drivers/staging/lustre/lustre/ptlrpc/gss/sec_gss.c

/*
 * Modifications for Lustre
 *
 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
 *
 * Copyright (c) 2011, 2012, Intel Corporation.
 *
 * Author: Eric Mei <ericm@clusterfs.com>
 */

/*
 * linux/net/sunrpc/auth_gss.c
 *
 * RPCSEC_GSS client authentication.
 *
 *  Copyright (c) 2000 The Regents of the University of Michigan.
 *  All rights reserved.
 *
 *  Dug Song       <dugsong@monkey.org>
 *  Andy Adamson   <andros@umich.edu>
 *
 *  Redistribution and use in source and binary forms, with or without
 *  modification, are permitted provided that the following conditions
 *  are met:
 *
 *  1. Redistributions of source code must retain the above copyright
 *     notice, this list of conditions and the following disclaimer.
 *  2. Redistributions in binary form must reproduce the above copyright
 *     notice, this list of conditions and the following disclaimer in the
 *     documentation and/or other materials provided with the distribution.
 *  3. Neither the name of the University nor the names of its
 *     contributors may be used to endorse or promote products derived
 *     from this software without specific prior written permission.
 *
 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 */

#define DEBUG_SUBSYSTEM S_SEC
#include <linux/init.h>
#include <linux/module.h>
#include <linux/slab.h>
#include <linux/dcache.h>
#include <linux/fs.h>
#include <linux/mutex.h>
#include <asm/atomic.h>

#include <obd.h>
#include <obd_class.h>
#include <obd_support.h>
#include <obd_cksum.h>
#include <lustre/lustre_idl.h>
#include <lustre_net.h>
#include <lustre_import.h>
#include <lustre_sec.h>

#include "gss_err.h"
#include "gss_internal.h"
#include "gss_api.h"

#include <linux/crypto.h>
#include <linux/crc32.h>

/*
 * early reply have fixed size, respectively in privacy and integrity mode.
 * so we calculate them only once.
 */
static int gss_at_reply_off_integ;
static int gss_at_reply_off_priv;


static inline int msg_last_segidx(struct lustre_msg *msg)
{
	LASSERT(msg->lm_bufcount > 0);
	return msg->lm_bufcount - 1;
}
static inline int msg_last_seglen(struct lustre_msg *msg)
{
	return msg->lm_buflens[msg_last_segidx(msg)];
}

/********************************************
 * wire data swabber			*
 ********************************************/

static
void gss_header_swabber(struct gss_header *ghdr)
{
	__swab32s(&ghdr->gh_flags);
	__swab32s(&ghdr->gh_proc);
	__swab32s(&ghdr->gh_seq);
	__swab32s(&ghdr->gh_svc);
	__swab32s(&ghdr->gh_pad1);
	__swab32s(&ghdr->gh_handle.len);
}

struct gss_header *gss_swab_header(struct lustre_msg *msg, int segment,
				   int swabbed)
{
	struct gss_header *ghdr;

	ghdr = lustre_msg_buf(msg, segment, sizeof(*ghdr));
	if (ghdr == NULL)
		return NULL;

	if (swabbed)
		gss_header_swabber(ghdr);

	if (sizeof(*ghdr) + ghdr->gh_handle.len > msg->lm_buflens[segment]) {
		CERROR("gss header has length %d, now %u received\n",
		       (int) sizeof(*ghdr) + ghdr->gh_handle.len,
		       msg->lm_buflens[segment]);
		return NULL;
	}

	return ghdr;
}

#if 0
static
void gss_netobj_swabber(netobj_t *obj)
{
	__swab32s(&obj->len);
}

netobj_t *gss_swab_netobj(struct lustre_msg *msg, int segment)
{
	netobj_t  *obj;

	obj = lustre_swab_buf(msg, segment, sizeof(*obj), gss_netobj_swabber);
	if (obj && sizeof(*obj) + obj->len > msg->lm_buflens[segment]) {
		CERROR("netobj require length %u but only %u received\n",
		       (unsigned int) sizeof(*obj) + obj->len,
		       msg->lm_buflens[segment]);
		return NULL;
	}

	return obj;
}
#endif

/*
 * payload should be obtained from mechanism. but currently since we
 * only support kerberos, we could simply use fixed value.
 * krb5 "meta" data:
 *  - krb5 header:      16
 *  - krb5 checksum:    20
 *
 * for privacy mode, payload also include the cipher text which has the same
 * size as plain text, plus possible confounder, padding both at maximum cipher
 * block size.
 */
#define GSS_KRB5_INTEG_MAX_PAYLOAD      (40)

static inline
int gss_mech_payload(struct gss_ctx *mechctx, int msgsize, int privacy)
{
	if (privacy)
		return GSS_KRB5_INTEG_MAX_PAYLOAD + 16 + 16 + 16 + msgsize;
	else
		return GSS_KRB5_INTEG_MAX_PAYLOAD;
}

/*
 * return signature size, otherwise < 0 to indicate error
 */
static int gss_sign_msg(struct lustre_msg *msg,
			struct gss_ctx *mechctx,
			enum lustre_sec_part sp,
			__u32 flags, __u32 proc, __u32 seq, __u32 svc,
			rawobj_t *handle)
{
	struct gss_header      *ghdr;
	rawobj_t		text[4], mic;
	int		     textcnt, max_textcnt, mic_idx;
	__u32		   major;

	LASSERT(msg->lm_bufcount >= 2);

	/* gss hdr */
	LASSERT(msg->lm_buflens[0] >=
		sizeof(*ghdr) + (handle ? handle->len : 0));
	ghdr = lustre_msg_buf(msg, 0, 0);

	ghdr->gh_version = PTLRPC_GSS_VERSION;
	ghdr->gh_sp = (__u8) sp;
	ghdr->gh_flags = flags;
	ghdr->gh_proc = proc;
	ghdr->gh_seq = seq;
	ghdr->gh_svc = svc;
	if (!handle) {
		/* fill in a fake one */
		ghdr->gh_handle.len = 0;
	} else {
		ghdr->gh_handle.len = handle->len;
		memcpy(ghdr->gh_handle.data, handle->data, handle->len);
	}

	/* no actual signature for null mode */
	if (svc == SPTLRPC_SVC_NULL)
		return lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);

	/* MIC */
	mic_idx = msg_last_segidx(msg);
	max_textcnt = (svc == SPTLRPC_SVC_AUTH) ? 1 : mic_idx;

	for (textcnt = 0; textcnt < max_textcnt; textcnt++) {
		text[textcnt].len = msg->lm_buflens[textcnt];
		text[textcnt].data = lustre_msg_buf(msg, textcnt, 0);
	}

	mic.len = msg->lm_buflens[mic_idx];
	mic.data = lustre_msg_buf(msg, mic_idx, 0);

	major = lgss_get_mic(mechctx, textcnt, text, 0, NULL, &mic);
	if (major != GSS_S_COMPLETE) {
		CERROR("fail to generate MIC: %08x\n", major);
		return -EPERM;
	}
	LASSERT(mic.len <= msg->lm_buflens[mic_idx]);

	return lustre_shrink_msg(msg, mic_idx, mic.len, 0);
}

/*
 * return gss error
 */
static
__u32 gss_verify_msg(struct lustre_msg *msg,
		     struct gss_ctx *mechctx,
		     __u32 svc)
{
	rawobj_t	text[4], mic;
	int	     textcnt, max_textcnt;
	int	     mic_idx;
	__u32	   major;

	LASSERT(msg->lm_bufcount >= 2);

	if (svc == SPTLRPC_SVC_NULL)
		return GSS_S_COMPLETE;

	mic_idx = msg_last_segidx(msg);
	max_textcnt = (svc == SPTLRPC_SVC_AUTH) ? 1 : mic_idx;

	for (textcnt = 0; textcnt < max_textcnt; textcnt++) {
		text[textcnt].len = msg->lm_buflens[textcnt];
		text[textcnt].data = lustre_msg_buf(msg, textcnt, 0);
	}

	mic.len = msg->lm_buflens[mic_idx];
	mic.data = lustre_msg_buf(msg, mic_idx, 0);

	major = lgss_verify_mic(mechctx, textcnt, text, 0, NULL, &mic);
	if (major != GSS_S_COMPLETE)
		CERROR("mic verify error: %08x\n", major);

	return major;
}

/*
 * return gss error code
 */
static
__u32 gss_unseal_msg(struct gss_ctx *mechctx,
		   struct lustre_msg *msgbuf,
		   int *msg_len, int msgbuf_len)
{
	rawobj_t		 clear_obj, hdrobj, token;
	__u8		    *clear_buf;
	int		      clear_buflen;
	__u32		    major;

	if (msgbuf->lm_bufcount != 2) {
		CERROR("invalid bufcount %d\n", msgbuf->lm_bufcount);
		return GSS_S_FAILURE;
	}

	/* allocate a temporary clear text buffer, same sized as token,
	 * we assume the final clear text size <= token size */
	clear_buflen = lustre_msg_buflen(msgbuf, 1);
	OBD_ALLOC_LARGE(clear_buf, clear_buflen);
	if (!clear_buf)
		return GSS_S_FAILURE;

	/* buffer objects */
	hdrobj.len = lustre_msg_buflen(msgbuf, 0);
	hdrobj.data = lustre_msg_buf(msgbuf, 0, 0);
	token.len = lustre_msg_buflen(msgbuf, 1);
	token.data = lustre_msg_buf(msgbuf, 1, 0);
	clear_obj.len = clear_buflen;
	clear_obj.data = clear_buf;

	major = lgss_unwrap(mechctx, &hdrobj, &token, &clear_obj);
	if (major != GSS_S_COMPLETE) {
		CERROR("unwrap message error: %08x\n", major);
		GOTO(out_free, major = GSS_S_FAILURE);
	}
	LASSERT(clear_obj.len <= clear_buflen);
	LASSERT(clear_obj.len <= msgbuf_len);

	/* now the decrypted message */
	memcpy(msgbuf, clear_obj.data, clear_obj.len);
	*msg_len = clear_obj.len;

	major = GSS_S_COMPLETE;
out_free:
	OBD_FREE_LARGE(clear_buf, clear_buflen);
	return major;
}

/********************************************
 * gss client context manipulation helpers  *
 ********************************************/

int cli_ctx_expire(struct ptlrpc_cli_ctx *ctx)
{
	LASSERT(atomic_read(&ctx->cc_refcount));

	if (!test_and_set_bit(PTLRPC_CTX_DEAD_BIT, &ctx->cc_flags)) {
		if (!ctx->cc_early_expire)
			clear_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags);

		CWARN("ctx %p(%u->%s) get expired: %lu(%+lds)\n",
		      ctx, ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec),
		      ctx->cc_expire,
		      ctx->cc_expire == 0 ? 0 :
		      cfs_time_sub(ctx->cc_expire, cfs_time_current_sec()));

		sptlrpc_cli_ctx_wakeup(ctx);
		return 1;
	}

	return 0;
}

/*
 * return 1 if the context is dead.
 */
int cli_ctx_check_death(struct ptlrpc_cli_ctx *ctx)
{
	if (unlikely(cli_ctx_is_dead(ctx)))
		return 1;

	/* expire is 0 means never expire. a newly created gss context
	 * which during upcall may has 0 expiration */
	if (ctx->cc_expire == 0)
		return 0;

	/* check real expiration */
	if (cfs_time_after(ctx->cc_expire, cfs_time_current_sec()))
		return 0;

	cli_ctx_expire(ctx);
	return 1;
}

void gss_cli_ctx_uptodate(struct gss_cli_ctx *gctx)
{
	struct ptlrpc_cli_ctx  *ctx = &gctx->gc_base;
	unsigned long	   ctx_expiry;

	if (lgss_inquire_context(gctx->gc_mechctx, &ctx_expiry)) {
		CERROR("ctx %p(%u): unable to inquire, expire it now\n",
		       gctx, ctx->cc_vcred.vc_uid);
		ctx_expiry = 1; /* make it expired now */
	}

	ctx->cc_expire = gss_round_ctx_expiry(ctx_expiry,
					      ctx->cc_sec->ps_flvr.sf_flags);

	/* At this point this ctx might have been marked as dead by
	 * someone else, in which case nobody will make further use
	 * of it. we don't care, and mark it UPTODATE will help
	 * destroying server side context when it be destroied. */
	set_bit(PTLRPC_CTX_UPTODATE_BIT, &ctx->cc_flags);

	if (sec_is_reverse(ctx->cc_sec)) {
		CWARN("server installed reverse ctx %p idx "LPX64", "
		      "expiry %lu(%+lds)\n", ctx,
		      gss_handle_to_u64(&gctx->gc_handle),
		      ctx->cc_expire, ctx->cc_expire - cfs_time_current_sec());
	} else {
		CWARN("client refreshed ctx %p idx "LPX64" (%u->%s), "
		      "expiry %lu(%+lds)\n", ctx,
		      gss_handle_to_u64(&gctx->gc_handle),
		      ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec),
		      ctx->cc_expire, ctx->cc_expire - cfs_time_current_sec());

