nfs4state.c

   1/*
   2*  Copyright (c) 2001 The Regents of the University of Michigan.
   3*  All rights reserved.
   4*
   5*  Kendrick Smith <kmsmith@umich.edu>
   6*  Andy Adamson <kandros@umich.edu>
   7*
   8*  Redistribution and use in source and binary forms, with or without
   9*  modification, are permitted provided that the following conditions
  10*  are met:
  11*
  12*  1. Redistributions of source code must retain the above copyright
  13*     notice, this list of conditions and the following disclaimer.
  14*  2. Redistributions in binary form must reproduce the above copyright
  15*     notice, this list of conditions and the following disclaimer in the
  16*     documentation and/or other materials provided with the distribution.
  17*  3. Neither the name of the University nor the names of its
  18*     contributors may be used to endorse or promote products derived
  19*     from this software without specific prior written permission.
  20*
  21*  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22*  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23*  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24*  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25*  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26*  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27*  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28*  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29*  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30*  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31*  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32*
  33*/
  34
  35#include <linux/file.h>
  36#include <linux/fs.h>
  37#include <linux/slab.h>
  38#include <linux/namei.h>
  39#include <linux/swap.h>
  40#include <linux/pagemap.h>
  41#include <linux/ratelimit.h>
  42#include <linux/sunrpc/svcauth_gss.h>
  43#include <linux/sunrpc/addr.h>
  44#include <linux/jhash.h>
  45#include <linux/string_helpers.h>
  46#include <linux/fsnotify.h>
  47#include <linux/rhashtable.h>
  48#include <linux/nfs_ssc.h>
  49
  50#include "xdr4.h"
  51#include "xdr4cb.h"
  52#include "vfs.h"
  53#include "current_stateid.h"
  54
  55#include "netns.h"
  56#include "pnfs.h"
  57#include "filecache.h"
  58#include "trace.h"
  59
  60#define NFSDDBG_FACILITY                NFSDDBG_PROC
  61
  62#define all_ones {{ ~0, ~0}, ~0}
  63static const stateid_t one_stateid = {
  64	.si_generation = ~0,
  65	.si_opaque = all_ones,
  66};
  67static const stateid_t zero_stateid = {
  68	/* all fields zero */
  69};
  70static const stateid_t currentstateid = {
  71	.si_generation = 1,
  72};
  73static const stateid_t close_stateid = {
  74	.si_generation = 0xffffffffU,
  75};
  76
  77static u64 current_sessionid = 1;
  78
  79#define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  80#define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  81#define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  82#define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  83
  84/* forward declarations */
  85static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  86static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  87static void nfsd4_end_grace(struct nfsd_net *nn);
  88static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
  89static void nfsd4_file_hash_remove(struct nfs4_file *fi);
  90static void deleg_reaper(struct nfsd_net *nn);
  91
  92/* Locking: */
  93
  94/*
  95 * Currently used for the del_recall_lru and file hash table.  In an
  96 * effort to decrease the scope of the client_mutex, this spinlock may
  97 * eventually cover more:
  98 */
  99static DEFINE_SPINLOCK(state_lock);
 100
 101enum nfsd4_st_mutex_lock_subclass {
 102	OPEN_STATEID_MUTEX = 0,
 103	LOCK_STATEID_MUTEX = 1,
 104};
 105
 106/*
 107 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 108 * the refcount on the open stateid to drop.
 109 */
 110static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 111
 112/*
 113 * A waitqueue where a writer to clients/#/ctl destroying a client can
 114 * wait for cl_rpc_users to drop to 0 and then for the client to be
 115 * unhashed.
 116 */
 117static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 118
 119static struct kmem_cache *client_slab;
 120static struct kmem_cache *openowner_slab;
 121static struct kmem_cache *lockowner_slab;
 122static struct kmem_cache *file_slab;
 123static struct kmem_cache *stateid_slab;
 124static struct kmem_cache *deleg_slab;
 125static struct kmem_cache *odstate_slab;
 126
 127static void free_session(struct nfsd4_session *);
 128
 129static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 130static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 131static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops;
 132
 133static struct workqueue_struct *laundry_wq;
 134
 135int nfsd4_create_laundry_wq(void)
 136{
 137	int rc = 0;
 138
 139	laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
 140	if (laundry_wq == NULL)
 141		rc = -ENOMEM;
 142	return rc;
 143}
 144
 145void nfsd4_destroy_laundry_wq(void)
 146{
 147	destroy_workqueue(laundry_wq);
 148}
 149
 150static bool is_session_dead(struct nfsd4_session *ses)
 151{
 152	return ses->se_dead;
 153}
 154
 155static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 156{
 157	if (atomic_read(&ses->se_ref) > ref_held_by_me)
 158		return nfserr_jukebox;
 159	ses->se_dead = true;
 160	return nfs_ok;
 161}
 162
 163static bool is_client_expired(struct nfs4_client *clp)
 164{
 165	return clp->cl_time == 0;
 166}
 167
 168static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
 169					struct nfs4_client *clp)
 170{
 171	if (clp->cl_state != NFSD4_ACTIVE)
 172		atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
 173}
 174
 175static __be32 get_client_locked(struct nfs4_client *clp)
 176{
 177	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 178
 179	lockdep_assert_held(&nn->client_lock);
 180
 181	if (is_client_expired(clp))
 182		return nfserr_expired;
 183	atomic_inc(&clp->cl_rpc_users);
 184	nfsd4_dec_courtesy_client_count(nn, clp);
 185	clp->cl_state = NFSD4_ACTIVE;
 186	return nfs_ok;
 187}
 188
 189/* must be called under the client_lock */
 190static inline void
 191renew_client_locked(struct nfs4_client *clp)
 192{
 193	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 194
 195	if (is_client_expired(clp)) {
 196		WARN_ON(1);
 197		printk("%s: client (clientid %08x/%08x) already expired\n",
 198			__func__,
 199			clp->cl_clientid.cl_boot,
 200			clp->cl_clientid.cl_id);
 201		return;
 202	}
 203
 204	list_move_tail(&clp->cl_lru, &nn->client_lru);
 205	clp->cl_time = ktime_get_boottime_seconds();
 206	nfsd4_dec_courtesy_client_count(nn, clp);
 207	clp->cl_state = NFSD4_ACTIVE;
 208}
 209
 210static void put_client_renew_locked(struct nfs4_client *clp)
 211{
 212	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 213
 214	lockdep_assert_held(&nn->client_lock);
 215
 216	if (!atomic_dec_and_test(&clp->cl_rpc_users))
 217		return;
 218	if (!is_client_expired(clp))
 219		renew_client_locked(clp);
 220	else
 221		wake_up_all(&expiry_wq);
 222}
 223
 224static void put_client_renew(struct nfs4_client *clp)
 225{
 226	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 227
 228	if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 229		return;
 230	if (!is_client_expired(clp))
 231		renew_client_locked(clp);
 232	else
 233		wake_up_all(&expiry_wq);
 234	spin_unlock(&nn->client_lock);
 235}
 236
 237static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 238{
 239	__be32 status;
 240
 241	if (is_session_dead(ses))
 242		return nfserr_badsession;
 243	status = get_client_locked(ses->se_client);
 244	if (status)
 245		return status;
 246	atomic_inc(&ses->se_ref);
 247	return nfs_ok;
 248}
 249
 250static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 251{
 252	struct nfs4_client *clp = ses->se_client;
 253	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 254
 255	lockdep_assert_held(&nn->client_lock);
 256
 257	if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 258		free_session(ses);
 259	put_client_renew_locked(clp);
 260}
 261
 262static void nfsd4_put_session(struct nfsd4_session *ses)
 263{
 264	struct nfs4_client *clp = ses->se_client;
 265	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 266
 267	spin_lock(&nn->client_lock);
 268	nfsd4_put_session_locked(ses);
 269	spin_unlock(&nn->client_lock);
 270}
 271
 272static struct nfsd4_blocked_lock *
 273find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 274			struct nfsd_net *nn)
 275{
 276	struct nfsd4_blocked_lock *cur, *found = NULL;
 277
 278	spin_lock(&nn->blocked_locks_lock);
 279	list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 280		if (fh_match(fh, &cur->nbl_fh)) {
 281			list_del_init(&cur->nbl_list);
 282			WARN_ON(list_empty(&cur->nbl_lru));
 283			list_del_init(&cur->nbl_lru);
 284			found = cur;
 285			break;
 286		}
 287	}
 288	spin_unlock(&nn->blocked_locks_lock);
 289	if (found)
 290		locks_delete_block(&found->nbl_lock);
 291	return found;
 292}
 293
 294static struct nfsd4_blocked_lock *
 295find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 296			struct nfsd_net *nn)
 297{
 298	struct nfsd4_blocked_lock *nbl;
 299
 300	nbl = find_blocked_lock(lo, fh, nn);
 301	if (!nbl) {
 302		nbl = kmalloc(sizeof(*nbl), GFP_KERNEL);
 303		if (nbl) {
 304			INIT_LIST_HEAD(&nbl->nbl_list);
 305			INIT_LIST_HEAD(&nbl->nbl_lru);
 306			fh_copy_shallow(&nbl->nbl_fh, fh);
 307			locks_init_lock(&nbl->nbl_lock);
 308			kref_init(&nbl->nbl_kref);
 309			nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 310					&nfsd4_cb_notify_lock_ops,
 311					NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 312		}
 313	}
 314	return nbl;
 315}
 316
 317static void
 318free_nbl(struct kref *kref)
 319{
 320	struct nfsd4_blocked_lock *nbl;
 321
 322	nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
 323	locks_release_private(&nbl->nbl_lock);
 324	kfree(nbl);
 325}
 326
 327static void
 328free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 329{
 330	locks_delete_block(&nbl->nbl_lock);
 331	kref_put(&nbl->nbl_kref, free_nbl);
 332}
 333
 334static void
 335remove_blocked_locks(struct nfs4_lockowner *lo)
 336{
 337	struct nfs4_client *clp = lo->lo_owner.so_client;
 338	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 339	struct nfsd4_blocked_lock *nbl;
 340	LIST_HEAD(reaplist);
 341
 342	/* Dequeue all blocked locks */
 343	spin_lock(&nn->blocked_locks_lock);
 344	while (!list_empty(&lo->lo_blocked)) {
 345		nbl = list_first_entry(&lo->lo_blocked,
 346					struct nfsd4_blocked_lock,
 347					nbl_list);
 348		list_del_init(&nbl->nbl_list);
 349		WARN_ON(list_empty(&nbl->nbl_lru));
 350		list_move(&nbl->nbl_lru, &reaplist);
 351	}
 352	spin_unlock(&nn->blocked_locks_lock);
 353
 354	/* Now free them */
 355	while (!list_empty(&reaplist)) {
 356		nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 357					nbl_lru);
 358		list_del_init(&nbl->nbl_lru);
 359		free_blocked_lock(nbl);
 360	}
 361}
 362
 363static void
 364nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 365{
 366	struct nfsd4_blocked_lock	*nbl = container_of(cb,
 367						struct nfsd4_blocked_lock, nbl_cb);
 368	locks_delete_block(&nbl->nbl_lock);
 369}
 370
 371static int
 372nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 373{
 374	trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
 375
 376	/*
 377	 * Since this is just an optimization, we don't try very hard if it
 378	 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 379	 * just quit trying on anything else.
 380	 */
 381	switch (task->tk_status) {
 382	case -NFS4ERR_DELAY:
 383		rpc_delay(task, 1 * HZ);
 384		return 0;
 385	default:
 386		return 1;
 387	}
 388}
 389
 390static void
 391nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 392{
 393	struct nfsd4_blocked_lock	*nbl = container_of(cb,
 394						struct nfsd4_blocked_lock, nbl_cb);
 395
 396	free_blocked_lock(nbl);
 397}
 398
 399static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 400	.prepare	= nfsd4_cb_notify_lock_prepare,
 401	.done		= nfsd4_cb_notify_lock_done,
 402	.release	= nfsd4_cb_notify_lock_release,
 403	.opcode		= OP_CB_NOTIFY_LOCK,
 404};
 405
 406/*
 407 * We store the NONE, READ, WRITE, and BOTH bits separately in the
 408 * st_{access,deny}_bmap field of the stateid, in order to track not
 409 * only what share bits are currently in force, but also what
 410 * combinations of share bits previous opens have used.  This allows us
 411 * to enforce the recommendation in
 412 * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
 413 * the server return an error if the client attempt to downgrade to a
 414 * combination of share bits not explicable by closing some of its
 415 * previous opens.
 416 *
 417 * This enforcement is arguably incomplete, since we don't keep
 418 * track of access/deny bit combinations; so, e.g., we allow:
 419 *
 420 *	OPEN allow read, deny write
 421 *	OPEN allow both, deny none
 422 *	DOWNGRADE allow read, deny none
 423 *
 424 * which we should reject.
 425 *
 426 * But you could also argue that our current code is already overkill,
 427 * since it only exists to return NFS4ERR_INVAL on incorrect client
 428 * behavior.
 429 */
 430static unsigned int
 431bmap_to_share_mode(unsigned long bmap)
 432{
 433	int i;
 434	unsigned int access = 0;
 435
 436	for (i = 1; i < 4; i++) {
 437		if (test_bit(i, &bmap))
 438			access |= i;
 439	}
 440	return access;
 441}
 442
 443/* set share access for a given stateid */
 444static inline void
 445set_access(u32 access, struct nfs4_ol_stateid *stp)
 446{
 447	unsigned char mask = 1 << access;
 448
 449	WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 450	stp->st_access_bmap |= mask;
 451}
 452
 453/* clear share access for a given stateid */
 454static inline void
 455clear_access(u32 access, struct nfs4_ol_stateid *stp)
 456{
 457	unsigned char mask = 1 << access;
 458
 459	WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 460	stp->st_access_bmap &= ~mask;
 461}
 462
 463/* test whether a given stateid has access */
 464static inline bool
 465test_access(u32 access, struct nfs4_ol_stateid *stp)
 466{
 467	unsigned char mask = 1 << access;
 468
 469	return (bool)(stp->st_access_bmap & mask);
 470}
 471
 472/* set share deny for a given stateid */
 473static inline void
 474set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 475{
 476	unsigned char mask = 1 << deny;
 477
 478	WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 479	stp->st_deny_bmap |= mask;
 480}
 481
 482/* clear share deny for a given stateid */
 483static inline void
 484clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 485{
 486	unsigned char mask = 1 << deny;
 487
 488	WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 489	stp->st_deny_bmap &= ~mask;
 490}
 491
 492/* test whether a given stateid is denying specific access */
 493static inline bool
 494test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 495{
 496	unsigned char mask = 1 << deny;
 497
 498	return (bool)(stp->st_deny_bmap & mask);
 499}
 500
 501static int nfs4_access_to_omode(u32 access)
 502{
 503	switch (access & NFS4_SHARE_ACCESS_BOTH) {
 504	case NFS4_SHARE_ACCESS_READ:
 505		return O_RDONLY;
 506	case NFS4_SHARE_ACCESS_WRITE:
 507		return O_WRONLY;
 508	case NFS4_SHARE_ACCESS_BOTH:
 509		return O_RDWR;
 510	}
 511	WARN_ON_ONCE(1);
 512	return O_RDONLY;
 513}
 514
 515static inline int
 516access_permit_read(struct nfs4_ol_stateid *stp)
 517{
 518	return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
 519		test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
 520		test_access(NFS4_SHARE_ACCESS_WRITE, stp);
 521}
 522
 523static inline int
 524access_permit_write(struct nfs4_ol_stateid *stp)
 525{
 526	return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
 527		test_access(NFS4_SHARE_ACCESS_BOTH, stp);
 528}
 529
 530static inline struct nfs4_stateowner *
 531nfs4_get_stateowner(struct nfs4_stateowner *sop)
 532{
 533	atomic_inc(&sop->so_count);
 534	return sop;
 535}
 536
 537static int
 538same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 539{
 540	return (sop->so_owner.len == owner->len) &&
 541		0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 542}
 543
 544static struct nfs4_openowner *
 545find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 546			struct nfs4_client *clp)
 547{
 548	struct nfs4_stateowner *so;
 549
 550	lockdep_assert_held(&clp->cl_lock);
 551
 552	list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 553			    so_strhash) {
 554		if (!so->so_is_open_owner)
 555			continue;
 556		if (same_owner_str(so, &open->op_owner))
 557			return openowner(nfs4_get_stateowner(so));
 558	}
 559	return NULL;
 560}
 561
 562static inline u32
 563opaque_hashval(const void *ptr, int nbytes)
 564{
 565	unsigned char *cptr = (unsigned char *) ptr;
 566
 567	u32 x = 0;
 568	while (nbytes--) {
 569		x *= 37;
 570		x += *cptr++;
 571	}
 572	return x;
 573}
 574
 575void
 576put_nfs4_file(struct nfs4_file *fi)
 577{
 578	if (refcount_dec_and_test(&fi->fi_ref)) {
 579		nfsd4_file_hash_remove(fi);
 580		WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 581		WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 582		kfree_rcu(fi, fi_rcu);
 583	}
 584}
 585
 586static struct nfsd_file *
 587find_writeable_file_locked(struct nfs4_file *f)
 588{
 589	struct nfsd_file *ret;
 590
 591	lockdep_assert_held(&f->fi_lock);
 592
 593	ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 594	if (!ret)
 595		ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 596	return ret;
 597}
 598
 599static struct nfsd_file *
 600find_writeable_file(struct nfs4_file *f)
 601{
 602	struct nfsd_file *ret;
 603
 604	spin_lock(&f->fi_lock);
 605	ret = find_writeable_file_locked(f);
 606	spin_unlock(&f->fi_lock);
 607
 608	return ret;
 609}
 610
 611static struct nfsd_file *
 612find_readable_file_locked(struct nfs4_file *f)
 613{
 614	struct nfsd_file *ret;
 615
 616	lockdep_assert_held(&f->fi_lock);
 617
 618	ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 619	if (!ret)
 620		ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 621	return ret;
 622}
 623
 624static struct nfsd_file *
 625find_readable_file(struct nfs4_file *f)
 626{
 627	struct nfsd_file *ret;
 628
 629	spin_lock(&f->fi_lock);
 630	ret = find_readable_file_locked(f);
 631	spin_unlock(&f->fi_lock);
 632
 633	return ret;
 634}
 635
 636struct nfsd_file *
 637find_any_file(struct nfs4_file *f)
 638{
 639	struct nfsd_file *ret;
 640
 641	if (!f)
 642		return NULL;
 643	spin_lock(&f->fi_lock);
 644	ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 645	if (!ret) {
 646		ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 647		if (!ret)
 648			ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 649	}
 650	spin_unlock(&f->fi_lock);
 651	return ret;
 652}
 653
 654static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
 655{
 656	lockdep_assert_held(&f->fi_lock);
 657
 658	if (f->fi_fds[O_RDWR])
 659		return f->fi_fds[O_RDWR];
 660	if (f->fi_fds[O_WRONLY])
 661		return f->fi_fds[O_WRONLY];
 662	if (f->fi_fds[O_RDONLY])
 663		return f->fi_fds[O_RDONLY];
 664	return NULL;
 665}
 666
 667static atomic_long_t num_delegations;
 668unsigned long max_delegations;
 669
 670/*
 671 * Open owner state (share locks)
 672 */
 673
 674/* hash tables for lock and open owners */
 675#define OWNER_HASH_BITS              8
 676#define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 677#define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 678
 679static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 680{
 681	unsigned int ret;
 682
 683	ret = opaque_hashval(ownername->data, ownername->len);
 684	return ret & OWNER_HASH_MASK;
 685}
 686
 687static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
 688
 689static const struct rhashtable_params nfs4_file_rhash_params = {
 690	.key_len		= sizeof_field(struct nfs4_file, fi_inode),
 691	.key_offset		= offsetof(struct nfs4_file, fi_inode),
 692	.head_offset		= offsetof(struct nfs4_file, fi_rlist),
 693
 694	/*
 695	 * Start with a single page hash table to reduce resizing churn
 696	 * on light workloads.
 697	 */
 698	.min_size		= 256,
 699	.automatic_shrinking	= true,
 700};
 701
 702/*
 703 * Check if courtesy clients have conflicting access and resolve it if possible
 704 *
 705 * access:  is op_share_access if share_access is true.
 706 *	    Check if access mode, op_share_access, would conflict with
 707 *	    the current deny mode of the file 'fp'.
 708 * access:  is op_share_deny if share_access is false.
 709 *	    Check if the deny mode, op_share_deny, would conflict with
 710 *	    current access of the file 'fp'.
 711 * stp:     skip checking this entry.
 712 * new_stp: normal open, not open upgrade.
 713 *
 714 * Function returns:
 715 *	false - access/deny mode conflict with normal client.
 716 *	true  - no conflict or conflict with courtesy client(s) is resolved.
 717 */
 718static bool
 719nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
 720		struct nfs4_ol_stateid *stp, u32 access, bool share_access)
 721{
 722	struct nfs4_ol_stateid *st;
 723	bool resolvable = true;
 724	unsigned char bmap;
 725	struct nfsd_net *nn;
 726	struct nfs4_client *clp;
 727
 728	lockdep_assert_held(&fp->fi_lock);
 729	list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
 730		/* ignore lock stateid */
 731		if (st->st_openstp)
 732			continue;
 733		if (st == stp && new_stp)
 734			continue;
 735		/* check file access against deny mode or vice versa */
 736		bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
 737		if (!(access & bmap_to_share_mode(bmap)))
 738			continue;
 739		clp = st->st_stid.sc_client;
 740		if (try_to_expire_client(clp))
 741			continue;
 742		resolvable = false;
 743		break;
 744	}
 745	if (resolvable) {
 746		clp = stp->st_stid.sc_client;
 747		nn = net_generic(clp->net, nfsd_net_id);
 748		mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
 749	}
 750	return resolvable;
 751}
 752
 753static void
 754__nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 755{
 756	lockdep_assert_held(&fp->fi_lock);
 757
 758	if (access & NFS4_SHARE_ACCESS_WRITE)
 759		atomic_inc(&fp->fi_access[O_WRONLY]);
 760	if (access & NFS4_SHARE_ACCESS_READ)
 761		atomic_inc(&fp->fi_access[O_RDONLY]);
 762}
 763
 764static __be32
 765nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 766{
 767	lockdep_assert_held(&fp->fi_lock);
 768
 769	/* Does this access mode make sense? */
 770	if (access & ~NFS4_SHARE_ACCESS_BOTH)
 771		return nfserr_inval;
 772
 773	/* Does it conflict with a deny mode already set? */
 774	if ((access & fp->fi_share_deny) != 0)
 775		return nfserr_share_denied;
 776
 777	__nfs4_file_get_access(fp, access);
 778	return nfs_ok;
 779}
 780
 781static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 782{
 783	/* Common case is that there is no deny mode. */
 784	if (deny) {
 785		/* Does this deny mode make sense? */
 786		if (deny & ~NFS4_SHARE_DENY_BOTH)
 787			return nfserr_inval;
 788
 789		if ((deny & NFS4_SHARE_DENY_READ) &&
 790		    atomic_read(&fp->fi_access[O_RDONLY]))
 791			return nfserr_share_denied;
 792
 793		if ((deny & NFS4_SHARE_DENY_WRITE) &&
 794		    atomic_read(&fp->fi_access[O_WRONLY]))
 795			return nfserr_share_denied;
 796	}
 797	return nfs_ok;
 798}
 799
 800static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 801{
 802	might_lock(&fp->fi_lock);
 803
 804	if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 805		struct nfsd_file *f1 = NULL;
 806		struct nfsd_file *f2 = NULL;
 807
 808		swap(f1, fp->fi_fds[oflag]);
 809		if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 810			swap(f2, fp->fi_fds[O_RDWR]);
 811		spin_unlock(&fp->fi_lock);
 812		if (f1)
 813			nfsd_file_put(f1);
 814		if (f2)
 815			nfsd_file_put(f2);
 816	}
 817}
 818
 819static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 820{
 821	WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 822
 823	if (access & NFS4_SHARE_ACCESS_WRITE)
 824		__nfs4_file_put_access(fp, O_WRONLY);
 825	if (access & NFS4_SHARE_ACCESS_READ)
 826		__nfs4_file_put_access(fp, O_RDONLY);
 827}
 828
 829/*
 830 * Allocate a new open/delegation state counter. This is needed for
 831 * pNFS for proper return on close semantics.
 832 *
 833 * Note that we only allocate it for pNFS-enabled exports, otherwise
 834 * all pointers to struct nfs4_clnt_odstate are always NULL.
 835 */
 836static struct nfs4_clnt_odstate *
 837alloc_clnt_odstate(struct nfs4_client *clp)
 838{
 839	struct nfs4_clnt_odstate *co;
 840
 841	co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 842	if (co) {
 843		co->co_client = clp;
 844		refcount_set(&co->co_odcount, 1);
 845	}
 846	return co;
 847}
 848
 849static void
 850hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 851{
 852	struct nfs4_file *fp = co->co_file;
 853
 854	lockdep_assert_held(&fp->fi_lock);
 855	list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 856}
 857
 858static inline void
 859get_clnt_odstate(struct nfs4_clnt_odstate *co)
 860{
 861	if (co)
 862		refcount_inc(&co->co_odcount);
 863}
 864
 865static void
 866put_clnt_odstate(struct nfs4_clnt_odstate *co)
 867{
 868	struct nfs4_file *fp;
 869
 870	if (!co)
 871		return;
 872
 873	fp = co->co_file;
 874	if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 875		list_del(&co->co_perfile);
 876		spin_unlock(&fp->fi_lock);
 877
 878		nfsd4_return_all_file_layouts(co->co_client, fp);
 879		kmem_cache_free(odstate_slab, co);
 880	}
 881}
 882
 883static struct nfs4_clnt_odstate *
 884find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 885{
 886	struct nfs4_clnt_odstate *co;
 887	struct nfs4_client *cl;
 888
 889	if (!new)
 890		return NULL;
 891
 892	cl = new->co_client;
 893
 894	spin_lock(&fp->fi_lock);
 895	list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 896		if (co->co_client == cl) {
 897			get_clnt_odstate(co);
 898			goto out;
 899		}
 900	}
 901	co = new;
 902	co->co_file = fp;
 903	hash_clnt_odstate_locked(new);
 904out:
 905	spin_unlock(&fp->fi_lock);
 906	return co;
 907}
 908
 909struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 910				  void (*sc_free)(struct nfs4_stid *))
 911{
 912	struct nfs4_stid *stid;
 913	int new_id;
 914
 915	stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 916	if (!stid)
 917		return NULL;
 918
 919	idr_preload(GFP_KERNEL);
 920	spin_lock(&cl->cl_lock);
 921	/* Reserving 0 for start of file in nfsdfs "states" file: */
 922	new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 923	spin_unlock(&cl->cl_lock);
 924	idr_preload_end();
 925	if (new_id < 0)
 926		goto out_free;
 927
 928	stid->sc_free = sc_free;
 929	stid->sc_client = cl;
 930	stid->sc_stateid.si_opaque.so_id = new_id;
 931	stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 932	/* Will be incremented before return to client: */
 933	refcount_set(&stid->sc_count, 1);
 934	spin_lock_init(&stid->sc_lock);
 935	INIT_LIST_HEAD(&stid->sc_cp_list);
 936
 937	return stid;
 938out_free:
 939	kmem_cache_free(slab, stid);
 940	return NULL;
 941}
 942
 943/*
 944 * Create a unique stateid_t to represent each COPY.
 945 */
 946static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 947			      unsigned char cs_type)
 948{
 949	int new_id;
 950
 951	stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 952	stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 953
 954	idr_preload(GFP_KERNEL);
 955	spin_lock(&nn->s2s_cp_lock);
 956	new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 957	stid->cs_stid.si_opaque.so_id = new_id;
 958	stid->cs_stid.si_generation = 1;
 959	spin_unlock(&nn->s2s_cp_lock);
 960	idr_preload_end();
 961	if (new_id < 0)
 962		return 0;
 963	stid->cs_type = cs_type;
 964	return 1;
 965}
 966
 967int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 968{
 969	return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
 970}
 971
 972struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
 973						     struct nfs4_stid *p_stid)
 974{
 975	struct nfs4_cpntf_state *cps;
 976
 977	cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
 978	if (!cps)
 979		return NULL;
 980	cps->cpntf_time = ktime_get_boottime_seconds();
 981	refcount_set(&cps->cp_stateid.cs_count, 1);
 982	if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
 983		goto out_free;
 984	spin_lock(&nn->s2s_cp_lock);
 985	list_add(&cps->cp_list, &p_stid->sc_cp_list);
 986	spin_unlock(&nn->s2s_cp_lock);
 987	return cps;
 988out_free:
 989	kfree(cps);
 990	return NULL;
 991}
 992
 993void nfs4_free_copy_state(struct nfsd4_copy *copy)
 994{
 995	struct nfsd_net *nn;
 996
 997	if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
 998		return;
 999	nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1000	spin_lock(&nn->s2s_cp_lock);
1001	idr_remove(&nn->s2s_cp_stateids,
1002		   copy->cp_stateid.cs_stid.si_opaque.so_id);
1003	spin_unlock(&nn->s2s_cp_lock);
1004}
1005
1006static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1007{
1008	struct nfs4_cpntf_state *cps;
1009	struct nfsd_net *nn;
1010
1011	nn = net_generic(net, nfsd_net_id);
1012	spin_lock(&nn->s2s_cp_lock);
1013	while (!list_empty(&stid->sc_cp_list)) {
1014		cps = list_first_entry(&stid->sc_cp_list,
1015				       struct nfs4_cpntf_state, cp_list);
1016		_free_cpntf_state_locked(nn, cps);
1017	}
1018	spin_unlock(&nn->s2s_cp_lock);
1019}
1020
1021static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1022{
1023	struct nfs4_stid *stid;
1024
1025	stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1026	if (!stid)
1027		return NULL;
1028
1029	return openlockstateid(stid);
1030}
1031
1032/*
1033 * As the sc_free callback of deleg, this may be called by nfs4_put_stid
1034 * in nfsd_break_one_deleg.
1035 * Considering nfsd_break_one_deleg is called with the flc->flc_lock held,
1036 * this function mustn't ever sleep.
1037 */
1038static void nfs4_free_deleg(struct nfs4_stid *stid)
1039{
1040	struct nfs4_delegation *dp = delegstateid(stid);
1041
1042	WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1043	WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1044	WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1045	WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1046	kmem_cache_free(deleg_slab, stid);
1047	atomic_long_dec(&num_delegations);
1048}
1049
1050/*
1051 * When we recall a delegation, we should be careful not to hand it
1052 * out again straight away.
1053 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1054 * in which the filehandles of recalled delegations are "stored".
1055 * If a filehandle appear in either filter, a delegation is blocked.
1056 * When a delegation is recalled, the filehandle is stored in the "new"
1057 * filter.
1058 * Every 30 seconds we swap the filters and clear the "new" one,
1059 * unless both are empty of course.  This results in delegations for a
1060 * given filehandle being blocked for between 30 and 60 seconds.
1061 *
1062 * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
1063 * low 3 bytes as hash-table indices.
1064 *
1065 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1066 * is used to manage concurrent access.  Testing does not need the lock
1067 * except when swapping the two filters.
1068 */
1069static DEFINE_SPINLOCK(blocked_delegations_lock);
1070static struct bloom_pair {
1071	int	entries, old_entries;
1072	time64_t swap_time;
1073	int	new; /* index into 'set' */
1074	DECLARE_BITMAP(set[2], 256);
1075} blocked_delegations;
1076
1077static int delegation_blocked(struct knfsd_fh *fh)
1078{
1079	u32 hash;
1080	struct bloom_pair *bd = &blocked_delegations;
1081
1082	if (bd->entries == 0)
1083		return 0;
1084	if (ktime_get_seconds() - bd->swap_time > 30) {
1085		spin_lock(&blocked_delegations_lock);
1086		if (ktime_get_seconds() - bd->swap_time > 30) {
1087			bd->entries -= bd->old_entries;
1088			bd->old_entries = bd->entries;
1089			bd->new = 1-bd->new;
1090			memset(bd->set[bd->new], 0,
1091			       sizeof(bd->set[0]));
1092			bd->swap_time = ktime_get_seconds();
1093		}
1094		spin_unlock(&blocked_delegations_lock);
1095	}
1096	hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1097	if (test_bit(hash&255, bd->set[0]) &&
1098	    test_bit((hash>>8)&255, bd->set[0]) &&
1099	    test_bit((hash>>16)&255, bd->set[0]))
1100		return 1;
1101
1102	if (test_bit(hash&255, bd->set[1]) &&
1103	    test_bit((hash>>8)&255, bd->set[1]) &&
1104	    test_bit((hash>>16)&255, bd->set[1]))
1105		return 1;
1106
1107	return 0;
1108}
1109
1110static void block_delegations(struct knfsd_fh *fh)
1111{
1112	u32 hash;
1113	struct bloom_pair *bd = &blocked_delegations;
1114
1115	hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1116
1117	spin_lock(&blocked_delegations_lock);
1118	__set_bit(hash&255, bd->set[bd->new]);
1119	__set_bit((hash>>8)&255, bd->set[bd->new]);
1120	__set_bit((hash>>16)&255, bd->set[bd->new]);
1121	if (bd->entries == 0)
1122		bd->swap_time = ktime_get_seconds();
1123	bd->entries += 1;
1124	spin_unlock(&blocked_delegations_lock);
1125}
1126
1127static struct nfs4_delegation *
1128alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1129		 struct nfs4_clnt_odstate *odstate, u32 dl_type)
1130{
1131	struct nfs4_delegation *dp;
1132	struct nfs4_stid *stid;
1133	long n;
1134
1135	dprintk("NFSD alloc_init_deleg\n");
1136	n = atomic_long_inc_return(&num_delegations);
1137	if (n < 0 || n > max_delegations)
1138		goto out_dec;
1139	if (delegation_blocked(&fp->fi_fhandle))
1140		goto out_dec;
1141	stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1142	if (stid == NULL)
1143		goto out_dec;
1144	dp = delegstateid(stid);
1145
1146	/*
1147	 * delegation seqid's are never incremented.  The 4.1 special
1148	 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1149	 * 0 anyway just for consistency and use 1:
1150	 */
1151	dp->dl_stid.sc_stateid.si_generation = 1;
1152	INIT_LIST_HEAD(&dp->dl_perfile);
1153	INIT_LIST_HEAD(&dp->dl_perclnt);
1154	INIT_LIST_HEAD(&dp->dl_recall_lru);
1155	dp->dl_clnt_odstate = odstate;
1156	get_clnt_odstate(odstate);
1157	dp->dl_type = dl_type;
1158	dp->dl_retries = 1;
1159	dp->dl_recalled = false;
1160	nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1161		      &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1162	nfsd4_init_cb(&dp->dl_cb_fattr.ncf_getattr, dp->dl_stid.sc_client,
1163			&nfsd4_cb_getattr_ops, NFSPROC4_CLNT_CB_GETATTR);
1164	dp->dl_cb_fattr.ncf_file_modified = false;
1165	get_nfs4_file(fp);
1166	dp->dl_stid.sc_file = fp;
1167	return dp;
1168out_dec:
1169	atomic_long_dec(&num_delegations);
1170	return NULL;
1171}
1172
1173void
1174nfs4_put_stid(struct nfs4_stid *s)
1175{
1176	struct nfs4_file *fp = s->sc_file;
1177	struct nfs4_client *clp = s->sc_client;
1178
1179	might_lock(&clp->cl_lock);
1180
1181	if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1182		wake_up_all(&close_wq);
1183		return;
1184	}
1185	idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1186	if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1187		atomic_dec(&s->sc_client->cl_admin_revoked);
1188	nfs4_free_cpntf_statelist(clp->net, s);
1189	spin_unlock(&clp->cl_lock);
1190	s->sc_free(s);
1191	if (fp)
1192		put_nfs4_file(fp);
1193}
1194
1195void
1196nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1197{
1198	stateid_t *src = &stid->sc_stateid;
1199
1200	spin_lock(&stid->sc_lock);
1201	if (unlikely(++src->si_generation == 0))
1202		src->si_generation = 1;
1203	memcpy(dst, src, sizeof(*dst));
1204	spin_unlock(&stid->sc_lock);
1205}
1206
1207static void put_deleg_file(struct nfs4_file *fp)
1208{
1209	struct nfsd_file *rnf = NULL;
1210	struct nfsd_file *nf = NULL;
1211
1212	spin_lock(&fp->fi_lock);
1213	if (--fp->fi_delegees == 0) {
1214		swap(nf, fp->fi_deleg_file);
1215		swap(rnf, fp->fi_rdeleg_file);
1216	}
1217	spin_unlock(&fp->fi_lock);
1218
1219	if (nf)
1220		nfsd_file_put(nf);
1221	if (rnf) {
1222		nfsd_file_put(rnf);
1223		nfs4_file_put_access(fp, NFS4_SHARE_ACCESS_READ);
1224	}
1225}
1226
1227static void nfsd4_finalize_deleg_timestamps(struct nfs4_delegation *dp, struct file *f)
1228{
1229	struct iattr ia = { .ia_valid = ATTR_ATIME | ATTR_CTIME | ATTR_MTIME | ATTR_DELEG };
1230	struct inode *inode = file_inode(f);
1231	int ret;
1232
1233	/* don't do anything if FMODE_NOCMTIME isn't set */
1234	if ((READ_ONCE(f->f_mode) & FMODE_NOCMTIME) == 0)
1235		return;
1236
1237	spin_lock(&f->f_lock);
1238	f->f_mode &= ~FMODE_NOCMTIME;
1239	spin_unlock(&f->f_lock);
1240
1241	/* was it never written? */
1242	if (!dp->dl_written)
1243		return;
1244
1245	/* did it get a setattr for the timestamps at some point? */
1246	if (dp->dl_setattr)
1247		return;
1248
1249	/* Stamp everything to "now" */
1250	inode_lock(inode);
1251	ret = notify_change(&nop_mnt_idmap, f->f_path.dentry, &ia, NULL);
1252	inode_unlock(inode);
1253	if (ret) {
1254		struct inode *inode = file_inode(f);
1255
1256		pr_notice_ratelimited("Unable to update timestamps on inode %02x:%02x:%lu: %d\n",
1257					MAJOR(inode->i_sb->s_dev),
1258					MINOR(inode->i_sb->s_dev),
1259					inode->i_ino, ret);
1260	}
1261}
1262
1263static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1264{
1265	struct nfs4_file *fp = dp->dl_stid.sc_file;
1266	struct nfsd_file *nf = fp->fi_deleg_file;
1267
1268	WARN_ON_ONCE(!fp->fi_delegees);
1269
1270	nfsd4_finalize_deleg_timestamps(dp, nf->nf_file);
1271	kernel_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1272	put_deleg_file(fp);
1273}
1274
1275static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1276{
1277	put_clnt_odstate(dp->dl_clnt_odstate);
1278	nfs4_unlock_deleg_lease(dp);
1279	nfs4_put_stid(&dp->dl_stid);
1280}
1281
1282/**
1283 * nfs4_delegation_exists - Discover if this delegation already exists
1284 * @clp:     a pointer to the nfs4_client we're granting a delegation to
1285 * @fp:      a pointer to the nfs4_file we're granting a delegation on
1286 *
1287 * Return:
1288 *      On success: true iff an existing delegation is found
1289 */
1290
1291static bool
1292nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1293{
1294	struct nfs4_delegation *searchdp = NULL;
1295	struct nfs4_client *searchclp = NULL;
1296
1297	lockdep_assert_held(&state_lock);
1298	lockdep_assert_held(&fp->fi_lock);
1299
1300	list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1301		searchclp = searchdp->dl_stid.sc_client;
1302		if (clp == searchclp) {
1303			return true;
1304		}
1305	}
1306	return false;
1307}
1308
1309/**
1310 * hash_delegation_locked - Add a delegation to the appropriate lists
1311 * @dp:     a pointer to the nfs4_delegation we are adding.
