fs/nfsd/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}
  63 static const stateid_t one_stateid = {
  64         .si_generation = ~0,
  65         .si_opaque = all_ones,
  66 };
  67 static const stateid_t zero_stateid = {
  68         /* all fields zero */
  69 };
  70 static const stateid_t currentstateid = {
  71         .si_generation = 1,
  72 };
  73 static const stateid_t close_stateid = {
  74         .si_generation = 0xffffffffU,
  75 };
  76
  77 static 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 */
  85 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  86 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  87 void nfsd4_end_grace(struct nfsd_net *nn);
  88 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
  89 static void nfsd4_file_hash_remove(struct nfs4_file *fi);
  90 static 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  */
  99 static DEFINE_SPINLOCK(state_lock);
 100
 101 enum 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  */
 110 static 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  */
 117 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 118
 119 static struct kmem_cache *client_slab;
 120 static struct kmem_cache *openowner_slab;
 121 static struct kmem_cache *lockowner_slab;
 122 static struct kmem_cache *file_slab;
 123 static struct kmem_cache *stateid_slab;
 124 static struct kmem_cache *deleg_slab;
 125 static struct kmem_cache *odstate_slab;
 126
 127 static void free_session(struct nfsd4_session *);
 128
 129 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 130 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 131 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops;
 132
 133 static struct workqueue_struct *laundry_wq;
 134
 135 int 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
 145 void nfsd4_destroy_laundry_wq(void)
 146 {
 147         destroy_workqueue(laundry_wq);
 148 }
 149
 150 static bool is_session_dead(struct nfsd4_session *ses)
 151 {
 152         return ses->se_flags & NFS4_SESSION_DEAD;
 153 }
 154
 155 static __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_flags |= NFS4_SESSION_DEAD;
 160         return nfs_ok;
 161 }
 162
 163 static bool is_client_expired(struct nfs4_client *clp)
 164 {
 165         return clp->cl_time == 0;
 166 }
 167
 168 static 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
 175 static __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 */
 190 static inline void
 191 renew_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
 210 static 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
 224 static 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
 237 static __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
 250 static 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
 262 static 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
 272 static struct nfsd4_blocked_lock *
 273 find_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
 294 static struct nfsd4_blocked_lock *
 295 find_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
 317 static void
 318 free_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
 327 static void
 328 free_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
 334 static void
 335 remove_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
 363 static void
 364 nfsd4_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
 371 static int
 372 nfsd4_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
 390 static void
 391 nfsd4_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
 399 static 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 };
 404
 405 /*
 406  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 407  * st_{access,deny}_bmap field of the stateid, in order to track not
 408  * only what share bits are currently in force, but also what
 409  * combinations of share bits previous opens have used.  This allows us
 410  * to enforce the recommendation in
 411  * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
 412  * the server return an error if the client attempt to downgrade to a
 413  * combination of share bits not explicable by closing some of its
 414  * previous opens.
 415  *
 416  * This enforcement is arguably incomplete, since we don't keep
 417  * track of access/deny bit combinations; so, e.g., we allow:
 418  *
 419  *      OPEN allow read, deny write
 420  *      OPEN allow both, deny none
 421  *      DOWNGRADE allow read, deny none
 422  *
 423  * which we should reject.
 424  *
 425  * But you could also argue that our current code is already overkill,
 426  * since it only exists to return NFS4ERR_INVAL on incorrect client
 427  * behavior.
 428  */
 429 static unsigned int
 430 bmap_to_share_mode(unsigned long bmap)
 431 {
 432         int i;
 433         unsigned int access = 0;
 434
 435         for (i = 1; i < 4; i++) {
 436                 if (test_bit(i, &bmap))
 437                         access |= i;
 438         }
 439         return access;
 440 }
 441
 442 /* set share access for a given stateid */
 443 static inline void
 444 set_access(u32 access, struct nfs4_ol_stateid *stp)
 445 {
 446         unsigned char mask = 1 << access;
 447
 448         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 449         stp->st_access_bmap |= mask;
 450 }
 451
 452 /* clear share access for a given stateid */
 453 static inline void
 454 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 455 {
 456         unsigned char mask = 1 << access;
 457
 458         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 459         stp->st_access_bmap &= ~mask;
 460 }
 461
 462 /* test whether a given stateid has access */
 463 static inline bool
 464 test_access(u32 access, struct nfs4_ol_stateid *stp)
 465 {
 466         unsigned char mask = 1 << access;
 467
 468         return (bool)(stp->st_access_bmap & mask);
 469 }
 470
 471 /* set share deny for a given stateid */
 472 static inline void
 473 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 474 {
 475         unsigned char mask = 1 << deny;
 476
 477         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 478         stp->st_deny_bmap |= mask;
 479 }
 480
 481 /* clear share deny for a given stateid */
 482 static inline void
 483 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 484 {
 485         unsigned char mask = 1 << deny;
 486
 487         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 488         stp->st_deny_bmap &= ~mask;
 489 }
 490
 491 /* test whether a given stateid is denying specific access */
 492 static inline bool
 493 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 494 {
 495         unsigned char mask = 1 << deny;
 496
 497         return (bool)(stp->st_deny_bmap & mask);
 498 }
 499
 500 static int nfs4_access_to_omode(u32 access)
 501 {
 502         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 503         case NFS4_SHARE_ACCESS_READ:
 504                 return O_RDONLY;
 505         case NFS4_SHARE_ACCESS_WRITE:
 506                 return O_WRONLY;
 507         case NFS4_SHARE_ACCESS_BOTH:
 508                 return O_RDWR;
 509         }
 510         WARN_ON_ONCE(1);
 511         return O_RDONLY;
 512 }
 513
 514 static inline int
 515 access_permit_read(struct nfs4_ol_stateid *stp)
 516 {
 517         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
 518                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
 519                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
 520 }
 521
 522 static inline int
 523 access_permit_write(struct nfs4_ol_stateid *stp)
 524 {
 525         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
 526                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
 527 }
 528
 529 static inline struct nfs4_stateowner *
 530 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 531 {
 532         atomic_inc(&sop->so_count);
 533         return sop;
 534 }
 535
 536 static int
 537 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 538 {
 539         return (sop->so_owner.len == owner->len) &&
 540                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 541 }
 542
 543 static struct nfs4_openowner *
 544 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 545                         struct nfs4_client *clp)
 546 {
 547         struct nfs4_stateowner *so;
 548
 549         lockdep_assert_held(&clp->cl_lock);
 550
 551         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 552                             so_strhash) {
 553                 if (!so->so_is_open_owner)
 554                         continue;
 555                 if (same_owner_str(so, &open->op_owner))
 556                         return openowner(nfs4_get_stateowner(so));
 557         }
 558         return NULL;
 559 }
 560
 561 static struct nfs4_openowner *
 562 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 563                         struct nfs4_client *clp)
 564 {
 565         struct nfs4_openowner *oo;
 566
 567         spin_lock(&clp->cl_lock);
 568         oo = find_openstateowner_str_locked(hashval, open, clp);
 569         spin_unlock(&clp->cl_lock);
 570         return oo;
 571 }
 572
 573 static inline u32
 574 opaque_hashval(const void *ptr, int nbytes)
 575 {
 576         unsigned char *cptr = (unsigned char *) ptr;
 577
 578         u32 x = 0;
 579         while (nbytes--) {
 580                 x *= 37;
 581                 x += *cptr++;
 582         }
 583         return x;
 584 }
 585
 586 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 587 {
 588         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 589
 590         kmem_cache_free(file_slab, fp);
 591 }
 592
 593 void
 594 put_nfs4_file(struct nfs4_file *fi)
 595 {
 596         if (refcount_dec_and_test(&fi->fi_ref)) {
 597                 nfsd4_file_hash_remove(fi);
 598                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 599                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 600                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 601         }
 602 }
 603
 604 static struct nfsd_file *
 605 find_writeable_file_locked(struct nfs4_file *f)
 606 {
 607         struct nfsd_file *ret;
 608
 609         lockdep_assert_held(&f->fi_lock);
 610
 611         ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 612         if (!ret)
 613                 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 614         return ret;
 615 }
 616
 617 static struct nfsd_file *
 618 find_writeable_file(struct nfs4_file *f)
 619 {
 620         struct nfsd_file *ret;
 621
 622         spin_lock(&f->fi_lock);
 623         ret = find_writeable_file_locked(f);
 624         spin_unlock(&f->fi_lock);
 625
 626         return ret;
 627 }
 628
 629 static struct nfsd_file *
 630 find_readable_file_locked(struct nfs4_file *f)
 631 {
 632         struct nfsd_file *ret;
 633
 634         lockdep_assert_held(&f->fi_lock);
 635
 636         ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 637         if (!ret)
 638                 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 639         return ret;
 640 }
 641
 642 static struct nfsd_file *
 643 find_readable_file(struct nfs4_file *f)
 644 {
 645         struct nfsd_file *ret;
 646
 647         spin_lock(&f->fi_lock);
 648         ret = find_readable_file_locked(f);
 649         spin_unlock(&f->fi_lock);
 650
 651         return ret;
 652 }
 653
 654 static struct nfsd_file *
 655 find_rw_file(struct nfs4_file *f)
 656 {
 657         struct nfsd_file *ret;
 658
 659         spin_lock(&f->fi_lock);
 660         ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 661         spin_unlock(&f->fi_lock);
 662
 663         return ret;
 664 }
 665
 666 struct nfsd_file *
 667 find_any_file(struct nfs4_file *f)
 668 {
 669         struct nfsd_file *ret;
 670
 671         if (!f)
 672                 return NULL;
 673         spin_lock(&f->fi_lock);
 674         ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 675         if (!ret) {
 676                 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 677                 if (!ret)
 678                         ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 679         }
 680         spin_unlock(&f->fi_lock);
 681         return ret;
 682 }
 683
 684 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
 685 {
 686         lockdep_assert_held(&f->fi_lock);
 687
 688         if (f->fi_fds[O_RDWR])
 689                 return f->fi_fds[O_RDWR];
 690         if (f->fi_fds[O_WRONLY])
 691                 return f->fi_fds[O_WRONLY];
 692         if (f->fi_fds[O_RDONLY])
 693                 return f->fi_fds[O_RDONLY];
 694         return NULL;
 695 }
 696
 697 static atomic_long_t num_delegations;
 698 unsigned long max_delegations;
 699
 700 /*
 701  * Open owner state (share locks)
 702  */
 703
 704 /* hash tables for lock and open owners */
 705 #define OWNER_HASH_BITS              8
 706 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 707 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 708
 709 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 710 {
 711         unsigned int ret;
 712
 713         ret = opaque_hashval(ownername->data, ownername->len);
 714         return ret & OWNER_HASH_MASK;
 715 }
 716
 717 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
 718
 719 static const struct rhashtable_params nfs4_file_rhash_params = {
 720         .key_len                = sizeof_field(struct nfs4_file, fi_inode),
 721         .key_offset             = offsetof(struct nfs4_file, fi_inode),
 722         .head_offset            = offsetof(struct nfs4_file, fi_rlist),
 723
 724         /*
 725          * Start with a single page hash table to reduce resizing churn
 726          * on light workloads.
 727          */
 728         .min_size               = 256,
 729         .automatic_shrinking    = true,
 730 };
 731
 732 /*
 733  * Check if courtesy clients have conflicting access and resolve it if possible
 734  *
 735  * access:  is op_share_access if share_access is true.
 736  *          Check if access mode, op_share_access, would conflict with
 737  *          the current deny mode of the file 'fp'.
 738  * access:  is op_share_deny if share_access is false.
 739  *          Check if the deny mode, op_share_deny, would conflict with
 740  *          current access of the file 'fp'.
 741  * stp:     skip checking this entry.
 742  * new_stp: normal open, not open upgrade.
 743  *
 744  * Function returns:
 745  *      false - access/deny mode conflict with normal client.
 746  *      true  - no conflict or conflict with courtesy client(s) is resolved.
 747  */
 748 static bool
 749 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
 750                 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
 751 {
 752         struct nfs4_ol_stateid *st;
 753         bool resolvable = true;
 754         unsigned char bmap;
 755         struct nfsd_net *nn;
 756         struct nfs4_client *clp;
 757
 758         lockdep_assert_held(&fp->fi_lock);
 759         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
 760                 /* ignore lock stateid */
 761                 if (st->st_openstp)
 762                         continue;
 763                 if (st == stp && new_stp)
 764                         continue;
 765                 /* check file access against deny mode or vice versa */
 766                 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
 767                 if (!(access & bmap_to_share_mode(bmap)))
 768                         continue;
 769                 clp = st->st_stid.sc_client;
 770                 if (try_to_expire_client(clp))
 771                         continue;
 772                 resolvable = false;
 773                 break;
 774         }
 775         if (resolvable) {
 776                 clp = stp->st_stid.sc_client;
 777                 nn = net_generic(clp->net, nfsd_net_id);
 778                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
 779         }
 780         return resolvable;
 781 }
 782
 783 static void
 784 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 785 {
 786         lockdep_assert_held(&fp->fi_lock);
 787
 788         if (access & NFS4_SHARE_ACCESS_WRITE)
 789                 atomic_inc(&fp->fi_access[O_WRONLY]);
 790         if (access & NFS4_SHARE_ACCESS_READ)
 791                 atomic_inc(&fp->fi_access[O_RDONLY]);
 792 }
 793
 794 static __be32
 795 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 796 {
 797         lockdep_assert_held(&fp->fi_lock);
 798
 799         /* Does this access mode make sense? */
 800         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 801                 return nfserr_inval;
 802
 803         /* Does it conflict with a deny mode already set? */
 804         if ((access & fp->fi_share_deny) != 0)
 805                 return nfserr_share_denied;
 806
 807         __nfs4_file_get_access(fp, access);
 808         return nfs_ok;
 809 }
 810
 811 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 812 {
 813         /* Common case is that there is no deny mode. */
 814         if (deny) {
 815                 /* Does this deny mode make sense? */
 816                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 817                         return nfserr_inval;
 818
 819                 if ((deny & NFS4_SHARE_DENY_READ) &&
 820                     atomic_read(&fp->fi_access[O_RDONLY]))
 821                         return nfserr_share_denied;
 822
 823                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 824                     atomic_read(&fp->fi_access[O_WRONLY]))
 825                         return nfserr_share_denied;
 826         }
 827         return nfs_ok;
 828 }
 829
 830 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 831 {
 832         might_lock(&fp->fi_lock);
 833
 834         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 835                 struct nfsd_file *f1 = NULL;
 836                 struct nfsd_file *f2 = NULL;
 837
 838                 swap(f1, fp->fi_fds[oflag]);
 839                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 840                         swap(f2, fp->fi_fds[O_RDWR]);
 841                 spin_unlock(&fp->fi_lock);
 842                 if (f1)
 843                         nfsd_file_put(f1);
 844                 if (f2)
 845                         nfsd_file_put(f2);
 846         }
 847 }
 848
 849 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 850 {
 851         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 852
 853         if (access & NFS4_SHARE_ACCESS_WRITE)
 854                 __nfs4_file_put_access(fp, O_WRONLY);
 855         if (access & NFS4_SHARE_ACCESS_READ)
 856                 __nfs4_file_put_access(fp, O_RDONLY);
 857 }
 858
 859 /*
 860  * Allocate a new open/delegation state counter. This is needed for
 861  * pNFS for proper return on close semantics.
 862  *
 863  * Note that we only allocate it for pNFS-enabled exports, otherwise
 864  * all pointers to struct nfs4_clnt_odstate are always NULL.
 865  */
 866 static struct nfs4_clnt_odstate *
 867 alloc_clnt_odstate(struct nfs4_client *clp)
 868 {
 869         struct nfs4_clnt_odstate *co;
 870
 871         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 872         if (co) {
 873                 co->co_client = clp;
 874                 refcount_set(&co->co_odcount, 1);
 875         }
 876         return co;
 877 }
 878
 879 static void
 880 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 881 {
 882         struct nfs4_file *fp = co->co_file;
 883
 884         lockdep_assert_held(&fp->fi_lock);
 885         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 886 }
 887
 888 static inline void
 889 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 890 {
 891         if (co)
 892                 refcount_inc(&co->co_odcount);
 893 }
 894
 895 static void
 896 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 897 {
 898         struct nfs4_file *fp;
 899
 900         if (!co)
 901                 return;
 902
 903         fp = co->co_file;
 904         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 905                 list_del(&co->co_perfile);
 906                 spin_unlock(&fp->fi_lock);
 907
 908                 nfsd4_return_all_file_layouts(co->co_client, fp);
 909                 kmem_cache_free(odstate_slab, co);
 910         }
 911 }
 912
 913 static struct nfs4_clnt_odstate *
 914 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 915 {
 916         struct nfs4_clnt_odstate *co;
 917         struct nfs4_client *cl;
 918
 919         if (!new)
 920                 return NULL;
 921
 922         cl = new->co_client;
 923
 924         spin_lock(&fp->fi_lock);
 925         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 926                 if (co->co_client == cl) {
 927                         get_clnt_odstate(co);
 928                         goto out;
 929                 }
 930         }
 931         co = new;
 932         co->co_file = fp;
 933         hash_clnt_odstate_locked(new);
 934 out:
 935         spin_unlock(&fp->fi_lock);
 936         return co;
 937 }
 938
 939 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 940                                   void (*sc_free)(struct nfs4_stid *))
 941 {
 942         struct nfs4_stid *stid;
 943         int new_id;
 944
 945         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 946         if (!stid)
 947                 return NULL;
 948
 949         idr_preload(GFP_KERNEL);
 950         spin_lock(&cl->cl_lock);
 951         /* Reserving 0 for start of file in nfsdfs "states" file: */
 952         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 953         spin_unlock(&cl->cl_lock);
 954         idr_preload_end();
 955         if (new_id < 0)
 956                 goto out_free;
 957
 958         stid->sc_free = sc_free;
 959         stid->sc_client = cl;
 960         stid->sc_stateid.si_opaque.so_id = new_id;
 961         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 962         /* Will be incremented before return to client: */
 963         refcount_set(&stid->sc_count, 1);
 964         spin_lock_init(&stid->sc_lock);
 965         INIT_LIST_HEAD(&stid->sc_cp_list);
 966
 967         /*
 968          * It shouldn't be a problem to reuse an opaque stateid value.
 969          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 970          * example, a stray write retransmission could be accepted by
 971          * the server when it should have been rejected.  Therefore,
 972          * adopt a trick from the sctp code to attempt to maximize the
 973          * amount of time until an id is reused, by ensuring they always
 974          * "increase" (mod INT_MAX):
 975          */
 976         return stid;
 977 out_free:
 978         kmem_cache_free(slab, stid);
 979         return NULL;
 980 }
 981
 982 /*
 983  * Create a unique stateid_t to represent each COPY.
 984  */
 985 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 986                               unsigned char cs_type)
 987 {
 988         int new_id;
 989
 990         stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 991         stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 992
 993         idr_preload(GFP_KERNEL);
 994         spin_lock(&nn->s2s_cp_lock);
 995         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 996         stid->cs_stid.si_opaque.so_id = new_id;
 997         stid->cs_stid.si_generation = 1;
 998         spin_unlock(&nn->s2s_cp_lock);
 999         idr_preload_end();
1000         if (new_id < 0)
1001                 return 0;
1002         stid->cs_type = cs_type;
1003         return 1;
1004 }
1005
1006 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1007 {
1008         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
1009 }
1010
1011 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1012                                                      struct nfs4_stid *p_stid)
1013 {
1014         struct nfs4_cpntf_state *cps;
1015
1016         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1017         if (!cps)
1018                 return NULL;
1019         cps->cpntf_time = ktime_get_boottime_seconds();
1020         refcount_set(&cps->cp_stateid.cs_count, 1);
1021         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1022                 goto out_free;
1023         spin_lock(&nn->s2s_cp_lock);
1024         list_add(&cps->cp_list, &p_stid->sc_cp_list);
1025         spin_unlock(&nn->s2s_cp_lock);
1026         return cps;
1027 out_free:
1028         kfree(cps);
1029         return NULL;
1030 }
1031
1032 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1033 {
1034         struct nfsd_net *nn;
1035
1036         if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1037                 return;
1038         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1039         spin_lock(&nn->s2s_cp_lock);
1040         idr_remove(&nn->s2s_cp_stateids,
1041                    copy->cp_stateid.cs_stid.si_opaque.so_id);
1042         spin_unlock(&nn->s2s_cp_lock);
1043 }
1044
1045 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1046 {
1047         struct nfs4_cpntf_state *cps;
1048         struct nfsd_net *nn;
1049
1050         nn = net_generic(net, nfsd_net_id);
1051         spin_lock(&nn->s2s_cp_lock);
1052         while (!list_empty(&stid->sc_cp_list)) {
1053                 cps = list_first_entry(&stid->sc_cp_list,
1054                                        struct nfs4_cpntf_state, cp_list);
1055                 _free_cpntf_state_locked(nn, cps);
1056         }
1057         spin_unlock(&nn->s2s_cp_lock);
1058 }
1059
1060 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1061 {
1062         struct nfs4_stid *stid;
1063
1064         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1065         if (!stid)
1066                 return NULL;
1067
1068         return openlockstateid(stid);
1069 }
1070
1071 static void nfs4_free_deleg(struct nfs4_stid *stid)
1072 {
1073         struct nfs4_delegation *dp = delegstateid(stid);
1074
1075         WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1076         WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1077         WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1078         WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1079         kmem_cache_free(deleg_slab, stid);
1080         atomic_long_dec(&num_delegations);
1081 }
1082
1083 /*
1084  * When we recall a delegation, we should be careful not to hand it
1085  * out again straight away.
1086  * To ensure this we keep a pair of bloom filters ('new' and 'old')
1087  * in which the filehandles of recalled delegations are "stored".
1088  * If a filehandle appear in either filter, a delegation is blocked.
1089  * When a delegation is recalled, the filehandle is stored in the "new"
1090  * filter.
1091  * Every 30 seconds we swap the filters and clear the "new" one,
1092  * unless both are empty of course.
1093  *
1094  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
1095  * low 3 bytes as hash-table indices.
1096  *
1097  * 'blocked_delegations_lock', which is always taken in block_delegations(),
1098  * is used to manage concurrent access.  Testing does not need the lock
1099  * except when swapping the two filters.
1100  */
1101 static DEFINE_SPINLOCK(blocked_delegations_lock);
1102 static struct bloom_pair {
1103         int     entries, old_entries;
1104         time64_t swap_time;
1105         int     new; /* index into 'set' */
1106         DECLARE_BITMAP(set[2], 256);
1107 } blocked_delegations;
1108
1109 static int delegation_blocked(struct knfsd_fh *fh)
1110 {
1111         u32 hash;
1112         struct bloom_pair *bd = &blocked_delegations;
1113
1114         if (bd->entries == 0)
1115                 return 0;
1116         if (ktime_get_seconds() - bd->swap_time > 30) {
1117                 spin_lock(&blocked_delegations_lock);
1118                 if (ktime_get_seconds() - bd->swap_time > 30) {
1119                         bd->entries -= bd->old_entries;
1120                         bd->old_entries = bd->entries;
1121                         memset(bd->set[bd->new], 0,
1122                                sizeof(bd->set[0]));
1123                         bd->new = 1-bd->new;
1124                         bd->swap_time = ktime_get_seconds();
1125                 }
1126                 spin_unlock(&blocked_delegations_lock);
1127         }
1128         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1129         if (test_bit(hash&255, bd->set[0]) &&
1130             test_bit((hash>>8)&255, bd->set[0]) &&
1131             test_bit((hash>>16)&255, bd->set[0]))
1132                 return 1;
1133
1134         if (test_bit(hash&255, bd->set[1]) &&
1135             test_bit((hash>>8)&255, bd->set[1]) &&
1136             test_bit((hash>>16)&255, bd->set[1]))
1137                 return 1;
1138
1139         return 0;
1140 }
1141
1142 static void block_delegations(struct knfsd_fh *fh)
1143 {
1144         u32 hash;
1145         struct bloom_pair *bd = &blocked_delegations;
1146
1147         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1148
1149         spin_lock(&blocked_delegations_lock);
1150         __set_bit(hash&255, bd->set[bd->new]);
1151         __set_bit((hash>>8)&255, bd->set[bd->new]);
1152         __set_bit((hash>>16)&255, bd->set[bd->new]);
1153         if (bd->entries == 0)
1154                 bd->swap_time = ktime_get_seconds();
1155         bd->entries += 1;
1156         spin_unlock(&blocked_delegations_lock);
1157 }
1158
1159 static struct nfs4_delegation *
1160 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1161                  struct nfs4_clnt_odstate *odstate, u32 dl_type)
1162 {
1163         struct nfs4_delegation *dp;
1164         struct nfs4_stid *stid;
1165         long n;
1166
1167         dprintk("NFSD alloc_init_deleg\n");
1168         n = atomic_long_inc_return(&num_delegations);
1169         if (n < 0 || n > max_delegations)
1170                 goto out_dec;
1171         if (delegation_blocked(&fp->fi_fhandle))
1172                 goto out_dec;
1173         stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1174         if (stid == NULL)
1175                 goto out_dec;
1176         dp = delegstateid(stid);
1177
1178         /*
1179          * delegation seqid's are never incremented.  The 4.1 special
1180          * meaning of seqid 0 isn't meaningful, really, but let's avoid
1181          * 0 anyway just for consistency and use 1:
1182          */
1183         dp->dl_stid.sc_stateid.si_generation = 1;
1184         INIT_LIST_HEAD(&dp->dl_perfile);
1185         INIT_LIST_HEAD(&dp->dl_perclnt);
1186         INIT_LIST_HEAD(&dp->dl_recall_lru);
1187         dp->dl_clnt_odstate = odstate;
1188         get_clnt_odstate(odstate);
1189         dp->dl_type = dl_type;
1190         dp->dl_retries = 1;
1191         dp->dl_recalled = false;
1192         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1193                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1194         nfsd4_init_cb(&dp->dl_cb_fattr.ncf_getattr, dp->dl_stid.sc_client,
1195                         &nfsd4_cb_getattr_ops, NFSPROC4_CLNT_CB_GETATTR);
1196         dp->dl_cb_fattr.ncf_file_modified = false;
1197         dp->dl_cb_fattr.ncf_cb_bmap[0] = FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE;
1198         get_nfs4_file(fp);
1199         dp->dl_stid.sc_file = fp;
1200         return dp;
1201 out_dec:
1202         atomic_long_dec(&num_delegations);
1203         return NULL;
1204 }
1205
1206 void
1207 nfs4_put_stid(struct nfs4_stid *s)
1208 {
1209         struct nfs4_file *fp = s->sc_file;
1210         struct nfs4_client *clp = s->sc_client;
1211
1212         might_lock(&clp->cl_lock);
1213
1214         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1215                 wake_up_all(&close_wq);
1216                 return;
1217         }
1218         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1219         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1220                 atomic_dec(&s->sc_client->cl_admin_revoked);
1221         nfs4_free_cpntf_statelist(clp->net, s);
1222         spin_unlock(&clp->cl_lock);
1223         s->sc_free(s);
1224         if (fp)
1225                 put_nfs4_file(fp);
1226 }
1227
1228 void
1229 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1230 {
1231         stateid_t *src = &stid->sc_stateid;
1232
1233         spin_lock(&stid->sc_lock);
1234         if (unlikely(++src->si_generation == 0))
1235                 src->si_generation = 1;
1236         memcpy(dst, src, sizeof(*dst));
1237         spin_unlock(&stid->sc_lock);
1238 }
1239
1240 static void put_deleg_file(struct nfs4_file *fp)
1241 {
1242         struct nfsd_file *nf = NULL;
1243
1244         spin_lock(&fp->fi_lock);
1245         if (--fp->fi_delegees == 0)
1246                 swap(nf, fp->fi_deleg_file);
1247         spin_unlock(&fp->fi_lock);
1248
1249         if (nf)
1250                 nfsd_file_put(nf);
1251 }
1252
1253 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1254 {
1255         struct nfs4_file *fp = dp->dl_stid.sc_file;
1256         struct nfsd_file *nf = fp->fi_deleg_file;
1257
1258         WARN_ON_ONCE(!fp->fi_delegees);
1259
1260         kernel_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1261         put_deleg_file(fp);
1262 }
1263
1264 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1265 {
1266         put_clnt_odstate(dp->dl_clnt_odstate);
1267         nfs4_unlock_deleg_lease(dp);
1268         nfs4_put_stid(&dp->dl_stid);
1269 }
1270
1271 /**
1272  * nfs4_delegation_exists - Discover if this delegation already exists
1273  * @clp:     a pointer to the nfs4_client we're granting a delegation to
1274  * @fp:      a pointer to the nfs4_file we're granting a delegation on
1275  *
1276  * Return:
1277  *      On success: true iff an existing delegation is found
1278  */
1279
1280 static bool
1281 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1282 {
1283         struct nfs4_delegation *searchdp = NULL;
1284         struct nfs4_client *searchclp = NULL;
1285
1286         lockdep_assert_held(&state_lock);
1287         lockdep_assert_held(&fp->fi_lock);
1288
1289         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1290                 searchclp = searchdp->dl_stid.sc_client;
1291                 if (clp == searchclp) {
1292                         return true;
1293                 }
1294         }
1295         return false;
1296 }
1297
1298 /**
1299  * hash_delegation_locked - Add a delegation to the appropriate lists
1300  * @dp:     a pointer to the nfs4_delegation we are adding.