		/* install reverse svc ctx for root context */
		if (ctx->cc_vcred.vc_uid == 0)
			gss_sec_install_rctx(ctx->cc_sec->ps_import,
					     ctx->cc_sec, ctx);
	}

	sptlrpc_cli_ctx_wakeup(ctx);
}

static void gss_cli_ctx_finalize(struct gss_cli_ctx *gctx)
{
	LASSERT(gctx->gc_base.cc_sec);

	if (gctx->gc_mechctx) {
		lgss_delete_sec_context(&gctx->gc_mechctx);
		gctx->gc_mechctx = NULL;
	}

	if (!rawobj_empty(&gctx->gc_svc_handle)) {
		/* forward ctx: mark buddy reverse svcctx soon-expire. */
		if (!sec_is_reverse(gctx->gc_base.cc_sec) &&
		    !rawobj_empty(&gctx->gc_svc_handle))
			gss_svc_upcall_expire_rvs_ctx(&gctx->gc_svc_handle);

		rawobj_free(&gctx->gc_svc_handle);
	}

	rawobj_free(&gctx->gc_handle);
}

/*
 * Based on sequence number algorithm as specified in RFC 2203.
 *
 * modified for our own problem: arriving request has valid sequence number,
 * but unwrapping request might cost a long time, after that its sequence
 * are not valid anymore (fall behind the window). It rarely happen, mostly
 * under extreme load.
 *
 * note we should not check sequence before verify the integrity of incoming
 * request, because just one attacking request with high sequence number might
 * cause all following request be dropped.
 *
 * so here we use a multi-phase approach: prepare 2 sequence windows,
 * "main window" for normal sequence and "back window" for fall behind sequence.
 * and 3-phase checking mechanism:
 *  0 - before integrity verification, perform a initial sequence checking in
 *      main window, which only try and don't actually set any bits. if the
 *      sequence is high above the window or fit in the window and the bit
 *      is 0, then accept and proceed to integrity verification. otherwise
 *      reject this sequence.
 *  1 - after integrity verification, check in main window again. if this
 *      sequence is high above the window or fit in the window and the bit
 *      is 0, then set the bit and accept; if it fit in the window but bit
 *      already set, then reject; if it fall behind the window, then proceed
 *      to phase 2.
 *  2 - check in back window. if it is high above the window or fit in the
 *      window and the bit is 0, then set the bit and accept. otherwise reject.
 *
 * return value:
 *   1: looks like a replay
 *   0: is ok
 *  -1: is a replay
 *
 * note phase 0 is necessary, because otherwise replay attacking request of
 * sequence which between the 2 windows can't be detected.
 *
 * this mechanism can't totally solve the problem, but could help much less
 * number of valid requests be dropped.
 */
static
int gss_do_check_seq(unsigned long *window, __u32 win_size, __u32 *max_seq,
		     __u32 seq_num, int phase)
{
	LASSERT(phase >= 0 && phase <= 2);

	if (seq_num > *max_seq) {
		/*
		 * 1. high above the window
		 */
		if (phase == 0)
			return 0;

		if (seq_num >= *max_seq + win_size) {
			memset(window, 0, win_size / 8);
			*max_seq = seq_num;
		} else {
			while (*max_seq < seq_num) {
				(*max_seq)++;
				__clear_bit((*max_seq) % win_size, window);
			}
		}
		__set_bit(seq_num % win_size, window);
	} else if (seq_num + win_size <= *max_seq) {
		/*
		 * 2. low behind the window
		 */
		if (phase == 0 || phase == 2)
			goto replay;

		CWARN("seq %u is %u behind (size %d), check backup window\n",
		      seq_num, *max_seq - win_size - seq_num, win_size);
		return 1;
	} else {
		/*
		 * 3. fit into the window
		 */
		switch (phase) {
		case 0:
			if (test_bit(seq_num % win_size, window))
				goto replay;
			break;
		case 1:
		case 2:
		     if (__test_and_set_bit(seq_num % win_size, window))
				goto replay;
			break;
		}
	}

	return 0;

replay:
	CERROR("seq %u (%s %s window) is a replay: max %u, winsize %d\n",
	       seq_num,
	       seq_num + win_size > *max_seq ? "in" : "behind",
	       phase == 2 ? "backup " : "main",
	       *max_seq, win_size);
	return -1;
}

/*
 * Based on sequence number algorithm as specified in RFC 2203.
 *
 * if @set == 0: initial check, don't set any bit in window
 * if @sec == 1: final check, set bit in window
 */
int gss_check_seq_num(struct gss_svc_seq_data *ssd, __u32 seq_num, int set)
{
	int rc = 0;

	spin_lock(&ssd->ssd_lock);

	if (set == 0) {
		/*
		 * phase 0 testing
		 */
		rc = gss_do_check_seq(ssd->ssd_win_main, GSS_SEQ_WIN_MAIN,
				      &ssd->ssd_max_main, seq_num, 0);
		if (unlikely(rc))
			gss_stat_oos_record_svc(0, 1);
	} else {
		/*
		 * phase 1 checking main window
		 */
		rc = gss_do_check_seq(ssd->ssd_win_main, GSS_SEQ_WIN_MAIN,
				      &ssd->ssd_max_main, seq_num, 1);
		switch (rc) {
		case -1:
			gss_stat_oos_record_svc(1, 1);
			/* fall through */
		case 0:
			goto exit;
		}
		/*
		 * phase 2 checking back window
		 */
		rc = gss_do_check_seq(ssd->ssd_win_back, GSS_SEQ_WIN_BACK,
				      &ssd->ssd_max_back, seq_num, 2);
		if (rc)
			gss_stat_oos_record_svc(2, 1);
		else
			gss_stat_oos_record_svc(2, 0);
	}
exit:
	spin_unlock(&ssd->ssd_lock);
	return rc;
}

/***************************************
 * cred APIs			   *
 ***************************************/

static inline int gss_cli_payload(struct ptlrpc_cli_ctx *ctx,
				  int msgsize, int privacy)
{
	return gss_mech_payload(NULL, msgsize, privacy);
}

static int gss_cli_bulk_payload(struct ptlrpc_cli_ctx *ctx,
				struct sptlrpc_flavor *flvr,
				int reply, int read)
{
	int     payload = sizeof(struct ptlrpc_bulk_sec_desc);

	LASSERT(SPTLRPC_FLVR_BULK_TYPE(flvr->sf_rpc) == SPTLRPC_BULK_DEFAULT);

	if ((!reply && !read) || (reply && read)) {
		switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
		case SPTLRPC_BULK_SVC_NULL:
			break;
		case SPTLRPC_BULK_SVC_INTG:
			payload += gss_cli_payload(ctx, 0, 0);
			break;
		case SPTLRPC_BULK_SVC_PRIV:
			payload += gss_cli_payload(ctx, 0, 1);
			break;
		case SPTLRPC_BULK_SVC_AUTH:
		default:
			LBUG();
		}
	}

	return payload;
}

int gss_cli_ctx_match(struct ptlrpc_cli_ctx *ctx, struct vfs_cred *vcred)
{
	return (ctx->cc_vcred.vc_uid == vcred->vc_uid);
}

void gss_cli_ctx_flags2str(unsigned long flags, char *buf, int bufsize)
{
	buf[0] = '\0';

	if (flags & PTLRPC_CTX_NEW)
		strncat(buf, "new,", bufsize);
	if (flags & PTLRPC_CTX_UPTODATE)
		strncat(buf, "uptodate,", bufsize);
	if (flags & PTLRPC_CTX_DEAD)
		strncat(buf, "dead,", bufsize);
	if (flags & PTLRPC_CTX_ERROR)
		strncat(buf, "error,", bufsize);
	if (flags & PTLRPC_CTX_CACHED)
		strncat(buf, "cached,", bufsize);
	if (flags & PTLRPC_CTX_ETERNAL)
		strncat(buf, "eternal,", bufsize);
	if (buf[0] == '\0')
		strncat(buf, "-,", bufsize);

	buf[strlen(buf) - 1] = '\0';
}

int gss_cli_ctx_sign(struct ptlrpc_cli_ctx *ctx,
		     struct ptlrpc_request *req)
{
	struct gss_cli_ctx      *gctx = ctx2gctx(ctx);
	__u32		    flags = 0, seq, svc;
	int		      rc;

	LASSERT(req->rq_reqbuf);
	LASSERT(req->rq_reqbuf->lm_bufcount >= 2);
	LASSERT(req->rq_cli_ctx == ctx);

	/* nothing to do for context negotiation RPCs */
	if (req->rq_ctx_init)
		return 0;

	svc = SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc);
	if (req->rq_pack_bulk)
		flags |= LUSTRE_GSS_PACK_BULK;
	if (req->rq_pack_udesc)
		flags |= LUSTRE_GSS_PACK_USER;

redo:
	seq = atomic_inc_return(&gctx->gc_seq);

	rc = gss_sign_msg(req->rq_reqbuf, gctx->gc_mechctx,
			  ctx->cc_sec->ps_part,
			  flags, gctx->gc_proc, seq, svc,
			  &gctx->gc_handle);
	if (rc < 0)
		return rc;

	/* gss_sign_msg() msg might take long time to finish, in which period
	 * more rpcs could be wrapped up and sent out. if we found too many
	 * of them we should repack this rpc, because sent it too late might
	 * lead to the sequence number fall behind the window on server and
	 * be dropped. also applies to gss_cli_ctx_seal().
	 *
	 * Note: null mode dosen't check sequence number. */
	if (svc != SPTLRPC_SVC_NULL &&
	    atomic_read(&gctx->gc_seq) - seq > GSS_SEQ_REPACK_THRESHOLD) {
		int behind = atomic_read(&gctx->gc_seq) - seq;

		gss_stat_oos_record_cli(behind);
		CWARN("req %p: %u behind, retry signing\n", req, behind);
		goto redo;
	}

	req->rq_reqdata_len = rc;
	return 0;
}

static
int gss_cli_ctx_handle_err_notify(struct ptlrpc_cli_ctx *ctx,
				  struct ptlrpc_request *req,
				  struct gss_header *ghdr)
{
	struct gss_err_header *errhdr;
	int rc;

	LASSERT(ghdr->gh_proc == PTLRPC_GSS_PROC_ERR);

	errhdr = (struct gss_err_header *) ghdr;

	CWARN("req x"LPU64"/t"LPU64", ctx %p idx "LPX64"(%u->%s): "
	      "%sserver respond (%08x/%08x)\n",
	      req->rq_xid, req->rq_transno, ctx,
	      gss_handle_to_u64(&ctx2gctx(ctx)->gc_handle),
	      ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec),
	      sec_is_reverse(ctx->cc_sec) ? "reverse" : "",
	      errhdr->gh_major, errhdr->gh_minor);

	/* context fini rpc, let it failed */
	if (req->rq_ctx_fini) {
		CWARN("context fini rpc failed\n");
		return -EINVAL;
	}

	/* reverse sec, just return error, don't expire this ctx because it's
	 * crucial to callback rpcs. note if the callback rpc failed because
	 * of bit flip during network transfer, the client will be evicted
	 * directly. so more gracefully we probably want let it retry for
	 * number of times. */
	if (sec_is_reverse(ctx->cc_sec))
		return -EINVAL;

	if (errhdr->gh_major != GSS_S_NO_CONTEXT &&
	    errhdr->gh_major != GSS_S_BAD_SIG)
		return -EACCES;

	/* server return NO_CONTEXT might be caused by context expire
	 * or server reboot/failover. we try to refresh a new ctx which
	 * be transparent to upper layer.
	 *
	 * In some cases, our gss handle is possible to be incidentally
	 * identical to another handle since the handle itself is not
	 * fully random. In krb5 case, the GSS_S_BAD_SIG will be
	 * returned, maybe other gss error for other mechanism.
	 *
	 * if we add new mechanism, make sure the correct error are
	 * returned in this case. */
	CWARN("%s: server might lost the context, retrying\n",
	      errhdr->gh_major == GSS_S_NO_CONTEXT ?  "NO_CONTEXT" : "BAD_SIG");

	sptlrpc_cli_ctx_expire(ctx);

	/* we need replace the ctx right here, otherwise during
	 * resent we'll hit the logic in sptlrpc_req_refresh_ctx()
	 * which keep the ctx with RESEND flag, thus we'll never
	 * get rid of this ctx. */
	rc = sptlrpc_req_replace_dead_ctx(req);
	if (rc == 0)
		req->rq_resend = 1;

	return rc;
}

int gss_cli_ctx_verify(struct ptlrpc_cli_ctx *ctx,
		       struct ptlrpc_request *req)
{
	struct gss_cli_ctx     *gctx;
	struct gss_header      *ghdr, *reqhdr;
	struct lustre_msg      *msg = req->rq_repdata;
	__u32		   major;
	int		     pack_bulk, swabbed, rc = 0;

	LASSERT(req->rq_cli_ctx == ctx);
	LASSERT(msg);

	gctx = container_of(ctx, struct gss_cli_ctx, gc_base);