1312 * @fp:     a pointer to the nfs4_file we're granting a delegation on
1313 *
1314 * Return:
1315 *      On success: NULL if the delegation was successfully hashed.
1316 *
1317 *      On error: -EAGAIN if one was previously granted to this
1318 *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1319 *
1320 */
1321
1322static int
1323hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1324{
1325	struct nfs4_client *clp = dp->dl_stid.sc_client;
1326
1327	lockdep_assert_held(&state_lock);
1328	lockdep_assert_held(&fp->fi_lock);
1329	lockdep_assert_held(&clp->cl_lock);
1330
1331	if (nfs4_delegation_exists(clp, fp))
1332		return -EAGAIN;
1333	refcount_inc(&dp->dl_stid.sc_count);
1334	dp->dl_stid.sc_type = SC_TYPE_DELEG;
1335	list_add(&dp->dl_perfile, &fp->fi_delegations);
1336	list_add(&dp->dl_perclnt, &clp->cl_delegations);
1337	return 0;
1338}
1339
1340static bool delegation_hashed(struct nfs4_delegation *dp)
1341{
1342	return !(list_empty(&dp->dl_perfile));
1343}
1344
1345static bool
1346unhash_delegation_locked(struct nfs4_delegation *dp, unsigned short statusmask)
1347{
1348	struct nfs4_file *fp = dp->dl_stid.sc_file;
1349
1350	lockdep_assert_held(&state_lock);
1351
1352	if (!delegation_hashed(dp))
1353		return false;
1354
1355	if (statusmask == SC_STATUS_REVOKED &&
1356	    dp->dl_stid.sc_client->cl_minorversion == 0)
1357		statusmask = SC_STATUS_CLOSED;
1358	dp->dl_stid.sc_status |= statusmask;
1359	if (statusmask & SC_STATUS_ADMIN_REVOKED)
1360		atomic_inc(&dp->dl_stid.sc_client->cl_admin_revoked);
1361
1362	/* Ensure that deleg break won't try to requeue it */
1363	++dp->dl_time;
1364	spin_lock(&fp->fi_lock);
1365	list_del_init(&dp->dl_perclnt);
1366	list_del_init(&dp->dl_recall_lru);
1367	list_del_init(&dp->dl_perfile);
1368	spin_unlock(&fp->fi_lock);
1369	return true;
1370}
1371
1372static void destroy_delegation(struct nfs4_delegation *dp)
1373{
1374	bool unhashed;
1375
1376	spin_lock(&state_lock);
1377	unhashed = unhash_delegation_locked(dp, SC_STATUS_CLOSED);
1378	spin_unlock(&state_lock);
1379	if (unhashed)
1380		destroy_unhashed_deleg(dp);
1381}
1382
1383/**
1384 * revoke_delegation - perform nfs4 delegation structure cleanup
1385 * @dp: pointer to the delegation
1386 *
1387 * This function assumes that it's called either from the administrative
1388 * interface (nfsd4_revoke_states()) that's revoking a specific delegation
1389 * stateid or it's called from a laundromat thread (nfsd4_landromat()) that
1390 * determined that this specific state has expired and needs to be revoked
1391 * (both mark state with the appropriate stid sc_status mode). It is also
1392 * assumed that a reference was taken on the @dp state.
1393 *
1394 * If this function finds that the @dp state is SC_STATUS_FREED it means
1395 * that a FREE_STATEID operation for this stateid has been processed and
1396 * we can proceed to removing it from recalled list. However, if @dp state
1397 * isn't marked SC_STATUS_FREED, it means we need place it on the cl_revoked
1398 * list and wait for the FREE_STATEID to arrive from the client. At the same
1399 * time, we need to mark it as SC_STATUS_FREEABLE to indicate to the
1400 * nfsd4_free_stateid() function that this stateid has already been added
1401 * to the cl_revoked list and that nfsd4_free_stateid() is now responsible
1402 * for removing it from the list. Inspection of where the delegation state
1403 * in the revocation process is protected by the clp->cl_lock.
1404 */
1405static void revoke_delegation(struct nfs4_delegation *dp)
1406{
1407	struct nfs4_client *clp = dp->dl_stid.sc_client;
1408
1409	WARN_ON(!list_empty(&dp->dl_recall_lru));
1410	WARN_ON_ONCE(dp->dl_stid.sc_client->cl_minorversion > 0 &&
1411		     !(dp->dl_stid.sc_status &
1412		     (SC_STATUS_REVOKED | SC_STATUS_ADMIN_REVOKED)));
1413
1414	trace_nfsd_stid_revoke(&dp->dl_stid);
1415
1416	spin_lock(&clp->cl_lock);
1417	if (dp->dl_stid.sc_status & SC_STATUS_FREED) {
1418		list_del_init(&dp->dl_recall_lru);
1419		goto out;
1420	}
1421	list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1422	dp->dl_stid.sc_status |= SC_STATUS_FREEABLE;
1423out:
1424	spin_unlock(&clp->cl_lock);
1425	destroy_unhashed_deleg(dp);
1426}
1427
1428/*
1429 * SETCLIENTID state
1430 */
1431
1432static unsigned int clientid_hashval(u32 id)
1433{
1434	return id & CLIENT_HASH_MASK;
1435}
1436
1437static unsigned int clientstr_hashval(struct xdr_netobj name)
1438{
1439	return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1440}
1441
1442/*
1443 * A stateid that had a deny mode associated with it is being released
1444 * or downgraded. Recalculate the deny mode on the file.
1445 */
1446static void
1447recalculate_deny_mode(struct nfs4_file *fp)
1448{
1449	struct nfs4_ol_stateid *stp;
1450	u32 old_deny;
1451
1452	spin_lock(&fp->fi_lock);
1453	old_deny = fp->fi_share_deny;
1454	fp->fi_share_deny = 0;
1455	list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1456		fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1457		if (fp->fi_share_deny == old_deny)
1458			break;
1459	}
1460	spin_unlock(&fp->fi_lock);
1461}
1462
1463static void
1464reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1465{
1466	int i;
1467	bool change = false;
1468
1469	for (i = 1; i < 4; i++) {
1470		if ((i & deny) != i) {
1471			change = true;
1472			clear_deny(i, stp);
1473		}
1474	}
1475
1476	/* Recalculate per-file deny mode if there was a change */
1477	if (change)
1478		recalculate_deny_mode(stp->st_stid.sc_file);
1479}
1480
1481/* release all access and file references for a given stateid */
1482static void
1483release_all_access(struct nfs4_ol_stateid *stp)
1484{
1485	int i;
1486	struct nfs4_file *fp = stp->st_stid.sc_file;
1487
1488	if (fp && stp->st_deny_bmap != 0)
1489		recalculate_deny_mode(fp);
1490
1491	for (i = 1; i < 4; i++) {
1492		if (test_access(i, stp))
1493			nfs4_file_put_access(stp->st_stid.sc_file, i);
1494		clear_access(i, stp);
1495	}
1496}
1497
1498static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1499{
1500	kfree(sop->so_owner.data);
1501	sop->so_ops->so_free(sop);
1502}
1503
1504static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1505{
1506	struct nfs4_client *clp = sop->so_client;
1507
1508	might_lock(&clp->cl_lock);
1509
1510	if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1511		return;
1512	sop->so_ops->so_unhash(sop);
1513	spin_unlock(&clp->cl_lock);
1514	nfs4_free_stateowner(sop);
1515}
1516
1517static bool
1518nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1519{
1520	return list_empty(&stp->st_perfile);
1521}
1522
1523static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1524{
1525	struct nfs4_file *fp = stp->st_stid.sc_file;
1526
1527	lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1528
1529	if (list_empty(&stp->st_perfile))
1530		return false;
1531
1532	spin_lock(&fp->fi_lock);
1533	list_del_init(&stp->st_perfile);
1534	spin_unlock(&fp->fi_lock);
1535	list_del(&stp->st_perstateowner);
1536	return true;
1537}
1538
1539static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1540{
1541	struct nfs4_ol_stateid *stp = openlockstateid(stid);
1542
1543	put_clnt_odstate(stp->st_clnt_odstate);
1544	release_all_access(stp);
1545	if (stp->st_stateowner)
1546		nfs4_put_stateowner(stp->st_stateowner);
1547	if (!list_empty(&stid->sc_cp_list))
1548		nfs4_free_cpntf_statelist(stid->sc_client->net, stid);
1549	kmem_cache_free(stateid_slab, stid);
1550}
1551
1552static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1553{
1554	struct nfs4_ol_stateid *stp = openlockstateid(stid);
1555	struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1556	struct nfsd_file *nf;
1557
1558	nf = find_any_file(stp->st_stid.sc_file);
1559	if (nf) {
1560		get_file(nf->nf_file);
1561		filp_close(nf->nf_file, (fl_owner_t)lo);
1562		nfsd_file_put(nf);
1563	}
1564	nfs4_free_ol_stateid(stid);
1565}
1566
1567/*
1568 * Put the persistent reference to an already unhashed generic stateid, while
1569 * holding the cl_lock. If it's the last reference, then put it onto the
1570 * reaplist for later destruction.
1571 */
1572static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1573				       struct list_head *reaplist)
1574{
1575	struct nfs4_stid *s = &stp->st_stid;
1576	struct nfs4_client *clp = s->sc_client;
1577
1578	lockdep_assert_held(&clp->cl_lock);
1579
1580	WARN_ON_ONCE(!list_empty(&stp->st_locks));
1581
1582	if (!refcount_dec_and_test(&s->sc_count)) {
1583		wake_up_all(&close_wq);
1584		return;
1585	}
1586
1587	idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1588	if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1589		atomic_dec(&s->sc_client->cl_admin_revoked);
1590	list_add(&stp->st_locks, reaplist);
1591}
1592
1593static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1594{
1595	lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1596
1597	if (!unhash_ol_stateid(stp))
1598		return false;
1599	list_del_init(&stp->st_locks);
1600	stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1601	return true;
1602}
1603
1604static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1605{
1606	struct nfs4_client *clp = stp->st_stid.sc_client;
1607	bool unhashed;
1608
1609	spin_lock(&clp->cl_lock);
1610	unhashed = unhash_lock_stateid(stp);
1611	spin_unlock(&clp->cl_lock);
1612	if (unhashed)
1613		nfs4_put_stid(&stp->st_stid);
1614}
1615
1616static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1617{
1618	struct nfs4_client *clp = lo->lo_owner.so_client;
1619
1620	lockdep_assert_held(&clp->cl_lock);
1621
1622	list_del_init(&lo->lo_owner.so_strhash);
1623}
1624
1625/*
1626 * Free a list of generic stateids that were collected earlier after being
1627 * fully unhashed.
1628 */
1629static void
1630free_ol_stateid_reaplist(struct list_head *reaplist)
1631{
1632	struct nfs4_ol_stateid *stp;
1633	struct nfs4_file *fp;
1634
1635	might_sleep();
1636
1637	while (!list_empty(reaplist)) {
1638		stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1639				       st_locks);
1640		list_del(&stp->st_locks);
1641		fp = stp->st_stid.sc_file;
1642		stp->st_stid.sc_free(&stp->st_stid);
1643		if (fp)
1644			put_nfs4_file(fp);
1645	}
1646}
1647
1648static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1649				       struct list_head *reaplist)
1650{
1651	struct nfs4_ol_stateid *stp;
1652
1653	lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1654
1655	while (!list_empty(&open_stp->st_locks)) {
1656		stp = list_entry(open_stp->st_locks.next,
1657				struct nfs4_ol_stateid, st_locks);
1658		unhash_lock_stateid(stp);
1659		put_ol_stateid_locked(stp, reaplist);
1660	}
1661}
1662
1663static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1664				struct list_head *reaplist)
1665{
1666	lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1667
1668	if (!unhash_ol_stateid(stp))
1669		return false;
1670	release_open_stateid_locks(stp, reaplist);
1671	return true;
1672}
1673
1674static void release_open_stateid(struct nfs4_ol_stateid *stp)
1675{
1676	LIST_HEAD(reaplist);
1677
1678	spin_lock(&stp->st_stid.sc_client->cl_lock);
1679	stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1680	if (unhash_open_stateid(stp, &reaplist))
1681		put_ol_stateid_locked(stp, &reaplist);
1682	spin_unlock(&stp->st_stid.sc_client->cl_lock);
1683	free_ol_stateid_reaplist(&reaplist);
1684}
1685
1686static bool nfs4_openowner_unhashed(struct nfs4_openowner *oo)
1687{
1688	lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1689
1690	return list_empty(&oo->oo_owner.so_strhash) &&
1691		list_empty(&oo->oo_perclient);
1692}
1693
1694static void unhash_openowner_locked(struct nfs4_openowner *oo)
1695{
1696	struct nfs4_client *clp = oo->oo_owner.so_client;
1697
1698	lockdep_assert_held(&clp->cl_lock);
1699
1700	list_del_init(&oo->oo_owner.so_strhash);
1701	list_del_init(&oo->oo_perclient);
1702}
1703
1704static void release_last_closed_stateid(struct nfs4_openowner *oo)
1705{
1706	struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1707					  nfsd_net_id);
1708	struct nfs4_ol_stateid *s;
1709
1710	spin_lock(&nn->client_lock);
1711	s = oo->oo_last_closed_stid;
1712	if (s) {
1713		list_del_init(&oo->oo_close_lru);
1714		oo->oo_last_closed_stid = NULL;
1715	}
1716	spin_unlock(&nn->client_lock);
1717	if (s)
1718		nfs4_put_stid(&s->st_stid);
1719}
1720
1721static void release_openowner(struct nfs4_openowner *oo)
1722{
1723	struct nfs4_ol_stateid *stp;
1724	struct nfs4_client *clp = oo->oo_owner.so_client;
1725	LIST_HEAD(reaplist);
1726
1727	spin_lock(&clp->cl_lock);
1728	unhash_openowner_locked(oo);
1729	while (!list_empty(&oo->oo_owner.so_stateids)) {
1730		stp = list_first_entry(&oo->oo_owner.so_stateids,
1731				struct nfs4_ol_stateid, st_perstateowner);
1732		if (unhash_open_stateid(stp, &reaplist))
1733			put_ol_stateid_locked(stp, &reaplist);
1734	}
1735	spin_unlock(&clp->cl_lock);
1736	free_ol_stateid_reaplist(&reaplist);
1737	release_last_closed_stateid(oo);
1738	nfs4_put_stateowner(&oo->oo_owner);
1739}
1740
1741static struct nfs4_stid *find_one_sb_stid(struct nfs4_client *clp,
1742					  struct super_block *sb,
1743					  unsigned int sc_types)
1744{
1745	unsigned long id, tmp;
1746	struct nfs4_stid *stid;
1747
1748	spin_lock(&clp->cl_lock);
1749	idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
1750		if ((stid->sc_type & sc_types) &&
1751		    stid->sc_status == 0 &&
1752		    stid->sc_file->fi_inode->i_sb == sb) {
1753			refcount_inc(&stid->sc_count);
1754			break;
1755		}
1756	spin_unlock(&clp->cl_lock);
1757	return stid;
1758}
1759
1760/**
1761 * nfsd4_revoke_states - revoke all nfsv4 states associated with given filesystem
1762 * @nn:   used to identify instance of nfsd (there is one per net namespace)
1763 * @sb:   super_block used to identify target filesystem
1764 *
1765 * All nfs4 states (open, lock, delegation, layout) held by the server instance
1766 * and associated with a file on the given filesystem will be revoked resulting
1767 * in any files being closed and so all references from nfsd to the filesystem
1768 * being released.  Thus nfsd will no longer prevent the filesystem from being
1769 * unmounted.
1770 *
1771 * The clients which own the states will subsequently being notified that the
1772 * states have been "admin-revoked".
1773 */
1774void nfsd4_revoke_states(struct nfsd_net *nn, struct super_block *sb)
1775{
1776	unsigned int idhashval;
1777	unsigned int sc_types;
1778
1779	sc_types = SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG | SC_TYPE_LAYOUT;
1780
1781	spin_lock(&nn->client_lock);
1782	for (idhashval = 0; idhashval < CLIENT_HASH_SIZE; idhashval++) {
1783		struct list_head *head = &nn->conf_id_hashtbl[idhashval];
1784		struct nfs4_client *clp;
1785	retry:
1786		list_for_each_entry(clp, head, cl_idhash) {
1787			struct nfs4_stid *stid = find_one_sb_stid(clp, sb,
1788								  sc_types);
1789			if (stid) {
1790				struct nfs4_ol_stateid *stp;
1791				struct nfs4_delegation *dp;
1792				struct nfs4_layout_stateid *ls;
1793
1794				spin_unlock(&nn->client_lock);
1795				switch (stid->sc_type) {
1796				case SC_TYPE_OPEN:
1797					stp = openlockstateid(stid);
1798					mutex_lock_nested(&stp->st_mutex,
1799							  OPEN_STATEID_MUTEX);
1800
1801					spin_lock(&clp->cl_lock);
1802					if (stid->sc_status == 0) {
1803						stid->sc_status |=
1804							SC_STATUS_ADMIN_REVOKED;
1805						atomic_inc(&clp->cl_admin_revoked);
1806						spin_unlock(&clp->cl_lock);
1807						release_all_access(stp);
1808					} else
1809						spin_unlock(&clp->cl_lock);
1810					mutex_unlock(&stp->st_mutex);
1811					break;
1812				case SC_TYPE_LOCK:
1813					stp = openlockstateid(stid);
1814					mutex_lock_nested(&stp->st_mutex,
1815							  LOCK_STATEID_MUTEX);
1816					spin_lock(&clp->cl_lock);
1817					if (stid->sc_status == 0) {
1818						struct nfs4_lockowner *lo =
1819							lockowner(stp->st_stateowner);
1820						struct nfsd_file *nf;
1821
1822						stid->sc_status |=
1823							SC_STATUS_ADMIN_REVOKED;
1824						atomic_inc(&clp->cl_admin_revoked);
1825						spin_unlock(&clp->cl_lock);
1826						nf = find_any_file(stp->st_stid.sc_file);
1827						if (nf) {
1828							get_file(nf->nf_file);
1829							filp_close(nf->nf_file,
1830								   (fl_owner_t)lo);
1831							nfsd_file_put(nf);
1832						}
1833						release_all_access(stp);
1834					} else
1835						spin_unlock(&clp->cl_lock);
1836					mutex_unlock(&stp->st_mutex);
1837					break;
1838				case SC_TYPE_DELEG:
1839					refcount_inc(&stid->sc_count);
1840					dp = delegstateid(stid);
1841					spin_lock(&state_lock);
1842					if (!unhash_delegation_locked(
1843						    dp, SC_STATUS_ADMIN_REVOKED))
1844						dp = NULL;
1845					spin_unlock(&state_lock);
1846					if (dp)
1847						revoke_delegation(dp);
1848					break;
1849				case SC_TYPE_LAYOUT:
1850					ls = layoutstateid(stid);
1851					nfsd4_close_layout(ls);
1852					break;
1853				}
1854				nfs4_put_stid(stid);
1855				spin_lock(&nn->client_lock);
1856				if (clp->cl_minorversion == 0)
1857					/* Allow cleanup after a lease period.
1858					 * store_release ensures cleanup will
1859					 * see any newly revoked states if it
1860					 * sees the time updated.
1861					 */
1862					nn->nfs40_last_revoke =
1863						ktime_get_boottime_seconds();
1864				goto retry;
1865			}
1866		}
1867	}
1868	spin_unlock(&nn->client_lock);
1869}
1870
1871static inline int
1872hash_sessionid(struct nfs4_sessionid *sessionid)
1873{
1874	struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1875
1876	return sid->sequence % SESSION_HASH_SIZE;
1877}
1878
1879#ifdef CONFIG_SUNRPC_DEBUG
1880static inline void
1881dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1882{
1883	u32 *ptr = (u32 *)(&sessionid->data[0]);
1884	dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1885}
1886#else
1887static inline void
1888dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1889{
1890}
1891#endif
1892
1893/*
1894 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1895 * won't be used for replay.
1896 */
1897void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1898{
1899	struct nfs4_stateowner *so = cstate->replay_owner;
1900
1901	if (nfserr == nfserr_replay_me)
1902		return;
1903
1904	if (!seqid_mutating_err(ntohl(nfserr))) {
1905		nfsd4_cstate_clear_replay(cstate);
1906		return;
1907	}
1908	if (!so)
1909		return;
1910	if (so->so_is_open_owner)
1911		release_last_closed_stateid(openowner(so));
1912	so->so_seqid++;
1913	return;
1914}
1915
1916static void
1917gen_sessionid(struct nfsd4_session *ses)
1918{
1919	struct nfs4_client *clp = ses->se_client;
1920	struct nfsd4_sessionid *sid;
1921
1922	sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1923	sid->clientid = clp->cl_clientid;
1924	sid->sequence = current_sessionid++;
1925	sid->reserved = 0;
1926}
1927
1928/*
1929 * The protocol defines ca_maxresponssize_cached to include the size of
1930 * the rpc header, but all we need to cache is the data starting after
1931 * the end of the initial SEQUENCE operation--the rest we regenerate
1932 * each time.  Therefore we can advertise a ca_maxresponssize_cached
1933 * value that is the number of bytes in our cache plus a few additional
1934 * bytes.  In order to stay on the safe side, and not promise more than
1935 * we can cache, those additional bytes must be the minimum possible: 24
1936 * bytes of rpc header (xid through accept state, with AUTH_NULL
1937 * verifier), 12 for the compound header (with zero-length tag), and 44
1938 * for the SEQUENCE op response:
1939 */
1940#define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1941
1942static struct shrinker *nfsd_slot_shrinker;
1943static DEFINE_SPINLOCK(nfsd_session_list_lock);
1944static LIST_HEAD(nfsd_session_list);
1945/* The sum of "target_slots-1" on every session.  The shrinker can push this
1946 * down, though it can take a little while for the memory to actually
1947 * be freed.  The "-1" is because we can never free slot 0 while the
1948 * session is active.
1949 */
1950static atomic_t nfsd_total_target_slots = ATOMIC_INIT(0);
1951
1952static void
1953free_session_slots(struct nfsd4_session *ses, int from)
1954{
1955	int i;
1956
1957	if (from >= ses->se_fchannel.maxreqs)
1958		return;
1959
1960	for (i = from; i < ses->se_fchannel.maxreqs; i++) {
1961		struct nfsd4_slot *slot = xa_load(&ses->se_slots, i);
1962
1963		/*
1964		 * Save the seqid in case we reactivate this slot.
1965		 * This will never require a memory allocation so GFP
1966		 * flag is irrelevant
1967		 */
1968		xa_store(&ses->se_slots, i, xa_mk_value(slot->sl_seqid), 0);
1969		free_svc_cred(&slot->sl_cred);
1970		kfree(slot);
1971	}
1972	ses->se_fchannel.maxreqs = from;
1973	if (ses->se_target_maxslots > from) {
1974		int new_target = from ?: 1;
1975		atomic_sub(ses->se_target_maxslots - new_target, &nfsd_total_target_slots);
1976		ses->se_target_maxslots = new_target;
1977	}
1978}
1979
1980/**
1981 * reduce_session_slots - reduce the target max-slots of a session if possible
1982 * @ses:  The session to affect
1983 * @dec:  how much to decrease the target by
1984 *
1985 * This interface can be used by a shrinker to reduce the target max-slots
1986 * for a session so that some slots can eventually be freed.
1987 * It uses spin_trylock() as it may be called in a context where another
1988 * spinlock is held that has a dependency on client_lock.  As shrinkers are
1989 * best-effort, skiping a session is client_lock is already held has no
1990 * great coast
1991 *
1992 * Return value:
1993 *   The number of slots that the target was reduced by.
1994 */
1995static int
1996reduce_session_slots(struct nfsd4_session *ses, int dec)
1997{
1998	struct nfsd_net *nn = net_generic(ses->se_client->net,
1999					  nfsd_net_id);
2000	int ret = 0;
2001
2002	if (ses->se_target_maxslots <= 1)
2003		return ret;
2004	if (!spin_trylock(&nn->client_lock))
2005		return ret;
2006	ret = min(dec, ses->se_target_maxslots-1);
2007	ses->se_target_maxslots -= ret;
2008	atomic_sub(ret, &nfsd_total_target_slots);
2009	ses->se_slot_gen += 1;
2010	if (ses->se_slot_gen == 0) {
2011		int i;
2012		ses->se_slot_gen = 1;
2013		for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
2014			struct nfsd4_slot *slot = xa_load(&ses->se_slots, i);
2015			slot->sl_generation = 0;
2016		}
2017	}
2018	spin_unlock(&nn->client_lock);
2019	return ret;
2020}
2021
2022static struct nfsd4_slot *nfsd4_alloc_slot(struct nfsd4_channel_attrs *fattrs,
2023					   int index, gfp_t gfp)
2024{
2025	struct nfsd4_slot *slot;
2026	size_t size;
2027
2028	/*
2029	 * The RPC and NFS session headers are never saved in
2030	 * the slot reply cache buffer.
2031	 */
2032	size = fattrs->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ ?
2033		0 : fattrs->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
2034
2035	slot = kzalloc(struct_size(slot, sl_data, size), gfp);
2036	if (!slot)
2037		return NULL;
2038	slot->sl_index = index;
2039	return slot;
2040}
2041
2042static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
2043					   struct nfsd4_channel_attrs *battrs)
2044{
2045	int numslots = fattrs->maxreqs;
2046	struct nfsd4_session *new;
2047	struct nfsd4_slot *slot;
2048	int i;
2049
2050	new = kzalloc(sizeof(*new), GFP_KERNEL);
2051	if (!new)
2052		return NULL;
2053	xa_init(&new->se_slots);
2054
2055	slot = nfsd4_alloc_slot(fattrs, 0, GFP_KERNEL);
2056	if (!slot || xa_is_err(xa_store(&new->se_slots, 0, slot, GFP_KERNEL)))
2057		goto out_free;
2058
2059	for (i = 1; i < numslots; i++) {
2060		const gfp_t gfp = GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN;
2061		slot = nfsd4_alloc_slot(fattrs, i, gfp);
2062		if (!slot)
2063			break;
2064		if (xa_is_err(xa_store(&new->se_slots, i, slot, gfp))) {
2065			kfree(slot);
2066			break;
2067		}
2068	}
2069	fattrs->maxreqs = i;
2070	memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
2071	new->se_target_maxslots = i;
2072	atomic_add(i - 1, &nfsd_total_target_slots);
2073	new->se_cb_slot_avail = ~0U;
2074	new->se_cb_highest_slot = min(battrs->maxreqs - 1,
2075				      NFSD_BC_SLOT_TABLE_SIZE - 1);
2076	spin_lock_init(&new->se_lock);
2077	return new;
2078out_free:
2079	kfree(slot);
2080	xa_destroy(&new->se_slots);
2081	kfree(new);
2082	return NULL;
2083}
2084
2085static void free_conn(struct nfsd4_conn *c)
2086{
2087	svc_xprt_put(c->cn_xprt);
2088	kfree(c);
2089}
2090
2091static void nfsd4_conn_lost(struct svc_xpt_user *u)
2092{
2093	struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
2094	struct nfs4_client *clp = c->cn_session->se_client;
2095
2096	trace_nfsd_cb_lost(clp);
2097
2098	spin_lock(&clp->cl_lock);
2099	if (!list_empty(&c->cn_persession)) {
2100		list_del(&c->cn_persession);
2101		free_conn(c);
2102	}
2103	nfsd4_probe_callback(clp);
2104	spin_unlock(&clp->cl_lock);
2105}
2106
2107static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
2108{
2109	struct nfsd4_conn *conn;
2110
2111	conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
2112	if (!conn)
2113		return NULL;
2114	svc_xprt_get(rqstp->rq_xprt);
2115	conn->cn_xprt = rqstp->rq_xprt;
2116	conn->cn_flags = flags;
2117	INIT_LIST_HEAD(&conn->cn_xpt_user.list);
2118	return conn;
2119}
2120
2121static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2122{
2123	conn->cn_session = ses;
2124	list_add(&conn->cn_persession, &ses->se_conns);
2125}
2126
2127static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2128{
2129	struct nfs4_client *clp = ses->se_client;
2130
2131	spin_lock(&clp->cl_lock);
2132	__nfsd4_hash_conn(conn, ses);
2133	spin_unlock(&clp->cl_lock);
2134}
2135
2136static int nfsd4_register_conn(struct nfsd4_conn *conn)
2137{
2138	conn->cn_xpt_user.callback = nfsd4_conn_lost;
2139	return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
2140}
2141
2142static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
2143{
2144	int ret;
2145
2146	nfsd4_hash_conn(conn, ses);
2147	ret = nfsd4_register_conn(conn);
2148	if (ret)
2149		/* oops; xprt is already down: */
2150		nfsd4_conn_lost(&conn->cn_xpt_user);
2151	/* We may have gained or lost a callback channel: */
2152	nfsd4_probe_callback_sync(ses->se_client);
2153}
2154
2155static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
2156{
2157	u32 dir = NFS4_CDFC4_FORE;
2158
2159	if (cses->flags & SESSION4_BACK_CHAN)
2160		dir |= NFS4_CDFC4_BACK;
2161	return alloc_conn(rqstp, dir);
2162}
2163
2164/* must be called under client_lock */
2165static void nfsd4_del_conns(struct nfsd4_session *s)
2166{
2167	struct nfs4_client *clp = s->se_client;
2168	struct nfsd4_conn *c;
2169
2170	spin_lock(&clp->cl_lock);
2171	while (!list_empty(&s->se_conns)) {
2172		c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
2173		list_del_init(&c->cn_persession);
2174		spin_unlock(&clp->cl_lock);
2175
2176		unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
2177		free_conn(c);
2178
2179		spin_lock(&clp->cl_lock);
2180	}
2181	spin_unlock(&clp->cl_lock);
2182}
2183
2184static void __free_session(struct nfsd4_session *ses)
2185{
2186	free_session_slots(ses, 0);
2187	xa_destroy(&ses->se_slots);
2188	kfree(ses);
2189}
2190
2191static void free_session(struct nfsd4_session *ses)
2192{
2193	nfsd4_del_conns(ses);
2194	__free_session(ses);
2195}
2196
2197static unsigned long
2198nfsd_slot_count(struct shrinker *s, struct shrink_control *sc)
2199{
2200	unsigned long cnt = atomic_read(&nfsd_total_target_slots);
2201
2202	return cnt ? cnt : SHRINK_EMPTY;
2203}
2204
2205static unsigned long
2206nfsd_slot_scan(struct shrinker *s, struct shrink_control *sc)
2207{
2208	struct nfsd4_session *ses;
2209	unsigned long scanned = 0;
2210	unsigned long freed = 0;
2211
2212	spin_lock(&nfsd_session_list_lock);
2213	list_for_each_entry(ses, &nfsd_session_list, se_all_sessions) {
2214		freed += reduce_session_slots(ses, 1);
2215		scanned += 1;
2216		if (scanned >= sc->nr_to_scan) {
2217			/* Move starting point for next scan */
2218			list_move(&nfsd_session_list, &ses->se_all_sessions);
2219			break;
2220		}
2221	}
2222	spin_unlock(&nfsd_session_list_lock);
2223	sc->nr_scanned = scanned;
2224	return freed;
2225}
2226
2227static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
2228{
2229	int idx;
2230	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2231
2232	new->se_client = clp;
2233	gen_sessionid(new);
2234
2235	INIT_LIST_HEAD(&new->se_conns);
2236
2237	atomic_set(&new->se_ref, 0);
2238	new->se_dead = false;
2239	new->se_cb_prog = cses->callback_prog;
2240	new->se_cb_sec = cses->cb_sec;
2241
2242	for (idx = 0; idx < NFSD_BC_SLOT_TABLE_SIZE; ++idx)
2243		new->se_cb_seq_nr[idx] = 1;
2244
2245	idx = hash_sessionid(&new->se_sessionid);
2246	list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
2247	spin_lock(&clp->cl_lock);
2248	list_add(&new->se_perclnt, &clp->cl_sessions);
2249	spin_unlock(&clp->cl_lock);
2250
2251	spin_lock(&nfsd_session_list_lock);
2252	list_add_tail(&new->se_all_sessions, &nfsd_session_list);
2253	spin_unlock(&nfsd_session_list_lock);
2254
2255	{
2256		struct sockaddr *sa = svc_addr(rqstp);
2257		/*
2258		 * This is a little silly; with sessions there's no real
2259		 * use for the callback address.  Use the peer address
2260		 * as a reasonable default for now, but consider fixing
2261		 * the rpc client not to require an address in the
2262		 * future:
2263		 */
2264		rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2265		clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2266	}
2267}
2268
2269/* caller must hold client_lock */
2270static struct nfsd4_session *
2271__find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2272{
2273	struct nfsd4_session *elem;
2274	int idx;
2275	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2276
2277	lockdep_assert_held(&nn->client_lock);
2278
2279	dump_sessionid(__func__, sessionid);
2280	idx = hash_sessionid(sessionid);
2281	/* Search in the appropriate list */
2282	list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2283		if (!memcmp(elem->se_sessionid.data, sessionid->data,
2284			    NFS4_MAX_SESSIONID_LEN)) {
2285			return elem;
2286		}
2287	}
2288
2289	dprintk("%s: session not found\n", __func__);
2290	return NULL;
2291}
2292
2293static struct nfsd4_session *
2294find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2295		__be32 *ret)
2296{
2297	struct nfsd4_session *session;
2298	__be32 status = nfserr_badsession;
2299
2300	session = __find_in_sessionid_hashtbl(sessionid, net);
2301	if (!session)
2302		goto out;
2303	status = nfsd4_get_session_locked(session);
2304	if (status)
2305		session = NULL;
2306out:
2307	*ret = status;
2308	return session;
2309}
2310
2311/* caller must hold client_lock */
2312static void
2313unhash_session(struct nfsd4_session *ses)
2314{
2315	struct nfs4_client *clp = ses->se_client;
2316	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2317
2318	lockdep_assert_held(&nn->client_lock);
2319
2320	list_del(&ses->se_hash);
2321	spin_lock(&ses->se_client->cl_lock);
2322	list_del(&ses->se_perclnt);
2323	spin_unlock(&ses->se_client->cl_lock);
2324	spin_lock(&nfsd_session_list_lock);
2325	list_del(&ses->se_all_sessions);
2326	spin_unlock(&nfsd_session_list_lock);
2327}
2328
2329/* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2330static int
2331STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2332{
2333	/*
2334	 * We're assuming the clid was not given out from a boot
2335	 * precisely 2^32 (about 136 years) before this one.  That seems
2336	 * a safe assumption:
2337	 */
2338	if (clid->cl_boot == (u32)nn->boot_time)
2339		return 0;
2340	trace_nfsd_clid_stale(clid);
2341	return 1;
2342}
2343
2344static struct nfs4_client *alloc_client(struct xdr_netobj name,
2345				struct nfsd_net *nn)
2346{
2347	struct nfs4_client *clp;
2348	int i;
2349
2350	if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients &&
2351	    atomic_read(&nn->nfsd_courtesy_clients) > 0)
2352		mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2353
2354	clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2355	if (clp == NULL)
2356		return NULL;
2357	xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2358	if (clp->cl_name.data == NULL)
2359		goto err_no_name;
2360	clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2361						 sizeof(struct list_head),
2362						 GFP_KERNEL);
2363	if (!clp->cl_ownerstr_hashtbl)
2364		goto err_no_hashtbl;
2365	clp->cl_callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
2366	if (!clp->cl_callback_wq)
2367		goto err_no_callback_wq;
2368
2369	for (i = 0; i < OWNER_HASH_SIZE; i++)
2370		INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2371	INIT_LIST_HEAD(&clp->cl_sessions);
2372	idr_init(&clp->cl_stateids);
2373	atomic_set(&clp->cl_rpc_users, 0);
2374	clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2375	clp->cl_state = NFSD4_ACTIVE;
2376	atomic_inc(&nn->nfs4_client_count);
2377	atomic_set(&clp->cl_delegs_in_recall, 0);
2378	INIT_LIST_HEAD(&clp->cl_idhash);
2379	INIT_LIST_HEAD(&clp->cl_openowners);
2380	INIT_LIST_HEAD(&clp->cl_delegations);
2381	INIT_LIST_HEAD(&clp->cl_lru);
2382	INIT_LIST_HEAD(&clp->cl_revoked);
2383#ifdef CONFIG_NFSD_PNFS
2384	INIT_LIST_HEAD(&clp->cl_lo_states);
2385#endif
2386	INIT_LIST_HEAD(&clp->async_copies);
2387	spin_lock_init(&clp->async_lock);
2388	spin_lock_init(&clp->cl_lock);
2389	rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2390	return clp;
2391err_no_callback_wq:
2392	kfree(clp->cl_ownerstr_hashtbl);
2393err_no_hashtbl:
2394	kfree(clp->cl_name.data);
2395err_no_name:
2396	kmem_cache_free(client_slab, clp);
2397	return NULL;
2398}
2399
2400static void __free_client(struct kref *k)
2401{
2402	struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2403	struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2404
2405	free_svc_cred(&clp->cl_cred);
2406	destroy_workqueue(clp->cl_callback_wq);
2407	kfree(clp->cl_ownerstr_hashtbl);
2408	kfree(clp->cl_name.data);
2409	kfree(clp->cl_nii_domain.data);
2410	kfree(clp->cl_nii_name.data);
2411	idr_destroy(&clp->cl_stateids);
2412	kfree(clp->cl_ra);
2413	kmem_cache_free(client_slab, clp);
2414}
2415
2416static void drop_client(struct nfs4_client *clp)
2417{
2418	kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2419}
2420
2421static void
2422free_client(struct nfs4_client *clp)
2423{
2424	while (!list_empty(&clp->cl_sessions)) {
2425		struct nfsd4_session *ses;
2426		ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2427				se_perclnt);
2428		list_del(&ses->se_perclnt);
2429		WARN_ON_ONCE(atomic_read(&ses->se_ref));
2430		free_session(ses);
2431	}
2432	rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2433	if (clp->cl_nfsd_dentry) {
2434		nfsd_client_rmdir(clp->cl_nfsd_dentry);
2435		clp->cl_nfsd_dentry = NULL;
2436		wake_up_all(&expiry_wq);
2437	}
2438	drop_client(clp);
2439}
2440
2441/* must be called under the client_lock */
2442static void
2443unhash_client_locked(struct nfs4_client *clp)
2444{
2445	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2446	struct nfsd4_session *ses;
2447
2448	lockdep_assert_held(&nn->client_lock);
2449
2450	/* Mark the client as expired! */
2451	clp->cl_time = 0;
2452	/* Make it invisible */
2453	if (!list_empty(&clp->cl_idhash)) {
2454		list_del_init(&clp->cl_idhash);
2455		if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2456			rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2457		else
2458			rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2459	}
2460	list_del_init(&clp->cl_lru);
2461	spin_lock(&clp->cl_lock);
2462	spin_lock(&nfsd_session_list_lock);
2463	list_for_each_entry(ses, &clp->cl_sessions, se_perclnt) {
2464		list_del_init(&ses->se_hash);
2465		list_del_init(&ses->se_all_sessions);
2466	}
2467	spin_unlock(&nfsd_session_list_lock);
2468	spin_unlock(&clp->cl_lock);
2469}
2470
2471static void
2472unhash_client(struct nfs4_client *clp)
2473{
2474	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2475
2476	spin_lock(&nn->client_lock);
2477	unhash_client_locked(clp);
2478	spin_unlock(&nn->client_lock);
2479}
2480
2481static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2482{
2483	int users = atomic_read(&clp->cl_rpc_users);
2484
2485	trace_nfsd_mark_client_expired(clp, users);
2486
2487	if (users)
2488		return nfserr_jukebox;
2489	unhash_client_locked(clp);
2490	return nfs_ok;
2491}
2492
2493static void
2494__destroy_client(struct nfs4_client *clp)
2495{
2496	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2497	int i;
2498	struct nfs4_openowner *oo;
2499	struct nfs4_delegation *dp;
2500	LIST_HEAD(reaplist);
2501
2502	spin_lock(&state_lock);
2503	while (!list_empty(&clp->cl_delegations)) {
2504		dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2505		unhash_delegation_locked(dp, SC_STATUS_CLOSED);
2506		list_add(&dp->dl_recall_lru, &reaplist);
2507	}
2508	spin_unlock(&state_lock);
2509	while (!list_empty(&reaplist)) {
2510		dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2511		list_del_init(&dp->dl_recall_lru);
2512		destroy_unhashed_deleg(dp);
2513	}
2514	while (!list_empty(&clp->cl_revoked)) {
2515		dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2516		list_del_init(&dp->dl_recall_lru);
2517		nfs4_put_stid(&dp->dl_stid);
2518	}
2519	while (!list_empty(&clp->cl_openowners)) {
2520		oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2521		nfs4_get_stateowner(&oo->oo_owner);
2522		release_openowner(oo);
2523	}
2524	for (i = 0; i < OWNER_HASH_SIZE; i++) {
2525		struct nfs4_stateowner *so, *tmp;
2526
2527		list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2528					 so_strhash) {
2529			/* Should be no openowners at this point */
2530			WARN_ON_ONCE(so->so_is_open_owner);
2531			remove_blocked_locks(lockowner(so));
2532		}
2533	}
2534	nfsd4_return_all_client_layouts(clp);
2535	nfsd4_shutdown_copy(clp);
2536	nfsd4_shutdown_callback(clp);
2537	if (clp->cl_cb_conn.cb_xprt)
2538		svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2539	atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2540	nfsd4_dec_courtesy_client_count(nn, clp);
2541	free_client(clp);
2542	wake_up_all(&expiry_wq);
2543}
2544
2545static void
2546destroy_client(struct nfs4_client *clp)
2547{
2548	unhash_client(clp);
2549	__destroy_client(clp);
2550}
2551
2552static void inc_reclaim_complete(struct nfs4_client *clp)
2553{
2554	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2555
2556	if (!nn->track_reclaim_completes)
2557		return;
2558	if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2559		return;
2560	if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2561			nn->reclaim_str_hashtbl_size) {
2562		printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2563				clp->net->ns.inum);
2564		nfsd4_end_grace(nn);
2565	}
2566}
2567
2568static void expire_client(struct nfs4_client *clp)
2569{
2570	unhash_client(clp);
2571	nfsd4_client_record_remove(clp);
2572	__destroy_client(clp);
2573}
2574
2575static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2576{
2577	memcpy(target->cl_verifier.data, source->data,
2578			sizeof(target->cl_verifier.data));
2579}
2580
2581static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2582{
2583	target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
2584	target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
2585}
2586
2587static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2588{
2589	target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2590	target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2591								GFP_KERNEL);
2592	target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2593	if ((source->cr_principal && !target->cr_principal) ||
2594	    (source->cr_raw_principal && !target->cr_raw_principal) ||
2595	    (source->cr_targ_princ && !target->cr_targ_princ))
2596		return -ENOMEM;
2597
2598	target->cr_flavor = source->cr_flavor;
2599	target->cr_uid = source->cr_uid;
2600	target->cr_gid = source->cr_gid;
2601	target->cr_group_info = source->cr_group_info;
2602	get_group_info(target->cr_group_info);
2603	target->cr_gss_mech = source->cr_gss_mech;
2604	if (source->cr_gss_mech)
2605		gss_mech_get(source->cr_gss_mech);
2606	return 0;
2607}
2608
2609static int
2610compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2611{
2612	if (o1->len < o2->len)
2613		return -1;
2614	if (o1->len > o2->len)
2615		return 1;
2616	return memcmp(o1->data, o2->data, o1->len);
2617}
2618
2619static int
2620same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2621{
2622	return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2623}
2624
2625static int
2626same_clid(clientid_t *cl1, clientid_t *cl2)
2627{
2628	return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2629}
2630
2631static bool groups_equal(struct group_info *g1, struct group_info *g2)
2632{
2633	int i;
2634
2635	if (g1->ngroups != g2->ngroups)
2636		return false;
2637	for (i=0; i<g1->ngroups; i++)
2638		if (!gid_eq(g1->gid[i], g2->gid[i]))
2639			return false;
2640	return true;
2641}
2642
2643/*
2644 * RFC 3530 language requires clid_inuse be returned when the
2645 * "principal" associated with a requests differs from that previously
2646 * used.  We use uid, gid's, and gss principal string as our best
2647 * approximation.  We also don't want to allow non-gss use of a client
2648 * established using gss: in theory cr_principal should catch that
2649 * change, but in practice cr_principal can be null even in the gss case
2650 * since gssd doesn't always pass down a principal string.
