1// SPDX-License-Identifier: BSD-3-Clause
   2/*
   3 * linux/net/sunrpc/auth_gss/auth_gss.c
   4 *
   5 * RPCSEC_GSS client authentication.
   6 *
   7 *  Copyright (c) 2000 The Regents of the University of Michigan.
   8 *  All rights reserved.
   9 *
  10 *  Dug Song       <dugsong@monkey.org>
  11 *  Andy Adamson   <andros@umich.edu>
  12 */
  13
  14#include <linux/module.h>
  15#include <linux/init.h>
  16#include <linux/types.h>
  17#include <linux/slab.h>
  18#include <linux/sched.h>
  19#include <linux/pagemap.h>
  20#include <linux/sunrpc/clnt.h>
  21#include <linux/sunrpc/auth.h>
  22#include <linux/sunrpc/auth_gss.h>
  23#include <linux/sunrpc/gss_krb5.h>
  24#include <linux/sunrpc/svcauth_gss.h>
  25#include <linux/sunrpc/gss_err.h>
  26#include <linux/workqueue.h>
  27#include <linux/sunrpc/rpc_pipe_fs.h>
  28#include <linux/sunrpc/gss_api.h>
  29#include <linux/uaccess.h>
  30#include <linux/hashtable.h>
  31
  32#include "auth_gss_internal.h"
  33#include "../netns.h"
  34
  35#include <trace/events/rpcgss.h>
  36
  37static const struct rpc_authops authgss_ops;
  38
  39static const struct rpc_credops gss_credops;
  40static const struct rpc_credops gss_nullops;
  41
  42#define GSS_RETRY_EXPIRED 5
  43static unsigned int gss_expired_cred_retry_delay = GSS_RETRY_EXPIRED;
  44
  45#define GSS_KEY_EXPIRE_TIMEO 240
  46static unsigned int gss_key_expire_timeo = GSS_KEY_EXPIRE_TIMEO;
  47
  48#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
  49# define RPCDBG_FACILITY	RPCDBG_AUTH
  50#endif
  51
  52/*
  53 * This compile-time check verifies that we will not exceed the
  54 * slack space allotted by the client and server auth_gss code
  55 * before they call gss_wrap().
  56 */
  57#define GSS_KRB5_MAX_SLACK_NEEDED					\
  58	(GSS_KRB5_TOK_HDR_LEN		/* gss token header */		\
  59	+ GSS_KRB5_MAX_CKSUM_LEN	/* gss token checksum */	\
  60	+ GSS_KRB5_MAX_BLOCKSIZE	/* confounder */		\
  61	+ GSS_KRB5_MAX_BLOCKSIZE	/* possible padding */		\
  62	+ GSS_KRB5_TOK_HDR_LEN		/* encrypted hdr in v2 token */	\
  63	+ GSS_KRB5_MAX_CKSUM_LEN	/* encryption hmac */		\
  64	+ XDR_UNIT * 2			/* RPC verifier */		\
  65	+ GSS_KRB5_TOK_HDR_LEN						\
  66	+ GSS_KRB5_MAX_CKSUM_LEN)
  67
  68#define GSS_CRED_SLACK		(RPC_MAX_AUTH_SIZE * 2)
  69/* length of a krb5 verifier (48), plus data added before arguments when
  70 * using integrity (two 4-byte integers): */
  71#define GSS_VERF_SLACK		100
  72
  73static DEFINE_HASHTABLE(gss_auth_hash_table, 4);
  74static DEFINE_SPINLOCK(gss_auth_hash_lock);
  75
  76struct gss_pipe {
  77	struct rpc_pipe_dir_object pdo;
  78	struct rpc_pipe *pipe;
  79	struct rpc_clnt *clnt;
  80	const char *name;
  81	struct kref kref;
  82};
  83
  84struct gss_auth {
  85	struct kref kref;
  86	struct hlist_node hash;
  87	struct rpc_auth rpc_auth;
  88	struct gss_api_mech *mech;
  89	enum rpc_gss_svc service;
  90	struct rpc_clnt *client;
  91	struct net	*net;
  92	netns_tracker	ns_tracker;
  93	/*
  94	 * There are two upcall pipes; dentry[1], named "gssd", is used
  95	 * for the new text-based upcall; dentry[0] is named after the
  96	 * mechanism (for example, "krb5") and exists for
  97	 * backwards-compatibility with older gssd's.
  98	 */
  99	struct gss_pipe *gss_pipe[2];
 100	const char *target_name;
 101};
 102
 103/* pipe_version >= 0 if and only if someone has a pipe open. */
 104static DEFINE_SPINLOCK(pipe_version_lock);
 105static struct rpc_wait_queue pipe_version_rpc_waitqueue;
 106static DECLARE_WAIT_QUEUE_HEAD(pipe_version_waitqueue);
 107static void gss_put_auth(struct gss_auth *gss_auth);
 108
 109static void gss_free_ctx(struct gss_cl_ctx *);
 110static const struct rpc_pipe_ops gss_upcall_ops_v0;
 111static const struct rpc_pipe_ops gss_upcall_ops_v1;
 112
 113static inline struct gss_cl_ctx *
 114gss_get_ctx(struct gss_cl_ctx *ctx)
 115{
 116	refcount_inc(&ctx->count);
 117	return ctx;
 118}
 119
 120static inline void
 121gss_put_ctx(struct gss_cl_ctx *ctx)
 122{
 123	if (refcount_dec_and_test(&ctx->count))
 124		gss_free_ctx(ctx);
 125}
 126
 127/* gss_cred_set_ctx:
 128 * called by gss_upcall_callback and gss_create_upcall in order
 129 * to set the gss context. The actual exchange of an old context
 130 * and a new one is protected by the pipe->lock.
 131 */
 132static void
 133gss_cred_set_ctx(struct rpc_cred *cred, struct gss_cl_ctx *ctx)
 134{
 135	struct gss_cred *gss_cred = container_of(cred, struct gss_cred, gc_base);
 136
 137	if (!test_bit(RPCAUTH_CRED_NEW, &cred->cr_flags))
 138		return;
 139	gss_get_ctx(ctx);
 140	rcu_assign_pointer(gss_cred->gc_ctx, ctx);
 141	set_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
 142	smp_mb__before_atomic();
 143	clear_bit(RPCAUTH_CRED_NEW, &cred->cr_flags);
 144}
 145
 146static struct gss_cl_ctx *
 147gss_cred_get_ctx(struct rpc_cred *cred)
 148{
 149	struct gss_cred *gss_cred = container_of(cred, struct gss_cred, gc_base);
 150	struct gss_cl_ctx *ctx = NULL;
 151
 152	rcu_read_lock();
 153	ctx = rcu_dereference(gss_cred->gc_ctx);
 154	if (ctx)
 155		gss_get_ctx(ctx);
 156	rcu_read_unlock();
 157	return ctx;
 158}
 159
 160static struct gss_cl_ctx *
 161gss_alloc_context(void)
 162{
 163	struct gss_cl_ctx *ctx;
 164
 165	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
 166	if (ctx != NULL) {
 167		ctx->gc_proc = RPC_GSS_PROC_DATA;
 168		ctx->gc_seq = 1;	/* NetApp 6.4R1 doesn't accept seq. no. 0 */
 169		spin_lock_init(&ctx->gc_seq_lock);
 170		refcount_set(&ctx->count,1);
 171	}
 172	return ctx;
 173}
 174
 175#define GSSD_MIN_TIMEOUT (60 * 60)
 176static const void *
 177gss_fill_context(const void *p, const void *end, struct gss_cl_ctx *ctx, struct gss_api_mech *gm)
 178{
 179	const void *q;
 180	unsigned int seclen;
 181	unsigned int timeout;
 182	unsigned long now = jiffies;
 183	u32 window_size;
 184	int ret;
 185
 186	/* First unsigned int gives the remaining lifetime in seconds of the
 187	 * credential - e.g. the remaining TGT lifetime for Kerberos or
 188	 * the -t value passed to GSSD.
 189	 */
 190	p = simple_get_bytes(p, end, &timeout, sizeof(timeout));
 191	if (IS_ERR(p))
 192		goto err;
 193	if (timeout == 0)
 194		timeout = GSSD_MIN_TIMEOUT;
 195	ctx->gc_expiry = now + ((unsigned long)timeout * HZ);
 196	/* Sequence number window. Determines the maximum number of
 197	 * simultaneous requests
 198	 */
 199	p = simple_get_bytes(p, end, &window_size, sizeof(window_size));
 200	if (IS_ERR(p))
 201		goto err;
 202	ctx->gc_win = window_size;
 203	/* gssd signals an error by passing ctx->gc_win = 0: */
 204	if (ctx->gc_win == 0) {
 205		/*
 206		 * in which case, p points to an error code. Anything other
 207		 * than -EKEYEXPIRED gets converted to -EACCES.
 208		 */
 209		p = simple_get_bytes(p, end, &ret, sizeof(ret));
 210		if (!IS_ERR(p))
 211			p = (ret == -EKEYEXPIRED) ? ERR_PTR(-EKEYEXPIRED) :
 212						    ERR_PTR(-EACCES);
 213		goto err;
 214	}
 215	/* copy the opaque wire context */
 216	p = simple_get_netobj(p, end, &ctx->gc_wire_ctx);
 217	if (IS_ERR(p))
 218		goto err;
 219	/* import the opaque security context */
 220	p  = simple_get_bytes(p, end, &seclen, sizeof(seclen));
 221	if (IS_ERR(p))
 222		goto err;
 223	q = (const void *)((const char *)p + seclen);
 224	if (unlikely(q > end || q < p)) {
 225		p = ERR_PTR(-EFAULT);
 226		goto err;
 227	}
 228	ret = gss_import_sec_context(p, seclen, gm, &ctx->gc_gss_ctx, NULL, GFP_KERNEL);
 229	if (ret < 0) {
 230		trace_rpcgss_import_ctx(ret);
 231		p = ERR_PTR(ret);
 232		goto err;
 233	}
 234
 235	/* is there any trailing data? */
 236	if (q == end) {
 237		p = q;
 238		goto done;
 239	}
 240
 241	/* pull in acceptor name (if there is one) */
 242	p = simple_get_netobj(q, end, &ctx->gc_acceptor);
 243	if (IS_ERR(p))
 244		goto err;
 245done:
 246	trace_rpcgss_context(window_size, ctx->gc_expiry, now, timeout,
 247			     ctx->gc_acceptor.len, ctx->gc_acceptor.data);
 248err:
 249	return p;
 250}
 251
 252/* XXX: Need some documentation about why UPCALL_BUF_LEN is so small.
 253 *	Is user space expecting no more than UPCALL_BUF_LEN bytes?
 254 *	Note that there are now _two_ NI_MAXHOST sized data items
 255 *	being passed in this string.