	/* special case for context negotiation, rq_repmsg/rq_replen actually
	 * are not used currently. but early reply always be treated normally */
	if (req->rq_ctx_init && !req->rq_early) {
		req->rq_repmsg = lustre_msg_buf(msg, 1, 0);
		req->rq_replen = msg->lm_buflens[1];
		return 0;
	}

	if (msg->lm_bufcount < 2 || msg->lm_bufcount > 4) {
		CERROR("unexpected bufcount %u\n", msg->lm_bufcount);
		return -EPROTO;
	}

	swabbed = ptlrpc_rep_need_swab(req);

	ghdr = gss_swab_header(msg, 0, swabbed);
	if (ghdr == NULL) {
		CERROR("can't decode gss header\n");
		return -EPROTO;
	}

	/* sanity checks */
	reqhdr = lustre_msg_buf(msg, 0, sizeof(*reqhdr));
	LASSERT(reqhdr);

	if (ghdr->gh_version != reqhdr->gh_version) {
		CERROR("gss version %u mismatch, expect %u\n",
		       ghdr->gh_version, reqhdr->gh_version);
		return -EPROTO;
	}

	switch (ghdr->gh_proc) {
	case PTLRPC_GSS_PROC_DATA:
		pack_bulk = ghdr->gh_flags & LUSTRE_GSS_PACK_BULK;

		if (!req->rq_early &&
		    !equi(req->rq_pack_bulk == 1, pack_bulk)) {
			CERROR("%s bulk flag in reply\n",
			       req->rq_pack_bulk ? "missing" : "unexpected");
			return -EPROTO;
		}

		if (ghdr->gh_seq != reqhdr->gh_seq) {
			CERROR("seqnum %u mismatch, expect %u\n",
			       ghdr->gh_seq, reqhdr->gh_seq);
			return -EPROTO;
		}

		if (ghdr->gh_svc != reqhdr->gh_svc) {
			CERROR("svc %u mismatch, expect %u\n",
			       ghdr->gh_svc, reqhdr->gh_svc);
			return -EPROTO;
		}

		if (swabbed)
			gss_header_swabber(ghdr);

		major = gss_verify_msg(msg, gctx->gc_mechctx, reqhdr->gh_svc);
		if (major != GSS_S_COMPLETE) {
			CERROR("failed to verify reply: %x\n", major);
			return -EPERM;
		}

		if (req->rq_early && reqhdr->gh_svc == SPTLRPC_SVC_NULL) {
			__u32 cksum;

			cksum = crc32_le(!(__u32) 0,
					 lustre_msg_buf(msg, 1, 0),
					 lustre_msg_buflen(msg, 1));
			if (cksum != msg->lm_cksum) {
				CWARN("early reply checksum mismatch: "
				      "%08x != %08x\n", cksum, msg->lm_cksum);
				return -EPROTO;
			}
		}

		if (pack_bulk) {
			/* bulk checksum is right after the lustre msg */
			if (msg->lm_bufcount < 3) {
				CERROR("Invalid reply bufcount %u\n",
				       msg->lm_bufcount);
				return -EPROTO;
			}

			rc = bulk_sec_desc_unpack(msg, 2, swabbed);
			if (rc) {
				CERROR("unpack bulk desc: %d\n", rc);
				return rc;
			}
		}

		req->rq_repmsg = lustre_msg_buf(msg, 1, 0);
		req->rq_replen = msg->lm_buflens[1];
		break;
	case PTLRPC_GSS_PROC_ERR:
		if (req->rq_early) {
			CERROR("server return error with early reply\n");
			rc = -EPROTO;
		} else {
			rc = gss_cli_ctx_handle_err_notify(ctx, req, ghdr);
		}
		break;
	default:
		CERROR("unknown gss proc %d\n", ghdr->gh_proc);
		rc = -EPROTO;
	}

	return rc;
}

int gss_cli_ctx_seal(struct ptlrpc_cli_ctx *ctx,
		     struct ptlrpc_request *req)
{
	struct gss_cli_ctx      *gctx;
	rawobj_t		 hdrobj, msgobj, token;
	struct gss_header       *ghdr;
	__u32		    buflens[2], major;
	int		      wiresize, rc;

	LASSERT(req->rq_clrbuf);
	LASSERT(req->rq_cli_ctx == ctx);
	LASSERT(req->rq_reqlen);

	gctx = container_of(ctx, struct gss_cli_ctx, gc_base);

	/* final clear data length */
	req->rq_clrdata_len = lustre_msg_size_v2(req->rq_clrbuf->lm_bufcount,
						 req->rq_clrbuf->lm_buflens);

	/* calculate wire data length */
	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	buflens[1] = gss_cli_payload(&gctx->gc_base, req->rq_clrdata_len, 1);
	wiresize = lustre_msg_size_v2(2, buflens);

	/* allocate wire buffer */
	if (req->rq_pool) {
		/* pre-allocated */
		LASSERT(req->rq_reqbuf);
		LASSERT(req->rq_reqbuf != req->rq_clrbuf);
		LASSERT(req->rq_reqbuf_len >= wiresize);
	} else {
		OBD_ALLOC_LARGE(req->rq_reqbuf, wiresize);
		if (!req->rq_reqbuf)
			return -ENOMEM;
		req->rq_reqbuf_len = wiresize;
	}

	lustre_init_msg_v2(req->rq_reqbuf, 2, buflens, NULL);
	req->rq_reqbuf->lm_secflvr = req->rq_flvr.sf_rpc;

	/* gss header */
	ghdr = lustre_msg_buf(req->rq_reqbuf, 0, 0);
	ghdr->gh_version = PTLRPC_GSS_VERSION;
	ghdr->gh_sp = (__u8) ctx->cc_sec->ps_part;
	ghdr->gh_flags = 0;
	ghdr->gh_proc = gctx->gc_proc;
	ghdr->gh_svc = SPTLRPC_SVC_PRIV;
	ghdr->gh_handle.len = gctx->gc_handle.len;
	memcpy(ghdr->gh_handle.data, gctx->gc_handle.data, gctx->gc_handle.len);
	if (req->rq_pack_bulk)
		ghdr->gh_flags |= LUSTRE_GSS_PACK_BULK;
	if (req->rq_pack_udesc)
		ghdr->gh_flags |= LUSTRE_GSS_PACK_USER;

redo:
	ghdr->gh_seq = atomic_inc_return(&gctx->gc_seq);

	/* buffer objects */
	hdrobj.len = PTLRPC_GSS_HEADER_SIZE;
	hdrobj.data = (__u8 *) ghdr;
	msgobj.len = req->rq_clrdata_len;
	msgobj.data = (__u8 *) req->rq_clrbuf;
	token.len = lustre_msg_buflen(req->rq_reqbuf, 1);
	token.data = lustre_msg_buf(req->rq_reqbuf, 1, 0);

	major = lgss_wrap(gctx->gc_mechctx, &hdrobj, &msgobj,
			  req->rq_clrbuf_len, &token);
	if (major != GSS_S_COMPLETE) {
		CERROR("priv: wrap message error: %08x\n", major);
		GOTO(err_free, rc = -EPERM);
	}
	LASSERT(token.len <= buflens[1]);

	/* see explain in gss_cli_ctx_sign() */
	if (unlikely(atomic_read(&gctx->gc_seq) - ghdr->gh_seq >
		     GSS_SEQ_REPACK_THRESHOLD)) {
		int behind = atomic_read(&gctx->gc_seq) - ghdr->gh_seq;

		gss_stat_oos_record_cli(behind);
		CWARN("req %p: %u behind, retry sealing\n", req, behind);

		ghdr->gh_seq = atomic_inc_return(&gctx->gc_seq);
		goto redo;
	}

	/* now set the final wire data length */
	req->rq_reqdata_len = lustre_shrink_msg(req->rq_reqbuf, 1, token.len,0);
	return 0;

err_free:
	if (!req->rq_pool) {
		OBD_FREE_LARGE(req->rq_reqbuf, req->rq_reqbuf_len);
		req->rq_reqbuf = NULL;
		req->rq_reqbuf_len = 0;
	}
	return rc;
}

int gss_cli_ctx_unseal(struct ptlrpc_cli_ctx *ctx,
		       struct ptlrpc_request *req)
{
	struct gss_cli_ctx      *gctx;
	struct gss_header       *ghdr;
	struct lustre_msg       *msg = req->rq_repdata;
	int		      msglen, pack_bulk, swabbed, rc;
	__u32		    major;

	LASSERT(req->rq_cli_ctx == ctx);
	LASSERT(req->rq_ctx_init == 0);
	LASSERT(msg);

	gctx = container_of(ctx, struct gss_cli_ctx, gc_base);
	swabbed = ptlrpc_rep_need_swab(req);

	ghdr = gss_swab_header(msg, 0, swabbed);
	if (ghdr == NULL) {
		CERROR("can't decode gss header\n");
		return -EPROTO;
	}

	/* sanity checks */
	if (ghdr->gh_version != PTLRPC_GSS_VERSION) {
		CERROR("gss version %u mismatch, expect %u\n",
		       ghdr->gh_version, PTLRPC_GSS_VERSION);
		return -EPROTO;
	}

	switch (ghdr->gh_proc) {
	case PTLRPC_GSS_PROC_DATA:
		pack_bulk = ghdr->gh_flags & LUSTRE_GSS_PACK_BULK;

		if (!req->rq_early &&
		    !equi(req->rq_pack_bulk == 1, pack_bulk)) {
			CERROR("%s bulk flag in reply\n",
			       req->rq_pack_bulk ? "missing" : "unexpected");
			return -EPROTO;
		}

		if (swabbed)
			gss_header_swabber(ghdr);

		/* use rq_repdata_len as buffer size, which assume unseal
		 * doesn't need extra memory space. for precise control, we'd
		 * better calculate out actual buffer size as
		 * (repbuf_len - offset - repdata_len) */
		major = gss_unseal_msg(gctx->gc_mechctx, msg,
				       &msglen, req->rq_repdata_len);
		if (major != GSS_S_COMPLETE) {
			CERROR("failed to unwrap reply: %x\n", major);
			rc = -EPERM;
			break;
		}

		swabbed = __lustre_unpack_msg(msg, msglen);
		if (swabbed < 0) {
			CERROR("Failed to unpack after decryption\n");
			return -EPROTO;
		}

		if (msg->lm_bufcount < 1) {
			CERROR("Invalid reply buffer: empty\n");
			return -EPROTO;
		}

		if (pack_bulk) {
			if (msg->lm_bufcount < 2) {
				CERROR("bufcount %u: missing bulk sec desc\n",
				       msg->lm_bufcount);
				return -EPROTO;
			}

			/* bulk checksum is the last segment */
			if (bulk_sec_desc_unpack(msg, msg->lm_bufcount - 1,
						 swabbed))
				return -EPROTO;
		}

		req->rq_repmsg = lustre_msg_buf(msg, 0, 0);
		req->rq_replen = msg->lm_buflens[0];

		rc = 0;
		break;
	case PTLRPC_GSS_PROC_ERR:
		if (req->rq_early) {
			CERROR("server return error with early reply\n");
			rc = -EPROTO;
		} else {
			rc = gss_cli_ctx_handle_err_notify(ctx, req, ghdr);
		}
		break;
	default:
		CERROR("unexpected proc %d\n", ghdr->gh_proc);
		rc = -EPERM;
	}

	return rc;
}

/*********************************************
 * reverse context installation	      *
 *********************************************/

static inline
int gss_install_rvs_svc_ctx(struct obd_import *imp,
			    struct gss_sec *gsec,
			    struct gss_cli_ctx *gctx)
{
	return gss_svc_upcall_install_rvs_ctx(imp, gsec, gctx);
}

/*********************************************
 * GSS security APIs			 *
 *********************************************/
int gss_sec_create_common(struct gss_sec *gsec,
			  struct ptlrpc_sec_policy *policy,
			  struct obd_import *imp,
			  struct ptlrpc_svc_ctx *svcctx,
			  struct sptlrpc_flavor *sf)
{
	struct ptlrpc_sec   *sec;

	LASSERT(imp);
	LASSERT(SPTLRPC_FLVR_POLICY(sf->sf_rpc) == SPTLRPC_POLICY_GSS);

	gsec->gs_mech = lgss_subflavor_to_mech(
				SPTLRPC_FLVR_BASE_SUB(sf->sf_rpc));
	if (!gsec->gs_mech) {
		CERROR("gss backend 0x%x not found\n",
		       SPTLRPC_FLVR_BASE_SUB(sf->sf_rpc));
		return -EOPNOTSUPP;
	}

	spin_lock_init(&gsec->gs_lock);
	gsec->gs_rvs_hdl = 0ULL;

	/* initialize upper ptlrpc_sec */
	sec = &gsec->gs_base;
	sec->ps_policy = policy;
	atomic_set(&sec->ps_refcount, 0);
	atomic_set(&sec->ps_nctx, 0);
	sec->ps_id = sptlrpc_get_next_secid();
	sec->ps_flvr = *sf;
	sec->ps_import = class_import_get(imp);
	spin_lock_init(&sec->ps_lock);
	INIT_LIST_HEAD(&sec->ps_gc_list);

	if (!svcctx) {
		sec->ps_gc_interval = GSS_GC_INTERVAL;
	} else {
		LASSERT(sec_is_reverse(sec));