2651 */
2652static bool is_gss_cred(struct svc_cred *cr)
2653{
2654	/* Is cr_flavor one of the gss "pseudoflavors"?: */
2655	return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2656}
2657
2658
2659static bool
2660same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2661{
2662	if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2663		|| (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2664		|| (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2665		|| !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2666		return false;
2667	/* XXX: check that cr_targ_princ fields match ? */
2668	if (cr1->cr_principal == cr2->cr_principal)
2669		return true;
2670	if (!cr1->cr_principal || !cr2->cr_principal)
2671		return false;
2672	return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2673}
2674
2675static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2676{
2677	struct svc_cred *cr = &rqstp->rq_cred;
2678	u32 service;
2679
2680	if (!cr->cr_gss_mech)
2681		return false;
2682	service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2683	return service == RPC_GSS_SVC_INTEGRITY ||
2684	       service == RPC_GSS_SVC_PRIVACY;
2685}
2686
2687bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2688{
2689	struct svc_cred *cr = &rqstp->rq_cred;
2690
2691	if (!cl->cl_mach_cred)
2692		return true;
2693	if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2694		return false;
2695	if (!svc_rqst_integrity_protected(rqstp))
2696		return false;
2697	if (cl->cl_cred.cr_raw_principal)
2698		return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2699						cr->cr_raw_principal);
2700	if (!cr->cr_principal)
2701		return false;
2702	return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2703}
2704
2705static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2706{
2707	__be32 verf[2];
2708
2709	/*
2710	 * This is opaque to client, so no need to byte-swap. Use
2711	 * __force to keep sparse happy
2712	 */
2713	verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2714	verf[1] = (__force __be32)nn->clverifier_counter++;
2715	memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2716}
2717
2718static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2719{
2720	clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2721	clp->cl_clientid.cl_id = nn->clientid_counter++;
2722	gen_confirm(clp, nn);
2723}
2724
2725static struct nfs4_stid *
2726find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2727{
2728	struct nfs4_stid *ret;
2729
2730	ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2731	if (!ret || !ret->sc_type)
2732		return NULL;
2733	return ret;
2734}
2735
2736static struct nfs4_stid *
2737find_stateid_by_type(struct nfs4_client *cl, stateid_t *t,
2738		     unsigned short typemask, unsigned short ok_states)
2739{
2740	struct nfs4_stid *s;
2741
2742	spin_lock(&cl->cl_lock);
2743	s = find_stateid_locked(cl, t);
2744	if (s != NULL) {
2745		if ((s->sc_status & ~ok_states) == 0 &&
2746		    (typemask & s->sc_type))
2747			refcount_inc(&s->sc_count);
2748		else
2749			s = NULL;
2750	}
2751	spin_unlock(&cl->cl_lock);
2752	return s;
2753}
2754
2755static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2756{
2757	struct nfsdfs_client *nc;
2758	nc = get_nfsdfs_client(inode);
2759	if (!nc)
2760		return NULL;
2761	return container_of(nc, struct nfs4_client, cl_nfsdfs);
2762}
2763
2764static void seq_quote_mem(struct seq_file *m, char *data, int len)
2765{
2766	seq_puts(m, "\"");
2767	seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2768	seq_puts(m, "\"");
2769}
2770
2771static const char *cb_state2str(int state)
2772{
2773	switch (state) {
2774	case NFSD4_CB_UP:
2775		return "UP";
2776	case NFSD4_CB_UNKNOWN:
2777		return "UNKNOWN";
2778	case NFSD4_CB_DOWN:
2779		return "DOWN";
2780	case NFSD4_CB_FAULT:
2781		return "FAULT";
2782	}
2783	return "UNDEFINED";
2784}
2785
2786static int client_info_show(struct seq_file *m, void *v)
2787{
2788	struct inode *inode = file_inode(m->file);
2789	struct nfsd4_session *ses;
2790	struct nfs4_client *clp;
2791	u64 clid;
2792
2793	clp = get_nfsdfs_clp(inode);
2794	if (!clp)
2795		return -ENXIO;
2796	memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2797	seq_printf(m, "clientid: 0x%llx\n", clid);
2798	seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2799
2800	if (clp->cl_state == NFSD4_COURTESY)
2801		seq_puts(m, "status: courtesy\n");
2802	else if (clp->cl_state == NFSD4_EXPIRABLE)
2803		seq_puts(m, "status: expirable\n");
2804	else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2805		seq_puts(m, "status: confirmed\n");
2806	else
2807		seq_puts(m, "status: unconfirmed\n");
2808	seq_printf(m, "seconds from last renew: %lld\n",
2809		ktime_get_boottime_seconds() - clp->cl_time);
2810	seq_puts(m, "name: ");
2811	seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2812	seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2813	if (clp->cl_nii_domain.data) {
2814		seq_puts(m, "Implementation domain: ");
2815		seq_quote_mem(m, clp->cl_nii_domain.data,
2816					clp->cl_nii_domain.len);
2817		seq_puts(m, "\nImplementation name: ");
2818		seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2819		seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2820			clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2821	}
2822	seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2823	seq_printf(m, "callback address: \"%pISpc\"\n", &clp->cl_cb_conn.cb_addr);
2824	seq_printf(m, "admin-revoked states: %d\n",
2825		   atomic_read(&clp->cl_admin_revoked));
2826	spin_lock(&clp->cl_lock);
2827	seq_printf(m, "session slots:");
2828	list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2829		seq_printf(m, " %u", ses->se_fchannel.maxreqs);
2830	seq_printf(m, "\nsession target slots:");
2831	list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2832		seq_printf(m, " %u", ses->se_target_maxslots);
2833	spin_unlock(&clp->cl_lock);
2834	seq_puts(m, "\n");
2835
2836	drop_client(clp);
2837
2838	return 0;
2839}
2840
2841DEFINE_SHOW_ATTRIBUTE(client_info);
2842
2843static void *states_start(struct seq_file *s, loff_t *pos)
2844	__acquires(&clp->cl_lock)
2845{
2846	struct nfs4_client *clp = s->private;
2847	unsigned long id = *pos;
2848	void *ret;
2849
2850	spin_lock(&clp->cl_lock);
2851	ret = idr_get_next_ul(&clp->cl_stateids, &id);
2852	*pos = id;
2853	return ret;
2854}
2855
2856static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2857{
2858	struct nfs4_client *clp = s->private;
2859	unsigned long id = *pos;
2860	void *ret;
2861
2862	id = *pos;
2863	id++;
2864	ret = idr_get_next_ul(&clp->cl_stateids, &id);
2865	*pos = id;
2866	return ret;
2867}
2868
2869static void states_stop(struct seq_file *s, void *v)
2870	__releases(&clp->cl_lock)
2871{
2872	struct nfs4_client *clp = s->private;
2873
2874	spin_unlock(&clp->cl_lock);
2875}
2876
2877static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2878{
2879         seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2880}
2881
2882static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2883{
2884	struct inode *inode = file_inode(f->nf_file);
2885
2886	seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2887					MAJOR(inode->i_sb->s_dev),
2888					 MINOR(inode->i_sb->s_dev),
2889					 inode->i_ino);
2890}
2891
2892static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2893{
2894	seq_puts(s, "owner: ");
2895	seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2896}
2897
2898static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2899{
2900	seq_printf(s, "0x%.8x", stid->si_generation);
2901	seq_printf(s, "%12phN", &stid->si_opaque);
2902}
2903
2904static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2905{
2906	struct nfs4_ol_stateid *ols;
2907	struct nfs4_file *nf;
2908	struct nfsd_file *file;
2909	struct nfs4_stateowner *oo;
2910	unsigned int access, deny;
2911
2912	ols = openlockstateid(st);
2913	oo = ols->st_stateowner;
2914	nf = st->sc_file;
2915
2916	seq_puts(s, "- ");
2917	nfs4_show_stateid(s, &st->sc_stateid);
2918	seq_puts(s, ": { type: open, ");
2919
2920	access = bmap_to_share_mode(ols->st_access_bmap);
2921	deny   = bmap_to_share_mode(ols->st_deny_bmap);
2922
2923	seq_printf(s, "access: %s%s, ",
2924		access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2925		access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2926	seq_printf(s, "deny: %s%s, ",
2927		deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2928		deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2929
2930	if (nf) {
2931		spin_lock(&nf->fi_lock);
2932		file = find_any_file_locked(nf);
2933		if (file) {
2934			nfs4_show_superblock(s, file);
2935			seq_puts(s, ", ");
2936			nfs4_show_fname(s, file);
2937			seq_puts(s, ", ");
2938		}
2939		spin_unlock(&nf->fi_lock);
2940	} else
2941		seq_puts(s, "closed, ");
2942	nfs4_show_owner(s, oo);
2943	if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2944		seq_puts(s, ", admin-revoked");
2945	seq_puts(s, " }\n");
2946	return 0;
2947}
2948
2949static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2950{
2951	struct nfs4_ol_stateid *ols;
2952	struct nfs4_file *nf;
2953	struct nfsd_file *file;
2954	struct nfs4_stateowner *oo;
2955
2956	ols = openlockstateid(st);
2957	oo = ols->st_stateowner;
2958	nf = st->sc_file;
2959
2960	seq_puts(s, "- ");
2961	nfs4_show_stateid(s, &st->sc_stateid);
2962	seq_puts(s, ": { type: lock, ");
2963
2964	spin_lock(&nf->fi_lock);
2965	file = find_any_file_locked(nf);
2966	if (file) {
2967		/*
2968		 * Note: a lock stateid isn't really the same thing as a lock,
2969		 * it's the locking state held by one owner on a file, and there
2970		 * may be multiple (or no) lock ranges associated with it.
2971		 * (Same for the matter is true of open stateids.)
2972		 */
2973
2974		nfs4_show_superblock(s, file);
2975		/* XXX: open stateid? */
2976		seq_puts(s, ", ");
2977		nfs4_show_fname(s, file);
2978		seq_puts(s, ", ");
2979	}
2980	nfs4_show_owner(s, oo);
2981	if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2982		seq_puts(s, ", admin-revoked");
2983	seq_puts(s, " }\n");
2984	spin_unlock(&nf->fi_lock);
2985	return 0;
2986}
2987
2988static char *nfs4_show_deleg_type(u32 dl_type)
2989{
2990	switch (dl_type) {
2991	case OPEN_DELEGATE_READ:
2992		return "r";
2993	case OPEN_DELEGATE_WRITE:
2994		return "w";
2995	case OPEN_DELEGATE_READ_ATTRS_DELEG:
2996		return "ra";
2997	case OPEN_DELEGATE_WRITE_ATTRS_DELEG:
2998		return "wa";
2999	}
3000	return "?";
3001}
3002
3003static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
3004{
3005	struct nfs4_delegation *ds;
3006	struct nfs4_file *nf;
3007	struct nfsd_file *file;
3008
3009	ds = delegstateid(st);
3010	nf = st->sc_file;
3011
3012	seq_puts(s, "- ");
3013	nfs4_show_stateid(s, &st->sc_stateid);
3014	seq_puts(s, ": { type: deleg, ");
3015
3016	seq_printf(s, "access: %s", nfs4_show_deleg_type(ds->dl_type));
3017
3018	/* XXX: lease time, whether it's being recalled. */
3019
3020	spin_lock(&nf->fi_lock);
3021	file = nf->fi_deleg_file;
3022	if (file) {
3023		seq_puts(s, ", ");
3024		nfs4_show_superblock(s, file);
3025		seq_puts(s, ", ");
3026		nfs4_show_fname(s, file);
3027	}
3028	spin_unlock(&nf->fi_lock);
3029	if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
3030		seq_puts(s, ", admin-revoked");
3031	seq_puts(s, " }\n");
3032	return 0;
3033}
3034
3035static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
3036{
3037	struct nfs4_layout_stateid *ls;
3038	struct nfsd_file *file;
3039
3040	ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
3041
3042	seq_puts(s, "- ");
3043	nfs4_show_stateid(s, &st->sc_stateid);
3044	seq_puts(s, ": { type: layout");
3045
3046	/* XXX: What else would be useful? */
3047
3048	spin_lock(&ls->ls_stid.sc_file->fi_lock);
3049	file = ls->ls_file;
3050	if (file) {
3051		seq_puts(s, ", ");
3052		nfs4_show_superblock(s, file);
3053		seq_puts(s, ", ");
3054		nfs4_show_fname(s, file);
3055	}
3056	spin_unlock(&ls->ls_stid.sc_file->fi_lock);
3057	if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
3058		seq_puts(s, ", admin-revoked");
3059	seq_puts(s, " }\n");
3060
3061	return 0;
3062}
3063
3064static int states_show(struct seq_file *s, void *v)
3065{
3066	struct nfs4_stid *st = v;
3067
3068	switch (st->sc_type) {
3069	case SC_TYPE_OPEN:
3070		return nfs4_show_open(s, st);
3071	case SC_TYPE_LOCK:
3072		return nfs4_show_lock(s, st);
3073	case SC_TYPE_DELEG:
3074		return nfs4_show_deleg(s, st);
3075	case SC_TYPE_LAYOUT:
3076		return nfs4_show_layout(s, st);
3077	default:
3078		return 0; /* XXX: or SEQ_SKIP? */
3079	}
3080	/* XXX: copy stateids? */
3081}
3082
3083static struct seq_operations states_seq_ops = {
3084	.start = states_start,
3085	.next = states_next,
3086	.stop = states_stop,
3087	.show = states_show
3088};
3089
3090static int client_states_open(struct inode *inode, struct file *file)
3091{
3092	struct seq_file *s;
3093	struct nfs4_client *clp;
3094	int ret;
3095
3096	clp = get_nfsdfs_clp(inode);
3097	if (!clp)
3098		return -ENXIO;
3099
3100	ret = seq_open(file, &states_seq_ops);
3101	if (ret) {
3102		drop_client(clp);
3103		return ret;
3104	}
3105	s = file->private_data;
3106	s->private = clp;
3107	return 0;
3108}
3109
3110static int client_opens_release(struct inode *inode, struct file *file)
3111{
3112	struct seq_file *m = file->private_data;
3113	struct nfs4_client *clp = m->private;
3114
3115	/* XXX: alternatively, we could get/drop in seq start/stop */
3116	drop_client(clp);
3117	return seq_release(inode, file);
3118}
3119
3120static const struct file_operations client_states_fops = {
3121	.open		= client_states_open,
3122	.read		= seq_read,
3123	.llseek		= seq_lseek,
3124	.release	= client_opens_release,
3125};
3126
3127/*
3128 * Normally we refuse to destroy clients that are in use, but here the
3129 * administrator is telling us to just do it.  We also want to wait
3130 * so the caller has a guarantee that the client's locks are gone by
3131 * the time the write returns:
3132 */
3133static void force_expire_client(struct nfs4_client *clp)
3134{
3135	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3136	bool already_expired;
3137
3138	trace_nfsd_clid_admin_expired(&clp->cl_clientid);
3139
3140	spin_lock(&nn->client_lock);
3141	clp->cl_time = 0;
3142	spin_unlock(&nn->client_lock);
3143
3144	wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
3145	spin_lock(&nn->client_lock);
3146	already_expired = list_empty(&clp->cl_lru);
3147	if (!already_expired)
3148		unhash_client_locked(clp);
3149	spin_unlock(&nn->client_lock);
3150
3151	if (!already_expired)
3152		expire_client(clp);
3153	else
3154		wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
3155}
3156
3157static ssize_t client_ctl_write(struct file *file, const char __user *buf,
3158				   size_t size, loff_t *pos)
3159{
3160	char *data;
3161	struct nfs4_client *clp;
3162
3163	data = simple_transaction_get(file, buf, size);
3164	if (IS_ERR(data))
3165		return PTR_ERR(data);
3166	if (size != 7 || 0 != memcmp(data, "expire\n", 7))
3167		return -EINVAL;
3168	clp = get_nfsdfs_clp(file_inode(file));
3169	if (!clp)
3170		return -ENXIO;
3171	force_expire_client(clp);
3172	drop_client(clp);
3173	return 7;
3174}
3175
3176static const struct file_operations client_ctl_fops = {
3177	.write		= client_ctl_write,
3178	.release	= simple_transaction_release,
3179};
3180
3181static const struct tree_descr client_files[] = {
3182	[0] = {"info", &client_info_fops, S_IRUSR},
3183	[1] = {"states", &client_states_fops, S_IRUSR},
3184	[2] = {"ctl", &client_ctl_fops, S_IWUSR},
3185	[3] = {""},
3186};
3187
3188static int
3189nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
3190				struct rpc_task *task)
3191{
3192	trace_nfsd_cb_recall_any_done(cb, task);
3193	switch (task->tk_status) {
3194	case -NFS4ERR_DELAY:
3195		rpc_delay(task, 2 * HZ);
3196		return 0;
3197	default:
3198		return 1;
3199	}
3200}
3201
3202static void
3203nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
3204{
3205	struct nfs4_client *clp = cb->cb_clp;
3206
3207	drop_client(clp);
3208}
3209
3210static int
3211nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3212{
3213	struct nfs4_cb_fattr *ncf =
3214			container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3215	struct nfs4_delegation *dp =
3216			container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3217
3218	trace_nfsd_cb_getattr_done(&dp->dl_stid.sc_stateid, task);
3219	ncf->ncf_cb_status = task->tk_status;
3220	switch (task->tk_status) {
3221	case -NFS4ERR_DELAY:
3222		rpc_delay(task, 2 * HZ);
3223		return 0;
3224	default:
3225		return 1;
3226	}
3227}
3228
3229static void
3230nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3231{
3232	struct nfs4_cb_fattr *ncf =
3233			container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3234	struct nfs4_delegation *dp =
3235			container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3236
3237	nfs4_put_stid(&dp->dl_stid);
3238}
3239
3240static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3241	.done		= nfsd4_cb_recall_any_done,
3242	.release	= nfsd4_cb_recall_any_release,
3243	.opcode		= OP_CB_RECALL_ANY,
3244};
3245
3246static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3247	.done		= nfsd4_cb_getattr_done,
3248	.release	= nfsd4_cb_getattr_release,
3249	.opcode		= OP_CB_GETATTR,
3250};
3251
3252static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3253{
3254	struct nfs4_delegation *dp =
3255			container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3256
3257	if (test_and_set_bit(NFSD4_CALLBACK_RUNNING, &ncf->ncf_getattr.cb_flags))
3258		return;
3259
3260	/* set to proper status when nfsd4_cb_getattr_done runs */
3261	ncf->ncf_cb_status = NFS4ERR_IO;
3262
3263	/* ensure that wake_bit is done when RUNNING is cleared */
3264	set_bit(NFSD4_CALLBACK_WAKE, &ncf->ncf_getattr.cb_flags);
3265
3266	refcount_inc(&dp->dl_stid.sc_count);
3267	nfsd4_run_cb(&ncf->ncf_getattr);
3268}
3269
3270static struct nfs4_client *create_client(struct xdr_netobj name,
3271		struct svc_rqst *rqstp, nfs4_verifier *verf)
3272{
3273	struct nfs4_client *clp;
3274	struct sockaddr *sa = svc_addr(rqstp);
3275	int ret;
3276	struct net *net = SVC_NET(rqstp);
3277	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3278	struct dentry *dentries[ARRAY_SIZE(client_files)];
3279
3280	clp = alloc_client(name, nn);
3281	if (clp == NULL)
3282		return NULL;
3283
3284	ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3285	if (ret) {
3286		free_client(clp);
3287		return NULL;
3288	}
3289	gen_clid(clp, nn);
3290	kref_init(&clp->cl_nfsdfs.cl_ref);
3291	nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3292	clp->cl_time = ktime_get_boottime_seconds();
3293	copy_verf(clp, verf);
3294	memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3295	clp->cl_cb_session = NULL;
3296	clp->net = net;
3297	clp->cl_nfsd_dentry = nfsd_client_mkdir(
3298		nn, &clp->cl_nfsdfs,
3299		clp->cl_clientid.cl_id - nn->clientid_base,
3300		client_files, dentries);
3301	clp->cl_nfsd_info_dentry = dentries[0];
3302	if (!clp->cl_nfsd_dentry) {
3303		free_client(clp);
3304		return NULL;
3305	}
3306	clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3307	if (!clp->cl_ra) {
3308		free_client(clp);
3309		return NULL;
3310	}
3311	clp->cl_ra_time = 0;
3312	nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3313			NFSPROC4_CLNT_CB_RECALL_ANY);
3314	return clp;
3315}
3316
3317static void
3318add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3319{
3320	struct rb_node **new = &(root->rb_node), *parent = NULL;
3321	struct nfs4_client *clp;
3322
3323	while (*new) {
3324		clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3325		parent = *new;
3326
3327		if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3328			new = &((*new)->rb_left);
3329		else
3330			new = &((*new)->rb_right);
3331	}
3332
3333	rb_link_node(&new_clp->cl_namenode, parent, new);
3334	rb_insert_color(&new_clp->cl_namenode, root);
3335}
3336
3337static struct nfs4_client *
3338find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3339{
3340	int cmp;
3341	struct rb_node *node = root->rb_node;
3342	struct nfs4_client *clp;
3343
3344	while (node) {
3345		clp = rb_entry(node, struct nfs4_client, cl_namenode);
3346		cmp = compare_blob(&clp->cl_name, name);
3347		if (cmp > 0)
3348			node = node->rb_left;
3349		else if (cmp < 0)
3350			node = node->rb_right;
3351		else
3352			return clp;
3353	}
3354	return NULL;
3355}
3356
3357static void
3358add_to_unconfirmed(struct nfs4_client *clp)
3359{
3360	unsigned int idhashval;
3361	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3362
3363	lockdep_assert_held(&nn->client_lock);
3364
3365	clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3366	add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3367	idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3368	list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3369	renew_client_locked(clp);
3370}
3371
3372static void
3373move_to_confirmed(struct nfs4_client *clp)
3374{
3375	unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3376	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3377
3378	lockdep_assert_held(&nn->client_lock);
3379
3380	list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3381	rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3382	add_clp_to_name_tree(clp, &nn->conf_name_tree);
3383	set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3384	trace_nfsd_clid_confirmed(&clp->cl_clientid);
3385	renew_client_locked(clp);
3386}
3387
3388static struct nfs4_client *
3389find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3390{
3391	struct nfs4_client *clp;
3392	unsigned int idhashval = clientid_hashval(clid->cl_id);
3393
3394	list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3395		if (same_clid(&clp->cl_clientid, clid)) {
3396			if ((bool)clp->cl_minorversion != sessions)
3397				return NULL;
3398			renew_client_locked(clp);
3399			return clp;
3400		}
3401	}
3402	return NULL;
3403}
3404
3405static struct nfs4_client *
3406find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3407{
3408	struct list_head *tbl = nn->conf_id_hashtbl;
3409
3410	lockdep_assert_held(&nn->client_lock);
3411	return find_client_in_id_table(tbl, clid, sessions);
3412}
3413
3414static struct nfs4_client *
3415find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3416{
3417	struct list_head *tbl = nn->unconf_id_hashtbl;
3418
3419	lockdep_assert_held(&nn->client_lock);
3420	return find_client_in_id_table(tbl, clid, sessions);
3421}
3422
3423static bool clp_used_exchangeid(struct nfs4_client *clp)
3424{
3425	return clp->cl_exchange_flags != 0;
3426} 
3427
3428static struct nfs4_client *
3429find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3430{
3431	lockdep_assert_held(&nn->client_lock);
3432	return find_clp_in_name_tree(name, &nn->conf_name_tree);
3433}
3434
3435static struct nfs4_client *
3436find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3437{
3438	lockdep_assert_held(&nn->client_lock);
3439	return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3440}
3441
3442static void
3443gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3444{
3445	struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3446	struct sockaddr	*sa = svc_addr(rqstp);
3447	u32 scopeid = rpc_get_scope_id(sa);
3448	unsigned short expected_family;
3449
3450	/* Currently, we only support tcp and tcp6 for the callback channel */
3451	if (se->se_callback_netid_len == 3 &&
3452	    !memcmp(se->se_callback_netid_val, "tcp", 3))
3453		expected_family = AF_INET;
3454	else if (se->se_callback_netid_len == 4 &&
3455		 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3456		expected_family = AF_INET6;
3457	else
3458		goto out_err;
3459
3460	conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3461					    se->se_callback_addr_len,
3462					    (struct sockaddr *)&conn->cb_addr,
3463					    sizeof(conn->cb_addr));
3464
3465	if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3466		goto out_err;
3467
3468	if (conn->cb_addr.ss_family == AF_INET6)
3469		((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3470
3471	conn->cb_prog = se->se_callback_prog;
3472	conn->cb_ident = se->se_callback_ident;
3473	memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3474	trace_nfsd_cb_args(clp, conn);
3475	return;
3476out_err:
3477	conn->cb_addr.ss_family = AF_UNSPEC;
3478	conn->cb_addrlen = 0;
3479	trace_nfsd_cb_nodelegs(clp);
3480	return;
3481}
3482
3483/*
3484 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3485 */
3486static void
3487nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3488{
3489	struct xdr_buf *buf = resp->xdr->buf;
3490	struct nfsd4_slot *slot = resp->cstate.slot;
3491	unsigned int base;
3492
3493	/*
3494	 * RFC 5661 Section 2.10.6.1.2:
3495	 *
3496	 * Any time SEQUENCE ... returns an error ... [t]he replier MUST NOT
3497	 * modify the reply cache entry for the slot whenever an error is
3498	 * returned from SEQUENCE ...
3499	 *
3500	 * Because nfsd4_store_cache_entry is called only by
3501	 * nfsd4_sequence_done(), nfsd4_store_cache_entry() is called only
3502	 * when a SEQUENCE operation was part of the COMPOUND.
3503	 * nfs41_check_op_ordering() ensures SEQUENCE is the first op.
3504	 */
3505	if (resp->opcnt == 1 && resp->cstate.status != nfs_ok)
3506		return;
3507
3508	slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3509	slot->sl_opcnt = resp->opcnt;
3510	slot->sl_status = resp->cstate.status;
3511	free_svc_cred(&slot->sl_cred);
3512	copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3513
3514	if (!(resp->cstate.slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
3515		slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3516		return;
3517	}
3518	slot->sl_flags |= NFSD4_SLOT_CACHED;
3519
3520	base = resp->cstate.data_offset;
3521	slot->sl_datalen = buf->len - base;
3522	if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3523		WARN(1, "%s: sessions DRC could not cache compound\n",
3524		     __func__);
3525	return;
3526}
3527
3528/*
3529 * The sequence operation is not cached because we can use the slot and
3530 * session values.
3531 */
3532static __be32
3533nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3534			 struct nfsd4_sequence *seq)
3535{
3536	struct nfsd4_compoundargs *args = resp->rqstp->rq_argp;
3537	struct nfsd4_slot *slot = resp->cstate.slot;
3538	struct xdr_stream *xdr = resp->xdr;
3539	__be32 *p;
3540
3541	dprintk("--> %s slot %p\n", __func__, slot);
3542
3543	/* Always encode the SEQUENCE response. */
3544	nfsd4_encode_operation(resp, &args->ops[0]);
3545	if (args->opcnt == 1)
3546		/* A solo SEQUENCE - nothing was cached */
3547		return args->ops[0].status;
3548
3549	if (!(slot->sl_flags & NFSD4_SLOT_CACHED)) {
3550		/* We weren't asked to cache this. */
3551		struct nfsd4_op *op;
3552
3553		op = &args->ops[resp->opcnt++];
3554		op->status = nfserr_retry_uncached_rep;
3555		nfsd4_encode_operation(resp, op);
3556		return op->status;
3557	}
3558
3559	/* return reply from cache */
3560	p = xdr_reserve_space(xdr, slot->sl_datalen);
3561	if (!p) {
3562		WARN_ON_ONCE(1);
3563		return nfserr_serverfault;
3564	}
3565	xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3566	xdr_commit_encode(xdr);
3567
3568	resp->opcnt = slot->sl_opcnt;
3569	return slot->sl_status;
3570}
3571
3572/*
3573 * Set the exchange_id flags returned by the server.
3574 */
3575static void
3576nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3577{
3578#ifdef CONFIG_NFSD_PNFS
3579	new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3580#else
3581	new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3582#endif
3583
3584	/* Referrals are supported, Migration is not. */
3585	new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3586
3587	/* set the wire flags to return to client. */
3588	clid->flags = new->cl_exchange_flags;
3589}
3590
3591static bool client_has_openowners(struct nfs4_client *clp)
3592{
3593	struct nfs4_openowner *oo;
3594
3595	list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3596		if (!list_empty(&oo->oo_owner.so_stateids))
3597			return true;
3598	}
3599	return false;
3600}
3601
3602static bool client_has_state(struct nfs4_client *clp)
3603{
3604	return client_has_openowners(clp)
3605#ifdef CONFIG_NFSD_PNFS
3606		|| !list_empty(&clp->cl_lo_states)
3607#endif
3608		|| !list_empty(&clp->cl_delegations)
3609		|| !list_empty(&clp->cl_sessions)
3610		|| nfsd4_has_active_async_copies(clp);
3611}
3612
3613static __be32 copy_impl_id(struct nfs4_client *clp,
3614				struct nfsd4_exchange_id *exid)
3615{
3616	if (!exid->nii_domain.data)
3617		return 0;
3618	xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3619	if (!clp->cl_nii_domain.data)
3620		return nfserr_jukebox;
3621	xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3622	if (!clp->cl_nii_name.data)
3623		return nfserr_jukebox;
3624	clp->cl_nii_time = exid->nii_time;
3625	return 0;
3626}
3627
3628__be32
3629nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3630		union nfsd4_op_u *u)
3631{
3632	struct nfsd4_exchange_id *exid = &u->exchange_id;
3633	struct nfs4_client *conf, *new;
3634	struct nfs4_client *unconf = NULL;
3635	__be32 status;
3636	char			addr_str[INET6_ADDRSTRLEN];
3637	nfs4_verifier		verf = exid->verifier;
3638	struct sockaddr		*sa = svc_addr(rqstp);
3639	bool	update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3640	struct nfsd_net		*nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3641
3642	rpc_ntop(sa, addr_str, sizeof(addr_str));
3643	dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3644		"ip_addr=%s flags %x, spa_how %u\n",
3645		__func__, rqstp, exid, exid->clname.len, exid->clname.data,
3646		addr_str, exid->flags, exid->spa_how);
3647
3648	exid->server_impl_name = kasprintf(GFP_KERNEL, "%s %s %s %s",
3649					   utsname()->sysname, utsname()->release,
3650					   utsname()->version, utsname()->machine);
3651	if (!exid->server_impl_name)
3652		return nfserr_jukebox;
3653
3654	if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3655		return nfserr_inval;
3656
3657	new = create_client(exid->clname, rqstp, &verf);
3658	if (new == NULL)
3659		return nfserr_jukebox;
3660	status = copy_impl_id(new, exid);
3661	if (status)
3662		goto out_nolock;
3663
3664	switch (exid->spa_how) {
3665	case SP4_MACH_CRED:
3666		exid->spo_must_enforce[0] = 0;
3667		exid->spo_must_enforce[1] = (
3668			1 << (OP_BIND_CONN_TO_SESSION - 32) |
3669			1 << (OP_EXCHANGE_ID - 32) |
3670			1 << (OP_CREATE_SESSION - 32) |
3671			1 << (OP_DESTROY_SESSION - 32) |
3672			1 << (OP_DESTROY_CLIENTID - 32));
3673
3674		exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3675					1 << (OP_OPEN_DOWNGRADE) |
3676					1 << (OP_LOCKU) |
3677					1 << (OP_DELEGRETURN));
3678
3679		exid->spo_must_allow[1] &= (
3680					1 << (OP_TEST_STATEID - 32) |
3681					1 << (OP_FREE_STATEID - 32));
3682		if (!svc_rqst_integrity_protected(rqstp)) {
3683			status = nfserr_inval;
3684			goto out_nolock;
3685		}
3686		/*
3687		 * Sometimes userspace doesn't give us a principal.
