga[i]->count;
-        }
-}
-
+/**
+ * This is called after unwrap the reply message.
+ * return nob of actual plain text size received, or error code.
+ */
 int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
-                                 int nob, obd_count pg_count,
-                                 struct brw_page **pga)
+                                 struct ptlrpc_bulk_desc *desc,
+                                 int nob)
 {
-        struct ptlrpc_bulk_desc *desc;
-        struct ptlrpc_cli_ctx *ctx;
-        int rc = 0;
-
-        if (!req->rq_pack_bulk)
-                return 0;
+        struct ptlrpc_cli_ctx  *ctx;
+        int                     rc;
 
         LASSERT(req->rq_bulk_read && !req->rq_bulk_write);
 
-        OBD_ALLOC(desc, offsetof(struct ptlrpc_bulk_desc, bd_iov[pg_count]));
-        if (desc == NULL) {
-                CERROR("out of memory, can't verify bulk read data\n");
-                return -ENOMEM;
-        }
-
-        pga_to_bulk_desc(nob, pg_count, pga, desc);
+        if (!req->rq_pack_bulk)
+                return desc->bd_nob_transferred;
 
         ctx = req->rq_cli_ctx;
-        if (ctx->cc_ops->unwrap_bulk)
+        if (ctx->cc_ops->unwrap_bulk) {
                 rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
-
-        OBD_FREE(desc, offsetof(struct ptlrpc_bulk_desc, bd_iov[pg_count]));
-
-        return rc;
+                if (rc < 0)
+                        return rc;
+        }
+        return desc->bd_nob_transferred;
 }
 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_read);
 
+/**
+ * This is called after unwrap the reply message.
+ * return 0 for success or error code.
+ */
 int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
                                   struct ptlrpc_bulk_desc *desc)
 {
-        struct ptlrpc_cli_ctx *ctx;
+        struct ptlrpc_cli_ctx  *ctx;
+        int                     rc;
+
+        LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
 
         if (!req->rq_pack_bulk)
                 return 0;
 
-        LASSERT(!req->rq_bulk_read && req->rq_bulk_write);
-
         ctx = req->rq_cli_ctx;
-        if (ctx->cc_ops->unwrap_bulk)
-                return ctx->cc_ops->unwrap_bulk(ctx, req, desc);
+        if (ctx->cc_ops->unwrap_bulk) {
+                rc = ctx->cc_ops->unwrap_bulk(ctx, req, desc);
+                if (rc < 0)
+                        return rc;
+        }
+
+        /*
+         * if everything is going right, nob should equals to nob_transferred.
+         * in case of privacy mode, nob_transferred needs to be adjusted.
+         */
+        if (desc->bd_nob != desc->bd_nob_transferred) {
+                CERROR("nob %d doesn't match transferred nob %d",
+                       desc->bd_nob, desc->bd_nob_transferred);
+                return -EPROTO;
+        }
 
         return 0;
 }
 EXPORT_SYMBOL(sptlrpc_cli_unwrap_bulk_write);
 
+#ifdef HAVE_SERVER_SUPPORT
+/**
+ * Performe transformation upon outgoing bulk read.
+ */
 int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
                           struct ptlrpc_bulk_desc *desc)
 {
         struct ptlrpc_svc_ctx *ctx;
 
+        LASSERT(req->rq_bulk_read);
+
         if (!req->rq_pack_bulk)
                 return 0;
 
-        LASSERT(req->rq_bulk_read || req->rq_bulk_write);
-
         ctx = req->rq_svc_ctx;
         if (ctx->sc_policy->sp_sops->wrap_bulk)
                 return ctx->sc_policy->sp_sops->wrap_bulk(req, desc);
@@ -2134,24 +2318,67 @@ int sptlrpc_svc_wrap_bulk(struct ptlrpc_request *req,
 }
 EXPORT_SYMBOL(sptlrpc_svc_wrap_bulk);
 