 256 */
 257#define UPCALL_BUF_LEN	256
 258
 259struct gss_upcall_msg {
 260	refcount_t count;
 261	kuid_t	uid;
 262	const char *service_name;
 263	struct rpc_pipe_msg msg;
 264	struct list_head list;
 265	struct gss_auth *auth;
 266	struct rpc_pipe *pipe;
 267	struct rpc_wait_queue rpc_waitqueue;
 268	wait_queue_head_t waitqueue;
 269	struct gss_cl_ctx *ctx;
 270	char databuf[UPCALL_BUF_LEN];
 271};
 272
 273static int get_pipe_version(struct net *net)
 274{
 275	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
 276	int ret;
 277
 278	spin_lock(&pipe_version_lock);
 279	if (sn->pipe_version >= 0) {
 280		atomic_inc(&sn->pipe_users);
 281		ret = sn->pipe_version;
 282	} else
 283		ret = -EAGAIN;
 284	spin_unlock(&pipe_version_lock);
 285	return ret;
 286}
 287
 288static void put_pipe_version(struct net *net)
 289{
 290	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
 291
 292	if (atomic_dec_and_lock(&sn->pipe_users, &pipe_version_lock)) {
 293		sn->pipe_version = -1;
 294		spin_unlock(&pipe_version_lock);
 295	}
 296}
 297
 298static void
 299gss_release_msg(struct gss_upcall_msg *gss_msg)
 300{
 301	struct net *net = gss_msg->auth->net;
 302	if (!refcount_dec_and_test(&gss_msg->count))
 303		return;
 304	put_pipe_version(net);
 305	BUG_ON(!list_empty(&gss_msg->list));
 306	if (gss_msg->ctx != NULL)
 307		gss_put_ctx(gss_msg->ctx);
 308	rpc_destroy_wait_queue(&gss_msg->rpc_waitqueue);
 309	gss_put_auth(gss_msg->auth);
 310	kfree_const(gss_msg->service_name);
 311	kfree(gss_msg);
 312}
 313
 314static struct gss_upcall_msg *
 315__gss_find_upcall(struct rpc_pipe *pipe, kuid_t uid, const struct gss_auth *auth)
 316{
 317	struct gss_upcall_msg *pos;
 318	list_for_each_entry(pos, &pipe->in_downcall, list) {
 319		if (!uid_eq(pos->uid, uid))
 320			continue;
 321		if (pos->auth->service != auth->service)
 322			continue;
 323		refcount_inc(&pos->count);
 324		return pos;
 325	}
 326	return NULL;
 327}
 328
 329/* Try to add an upcall to the pipefs queue.
 330 * If an upcall owned by our uid already exists, then we return a reference
 331 * to that upcall instead of adding the new upcall.
 332 */
 333static inline struct gss_upcall_msg *
 334gss_add_msg(struct gss_upcall_msg *gss_msg)
 335{
 336	struct rpc_pipe *pipe = gss_msg->pipe;
 337	struct gss_upcall_msg *old;
 338
 339	spin_lock(&pipe->lock);
 340	old = __gss_find_upcall(pipe, gss_msg->uid, gss_msg->auth);
 341	if (old == NULL) {
 342		refcount_inc(&gss_msg->count);
 343		list_add(&gss_msg->list, &pipe->in_downcall);
 344	} else
 345		gss_msg = old;
 346	spin_unlock(&pipe->lock);
 347	return gss_msg;
 348}
 349
 350static void
 351__gss_unhash_msg(struct gss_upcall_msg *gss_msg)
 352{
 353	list_del_init(&gss_msg->list);
 354	rpc_wake_up_status(&gss_msg->rpc_waitqueue, gss_msg->msg.errno);
 355	wake_up_all(&gss_msg->waitqueue);
 356	refcount_dec(&gss_msg->count);
 357}
 358
 359static void
 360gss_unhash_msg(struct gss_upcall_msg *gss_msg)
 361{
 362	struct rpc_pipe *pipe = gss_msg->pipe;
 363
 364	if (list_empty(&gss_msg->list))
 365		return;
 366	spin_lock(&pipe->lock);
 367	if (!list_empty(&gss_msg->list))
 368		__gss_unhash_msg(gss_msg);
 369	spin_unlock(&pipe->lock);
 370}
 371
 372static void
 373gss_handle_downcall_result(struct gss_cred *gss_cred, struct gss_upcall_msg *gss_msg)
 374{
 375	switch (gss_msg->msg.errno) {
 376	case 0:
 377		if (gss_msg->ctx == NULL)
 378			break;
 379		clear_bit(RPCAUTH_CRED_NEGATIVE, &gss_cred->gc_base.cr_flags);
 380		gss_cred_set_ctx(&gss_cred->gc_base, gss_msg->ctx);
 381		break;
 382	case -EKEYEXPIRED:
 383		set_bit(RPCAUTH_CRED_NEGATIVE, &gss_cred->gc_base.cr_flags);
 384	}
 385	gss_cred->gc_upcall_timestamp = jiffies;
 386	gss_cred->gc_upcall = NULL;
 387	rpc_wake_up_status(&gss_msg->rpc_waitqueue, gss_msg->msg.errno);
 388}
 389
 390static void
 391gss_upcall_callback(struct rpc_task *task)
 392{
 393	struct gss_cred *gss_cred = container_of(task->tk_rqstp->rq_cred,
 394			struct gss_cred, gc_base);
 395	struct gss_upcall_msg *gss_msg = gss_cred->gc_upcall;
 396	struct rpc_pipe *pipe = gss_msg->pipe;
 397
 398	spin_lock(&pipe->lock);
 399	gss_handle_downcall_result(gss_cred, gss_msg);
 400	spin_unlock(&pipe->lock);
 401	task->tk_status = gss_msg->msg.errno;
 402	gss_release_msg(gss_msg);
 403}
 404
 405static void gss_encode_v0_msg(struct gss_upcall_msg *gss_msg,
 406			      const struct cred *cred)
 407{
 408	struct user_namespace *userns = cred->user_ns;
 409
 410	uid_t uid = from_kuid_munged(userns, gss_msg->uid);
 411	memcpy(gss_msg->databuf, &uid, sizeof(uid));
 412	gss_msg->msg.data = gss_msg->databuf;
 413	gss_msg->msg.len = sizeof(uid);
 414
 415	BUILD_BUG_ON(sizeof(uid) > sizeof(gss_msg->databuf));
 416}
 417
 418static ssize_t
 419gss_v0_upcall(struct file *file, struct rpc_pipe_msg *msg,
 420		char __user *buf, size_t buflen)
 421{
 422	struct gss_upcall_msg *gss_msg = container_of(msg,
 423						      struct gss_upcall_msg,
 424						      msg);
 425	if (msg->copied == 0)
 426		gss_encode_v0_msg(gss_msg, file->f_cred);
 427	return rpc_pipe_generic_upcall(file, msg, buf, buflen);
 428}
 429
 430static int gss_encode_v1_msg(struct gss_upcall_msg *gss_msg,
 431				const char *service_name,
 432				const char *target_name,
 433				const struct cred *cred)
 434{
 435	struct user_namespace *userns = cred->user_ns;
 436	struct gss_api_mech *mech = gss_msg->auth->mech;
 437	char *p = gss_msg->databuf;
 438	size_t buflen = sizeof(gss_msg->databuf);
 439	int len;
 440
 441	len = scnprintf(p, buflen, "mech=%s uid=%d", mech->gm_name,
 442			from_kuid_munged(userns, gss_msg->uid));
 443	buflen -= len;
 444	p += len;
 445	gss_msg->msg.len = len;
 446
 447	/*
 448	 * target= is a full service principal that names the remote
 449	 * identity that we are authenticating to.
 450	 */
 451	if (target_name) {
 452		len = scnprintf(p, buflen, " target=%s", target_name);
 453		buflen -= len;
 454		p += len;
 455		gss_msg->msg.len += len;
 456	}
 457
 458	/*
 459	 * gssd uses service= and srchost= to select a matching key from
 460	 * the system's keytab to use as the source principal.
 461	 *
 462	 * service= is the service name part of the source principal,
 463	 * or "*" (meaning choose any).
 464	 *
 465	 * srchost= is the hostname part of the source principal. When
 466	 * not provided, gssd uses the local hostname.
 467	 */
 468	if (service_name) {
 469		char *c = strchr(service_name, '@');
 470
 471		if (!c)
 472			len = scnprintf(p, buflen, " service=%s",
 473					service_name);
 474		else
 475			len = scnprintf(p, buflen,
 476					" service=%.*s srchost=%s",
 477					(int)(c - service_name),
 478					service_name, c + 1);
 479		buflen -= len;
 480		p += len;
 481		gss_msg->msg.len += len;
 482	}
 483
 484	if (mech->gm_upcall_enctypes) {
 485		len = scnprintf(p, buflen, " enctypes=%s",
 486				mech->gm_upcall_enctypes);
 487		buflen -= len;
 488		p += len;
 489		gss_msg->msg.len += len;
 490	}
 491	trace_rpcgss_upcall_msg(gss_msg->databuf);
 492	len = scnprintf(p, buflen, "\n");
 493	if (len == 0)
 494		goto out_overflow;
 495	gss_msg->msg.len += len;
 496	gss_msg->msg.data = gss_msg->databuf;
 497	return 0;
 498out_overflow:
 499	WARN_ON_ONCE(1);
 500	return -ENOMEM;
 501}
 502
 503static ssize_t
 504gss_v1_upcall(struct file *file, struct rpc_pipe_msg *msg,
 505		char __user *buf, size_t buflen)
 506{
 507	struct gss_upcall_msg *gss_msg = container_of(msg,
 508						      struct gss_upcall_msg,
 509						      msg);
 510	int err;
 511	if (msg->copied == 0) {
 512		err = gss_encode_v1_msg(gss_msg,
 513					gss_msg->service_name,
 514					gss_msg->auth->target_name,
 515					file->f_cred);
 516		if (err)
 517			return err;
 518	}
 519	return rpc_pipe_generic_upcall(file, msg, buf, buflen);
 520}
 521
 522static struct gss_upcall_msg *
 523gss_alloc_msg(struct gss_auth *gss_auth,
 524		kuid_t uid, const char *service_name)
 525{
 526	struct gss_upcall_msg *gss_msg;
 527	int vers;
 528	int err = -ENOMEM;
 529
 530	gss_msg = kzalloc(sizeof(*gss_msg), GFP_KERNEL);
 531	if (gss_msg == NULL)
 532		goto err;
 533	vers = get_pipe_version(gss_auth->net);
 534	err = vers;
 535	if (err < 0)
 536		goto err_free_msg;
 537	gss_msg->pipe = gss_auth->gss_pipe[vers]->pipe;
 538	INIT_LIST_HEAD(&gss_msg->list);
 539	rpc_init_wait_queue(&gss_msg->rpc_waitqueue, "RPCSEC_GSS upcall waitq");
 540	init_waitqueue_head(&gss_msg->waitqueue);
 541	refcount_set(&gss_msg->count, 1);
 542	gss_msg->uid = uid;
 543	gss_msg->auth = gss_auth;
 544	kref_get(&gss_auth->kref);
 545	if (service_name) {
 546		gss_msg->service_name = kstrdup_const(service_name, GFP_KERNEL);