3688		 * Which is a bug, really.  Anyway, we can't enforce
3689		 * MACH_CRED in that case, better to give up now:
3690		 */
3691		if (!new->cl_cred.cr_principal &&
3692					!new->cl_cred.cr_raw_principal) {
3693			status = nfserr_serverfault;
3694			goto out_nolock;
3695		}
3696		new->cl_mach_cred = true;
3697		break;
3698	case SP4_NONE:
3699		break;
3700	default:				/* checked by xdr code */
3701		WARN_ON_ONCE(1);
3702		fallthrough;
3703	case SP4_SSV:
3704		status = nfserr_encr_alg_unsupp;
3705		goto out_nolock;
3706	}
3707
3708	/* Cases below refer to rfc 5661 section 18.35.4: */
3709	spin_lock(&nn->client_lock);
3710	conf = find_confirmed_client_by_name(&exid->clname, nn);
3711	if (conf) {
3712		bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3713		bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3714
3715		if (update) {
3716			if (!clp_used_exchangeid(conf)) { /* buggy client */
3717				status = nfserr_inval;
3718				goto out;
3719			}
3720			if (!nfsd4_mach_creds_match(conf, rqstp)) {
3721				status = nfserr_wrong_cred;
3722				goto out;
3723			}
3724			if (!creds_match) { /* case 9 */
3725				status = nfserr_perm;
3726				goto out;
3727			}
3728			if (!verfs_match) { /* case 8 */
3729				status = nfserr_not_same;
3730				goto out;
3731			}
3732			/* case 6 */
3733			exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3734			trace_nfsd_clid_confirmed_r(conf);
3735			goto out_copy;
3736		}
3737		if (!creds_match) { /* case 3 */
3738			if (client_has_state(conf)) {
3739				status = nfserr_clid_inuse;
3740				trace_nfsd_clid_cred_mismatch(conf, rqstp);
3741				goto out;
3742			}
3743			goto out_new;
3744		}
3745		if (verfs_match) { /* case 2 */
3746			conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3747			trace_nfsd_clid_confirmed_r(conf);
3748			goto out_copy;
3749		}
3750		/* case 5, client reboot */
3751		trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3752		conf = NULL;
3753		goto out_new;
3754	}
3755
3756	if (update) { /* case 7 */
3757		status = nfserr_noent;
3758		goto out;
3759	}
3760
3761	unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3762	if (unconf) /* case 4, possible retry or client restart */
3763		unhash_client_locked(unconf);
3764
3765	/* case 1, new owner ID */
3766	trace_nfsd_clid_fresh(new);
3767
3768out_new:
3769	if (conf) {
3770		status = mark_client_expired_locked(conf);
3771		if (status)
3772			goto out;
3773		trace_nfsd_clid_replaced(&conf->cl_clientid);
3774	}
3775	new->cl_minorversion = cstate->minorversion;
3776	new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3777	new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3778
3779	/* Contrived initial CREATE_SESSION response */
3780	new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3781
3782	add_to_unconfirmed(new);
3783	swap(new, conf);
3784out_copy:
3785	exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3786	exid->clientid.cl_id = conf->cl_clientid.cl_id;
3787
3788	exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3789	nfsd4_set_ex_flags(conf, exid);
3790
3791	exid->nii_domain.len = sizeof("kernel.org") - 1;
3792	exid->nii_domain.data = "kernel.org";
3793
3794	/*
3795	 * Note that RFC 8881 places no length limit on
3796	 * nii_name, but this implementation permits no
3797	 * more than NFS4_OPAQUE_LIMIT bytes.
3798	 */
3799	exid->nii_name.len = strlen(exid->server_impl_name);
3800	if (exid->nii_name.len > NFS4_OPAQUE_LIMIT)
3801		exid->nii_name.len = NFS4_OPAQUE_LIMIT;
3802	exid->nii_name.data = exid->server_impl_name;
3803
3804	/* just send zeros - the date is in nii_name */
3805	exid->nii_time.tv_sec = 0;
3806	exid->nii_time.tv_nsec = 0;
3807
3808	dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3809		conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3810	status = nfs_ok;
3811
3812out:
3813	spin_unlock(&nn->client_lock);
3814out_nolock:
3815	if (new)
3816		expire_client(new);
3817	if (unconf) {
3818		trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3819		expire_client(unconf);
3820	}
3821	return status;
3822}
3823
3824void
3825nfsd4_exchange_id_release(union nfsd4_op_u *u)
3826{
3827	struct nfsd4_exchange_id *exid = &u->exchange_id;
3828
3829	kfree(exid->server_impl_name);
3830}
3831
3832static __be32 check_slot_seqid(u32 seqid, u32 slot_seqid, u8 flags)
3833{
3834	/* The slot is in use, and no response has been sent. */
3835	if (flags & NFSD4_SLOT_INUSE) {
3836		if (seqid == slot_seqid)
3837			return nfserr_jukebox;
3838		else
3839			return nfserr_seq_misordered;
3840	}
3841	/* Note unsigned 32-bit arithmetic handles wraparound: */
3842	if (likely(seqid == slot_seqid + 1))
3843		return nfs_ok;
3844	if ((flags & NFSD4_SLOT_REUSED) && seqid == 1)
3845		return nfs_ok;
3846	if (seqid == slot_seqid)
3847		return nfserr_replay_cache;
3848	return nfserr_seq_misordered;
3849}
3850
3851/*
3852 * Cache the create session result into the create session single DRC
3853 * slot cache by saving the xdr structure. sl_seqid has been set.
3854 * Do this for solo or embedded create session operations.
3855 */
3856static void
3857nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3858			   struct nfsd4_clid_slot *slot, __be32 nfserr)
3859{
3860	slot->sl_status = nfserr;
3861	memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3862}
3863
3864static __be32
3865nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3866			    struct nfsd4_clid_slot *slot)
3867{
3868	memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3869	return slot->sl_status;
3870}
3871
3872#define NFSD_MIN_REQ_HDR_SEQ_SZ	((\
3873			2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3874			1 +	/* MIN tag is length with zero, only length */ \
3875			3 +	/* version, opcount, opcode */ \
3876			XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3877				/* seqid, slotID, slotID, cache */ \
3878			4 ) * sizeof(__be32))
3879
3880#define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3881			2 +	/* verifier: AUTH_NULL, length 0 */\
3882			1 +	/* status */ \
3883			1 +	/* MIN tag is length with zero, only length */ \
3884			3 +	/* opcount, opcode, opstatus*/ \
3885			XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3886				/* seqid, slotID, slotID, slotID, status */ \
3887			5 ) * sizeof(__be32))
3888
3889static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3890{
3891	u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3892
3893	if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3894		return nfserr_toosmall;
3895	if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3896		return nfserr_toosmall;
3897	ca->headerpadsz = 0;
3898	ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3899	ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3900	ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3901	ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3902			NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3903	ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3904
3905	return nfs_ok;
3906}
3907
3908/*
3909 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3910 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3911 */
3912#define RPC_MAX_HEADER_WITH_AUTH_SYS \
3913	(RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3914
3915#define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3916	(RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3917
3918#define NFSD_CB_MAX_REQ_SZ	((NFS4_enc_cb_recall_sz + \
3919				 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3920#define NFSD_CB_MAX_RESP_SZ	((NFS4_dec_cb_recall_sz + \
3921				 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3922				 sizeof(__be32))
3923
3924static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3925{
3926	ca->headerpadsz = 0;
3927
3928	if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3929		return nfserr_toosmall;
3930	if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3931		return nfserr_toosmall;
3932	ca->maxresp_cached = 0;
3933	if (ca->maxops < 2)
3934		return nfserr_toosmall;
3935
3936	return nfs_ok;
3937}
3938
3939static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3940{
3941	switch (cbs->flavor) {
3942	case RPC_AUTH_NULL:
3943	case RPC_AUTH_UNIX:
3944		return nfs_ok;
3945	default:
3946		/*
3947		 * GSS case: the spec doesn't allow us to return this
3948		 * error.  But it also doesn't allow us not to support
3949		 * GSS.
3950		 * I'd rather this fail hard than return some error the
3951		 * client might think it can already handle:
3952		 */
3953		return nfserr_encr_alg_unsupp;
3954	}
3955}
3956
3957__be32
3958nfsd4_create_session(struct svc_rqst *rqstp,
3959		struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3960{
3961	struct nfsd4_create_session *cr_ses = &u->create_session;
3962	struct sockaddr *sa = svc_addr(rqstp);
3963	struct nfs4_client *conf, *unconf;
3964	struct nfsd4_clid_slot *cs_slot;
3965	struct nfs4_client *old = NULL;
3966	struct nfsd4_session *new;
3967	struct nfsd4_conn *conn;
3968	__be32 status = 0;
3969	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3970
3971	if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3972		return nfserr_inval;
3973	status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3974	if (status)
3975		return status;
3976	status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3977	if (status)
3978		return status;
3979	status = check_backchannel_attrs(&cr_ses->back_channel);
3980	if (status)
3981		goto out_err;
3982	status = nfserr_jukebox;
3983	new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3984	if (!new)
3985		goto out_err;
3986	conn = alloc_conn_from_crses(rqstp, cr_ses);
3987	if (!conn)
3988		goto out_free_session;
3989
3990	spin_lock(&nn->client_lock);
3991
3992	/* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3993	unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3994	conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3995	if (!conf && !unconf) {
3996		status = nfserr_stale_clientid;
3997		goto out_free_conn;
3998	}
3999
4000	/* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
4001	if (conf) {
4002		cs_slot = &conf->cl_cs_slot;
4003		trace_nfsd_slot_seqid_conf(conf, cr_ses);
4004	} else {
4005		cs_slot = &unconf->cl_cs_slot;
4006		trace_nfsd_slot_seqid_unconf(unconf, cr_ses);
4007	}
4008	status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
4009	switch (status) {
4010	case nfs_ok:
4011		cs_slot->sl_seqid++;
4012		cr_ses->seqid = cs_slot->sl_seqid;
4013		break;
4014	case nfserr_replay_cache:
4015		status = nfsd4_replay_create_session(cr_ses, cs_slot);
4016		fallthrough;
4017	case nfserr_jukebox:
4018		/* The server MUST NOT cache NFS4ERR_DELAY */
4019		goto out_free_conn;
4020	default:
4021		goto out_cache_error;
4022	}
4023
4024	/* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
4025	if (conf) {
4026		status = nfserr_wrong_cred;
4027		if (!nfsd4_mach_creds_match(conf, rqstp))
4028			goto out_cache_error;
4029	} else {
4030		status = nfserr_clid_inuse;
4031		if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
4032		    !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
4033			trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4034			goto out_cache_error;
4035		}
4036		status = nfserr_wrong_cred;
4037		if (!nfsd4_mach_creds_match(unconf, rqstp))
4038			goto out_cache_error;
4039		old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4040		if (old) {
4041			status = mark_client_expired_locked(old);
4042			if (status)
4043				goto out_expired_error;
4044			trace_nfsd_clid_replaced(&old->cl_clientid);
4045		}
4046		move_to_confirmed(unconf);
4047		conf = unconf;
4048	}
4049
4050	/* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
4051	status = nfs_ok;
4052	/* Persistent sessions are not supported */
4053	cr_ses->flags &= ~SESSION4_PERSIST;
4054	/* Upshifting from TCP to RDMA is not supported */
4055	cr_ses->flags &= ~SESSION4_RDMA;
4056	/* Report the correct number of backchannel slots */
4057	cr_ses->back_channel.maxreqs = new->se_cb_highest_slot + 1;
4058
4059	init_session(rqstp, new, conf, cr_ses);
4060	nfsd4_get_session_locked(new);
4061
4062	memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
4063	       NFS4_MAX_SESSIONID_LEN);
4064
4065	/* cache solo and embedded create sessions under the client_lock */
4066	nfsd4_cache_create_session(cr_ses, cs_slot, status);
4067	spin_unlock(&nn->client_lock);
4068	if (conf == unconf)
4069		fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4070	/* init connection and backchannel */
4071	nfsd4_init_conn(rqstp, conn, new);
4072	nfsd4_put_session(new);
4073	if (old)
4074		expire_client(old);
4075	return status;
4076
4077out_expired_error:
4078	/*
4079	 * Revert the slot seq_nr change so the server will process
4080	 * the client's resend instead of returning a cached response.
4081	 */
4082	if (status == nfserr_jukebox) {
4083		cs_slot->sl_seqid--;
4084		cr_ses->seqid = cs_slot->sl_seqid;
4085		goto out_free_conn;
4086	}
4087out_cache_error:
4088	nfsd4_cache_create_session(cr_ses, cs_slot, status);
4089out_free_conn:
4090	spin_unlock(&nn->client_lock);
4091	free_conn(conn);
4092out_free_session:
4093	__free_session(new);
4094out_err:
4095	return status;
4096}
4097
4098static __be32 nfsd4_map_bcts_dir(u32 *dir)
4099{
4100	switch (*dir) {
4101	case NFS4_CDFC4_FORE:
4102	case NFS4_CDFC4_BACK:
4103		return nfs_ok;
4104	case NFS4_CDFC4_FORE_OR_BOTH:
4105	case NFS4_CDFC4_BACK_OR_BOTH:
4106		*dir = NFS4_CDFC4_BOTH;
4107		return nfs_ok;
4108	}
4109	return nfserr_inval;
4110}
4111
4112__be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
4113		struct nfsd4_compound_state *cstate,
4114		union nfsd4_op_u *u)
4115{
4116	struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
4117	struct nfsd4_session *session = cstate->session;
4118	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4119	__be32 status;
4120
4121	status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
4122	if (status)
4123		return status;
4124	spin_lock(&nn->client_lock);
4125	session->se_cb_prog = bc->bc_cb_program;
4126	session->se_cb_sec = bc->bc_cb_sec;
4127	spin_unlock(&nn->client_lock);
4128
4129	nfsd4_probe_callback(session->se_client);
4130
4131	return nfs_ok;
4132}
4133
4134static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
4135{
4136	struct nfsd4_conn *c;
4137
4138	list_for_each_entry(c, &s->se_conns, cn_persession) {
4139		if (c->cn_xprt == xpt) {
4140			return c;
4141		}
4142	}
4143	return NULL;
4144}
4145
4146static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
4147		struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
4148{
4149	struct nfs4_client *clp = session->se_client;
4150	struct svc_xprt *xpt = rqst->rq_xprt;
4151	struct nfsd4_conn *c;
4152	__be32 status;
4153
4154	/* Following the last paragraph of RFC 5661 Section 18.34.3: */
4155	spin_lock(&clp->cl_lock);
4156	c = __nfsd4_find_conn(xpt, session);
4157	if (!c)
4158		status = nfserr_noent;
4159	else if (req == c->cn_flags)
4160		status = nfs_ok;
4161	else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
4162				c->cn_flags != NFS4_CDFC4_BACK)
4163		status = nfs_ok;
4164	else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
4165				c->cn_flags != NFS4_CDFC4_FORE)
4166		status = nfs_ok;
4167	else
4168		status = nfserr_inval;
4169	spin_unlock(&clp->cl_lock);
4170	if (status == nfs_ok && conn)
4171		*conn = c;
4172	return status;
4173}
4174
4175__be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
4176		     struct nfsd4_compound_state *cstate,
4177		     union nfsd4_op_u *u)
4178{
4179	struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
4180	__be32 status;
4181	struct nfsd4_conn *conn;
4182	struct nfsd4_session *session;
4183	struct net *net = SVC_NET(rqstp);
4184	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4185
4186	if (!nfsd4_last_compound_op(rqstp))
4187		return nfserr_not_only_op;
4188	spin_lock(&nn->client_lock);
4189	session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4190	spin_unlock(&nn->client_lock);
4191	if (!session)
4192		goto out_no_session;
4193	status = nfserr_wrong_cred;
4194	if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4195		goto out;
4196	status = nfsd4_match_existing_connection(rqstp, session,
4197			bcts->dir, &conn);
4198	if (status == nfs_ok) {
4199		if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4200				bcts->dir == NFS4_CDFC4_BACK)
4201			conn->cn_flags |= NFS4_CDFC4_BACK;
4202		nfsd4_probe_callback(session->se_client);
4203		goto out;
4204	}
4205	if (status == nfserr_inval)
4206		goto out;
4207	status = nfsd4_map_bcts_dir(&bcts->dir);
4208	if (status)
4209		goto out;
4210	conn = alloc_conn(rqstp, bcts->dir);
4211	status = nfserr_jukebox;
4212	if (!conn)
4213		goto out;
4214	nfsd4_init_conn(rqstp, conn, session);
4215	status = nfs_ok;
4216out:
4217	nfsd4_put_session(session);
4218out_no_session:
4219	return status;
4220}
4221
4222static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4223{
4224	if (!cstate->session)
4225		return false;
4226	return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4227}
4228
4229__be32
4230nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4231		union nfsd4_op_u *u)
4232{
4233	struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4234	struct nfsd4_session *ses;
4235	__be32 status;
4236	int ref_held_by_me = 0;
4237	struct net *net = SVC_NET(r);
4238	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4239
4240	status = nfserr_not_only_op;
4241	if (nfsd4_compound_in_session(cstate, sessionid)) {
4242		if (!nfsd4_last_compound_op(r))
4243			goto out;
4244		ref_held_by_me++;
4245	}
4246	dump_sessionid(__func__, sessionid);
4247	spin_lock(&nn->client_lock);
4248	ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4249	if (!ses)
4250		goto out_client_lock;
4251	status = nfserr_wrong_cred;
4252	if (!nfsd4_mach_creds_match(ses->se_client, r))
4253		goto out_put_session;
4254	status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4255	if (status)
4256		goto out_put_session;
4257	unhash_session(ses);
4258	spin_unlock(&nn->client_lock);
4259
4260	nfsd4_probe_callback_sync(ses->se_client);
4261
4262	spin_lock(&nn->client_lock);
4263	status = nfs_ok;
4264out_put_session:
4265	nfsd4_put_session_locked(ses);
4266out_client_lock:
4267	spin_unlock(&nn->client_lock);
4268out:
4269	return status;
4270}
4271
4272static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4273{
4274	struct nfs4_client *clp = ses->se_client;
4275	struct nfsd4_conn *c;
4276	__be32 status = nfs_ok;
4277	int ret;
4278
4279	spin_lock(&clp->cl_lock);
4280	c = __nfsd4_find_conn(new->cn_xprt, ses);
4281	if (c)
4282		goto out_free;
4283	status = nfserr_conn_not_bound_to_session;
4284	if (clp->cl_mach_cred)
4285		goto out_free;
4286	__nfsd4_hash_conn(new, ses);
4287	spin_unlock(&clp->cl_lock);
4288	ret = nfsd4_register_conn(new);
4289	if (ret)
4290		/* oops; xprt is already down: */
4291		nfsd4_conn_lost(&new->cn_xpt_user);
4292	return nfs_ok;
4293out_free:
4294	spin_unlock(&clp->cl_lock);
4295	free_conn(new);
4296	return status;
4297}
4298
4299static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4300{
4301	struct nfsd4_compoundargs *args = rqstp->rq_argp;
4302
4303	return args->opcnt > session->se_fchannel.maxops;
4304}
4305
4306static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4307				  struct nfsd4_session *session)
4308{
4309	struct xdr_buf *xb = &rqstp->rq_arg;
4310
4311	return xb->len > session->se_fchannel.maxreq_sz;
4312}
4313
4314static bool replay_matches_cache(struct svc_rqst *rqstp,
4315		 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4316{
4317	struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4318
4319	if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4320	    (bool)seq->cachethis)
4321		return false;
4322	/*
4323	 * If there's an error then the reply can have fewer ops than
4324	 * the call.
4325	 */
4326	if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4327		return false;
4328	/*
4329	 * But if we cached a reply with *more* ops than the call you're
4330	 * sending us now, then this new call is clearly not really a
4331	 * replay of the old one:
4332	 */
4333	if (slot->sl_opcnt > argp->opcnt)
4334		return false;
4335	/* This is the only check explicitly called by spec: */
4336	if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4337		return false;
4338	/*
4339	 * There may be more comparisons we could actually do, but the
4340	 * spec doesn't require us to catch every case where the calls
4341	 * don't match (that would require caching the call as well as
4342	 * the reply), so we don't bother.
4343	 */
4344	return true;
4345}
4346
4347/*
4348 * Note that the response is constructed here both for the case
4349 * of a new SEQUENCE request and for a replayed SEQUENCE request.
4350 * We do not cache SEQUENCE responses as SEQUENCE is idempotent.
4351 */
4352static void nfsd4_construct_sequence_response(struct nfsd4_session *session,
4353					      struct nfsd4_sequence *seq)
4354{
4355	struct nfs4_client *clp = session->se_client;
4356
4357	seq->maxslots_response = max(session->se_target_maxslots,
4358				     seq->maxslots);
4359	seq->target_maxslots = session->se_target_maxslots;
4360
4361	switch (clp->cl_cb_state) {
4362	case NFSD4_CB_DOWN:
4363		seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4364		break;
4365	case NFSD4_CB_FAULT:
4366		seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4367		break;
4368	default:
4369		seq->status_flags = 0;
4370	}
4371	if (!list_empty(&clp->cl_revoked))
4372		seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4373	if (atomic_read(&clp->cl_admin_revoked))
4374		seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4375}
4376
4377__be32
4378nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4379		union nfsd4_op_u *u)
4380{
4381	struct nfsd4_sequence *seq = &u->sequence;
4382	struct nfsd4_compoundres *resp = rqstp->rq_resp;
4383	struct xdr_stream *xdr = resp->xdr;
4384	struct nfsd4_session *session;
4385	struct nfs4_client *clp;
4386	struct nfsd4_slot *slot;
4387	struct nfsd4_conn *conn;
4388	__be32 status;
4389	int buflen;
4390	struct net *net = SVC_NET(rqstp);
4391	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4392
4393	if (resp->opcnt != 1)
4394		return nfserr_sequence_pos;
4395
4396	/*
4397	 * Will be either used or freed by nfsd4_sequence_check_conn
4398	 * below.
4399	 */
4400	conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4401	if (!conn)
4402		return nfserr_jukebox;
4403
4404	spin_lock(&nn->client_lock);
4405	session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4406	if (!session)
4407		goto out_no_session;
4408	clp = session->se_client;
4409
4410	status = nfserr_too_many_ops;
4411	if (nfsd4_session_too_many_ops(rqstp, session))
4412		goto out_put_session;
4413
4414	status = nfserr_req_too_big;
4415	if (nfsd4_request_too_big(rqstp, session))
4416		goto out_put_session;
4417
4418	status = nfserr_badslot;
4419	if (seq->slotid >= session->se_fchannel.maxreqs)
4420		goto out_put_session;
4421
4422	slot = xa_load(&session->se_slots, seq->slotid);
4423	dprintk("%s: slotid %d\n", __func__, seq->slotid);
4424
4425	trace_nfsd_slot_seqid_sequence(clp, seq, slot);
4426
4427	nfsd4_construct_sequence_response(session, seq);
4428
4429	status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_flags);
4430	if (status == nfserr_replay_cache) {
4431		status = nfserr_seq_misordered;
4432		if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4433			goto out_put_session;
4434		status = nfserr_seq_false_retry;
4435		if (!replay_matches_cache(rqstp, seq, slot))
4436			goto out_put_session;
4437		cstate->slot = slot;
4438		cstate->session = session;
4439		cstate->clp = clp;
4440		/* Return the cached reply status and set cstate->status
4441		 * for nfsd4_proc_compound processing */
4442		status = nfsd4_replay_cache_entry(resp, seq);
4443		cstate->status = nfserr_replay_cache;
4444		goto out;
4445	}
4446	if (status)
4447		goto out_put_session;
4448
4449	status = nfsd4_sequence_check_conn(conn, session);
4450	conn = NULL;
4451	if (status)
4452		goto out_put_session;
4453
4454	if (session->se_target_maxslots < session->se_fchannel.maxreqs &&
4455	    slot->sl_generation == session->se_slot_gen &&
4456	    seq->maxslots <= session->se_target_maxslots)
4457		/* Client acknowledged our reduce maxreqs */
4458		free_session_slots(session, session->se_target_maxslots);
4459
4460	buflen = (seq->cachethis) ?
4461			session->se_fchannel.maxresp_cached :
4462			session->se_fchannel.maxresp_sz;
4463	status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4464				    nfserr_rep_too_big;
4465	if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4466		goto out_put_session;
4467	svc_reserve_auth(rqstp, buflen);
4468
4469	status = nfs_ok;
4470	/* Success! accept new slot seqid */
4471	slot->sl_seqid = seq->seqid;
4472	slot->sl_flags &= ~NFSD4_SLOT_REUSED;
4473	slot->sl_flags |= NFSD4_SLOT_INUSE;
4474	slot->sl_generation = session->se_slot_gen;
4475	if (seq->cachethis)
4476		slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4477	else
4478		slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4479
4480	cstate->slot = slot;
4481	cstate->session = session;
4482	cstate->clp = clp;
4483
4484	/*
4485	 * If the client ever uses the highest available slot,
4486	 * gently try to allocate another 20%.  This allows
4487	 * fairly quick growth without grossly over-shooting what
4488	 * the client might use.
4489	 */
4490	if (seq->slotid == session->se_fchannel.maxreqs - 1 &&
4491	    session->se_target_maxslots >= session->se_fchannel.maxreqs &&
4492	    session->se_fchannel.maxreqs < NFSD_MAX_SLOTS_PER_SESSION) {
4493		int s = session->se_fchannel.maxreqs;
4494		int cnt = DIV_ROUND_UP(s, 5);
4495		void *prev_slot;
4496
4497		do {
4498			/*
4499			 * GFP_NOWAIT both allows allocation under a
4500			 * spinlock, and only succeeds if there is
4501			 * plenty of memory.
4502			 */
4503			slot = nfsd4_alloc_slot(&session->se_fchannel, s,
4504						GFP_NOWAIT);
4505			prev_slot = xa_load(&session->se_slots, s);
4506			if (xa_is_value(prev_slot) && slot) {
4507				slot->sl_seqid = xa_to_value(prev_slot);
4508				slot->sl_flags |= NFSD4_SLOT_REUSED;
4509			}
4510			if (slot &&
4511			    !xa_is_err(xa_store(&session->se_slots, s, slot,
4512						GFP_NOWAIT))) {
4513				s += 1;
4514				session->se_fchannel.maxreqs = s;
4515				atomic_add(s - session->se_target_maxslots,
4516					   &nfsd_total_target_slots);
4517				session->se_target_maxslots = s;
4518			} else {
4519				kfree(slot);
4520				slot = NULL;
4521			}
4522		} while (slot && --cnt > 0);
4523	}
4524
4525out:
4526	trace_nfsd_seq4_status(rqstp, seq);
4527out_no_session:
4528	if (conn)
4529		free_conn(conn);
4530	spin_unlock(&nn->client_lock);
4531	return status;
4532out_put_session:
4533	nfsd4_put_session_locked(session);
4534	goto out_no_session;
4535}
4536
4537void
4538nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4539{
4540	struct nfsd4_compound_state *cs = &resp->cstate;
4541
4542	if (nfsd4_has_session(cs)) {
4543		if (cs->status != nfserr_replay_cache) {
4544			nfsd4_store_cache_entry(resp);
4545			cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4546		}
4547		/* Drop session reference that was taken in nfsd4_sequence() */
4548		nfsd4_put_session(cs->session);
4549	} else if (cs->clp)
4550		put_client_renew(cs->clp);
4551}
4552
4553__be32
4554nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4555		struct nfsd4_compound_state *cstate,
4556		union nfsd4_op_u *u)
4557{
4558	struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4559	struct nfs4_client *conf, *unconf;
4560	struct nfs4_client *clp = NULL;
4561	__be32 status = 0;
4562	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4563
4564	spin_lock(&nn->client_lock);
4565	unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4566	conf = find_confirmed_client(&dc->clientid, true, nn);
4567	WARN_ON_ONCE(conf && unconf);
4568
4569	if (conf) {
4570		if (client_has_state(conf)) {
4571			status = nfserr_clientid_busy;
4572			goto out;
4573		}
4574		status = mark_client_expired_locked(conf);
4575		if (status)
4576			goto out;
4577		clp = conf;
4578	} else if (unconf)
4579		clp = unconf;
4580	else {
4581		status = nfserr_stale_clientid;
4582		goto out;
4583	}
4584	if (!nfsd4_mach_creds_match(clp, rqstp)) {
4585		clp = NULL;
4586		status = nfserr_wrong_cred;
4587		goto out;
4588	}
4589	trace_nfsd_clid_destroyed(&clp->cl_clientid);
4590	unhash_client_locked(clp);
4591out:
4592	spin_unlock(&nn->client_lock);
4593	if (clp)
4594		expire_client(clp);
4595	return status;
4596}
4597
4598__be32
4599nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4600		struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4601{
4602	struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4603	struct nfs4_client *clp = cstate->clp;
4604	__be32 status = 0;
4605
4606	if (rc->rca_one_fs) {
4607		if (!cstate->current_fh.fh_dentry)
4608			return nfserr_nofilehandle;
4609		/*
4610		 * We don't take advantage of the rca_one_fs case.
4611		 * That's OK, it's optional, we can safely ignore it.
4612		 */
4613		return nfs_ok;
4614	}
4615
4616	status = nfserr_complete_already;
4617	if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4618		goto out;
4619
4620	status = nfserr_stale_clientid;
4621	if (is_client_expired(clp))
4622		/*
4623		 * The following error isn't really legal.
4624		 * But we only get here if the client just explicitly
4625		 * destroyed the client.  Surely it no longer cares what
4626		 * error it gets back on an operation for the dead
4627		 * client.
4628		 */
4629		goto out;
4630
4631	status = nfs_ok;
4632	trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4633	nfsd4_client_record_create(clp);
4634	inc_reclaim_complete(clp);
4635out:
4636	return status;
4637}
4638
4639__be32
4640nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4641		  union nfsd4_op_u *u)
4642{
4643	struct nfsd4_setclientid *setclid = &u->setclientid;
4644	struct xdr_netobj 	clname = setclid->se_name;
4645	nfs4_verifier		clverifier = setclid->se_verf;
4646	struct nfs4_client	*conf, *new;
4647	struct nfs4_client	*unconf = NULL;
4648	__be32 			status;
4649	struct nfsd_net		*nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4650
4651	new = create_client(clname, rqstp, &clverifier);
4652	if (new == NULL)
4653		return nfserr_jukebox;
4654	spin_lock(&nn->client_lock);
4655	conf = find_confirmed_client_by_name(&clname, nn);
4656	if (conf && client_has_state(conf)) {
4657		status = nfserr_clid_inuse;
4658		if (clp_used_exchangeid(conf))
4659			goto out;
4660		if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4661			trace_nfsd_clid_cred_mismatch(conf, rqstp);
4662			goto out;
4663		}
4664	}
4665	unconf = find_unconfirmed_client_by_name(&clname, nn);
4666	if (unconf)
4667		unhash_client_locked(unconf);
4668	if (conf) {
4669		if (same_verf(&conf->cl_verifier, &clverifier)) {
4670			copy_clid(new, conf);
4671			gen_confirm(new, nn);
4672		} else
4673			trace_nfsd_clid_verf_mismatch(conf, rqstp,
4674						      &clverifier);
4675	} else
4676		trace_nfsd_clid_fresh(new);
4677	new->cl_minorversion = 0;
4678	gen_callback(new, setclid, rqstp);
4679	add_to_unconfirmed(new);
4680	setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4681	setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4682	memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4683	new = NULL;
4684	status = nfs_ok;
4685out:
4686	spin_unlock(&nn->client_lock);
4687	if (new)
4688		free_client(new);
4689	if (unconf) {
4690		trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4691		expire_client(unconf);
4692	}
4693	return status;
4694}
4695
4696__be32
4697nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4698			struct nfsd4_compound_state *cstate,
4699			union nfsd4_op_u *u)
4700{
4701	struct nfsd4_setclientid_confirm *setclientid_confirm =
4702			&u->setclientid_confirm;
4703	struct nfs4_client *conf, *unconf;
4704	struct nfs4_client *old = NULL;
4705	nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
4706	clientid_t * clid = &setclientid_confirm->sc_clientid;
4707	__be32 status;
4708	struct nfsd_net	*nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4709
4710	if (STALE_CLIENTID(clid, nn))
4711		return nfserr_stale_clientid;
4712
4713	spin_lock(&nn->client_lock);
4714	conf = find_confirmed_client(clid, false, nn);
4715	unconf = find_unconfirmed_client(clid, false, nn);
4716	/*
4717	 * We try hard to give out unique clientid's, so if we get an
4718	 * attempt to confirm the same clientid with a different cred,
4719	 * the client may be buggy; this should never happen.
