Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / staging / lustre / lustre / include / lustre_sec.h
diff --git a/kernel/drivers/staging/lustre/lustre/include/lustre_sec.h b/kernel/drivers/staging/lustre/lustre/include/lustre_sec.h
new file mode 100644 (file)
index 0000000..dff70a5
--- /dev/null
@@ -0,0 +1,1147 @@
+/*
+ * 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.
+ *
+ * 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).
+ *
+ * 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.
+ *
+ * Copyright (c) 2012, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ */
+
+#ifndef _LUSTRE_SEC_H_
+#define _LUSTRE_SEC_H_
+
+/** \defgroup sptlrpc sptlrpc
+ *
+ * @{
+ */
+
+/*
+ * to avoid include
+ */
+struct obd_import;
+struct obd_export;
+struct ptlrpc_request;
+struct ptlrpc_reply_state;
+struct ptlrpc_bulk_desc;
+struct brw_page;
+/* Linux specific */
+struct key;
+struct seq_file;
+
+/*
+ * forward declaration
+ */
+struct ptlrpc_sec_policy;
+struct ptlrpc_sec_cops;
+struct ptlrpc_sec_sops;
+struct ptlrpc_sec;
+struct ptlrpc_svc_ctx;
+struct ptlrpc_cli_ctx;
+struct ptlrpc_ctx_ops;
+
+/**
+ * \addtogroup flavor flavor
+ *
+ * RPC flavor is represented by a 32 bits integer. Currently the high 12 bits
+ * are unused, must be set to 0 for future expansion.
+ * <pre>
+ * ------------------------------------------------------------------------
+ * | 4b (bulk svc) | 4b (bulk type) | 4b (svc) | 4b (mech)  | 4b (policy) |
+ * ------------------------------------------------------------------------
+ * </pre>
+ *
+ * @{
+ */
+
+/*
+ * flavor constants
+ */
+enum sptlrpc_policy {
+       SPTLRPC_POLICY_NULL          = 0,
+       SPTLRPC_POLICY_PLAIN        = 1,
+       SPTLRPC_POLICY_GSS            = 2,
+       SPTLRPC_POLICY_MAX,
+};
+
+enum sptlrpc_mech_null {
+       SPTLRPC_MECH_NULL              = 0,
+       SPTLRPC_MECH_NULL_MAX,
+};
+
+enum sptlrpc_mech_plain {
+       SPTLRPC_MECH_PLAIN            = 0,
+       SPTLRPC_MECH_PLAIN_MAX,
+};
+
+enum sptlrpc_mech_gss {
+       SPTLRPC_MECH_GSS_NULL      = 0,
+       SPTLRPC_MECH_GSS_KRB5      = 1,
+       SPTLRPC_MECH_GSS_MAX,
+};
+
+enum sptlrpc_service_type {
+       SPTLRPC_SVC_NULL                = 0,    /**< no security */
+       SPTLRPC_SVC_AUTH                = 1,    /**< authentication only */
+       SPTLRPC_SVC_INTG                = 2,    /**< integrity */
+       SPTLRPC_SVC_PRIV                = 3,    /**< privacy */
+       SPTLRPC_SVC_MAX,
+};
+
+enum sptlrpc_bulk_type {
+       SPTLRPC_BULK_DEFAULT        = 0,    /**< follow rpc flavor */
+       SPTLRPC_BULK_HASH              = 1,    /**< hash integrity */
+       SPTLRPC_BULK_MAX,
+};
+
+enum sptlrpc_bulk_service {
+       SPTLRPC_BULK_SVC_NULL      = 0,    /**< no security */
+       SPTLRPC_BULK_SVC_AUTH      = 1,    /**< authentication only */
+       SPTLRPC_BULK_SVC_INTG      = 2,    /**< integrity */
+       SPTLRPC_BULK_SVC_PRIV      = 3,    /**< privacy */
+       SPTLRPC_BULK_SVC_MAX,
+};
+
+/*
+ * compose/extract macros
+ */
+#define FLVR_POLICY_OFFSET           (0)
+#define FLVR_MECH_OFFSET               (4)
+#define FLVR_SVC_OFFSET                 (8)
+#define FLVR_BULK_TYPE_OFFSET     (12)
+#define FLVR_BULK_SVC_OFFSET       (16)
+
+#define MAKE_FLVR(policy, mech, svc, btype, bsvc)                     \
+       (((__u32)(policy) << FLVR_POLICY_OFFSET) |                    \
+        ((__u32)(mech) << FLVR_MECH_OFFSET) |                    \
+        ((__u32)(svc) << FLVR_SVC_OFFSET) |                        \
+        ((__u32)(btype) << FLVR_BULK_TYPE_OFFSET) |                \
+        ((__u32)(bsvc) << FLVR_BULK_SVC_OFFSET))
+
+/*
+ * extraction
+ */
+#define SPTLRPC_FLVR_POLICY(flavor)                                 \
+       ((((__u32)(flavor)) >> FLVR_POLICY_OFFSET) & 0xF)
+#define SPTLRPC_FLVR_MECH(flavor)                                     \
+       ((((__u32)(flavor)) >> FLVR_MECH_OFFSET) & 0xF)
+#define SPTLRPC_FLVR_SVC(flavor)                                       \
+       ((((__u32)(flavor)) >> FLVR_SVC_OFFSET) & 0xF)
+#define SPTLRPC_FLVR_BULK_TYPE(flavor)                           \
+       ((((__u32)(flavor)) >> FLVR_BULK_TYPE_OFFSET) & 0xF)
+#define SPTLRPC_FLVR_BULK_SVC(flavor)                             \
+       ((((__u32)(flavor)) >> FLVR_BULK_SVC_OFFSET) & 0xF)
+
+#define SPTLRPC_FLVR_BASE(flavor)                                     \
+       ((((__u32)(flavor)) >> FLVR_POLICY_OFFSET) & 0xFFF)
+#define SPTLRPC_FLVR_BASE_SUB(flavor)                             \
+       ((((__u32)(flavor)) >> FLVR_MECH_OFFSET) & 0xFF)