 547		if (!gss_msg->service_name) {
 548			err = -ENOMEM;
 549			goto err_put_pipe_version;
 550		}
 551	}
 552	return gss_msg;
 553err_put_pipe_version:
 554	put_pipe_version(gss_auth->net);
 555err_free_msg:
 556	kfree(gss_msg);
 557err:
 558	return ERR_PTR(err);
 559}
 560
 561static struct gss_upcall_msg *
 562gss_setup_upcall(struct gss_auth *gss_auth, struct rpc_cred *cred)
 563{
 564	struct gss_cred *gss_cred = container_of(cred,
 565			struct gss_cred, gc_base);
 566	struct gss_upcall_msg *gss_new, *gss_msg;
 567	kuid_t uid = cred->cr_cred->fsuid;
 568
 569	gss_new = gss_alloc_msg(gss_auth, uid, gss_cred->gc_principal);
 570	if (IS_ERR(gss_new))
 571		return gss_new;
 572	gss_msg = gss_add_msg(gss_new);
 573	if (gss_msg == gss_new) {
 574		int res;
 575		refcount_inc(&gss_msg->count);
 576		res = rpc_queue_upcall(gss_new->pipe, &gss_new->msg);
 577		if (res) {
 578			gss_unhash_msg(gss_new);
 579			refcount_dec(&gss_msg->count);
 580			gss_release_msg(gss_new);
 581			gss_msg = ERR_PTR(res);
 582		}
 583	} else
 584		gss_release_msg(gss_new);
 585	return gss_msg;
 586}
 587
 588static void warn_gssd(void)
 589{
 590	dprintk("AUTH_GSS upcall failed. Please check user daemon is running.\n");
 591}
 592
 593static inline int
 594gss_refresh_upcall(struct rpc_task *task)
 595{
 596	struct rpc_cred *cred = task->tk_rqstp->rq_cred;
 597	struct gss_auth *gss_auth = container_of(cred->cr_auth,
 598			struct gss_auth, rpc_auth);
 599	struct gss_cred *gss_cred = container_of(cred,
 600			struct gss_cred, gc_base);
 601	struct gss_upcall_msg *gss_msg;
 602	struct rpc_pipe *pipe;
 603	int err = 0;
 604
 605	gss_msg = gss_setup_upcall(gss_auth, cred);
 606	if (PTR_ERR(gss_msg) == -EAGAIN) {
 607		/* XXX: warning on the first, under the assumption we
 608		 * shouldn't normally hit this case on a refresh. */
 609		warn_gssd();
 610		rpc_sleep_on_timeout(&pipe_version_rpc_waitqueue,
 611				task, NULL, jiffies + (15 * HZ));
 612		err = -EAGAIN;
 613		goto out;
 614	}
 615	if (IS_ERR(gss_msg)) {
 616		err = PTR_ERR(gss_msg);
 617		goto out;
 618	}
 619	pipe = gss_msg->pipe;
 620	spin_lock(&pipe->lock);
 621	if (gss_cred->gc_upcall != NULL)
 622		rpc_sleep_on(&gss_cred->gc_upcall->rpc_waitqueue, task, NULL);
 623	else if (gss_msg->ctx == NULL && gss_msg->msg.errno >= 0) {
 624		gss_cred->gc_upcall = gss_msg;
 625		/* gss_upcall_callback will release the reference to gss_upcall_msg */
 626		refcount_inc(&gss_msg->count);
 627		rpc_sleep_on(&gss_msg->rpc_waitqueue, task, gss_upcall_callback);
 628	} else {
 629		gss_handle_downcall_result(gss_cred, gss_msg);
 630		err = gss_msg->msg.errno;
 631	}
 632	spin_unlock(&pipe->lock);
 633	gss_release_msg(gss_msg);
 634out:
 635	trace_rpcgss_upcall_result(from_kuid(&init_user_ns,
 636					     cred->cr_cred->fsuid), err);
 637	return err;
 638}
 639
 640static inline int
 641gss_create_upcall(struct gss_auth *gss_auth, struct gss_cred *gss_cred)
 642{
 643	struct net *net = gss_auth->net;
 644	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
 645	struct rpc_pipe *pipe;
 646	struct rpc_cred *cred = &gss_cred->gc_base;
 647	struct gss_upcall_msg *gss_msg;
 648	DEFINE_WAIT(wait);
 649	int err;
 650
 651retry:
 652	err = 0;
 653	/* if gssd is down, just skip upcalling altogether */
 654	if (!gssd_running(net)) {
 655		warn_gssd();
 656		err = -EACCES;
 657		goto out;
 658	}
 659	gss_msg = gss_setup_upcall(gss_auth, cred);
 660	if (PTR_ERR(gss_msg) == -EAGAIN) {
 661		err = wait_event_interruptible_timeout(pipe_version_waitqueue,
 662				sn->pipe_version >= 0, 15 * HZ);
 663		if (sn->pipe_version < 0) {
 664			warn_gssd();
 665			err = -EACCES;
 666		}
 667		if (err < 0)
 668			goto out;
 669		goto retry;
 670	}
 671	if (IS_ERR(gss_msg)) {
 672		err = PTR_ERR(gss_msg);
 673		goto out;
 674	}
 675	pipe = gss_msg->pipe;
 676	for (;;) {
 677		prepare_to_wait(&gss_msg->waitqueue, &wait, TASK_KILLABLE);
 678		spin_lock(&pipe->lock);
 679		if (gss_msg->ctx != NULL || gss_msg->msg.errno < 0) {
 680			break;
 681		}
 682		spin_unlock(&pipe->lock);
 683		if (fatal_signal_pending(current)) {
 684			err = -ERESTARTSYS;
 685			goto out_intr;
 686		}
 687		schedule();
 688	}
 689	if (gss_msg->ctx) {
 690		trace_rpcgss_ctx_init(gss_cred);
 691		gss_cred_set_ctx(cred, gss_msg->ctx);
 692	} else {
 693		err = gss_msg->msg.errno;
 694	}
 695	spin_unlock(&pipe->lock);
 696out_intr:
 697	finish_wait(&gss_msg->waitqueue, &wait);
 698	gss_release_msg(gss_msg);
 699out:
 700	trace_rpcgss_upcall_result(from_kuid(&init_user_ns,
 701					     cred->cr_cred->fsuid), err);
 702	return err;
 703}
 704
 705static struct gss_upcall_msg *
 706gss_find_downcall(struct rpc_pipe *pipe, kuid_t uid)
 707{
 708	struct gss_upcall_msg *pos;
 709	list_for_each_entry(pos, &pipe->in_downcall, list) {
 710		if (!uid_eq(pos->uid, uid))
 711			continue;
 712		if (!rpc_msg_is_inflight(&pos->msg))
 713			continue;
 714		refcount_inc(&pos->count);
 715		return pos;
 716	}
 717	return NULL;
 718}
 719
 720#define MSG_BUF_MAXSIZE 1024
 721
 722static ssize_t
 723gss_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 724{
 725	const void *p, *end;
 726	void *buf;
 727	struct gss_upcall_msg *gss_msg;
 728	struct rpc_pipe *pipe = RPC_I(file_inode(filp))->pipe;
 729	struct gss_cl_ctx *ctx;
 730	uid_t id;
 731	kuid_t uid;
 732	ssize_t err = -EFBIG;
 733
 734	if (mlen > MSG_BUF_MAXSIZE)
 735		goto out;
 736	err = -ENOMEM;
 737	buf = kmalloc(mlen, GFP_KERNEL);
 738	if (!buf)
 739		goto out;
 740
 741	err = -EFAULT;
 742	if (copy_from_user(buf, src, mlen))
 743		goto err;
 744
 745	end = (const void *)((char *)buf + mlen);
 746	p = simple_get_bytes(buf, end, &id, sizeof(id));
 747	if (IS_ERR(p)) {
 748		err = PTR_ERR(p);
 749		goto err;
 750	}
 751
 752	uid = make_kuid(current_user_ns(), id);
 753	if (!uid_valid(uid)) {
 754		err = -EINVAL;
 755		goto err;
 756	}
 757
 758	err = -ENOMEM;
 759	ctx = gss_alloc_context();
 760	if (ctx == NULL)
 761		goto err;
 762
 763	err = -ENOENT;
 764	/* Find a matching upcall */
 765	spin_lock(&pipe->lock);
 766	gss_msg = gss_find_downcall(pipe, uid);
 767	if (gss_msg == NULL) {
 768		spin_unlock(&pipe->lock);
 769		goto err_put_ctx;
 770	}
 771	list_del_init(&gss_msg->list);
 772	spin_unlock(&pipe->lock);
 773
 774	p = gss_fill_context(p, end, ctx, gss_msg->auth->mech);
 775	if (IS_ERR(p)) {
 776		err = PTR_ERR(p);
 777		switch (err) {
 778		case -EACCES:
 779		case -EKEYEXPIRED:
 780			gss_msg->msg.errno = err;
 781			err = mlen;
 782			break;
 783		case -EFAULT:
 784		case -ENOMEM:
 785		case -EINVAL:
 786		case -ENOSYS:
 787			gss_msg->msg.errno = -EAGAIN;
 788			break;
 789		default:
 790			printk(KERN_CRIT "%s: bad return from "
 791				"gss_fill_context: %zd\n", __func__, err);
 792			gss_msg->msg.errno = -EIO;
 793		}
 794		goto err_release_msg;
 795	}
 796	gss_msg->ctx = gss_get_ctx(ctx);
 797	err = mlen;
 798
 799err_release_msg:
 800	spin_lock(&pipe->lock);
 801	__gss_unhash_msg(gss_msg);
 802	spin_unlock(&pipe->lock);
 803	gss_release_msg(gss_msg);
 804err_put_ctx:
 805	gss_put_ctx(ctx);
 806err:
 807	kfree(buf);
 808out:
 809	return err;
 810}
 811
 812static int gss_pipe_open(struct inode *inode, int new_version)
 813{
 814	struct net *net = inode->i_sb->s_fs_info;
 815	struct sunrpc_net *sn = net_generic(net, sunrpc_net_id);
 816	int ret = 0;
 817
 818	spin_lock(&pipe_version_lock);
 819	if (sn->pipe_version < 0) {
 820		/* First open of any gss pipe determines the version: */
 821		sn->pipe_version = new_version;
 822		rpc_wake_up(&pipe_version_rpc_waitqueue);
 823		wake_up(&pipe_version_waitqueue);
 824	} else if (sn->pipe_version != new_version) {
 825		/* Trying to open a pipe of a different version */
 826		ret = -EBUSY;
 827		goto out;
 828	}
 829	atomic_inc(&sn->pipe_users);
 830out:
 831	spin_unlock(&pipe_version_lock);
 832	return ret;
 833
 834}
 835
 836static int gss_pipe_open_v0(struct inode *inode)
 837{
 838	return gss_pipe_open(inode, 0);
 839}
 840
 841static int gss_pipe_open_v1(struct inode *inode)
 842{
 843	return gss_pipe_open(inode, 1);
 844}
 845
 846static void
 847gss_pipe_release(struct inode *inode)
 848{
 849	struct net *net = inode->i_sb->s_fs_info;
 850	struct rpc_pipe *pipe = RPC_I(inode)->pipe;
 851	struct gss_upcall_msg *gss_msg;
 852
 853restart:
 854	spin_lock(&pipe->lock);
 855	list_for_each_entry(gss_msg, &pipe->in_downcall, list) {
 856
 857		if (!list_empty(&gss_msg->msg.list))
 858			continue;
 859		gss_msg->msg.errno = -EPIPE;
 860		refcount_inc(&gss_msg->count);
 861		__gss_unhash_msg(gss_msg);
 862		spin_unlock(&pipe->lock);
 863		gss_release_msg(gss_msg);
 864		goto restart;
 865	}
 866	spin_unlock(&pipe->lock);