		/* never do gc on reverse sec */
		sec->ps_gc_interval = 0;
	}

	if (SPTLRPC_FLVR_BULK_SVC(sec->ps_flvr.sf_rpc) == SPTLRPC_BULK_SVC_PRIV)
		sptlrpc_enc_pool_add_user();

	CDEBUG(D_SEC, "create %s%s@%p\n", (svcctx ? "reverse " : ""),
	       policy->sp_name, gsec);
	return 0;
}

void gss_sec_destroy_common(struct gss_sec *gsec)
{
	struct ptlrpc_sec      *sec = &gsec->gs_base;

	LASSERT(sec->ps_import);
	LASSERT(atomic_read(&sec->ps_refcount) == 0);
	LASSERT(atomic_read(&sec->ps_nctx) == 0);

	if (gsec->gs_mech) {
		lgss_mech_put(gsec->gs_mech);
		gsec->gs_mech = NULL;
	}

	class_import_put(sec->ps_import);

	if (SPTLRPC_FLVR_BULK_SVC(sec->ps_flvr.sf_rpc) == SPTLRPC_BULK_SVC_PRIV)
		sptlrpc_enc_pool_del_user();
}

void gss_sec_kill(struct ptlrpc_sec *sec)
{
	sec->ps_dying = 1;
}

int gss_cli_ctx_init_common(struct ptlrpc_sec *sec,
			    struct ptlrpc_cli_ctx *ctx,
			    struct ptlrpc_ctx_ops *ctxops,
			    struct vfs_cred *vcred)
{
	struct gss_cli_ctx    *gctx = ctx2gctx(ctx);

	gctx->gc_win = 0;
	atomic_set(&gctx->gc_seq, 0);

	INIT_HLIST_NODE(&ctx->cc_cache);
	atomic_set(&ctx->cc_refcount, 0);
	ctx->cc_sec = sec;
	ctx->cc_ops = ctxops;
	ctx->cc_expire = 0;
	ctx->cc_flags = PTLRPC_CTX_NEW;
	ctx->cc_vcred = *vcred;
	spin_lock_init(&ctx->cc_lock);
	INIT_LIST_HEAD(&ctx->cc_req_list);
	INIT_LIST_HEAD(&ctx->cc_gc_chain);

	/* take a ref on belonging sec, balanced in ctx destroying */
	atomic_inc(&sec->ps_refcount);
	/* statistic only */
	atomic_inc(&sec->ps_nctx);

	CDEBUG(D_SEC, "%s@%p: create ctx %p(%u->%s)\n",
	       sec->ps_policy->sp_name, ctx->cc_sec,
	       ctx, ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec));
	return 0;
}

/*
 * return value:
 *   1: the context has been taken care of by someone else
 *   0: proceed to really destroy the context locally
 */
int gss_cli_ctx_fini_common(struct ptlrpc_sec *sec,
			    struct ptlrpc_cli_ctx *ctx)
{
	struct gss_cli_ctx *gctx = ctx2gctx(ctx);

	LASSERT(atomic_read(&sec->ps_nctx) > 0);
	LASSERT(atomic_read(&ctx->cc_refcount) == 0);
	LASSERT(ctx->cc_sec == sec);

	/*
	 * remove UPTODATE flag of reverse ctx thus we won't send fini rpc,
	 * this is to avoid potential problems of client side reverse svc ctx
	 * be mis-destroyed in various recovery senarios. anyway client can
	 * manage its reverse ctx well by associating it with its buddy ctx.
	 */
	if (sec_is_reverse(sec))
		ctx->cc_flags &= ~PTLRPC_CTX_UPTODATE;

	if (gctx->gc_mechctx) {
		/* the final context fini rpc will use this ctx too, and it's
		 * asynchronous which finished by request_out_callback(). so
		 * we add refcount, whoever drop finally drop the refcount to
		 * 0 should responsible for the rest of destroy. */
		atomic_inc(&ctx->cc_refcount);

		gss_do_ctx_fini_rpc(gctx);
		gss_cli_ctx_finalize(gctx);

		if (!atomic_dec_and_test(&ctx->cc_refcount))
			return 1;
	}

	if (sec_is_reverse(sec))
		CWARN("reverse sec %p: destroy ctx %p\n",
		      ctx->cc_sec, ctx);
	else
		CWARN("%s@%p: destroy ctx %p(%u->%s)\n",
		      sec->ps_policy->sp_name, ctx->cc_sec,
		      ctx, ctx->cc_vcred.vc_uid, sec2target_str(ctx->cc_sec));

	return 0;
}

static
int gss_alloc_reqbuf_intg(struct ptlrpc_sec *sec,
			  struct ptlrpc_request *req,
			  int svc, int msgsize)
{
	int		       bufsize, txtsize;
	int		       bufcnt = 2;
	__u32		     buflens[5];

	/*
	 * on-wire data layout:
	 * - gss header
	 * - lustre message
	 * - user descriptor (optional)
	 * - bulk sec descriptor (optional)
	 * - signature (optional)
	 *   - svc == NULL: NULL
	 *   - svc == AUTH: signature of gss header
	 *   - svc == INTG: signature of all above
	 *
	 * if this is context negotiation, reserver fixed space
	 * at the last (signature) segment regardless of svc mode.
	 */

	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	txtsize = buflens[0];

	buflens[1] = msgsize;
	if (svc == SPTLRPC_SVC_INTG)
		txtsize += buflens[1];

	if (req->rq_pack_udesc) {
		buflens[bufcnt] = sptlrpc_current_user_desc_size();
		if (svc == SPTLRPC_SVC_INTG)
			txtsize += buflens[bufcnt];
		bufcnt++;
	}

	if (req->rq_pack_bulk) {
		buflens[bufcnt] = gss_cli_bulk_payload(req->rq_cli_ctx,
						       &req->rq_flvr,
						       0, req->rq_bulk_read);
		if (svc == SPTLRPC_SVC_INTG)
			txtsize += buflens[bufcnt];
		bufcnt++;
	}

	if (req->rq_ctx_init)
		buflens[bufcnt++] = GSS_CTX_INIT_MAX_LEN;
	else if (svc != SPTLRPC_SVC_NULL)
		buflens[bufcnt++] = gss_cli_payload(req->rq_cli_ctx, txtsize,0);

	bufsize = lustre_msg_size_v2(bufcnt, buflens);

	if (!req->rq_reqbuf) {
		bufsize = size_roundup_power2(bufsize);

		OBD_ALLOC_LARGE(req->rq_reqbuf, bufsize);
		if (!req->rq_reqbuf)
			return -ENOMEM;

		req->rq_reqbuf_len = bufsize;
	} else {
		LASSERT(req->rq_pool);
		LASSERT(req->rq_reqbuf_len >= bufsize);
		memset(req->rq_reqbuf, 0, bufsize);
	}

	lustre_init_msg_v2(req->rq_reqbuf, bufcnt, buflens, NULL);
	req->rq_reqbuf->lm_secflvr = req->rq_flvr.sf_rpc;

	req->rq_reqmsg = lustre_msg_buf(req->rq_reqbuf, 1, msgsize);
	LASSERT(req->rq_reqmsg);

	/* pack user desc here, later we might leave current user's process */
	if (req->rq_pack_udesc)
		sptlrpc_pack_user_desc(req->rq_reqbuf, 2);

	return 0;
}

static
int gss_alloc_reqbuf_priv(struct ptlrpc_sec *sec,
			  struct ptlrpc_request *req,
			  int msgsize)
{
	__u32		     ibuflens[3], wbuflens[2];
	int		       ibufcnt;
	int		       clearsize, wiresize;

	LASSERT(req->rq_clrbuf == NULL);
	LASSERT(req->rq_clrbuf_len == 0);

	/* Inner (clear) buffers
	 *  - lustre message
	 *  - user descriptor (optional)
	 *  - bulk checksum (optional)
	 */
	ibufcnt = 1;
	ibuflens[0] = msgsize;

	if (req->rq_pack_udesc)
		ibuflens[ibufcnt++] = sptlrpc_current_user_desc_size();
	if (req->rq_pack_bulk)
		ibuflens[ibufcnt++] = gss_cli_bulk_payload(req->rq_cli_ctx,
							   &req->rq_flvr, 0,
							   req->rq_bulk_read);

	clearsize = lustre_msg_size_v2(ibufcnt, ibuflens);
	/* to allow append padding during encryption */
	clearsize += GSS_MAX_CIPHER_BLOCK;

	/* Wrapper (wire) buffers
	 *  - gss header
	 *  - cipher text
	 */
	wbuflens[0] = PTLRPC_GSS_HEADER_SIZE;
	wbuflens[1] = gss_cli_payload(req->rq_cli_ctx, clearsize, 1);
	wiresize = lustre_msg_size_v2(2, wbuflens);

	if (req->rq_pool) {
		/* rq_reqbuf is preallocated */
		LASSERT(req->rq_reqbuf);
		LASSERT(req->rq_reqbuf_len >= wiresize);

		memset(req->rq_reqbuf, 0, req->rq_reqbuf_len);

		/* if the pre-allocated buffer is big enough, we just pack
		 * both clear buf & request buf in it, to avoid more alloc. */
		if (clearsize + wiresize <= req->rq_reqbuf_len) {
			req->rq_clrbuf =
				(void *) (((char *) req->rq_reqbuf) + wiresize);
		} else {
			CWARN("pre-allocated buf size %d is not enough for "
			      "both clear (%d) and cipher (%d) text, proceed "
			      "with extra allocation\n", req->rq_reqbuf_len,
			      clearsize, wiresize);
		}
	}

	if (!req->rq_clrbuf) {
		clearsize = size_roundup_power2(clearsize);

		OBD_ALLOC_LARGE(req->rq_clrbuf, clearsize);
		if (!req->rq_clrbuf)
			return -ENOMEM;
	}
	req->rq_clrbuf_len = clearsize;

	lustre_init_msg_v2(req->rq_clrbuf, ibufcnt, ibuflens, NULL);
	req->rq_reqmsg = lustre_msg_buf(req->rq_clrbuf, 0, msgsize);

	if (req->rq_pack_udesc)
		sptlrpc_pack_user_desc(req->rq_clrbuf, 1);

	return 0;
}

/*
 * NOTE: any change of request buffer allocation should also consider
 * changing enlarge_reqbuf() series functions.
 */
int gss_alloc_reqbuf(struct ptlrpc_sec *sec,
		     struct ptlrpc_request *req,
		     int msgsize)
{
	int     svc = SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc);

	LASSERT(!req->rq_pack_bulk ||
		(req->rq_bulk_read || req->rq_bulk_write));

	switch (svc) {
	case SPTLRPC_SVC_NULL:
	case SPTLRPC_SVC_AUTH:
	case SPTLRPC_SVC_INTG:
		return gss_alloc_reqbuf_intg(sec, req, svc, msgsize);
	case SPTLRPC_SVC_PRIV:
		return gss_alloc_reqbuf_priv(sec, req, msgsize);
	default:
		LASSERTF(0, "bad rpc flavor %x\n", req->rq_flvr.sf_rpc);
		return 0;
	}
}

void gss_free_reqbuf(struct ptlrpc_sec *sec,
		     struct ptlrpc_request *req)
{
	int     privacy;

	LASSERT(!req->rq_pool || req->rq_reqbuf);
	privacy = SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc) == SPTLRPC_SVC_PRIV;

	if (!req->rq_clrbuf)
		goto release_reqbuf;

	/* release clear buffer */
	LASSERT(privacy);
	LASSERT(req->rq_clrbuf_len);

	if (req->rq_pool == NULL ||
	    req->rq_clrbuf < req->rq_reqbuf ||
	    (char *) req->rq_clrbuf >=
	    (char *) req->rq_reqbuf + req->rq_reqbuf_len)
		OBD_FREE_LARGE(req->rq_clrbuf, req->rq_clrbuf_len);

	req->rq_clrbuf = NULL;
	req->rq_clrbuf_len = 0;

release_reqbuf:
	if (!req->rq_pool && req->rq_reqbuf) {
		LASSERT(req->rq_reqbuf_len);

		OBD_FREE_LARGE(req->rq_reqbuf, req->rq_reqbuf_len);
		req->rq_reqbuf = NULL;
		req->rq_reqbuf_len = 0;
	}
}

static int do_alloc_repbuf(struct ptlrpc_request *req, int bufsize)
{
	bufsize = size_roundup_power2(bufsize);

	OBD_ALLOC_LARGE(req->rq_repbuf, bufsize);
	if (!req->rq_repbuf)
		return -ENOMEM;

	req->rq_repbuf_len = bufsize;
	return 0;
}

static
int gss_alloc_repbuf_intg(struct ptlrpc_sec *sec,
			  struct ptlrpc_request *req,
			  int svc, int msgsize)
{
	int	     txtsize;
	__u32	   buflens[4];
	int	     bufcnt = 2;
	int	     alloc_size;