+
+/*
+ * gss subflavors
+ */
+#define MAKE_BASE_SUBFLVR(mech, svc)                               \
+       ((__u32)(mech) |                                                \
+        ((__u32)(svc) << (FLVR_SVC_OFFSET - FLVR_MECH_OFFSET)))
+
+#define SPTLRPC_SUBFLVR_KRB5N                                     \
+       MAKE_BASE_SUBFLVR(SPTLRPC_MECH_GSS_KRB5, SPTLRPC_SVC_NULL)
+#define SPTLRPC_SUBFLVR_KRB5A                                     \
+       MAKE_BASE_SUBFLVR(SPTLRPC_MECH_GSS_KRB5, SPTLRPC_SVC_AUTH)
+#define SPTLRPC_SUBFLVR_KRB5I                                     \
+       MAKE_BASE_SUBFLVR(SPTLRPC_MECH_GSS_KRB5, SPTLRPC_SVC_INTG)
+#define SPTLRPC_SUBFLVR_KRB5P                                     \
+       MAKE_BASE_SUBFLVR(SPTLRPC_MECH_GSS_KRB5, SPTLRPC_SVC_PRIV)
+
+/*
+ * "end user" flavors
+ */
+#define SPTLRPC_FLVR_NULL                             \
+       MAKE_FLVR(SPTLRPC_POLICY_NULL,            \
+                 SPTLRPC_MECH_NULL,                \
+                 SPTLRPC_SVC_NULL,                  \
+                 SPTLRPC_BULK_DEFAULT,          \
+                 SPTLRPC_BULK_SVC_NULL)
+#define SPTLRPC_FLVR_PLAIN                           \
+       MAKE_FLVR(SPTLRPC_POLICY_PLAIN,          \
+                 SPTLRPC_MECH_PLAIN,              \
+                 SPTLRPC_SVC_NULL,                  \
+                 SPTLRPC_BULK_HASH,                \
+                 SPTLRPC_BULK_SVC_INTG)
+#define SPTLRPC_FLVR_KRB5N                           \
+       MAKE_FLVR(SPTLRPC_POLICY_GSS,              \
+                 SPTLRPC_MECH_GSS_KRB5,                \
+                 SPTLRPC_SVC_NULL,                  \
+                 SPTLRPC_BULK_DEFAULT,          \
+                 SPTLRPC_BULK_SVC_NULL)
+#define SPTLRPC_FLVR_KRB5A                           \
+       MAKE_FLVR(SPTLRPC_POLICY_GSS,              \
+                 SPTLRPC_MECH_GSS_KRB5,                \
+                 SPTLRPC_SVC_AUTH,                  \
+                 SPTLRPC_BULK_DEFAULT,          \
+                 SPTLRPC_BULK_SVC_NULL)
+#define SPTLRPC_FLVR_KRB5I                           \
+       MAKE_FLVR(SPTLRPC_POLICY_GSS,              \
+                 SPTLRPC_MECH_GSS_KRB5,                \
+                 SPTLRPC_SVC_INTG,                  \
+                 SPTLRPC_BULK_DEFAULT,          \
+                 SPTLRPC_BULK_SVC_INTG)
+#define SPTLRPC_FLVR_KRB5P                           \
+       MAKE_FLVR(SPTLRPC_POLICY_GSS,              \
+                 SPTLRPC_MECH_GSS_KRB5,                \
+                 SPTLRPC_SVC_PRIV,                  \
+                 SPTLRPC_BULK_DEFAULT,          \
+                 SPTLRPC_BULK_SVC_PRIV)
+
+#define SPTLRPC_FLVR_DEFAULT       SPTLRPC_FLVR_NULL
+
+#define SPTLRPC_FLVR_INVALID       ((__u32) 0xFFFFFFFF)
+#define SPTLRPC_FLVR_ANY               ((__u32) 0xFFF00000)
+
+/**
+ * extract the useful part from wire flavor
+ */
+#define WIRE_FLVR(wflvr)               (((__u32) (wflvr)) & 0x000FFFFF)
+
+/** @} flavor */
+
+static inline void flvr_set_svc(__u32 *flvr, __u32 svc)
+{
+       LASSERT(svc < SPTLRPC_SVC_MAX);
+       *flvr = MAKE_FLVR(SPTLRPC_FLVR_POLICY(*flvr),
+                         SPTLRPC_FLVR_MECH(*flvr),
+                         svc,
+                         SPTLRPC_FLVR_BULK_TYPE(*flvr),
+                         SPTLRPC_FLVR_BULK_SVC(*flvr));
+}
+
+static inline void flvr_set_bulk_svc(__u32 *flvr, __u32 svc)
+{
+       LASSERT(svc < SPTLRPC_BULK_SVC_MAX);
+       *flvr = MAKE_FLVR(SPTLRPC_FLVR_POLICY(*flvr),
+                         SPTLRPC_FLVR_MECH(*flvr),
+                         SPTLRPC_FLVR_SVC(*flvr),
+                         SPTLRPC_FLVR_BULK_TYPE(*flvr),
+                         svc);
+}
+
+struct bulk_spec_hash {
+       __u8    hash_alg;
+};
+
+/**
+ * Full description of flavors being used on a ptlrpc connection, include
+ * both regular RPC and bulk transfer parts.
+ */
+struct sptlrpc_flavor {
+       /**
+        * wire flavor, should be renamed to sf_wire.
+        */
+       __u32   sf_rpc;
+       /**
+        * general flags of PTLRPC_SEC_FL_*
+        */
+       __u32   sf_flags;
+       /**
+        * rpc flavor specification
+        */
+       union {
+               /* nothing for now */
+       } u_rpc;
+       /**
+        * bulk flavor specification
+        */
+       union {
+               struct bulk_spec_hash hash;
+       } u_bulk;
+};
+
+/**
+ * identify the RPC is generated from what part of Lustre. It's encoded into
+ * RPC requests and to be checked by ptlrpc service.