 867
 868	put_pipe_version(net);
 869}
 870
 871static void
 872gss_pipe_destroy_msg(struct rpc_pipe_msg *msg)
 873{
 874	struct gss_upcall_msg *gss_msg = container_of(msg, struct gss_upcall_msg, msg);
 875
 876	if (msg->errno < 0) {
 877		refcount_inc(&gss_msg->count);
 878		gss_unhash_msg(gss_msg);
 879		if (msg->errno == -ETIMEDOUT)
 880			warn_gssd();
 881		gss_release_msg(gss_msg);
 882	}
 883	gss_release_msg(gss_msg);
 884}
 885
 886static void gss_pipe_dentry_destroy(struct dentry *dir,
 887		struct rpc_pipe_dir_object *pdo)
 888{
 889	struct gss_pipe *gss_pipe = pdo->pdo_data;
 890	struct rpc_pipe *pipe = gss_pipe->pipe;
 891
 892	if (pipe->dentry != NULL) {
 893		rpc_unlink(pipe->dentry);
 894		pipe->dentry = NULL;
 895	}
 896}
 897
 898static int gss_pipe_dentry_create(struct dentry *dir,
 899		struct rpc_pipe_dir_object *pdo)
 900{
 901	struct gss_pipe *p = pdo->pdo_data;
 902	struct dentry *dentry;
 903
 904	dentry = rpc_mkpipe_dentry(dir, p->name, p->clnt, p->pipe);
 905	if (IS_ERR(dentry))
 906		return PTR_ERR(dentry);
 907	p->pipe->dentry = dentry;
 908	return 0;
 909}
 910
 911static const struct rpc_pipe_dir_object_ops gss_pipe_dir_object_ops = {
 912	.create = gss_pipe_dentry_create,
 913	.destroy = gss_pipe_dentry_destroy,
 914};
 915
 916static struct gss_pipe *gss_pipe_alloc(struct rpc_clnt *clnt,
 917		const char *name,
 918		const struct rpc_pipe_ops *upcall_ops)
 919{
 920	struct gss_pipe *p;
 921	int err = -ENOMEM;
 922
 923	p = kmalloc(sizeof(*p), GFP_KERNEL);
 924	if (p == NULL)
 925		goto err;
 926	p->pipe = rpc_mkpipe_data(upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
 927	if (IS_ERR(p->pipe)) {
 928		err = PTR_ERR(p->pipe);
 929		goto err_free_gss_pipe;
 930	}
 931	p->name = name;
 932	p->clnt = clnt;
 933	kref_init(&p->kref);
 934	rpc_init_pipe_dir_object(&p->pdo,
 935			&gss_pipe_dir_object_ops,
 936			p);
 937	return p;
 938err_free_gss_pipe:
 939	kfree(p);
 940err:
 941	return ERR_PTR(err);
 942}
 943
 944struct gss_alloc_pdo {
 945	struct rpc_clnt *clnt;
 946	const char *name;
 947	const struct rpc_pipe_ops *upcall_ops;
 948};
 949
 950static int gss_pipe_match_pdo(struct rpc_pipe_dir_object *pdo, void *data)
 951{
 952	struct gss_pipe *gss_pipe;
 953	struct gss_alloc_pdo *args = data;
 954
 955	if (pdo->pdo_ops != &gss_pipe_dir_object_ops)
 956		return 0;
 957	gss_pipe = container_of(pdo, struct gss_pipe, pdo);
 958	if (strcmp(gss_pipe->name, args->name) != 0)
 959		return 0;
 960	if (!kref_get_unless_zero(&gss_pipe->kref))
 961		return 0;
 962	return 1;
 963}
 964
 965static struct rpc_pipe_dir_object *gss_pipe_alloc_pdo(void *data)
 966{
 967	struct gss_pipe *gss_pipe;
 968	struct gss_alloc_pdo *args = data;
 969
 970	gss_pipe = gss_pipe_alloc(args->clnt, args->name, args->upcall_ops);
 971	if (!IS_ERR(gss_pipe))
 972		return &gss_pipe->pdo;
 973	return NULL;
 974}
 975
 976static struct gss_pipe *gss_pipe_get(struct rpc_clnt *clnt,
 977		const char *name,
 978		const struct rpc_pipe_ops *upcall_ops)
 979{
 980	struct net *net = rpc_net_ns(clnt);
 981	struct rpc_pipe_dir_object *pdo;
 982	struct gss_alloc_pdo args = {
 983		.clnt = clnt,
 984		.name = name,
 985		.upcall_ops = upcall_ops,
 986	};
 987
 988	pdo = rpc_find_or_alloc_pipe_dir_object(net,
 989			&clnt->cl_pipedir_objects,
 990			gss_pipe_match_pdo,
 991			gss_pipe_alloc_pdo,
 992			&args);
 993	if (pdo != NULL)
 994		return container_of(pdo, struct gss_pipe, pdo);
 995	return ERR_PTR(-ENOMEM);
 996}
 997
 998static void __gss_pipe_free(struct gss_pipe *p)
 999{
1000	struct rpc_clnt *clnt = p->clnt;
1001	struct net *net = rpc_net_ns(clnt);
1002
1003	rpc_remove_pipe_dir_object(net,
1004			&clnt->cl_pipedir_objects,
1005			&p->pdo);
1006	rpc_destroy_pipe_data(p->pipe);
1007	kfree(p);
1008}
1009
1010static void __gss_pipe_release(struct kref *kref)
1011{
1012	struct gss_pipe *p = container_of(kref, struct gss_pipe, kref);
1013
1014	__gss_pipe_free(p);
1015}
1016
1017static void gss_pipe_free(struct gss_pipe *p)
1018{
1019	if (p != NULL)
1020		kref_put(&p->kref, __gss_pipe_release);
1021}
1022
1023/*
1024 * NOTE: we have the opportunity to use different
1025 * parameters based on the input flavor (which must be a pseudoflavor)
1026 */
1027static struct gss_auth *
1028gss_create_new(const struct rpc_auth_create_args *args, struct rpc_clnt *clnt)
1029{
1030	rpc_authflavor_t flavor = args->pseudoflavor;
1031	struct gss_auth *gss_auth;
1032	struct gss_pipe *gss_pipe;
1033	struct rpc_auth * auth;
1034	int err = -ENOMEM; /* XXX? */
1035
1036	if (!try_module_get(THIS_MODULE))
1037		return ERR_PTR(err);
1038	if (!(gss_auth = kmalloc(sizeof(*gss_auth), GFP_KERNEL)))
1039		goto out_dec;
1040	INIT_HLIST_NODE(&gss_auth->hash);
1041	gss_auth->target_name = NULL;
1042	if (args->target_name) {
1043		gss_auth->target_name = kstrdup(args->target_name, GFP_KERNEL);
1044		if (gss_auth->target_name == NULL)
1045			goto err_free;
1046	}
1047	gss_auth->client = clnt;
1048	gss_auth->net = get_net_track(rpc_net_ns(clnt), &gss_auth->ns_tracker,
1049				      GFP_KERNEL);
1050	err = -EINVAL;
1051	gss_auth->mech = gss_mech_get_by_pseudoflavor(flavor);
1052	if (!gss_auth->mech)
1053		goto err_put_net;
1054	gss_auth->service = gss_pseudoflavor_to_service(gss_auth->mech, flavor);
1055	if (gss_auth->service == 0)
1056		goto err_put_mech;
1057	if (!gssd_running(gss_auth->net))
1058		goto err_put_mech;
1059	auth = &gss_auth->rpc_auth;
1060	auth->au_cslack = GSS_CRED_SLACK >> 2;
1061	BUILD_BUG_ON(GSS_KRB5_MAX_SLACK_NEEDED > RPC_MAX_AUTH_SIZE);
1062	auth->au_rslack = GSS_KRB5_MAX_SLACK_NEEDED >> 2;
1063	auth->au_verfsize = GSS_VERF_SLACK >> 2;
1064	auth->au_ralign = GSS_VERF_SLACK >> 2;
1065	__set_bit(RPCAUTH_AUTH_UPDATE_SLACK, &auth->au_flags);
1066	auth->au_ops = &authgss_ops;
1067	auth->au_flavor = flavor;
1068	if (gss_pseudoflavor_to_datatouch(gss_auth->mech, flavor))
1069		__set_bit(RPCAUTH_AUTH_DATATOUCH, &auth->au_flags);
1070	refcount_set(&auth->au_count, 1);
1071	kref_init(&gss_auth->kref);
1072
1073	err = rpcauth_init_credcache(auth);
1074	if (err)
1075		goto err_put_mech;
1076	/*
1077	 * Note: if we created the old pipe first, then someone who
1078	 * examined the directory at the right moment might conclude
1079	 * that we supported only the old pipe.  So we instead create
1080	 * the new pipe first.
1081	 */
1082	gss_pipe = gss_pipe_get(clnt, "gssd", &gss_upcall_ops_v1);
1083	if (IS_ERR(gss_pipe)) {
1084		err = PTR_ERR(gss_pipe);
1085		goto err_destroy_credcache;
1086	}
1087	gss_auth->gss_pipe[1] = gss_pipe;
1088
1089	gss_pipe = gss_pipe_get(clnt, gss_auth->mech->gm_name,
1090			&gss_upcall_ops_v0);
1091	if (IS_ERR(gss_pipe)) {
1092		err = PTR_ERR(gss_pipe);
1093		goto err_destroy_pipe_1;
1094	}
1095	gss_auth->gss_pipe[0] = gss_pipe;
1096
1097	return gss_auth;
1098err_destroy_pipe_1:
1099	gss_pipe_free(gss_auth->gss_pipe[1]);
1100err_destroy_credcache:
1101	rpcauth_destroy_credcache(auth);
1102err_put_mech:
1103	gss_mech_put(gss_auth->mech);
1104err_put_net:
1105	put_net_track(gss_auth->net, &gss_auth->ns_tracker);
1106err_free:
1107	kfree(gss_auth->target_name);
1108	kfree(gss_auth);
1109out_dec:
1110	module_put(THIS_MODULE);
1111	trace_rpcgss_createauth(flavor, err);
1112	return ERR_PTR(err);
1113}
1114
1115static void
1116gss_free(struct gss_auth *gss_auth)
1117{
1118	gss_pipe_free(gss_auth->gss_pipe[0]);
1119	gss_pipe_free(gss_auth->gss_pipe[1]);
1120	gss_mech_put(gss_auth->mech);
1121	put_net_track(gss_auth->net, &gss_auth->ns_tracker);
1122	kfree(gss_auth->target_name);
1123
1124	kfree(gss_auth);
1125	module_put(THIS_MODULE);
1126}
1127
1128static void
1129gss_free_callback(struct kref *kref)
1130{
1131	struct gss_auth *gss_auth = container_of(kref, struct gss_auth, kref);
1132
1133	gss_free(gss_auth);
1134}
1135
1136static void
1137gss_put_auth(struct gss_auth *gss_auth)
1138{
1139	kref_put(&gss_auth->kref, gss_free_callback);
1140}
1141
1142static void
1143gss_destroy(struct rpc_auth *auth)
1144{
1145	struct gss_auth *gss_auth = container_of(auth,
1146			struct gss_auth, rpc_auth);
1147
1148	if (hash_hashed(&gss_auth->hash)) {
1149		spin_lock(&gss_auth_hash_lock);
1150		hash_del(&gss_auth->hash);
1151		spin_unlock(&gss_auth_hash_lock);
1152	}
1153
1154	gss_pipe_free(gss_auth->gss_pipe[0]);
1155	gss_auth->gss_pipe[0] = NULL;
1156	gss_pipe_free(gss_auth->gss_pipe[1]);
1157	gss_auth->gss_pipe[1] = NULL;
1158	rpcauth_destroy_credcache(auth);
1159
1160	gss_put_auth(gss_auth);
1161}
1162
1163/*
1164 * Auths may be shared between rpc clients that were cloned from a
1165 * common client with the same xprt, if they also share the flavor and
1166 * target_name.