	/*
	 * on-wire data layout:
	 * - gss header
	 * - lustre message
	 * - bulk sec descriptor (optional)
	 * - signature (optional)
	 *   - svc == NULL: NULL
	 *   - svc == AUTH: signature of gss header
	 *   - svc == INTG: signature of all above
	 *
	 * if this is context negotiation, reserver fixed space
	 * at the last (signature) segment regardless of svc mode.
	 */

	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	txtsize = buflens[0];

	buflens[1] = msgsize;
	if (svc == SPTLRPC_SVC_INTG)
		txtsize += buflens[1];

	if (req->rq_pack_bulk) {
		buflens[bufcnt] = gss_cli_bulk_payload(req->rq_cli_ctx,
						       &req->rq_flvr,
						       1, req->rq_bulk_read);
		if (svc == SPTLRPC_SVC_INTG)
			txtsize += buflens[bufcnt];
		bufcnt++;
	}

	if (req->rq_ctx_init)
		buflens[bufcnt++] = GSS_CTX_INIT_MAX_LEN;
	else if (svc != SPTLRPC_SVC_NULL)
		buflens[bufcnt++] = gss_cli_payload(req->rq_cli_ctx, txtsize,0);

	alloc_size = lustre_msg_size_v2(bufcnt, buflens);

	/* add space for early reply */
	alloc_size += gss_at_reply_off_integ;

	return do_alloc_repbuf(req, alloc_size);
}

static
int gss_alloc_repbuf_priv(struct ptlrpc_sec *sec,
			  struct ptlrpc_request *req,
			  int msgsize)
{
	int	     txtsize;
	__u32	   buflens[2];
	int	     bufcnt;
	int	     alloc_size;

	/* inner buffers */
	bufcnt = 1;
	buflens[0] = msgsize;

	if (req->rq_pack_bulk)
		buflens[bufcnt++] = gss_cli_bulk_payload(req->rq_cli_ctx,
							 &req->rq_flvr,
							 1, req->rq_bulk_read);
	txtsize = lustre_msg_size_v2(bufcnt, buflens);
	txtsize += GSS_MAX_CIPHER_BLOCK;

	/* wrapper buffers */
	bufcnt = 2;
	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	buflens[1] = gss_cli_payload(req->rq_cli_ctx, txtsize, 1);

	alloc_size = lustre_msg_size_v2(bufcnt, buflens);
	/* add space for early reply */
	alloc_size += gss_at_reply_off_priv;

	return do_alloc_repbuf(req, alloc_size);
}

int gss_alloc_repbuf(struct ptlrpc_sec *sec,
		     struct ptlrpc_request *req,
		     int msgsize)
{
	int     svc = SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc);

	LASSERT(!req->rq_pack_bulk ||
		(req->rq_bulk_read || req->rq_bulk_write));

	switch (svc) {
	case SPTLRPC_SVC_NULL:
	case SPTLRPC_SVC_AUTH:
	case SPTLRPC_SVC_INTG:
		return gss_alloc_repbuf_intg(sec, req, svc, msgsize);
	case SPTLRPC_SVC_PRIV:
		return gss_alloc_repbuf_priv(sec, req, msgsize);
	default:
		LASSERTF(0, "bad rpc flavor %x\n", req->rq_flvr.sf_rpc);
		return 0;
	}
}

void gss_free_repbuf(struct ptlrpc_sec *sec,
		     struct ptlrpc_request *req)
{
	OBD_FREE_LARGE(req->rq_repbuf, req->rq_repbuf_len);
	req->rq_repbuf = NULL;
	req->rq_repbuf_len = 0;
	req->rq_repdata = NULL;
	req->rq_repdata_len = 0;
}

static int get_enlarged_msgsize(struct lustre_msg *msg,
				int segment, int newsize)
{
	int save, newmsg_size;

	LASSERT(newsize >= msg->lm_buflens[segment]);

	save = msg->lm_buflens[segment];
	msg->lm_buflens[segment] = newsize;
	newmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
	msg->lm_buflens[segment] = save;

	return newmsg_size;
}

static int get_enlarged_msgsize2(struct lustre_msg *msg,
				 int segment1, int newsize1,
				 int segment2, int newsize2)
{
	int save1, save2, newmsg_size;

	LASSERT(newsize1 >= msg->lm_buflens[segment1]);
	LASSERT(newsize2 >= msg->lm_buflens[segment2]);

	save1 = msg->lm_buflens[segment1];
	save2 = msg->lm_buflens[segment2];
	msg->lm_buflens[segment1] = newsize1;
	msg->lm_buflens[segment2] = newsize2;
	newmsg_size = lustre_msg_size_v2(msg->lm_bufcount, msg->lm_buflens);
	msg->lm_buflens[segment1] = save1;
	msg->lm_buflens[segment2] = save2;

	return newmsg_size;
}

static
int gss_enlarge_reqbuf_intg(struct ptlrpc_sec *sec,
			    struct ptlrpc_request *req,
			    int svc,
			    int segment, int newsize)
{
	struct lustre_msg      *newbuf;
	int		     txtsize, sigsize = 0, i;
	int		     newmsg_size, newbuf_size;

	/*
	 * gss header is at seg 0;
	 * embedded msg is at seg 1;
	 * signature (if any) is at the last seg
	 */
	LASSERT(req->rq_reqbuf);
	LASSERT(req->rq_reqbuf_len > req->rq_reqlen);
	LASSERT(req->rq_reqbuf->lm_bufcount >= 2);
	LASSERT(lustre_msg_buf(req->rq_reqbuf, 1, 0) == req->rq_reqmsg);

	/* 1. compute new embedded msg size */
	newmsg_size = get_enlarged_msgsize(req->rq_reqmsg, segment, newsize);
	LASSERT(newmsg_size >= req->rq_reqbuf->lm_buflens[1]);

	/* 2. compute new wrapper msg size */
	if (svc == SPTLRPC_SVC_NULL) {
		/* no signature, get size directly */
		newbuf_size = get_enlarged_msgsize(req->rq_reqbuf,
						   1, newmsg_size);
	} else {
		txtsize = req->rq_reqbuf->lm_buflens[0];

		if (svc == SPTLRPC_SVC_INTG) {
			for (i = 1; i < req->rq_reqbuf->lm_bufcount; i++)
				txtsize += req->rq_reqbuf->lm_buflens[i];
			txtsize += newmsg_size - req->rq_reqbuf->lm_buflens[1];
		}

		sigsize = gss_cli_payload(req->rq_cli_ctx, txtsize, 0);
		LASSERT(sigsize >= msg_last_seglen(req->rq_reqbuf));

		newbuf_size = get_enlarged_msgsize2(
					req->rq_reqbuf,
					1, newmsg_size,
					msg_last_segidx(req->rq_reqbuf),
					sigsize);
	}

	/* request from pool should always have enough buffer */
	LASSERT(!req->rq_pool || req->rq_reqbuf_len >= newbuf_size);

	if (req->rq_reqbuf_len < newbuf_size) {
		newbuf_size = size_roundup_power2(newbuf_size);

		OBD_ALLOC_LARGE(newbuf, newbuf_size);
		if (newbuf == NULL)
			return -ENOMEM;

		memcpy(newbuf, req->rq_reqbuf, req->rq_reqbuf_len);

		OBD_FREE_LARGE(req->rq_reqbuf, req->rq_reqbuf_len);
		req->rq_reqbuf = newbuf;
		req->rq_reqbuf_len = newbuf_size;
		req->rq_reqmsg = lustre_msg_buf(req->rq_reqbuf, 1, 0);
	}

	/* do enlargement, from wrapper to embedded, from end to begin */
	if (svc != SPTLRPC_SVC_NULL)
		_sptlrpc_enlarge_msg_inplace(req->rq_reqbuf,
					     msg_last_segidx(req->rq_reqbuf),
					     sigsize);

	_sptlrpc_enlarge_msg_inplace(req->rq_reqbuf, 1, newmsg_size);
	_sptlrpc_enlarge_msg_inplace(req->rq_reqmsg, segment, newsize);

	req->rq_reqlen = newmsg_size;
	return 0;
}

static
int gss_enlarge_reqbuf_priv(struct ptlrpc_sec *sec,
			    struct ptlrpc_request *req,
			    int segment, int newsize)
{
	struct lustre_msg      *newclrbuf;
	int		     newmsg_size, newclrbuf_size, newcipbuf_size;
	__u32		   buflens[3];

	/*
	 * embedded msg is at seg 0 of clear buffer;
	 * cipher text is at seg 2 of cipher buffer;
	 */
	LASSERT(req->rq_pool ||
		(req->rq_reqbuf == NULL && req->rq_reqbuf_len == 0));
	LASSERT(req->rq_reqbuf == NULL ||
		(req->rq_pool && req->rq_reqbuf->lm_bufcount == 3));
	LASSERT(req->rq_clrbuf);
	LASSERT(req->rq_clrbuf_len > req->rq_reqlen);
	LASSERT(lustre_msg_buf(req->rq_clrbuf, 0, 0) == req->rq_reqmsg);

	/* compute new embedded msg size */
	newmsg_size = get_enlarged_msgsize(req->rq_reqmsg, segment, newsize);

	/* compute new clear buffer size */
	newclrbuf_size = get_enlarged_msgsize(req->rq_clrbuf, 0, newmsg_size);
	newclrbuf_size += GSS_MAX_CIPHER_BLOCK;

	/* compute new cipher buffer size */
	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	buflens[1] = gss_cli_payload(req->rq_cli_ctx, buflens[0], 0);
	buflens[2] = gss_cli_payload(req->rq_cli_ctx, newclrbuf_size, 1);
	newcipbuf_size = lustre_msg_size_v2(3, buflens);

	/* handle the case that we put both clear buf and cipher buf into
	 * pre-allocated single buffer. */
	if (unlikely(req->rq_pool) &&
	    req->rq_clrbuf >= req->rq_reqbuf &&
	    (char *) req->rq_clrbuf <
	    (char *) req->rq_reqbuf + req->rq_reqbuf_len) {
		/* it couldn't be better we still fit into the
		 * pre-allocated buffer. */
		if (newclrbuf_size + newcipbuf_size <= req->rq_reqbuf_len) {
			void *src, *dst;

			/* move clear text backward. */
			src = req->rq_clrbuf;
			dst = (char *) req->rq_reqbuf + newcipbuf_size;

			memmove(dst, src, req->rq_clrbuf_len);

			req->rq_clrbuf = (struct lustre_msg *) dst;
			req->rq_clrbuf_len = newclrbuf_size;
			req->rq_reqmsg = lustre_msg_buf(req->rq_clrbuf, 0, 0);
		} else {
			/* sadly we have to split out the clear buffer */
			LASSERT(req->rq_reqbuf_len >= newcipbuf_size);
			LASSERT(req->rq_clrbuf_len < newclrbuf_size);
		}
	}

	if (req->rq_clrbuf_len < newclrbuf_size) {
		newclrbuf_size = size_roundup_power2(newclrbuf_size);

		OBD_ALLOC_LARGE(newclrbuf, newclrbuf_size);
		if (newclrbuf == NULL)
			return -ENOMEM;

		memcpy(newclrbuf, req->rq_clrbuf, req->rq_clrbuf_len);

		if (req->rq_reqbuf == NULL ||
		    req->rq_clrbuf < req->rq_reqbuf ||
		    (char *) req->rq_clrbuf >=
		    (char *) req->rq_reqbuf + req->rq_reqbuf_len) {
			OBD_FREE_LARGE(req->rq_clrbuf, req->rq_clrbuf_len);
		}

		req->rq_clrbuf = newclrbuf;
		req->rq_clrbuf_len = newclrbuf_size;
		req->rq_reqmsg = lustre_msg_buf(req->rq_clrbuf, 0, 0);
	}

	_sptlrpc_enlarge_msg_inplace(req->rq_clrbuf, 0, newmsg_size);
	_sptlrpc_enlarge_msg_inplace(req->rq_reqmsg, segment, newsize);
	req->rq_reqlen = newmsg_size;

	return 0;
}

int gss_enlarge_reqbuf(struct ptlrpc_sec *sec,
		       struct ptlrpc_request *req,
		       int segment, int newsize)
{
	int     svc = SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc);