+ */
+enum lustre_sec_part {
+       LUSTRE_SP_CLI      = 0,
+       LUSTRE_SP_MDT,
+       LUSTRE_SP_OST,
+       LUSTRE_SP_MGC,
+       LUSTRE_SP_MGS,
+       LUSTRE_SP_ANY      = 0xFF
+};
+
+const char *sptlrpc_part2name(enum lustre_sec_part sp);
+enum lustre_sec_part sptlrpc_target_sec_part(struct obd_device *obd);
+
+/**
+ * A rule specifies a flavor to be used by a ptlrpc connection between
+ * two Lustre parts.
+ */
+struct sptlrpc_rule {
+       __u32              sr_netid;   /* LNET network ID */
+       __u8                sr_from;    /* sec_part */
+       __u8                sr_to;      /* sec_part */
+       __u16              sr_padding;
+       struct sptlrpc_flavor   sr_flvr;
+};
+
+/**
+ * A set of rules in memory.
+ *
+ * Rules are generated and stored on MGS, and propagated to MDT, OST,
+ * and client when needed.
+ */
+struct sptlrpc_rule_set {
+       int                  srs_nslot;
+       int                  srs_nrule;
+       struct sptlrpc_rule    *srs_rules;
+};
+
+int sptlrpc_parse_flavor(const char *str, struct sptlrpc_flavor *flvr);
+int sptlrpc_flavor_has_bulk(struct sptlrpc_flavor *flvr);
+
+static inline void sptlrpc_rule_set_init(struct sptlrpc_rule_set *set)
+{
+       memset(set, 0, sizeof(*set));
+}
+
+void sptlrpc_rule_set_free(struct sptlrpc_rule_set *set);
+int  sptlrpc_rule_set_expand(struct sptlrpc_rule_set *set);
+int  sptlrpc_rule_set_merge(struct sptlrpc_rule_set *set,
+                           struct sptlrpc_rule *rule);
+int sptlrpc_rule_set_choose(struct sptlrpc_rule_set *rset,
+                           enum lustre_sec_part from,
+                           enum lustre_sec_part to,
+                           lnet_nid_t nid,
+                           struct sptlrpc_flavor *sf);
+void sptlrpc_rule_set_dump(struct sptlrpc_rule_set *set);
+
+int  sptlrpc_process_config(struct lustre_cfg *lcfg);
+void sptlrpc_conf_log_start(const char *logname);
+void sptlrpc_conf_log_stop(const char *logname);
+void sptlrpc_conf_log_update_begin(const char *logname);
+void sptlrpc_conf_log_update_end(const char *logname);
+void sptlrpc_conf_client_adapt(struct obd_device *obd);
+void sptlrpc_target_choose_flavor(struct sptlrpc_rule_set *rset,
+                                 enum lustre_sec_part from,
+                                 lnet_nid_t nid,
+                                 struct sptlrpc_flavor *flavor);
+
+/* The maximum length of security payload. 1024 is enough for Kerberos 5,
+ * and should be enough for other future mechanisms but not sure.
+ * Only used by pre-allocated request/reply pool.
+ */
+#define SPTLRPC_MAX_PAYLOAD     (1024)
+
+
+struct vfs_cred {
+       uint32_t        vc_uid;
+       uint32_t        vc_gid;
+};
+
+struct ptlrpc_ctx_ops {
+       /**
+        * To determine whether it's suitable to use the \a ctx for \a vcred.
+        */
+       int     (*match)       (struct ptlrpc_cli_ctx *ctx,
+                               struct vfs_cred *vcred);
+
+       /**
+        * To bring the \a ctx uptodate.
+        */
+       int     (*refresh)     (struct ptlrpc_cli_ctx *ctx);
+
+       /**
+        * Validate the \a ctx.
+        */
+       int     (*validate)    (struct ptlrpc_cli_ctx *ctx);
+
+       /**
+        * Force the \a ctx to die.
+        */
+       void    (*force_die)   (struct ptlrpc_cli_ctx *ctx,
+                               int grace);
+       int     (*display)     (struct ptlrpc_cli_ctx *ctx,
+                               char *buf, int bufsize);
+
+       /**
+        * Sign the request message using \a ctx.
+        *
+        * \pre req->rq_reqmsg point to request message.
+        * \pre req->rq_reqlen is the request message length.
+        * \post req->rq_reqbuf point to request message with signature.
+        * \post req->rq_reqdata_len is set to the final request message size.
+        *
+        * \see null_ctx_sign(), plain_ctx_sign(), gss_cli_ctx_sign().
+        */
+       int     (*sign) (struct ptlrpc_cli_ctx *ctx,
+                               struct ptlrpc_request *req);
+
+       /**
+        * Verify the reply message using \a ctx.
+        *
+        * \pre req->rq_repdata point to reply message with signature.
+        * \pre req->rq_repdata_len is the total reply message length.
+        * \post req->rq_repmsg point to reply message without signature.
+        * \post req->rq_replen is the reply message length.
+        *
+        * \see null_ctx_verify(), plain_ctx_verify(), gss_cli_ctx_verify().
+        */
+       int     (*verify)      (struct ptlrpc_cli_ctx *ctx,
+                               struct ptlrpc_request *req);
+
+       /**
+        * Encrypt the request message using \a ctx.
+        *
+        * \pre req->rq_reqmsg point to request message in clear text.
+        * \pre req->rq_reqlen is the request message length.
+        * \post req->rq_reqbuf point to request message.
+        * \post req->rq_reqdata_len is set to the final request message size.
+        *
+        * \see gss_cli_ctx_seal().
+        */
+       int     (*seal) (struct ptlrpc_cli_ctx *ctx,
+                               struct ptlrpc_request *req);
+
+       /**
+        * Decrypt the reply message using \a ctx.