1167 *
1168 * The auth is looked up from the oldest parent sharing the same
1169 * cl_xprt, and the auth itself references only that common parent
1170 * (which is guaranteed to last as long as any of its descendants).
1171 */
1172static struct gss_auth *
1173gss_auth_find_or_add_hashed(const struct rpc_auth_create_args *args,
1174		struct rpc_clnt *clnt,
1175		struct gss_auth *new)
1176{
1177	struct gss_auth *gss_auth;
1178	unsigned long hashval = (unsigned long)clnt;
1179
1180	spin_lock(&gss_auth_hash_lock);
1181	hash_for_each_possible(gss_auth_hash_table,
1182			gss_auth,
1183			hash,
1184			hashval) {
1185		if (gss_auth->client != clnt)
1186			continue;
1187		if (gss_auth->rpc_auth.au_flavor != args->pseudoflavor)
1188			continue;
1189		if (gss_auth->target_name != args->target_name) {
1190			if (gss_auth->target_name == NULL)
1191				continue;
1192			if (args->target_name == NULL)
1193				continue;
1194			if (strcmp(gss_auth->target_name, args->target_name))
1195				continue;
1196		}
1197		if (!refcount_inc_not_zero(&gss_auth->rpc_auth.au_count))
1198			continue;
1199		goto out;
1200	}
1201	if (new)
1202		hash_add(gss_auth_hash_table, &new->hash, hashval);
1203	gss_auth = new;
1204out:
1205	spin_unlock(&gss_auth_hash_lock);
1206	return gss_auth;
1207}
1208
1209static struct gss_auth *
1210gss_create_hashed(const struct rpc_auth_create_args *args,
1211		  struct rpc_clnt *clnt)
1212{
1213	struct gss_auth *gss_auth;
1214	struct gss_auth *new;
1215
1216	gss_auth = gss_auth_find_or_add_hashed(args, clnt, NULL);
1217	if (gss_auth != NULL)
1218		goto out;
1219	new = gss_create_new(args, clnt);
1220	if (IS_ERR(new))
1221		return new;
1222	gss_auth = gss_auth_find_or_add_hashed(args, clnt, new);
1223	if (gss_auth != new)
1224		gss_destroy(&new->rpc_auth);
1225out:
1226	return gss_auth;
1227}
1228
1229static struct rpc_auth *
1230gss_create(const struct rpc_auth_create_args *args, struct rpc_clnt *clnt)
1231{
1232	struct gss_auth *gss_auth;
1233	struct rpc_xprt_switch *xps = rcu_access_pointer(clnt->cl_xpi.xpi_xpswitch);
1234
1235	while (clnt != clnt->cl_parent) {
1236		struct rpc_clnt *parent = clnt->cl_parent;
1237		/* Find the original parent for this transport */
1238		if (rcu_access_pointer(parent->cl_xpi.xpi_xpswitch) != xps)
1239			break;
1240		clnt = parent;
1241	}
1242
1243	gss_auth = gss_create_hashed(args, clnt);
1244	if (IS_ERR(gss_auth))
1245		return ERR_CAST(gss_auth);
1246	return &gss_auth->rpc_auth;
1247}
1248
1249static struct gss_cred *
1250gss_dup_cred(struct gss_auth *gss_auth, struct gss_cred *gss_cred)
1251{
1252	struct gss_cred *new;
1253
1254	/* Make a copy of the cred so that we can reference count it */
1255	new = kzalloc(sizeof(*gss_cred), GFP_KERNEL);
1256	if (new) {
1257		struct auth_cred acred = {
1258			.cred = gss_cred->gc_base.cr_cred,
1259		};
1260		struct gss_cl_ctx *ctx =
1261			rcu_dereference_protected(gss_cred->gc_ctx, 1);
1262
1263		rpcauth_init_cred(&new->gc_base, &acred,
1264				&gss_auth->rpc_auth,
1265				&gss_nullops);
1266		new->gc_base.cr_flags = 1UL << RPCAUTH_CRED_UPTODATE;
1267		new->gc_service = gss_cred->gc_service;
1268		new->gc_principal = gss_cred->gc_principal;
1269		kref_get(&gss_auth->kref);
1270		rcu_assign_pointer(new->gc_ctx, ctx);
1271		gss_get_ctx(ctx);
1272	}
1273	return new;
1274}
1275
1276/*
1277 * gss_send_destroy_context will cause the RPCSEC_GSS to send a NULL RPC call
1278 * to the server with the GSS control procedure field set to
1279 * RPC_GSS_PROC_DESTROY. This should normally cause the server to release
1280 * all RPCSEC_GSS state associated with that context.
1281 */
1282static void
1283gss_send_destroy_context(struct rpc_cred *cred)
1284{
1285	struct gss_cred *gss_cred = container_of(cred, struct gss_cred, gc_base);
1286	struct gss_auth *gss_auth = container_of(cred->cr_auth, struct gss_auth, rpc_auth);
1287	struct gss_cl_ctx *ctx = rcu_dereference_protected(gss_cred->gc_ctx, 1);
1288	struct gss_cred *new;
1289	struct rpc_task *task;
1290
1291	new = gss_dup_cred(gss_auth, gss_cred);
1292	if (new) {
1293		ctx->gc_proc = RPC_GSS_PROC_DESTROY;
1294
1295		trace_rpcgss_ctx_destroy(gss_cred);
1296		task = rpc_call_null(gss_auth->client, &new->gc_base,
1297				     RPC_TASK_ASYNC);
1298		if (!IS_ERR(task))
1299			rpc_put_task(task);
1300
1301		put_rpccred(&new->gc_base);
1302	}
1303}
1304
1305/* gss_destroy_cred (and gss_free_ctx) are used to clean up after failure
1306 * to create a new cred or context, so they check that things have been
1307 * allocated before freeing them. */
1308static void
1309gss_do_free_ctx(struct gss_cl_ctx *ctx)
1310{
1311	gss_delete_sec_context(&ctx->gc_gss_ctx);
1312	kfree(ctx->gc_wire_ctx.data);
1313	kfree(ctx->gc_acceptor.data);
1314	kfree(ctx);
1315}
1316
1317static void
1318gss_free_ctx_callback(struct rcu_head *head)
1319{
1320	struct gss_cl_ctx *ctx = container_of(head, struct gss_cl_ctx, gc_rcu);
1321	gss_do_free_ctx(ctx);
1322}
1323
1324static void
1325gss_free_ctx(struct gss_cl_ctx *ctx)
1326{
1327	call_rcu(&ctx->gc_rcu, gss_free_ctx_callback);
1328}
1329
1330static void
1331gss_free_cred(struct gss_cred *gss_cred)
1332{
1333	kfree(gss_cred);
1334}
1335
1336static void
1337gss_free_cred_callback(struct rcu_head *head)
1338{
1339	struct gss_cred *gss_cred = container_of(head, struct gss_cred, gc_base.cr_rcu);
1340	gss_free_cred(gss_cred);
1341}
1342
1343static void
1344gss_destroy_nullcred(struct rpc_cred *cred)
1345{
1346	struct gss_cred *gss_cred = container_of(cred, struct gss_cred, gc_base);
1347	struct gss_auth *gss_auth = container_of(cred->cr_auth, struct gss_auth, rpc_auth);
1348	struct gss_cl_ctx *ctx = rcu_dereference_protected(gss_cred->gc_ctx, 1);
1349
1350	RCU_INIT_POINTER(gss_cred->gc_ctx, NULL);
1351	put_cred(cred->cr_cred);
1352	call_rcu(&cred->cr_rcu, gss_free_cred_callback);
1353	if (ctx)
1354		gss_put_ctx(ctx);
1355	gss_put_auth(gss_auth);
1356}
1357
1358static void
1359gss_destroy_cred(struct rpc_cred *cred)
1360{
1361	if (test_and_clear_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags) != 0)
1362		gss_send_destroy_context(cred);
1363	gss_destroy_nullcred(cred);
1364}
1365
1366static int
1367gss_hash_cred(struct auth_cred *acred, unsigned int hashbits)
1368{
1369	return hash_64(from_kuid(&init_user_ns, acred->cred->fsuid), hashbits);
1370}
1371
1372/*
1373 * Lookup RPCSEC_GSS cred for the current process
1374 */
1375static struct rpc_cred *gss_lookup_cred(struct rpc_auth *auth,
1376					struct auth_cred *acred, int flags)
1377{
1378	return rpcauth_lookup_credcache(auth, acred, flags,
1379					rpc_task_gfp_mask());
1380}
1381
1382static struct rpc_cred *
1383gss_create_cred(struct rpc_auth *auth, struct auth_cred *acred, int flags, gfp_t gfp)
1384{
1385	struct gss_auth *gss_auth = container_of(auth, struct gss_auth, rpc_auth);
1386	struct gss_cred	*cred = NULL;
1387	int err = -ENOMEM;
1388
1389	if (!(cred = kzalloc(sizeof(*cred), gfp)))
1390		goto out_err;
1391
1392	rpcauth_init_cred(&cred->gc_base, acred, auth, &gss_credops);
1393	/*
1394	 * Note: in order to force a call to call_refresh(), we deliberately
1395	 * fail to flag the credential as RPCAUTH_CRED_UPTODATE.
1396	 */
1397	cred->gc_base.cr_flags = 1UL << RPCAUTH_CRED_NEW;
1398	cred->gc_service = gss_auth->service;
1399	cred->gc_principal = acred->principal;
1400	kref_get(&gss_auth->kref);
1401	return &cred->gc_base;
1402
1403out_err:
1404	return ERR_PTR(err);
1405}
1406
1407static int
1408gss_cred_init(struct rpc_auth *auth, struct rpc_cred *cred)
1409{
1410	struct gss_auth *gss_auth = container_of(auth, struct gss_auth, rpc_auth);
1411	struct gss_cred *gss_cred = container_of(cred,struct gss_cred, gc_base);
1412	int err;
1413
1414	do {
1415		err = gss_create_upcall(gss_auth, gss_cred);
1416	} while (err == -EAGAIN);
1417	return err;
1418}
1419
1420static char *
1421gss_stringify_acceptor(struct rpc_cred *cred)
1422{
1423	char *string = NULL;
1424	struct gss_cred *gss_cred = container_of(cred, struct gss_cred, gc_base);
1425	struct gss_cl_ctx *ctx;
1426	unsigned int len;
1427	struct xdr_netobj *acceptor;
1428
1429	rcu_read_lock();
1430	ctx = rcu_dereference(gss_cred->gc_ctx);
1431	if (!ctx)
1432		goto out;
1433
1434	len = ctx->gc_acceptor.len;
1435	rcu_read_unlock();
1436
1437	/* no point if there's no string */
1438	if (!len)
1439		return NULL;
1440realloc:
1441	string = kmalloc(len + 1, GFP_KERNEL);
1442	if (!string)
1443		return NULL;
1444
1445	rcu_read_lock();
1446	ctx = rcu_dereference(gss_cred->gc_ctx);
1447
1448	/* did the ctx disappear or was it replaced by one with no acceptor? */
1449	if (!ctx || !ctx->gc_acceptor.len) {
1450		kfree(string);
1451		string = NULL;
1452		goto out;
1453	}
1454
1455	acceptor = &ctx->gc_acceptor;
1456
1457	/*
1458	 * Did we find a new acceptor that's longer than the original? Allocate
1459	 * a longer buffer and try again.