1301  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1302  *
1303  * Return:
1304  *      On success: NULL if the delegation was successfully hashed.
1305  *
1306  *      On error: -EAGAIN if one was previously granted to this
1307  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1308  *
1309  */
1310
1311 static int
1312 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1313 {
1314         struct nfs4_client *clp = dp->dl_stid.sc_client;
1315
1316         lockdep_assert_held(&state_lock);
1317         lockdep_assert_held(&fp->fi_lock);
1318         lockdep_assert_held(&clp->cl_lock);
1319
1320         if (nfs4_delegation_exists(clp, fp))
1321                 return -EAGAIN;
1322         refcount_inc(&dp->dl_stid.sc_count);
1323         dp->dl_stid.sc_type = SC_TYPE_DELEG;
1324         list_add(&dp->dl_perfile, &fp->fi_delegations);
1325         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1326         return 0;
1327 }
1328
1329 static bool delegation_hashed(struct nfs4_delegation *dp)
1330 {
1331         return !(list_empty(&dp->dl_perfile));
1332 }
1333
1334 static bool
1335 unhash_delegation_locked(struct nfs4_delegation *dp, unsigned short statusmask)
1336 {
1337         struct nfs4_file *fp = dp->dl_stid.sc_file;
1338
1339         lockdep_assert_held(&state_lock);
1340
1341         if (!delegation_hashed(dp))
1342                 return false;
1343
1344         if (statusmask == SC_STATUS_REVOKED &&
1345             dp->dl_stid.sc_client->cl_minorversion == 0)
1346                 statusmask = SC_STATUS_CLOSED;
1347         dp->dl_stid.sc_status |= statusmask;
1348         if (statusmask & SC_STATUS_ADMIN_REVOKED)
1349                 atomic_inc(&dp->dl_stid.sc_client->cl_admin_revoked);
1350
1351         /* Ensure that deleg break won't try to requeue it */
1352         ++dp->dl_time;
1353         spin_lock(&fp->fi_lock);
1354         list_del_init(&dp->dl_perclnt);
1355         list_del_init(&dp->dl_recall_lru);
1356         list_del_init(&dp->dl_perfile);
1357         spin_unlock(&fp->fi_lock);
1358         return true;
1359 }
1360
1361 static void destroy_delegation(struct nfs4_delegation *dp)
1362 {
1363         bool unhashed;
1364
1365         spin_lock(&state_lock);
1366         unhashed = unhash_delegation_locked(dp, SC_STATUS_CLOSED);
1367         spin_unlock(&state_lock);
1368         if (unhashed)
1369                 destroy_unhashed_deleg(dp);
1370 }
1371
1372 static void revoke_delegation(struct nfs4_delegation *dp)
1373 {
1374         struct nfs4_client *clp = dp->dl_stid.sc_client;
1375
1376         WARN_ON(!list_empty(&dp->dl_recall_lru));
1377
1378         trace_nfsd_stid_revoke(&dp->dl_stid);
1379
1380         if (dp->dl_stid.sc_status &
1381             (SC_STATUS_REVOKED | SC_STATUS_ADMIN_REVOKED)) {
1382                 spin_lock(&clp->cl_lock);
1383                 refcount_inc(&dp->dl_stid.sc_count);
1384                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1385                 spin_unlock(&clp->cl_lock);
1386         }
1387         destroy_unhashed_deleg(dp);
1388 }
1389
1390 /*
1391  * SETCLIENTID state
1392  */
1393
1394 static unsigned int clientid_hashval(u32 id)
1395 {
1396         return id & CLIENT_HASH_MASK;
1397 }
1398
1399 static unsigned int clientstr_hashval(struct xdr_netobj name)
1400 {
1401         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1402 }
1403
1404 /*
1405  * A stateid that had a deny mode associated with it is being released
1406  * or downgraded. Recalculate the deny mode on the file.
1407  */
1408 static void
1409 recalculate_deny_mode(struct nfs4_file *fp)
1410 {
1411         struct nfs4_ol_stateid *stp;
1412
1413         spin_lock(&fp->fi_lock);
1414         fp->fi_share_deny = 0;
1415         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1416                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1417         spin_unlock(&fp->fi_lock);
1418 }
1419
1420 static void
1421 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1422 {
1423         int i;
1424         bool change = false;
1425
1426         for (i = 1; i < 4; i++) {
1427                 if ((i & deny) != i) {
1428                         change = true;
1429                         clear_deny(i, stp);
1430                 }
1431         }
1432
1433         /* Recalculate per-file deny mode if there was a change */
1434         if (change)
1435                 recalculate_deny_mode(stp->st_stid.sc_file);
1436 }
1437
1438 /* release all access and file references for a given stateid */
1439 static void
1440 release_all_access(struct nfs4_ol_stateid *stp)
1441 {
1442         int i;
1443         struct nfs4_file *fp = stp->st_stid.sc_file;
1444
1445         if (fp && stp->st_deny_bmap != 0)
1446                 recalculate_deny_mode(fp);
1447
1448         for (i = 1; i < 4; i++) {
1449                 if (test_access(i, stp))
1450                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1451                 clear_access(i, stp);
1452         }
1453 }
1454
1455 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1456 {
1457         kfree(sop->so_owner.data);
1458         sop->so_ops->so_free(sop);
1459 }
1460
1461 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1462 {
1463         struct nfs4_client *clp = sop->so_client;
1464
1465         might_lock(&clp->cl_lock);
1466
1467         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1468                 return;
1469         sop->so_ops->so_unhash(sop);
1470         spin_unlock(&clp->cl_lock);
1471         nfs4_free_stateowner(sop);
1472 }
1473
1474 static bool
1475 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1476 {
1477         return list_empty(&stp->st_perfile);
1478 }
1479
1480 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1481 {
1482         struct nfs4_file *fp = stp->st_stid.sc_file;
1483
1484         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1485
1486         if (list_empty(&stp->st_perfile))
1487                 return false;
1488
1489         spin_lock(&fp->fi_lock);
1490         list_del_init(&stp->st_perfile);
1491         spin_unlock(&fp->fi_lock);
1492         list_del(&stp->st_perstateowner);
1493         return true;
1494 }
1495
1496 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1497 {
1498         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1499
1500         put_clnt_odstate(stp->st_clnt_odstate);
1501         release_all_access(stp);
1502         if (stp->st_stateowner)
1503                 nfs4_put_stateowner(stp->st_stateowner);
1504         WARN_ON(!list_empty(&stid->sc_cp_list));
1505         kmem_cache_free(stateid_slab, stid);
1506 }
1507
1508 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1509 {
1510         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1511         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1512         struct nfsd_file *nf;
1513
1514         nf = find_any_file(stp->st_stid.sc_file);
1515         if (nf) {
1516                 get_file(nf->nf_file);
1517                 filp_close(nf->nf_file, (fl_owner_t)lo);
1518                 nfsd_file_put(nf);
1519         }
1520         nfs4_free_ol_stateid(stid);
1521 }
1522
1523 /*
1524  * Put the persistent reference to an already unhashed generic stateid, while
1525  * holding the cl_lock. If it's the last reference, then put it onto the
1526  * reaplist for later destruction.
1527  */
1528 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1529                                        struct list_head *reaplist)
1530 {
1531         struct nfs4_stid *s = &stp->st_stid;
1532         struct nfs4_client *clp = s->sc_client;
1533
1534         lockdep_assert_held(&clp->cl_lock);
1535
1536         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1537
1538         if (!refcount_dec_and_test(&s->sc_count)) {
1539                 wake_up_all(&close_wq);
1540                 return;
1541         }
1542
1543         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1544         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1545                 atomic_dec(&s->sc_client->cl_admin_revoked);
1546         list_add(&stp->st_locks, reaplist);
1547 }
1548
1549 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1550 {
1551         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1552
1553         if (!unhash_ol_stateid(stp))
1554                 return false;
1555         list_del_init(&stp->st_locks);
1556         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1557         return true;
1558 }
1559
1560 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1561 {
1562         struct nfs4_client *clp = stp->st_stid.sc_client;
1563         bool unhashed;
1564
1565         spin_lock(&clp->cl_lock);
1566         unhashed = unhash_lock_stateid(stp);
1567         spin_unlock(&clp->cl_lock);
1568         if (unhashed)
1569                 nfs4_put_stid(&stp->st_stid);
1570 }
1571
1572 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1573 {
1574         struct nfs4_client *clp = lo->lo_owner.so_client;
1575
1576         lockdep_assert_held(&clp->cl_lock);
1577
1578         list_del_init(&lo->lo_owner.so_strhash);
1579 }
1580
1581 /*
1582  * Free a list of generic stateids that were collected earlier after being
1583  * fully unhashed.
1584  */
1585 static void
1586 free_ol_stateid_reaplist(struct list_head *reaplist)
1587 {
1588         struct nfs4_ol_stateid *stp;
1589         struct nfs4_file *fp;
1590
1591         might_sleep();
1592
1593         while (!list_empty(reaplist)) {
1594                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1595                                        st_locks);
1596                 list_del(&stp->st_locks);
1597                 fp = stp->st_stid.sc_file;
1598                 stp->st_stid.sc_free(&stp->st_stid);
1599                 if (fp)
1600                         put_nfs4_file(fp);
1601         }
1602 }
1603
1604 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1605                                        struct list_head *reaplist)
1606 {
1607         struct nfs4_ol_stateid *stp;
1608
1609         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1610
1611         while (!list_empty(&open_stp->st_locks)) {
1612                 stp = list_entry(open_stp->st_locks.next,
1613                                 struct nfs4_ol_stateid, st_locks);
1614                 unhash_lock_stateid(stp);
1615                 put_ol_stateid_locked(stp, reaplist);
1616         }
1617 }
1618
1619 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1620                                 struct list_head *reaplist)
1621 {
1622         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1623
1624         if (!unhash_ol_stateid(stp))
1625                 return false;
1626         release_open_stateid_locks(stp, reaplist);
1627         return true;
1628 }
1629
1630 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1631 {
1632         LIST_HEAD(reaplist);
1633
1634         spin_lock(&stp->st_stid.sc_client->cl_lock);
1635         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1636         if (unhash_open_stateid(stp, &reaplist))
1637                 put_ol_stateid_locked(stp, &reaplist);
1638         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1639         free_ol_stateid_reaplist(&reaplist);
1640 }
1641
1642 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1643 {
1644         struct nfs4_client *clp = oo->oo_owner.so_client;
1645
1646         lockdep_assert_held(&clp->cl_lock);
1647
1648         list_del_init(&oo->oo_owner.so_strhash);
1649         list_del_init(&oo->oo_perclient);
1650 }
1651
1652 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1653 {
1654         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1655                                           nfsd_net_id);
1656         struct nfs4_ol_stateid *s;
1657
1658         spin_lock(&nn->client_lock);
1659         s = oo->oo_last_closed_stid;
1660         if (s) {
1661                 list_del_init(&oo->oo_close_lru);
1662                 oo->oo_last_closed_stid = NULL;
1663         }
1664         spin_unlock(&nn->client_lock);
1665         if (s)
1666                 nfs4_put_stid(&s->st_stid);
1667 }
1668
1669 static void release_openowner(struct nfs4_openowner *oo)
1670 {
1671         struct nfs4_ol_stateid *stp;
1672         struct nfs4_client *clp = oo->oo_owner.so_client;
1673         struct list_head reaplist;
1674
1675         INIT_LIST_HEAD(&reaplist);
1676
1677         spin_lock(&clp->cl_lock);
1678         unhash_openowner_locked(oo);
1679         while (!list_empty(&oo->oo_owner.so_stateids)) {
1680                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1681                                 struct nfs4_ol_stateid, st_perstateowner);
1682                 if (unhash_open_stateid(stp, &reaplist))
1683                         put_ol_stateid_locked(stp, &reaplist);
1684         }
1685         spin_unlock(&clp->cl_lock);
1686         free_ol_stateid_reaplist(&reaplist);
1687         release_last_closed_stateid(oo);
1688         nfs4_put_stateowner(&oo->oo_owner);
1689 }
1690
1691 static struct nfs4_stid *find_one_sb_stid(struct nfs4_client *clp,
1692                                           struct super_block *sb,
1693                                           unsigned int sc_types)
1694 {
1695         unsigned long id, tmp;
1696         struct nfs4_stid *stid;
1697
1698         spin_lock(&clp->cl_lock);
1699         idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
1700                 if ((stid->sc_type & sc_types) &&
1701                     stid->sc_status == 0 &&
1702                     stid->sc_file->fi_inode->i_sb == sb) {
1703                         refcount_inc(&stid->sc_count);
1704                         break;
1705                 }
1706         spin_unlock(&clp->cl_lock);
1707         return stid;
1708 }
1709
1710 /**
1711  * nfsd4_revoke_states - revoke all nfsv4 states associated with given filesystem
1712  * @net:  used to identify instance of nfsd (there is one per net namespace)
1713  * @sb:   super_block used to identify target filesystem
1714  *
1715  * All nfs4 states (open, lock, delegation, layout) held by the server instance
1716  * and associated with a file on the given filesystem will be revoked resulting
1717  * in any files being closed and so all references from nfsd to the filesystem
1718  * being released.  Thus nfsd will no longer prevent the filesystem from being
1719  * unmounted.
1720  *
1721  * The clients which own the states will subsequently being notified that the
1722  * states have been "admin-revoked".
1723  */
1724 void nfsd4_revoke_states(struct net *net, struct super_block *sb)
1725 {
1726         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1727         unsigned int idhashval;
1728         unsigned int sc_types;
1729
1730         sc_types = SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG | SC_TYPE_LAYOUT;
1731
1732         spin_lock(&nn->client_lock);
1733         for (idhashval = 0; idhashval < CLIENT_HASH_MASK; idhashval++) {
1734                 struct list_head *head = &nn->conf_id_hashtbl[idhashval];
1735                 struct nfs4_client *clp;
1736         retry:
1737                 list_for_each_entry(clp, head, cl_idhash) {
1738                         struct nfs4_stid *stid = find_one_sb_stid(clp, sb,
1739                                                                   sc_types);
1740                         if (stid) {
1741                                 struct nfs4_ol_stateid *stp;
1742                                 struct nfs4_delegation *dp;
1743                                 struct nfs4_layout_stateid *ls;
1744
1745                                 spin_unlock(&nn->client_lock);
1746                                 switch (stid->sc_type) {
1747                                 case SC_TYPE_OPEN:
1748                                         stp = openlockstateid(stid);
1749                                         mutex_lock_nested(&stp->st_mutex,
1750                                                           OPEN_STATEID_MUTEX);
1751
1752                                         spin_lock(&clp->cl_lock);
1753                                         if (stid->sc_status == 0) {
1754                                                 stid->sc_status |=
1755                                                         SC_STATUS_ADMIN_REVOKED;
1756                                                 atomic_inc(&clp->cl_admin_revoked);
1757                                                 spin_unlock(&clp->cl_lock);
1758                                                 release_all_access(stp);
1759                                         } else
1760                                                 spin_unlock(&clp->cl_lock);
1761                                         mutex_unlock(&stp->st_mutex);
1762                                         break;
1763                                 case SC_TYPE_LOCK:
1764                                         stp = openlockstateid(stid);
1765                                         mutex_lock_nested(&stp->st_mutex,
1766                                                           LOCK_STATEID_MUTEX);
1767                                         spin_lock(&clp->cl_lock);
1768                                         if (stid->sc_status == 0) {
1769                                                 struct nfs4_lockowner *lo =
1770                                                         lockowner(stp->st_stateowner);
1771                                                 struct nfsd_file *nf;
1772
1773                                                 stid->sc_status |=
1774                                                         SC_STATUS_ADMIN_REVOKED;
1775                                                 atomic_inc(&clp->cl_admin_revoked);
1776                                                 spin_unlock(&clp->cl_lock);
1777                                                 nf = find_any_file(stp->st_stid.sc_file);
1778                                                 if (nf) {
1779                                                         get_file(nf->nf_file);
1780                                                         filp_close(nf->nf_file,
1781                                                                    (fl_owner_t)lo);
1782                                                         nfsd_file_put(nf);
1783                                                 }
1784                                                 release_all_access(stp);
1785                                         } else
1786                                                 spin_unlock(&clp->cl_lock);
1787                                         mutex_unlock(&stp->st_mutex);
1788                                         break;
1789                                 case SC_TYPE_DELEG:
1790                                         dp = delegstateid(stid);
1791                                         spin_lock(&state_lock);
1792                                         if (!unhash_delegation_locked(
1793                                                     dp, SC_STATUS_ADMIN_REVOKED))
1794                                                 dp = NULL;
1795                                         spin_unlock(&state_lock);
1796                                         if (dp)
1797                                                 revoke_delegation(dp);
1798                                         break;
1799                                 case SC_TYPE_LAYOUT:
1800                                         ls = layoutstateid(stid);
1801                                         nfsd4_close_layout(ls);
1802                                         break;
1803                                 }
1804                                 nfs4_put_stid(stid);
1805                                 spin_lock(&nn->client_lock);
1806                                 if (clp->cl_minorversion == 0)
1807                                         /* Allow cleanup after a lease period.
1808                                          * store_release ensures cleanup will
1809                                          * see any newly revoked states if it
1810                                          * sees the time updated.
1811                                          */
1812                                         nn->nfs40_last_revoke =
1813                                                 ktime_get_boottime_seconds();
1814                                 goto retry;
1815                         }
1816                 }
1817         }
1818         spin_unlock(&nn->client_lock);
1819 }
1820
1821 static inline int
1822 hash_sessionid(struct nfs4_sessionid *sessionid)
1823 {
1824         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1825
1826         return sid->sequence % SESSION_HASH_SIZE;
1827 }
1828
1829 #ifdef CONFIG_SUNRPC_DEBUG
1830 static inline void
1831 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1832 {
1833         u32 *ptr = (u32 *)(&sessionid->data[0]);
1834         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1835 }
1836 #else
1837 static inline void
1838 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1839 {
1840 }
1841 #endif
1842
1843 /*
1844  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1845  * won't be used for replay.
1846  */
1847 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1848 {
1849         struct nfs4_stateowner *so = cstate->replay_owner;
1850
1851         if (nfserr == nfserr_replay_me)
1852                 return;
1853
1854         if (!seqid_mutating_err(ntohl(nfserr))) {
1855                 nfsd4_cstate_clear_replay(cstate);
1856                 return;
1857         }
1858         if (!so)
1859                 return;
1860         if (so->so_is_open_owner)
1861                 release_last_closed_stateid(openowner(so));
1862         so->so_seqid++;
1863         return;
1864 }
1865
1866 static void
1867 gen_sessionid(struct nfsd4_session *ses)
1868 {
1869         struct nfs4_client *clp = ses->se_client;
1870         struct nfsd4_sessionid *sid;
1871
1872         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1873         sid->clientid = clp->cl_clientid;
1874         sid->sequence = current_sessionid++;
1875         sid->reserved = 0;
1876 }
1877
1878 /*
1879  * The protocol defines ca_maxresponssize_cached to include the size of
1880  * the rpc header, but all we need to cache is the data starting after
1881  * the end of the initial SEQUENCE operation--the rest we regenerate
1882  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1883  * value that is the number of bytes in our cache plus a few additional
1884  * bytes.  In order to stay on the safe side, and not promise more than
1885  * we can cache, those additional bytes must be the minimum possible: 24
1886  * bytes of rpc header (xid through accept state, with AUTH_NULL
1887  * verifier), 12 for the compound header (with zero-length tag), and 44
1888  * for the SEQUENCE op response:
1889  */
1890 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1891
1892 static void
1893 free_session_slots(struct nfsd4_session *ses)
1894 {
1895         int i;
1896
1897         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1898                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1899                 kfree(ses->se_slots[i]);
1900         }
1901 }
1902
1903 /*
1904  * We don't actually need to cache the rpc and session headers, so we
1905  * can allocate a little less for each slot:
1906  */
1907 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1908 {
1909         u32 size;
1910
1911         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1912                 size = 0;
1913         else
1914                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1915         return size + sizeof(struct nfsd4_slot);
1916 }
1917
1918 /*
1919  * XXX: If we run out of reserved DRC memory we could (up to a point)
1920  * re-negotiate active sessions and reduce their slot usage to make
1921  * room for new connections. For now we just fail the create session.
1922  */
1923 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1924 {
1925         u32 slotsize = slot_bytes(ca);
1926         u32 num = ca->maxreqs;
1927         unsigned long avail, total_avail;
1928         unsigned int scale_factor;
1929
1930         spin_lock(&nfsd_drc_lock);
1931         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1932                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1933         else
1934                 /* We have handed out more space than we chose in
1935                  * set_max_drc() to allow.  That isn't really a
1936                  * problem as long as that doesn't make us think we
1937                  * have lots more due to integer overflow.
1938                  */
1939                 total_avail = 0;
1940         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1941         /*
1942          * Never use more than a fraction of the remaining memory,
1943          * unless it's the only way to give this client a slot.
1944          * The chosen fraction is either 1/8 or 1/number of threads,
1945          * whichever is smaller.  This ensures there are adequate
1946          * slots to support multiple clients per thread.
1947          * Give the client one slot even if that would require
1948          * over-allocation--it is better than failure.
1949          */
1950         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1951
1952         avail = clamp_t(unsigned long, avail, slotsize,
1953                         total_avail/scale_factor);
1954         num = min_t(int, num, avail / slotsize);
1955         num = max_t(int, num, 1);
1956         nfsd_drc_mem_used += num * slotsize;
1957         spin_unlock(&nfsd_drc_lock);
1958
1959         return num;
1960 }
1961
1962 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1963 {
1964         int slotsize = slot_bytes(ca);
1965
1966         spin_lock(&nfsd_drc_lock);
1967         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1968         spin_unlock(&nfsd_drc_lock);
1969 }
1970
1971 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1972                                            struct nfsd4_channel_attrs *battrs)
1973 {
1974         int numslots = fattrs->maxreqs;
1975         int slotsize = slot_bytes(fattrs);
1976         struct nfsd4_session *new;
1977         int i;
1978
1979         BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1980                      > PAGE_SIZE);
1981
1982         new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1983         if (!new)
1984                 return NULL;
1985         /* allocate each struct nfsd4_slot and data cache in one piece */
1986         for (i = 0; i < numslots; i++) {
1987                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1988                 if (!new->se_slots[i])
1989                         goto out_free;
1990         }
1991
1992         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1993         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1994
1995         return new;
1996 out_free:
1997         while (i--)
1998                 kfree(new->se_slots[i]);
1999         kfree(new);
2000         return NULL;
2001 }
2002
2003 static void free_conn(struct nfsd4_conn *c)
2004 {
2005         svc_xprt_put(c->cn_xprt);
2006         kfree(c);
2007 }
2008
2009 static void nfsd4_conn_lost(struct svc_xpt_user *u)
2010 {
2011         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
2012         struct nfs4_client *clp = c->cn_session->se_client;
2013
2014         trace_nfsd_cb_lost(clp);
2015
2016         spin_lock(&clp->cl_lock);
2017         if (!list_empty(&c->cn_persession)) {
2018                 list_del(&c->cn_persession);
2019                 free_conn(c);
2020         }
2021         nfsd4_probe_callback(clp);
2022         spin_unlock(&clp->cl_lock);
2023 }
2024
2025 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
2026 {
2027         struct nfsd4_conn *conn;
2028
2029         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
2030         if (!conn)
2031                 return NULL;
2032         svc_xprt_get(rqstp->rq_xprt);
2033         conn->cn_xprt = rqstp->rq_xprt;
2034         conn->cn_flags = flags;
2035         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
2036         return conn;
2037 }
2038
2039 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2040 {
2041         conn->cn_session = ses;
2042         list_add(&conn->cn_persession, &ses->se_conns);
2043 }
2044
2045 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2046 {
2047         struct nfs4_client *clp = ses->se_client;
2048
2049         spin_lock(&clp->cl_lock);
2050         __nfsd4_hash_conn(conn, ses);
2051         spin_unlock(&clp->cl_lock);
2052 }
2053
2054 static int nfsd4_register_conn(struct nfsd4_conn *conn)
2055 {
2056         conn->cn_xpt_user.callback = nfsd4_conn_lost;
2057         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
2058 }
2059
2060 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
2061 {
2062         int ret;
2063
2064         nfsd4_hash_conn(conn, ses);
2065         ret = nfsd4_register_conn(conn);
2066         if (ret)
2067                 /* oops; xprt is already down: */
2068                 nfsd4_conn_lost(&conn->cn_xpt_user);
2069         /* We may have gained or lost a callback channel: */
2070         nfsd4_probe_callback_sync(ses->se_client);
2071 }
2072
2073 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
2074 {
2075         u32 dir = NFS4_CDFC4_FORE;
2076
2077         if (cses->flags & SESSION4_BACK_CHAN)
2078                 dir |= NFS4_CDFC4_BACK;
2079         return alloc_conn(rqstp, dir);
2080 }
2081
2082 /* must be called under client_lock */
2083 static void nfsd4_del_conns(struct nfsd4_session *s)
2084 {
2085         struct nfs4_client *clp = s->se_client;
2086         struct nfsd4_conn *c;
2087
2088         spin_lock(&clp->cl_lock);
2089         while (!list_empty(&s->se_conns)) {
2090                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
2091                 list_del_init(&c->cn_persession);
2092                 spin_unlock(&clp->cl_lock);
2093
2094                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
2095                 free_conn(c);
2096
2097                 spin_lock(&clp->cl_lock);
2098         }
2099         spin_unlock(&clp->cl_lock);
2100 }
2101
2102 static void __free_session(struct nfsd4_session *ses)
2103 {
2104         free_session_slots(ses);
2105         kfree(ses);
2106 }
2107
2108 static void free_session(struct nfsd4_session *ses)
2109 {
2110         nfsd4_del_conns(ses);
2111         nfsd4_put_drc_mem(&ses->se_fchannel);
2112         __free_session(ses);
2113 }
2114
2115 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
2116 {
2117         int idx;
2118         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2119
2120         new->se_client = clp;
2121         gen_sessionid(new);
2122
2123         INIT_LIST_HEAD(&new->se_conns);
2124
2125         new->se_cb_seq_nr = 1;
2126         new->se_flags = cses->flags;
2127         new->se_cb_prog = cses->callback_prog;
2128         new->se_cb_sec = cses->cb_sec;
2129         atomic_set(&new->se_ref, 0);
2130         idx = hash_sessionid(&new->se_sessionid);
2131         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
2132         spin_lock(&clp->cl_lock);
2133         list_add(&new->se_perclnt, &clp->cl_sessions);
2134         spin_unlock(&clp->cl_lock);
2135
2136         {
2137                 struct sockaddr *sa = svc_addr(rqstp);
2138                 /*
2139                  * This is a little silly; with sessions there's no real
2140                  * use for the callback address.  Use the peer address
2141                  * as a reasonable default for now, but consider fixing
2142                  * the rpc client not to require an address in the
2143                  * future:
2144                  */
2145                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2146                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2147         }
2148 }
2149
2150 /* caller must hold client_lock */
2151 static struct nfsd4_session *
2152 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2153 {
2154         struct nfsd4_session *elem;
2155         int idx;
2156         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2157
2158         lockdep_assert_held(&nn->client_lock);
2159
2160         dump_sessionid(__func__, sessionid);
2161         idx = hash_sessionid(sessionid);
2162         /* Search in the appropriate list */
2163         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2164                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2165                             NFS4_MAX_SESSIONID_LEN)) {
2166                         return elem;
2167                 }
2168         }
2169
2170         dprintk("%s: session not found\n", __func__);
2171         return NULL;
2172 }
2173
2174 static struct nfsd4_session *
2175 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2176                 __be32 *ret)
2177 {
2178         struct nfsd4_session *session;
2179         __be32 status = nfserr_badsession;
2180
2181         session = __find_in_sessionid_hashtbl(sessionid, net);
2182         if (!session)
2183                 goto out;
2184         status = nfsd4_get_session_locked(session);
2185         if (status)
2186                 session = NULL;
2187 out:
2188         *ret = status;
2189         return session;
2190 }
2191
2192 /* caller must hold client_lock */
2193 static void
2194 unhash_session(struct nfsd4_session *ses)
2195 {
2196         struct nfs4_client *clp = ses->se_client;
2197         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2198
2199         lockdep_assert_held(&nn->client_lock);
2200
2201         list_del(&ses->se_hash);
2202         spin_lock(&ses->se_client->cl_lock);
2203         list_del(&ses->se_perclnt);
2204         spin_unlock(&ses->se_client->cl_lock);
2205 }
2206
2207 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2208 static int
2209 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2210 {
2211         /*
2212          * We're assuming the clid was not given out from a boot
2213          * precisely 2^32 (about 136 years) before this one.  That seems
2214          * a safe assumption:
2215          */
2216         if (clid->cl_boot == (u32)nn->boot_time)
2217                 return 0;
2218         trace_nfsd_clid_stale(clid);
2219         return 1;
2220 }
2221
2222 /*
2223  * XXX Should we use a slab cache ?