+        *
+        * \pre req->rq_repdata point to encrypted reply message.
+        * \pre req->rq_repdata_len is the total cipher text length.
+        * \post req->rq_repmsg point to reply message in clear text.
+        * \post req->rq_replen is the reply message length in clear text.
+        *
+        * \see gss_cli_ctx_unseal().
+        */
+       int     (*unseal)      (struct ptlrpc_cli_ctx *ctx,
+                               struct ptlrpc_request *req);
+
+       /**
+        * Wrap bulk request data. This is called before wrapping RPC
+        * request message.
+        *
+        * \pre bulk buffer is descripted by desc->bd_iov and
+        * desc->bd_iov_count. note for read it's just buffer, no data
+        * need to be sent;  for write it contains data in clear text.
+        * \post when necessary, ptlrpc_bulk_sec_desc was properly prepared
+        * (usually inside of RPC request message).
+        * - encryption: cipher text bulk buffer is descripted by
+        *   desc->bd_enc_iov and desc->bd_iov_count (currently assume iov
+        *   count remains the same).
+        * - otherwise: bulk buffer is still desc->bd_iov and
+        *   desc->bd_iov_count.
+        *
+        * \return 0: success.
+        * \return -ev: error code.
+        *
+        * \see plain_cli_wrap_bulk(), gss_cli_ctx_wrap_bulk().
+        */
+       int     (*wrap_bulk)   (struct ptlrpc_cli_ctx *ctx,
+                               struct ptlrpc_request *req,
+                               struct ptlrpc_bulk_desc *desc);
+
+       /**
+        * Unwrap bulk reply data. This is called after wrapping RPC
+        * reply message.
+        *
+        * \pre bulk buffer is descripted by desc->bd_iov/desc->bd_enc_iov and
+        * desc->bd_iov_count, according to wrap_bulk().
+        * \post final bulk data in clear text is placed in buffer described
+        * by desc->bd_iov and desc->bd_iov_count.
+        * \return +ve nob of actual bulk data in clear text.
+        * \return -ve error code.
+        *
+        * \see plain_cli_unwrap_bulk(), gss_cli_ctx_unwrap_bulk().
+        */
+       int     (*unwrap_bulk) (struct ptlrpc_cli_ctx *ctx,
+                               struct ptlrpc_request *req,
+                               struct ptlrpc_bulk_desc *desc);
+};
+
+#define PTLRPC_CTX_NEW_BIT          (0)  /* newly created */
+#define PTLRPC_CTX_UPTODATE_BIT        (1)  /* uptodate */
+#define PTLRPC_CTX_DEAD_BIT        (2)  /* mark expired gracefully */
+#define PTLRPC_CTX_ERROR_BIT      (3)  /* fatal error (refresh, etc.) */
+#define PTLRPC_CTX_CACHED_BIT    (8)  /* in ctx cache (hash etc.) */
+#define PTLRPC_CTX_ETERNAL_BIT  (9)  /* always valid */
+
+#define PTLRPC_CTX_NEW          (1 << PTLRPC_CTX_NEW_BIT)
+#define PTLRPC_CTX_UPTODATE        (1 << PTLRPC_CTX_UPTODATE_BIT)
+#define PTLRPC_CTX_DEAD                (1 << PTLRPC_CTX_DEAD_BIT)
+#define PTLRPC_CTX_ERROR              (1 << PTLRPC_CTX_ERROR_BIT)
+#define PTLRPC_CTX_CACHED            (1 << PTLRPC_CTX_CACHED_BIT)
+#define PTLRPC_CTX_ETERNAL          (1 << PTLRPC_CTX_ETERNAL_BIT)
+
+#define PTLRPC_CTX_STATUS_MASK  (PTLRPC_CTX_NEW_BIT    |       \
+                                       PTLRPC_CTX_UPTODATE   |       \
+                                       PTLRPC_CTX_DEAD       |       \
+                                       PTLRPC_CTX_ERROR)
+
+struct ptlrpc_cli_ctx {
+       struct hlist_node       cc_cache;      /* linked into ctx cache */
+       atomic_t            cc_refcount;
+       struct ptlrpc_sec      *cc_sec;
+       struct ptlrpc_ctx_ops  *cc_ops;
+       unsigned long         cc_expire;     /* in seconds */
+       unsigned int        cc_early_expire:1;
+       unsigned long      cc_flags;
+       struct vfs_cred  cc_vcred;
+       spinlock_t              cc_lock;
+       struct list_head              cc_req_list;   /* waiting reqs linked here */
+       struct list_head              cc_gc_chain;   /* linked to gc chain */
+};
+
+/**
+ * client side policy operation vector.
+ */
+struct ptlrpc_sec_cops {
+       /**
+        * Given an \a imp, create and initialize a ptlrpc_sec structure.
+        * \param ctx service context:
+        * - regular import: \a ctx should be NULL;
+        * - reverse import: \a ctx is obtained from incoming request.
+        * \param flavor specify what flavor to use.
+        *
+        * When necessary, policy module is responsible for taking reference
+        * on the import.
+        *
+        * \see null_create_sec(), plain_create_sec(), gss_sec_create_kr().
+        */
+       struct ptlrpc_sec *     (*create_sec)  (struct obd_import *imp,
+                                               struct ptlrpc_svc_ctx *ctx,
+                                               struct sptlrpc_flavor *flavor);
+
+       /**
+        * Destructor of ptlrpc_sec. When called, refcount has been dropped
+        * to 0 and all contexts has been destroyed.
+        *
+        * \see null_destroy_sec(), plain_destroy_sec(), gss_sec_destroy_kr().
+        */
+       void                (*destroy_sec) (struct ptlrpc_sec *sec);
+
+       /**
+        * Notify that this ptlrpc_sec is going to die. Optionally, policy
+        * module is supposed to set sec->ps_dying and whatever necessary
+        * actions.
+        *
+        * \see plain_kill_sec(), gss_sec_kill().