1460	 */
1461	if (len < acceptor->len) {
1462		len = acceptor->len;
1463		rcu_read_unlock();
1464		kfree(string);
1465		goto realloc;
1466	}
1467
1468	memcpy(string, acceptor->data, acceptor->len);
1469	string[acceptor->len] = '\0';
1470out:
1471	rcu_read_unlock();
1472	return string;
1473}
1474
1475/*
1476 * Returns -EACCES if GSS context is NULL or will expire within the
1477 * timeout (miliseconds)
1478 */
1479static int
1480gss_key_timeout(struct rpc_cred *rc)
1481{
1482	struct gss_cred *gss_cred = container_of(rc, struct gss_cred, gc_base);
1483	struct gss_cl_ctx *ctx;
1484	unsigned long timeout = jiffies + (gss_key_expire_timeo * HZ);
1485	int ret = 0;
1486
1487	rcu_read_lock();
1488	ctx = rcu_dereference(gss_cred->gc_ctx);
1489	if (!ctx || time_after(timeout, ctx->gc_expiry))
1490		ret = -EACCES;
1491	rcu_read_unlock();
1492
1493	return ret;
1494}
1495
1496static int
1497gss_match(struct auth_cred *acred, struct rpc_cred *rc, int flags)
1498{
1499	struct gss_cred *gss_cred = container_of(rc, struct gss_cred, gc_base);
1500	struct gss_cl_ctx *ctx;
1501	int ret;
1502
1503	if (test_bit(RPCAUTH_CRED_NEW, &rc->cr_flags))
1504		goto out;
1505	/* Don't match with creds that have expired. */
1506	rcu_read_lock();
1507	ctx = rcu_dereference(gss_cred->gc_ctx);
1508	if (!ctx || time_after(jiffies, ctx->gc_expiry)) {
1509		rcu_read_unlock();
1510		return 0;
1511	}
1512	rcu_read_unlock();
1513	if (!test_bit(RPCAUTH_CRED_UPTODATE, &rc->cr_flags))
1514		return 0;
1515out:
1516	if (acred->principal != NULL) {
1517		if (gss_cred->gc_principal == NULL)
1518			return 0;
1519		ret = strcmp(acred->principal, gss_cred->gc_principal) == 0;
1520	} else {
1521		if (gss_cred->gc_principal != NULL)
1522			return 0;
1523		ret = uid_eq(rc->cr_cred->fsuid, acred->cred->fsuid);
1524	}
1525	return ret;
1526}
1527
1528/*
1529 * Marshal credentials.
1530 *
1531 * The expensive part is computing the verifier. We can't cache a
1532 * pre-computed version of the verifier because the seqno, which
1533 * is different every time, is included in the MIC.
1534 */
1535static int gss_marshal(struct rpc_task *task, struct xdr_stream *xdr)
1536{
1537	struct rpc_rqst *req = task->tk_rqstp;
1538	struct rpc_cred *cred = req->rq_cred;
1539	struct gss_cred	*gss_cred = container_of(cred, struct gss_cred,
1540						 gc_base);
1541	struct gss_cl_ctx	*ctx = gss_cred_get_ctx(cred);
1542	__be32		*p, *cred_len;
1543	u32             maj_stat = 0;
1544	struct xdr_netobj mic;
1545	struct kvec	iov;
1546	struct xdr_buf	verf_buf;
1547	int status;
1548
1549	/* Credential */
1550
1551	p = xdr_reserve_space(xdr, 7 * sizeof(*p) +
1552			      ctx->gc_wire_ctx.len);
1553	if (!p)
1554		goto marshal_failed;
1555	*p++ = rpc_auth_gss;
1556	cred_len = p++;
1557
1558	spin_lock(&ctx->gc_seq_lock);
1559	req->rq_seqno = (ctx->gc_seq < MAXSEQ) ? ctx->gc_seq++ : MAXSEQ;
1560	spin_unlock(&ctx->gc_seq_lock);
1561	if (req->rq_seqno == MAXSEQ)
1562		goto expired;
1563	trace_rpcgss_seqno(task);
1564
1565	*p++ = cpu_to_be32(RPC_GSS_VERSION);
1566	*p++ = cpu_to_be32(ctx->gc_proc);
1567	*p++ = cpu_to_be32(req->rq_seqno);
1568	*p++ = cpu_to_be32(gss_cred->gc_service);
1569	p = xdr_encode_netobj(p, &ctx->gc_wire_ctx);
1570	*cred_len = cpu_to_be32((p - (cred_len + 1)) << 2);
1571
1572	/* Verifier */
1573
1574	/* We compute the checksum for the verifier over the xdr-encoded bytes
1575	 * starting with the xid and ending at the end of the credential: */
1576	iov.iov_base = req->rq_snd_buf.head[0].iov_base;
1577	iov.iov_len = (u8 *)p - (u8 *)iov.iov_base;
1578	xdr_buf_from_iov(&iov, &verf_buf);
1579
1580	p = xdr_reserve_space(xdr, sizeof(*p));
1581	if (!p)
1582		goto marshal_failed;
1583	*p++ = rpc_auth_gss;
1584	mic.data = (u8 *)(p + 1);
1585	maj_stat = gss_get_mic(ctx->gc_gss_ctx, &verf_buf, &mic);
1586	if (maj_stat == GSS_S_CONTEXT_EXPIRED)
1587		goto expired;
1588	else if (maj_stat != 0)
1589		goto bad_mic;
1590	if (xdr_stream_encode_opaque_inline(xdr, (void **)&p, mic.len) < 0)
1591		goto marshal_failed;
1592	status = 0;
1593out:
1594	gss_put_ctx(ctx);
1595	return status;
1596expired:
1597	clear_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
1598	status = -EKEYEXPIRED;
1599	goto out;
1600marshal_failed:
1601	status = -EMSGSIZE;
1602	goto out;
1603bad_mic:
1604	trace_rpcgss_get_mic(task, maj_stat);
1605	status = -EIO;
1606	goto out;
1607}
1608
1609static int gss_renew_cred(struct rpc_task *task)
1610{
1611	struct rpc_cred *oldcred = task->tk_rqstp->rq_cred;
1612	struct gss_cred *gss_cred = container_of(oldcred,
1613						 struct gss_cred,
1614						 gc_base);
1615	struct rpc_auth *auth = oldcred->cr_auth;
1616	struct auth_cred acred = {
1617		.cred = oldcred->cr_cred,
1618		.principal = gss_cred->gc_principal,
1619	};
1620	struct rpc_cred *new;
1621
1622	new = gss_lookup_cred(auth, &acred, RPCAUTH_LOOKUP_NEW);
1623	if (IS_ERR(new))
1624		return PTR_ERR(new);
1625
1626	task->tk_rqstp->rq_cred = new;
1627	put_rpccred(oldcred);
1628	return 0;
1629}
1630
1631static int gss_cred_is_negative_entry(struct rpc_cred *cred)
1632{
1633	if (test_bit(RPCAUTH_CRED_NEGATIVE, &cred->cr_flags)) {
1634		unsigned long now = jiffies;
1635		unsigned long begin, expire;
1636		struct gss_cred *gss_cred;
1637
1638		gss_cred = container_of(cred, struct gss_cred, gc_base);
1639		begin = gss_cred->gc_upcall_timestamp;
1640		expire = begin + gss_expired_cred_retry_delay * HZ;
1641
1642		if (time_in_range_open(now, begin, expire))
1643			return 1;
1644	}
1645	return 0;
1646}
1647
1648/*
1649* Refresh credentials. XXX - finish
1650*/
1651static int
1652gss_refresh(struct rpc_task *task)
1653{
1654	struct rpc_cred *cred = task->tk_rqstp->rq_cred;
1655	int ret = 0;
1656
1657	if (gss_cred_is_negative_entry(cred))
1658		return -EKEYEXPIRED;
1659
1660	if (!test_bit(RPCAUTH_CRED_NEW, &cred->cr_flags) &&
1661			!test_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags)) {
1662		ret = gss_renew_cred(task);
1663		if (ret < 0)
1664			goto out;
1665		cred = task->tk_rqstp->rq_cred;
1666	}
1667
1668	if (test_bit(RPCAUTH_CRED_NEW, &cred->cr_flags))
1669		ret = gss_refresh_upcall(task);
1670out:
1671	return ret;
1672}
1673
1674/* Dummy refresh routine: used only when destroying the context */
1675static int
1676gss_refresh_null(struct rpc_task *task)
1677{
1678	return 0;
1679}
1680
1681static int
1682gss_validate(struct rpc_task *task, struct xdr_stream *xdr)
1683{
1684	struct rpc_cred *cred = task->tk_rqstp->rq_cred;
1685	struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred);
1686	__be32		*p, *seq = NULL;
1687	struct kvec	iov;
1688	struct xdr_buf	verf_buf;
1689	struct xdr_netobj mic;
1690	u32		len, maj_stat;
1691	int		status;
1692
1693	p = xdr_inline_decode(xdr, 2 * sizeof(*p));
1694	if (!p)
1695		goto validate_failed;
1696	if (*p++ != rpc_auth_gss)
1697		goto validate_failed;
1698	len = be32_to_cpup(p);
1699	if (len > RPC_MAX_AUTH_SIZE)
1700		goto validate_failed;
1701	p = xdr_inline_decode(xdr, len);
1702	if (!p)
1703		goto validate_failed;
1704
1705	seq = kmalloc(4, GFP_KERNEL);
1706	if (!seq)
1707		goto validate_failed;
1708	*seq = cpu_to_be32(task->tk_rqstp->rq_seqno);
1709	iov.iov_base = seq;
1710	iov.iov_len = 4;
1711	xdr_buf_from_iov(&iov, &verf_buf);
1712	mic.data = (u8 *)p;
1713	mic.len = len;
1714	maj_stat = gss_verify_mic(ctx->gc_gss_ctx, &verf_buf, &mic);
1715	if (maj_stat == GSS_S_CONTEXT_EXPIRED)
1716		clear_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
1717	if (maj_stat)
1718		goto bad_mic;
1719
1720	/* We leave it to unwrap to calculate au_rslack. For now we just
1721	 * calculate the length of the verifier: */
1722	if (test_bit(RPCAUTH_AUTH_UPDATE_SLACK, &cred->cr_auth->au_flags))
1723		cred->cr_auth->au_verfsize = XDR_QUADLEN(len) + 2;
1724	status = 0;
1725out:
1726	gss_put_ctx(ctx);
1727	kfree(seq);
1728	return status;
1729
1730validate_failed:
1731	status = -EIO;
1732	goto out;
1733bad_mic:
1734	trace_rpcgss_verify_mic(task, maj_stat);
1735	status = -EACCES;
1736	goto out;
1737}
1738
1739static noinline_for_stack int
1740gss_wrap_req_integ(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
1741		   struct rpc_task *task, struct xdr_stream *xdr)
1742{
1743	struct rpc_rqst *rqstp = task->tk_rqstp;
1744	struct xdr_buf integ_buf, *snd_buf = &rqstp->rq_snd_buf;
1745	struct xdr_netobj mic;