	LASSERT(!req->rq_ctx_init && !req->rq_ctx_fini);

	switch (svc) {
	case SPTLRPC_SVC_NULL:
	case SPTLRPC_SVC_AUTH:
	case SPTLRPC_SVC_INTG:
		return gss_enlarge_reqbuf_intg(sec, req, svc, segment, newsize);
	case SPTLRPC_SVC_PRIV:
		return gss_enlarge_reqbuf_priv(sec, req, segment, newsize);
	default:
		LASSERTF(0, "bad rpc flavor %x\n", req->rq_flvr.sf_rpc);
		return 0;
	}
}

int gss_sec_install_rctx(struct obd_import *imp,
			 struct ptlrpc_sec *sec,
			 struct ptlrpc_cli_ctx *ctx)
{
	struct gss_sec     *gsec;
	struct gss_cli_ctx *gctx;
	int		 rc;

	gsec = container_of(sec, struct gss_sec, gs_base);
	gctx = container_of(ctx, struct gss_cli_ctx, gc_base);

	rc = gss_install_rvs_svc_ctx(imp, gsec, gctx);
	return rc;
}

/********************************************
 * server side API			  *
 ********************************************/

static inline
int gss_svc_reqctx_is_special(struct gss_svc_reqctx *grctx)
{
	LASSERT(grctx);
	return (grctx->src_init || grctx->src_init_continue ||
		grctx->src_err_notify);
}

static
void gss_svc_reqctx_free(struct gss_svc_reqctx *grctx)
{
	if (grctx->src_ctx)
		gss_svc_upcall_put_ctx(grctx->src_ctx);

	sptlrpc_policy_put(grctx->src_base.sc_policy);
	OBD_FREE_PTR(grctx);
}

static inline
void gss_svc_reqctx_addref(struct gss_svc_reqctx *grctx)
{
	LASSERT(atomic_read(&grctx->src_base.sc_refcount) > 0);
	atomic_inc(&grctx->src_base.sc_refcount);
}

static inline
void gss_svc_reqctx_decref(struct gss_svc_reqctx *grctx)
{
	LASSERT(atomic_read(&grctx->src_base.sc_refcount) > 0);

	if (atomic_dec_and_test(&grctx->src_base.sc_refcount))
		gss_svc_reqctx_free(grctx);
}

static
int gss_svc_sign(struct ptlrpc_request *req,
		 struct ptlrpc_reply_state *rs,
		 struct gss_svc_reqctx *grctx,
		 __u32 svc)
{
	__u32   flags = 0;
	int     rc;

	LASSERT(rs->rs_msg == lustre_msg_buf(rs->rs_repbuf, 1, 0));

	/* embedded lustre_msg might have been shrinked */
	if (req->rq_replen != rs->rs_repbuf->lm_buflens[1])
		lustre_shrink_msg(rs->rs_repbuf, 1, req->rq_replen, 1);

	if (req->rq_pack_bulk)
		flags |= LUSTRE_GSS_PACK_BULK;

	rc = gss_sign_msg(rs->rs_repbuf, grctx->src_ctx->gsc_mechctx,
			  LUSTRE_SP_ANY, flags, PTLRPC_GSS_PROC_DATA,
			  grctx->src_wirectx.gw_seq, svc, NULL);
	if (rc < 0)
		return rc;

	rs->rs_repdata_len = rc;

	if (likely(req->rq_packed_final)) {
		if (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT)
			req->rq_reply_off = gss_at_reply_off_integ;
		else
			req->rq_reply_off = 0;
	} else {
		if (svc == SPTLRPC_SVC_NULL)
			rs->rs_repbuf->lm_cksum = crc32_le(!(__u32) 0,
					lustre_msg_buf(rs->rs_repbuf, 1, 0),
					lustre_msg_buflen(rs->rs_repbuf, 1));
		req->rq_reply_off = 0;
	}

	return 0;
}

int gss_pack_err_notify(struct ptlrpc_request *req, __u32 major, __u32 minor)
{
	struct gss_svc_reqctx     *grctx = gss_svc_ctx2reqctx(req->rq_svc_ctx);
	struct ptlrpc_reply_state *rs;
	struct gss_err_header     *ghdr;
	int			replen = sizeof(struct ptlrpc_body);
	int			rc;

	//if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_SVCGSS_ERR_NOTIFY, OBD_FAIL_ONCE))
	//      return -EINVAL;

	grctx->src_err_notify = 1;
	grctx->src_reserve_len = 0;

	rc = lustre_pack_reply_v2(req, 1, &replen, NULL, 0);
	if (rc) {
		CERROR("could not pack reply, err %d\n", rc);
		return rc;
	}

	/* gss hdr */
	rs = req->rq_reply_state;
	LASSERT(rs->rs_repbuf->lm_buflens[1] >= sizeof(*ghdr));
	ghdr = lustre_msg_buf(rs->rs_repbuf, 0, 0);
	ghdr->gh_version = PTLRPC_GSS_VERSION;
	ghdr->gh_flags = 0;
	ghdr->gh_proc = PTLRPC_GSS_PROC_ERR;
	ghdr->gh_major = major;
	ghdr->gh_minor = minor;
	ghdr->gh_handle.len = 0; /* fake context handle */

	rs->rs_repdata_len = lustre_msg_size_v2(rs->rs_repbuf->lm_bufcount,
						rs->rs_repbuf->lm_buflens);

	CDEBUG(D_SEC, "prepare gss error notify(0x%x/0x%x) to %s\n",
	       major, minor, libcfs_nid2str(req->rq_peer.nid));
	return 0;
}

static
int gss_svc_handle_init(struct ptlrpc_request *req,
			struct gss_wire_ctx *gw)
{
	struct gss_svc_reqctx     *grctx = gss_svc_ctx2reqctx(req->rq_svc_ctx);
	struct lustre_msg	 *reqbuf = req->rq_reqbuf;
	struct obd_uuid	   *uuid;
	struct obd_device	 *target;
	rawobj_t		   uuid_obj, rvs_hdl, in_token;
	__u32		      lustre_svc;
	__u32		     *secdata, seclen;
	int			swabbed, rc;

	CDEBUG(D_SEC, "processing gss init(%d) request from %s\n", gw->gw_proc,
	       libcfs_nid2str(req->rq_peer.nid));

	req->rq_ctx_init = 1;

	if (gw->gw_flags & LUSTRE_GSS_PACK_BULK) {
		CERROR("unexpected bulk flag\n");
		return SECSVC_DROP;
	}

	if (gw->gw_proc == PTLRPC_GSS_PROC_INIT && gw->gw_handle.len != 0) {
		CERROR("proc %u: invalid handle length %u\n",
		       gw->gw_proc, gw->gw_handle.len);
		return SECSVC_DROP;
	}

	if (reqbuf->lm_bufcount < 3 || reqbuf->lm_bufcount > 4) {
		CERROR("Invalid bufcount %d\n", reqbuf->lm_bufcount);
		return SECSVC_DROP;
	}

	swabbed = ptlrpc_req_need_swab(req);

	/* ctx initiate payload is in last segment */
	secdata = lustre_msg_buf(reqbuf, reqbuf->lm_bufcount - 1, 0);
	seclen = reqbuf->lm_buflens[reqbuf->lm_bufcount - 1];

	if (seclen < 4 + 4) {
		CERROR("sec size %d too small\n", seclen);
		return SECSVC_DROP;
	}

	/* lustre svc type */
	lustre_svc = le32_to_cpu(*secdata++);
	seclen -= 4;

	/* extract target uuid, note this code is somewhat fragile
	 * because touched internal structure of obd_uuid */
	if (rawobj_extract(&uuid_obj, &secdata, &seclen)) {
		CERROR("failed to extract target uuid\n");
		return SECSVC_DROP;
	}
	uuid_obj.data[uuid_obj.len - 1] = '\0';

	uuid = (struct obd_uuid *) uuid_obj.data;
	target = class_uuid2obd(uuid);
	if (!target || target->obd_stopping || !target->obd_set_up) {
		CERROR("target '%s' is not available for context init (%s)\n",
		       uuid->uuid, target == NULL ? "no target" :
		       (target->obd_stopping ? "stopping" : "not set up"));
		return SECSVC_DROP;
	}

	/* extract reverse handle */
	if (rawobj_extract(&rvs_hdl, &secdata, &seclen)) {
		CERROR("failed extract reverse handle\n");
		return SECSVC_DROP;
	}

	/* extract token */
	if (rawobj_extract(&in_token, &secdata, &seclen)) {
		CERROR("can't extract token\n");
		return SECSVC_DROP;
	}

	rc = gss_svc_upcall_handle_init(req, grctx, gw, target, lustre_svc,
					&rvs_hdl, &in_token);
	if (rc != SECSVC_OK)
		return rc;

	if (grctx->src_ctx->gsc_usr_mds || grctx->src_ctx->gsc_usr_oss ||
	    grctx->src_ctx->gsc_usr_root)
		CWARN("create svc ctx %p: user from %s authenticated as %s\n",
		      grctx->src_ctx, libcfs_nid2str(req->rq_peer.nid),
		      grctx->src_ctx->gsc_usr_mds ? "mds" :
			(grctx->src_ctx->gsc_usr_oss ? "oss" : "root"));
	else
		CWARN("create svc ctx %p: accept user %u from %s\n",
		      grctx->src_ctx, grctx->src_ctx->gsc_uid,
		      libcfs_nid2str(req->rq_peer.nid));

	if (gw->gw_flags & LUSTRE_GSS_PACK_USER) {
		if (reqbuf->lm_bufcount < 4) {
			CERROR("missing user descriptor\n");
			return SECSVC_DROP;
		}
		if (sptlrpc_unpack_user_desc(reqbuf, 2, swabbed)) {
			CERROR("Mal-formed user descriptor\n");
			return SECSVC_DROP;
		}

		req->rq_pack_udesc = 1;
		req->rq_user_desc = lustre_msg_buf(reqbuf, 2, 0);
	}

	req->rq_reqmsg = lustre_msg_buf(reqbuf, 1, 0);
	req->rq_reqlen = lustre_msg_buflen(reqbuf, 1);

	return rc;
}

/*
 * last segment must be the gss signature.
 */
static
int gss_svc_verify_request(struct ptlrpc_request *req,
			   struct gss_svc_reqctx *grctx,
			   struct gss_wire_ctx *gw,
			   __u32 *major)
{
	struct gss_svc_ctx *gctx = grctx->src_ctx;
	struct lustre_msg  *msg = req->rq_reqbuf;
	int		 offset = 2;
	int		 swabbed;

	*major = GSS_S_COMPLETE;

	if (msg->lm_bufcount < 2) {
		CERROR("Too few segments (%u) in request\n", msg->lm_bufcount);
		return -EINVAL;
	}

	if (gw->gw_svc == SPTLRPC_SVC_NULL)
		goto verified;

	if (gss_check_seq_num(&gctx->gsc_seqdata, gw->gw_seq, 0)) {
		CERROR("phase 0: discard replayed req: seq %u\n", gw->gw_seq);
		*major = GSS_S_DUPLICATE_TOKEN;
		return -EACCES;
	}

	*major = gss_verify_msg(msg, gctx->gsc_mechctx, gw->gw_svc);
	if (*major != GSS_S_COMPLETE) {
		CERROR("failed to verify request: %x\n", *major);
		return -EACCES;
	}

	if (gctx->gsc_reverse == 0 &&
	    gss_check_seq_num(&gctx->gsc_seqdata, gw->gw_seq, 1)) {
		CERROR("phase 1+: discard replayed req: seq %u\n", gw->gw_seq);
		*major = GSS_S_DUPLICATE_TOKEN;
		return -EACCES;
	}

verified:
	swabbed = ptlrpc_req_need_swab(req);

	/* user descriptor */
	if (gw->gw_flags & LUSTRE_GSS_PACK_USER) {
		if (msg->lm_bufcount < (offset + 1)) {
			CERROR("no user desc included\n");
			return -EINVAL;
		}

		if (sptlrpc_unpack_user_desc(msg, offset, swabbed)) {
			CERROR("Mal-formed user descriptor\n");
			return -EINVAL;
		}

		req->rq_pack_udesc = 1;
		req->rq_user_desc = lustre_msg_buf(msg, offset, 0);
		offset++;
	}

	/* check bulk_sec_desc data */
	if (gw->gw_flags & LUSTRE_GSS_PACK_BULK) {
		if (msg->lm_bufcount < (offset + 1)) {
			CERROR("missing bulk sec descriptor\n");
			return -EINVAL;
		}

		if (bulk_sec_desc_unpack(msg, offset, swabbed))
			return -EINVAL;

		req->rq_pack_bulk = 1;
		grctx->src_reqbsd = lustre_msg_buf(msg, offset, 0);
		grctx->src_reqbsd_size = lustre_msg_buflen(msg, offset);
	}

	req->rq_reqmsg = lustre_msg_buf(msg, 1, 0);
	req->rq_reqlen = msg->lm_buflens[1];
	return 0;
}

static
int gss_svc_unseal_request(struct ptlrpc_request *req,
			   struct gss_svc_reqctx *grctx,
			   struct gss_wire_ctx *gw,
			   __u32 *major)
{
	struct gss_svc_ctx *gctx = grctx->src_ctx;
	struct lustre_msg  *msg = req->rq_reqbuf;
	int		 swabbed, msglen, offset = 1;

	if (gss_check_seq_num(&gctx->gsc_seqdata, gw->gw_seq, 0)) {
		CERROR("phase 0: discard replayed req: seq %u\n", gw->gw_seq);
		*major = GSS_S_DUPLICATE_TOKEN;
		return -EACCES;
	}