+        */
+       void                (*kill_sec)    (struct ptlrpc_sec *sec);
+
+       /**
+        * Given \a vcred, lookup and/or create its context. The policy module
+        * is supposed to maintain its own context cache.
+        * XXX currently \a create and \a remove_dead is always 1, perhaps
+        * should be removed completely.
+        *
+        * \see null_lookup_ctx(), plain_lookup_ctx(), gss_sec_lookup_ctx_kr().
+        */
+       struct ptlrpc_cli_ctx * (*lookup_ctx)  (struct ptlrpc_sec *sec,
+                                               struct vfs_cred *vcred,
+                                               int create,
+                                               int remove_dead);
+
+       /**
+        * Called then the reference of \a ctx dropped to 0. The policy module
+        * is supposed to destroy this context or whatever else according to
+        * its cache maintenance mechanism.
+        *
+        * \param sync if zero, we shouldn't wait for the context being
+        * destroyed completely.
+        *
+        * \see plain_release_ctx(), gss_sec_release_ctx_kr().
+        */
+       void                (*release_ctx) (struct ptlrpc_sec *sec,
+                                               struct ptlrpc_cli_ctx *ctx,
+                                               int sync);
+
+       /**
+        * Flush the context cache.
+        *
+        * \param uid context of which user, -1 means all contexts.
+        * \param grace if zero, the PTLRPC_CTX_UPTODATE_BIT of affected
+        * contexts should be cleared immediately.
+        * \param force if zero, only idle contexts will be flushed.
+        *
+        * \see plain_flush_ctx_cache(), gss_sec_flush_ctx_cache_kr().
+        */
+       int                  (*flush_ctx_cache)
+                                              (struct ptlrpc_sec *sec,
+                                               uid_t uid,
+                                               int grace,
+                                               int force);
+
+       /**
+        * Called periodically by garbage collector to remove dead contexts
+        * from cache.
+        *
+        * \see gss_sec_gc_ctx_kr().
+        */
+       void                (*gc_ctx)      (struct ptlrpc_sec *sec);
+
+       /**
+        * Given an context \a ctx, install a corresponding reverse service
+        * context on client side.
+        * XXX currently it's only used by GSS module, maybe we should remove
+        * this from general API.
+        */
+       int                  (*install_rctx)(struct obd_import *imp,
+                                               struct ptlrpc_sec *sec,
+                                               struct ptlrpc_cli_ctx *ctx);
+
+       /**
+        * To allocate request buffer for \a req.
+        *
+        * \pre req->rq_reqmsg == NULL.
+        * \pre req->rq_reqbuf == NULL, otherwise it must be pre-allocated,
+        * we are not supposed to free it.
+        * \post if success, req->rq_reqmsg point to a buffer with size
+        * at least \a lustre_msg_size.
+        *
+        * \see null_alloc_reqbuf(), plain_alloc_reqbuf(), gss_alloc_reqbuf().
+        */
+       int                  (*alloc_reqbuf)(struct ptlrpc_sec *sec,
+                                               struct ptlrpc_request *req,
+                                               int lustre_msg_size);
+
+       /**
+        * To free request buffer for \a req.
+        *
+        * \pre req->rq_reqbuf != NULL.
+        *
+        * \see null_free_reqbuf(), plain_free_reqbuf(), gss_free_reqbuf().
+        */
+       void                (*free_reqbuf) (struct ptlrpc_sec *sec,
+                                               struct ptlrpc_request *req);
+
+       /**
+        * To allocate reply buffer for \a req.
+        *
+        * \pre req->rq_repbuf == NULL.
+        * \post if success, req->rq_repbuf point to a buffer with size
+        * req->rq_repbuf_len, the size should be large enough to receive
+        * reply which be transformed from \a lustre_msg_size of clear text.
+        *
+        * \see null_alloc_repbuf(), plain_alloc_repbuf(), gss_alloc_repbuf().
+        */
+       int                  (*alloc_repbuf)(struct ptlrpc_sec *sec,
+                                               struct ptlrpc_request *req,
+                                               int lustre_msg_size);
+
+       /**
+        * To free reply buffer for \a req.
+        *
+        * \pre req->rq_repbuf != NULL.
+        * \post req->rq_repbuf == NULL.
+        * \post req->rq_repbuf_len == 0.
+        *
+        * \see null_free_repbuf(), plain_free_repbuf(), gss_free_repbuf().
+        */
+       void                (*free_repbuf) (struct ptlrpc_sec *sec,
+                                               struct ptlrpc_request *req);
+
+       /**
+        * To expand the request buffer of \a req, thus the \a segment in
+        * the request message pointed by req->rq_reqmsg can accommodate
+        * at least \a newsize of data.
+        *
+        * \pre req->rq_reqmsg->lm_buflens[segment] < newsize.
+        *
+        * \see null_enlarge_reqbuf(), plain_enlarge_reqbuf(),
+        * gss_enlarge_reqbuf().
+        */
+       int                  (*enlarge_reqbuf)
+                                              (struct ptlrpc_sec *sec,
+                                               struct ptlrpc_request *req,
+                                               int segment, int newsize);
+       /*
+        * misc
+        */
+       int                  (*display)     (struct ptlrpc_sec *sec,
+                                               struct seq_file *seq);
+};
+
+/**
+ * server side policy operation vector.
+ */
+struct ptlrpc_sec_sops {
+       /**
+        * verify an incoming request.
+        *
+        * \pre request message is pointed by req->rq_reqbuf, size is
+        * req->rq_reqdata_len; and the message has been unpacked to
+        * host byte order.
+        *
+        * \retval SECSVC_OK success, req->rq_reqmsg point to request message
+        * in clear text, size is req->rq_reqlen; req->rq_svc_ctx is set;
+        * req->rq_sp_from is decoded from request.
+        * \retval SECSVC_COMPLETE success, the request has been fully
+        * processed, and reply message has been prepared; req->rq_sp_from is
+        * decoded from request.