2224  * This type of memory management is somewhat inefficient, but we use it
2225  * anyway since SETCLIENTID is not a common operation.
2226  */
2227 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2228                                 struct nfsd_net *nn)
2229 {
2230         struct nfs4_client *clp;
2231         int i;
2232
2233         if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2234                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2235                 return NULL;
2236         }
2237         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2238         if (clp == NULL)
2239                 return NULL;
2240         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2241         if (clp->cl_name.data == NULL)
2242                 goto err_no_name;
2243         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2244                                                  sizeof(struct list_head),
2245                                                  GFP_KERNEL);
2246         if (!clp->cl_ownerstr_hashtbl)
2247                 goto err_no_hashtbl;
2248         for (i = 0; i < OWNER_HASH_SIZE; i++)
2249                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2250         INIT_LIST_HEAD(&clp->cl_sessions);
2251         idr_init(&clp->cl_stateids);
2252         atomic_set(&clp->cl_rpc_users, 0);
2253         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2254         clp->cl_state = NFSD4_ACTIVE;
2255         atomic_inc(&nn->nfs4_client_count);
2256         atomic_set(&clp->cl_delegs_in_recall, 0);
2257         INIT_LIST_HEAD(&clp->cl_idhash);
2258         INIT_LIST_HEAD(&clp->cl_openowners);
2259         INIT_LIST_HEAD(&clp->cl_delegations);
2260         INIT_LIST_HEAD(&clp->cl_lru);
2261         INIT_LIST_HEAD(&clp->cl_revoked);
2262 #ifdef CONFIG_NFSD_PNFS
2263         INIT_LIST_HEAD(&clp->cl_lo_states);
2264 #endif
2265         INIT_LIST_HEAD(&clp->async_copies);
2266         spin_lock_init(&clp->async_lock);
2267         spin_lock_init(&clp->cl_lock);
2268         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2269         return clp;
2270 err_no_hashtbl:
2271         kfree(clp->cl_name.data);
2272 err_no_name:
2273         kmem_cache_free(client_slab, clp);
2274         return NULL;
2275 }
2276
2277 static void __free_client(struct kref *k)
2278 {
2279         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2280         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2281
2282         free_svc_cred(&clp->cl_cred);
2283         kfree(clp->cl_ownerstr_hashtbl);
2284         kfree(clp->cl_name.data);
2285         kfree(clp->cl_nii_domain.data);
2286         kfree(clp->cl_nii_name.data);
2287         idr_destroy(&clp->cl_stateids);
2288         kfree(clp->cl_ra);
2289         kmem_cache_free(client_slab, clp);
2290 }
2291
2292 static void drop_client(struct nfs4_client *clp)
2293 {
2294         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2295 }
2296
2297 static void
2298 free_client(struct nfs4_client *clp)
2299 {
2300         while (!list_empty(&clp->cl_sessions)) {
2301                 struct nfsd4_session *ses;
2302                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2303                                 se_perclnt);
2304                 list_del(&ses->se_perclnt);
2305                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2306                 free_session(ses);
2307         }
2308         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2309         if (clp->cl_nfsd_dentry) {
2310                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2311                 clp->cl_nfsd_dentry = NULL;
2312                 wake_up_all(&expiry_wq);
2313         }
2314         drop_client(clp);
2315 }
2316
2317 /* must be called under the client_lock */
2318 static void
2319 unhash_client_locked(struct nfs4_client *clp)
2320 {
2321         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2322         struct nfsd4_session *ses;
2323
2324         lockdep_assert_held(&nn->client_lock);
2325
2326         /* Mark the client as expired! */
2327         clp->cl_time = 0;
2328         /* Make it invisible */
2329         if (!list_empty(&clp->cl_idhash)) {
2330                 list_del_init(&clp->cl_idhash);
2331                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2332                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2333                 else
2334                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2335         }
2336         list_del_init(&clp->cl_lru);
2337         spin_lock(&clp->cl_lock);
2338         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2339                 list_del_init(&ses->se_hash);
2340         spin_unlock(&clp->cl_lock);
2341 }
2342
2343 static void
2344 unhash_client(struct nfs4_client *clp)
2345 {
2346         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2347
2348         spin_lock(&nn->client_lock);
2349         unhash_client_locked(clp);
2350         spin_unlock(&nn->client_lock);
2351 }
2352
2353 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2354 {
2355         if (atomic_read(&clp->cl_rpc_users))
2356                 return nfserr_jukebox;
2357         unhash_client_locked(clp);
2358         return nfs_ok;
2359 }
2360
2361 static void
2362 __destroy_client(struct nfs4_client *clp)
2363 {
2364         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2365         int i;
2366         struct nfs4_openowner *oo;
2367         struct nfs4_delegation *dp;
2368         struct list_head reaplist;
2369
2370         INIT_LIST_HEAD(&reaplist);
2371         spin_lock(&state_lock);
2372         while (!list_empty(&clp->cl_delegations)) {
2373                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2374                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
2375                 list_add(&dp->dl_recall_lru, &reaplist);
2376         }
2377         spin_unlock(&state_lock);
2378         while (!list_empty(&reaplist)) {
2379                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2380                 list_del_init(&dp->dl_recall_lru);
2381                 destroy_unhashed_deleg(dp);
2382         }
2383         while (!list_empty(&clp->cl_revoked)) {
2384                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2385                 list_del_init(&dp->dl_recall_lru);
2386                 nfs4_put_stid(&dp->dl_stid);
2387         }
2388         while (!list_empty(&clp->cl_openowners)) {
2389                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2390                 nfs4_get_stateowner(&oo->oo_owner);
2391                 release_openowner(oo);
2392         }
2393         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2394                 struct nfs4_stateowner *so, *tmp;
2395
2396                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2397                                          so_strhash) {
2398                         /* Should be no openowners at this point */
2399                         WARN_ON_ONCE(so->so_is_open_owner);
2400                         remove_blocked_locks(lockowner(so));
2401                 }
2402         }
2403         nfsd4_return_all_client_layouts(clp);
2404         nfsd4_shutdown_copy(clp);
2405         nfsd4_shutdown_callback(clp);
2406         if (clp->cl_cb_conn.cb_xprt)
2407                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2408         atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2409         nfsd4_dec_courtesy_client_count(nn, clp);
2410         free_client(clp);
2411         wake_up_all(&expiry_wq);
2412 }
2413
2414 static void
2415 destroy_client(struct nfs4_client *clp)
2416 {
2417         unhash_client(clp);
2418         __destroy_client(clp);
2419 }
2420
2421 static void inc_reclaim_complete(struct nfs4_client *clp)
2422 {
2423         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2424
2425         if (!nn->track_reclaim_completes)
2426                 return;
2427         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2428                 return;
2429         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2430                         nn->reclaim_str_hashtbl_size) {
2431                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2432                                 clp->net->ns.inum);
2433                 nfsd4_end_grace(nn);
2434         }
2435 }
2436
2437 static void expire_client(struct nfs4_client *clp)
2438 {
2439         unhash_client(clp);
2440         nfsd4_client_record_remove(clp);
2441         __destroy_client(clp);
2442 }
2443
2444 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2445 {
2446         memcpy(target->cl_verifier.data, source->data,
2447                         sizeof(target->cl_verifier.data));
2448 }
2449
2450 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2451 {
2452         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2453         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2454 }
2455
2456 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2457 {
2458         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2459         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2460                                                                 GFP_KERNEL);
2461         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2462         if ((source->cr_principal && !target->cr_principal) ||
2463             (source->cr_raw_principal && !target->cr_raw_principal) ||
2464             (source->cr_targ_princ && !target->cr_targ_princ))
2465                 return -ENOMEM;
2466
2467         target->cr_flavor = source->cr_flavor;
2468         target->cr_uid = source->cr_uid;
2469         target->cr_gid = source->cr_gid;
2470         target->cr_group_info = source->cr_group_info;
2471         get_group_info(target->cr_group_info);
2472         target->cr_gss_mech = source->cr_gss_mech;
2473         if (source->cr_gss_mech)
2474                 gss_mech_get(source->cr_gss_mech);
2475         return 0;
2476 }
2477
2478 static int
2479 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2480 {
2481         if (o1->len < o2->len)
2482                 return -1;
2483         if (o1->len > o2->len)
2484                 return 1;
2485         return memcmp(o1->data, o2->data, o1->len);
2486 }
2487
2488 static int
2489 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2490 {
2491         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2492 }
2493
2494 static int
2495 same_clid(clientid_t *cl1, clientid_t *cl2)
2496 {
2497         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2498 }
2499
2500 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2501 {
2502         int i;
2503
2504         if (g1->ngroups != g2->ngroups)
2505                 return false;
2506         for (i=0; i<g1->ngroups; i++)
2507                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2508                         return false;
2509         return true;
2510 }
2511
2512 /*
2513  * RFC 3530 language requires clid_inuse be returned when the
2514  * "principal" associated with a requests differs from that previously
2515  * used.  We use uid, gid's, and gss principal string as our best
2516  * approximation.  We also don't want to allow non-gss use of a client
2517  * established using gss: in theory cr_principal should catch that
2518  * change, but in practice cr_principal can be null even in the gss case
2519  * since gssd doesn't always pass down a principal string.
2520  */
2521 static bool is_gss_cred(struct svc_cred *cr)
2522 {
2523         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2524         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2525 }
2526
2527
2528 static bool
2529 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2530 {
2531         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2532                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2533                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2534                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2535                 return false;
2536         /* XXX: check that cr_targ_princ fields match ? */
2537         if (cr1->cr_principal == cr2->cr_principal)
2538                 return true;
2539         if (!cr1->cr_principal || !cr2->cr_principal)
2540                 return false;
2541         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2542 }
2543
2544 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2545 {
2546         struct svc_cred *cr = &rqstp->rq_cred;
2547         u32 service;
2548
2549         if (!cr->cr_gss_mech)
2550                 return false;
2551         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2552         return service == RPC_GSS_SVC_INTEGRITY ||
2553                service == RPC_GSS_SVC_PRIVACY;
2554 }
2555
2556 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2557 {
2558         struct svc_cred *cr = &rqstp->rq_cred;
2559
2560         if (!cl->cl_mach_cred)
2561                 return true;
2562         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2563                 return false;
2564         if (!svc_rqst_integrity_protected(rqstp))
2565                 return false;
2566         if (cl->cl_cred.cr_raw_principal)
2567                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2568                                                 cr->cr_raw_principal);
2569         if (!cr->cr_principal)
2570                 return false;
2571         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2572 }
2573
2574 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2575 {
2576         __be32 verf[2];
2577
2578         /*
2579          * This is opaque to client, so no need to byte-swap. Use
2580          * __force to keep sparse happy
2581          */
2582         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2583         verf[1] = (__force __be32)nn->clverifier_counter++;
2584         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2585 }
2586
2587 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2588 {
2589         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2590         clp->cl_clientid.cl_id = nn->clientid_counter++;
2591         gen_confirm(clp, nn);
2592 }
2593
2594 static struct nfs4_stid *
2595 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2596 {
2597         struct nfs4_stid *ret;
2598
2599         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2600         if (!ret || !ret->sc_type)
2601                 return NULL;
2602         return ret;
2603 }
2604
2605 static struct nfs4_stid *
2606 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t,
2607                      unsigned short typemask, unsigned short ok_states)
2608 {
2609         struct nfs4_stid *s;
2610
2611         spin_lock(&cl->cl_lock);
2612         s = find_stateid_locked(cl, t);
2613         if (s != NULL) {
2614                 if ((s->sc_status & ~ok_states) == 0 &&
2615                     (typemask & s->sc_type))
2616                         refcount_inc(&s->sc_count);
2617                 else
2618                         s = NULL;
2619         }
2620         spin_unlock(&cl->cl_lock);
2621         return s;
2622 }
2623
2624 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2625 {
2626         struct nfsdfs_client *nc;
2627         nc = get_nfsdfs_client(inode);
2628         if (!nc)
2629                 return NULL;
2630         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2631 }
2632
2633 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2634 {
2635         seq_puts(m, "\"");
2636         seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2637         seq_puts(m, "\"");
2638 }
2639
2640 static const char *cb_state2str(int state)
2641 {
2642         switch (state) {
2643         case NFSD4_CB_UP:
2644                 return "UP";
2645         case NFSD4_CB_UNKNOWN:
2646                 return "UNKNOWN";
2647         case NFSD4_CB_DOWN:
2648                 return "DOWN";
2649         case NFSD4_CB_FAULT:
2650                 return "FAULT";
2651         }
2652         return "UNDEFINED";
2653 }
2654
2655 static int client_info_show(struct seq_file *m, void *v)
2656 {
2657         struct inode *inode = file_inode(m->file);
2658         struct nfs4_client *clp;
2659         u64 clid;
2660
2661         clp = get_nfsdfs_clp(inode);
2662         if (!clp)
2663                 return -ENXIO;
2664         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2665         seq_printf(m, "clientid: 0x%llx\n", clid);
2666         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2667
2668         if (clp->cl_state == NFSD4_COURTESY)
2669                 seq_puts(m, "status: courtesy\n");
2670         else if (clp->cl_state == NFSD4_EXPIRABLE)
2671                 seq_puts(m, "status: expirable\n");
2672         else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2673                 seq_puts(m, "status: confirmed\n");
2674         else
2675                 seq_puts(m, "status: unconfirmed\n");
2676         seq_printf(m, "seconds from last renew: %lld\n",
2677                 ktime_get_boottime_seconds() - clp->cl_time);
2678         seq_puts(m, "name: ");
2679         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2680         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2681         if (clp->cl_nii_domain.data) {
2682                 seq_puts(m, "Implementation domain: ");
2683                 seq_quote_mem(m, clp->cl_nii_domain.data,
2684                                         clp->cl_nii_domain.len);
2685                 seq_puts(m, "\nImplementation name: ");
2686                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2687                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2688                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2689         }
2690         seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2691         seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2692         seq_printf(m, "admin-revoked states: %d\n",
2693                    atomic_read(&clp->cl_admin_revoked));
2694         drop_client(clp);
2695
2696         return 0;
2697 }
2698
2699 DEFINE_SHOW_ATTRIBUTE(client_info);
2700
2701 static void *states_start(struct seq_file *s, loff_t *pos)
2702         __acquires(&clp->cl_lock)
2703 {
2704         struct nfs4_client *clp = s->private;
2705         unsigned long id = *pos;
2706         void *ret;
2707
2708         spin_lock(&clp->cl_lock);
2709         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2710         *pos = id;
2711         return ret;
2712 }
2713
2714 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2715 {
2716         struct nfs4_client *clp = s->private;
2717         unsigned long id = *pos;
2718         void *ret;
2719
2720         id = *pos;
2721         id++;
2722         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2723         *pos = id;
2724         return ret;
2725 }
2726
2727 static void states_stop(struct seq_file *s, void *v)
2728         __releases(&clp->cl_lock)
2729 {
2730         struct nfs4_client *clp = s->private;
2731
2732         spin_unlock(&clp->cl_lock);
2733 }
2734
2735 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2736 {
2737          seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2738 }
2739
2740 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2741 {
2742         struct inode *inode = file_inode(f->nf_file);
2743
2744         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2745                                         MAJOR(inode->i_sb->s_dev),
2746                                          MINOR(inode->i_sb->s_dev),
2747                                          inode->i_ino);
2748 }
2749
2750 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2751 {
2752         seq_puts(s, "owner: ");
2753         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2754 }
2755
2756 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2757 {
2758         seq_printf(s, "0x%.8x", stid->si_generation);
2759         seq_printf(s, "%12phN", &stid->si_opaque);
2760 }
2761
2762 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2763 {
2764         struct nfs4_ol_stateid *ols;
2765         struct nfs4_file *nf;
2766         struct nfsd_file *file;
2767         struct nfs4_stateowner *oo;
2768         unsigned int access, deny;
2769
2770         ols = openlockstateid(st);
2771         oo = ols->st_stateowner;
2772         nf = st->sc_file;
2773
2774         seq_puts(s, "- ");
2775         nfs4_show_stateid(s, &st->sc_stateid);
2776         seq_puts(s, ": { type: open, ");
2777
2778         access = bmap_to_share_mode(ols->st_access_bmap);
2779         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2780
2781         seq_printf(s, "access: %s%s, ",
2782                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2783                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2784         seq_printf(s, "deny: %s%s, ",
2785                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2786                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2787
2788         spin_lock(&nf->fi_lock);
2789         file = find_any_file_locked(nf);
2790         if (file) {
2791                 nfs4_show_superblock(s, file);
2792                 seq_puts(s, ", ");
2793                 nfs4_show_fname(s, file);
2794                 seq_puts(s, ", ");
2795         }
2796         spin_unlock(&nf->fi_lock);
2797         nfs4_show_owner(s, oo);
2798         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2799                 seq_puts(s, ", admin-revoked");
2800         seq_puts(s, " }\n");
2801         return 0;
2802 }
2803
2804 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2805 {
2806         struct nfs4_ol_stateid *ols;
2807         struct nfs4_file *nf;
2808         struct nfsd_file *file;
2809         struct nfs4_stateowner *oo;
2810
2811         ols = openlockstateid(st);
2812         oo = ols->st_stateowner;
2813         nf = st->sc_file;
2814
2815         seq_puts(s, "- ");
2816         nfs4_show_stateid(s, &st->sc_stateid);
2817         seq_puts(s, ": { type: lock, ");
2818
2819         spin_lock(&nf->fi_lock);
2820         file = find_any_file_locked(nf);
2821         if (file) {
2822                 /*
2823                  * Note: a lock stateid isn't really the same thing as a lock,
2824                  * it's the locking state held by one owner on a file, and there
2825                  * may be multiple (or no) lock ranges associated with it.
2826                  * (Same for the matter is true of open stateids.)
2827                  */
2828
2829                 nfs4_show_superblock(s, file);
2830                 /* XXX: open stateid? */
2831                 seq_puts(s, ", ");
2832                 nfs4_show_fname(s, file);
2833                 seq_puts(s, ", ");
2834         }
2835         nfs4_show_owner(s, oo);
2836         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2837                 seq_puts(s, ", admin-revoked");
2838         seq_puts(s, " }\n");
2839         spin_unlock(&nf->fi_lock);
2840         return 0;
2841 }
2842
2843 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2844 {
2845         struct nfs4_delegation *ds;
2846         struct nfs4_file *nf;
2847         struct nfsd_file *file;
2848
2849         ds = delegstateid(st);
2850         nf = st->sc_file;
2851
2852         seq_puts(s, "- ");
2853         nfs4_show_stateid(s, &st->sc_stateid);
2854         seq_puts(s, ": { type: deleg, ");
2855
2856         seq_printf(s, "access: %s",
2857                    ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2858
2859         /* XXX: lease time, whether it's being recalled. */
2860
2861         spin_lock(&nf->fi_lock);
2862         file = nf->fi_deleg_file;
2863         if (file) {
2864                 seq_puts(s, ", ");
2865                 nfs4_show_superblock(s, file);
2866                 seq_puts(s, ", ");
2867                 nfs4_show_fname(s, file);
2868         }
2869         spin_unlock(&nf->fi_lock);
2870         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2871                 seq_puts(s, ", admin-revoked");
2872         seq_puts(s, " }\n");
2873         return 0;
2874 }
2875
2876 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2877 {
2878         struct nfs4_layout_stateid *ls;
2879         struct nfsd_file *file;
2880
2881         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2882
2883         seq_puts(s, "- ");
2884         nfs4_show_stateid(s, &st->sc_stateid);
2885         seq_puts(s, ": { type: layout");
2886
2887         /* XXX: What else would be useful? */
2888
2889         spin_lock(&ls->ls_stid.sc_file->fi_lock);
2890         file = ls->ls_file;
2891         if (file) {
2892                 seq_puts(s, ", ");
2893                 nfs4_show_superblock(s, file);
2894                 seq_puts(s, ", ");
2895                 nfs4_show_fname(s, file);
2896         }
2897         spin_unlock(&ls->ls_stid.sc_file->fi_lock);
2898         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2899                 seq_puts(s, ", admin-revoked");
2900         seq_puts(s, " }\n");
2901
2902         return 0;
2903 }
2904
2905 static int states_show(struct seq_file *s, void *v)
2906 {
2907         struct nfs4_stid *st = v;
2908
2909         switch (st->sc_type) {
2910         case SC_TYPE_OPEN:
2911                 return nfs4_show_open(s, st);
2912         case SC_TYPE_LOCK:
2913                 return nfs4_show_lock(s, st);
2914         case SC_TYPE_DELEG:
2915                 return nfs4_show_deleg(s, st);
2916         case SC_TYPE_LAYOUT:
2917                 return nfs4_show_layout(s, st);
2918         default:
2919                 return 0; /* XXX: or SEQ_SKIP? */
2920         }
2921         /* XXX: copy stateids? */
2922 }
2923
2924 static struct seq_operations states_seq_ops = {
2925         .start = states_start,
2926         .next = states_next,
2927         .stop = states_stop,
2928         .show = states_show
2929 };
2930
2931 static int client_states_open(struct inode *inode, struct file *file)
2932 {
2933         struct seq_file *s;
2934         struct nfs4_client *clp;
2935         int ret;
2936
2937         clp = get_nfsdfs_clp(inode);
2938         if (!clp)
2939                 return -ENXIO;
2940
2941         ret = seq_open(file, &states_seq_ops);
2942         if (ret)
2943                 return ret;
2944         s = file->private_data;
2945         s->private = clp;
2946         return 0;
2947 }
2948
2949 static int client_opens_release(struct inode *inode, struct file *file)
2950 {
2951         struct seq_file *m = file->private_data;
2952         struct nfs4_client *clp = m->private;
2953
2954         /* XXX: alternatively, we could get/drop in seq start/stop */
2955         drop_client(clp);
2956         return seq_release(inode, file);
2957 }
2958
2959 static const struct file_operations client_states_fops = {
2960         .open           = client_states_open,
2961         .read           = seq_read,
2962         .llseek         = seq_lseek,
2963         .release        = client_opens_release,
2964 };
2965
2966 /*
2967  * Normally we refuse to destroy clients that are in use, but here the
2968  * administrator is telling us to just do it.  We also want to wait
2969  * so the caller has a guarantee that the client's locks are gone by
2970  * the time the write returns:
2971  */
2972 static void force_expire_client(struct nfs4_client *clp)
2973 {
2974         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2975         bool already_expired;
2976
2977         trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2978
2979         spin_lock(&nn->client_lock);
2980         clp->cl_time = 0;
2981         spin_unlock(&nn->client_lock);
2982
2983         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2984         spin_lock(&nn->client_lock);
2985         already_expired = list_empty(&clp->cl_lru);
2986         if (!already_expired)
2987                 unhash_client_locked(clp);
2988         spin_unlock(&nn->client_lock);
2989
2990         if (!already_expired)
2991                 expire_client(clp);
2992         else
2993                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2994 }
2995
2996 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2997                                    size_t size, loff_t *pos)
2998 {
2999         char *data;
3000         struct nfs4_client *clp;
3001
3002         data = simple_transaction_get(file, buf, size);
3003         if (IS_ERR(data))
3004                 return PTR_ERR(data);
3005         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
3006                 return -EINVAL;
3007         clp = get_nfsdfs_clp(file_inode(file));
3008         if (!clp)
3009                 return -ENXIO;
3010         force_expire_client(clp);
3011         drop_client(clp);
3012         return 7;
3013 }
3014
3015 static const struct file_operations client_ctl_fops = {
3016         .write          = client_ctl_write,
3017         .release        = simple_transaction_release,
3018 };
3019
3020 static const struct tree_descr client_files[] = {
3021         [0] = {"info", &client_info_fops, S_IRUSR},
3022         [1] = {"states", &client_states_fops, S_IRUSR},
3023         [2] = {"ctl", &client_ctl_fops, S_IWUSR},
3024         [3] = {""},
3025 };
3026
3027 static int
3028 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
3029                                 struct rpc_task *task)
3030 {
3031         trace_nfsd_cb_recall_any_done(cb, task);
3032         switch (task->tk_status) {
3033         case -NFS4ERR_DELAY:
3034                 rpc_delay(task, 2 * HZ);
3035                 return 0;
3036         default:
3037                 return 1;
3038         }
3039 }
3040
3041 static void
3042 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
3043 {
3044         struct nfs4_client *clp = cb->cb_clp;
3045         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3046
3047         spin_lock(&nn->client_lock);
3048         clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
3049         put_client_renew_locked(clp);
3050         spin_unlock(&nn->client_lock);
3051 }
3052
3053 static int
3054 nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3055 {
3056         struct nfs4_cb_fattr *ncf =
3057                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3058
3059         ncf->ncf_cb_status = task->tk_status;
3060         switch (task->tk_status) {
3061         case -NFS4ERR_DELAY:
3062                 rpc_delay(task, 2 * HZ);
3063                 return 0;
3064         default:
3065                 return 1;
3066         }
3067 }
3068
3069 static void
3070 nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3071 {
3072         struct nfs4_cb_fattr *ncf =
3073                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3074         struct nfs4_delegation *dp =
3075                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3076
3077         nfs4_put_stid(&dp->dl_stid);
3078         clear_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags);
3079         wake_up_bit(&ncf->ncf_cb_flags, CB_GETATTR_BUSY);
3080 }
3081
3082 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3083         .done           = nfsd4_cb_recall_any_done,
3084         .release        = nfsd4_cb_recall_any_release,
3085 };
3086
3087 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3088         .done           = nfsd4_cb_getattr_done,
3089         .release        = nfsd4_cb_getattr_release,
3090 };
3091
3092 static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3093 {
3094         struct nfs4_delegation *dp =
3095                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3096
3097         if (test_and_set_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags))
3098                 return;
3099         /* set to proper status when nfsd4_cb_getattr_done runs */
3100         ncf->ncf_cb_status = NFS4ERR_IO;
3101
3102         refcount_inc(&dp->dl_stid.sc_count);
3103         nfsd4_run_cb(&ncf->ncf_getattr);
3104 }
3105
3106 static struct nfs4_client *create_client(struct xdr_netobj name,
3107                 struct svc_rqst *rqstp, nfs4_verifier *verf)
3108 {
3109         struct nfs4_client *clp;
3110         struct sockaddr *sa = svc_addr(rqstp);
3111         int ret;
3112         struct net *net = SVC_NET(rqstp);
3113         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3114         struct dentry *dentries[ARRAY_SIZE(client_files)];
3115
3116         clp = alloc_client(name, nn);
3117         if (clp == NULL)
3118                 return NULL;
3119
3120         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3121         if (ret) {
3122                 free_client(clp);
3123                 return NULL;
3124         }
3125         gen_clid(clp, nn);
3126         kref_init(&clp->cl_nfsdfs.cl_ref);
3127         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3128         clp->cl_time = ktime_get_boottime_seconds();
3129         clear_bit(0, &clp->cl_cb_slot_busy);
3130         copy_verf(clp, verf);
3131         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3132         clp->cl_cb_session = NULL;
3133         clp->net = net;
3134         clp->cl_nfsd_dentry = nfsd_client_mkdir(
3135                 nn, &clp->cl_nfsdfs,
3136                 clp->cl_clientid.cl_id - nn->clientid_base,
3137                 client_files, dentries);
3138         clp->cl_nfsd_info_dentry = dentries[0];
3139         if (!clp->cl_nfsd_dentry) {
3140                 free_client(clp);
3141                 return NULL;
3142         }
3143         clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3144         if (!clp->cl_ra) {
3145                 free_client(clp);
3146                 return NULL;
3147         }
3148         clp->cl_ra_time = 0;
3149         nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3150                         NFSPROC4_CLNT_CB_RECALL_ANY);
3151         return clp;
3152 }
3153
3154 static void
3155 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3156 {
3157         struct rb_node **new = &(root->rb_node), *parent = NULL;
3158         struct nfs4_client *clp;
3159
3160         while (*new) {
3161                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3162                 parent = *new;
3163
3164                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3165                         new = &((*new)->rb_left);
3166                 else
3167                         new = &((*new)->rb_right);
3168         }
3169
3170         rb_link_node(&new_clp->cl_namenode, parent, new);
3171         rb_insert_color(&new_clp->cl_namenode, root);
3172 }
3173
3174 static struct nfs4_client *
3175 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3176 {
3177         int cmp;
3178         struct rb_node *node = root->rb_node;
3179         struct nfs4_client *clp;
3180
3181         while (node) {
3182                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
3183                 cmp = compare_blob(&clp->cl_name, name);
3184                 if (cmp > 0)
3185                         node = node->rb_left;
3186                 else if (cmp < 0)
3187                         node = node->rb_right;
3188                 else
3189                         return clp;
3190         }
3191         return NULL;
3192 }
3193
3194 static void
3195 add_to_unconfirmed(struct nfs4_client *clp)
3196 {
3197         unsigned int idhashval;
3198         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3199
3200         lockdep_assert_held(&nn->client_lock);
3201
3202         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3203         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3204         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3205         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3206         renew_client_locked(clp);
3207 }
3208
3209 static void
3210 move_to_confirmed(struct nfs4_client *clp)
3211 {
3212         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3213         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3214
3215         lockdep_assert_held(&nn->client_lock);
3216
3217         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3218         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3219         add_clp_to_name_tree(clp, &nn->conf_name_tree);
3220         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3221         trace_nfsd_clid_confirmed(&clp->cl_clientid);
3222         renew_client_locked(clp);
3223 }
3224
3225 static struct nfs4_client *
3226 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3227 {
3228         struct nfs4_client *clp;
3229         unsigned int idhashval = clientid_hashval(clid->cl_id);
3230
3231         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3232                 if (same_clid(&clp->cl_clientid, clid)) {
3233                         if ((bool)clp->cl_minorversion != sessions)
3234                                 return NULL;
3235                         renew_client_locked(clp);
3236                         return clp;
3237                 }
3238         }
3239         return NULL;
3240 }
3241
3242 static struct nfs4_client *
3243 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3244 {
3245         struct list_head *tbl = nn->conf_id_hashtbl;
3246
3247         lockdep_assert_held(&nn->client_lock);
3248         return find_client_in_id_table(tbl, clid, sessions);
3249 }
3250
3251 static struct nfs4_client *
3252 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3253 {
3254         struct list_head *tbl = nn->unconf_id_hashtbl;
3255
3256         lockdep_assert_held(&nn->client_lock);
3257         return find_client_in_id_table(tbl, clid, sessions);
3258 }
3259
3260 static bool clp_used_exchangeid(struct nfs4_client *clp)
3261 {
3262         return clp->cl_exchange_flags != 0;
3263 }
3264
3265 static struct nfs4_client *
3266 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3267 {
3268         lockdep_assert_held(&nn->client_lock);
3269         return find_clp_in_name_tree(name, &nn->conf_name_tree);
3270 }
3271
3272 static struct nfs4_client *
3273 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3274 {
3275         lockdep_assert_held(&nn->client_lock);
3276         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3277 }
3278
3279 static void
3280 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3281 {
3282         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3283         struct sockaddr *sa = svc_addr(rqstp);
3284         u32 scopeid = rpc_get_scope_id(sa);
3285         unsigned short expected_family;
3286
3287         /* Currently, we only support tcp and tcp6 for the callback channel */
3288         if (se->se_callback_netid_len == 3 &&
3289             !memcmp(se->se_callback_netid_val, "tcp", 3))
3290                 expected_family = AF_INET;
3291         else if (se->se_callback_netid_len == 4 &&
3292                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
3293                 expected_family = AF_INET6;
3294         else
3295                 goto out_err;
3296
3297         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3298                                             se->se_callback_addr_len,
3299                                             (struct sockaddr *)&conn->cb_addr,
3300                                             sizeof(conn->cb_addr));
3301
3302         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3303                 goto out_err;
3304
3305         if (conn->cb_addr.ss_family == AF_INET6)
3306                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3307
3308         conn->cb_prog = se->se_callback_prog;
3309         conn->cb_ident = se->se_callback_ident;
3310         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3311         trace_nfsd_cb_args(clp, conn);
3312         return;
3313 out_err:
3314         conn->cb_addr.ss_family = AF_UNSPEC;
3315         conn->cb_addrlen = 0;
3316         trace_nfsd_cb_nodelegs(clp);
3317         return;
3318 }
3319
3320 /*
3321  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3322  */
3323 static void
3324 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3325 {
3326         struct xdr_buf *buf = resp->xdr->buf;
3327         struct nfsd4_slot *slot = resp->cstate.slot;
3328         unsigned int base;
3329
3330         dprintk("--> %s slot %p\n", __func__, slot);
3331
3332         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3333         slot->sl_opcnt = resp->opcnt;
3334         slot->sl_status = resp->cstate.status;
3335         free_svc_cred(&slot->sl_cred);
3336         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3337
3338         if (!nfsd4_cache_this(resp)) {
3339                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3340                 return;
3341         }
3342         slot->sl_flags |= NFSD4_SLOT_CACHED;
3343
3344         base = resp->cstate.data_offset;
3345         slot->sl_datalen = buf->len - base;
3346         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3347                 WARN(1, "%s: sessions DRC could not cache compound\n",
3348                      __func__);
3349         return;
3350 }
3351
3352 /*
3353  * Encode the replay sequence operation from the slot values.