1746	__be32 *p, *integ_len;
1747	u32 offset, maj_stat;
1748
1749	p = xdr_reserve_space(xdr, 2 * sizeof(*p));
1750	if (!p)
1751		goto wrap_failed;
1752	integ_len = p++;
1753	*p = cpu_to_be32(rqstp->rq_seqno);
1754
1755	if (rpcauth_wrap_req_encode(task, xdr))
1756		goto wrap_failed;
1757
1758	offset = (u8 *)p - (u8 *)snd_buf->head[0].iov_base;
1759	if (xdr_buf_subsegment(snd_buf, &integ_buf,
1760				offset, snd_buf->len - offset))
1761		goto wrap_failed;
1762	*integ_len = cpu_to_be32(integ_buf.len);
1763
1764	p = xdr_reserve_space(xdr, 0);
1765	if (!p)
1766		goto wrap_failed;
1767	mic.data = (u8 *)(p + 1);
1768	maj_stat = gss_get_mic(ctx->gc_gss_ctx, &integ_buf, &mic);
1769	if (maj_stat == GSS_S_CONTEXT_EXPIRED)
1770		clear_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
1771	else if (maj_stat)
1772		goto bad_mic;
1773	/* Check that the trailing MIC fit in the buffer, after the fact */
1774	if (xdr_stream_encode_opaque_inline(xdr, (void **)&p, mic.len) < 0)
1775		goto wrap_failed;
1776	return 0;
1777wrap_failed:
1778	return -EMSGSIZE;
1779bad_mic:
1780	trace_rpcgss_get_mic(task, maj_stat);
1781	return -EIO;
1782}
1783
1784static void
1785priv_release_snd_buf(struct rpc_rqst *rqstp)
1786{
1787	int i;
1788
1789	for (i=0; i < rqstp->rq_enc_pages_num; i++)
1790		__free_page(rqstp->rq_enc_pages[i]);
1791	kfree(rqstp->rq_enc_pages);
1792	rqstp->rq_release_snd_buf = NULL;
1793}
1794
1795static int
1796alloc_enc_pages(struct rpc_rqst *rqstp)
1797{
1798	struct xdr_buf *snd_buf = &rqstp->rq_snd_buf;
1799	int first, last, i;
1800
1801	if (rqstp->rq_release_snd_buf)
1802		rqstp->rq_release_snd_buf(rqstp);
1803
1804	if (snd_buf->page_len == 0) {
1805		rqstp->rq_enc_pages_num = 0;
1806		return 0;
1807	}
1808
1809	first = snd_buf->page_base >> PAGE_SHIFT;
1810	last = (snd_buf->page_base + snd_buf->page_len - 1) >> PAGE_SHIFT;
1811	rqstp->rq_enc_pages_num = last - first + 1 + 1;
1812	rqstp->rq_enc_pages
1813		= kmalloc_array(rqstp->rq_enc_pages_num,
1814				sizeof(struct page *),
1815				GFP_KERNEL);
1816	if (!rqstp->rq_enc_pages)
1817		goto out;
1818	for (i=0; i < rqstp->rq_enc_pages_num; i++) {
1819		rqstp->rq_enc_pages[i] = alloc_page(GFP_KERNEL);
1820		if (rqstp->rq_enc_pages[i] == NULL)
1821			goto out_free;
1822	}
1823	rqstp->rq_release_snd_buf = priv_release_snd_buf;
1824	return 0;
1825out_free:
1826	rqstp->rq_enc_pages_num = i;
1827	priv_release_snd_buf(rqstp);
1828out:
1829	return -EAGAIN;
1830}
1831
1832static noinline_for_stack int
1833gss_wrap_req_priv(struct rpc_cred *cred, struct gss_cl_ctx *ctx,
1834		  struct rpc_task *task, struct xdr_stream *xdr)
1835{
1836	struct rpc_rqst *rqstp = task->tk_rqstp;
1837	struct xdr_buf	*snd_buf = &rqstp->rq_snd_buf;
1838	u32		pad, offset, maj_stat;
1839	int		status;
1840	__be32		*p, *opaque_len;
1841	struct page	**inpages;
1842	int		first;
1843	struct kvec	*iov;
1844
1845	status = -EIO;
1846	p = xdr_reserve_space(xdr, 2 * sizeof(*p));
1847	if (!p)
1848		goto wrap_failed;
1849	opaque_len = p++;
1850	*p = cpu_to_be32(rqstp->rq_seqno);
1851
1852	if (rpcauth_wrap_req_encode(task, xdr))
1853		goto wrap_failed;
1854
1855	status = alloc_enc_pages(rqstp);
1856	if (unlikely(status))
1857		goto wrap_failed;
1858	first = snd_buf->page_base >> PAGE_SHIFT;
1859	inpages = snd_buf->pages + first;
1860	snd_buf->pages = rqstp->rq_enc_pages;
1861	snd_buf->page_base -= first << PAGE_SHIFT;
1862	/*
1863	 * Move the tail into its own page, in case gss_wrap needs
1864	 * more space in the head when wrapping.
1865	 *
1866	 * Still... Why can't gss_wrap just slide the tail down?
1867	 */
1868	if (snd_buf->page_len || snd_buf->tail[0].iov_len) {
1869		char *tmp;
1870
1871		tmp = page_address(rqstp->rq_enc_pages[rqstp->rq_enc_pages_num - 1]);
1872		memcpy(tmp, snd_buf->tail[0].iov_base, snd_buf->tail[0].iov_len);
1873		snd_buf->tail[0].iov_base = tmp;
1874	}
1875	offset = (u8 *)p - (u8 *)snd_buf->head[0].iov_base;
1876	maj_stat = gss_wrap(ctx->gc_gss_ctx, offset, snd_buf, inpages);
1877	/* slack space should prevent this ever happening: */
1878	if (unlikely(snd_buf->len > snd_buf->buflen))
1879		goto wrap_failed;
1880	/* We're assuming that when GSS_S_CONTEXT_EXPIRED, the encryption was
1881	 * done anyway, so it's safe to put the request on the wire: */
1882	if (maj_stat == GSS_S_CONTEXT_EXPIRED)
1883		clear_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
1884	else if (maj_stat)
1885		goto bad_wrap;
1886
1887	*opaque_len = cpu_to_be32(snd_buf->len - offset);
1888	/* guess whether the pad goes into the head or the tail: */
1889	if (snd_buf->page_len || snd_buf->tail[0].iov_len)
1890		iov = snd_buf->tail;
1891	else
1892		iov = snd_buf->head;
1893	p = iov->iov_base + iov->iov_len;
1894	pad = xdr_pad_size(snd_buf->len - offset);
1895	memset(p, 0, pad);
1896	iov->iov_len += pad;
1897	snd_buf->len += pad;
1898
1899	return 0;
1900wrap_failed:
1901	return status;
1902bad_wrap:
1903	trace_rpcgss_wrap(task, maj_stat);
1904	return -EIO;
1905}
1906
1907static int gss_wrap_req(struct rpc_task *task, struct xdr_stream *xdr)
1908{
1909	struct rpc_cred *cred = task->tk_rqstp->rq_cred;
1910	struct gss_cred	*gss_cred = container_of(cred, struct gss_cred,
1911			gc_base);
1912	struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred);
1913	int status;
1914
1915	status = -EIO;
1916	if (ctx->gc_proc != RPC_GSS_PROC_DATA) {
1917		/* The spec seems a little ambiguous here, but I think that not
1918		 * wrapping context destruction requests makes the most sense.
1919		 */
1920		status = rpcauth_wrap_req_encode(task, xdr);
1921		goto out;
1922	}
1923	switch (gss_cred->gc_service) {
1924	case RPC_GSS_SVC_NONE:
1925		status = rpcauth_wrap_req_encode(task, xdr);
1926		break;
1927	case RPC_GSS_SVC_INTEGRITY:
1928		status = gss_wrap_req_integ(cred, ctx, task, xdr);
1929		break;
1930	case RPC_GSS_SVC_PRIVACY:
1931		status = gss_wrap_req_priv(cred, ctx, task, xdr);
1932		break;
1933	default:
1934		status = -EIO;
1935	}
1936out:
1937	gss_put_ctx(ctx);
1938	return status;
1939}
1940
1941/**
1942 * gss_update_rslack - Possibly update RPC receive buffer size estimates
1943 * @task: rpc_task for incoming RPC Reply being unwrapped
1944 * @cred: controlling rpc_cred for @task
1945 * @before: XDR words needed before each RPC Reply message
1946 * @after: XDR words needed following each RPC Reply message
1947 *
1948 */
1949static void gss_update_rslack(struct rpc_task *task, struct rpc_cred *cred,
1950			      unsigned int before, unsigned int after)
1951{
1952	struct rpc_auth *auth = cred->cr_auth;
1953
1954	if (test_and_clear_bit(RPCAUTH_AUTH_UPDATE_SLACK, &auth->au_flags)) {
1955		auth->au_ralign = auth->au_verfsize + before;
1956		auth->au_rslack = auth->au_verfsize + after;
1957		trace_rpcgss_update_slack(task, auth);
1958	}
1959}
1960
1961static int
1962gss_unwrap_resp_auth(struct rpc_task *task, struct rpc_cred *cred)
1963{
1964	gss_update_rslack(task, cred, 0, 0);
1965	return 0;
1966}
1967
1968/*
1969 * RFC 2203, Section 5.3.2.2
1970 *
1971 *	struct rpc_gss_integ_data {
1972 *		opaque databody_integ<>;
1973 *		opaque checksum<>;
1974 *	};
1975 *
1976 *	struct rpc_gss_data_t {
1977 *		unsigned int seq_num;
1978 *		proc_req_arg_t arg;
1979 *	};
1980 */
1981static noinline_for_stack int
1982gss_unwrap_resp_integ(struct rpc_task *task, struct rpc_cred *cred,
1983		      struct gss_cl_ctx *ctx, struct rpc_rqst *rqstp,
1984		      struct xdr_stream *xdr)
1985{
1986	struct xdr_buf gss_data, *rcv_buf = &rqstp->rq_rcv_buf;
1987	u32 len, offset, seqno, maj_stat;
1988	struct xdr_netobj mic;
1989	int ret;
1990
1991	ret = -EIO;
1992	mic.data = NULL;
1993
1994	/* opaque databody_integ<>; */
1995	if (xdr_stream_decode_u32(xdr, &len))
1996		goto unwrap_failed;
1997	if (len & 3)
1998		goto unwrap_failed;
1999	offset = rcv_buf->len - xdr_stream_remaining(xdr);
2000	if (xdr_stream_decode_u32(xdr, &seqno))
2001		goto unwrap_failed;
2002	if (seqno != rqstp->rq_seqno)
2003		goto bad_seqno;
2004	if (xdr_buf_subsegment(rcv_buf, &gss_data, offset, len))
2005		goto unwrap_failed;
2006
2007	/*
2008	 * The xdr_stream now points to the beginning of the
2009	 * upper layer payload, to be passed below to
2010	 * rpcauth_unwrap_resp_decode(). The checksum, which
2011	 * follows the upper layer payload in @rcv_buf, is
2012	 * located and parsed without updating the xdr_stream.