+        * \retval SECSVC_DROP failed, this request should be dropped.
+        *
+        * \see null_accept(), plain_accept(), gss_svc_accept_kr().
+        */
+       int                  (*accept)      (struct ptlrpc_request *req);
+
+       /**
+        * Perform security transformation upon reply message.
+        *
+        * \pre reply message is pointed by req->rq_reply_state->rs_msg, size
+        * is req->rq_replen.
+        * \post req->rs_repdata_len is the final message size.
+        * \post req->rq_reply_off is set.
+        *
+        * \see null_authorize(), plain_authorize(), gss_svc_authorize().
+        */
+       int                  (*authorize)   (struct ptlrpc_request *req);
+
+       /**
+        * Invalidate server context \a ctx.
+        *
+        * \see gss_svc_invalidate_ctx().
+        */
+       void                (*invalidate_ctx)
+                                              (struct ptlrpc_svc_ctx *ctx);
+
+       /**
+        * Allocate a ptlrpc_reply_state.
+        *
+        * \param msgsize size of the reply message in clear text.
+        * \pre if req->rq_reply_state != NULL, then it's pre-allocated, we
+        * should simply use it; otherwise we'll responsible for allocating
+        * a new one.
+        * \post req->rq_reply_state != NULL;
+        * \post req->rq_reply_state->rs_msg != NULL;
+        *
+        * \see null_alloc_rs(), plain_alloc_rs(), gss_svc_alloc_rs().
+        */
+       int                  (*alloc_rs)    (struct ptlrpc_request *req,
+                                               int msgsize);
+
+       /**
+        * Free a ptlrpc_reply_state.
+        */
+       void                (*free_rs)     (struct ptlrpc_reply_state *rs);
+
+       /**
+        * Release the server context \a ctx.
+        *
+        * \see gss_svc_free_ctx().
+        */
+       void                (*free_ctx)    (struct ptlrpc_svc_ctx *ctx);
+
+       /**
+        * Install a reverse context based on the server context \a ctx.
+        *
+        * \see gss_svc_install_rctx_kr().
+        */
+       int                  (*install_rctx)(struct obd_import *imp,
+                                               struct ptlrpc_svc_ctx *ctx);
+
+       /**
+        * Prepare buffer for incoming bulk write.
+        *
+        * \pre desc->bd_iov and desc->bd_iov_count describes the buffer
+        * intended to receive the write.
+        *
+        * \see gss_svc_prep_bulk().
+        */
+       int                  (*prep_bulk)   (struct ptlrpc_request *req,
+                                               struct ptlrpc_bulk_desc *desc);
+
+       /**
+        * Unwrap the bulk write data.
+        *
+        * \see plain_svc_unwrap_bulk(), gss_svc_unwrap_bulk().
+        */
+       int                  (*unwrap_bulk) (struct ptlrpc_request *req,
+                                               struct ptlrpc_bulk_desc *desc);
+
+       /**
+        * Wrap the bulk read data.
+        *
+        * \see plain_svc_wrap_bulk(), gss_svc_wrap_bulk().
+        */
+       int                  (*wrap_bulk)   (struct ptlrpc_request *req,
+                                               struct ptlrpc_bulk_desc *desc);
+};
+
+struct ptlrpc_sec_policy {
+       struct module              *sp_owner;
+       char                       *sp_name;
+       __u16                      sp_policy; /* policy number */
+       struct ptlrpc_sec_cops   *sp_cops;   /* client ops */
+       struct ptlrpc_sec_sops   *sp_sops;   /* server ops */
+};
+
+#define PTLRPC_SEC_FL_REVERSE     0x0001 /* reverse sec */
+#define PTLRPC_SEC_FL_ROOTONLY   0x0002 /* treat everyone as root */
+#define PTLRPC_SEC_FL_UDESC         0x0004 /* ship udesc */
+#define PTLRPC_SEC_FL_BULK           0x0008 /* intensive bulk i/o expected */
+#define PTLRPC_SEC_FL_PAG             0x0010 /* PAG mode */
+
+/**
+ * The ptlrpc_sec represents the client side ptlrpc security facilities,
+ * each obd_import (both regular and reverse import) must associate with
+ * a ptlrpc_sec.
+ *
+ * \see sptlrpc_import_sec_adapt().
+ */
+struct ptlrpc_sec {
+       struct ptlrpc_sec_policy       *ps_policy;
+       atomic_t                    ps_refcount;
+       /** statistic only */
+       atomic_t                    ps_nctx;
+       /** unique identifier */
+       int                          ps_id;
+       struct sptlrpc_flavor      ps_flvr;
+       enum lustre_sec_part        ps_part;
+       /** after set, no more new context will be created */
+       unsigned int                ps_dying:1;
+       /** owning import */
+       struct obd_import             *ps_import;
+       spinlock_t                      ps_lock;
+
+       /*
+        * garbage collection
+        */
+       struct list_head                      ps_gc_list;
+       unsigned long                 ps_gc_interval; /* in seconds */
+       unsigned long                 ps_gc_next;     /* in seconds */
+};
+
+static inline int sec_is_reverse(struct ptlrpc_sec *sec)
+{
+       return (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_REVERSE);
+}
+
+static inline int sec_is_rootonly(struct ptlrpc_sec *sec)
+{
+       return (sec->ps_flvr.sf_flags & PTLRPC_SEC_FL_ROOTONLY);
+}
+
+
+struct ptlrpc_svc_ctx {
+       atomic_t                    sc_refcount;
+       struct ptlrpc_sec_policy       *sc_policy;
+};
+
+/*
+ * user identity descriptor
+ */
+#define LUSTRE_MAX_GROUPS             (128)
+
+struct ptlrpc_user_desc {
+       __u32      pud_uid;
+       __u32      pud_gid;
+       __u32      pud_fsuid;
+       __u32      pud_fsgid;
+       __u32      pud_cap;
+       __u32      pud_ngroups;
+       __u32      pud_groups[0];
+};
+
+/*
+ * bulk flavors
+ */
+enum sptlrpc_bulk_hash_alg {
+       BULK_HASH_ALG_NULL      = 0,
+       BULK_HASH_ALG_ADLER32,
+       BULK_HASH_ALG_CRC32,
+       BULK_HASH_ALG_MD5,
+       BULK_HASH_ALG_SHA1,
+       BULK_HASH_ALG_SHA256,
+       BULK_HASH_ALG_SHA384,
+       BULK_HASH_ALG_SHA512,
+       BULK_HASH_ALG_MAX
+};
+
+const char *sptlrpc_get_hash_name(__u8 hash_alg);
+__u8 sptlrpc_get_hash_alg(const char *algname);
+
+enum {
+       BSD_FL_ERR      = 1,
+};
+
+struct ptlrpc_bulk_sec_desc {
+       __u8        bsd_version;    /* 0 */
+       __u8        bsd_type;       /* SPTLRPC_BULK_XXX */
+       __u8        bsd_svc;    /* SPTLRPC_BULK_SVC_XXXX */
+       __u8        bsd_flags;      /* flags */
+       __u32      bsd_nob;     /* nob of bulk data */
+       __u8        bsd_data[0];    /* policy-specific token */
+};
+
+
+/*
+ * round size up to next power of 2, for slab allocation.