3354  * If cachethis is FALSE encode the uncached rep error on the next
3355  * operation which sets resp->p and increments resp->opcnt for
3356  * nfs4svc_encode_compoundres.
3357  *
3358  */
3359 static __be32
3360 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3361                           struct nfsd4_compoundres *resp)
3362 {
3363         struct nfsd4_op *op;
3364         struct nfsd4_slot *slot = resp->cstate.slot;
3365
3366         /* Encode the replayed sequence operation */
3367         op = &args->ops[resp->opcnt - 1];
3368         nfsd4_encode_operation(resp, op);
3369
3370         if (slot->sl_flags & NFSD4_SLOT_CACHED)
3371                 return op->status;
3372         if (args->opcnt == 1) {
3373                 /*
3374                  * The original operation wasn't a solo sequence--we
3375                  * always cache those--so this retry must not match the
3376                  * original:
3377                  */
3378                 op->status = nfserr_seq_false_retry;
3379         } else {
3380                 op = &args->ops[resp->opcnt++];
3381                 op->status = nfserr_retry_uncached_rep;
3382                 nfsd4_encode_operation(resp, op);
3383         }
3384         return op->status;
3385 }
3386
3387 /*
3388  * The sequence operation is not cached because we can use the slot and
3389  * session values.
3390  */
3391 static __be32
3392 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3393                          struct nfsd4_sequence *seq)
3394 {
3395         struct nfsd4_slot *slot = resp->cstate.slot;
3396         struct xdr_stream *xdr = resp->xdr;
3397         __be32 *p;
3398         __be32 status;
3399
3400         dprintk("--> %s slot %p\n", __func__, slot);
3401
3402         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3403         if (status)
3404                 return status;
3405
3406         p = xdr_reserve_space(xdr, slot->sl_datalen);
3407         if (!p) {
3408                 WARN_ON_ONCE(1);
3409                 return nfserr_serverfault;
3410         }
3411         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3412         xdr_commit_encode(xdr);
3413
3414         resp->opcnt = slot->sl_opcnt;
3415         return slot->sl_status;
3416 }
3417
3418 /*
3419  * Set the exchange_id flags returned by the server.
3420  */
3421 static void
3422 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3423 {
3424 #ifdef CONFIG_NFSD_PNFS
3425         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3426 #else
3427         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3428 #endif
3429
3430         /* Referrals are supported, Migration is not. */
3431         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3432
3433         /* set the wire flags to return to client. */
3434         clid->flags = new->cl_exchange_flags;
3435 }
3436
3437 static bool client_has_openowners(struct nfs4_client *clp)
3438 {
3439         struct nfs4_openowner *oo;
3440
3441         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3442                 if (!list_empty(&oo->oo_owner.so_stateids))
3443                         return true;
3444         }
3445         return false;
3446 }
3447
3448 static bool client_has_state(struct nfs4_client *clp)
3449 {
3450         return client_has_openowners(clp)
3451 #ifdef CONFIG_NFSD_PNFS
3452                 || !list_empty(&clp->cl_lo_states)
3453 #endif
3454                 || !list_empty(&clp->cl_delegations)
3455                 || !list_empty(&clp->cl_sessions)
3456                 || !list_empty(&clp->async_copies);
3457 }
3458
3459 static __be32 copy_impl_id(struct nfs4_client *clp,
3460                                 struct nfsd4_exchange_id *exid)
3461 {
3462         if (!exid->nii_domain.data)
3463                 return 0;
3464         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3465         if (!clp->cl_nii_domain.data)
3466                 return nfserr_jukebox;
3467         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3468         if (!clp->cl_nii_name.data)
3469                 return nfserr_jukebox;
3470         clp->cl_nii_time = exid->nii_time;
3471         return 0;
3472 }
3473
3474 __be32
3475 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3476                 union nfsd4_op_u *u)
3477 {
3478         struct nfsd4_exchange_id *exid = &u->exchange_id;
3479         struct nfs4_client *conf, *new;
3480         struct nfs4_client *unconf = NULL;
3481         __be32 status;
3482         char                    addr_str[INET6_ADDRSTRLEN];
3483         nfs4_verifier           verf = exid->verifier;
3484         struct sockaddr         *sa = svc_addr(rqstp);
3485         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3486         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3487
3488         rpc_ntop(sa, addr_str, sizeof(addr_str));
3489         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3490                 "ip_addr=%s flags %x, spa_how %u\n",
3491                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3492                 addr_str, exid->flags, exid->spa_how);
3493
3494         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3495                 return nfserr_inval;
3496
3497         new = create_client(exid->clname, rqstp, &verf);
3498         if (new == NULL)
3499                 return nfserr_jukebox;
3500         status = copy_impl_id(new, exid);
3501         if (status)
3502                 goto out_nolock;
3503
3504         switch (exid->spa_how) {
3505         case SP4_MACH_CRED:
3506                 exid->spo_must_enforce[0] = 0;
3507                 exid->spo_must_enforce[1] = (
3508                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3509                         1 << (OP_EXCHANGE_ID - 32) |
3510                         1 << (OP_CREATE_SESSION - 32) |
3511                         1 << (OP_DESTROY_SESSION - 32) |
3512                         1 << (OP_DESTROY_CLIENTID - 32));
3513
3514                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3515                                         1 << (OP_OPEN_DOWNGRADE) |
3516                                         1 << (OP_LOCKU) |
3517                                         1 << (OP_DELEGRETURN));
3518
3519                 exid->spo_must_allow[1] &= (
3520                                         1 << (OP_TEST_STATEID - 32) |
3521                                         1 << (OP_FREE_STATEID - 32));
3522                 if (!svc_rqst_integrity_protected(rqstp)) {
3523                         status = nfserr_inval;
3524                         goto out_nolock;
3525                 }
3526                 /*
3527                  * Sometimes userspace doesn't give us a principal.
3528                  * Which is a bug, really.  Anyway, we can't enforce
3529                  * MACH_CRED in that case, better to give up now:
3530                  */
3531                 if (!new->cl_cred.cr_principal &&
3532                                         !new->cl_cred.cr_raw_principal) {
3533                         status = nfserr_serverfault;
3534                         goto out_nolock;
3535                 }
3536                 new->cl_mach_cred = true;
3537                 break;
3538         case SP4_NONE:
3539                 break;
3540         default:                                /* checked by xdr code */
3541                 WARN_ON_ONCE(1);
3542                 fallthrough;
3543         case SP4_SSV:
3544                 status = nfserr_encr_alg_unsupp;
3545                 goto out_nolock;
3546         }
3547
3548         /* Cases below refer to rfc 5661 section 18.35.4: */
3549         spin_lock(&nn->client_lock);
3550         conf = find_confirmed_client_by_name(&exid->clname, nn);
3551         if (conf) {
3552                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3553                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3554
3555                 if (update) {
3556                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3557                                 status = nfserr_inval;
3558                                 goto out;
3559                         }
3560                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3561                                 status = nfserr_wrong_cred;
3562                                 goto out;
3563                         }
3564                         if (!creds_match) { /* case 9 */
3565                                 status = nfserr_perm;
3566                                 goto out;
3567                         }
3568                         if (!verfs_match) { /* case 8 */
3569                                 status = nfserr_not_same;
3570                                 goto out;
3571                         }
3572                         /* case 6 */
3573                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3574                         trace_nfsd_clid_confirmed_r(conf);
3575                         goto out_copy;
3576                 }
3577                 if (!creds_match) { /* case 3 */
3578                         if (client_has_state(conf)) {
3579                                 status = nfserr_clid_inuse;
3580                                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3581                                 goto out;
3582                         }
3583                         goto out_new;
3584                 }
3585                 if (verfs_match) { /* case 2 */
3586                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3587                         trace_nfsd_clid_confirmed_r(conf);
3588                         goto out_copy;
3589                 }
3590                 /* case 5, client reboot */
3591                 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3592                 conf = NULL;
3593                 goto out_new;
3594         }
3595
3596         if (update) { /* case 7 */
3597                 status = nfserr_noent;
3598                 goto out;
3599         }
3600
3601         unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3602         if (unconf) /* case 4, possible retry or client restart */
3603                 unhash_client_locked(unconf);
3604
3605         /* case 1, new owner ID */
3606         trace_nfsd_clid_fresh(new);
3607
3608 out_new:
3609         if (conf) {
3610                 status = mark_client_expired_locked(conf);
3611                 if (status)
3612                         goto out;
3613                 trace_nfsd_clid_replaced(&conf->cl_clientid);
3614         }
3615         new->cl_minorversion = cstate->minorversion;
3616         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3617         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3618
3619         /* Contrived initial CREATE_SESSION response */
3620         new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3621
3622         add_to_unconfirmed(new);
3623         swap(new, conf);
3624 out_copy:
3625         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3626         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3627
3628         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3629         nfsd4_set_ex_flags(conf, exid);
3630
3631         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3632                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3633         status = nfs_ok;
3634
3635 out:
3636         spin_unlock(&nn->client_lock);
3637 out_nolock:
3638         if (new)
3639                 expire_client(new);
3640         if (unconf) {
3641                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3642                 expire_client(unconf);
3643         }
3644         return status;
3645 }
3646
3647 static __be32
3648 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3649 {
3650         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3651                 slot_seqid);
3652
3653         /* The slot is in use, and no response has been sent. */
3654         if (slot_inuse) {
3655                 if (seqid == slot_seqid)
3656                         return nfserr_jukebox;
3657                 else
3658                         return nfserr_seq_misordered;
3659         }
3660         /* Note unsigned 32-bit arithmetic handles wraparound: */
3661         if (likely(seqid == slot_seqid + 1))
3662                 return nfs_ok;
3663         if (seqid == slot_seqid)
3664                 return nfserr_replay_cache;
3665         return nfserr_seq_misordered;
3666 }
3667
3668 /*
3669  * Cache the create session result into the create session single DRC
3670  * slot cache by saving the xdr structure. sl_seqid has been set.
3671  * Do this for solo or embedded create session operations.
3672  */
3673 static void
3674 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3675                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3676 {
3677         slot->sl_status = nfserr;
3678         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3679 }
3680
3681 static __be32
3682 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3683                             struct nfsd4_clid_slot *slot)
3684 {
3685         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3686         return slot->sl_status;
3687 }
3688
3689 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3690                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3691                         1 +     /* MIN tag is length with zero, only length */ \
3692                         3 +     /* version, opcount, opcode */ \
3693                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3694                                 /* seqid, slotID, slotID, cache */ \
3695                         4 ) * sizeof(__be32))
3696
3697 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3698                         2 +     /* verifier: AUTH_NULL, length 0 */\
3699                         1 +     /* status */ \
3700                         1 +     /* MIN tag is length with zero, only length */ \
3701                         3 +     /* opcount, opcode, opstatus*/ \
3702                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3703                                 /* seqid, slotID, slotID, slotID, status */ \
3704                         5 ) * sizeof(__be32))
3705
3706 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3707 {
3708         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3709
3710         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3711                 return nfserr_toosmall;
3712         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3713                 return nfserr_toosmall;
3714         ca->headerpadsz = 0;
3715         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3716         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3717         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3718         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3719                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3720         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3721         /*
3722          * Note decreasing slot size below client's request may make it
3723          * difficult for client to function correctly, whereas
3724          * decreasing the number of slots will (just?) affect
3725          * performance.  When short on memory we therefore prefer to
3726          * decrease number of slots instead of their size.  Clients that
3727          * request larger slots than they need will get poor results:
3728          * Note that we always allow at least one slot, because our
3729          * accounting is soft and provides no guarantees either way.
3730          */
3731         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3732
3733         return nfs_ok;
3734 }
3735
3736 /*
3737  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3738  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3739  */
3740 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3741         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3742
3743 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3744         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3745
3746 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3747                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3748 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3749                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3750                                  sizeof(__be32))
3751
3752 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3753 {
3754         ca->headerpadsz = 0;
3755
3756         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3757                 return nfserr_toosmall;
3758         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3759                 return nfserr_toosmall;
3760         ca->maxresp_cached = 0;
3761         if (ca->maxops < 2)
3762                 return nfserr_toosmall;
3763
3764         return nfs_ok;
3765 }
3766
3767 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3768 {
3769         switch (cbs->flavor) {
3770         case RPC_AUTH_NULL:
3771         case RPC_AUTH_UNIX:
3772                 return nfs_ok;
3773         default:
3774                 /*
3775                  * GSS case: the spec doesn't allow us to return this
3776                  * error.  But it also doesn't allow us not to support
3777                  * GSS.
3778                  * I'd rather this fail hard than return some error the
3779                  * client might think it can already handle:
3780                  */
3781                 return nfserr_encr_alg_unsupp;
3782         }
3783 }
3784
3785 __be32
3786 nfsd4_create_session(struct svc_rqst *rqstp,
3787                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3788 {
3789         struct nfsd4_create_session *cr_ses = &u->create_session;
3790         struct sockaddr *sa = svc_addr(rqstp);
3791         struct nfs4_client *conf, *unconf;
3792         struct nfsd4_clid_slot *cs_slot;
3793         struct nfs4_client *old = NULL;
3794         struct nfsd4_session *new;
3795         struct nfsd4_conn *conn;
3796         __be32 status = 0;
3797         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3798
3799         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3800                 return nfserr_inval;
3801         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3802         if (status)
3803                 return status;
3804         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3805         if (status)
3806                 return status;
3807         status = check_backchannel_attrs(&cr_ses->back_channel);
3808         if (status)
3809                 goto out_release_drc_mem;
3810         status = nfserr_jukebox;
3811         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3812         if (!new)
3813                 goto out_release_drc_mem;
3814         conn = alloc_conn_from_crses(rqstp, cr_ses);
3815         if (!conn)
3816                 goto out_free_session;
3817
3818         spin_lock(&nn->client_lock);
3819
3820         /* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3821         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3822         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3823         if (!conf && !unconf) {
3824                 status = nfserr_stale_clientid;
3825                 goto out_free_conn;
3826         }
3827
3828         /* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
3829         if (conf)
3830                 cs_slot = &conf->cl_cs_slot;
3831         else
3832                 cs_slot = &unconf->cl_cs_slot;
3833         status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3834         if (status) {
3835                 if (status == nfserr_replay_cache) {
3836                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
3837                         goto out_free_conn;
3838                 }
3839                 goto out_cache_error;
3840         }
3841         cs_slot->sl_seqid++;
3842         cr_ses->seqid = cs_slot->sl_seqid;
3843
3844         /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3845         if (conf) {
3846                 status = nfserr_wrong_cred;
3847                 if (!nfsd4_mach_creds_match(conf, rqstp))
3848                         goto out_cache_error;
3849         } else {
3850                 status = nfserr_clid_inuse;
3851                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3852                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3853                         trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3854                         goto out_cache_error;
3855                 }
3856                 status = nfserr_wrong_cred;
3857                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3858                         goto out_cache_error;
3859                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3860                 if (old) {
3861                         status = mark_client_expired_locked(old);
3862                         if (status) {
3863                                 old = NULL;
3864                                 goto out_cache_error;
3865                         }
3866                         trace_nfsd_clid_replaced(&old->cl_clientid);
3867                 }
3868                 move_to_confirmed(unconf);
3869                 conf = unconf;
3870         }
3871
3872         /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3873         status = nfs_ok;
3874         /* Persistent sessions are not supported */
3875         cr_ses->flags &= ~SESSION4_PERSIST;
3876         /* Upshifting from TCP to RDMA is not supported */
3877         cr_ses->flags &= ~SESSION4_RDMA;
3878
3879         init_session(rqstp, new, conf, cr_ses);
3880         nfsd4_get_session_locked(new);
3881
3882         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3883                NFS4_MAX_SESSIONID_LEN);
3884
3885         /* cache solo and embedded create sessions under the client_lock */
3886         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3887         spin_unlock(&nn->client_lock);
3888         if (conf == unconf)
3889                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3890         /* init connection and backchannel */
3891         nfsd4_init_conn(rqstp, conn, new);
3892         nfsd4_put_session(new);
3893         if (old)
3894                 expire_client(old);
3895         return status;
3896
3897 out_cache_error:
3898         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3899 out_free_conn:
3900         spin_unlock(&nn->client_lock);
3901         free_conn(conn);
3902         if (old)
3903                 expire_client(old);
3904 out_free_session:
3905         __free_session(new);
3906 out_release_drc_mem:
3907         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3908         return status;
3909 }
3910
3911 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3912 {
3913         switch (*dir) {
3914         case NFS4_CDFC4_FORE:
3915         case NFS4_CDFC4_BACK:
3916                 return nfs_ok;
3917         case NFS4_CDFC4_FORE_OR_BOTH:
3918         case NFS4_CDFC4_BACK_OR_BOTH:
3919                 *dir = NFS4_CDFC4_BOTH;
3920                 return nfs_ok;
3921         }
3922         return nfserr_inval;
3923 }
3924
3925 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3926                 struct nfsd4_compound_state *cstate,
3927                 union nfsd4_op_u *u)
3928 {
3929         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3930         struct nfsd4_session *session = cstate->session;
3931         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3932         __be32 status;
3933
3934         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3935         if (status)
3936                 return status;
3937         spin_lock(&nn->client_lock);
3938         session->se_cb_prog = bc->bc_cb_program;
3939         session->se_cb_sec = bc->bc_cb_sec;
3940         spin_unlock(&nn->client_lock);
3941
3942         nfsd4_probe_callback(session->se_client);
3943
3944         return nfs_ok;
3945 }
3946
3947 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3948 {
3949         struct nfsd4_conn *c;
3950
3951         list_for_each_entry(c, &s->se_conns, cn_persession) {
3952                 if (c->cn_xprt == xpt) {
3953                         return c;
3954                 }
3955         }
3956         return NULL;
3957 }
3958
3959 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3960                 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3961 {
3962         struct nfs4_client *clp = session->se_client;
3963         struct svc_xprt *xpt = rqst->rq_xprt;
3964         struct nfsd4_conn *c;
3965         __be32 status;
3966
3967         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3968         spin_lock(&clp->cl_lock);
3969         c = __nfsd4_find_conn(xpt, session);
3970         if (!c)
3971                 status = nfserr_noent;
3972         else if (req == c->cn_flags)
3973                 status = nfs_ok;
3974         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3975                                 c->cn_flags != NFS4_CDFC4_BACK)
3976                 status = nfs_ok;
3977         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3978                                 c->cn_flags != NFS4_CDFC4_FORE)
3979                 status = nfs_ok;
3980         else
3981                 status = nfserr_inval;
3982         spin_unlock(&clp->cl_lock);
3983         if (status == nfs_ok && conn)
3984                 *conn = c;
3985         return status;
3986 }
3987
3988 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3989                      struct nfsd4_compound_state *cstate,
3990                      union nfsd4_op_u *u)
3991 {
3992         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3993         __be32 status;
3994         struct nfsd4_conn *conn;
3995         struct nfsd4_session *session;
3996         struct net *net = SVC_NET(rqstp);
3997         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3998
3999         if (!nfsd4_last_compound_op(rqstp))
4000                 return nfserr_not_only_op;
4001         spin_lock(&nn->client_lock);
4002         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4003         spin_unlock(&nn->client_lock);
4004         if (!session)
4005                 goto out_no_session;
4006         status = nfserr_wrong_cred;
4007         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4008                 goto out;
4009         status = nfsd4_match_existing_connection(rqstp, session,
4010                         bcts->dir, &conn);
4011         if (status == nfs_ok) {
4012                 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4013                                 bcts->dir == NFS4_CDFC4_BACK)
4014                         conn->cn_flags |= NFS4_CDFC4_BACK;
4015                 nfsd4_probe_callback(session->se_client);
4016                 goto out;
4017         }
4018         if (status == nfserr_inval)
4019                 goto out;
4020         status = nfsd4_map_bcts_dir(&bcts->dir);
4021         if (status)
4022                 goto out;
4023         conn = alloc_conn(rqstp, bcts->dir);
4024         status = nfserr_jukebox;
4025         if (!conn)
4026                 goto out;
4027         nfsd4_init_conn(rqstp, conn, session);
4028         status = nfs_ok;
4029 out:
4030         nfsd4_put_session(session);
4031 out_no_session:
4032         return status;
4033 }
4034
4035 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4036 {
4037         if (!cstate->session)
4038                 return false;
4039         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4040 }
4041
4042 __be32
4043 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4044                 union nfsd4_op_u *u)
4045 {
4046         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4047         struct nfsd4_session *ses;
4048         __be32 status;
4049         int ref_held_by_me = 0;
4050         struct net *net = SVC_NET(r);
4051         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4052
4053         status = nfserr_not_only_op;
4054         if (nfsd4_compound_in_session(cstate, sessionid)) {
4055                 if (!nfsd4_last_compound_op(r))
4056                         goto out;
4057                 ref_held_by_me++;
4058         }
4059         dump_sessionid(__func__, sessionid);
4060         spin_lock(&nn->client_lock);
4061         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4062         if (!ses)
4063                 goto out_client_lock;
4064         status = nfserr_wrong_cred;
4065         if (!nfsd4_mach_creds_match(ses->se_client, r))
4066                 goto out_put_session;
4067         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4068         if (status)
4069                 goto out_put_session;
4070         unhash_session(ses);
4071         spin_unlock(&nn->client_lock);
4072
4073         nfsd4_probe_callback_sync(ses->se_client);
4074
4075         spin_lock(&nn->client_lock);
4076         status = nfs_ok;
4077 out_put_session:
4078         nfsd4_put_session_locked(ses);
4079 out_client_lock:
4080         spin_unlock(&nn->client_lock);
4081 out:
4082         return status;
4083 }
4084
4085 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4086 {
4087         struct nfs4_client *clp = ses->se_client;
4088         struct nfsd4_conn *c;
4089         __be32 status = nfs_ok;
4090         int ret;
4091
4092         spin_lock(&clp->cl_lock);
4093         c = __nfsd4_find_conn(new->cn_xprt, ses);
4094         if (c)
4095                 goto out_free;
4096         status = nfserr_conn_not_bound_to_session;
4097         if (clp->cl_mach_cred)
4098                 goto out_free;
4099         __nfsd4_hash_conn(new, ses);
4100         spin_unlock(&clp->cl_lock);
4101         ret = nfsd4_register_conn(new);
4102         if (ret)
4103                 /* oops; xprt is already down: */
4104                 nfsd4_conn_lost(&new->cn_xpt_user);
4105         return nfs_ok;
4106 out_free:
4107         spin_unlock(&clp->cl_lock);
4108         free_conn(new);
4109         return status;
4110 }
4111
4112 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4113 {
4114         struct nfsd4_compoundargs *args = rqstp->rq_argp;
4115
4116         return args->opcnt > session->se_fchannel.maxops;
4117 }
4118
4119 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4120                                   struct nfsd4_session *session)
4121 {
4122         struct xdr_buf *xb = &rqstp->rq_arg;
4123
4124         return xb->len > session->se_fchannel.maxreq_sz;
4125 }
4126
4127 static bool replay_matches_cache(struct svc_rqst *rqstp,
4128                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4129 {
4130         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4131
4132         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4133             (bool)seq->cachethis)
4134                 return false;
4135         /*
4136          * If there's an error then the reply can have fewer ops than
4137          * the call.
4138          */
4139         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4140                 return false;
4141         /*
4142          * But if we cached a reply with *more* ops than the call you're
4143          * sending us now, then this new call is clearly not really a
4144          * replay of the old one:
4145          */
4146         if (slot->sl_opcnt > argp->opcnt)
4147                 return false;
4148         /* This is the only check explicitly called by spec: */
4149         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4150                 return false;
4151         /*
4152          * There may be more comparisons we could actually do, but the
4153          * spec doesn't require us to catch every case where the calls
4154          * don't match (that would require caching the call as well as
4155          * the reply), so we don't bother.