4720	 *
4721	 * Nevertheless, RFC 7530 recommends INUSE for this case:
4722	 */
4723	status = nfserr_clid_inuse;
4724	if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4725		trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4726		goto out;
4727	}
4728	if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4729		trace_nfsd_clid_cred_mismatch(conf, rqstp);
4730		goto out;
4731	}
4732	if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4733		if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4734			status = nfs_ok;
4735		} else
4736			status = nfserr_stale_clientid;
4737		goto out;
4738	}
4739	status = nfs_ok;
4740	if (conf) {
4741		if (get_client_locked(conf) == nfs_ok) {
4742			old = unconf;
4743			unhash_client_locked(old);
4744			nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4745		} else {
4746			conf = NULL;
4747		}
4748	}
4749
4750	if (!conf) {
4751		old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4752		if (old) {
4753			status = nfserr_clid_inuse;
4754			if (client_has_state(old)
4755					&& !same_creds(&unconf->cl_cred,
4756							&old->cl_cred)) {
4757				old = NULL;
4758				goto out;
4759			}
4760			status = mark_client_expired_locked(old);
4761			if (status) {
4762				old = NULL;
4763				goto out;
4764			}
4765			trace_nfsd_clid_replaced(&old->cl_clientid);
4766		}
4767		status = get_client_locked(unconf);
4768		if (status != nfs_ok) {
4769			old = NULL;
4770			goto out;
4771		}
4772		move_to_confirmed(unconf);
4773		conf = unconf;
4774	}
4775	spin_unlock(&nn->client_lock);
4776	if (conf == unconf)
4777		fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4778	nfsd4_probe_callback(conf);
4779	spin_lock(&nn->client_lock);
4780	put_client_renew_locked(conf);
4781out:
4782	spin_unlock(&nn->client_lock);
4783	if (old)
4784		expire_client(old);
4785	return status;
4786}
4787
4788static struct nfs4_file *nfsd4_alloc_file(void)
4789{
4790	return kmem_cache_alloc(file_slab, GFP_KERNEL);
4791}
4792
4793/* OPEN Share state helper functions */
4794
4795static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4796{
4797	refcount_set(&fp->fi_ref, 1);
4798	spin_lock_init(&fp->fi_lock);
4799	INIT_LIST_HEAD(&fp->fi_stateids);
4800	INIT_LIST_HEAD(&fp->fi_delegations);
4801	INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4802	fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4803	fp->fi_deleg_file = NULL;
4804	fp->fi_rdeleg_file = NULL;
4805	fp->fi_had_conflict = false;
4806	fp->fi_share_deny = 0;
4807	memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4808	memset(fp->fi_access, 0, sizeof(fp->fi_access));
4809	fp->fi_aliased = false;
4810	fp->fi_inode = d_inode(fh->fh_dentry);
4811#ifdef CONFIG_NFSD_PNFS
4812	INIT_LIST_HEAD(&fp->fi_lo_states);
4813	atomic_set(&fp->fi_lo_recalls, 0);
4814#endif
4815}
4816
4817void
4818nfsd4_free_slabs(void)
4819{
4820	kmem_cache_destroy(client_slab);
4821	kmem_cache_destroy(openowner_slab);
4822	kmem_cache_destroy(lockowner_slab);
4823	kmem_cache_destroy(file_slab);
4824	kmem_cache_destroy(stateid_slab);
4825	kmem_cache_destroy(deleg_slab);
4826	kmem_cache_destroy(odstate_slab);
4827}
4828
4829int
4830nfsd4_init_slabs(void)
4831{
4832	client_slab = KMEM_CACHE(nfs4_client, 0);
4833	if (client_slab == NULL)
4834		goto out;
4835	openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4836	if (openowner_slab == NULL)
4837		goto out_free_client_slab;
4838	lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4839	if (lockowner_slab == NULL)
4840		goto out_free_openowner_slab;
4841	file_slab = KMEM_CACHE(nfs4_file, 0);
4842	if (file_slab == NULL)
4843		goto out_free_lockowner_slab;
4844	stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4845	if (stateid_slab == NULL)
4846		goto out_free_file_slab;
4847	deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4848	if (deleg_slab == NULL)
4849		goto out_free_stateid_slab;
4850	odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4851	if (odstate_slab == NULL)
4852		goto out_free_deleg_slab;
4853	return 0;
4854
4855out_free_deleg_slab:
4856	kmem_cache_destroy(deleg_slab);
4857out_free_stateid_slab:
4858	kmem_cache_destroy(stateid_slab);
4859out_free_file_slab:
4860	kmem_cache_destroy(file_slab);
4861out_free_lockowner_slab:
4862	kmem_cache_destroy(lockowner_slab);
4863out_free_openowner_slab:
4864	kmem_cache_destroy(openowner_slab);
4865out_free_client_slab:
4866	kmem_cache_destroy(client_slab);
4867out:
4868	return -ENOMEM;
4869}
4870
4871static unsigned long
4872nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4873{
4874	struct nfsd_net *nn = shrink->private_data;
4875	long count;
4876
4877	count = atomic_read(&nn->nfsd_courtesy_clients);
4878	if (!count)
4879		count = atomic_long_read(&num_delegations);
4880	if (count)
4881		queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4882	return (unsigned long)count;
4883}
4884
4885static unsigned long
4886nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4887{
4888	return SHRINK_STOP;
4889}
4890
4891void
4892nfsd4_init_leases_net(struct nfsd_net *nn)
4893{
4894	struct sysinfo si;
4895	u64 max_clients;
4896
4897	nn->nfsd4_lease = 90;	/* default lease time */
4898	nn->nfsd4_grace = 90;
4899	nn->somebody_reclaimed = false;
4900	nn->track_reclaim_completes = false;
4901	nn->clverifier_counter = get_random_u32();
4902	nn->clientid_base = get_random_u32();
4903	nn->clientid_counter = nn->clientid_base + 1;
4904	nn->s2s_cp_cl_id = nn->clientid_counter++;
4905
4906	atomic_set(&nn->nfs4_client_count, 0);
4907	si_meminfo(&si);
4908	max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4909	max_clients *= NFS4_CLIENTS_PER_GB;
4910	nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4911
4912	atomic_set(&nn->nfsd_courtesy_clients, 0);
4913}
4914
4915enum rp_lock {
4916	RP_UNLOCKED,
4917	RP_LOCKED,
4918	RP_UNHASHED,
4919};
4920
4921static void init_nfs4_replay(struct nfs4_replay *rp)
4922{
4923	rp->rp_status = nfserr_serverfault;
4924	rp->rp_buflen = 0;
4925	rp->rp_buf = rp->rp_ibuf;
4926	rp->rp_locked = RP_UNLOCKED;
4927}
4928
4929static int nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4930				      struct nfs4_stateowner *so)
4931{
4932	if (!nfsd4_has_session(cstate)) {
4933		wait_var_event(&so->so_replay.rp_locked,
4934			       cmpxchg(&so->so_replay.rp_locked,
4935				       RP_UNLOCKED, RP_LOCKED) != RP_LOCKED);
4936		if (so->so_replay.rp_locked == RP_UNHASHED)
4937			return -EAGAIN;
4938		cstate->replay_owner = nfs4_get_stateowner(so);
4939	}
4940	return 0;
4941}
4942
4943void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4944{
4945	struct nfs4_stateowner *so = cstate->replay_owner;
4946
4947	if (so != NULL) {
4948		cstate->replay_owner = NULL;
4949		store_release_wake_up(&so->so_replay.rp_locked, RP_UNLOCKED);
4950		nfs4_put_stateowner(so);
4951	}
4952}
4953
4954static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4955{
4956	struct nfs4_stateowner *sop;
4957
4958	sop = kmem_cache_alloc(slab, GFP_KERNEL);
4959	if (!sop)
4960		return NULL;
4961
4962	xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4963	if (!sop->so_owner.data) {
4964		kmem_cache_free(slab, sop);
4965		return NULL;
4966	}
4967
4968	INIT_LIST_HEAD(&sop->so_stateids);
4969	sop->so_client = clp;
4970	init_nfs4_replay(&sop->so_replay);
4971	atomic_set(&sop->so_count, 1);
4972	return sop;
4973}
4974
4975static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4976{
4977	lockdep_assert_held(&clp->cl_lock);
4978
4979	list_add(&oo->oo_owner.so_strhash,
4980		 &clp->cl_ownerstr_hashtbl[strhashval]);
4981	list_add(&oo->oo_perclient, &clp->cl_openowners);
4982}
4983
4984static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4985{
4986	unhash_openowner_locked(openowner(so));
4987}
4988
4989static void nfs4_free_openowner(struct nfs4_stateowner *so)
4990{
4991	struct nfs4_openowner *oo = openowner(so);
4992
4993	kmem_cache_free(openowner_slab, oo);
4994}
4995
4996static const struct nfs4_stateowner_operations openowner_ops = {
4997	.so_unhash =	nfs4_unhash_openowner,
4998	.so_free =	nfs4_free_openowner,
4999};
5000
5001static struct nfs4_ol_stateid *
5002nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
5003{
5004	struct nfs4_ol_stateid *local, *ret = NULL;
5005	struct nfs4_openowner *oo = open->op_openowner;
5006
5007	lockdep_assert_held(&fp->fi_lock);
5008
5009	list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
5010		/* ignore lock owners */
5011		if (local->st_stateowner->so_is_open_owner == 0)
5012			continue;
5013		if (local->st_stateowner != &oo->oo_owner)
5014			continue;
5015		if (local->st_stid.sc_type == SC_TYPE_OPEN &&
5016		    !local->st_stid.sc_status) {
5017			ret = local;
5018			refcount_inc(&ret->st_stid.sc_count);
5019			break;
5020		}
5021	}
5022	return ret;
5023}
5024
5025static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
5026	__releases(&s->sc_client->cl_lock)
5027{
5028	struct nfs4_client *cl = s->sc_client;
5029	LIST_HEAD(reaplist);
5030	struct nfs4_ol_stateid *stp;
5031	struct nfs4_delegation *dp;
5032	bool unhashed;
5033
5034	switch (s->sc_type) {
5035	case SC_TYPE_OPEN:
5036		stp = openlockstateid(s);
5037		if (unhash_open_stateid(stp, &reaplist))
5038			put_ol_stateid_locked(stp, &reaplist);
5039		spin_unlock(&cl->cl_lock);
5040		free_ol_stateid_reaplist(&reaplist);
5041		break;
5042	case SC_TYPE_LOCK:
5043		stp = openlockstateid(s);
5044		unhashed = unhash_lock_stateid(stp);
5045		spin_unlock(&cl->cl_lock);
5046		if (unhashed)
5047			nfs4_put_stid(s);
5048		break;
5049	case SC_TYPE_DELEG:
5050		dp = delegstateid(s);
5051		list_del_init(&dp->dl_recall_lru);
5052		spin_unlock(&cl->cl_lock);
5053		nfs4_put_stid(s);
5054		break;
5055	default:
5056		spin_unlock(&cl->cl_lock);
5057	}
5058}
5059
5060static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
5061				    stateid_t *stid)
5062{
5063	/* NFSv4.0 has no way for the client to tell the server
5064	 * that it can forget an admin-revoked stateid.
5065	 * So we keep it around until the first time that the
5066	 * client uses it, and drop it the first time
5067	 * nfserr_admin_revoked is returned.
5068	 * For v4.1 and later we wait until explicitly told
5069	 * to free the stateid.
5070	 */
5071	if (cl->cl_minorversion == 0) {
5072		struct nfs4_stid *st;
5073
5074		spin_lock(&cl->cl_lock);
5075		st = find_stateid_locked(cl, stid);
5076		if (st)
5077			nfsd4_drop_revoked_stid(st);
5078		else
5079			spin_unlock(&cl->cl_lock);
5080	}
5081}
5082
5083static __be32
5084nfsd4_verify_open_stid(struct nfs4_stid *s)
5085{
5086	__be32 ret = nfs_ok;
5087
5088	if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
5089		ret = nfserr_admin_revoked;
5090	else if (s->sc_status & SC_STATUS_REVOKED)
5091		ret = nfserr_deleg_revoked;
5092	else if (s->sc_status & SC_STATUS_CLOSED)
5093		ret = nfserr_bad_stateid;
5094	return ret;
5095}
5096
5097/* Lock the stateid st_mutex, and deal with races with CLOSE */
5098static __be32
5099nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
5100{
5101	__be32 ret;
5102
5103	mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
5104	ret = nfsd4_verify_open_stid(&stp->st_stid);
5105	if (ret == nfserr_admin_revoked)
5106		nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
5107					&stp->st_stid.sc_stateid);
5108
5109	if (ret != nfs_ok)
5110		mutex_unlock(&stp->st_mutex);
5111	return ret;
5112}
5113
5114static struct nfs4_ol_stateid *
5115nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
5116{
5117	struct nfs4_ol_stateid *stp;
5118	for (;;) {
5119		spin_lock(&fp->fi_lock);
5120		stp = nfsd4_find_existing_open(fp, open);
5121		spin_unlock(&fp->fi_lock);
5122		if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
5123			break;
5124		nfs4_put_stid(&stp->st_stid);
5125	}
5126	return stp;
5127}
5128
5129static struct nfs4_openowner *
5130find_or_alloc_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
5131			      struct nfsd4_compound_state *cstate)
5132{
5133	struct nfs4_client *clp = cstate->clp;
5134	struct nfs4_openowner *oo, *new = NULL;
5135
5136retry:
5137	spin_lock(&clp->cl_lock);
5138	oo = find_openstateowner_str(strhashval, open, clp);
5139	if (!oo && new) {
5140		hash_openowner(new, clp, strhashval);
5141		spin_unlock(&clp->cl_lock);
5142		return new;
5143	}
5144	spin_unlock(&clp->cl_lock);
5145
5146	if (oo && !(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5147		/* Replace unconfirmed owners without checking for replay. */
5148		release_openowner(oo);
5149		oo = NULL;
5150	}
5151	if (oo) {
5152		if (new)
5153			nfs4_free_stateowner(&new->oo_owner);
5154		return oo;
5155	}
5156
5157	new = alloc_stateowner(openowner_slab, &open->op_owner, clp);
5158	if (!new)
5159		return NULL;
5160	new->oo_owner.so_ops = &openowner_ops;
5161	new->oo_owner.so_is_open_owner = 1;
5162	new->oo_owner.so_seqid = open->op_seqid;
5163	new->oo_flags = 0;
5164	if (nfsd4_has_session(cstate))
5165		new->oo_flags |= NFS4_OO_CONFIRMED;
5166	new->oo_time = 0;
5167	new->oo_last_closed_stid = NULL;
5168	INIT_LIST_HEAD(&new->oo_close_lru);
5169	goto retry;
5170}
5171
5172static struct nfs4_ol_stateid *
5173init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
5174{
5175
5176	struct nfs4_openowner *oo = open->op_openowner;
5177	struct nfs4_ol_stateid *retstp = NULL;
5178	struct nfs4_ol_stateid *stp;
5179
5180	stp = open->op_stp;
5181	/* We are moving these outside of the spinlocks to avoid the warnings */
5182	mutex_init(&stp->st_mutex);
5183	mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5184
5185retry:
5186	spin_lock(&oo->oo_owner.so_client->cl_lock);
5187	spin_lock(&fp->fi_lock);
5188
5189	if (nfs4_openowner_unhashed(oo)) {
5190		mutex_unlock(&stp->st_mutex);
5191		stp = NULL;
5192		goto out_unlock;
5193	}
5194
5195	retstp = nfsd4_find_existing_open(fp, open);
5196	if (retstp)
5197		goto out_unlock;
5198
5199	open->op_stp = NULL;
5200	refcount_inc(&stp->st_stid.sc_count);
5201	stp->st_stid.sc_type = SC_TYPE_OPEN;
5202	INIT_LIST_HEAD(&stp->st_locks);
5203	stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
5204	get_nfs4_file(fp);
5205	stp->st_stid.sc_file = fp;
5206	stp->st_access_bmap = 0;
5207	stp->st_deny_bmap = 0;
5208	stp->st_openstp = NULL;
5209	list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
5210	list_add(&stp->st_perfile, &fp->fi_stateids);
5211
5212out_unlock:
5213	spin_unlock(&fp->fi_lock);
5214	spin_unlock(&oo->oo_owner.so_client->cl_lock);
5215	if (retstp) {
5216		/* Handle races with CLOSE */
5217		if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5218			nfs4_put_stid(&retstp->st_stid);
5219			goto retry;
5220		}
5221		/* To keep mutex tracking happy */
5222		mutex_unlock(&stp->st_mutex);
5223		stp = retstp;
5224	}
5225	return stp;
5226}
5227
5228/*
5229 * In the 4.0 case we need to keep the owners around a little while to handle
5230 * CLOSE replay. We still do need to release any file access that is held by
5231 * them before returning however.
5232 */
5233static void
5234move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
5235{
5236	struct nfs4_ol_stateid *last;
5237	struct nfs4_openowner *oo = openowner(s->st_stateowner);
5238	struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
5239						nfsd_net_id);
5240
5241	dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
5242
5243	/*
5244	 * We know that we hold one reference via nfsd4_close, and another
5245	 * "persistent" reference for the client. If the refcount is higher
5246	 * than 2, then there are still calls in progress that are using this
5247	 * stateid. We can't put the sc_file reference until they are finished.
5248	 * Wait for the refcount to drop to 2. Since it has been unhashed,
5249	 * there should be no danger of the refcount going back up again at
5250	 * this point.
5251	 * Some threads with a reference might be waiting for rp_locked,
5252	 * so tell them to stop waiting.
5253	 */
5254	store_release_wake_up(&oo->oo_owner.so_replay.rp_locked, RP_UNHASHED);
5255	wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
5256
5257	release_all_access(s);
5258	if (s->st_stid.sc_file) {
5259		put_nfs4_file(s->st_stid.sc_file);
5260		s->st_stid.sc_file = NULL;
5261	}
5262
5263	spin_lock(&nn->client_lock);
5264	last = oo->oo_last_closed_stid;
5265	oo->oo_last_closed_stid = s;
5266	list_move_tail(&oo->oo_close_lru, &nn->close_lru);
5267	oo->oo_time = ktime_get_boottime_seconds();
5268	spin_unlock(&nn->client_lock);
5269	if (last)
5270		nfs4_put_stid(&last->st_stid);
5271}
5272
5273static noinline_for_stack struct nfs4_file *
5274nfsd4_file_hash_lookup(const struct svc_fh *fhp)
5275{
5276	struct inode *inode = d_inode(fhp->fh_dentry);
5277	struct rhlist_head *tmp, *list;
5278	struct nfs4_file *fi;
5279
5280	rcu_read_lock();
5281	list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5282			       nfs4_file_rhash_params);
5283	rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5284		if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5285			if (refcount_inc_not_zero(&fi->fi_ref)) {
5286				rcu_read_unlock();
5287				return fi;
5288			}
5289		}
5290	}
5291	rcu_read_unlock();
5292	return NULL;
5293}
5294
5295/*
5296 * On hash insertion, identify entries with the same inode but
5297 * distinct filehandles. They will all be on the list returned
5298 * by rhltable_lookup().
5299 *
5300 * inode->i_lock prevents racing insertions from adding an entry
5301 * for the same inode/fhp pair twice.
5302 */
5303static noinline_for_stack struct nfs4_file *
5304nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5305{
5306	struct inode *inode = d_inode(fhp->fh_dentry);
5307	struct rhlist_head *tmp, *list;
5308	struct nfs4_file *ret = NULL;
5309	bool alias_found = false;
5310	struct nfs4_file *fi;
5311	int err;
5312
5313	rcu_read_lock();
5314	spin_lock(&inode->i_lock);
5315
5316	list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5317			       nfs4_file_rhash_params);
5318	rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5319		if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5320			if (refcount_inc_not_zero(&fi->fi_ref))
5321				ret = fi;
5322		} else
5323			fi->fi_aliased = alias_found = true;
5324	}
5325	if (ret)
5326		goto out_unlock;
5327
5328	nfsd4_file_init(fhp, new);
5329	err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5330			      nfs4_file_rhash_params);
5331	if (err)
5332		goto out_unlock;
5333
5334	new->fi_aliased = alias_found;
5335	ret = new;
5336
5337out_unlock:
5338	spin_unlock(&inode->i_lock);
5339	rcu_read_unlock();
5340	return ret;
5341}
5342
5343static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5344{
5345	rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5346			nfs4_file_rhash_params);
5347}
5348
5349/*
5350 * Called to check deny when READ with all zero stateid or
5351 * WRITE with all zero or all one stateid
5352 */
5353static __be32
5354nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5355{
5356	struct nfs4_file *fp;
5357	__be32 ret = nfs_ok;
5358
5359	fp = nfsd4_file_hash_lookup(current_fh);
5360	if (!fp)
5361		return ret;
5362
5363	/* Check for conflicting share reservations */
5364	spin_lock(&fp->fi_lock);
5365	if (fp->fi_share_deny & deny_type)
5366		ret = nfserr_locked;
5367	spin_unlock(&fp->fi_lock);
5368	put_nfs4_file(fp);
5369	return ret;
5370}
5371
5372static bool nfsd4_deleg_present(const struct inode *inode)
5373{
5374	struct file_lock_context *ctx = locks_inode_context(inode);
5375
5376	return ctx && !list_empty_careful(&ctx->flc_lease);
5377}
5378
5379/**
5380 * nfsd_wait_for_delegreturn - wait for delegations to be returned
5381 * @rqstp: the RPC transaction being executed
5382 * @inode: in-core inode of the file being waited for
5383 *
5384 * The timeout prevents deadlock if all nfsd threads happen to be
5385 * tied up waiting for returning delegations.
5386 *
5387 * Return values:
5388 *   %true: delegation was returned
5389 *   %false: timed out waiting for delegreturn
5390 */
5391bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5392{
5393	long __maybe_unused timeo;
5394
5395	timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5396				       NFSD_DELEGRETURN_TIMEOUT);
5397	trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5398	return timeo > 0;
5399}
5400
5401static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5402{
5403	struct nfs4_delegation *dp = cb_to_delegation(cb);
5404	struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5405					  nfsd_net_id);
5406
5407	block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5408
5409	/*
5410	 * We can't do this in nfsd_break_deleg_cb because it is
5411	 * already holding inode->i_lock.
5412	 *
5413	 * If the dl_time != 0, then we know that it has already been
5414	 * queued for a lease break. Don't queue it again.
5415	 */
5416	spin_lock(&state_lock);
5417	if (delegation_hashed(dp) && dp->dl_time == 0) {
5418		dp->dl_time = ktime_get_boottime_seconds();
5419		list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5420	}
5421	spin_unlock(&state_lock);
5422}
5423
5424static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5425		struct rpc_task *task)
5426{
5427	struct nfs4_delegation *dp = cb_to_delegation(cb);
5428
5429	trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5430
5431	if (dp->dl_stid.sc_status)
5432		/* CLOSED or REVOKED */
5433		return 1;
5434
5435	switch (task->tk_status) {
5436	case 0:
5437		return 1;
5438	case -NFS4ERR_DELAY:
5439		rpc_delay(task, 2 * HZ);
5440		return 0;
5441	case -EBADHANDLE:
5442	case -NFS4ERR_BAD_STATEID:
5443		/*
5444		 * Race: client probably got cb_recall before open reply
5445		 * granting delegation.
5446		 */
5447		if (dp->dl_retries--) {
5448			rpc_delay(task, 2 * HZ);
5449			return 0;
5450		}
5451		fallthrough;
5452	default:
5453		return 1;
5454	}
5455}
5456
5457static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5458{
5459	struct nfs4_delegation *dp = cb_to_delegation(cb);
5460
5461	nfs4_put_stid(&dp->dl_stid);
5462}
5463
5464static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5465	.prepare	= nfsd4_cb_recall_prepare,
5466	.done		= nfsd4_cb_recall_done,
5467	.release	= nfsd4_cb_recall_release,
5468	.opcode		= OP_CB_RECALL,
5469};
5470
5471static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5472{
5473	bool queued;
5474
5475	if (test_and_set_bit(NFSD4_CALLBACK_RUNNING, &dp->dl_recall.cb_flags))
5476		return;
5477
5478	/*
5479	 * We're assuming the state code never drops its reference
5480	 * without first removing the lease.  Since we're in this lease
5481	 * callback (and since the lease code is serialized by the
5482	 * flc_lock) we know the server hasn't removed the lease yet, and
5483	 * we know it's safe to take a reference.
5484	 */
5485	refcount_inc(&dp->dl_stid.sc_count);
5486	queued = nfsd4_run_cb(&dp->dl_recall);
5487	WARN_ON_ONCE(!queued);
5488	if (!queued)
5489		refcount_dec(&dp->dl_stid.sc_count);
5490}
5491
5492/* Called from break_lease() with flc_lock held. */
5493static bool
5494nfsd_break_deleg_cb(struct file_lease *fl)
5495{
5496	struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5497	struct nfs4_file *fp = dp->dl_stid.sc_file;
5498	struct nfs4_client *clp = dp->dl_stid.sc_client;
5499	struct nfsd_net *nn;
5500
5501	trace_nfsd_cb_recall(&dp->dl_stid);
5502
5503	dp->dl_recalled = true;
5504	atomic_inc(&clp->cl_delegs_in_recall);
5505	if (try_to_expire_client(clp)) {
5506		nn = net_generic(clp->net, nfsd_net_id);
5507		mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5508	}
5509
5510	/*
5511	 * We don't want the locks code to timeout the lease for us;
5512	 * we'll remove it ourself if a delegation isn't returned
5513	 * in time:
5514	 */
5515	fl->fl_break_time = 0;
5516
5517	fp->fi_had_conflict = true;
5518	nfsd_break_one_deleg(dp);
5519	return false;
5520}
5521
5522/**
5523 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5524 * @fl: Lock state to check
5525 *
5526 * Return values:
5527 *   %true: Lease conflict was resolved
5528 *   %false: Lease conflict was not resolved.
5529 */
5530static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5531{
5532	struct nfs4_delegation *dl = fl->c.flc_owner;
5533	struct svc_rqst *rqst;
5534	struct nfs4_client *clp;
5535
5536	rqst = nfsd_current_rqst();
5537	if (!nfsd_v4client(rqst))
5538		return false;
5539	clp = *(rqst->rq_lease_breaker);
5540	return dl->dl_stid.sc_client == clp;
5541}
5542
5543static int
5544nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5545		     struct list_head *dispose)
5546{
5547	struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5548	struct nfs4_client *clp = dp->dl_stid.sc_client;
5549
5550	if (arg & F_UNLCK) {
5551		if (dp->dl_recalled)
5552			atomic_dec(&clp->cl_delegs_in_recall);
5553		return lease_modify(onlist, arg, dispose);
5554	} else
5555		return -EAGAIN;
5556}
5557
5558/**
5559 *  nfsd4_deleg_lm_open_conflict - see if the given file points to an inode that has
5560 *				   an existing open that would conflict with the
5561 *				   desired lease.
5562 * @filp:	file to check
5563 * @arg:	type of lease that we're trying to acquire
5564 *
5565 * The kernel will call into this operation to determine whether there
5566 * are conflicting opens that may prevent the deleg from being granted.
5567 * For nfsd, that check is done at a higher level, so this trivially
5568 * returns 0.
5569 */
5570static int
5571nfsd4_deleg_lm_open_conflict(struct file *filp, int arg)
5572{
5573	return 0;
5574}
5575
5576static const struct lease_manager_operations nfsd_lease_mng_ops = {
5577	.lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5578	.lm_break = nfsd_break_deleg_cb,
5579	.lm_change = nfsd_change_deleg_cb,
5580	.lm_open_conflict = nfsd4_deleg_lm_open_conflict,
5581};
5582
5583static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5584{
5585	if (nfsd4_has_session(cstate))
5586		return nfs_ok;
5587	if (seqid == so->so_seqid - 1)
5588		return nfserr_replay_me;
5589	if (seqid == so->so_seqid)
5590		return nfs_ok;
5591	return nfserr_bad_seqid;
5592}
5593
5594static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5595						struct nfsd_net *nn)
5596{
5597	struct nfs4_client *found;
5598
5599	spin_lock(&nn->client_lock);
5600	found = find_confirmed_client(clid, sessions, nn);
5601	if (found)
5602		atomic_inc(&found->cl_rpc_users);
5603	spin_unlock(&nn->client_lock);
5604	return found;
5605}
5606
5607static __be32 set_client(clientid_t *clid,
5608		struct nfsd4_compound_state *cstate,
5609		struct nfsd_net *nn)
5610{
5611	if (cstate->clp) {
5612		if (!same_clid(&cstate->clp->cl_clientid, clid))
5613			return nfserr_stale_clientid;
5614		return nfs_ok;
5615	}
5616	if (STALE_CLIENTID(clid, nn))
5617		return nfserr_stale_clientid;
5618	/*
5619	 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5620	 * set cstate->clp), so session = false:
5621	 */
5622	cstate->clp = lookup_clientid(clid, false, nn);
5623	if (!cstate->clp)
5624		return nfserr_expired;
5625	return nfs_ok;
5626}
5627
5628__be32
5629nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5630		    struct nfsd4_open *open, struct nfsd_net *nn)
5631{
5632	clientid_t *clientid = &open->op_clientid;
5633	struct nfs4_client *clp = NULL;
5634	unsigned int strhashval;
5635	struct nfs4_openowner *oo = NULL;
5636	__be32 status;
5637
5638	/*
5639	 * In case we need it later, after we've already created the
5640	 * file and don't want to risk a further failure:
5641	 */
5642	open->op_file = nfsd4_alloc_file();
5643	if (open->op_file == NULL)
5644		return nfserr_jukebox;
5645
5646	status = set_client(clientid, cstate, nn);
5647	if (status)
5648		return status;
5649	clp = cstate->clp;
5650
5651	strhashval = ownerstr_hashval(&open->op_owner);
5652retry:
5653	oo = find_or_alloc_open_stateowner(strhashval, open, cstate);
5654	open->op_openowner = oo;
5655	if (!oo)
5656		return nfserr_jukebox;
5657	if (nfsd4_cstate_assign_replay(cstate, &oo->oo_owner) == -EAGAIN) {
5658		nfs4_put_stateowner(&oo->oo_owner);
5659		goto retry;
5660	}
5661	status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5662	if (status)
5663		return status;
5664
5665	open->op_stp = nfs4_alloc_open_stateid(clp);
5666	if (!open->op_stp)
5667		return nfserr_jukebox;
5668
5669	if (nfsd4_has_session(cstate) &&
5670	    (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5671		open->op_odstate = alloc_clnt_odstate(clp);
5672		if (!open->op_odstate)
5673			return nfserr_jukebox;
5674	}
5675
5676	return nfs_ok;
5677}
5678
5679static inline __be32
5680nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5681{
5682	if (!(flags & RD_STATE) && deleg_is_read(dp->dl_type))
5683		return nfserr_openmode;
5684	else
5685		return nfs_ok;
5686}
5687
5688static int share_access_to_flags(u32 share_access)
5689{
5690	return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5691}
5692
5693static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5694						  stateid_t *s)
5695{
5696	struct nfs4_stid *ret;
5697
5698	ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5699	if (!ret)
5700		return NULL;
5701	return delegstateid(ret);
5702}
5703
5704static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5705{
5706	return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5707	       open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5708}
5709
5710static __be32
5711nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5712		struct nfs4_delegation **dp)
5713{
5714	int flags;
5715	__be32 status = nfserr_bad_stateid;
5716	struct nfs4_delegation *deleg;
5717
5718	deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5719	if (deleg == NULL)
5720		goto out;
5721	if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5722		nfs4_put_stid(&deleg->dl_stid);
5723		status = nfserr_admin_revoked;
5724		goto out;
5725	}
5726	if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5727		nfs4_put_stid(&deleg->dl_stid);
5728		nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5729		status = nfserr_deleg_revoked;
5730		goto out;
5731	}
5732	flags = share_access_to_flags(open->op_share_access);
5733	status = nfs4_check_delegmode(deleg, flags);
5734	if (status) {
5735		nfs4_put_stid(&deleg->dl_stid);
5736		goto out;
5737	}
5738	*dp = deleg;
5739out:
5740	if (!nfsd4_is_deleg_cur(open))
5741		return nfs_ok;
5742	if (status)
5743		return status;
5744	open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5745	return nfs_ok;
5746}
5747
5748static inline int nfs4_access_to_access(u32 nfs4_access)
5749{
5750	int flags = 0;
5751
5752	if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5753		flags |= NFSD_MAY_READ;
5754	if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5755		flags |= NFSD_MAY_WRITE;
5756	return flags;
5757}
5758
5759static inline __be32
5760nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5761		struct nfsd4_open *open)
5762{
5763	struct iattr iattr = {
5764		.ia_valid = ATTR_SIZE,
5765		.ia_size = 0,
5766	};
5767	struct nfsd_attrs attrs = {
5768		.na_iattr	= &iattr,
5769	};
5770	if (!open->op_truncate)
5771		return 0;
5772	if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5773		return nfserr_inval;
5774	return nfsd_setattr(rqstp, fh, &attrs, NULL);
5775}
5776
5777static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5778		struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5779		struct nfsd4_open *open, bool new_stp)
5780{
5781	struct nfsd_file *nf = NULL;
5782	__be32 status;
5783	int oflag = nfs4_access_to_omode(open->op_share_access);
5784	int access = nfs4_access_to_access(open->op_share_access);
5785	unsigned char old_access_bmap, old_deny_bmap;
5786
5787	spin_lock(&fp->fi_lock);
5788
5789	/*
5790	 * Are we trying to set a deny mode that would conflict with
5791	 * current access?
5792	 */
5793	status = nfs4_file_check_deny(fp, open->op_share_deny);
5794	if (status != nfs_ok) {
5795		if (status != nfserr_share_denied) {
5796			spin_unlock(&fp->fi_lock);
5797			goto out;
5798		}
5799		if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5800				stp, open->op_share_deny, false))
5801			status = nfserr_jukebox;
5802		spin_unlock(&fp->fi_lock);
5803		goto out;
5804	}
5805
5806	/* set access to the file */
5807	status = nfs4_file_get_access(fp, open->op_share_access);
5808	if (status != nfs_ok) {
5809		if (status != nfserr_share_denied) {
5810			spin_unlock(&fp->fi_lock);
5811			goto out;
5812		}
5813		if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5814				stp, open->op_share_access, true))
5815			status = nfserr_jukebox;
5816		spin_unlock(&fp->fi_lock);
5817		goto out;
5818	}
5819
5820	/* Set access bits in stateid */
5821	old_access_bmap = stp->st_access_bmap;
5822	set_access(open->op_share_access, stp);
5823
5824	/* Set new deny mask */
5825	old_deny_bmap = stp->st_deny_bmap;
5826	set_deny(open->op_share_deny, stp);
5827	fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5828
5829	if (!fp->fi_fds[oflag]) {
5830		spin_unlock(&fp->fi_lock);
5831
5832		status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5833						  open->op_filp, &nf);
5834		if (status != nfs_ok)
5835			goto out_put_access;
5836
5837		spin_lock(&fp->fi_lock);
5838		if (!fp->fi_fds[oflag]) {
5839			fp->fi_fds[oflag] = nf;
5840			nf = NULL;
5841		}
5842	}
5843	spin_unlock(&fp->fi_lock);
5844	if (nf)
5845		nfsd_file_put(nf);
5846
5847	status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5848								access));
5849	if (status)
5850		goto out_put_access;
5851
5852	status = nfsd4_truncate(rqstp, cur_fh, open);
5853	if (status)
5854		goto out_put_access;
5855out:
5856	return status;
5857out_put_access:
5858	stp->st_access_bmap = old_access_bmap;
5859	nfs4_file_put_access(fp, open->op_share_access);
5860	reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5861	goto out;
5862}
5863
5864static __be32
5865nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5866		struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5867		struct nfsd4_open *open)
5868{
5869	__be32 status;
5870	unsigned char old_deny_bmap = stp->st_deny_bmap;
5871
5872	if (!test_access(open->op_share_access, stp))
5873		return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5874
5875	/* test and set deny mode */
5876	spin_lock(&fp->fi_lock);
5877	status = nfs4_file_check_deny(fp, open->op_share_deny);
5878	switch (status) {
5879	case nfs_ok:
5880		set_deny(open->op_share_deny, stp);
5881		fp->fi_share_deny |=
5882			(open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5883		break;
5884	case nfserr_share_denied:
5885		if (nfs4_resolve_deny_conflicts_locked(fp, false,
5886				stp, open->op_share_deny, false))
5887			status = nfserr_jukebox;
5888		break;
5889	}
5890	spin_unlock(&fp->fi_lock);
5891
5892	if (status != nfs_ok)
5893		return status;
5894
5895	status = nfsd4_truncate(rqstp, cur_fh, open);
5896	if (status != nfs_ok)
5897		reset_union_bmap_deny(old_deny_bmap, stp);
5898	return status;
5899}
5900
5901/* Should we give out recallable state?: */
5902static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5903{
5904	if (clp->cl_cb_state == NFSD4_CB_UP)
5905		return true;
5906	/*
5907	 * In the sessions case, since we don't have to establish a
5908	 * separate connection for callbacks, we assume it's OK
5909	 * until we hear otherwise:
5910	 */
5911	return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5912}
5913
5914static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp)
5915{
5916	struct file_lease *fl;
5917
5918	fl = locks_alloc_lease();
5919	if (!fl)
5920		return NULL;
5921	fl->fl_lmops = &nfsd_lease_mng_ops;
5922	fl->c.flc_flags = FL_DELEG;
5923	fl->c.flc_type = deleg_is_read(dp->dl_type) ? F_RDLCK : F_WRLCK;
5924	fl->c.flc_owner = (fl_owner_t)dp;
5925	fl->c.flc_pid = current->tgid;
5926	fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5927	return fl;
5928}
5929
5930static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5931					 struct nfs4_file *fp)
5932{
5933	struct nfs4_ol_stateid *st;
5934	struct file *f = fp->fi_deleg_file->nf_file;
5935	struct inode *ino = file_inode(f);
5936	int writes;
5937
5938	writes = atomic_read(&ino->i_writecount);
5939	if (!writes)
5940		return 0;
5941	/*
5942	 * There could be multiple filehandles (hence multiple
5943	 * nfs4_files) referencing this file, but that's not too
5944	 * common; let's just give up in that case rather than
5945	 * trying to go look up all the clients using that other
5946	 * nfs4_file as well:
5947	 */
5948	if (fp->fi_aliased)
5949		return -EAGAIN;
5950	/*
5951	 * If there's a close in progress, make sure that we see it
5952	 * clear any fi_fds[] entries before we see it decrement
5953	 * i_writecount:
5954	 */
5955	smp_mb__after_atomic();
5956
5957	if (fp->fi_fds[O_WRONLY])
5958		writes--;
5959	if (fp->fi_fds[O_RDWR])
5960		writes--;
5961	if (writes > 0)
5962		return -EAGAIN; /* There may be non-NFSv4 writers */
5963	/*
5964	 * It's possible there are non-NFSv4 write opens in progress,
5965	 * but if they haven't incremented i_writecount yet then they
5966	 * also haven't called break lease yet; so, they'll break this
5967	 * lease soon enough.  So, all that's left to check for is NFSv4
5968	 * opens:
5969	 */
5970	spin_lock(&fp->fi_lock);
5971	list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5972		if (st->st_openstp == NULL /* it's an open */ &&
5973		    access_permit_write(st) &&
5974		    st->st_stid.sc_client != clp) {
5975			spin_unlock(&fp->fi_lock);
5976			return -EAGAIN;
5977		}
5978	}
5979	spin_unlock(&fp->fi_lock);
5980	/*
5981	 * There's a small chance that we could be racing with another
5982	 * NFSv4 open.  However, any open that hasn't added itself to
5983	 * the fi_stateids list also hasn't called break_lease yet; so,
5984	 * they'll break this lease soon enough.
5985	 */
5986	return 0;
5987}
5988
5989/*
5990 * It's possible that between opening the dentry and setting the delegation,
5991 * that it has been renamed or unlinked. Redo the lookup to verify that this
5992 * hasn't happened.
5993 */
5994static int
5995nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5996			  struct svc_fh *parent)
5997{
5998	struct svc_export *exp;
5999	struct dentry *child;
6000	__be32 err;
6001
6002	err = nfsd_lookup_dentry(open->op_rqstp, parent,
6003				 open->op_fname, open->op_fnamelen,
6004				 &exp, &child);
6005
6006	if (err)
6007		return -EAGAIN;
6008
6009	exp_put(exp);
6010	dput(child);
6011	if (child != file_dentry(fp->fi_deleg_file->nf_file))
6012		return -EAGAIN;
6013
6014	return 0;
6015}
6016
6017/*
6018 * We avoid breaking delegations held by a client due to its own activity, but
6019 * clearing setuid/setgid bits on a write is an implicit activity and the client
6020 * may not notice and continue using the old mode. Avoid giving out a delegation
6021 * on setuid/setgid files when the client is requesting an open for write.
6022 */
6023static int
6024nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
6025{
6026	struct inode *inode = file_inode(nf->nf_file);
6027
6028	if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
6029	    (inode->i_mode & (S_ISUID|S_ISGID)))
6030		return -EAGAIN;
6031	return 0;
6032}
6033
6034#ifdef CONFIG_NFSD_V4_DELEG_TIMESTAMPS
6035static bool nfsd4_want_deleg_timestamps(const struct nfsd4_open *open)
6036{
6037	return open->op_deleg_want & OPEN4_SHARE_ACCESS_WANT_DELEG_TIMESTAMPS;
6038}
6039#else /* CONFIG_NFSD_V4_DELEG_TIMESTAMPS */
6040static bool nfsd4_want_deleg_timestamps(const struct nfsd4_open *open)
6041{
6042	return false;
6043}
6044#endif /* CONFIG NFSD_V4_DELEG_TIMESTAMPS */
6045
6046static struct nfs4_delegation *
6047nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
6048		    struct svc_fh *parent)
6049{
6050	bool deleg_ts = nfsd4_want_deleg_timestamps(open);
6051	struct nfs4_client *clp = stp->st_stid.sc_client;
6052	struct nfs4_file *fp = stp->st_stid.sc_file;
6053	struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
6054	struct nfs4_delegation *dp;
6055	struct nfsd_file *nf = NULL;
6056	struct file_lease *fl;
6057	int status = 0;
6058	u32 dl_type;
6059
6060	/*
6061	 * The fi_had_conflict and nfs_get_existing_delegation checks
6062	 * here are just optimizations; we'll need to recheck them at
6063	 * the end:
6064	 */
6065	if (fp->fi_had_conflict)
6066		return ERR_PTR(-EAGAIN);
6067
6068	/*
6069	 * Try for a write delegation first. RFC8881 section 10.4 says:
6070	 *
6071	 *  "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
6072	 *   on its own, all opens."