	*major = gss_unseal_msg(gctx->gsc_mechctx, msg,
			       &msglen, req->rq_reqdata_len);
	if (*major != GSS_S_COMPLETE) {
		CERROR("failed to unwrap request: %x\n", *major);
		return -EACCES;
	}

	if (gss_check_seq_num(&gctx->gsc_seqdata, gw->gw_seq, 1)) {
		CERROR("phase 1+: discard replayed req: seq %u\n", gw->gw_seq);
		*major = GSS_S_DUPLICATE_TOKEN;
		return -EACCES;
	}

	swabbed = __lustre_unpack_msg(msg, msglen);
	if (swabbed < 0) {
		CERROR("Failed to unpack after decryption\n");
		return -EINVAL;
	}
	req->rq_reqdata_len = msglen;

	if (msg->lm_bufcount < 1) {
		CERROR("Invalid buffer: is empty\n");
		return -EINVAL;
	}

	if (gw->gw_flags & LUSTRE_GSS_PACK_USER) {
		if (msg->lm_bufcount < offset + 1) {
			CERROR("no user descriptor included\n");
			return -EINVAL;
		}

		if (sptlrpc_unpack_user_desc(msg, offset, swabbed)) {
			CERROR("Mal-formed user descriptor\n");
			return -EINVAL;
		}

		req->rq_pack_udesc = 1;
		req->rq_user_desc = lustre_msg_buf(msg, offset, 0);
		offset++;
	}

	if (gw->gw_flags & LUSTRE_GSS_PACK_BULK) {
		if (msg->lm_bufcount < offset + 1) {
			CERROR("no bulk checksum included\n");
			return -EINVAL;
		}

		if (bulk_sec_desc_unpack(msg, offset, swabbed))
			return -EINVAL;

		req->rq_pack_bulk = 1;
		grctx->src_reqbsd = lustre_msg_buf(msg, offset, 0);
		grctx->src_reqbsd_size = lustre_msg_buflen(msg, offset);
	}

	req->rq_reqmsg = lustre_msg_buf(req->rq_reqbuf, 0, 0);
	req->rq_reqlen = req->rq_reqbuf->lm_buflens[0];
	return 0;
}

static
int gss_svc_handle_data(struct ptlrpc_request *req,
			struct gss_wire_ctx *gw)
{
	struct gss_svc_reqctx *grctx = gss_svc_ctx2reqctx(req->rq_svc_ctx);
	__u32		  major = 0;
	int		    rc = 0;

	grctx->src_ctx = gss_svc_upcall_get_ctx(req, gw);
	if (!grctx->src_ctx) {
		major = GSS_S_NO_CONTEXT;
		goto error;
	}

	switch (gw->gw_svc) {
	case SPTLRPC_SVC_NULL:
	case SPTLRPC_SVC_AUTH:
	case SPTLRPC_SVC_INTG:
		rc = gss_svc_verify_request(req, grctx, gw, &major);
		break;
	case SPTLRPC_SVC_PRIV:
		rc = gss_svc_unseal_request(req, grctx, gw, &major);
		break;
	default:
		CERROR("unsupported gss service %d\n", gw->gw_svc);
		rc = -EINVAL;
	}

	if (rc == 0)
		return SECSVC_OK;

	CERROR("svc %u failed: major 0x%08x: req xid "LPU64" ctx %p idx "
	       LPX64"(%u->%s)\n", gw->gw_svc, major, req->rq_xid,
	       grctx->src_ctx, gss_handle_to_u64(&gw->gw_handle),
	       grctx->src_ctx->gsc_uid, libcfs_nid2str(req->rq_peer.nid));
error:
	/* we only notify client in case of NO_CONTEXT/BAD_SIG, which
	 * might happen after server reboot, to allow recovery. */
	if ((major == GSS_S_NO_CONTEXT || major == GSS_S_BAD_SIG) &&
	    gss_pack_err_notify(req, major, 0) == 0)
		return SECSVC_COMPLETE;

	return SECSVC_DROP;
}

static
int gss_svc_handle_destroy(struct ptlrpc_request *req,
			   struct gss_wire_ctx *gw)
{
	struct gss_svc_reqctx  *grctx = gss_svc_ctx2reqctx(req->rq_svc_ctx);
	__u32		   major;

	req->rq_ctx_fini = 1;
	req->rq_no_reply = 1;

	grctx->src_ctx = gss_svc_upcall_get_ctx(req, gw);
	if (!grctx->src_ctx) {
		CDEBUG(D_SEC, "invalid gss context handle for destroy.\n");
		return SECSVC_DROP;
	}

	if (gw->gw_svc != SPTLRPC_SVC_INTG) {
		CERROR("svc %u is not supported in destroy.\n", gw->gw_svc);
		return SECSVC_DROP;
	}

	if (gss_svc_verify_request(req, grctx, gw, &major))
		return SECSVC_DROP;

	CWARN("destroy svc ctx %p idx "LPX64" (%u->%s)\n",
	      grctx->src_ctx, gss_handle_to_u64(&gw->gw_handle),
	      grctx->src_ctx->gsc_uid, libcfs_nid2str(req->rq_peer.nid));

	gss_svc_upcall_destroy_ctx(grctx->src_ctx);

	if (gw->gw_flags & LUSTRE_GSS_PACK_USER) {
		if (req->rq_reqbuf->lm_bufcount < 4) {
			CERROR("missing user descriptor, ignore it\n");
			return SECSVC_OK;
		}
		if (sptlrpc_unpack_user_desc(req->rq_reqbuf, 2,
					     ptlrpc_req_need_swab(req))) {
			CERROR("Mal-formed user descriptor, ignore it\n");
			return SECSVC_OK;
		}

		req->rq_pack_udesc = 1;
		req->rq_user_desc = lustre_msg_buf(req->rq_reqbuf, 2, 0);
	}

	return SECSVC_OK;
}

int gss_svc_accept(struct ptlrpc_sec_policy *policy, struct ptlrpc_request *req)
{
	struct gss_header      *ghdr;
	struct gss_svc_reqctx  *grctx;
	struct gss_wire_ctx    *gw;
	int		     swabbed, rc;

	LASSERT(req->rq_reqbuf);
	LASSERT(req->rq_svc_ctx == NULL);

	if (req->rq_reqbuf->lm_bufcount < 2) {
		CERROR("buf count only %d\n", req->rq_reqbuf->lm_bufcount);
		return SECSVC_DROP;
	}

	swabbed = ptlrpc_req_need_swab(req);

	ghdr = gss_swab_header(req->rq_reqbuf, 0, swabbed);
	if (ghdr == NULL) {
		CERROR("can't decode gss header\n");
		return SECSVC_DROP;
	}

	/* sanity checks */
	if (ghdr->gh_version != PTLRPC_GSS_VERSION) {
		CERROR("gss version %u, expect %u\n", ghdr->gh_version,
		       PTLRPC_GSS_VERSION);
		return SECSVC_DROP;
	}

	req->rq_sp_from = ghdr->gh_sp;

	/* alloc grctx data */
	OBD_ALLOC_PTR(grctx);
	if (!grctx)
		return SECSVC_DROP;

	grctx->src_base.sc_policy = sptlrpc_policy_get(policy);
	atomic_set(&grctx->src_base.sc_refcount, 1);
	req->rq_svc_ctx = &grctx->src_base;
	gw = &grctx->src_wirectx;

	/* save wire context */
	gw->gw_flags = ghdr->gh_flags;
	gw->gw_proc = ghdr->gh_proc;
	gw->gw_seq = ghdr->gh_seq;
	gw->gw_svc = ghdr->gh_svc;
	rawobj_from_netobj(&gw->gw_handle, &ghdr->gh_handle);

	/* keep original wire header which subject to checksum verification */
	if (swabbed)
		gss_header_swabber(ghdr);

	switch (ghdr->gh_proc) {
	case PTLRPC_GSS_PROC_INIT:
	case PTLRPC_GSS_PROC_CONTINUE_INIT:
		rc = gss_svc_handle_init(req, gw);
		break;
	case PTLRPC_GSS_PROC_DATA:
		rc = gss_svc_handle_data(req, gw);
		break;
	case PTLRPC_GSS_PROC_DESTROY:
		rc = gss_svc_handle_destroy(req, gw);
		break;
	default:
		CERROR("unknown proc %u\n", gw->gw_proc);
		rc = SECSVC_DROP;
		break;
	}

	switch (rc) {
	case SECSVC_OK:
		LASSERT(grctx->src_ctx);

		req->rq_auth_gss = 1;
		req->rq_auth_remote = grctx->src_ctx->gsc_remote;
		req->rq_auth_usr_mdt = grctx->src_ctx->gsc_usr_mds;
		req->rq_auth_usr_ost = grctx->src_ctx->gsc_usr_oss;
		req->rq_auth_usr_root = grctx->src_ctx->gsc_usr_root;
		req->rq_auth_uid = grctx->src_ctx->gsc_uid;
		req->rq_auth_mapped_uid = grctx->src_ctx->gsc_mapped_uid;
		break;
	case SECSVC_COMPLETE:
		break;
	case SECSVC_DROP:
		gss_svc_reqctx_free(grctx);
		req->rq_svc_ctx = NULL;
		break;
	}

	return rc;
}

void gss_svc_invalidate_ctx(struct ptlrpc_svc_ctx *svc_ctx)
{
	struct gss_svc_reqctx  *grctx;

	if (svc_ctx == NULL) {
		return;
	}

	grctx = gss_svc_ctx2reqctx(svc_ctx);

	CWARN("gss svc invalidate ctx %p(%u)\n",
	      grctx->src_ctx, grctx->src_ctx->gsc_uid);
	gss_svc_upcall_destroy_ctx(grctx->src_ctx);
}

static inline
int gss_svc_payload(struct gss_svc_reqctx *grctx, int early,
		    int msgsize, int privacy)
{
	/* we should treat early reply normally, but which is actually sharing
	 * the same ctx with original request, so in this case we should
	 * ignore the special ctx's special flags */
	if (early == 0 && gss_svc_reqctx_is_special(grctx))
		return grctx->src_reserve_len;

	return gss_mech_payload(NULL, msgsize, privacy);
}

static int gss_svc_bulk_payload(struct gss_svc_ctx *gctx,
				struct sptlrpc_flavor *flvr,
				int read)
{
	int     payload = sizeof(struct ptlrpc_bulk_sec_desc);

	if (read) {
		switch (SPTLRPC_FLVR_BULK_SVC(flvr->sf_rpc)) {
		case SPTLRPC_BULK_SVC_NULL:
			break;
		case SPTLRPC_BULK_SVC_INTG:
			payload += gss_mech_payload(NULL, 0, 0);
			break;
		case SPTLRPC_BULK_SVC_PRIV:
			payload += gss_mech_payload(NULL, 0, 1);
			break;
		case SPTLRPC_BULK_SVC_AUTH:
		default:
			LBUG();
		}
	}

	return payload;
}

int gss_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
{
	struct gss_svc_reqctx       *grctx;
	struct ptlrpc_reply_state   *rs;
	int			  early, privacy, svc, bsd_off = 0;
	__u32			ibuflens[2], buflens[4];
	int			  ibufcnt = 0, bufcnt;
	int			  txtsize, wmsg_size, rs_size;

	LASSERT(msglen % 8 == 0);

	if (req->rq_pack_bulk && !req->rq_bulk_read && !req->rq_bulk_write) {
		CERROR("client request bulk sec on non-bulk rpc\n");
		return -EPROTO;
	}

	svc = SPTLRPC_FLVR_SVC(req->rq_flvr.sf_rpc);
	early = (req->rq_packed_final == 0);

	grctx = gss_svc_ctx2reqctx(req->rq_svc_ctx);
	if (!early && gss_svc_reqctx_is_special(grctx))
		privacy = 0;
	else
		privacy = (svc == SPTLRPC_SVC_PRIV);

	if (privacy) {
		/* inner clear buffers */
		ibufcnt = 1;
		ibuflens[0] = msglen;

		if (req->rq_pack_bulk) {
			LASSERT(grctx->src_reqbsd);

			bsd_off = ibufcnt;
			ibuflens[ibufcnt++] = gss_svc_bulk_payload(
							grctx->src_ctx,
							&req->rq_flvr,
							req->rq_bulk_read);
		}

		txtsize = lustre_msg_size_v2(ibufcnt, ibuflens);
		txtsize += GSS_MAX_CIPHER_BLOCK;