+ * @size must be sane (can't overflow after round up)
+ */
+static inline int size_roundup_power2(int size)
+{
+       size--;
+       size |= size >> 1;
+       size |= size >> 2;
+       size |= size >> 4;
+       size |= size >> 8;
+       size |= size >> 16;
+       size++;
+       return size;
+}
+
+/*
+ * internal support libraries
+ */
+void _sptlrpc_enlarge_msg_inplace(struct lustre_msg *msg,
+                                 int segment, int newsize);
+
+/*
+ * security policies
+ */
+int sptlrpc_register_policy(struct ptlrpc_sec_policy *policy);
+int sptlrpc_unregister_policy(struct ptlrpc_sec_policy *policy);
+
+__u32 sptlrpc_name2flavor_base(const char *name);
+const char *sptlrpc_flavor2name_base(__u32 flvr);
+char *sptlrpc_flavor2name_bulk(struct sptlrpc_flavor *sf,
+                              char *buf, int bufsize);
+char *sptlrpc_flavor2name(struct sptlrpc_flavor *sf, char *buf, int bufsize);
+char *sptlrpc_secflags2str(__u32 flags, char *buf, int bufsize);
+
+static inline
+struct ptlrpc_sec_policy *sptlrpc_policy_get(struct ptlrpc_sec_policy *policy)
+{
+       __module_get(policy->sp_owner);
+       return policy;
+}
+
+static inline
+void sptlrpc_policy_put(struct ptlrpc_sec_policy *policy)
+{
+       module_put(policy->sp_owner);
+}
+
+/*
+ * client credential
+ */
+static inline
+unsigned long cli_ctx_status(struct ptlrpc_cli_ctx *ctx)
+{
+       return (ctx->cc_flags & PTLRPC_CTX_STATUS_MASK);
+}
+
+static inline
+int cli_ctx_is_ready(struct ptlrpc_cli_ctx *ctx)
+{
+       return (cli_ctx_status(ctx) == PTLRPC_CTX_UPTODATE);
+}
+
+static inline
+int cli_ctx_is_refreshed(struct ptlrpc_cli_ctx *ctx)
+{
+       return (cli_ctx_status(ctx) != 0);
+}
+
+static inline
+int cli_ctx_is_uptodate(struct ptlrpc_cli_ctx *ctx)
+{
+       return ((ctx->cc_flags & PTLRPC_CTX_UPTODATE) != 0);
+}
+
+static inline
+int cli_ctx_is_error(struct ptlrpc_cli_ctx *ctx)
+{
+       return ((ctx->cc_flags & PTLRPC_CTX_ERROR) != 0);
+}
+
+static inline
+int cli_ctx_is_dead(struct ptlrpc_cli_ctx *ctx)
+{
+       return ((ctx->cc_flags & (PTLRPC_CTX_DEAD | PTLRPC_CTX_ERROR)) != 0);
+}
+
+static inline
+int cli_ctx_is_eternal(struct ptlrpc_cli_ctx *ctx)
+{
+       return ((ctx->cc_flags & PTLRPC_CTX_ETERNAL) != 0);
+}
+
+/*
+ * sec get/put
+ */
+struct ptlrpc_sec *sptlrpc_sec_get(struct ptlrpc_sec *sec);
+void sptlrpc_sec_put(struct ptlrpc_sec *sec);
+
+/*
+ * internal apis which only used by policy implementation
+ */
+int  sptlrpc_get_next_secid(void);
+void sptlrpc_sec_destroy(struct ptlrpc_sec *sec);
+
+/*
+ * exported client context api
+ */
+struct ptlrpc_cli_ctx *sptlrpc_cli_ctx_get(struct ptlrpc_cli_ctx *ctx);
+void sptlrpc_cli_ctx_put(struct ptlrpc_cli_ctx *ctx, int sync);
+void sptlrpc_cli_ctx_expire(struct ptlrpc_cli_ctx *ctx);
+void sptlrpc_cli_ctx_wakeup(struct ptlrpc_cli_ctx *ctx);
+int sptlrpc_cli_ctx_display(struct ptlrpc_cli_ctx *ctx, char *buf, int bufsize);
+
+/*
+ * exported client context wrap/buffers
+ */
+int sptlrpc_cli_wrap_request(struct ptlrpc_request *req);
+int sptlrpc_cli_unwrap_reply(struct ptlrpc_request *req);
+int sptlrpc_cli_alloc_reqbuf(struct ptlrpc_request *req, int msgsize);
+void sptlrpc_cli_free_reqbuf(struct ptlrpc_request *req);
+int sptlrpc_cli_alloc_repbuf(struct ptlrpc_request *req, int msgsize);
+void sptlrpc_cli_free_repbuf(struct ptlrpc_request *req);
+int sptlrpc_cli_enlarge_reqbuf(struct ptlrpc_request *req,
+                              int segment, int newsize);
+int  sptlrpc_cli_unwrap_early_reply(struct ptlrpc_request *req,
+                                   struct ptlrpc_request **req_ret);
+void sptlrpc_cli_finish_early_reply(struct ptlrpc_request *early_req);
+
+void sptlrpc_request_out_callback(struct ptlrpc_request *req);
+
+/*
+ * exported higher interface of import & request
+ */
+int sptlrpc_import_sec_adapt(struct obd_import *imp,
+                            struct ptlrpc_svc_ctx *ctx,
+                            struct sptlrpc_flavor *flvr);