4156          */
4157         return true;
4158 }
4159
4160 __be32
4161 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4162                 union nfsd4_op_u *u)
4163 {
4164         struct nfsd4_sequence *seq = &u->sequence;
4165         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4166         struct xdr_stream *xdr = resp->xdr;
4167         struct nfsd4_session *session;
4168         struct nfs4_client *clp;
4169         struct nfsd4_slot *slot;
4170         struct nfsd4_conn *conn;
4171         __be32 status;
4172         int buflen;
4173         struct net *net = SVC_NET(rqstp);
4174         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4175
4176         if (resp->opcnt != 1)
4177                 return nfserr_sequence_pos;
4178
4179         /*
4180          * Will be either used or freed by nfsd4_sequence_check_conn
4181          * below.
4182          */
4183         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4184         if (!conn)
4185                 return nfserr_jukebox;
4186
4187         spin_lock(&nn->client_lock);
4188         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4189         if (!session)
4190                 goto out_no_session;
4191         clp = session->se_client;
4192
4193         status = nfserr_too_many_ops;
4194         if (nfsd4_session_too_many_ops(rqstp, session))
4195                 goto out_put_session;
4196
4197         status = nfserr_req_too_big;
4198         if (nfsd4_request_too_big(rqstp, session))
4199                 goto out_put_session;
4200
4201         status = nfserr_badslot;
4202         if (seq->slotid >= session->se_fchannel.maxreqs)
4203                 goto out_put_session;
4204
4205         slot = session->se_slots[seq->slotid];
4206         dprintk("%s: slotid %d\n", __func__, seq->slotid);
4207
4208         /* We do not negotiate the number of slots yet, so set the
4209          * maxslots to the session maxreqs which is used to encode
4210          * sr_highest_slotid and the sr_target_slot id to maxslots */
4211         seq->maxslots = session->se_fchannel.maxreqs;
4212
4213         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4214                                         slot->sl_flags & NFSD4_SLOT_INUSE);
4215         if (status == nfserr_replay_cache) {
4216                 status = nfserr_seq_misordered;
4217                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4218                         goto out_put_session;
4219                 status = nfserr_seq_false_retry;
4220                 if (!replay_matches_cache(rqstp, seq, slot))
4221                         goto out_put_session;
4222                 cstate->slot = slot;
4223                 cstate->session = session;
4224                 cstate->clp = clp;
4225                 /* Return the cached reply status and set cstate->status
4226                  * for nfsd4_proc_compound processing */
4227                 status = nfsd4_replay_cache_entry(resp, seq);
4228                 cstate->status = nfserr_replay_cache;
4229                 goto out;
4230         }
4231         if (status)
4232                 goto out_put_session;
4233
4234         status = nfsd4_sequence_check_conn(conn, session);
4235         conn = NULL;
4236         if (status)
4237                 goto out_put_session;
4238
4239         buflen = (seq->cachethis) ?
4240                         session->se_fchannel.maxresp_cached :
4241                         session->se_fchannel.maxresp_sz;
4242         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4243                                     nfserr_rep_too_big;
4244         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4245                 goto out_put_session;
4246         svc_reserve(rqstp, buflen);
4247
4248         status = nfs_ok;
4249         /* Success! bump slot seqid */
4250         slot->sl_seqid = seq->seqid;
4251         slot->sl_flags |= NFSD4_SLOT_INUSE;
4252         if (seq->cachethis)
4253                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4254         else
4255                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4256
4257         cstate->slot = slot;
4258         cstate->session = session;
4259         cstate->clp = clp;
4260
4261 out:
4262         switch (clp->cl_cb_state) {
4263         case NFSD4_CB_DOWN:
4264                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4265                 break;
4266         case NFSD4_CB_FAULT:
4267                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4268                 break;
4269         default:
4270                 seq->status_flags = 0;
4271         }
4272         if (!list_empty(&clp->cl_revoked))
4273                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4274         if (atomic_read(&clp->cl_admin_revoked))
4275                 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4276         trace_nfsd_seq4_status(rqstp, seq);
4277 out_no_session:
4278         if (conn)
4279                 free_conn(conn);
4280         spin_unlock(&nn->client_lock);
4281         return status;
4282 out_put_session:
4283         nfsd4_put_session_locked(session);
4284         goto out_no_session;
4285 }
4286
4287 void
4288 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4289 {
4290         struct nfsd4_compound_state *cs = &resp->cstate;
4291
4292         if (nfsd4_has_session(cs)) {
4293                 if (cs->status != nfserr_replay_cache) {
4294                         nfsd4_store_cache_entry(resp);
4295                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4296                 }
4297                 /* Drop session reference that was taken in nfsd4_sequence() */
4298                 nfsd4_put_session(cs->session);
4299         } else if (cs->clp)
4300                 put_client_renew(cs->clp);
4301 }
4302
4303 __be32
4304 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4305                 struct nfsd4_compound_state *cstate,
4306                 union nfsd4_op_u *u)
4307 {
4308         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4309         struct nfs4_client *conf, *unconf;
4310         struct nfs4_client *clp = NULL;
4311         __be32 status = 0;
4312         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4313
4314         spin_lock(&nn->client_lock);
4315         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4316         conf = find_confirmed_client(&dc->clientid, true, nn);
4317         WARN_ON_ONCE(conf && unconf);
4318
4319         if (conf) {
4320                 if (client_has_state(conf)) {
4321                         status = nfserr_clientid_busy;
4322                         goto out;
4323                 }
4324                 status = mark_client_expired_locked(conf);
4325                 if (status)
4326                         goto out;
4327                 clp = conf;
4328         } else if (unconf)
4329                 clp = unconf;
4330         else {
4331                 status = nfserr_stale_clientid;
4332                 goto out;
4333         }
4334         if (!nfsd4_mach_creds_match(clp, rqstp)) {
4335                 clp = NULL;
4336                 status = nfserr_wrong_cred;
4337                 goto out;
4338         }
4339         trace_nfsd_clid_destroyed(&clp->cl_clientid);
4340         unhash_client_locked(clp);
4341 out:
4342         spin_unlock(&nn->client_lock);
4343         if (clp)
4344                 expire_client(clp);
4345         return status;
4346 }
4347
4348 __be32
4349 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4350                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4351 {
4352         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4353         struct nfs4_client *clp = cstate->clp;
4354         __be32 status = 0;
4355
4356         if (rc->rca_one_fs) {
4357                 if (!cstate->current_fh.fh_dentry)
4358                         return nfserr_nofilehandle;
4359                 /*
4360                  * We don't take advantage of the rca_one_fs case.
4361                  * That's OK, it's optional, we can safely ignore it.
4362                  */
4363                 return nfs_ok;
4364         }
4365
4366         status = nfserr_complete_already;
4367         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4368                 goto out;
4369
4370         status = nfserr_stale_clientid;
4371         if (is_client_expired(clp))
4372                 /*
4373                  * The following error isn't really legal.
4374                  * But we only get here if the client just explicitly
4375                  * destroyed the client.  Surely it no longer cares what
4376                  * error it gets back on an operation for the dead
4377                  * client.
4378                  */
4379                 goto out;
4380
4381         status = nfs_ok;
4382         trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4383         nfsd4_client_record_create(clp);
4384         inc_reclaim_complete(clp);
4385 out:
4386         return status;
4387 }
4388
4389 __be32
4390 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4391                   union nfsd4_op_u *u)
4392 {
4393         struct nfsd4_setclientid *setclid = &u->setclientid;
4394         struct xdr_netobj       clname = setclid->se_name;
4395         nfs4_verifier           clverifier = setclid->se_verf;
4396         struct nfs4_client      *conf, *new;
4397         struct nfs4_client      *unconf = NULL;
4398         __be32                  status;
4399         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4400
4401         new = create_client(clname, rqstp, &clverifier);
4402         if (new == NULL)
4403                 return nfserr_jukebox;
4404         spin_lock(&nn->client_lock);
4405         conf = find_confirmed_client_by_name(&clname, nn);
4406         if (conf && client_has_state(conf)) {
4407                 status = nfserr_clid_inuse;
4408                 if (clp_used_exchangeid(conf))
4409                         goto out;
4410                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4411                         trace_nfsd_clid_cred_mismatch(conf, rqstp);
4412                         goto out;
4413                 }
4414         }
4415         unconf = find_unconfirmed_client_by_name(&clname, nn);
4416         if (unconf)
4417                 unhash_client_locked(unconf);
4418         if (conf) {
4419                 if (same_verf(&conf->cl_verifier, &clverifier)) {
4420                         copy_clid(new, conf);
4421                         gen_confirm(new, nn);
4422                 } else
4423                         trace_nfsd_clid_verf_mismatch(conf, rqstp,
4424                                                       &clverifier);
4425         } else
4426                 trace_nfsd_clid_fresh(new);
4427         new->cl_minorversion = 0;
4428         gen_callback(new, setclid, rqstp);
4429         add_to_unconfirmed(new);
4430         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4431         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4432         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4433         new = NULL;
4434         status = nfs_ok;
4435 out:
4436         spin_unlock(&nn->client_lock);
4437         if (new)
4438                 free_client(new);
4439         if (unconf) {
4440                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4441                 expire_client(unconf);
4442         }
4443         return status;
4444 }
4445
4446 __be32
4447 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4448                         struct nfsd4_compound_state *cstate,
4449                         union nfsd4_op_u *u)
4450 {
4451         struct nfsd4_setclientid_confirm *setclientid_confirm =
4452                         &u->setclientid_confirm;
4453         struct nfs4_client *conf, *unconf;
4454         struct nfs4_client *old = NULL;
4455         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4456         clientid_t * clid = &setclientid_confirm->sc_clientid;
4457         __be32 status;
4458         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4459
4460         if (STALE_CLIENTID(clid, nn))
4461                 return nfserr_stale_clientid;
4462
4463         spin_lock(&nn->client_lock);
4464         conf = find_confirmed_client(clid, false, nn);
4465         unconf = find_unconfirmed_client(clid, false, nn);
4466         /*
4467          * We try hard to give out unique clientid's, so if we get an
4468          * attempt to confirm the same clientid with a different cred,
4469          * the client may be buggy; this should never happen.
4470          *
4471          * Nevertheless, RFC 7530 recommends INUSE for this case:
4472          */
4473         status = nfserr_clid_inuse;
4474         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4475                 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4476                 goto out;
4477         }
4478         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4479                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4480                 goto out;
4481         }
4482         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4483                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4484                         status = nfs_ok;
4485                 } else
4486                         status = nfserr_stale_clientid;
4487                 goto out;
4488         }
4489         status = nfs_ok;
4490         if (conf) {
4491                 old = unconf;
4492                 unhash_client_locked(old);
4493                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4494         } else {
4495                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4496                 if (old) {
4497                         status = nfserr_clid_inuse;
4498                         if (client_has_state(old)
4499                                         && !same_creds(&unconf->cl_cred,
4500                                                         &old->cl_cred)) {
4501                                 old = NULL;
4502                                 goto out;
4503                         }
4504                         status = mark_client_expired_locked(old);
4505                         if (status) {
4506                                 old = NULL;
4507                                 goto out;
4508                         }
4509                         trace_nfsd_clid_replaced(&old->cl_clientid);
4510                 }
4511                 move_to_confirmed(unconf);
4512                 conf = unconf;
4513         }
4514         get_client_locked(conf);
4515         spin_unlock(&nn->client_lock);
4516         if (conf == unconf)
4517                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4518         nfsd4_probe_callback(conf);
4519         spin_lock(&nn->client_lock);
4520         put_client_renew_locked(conf);
4521 out:
4522         spin_unlock(&nn->client_lock);
4523         if (old)
4524                 expire_client(old);
4525         return status;
4526 }
4527
4528 static struct nfs4_file *nfsd4_alloc_file(void)
4529 {
4530         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4531 }
4532
4533 /* OPEN Share state helper functions */
4534
4535 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4536 {
4537         refcount_set(&fp->fi_ref, 1);
4538         spin_lock_init(&fp->fi_lock);
4539         INIT_LIST_HEAD(&fp->fi_stateids);
4540         INIT_LIST_HEAD(&fp->fi_delegations);
4541         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4542         fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4543         fp->fi_deleg_file = NULL;
4544         fp->fi_had_conflict = false;
4545         fp->fi_share_deny = 0;
4546         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4547         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4548         fp->fi_aliased = false;
4549         fp->fi_inode = d_inode(fh->fh_dentry);
4550 #ifdef CONFIG_NFSD_PNFS
4551         INIT_LIST_HEAD(&fp->fi_lo_states);
4552         atomic_set(&fp->fi_lo_recalls, 0);
4553 #endif
4554 }
4555
4556 void
4557 nfsd4_free_slabs(void)
4558 {
4559         kmem_cache_destroy(client_slab);
4560         kmem_cache_destroy(openowner_slab);
4561         kmem_cache_destroy(lockowner_slab);
4562         kmem_cache_destroy(file_slab);
4563         kmem_cache_destroy(stateid_slab);
4564         kmem_cache_destroy(deleg_slab);
4565         kmem_cache_destroy(odstate_slab);
4566 }
4567
4568 int
4569 nfsd4_init_slabs(void)
4570 {
4571         client_slab = KMEM_CACHE(nfs4_client, 0);
4572         if (client_slab == NULL)
4573                 goto out;
4574         openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4575         if (openowner_slab == NULL)
4576                 goto out_free_client_slab;
4577         lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4578         if (lockowner_slab == NULL)
4579                 goto out_free_openowner_slab;
4580         file_slab = KMEM_CACHE(nfs4_file, 0);
4581         if (file_slab == NULL)
4582                 goto out_free_lockowner_slab;
4583         stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4584         if (stateid_slab == NULL)
4585                 goto out_free_file_slab;
4586         deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4587         if (deleg_slab == NULL)
4588                 goto out_free_stateid_slab;
4589         odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4590         if (odstate_slab == NULL)
4591                 goto out_free_deleg_slab;
4592         return 0;
4593
4594 out_free_deleg_slab:
4595         kmem_cache_destroy(deleg_slab);
4596 out_free_stateid_slab:
4597         kmem_cache_destroy(stateid_slab);
4598 out_free_file_slab:
4599         kmem_cache_destroy(file_slab);
4600 out_free_lockowner_slab:
4601         kmem_cache_destroy(lockowner_slab);
4602 out_free_openowner_slab:
4603         kmem_cache_destroy(openowner_slab);
4604 out_free_client_slab:
4605         kmem_cache_destroy(client_slab);
4606 out:
4607         return -ENOMEM;
4608 }
4609
4610 static unsigned long
4611 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4612 {
4613         int count;
4614         struct nfsd_net *nn = shrink->private_data;
4615
4616         count = atomic_read(&nn->nfsd_courtesy_clients);
4617         if (!count)
4618                 count = atomic_long_read(&num_delegations);
4619         if (count)
4620                 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4621         return (unsigned long)count;
4622 }
4623
4624 static unsigned long
4625 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4626 {
4627         return SHRINK_STOP;
4628 }
4629
4630 void
4631 nfsd4_init_leases_net(struct nfsd_net *nn)
4632 {
4633         struct sysinfo si;
4634         u64 max_clients;
4635
4636         nn->nfsd4_lease = 90;   /* default lease time */
4637         nn->nfsd4_grace = 90;
4638         nn->somebody_reclaimed = false;
4639         nn->track_reclaim_completes = false;
4640         nn->clverifier_counter = get_random_u32();
4641         nn->clientid_base = get_random_u32();
4642         nn->clientid_counter = nn->clientid_base + 1;
4643         nn->s2s_cp_cl_id = nn->clientid_counter++;
4644
4645         atomic_set(&nn->nfs4_client_count, 0);
4646         si_meminfo(&si);
4647         max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4648         max_clients *= NFS4_CLIENTS_PER_GB;
4649         nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4650
4651         atomic_set(&nn->nfsd_courtesy_clients, 0);
4652 }
4653
4654 static void init_nfs4_replay(struct nfs4_replay *rp)
4655 {
4656         rp->rp_status = nfserr_serverfault;
4657         rp->rp_buflen = 0;
4658         rp->rp_buf = rp->rp_ibuf;
4659         mutex_init(&rp->rp_mutex);
4660 }
4661
4662 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4663                 struct nfs4_stateowner *so)
4664 {
4665         if (!nfsd4_has_session(cstate)) {
4666                 mutex_lock(&so->so_replay.rp_mutex);
4667                 cstate->replay_owner = nfs4_get_stateowner(so);
4668         }
4669 }
4670
4671 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4672 {
4673         struct nfs4_stateowner *so = cstate->replay_owner;
4674
4675         if (so != NULL) {
4676                 cstate->replay_owner = NULL;
4677                 mutex_unlock(&so->so_replay.rp_mutex);
4678                 nfs4_put_stateowner(so);
4679         }
4680 }
4681
4682 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4683 {
4684         struct nfs4_stateowner *sop;
4685
4686         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4687         if (!sop)
4688                 return NULL;
4689
4690         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4691         if (!sop->so_owner.data) {
4692                 kmem_cache_free(slab, sop);
4693                 return NULL;
4694         }
4695
4696         INIT_LIST_HEAD(&sop->so_stateids);
4697         sop->so_client = clp;
4698         init_nfs4_replay(&sop->so_replay);
4699         atomic_set(&sop->so_count, 1);
4700         return sop;
4701 }
4702
4703 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4704 {
4705         lockdep_assert_held(&clp->cl_lock);
4706
4707         list_add(&oo->oo_owner.so_strhash,
4708                  &clp->cl_ownerstr_hashtbl[strhashval]);
4709         list_add(&oo->oo_perclient, &clp->cl_openowners);
4710 }
4711
4712 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4713 {
4714         unhash_openowner_locked(openowner(so));
4715 }
4716
4717 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4718 {
4719         struct nfs4_openowner *oo = openowner(so);
4720
4721         kmem_cache_free(openowner_slab, oo);
4722 }
4723
4724 static const struct nfs4_stateowner_operations openowner_ops = {
4725         .so_unhash =    nfs4_unhash_openowner,
4726         .so_free =      nfs4_free_openowner,
4727 };
4728
4729 static struct nfs4_ol_stateid *
4730 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4731 {
4732         struct nfs4_ol_stateid *local, *ret = NULL;
4733         struct nfs4_openowner *oo = open->op_openowner;
4734
4735         lockdep_assert_held(&fp->fi_lock);
4736
4737         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4738                 /* ignore lock owners */
4739                 if (local->st_stateowner->so_is_open_owner == 0)
4740                         continue;
4741                 if (local->st_stateowner != &oo->oo_owner)
4742                         continue;
4743                 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4744                     !local->st_stid.sc_status) {
4745                         ret = local;
4746                         refcount_inc(&ret->st_stid.sc_count);
4747                         break;
4748                 }
4749         }
4750         return ret;
4751 }
4752
4753 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4754         __releases(&s->sc_client->cl_lock)
4755 {
4756         struct nfs4_client *cl = s->sc_client;
4757         LIST_HEAD(reaplist);
4758         struct nfs4_ol_stateid *stp;
4759         struct nfs4_delegation *dp;
4760         bool unhashed;
4761
4762         switch (s->sc_type) {
4763         case SC_TYPE_OPEN:
4764                 stp = openlockstateid(s);
4765                 if (unhash_open_stateid(stp, &reaplist))
4766                         put_ol_stateid_locked(stp, &reaplist);
4767                 spin_unlock(&cl->cl_lock);
4768                 free_ol_stateid_reaplist(&reaplist);
4769                 break;
4770         case SC_TYPE_LOCK:
4771                 stp = openlockstateid(s);
4772                 unhashed = unhash_lock_stateid(stp);
4773                 spin_unlock(&cl->cl_lock);
4774                 if (unhashed)
4775                         nfs4_put_stid(s);
4776                 break;
4777         case SC_TYPE_DELEG:
4778                 dp = delegstateid(s);
4779                 list_del_init(&dp->dl_recall_lru);
4780                 spin_unlock(&cl->cl_lock);
4781                 nfs4_put_stid(s);
4782                 break;
4783         default:
4784                 spin_unlock(&cl->cl_lock);
4785         }
4786 }
4787
4788 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4789                                     stateid_t *stid)
4790 {
4791         /* NFSv4.0 has no way for the client to tell the server
4792          * that it can forget an admin-revoked stateid.
4793          * So we keep it around until the first time that the
4794          * client uses it, and drop it the first time
4795          * nfserr_admin_revoked is returned.
4796          * For v4.1 and later we wait until explicitly told
4797          * to free the stateid.
4798          */
4799         if (cl->cl_minorversion == 0) {
4800                 struct nfs4_stid *st;
4801
4802                 spin_lock(&cl->cl_lock);
4803                 st = find_stateid_locked(cl, stid);
4804                 if (st)
4805                         nfsd4_drop_revoked_stid(st);
4806                 else
4807                         spin_unlock(&cl->cl_lock);
4808         }
4809 }
4810
4811 static __be32
4812 nfsd4_verify_open_stid(struct nfs4_stid *s)
4813 {
4814         __be32 ret = nfs_ok;
4815
4816         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4817                 ret = nfserr_admin_revoked;
4818         else if (s->sc_status & SC_STATUS_REVOKED)
4819                 ret = nfserr_deleg_revoked;
4820         else if (s->sc_status & SC_STATUS_CLOSED)
4821                 ret = nfserr_bad_stateid;
4822         return ret;
4823 }
4824
4825 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4826 static __be32
4827 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4828 {
4829         __be32 ret;
4830
4831         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4832         ret = nfsd4_verify_open_stid(&stp->st_stid);
4833         if (ret == nfserr_admin_revoked)
4834                 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4835                                         &stp->st_stid.sc_stateid);
4836
4837         if (ret != nfs_ok)
4838                 mutex_unlock(&stp->st_mutex);
4839         return ret;
4840 }
4841
4842 static struct nfs4_ol_stateid *
4843 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4844 {
4845         struct nfs4_ol_stateid *stp;
4846         for (;;) {
4847                 spin_lock(&fp->fi_lock);
4848                 stp = nfsd4_find_existing_open(fp, open);
4849                 spin_unlock(&fp->fi_lock);
4850                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4851                         break;
4852                 nfs4_put_stid(&stp->st_stid);
4853         }
4854         return stp;
4855 }
4856
4857 static struct nfs4_openowner *
4858 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4859                            struct nfsd4_compound_state *cstate)
4860 {
4861         struct nfs4_client *clp = cstate->clp;
4862         struct nfs4_openowner *oo, *ret;
4863
4864         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4865         if (!oo)
4866                 return NULL;
4867         oo->oo_owner.so_ops = &openowner_ops;
4868         oo->oo_owner.so_is_open_owner = 1;
4869         oo->oo_owner.so_seqid = open->op_seqid;
4870         oo->oo_flags = 0;
4871         if (nfsd4_has_session(cstate))
4872                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4873         oo->oo_time = 0;
4874         oo->oo_last_closed_stid = NULL;
4875         INIT_LIST_HEAD(&oo->oo_close_lru);
4876         spin_lock(&clp->cl_lock);
4877         ret = find_openstateowner_str_locked(strhashval, open, clp);
4878         if (ret == NULL) {
4879                 hash_openowner(oo, clp, strhashval);
4880                 ret = oo;
4881         } else
4882                 nfs4_free_stateowner(&oo->oo_owner);
4883
4884         spin_unlock(&clp->cl_lock);
4885         return ret;
4886 }
4887
4888 static struct nfs4_ol_stateid *
4889 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4890 {
4891
4892         struct nfs4_openowner *oo = open->op_openowner;
4893         struct nfs4_ol_stateid *retstp = NULL;
4894         struct nfs4_ol_stateid *stp;
4895
4896         stp = open->op_stp;
4897         /* We are moving these outside of the spinlocks to avoid the warnings */
4898         mutex_init(&stp->st_mutex);
4899         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4900
4901 retry:
4902         spin_lock(&oo->oo_owner.so_client->cl_lock);
4903         spin_lock(&fp->fi_lock);
4904
4905         retstp = nfsd4_find_existing_open(fp, open);
4906         if (retstp)
4907                 goto out_unlock;
4908
4909         open->op_stp = NULL;
4910         refcount_inc(&stp->st_stid.sc_count);
4911         stp->st_stid.sc_type = SC_TYPE_OPEN;
4912         INIT_LIST_HEAD(&stp->st_locks);
4913         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4914         get_nfs4_file(fp);
4915         stp->st_stid.sc_file = fp;
4916         stp->st_access_bmap = 0;
4917         stp->st_deny_bmap = 0;
4918         stp->st_openstp = NULL;
4919         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4920         list_add(&stp->st_perfile, &fp->fi_stateids);
4921
4922 out_unlock:
4923         spin_unlock(&fp->fi_lock);
4924         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4925         if (retstp) {
4926                 /* Handle races with CLOSE */
4927                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4928                         nfs4_put_stid(&retstp->st_stid);
4929                         goto retry;
4930                 }
4931                 /* To keep mutex tracking happy */
4932                 mutex_unlock(&stp->st_mutex);
4933                 stp = retstp;
4934         }
4935         return stp;
4936 }
4937
4938 /*
4939  * In the 4.0 case we need to keep the owners around a little while to handle
4940  * CLOSE replay. We still do need to release any file access that is held by
4941  * them before returning however.
4942  */
4943 static void
4944 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4945 {
4946         struct nfs4_ol_stateid *last;
4947         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4948         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4949                                                 nfsd_net_id);
4950
4951         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4952
4953         /*
4954          * We know that we hold one reference via nfsd4_close, and another
4955          * "persistent" reference for the client. If the refcount is higher
4956          * than 2, then there are still calls in progress that are using this
4957          * stateid. We can't put the sc_file reference until they are finished.
4958          * Wait for the refcount to drop to 2. Since it has been unhashed,
4959          * there should be no danger of the refcount going back up again at
4960          * this point.
4961          */
4962         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4963
4964         release_all_access(s);
4965         if (s->st_stid.sc_file) {
4966                 put_nfs4_file(s->st_stid.sc_file);
4967                 s->st_stid.sc_file = NULL;
4968         }
4969
4970         spin_lock(&nn->client_lock);
4971         last = oo->oo_last_closed_stid;
4972         oo->oo_last_closed_stid = s;
4973         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4974         oo->oo_time = ktime_get_boottime_seconds();
4975         spin_unlock(&nn->client_lock);
4976         if (last)
4977                 nfs4_put_stid(&last->st_stid);
4978 }
4979
4980 static noinline_for_stack struct nfs4_file *
4981 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4982 {
4983         struct inode *inode = d_inode(fhp->fh_dentry);
4984         struct rhlist_head *tmp, *list;
4985         struct nfs4_file *fi;
4986
4987         rcu_read_lock();
4988         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4989                                nfs4_file_rhash_params);
4990         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4991                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4992                         if (refcount_inc_not_zero(&fi->fi_ref)) {
4993                                 rcu_read_unlock();
4994                                 return fi;
4995                         }
4996                 }
4997         }
4998         rcu_read_unlock();
4999         return NULL;
5000 }
5001
5002 /*
5003  * On hash insertion, identify entries with the same inode but
5004  * distinct filehandles. They will all be on the list returned
5005  * by rhltable_lookup().
5006  *
5007  * inode->i_lock prevents racing insertions from adding an entry
5008  * for the same inode/fhp pair twice.
5009  */
5010 static noinline_for_stack struct nfs4_file *
5011 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5012 {
5013         struct inode *inode = d_inode(fhp->fh_dentry);
5014         struct rhlist_head *tmp, *list;
5015         struct nfs4_file *ret = NULL;
5016         bool alias_found = false;
5017         struct nfs4_file *fi;
5018         int err;
5019
5020         rcu_read_lock();
5021         spin_lock(&inode->i_lock);
5022
5023         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5024                                nfs4_file_rhash_params);
5025         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5026                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5027                         if (refcount_inc_not_zero(&fi->fi_ref))
5028                                 ret = fi;
5029                 } else
5030                         fi->fi_aliased = alias_found = true;
5031         }
5032         if (ret)
5033                 goto out_unlock;
5034
5035         nfsd4_file_init(fhp, new);
5036         err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5037                               nfs4_file_rhash_params);
5038         if (err)
5039                 goto out_unlock;
5040
5041         new->fi_aliased = alias_found;
5042         ret = new;
5043
5044 out_unlock:
5045         spin_unlock(&inode->i_lock);
5046         rcu_read_unlock();
5047         return ret;
5048 }
5049
5050 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5051 {
5052         rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5053                         nfs4_file_rhash_params);
5054 }
5055
5056 /*
5057  * Called to check deny when READ with all zero stateid or
5058  * WRITE with all zero or all one stateid
5059  */
5060 static __be32
5061 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5062 {
5063         struct nfs4_file *fp;
5064         __be32 ret = nfs_ok;
5065
5066         fp = nfsd4_file_hash_lookup(current_fh);
5067         if (!fp)
5068                 return ret;
5069
5070         /* Check for conflicting share reservations */
5071         spin_lock(&fp->fi_lock);
5072         if (fp->fi_share_deny & deny_type)
5073                 ret = nfserr_locked;
5074         spin_unlock(&fp->fi_lock);
5075         put_nfs4_file(fp);
5076         return ret;
5077 }
5078
5079 static bool nfsd4_deleg_present(const struct inode *inode)
5080 {
5081         struct file_lock_context *ctx = locks_inode_context(inode);
5082
5083         return ctx && !list_empty_careful(&ctx->flc_lease);
5084 }
5085
5086 /**
5087  * nfsd_wait_for_delegreturn - wait for delegations to be returned
5088  * @rqstp: the RPC transaction being executed
5089  * @inode: in-core inode of the file being waited for
5090  *
5091  * The timeout prevents deadlock if all nfsd threads happen to be
5092  * tied up waiting for returning delegations.