+/**
+ * Performe transformation upon incoming bulk write.
+ */
 int sptlrpc_svc_unwrap_bulk(struct ptlrpc_request *req,
                             struct ptlrpc_bulk_desc *desc)
 {
         struct ptlrpc_svc_ctx *ctx;
+        int                    rc;
+
+        LASSERT(req->rq_bulk_write);
+
+        /*
+         * if it's in privacy mode, transferred should >= expected; otherwise
+         * transferred should == expected.
+         */
+        if (desc->bd_nob_transferred < desc->bd_nob ||
+            (desc->bd_nob_transferred > desc->bd_nob &&
+             SPTLRPC_FLVR_BULK_SVC(req->rq_flvr.sf_rpc) !=
+             SPTLRPC_BULK_SVC_PRIV)) {
+                DEBUG_REQ(D_ERROR, req, "truncated bulk GET %d(%d)",
+                          desc->bd_nob_transferred, desc->bd_nob);
+                return -ETIMEDOUT;
+        }
 
         if (!req->rq_pack_bulk)
                 return 0;
 
-        LASSERT(req->rq_bulk_read || req->rq_bulk_write);
-
         ctx = req->rq_svc_ctx;
-        if (ctx->sc_policy->sp_sops->unwrap_bulk);
-                return ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
+        if (ctx->sc_policy->sp_sops->unwrap_bulk) {
+                rc = ctx->sc_policy->sp_sops->unwrap_bulk(req, desc);
+                if (rc)
+                        CERROR("error unwrap bulk: %d\n", rc);
+        }
 
+        /* return 0 to allow reply be sent */
         return 0;
 }
 EXPORT_SYMBOL(sptlrpc_svc_unwrap_bulk);
 
+/**
+ * Prepare buffers for incoming bulk write.
+ */
+int sptlrpc_svc_prep_bulk(struct ptlrpc_request *req,
+                          struct ptlrpc_bulk_desc *desc)
+{
+        struct ptlrpc_svc_ctx *ctx;
+
+        LASSERT(req->rq_bulk_write);
+
+        if (!req->rq_pack_bulk)
+                return 0;
+
+        ctx = req->rq_svc_ctx;
+        if (ctx->sc_policy->sp_sops->prep_bulk)
+                return ctx->sc_policy->sp_sops->prep_bulk(req, desc);
+
+        return 0;
+}
+EXPORT_SYMBOL(sptlrpc_svc_prep_bulk);
+
+#endif /* HAVE_SERVER_SUPPORT */
 
 /****************************************
  * user descriptor helpers              *
@@ -2179,18 +2406,18 @@ int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
 
         pud = lustre_msg_buf(msg, offset, 0);
 
-        pud->pud_uid = cfs_current()->uid;
-        pud->pud_gid = cfs_current()->gid;
-        pud->pud_fsuid = cfs_current()->fsuid;
-        pud->pud_fsgid = cfs_current()->fsgid;
-        pud->pud_cap = cfs_current()->cap_effective;
+        pud->pud_uid = cfs_curproc_uid();
+        pud->pud_gid = cfs_curproc_gid();
+        pud->pud_fsuid = cfs_curproc_fsuid();
+        pud->pud_fsgid = cfs_curproc_fsgid();
+        pud->pud_cap = cfs_curproc_cap_pack();
         pud->pud_ngroups = (msg->lm_buflens[offset] - sizeof(*pud)) / 4;
 
 #ifdef __KERNEL__
         task_lock(current);
         if (pud->pud_ngroups > current_ngroups)
                 pud->pud_ngroups = current_ngroups;
-        memcpy(pud->pud_groups, cfs_current()->group_info->blocks[0],
+        memcpy(pud->pud_groups, current_cred()->group_info->blocks[0],
                pud->pud_ngroups * sizeof(__u32));
         task_unlock(current);
 #endif
@@ -2199,7 +2426,7 @@ int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset)
 }
 EXPORT_SYMBOL(sptlrpc_pack_user_desc);
 