6073	 *
6074	 * Furthermore, section 9.1.2 says:
6075	 *
6076	 *  "In the case of READ, the server may perform the corresponding
6077	 *  check on the access mode, or it may choose to allow READ for
6078	 *  OPEN4_SHARE_ACCESS_WRITE, to accommodate clients whose WRITE
6079	 *  implementation may unavoidably do reads (e.g., due to buffer
6080	 *  cache constraints)."
6081	 *
6082	 *  We choose to offer a write delegation for OPEN with the
6083	 *  OPEN4_SHARE_ACCESS_WRITE access mode to accommodate such clients.
6084	 */
6085	if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
6086		nf = find_writeable_file(fp);
6087		dl_type = deleg_ts ? OPEN_DELEGATE_WRITE_ATTRS_DELEG : OPEN_DELEGATE_WRITE;
6088	}
6089
6090	/*
6091	 * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
6092	 * file for some reason, then try for a read delegation instead.
6093	 */
6094	if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
6095		nf = find_readable_file(fp);
6096		dl_type = deleg_ts ? OPEN_DELEGATE_READ_ATTRS_DELEG : OPEN_DELEGATE_READ;
6097	}
6098
6099	if (!nf)
6100		return ERR_PTR(-EAGAIN);
6101
6102	/*
6103	 * File delegations and associated locks cannot be recovered if the
6104	 * export is from an NFS proxy server.
6105	 */
6106	if (exportfs_cannot_lock(nf->nf_file->f_path.mnt->mnt_sb->s_export_op)) {
6107		nfsd_file_put(nf);
6108		return ERR_PTR(-EOPNOTSUPP);
6109	}
6110
6111	spin_lock(&state_lock);
6112	spin_lock(&fp->fi_lock);
6113	if (nfs4_delegation_exists(clp, fp))
6114		status = -EAGAIN;
6115	else if (nfsd4_verify_setuid_write(open, nf))
6116		status = -EAGAIN;
6117	else if (!fp->fi_deleg_file) {
6118		fp->fi_deleg_file = nf;
6119		/* increment early to prevent fi_deleg_file from being
6120		 * cleared */
6121		fp->fi_delegees = 1;
6122		nf = NULL;
6123	} else
6124		fp->fi_delegees++;
6125	spin_unlock(&fp->fi_lock);
6126	spin_unlock(&state_lock);
6127	if (nf)
6128		nfsd_file_put(nf);
6129	if (status)
6130		return ERR_PTR(status);
6131
6132	status = -ENOMEM;
6133	dp = alloc_init_deleg(clp, fp, odstate, dl_type);
6134	if (!dp)
6135		goto out_delegees;
6136
6137	fl = nfs4_alloc_init_lease(dp);
6138	if (!fl)
6139		goto out_clnt_odstate;
6140
6141	status = kernel_setlease(fp->fi_deleg_file->nf_file,
6142				      fl->c.flc_type, &fl, NULL);
6143	if (fl)
6144		locks_free_lease(fl);
6145	if (status)
6146		goto out_clnt_odstate;
6147
6148	if (parent) {
6149		status = nfsd4_verify_deleg_dentry(open, fp, parent);
6150		if (status)
6151			goto out_unlock;
6152	}
6153
6154	status = nfsd4_check_conflicting_opens(clp, fp);
6155	if (status)
6156		goto out_unlock;
6157
6158	/*
6159	 * Now that the deleg is set, check again to ensure that nothing
6160	 * raced in and changed the mode while we weren't looking.
6161	 */
6162	status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
6163	if (status)
6164		goto out_unlock;
6165
6166	status = -EAGAIN;
6167	if (fp->fi_had_conflict)
6168		goto out_unlock;
6169
6170	spin_lock(&state_lock);
6171	spin_lock(&clp->cl_lock);
6172	spin_lock(&fp->fi_lock);
6173	status = hash_delegation_locked(dp, fp);
6174	spin_unlock(&fp->fi_lock);
6175	spin_unlock(&clp->cl_lock);
6176	spin_unlock(&state_lock);
6177
6178	if (status)
6179		goto out_unlock;
6180
6181	return dp;
6182out_unlock:
6183	kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
6184out_clnt_odstate:
6185	put_clnt_odstate(dp->dl_clnt_odstate);
6186	nfs4_put_stid(&dp->dl_stid);
6187out_delegees:
6188	put_deleg_file(fp);
6189	return ERR_PTR(status);
6190}
6191
6192static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
6193{
6194	open->op_delegate_type = OPEN_DELEGATE_NONE_EXT;
6195	if (status == -EAGAIN)
6196		open->op_why_no_deleg = WND4_CONTENTION;
6197	else {
6198		open->op_why_no_deleg = WND4_RESOURCE;
6199		switch (open->op_deleg_want) {
6200		case OPEN4_SHARE_ACCESS_WANT_READ_DELEG:
6201		case OPEN4_SHARE_ACCESS_WANT_WRITE_DELEG:
6202		case OPEN4_SHARE_ACCESS_WANT_ANY_DELEG:
6203			break;
6204		case OPEN4_SHARE_ACCESS_WANT_CANCEL:
6205			open->op_why_no_deleg = WND4_CANCELLED;
6206			break;
6207		case OPEN4_SHARE_ACCESS_WANT_NO_DELEG:
6208			WARN_ON_ONCE(1);
6209		}
6210	}
6211}
6212
6213static bool
6214nfs4_delegation_stat(struct nfs4_delegation *dp, struct svc_fh *currentfh,
6215		     struct kstat *stat)
6216{
6217	struct nfsd_file *nf = find_writeable_file(dp->dl_stid.sc_file);
6218	struct path path;
6219	int rc;
6220
6221	if (!nf)
6222		return false;
6223
6224	path.mnt = currentfh->fh_export->ex_path.mnt;
6225	path.dentry = file_dentry(nf->nf_file);
6226
6227	rc = vfs_getattr(&path, stat,
6228			 STATX_MODE | STATX_SIZE | STATX_ATIME |
6229			 STATX_MTIME | STATX_CTIME | STATX_CHANGE_COOKIE,
6230			 AT_STATX_SYNC_AS_STAT);
6231
6232	nfsd_file_put(nf);
6233	return rc == 0;
6234}
6235
6236/*
6237 * Add NFS4_SHARE_ACCESS_READ to the write delegation granted on OPEN
6238 * with NFS4_SHARE_ACCESS_WRITE by allocating separate nfsd_file and
6239 * struct file to be used for read with delegation stateid.
6240 *
6241 */
6242static bool
6243nfsd4_add_rdaccess_to_wrdeleg(struct svc_rqst *rqstp, struct nfsd4_open *open,
6244			      struct svc_fh *fh, struct nfs4_ol_stateid *stp)
6245{
6246	struct nfs4_file *fp;
6247	struct nfsd_file *nf = NULL;
6248
6249	if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) ==
6250			NFS4_SHARE_ACCESS_WRITE) {
6251		if (nfsd_file_acquire_opened(rqstp, fh, NFSD_MAY_READ, NULL, &nf))
6252			return (false);
6253		fp = stp->st_stid.sc_file;
6254		spin_lock(&fp->fi_lock);
6255		__nfs4_file_get_access(fp, NFS4_SHARE_ACCESS_READ);
6256		if (!fp->fi_fds[O_RDONLY]) {
6257			fp->fi_fds[O_RDONLY] = nf;
6258			nf = NULL;
6259		}
6260		fp->fi_rdeleg_file = nfsd_file_get(fp->fi_fds[O_RDONLY]);
6261		spin_unlock(&fp->fi_lock);
6262		if (nf)
6263			nfsd_file_put(nf);
6264	}
6265	return true;
6266}
6267
6268/*
6269 * The Linux NFS server does not offer write delegations to NFSv4.0
6270 * clients in order to avoid conflicts between write delegations and
6271 * GETATTRs requesting CHANGE or SIZE attributes.
6272 *
6273 * With NFSv4.1 and later minorversions, the SEQUENCE operation that
6274 * begins each COMPOUND contains a client ID. Delegation recall can
6275 * be avoided when the server recognizes the client sending a
6276 * GETATTR also holds write delegation it conflicts with.
6277 *
6278 * However, the NFSv4.0 protocol does not enable a server to
6279 * determine that a GETATTR originated from the client holding the
6280 * conflicting delegation versus coming from some other client. Per
6281 * RFC 7530 Section 16.7.5, the server must recall or send a
6282 * CB_GETATTR even when the GETATTR originates from the client that
6283 * holds the conflicting delegation.
6284 *
6285 * An NFSv4.0 client can trigger a pathological situation if it
6286 * always sends a DELEGRETURN preceded by a conflicting GETATTR in
6287 * the same COMPOUND. COMPOUND execution will always stop at the
6288 * GETATTR and the DELEGRETURN will never get executed. The server
6289 * eventually revokes the delegation, which can result in loss of
6290 * open or lock state.
6291 */
6292static void
6293nfs4_open_delegation(struct svc_rqst *rqstp, struct nfsd4_open *open,
6294		     struct nfs4_ol_stateid *stp, struct svc_fh *currentfh,
6295		     struct svc_fh *fh)
6296{
6297	struct nfs4_openowner *oo = openowner(stp->st_stateowner);
6298	bool deleg_ts = nfsd4_want_deleg_timestamps(open);
6299	struct nfs4_client *clp = stp->st_stid.sc_client;
6300	struct svc_fh *parent = NULL;
6301	struct nfs4_delegation *dp;
6302	struct kstat stat;
6303	int status = 0;
6304	int cb_up;
6305
6306	cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
6307	open->op_recall = false;
6308	switch (open->op_claim_type) {
6309		case NFS4_OPEN_CLAIM_PREVIOUS:
6310			if (!cb_up)
6311				open->op_recall = true;
6312			break;
6313		case NFS4_OPEN_CLAIM_NULL:
6314			parent = currentfh;
6315			fallthrough;
6316		case NFS4_OPEN_CLAIM_FH:
6317			/*
6318			 * Let's not give out any delegations till everyone's
6319			 * had the chance to reclaim theirs, *and* until
6320			 * NLM locks have all been reclaimed:
6321			 */
6322			if (locks_in_grace(clp->net))
6323				goto out_no_deleg;
6324			if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
6325				goto out_no_deleg;
6326			if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
6327					!clp->cl_minorversion)
6328				goto out_no_deleg;
6329			break;
6330		default:
6331			goto out_no_deleg;
6332	}
6333	dp = nfs4_set_delegation(open, stp, parent);
6334	if (IS_ERR(dp))
6335		goto out_no_deleg;
6336
6337	memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
6338
6339	if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
6340		struct file *f = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
6341
6342		if (!nfsd4_add_rdaccess_to_wrdeleg(rqstp, open, fh, stp) ||
6343				!nfs4_delegation_stat(dp, currentfh, &stat)) {
6344			nfs4_put_stid(&dp->dl_stid);
6345			destroy_delegation(dp);
6346			goto out_no_deleg;
6347		}
6348		open->op_delegate_type = deleg_ts ? OPEN_DELEGATE_WRITE_ATTRS_DELEG :
6349						    OPEN_DELEGATE_WRITE;
6350		dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
6351		dp->dl_cb_fattr.ncf_initial_cinfo = nfsd4_change_attribute(&stat);
6352		dp->dl_atime = stat.atime;
6353		dp->dl_ctime = stat.ctime;
6354		dp->dl_mtime = stat.mtime;
6355		spin_lock(&f->f_lock);
6356		if (deleg_ts)
6357			f->f_mode |= FMODE_NOCMTIME;
6358		spin_unlock(&f->f_lock);
6359		trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
6360	} else {
6361		open->op_delegate_type = deleg_ts && nfs4_delegation_stat(dp, currentfh, &stat) ?
6362					 OPEN_DELEGATE_READ_ATTRS_DELEG : OPEN_DELEGATE_READ;
6363		dp->dl_atime = stat.atime;
6364		trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
6365	}
6366	nfs4_put_stid(&dp->dl_stid);
6367	return;
6368out_no_deleg:
6369	open->op_delegate_type = OPEN_DELEGATE_NONE;
6370
6371	/* 4.1 client asking for a delegation? */
6372	if (open->op_deleg_want)
6373		nfsd4_open_deleg_none_ext(open, status);
6374	return;
6375}
6376
6377static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
6378					struct nfs4_delegation *dp)
6379{
6380	if (deleg_is_write(dp->dl_type)) {
6381		if (open->op_deleg_want & OPEN4_SHARE_ACCESS_WANT_READ_DELEG) {
6382			open->op_delegate_type = OPEN_DELEGATE_NONE_EXT;
6383			open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
6384		} else if (open->op_deleg_want & OPEN4_SHARE_ACCESS_WANT_WRITE_DELEG) {
6385			open->op_delegate_type = OPEN_DELEGATE_NONE_EXT;
6386			open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
6387		}
6388	}
6389	/* Otherwise the client must be confused wanting a delegation
6390	 * it already has, therefore we don't return
6391	 * OPEN_DELEGATE_NONE_EXT and reason.
6392	 */
6393}
6394
6395/* Are we returning only a delegation stateid? */
6396static bool open_xor_delegation(struct nfsd4_open *open)
6397{
6398	if (!(open->op_deleg_want & OPEN4_SHARE_ACCESS_WANT_OPEN_XOR_DELEGATION))
6399		return false;
6400	/* Did we actually get a delegation? */
6401	if (!deleg_is_read(open->op_delegate_type) && !deleg_is_write(open->op_delegate_type))
6402		return false;
6403	return true;
6404}
6405
6406/**
6407 * nfsd4_process_open2 - finish open processing
6408 * @rqstp: the RPC transaction being executed
6409 * @current_fh: NFSv4 COMPOUND's current filehandle
6410 * @open: OPEN arguments
6411 *
6412 * If successful, (1) truncate the file if open->op_truncate was
6413 * set, (2) set open->op_stateid, (3) set open->op_delegation.
6414 *
6415 * Returns %nfs_ok on success; otherwise an nfs4stat value in
6416 * network byte order is returned.
6417 */
6418__be32
6419nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6420{
6421	struct nfsd4_compoundres *resp = rqstp->rq_resp;
6422	struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6423	struct nfs4_file *fp = NULL;
6424	struct nfs4_ol_stateid *stp = NULL;
6425	struct nfs4_delegation *dp = NULL;
6426	__be32 status;
6427	bool new_stp = false;
6428
6429	/*
6430	 * Lookup file; if found, lookup stateid and check open request,
6431	 * and check for delegations in the process of being recalled.
6432	 * If not found, create the nfs4_file struct
6433	 */
6434	fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6435	if (unlikely(!fp))
6436		return nfserr_jukebox;
6437	if (fp != open->op_file) {
6438		status = nfs4_check_deleg(cl, open, &dp);
6439		if (status)
6440			goto out;
6441		if (dp && nfsd4_is_deleg_cur(open) &&
6442				(dp->dl_stid.sc_file != fp)) {
6443			/*
6444			 * RFC8881 section 8.2.4 mandates the server to return
6445			 * NFS4ERR_BAD_STATEID if the selected table entry does
6446			 * not match the current filehandle. However returning
6447			 * NFS4ERR_BAD_STATEID in the OPEN can cause the client
6448			 * to repeatedly retry the operation with the same
6449			 * stateid, since the stateid itself is valid. To avoid
6450			 * this situation NFSD returns NFS4ERR_INVAL instead.
6451			 */
6452			status = nfserr_inval;
6453			goto out;
6454		}
6455		stp = nfsd4_find_and_lock_existing_open(fp, open);
6456	} else {
6457		open->op_file = NULL;
6458		status = nfserr_bad_stateid;
6459		if (nfsd4_is_deleg_cur(open))
6460			goto out;
6461	}
6462
6463	if (!stp) {
6464		stp = init_open_stateid(fp, open);
6465		if (!stp) {
6466			status = nfserr_jukebox;
6467			goto out;
6468		}
6469
6470		if (!open->op_stp)
6471			new_stp = true;
6472	}
6473
6474	/*
6475	 * OPEN the file, or upgrade an existing OPEN.
6476	 * If truncate fails, the OPEN fails.
6477	 *
6478	 * stp is already locked.
6479	 */
6480	if (!new_stp) {
6481		/* Stateid was found, this is an OPEN upgrade */
6482		status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6483		if (status) {
6484			mutex_unlock(&stp->st_mutex);
6485			goto out;
6486		}
6487	} else {
6488		status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6489		if (status) {
6490			release_open_stateid(stp);
6491			mutex_unlock(&stp->st_mutex);
6492			goto out;
6493		}
6494
6495		stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6496							open->op_odstate);
6497		if (stp->st_clnt_odstate == open->op_odstate)
6498			open->op_odstate = NULL;
6499	}
6500
6501	nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6502	mutex_unlock(&stp->st_mutex);
6503
6504	if (nfsd4_has_session(&resp->cstate)) {
6505		if (open->op_deleg_want & OPEN4_SHARE_ACCESS_WANT_NO_DELEG) {
6506			open->op_delegate_type = OPEN_DELEGATE_NONE_EXT;
6507			open->op_why_no_deleg = WND4_NOT_WANTED;
6508			goto nodeleg;
6509		}
6510	}
6511
6512	/*
6513	* Attempt to hand out a delegation. No error return, because the
6514	* OPEN succeeds even if we fail.
6515	*/
6516	nfs4_open_delegation(rqstp, open, stp,
6517		&resp->cstate.current_fh, current_fh);
6518
6519	/*
6520	 * If there is an existing open stateid, it must be updated and
6521	 * returned. Only respect WANT_OPEN_XOR_DELEGATION when a new
6522	 * open stateid would have to be created.
6523	 */
6524	if (new_stp && open_xor_delegation(open)) {
6525		memcpy(&open->op_stateid, &zero_stateid, sizeof(open->op_stateid));
6526		open->op_rflags |= OPEN4_RESULT_NO_OPEN_STATEID;
6527		release_open_stateid(stp);
6528	}
6529nodeleg:
6530	status = nfs_ok;
6531	trace_nfsd_open(&stp->st_stid.sc_stateid);
6532out:
6533	/* 4.1 client trying to upgrade/downgrade delegation? */
6534	if (open->op_delegate_type == OPEN_DELEGATE_NONE && dp &&
6535	    open->op_deleg_want)
6536		nfsd4_deleg_xgrade_none_ext(open, dp);
6537
6538	if (fp)
6539		put_nfs4_file(fp);
6540	if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6541		open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6542	/*
6543	* To finish the open response, we just need to set the rflags.
6544	*/
6545	open->op_rflags |= NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6546	if (nfsd4_has_session(&resp->cstate))
6547		open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6548	else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6549		open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6550
6551	if (dp)
6552		nfs4_put_stid(&dp->dl_stid);
6553	if (stp)
6554		nfs4_put_stid(&stp->st_stid);
6555
6556	return status;
6557}
6558
6559void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6560			      struct nfsd4_open *open)
6561{
6562	if (open->op_openowner)
6563		nfs4_put_stateowner(&open->op_openowner->oo_owner);
6564	if (open->op_file)
6565		kmem_cache_free(file_slab, open->op_file);
6566	if (open->op_stp)
6567		nfs4_put_stid(&open->op_stp->st_stid);
6568	if (open->op_odstate)
6569		kmem_cache_free(odstate_slab, open->op_odstate);
6570}
6571
6572__be32
6573nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6574	    union nfsd4_op_u *u)
6575{
6576	clientid_t *clid = &u->renew;
6577	struct nfs4_client *clp;
6578	__be32 status;
6579	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6580
6581	trace_nfsd_clid_renew(clid);
6582	status = set_client(clid, cstate, nn);
6583	if (status)
6584		return status;
6585	clp = cstate->clp;
6586	if (!list_empty(&clp->cl_delegations)
6587			&& clp->cl_cb_state != NFSD4_CB_UP)
6588		return nfserr_cb_path_down;
6589	return nfs_ok;
6590}
6591
6592static void
6593nfsd4_end_grace(struct nfsd_net *nn)
6594{
6595	/* do nothing if grace period already ended */
6596	if (nn->grace_ended)
6597		return;
6598
6599	trace_nfsd_grace_complete(nn);
6600	nn->grace_ended = true;
6601	/*
6602	 * If the server goes down again right now, an NFSv4
6603	 * client will still be allowed to reclaim after it comes back up,
6604	 * even if it hasn't yet had a chance to reclaim state this time.
6605	 *
6606	 */
6607	nfsd4_record_grace_done(nn);
6608	/*
6609	 * At this point, NFSv4 clients can still reclaim.  But if the
6610	 * server crashes, any that have not yet reclaimed will be out
6611	 * of luck on the next boot.
6612	 *
6613	 * (NFSv4.1+ clients are considered to have reclaimed once they
6614	 * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
6615	 * have reclaimed after their first OPEN.)
6616	 */
6617	locks_end_grace(&nn->nfsd4_manager);
6618	/*
6619	 * At this point, and once lockd and/or any other containers
6620	 * exit their grace period, further reclaims will fail and
6621	 * regular locking can resume.
6622	 */
6623}
6624
6625/**
6626 * nfsd4_force_end_grace - forcibly end the NFSv4 grace period
6627 * @nn: network namespace for the server instance to be updated
6628 *
6629 * Forces bypass of normal grace period completion, then schedules
6630 * the laundromat to end the grace period immediately. Does not wait
6631 * for the grace period to fully terminate before returning.
6632 *
6633 * Return values:
6634 *   %true: Grace termination schedule
6635 *   %false: No action was taken
6636 */
6637bool nfsd4_force_end_grace(struct nfsd_net *nn)
6638{
6639	if (!nn->client_tracking_ops)
6640		return false;
6641	spin_lock(&nn->client_lock);
6642	if (nn->grace_ended || !nn->client_tracking_active) {
6643		spin_unlock(&nn->client_lock);
6644		return false;
6645	}
6646	WRITE_ONCE(nn->grace_end_forced, true);
6647	mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
6648	spin_unlock(&nn->client_lock);
6649	return true;
6650}
6651
6652/*
6653 * If we've waited a lease period but there are still clients trying to
6654 * reclaim, wait a little longer to give them a chance to finish.
6655 */
6656static bool clients_still_reclaiming(struct nfsd_net *nn)
6657{
6658	time64_t double_grace_period_end = nn->boot_time +
6659					   2 * nn->nfsd4_lease;
6660
6661	if (READ_ONCE(nn->grace_end_forced))
6662		return false;
6663	if (nn->track_reclaim_completes &&
6664			atomic_read(&nn->nr_reclaim_complete) ==
6665			nn->reclaim_str_hashtbl_size)
6666		return false;
6667	if (!nn->somebody_reclaimed)
6668		return false;
6669	nn->somebody_reclaimed = false;
6670	/*
6671	 * If we've given them *two* lease times to reclaim, and they're
6672	 * still not done, give up:
6673	 */
6674	if (ktime_get_boottime_seconds() > double_grace_period_end)
6675		return false;
6676	return true;
6677}
6678
6679struct laundry_time {
6680	time64_t cutoff;
6681	time64_t new_timeo;
6682};
6683
6684static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6685{
6686	time64_t time_remaining;
6687
6688	if (last_refresh < lt->cutoff)
6689		return true;
6690	time_remaining = last_refresh - lt->cutoff;
6691	lt->new_timeo = min(lt->new_timeo, time_remaining);
6692	return false;
6693}
6694
6695#ifdef CONFIG_NFSD_V4_2_INTER_SSC
6696void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6697{
6698	spin_lock_init(&nn->nfsd_ssc_lock);
6699	INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6700	init_waitqueue_head(&nn->nfsd_ssc_waitq);
6701}
6702
6703/*
6704 * This is called when nfsd is being shutdown, after all inter_ssc
6705 * cleanup were done, to destroy the ssc delayed unmount list.
6706 */
6707static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6708{
6709	struct nfsd4_ssc_umount_item *ni = NULL;
6710	struct nfsd4_ssc_umount_item *tmp;
6711
6712	spin_lock(&nn->nfsd_ssc_lock);
6713	list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6714		list_del(&ni->nsui_list);
6715		spin_unlock(&nn->nfsd_ssc_lock);
6716		mntput(ni->nsui_vfsmount);
6717		kfree(ni);
6718		spin_lock(&nn->nfsd_ssc_lock);
6719	}
6720	spin_unlock(&nn->nfsd_ssc_lock);
6721}
6722
6723static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6724{
6725	bool do_wakeup = false;
6726	struct nfsd4_ssc_umount_item *ni = NULL;
6727	struct nfsd4_ssc_umount_item *tmp;
6728
6729	spin_lock(&nn->nfsd_ssc_lock);
6730	list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6731		if (time_after(jiffies, ni->nsui_expire)) {
6732			if (refcount_read(&ni->nsui_refcnt) > 1)
6733				continue;
6734
6735			/* mark being unmount */
6736			ni->nsui_busy = true;
6737			spin_unlock(&nn->nfsd_ssc_lock);
6738			mntput(ni->nsui_vfsmount);
6739			spin_lock(&nn->nfsd_ssc_lock);
6740
6741			/* waiters need to start from begin of list */
6742			list_del(&ni->nsui_list);
6743			kfree(ni);
6744
6745			/* wakeup ssc_connect waiters */
6746			do_wakeup = true;
6747			continue;
6748		}
6749		break;
6750	}
6751	if (do_wakeup)
6752		wake_up_all(&nn->nfsd_ssc_waitq);
6753	spin_unlock(&nn->nfsd_ssc_lock);
6754}
6755#endif
6756
6757/* Check if any lock belonging to this lockowner has any blockers */
6758static bool
6759nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6760{
6761	struct file_lock_context *ctx;
6762	struct nfs4_ol_stateid *stp;
6763	struct nfs4_file *nf;
6764
6765	list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6766		nf = stp->st_stid.sc_file;
6767		ctx = locks_inode_context(nf->fi_inode);
6768		if (!ctx)
6769			continue;
6770		if (locks_owner_has_blockers(ctx, lo))
6771			return true;
6772	}
6773	return false;
6774}
6775
6776static bool
6777nfs4_anylock_blockers(struct nfs4_client *clp)
6778{
6779	int i;
6780	struct nfs4_stateowner *so;
6781	struct nfs4_lockowner *lo;
6782
6783	if (atomic_read(&clp->cl_delegs_in_recall))
6784		return true;
6785	spin_lock(&clp->cl_lock);
6786	for (i = 0; i < OWNER_HASH_SIZE; i++) {
6787		list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6788				so_strhash) {
6789			if (so->so_is_open_owner)
6790				continue;
6791			lo = lockowner(so);
6792			if (nfs4_lockowner_has_blockers(lo)) {
6793				spin_unlock(&clp->cl_lock);
6794				return true;
6795			}
6796		}
6797	}
6798	spin_unlock(&clp->cl_lock);
6799	return false;
6800}
6801
6802static void
6803nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6804				struct laundry_time *lt)
6805{
6806	unsigned int maxreap, reapcnt = 0;
6807	struct list_head *pos, *next;
6808	struct nfs4_client *clp;
6809
6810	maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6811			NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6812	INIT_LIST_HEAD(reaplist);
6813	spin_lock(&nn->client_lock);
6814	list_for_each_safe(pos, next, &nn->client_lru) {
6815		clp = list_entry(pos, struct nfs4_client, cl_lru);
6816		if (clp->cl_state == NFSD4_EXPIRABLE)
6817			goto exp_client;
6818		if (!state_expired(lt, clp->cl_time))
6819			break;
6820		if (!atomic_read(&clp->cl_rpc_users)) {
6821			if (clp->cl_state == NFSD4_ACTIVE)
6822				atomic_inc(&nn->nfsd_courtesy_clients);
6823			clp->cl_state = NFSD4_COURTESY;
6824		}
6825		if (!client_has_state(clp))
6826			goto exp_client;
6827		if (!nfs4_anylock_blockers(clp))
6828			if (reapcnt >= maxreap)
6829				continue;
6830exp_client:
6831		if (!mark_client_expired_locked(clp)) {
6832			list_add(&clp->cl_lru, reaplist);
6833			reapcnt++;
6834		}
6835	}
6836	spin_unlock(&nn->client_lock);
6837}
6838
6839static void
6840nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6841				struct list_head *reaplist)
6842{
6843	unsigned int maxreap = 0, reapcnt = 0;
6844	struct list_head *pos, *next;
6845	struct nfs4_client *clp;
6846
6847	maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6848	INIT_LIST_HEAD(reaplist);
6849
6850	spin_lock(&nn->client_lock);
6851	list_for_each_safe(pos, next, &nn->client_lru) {
6852		clp = list_entry(pos, struct nfs4_client, cl_lru);
6853		if (clp->cl_state == NFSD4_ACTIVE)
6854			break;
6855		if (reapcnt >= maxreap)
6856			break;
6857		if (!mark_client_expired_locked(clp)) {
6858			list_add(&clp->cl_lru, reaplist);
6859			reapcnt++;
6860		}
6861	}
6862	spin_unlock(&nn->client_lock);
6863}
6864
6865static void
6866nfs4_process_client_reaplist(struct list_head *reaplist)
6867{
6868	struct list_head *pos, *next;
6869	struct nfs4_client *clp;
6870
6871	list_for_each_safe(pos, next, reaplist) {
6872		clp = list_entry(pos, struct nfs4_client, cl_lru);
6873		trace_nfsd_clid_purged(&clp->cl_clientid);
6874		list_del_init(&clp->cl_lru);
6875		expire_client(clp);
6876	}
6877}
6878
6879static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6880				      struct laundry_time *lt)
6881{
6882	struct nfs4_client *clp;
6883
6884	spin_lock(&nn->client_lock);
6885	if (nn->nfs40_last_revoke == 0 ||
6886	    nn->nfs40_last_revoke > lt->cutoff) {
6887		spin_unlock(&nn->client_lock);
6888		return;
6889	}
6890	nn->nfs40_last_revoke = 0;
6891
6892retry:
6893	list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6894		unsigned long id, tmp;
6895		struct nfs4_stid *stid;
6896
6897		if (atomic_read(&clp->cl_admin_revoked) == 0)
6898			continue;
6899
6900		spin_lock(&clp->cl_lock);
6901		idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6902			if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6903				refcount_inc(&stid->sc_count);
6904				spin_unlock(&nn->client_lock);
6905				/* this function drops ->cl_lock */
6906				nfsd4_drop_revoked_stid(stid);
6907				nfs4_put_stid(stid);
6908				spin_lock(&nn->client_lock);
6909				goto retry;
6910			}
6911		spin_unlock(&clp->cl_lock);
6912	}
6913	spin_unlock(&nn->client_lock);
6914}
6915
6916static time64_t
6917nfs4_laundromat(struct nfsd_net *nn)
6918{
6919	struct nfs4_openowner *oo;
6920	struct nfs4_delegation *dp;
6921	struct nfs4_ol_stateid *stp;
6922	struct nfsd4_blocked_lock *nbl;
6923	struct list_head *pos, *next, reaplist;
6924	struct laundry_time lt = {
6925		.cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6926		.new_timeo = nn->nfsd4_lease
6927	};
6928	struct nfs4_cpntf_state *cps;
6929	copy_stateid_t *cps_t;
6930	int i;
6931
6932	if (clients_still_reclaiming(nn)) {
6933		lt.new_timeo = 0;
6934		goto out;
6935	}
6936	nfsd4_end_grace(nn);
6937
6938	spin_lock(&nn->s2s_cp_lock);
6939	idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6940		cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6941		if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6942				state_expired(&lt, cps->cpntf_time))
6943			_free_cpntf_state_locked(nn, cps);
6944	}
6945	spin_unlock(&nn->s2s_cp_lock);
6946	nfsd4_async_copy_reaper(nn);
6947	nfs4_get_client_reaplist(nn, &reaplist, &lt);
6948	nfs4_process_client_reaplist(&reaplist);
6949
6950	nfs40_clean_admin_revoked(nn, &lt);
6951
6952	spin_lock(&state_lock);
6953	list_for_each_safe(pos, next, &nn->del_recall_lru) {
6954		dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6955		if (!state_expired(&lt, dp->dl_time))
6956			break;
6957		refcount_inc(&dp->dl_stid.sc_count);
6958		unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6959		list_add(&dp->dl_recall_lru, &reaplist);
6960	}
6961	spin_unlock(&state_lock);
6962	while (!list_empty(&reaplist)) {
6963		dp = list_first_entry(&reaplist, struct nfs4_delegation,
6964					dl_recall_lru);
6965		list_del_init(&dp->dl_recall_lru);
6966		revoke_delegation(dp);
6967	}
6968
6969	spin_lock(&nn->client_lock);
6970	while (!list_empty(&nn->close_lru)) {
6971		oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6972					oo_close_lru);
6973		if (!state_expired(&lt, oo->oo_time))
6974			break;
6975		list_del_init(&oo->oo_close_lru);
6976		stp = oo->oo_last_closed_stid;
6977		oo->oo_last_closed_stid = NULL;
6978		spin_unlock(&nn->client_lock);
6979		nfs4_put_stid(&stp->st_stid);
6980		spin_lock(&nn->client_lock);
6981	}
6982	spin_unlock(&nn->client_lock);
6983
6984	/*
6985	 * It's possible for a client to try and acquire an already held lock
6986	 * that is being held for a long time, and then lose interest in it.
6987	 * So, we clean out any un-revisited request after a lease period
6988	 * under the assumption that the client is no longer interested.
6989	 *
6990	 * RFC5661, sec. 9.6 states that the client must not rely on getting
6991	 * notifications and must continue to poll for locks, even when the
6992	 * server supports them. Thus this shouldn't lead to clients blocking
6993	 * indefinitely once the lock does become free.
6994	 */
6995	BUG_ON(!list_empty(&reaplist));
6996	spin_lock(&nn->blocked_locks_lock);
6997	while (!list_empty(&nn->blocked_locks_lru)) {
6998		nbl = list_first_entry(&nn->blocked_locks_lru,
6999					struct nfsd4_blocked_lock, nbl_lru);
7000		if (!state_expired(&lt, nbl->nbl_time))
7001			break;
7002		list_move(&nbl->nbl_lru, &reaplist);
7003		list_del_init(&nbl->nbl_list);
7004	}
7005	spin_unlock(&nn->blocked_locks_lock);
7006
7007	while (!list_empty(&reaplist)) {
7008		nbl = list_first_entry(&reaplist,
7009					struct nfsd4_blocked_lock, nbl_lru);
7010		list_del_init(&nbl->nbl_lru);
7011		free_blocked_lock(nbl);
7012	}
7013#ifdef CONFIG_NFSD_V4_2_INTER_SSC
7014	/* service the server-to-server copy delayed unmount list */
7015	nfsd4_ssc_expire_umount(nn);
7016#endif
7017	if (atomic_long_read(&num_delegations) >= max_delegations)
7018		deleg_reaper(nn);
7019out:
7020	return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
7021}
7022
7023static void laundromat_main(struct work_struct *);
7024
7025static void
7026laundromat_main(struct work_struct *laundry)
7027{
7028	time64_t t;
7029	struct delayed_work *dwork = to_delayed_work(laundry);
7030	struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
7031					   laundromat_work);
7032
7033	t = nfs4_laundromat(nn);
7034	queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
7035}
7036
7037static void
7038courtesy_client_reaper(struct nfsd_net *nn)
7039{
7040	struct list_head reaplist;
7041
7042	nfs4_get_courtesy_client_reaplist(nn, &reaplist);
7043	nfs4_process_client_reaplist(&reaplist);
7044}
7045
7046static void
7047deleg_reaper(struct nfsd_net *nn)
7048{
7049	struct list_head *pos, *next;
7050	struct nfs4_client *clp;
7051
7052	spin_lock(&nn->client_lock);
7053	list_for_each_safe(pos, next, &nn->client_lru) {
7054		clp = list_entry(pos, struct nfs4_client, cl_lru);
7055
7056		if (clp->cl_state != NFSD4_ACTIVE)
7057			continue;
7058		if (list_empty(&clp->cl_delegations))
7059			continue;
7060		if (atomic_read(&clp->cl_delegs_in_recall))
7061			continue;
7062		if (test_and_set_bit(NFSD4_CALLBACK_RUNNING, &clp->cl_ra->ra_cb.cb_flags))
7063			continue;
7064		if (ktime_get_boottime_seconds() - clp->cl_ra_time < 5)
7065			continue;
7066		if (clp->cl_cb_state != NFSD4_CB_UP)
7067			continue;
7068
7069		/* release in nfsd4_cb_recall_any_release */
7070		kref_get(&clp->cl_nfsdfs.cl_ref);
7071		clp->cl_ra_time = ktime_get_boottime_seconds();
7072		clp->cl_ra->ra_keep = 0;
7073		clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
7074						BIT(RCA4_TYPE_MASK_WDATA_DLG);
7075		trace_nfsd_cb_recall_any(clp->cl_ra);
7076		nfsd4_run_cb(&clp->cl_ra->ra_cb);
7077	}
7078	spin_unlock(&nn->client_lock);
7079}
7080
7081static void
7082nfsd4_state_shrinker_worker(struct work_struct *work)
7083{
7084	struct nfsd_net *nn = container_of(work, struct nfsd_net,
7085				nfsd_shrinker_work);
7086
7087	courtesy_client_reaper(nn);
7088	deleg_reaper(nn);
7089}
7090
7091static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
7092{
7093	if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
7094		return nfserr_bad_stateid;
7095	return nfs_ok;
7096}
7097
7098static
7099__be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
7100{
7101        __be32 status = nfserr_openmode;
7102
7103	/* For lock stateid's, we test the parent open, not the lock: */
7104	if (stp->st_openstp)
7105		stp = stp->st_openstp;
7106	if ((flags & WR_STATE) && !access_permit_write(stp))
7107                goto out;
7108	if ((flags & RD_STATE) && !access_permit_read(stp))
7109                goto out;
7110	status = nfs_ok;
7111out:
7112	return status;
7113}
7114
7115static inline __be32
7116check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
7117{
7118	if (ONE_STATEID(stateid) && (flags & RD_STATE))
7119		return nfs_ok;
7120	else if (opens_in_grace(net)) {
7121		/* Answer in remaining cases depends on existence of
7122		 * conflicting state; so we must wait out the grace period. */
7123		return nfserr_grace;
7124	} else if (flags & WR_STATE)
7125		return nfs4_share_conflict(current_fh,
7126				NFS4_SHARE_DENY_WRITE);
7127	else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
7128		return nfs4_share_conflict(current_fh,
7129				NFS4_SHARE_DENY_READ);
7130}
7131
7132static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
7133{
7134	/*
7135	 * When sessions are used the stateid generation number is ignored
7136	 * when it is zero.