5093  *
5094  * Return values:
5095  *   %true: delegation was returned
5096  *   %false: timed out waiting for delegreturn
5097  */
5098 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5099 {
5100         long __maybe_unused timeo;
5101
5102         timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5103                                        NFSD_DELEGRETURN_TIMEOUT);
5104         trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5105         return timeo > 0;
5106 }
5107
5108 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5109 {
5110         struct nfs4_delegation *dp = cb_to_delegation(cb);
5111         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5112                                           nfsd_net_id);
5113
5114         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5115
5116         /*
5117          * We can't do this in nfsd_break_deleg_cb because it is
5118          * already holding inode->i_lock.
5119          *
5120          * If the dl_time != 0, then we know that it has already been
5121          * queued for a lease break. Don't queue it again.
5122          */
5123         spin_lock(&state_lock);
5124         if (delegation_hashed(dp) && dp->dl_time == 0) {
5125                 dp->dl_time = ktime_get_boottime_seconds();
5126                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5127         }
5128         spin_unlock(&state_lock);
5129 }
5130
5131 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5132                 struct rpc_task *task)
5133 {
5134         struct nfs4_delegation *dp = cb_to_delegation(cb);
5135
5136         trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5137
5138         if (dp->dl_stid.sc_status)
5139                 /* CLOSED or REVOKED */
5140                 return 1;
5141
5142         switch (task->tk_status) {
5143         case 0:
5144                 return 1;
5145         case -NFS4ERR_DELAY:
5146                 rpc_delay(task, 2 * HZ);
5147                 return 0;
5148         case -EBADHANDLE:
5149         case -NFS4ERR_BAD_STATEID:
5150                 /*
5151                  * Race: client probably got cb_recall before open reply
5152                  * granting delegation.
5153                  */
5154                 if (dp->dl_retries--) {
5155                         rpc_delay(task, 2 * HZ);
5156                         return 0;
5157                 }
5158                 fallthrough;
5159         default:
5160                 return 1;
5161         }
5162 }
5163
5164 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5165 {
5166         struct nfs4_delegation *dp = cb_to_delegation(cb);
5167
5168         nfs4_put_stid(&dp->dl_stid);
5169 }
5170
5171 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5172         .prepare        = nfsd4_cb_recall_prepare,
5173         .done           = nfsd4_cb_recall_done,
5174         .release        = nfsd4_cb_recall_release,
5175 };
5176
5177 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5178 {
5179         /*
5180          * We're assuming the state code never drops its reference
5181          * without first removing the lease.  Since we're in this lease
5182          * callback (and since the lease code is serialized by the
5183          * flc_lock) we know the server hasn't removed the lease yet, and
5184          * we know it's safe to take a reference.
5185          */
5186         refcount_inc(&dp->dl_stid.sc_count);
5187         WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5188 }
5189
5190 /* Called from break_lease() with flc_lock held. */
5191 static bool
5192 nfsd_break_deleg_cb(struct file_lease *fl)
5193 {
5194         struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5195         struct nfs4_file *fp = dp->dl_stid.sc_file;
5196         struct nfs4_client *clp = dp->dl_stid.sc_client;
5197         struct nfsd_net *nn;
5198
5199         trace_nfsd_cb_recall(&dp->dl_stid);
5200
5201         dp->dl_recalled = true;
5202         atomic_inc(&clp->cl_delegs_in_recall);
5203         if (try_to_expire_client(clp)) {
5204                 nn = net_generic(clp->net, nfsd_net_id);
5205                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5206         }
5207
5208         /*
5209          * We don't want the locks code to timeout the lease for us;
5210          * we'll remove it ourself if a delegation isn't returned
5211          * in time:
5212          */
5213         fl->fl_break_time = 0;
5214
5215         fp->fi_had_conflict = true;
5216         nfsd_break_one_deleg(dp);
5217         return false;
5218 }
5219
5220 /**
5221  * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5222  * @fl: Lock state to check
5223  *
5224  * Return values:
5225  *   %true: Lease conflict was resolved
5226  *   %false: Lease conflict was not resolved.
5227  */
5228 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5229 {
5230         struct nfs4_delegation *dl = fl->c.flc_owner;
5231         struct svc_rqst *rqst;
5232         struct nfs4_client *clp;
5233
5234         if (!i_am_nfsd())
5235                 return false;
5236         rqst = kthread_data(current);
5237         /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
5238         if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
5239                 return false;
5240         clp = *(rqst->rq_lease_breaker);
5241         return dl->dl_stid.sc_client == clp;
5242 }
5243
5244 static int
5245 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5246                      struct list_head *dispose)
5247 {
5248         struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5249         struct nfs4_client *clp = dp->dl_stid.sc_client;
5250
5251         if (arg & F_UNLCK) {
5252                 if (dp->dl_recalled)
5253                         atomic_dec(&clp->cl_delegs_in_recall);
5254                 return lease_modify(onlist, arg, dispose);
5255         } else
5256                 return -EAGAIN;
5257 }
5258
5259 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5260         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5261         .lm_break = nfsd_break_deleg_cb,
5262         .lm_change = nfsd_change_deleg_cb,
5263 };
5264
5265 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5266 {
5267         if (nfsd4_has_session(cstate))
5268                 return nfs_ok;
5269         if (seqid == so->so_seqid - 1)
5270                 return nfserr_replay_me;
5271         if (seqid == so->so_seqid)
5272                 return nfs_ok;
5273         return nfserr_bad_seqid;
5274 }
5275
5276 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5277                                                 struct nfsd_net *nn)
5278 {
5279         struct nfs4_client *found;
5280
5281         spin_lock(&nn->client_lock);
5282         found = find_confirmed_client(clid, sessions, nn);
5283         if (found)
5284                 atomic_inc(&found->cl_rpc_users);
5285         spin_unlock(&nn->client_lock);
5286         return found;
5287 }
5288
5289 static __be32 set_client(clientid_t *clid,
5290                 struct nfsd4_compound_state *cstate,
5291                 struct nfsd_net *nn)
5292 {
5293         if (cstate->clp) {
5294                 if (!same_clid(&cstate->clp->cl_clientid, clid))
5295                         return nfserr_stale_clientid;
5296                 return nfs_ok;
5297         }
5298         if (STALE_CLIENTID(clid, nn))
5299                 return nfserr_stale_clientid;
5300         /*
5301          * We're in the 4.0 case (otherwise the SEQUENCE op would have
5302          * set cstate->clp), so session = false:
5303          */
5304         cstate->clp = lookup_clientid(clid, false, nn);
5305         if (!cstate->clp)
5306                 return nfserr_expired;
5307         return nfs_ok;
5308 }
5309
5310 __be32
5311 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5312                     struct nfsd4_open *open, struct nfsd_net *nn)
5313 {
5314         clientid_t *clientid = &open->op_clientid;
5315         struct nfs4_client *clp = NULL;
5316         unsigned int strhashval;
5317         struct nfs4_openowner *oo = NULL;
5318         __be32 status;
5319
5320         /*
5321          * In case we need it later, after we've already created the
5322          * file and don't want to risk a further failure:
5323          */
5324         open->op_file = nfsd4_alloc_file();
5325         if (open->op_file == NULL)
5326                 return nfserr_jukebox;
5327
5328         status = set_client(clientid, cstate, nn);
5329         if (status)
5330                 return status;
5331         clp = cstate->clp;
5332
5333         strhashval = ownerstr_hashval(&open->op_owner);
5334         oo = find_openstateowner_str(strhashval, open, clp);
5335         open->op_openowner = oo;
5336         if (!oo) {
5337                 goto new_owner;
5338         }
5339         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5340                 /* Replace unconfirmed owners without checking for replay. */
5341                 release_openowner(oo);
5342                 open->op_openowner = NULL;
5343                 goto new_owner;
5344         }
5345         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5346         if (status)
5347                 return status;
5348         goto alloc_stateid;
5349 new_owner:
5350         oo = alloc_init_open_stateowner(strhashval, open, cstate);
5351         if (oo == NULL)
5352                 return nfserr_jukebox;
5353         open->op_openowner = oo;
5354 alloc_stateid:
5355         open->op_stp = nfs4_alloc_open_stateid(clp);
5356         if (!open->op_stp)
5357                 return nfserr_jukebox;
5358
5359         if (nfsd4_has_session(cstate) &&
5360             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5361                 open->op_odstate = alloc_clnt_odstate(clp);
5362                 if (!open->op_odstate)
5363                         return nfserr_jukebox;
5364         }
5365
5366         return nfs_ok;
5367 }
5368
5369 static inline __be32
5370 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5371 {
5372         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5373                 return nfserr_openmode;
5374         else
5375                 return nfs_ok;
5376 }
5377
5378 static int share_access_to_flags(u32 share_access)
5379 {
5380         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5381 }
5382
5383 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5384                                                   stateid_t *s)
5385 {
5386         struct nfs4_stid *ret;
5387
5388         ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5389         if (!ret)
5390                 return NULL;
5391         return delegstateid(ret);
5392 }
5393
5394 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5395 {
5396         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5397                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5398 }
5399
5400 static __be32
5401 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5402                 struct nfs4_delegation **dp)
5403 {
5404         int flags;
5405         __be32 status = nfserr_bad_stateid;
5406         struct nfs4_delegation *deleg;
5407
5408         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5409         if (deleg == NULL)
5410                 goto out;
5411         if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5412                 nfs4_put_stid(&deleg->dl_stid);
5413                 status = nfserr_admin_revoked;
5414                 goto out;
5415         }
5416         if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5417                 nfs4_put_stid(&deleg->dl_stid);
5418                 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5419                 status = nfserr_deleg_revoked;
5420                 goto out;
5421         }
5422         flags = share_access_to_flags(open->op_share_access);
5423         status = nfs4_check_delegmode(deleg, flags);
5424         if (status) {
5425                 nfs4_put_stid(&deleg->dl_stid);
5426                 goto out;
5427         }
5428         *dp = deleg;
5429 out:
5430         if (!nfsd4_is_deleg_cur(open))
5431                 return nfs_ok;
5432         if (status)
5433                 return status;
5434         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5435         return nfs_ok;
5436 }
5437
5438 static inline int nfs4_access_to_access(u32 nfs4_access)
5439 {
5440         int flags = 0;
5441
5442         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5443                 flags |= NFSD_MAY_READ;
5444         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5445                 flags |= NFSD_MAY_WRITE;
5446         return flags;
5447 }
5448
5449 static inline __be32
5450 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5451                 struct nfsd4_open *open)
5452 {
5453         struct iattr iattr = {
5454                 .ia_valid = ATTR_SIZE,
5455                 .ia_size = 0,
5456         };
5457         struct nfsd_attrs attrs = {
5458                 .na_iattr       = &iattr,
5459         };
5460         if (!open->op_truncate)
5461                 return 0;
5462         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5463                 return nfserr_inval;
5464         return nfsd_setattr(rqstp, fh, &attrs, NULL);
5465 }
5466
5467 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5468                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5469                 struct nfsd4_open *open, bool new_stp)
5470 {
5471         struct nfsd_file *nf = NULL;
5472         __be32 status;
5473         int oflag = nfs4_access_to_omode(open->op_share_access);
5474         int access = nfs4_access_to_access(open->op_share_access);
5475         unsigned char old_access_bmap, old_deny_bmap;
5476
5477         spin_lock(&fp->fi_lock);
5478
5479         /*
5480          * Are we trying to set a deny mode that would conflict with
5481          * current access?
5482          */
5483         status = nfs4_file_check_deny(fp, open->op_share_deny);
5484         if (status != nfs_ok) {
5485                 if (status != nfserr_share_denied) {
5486                         spin_unlock(&fp->fi_lock);
5487                         goto out;
5488                 }
5489                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5490                                 stp, open->op_share_deny, false))
5491                         status = nfserr_jukebox;
5492                 spin_unlock(&fp->fi_lock);
5493                 goto out;
5494         }
5495
5496         /* set access to the file */
5497         status = nfs4_file_get_access(fp, open->op_share_access);
5498         if (status != nfs_ok) {
5499                 if (status != nfserr_share_denied) {
5500                         spin_unlock(&fp->fi_lock);
5501                         goto out;
5502                 }
5503                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5504                                 stp, open->op_share_access, true))
5505                         status = nfserr_jukebox;
5506                 spin_unlock(&fp->fi_lock);
5507                 goto out;
5508         }
5509
5510         /* Set access bits in stateid */
5511         old_access_bmap = stp->st_access_bmap;
5512         set_access(open->op_share_access, stp);
5513
5514         /* Set new deny mask */
5515         old_deny_bmap = stp->st_deny_bmap;
5516         set_deny(open->op_share_deny, stp);
5517         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5518
5519         if (!fp->fi_fds[oflag]) {
5520                 spin_unlock(&fp->fi_lock);
5521
5522                 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5523                                                   open->op_filp, &nf);
5524                 if (status != nfs_ok)
5525                         goto out_put_access;
5526
5527                 spin_lock(&fp->fi_lock);
5528                 if (!fp->fi_fds[oflag]) {
5529                         fp->fi_fds[oflag] = nf;
5530                         nf = NULL;
5531                 }
5532         }
5533         spin_unlock(&fp->fi_lock);
5534         if (nf)
5535                 nfsd_file_put(nf);
5536
5537         status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5538                                                                 access));
5539         if (status)
5540                 goto out_put_access;
5541
5542         status = nfsd4_truncate(rqstp, cur_fh, open);
5543         if (status)
5544                 goto out_put_access;
5545 out:
5546         return status;
5547 out_put_access:
5548         stp->st_access_bmap = old_access_bmap;
5549         nfs4_file_put_access(fp, open->op_share_access);
5550         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5551         goto out;
5552 }
5553
5554 static __be32
5555 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5556                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5557                 struct nfsd4_open *open)
5558 {
5559         __be32 status;
5560         unsigned char old_deny_bmap = stp->st_deny_bmap;
5561
5562         if (!test_access(open->op_share_access, stp))
5563                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5564
5565         /* test and set deny mode */
5566         spin_lock(&fp->fi_lock);
5567         status = nfs4_file_check_deny(fp, open->op_share_deny);
5568         switch (status) {
5569         case nfs_ok:
5570                 set_deny(open->op_share_deny, stp);
5571                 fp->fi_share_deny |=
5572                         (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5573                 break;
5574         case nfserr_share_denied:
5575                 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5576                                 stp, open->op_share_deny, false))
5577                         status = nfserr_jukebox;
5578                 break;
5579         }
5580         spin_unlock(&fp->fi_lock);
5581
5582         if (status != nfs_ok)
5583                 return status;
5584
5585         status = nfsd4_truncate(rqstp, cur_fh, open);
5586         if (status != nfs_ok)
5587                 reset_union_bmap_deny(old_deny_bmap, stp);
5588         return status;
5589 }
5590
5591 /* Should we give out recallable state?: */
5592 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5593 {
5594         if (clp->cl_cb_state == NFSD4_CB_UP)
5595                 return true;
5596         /*
5597          * In the sessions case, since we don't have to establish a
5598          * separate connection for callbacks, we assume it's OK
5599          * until we hear otherwise:
5600          */
5601         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5602 }
5603
5604 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5605                                                 int flag)
5606 {
5607         struct file_lease *fl;
5608
5609         fl = locks_alloc_lease();
5610         if (!fl)
5611                 return NULL;
5612         fl->fl_lmops = &nfsd_lease_mng_ops;
5613         fl->c.flc_flags = FL_DELEG;
5614         fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5615         fl->c.flc_owner = (fl_owner_t)dp;
5616         fl->c.flc_pid = current->tgid;
5617         fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5618         return fl;
5619 }
5620
5621 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5622                                          struct nfs4_file *fp)
5623 {
5624         struct nfs4_ol_stateid *st;
5625         struct file *f = fp->fi_deleg_file->nf_file;
5626         struct inode *ino = file_inode(f);
5627         int writes;
5628
5629         writes = atomic_read(&ino->i_writecount);
5630         if (!writes)
5631                 return 0;
5632         /*
5633          * There could be multiple filehandles (hence multiple
5634          * nfs4_files) referencing this file, but that's not too
5635          * common; let's just give up in that case rather than
5636          * trying to go look up all the clients using that other
5637          * nfs4_file as well:
5638          */
5639         if (fp->fi_aliased)
5640                 return -EAGAIN;
5641         /*
5642          * If there's a close in progress, make sure that we see it
5643          * clear any fi_fds[] entries before we see it decrement
5644          * i_writecount:
5645          */
5646         smp_mb__after_atomic();
5647
5648         if (fp->fi_fds[O_WRONLY])
5649                 writes--;
5650         if (fp->fi_fds[O_RDWR])
5651                 writes--;
5652         if (writes > 0)
5653                 return -EAGAIN; /* There may be non-NFSv4 writers */
5654         /*
5655          * It's possible there are non-NFSv4 write opens in progress,
5656          * but if they haven't incremented i_writecount yet then they
5657          * also haven't called break lease yet; so, they'll break this
5658          * lease soon enough.  So, all that's left to check for is NFSv4
5659          * opens:
5660          */
5661         spin_lock(&fp->fi_lock);
5662         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5663                 if (st->st_openstp == NULL /* it's an open */ &&
5664                     access_permit_write(st) &&
5665                     st->st_stid.sc_client != clp) {
5666                         spin_unlock(&fp->fi_lock);
5667                         return -EAGAIN;
5668                 }
5669         }
5670         spin_unlock(&fp->fi_lock);
5671         /*
5672          * There's a small chance that we could be racing with another
5673          * NFSv4 open.  However, any open that hasn't added itself to
5674          * the fi_stateids list also hasn't called break_lease yet; so,
5675          * they'll break this lease soon enough.
5676          */
5677         return 0;
5678 }
5679
5680 /*
5681  * It's possible that between opening the dentry and setting the delegation,
5682  * that it has been renamed or unlinked. Redo the lookup to verify that this
5683  * hasn't happened.
5684  */
5685 static int
5686 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5687                           struct svc_fh *parent)
5688 {
5689         struct svc_export *exp;
5690         struct dentry *child;
5691         __be32 err;
5692
5693         err = nfsd_lookup_dentry(open->op_rqstp, parent,
5694                                  open->op_fname, open->op_fnamelen,
5695                                  &exp, &child);
5696
5697         if (err)
5698                 return -EAGAIN;
5699
5700         exp_put(exp);
5701         dput(child);
5702         if (child != file_dentry(fp->fi_deleg_file->nf_file))
5703                 return -EAGAIN;
5704
5705         return 0;
5706 }
5707
5708 /*
5709  * We avoid breaking delegations held by a client due to its own activity, but
5710  * clearing setuid/setgid bits on a write is an implicit activity and the client
5711  * may not notice and continue using the old mode. Avoid giving out a delegation
5712  * on setuid/setgid files when the client is requesting an open for write.
5713  */
5714 static int
5715 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5716 {
5717         struct inode *inode = file_inode(nf->nf_file);
5718
5719         if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5720             (inode->i_mode & (S_ISUID|S_ISGID)))
5721                 return -EAGAIN;
5722         return 0;
5723 }
5724
5725 static struct nfs4_delegation *
5726 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5727                     struct svc_fh *parent)
5728 {
5729         int status = 0;
5730         struct nfs4_client *clp = stp->st_stid.sc_client;
5731         struct nfs4_file *fp = stp->st_stid.sc_file;
5732         struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5733         struct nfs4_delegation *dp;
5734         struct nfsd_file *nf = NULL;
5735         struct file_lease *fl;
5736         u32 dl_type;
5737
5738         /*
5739          * The fi_had_conflict and nfs_get_existing_delegation checks
5740          * here are just optimizations; we'll need to recheck them at
5741          * the end:
5742          */
5743         if (fp->fi_had_conflict)
5744                 return ERR_PTR(-EAGAIN);
5745
5746         /*
5747          * Try for a write delegation first. RFC8881 section 10.4 says:
5748          *
5749          *  "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5750          *   on its own, all opens."
5751          *
5752          * Furthermore the client can use a write delegation for most READ
5753          * operations as well, so we require a O_RDWR file here.
5754          *
5755          * Offer a write delegation in the case of a BOTH open, and ensure
5756          * we get the O_RDWR descriptor.
5757          */
5758         if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5759                 nf = find_rw_file(fp);
5760                 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5761         }
5762
5763         /*
5764          * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5765          * file for some reason, then try for a read delegation instead.
5766          */
5767         if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5768                 nf = find_readable_file(fp);
5769                 dl_type = NFS4_OPEN_DELEGATE_READ;
5770         }
5771
5772         if (!nf)
5773                 return ERR_PTR(-EAGAIN);
5774
5775         spin_lock(&state_lock);
5776         spin_lock(&fp->fi_lock);
5777         if (nfs4_delegation_exists(clp, fp))
5778                 status = -EAGAIN;
5779         else if (nfsd4_verify_setuid_write(open, nf))
5780                 status = -EAGAIN;
5781         else if (!fp->fi_deleg_file) {
5782                 fp->fi_deleg_file = nf;
5783                 /* increment early to prevent fi_deleg_file from being
5784                  * cleared */
5785                 fp->fi_delegees = 1;
5786                 nf = NULL;
5787         } else
5788                 fp->fi_delegees++;
5789         spin_unlock(&fp->fi_lock);
5790         spin_unlock(&state_lock);
5791         if (nf)
5792                 nfsd_file_put(nf);
5793         if (status)
5794                 return ERR_PTR(status);
5795
5796         status = -ENOMEM;
5797         dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5798         if (!dp)
5799                 goto out_delegees;
5800
5801         fl = nfs4_alloc_init_lease(dp, dl_type);
5802         if (!fl)
5803                 goto out_clnt_odstate;
5804
5805         status = kernel_setlease(fp->fi_deleg_file->nf_file,
5806                                       fl->c.flc_type, &fl, NULL);
5807         if (fl)
5808                 locks_free_lease(fl);
5809         if (status)
5810                 goto out_clnt_odstate;
5811
5812         if (parent) {
5813                 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5814                 if (status)
5815                         goto out_unlock;
5816         }
5817
5818         status = nfsd4_check_conflicting_opens(clp, fp);
5819         if (status)
5820                 goto out_unlock;
5821
5822         /*
5823          * Now that the deleg is set, check again to ensure that nothing
5824          * raced in and changed the mode while we weren't lookng.
5825          */
5826         status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5827         if (status)
5828                 goto out_unlock;
5829
5830         status = -EAGAIN;
5831         if (fp->fi_had_conflict)
5832                 goto out_unlock;
5833
5834         spin_lock(&state_lock);
5835         spin_lock(&clp->cl_lock);
5836         spin_lock(&fp->fi_lock);
5837         status = hash_delegation_locked(dp, fp);
5838         spin_unlock(&fp->fi_lock);
5839         spin_unlock(&clp->cl_lock);
5840         spin_unlock(&state_lock);
5841
5842         if (status)
5843                 goto out_unlock;
5844
5845         return dp;
5846 out_unlock:
5847         kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5848 out_clnt_odstate:
5849         put_clnt_odstate(dp->dl_clnt_odstate);
5850         nfs4_put_stid(&dp->dl_stid);
5851 out_delegees:
5852         put_deleg_file(fp);
5853         return ERR_PTR(status);
5854 }
5855
5856 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5857 {
5858         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5859         if (status == -EAGAIN)
5860                 open->op_why_no_deleg = WND4_CONTENTION;
5861         else {
5862                 open->op_why_no_deleg = WND4_RESOURCE;
5863                 switch (open->op_deleg_want) {
5864                 case NFS4_SHARE_WANT_READ_DELEG:
5865                 case NFS4_SHARE_WANT_WRITE_DELEG:
5866                 case NFS4_SHARE_WANT_ANY_DELEG:
5867                         break;
5868                 case NFS4_SHARE_WANT_CANCEL:
5869                         open->op_why_no_deleg = WND4_CANCELLED;
5870                         break;
5871                 case NFS4_SHARE_WANT_NO_DELEG:
5872                         WARN_ON_ONCE(1);
5873                 }
5874         }
5875 }
5876
5877 /*
5878  * The Linux NFS server does not offer write delegations to NFSv4.0
5879  * clients in order to avoid conflicts between write delegations and
5880  * GETATTRs requesting CHANGE or SIZE attributes.
5881  *
5882  * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5883  * begins each COMPOUND contains a client ID. Delegation recall can
5884  * be avoided when the server recognizes the client sending a
5885  * GETATTR also holds write delegation it conflicts with.
5886  *
5887  * However, the NFSv4.0 protocol does not enable a server to
5888  * determine that a GETATTR originated from the client holding the
5889  * conflicting delegation versus coming from some other client. Per
5890  * RFC 7530 Section 16.7.5, the server must recall or send a
5891  * CB_GETATTR even when the GETATTR originates from the client that
5892  * holds the conflicting delegation.
5893  *
5894  * An NFSv4.0 client can trigger a pathological situation if it
5895  * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5896  * the same COMPOUND. COMPOUND execution will always stop at the
5897  * GETATTR and the DELEGRETURN will never get executed. The server
5898  * eventually revokes the delegation, which can result in loss of
5899  * open or lock state.
5900  */
5901 static void
5902 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5903                      struct svc_fh *currentfh)
5904 {
5905         struct nfs4_delegation *dp;
5906         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5907         struct nfs4_client *clp = stp->st_stid.sc_client;
5908         struct svc_fh *parent = NULL;
5909         int cb_up;
5910         int status = 0;
5911         struct kstat stat;
5912         struct path path;
5913
5914         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5915         open->op_recall = false;
5916         switch (open->op_claim_type) {
5917                 case NFS4_OPEN_CLAIM_PREVIOUS:
5918                         if (!cb_up)
5919                                 open->op_recall = true;
5920                         break;
5921                 case NFS4_OPEN_CLAIM_NULL:
5922                         parent = currentfh;
5923                         fallthrough;
5924                 case NFS4_OPEN_CLAIM_FH:
5925                         /*
5926                          * Let's not give out any delegations till everyone's
5927                          * had the chance to reclaim theirs, *and* until
5928                          * NLM locks have all been reclaimed:
5929                          */
5930                         if (locks_in_grace(clp->net))
5931                                 goto out_no_deleg;
5932                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5933                                 goto out_no_deleg;
5934                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5935                                         !clp->cl_minorversion)
5936                                 goto out_no_deleg;
5937                         break;
5938                 default:
5939                         goto out_no_deleg;
5940         }
5941         dp = nfs4_set_delegation(open, stp, parent);
5942         if (IS_ERR(dp))
5943                 goto out_no_deleg;
5944
5945         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5946
5947         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5948                 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5949                 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5950                 path.mnt = currentfh->fh_export->ex_path.mnt;
5951                 path.dentry = currentfh->fh_dentry;
5952                 if (vfs_getattr(&path, &stat,
5953                                 (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5954                                 AT_STATX_SYNC_AS_STAT)) {
5955                         nfs4_put_stid(&dp->dl_stid);
5956                         destroy_delegation(dp);
5957                         goto out_no_deleg;
5958                 }
5959                 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
5960                 dp->dl_cb_fattr.ncf_initial_cinfo =
5961                         nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
5962         } else {
5963                 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5964                 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5965         }
5966         nfs4_put_stid(&dp->dl_stid);
5967         return;
5968 out_no_deleg:
5969         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5970         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5971             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5972                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5973                 open->op_recall = true;
5974         }
5975
5976         /* 4.1 client asking for a delegation? */
5977         if (open->op_deleg_want)
5978                 nfsd4_open_deleg_none_ext(open, status);
5979         return;
5980 }
5981
5982 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5983                                         struct nfs4_delegation *dp)
5984 {
5985         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5986             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5987                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5988                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5989         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5990                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5991                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5992                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5993         }
5994         /* Otherwise the client must be confused wanting a delegation
5995          * it already has, therefore we don't return
5996          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5997          */
5998 }
5999
6000 /**
6001  * nfsd4_process_open2 - finish open processing
6002  * @rqstp: the RPC transaction being executed
6003  * @current_fh: NFSv4 COMPOUND's current filehandle
6004  * @open: OPEN arguments
6005  *
6006  * If successful, (1) truncate the file if open->op_truncate was
6007  * set, (2) set open->op_stateid, (3) set open->op_delegation.