-int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset)
+int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset, int swabbed)
 {
         struct ptlrpc_user_desc *pud;
         int                      i;
@@ -2208,7 +2435,7 @@ int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset)
         if (!pud)
                 return -EINVAL;
 
-        if (lustre_msg_swabbed(msg)) {
+        if (swabbed) {
                 __swab32s(&pud->pud_uid);
                 __swab32s(&pud->pud_gid);
                 __swab32s(&pud->pud_fsuid);
@@ -2229,7 +2456,7 @@ int sptlrpc_unpack_user_desc(struct lustre_msg *msg, int offset)
                 return -EINVAL;
         }
 
-        if (lustre_msg_swabbed(msg)) {
+        if (swabbed) {
                 for (i = 0; i < pud->pud_ngroups; i++)
                         __swab32s(&pud->pud_groups[i]);
         }
@@ -2252,55 +2479,47 @@ const char * sec2target_str(struct ptlrpc_sec *sec)
 }
 EXPORT_SYMBOL(sec2target_str);
 
+/*
+ * return true if the bulk data is protected
+ */
+int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr)
+{
+        switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
+        case SPTLRPC_BULK_SVC_INTG:
+        case SPTLRPC_BULK_SVC_PRIV:
+                return 1;
+        default:
+                return 0;
+        }
+}
+EXPORT_SYMBOL(sptlrpc_flavor_has_bulk);
+
 /****************************************
  * crypto API helper/alloc blkciper     *
  ****************************************/
 
-#ifdef __KERNEL__
-#ifndef HAVE_ASYNC_BLOCK_CIPHER
-struct ll_crypto_cipher *ll_crypto_alloc_blkcipher(const char * algname,
-                                                   u32 type, u32 mask)
-{
-        char        buf[CRYPTO_MAX_ALG_NAME + 1];
-        const char *pan = algname;
-        u32         flag = 0; 
-
-        if (strncmp("cbc(", algname, 4) == 0)
-                flag |= CRYPTO_TFM_MODE_CBC;
-        else if (strncmp("ecb(", algname, 4) == 0)
-                flag |= CRYPTO_TFM_MODE_ECB;
-        if (flag) {
-                char *vp = strnchr(algname, CRYPTO_MAX_ALG_NAME, ')');
-                if (vp) {
-                        memcpy(buf, algname + 4, vp - algname - 4);
-                        buf[vp - algname - 4] = '\0';
-                        pan = buf;
-                } else {
-                        flag = 0;
-                }
-        }
-        return crypto_alloc_tfm(pan, flag);
-}
-EXPORT_SYMBOL(ll_crypto_alloc_blkcipher);
-#endif
-#endif
-
 /****************************************
  * initialize/finalize                  *
  ****************************************/
 
-int __init sptlrpc_init(void)
+int sptlrpc_init(void)
 {
         int rc;
 
-        rc = sptlrpc_gc_start_thread();
+        cfs_rwlock_init(&policy_lock);
+
+        rc = sptlrpc_gc_init();
         if (rc)
                 goto out;
 
-        rc = sptlrpc_enc_pool_init();
+        rc = sptlrpc_conf_init();
         if (rc)
                 goto out_gc;
 
+        rc = sptlrpc_enc_pool_init();
+        if (rc)
+                goto out_conf;
+
         rc = sptlrpc_null_init();
         if (rc)
                 goto out_pool;
@@ -2321,17 +2540,20 @@ out_null:
         sptlrpc_null_fini();
 out_pool:
         sptlrpc_enc_pool_fini();
+out_conf:
+        sptlrpc_conf_fini();
 out_gc:
-        sptlrpc_gc_stop_thread();
+        sptlrpc_gc_fini();
 out:
         return rc;
 }
 
-void __exit sptlrpc_fini(void)
+void sptlrpc_fini(void)
 {
         sptlrpc_lproc_fini();
         sptlrpc_plain_fini();
         sptlrpc_null_fini();
         sptlrpc_enc_pool_fini();
-        sptlrpc_gc_stop_thread();
+        sptlrpc_conf_fini();
+        sptlrpc_gc_fini();
 }