+struct ptlrpc_sec *sptlrpc_import_sec_ref(struct obd_import *imp);
+void sptlrpc_import_sec_put(struct obd_import *imp);
+
+int  sptlrpc_import_check_ctx(struct obd_import *imp);
+void sptlrpc_import_flush_root_ctx(struct obd_import *imp);
+void sptlrpc_import_flush_my_ctx(struct obd_import *imp);
+void sptlrpc_import_flush_all_ctx(struct obd_import *imp);
+int  sptlrpc_req_get_ctx(struct ptlrpc_request *req);
+void sptlrpc_req_put_ctx(struct ptlrpc_request *req, int sync);
+int  sptlrpc_req_refresh_ctx(struct ptlrpc_request *req, long timeout);
+int  sptlrpc_req_replace_dead_ctx(struct ptlrpc_request *req);
+void sptlrpc_req_set_flavor(struct ptlrpc_request *req, int opcode);
+
+int sptlrpc_parse_rule(char *param, struct sptlrpc_rule *rule);
+
+/* gc */
+void sptlrpc_gc_add_sec(struct ptlrpc_sec *sec);
+void sptlrpc_gc_del_sec(struct ptlrpc_sec *sec);
+void sptlrpc_gc_add_ctx(struct ptlrpc_cli_ctx *ctx);
+
+/* misc */
+const char *sec2target_str(struct ptlrpc_sec *sec);
+/*
+ * lprocfs
+ */
+#if defined (CONFIG_PROC_FS)
+struct proc_dir_entry;
+extern struct proc_dir_entry *sptlrpc_proc_root;
+int sptlrpc_lprocfs_cliobd_attach(struct obd_device *dev);
+#else
+#define sptlrpc_proc_root      NULL
+static inline int sptlrpc_lprocfs_cliobd_attach(struct obd_device *dev)
+{ return 0; }
+#endif
+
+/*
+ * server side
+ */
+enum secsvc_accept_res {
+       SECSVC_OK       = 0,
+       SECSVC_COMPLETE,
+       SECSVC_DROP,
+};
+
+int  sptlrpc_svc_unwrap_request(struct ptlrpc_request *req);
+int  sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen);
+int  sptlrpc_svc_wrap_reply(struct ptlrpc_request *req);
+void sptlrpc_svc_free_rs(struct ptlrpc_reply_state *rs);
+void sptlrpc_svc_ctx_addref(struct ptlrpc_request *req);
+void sptlrpc_svc_ctx_decref(struct ptlrpc_request *req);
+void sptlrpc_svc_ctx_invalidate(struct ptlrpc_request *req);
+
+int  sptlrpc_target_export_check(struct obd_export *exp,
+                                struct ptlrpc_request *req);
+void sptlrpc_target_update_exp_flavor(struct obd_device *obd,
+                                     struct sptlrpc_rule_set *rset);
+
+/*
+ * reverse context
+ */
+int sptlrpc_svc_install_rvs_ctx(struct obd_import *imp,
+                               struct ptlrpc_svc_ctx *ctx);
+int sptlrpc_cli_install_rvs_ctx(struct obd_import *imp,
+                               struct ptlrpc_cli_ctx *ctx);
+
+/* bulk security api */
+int sptlrpc_enc_pool_add_user(void);
+int sptlrpc_enc_pool_del_user(void);
+int  sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc);
+void sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc *desc);
+
+int sptlrpc_cli_wrap_bulk(struct ptlrpc_request *req,
+                         struct ptlrpc_bulk_desc *desc);
+int sptlrpc_cli_unwrap_bulk_read(struct ptlrpc_request *req,
+                                struct ptlrpc_bulk_desc *desc,
+                                int nob);
+int sptlrpc_cli_unwrap_bulk_write(struct ptlrpc_request *req,
+                                 struct ptlrpc_bulk_desc *desc);
+
+/* bulk helpers (internal use only by policies) */
+int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg,
+                             void *buf, int buflen);
+
+int bulk_sec_desc_unpack(struct lustre_msg *msg, int offset, int swabbed);
+
+/* user descriptor helpers */
+static inline int sptlrpc_user_desc_size(int ngroups)
+{
+       return sizeof(struct ptlrpc_user_desc) + ngroups * sizeof(__u32);
+}
+
+int sptlrpc_current_user_desc_size(void);
+int sptlrpc_pack_user_desc(struct lustre_msg *msg, int offset);
+int sptlrpc_unpack_user_desc(struct lustre_msg *req, int offset, int swabbed);
+
+
+#define CFS_CAP_CHOWN_MASK (1 << CFS_CAP_CHOWN)
+#define CFS_CAP_SYS_RESOURCE_MASK (1 << CFS_CAP_SYS_RESOURCE)
+
+enum {
+       LUSTRE_SEC_NONE  = 0,
+       LUSTRE_SEC_REMOTE       = 1,
+       LUSTRE_SEC_SPECIFY      = 2,
+       LUSTRE_SEC_ALL    = 3
+};
+
+/** @} sptlrpc */
+
+#endif /* _LUSTRE_SEC_H_ */