2013	 */
2014
2015	/* opaque checksum<>; */
2016	offset += len;
2017	if (xdr_decode_word(rcv_buf, offset, &len))
2018		goto unwrap_failed;
2019	offset += sizeof(__be32);
2020	if (offset + len > rcv_buf->len)
2021		goto unwrap_failed;
2022	mic.len = len;
2023	mic.data = kmalloc(len, GFP_KERNEL);
2024	if (ZERO_OR_NULL_PTR(mic.data))
2025		goto unwrap_failed;
2026	if (read_bytes_from_xdr_buf(rcv_buf, offset, mic.data, mic.len))
2027		goto unwrap_failed;
2028
2029	maj_stat = gss_verify_mic(ctx->gc_gss_ctx, &gss_data, &mic);
2030	if (maj_stat == GSS_S_CONTEXT_EXPIRED)
2031		clear_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
2032	if (maj_stat != GSS_S_COMPLETE)
2033		goto bad_mic;
2034
2035	gss_update_rslack(task, cred, 2, 2 + 1 + XDR_QUADLEN(mic.len));
2036	ret = 0;
2037
2038out:
2039	kfree(mic.data);
2040	return ret;
2041
2042unwrap_failed:
2043	trace_rpcgss_unwrap_failed(task);
2044	goto out;
2045bad_seqno:
2046	trace_rpcgss_bad_seqno(task, rqstp->rq_seqno, seqno);
2047	goto out;
2048bad_mic:
2049	trace_rpcgss_verify_mic(task, maj_stat);
2050	goto out;
2051}
2052
2053static noinline_for_stack int
2054gss_unwrap_resp_priv(struct rpc_task *task, struct rpc_cred *cred,
2055		     struct gss_cl_ctx *ctx, struct rpc_rqst *rqstp,
2056		     struct xdr_stream *xdr)
2057{
2058	struct xdr_buf *rcv_buf = &rqstp->rq_rcv_buf;
2059	struct kvec *head = rqstp->rq_rcv_buf.head;
2060	u32 offset, opaque_len, maj_stat;
2061	__be32 *p;
2062
2063	p = xdr_inline_decode(xdr, 2 * sizeof(*p));
2064	if (unlikely(!p))
2065		goto unwrap_failed;
2066	opaque_len = be32_to_cpup(p++);
2067	offset = (u8 *)(p) - (u8 *)head->iov_base;
2068	if (offset + opaque_len > rcv_buf->len)
2069		goto unwrap_failed;
2070
2071	maj_stat = gss_unwrap(ctx->gc_gss_ctx, offset,
2072			      offset + opaque_len, rcv_buf);
2073	if (maj_stat == GSS_S_CONTEXT_EXPIRED)
2074		clear_bit(RPCAUTH_CRED_UPTODATE, &cred->cr_flags);
2075	if (maj_stat != GSS_S_COMPLETE)
2076		goto bad_unwrap;
2077	/* gss_unwrap decrypted the sequence number */
2078	if (be32_to_cpup(p++) != rqstp->rq_seqno)
2079		goto bad_seqno;
2080
2081	/* gss_unwrap redacts the opaque blob from the head iovec.
2082	 * rcv_buf has changed, thus the stream needs to be reset.
2083	 */
2084	xdr_init_decode(xdr, rcv_buf, p, rqstp);
2085
2086	gss_update_rslack(task, cred, 2 + ctx->gc_gss_ctx->align,
2087			  2 + ctx->gc_gss_ctx->slack);
2088
2089	return 0;
2090unwrap_failed:
2091	trace_rpcgss_unwrap_failed(task);
2092	return -EIO;
2093bad_seqno:
2094	trace_rpcgss_bad_seqno(task, rqstp->rq_seqno, be32_to_cpup(--p));
2095	return -EIO;
2096bad_unwrap:
2097	trace_rpcgss_unwrap(task, maj_stat);
2098	return -EIO;
2099}
2100
2101static bool
2102gss_seq_is_newer(u32 new, u32 old)
2103{
2104	return (s32)(new - old) > 0;
2105}
2106
2107static bool
2108gss_xmit_need_reencode(struct rpc_task *task)
2109{
2110	struct rpc_rqst *req = task->tk_rqstp;
2111	struct rpc_cred *cred = req->rq_cred;
2112	struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred);
2113	u32 win, seq_xmit = 0;
2114	bool ret = true;
2115
2116	if (!ctx)
2117		goto out;
2118
2119	if (gss_seq_is_newer(req->rq_seqno, READ_ONCE(ctx->gc_seq)))
2120		goto out_ctx;
2121
2122	seq_xmit = READ_ONCE(ctx->gc_seq_xmit);
2123	while (gss_seq_is_newer(req->rq_seqno, seq_xmit)) {
2124		u32 tmp = seq_xmit;
2125
2126		seq_xmit = cmpxchg(&ctx->gc_seq_xmit, tmp, req->rq_seqno);
2127		if (seq_xmit == tmp) {
2128			ret = false;
2129			goto out_ctx;
2130		}
2131	}
2132
2133	win = ctx->gc_win;
2134	if (win > 0)
2135		ret = !gss_seq_is_newer(req->rq_seqno, seq_xmit - win);
2136
2137out_ctx:
2138	gss_put_ctx(ctx);
2139out:
2140	trace_rpcgss_need_reencode(task, seq_xmit, ret);
2141	return ret;
2142}
2143
2144static int
2145gss_unwrap_resp(struct rpc_task *task, struct xdr_stream *xdr)
2146{
2147	struct rpc_rqst *rqstp = task->tk_rqstp;
2148	struct rpc_cred *cred = rqstp->rq_cred;
2149	struct gss_cred *gss_cred = container_of(cred, struct gss_cred,
2150			gc_base);
2151	struct gss_cl_ctx *ctx = gss_cred_get_ctx(cred);
2152	int status = -EIO;
2153
2154	if (ctx->gc_proc != RPC_GSS_PROC_DATA)
2155		goto out_decode;
2156	switch (gss_cred->gc_service) {
2157	case RPC_GSS_SVC_NONE:
2158		status = gss_unwrap_resp_auth(task, cred);
2159		break;
2160	case RPC_GSS_SVC_INTEGRITY:
2161		status = gss_unwrap_resp_integ(task, cred, ctx, rqstp, xdr);
2162		break;
2163	case RPC_GSS_SVC_PRIVACY:
2164		status = gss_unwrap_resp_priv(task, cred, ctx, rqstp, xdr);
2165		break;
2166	}
2167	if (status)
2168		goto out;
2169
2170out_decode:
2171	status = rpcauth_unwrap_resp_decode(task, xdr);
2172out:
2173	gss_put_ctx(ctx);
2174	return status;
2175}
2176
2177static const struct rpc_authops authgss_ops = {
2178	.owner		= THIS_MODULE,
2179	.au_flavor	= RPC_AUTH_GSS,
2180	.au_name	= "RPCSEC_GSS",
2181	.create		= gss_create,
2182	.destroy	= gss_destroy,
2183	.hash_cred	= gss_hash_cred,
2184	.lookup_cred	= gss_lookup_cred,
2185	.crcreate	= gss_create_cred,
2186	.info2flavor	= gss_mech_info2flavor,
2187	.flavor2info	= gss_mech_flavor2info,
2188};
2189
2190static const struct rpc_credops gss_credops = {
2191	.cr_name		= "AUTH_GSS",
2192	.crdestroy		= gss_destroy_cred,
2193	.cr_init		= gss_cred_init,
2194	.crmatch		= gss_match,
2195	.crmarshal		= gss_marshal,
2196	.crrefresh		= gss_refresh,
2197	.crvalidate		= gss_validate,
2198	.crwrap_req		= gss_wrap_req,
2199	.crunwrap_resp		= gss_unwrap_resp,
2200	.crkey_timeout		= gss_key_timeout,
2201	.crstringify_acceptor	= gss_stringify_acceptor,
2202	.crneed_reencode	= gss_xmit_need_reencode,
2203};
2204
2205static const struct rpc_credops gss_nullops = {
2206	.cr_name		= "AUTH_GSS",
2207	.crdestroy		= gss_destroy_nullcred,
2208	.crmatch		= gss_match,
2209	.crmarshal		= gss_marshal,
2210	.crrefresh		= gss_refresh_null,
2211	.crvalidate		= gss_validate,
2212	.crwrap_req		= gss_wrap_req,
2213	.crunwrap_resp		= gss_unwrap_resp,
2214	.crstringify_acceptor	= gss_stringify_acceptor,
2215};
2216
2217static const struct rpc_pipe_ops gss_upcall_ops_v0 = {
2218	.upcall		= gss_v0_upcall,
2219	.downcall	= gss_pipe_downcall,
2220	.destroy_msg	= gss_pipe_destroy_msg,
2221	.open_pipe	= gss_pipe_open_v0,
2222	.release_pipe	= gss_pipe_release,
2223};
2224
2225static const struct rpc_pipe_ops gss_upcall_ops_v1 = {
2226	.upcall		= gss_v1_upcall,
2227	.downcall	= gss_pipe_downcall,
2228	.destroy_msg	= gss_pipe_destroy_msg,
2229	.open_pipe	= gss_pipe_open_v1,
2230	.release_pipe	= gss_pipe_release,
2231};
2232
2233static __net_init int rpcsec_gss_init_net(struct net *net)
2234{
2235	return gss_svc_init_net(net);
2236}
2237
2238static __net_exit void rpcsec_gss_exit_net(struct net *net)
2239{
2240	gss_svc_shutdown_net(net);
2241}
2242
2243static struct pernet_operations rpcsec_gss_net_ops = {
2244	.init = rpcsec_gss_init_net,
2245	.exit = rpcsec_gss_exit_net,
2246};
2247
2248/*
2249 * Initialize RPCSEC_GSS module
2250 */
2251static int __init init_rpcsec_gss(void)
2252{
2253	int err = 0;
2254
2255	err = rpcauth_register(&authgss_ops);
2256	if (err)
2257		goto out;
2258	err = gss_svc_init();
2259	if (err)
2260		goto out_unregister;
2261	err = register_pernet_subsys(&rpcsec_gss_net_ops);
2262	if (err)
2263		goto out_svc_exit;
2264	rpc_init_wait_queue(&pipe_version_rpc_waitqueue, "gss pipe version");
2265	return 0;
2266out_svc_exit:
2267	gss_svc_shutdown();
2268out_unregister:
2269	rpcauth_unregister(&authgss_ops);
2270out:
2271	return err;
2272}
2273
2274static void __exit exit_rpcsec_gss(void)
2275{
2276	unregister_pernet_subsys(&rpcsec_gss_net_ops);
2277	gss_svc_shutdown();
2278	rpcauth_unregister(&authgss_ops);
2279	rcu_barrier(); /* Wait for completion of call_rcu()'s */
2280}
2281
2282MODULE_ALIAS("rpc-auth-6");
2283MODULE_DESCRIPTION("Sun RPC Kerberos RPCSEC_GSS client authentication");
2284MODULE_LICENSE("GPL");
2285module_param_named(expired_cred_retry_delay,
2286		   gss_expired_cred_retry_delay,
2287		   uint, 0644);
2288MODULE_PARM_DESC(expired_cred_retry_delay, "Timeout (in seconds) until "
2289		"the RPC engine retries an expired credential");
2290
2291module_param_named(key_expire_timeo,
2292		   gss_key_expire_timeo,
2293		   uint, 0644);
2294MODULE_PARM_DESC(key_expire_timeo, "Time (in seconds) at the end of a "
2295		"credential keys lifetime where the NFS layer cleans up "
2296		"prior to key expiration");
2297
2298module_init(init_rpcsec_gss)
2299module_exit(exit_rpcsec_gss)