7137	 */
7138	if (has_session && in->si_generation == 0)
7139		return nfs_ok;
7140
7141	if (in->si_generation == ref->si_generation)
7142		return nfs_ok;
7143
7144	/* If the client sends us a stateid from the future, it's buggy: */
7145	if (nfsd4_stateid_generation_after(in, ref))
7146		return nfserr_bad_stateid;
7147	/*
7148	 * However, we could see a stateid from the past, even from a
7149	 * non-buggy client.  For example, if the client sends a lock
7150	 * while some IO is outstanding, the lock may bump si_generation
7151	 * while the IO is still in flight.  The client could avoid that
7152	 * situation by waiting for responses on all the IO requests,
7153	 * but better performance may result in retrying IO that
7154	 * receives an old_stateid error if requests are rarely
7155	 * reordered in flight:
7156	 */
7157	return nfserr_old_stateid;
7158}
7159
7160static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
7161{
7162	__be32 ret;
7163
7164	spin_lock(&s->sc_lock);
7165	ret = nfsd4_verify_open_stid(s);
7166	if (ret == nfs_ok)
7167		ret = check_stateid_generation(in, &s->sc_stateid, has_session);
7168	spin_unlock(&s->sc_lock);
7169	if (ret == nfserr_admin_revoked)
7170		nfsd40_drop_revoked_stid(s->sc_client,
7171					&s->sc_stateid);
7172	return ret;
7173}
7174
7175static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
7176{
7177	if (ols->st_stateowner->so_is_open_owner &&
7178	    !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
7179		return nfserr_bad_stateid;
7180	return nfs_ok;
7181}
7182
7183static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
7184{
7185	struct nfs4_stid *s;
7186	__be32 status = nfserr_bad_stateid;
7187
7188	if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
7189		CLOSE_STATEID(stateid))
7190		return status;
7191	spin_lock(&cl->cl_lock);
7192	s = find_stateid_locked(cl, stateid);
7193	if (!s)
7194		goto out_unlock;
7195	status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
7196	if (status)
7197		goto out_unlock;
7198	status = nfsd4_verify_open_stid(s);
7199	if (status)
7200		goto out_unlock;
7201
7202	switch (s->sc_type) {
7203	case SC_TYPE_DELEG:
7204		status = nfs_ok;
7205		break;
7206	case SC_TYPE_OPEN:
7207	case SC_TYPE_LOCK:
7208		status = nfsd4_check_openowner_confirmed(openlockstateid(s));
7209		break;
7210	default:
7211		printk("unknown stateid type %x\n", s->sc_type);
7212		status = nfserr_bad_stateid;
7213	}
7214out_unlock:
7215	spin_unlock(&cl->cl_lock);
7216	if (status == nfserr_admin_revoked)
7217		nfsd40_drop_revoked_stid(cl, stateid);
7218	return status;
7219}
7220
7221__be32
7222nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
7223		     stateid_t *stateid,
7224		     unsigned short typemask, unsigned short statusmask,
7225		     struct nfs4_stid **s, struct nfsd_net *nn)
7226{
7227	__be32 status;
7228	struct nfs4_stid *stid;
7229	bool return_revoked = false;
7230
7231	/*
7232	 *  only return revoked delegations if explicitly asked.
7233	 *  otherwise we report revoked or bad_stateid status.
7234	 */
7235	if (statusmask & SC_STATUS_REVOKED)
7236		return_revoked = true;
7237	if (typemask & SC_TYPE_DELEG)
7238		/* Always allow REVOKED for DELEG so we can
7239		 * return the appropriate error.
7240		 */
7241		statusmask |= SC_STATUS_REVOKED;
7242
7243	statusmask |= SC_STATUS_ADMIN_REVOKED | SC_STATUS_FREEABLE;
7244
7245	if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
7246		CLOSE_STATEID(stateid))
7247		return nfserr_bad_stateid;
7248	status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
7249	if (status == nfserr_stale_clientid) {
7250		if (cstate->session)
7251			return nfserr_bad_stateid;
7252		return nfserr_stale_stateid;
7253	}
7254	if (status)
7255		return status;
7256	stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
7257	if (!stid)
7258		return nfserr_bad_stateid;
7259	if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
7260		nfs4_put_stid(stid);
7261		return nfserr_deleg_revoked;
7262	}
7263	if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
7264		nfsd40_drop_revoked_stid(cstate->clp, stateid);
7265		nfs4_put_stid(stid);
7266		return nfserr_admin_revoked;
7267	}
7268	*s = stid;
7269	return nfs_ok;
7270}
7271
7272static struct nfsd_file *
7273nfs4_find_file(struct nfs4_stid *s, int flags)
7274{
7275	struct nfsd_file *ret = NULL;
7276
7277	if (!s || s->sc_status)
7278		return NULL;
7279
7280	switch (s->sc_type) {
7281	case SC_TYPE_DELEG:
7282	case SC_TYPE_OPEN:
7283	case SC_TYPE_LOCK:
7284		if (flags & RD_STATE)
7285			ret = find_readable_file(s->sc_file);
7286		else
7287			ret = find_writeable_file(s->sc_file);
7288	}
7289
7290	return ret;
7291}
7292
7293static __be32
7294nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
7295{
7296	__be32 status;
7297
7298	status = nfsd4_check_openowner_confirmed(ols);
7299	if (status)
7300		return status;
7301	return nfs4_check_openmode(ols, flags);
7302}
7303
7304static __be32
7305nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
7306		struct nfsd_file **nfp, int flags)
7307{
7308	int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
7309	struct nfsd_file *nf;
7310	__be32 status;
7311
7312	nf = nfs4_find_file(s, flags);
7313	if (nf) {
7314		status = nfsd_permission(&rqstp->rq_cred,
7315					 fhp->fh_export, fhp->fh_dentry,
7316				acc | NFSD_MAY_OWNER_OVERRIDE);
7317		if (status) {
7318			nfsd_file_put(nf);
7319			goto out;
7320		}
7321	} else {
7322		status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
7323		if (status)
7324			return status;
7325	}
7326	*nfp = nf;
7327out:
7328	return status;
7329}
7330static void
7331_free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
7332{
7333	WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
7334	if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
7335		return;
7336	list_del(&cps->cp_list);
7337	idr_remove(&nn->s2s_cp_stateids,
7338		   cps->cp_stateid.cs_stid.si_opaque.so_id);
7339	kfree(cps);
7340}
7341/*
7342 * A READ from an inter server to server COPY will have a
7343 * copy stateid. Look up the copy notify stateid from the
7344 * idr structure and take a reference on it.
7345 */
7346__be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
7347			  struct nfs4_client *clp,
7348			  struct nfs4_cpntf_state **cps)
7349{
7350	copy_stateid_t *cps_t;
7351	struct nfs4_cpntf_state *state = NULL;
7352
7353	if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
7354		return nfserr_bad_stateid;
7355	spin_lock(&nn->s2s_cp_lock);
7356	cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
7357	if (cps_t) {
7358		state = container_of(cps_t, struct nfs4_cpntf_state,
7359				     cp_stateid);
7360		if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
7361			state = NULL;
7362			goto unlock;
7363		}
7364		if (!clp)
7365			refcount_inc(&state->cp_stateid.cs_count);
7366		else
7367			_free_cpntf_state_locked(nn, state);
7368	}
7369unlock:
7370	spin_unlock(&nn->s2s_cp_lock);
7371	if (!state)
7372		return nfserr_bad_stateid;
7373	if (!clp)
7374		*cps = state;
7375	return 0;
7376}
7377
7378static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
7379			       struct nfs4_stid **stid)
7380{
7381	__be32 status;
7382	struct nfs4_cpntf_state *cps = NULL;
7383	struct nfs4_client *found;
7384
7385	status = manage_cpntf_state(nn, st, NULL, &cps);
7386	if (status)
7387		return status;
7388
7389	cps->cpntf_time = ktime_get_boottime_seconds();
7390
7391	status = nfserr_expired;
7392	found = lookup_clientid(&cps->cp_p_clid, true, nn);
7393	if (!found)
7394		goto out;
7395
7396	*stid = find_stateid_by_type(found, &cps->cp_p_stateid,
7397				     SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7398				     0);
7399	if (*stid)
7400		status = nfs_ok;
7401	else
7402		status = nfserr_bad_stateid;
7403
7404	put_client_renew(found);
7405out:
7406	nfs4_put_cpntf_state(nn, cps);
7407	return status;
7408}
7409
7410void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
7411{
7412	spin_lock(&nn->s2s_cp_lock);
7413	_free_cpntf_state_locked(nn, cps);
7414	spin_unlock(&nn->s2s_cp_lock);
7415}
7416
7417/**
7418 * nfs4_preprocess_stateid_op - find and prep stateid for an operation
7419 * @rqstp: incoming request from client
7420 * @cstate: current compound state
7421 * @fhp: filehandle associated with requested stateid
7422 * @stateid: stateid (provided by client)
7423 * @flags: flags describing type of operation to be done
7424 * @nfp: optional nfsd_file return pointer (may be NULL)
7425 * @cstid: optional returned nfs4_stid pointer (may be NULL)
7426 *
7427 * Given info from the client, look up a nfs4_stid for the operation. On
7428 * success, it returns a reference to the nfs4_stid and/or the nfsd_file
7429 * associated with it.
7430 */
7431__be32
7432nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
7433		struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
7434		stateid_t *stateid, int flags, struct nfsd_file **nfp,
7435		struct nfs4_stid **cstid)
7436{
7437	struct net *net = SVC_NET(rqstp);
7438	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7439	struct nfs4_stid *s = NULL;
7440	__be32 status;
7441
7442	if (nfp)
7443		*nfp = NULL;
7444
7445	if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
7446		status = check_special_stateids(net, fhp, stateid, flags);
7447		goto done;
7448	}
7449
7450	status = nfsd4_lookup_stateid(cstate, stateid,
7451				SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7452				0, &s, nn);
7453	if (status == nfserr_bad_stateid)
7454		status = find_cpntf_state(nn, stateid, &s);
7455	if (status)
7456		return status;
7457	status = nfsd4_stid_check_stateid_generation(stateid, s,
7458			nfsd4_has_session(cstate));
7459	if (status)
7460		goto out;
7461
7462	switch (s->sc_type) {
7463	case SC_TYPE_DELEG:
7464		status = nfs4_check_delegmode(delegstateid(s), flags);
7465		break;
7466	case SC_TYPE_OPEN:
7467	case SC_TYPE_LOCK:
7468		status = nfs4_check_olstateid(openlockstateid(s), flags);
7469		break;
7470	}
7471	if (status)
7472		goto out;
7473	status = nfs4_check_fh(fhp, s);
7474
7475done:
7476	if (status == nfs_ok && nfp)
7477		status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7478out:
7479	if (s) {
7480		if (!status && cstid)
7481			*cstid = s;
7482		else
7483			nfs4_put_stid(s);
7484	}
7485	return status;
7486}
7487
7488/*
7489 * Test if the stateid is valid
7490 */
7491__be32
7492nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7493		   union nfsd4_op_u *u)
7494{
7495	struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7496	struct nfsd4_test_stateid_id *stateid;
7497	struct nfs4_client *cl = cstate->clp;
7498
7499	list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7500		stateid->ts_id_status =
7501			nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7502
7503	return nfs_ok;
7504}
7505
7506static __be32
7507nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7508{
7509	struct nfs4_ol_stateid *stp = openlockstateid(s);
7510	__be32 ret;
7511
7512	ret = nfsd4_lock_ol_stateid(stp);
7513	if (ret)
7514		goto out_put_stid;
7515
7516	ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7517	if (ret)
7518		goto out;
7519
7520	ret = nfserr_locks_held;
7521	if (check_for_locks(stp->st_stid.sc_file,
7522			    lockowner(stp->st_stateowner)))
7523		goto out;
7524
7525	release_lock_stateid(stp);
7526	ret = nfs_ok;
7527
7528out:
7529	mutex_unlock(&stp->st_mutex);
7530out_put_stid:
7531	nfs4_put_stid(s);
7532	return ret;
7533}
7534
7535__be32
7536nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7537		   union nfsd4_op_u *u)
7538{
7539	struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7540	stateid_t *stateid = &free_stateid->fr_stateid;
7541	struct nfs4_stid *s;
7542	struct nfs4_delegation *dp;
7543	struct nfs4_client *cl = cstate->clp;
7544	__be32 ret = nfserr_bad_stateid;
7545
7546	spin_lock(&cl->cl_lock);
7547	s = find_stateid_locked(cl, stateid);
7548	if (!s || s->sc_status & SC_STATUS_CLOSED)
7549		goto out_unlock;
7550	if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7551		nfsd4_drop_revoked_stid(s);
7552		ret = nfs_ok;
7553		goto out;
7554	}
7555	spin_lock(&s->sc_lock);
7556	switch (s->sc_type) {
7557	case SC_TYPE_DELEG:
7558		if (s->sc_status & SC_STATUS_REVOKED) {
7559			s->sc_status |= SC_STATUS_CLOSED;
7560			spin_unlock(&s->sc_lock);
7561			dp = delegstateid(s);
7562			if (s->sc_status & SC_STATUS_FREEABLE)
7563				list_del_init(&dp->dl_recall_lru);
7564			s->sc_status |= SC_STATUS_FREED;
7565			spin_unlock(&cl->cl_lock);
7566			nfs4_put_stid(s);
7567			ret = nfs_ok;
7568			goto out;
7569		}
7570		ret = nfserr_locks_held;
7571		break;
7572	case SC_TYPE_OPEN:
7573		ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7574		if (ret)
7575			break;
7576		ret = nfserr_locks_held;
7577		break;
7578	case SC_TYPE_LOCK:
7579		spin_unlock(&s->sc_lock);
7580		refcount_inc(&s->sc_count);
7581		spin_unlock(&cl->cl_lock);
7582		ret = nfsd4_free_lock_stateid(stateid, s);
7583		goto out;
7584	}
7585	spin_unlock(&s->sc_lock);
7586out_unlock:
7587	spin_unlock(&cl->cl_lock);
7588out:
7589	return ret;
7590}
7591
7592static inline int
7593setlkflg (int type)
7594{
7595	return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7596		RD_STATE : WR_STATE;
7597}
7598
7599static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7600{
7601	struct svc_fh *current_fh = &cstate->current_fh;
7602	struct nfs4_stateowner *sop = stp->st_stateowner;
7603	__be32 status;
7604
7605	status = nfsd4_check_seqid(cstate, sop, seqid);
7606	if (status)
7607		return status;
7608	status = nfsd4_lock_ol_stateid(stp);
7609	if (status != nfs_ok)
7610		return status;
7611	status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7612	if (status == nfs_ok)
7613		status = nfs4_check_fh(current_fh, &stp->st_stid);
7614	if (status != nfs_ok)
7615		mutex_unlock(&stp->st_mutex);
7616	return status;
7617}
7618
7619/**
7620 * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7621 * @cstate: compund state
7622 * @seqid: seqid (provided by client)
7623 * @stateid: stateid (provided by client)
7624 * @typemask: mask of allowable types for this operation
7625 * @statusmask: mask of allowed states: 0 or STID_CLOSED
7626 * @stpp: return pointer for the stateid found
7627 * @nn: net namespace for request
7628 *
7629 * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7630 * return it in @stpp. On a nfs_ok return, the returned stateid will
7631 * have its st_mutex locked.
7632 */
7633static __be32
7634nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7635			 stateid_t *stateid,
7636			 unsigned short typemask, unsigned short statusmask,
7637			 struct nfs4_ol_stateid **stpp,
7638			 struct nfsd_net *nn)
7639{
7640	__be32 status;
7641	struct nfs4_stid *s;
7642	struct nfs4_ol_stateid *stp = NULL;
7643
7644	trace_nfsd_preprocess(seqid, stateid);
7645
7646	*stpp = NULL;
7647retry:
7648	status = nfsd4_lookup_stateid(cstate, stateid,
7649				      typemask, statusmask, &s, nn);
7650	if (status)
7651		return status;
7652	stp = openlockstateid(s);
7653	if (nfsd4_cstate_assign_replay(cstate, stp->st_stateowner) == -EAGAIN) {
7654		nfs4_put_stateowner(stp->st_stateowner);
7655		goto retry;
7656	}
7657
7658	status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7659	if (!status)
7660		*stpp = stp;
7661	else
7662		nfs4_put_stid(&stp->st_stid);
7663	return status;
7664}
7665
7666static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7667						 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7668{
7669	__be32 status;
7670	struct nfs4_openowner *oo;
7671	struct nfs4_ol_stateid *stp;
7672
7673	status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7674					  SC_TYPE_OPEN, 0, &stp, nn);
7675	if (status)
7676		return status;
7677	oo = openowner(stp->st_stateowner);
7678	if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7679		mutex_unlock(&stp->st_mutex);
7680		nfs4_put_stid(&stp->st_stid);
7681		return nfserr_bad_stateid;
7682	}
7683	*stpp = stp;
7684	return nfs_ok;
7685}
7686
7687__be32
7688nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7689		   union nfsd4_op_u *u)
7690{
7691	struct nfsd4_open_confirm *oc = &u->open_confirm;
7692	__be32 status;
7693	struct nfs4_openowner *oo;
7694	struct nfs4_ol_stateid *stp;
7695	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7696
7697	dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7698			cstate->current_fh.fh_dentry);
7699
7700	status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7701	if (status)
7702		return status;
7703
7704	status = nfs4_preprocess_seqid_op(cstate,
7705					  oc->oc_seqid, &oc->oc_req_stateid,
7706					  SC_TYPE_OPEN, 0, &stp, nn);
7707	if (status)
7708		goto out;
7709	oo = openowner(stp->st_stateowner);
7710	status = nfserr_bad_stateid;
7711	if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7712		mutex_unlock(&stp->st_mutex);
7713		goto put_stateid;
7714	}
7715	oo->oo_flags |= NFS4_OO_CONFIRMED;
7716	nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7717	mutex_unlock(&stp->st_mutex);
7718	trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7719	nfsd4_client_record_create(oo->oo_owner.so_client);
7720	status = nfs_ok;
7721put_stateid:
7722	nfs4_put_stid(&stp->st_stid);
7723out:
7724	nfsd4_bump_seqid(cstate, status);
7725	return status;
7726}
7727
7728static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7729{
7730	if (!test_access(access, stp))
7731		return;
7732	nfs4_file_put_access(stp->st_stid.sc_file, access);
7733	clear_access(access, stp);
7734}
7735
7736static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7737{
7738	switch (to_access) {
7739	case NFS4_SHARE_ACCESS_READ:
7740		nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7741		nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7742		break;
7743	case NFS4_SHARE_ACCESS_WRITE:
7744		nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7745		nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7746		break;
7747	case NFS4_SHARE_ACCESS_BOTH:
7748		break;
7749	default:
7750		WARN_ON_ONCE(1);
7751	}
7752}
7753
7754__be32
7755nfsd4_open_downgrade(struct svc_rqst *rqstp,
7756		     struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7757{
7758	struct nfsd4_open_downgrade *od = &u->open_downgrade;
7759	__be32 status;
7760	struct nfs4_ol_stateid *stp;
7761	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7762
7763	dprintk("NFSD: nfsd4_open_downgrade on file %pd\n", 
7764			cstate->current_fh.fh_dentry);
7765
7766	/* We don't yet support WANT bits: */
7767	if (od->od_deleg_want)
7768		dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7769			od->od_deleg_want);
7770
7771	status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7772					&od->od_stateid, &stp, nn);
7773	if (status)
7774		goto out; 
7775	status = nfserr_inval;
7776	if (!test_access(od->od_share_access, stp)) {
7777		dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7778			stp->st_access_bmap, od->od_share_access);
7779		goto put_stateid;
7780	}
7781	if (!test_deny(od->od_share_deny, stp)) {
7782		dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7783			stp->st_deny_bmap, od->od_share_deny);
7784		goto put_stateid;
7785	}
7786	nfs4_stateid_downgrade(stp, od->od_share_access);
7787	reset_union_bmap_deny(od->od_share_deny, stp);
7788	nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7789	status = nfs_ok;
7790put_stateid:
7791	mutex_unlock(&stp->st_mutex);
7792	nfs4_put_stid(&stp->st_stid);
7793out:
7794	nfsd4_bump_seqid(cstate, status);
7795	return status;
7796}
7797
7798static bool nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7799{
7800	struct nfs4_client *clp = s->st_stid.sc_client;
7801	bool unhashed;
7802	LIST_HEAD(reaplist);
7803	struct nfs4_ol_stateid *stp;
7804
7805	spin_lock(&clp->cl_lock);
7806	unhashed = unhash_open_stateid(s, &reaplist);
7807
7808	if (clp->cl_minorversion) {
7809		if (unhashed)
7810			put_ol_stateid_locked(s, &reaplist);
7811		spin_unlock(&clp->cl_lock);
7812		list_for_each_entry(stp, &reaplist, st_locks)
7813			nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7814		free_ol_stateid_reaplist(&reaplist);
7815		return false;
7816	} else {
7817		spin_unlock(&clp->cl_lock);
7818		free_ol_stateid_reaplist(&reaplist);
7819		return unhashed;
7820	}
7821}
7822
7823/*
7824 * nfs4_unlock_state() called after encode
7825 */
7826__be32
7827nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7828		union nfsd4_op_u *u)
7829{
7830	struct nfsd4_close *close = &u->close;
7831	__be32 status;
7832	struct nfs4_ol_stateid *stp;
7833	struct net *net = SVC_NET(rqstp);
7834	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7835	bool need_move_to_close_list;
7836
7837	dprintk("NFSD: nfsd4_close on file %pd\n",
7838			cstate->current_fh.fh_dentry);
7839
7840	status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7841					  &close->cl_stateid,
7842					  SC_TYPE_OPEN, SC_STATUS_CLOSED,
7843					  &stp, nn);
7844	nfsd4_bump_seqid(cstate, status);
7845	if (status)
7846		goto out;
7847
7848	spin_lock(&stp->st_stid.sc_client->cl_lock);
7849	stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7850	spin_unlock(&stp->st_stid.sc_client->cl_lock);
7851
7852	/*
7853	 * Technically we don't _really_ have to increment or copy it, since
7854	 * it should just be gone after this operation and we clobber the
7855	 * copied value below, but we continue to do so here just to ensure
7856	 * that racing ops see that there was a state change.
7857	 */
7858	nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7859
7860	need_move_to_close_list = nfsd4_close_open_stateid(stp);
7861	mutex_unlock(&stp->st_mutex);
7862	if (need_move_to_close_list)
7863		move_to_close_lru(stp, net);
7864
7865	/* v4.1+ suggests that we send a special stateid in here, since the
7866	 * clients should just ignore this anyway. Since this is not useful
7867	 * for v4.0 clients either, we set it to the special close_stateid
7868	 * universally.
7869	 *
7870	 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7871	 */
7872	memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7873
7874	/* put reference from nfs4_preprocess_seqid_op */
7875	nfs4_put_stid(&stp->st_stid);
7876out:
7877	return status;
7878}
7879
7880__be32
7881nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7882		  union nfsd4_op_u *u)
7883{
7884	struct nfsd4_delegreturn *dr = &u->delegreturn;
7885	struct nfs4_delegation *dp;
7886	stateid_t *stateid = &dr->dr_stateid;
7887	struct nfs4_stid *s;
7888	__be32 status;
7889	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7890
7891	status = fh_verify(rqstp, &cstate->current_fh, 0, 0);
7892	if (status)
7893		return status;
7894
7895	status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, SC_STATUS_REVOKED, &s, nn);
7896	if (status)
7897		goto out;
7898	dp = delegstateid(s);
7899	status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7900	if (status)
7901		goto put_stateid;
7902
7903	trace_nfsd_deleg_return(stateid);
7904	destroy_delegation(dp);
7905	smp_mb__after_atomic();
7906	wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7907put_stateid:
7908	nfs4_put_stid(&dp->dl_stid);
7909out:
7910	return status;
7911}
7912
7913/* last octet in a range */
7914static inline u64
7915last_byte_offset(u64 start, u64 len)
7916{
7917	u64 end;
7918
7919	WARN_ON_ONCE(!len);
7920	end = start + len;
7921	return end > start ? end - 1: NFS4_MAX_UINT64;
7922}
7923
7924/*
7925 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7926 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7927 * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
7928 * locking, this prevents us from being completely protocol-compliant.  The
7929 * real solution to this problem is to start using unsigned file offsets in
7930 * the VFS, but this is a very deep change!
7931 */
7932static inline void
7933nfs4_transform_lock_offset(struct file_lock *lock)
7934{
7935	if (lock->fl_start < 0)
7936		lock->fl_start = OFFSET_MAX;
7937	if (lock->fl_end < 0)
7938		lock->fl_end = OFFSET_MAX;
7939}
7940
7941static fl_owner_t
7942nfsd4_lm_get_owner(fl_owner_t owner)
7943{
7944	struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7945
7946	nfs4_get_stateowner(&lo->lo_owner);
7947	return owner;
7948}
7949
7950static void
7951nfsd4_lm_put_owner(fl_owner_t owner)
7952{
7953	struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7954
7955	if (lo)
7956		nfs4_put_stateowner(&lo->lo_owner);
7957}
7958
7959/* return pointer to struct nfs4_client if client is expirable */
7960static bool
7961nfsd4_lm_lock_expirable(struct file_lock *cfl)
7962{
7963	struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7964	struct nfs4_client *clp = lo->lo_owner.so_client;
7965	struct nfsd_net *nn;
7966
7967	if (try_to_expire_client(clp)) {
7968		nn = net_generic(clp->net, nfsd_net_id);
7969		mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7970		return true;
7971	}
7972	return false;
7973}
7974
7975/* schedule laundromat to run immediately and wait for it to complete */
7976static void
7977nfsd4_lm_expire_lock(void)
7978{
7979	flush_workqueue(laundry_wq);
7980}
7981
7982static void
7983nfsd4_lm_notify(struct file_lock *fl)
7984{
7985	struct nfs4_lockowner		*lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7986	struct net			*net = lo->lo_owner.so_client->net;
7987	struct nfsd_net			*nn = net_generic(net, nfsd_net_id);
7988	struct nfsd4_blocked_lock	*nbl = container_of(fl,
7989						struct nfsd4_blocked_lock, nbl_lock);
7990	bool queue = false;
7991
7992	/* An empty list means that something else is going to be using it */
7993	spin_lock(&nn->blocked_locks_lock);
7994	if (!list_empty(&nbl->nbl_list)) {
7995		list_del_init(&nbl->nbl_list);
7996		list_del_init(&nbl->nbl_lru);
7997		queue = true;
7998	}
7999	spin_unlock(&nn->blocked_locks_lock);
8000
8001	if (queue) {
8002		trace_nfsd_cb_notify_lock(lo, nbl);
8003		nfsd4_try_run_cb(&nbl->nbl_cb);
8004	}
8005}
8006
8007static const struct lock_manager_operations nfsd_posix_mng_ops  = {
8008	.lm_mod_owner = THIS_MODULE,
8009	.lm_notify = nfsd4_lm_notify,
8010	.lm_get_owner = nfsd4_lm_get_owner,
8011	.lm_put_owner = nfsd4_lm_put_owner,
8012	.lm_lock_expirable = nfsd4_lm_lock_expirable,
8013	.lm_expire_lock = nfsd4_lm_expire_lock,
8014};
8015
8016static inline void
8017nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
8018{
8019	struct nfs4_lockowner *lo;
8020
8021	if (fl->fl_lmops == &nfsd_posix_mng_ops) {
8022		lo = (struct nfs4_lockowner *) fl->c.flc_owner;
8023		xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
8024						GFP_KERNEL);
8025		if (!deny->ld_owner.data)
8026			/* We just don't care that much */
8027			goto nevermind;
8028		deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
8029	} else {
8030nevermind:
8031		deny->ld_owner.len = 0;
8032		deny->ld_owner.data = NULL;
8033		deny->ld_clientid.cl_boot = 0;
8034		deny->ld_clientid.cl_id = 0;
8035	}
8036	deny->ld_start = fl->fl_start;
8037	deny->ld_length = NFS4_MAX_UINT64;
8038	if (fl->fl_end != NFS4_MAX_UINT64)
8039		deny->ld_length = fl->fl_end - fl->fl_start + 1;        
8040	deny->ld_type = NFS4_READ_LT;
8041	if (fl->c.flc_type != F_RDLCK)
8042		deny->ld_type = NFS4_WRITE_LT;
8043}
8044
8045static struct nfs4_lockowner *
8046find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
8047{
8048	unsigned int strhashval = ownerstr_hashval(owner);
8049	struct nfs4_stateowner *so;
8050
8051	lockdep_assert_held(&clp->cl_lock);
8052
8053	list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
8054			    so_strhash) {
8055		if (so->so_is_open_owner)
8056			continue;
8057		if (same_owner_str(so, owner))
8058			return lockowner(nfs4_get_stateowner(so));
8059	}
8060	return NULL;
8061}
8062
8063static struct nfs4_lockowner *
8064find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
8065{
8066	struct nfs4_lockowner *lo;
8067
8068	spin_lock(&clp->cl_lock);
8069	lo = find_lockowner_str_locked(clp, owner);
8070	spin_unlock(&clp->cl_lock);
8071	return lo;
8072}
8073
8074static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
8075{
8076	unhash_lockowner_locked(lockowner(sop));
8077}
8078
8079static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
8080{
8081	struct nfs4_lockowner *lo = lockowner(sop);
8082
8083	kmem_cache_free(lockowner_slab, lo);
8084}
8085
8086static const struct nfs4_stateowner_operations lockowner_ops = {
8087	.so_unhash =	nfs4_unhash_lockowner,
8088	.so_free =	nfs4_free_lockowner,
8089};
8090
8091/*
8092 * Alloc a lock owner structure.
8093 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
8094 * occurred. 
8095 *
8096 * strhashval = ownerstr_hashval
8097 */
8098static struct nfs4_lockowner *
8099alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
8100			   struct nfs4_ol_stateid *open_stp,
8101			   struct nfsd4_lock *lock)
8102{
8103	struct nfs4_lockowner *lo, *ret;
8104
8105	lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
8106	if (!lo)
8107		return NULL;
8108	INIT_LIST_HEAD(&lo->lo_blocked);
8109	INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
8110	lo->lo_owner.so_is_open_owner = 0;
8111	lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
8112	lo->lo_owner.so_ops = &lockowner_ops;
8113	spin_lock(&clp->cl_lock);
8114	ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
8115	if (ret == NULL) {
8116		list_add(&lo->lo_owner.so_strhash,
8117			 &clp->cl_ownerstr_hashtbl[strhashval]);
8118		ret = lo;
8119	} else
8120		nfs4_free_stateowner(&lo->lo_owner);
8121
8122	spin_unlock(&clp->cl_lock);
8123	return ret;
8124}
8125
8126static struct nfs4_ol_stateid *
8127find_lock_stateid(const struct nfs4_lockowner *lo,
8128		  const struct nfs4_ol_stateid *ost)
8129{
8130	struct nfs4_ol_stateid *lst;
8131
8132	lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
8133
8134	/* If ost is not hashed, ost->st_locks will not be valid */
8135	if (!nfs4_ol_stateid_unhashed(ost))
8136		list_for_each_entry(lst, &ost->st_locks, st_locks) {
8137			if (lst->st_stateowner == &lo->lo_owner) {
8138				refcount_inc(&lst->st_stid.sc_count);
8139				return lst;
8140			}
8141		}
8142	return NULL;
8143}
8144
8145static struct nfs4_ol_stateid *
8146init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
8147		  struct nfs4_file *fp, struct inode *inode,
8148		  struct nfs4_ol_stateid *open_stp)
8149{
8150	struct nfs4_client *clp = lo->lo_owner.so_client;
8151	struct nfs4_ol_stateid *retstp;
8152
8153	mutex_init(&stp->st_mutex);
8154	mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
8155retry:
8156	spin_lock(&clp->cl_lock);
8157	if (nfs4_ol_stateid_unhashed(open_stp))
8158		goto out_close;
8159	retstp = find_lock_stateid(lo, open_stp);
8160	if (retstp)
8161		goto out_found;
8162	refcount_inc(&stp->st_stid.sc_count);
8163	stp->st_stid.sc_type = SC_TYPE_LOCK;
8164	stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
8165	get_nfs4_file(fp);
8166	stp->st_stid.sc_file = fp;
8167	stp->st_access_bmap = 0;
8168	stp->st_deny_bmap = open_stp->st_deny_bmap;
8169	stp->st_openstp = open_stp;
8170	spin_lock(&fp->fi_lock);
8171	list_add(&stp->st_locks, &open_stp->st_locks);
8172	list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
8173	list_add(&stp->st_perfile, &fp->fi_stateids);
8174	spin_unlock(&fp->fi_lock);
8175	spin_unlock(&clp->cl_lock);
8176	return stp;
8177out_found:
8178	spin_unlock(&clp->cl_lock);
8179	if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
8180		nfs4_put_stid(&retstp->st_stid);
8181		goto retry;
8182	}
8183	/* To keep mutex tracking happy */
8184	mutex_unlock(&stp->st_mutex);
8185	return retstp;
8186out_close:
8187	spin_unlock(&clp->cl_lock);
8188	mutex_unlock(&stp->st_mutex);
8189	return NULL;
8190}
8191
8192static struct nfs4_ol_stateid *
8193find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
8194			    struct inode *inode, struct nfs4_ol_stateid *ost,
8195			    bool *new)
8196{
8197	struct nfs4_stid *ns = NULL;
8198	struct nfs4_ol_stateid *lst;
8199	struct nfs4_openowner *oo = openowner(ost->st_stateowner);
8200	struct nfs4_client *clp = oo->oo_owner.so_client;
8201
8202	*new = false;
8203	spin_lock(&clp->cl_lock);
8204	lst = find_lock_stateid(lo, ost);
8205	spin_unlock(&clp->cl_lock);
8206	if (lst != NULL) {
8207		if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
8208			goto out;
8209		nfs4_put_stid(&lst->st_stid);
8210	}
8211	ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
8212	if (ns == NULL)
8213		return NULL;
8214
8215	lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
8216	if (lst == openlockstateid(ns))
8217		*new = true;
8218	else
8219		nfs4_put_stid(ns);
8220out:
8221	return lst;
8222}
8223
8224static int
8225check_lock_length(u64 offset, u64 length)
8226{
8227	return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
8228		(length > ~offset)));
8229}
8230
8231static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
8232{
8233	struct nfs4_file *fp = lock_stp->st_stid.sc_file;
8234
8235	lockdep_assert_held(&fp->fi_lock);
8236
8237	if (test_access(access, lock_stp))
8238		return;
8239	__nfs4_file_get_access(fp, access);
8240	set_access(access, lock_stp);
8241}
8242
8243static __be32
8244lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
8245			    struct nfs4_ol_stateid *ost,
8246			    struct nfsd4_lock *lock,
8247			    struct nfs4_ol_stateid **plst, bool *new)
8248{
8249	__be32 status;
8250	struct nfs4_file *fi = ost->st_stid.sc_file;
8251	struct nfs4_openowner *oo = openowner(ost->st_stateowner);
8252	struct nfs4_client *cl = oo->oo_owner.so_client;
8253	struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
8254	struct nfs4_lockowner *lo;
8255	struct nfs4_ol_stateid *lst;
8256	unsigned int strhashval;
8257
8258	lo = find_lockowner_str(cl, &lock->lk_new_owner);
8259	if (!lo) {
8260		strhashval = ownerstr_hashval(&lock->lk_new_owner);
8261		lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
8262		if (lo == NULL)
8263			return nfserr_jukebox;
8264	} else {
8265		/* with an existing lockowner, seqids must be the same */
8266		status = nfserr_bad_seqid;
8267		if (!cstate->minorversion &&
8268		    lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
8269			goto out;
8270	}
8271
8272	lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
8273	if (lst == NULL) {
8274		status = nfserr_jukebox;
8275		goto out;
8276	}
8277
8278	status = nfs_ok;
8279	*plst = lst;
8280out:
8281	nfs4_put_stateowner(&lo->lo_owner);
8282	return status;
8283}
8284
8285/*
8286 *  LOCK operation 
8287 */
8288__be32
8289nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8290	   union nfsd4_op_u *u)
8291{
8292	struct nfsd4_lock *lock = &u->lock;
8293	struct nfs4_openowner *open_sop = NULL;
8294	struct nfs4_lockowner *lock_sop = NULL;
8295	struct nfs4_ol_stateid *lock_stp = NULL;
8296	struct nfs4_ol_stateid *open_stp = NULL;
8297	struct nfs4_file *fp;
8298	struct nfsd_file *nf = NULL;
8299	struct nfsd4_blocked_lock *nbl = NULL;
8300	struct file_lock *file_lock = NULL;
8301	struct file_lock *conflock = NULL;
8302	__be32 status = 0;
8303	int lkflg;
8304	int err;
8305	bool new = false;
8306	unsigned char type;
8307	unsigned int flags = FL_POSIX;
8308	struct net *net = SVC_NET(rqstp);
8309	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8310
8311	dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
8312		(long long) lock->lk_offset,
8313		(long long) lock->lk_length);
8314
8315	if (check_lock_length(lock->lk_offset, lock->lk_length))
8316		 return nfserr_inval;
8317
8318	status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
8319	if (status != nfs_ok)
8320		return status;
8321	if (exportfs_cannot_lock(cstate->current_fh.fh_dentry->d_sb->s_export_op)) {
8322		status = nfserr_notsupp;
8323		goto out;
8324	}
8325
8326	if (lock->lk_is_new) {
8327		if (nfsd4_has_session(cstate))
8328			/* See rfc 5661 18.10.3: given clientid is ignored: */
8329			memcpy(&lock->lk_new_clientid,
8330				&cstate->clp->cl_clientid,
8331				sizeof(clientid_t));
8332
8333		/* validate and update open stateid and open seqid */
8334		status = nfs4_preprocess_confirmed_seqid_op(cstate,
8335				        lock->lk_new_open_seqid,
8336		                        &lock->lk_new_open_stateid,
8337					&open_stp, nn);
8338		if (status)
8339			goto out;
8340		mutex_unlock(&open_stp->st_mutex);
8341		open_sop = openowner(open_stp->st_stateowner);
8342		status = nfserr_bad_stateid;
8343		if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
8344						&lock->lk_new_clientid))
8345			goto out;
8346		status = lookup_or_create_lock_state(cstate, open_stp, lock,
8347							&lock_stp, &new);
8348	} else {
8349		status = nfs4_preprocess_seqid_op(cstate,
8350						  lock->lk_old_lock_seqid,
8351						  &lock->lk_old_lock_stateid,
8352						  SC_TYPE_LOCK, 0, &lock_stp,
8353						  nn);
8354	}
8355	if (status)
8356		goto out;
8357	lock_sop = lockowner(lock_stp->st_stateowner);
8358
8359	lkflg = setlkflg(lock->lk_type);
8360	status = nfs4_check_openmode(lock_stp, lkflg);
8361	if (status)
8362		goto out;
8363
8364	status = nfserr_grace;
8365	if (locks_in_grace(net) && !lock->lk_reclaim)
8366		goto out;
8367	status = nfserr_no_grace;
8368	if (!locks_in_grace(net) && lock->lk_reclaim)
8369		goto out;
8370
8371	if (lock->lk_reclaim)
8372		flags |= FL_RECLAIM;
8373
8374	fp = lock_stp->st_stid.sc_file;
8375	switch (lock->lk_type) {
8376		case NFS4_READW_LT:
8377			fallthrough;
8378		case NFS4_READ_LT:
8379			spin_lock(&fp->fi_lock);
8380			nf = find_readable_file_locked(fp);
8381			if (nf)
8382				get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
8383			spin_unlock(&fp->fi_lock);
8384			type = F_RDLCK;
8385			break;
8386		case NFS4_WRITEW_LT:
8387			fallthrough;
8388		case NFS4_WRITE_LT:
8389			spin_lock(&fp->fi_lock);
8390			nf = find_writeable_file_locked(fp);
8391			if (nf)
8392				get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
8393			spin_unlock(&fp->fi_lock);
8394			type = F_WRLCK;
8395			break;
8396		default:
8397			status = nfserr_inval;
8398		goto out;
8399	}
8400
8401	if (!nf) {
8402		status = nfserr_openmode;
8403		goto out;
8404	}
8405
8406	if (lock->lk_type & (NFS4_READW_LT | NFS4_WRITEW_LT) &&
8407		nfsd4_has_session(cstate) &&
8408		locks_can_async_lock(nf->nf_file->f_op))
8409			flags |= FL_SLEEP;
8410
8411	nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
8412	if (!nbl) {
8413		dprintk("NFSD: %s: unable to allocate block!\n", __func__);
8414		status = nfserr_jukebox;
8415		goto out;
8416	}
8417
8418	file_lock = &nbl->nbl_lock;
8419	file_lock->c.flc_type = type;
8420	file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
8421	file_lock->c.flc_pid = current->tgid;
8422	file_lock->c.flc_file = nf->nf_file;
8423	file_lock->c.flc_flags = flags;
8424	file_lock->fl_lmops = &nfsd_posix_mng_ops;
8425	file_lock->fl_start = lock->lk_offset;
8426	file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
8427	nfs4_transform_lock_offset(file_lock);
8428
8429	conflock = locks_alloc_lock();
8430	if (!conflock) {
8431		dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8432		status = nfserr_jukebox;
8433		goto out;
8434	}
8435
8436	if (flags & FL_SLEEP) {
8437		nbl->nbl_time = ktime_get_boottime_seconds();
8438		spin_lock(&nn->blocked_locks_lock);
8439		list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
8440		list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
8441		kref_get(&nbl->nbl_kref);
8442		spin_unlock(&nn->blocked_locks_lock);
8443	}
8444
8445	err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
8446	switch (err) {
8447	case 0: /* success! */
8448		nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
8449		status = 0;
8450		if (lock->lk_reclaim)
8451			nn->somebody_reclaimed = true;
8452		break;
8453	case FILE_LOCK_DEFERRED:
8454		kref_put(&nbl->nbl_kref, free_nbl);
8455		nbl = NULL;
8456		fallthrough;
8457	case -EAGAIN:		/* conflock holds conflicting lock */
8458		status = nfserr_denied;
8459		dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8460		nfs4_set_lock_denied(conflock, &lock->lk_denied);
8461		break;
8462	case -EDEADLK:
8463		status = nfserr_deadlock;
8464		break;
8465	default:
8466		dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8467		status = nfserrno(err);
8468		break;
8469	}
8470out:
8471	if (nbl) {
8472		/* dequeue it if we queued it before */
8473		if (flags & FL_SLEEP) {
8474			spin_lock(&nn->blocked_locks_lock);
8475			if (!list_empty(&nbl->nbl_list) &&
8476			    !list_empty(&nbl->nbl_lru)) {
8477				list_del_init(&nbl->nbl_list);
8478				list_del_init(&nbl->nbl_lru);
8479				kref_put(&nbl->nbl_kref, free_nbl);
8480			}
8481			/* nbl can use one of lists to be linked to reaplist */
8482			spin_unlock(&nn->blocked_locks_lock);
8483		}
8484		free_blocked_lock(nbl);
8485	}
8486	if (nf)
8487		nfsd_file_put(nf);
8488	if (lock_stp) {
8489		/* Bump seqid manually if the 4.0 replay owner is openowner */
8490		if (cstate->replay_owner &&
8491		    cstate->replay_owner != &lock_sop->lo_owner &&
8492		    seqid_mutating_err(ntohl(status)))
8493			lock_sop->lo_owner.so_seqid++;
8494
8495		/*
8496		 * If this is a new, never-before-used stateid, and we are
8497		 * returning an error, then just go ahead and release it.