		/* wrapper buffer */
		bufcnt = 2;
		buflens[0] = PTLRPC_GSS_HEADER_SIZE;
		buflens[1] = gss_svc_payload(grctx, early, txtsize, 1);
	} else {
		bufcnt = 2;
		buflens[0] = PTLRPC_GSS_HEADER_SIZE;
		buflens[1] = msglen;

		txtsize = buflens[0];
		if (svc == SPTLRPC_SVC_INTG)
			txtsize += buflens[1];

		if (req->rq_pack_bulk) {
			LASSERT(grctx->src_reqbsd);

			bsd_off = bufcnt;
			buflens[bufcnt] = gss_svc_bulk_payload(
							grctx->src_ctx,
							&req->rq_flvr,
							req->rq_bulk_read);
			if (svc == SPTLRPC_SVC_INTG)
				txtsize += buflens[bufcnt];
			bufcnt++;
		}

		if ((!early && gss_svc_reqctx_is_special(grctx)) ||
		    svc != SPTLRPC_SVC_NULL)
			buflens[bufcnt++] = gss_svc_payload(grctx, early,
							    txtsize, 0);
	}

	wmsg_size = lustre_msg_size_v2(bufcnt, buflens);

	rs_size = sizeof(*rs) + wmsg_size;
	rs = req->rq_reply_state;

	if (rs) {
		/* pre-allocated */
		LASSERT(rs->rs_size >= rs_size);
	} else {
		OBD_ALLOC_LARGE(rs, rs_size);
		if (rs == NULL)
			return -ENOMEM;

		rs->rs_size = rs_size;
	}

	rs->rs_repbuf = (struct lustre_msg *) (rs + 1);
	rs->rs_repbuf_len = wmsg_size;

	/* initialize the buffer */
	if (privacy) {
		lustre_init_msg_v2(rs->rs_repbuf, ibufcnt, ibuflens, NULL);
		rs->rs_msg = lustre_msg_buf(rs->rs_repbuf, 0, msglen);
	} else {
		lustre_init_msg_v2(rs->rs_repbuf, bufcnt, buflens, NULL);
		rs->rs_repbuf->lm_secflvr = req->rq_flvr.sf_rpc;

		rs->rs_msg = lustre_msg_buf(rs->rs_repbuf, 1, 0);
	}

	if (bsd_off) {
		grctx->src_repbsd = lustre_msg_buf(rs->rs_repbuf, bsd_off, 0);
		grctx->src_repbsd_size = lustre_msg_buflen(rs->rs_repbuf,
							   bsd_off);
	}

	gss_svc_reqctx_addref(grctx);
	rs->rs_svc_ctx = req->rq_svc_ctx;

	LASSERT(rs->rs_msg);
	req->rq_reply_state = rs;
	return 0;
}

static int gss_svc_seal(struct ptlrpc_request *req,
			struct ptlrpc_reply_state *rs,
			struct gss_svc_reqctx *grctx)
{
	struct gss_svc_ctx      *gctx = grctx->src_ctx;
	rawobj_t		 hdrobj, msgobj, token;
	struct gss_header       *ghdr;
	__u8		    *token_buf;
	int		      token_buflen;
	__u32		    buflens[2], major;
	int		      msglen, rc;

	/* get clear data length. note embedded lustre_msg might
	 * have been shrinked */
	if (req->rq_replen != lustre_msg_buflen(rs->rs_repbuf, 0))
		msglen = lustre_shrink_msg(rs->rs_repbuf, 0, req->rq_replen, 1);
	else
		msglen = lustre_msg_size_v2(rs->rs_repbuf->lm_bufcount,
					    rs->rs_repbuf->lm_buflens);

	/* temporarily use tail of buffer to hold gss header data */
	LASSERT(msglen + PTLRPC_GSS_HEADER_SIZE <= rs->rs_repbuf_len);
	ghdr = (struct gss_header *) ((char *) rs->rs_repbuf +
				rs->rs_repbuf_len - PTLRPC_GSS_HEADER_SIZE);
	ghdr->gh_version = PTLRPC_GSS_VERSION;
	ghdr->gh_sp = LUSTRE_SP_ANY;
	ghdr->gh_flags = 0;
	ghdr->gh_proc = PTLRPC_GSS_PROC_DATA;
	ghdr->gh_seq = grctx->src_wirectx.gw_seq;
	ghdr->gh_svc = SPTLRPC_SVC_PRIV;
	ghdr->gh_handle.len = 0;
	if (req->rq_pack_bulk)
		ghdr->gh_flags |= LUSTRE_GSS_PACK_BULK;

	/* allocate temporary cipher buffer */
	token_buflen = gss_mech_payload(gctx->gsc_mechctx, msglen, 1);
	OBD_ALLOC_LARGE(token_buf, token_buflen);
	if (token_buf == NULL)
		return -ENOMEM;

	hdrobj.len = PTLRPC_GSS_HEADER_SIZE;
	hdrobj.data = (__u8 *) ghdr;
	msgobj.len = msglen;
	msgobj.data = (__u8 *) rs->rs_repbuf;
	token.len = token_buflen;
	token.data = token_buf;

	major = lgss_wrap(gctx->gsc_mechctx, &hdrobj, &msgobj,
			  rs->rs_repbuf_len - PTLRPC_GSS_HEADER_SIZE, &token);
	if (major != GSS_S_COMPLETE) {
		CERROR("wrap message error: %08x\n", major);
		GOTO(out_free, rc = -EPERM);
	}
	LASSERT(token.len <= token_buflen);

	/* we are about to override data at rs->rs_repbuf, nullify pointers
	 * to which to catch further illegal usage. */
	if (req->rq_pack_bulk) {
		grctx->src_repbsd = NULL;
		grctx->src_repbsd_size = 0;
	}

	/* now fill the actual wire data
	 * - gss header
	 * - gss token
	 */
	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	buflens[1] = token.len;

	rs->rs_repdata_len = lustre_msg_size_v2(2, buflens);
	LASSERT(rs->rs_repdata_len <= rs->rs_repbuf_len);

	lustre_init_msg_v2(rs->rs_repbuf, 2, buflens, NULL);
	rs->rs_repbuf->lm_secflvr = req->rq_flvr.sf_rpc;

	memcpy(lustre_msg_buf(rs->rs_repbuf, 0, 0), ghdr,
	       PTLRPC_GSS_HEADER_SIZE);
	memcpy(lustre_msg_buf(rs->rs_repbuf, 1, 0), token.data, token.len);

	/* reply offset */
	if (req->rq_packed_final &&
	    (lustre_msghdr_get_flags(req->rq_reqmsg) & MSGHDR_AT_SUPPORT))
		req->rq_reply_off = gss_at_reply_off_priv;
	else
		req->rq_reply_off = 0;

	/* to catch upper layer's further access */
	rs->rs_msg = NULL;
	req->rq_repmsg = NULL;
	req->rq_replen = 0;

	rc = 0;
out_free:
	OBD_FREE_LARGE(token_buf, token_buflen);
	return rc;
}

int gss_svc_authorize(struct ptlrpc_request *req)
{
	struct ptlrpc_reply_state *rs = req->rq_reply_state;
	struct gss_svc_reqctx     *grctx = gss_svc_ctx2reqctx(req->rq_svc_ctx);
	struct gss_wire_ctx       *gw = &grctx->src_wirectx;
	int			early, rc;

	early = (req->rq_packed_final == 0);

	if (!early && gss_svc_reqctx_is_special(grctx)) {
		LASSERT(rs->rs_repdata_len != 0);

		req->rq_reply_off = gss_at_reply_off_integ;
		return 0;
	}

	/* early reply could happen in many cases */
	if (!early &&
	    gw->gw_proc != PTLRPC_GSS_PROC_DATA &&
	    gw->gw_proc != PTLRPC_GSS_PROC_DESTROY) {
		CERROR("proc %d not support\n", gw->gw_proc);
		return -EINVAL;
	}

	LASSERT(grctx->src_ctx);

	switch (gw->gw_svc) {
	case SPTLRPC_SVC_NULL:
	case SPTLRPC_SVC_AUTH:
	case SPTLRPC_SVC_INTG:
		rc = gss_svc_sign(req, rs, grctx, gw->gw_svc);
		break;
	case SPTLRPC_SVC_PRIV:
		rc = gss_svc_seal(req, rs, grctx);
		break;
	default:
		CERROR("Unknown service %d\n", gw->gw_svc);
		GOTO(out, rc = -EINVAL);
	}
	rc = 0;

out:
	return rc;
}

void gss_svc_free_rs(struct ptlrpc_reply_state *rs)
{
	struct gss_svc_reqctx *grctx;

	LASSERT(rs->rs_svc_ctx);
	grctx = container_of(rs->rs_svc_ctx, struct gss_svc_reqctx, src_base);

	gss_svc_reqctx_decref(grctx);
	rs->rs_svc_ctx = NULL;

	if (!rs->rs_prealloc)
		OBD_FREE_LARGE(rs, rs->rs_size);
}

void gss_svc_free_ctx(struct ptlrpc_svc_ctx *ctx)
{
	LASSERT(atomic_read(&ctx->sc_refcount) == 0);
	gss_svc_reqctx_free(gss_svc_ctx2reqctx(ctx));
}

int gss_copy_rvc_cli_ctx(struct ptlrpc_cli_ctx *cli_ctx,
			 struct ptlrpc_svc_ctx *svc_ctx)
{
	struct gss_cli_ctx     *cli_gctx = ctx2gctx(cli_ctx);
	struct gss_svc_ctx     *svc_gctx = gss_svc_ctx2gssctx(svc_ctx);
	struct gss_ctx	 *mechctx = NULL;

	LASSERT(cli_gctx);
	LASSERT(svc_gctx && svc_gctx->gsc_mechctx);

	cli_gctx->gc_proc = PTLRPC_GSS_PROC_DATA;
	cli_gctx->gc_win = GSS_SEQ_WIN;

	/* The problem is the reverse ctx might get lost in some recovery
	 * situations, and the same svc_ctx will be used to re-create it.
	 * if there's callback be sentout before that, new reverse ctx start
	 * with sequence 0 will lead to future callback rpc be treated as
	 * replay.
	 *
	 * each reverse root ctx will record its latest sequence number on its
	 * buddy svcctx before be destroied, so here we continue use it.
	 */
	atomic_set(&cli_gctx->gc_seq, svc_gctx->gsc_rvs_seq);

	if (gss_svc_upcall_dup_handle(&cli_gctx->gc_svc_handle, svc_gctx)) {
		CERROR("failed to dup svc handle\n");
		goto err_out;
	}

	if (lgss_copy_reverse_context(svc_gctx->gsc_mechctx, &mechctx) !=
	    GSS_S_COMPLETE) {
		CERROR("failed to copy mech context\n");
		goto err_svc_handle;
	}

	if (rawobj_dup(&cli_gctx->gc_handle, &svc_gctx->gsc_rvs_hdl)) {
		CERROR("failed to dup reverse handle\n");
		goto err_ctx;
	}

	cli_gctx->gc_mechctx = mechctx;
	gss_cli_ctx_uptodate(cli_gctx);

	return 0;

err_ctx:
	lgss_delete_sec_context(&mechctx);
err_svc_handle:
	rawobj_free(&cli_gctx->gc_svc_handle);
err_out:
	return -ENOMEM;
}

static void gss_init_at_reply_offset(void)
{
	__u32 buflens[3];
	int clearsize;

	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	buflens[1] = lustre_msg_early_size();
	buflens[2] = gss_cli_payload(NULL, buflens[1], 0);
	gss_at_reply_off_integ = lustre_msg_size_v2(3, buflens);

	buflens[0] = lustre_msg_early_size();
	clearsize = lustre_msg_size_v2(1, buflens);
	buflens[0] = PTLRPC_GSS_HEADER_SIZE;
	buflens[1] = gss_cli_payload(NULL, clearsize, 0);
	buflens[2] = gss_cli_payload(NULL, clearsize, 1);
	gss_at_reply_off_priv = lustre_msg_size_v2(3, buflens);
}

int __init sptlrpc_gss_init(void)
{
	int rc;

	rc = gss_init_lproc();
	if (rc)
		return rc;

	rc = gss_init_cli_upcall();
	if (rc)
		goto out_lproc;

	rc = gss_init_svc_upcall();
	if (rc)
		goto out_cli_upcall;

	rc = init_kerberos_module();
	if (rc)
		goto out_svc_upcall;

	/* register policy after all other stuff be intialized, because it
	 * might be in used immediately after the registration. */

	rc = gss_init_keyring();
	if (rc)
		goto out_kerberos;

#ifdef HAVE_GSS_PIPEFS
	rc = gss_init_pipefs();
	if (rc)
		goto out_keyring;
#endif

	gss_init_at_reply_offset();

	return 0;

#ifdef HAVE_GSS_PIPEFS
out_keyring:
	gss_exit_keyring();
#endif

out_kerberos:
	cleanup_kerberos_module();
out_svc_upcall:
	gss_exit_svc_upcall();
out_cli_upcall:
	gss_exit_cli_upcall();
out_lproc:
	gss_exit_lproc();
	return rc;
}

static void __exit sptlrpc_gss_exit(void)
{
	gss_exit_keyring();
#ifdef HAVE_GSS_PIPEFS
	gss_exit_pipefs();
#endif
	cleanup_kerberos_module();
	gss_exit_svc_upcall();
	gss_exit_cli_upcall();
	gss_exit_lproc();
}

MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
MODULE_DESCRIPTION("GSS security policy for Lustre");
MODULE_LICENSE("GPL");

module_init(sptlrpc_gss_init);
module_exit(sptlrpc_gss_exit);