6008  *
6009  * Returns %nfs_ok on success; otherwise an nfs4stat value in
6010  * network byte order is returned.
6011  */
6012 __be32
6013 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6014 {
6015         struct nfsd4_compoundres *resp = rqstp->rq_resp;
6016         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6017         struct nfs4_file *fp = NULL;
6018         struct nfs4_ol_stateid *stp = NULL;
6019         struct nfs4_delegation *dp = NULL;
6020         __be32 status;
6021         bool new_stp = false;
6022
6023         /*
6024          * Lookup file; if found, lookup stateid and check open request,
6025          * and check for delegations in the process of being recalled.
6026          * If not found, create the nfs4_file struct
6027          */
6028         fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6029         if (unlikely(!fp))
6030                 return nfserr_jukebox;
6031         if (fp != open->op_file) {
6032                 status = nfs4_check_deleg(cl, open, &dp);
6033                 if (status)
6034                         goto out;
6035                 stp = nfsd4_find_and_lock_existing_open(fp, open);
6036         } else {
6037                 open->op_file = NULL;
6038                 status = nfserr_bad_stateid;
6039                 if (nfsd4_is_deleg_cur(open))
6040                         goto out;
6041         }
6042
6043         if (!stp) {
6044                 stp = init_open_stateid(fp, open);
6045                 if (!open->op_stp)
6046                         new_stp = true;
6047         }
6048
6049         /*
6050          * OPEN the file, or upgrade an existing OPEN.
6051          * If truncate fails, the OPEN fails.
6052          *
6053          * stp is already locked.
6054          */
6055         if (!new_stp) {
6056                 /* Stateid was found, this is an OPEN upgrade */
6057                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6058                 if (status) {
6059                         mutex_unlock(&stp->st_mutex);
6060                         goto out;
6061                 }
6062         } else {
6063                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6064                 if (status) {
6065                         release_open_stateid(stp);
6066                         mutex_unlock(&stp->st_mutex);
6067                         goto out;
6068                 }
6069
6070                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6071                                                         open->op_odstate);
6072                 if (stp->st_clnt_odstate == open->op_odstate)
6073                         open->op_odstate = NULL;
6074         }
6075
6076         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6077         mutex_unlock(&stp->st_mutex);
6078
6079         if (nfsd4_has_session(&resp->cstate)) {
6080                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6081                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6082                         open->op_why_no_deleg = WND4_NOT_WANTED;
6083                         goto nodeleg;
6084                 }
6085         }
6086
6087         /*
6088         * Attempt to hand out a delegation. No error return, because the
6089         * OPEN succeeds even if we fail.
6090         */
6091         nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6092 nodeleg:
6093         status = nfs_ok;
6094         trace_nfsd_open(&stp->st_stid.sc_stateid);
6095 out:
6096         /* 4.1 client trying to upgrade/downgrade delegation? */
6097         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6098             open->op_deleg_want)
6099                 nfsd4_deleg_xgrade_none_ext(open, dp);
6100
6101         if (fp)
6102                 put_nfs4_file(fp);
6103         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6104                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6105         /*
6106         * To finish the open response, we just need to set the rflags.
6107         */
6108         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6109         if (nfsd4_has_session(&resp->cstate))
6110                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6111         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6112                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6113
6114         if (dp)
6115                 nfs4_put_stid(&dp->dl_stid);
6116         if (stp)
6117                 nfs4_put_stid(&stp->st_stid);
6118
6119         return status;
6120 }
6121
6122 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6123                               struct nfsd4_open *open)
6124 {
6125         if (open->op_openowner) {
6126                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
6127
6128                 nfsd4_cstate_assign_replay(cstate, so);
6129                 nfs4_put_stateowner(so);
6130         }
6131         if (open->op_file)
6132                 kmem_cache_free(file_slab, open->op_file);
6133         if (open->op_stp)
6134                 nfs4_put_stid(&open->op_stp->st_stid);
6135         if (open->op_odstate)
6136                 kmem_cache_free(odstate_slab, open->op_odstate);
6137 }
6138
6139 __be32
6140 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6141             union nfsd4_op_u *u)
6142 {
6143         clientid_t *clid = &u->renew;
6144         struct nfs4_client *clp;
6145         __be32 status;
6146         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6147
6148         trace_nfsd_clid_renew(clid);
6149         status = set_client(clid, cstate, nn);
6150         if (status)
6151                 return status;
6152         clp = cstate->clp;
6153         if (!list_empty(&clp->cl_delegations)
6154                         && clp->cl_cb_state != NFSD4_CB_UP)
6155                 return nfserr_cb_path_down;
6156         return nfs_ok;
6157 }
6158
6159 void
6160 nfsd4_end_grace(struct nfsd_net *nn)
6161 {
6162         /* do nothing if grace period already ended */
6163         if (nn->grace_ended)
6164                 return;
6165
6166         trace_nfsd_grace_complete(nn);
6167         nn->grace_ended = true;
6168         /*
6169          * If the server goes down again right now, an NFSv4
6170          * client will still be allowed to reclaim after it comes back up,
6171          * even if it hasn't yet had a chance to reclaim state this time.
6172          *
6173          */
6174         nfsd4_record_grace_done(nn);
6175         /*
6176          * At this point, NFSv4 clients can still reclaim.  But if the
6177          * server crashes, any that have not yet reclaimed will be out
6178          * of luck on the next boot.
6179          *
6180          * (NFSv4.1+ clients are considered to have reclaimed once they
6181          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
6182          * have reclaimed after their first OPEN.)
6183          */
6184         locks_end_grace(&nn->nfsd4_manager);
6185         /*
6186          * At this point, and once lockd and/or any other containers
6187          * exit their grace period, further reclaims will fail and
6188          * regular locking can resume.
6189          */
6190 }
6191
6192 /*
6193  * If we've waited a lease period but there are still clients trying to
6194  * reclaim, wait a little longer to give them a chance to finish.
6195  */
6196 static bool clients_still_reclaiming(struct nfsd_net *nn)
6197 {
6198         time64_t double_grace_period_end = nn->boot_time +
6199                                            2 * nn->nfsd4_lease;
6200
6201         if (nn->track_reclaim_completes &&
6202                         atomic_read(&nn->nr_reclaim_complete) ==
6203                         nn->reclaim_str_hashtbl_size)
6204                 return false;
6205         if (!nn->somebody_reclaimed)
6206                 return false;
6207         nn->somebody_reclaimed = false;
6208         /*
6209          * If we've given them *two* lease times to reclaim, and they're
6210          * still not done, give up:
6211          */
6212         if (ktime_get_boottime_seconds() > double_grace_period_end)
6213                 return false;
6214         return true;
6215 }
6216
6217 struct laundry_time {
6218         time64_t cutoff;
6219         time64_t new_timeo;
6220 };
6221
6222 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6223 {
6224         time64_t time_remaining;
6225
6226         if (last_refresh < lt->cutoff)
6227                 return true;
6228         time_remaining = last_refresh - lt->cutoff;
6229         lt->new_timeo = min(lt->new_timeo, time_remaining);
6230         return false;
6231 }
6232
6233 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6234 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6235 {
6236         spin_lock_init(&nn->nfsd_ssc_lock);
6237         INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6238         init_waitqueue_head(&nn->nfsd_ssc_waitq);
6239 }
6240 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
6241
6242 /*
6243  * This is called when nfsd is being shutdown, after all inter_ssc
6244  * cleanup were done, to destroy the ssc delayed unmount list.
6245  */
6246 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6247 {
6248         struct nfsd4_ssc_umount_item *ni = NULL;
6249         struct nfsd4_ssc_umount_item *tmp;
6250
6251         spin_lock(&nn->nfsd_ssc_lock);
6252         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6253                 list_del(&ni->nsui_list);
6254                 spin_unlock(&nn->nfsd_ssc_lock);
6255                 mntput(ni->nsui_vfsmount);
6256                 kfree(ni);
6257                 spin_lock(&nn->nfsd_ssc_lock);
6258         }
6259         spin_unlock(&nn->nfsd_ssc_lock);
6260 }
6261
6262 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6263 {
6264         bool do_wakeup = false;
6265         struct nfsd4_ssc_umount_item *ni = NULL;
6266         struct nfsd4_ssc_umount_item *tmp;
6267
6268         spin_lock(&nn->nfsd_ssc_lock);
6269         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6270                 if (time_after(jiffies, ni->nsui_expire)) {
6271                         if (refcount_read(&ni->nsui_refcnt) > 1)
6272                                 continue;
6273
6274                         /* mark being unmount */
6275                         ni->nsui_busy = true;
6276                         spin_unlock(&nn->nfsd_ssc_lock);
6277                         mntput(ni->nsui_vfsmount);
6278                         spin_lock(&nn->nfsd_ssc_lock);
6279
6280                         /* waiters need to start from begin of list */
6281                         list_del(&ni->nsui_list);
6282                         kfree(ni);
6283
6284                         /* wakeup ssc_connect waiters */
6285                         do_wakeup = true;
6286                         continue;
6287                 }
6288                 break;
6289         }
6290         if (do_wakeup)
6291                 wake_up_all(&nn->nfsd_ssc_waitq);
6292         spin_unlock(&nn->nfsd_ssc_lock);
6293 }
6294 #endif
6295
6296 /* Check if any lock belonging to this lockowner has any blockers */
6297 static bool
6298 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6299 {
6300         struct file_lock_context *ctx;
6301         struct nfs4_ol_stateid *stp;
6302         struct nfs4_file *nf;
6303
6304         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6305                 nf = stp->st_stid.sc_file;
6306                 ctx = locks_inode_context(nf->fi_inode);
6307                 if (!ctx)
6308                         continue;
6309                 if (locks_owner_has_blockers(ctx, lo))
6310                         return true;
6311         }
6312         return false;
6313 }
6314
6315 static bool
6316 nfs4_anylock_blockers(struct nfs4_client *clp)
6317 {
6318         int i;
6319         struct nfs4_stateowner *so;
6320         struct nfs4_lockowner *lo;
6321
6322         if (atomic_read(&clp->cl_delegs_in_recall))
6323                 return true;
6324         spin_lock(&clp->cl_lock);
6325         for (i = 0; i < OWNER_HASH_SIZE; i++) {
6326                 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6327                                 so_strhash) {
6328                         if (so->so_is_open_owner)
6329                                 continue;
6330                         lo = lockowner(so);
6331                         if (nfs4_lockowner_has_blockers(lo)) {
6332                                 spin_unlock(&clp->cl_lock);
6333                                 return true;
6334                         }
6335                 }
6336         }
6337         spin_unlock(&clp->cl_lock);
6338         return false;
6339 }
6340
6341 static void
6342 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6343                                 struct laundry_time *lt)
6344 {
6345         unsigned int maxreap, reapcnt = 0;
6346         struct list_head *pos, *next;
6347         struct nfs4_client *clp;
6348
6349         maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6350                         NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6351         INIT_LIST_HEAD(reaplist);
6352         spin_lock(&nn->client_lock);
6353         list_for_each_safe(pos, next, &nn->client_lru) {
6354                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6355                 if (clp->cl_state == NFSD4_EXPIRABLE)
6356                         goto exp_client;
6357                 if (!state_expired(lt, clp->cl_time))
6358                         break;
6359                 if (!atomic_read(&clp->cl_rpc_users)) {
6360                         if (clp->cl_state == NFSD4_ACTIVE)
6361                                 atomic_inc(&nn->nfsd_courtesy_clients);
6362                         clp->cl_state = NFSD4_COURTESY;
6363                 }
6364                 if (!client_has_state(clp))
6365                         goto exp_client;
6366                 if (!nfs4_anylock_blockers(clp))
6367                         if (reapcnt >= maxreap)
6368                                 continue;
6369 exp_client:
6370                 if (!mark_client_expired_locked(clp)) {
6371                         list_add(&clp->cl_lru, reaplist);
6372                         reapcnt++;
6373                 }
6374         }
6375         spin_unlock(&nn->client_lock);
6376 }
6377
6378 static void
6379 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6380                                 struct list_head *reaplist)
6381 {
6382         unsigned int maxreap = 0, reapcnt = 0;
6383         struct list_head *pos, *next;
6384         struct nfs4_client *clp;
6385
6386         maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6387         INIT_LIST_HEAD(reaplist);
6388
6389         spin_lock(&nn->client_lock);
6390         list_for_each_safe(pos, next, &nn->client_lru) {
6391                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6392                 if (clp->cl_state == NFSD4_ACTIVE)
6393                         break;
6394                 if (reapcnt >= maxreap)
6395                         break;
6396                 if (!mark_client_expired_locked(clp)) {
6397                         list_add(&clp->cl_lru, reaplist);
6398                         reapcnt++;
6399                 }
6400         }
6401         spin_unlock(&nn->client_lock);
6402 }
6403
6404 static void
6405 nfs4_process_client_reaplist(struct list_head *reaplist)
6406 {
6407         struct list_head *pos, *next;
6408         struct nfs4_client *clp;
6409
6410         list_for_each_safe(pos, next, reaplist) {
6411                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6412                 trace_nfsd_clid_purged(&clp->cl_clientid);
6413                 list_del_init(&clp->cl_lru);
6414                 expire_client(clp);
6415         }
6416 }
6417
6418 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6419                                       struct laundry_time *lt)
6420 {
6421         struct nfs4_client *clp;
6422
6423         spin_lock(&nn->client_lock);
6424         if (nn->nfs40_last_revoke == 0 ||
6425             nn->nfs40_last_revoke > lt->cutoff) {
6426                 spin_unlock(&nn->client_lock);
6427                 return;
6428         }
6429         nn->nfs40_last_revoke = 0;
6430
6431 retry:
6432         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6433                 unsigned long id, tmp;
6434                 struct nfs4_stid *stid;
6435
6436                 if (atomic_read(&clp->cl_admin_revoked) == 0)
6437                         continue;
6438
6439                 spin_lock(&clp->cl_lock);
6440                 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6441                         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6442                                 refcount_inc(&stid->sc_count);
6443                                 spin_unlock(&nn->client_lock);
6444                                 /* this function drops ->cl_lock */
6445                                 nfsd4_drop_revoked_stid(stid);
6446                                 nfs4_put_stid(stid);
6447                                 spin_lock(&nn->client_lock);
6448                                 goto retry;
6449                         }
6450                 spin_unlock(&clp->cl_lock);
6451         }
6452         spin_unlock(&nn->client_lock);
6453 }
6454
6455 static time64_t
6456 nfs4_laundromat(struct nfsd_net *nn)
6457 {
6458         struct nfs4_openowner *oo;
6459         struct nfs4_delegation *dp;
6460         struct nfs4_ol_stateid *stp;
6461         struct nfsd4_blocked_lock *nbl;
6462         struct list_head *pos, *next, reaplist;
6463         struct laundry_time lt = {
6464                 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6465                 .new_timeo = nn->nfsd4_lease
6466         };
6467         struct nfs4_cpntf_state *cps;
6468         copy_stateid_t *cps_t;
6469         int i;
6470
6471         if (clients_still_reclaiming(nn)) {
6472                 lt.new_timeo = 0;
6473                 goto out;
6474         }
6475         nfsd4_end_grace(nn);
6476
6477         spin_lock(&nn->s2s_cp_lock);
6478         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6479                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6480                 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6481                                 state_expired(&lt, cps->cpntf_time))
6482                         _free_cpntf_state_locked(nn, cps);
6483         }
6484         spin_unlock(&nn->s2s_cp_lock);
6485         nfs4_get_client_reaplist(nn, &reaplist, &lt);
6486         nfs4_process_client_reaplist(&reaplist);
6487
6488         nfs40_clean_admin_revoked(nn, &lt);
6489
6490         spin_lock(&state_lock);
6491         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6492                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6493                 if (!state_expired(&lt, dp->dl_time))
6494                         break;
6495                 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6496                 list_add(&dp->dl_recall_lru, &reaplist);
6497         }
6498         spin_unlock(&state_lock);
6499         while (!list_empty(&reaplist)) {
6500                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6501                                         dl_recall_lru);
6502                 list_del_init(&dp->dl_recall_lru);
6503                 revoke_delegation(dp);
6504         }
6505
6506         spin_lock(&nn->client_lock);
6507         while (!list_empty(&nn->close_lru)) {
6508                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6509                                         oo_close_lru);
6510                 if (!state_expired(&lt, oo->oo_time))
6511                         break;
6512                 list_del_init(&oo->oo_close_lru);
6513                 stp = oo->oo_last_closed_stid;
6514                 oo->oo_last_closed_stid = NULL;
6515                 spin_unlock(&nn->client_lock);
6516                 nfs4_put_stid(&stp->st_stid);
6517                 spin_lock(&nn->client_lock);
6518         }
6519         spin_unlock(&nn->client_lock);
6520
6521         /*
6522          * It's possible for a client to try and acquire an already held lock
6523          * that is being held for a long time, and then lose interest in it.
6524          * So, we clean out any un-revisited request after a lease period
6525          * under the assumption that the client is no longer interested.
6526          *
6527          * RFC5661, sec. 9.6 states that the client must not rely on getting
6528          * notifications and must continue to poll for locks, even when the
6529          * server supports them. Thus this shouldn't lead to clients blocking
6530          * indefinitely once the lock does become free.
6531          */
6532         BUG_ON(!list_empty(&reaplist));
6533         spin_lock(&nn->blocked_locks_lock);
6534         while (!list_empty(&nn->blocked_locks_lru)) {
6535                 nbl = list_first_entry(&nn->blocked_locks_lru,
6536                                         struct nfsd4_blocked_lock, nbl_lru);
6537                 if (!state_expired(&lt, nbl->nbl_time))
6538                         break;
6539                 list_move(&nbl->nbl_lru, &reaplist);
6540                 list_del_init(&nbl->nbl_list);
6541         }
6542         spin_unlock(&nn->blocked_locks_lock);
6543
6544         while (!list_empty(&reaplist)) {
6545                 nbl = list_first_entry(&reaplist,
6546                                         struct nfsd4_blocked_lock, nbl_lru);
6547                 list_del_init(&nbl->nbl_lru);
6548                 free_blocked_lock(nbl);
6549         }
6550 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6551         /* service the server-to-server copy delayed unmount list */
6552         nfsd4_ssc_expire_umount(nn);
6553 #endif
6554         if (atomic_long_read(&num_delegations) >= max_delegations)
6555                 deleg_reaper(nn);
6556 out:
6557         return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6558 }
6559
6560 static void laundromat_main(struct work_struct *);
6561
6562 static void
6563 laundromat_main(struct work_struct *laundry)
6564 {
6565         time64_t t;
6566         struct delayed_work *dwork = to_delayed_work(laundry);
6567         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6568                                            laundromat_work);
6569
6570         t = nfs4_laundromat(nn);
6571         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6572 }
6573
6574 static void
6575 courtesy_client_reaper(struct nfsd_net *nn)
6576 {
6577         struct list_head reaplist;
6578
6579         nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6580         nfs4_process_client_reaplist(&reaplist);
6581 }
6582
6583 static void
6584 deleg_reaper(struct nfsd_net *nn)
6585 {
6586         struct list_head *pos, *next;
6587         struct nfs4_client *clp;
6588         struct list_head cblist;
6589
6590         INIT_LIST_HEAD(&cblist);
6591         spin_lock(&nn->client_lock);
6592         list_for_each_safe(pos, next, &nn->client_lru) {
6593                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6594                 if (clp->cl_state != NFSD4_ACTIVE ||
6595                         list_empty(&clp->cl_delegations) ||
6596                         atomic_read(&clp->cl_delegs_in_recall) ||
6597                         test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6598                         (ktime_get_boottime_seconds() -
6599                                 clp->cl_ra_time < 5)) {
6600                         continue;
6601                 }
6602                 list_add(&clp->cl_ra_cblist, &cblist);
6603
6604                 /* release in nfsd4_cb_recall_any_release */
6605                 atomic_inc(&clp->cl_rpc_users);
6606                 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6607                 clp->cl_ra_time = ktime_get_boottime_seconds();
6608         }
6609         spin_unlock(&nn->client_lock);
6610
6611         while (!list_empty(&cblist)) {
6612                 clp = list_first_entry(&cblist, struct nfs4_client,
6613                                         cl_ra_cblist);
6614                 list_del_init(&clp->cl_ra_cblist);
6615                 clp->cl_ra->ra_keep = 0;
6616                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6617                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6618                                                 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6619                 trace_nfsd_cb_recall_any(clp->cl_ra);
6620                 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6621         }
6622 }
6623
6624 static void
6625 nfsd4_state_shrinker_worker(struct work_struct *work)
6626 {
6627         struct nfsd_net *nn = container_of(work, struct nfsd_net,
6628                                 nfsd_shrinker_work);
6629
6630         courtesy_client_reaper(nn);
6631         deleg_reaper(nn);
6632 }
6633
6634 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6635 {
6636         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6637                 return nfserr_bad_stateid;
6638         return nfs_ok;
6639 }
6640
6641 static
6642 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6643 {
6644         __be32 status = nfserr_openmode;
6645
6646         /* For lock stateid's, we test the parent open, not the lock: */
6647         if (stp->st_openstp)
6648                 stp = stp->st_openstp;
6649         if ((flags & WR_STATE) && !access_permit_write(stp))
6650                 goto out;
6651         if ((flags & RD_STATE) && !access_permit_read(stp))
6652                 goto out;
6653         status = nfs_ok;
6654 out:
6655         return status;
6656 }
6657
6658 static inline __be32
6659 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6660 {
6661         if (ONE_STATEID(stateid) && (flags & RD_STATE))
6662                 return nfs_ok;
6663         else if (opens_in_grace(net)) {
6664                 /* Answer in remaining cases depends on existence of
6665                  * conflicting state; so we must wait out the grace period. */
6666                 return nfserr_grace;
6667         } else if (flags & WR_STATE)
6668                 return nfs4_share_conflict(current_fh,
6669                                 NFS4_SHARE_DENY_WRITE);
6670         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6671                 return nfs4_share_conflict(current_fh,
6672                                 NFS4_SHARE_DENY_READ);
6673 }
6674
6675 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6676 {
6677         /*
6678          * When sessions are used the stateid generation number is ignored
6679          * when it is zero.
6680          */
6681         if (has_session && in->si_generation == 0)
6682                 return nfs_ok;
6683
6684         if (in->si_generation == ref->si_generation)
6685                 return nfs_ok;
6686
6687         /* If the client sends us a stateid from the future, it's buggy: */
6688         if (nfsd4_stateid_generation_after(in, ref))
6689                 return nfserr_bad_stateid;
6690         /*
6691          * However, we could see a stateid from the past, even from a
6692          * non-buggy client.  For example, if the client sends a lock
6693          * while some IO is outstanding, the lock may bump si_generation
6694          * while the IO is still in flight.  The client could avoid that
6695          * situation by waiting for responses on all the IO requests,
6696          * but better performance may result in retrying IO that
6697          * receives an old_stateid error if requests are rarely
6698          * reordered in flight:
6699          */
6700         return nfserr_old_stateid;
6701 }
6702
6703 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6704 {
6705         __be32 ret;
6706
6707         spin_lock(&s->sc_lock);
6708         ret = nfsd4_verify_open_stid(s);
6709         if (ret == nfs_ok)
6710                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6711         spin_unlock(&s->sc_lock);
6712         if (ret == nfserr_admin_revoked)
6713                 nfsd40_drop_revoked_stid(s->sc_client,
6714                                         &s->sc_stateid);
6715         return ret;
6716 }
6717
6718 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6719 {
6720         if (ols->st_stateowner->so_is_open_owner &&
6721             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6722                 return nfserr_bad_stateid;
6723         return nfs_ok;
6724 }
6725
6726 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6727 {
6728         struct nfs4_stid *s;
6729         __be32 status = nfserr_bad_stateid;
6730
6731         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6732                 CLOSE_STATEID(stateid))
6733                 return status;
6734         spin_lock(&cl->cl_lock);
6735         s = find_stateid_locked(cl, stateid);
6736         if (!s)
6737                 goto out_unlock;
6738         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6739         if (status)
6740                 goto out_unlock;
6741         status = nfsd4_verify_open_stid(s);
6742         if (status)
6743                 goto out_unlock;
6744
6745         switch (s->sc_type) {
6746         case SC_TYPE_DELEG:
6747                 status = nfs_ok;
6748                 break;
6749         case SC_TYPE_OPEN:
6750         case SC_TYPE_LOCK:
6751                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6752                 break;
6753         default:
6754                 printk("unknown stateid type %x\n", s->sc_type);
6755                 status = nfserr_bad_stateid;
6756         }
6757 out_unlock:
6758         spin_unlock(&cl->cl_lock);
6759         if (status == nfserr_admin_revoked)
6760                 nfsd40_drop_revoked_stid(cl, stateid);
6761         return status;
6762 }
6763
6764 __be32
6765 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6766                      stateid_t *stateid,
6767                      unsigned short typemask, unsigned short statusmask,
6768                      struct nfs4_stid **s, struct nfsd_net *nn)
6769 {
6770         __be32 status;
6771         struct nfs4_stid *stid;
6772         bool return_revoked = false;
6773
6774         /*
6775          *  only return revoked delegations if explicitly asked.
6776          *  otherwise we report revoked or bad_stateid status.
6777          */
6778         if (statusmask & SC_STATUS_REVOKED)
6779                 return_revoked = true;
6780         if (typemask & SC_TYPE_DELEG)
6781                 /* Always allow REVOKED for DELEG so we can
6782                  * retturn the appropriate error.
6783                  */
6784                 statusmask |= SC_STATUS_REVOKED;
6785
6786         statusmask |= SC_STATUS_ADMIN_REVOKED;
6787
6788         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6789                 CLOSE_STATEID(stateid))
6790                 return nfserr_bad_stateid;
6791         status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6792         if (status == nfserr_stale_clientid) {
6793                 if (cstate->session)
6794                         return nfserr_bad_stateid;
6795                 return nfserr_stale_stateid;
6796         }
6797         if (status)
6798                 return status;
6799         stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6800         if (!stid)
6801                 return nfserr_bad_stateid;
6802         if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6803                 nfs4_put_stid(stid);
6804                 return nfserr_deleg_revoked;
6805         }
6806         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6807                 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6808                 nfs4_put_stid(stid);
6809                 return nfserr_admin_revoked;
6810         }
6811         *s = stid;
6812         return nfs_ok;
6813 }
6814
6815 static struct nfsd_file *
6816 nfs4_find_file(struct nfs4_stid *s, int flags)
6817 {
6818         struct nfsd_file *ret = NULL;
6819
6820         if (!s || s->sc_status)
6821                 return NULL;
6822
6823         switch (s->sc_type) {
6824         case SC_TYPE_DELEG:
6825                 spin_lock(&s->sc_file->fi_lock);
6826                 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6827                 spin_unlock(&s->sc_file->fi_lock);
6828                 break;
6829         case SC_TYPE_OPEN:
6830         case SC_TYPE_LOCK:
6831                 if (flags & RD_STATE)
6832                         ret = find_readable_file(s->sc_file);
6833                 else
6834                         ret = find_writeable_file(s->sc_file);
6835         }
6836
6837         return ret;
6838 }
6839
6840 static __be32
6841 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6842 {
6843         __be32 status;
6844
6845         status = nfsd4_check_openowner_confirmed(ols);
6846         if (status)
6847                 return status;
6848         return nfs4_check_openmode(ols, flags);
6849 }
6850
6851 static __be32
6852 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6853                 struct nfsd_file **nfp, int flags)
6854 {
6855         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6856         struct nfsd_file *nf;
6857         __be32 status;
6858
6859         nf = nfs4_find_file(s, flags);
6860         if (nf) {
6861                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6862                                 acc | NFSD_MAY_OWNER_OVERRIDE);
6863                 if (status) {
6864                         nfsd_file_put(nf);
6865                         goto out;
6866                 }
6867         } else {
6868                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6869                 if (status)
6870                         return status;
6871         }
6872         *nfp = nf;
6873 out:
6874         return status;
6875 }
6876 static void
6877 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6878 {
6879         WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6880         if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6881                 return;
6882         list_del(&cps->cp_list);
6883         idr_remove(&nn->s2s_cp_stateids,
6884                    cps->cp_stateid.cs_stid.si_opaque.so_id);
6885         kfree(cps);
6886 }
6887 /*
6888  * A READ from an inter server to server COPY will have a
6889  * copy stateid. Look up the copy notify stateid from the
6890  * idr structure and take a reference on it.