8498		 */
8499		if (status && new)
8500			release_lock_stateid(lock_stp);
8501
8502		mutex_unlock(&lock_stp->st_mutex);
8503
8504		nfs4_put_stid(&lock_stp->st_stid);
8505	}
8506	if (open_stp)
8507		nfs4_put_stid(&open_stp->st_stid);
8508	nfsd4_bump_seqid(cstate, status);
8509	if (conflock)
8510		locks_free_lock(conflock);
8511	return status;
8512}
8513
8514void nfsd4_lock_release(union nfsd4_op_u *u)
8515{
8516	struct nfsd4_lock *lock = &u->lock;
8517	struct nfsd4_lock_denied *deny = &lock->lk_denied;
8518
8519	kfree(deny->ld_owner.data);
8520}
8521
8522/*
8523 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8524 * so we do a temporary open here just to get an open file to pass to
8525 * vfs_test_lock.
8526 */
8527static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8528{
8529	struct nfsd_file *nf;
8530	struct inode *inode;
8531	__be32 err;
8532
8533	err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8534	if (err)
8535		return err;
8536	inode = fhp->fh_dentry->d_inode;
8537	inode_lock(inode); /* to block new leases till after test_lock: */
8538	err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8539	if (err)
8540		goto out;
8541	lock->c.flc_file = nf->nf_file;
8542	err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8543	lock->c.flc_file = NULL;
8544out:
8545	inode_unlock(inode);
8546	nfsd_file_put(nf);
8547	return err;
8548}
8549
8550/*
8551 * LOCKT operation
8552 */
8553__be32
8554nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8555	    union nfsd4_op_u *u)
8556{
8557	struct nfsd4_lockt *lockt = &u->lockt;
8558	struct file_lock *file_lock = NULL;
8559	struct nfs4_lockowner *lo = NULL;
8560	__be32 status;
8561	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8562
8563	if (locks_in_grace(SVC_NET(rqstp)))
8564		return nfserr_grace;
8565
8566	if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8567		 return nfserr_inval;
8568
8569	if (!nfsd4_has_session(cstate)) {
8570		status = set_client(&lockt->lt_clientid, cstate, nn);
8571		if (status)
8572			goto out;
8573	}
8574
8575	if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8576		goto out;
8577
8578	file_lock = locks_alloc_lock();
8579	if (!file_lock) {
8580		dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8581		status = nfserr_jukebox;
8582		goto out;
8583	}
8584
8585	switch (lockt->lt_type) {
8586		case NFS4_READ_LT:
8587		case NFS4_READW_LT:
8588			file_lock->c.flc_type = F_RDLCK;
8589			break;
8590		case NFS4_WRITE_LT:
8591		case NFS4_WRITEW_LT:
8592			file_lock->c.flc_type = F_WRLCK;
8593			break;
8594		default:
8595			dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8596			status = nfserr_inval;
8597			goto out;
8598	}
8599
8600	lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8601	if (lo)
8602		file_lock->c.flc_owner = (fl_owner_t)lo;
8603	file_lock->c.flc_pid = current->tgid;
8604	file_lock->c.flc_flags = FL_POSIX;
8605
8606	file_lock->fl_start = lockt->lt_offset;
8607	file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8608
8609	nfs4_transform_lock_offset(file_lock);
8610
8611	status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8612	if (status)
8613		goto out;
8614
8615	if (file_lock->c.flc_type != F_UNLCK) {
8616		status = nfserr_denied;
8617		nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8618	}
8619out:
8620	if (lo)
8621		nfs4_put_stateowner(&lo->lo_owner);
8622	if (file_lock)
8623		locks_free_lock(file_lock);
8624	return status;
8625}
8626
8627void nfsd4_lockt_release(union nfsd4_op_u *u)
8628{
8629	struct nfsd4_lockt *lockt = &u->lockt;
8630	struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8631
8632	kfree(deny->ld_owner.data);
8633}
8634
8635__be32
8636nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8637	    union nfsd4_op_u *u)
8638{
8639	struct nfsd4_locku *locku = &u->locku;
8640	struct nfs4_ol_stateid *stp;
8641	struct nfsd_file *nf = NULL;
8642	struct file_lock *file_lock = NULL;
8643	__be32 status;
8644	int err;
8645	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8646
8647	dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8648		(long long) locku->lu_offset,
8649		(long long) locku->lu_length);
8650
8651	if (check_lock_length(locku->lu_offset, locku->lu_length))
8652		 return nfserr_inval;
8653
8654	status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8655					  &locku->lu_stateid, SC_TYPE_LOCK, 0,
8656					  &stp, nn);
8657	if (status)
8658		goto out;
8659	nf = find_any_file(stp->st_stid.sc_file);
8660	if (!nf) {
8661		status = nfserr_lock_range;
8662		goto put_stateid;
8663	}
8664	if (exportfs_cannot_lock(nf->nf_file->f_path.mnt->mnt_sb->s_export_op)) {
8665		status = nfserr_notsupp;
8666		goto put_file;
8667	}
8668
8669	file_lock = locks_alloc_lock();
8670	if (!file_lock) {
8671		dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8672		status = nfserr_jukebox;
8673		goto put_file;
8674	}
8675
8676	file_lock->c.flc_type = F_UNLCK;
8677	file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8678	file_lock->c.flc_pid = current->tgid;
8679	file_lock->c.flc_file = nf->nf_file;
8680	file_lock->c.flc_flags = FL_POSIX;
8681	file_lock->fl_lmops = &nfsd_posix_mng_ops;
8682	file_lock->fl_start = locku->lu_offset;
8683
8684	file_lock->fl_end = last_byte_offset(locku->lu_offset,
8685						locku->lu_length);
8686	nfs4_transform_lock_offset(file_lock);
8687
8688	err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8689	if (err) {
8690		dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8691		goto out_nfserr;
8692	}
8693	nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8694put_file:
8695	nfsd_file_put(nf);
8696put_stateid:
8697	mutex_unlock(&stp->st_mutex);
8698	nfs4_put_stid(&stp->st_stid);
8699out:
8700	nfsd4_bump_seqid(cstate, status);
8701	if (file_lock)
8702		locks_free_lock(file_lock);
8703	return status;
8704
8705out_nfserr:
8706	status = nfserrno(err);
8707	goto put_file;
8708}
8709
8710/*
8711 * returns
8712 * 	true:  locks held by lockowner
8713 * 	false: no locks held by lockowner
8714 */
8715static bool
8716check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8717{
8718	struct file_lock *fl;
8719	int status = false;
8720	struct nfsd_file *nf;
8721	struct inode *inode;
8722	struct file_lock_context *flctx;
8723
8724	spin_lock(&fp->fi_lock);
8725	nf = find_any_file_locked(fp);
8726	if (!nf) {
8727		/* Any valid lock stateid should have some sort of access */
8728		WARN_ON_ONCE(1);
8729		goto out;
8730	}
8731
8732	inode = file_inode(nf->nf_file);
8733	flctx = locks_inode_context(inode);
8734
8735	if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8736		spin_lock(&flctx->flc_lock);
8737		for_each_file_lock(fl, &flctx->flc_posix) {
8738			if (fl->c.flc_owner == (fl_owner_t)lowner) {
8739				status = true;
8740				break;
8741			}
8742		}
8743		spin_unlock(&flctx->flc_lock);
8744	}
8745out:
8746	spin_unlock(&fp->fi_lock);
8747	return status;
8748}
8749
8750/**
8751 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8752 * @rqstp: RPC transaction
8753 * @cstate: NFSv4 COMPOUND state
8754 * @u: RELEASE_LOCKOWNER arguments
8755 *
8756 * Check if there are any locks still held and if not, free the lockowner
8757 * and any lock state that is owned.
8758 *
8759 * Return values:
8760 *   %nfs_ok: lockowner released or not found
8761 *   %nfserr_locks_held: lockowner still in use
8762 *   %nfserr_stale_clientid: clientid no longer active
8763 *   %nfserr_expired: clientid not recognized
8764 */
8765__be32
8766nfsd4_release_lockowner(struct svc_rqst *rqstp,
8767			struct nfsd4_compound_state *cstate,
8768			union nfsd4_op_u *u)
8769{
8770	struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8771	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8772	clientid_t *clid = &rlockowner->rl_clientid;
8773	struct nfs4_ol_stateid *stp;
8774	struct nfs4_lockowner *lo;
8775	struct nfs4_client *clp;
8776	LIST_HEAD(reaplist);
8777	__be32 status;
8778
8779	dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8780		clid->cl_boot, clid->cl_id);
8781
8782	status = set_client(clid, cstate, nn);
8783	if (status)
8784		return status;
8785	clp = cstate->clp;
8786
8787	spin_lock(&clp->cl_lock);
8788	lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8789	if (!lo) {
8790		spin_unlock(&clp->cl_lock);
8791		return nfs_ok;
8792	}
8793
8794	list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8795		if (check_for_locks(stp->st_stid.sc_file, lo)) {
8796			spin_unlock(&clp->cl_lock);
8797			nfs4_put_stateowner(&lo->lo_owner);
8798			return nfserr_locks_held;
8799		}
8800	}
8801	unhash_lockowner_locked(lo);
8802	while (!list_empty(&lo->lo_owner.so_stateids)) {
8803		stp = list_first_entry(&lo->lo_owner.so_stateids,
8804				       struct nfs4_ol_stateid,
8805				       st_perstateowner);
8806		unhash_lock_stateid(stp);
8807		put_ol_stateid_locked(stp, &reaplist);
8808	}
8809	spin_unlock(&clp->cl_lock);
8810
8811	free_ol_stateid_reaplist(&reaplist);
8812	remove_blocked_locks(lo);
8813	nfs4_put_stateowner(&lo->lo_owner);
8814	return nfs_ok;
8815}
8816
8817static inline struct nfs4_client_reclaim *
8818alloc_reclaim(void)
8819{
8820	return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8821}
8822
8823bool
8824nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8825{
8826	struct nfs4_client_reclaim *crp;
8827
8828	crp = nfsd4_find_reclaim_client(name, nn);
8829	return (crp && crp->cr_clp);
8830}
8831
8832/*
8833 * failure => all reset bets are off, nfserr_no_grace...
8834 */
8835struct nfs4_client_reclaim *
8836nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8837		struct nfsd_net *nn)
8838{
8839	unsigned int strhashval;
8840	struct nfs4_client_reclaim *crp;
8841
8842	name.data = kmemdup(name.data, name.len, GFP_KERNEL);
8843	if (!name.data) {
8844		dprintk("%s: failed to allocate memory for name.data!\n",
8845			__func__);
8846		return NULL;
8847	}
8848	if (princhash.len) {
8849		princhash.data = kmemdup(princhash.data, princhash.len, GFP_KERNEL);
8850		if (!princhash.data) {
8851			dprintk("%s: failed to allocate memory for princhash.data!\n",
8852				__func__);
8853			kfree(name.data);
8854			return NULL;
8855		}
8856	} else
8857		princhash.data = NULL;
8858	crp = alloc_reclaim();
8859	if (crp) {
8860		strhashval = clientstr_hashval(name);
8861		INIT_LIST_HEAD(&crp->cr_strhash);
8862		list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8863		crp->cr_name.data = name.data;
8864		crp->cr_name.len = name.len;
8865		crp->cr_princhash.data = princhash.data;
8866		crp->cr_princhash.len = princhash.len;
8867		crp->cr_clp = NULL;
8868		nn->reclaim_str_hashtbl_size++;
8869	} else {
8870		kfree(name.data);
8871		kfree(princhash.data);
8872	}
8873	return crp;
8874}
8875
8876void
8877nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8878{
8879	list_del(&crp->cr_strhash);
8880	kfree(crp->cr_name.data);
8881	kfree(crp->cr_princhash.data);
8882	kfree(crp);
8883	nn->reclaim_str_hashtbl_size--;
8884}
8885
8886void
8887nfs4_release_reclaim(struct nfsd_net *nn)
8888{
8889	struct nfs4_client_reclaim *crp = NULL;
8890	int i;
8891
8892	for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8893		while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8894			crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8895			                struct nfs4_client_reclaim, cr_strhash);
8896			nfs4_remove_reclaim_record(crp, nn);
8897		}
8898	}
8899	WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8900}
8901
8902/*
8903 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8904struct nfs4_client_reclaim *
8905nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8906{
8907	unsigned int strhashval;
8908	struct nfs4_client_reclaim *crp = NULL;
8909
8910	strhashval = clientstr_hashval(name);
8911	list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8912		if (compare_blob(&crp->cr_name, &name) == 0) {
8913			return crp;
8914		}
8915	}
8916	return NULL;
8917}
8918
8919__be32
8920nfs4_check_open_reclaim(struct nfs4_client *clp)
8921{
8922	if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8923		return nfserr_no_grace;
8924
8925	if (nfsd4_client_record_check(clp))
8926		return nfserr_reclaim_bad;
8927
8928	return nfs_ok;
8929}
8930
8931/*
8932 * Since the lifetime of a delegation isn't limited to that of an open, a
8933 * client may quite reasonably hang on to a delegation as long as it has
8934 * the inode cached.  This becomes an obvious problem the first time a
8935 * client's inode cache approaches the size of the server's total memory.
8936 *
8937 * For now we avoid this problem by imposing a hard limit on the number
8938 * of delegations, which varies according to the server's memory size.
8939 */
8940static void
8941set_max_delegations(void)
8942{
8943	/*
8944	 * Allow at most 4 delegations per megabyte of RAM.  Quick
8945	 * estimates suggest that in the worst case (where every delegation
8946	 * is for a different inode), a delegation could take about 1.5K,
8947	 * giving a worst case usage of about 6% of memory.
8948	 */
8949	max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8950}
8951
8952static int nfs4_state_create_net(struct net *net)
8953{
8954	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8955	int i;
8956
8957	nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8958					    sizeof(struct list_head),
8959					    GFP_KERNEL);
8960	if (!nn->conf_id_hashtbl)
8961		goto err;
8962	nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8963					      sizeof(struct list_head),
8964					      GFP_KERNEL);
8965	if (!nn->unconf_id_hashtbl)
8966		goto err_unconf_id;
8967	nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8968					      sizeof(struct list_head),
8969					      GFP_KERNEL);
8970	if (!nn->sessionid_hashtbl)
8971		goto err_sessionid;
8972
8973	for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8974		INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8975		INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8976	}
8977	for (i = 0; i < SESSION_HASH_SIZE; i++)
8978		INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8979	nn->conf_name_tree = RB_ROOT;
8980	nn->unconf_name_tree = RB_ROOT;
8981	nn->boot_time = ktime_get_real_seconds();
8982	nn->grace_ended = false;
8983	nn->grace_end_forced = false;
8984	nn->client_tracking_active = false;
8985	nn->nfsd4_manager.block_opens = true;
8986	INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8987	INIT_LIST_HEAD(&nn->client_lru);
8988	INIT_LIST_HEAD(&nn->close_lru);
8989	INIT_LIST_HEAD(&nn->del_recall_lru);
8990	spin_lock_init(&nn->client_lock);
8991	spin_lock_init(&nn->s2s_cp_lock);
8992	idr_init(&nn->s2s_cp_stateids);
8993	atomic_set(&nn->pending_async_copies, 0);
8994
8995	spin_lock_init(&nn->blocked_locks_lock);
8996	INIT_LIST_HEAD(&nn->blocked_locks_lru);
8997
8998	INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8999	INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
9000	get_net(net);
9001
9002	nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
9003	if (!nn->nfsd_client_shrinker)
9004		goto err_shrinker;
9005
9006	nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
9007	nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
9008	nn->nfsd_client_shrinker->private_data = nn;
9009
9010	shrinker_register(nn->nfsd_client_shrinker);
9011
9012	return 0;
9013
9014err_shrinker:
9015	put_net(net);
9016	kfree(nn->sessionid_hashtbl);
9017err_sessionid:
9018	kfree(nn->unconf_id_hashtbl);
9019err_unconf_id:
9020	kfree(nn->conf_id_hashtbl);
9021err:
9022	return -ENOMEM;
9023}
9024
9025static void
9026nfs4_state_destroy_net(struct net *net)
9027{
9028	int i;
9029	struct nfs4_client *clp = NULL;
9030	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
9031
9032	for (i = 0; i < CLIENT_HASH_SIZE; i++) {
9033		while (!list_empty(&nn->conf_id_hashtbl[i])) {
9034			clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
9035			destroy_client(clp);
9036		}
9037	}
9038
9039	WARN_ON(!list_empty(&nn->blocked_locks_lru));
9040
9041	for (i = 0; i < CLIENT_HASH_SIZE; i++) {
9042		while (!list_empty(&nn->unconf_id_hashtbl[i])) {
9043			clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
9044			destroy_client(clp);
9045		}
9046	}
9047
9048	kfree(nn->sessionid_hashtbl);
9049	kfree(nn->unconf_id_hashtbl);
9050	kfree(nn->conf_id_hashtbl);
9051	put_net(net);
9052}
9053
9054int
9055nfs4_state_start_net(struct net *net)
9056{
9057	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
9058	int ret;
9059
9060	ret = nfs4_state_create_net(net);
9061	if (ret)
9062		return ret;
9063	locks_start_grace(net, &nn->nfsd4_manager);
9064	nfsd4_client_tracking_init(net);
9065	/* safe for laundromat to run now */
9066	spin_lock(&nn->client_lock);
9067	nn->client_tracking_active = true;
9068	spin_unlock(&nn->client_lock);
9069	if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
9070		goto skip_grace;
9071	printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
9072	       nn->nfsd4_grace, net->ns.inum);
9073	trace_nfsd_grace_start(nn);
9074	queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
9075	return 0;
9076
9077skip_grace:
9078	printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
9079			net->ns.inum);
9080	queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
9081	nfsd4_end_grace(nn);
9082	return 0;
9083}
9084
9085/* initialization to perform when the nfsd service is started: */
9086int
9087nfs4_state_start(void)
9088{
9089	int ret;
9090
9091	ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
9092	if (ret)
9093		return ret;
9094
9095	nfsd_slot_shrinker = shrinker_alloc(0, "nfsd-DRC-slot");
9096	if (!nfsd_slot_shrinker) {
9097		rhltable_destroy(&nfs4_file_rhltable);
9098		return -ENOMEM;
9099	}
9100	nfsd_slot_shrinker->count_objects = nfsd_slot_count;
9101	nfsd_slot_shrinker->scan_objects = nfsd_slot_scan;
9102	shrinker_register(nfsd_slot_shrinker);
9103
9104	set_max_delegations();
9105	return 0;
9106}
9107
9108void
9109nfs4_state_shutdown_net(struct net *net)
9110{
9111	struct nfs4_delegation *dp = NULL;
9112	struct list_head *pos, *next, reaplist;
9113	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
9114
9115	shrinker_free(nn->nfsd_client_shrinker);
9116	cancel_work_sync(&nn->nfsd_shrinker_work);
9117	spin_lock(&nn->client_lock);
9118	nn->client_tracking_active = false;
9119	spin_unlock(&nn->client_lock);
9120	cancel_delayed_work_sync(&nn->laundromat_work);
9121	locks_end_grace(&nn->nfsd4_manager);
9122
9123	INIT_LIST_HEAD(&reaplist);
9124	spin_lock(&state_lock);
9125	list_for_each_safe(pos, next, &nn->del_recall_lru) {
9126		dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
9127		unhash_delegation_locked(dp, SC_STATUS_CLOSED);
9128		list_add(&dp->dl_recall_lru, &reaplist);
9129	}
9130	spin_unlock(&state_lock);
9131	list_for_each_safe(pos, next, &reaplist) {
9132		dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
9133		list_del_init(&dp->dl_recall_lru);
9134		destroy_unhashed_deleg(dp);
9135	}
9136
9137	nfsd4_client_tracking_exit(net);
9138	nfs4_state_destroy_net(net);
9139#ifdef CONFIG_NFSD_V4_2_INTER_SSC
9140	nfsd4_ssc_shutdown_umount(nn);
9141#endif
9142}
9143
9144void
9145nfs4_state_shutdown(void)
9146{
9147	rhltable_destroy(&nfs4_file_rhltable);
9148	shrinker_free(nfsd_slot_shrinker);
9149}
9150
9151static void
9152get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
9153{
9154	if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
9155	    CURRENT_STATEID(stateid))
9156		memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
9157}
9158
9159static void
9160put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
9161{
9162	if (cstate->minorversion) {
9163		memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
9164		SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
9165	}
9166}
9167
9168void
9169clear_current_stateid(struct nfsd4_compound_state *cstate)
9170{
9171	CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
9172}
9173
9174/*
9175 * functions to set current state id
9176 */
9177void
9178nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
9179		union nfsd4_op_u *u)
9180{
9181	put_stateid(cstate, &u->open_downgrade.od_stateid);
9182}
9183
9184void
9185nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
9186		union nfsd4_op_u *u)
9187{
9188	put_stateid(cstate, &u->open.op_stateid);
9189}
9190
9191void
9192nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
9193		union nfsd4_op_u *u)
9194{
9195	put_stateid(cstate, &u->close.cl_stateid);
9196}
9197
9198void
9199nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
9200		union nfsd4_op_u *u)
9201{
9202	put_stateid(cstate, &u->lock.lk_resp_stateid);
9203}
9204
9205/*
9206 * functions to consume current state id
9207 */
9208
9209void
9210nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
9211		union nfsd4_op_u *u)
9212{
9213	get_stateid(cstate, &u->open_downgrade.od_stateid);
9214}
9215
9216void
9217nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
9218		union nfsd4_op_u *u)
9219{
9220	get_stateid(cstate, &u->delegreturn.dr_stateid);
9221}
9222
9223void
9224nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
9225		union nfsd4_op_u *u)
9226{
9227	get_stateid(cstate, &u->free_stateid.fr_stateid);
9228}
9229
9230void
9231nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
9232		union nfsd4_op_u *u)
9233{
9234	get_stateid(cstate, &u->setattr.sa_stateid);
9235}
9236
9237void
9238nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
9239		union nfsd4_op_u *u)
9240{
9241	get_stateid(cstate, &u->close.cl_stateid);
9242}
9243
9244void
9245nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
9246		union nfsd4_op_u *u)
9247{
9248	get_stateid(cstate, &u->locku.lu_stateid);
9249}
9250
9251void
9252nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
9253		union nfsd4_op_u *u)
9254{
9255	get_stateid(cstate, &u->read.rd_stateid);
9256}
9257
9258void
9259nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
9260		union nfsd4_op_u *u)
9261{
9262	get_stateid(cstate, &u->write.wr_stateid);
9263}
9264
9265/**
9266 * nfsd4_vet_deleg_time - vet and set the timespec for a delegated timestamp update
9267 * @req: timestamp from the client
9268 * @orig: original timestamp in the inode
9269 * @now: current time
9270 *
9271 * Given a timestamp from the client response, check it against the
9272 * current timestamp in the inode and the current time. Returns true
9273 * if the inode's timestamp needs to be updated, and false otherwise.
9274 * @req may also be changed if the timestamp needs to be clamped.
9275 */
9276bool nfsd4_vet_deleg_time(struct timespec64 *req, const struct timespec64 *orig,
9277			  const struct timespec64 *now)
9278{
9279
9280	/*
9281	 * "When the time presented is before the original time, then the
9282	 *  update is ignored." Also no need to update if there is no change.
9283	 */
9284	if (timespec64_compare(req, orig) <= 0)
9285		return false;
9286
9287	/*
9288	 * "When the time presented is in the future, the server can either
9289	 *  clamp the new time to the current time, or it may
9290	 *  return NFS4ERR_DELAY to the client, allowing it to retry."
9291	 */
9292	if (timespec64_compare(req, now) > 0)
9293		*req = *now;
9294
9295	return true;
9296}
9297
9298static int cb_getattr_update_times(struct dentry *dentry, struct nfs4_delegation *dp)
9299{
9300	struct inode *inode = d_inode(dentry);
9301	struct nfs4_cb_fattr *ncf = &dp->dl_cb_fattr;
9302	struct iattr attrs = { };
9303	int ret;
9304
9305	if (deleg_attrs_deleg(dp->dl_type)) {
9306		struct timespec64 now = current_time(inode);
9307
9308		attrs.ia_atime = ncf->ncf_cb_atime;
9309		attrs.ia_mtime = ncf->ncf_cb_mtime;
9310
9311		if (nfsd4_vet_deleg_time(&attrs.ia_atime, &dp->dl_atime, &now))
9312			attrs.ia_valid |= ATTR_ATIME | ATTR_ATIME_SET;
9313
9314		if (nfsd4_vet_deleg_time(&attrs.ia_mtime, &dp->dl_mtime, &now)) {
9315			attrs.ia_valid |= ATTR_MTIME | ATTR_MTIME_SET;
9316			attrs.ia_ctime = attrs.ia_mtime;
9317			if (nfsd4_vet_deleg_time(&attrs.ia_ctime, &dp->dl_ctime, &now))
9318				attrs.ia_valid |= ATTR_CTIME | ATTR_CTIME_SET;
9319		}
9320	} else {
9321		attrs.ia_valid |= ATTR_MTIME | ATTR_CTIME;
9322	}
9323
9324	if (!attrs.ia_valid)
9325		return 0;
9326
9327	attrs.ia_valid |= ATTR_DELEG;
9328	inode_lock(inode);
9329	ret = notify_change(&nop_mnt_idmap, dentry, &attrs, NULL);
9330	inode_unlock(inode);
9331	return ret;
9332}
9333
9334/**
9335 * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
9336 * @rqstp: RPC transaction context
9337 * @dentry: dentry of inode to be checked for a conflict
9338 * @pdp: returned WRITE delegation, if one was found
9339 *
9340 * This function is called when there is a conflict between a write
9341 * delegation and a change/size GETATTR from another client. The server
9342 * must either use the CB_GETATTR to get the current values of the
9343 * attributes from the client that holds the delegation or recall the
9344 * delegation before replying to the GETATTR. See RFC 8881 section
9345 * 18.7.4.
9346 *
9347 * Returns 0 if there is no conflict; otherwise an nfs_stat
9348 * code is returned. If @pdp is set to a non-NULL value, then the
9349 * caller must put the reference.
9350 */
9351__be32
9352nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct dentry *dentry,
9353			     struct nfs4_delegation **pdp)
9354{
9355	__be32 status;
9356	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
9357	struct file_lock_context *ctx;
9358	struct nfs4_delegation *dp = NULL;
9359	struct file_lease *fl;
9360	struct nfs4_cb_fattr *ncf;
9361	struct inode *inode = d_inode(dentry);
9362
9363	ctx = locks_inode_context(inode);
9364	if (!ctx)
9365		return nfs_ok;
9366
9367#define NON_NFSD_LEASE ((void *)1)
9368
9369	spin_lock(&ctx->flc_lock);
9370	for_each_file_lock(fl, &ctx->flc_lease) {
9371		if (fl->c.flc_flags == FL_LAYOUT)
9372			continue;
9373		if (fl->c.flc_type == F_WRLCK) {
9374			if (fl->fl_lmops == &nfsd_lease_mng_ops)
9375				dp = fl->c.flc_owner;
9376			else
9377				dp = NON_NFSD_LEASE;
9378		}
9379		break;
9380	}
9381	if (dp == NULL || dp == NON_NFSD_LEASE ||
9382	    dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
9383		spin_unlock(&ctx->flc_lock);
9384		if (dp == NON_NFSD_LEASE) {
9385			status = nfserrno(nfsd_open_break_lease(inode,
9386								NFSD_MAY_READ));
9387			if (status != nfserr_jukebox ||
9388			    !nfsd_wait_for_delegreturn(rqstp, inode))
9389				return status;
9390		}
9391		return 0;
9392	}
9393
9394	nfsd_stats_wdeleg_getattr_inc(nn);
9395	refcount_inc(&dp->dl_stid.sc_count);
9396	ncf = &dp->dl_cb_fattr;
9397	nfs4_cb_getattr(&dp->dl_cb_fattr);
9398	spin_unlock(&ctx->flc_lock);
9399
9400	wait_on_bit_timeout(&ncf->ncf_getattr.cb_flags, NFSD4_CALLBACK_RUNNING,
9401			    TASK_UNINTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
9402	if (ncf->ncf_cb_status) {
9403		/* Recall delegation only if client didn't respond */
9404		status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
9405		if (status != nfserr_jukebox ||
9406		    !nfsd_wait_for_delegreturn(rqstp, inode))
9407			goto out_status;
9408	}
9409	if (!ncf->ncf_file_modified &&
9410	    (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
9411	     ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
9412		ncf->ncf_file_modified = true;
9413	if (ncf->ncf_file_modified) {
9414		int err;
9415
9416		/*
9417		 * Per section 10.4.3 of RFC 8881, the server would
9418		 * not update the file's metadata with the client's
9419		 * modified size
9420		 */
9421		err = cb_getattr_update_times(dentry, dp);
9422		if (err) {
9423			status = nfserrno(err);
9424			goto out_status;
9425		}
9426		ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
9427		*pdp = dp;
9428		return nfs_ok;
9429	}
9430	status = nfs_ok;
9431out_status:
9432	nfs4_put_stid(&dp->dl_stid);
9433	return status;
9434}
9435
9436/**
9437 * nfsd_get_dir_deleg - attempt to get a directory delegation
9438 * @cstate: compound state
9439 * @gdd: GET_DIR_DELEGATION arg/resp structure
9440 * @nf: nfsd_file opened on the directory
9441 *
9442 * Given a GET_DIR_DELEGATION request @gdd, attempt to acquire a delegation
9443 * on the directory to which @nf refers. Note that this does not set up any
9444 * sort of async notifications for the delegation.
9445 */
9446struct nfs4_delegation *
9447nfsd_get_dir_deleg(struct nfsd4_compound_state *cstate,
9448		   struct nfsd4_get_dir_delegation *gdd,
9449		   struct nfsd_file *nf)
9450{
9451	struct nfs4_client *clp = cstate->clp;
9452	struct nfs4_delegation *dp;
9453	struct file_lease *fl;
9454	struct nfs4_file *fp, *rfp;
9455	int status = 0;
9456
9457	fp = nfsd4_alloc_file();
9458	if (!fp)
9459		return ERR_PTR(-ENOMEM);
9460
9461	nfsd4_file_init(&cstate->current_fh, fp);
9462
9463	rfp = nfsd4_file_hash_insert(fp, &cstate->current_fh);
9464	if (unlikely(!rfp)) {
9465		put_nfs4_file(fp);
9466		return ERR_PTR(-ENOMEM);
9467	}
9468
9469	if (rfp != fp) {
9470		put_nfs4_file(fp);
9471		fp = rfp;
9472	}
9473
9474	/* if this client already has one, return that it's unavailable */
9475	spin_lock(&state_lock);
9476	spin_lock(&fp->fi_lock);
9477	/* existing delegation? */
9478	if (nfs4_delegation_exists(clp, fp)) {
9479		status = -EAGAIN;
9480	} else if (!fp->fi_deleg_file) {
9481		fp->fi_deleg_file = nfsd_file_get(nf);
9482		fp->fi_delegees = 1;
9483	} else {
9484		++fp->fi_delegees;
9485	}
9486	spin_unlock(&fp->fi_lock);
9487	spin_unlock(&state_lock);
9488
9489	if (status) {
9490		put_nfs4_file(fp);
9491		return ERR_PTR(status);
9492	}
9493
9494	/* Try to set up the lease */
9495	status = -ENOMEM;
9496	dp = alloc_init_deleg(clp, fp, NULL, NFS4_OPEN_DELEGATE_READ);
9497	if (!dp)
9498		goto out_delegees;
9499
9500	fl = nfs4_alloc_init_lease(dp);
9501	if (!fl)
9502		goto out_put_stid;
9503
9504	status = kernel_setlease(nf->nf_file,
9505				 fl->c.flc_type, &fl, NULL);
9506	if (fl)
9507		locks_free_lease(fl);
9508	if (status)
9509		goto out_put_stid;
9510
9511	/*
9512	 * Now, try to hash it. This can fail if we race another nfsd task
9513	 * trying to set a delegation on the same file. If that happens,
9514	 * then just say UNAVAIL.
9515	 */
9516	spin_lock(&state_lock);
9517	spin_lock(&clp->cl_lock);
9518	spin_lock(&fp->fi_lock);
9519	status = hash_delegation_locked(dp, fp);
9520	spin_unlock(&fp->fi_lock);
9521	spin_unlock(&clp->cl_lock);
9522	spin_unlock(&state_lock);
9523
9524	if (!status) {
9525		put_nfs4_file(fp);
9526		return dp;
9527	}
9528
9529	/* Something failed. Drop the lease and clean up the stid */
9530	kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
9531out_put_stid:
9532	nfs4_put_stid(&dp->dl_stid);
9533out_delegees:
9534	put_deleg_file(fp);
9535	put_nfs4_file(fp);
9536	return ERR_PTR(status);
9537}