6891  */
6892 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6893                           struct nfs4_client *clp,
6894                           struct nfs4_cpntf_state **cps)
6895 {
6896         copy_stateid_t *cps_t;
6897         struct nfs4_cpntf_state *state = NULL;
6898
6899         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6900                 return nfserr_bad_stateid;
6901         spin_lock(&nn->s2s_cp_lock);
6902         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6903         if (cps_t) {
6904                 state = container_of(cps_t, struct nfs4_cpntf_state,
6905                                      cp_stateid);
6906                 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6907                         state = NULL;
6908                         goto unlock;
6909                 }
6910                 if (!clp)
6911                         refcount_inc(&state->cp_stateid.cs_count);
6912                 else
6913                         _free_cpntf_state_locked(nn, state);
6914         }
6915 unlock:
6916         spin_unlock(&nn->s2s_cp_lock);
6917         if (!state)
6918                 return nfserr_bad_stateid;
6919         if (!clp)
6920                 *cps = state;
6921         return 0;
6922 }
6923
6924 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6925                                struct nfs4_stid **stid)
6926 {
6927         __be32 status;
6928         struct nfs4_cpntf_state *cps = NULL;
6929         struct nfs4_client *found;
6930
6931         status = manage_cpntf_state(nn, st, NULL, &cps);
6932         if (status)
6933                 return status;
6934
6935         cps->cpntf_time = ktime_get_boottime_seconds();
6936
6937         status = nfserr_expired;
6938         found = lookup_clientid(&cps->cp_p_clid, true, nn);
6939         if (!found)
6940                 goto out;
6941
6942         *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6943                                      SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6944                                      0);
6945         if (*stid)
6946                 status = nfs_ok;
6947         else
6948                 status = nfserr_bad_stateid;
6949
6950         put_client_renew(found);
6951 out:
6952         nfs4_put_cpntf_state(nn, cps);
6953         return status;
6954 }
6955
6956 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6957 {
6958         spin_lock(&nn->s2s_cp_lock);
6959         _free_cpntf_state_locked(nn, cps);
6960         spin_unlock(&nn->s2s_cp_lock);
6961 }
6962
6963 /**
6964  * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6965  * @rqstp: incoming request from client
6966  * @cstate: current compound state
6967  * @fhp: filehandle associated with requested stateid
6968  * @stateid: stateid (provided by client)
6969  * @flags: flags describing type of operation to be done
6970  * @nfp: optional nfsd_file return pointer (may be NULL)
6971  * @cstid: optional returned nfs4_stid pointer (may be NULL)
6972  *
6973  * Given info from the client, look up a nfs4_stid for the operation. On
6974  * success, it returns a reference to the nfs4_stid and/or the nfsd_file
6975  * associated with it.
6976  */
6977 __be32
6978 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6979                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6980                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6981                 struct nfs4_stid **cstid)
6982 {
6983         struct net *net = SVC_NET(rqstp);
6984         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6985         struct nfs4_stid *s = NULL;
6986         __be32 status;
6987
6988         if (nfp)
6989                 *nfp = NULL;
6990
6991         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6992                 if (cstid)
6993                         status = nfserr_bad_stateid;
6994                 else
6995                         status = check_special_stateids(net, fhp, stateid,
6996                                                                         flags);
6997                 goto done;
6998         }
6999
7000         status = nfsd4_lookup_stateid(cstate, stateid,
7001                                 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7002                                 0, &s, nn);
7003         if (status == nfserr_bad_stateid)
7004                 status = find_cpntf_state(nn, stateid, &s);
7005         if (status)
7006                 return status;
7007         status = nfsd4_stid_check_stateid_generation(stateid, s,
7008                         nfsd4_has_session(cstate));
7009         if (status)
7010                 goto out;
7011
7012         switch (s->sc_type) {
7013         case SC_TYPE_DELEG:
7014                 status = nfs4_check_delegmode(delegstateid(s), flags);
7015                 break;
7016         case SC_TYPE_OPEN:
7017         case SC_TYPE_LOCK:
7018                 status = nfs4_check_olstateid(openlockstateid(s), flags);
7019                 break;
7020         }
7021         if (status)
7022                 goto out;
7023         status = nfs4_check_fh(fhp, s);
7024
7025 done:
7026         if (status == nfs_ok && nfp)
7027                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7028 out:
7029         if (s) {
7030                 if (!status && cstid)
7031                         *cstid = s;
7032                 else
7033                         nfs4_put_stid(s);
7034         }
7035         return status;
7036 }
7037
7038 /*
7039  * Test if the stateid is valid
7040  */
7041 __be32
7042 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7043                    union nfsd4_op_u *u)
7044 {
7045         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7046         struct nfsd4_test_stateid_id *stateid;
7047         struct nfs4_client *cl = cstate->clp;
7048
7049         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7050                 stateid->ts_id_status =
7051                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7052
7053         return nfs_ok;
7054 }
7055
7056 static __be32
7057 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7058 {
7059         struct nfs4_ol_stateid *stp = openlockstateid(s);
7060         __be32 ret;
7061
7062         ret = nfsd4_lock_ol_stateid(stp);
7063         if (ret)
7064                 goto out_put_stid;
7065
7066         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7067         if (ret)
7068                 goto out;
7069
7070         ret = nfserr_locks_held;
7071         if (check_for_locks(stp->st_stid.sc_file,
7072                             lockowner(stp->st_stateowner)))
7073                 goto out;
7074
7075         release_lock_stateid(stp);
7076         ret = nfs_ok;
7077
7078 out:
7079         mutex_unlock(&stp->st_mutex);
7080 out_put_stid:
7081         nfs4_put_stid(s);
7082         return ret;
7083 }
7084
7085 __be32
7086 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7087                    union nfsd4_op_u *u)
7088 {
7089         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7090         stateid_t *stateid = &free_stateid->fr_stateid;
7091         struct nfs4_stid *s;
7092         struct nfs4_delegation *dp;
7093         struct nfs4_client *cl = cstate->clp;
7094         __be32 ret = nfserr_bad_stateid;
7095
7096         spin_lock(&cl->cl_lock);
7097         s = find_stateid_locked(cl, stateid);
7098         if (!s || s->sc_status & SC_STATUS_CLOSED)
7099                 goto out_unlock;
7100         if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7101                 nfsd4_drop_revoked_stid(s);
7102                 ret = nfs_ok;
7103                 goto out;
7104         }
7105         spin_lock(&s->sc_lock);
7106         switch (s->sc_type) {
7107         case SC_TYPE_DELEG:
7108                 if (s->sc_status & SC_STATUS_REVOKED) {
7109                         spin_unlock(&s->sc_lock);
7110                         dp = delegstateid(s);
7111                         list_del_init(&dp->dl_recall_lru);
7112                         spin_unlock(&cl->cl_lock);
7113                         nfs4_put_stid(s);
7114                         ret = nfs_ok;
7115                         goto out;
7116                 }
7117                 ret = nfserr_locks_held;
7118                 break;
7119         case SC_TYPE_OPEN:
7120                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7121                 if (ret)
7122                         break;
7123                 ret = nfserr_locks_held;
7124                 break;
7125         case SC_TYPE_LOCK:
7126                 spin_unlock(&s->sc_lock);
7127                 refcount_inc(&s->sc_count);
7128                 spin_unlock(&cl->cl_lock);
7129                 ret = nfsd4_free_lock_stateid(stateid, s);
7130                 goto out;
7131         }
7132         spin_unlock(&s->sc_lock);
7133 out_unlock:
7134         spin_unlock(&cl->cl_lock);
7135 out:
7136         return ret;
7137 }
7138
7139 static inline int
7140 setlkflg (int type)
7141 {
7142         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7143                 RD_STATE : WR_STATE;
7144 }
7145
7146 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7147 {
7148         struct svc_fh *current_fh = &cstate->current_fh;
7149         struct nfs4_stateowner *sop = stp->st_stateowner;
7150         __be32 status;
7151
7152         status = nfsd4_check_seqid(cstate, sop, seqid);
7153         if (status)
7154                 return status;
7155         status = nfsd4_lock_ol_stateid(stp);
7156         if (status != nfs_ok)
7157                 return status;
7158         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7159         if (status == nfs_ok)
7160                 status = nfs4_check_fh(current_fh, &stp->st_stid);
7161         if (status != nfs_ok)
7162                 mutex_unlock(&stp->st_mutex);
7163         return status;
7164 }
7165
7166 /**
7167  * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7168  * @cstate: compund state
7169  * @seqid: seqid (provided by client)
7170  * @stateid: stateid (provided by client)
7171  * @typemask: mask of allowable types for this operation
7172  * @statusmask: mask of allowed states: 0 or STID_CLOSED
7173  * @stpp: return pointer for the stateid found
7174  * @nn: net namespace for request
7175  *
7176  * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7177  * return it in @stpp. On a nfs_ok return, the returned stateid will
7178  * have its st_mutex locked.
7179  */
7180 static __be32
7181 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7182                          stateid_t *stateid,
7183                          unsigned short typemask, unsigned short statusmask,
7184                          struct nfs4_ol_stateid **stpp,
7185                          struct nfsd_net *nn)
7186 {
7187         __be32 status;
7188         struct nfs4_stid *s;
7189         struct nfs4_ol_stateid *stp = NULL;
7190
7191         trace_nfsd_preprocess(seqid, stateid);
7192
7193         *stpp = NULL;
7194         status = nfsd4_lookup_stateid(cstate, stateid,
7195                                       typemask, statusmask, &s, nn);
7196         if (status)
7197                 return status;
7198         stp = openlockstateid(s);
7199         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
7200
7201         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7202         if (!status)
7203                 *stpp = stp;
7204         else
7205                 nfs4_put_stid(&stp->st_stid);
7206         return status;
7207 }
7208
7209 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7210                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7211 {
7212         __be32 status;
7213         struct nfs4_openowner *oo;
7214         struct nfs4_ol_stateid *stp;
7215
7216         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7217                                           SC_TYPE_OPEN, 0, &stp, nn);
7218         if (status)
7219                 return status;
7220         oo = openowner(stp->st_stateowner);
7221         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7222                 mutex_unlock(&stp->st_mutex);
7223                 nfs4_put_stid(&stp->st_stid);
7224                 return nfserr_bad_stateid;
7225         }
7226         *stpp = stp;
7227         return nfs_ok;
7228 }
7229
7230 __be32
7231 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7232                    union nfsd4_op_u *u)
7233 {
7234         struct nfsd4_open_confirm *oc = &u->open_confirm;
7235         __be32 status;
7236         struct nfs4_openowner *oo;
7237         struct nfs4_ol_stateid *stp;
7238         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7239
7240         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7241                         cstate->current_fh.fh_dentry);
7242
7243         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7244         if (status)
7245                 return status;
7246
7247         status = nfs4_preprocess_seqid_op(cstate,
7248                                           oc->oc_seqid, &oc->oc_req_stateid,
7249                                           SC_TYPE_OPEN, 0, &stp, nn);
7250         if (status)
7251                 goto out;
7252         oo = openowner(stp->st_stateowner);
7253         status = nfserr_bad_stateid;
7254         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7255                 mutex_unlock(&stp->st_mutex);
7256                 goto put_stateid;
7257         }
7258         oo->oo_flags |= NFS4_OO_CONFIRMED;
7259         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7260         mutex_unlock(&stp->st_mutex);
7261         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7262         nfsd4_client_record_create(oo->oo_owner.so_client);
7263         status = nfs_ok;
7264 put_stateid:
7265         nfs4_put_stid(&stp->st_stid);
7266 out:
7267         nfsd4_bump_seqid(cstate, status);
7268         return status;
7269 }
7270
7271 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7272 {
7273         if (!test_access(access, stp))
7274                 return;
7275         nfs4_file_put_access(stp->st_stid.sc_file, access);
7276         clear_access(access, stp);
7277 }
7278
7279 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7280 {
7281         switch (to_access) {
7282         case NFS4_SHARE_ACCESS_READ:
7283                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7284                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7285                 break;
7286         case NFS4_SHARE_ACCESS_WRITE:
7287                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7288                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7289                 break;
7290         case NFS4_SHARE_ACCESS_BOTH:
7291                 break;
7292         default:
7293                 WARN_ON_ONCE(1);
7294         }
7295 }
7296
7297 __be32
7298 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7299                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7300 {
7301         struct nfsd4_open_downgrade *od = &u->open_downgrade;
7302         __be32 status;
7303         struct nfs4_ol_stateid *stp;
7304         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7305
7306         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7307                         cstate->current_fh.fh_dentry);
7308
7309         /* We don't yet support WANT bits: */
7310         if (od->od_deleg_want)
7311                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7312                         od->od_deleg_want);
7313
7314         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7315                                         &od->od_stateid, &stp, nn);
7316         if (status)
7317                 goto out;
7318         status = nfserr_inval;
7319         if (!test_access(od->od_share_access, stp)) {
7320                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7321                         stp->st_access_bmap, od->od_share_access);
7322                 goto put_stateid;
7323         }
7324         if (!test_deny(od->od_share_deny, stp)) {
7325                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7326                         stp->st_deny_bmap, od->od_share_deny);
7327                 goto put_stateid;
7328         }
7329         nfs4_stateid_downgrade(stp, od->od_share_access);
7330         reset_union_bmap_deny(od->od_share_deny, stp);
7331         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7332         status = nfs_ok;
7333 put_stateid:
7334         mutex_unlock(&stp->st_mutex);
7335         nfs4_put_stid(&stp->st_stid);
7336 out:
7337         nfsd4_bump_seqid(cstate, status);
7338         return status;
7339 }
7340
7341 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7342 {
7343         struct nfs4_client *clp = s->st_stid.sc_client;
7344         bool unhashed;
7345         LIST_HEAD(reaplist);
7346         struct nfs4_ol_stateid *stp;
7347
7348         spin_lock(&clp->cl_lock);
7349         unhashed = unhash_open_stateid(s, &reaplist);
7350
7351         if (clp->cl_minorversion) {
7352                 if (unhashed)
7353                         put_ol_stateid_locked(s, &reaplist);
7354                 spin_unlock(&clp->cl_lock);
7355                 list_for_each_entry(stp, &reaplist, st_locks)
7356                         nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7357                 free_ol_stateid_reaplist(&reaplist);
7358         } else {
7359                 spin_unlock(&clp->cl_lock);
7360                 free_ol_stateid_reaplist(&reaplist);
7361                 if (unhashed)
7362                         move_to_close_lru(s, clp->net);
7363         }
7364 }
7365
7366 /*
7367  * nfs4_unlock_state() called after encode
7368  */
7369 __be32
7370 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7371                 union nfsd4_op_u *u)
7372 {
7373         struct nfsd4_close *close = &u->close;
7374         __be32 status;
7375         struct nfs4_ol_stateid *stp;
7376         struct net *net = SVC_NET(rqstp);
7377         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7378
7379         dprintk("NFSD: nfsd4_close on file %pd\n",
7380                         cstate->current_fh.fh_dentry);
7381
7382         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7383                                           &close->cl_stateid,
7384                                           SC_TYPE_OPEN, SC_STATUS_CLOSED,
7385                                           &stp, nn);
7386         nfsd4_bump_seqid(cstate, status);
7387         if (status)
7388                 goto out;
7389
7390         spin_lock(&stp->st_stid.sc_client->cl_lock);
7391         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7392         spin_unlock(&stp->st_stid.sc_client->cl_lock);
7393
7394         /*
7395          * Technically we don't _really_ have to increment or copy it, since
7396          * it should just be gone after this operation and we clobber the
7397          * copied value below, but we continue to do so here just to ensure
7398          * that racing ops see that there was a state change.
7399          */
7400         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7401
7402         nfsd4_close_open_stateid(stp);
7403         mutex_unlock(&stp->st_mutex);
7404
7405         /* v4.1+ suggests that we send a special stateid in here, since the
7406          * clients should just ignore this anyway. Since this is not useful
7407          * for v4.0 clients either, we set it to the special close_stateid
7408          * universally.
7409          *
7410          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7411          */
7412         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7413
7414         /* put reference from nfs4_preprocess_seqid_op */
7415         nfs4_put_stid(&stp->st_stid);
7416 out:
7417         return status;
7418 }
7419
7420 __be32
7421 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7422                   union nfsd4_op_u *u)
7423 {
7424         struct nfsd4_delegreturn *dr = &u->delegreturn;
7425         struct nfs4_delegation *dp;
7426         stateid_t *stateid = &dr->dr_stateid;
7427         struct nfs4_stid *s;
7428         __be32 status;
7429         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7430
7431         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7432                 return status;
7433
7434         status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7435         if (status)
7436                 goto out;
7437         dp = delegstateid(s);
7438         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7439         if (status)
7440                 goto put_stateid;
7441
7442         trace_nfsd_deleg_return(stateid);
7443         wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7444         destroy_delegation(dp);
7445 put_stateid:
7446         nfs4_put_stid(&dp->dl_stid);
7447 out:
7448         return status;
7449 }
7450
7451 /* last octet in a range */
7452 static inline u64
7453 last_byte_offset(u64 start, u64 len)
7454 {
7455         u64 end;
7456
7457         WARN_ON_ONCE(!len);
7458         end = start + len;
7459         return end > start ? end - 1: NFS4_MAX_UINT64;
7460 }
7461
7462 /*
7463  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7464  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7465  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
7466  * locking, this prevents us from being completely protocol-compliant.  The
7467  * real solution to this problem is to start using unsigned file offsets in
7468  * the VFS, but this is a very deep change!
7469  */
7470 static inline void
7471 nfs4_transform_lock_offset(struct file_lock *lock)
7472 {
7473         if (lock->fl_start < 0)
7474                 lock->fl_start = OFFSET_MAX;
7475         if (lock->fl_end < 0)
7476                 lock->fl_end = OFFSET_MAX;
7477 }
7478
7479 static fl_owner_t
7480 nfsd4_lm_get_owner(fl_owner_t owner)
7481 {
7482         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7483
7484         nfs4_get_stateowner(&lo->lo_owner);
7485         return owner;
7486 }
7487
7488 static void
7489 nfsd4_lm_put_owner(fl_owner_t owner)
7490 {
7491         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7492
7493         if (lo)
7494                 nfs4_put_stateowner(&lo->lo_owner);
7495 }
7496
7497 /* return pointer to struct nfs4_client if client is expirable */
7498 static bool
7499 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7500 {
7501         struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7502         struct nfs4_client *clp = lo->lo_owner.so_client;
7503         struct nfsd_net *nn;
7504
7505         if (try_to_expire_client(clp)) {
7506                 nn = net_generic(clp->net, nfsd_net_id);
7507                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7508                 return true;
7509         }
7510         return false;
7511 }
7512
7513 /* schedule laundromat to run immediately and wait for it to complete */
7514 static void
7515 nfsd4_lm_expire_lock(void)
7516 {
7517         flush_workqueue(laundry_wq);
7518 }
7519
7520 static void
7521 nfsd4_lm_notify(struct file_lock *fl)
7522 {
7523         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7524         struct net                      *net = lo->lo_owner.so_client->net;
7525         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
7526         struct nfsd4_blocked_lock       *nbl = container_of(fl,
7527                                                 struct nfsd4_blocked_lock, nbl_lock);
7528         bool queue = false;
7529
7530         /* An empty list means that something else is going to be using it */
7531         spin_lock(&nn->blocked_locks_lock);
7532         if (!list_empty(&nbl->nbl_list)) {
7533                 list_del_init(&nbl->nbl_list);
7534                 list_del_init(&nbl->nbl_lru);
7535                 queue = true;
7536         }
7537         spin_unlock(&nn->blocked_locks_lock);
7538
7539         if (queue) {
7540                 trace_nfsd_cb_notify_lock(lo, nbl);
7541                 nfsd4_run_cb(&nbl->nbl_cb);
7542         }
7543 }
7544
7545 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
7546         .lm_mod_owner = THIS_MODULE,
7547         .lm_notify = nfsd4_lm_notify,
7548         .lm_get_owner = nfsd4_lm_get_owner,
7549         .lm_put_owner = nfsd4_lm_put_owner,
7550         .lm_lock_expirable = nfsd4_lm_lock_expirable,
7551         .lm_expire_lock = nfsd4_lm_expire_lock,
7552 };
7553
7554 static inline void
7555 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7556 {
7557         struct nfs4_lockowner *lo;
7558
7559         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7560                 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7561                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7562                                                 GFP_KERNEL);
7563                 if (!deny->ld_owner.data)
7564                         /* We just don't care that much */
7565                         goto nevermind;
7566                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7567         } else {
7568 nevermind:
7569                 deny->ld_owner.len = 0;
7570                 deny->ld_owner.data = NULL;
7571                 deny->ld_clientid.cl_boot = 0;
7572                 deny->ld_clientid.cl_id = 0;
7573         }
7574         deny->ld_start = fl->fl_start;
7575         deny->ld_length = NFS4_MAX_UINT64;
7576         if (fl->fl_end != NFS4_MAX_UINT64)
7577                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7578         deny->ld_type = NFS4_READ_LT;
7579         if (fl->c.flc_type != F_RDLCK)
7580                 deny->ld_type = NFS4_WRITE_LT;
7581 }
7582
7583 static struct nfs4_lockowner *
7584 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7585 {
7586         unsigned int strhashval = ownerstr_hashval(owner);
7587         struct nfs4_stateowner *so;
7588
7589         lockdep_assert_held(&clp->cl_lock);
7590
7591         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7592                             so_strhash) {
7593                 if (so->so_is_open_owner)
7594                         continue;
7595                 if (same_owner_str(so, owner))
7596                         return lockowner(nfs4_get_stateowner(so));
7597         }
7598         return NULL;
7599 }
7600
7601 static struct nfs4_lockowner *
7602 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7603 {
7604         struct nfs4_lockowner *lo;
7605
7606         spin_lock(&clp->cl_lock);
7607         lo = find_lockowner_str_locked(clp, owner);
7608         spin_unlock(&clp->cl_lock);
7609         return lo;
7610 }
7611
7612 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7613 {
7614         unhash_lockowner_locked(lockowner(sop));
7615 }
7616
7617 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7618 {
7619         struct nfs4_lockowner *lo = lockowner(sop);
7620
7621         kmem_cache_free(lockowner_slab, lo);
7622 }
7623
7624 static const struct nfs4_stateowner_operations lockowner_ops = {
7625         .so_unhash =    nfs4_unhash_lockowner,
7626         .so_free =      nfs4_free_lockowner,
7627 };
7628
7629 /*
7630  * Alloc a lock owner structure.
7631  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7632  * occurred.
7633  *
7634  * strhashval = ownerstr_hashval
7635  */
7636 static struct nfs4_lockowner *
7637 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7638                            struct nfs4_ol_stateid *open_stp,
7639                            struct nfsd4_lock *lock)
7640 {
7641         struct nfs4_lockowner *lo, *ret;
7642
7643         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7644         if (!lo)
7645                 return NULL;
7646         INIT_LIST_HEAD(&lo->lo_blocked);
7647         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7648         lo->lo_owner.so_is_open_owner = 0;
7649         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7650         lo->lo_owner.so_ops = &lockowner_ops;
7651         spin_lock(&clp->cl_lock);
7652         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7653         if (ret == NULL) {
7654                 list_add(&lo->lo_owner.so_strhash,
7655                          &clp->cl_ownerstr_hashtbl[strhashval]);
7656                 ret = lo;
7657         } else
7658                 nfs4_free_stateowner(&lo->lo_owner);
7659
7660         spin_unlock(&clp->cl_lock);
7661         return ret;
7662 }
7663
7664 static struct nfs4_ol_stateid *
7665 find_lock_stateid(const struct nfs4_lockowner *lo,
7666                   const struct nfs4_ol_stateid *ost)
7667 {
7668         struct nfs4_ol_stateid *lst;
7669
7670         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7671
7672         /* If ost is not hashed, ost->st_locks will not be valid */
7673         if (!nfs4_ol_stateid_unhashed(ost))
7674                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7675                         if (lst->st_stateowner == &lo->lo_owner) {
7676                                 refcount_inc(&lst->st_stid.sc_count);
7677                                 return lst;
7678                         }
7679                 }
7680         return NULL;
7681 }
7682
7683 static struct nfs4_ol_stateid *
7684 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7685                   struct nfs4_file *fp, struct inode *inode,
7686                   struct nfs4_ol_stateid *open_stp)
7687 {
7688         struct nfs4_client *clp = lo->lo_owner.so_client;
7689         struct nfs4_ol_stateid *retstp;
7690
7691         mutex_init(&stp->st_mutex);
7692         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7693 retry:
7694         spin_lock(&clp->cl_lock);
7695         if (nfs4_ol_stateid_unhashed(open_stp))
7696                 goto out_close;
7697         retstp = find_lock_stateid(lo, open_stp);
7698         if (retstp)
7699                 goto out_found;
7700         refcount_inc(&stp->st_stid.sc_count);
7701         stp->st_stid.sc_type = SC_TYPE_LOCK;
7702         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7703         get_nfs4_file(fp);
7704         stp->st_stid.sc_file = fp;
7705         stp->st_access_bmap = 0;
7706         stp->st_deny_bmap = open_stp->st_deny_bmap;
7707         stp->st_openstp = open_stp;
7708         spin_lock(&fp->fi_lock);
7709         list_add(&stp->st_locks, &open_stp->st_locks);
7710         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7711         list_add(&stp->st_perfile, &fp->fi_stateids);
7712         spin_unlock(&fp->fi_lock);
7713         spin_unlock(&clp->cl_lock);
7714         return stp;
7715 out_found:
7716         spin_unlock(&clp->cl_lock);
7717         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7718                 nfs4_put_stid(&retstp->st_stid);
7719                 goto retry;
7720         }
7721         /* To keep mutex tracking happy */
7722         mutex_unlock(&stp->st_mutex);
7723         return retstp;
7724 out_close:
7725         spin_unlock(&clp->cl_lock);
7726         mutex_unlock(&stp->st_mutex);
7727         return NULL;
7728 }
7729
7730 static struct nfs4_ol_stateid *
7731 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7732                             struct inode *inode, struct nfs4_ol_stateid *ost,
7733                             bool *new)
7734 {
7735         struct nfs4_stid *ns = NULL;
7736         struct nfs4_ol_stateid *lst;
7737         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7738         struct nfs4_client *clp = oo->oo_owner.so_client;
7739
7740         *new = false;
7741         spin_lock(&clp->cl_lock);
7742         lst = find_lock_stateid(lo, ost);
7743         spin_unlock(&clp->cl_lock);
7744         if (lst != NULL) {
7745                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7746                         goto out;
7747                 nfs4_put_stid(&lst->st_stid);
7748         }
7749         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7750         if (ns == NULL)
7751                 return NULL;
7752
7753         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7754         if (lst == openlockstateid(ns))
7755                 *new = true;
7756         else
7757                 nfs4_put_stid(ns);
7758 out:
7759         return lst;
7760 }
7761
7762 static int
7763 check_lock_length(u64 offset, u64 length)
7764 {
7765         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7766                 (length > ~offset)));
7767 }
7768
7769 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7770 {
7771         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7772
7773         lockdep_assert_held(&fp->fi_lock);
7774
7775         if (test_access(access, lock_stp))
7776                 return;
7777         __nfs4_file_get_access(fp, access);
7778         set_access(access, lock_stp);
7779 }
7780
7781 static __be32
7782 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7783                             struct nfs4_ol_stateid *ost,
7784                             struct nfsd4_lock *lock,
7785                             struct nfs4_ol_stateid **plst, bool *new)
7786 {
7787         __be32 status;
7788         struct nfs4_file *fi = ost->st_stid.sc_file;
7789         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7790         struct nfs4_client *cl = oo->oo_owner.so_client;
7791         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7792         struct nfs4_lockowner *lo;
7793         struct nfs4_ol_stateid *lst;
7794         unsigned int strhashval;
7795
7796         lo = find_lockowner_str(cl, &lock->lk_new_owner);
7797         if (!lo) {
7798                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7799                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7800                 if (lo == NULL)
7801                         return nfserr_jukebox;
7802         } else {
7803                 /* with an existing lockowner, seqids must be the same */
7804                 status = nfserr_bad_seqid;
7805                 if (!cstate->minorversion &&
7806                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7807                         goto out;
7808         }
7809
7810         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7811         if (lst == NULL) {
7812                 status = nfserr_jukebox;
7813                 goto out;
7814         }
7815
7816         status = nfs_ok;
7817         *plst = lst;
7818 out:
7819         nfs4_put_stateowner(&lo->lo_owner);
7820         return status;
7821 }
7822
7823 /*
7824  *  LOCK operation
7825  */
7826 __be32
7827 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7828            union nfsd4_op_u *u)
7829 {
7830         struct nfsd4_lock *lock = &u->lock;
7831         struct nfs4_openowner *open_sop = NULL;
7832         struct nfs4_lockowner *lock_sop = NULL;
7833         struct nfs4_ol_stateid *lock_stp = NULL;
7834         struct nfs4_ol_stateid *open_stp = NULL;
7835         struct nfs4_file *fp;
7836         struct nfsd_file *nf = NULL;
7837         struct nfsd4_blocked_lock *nbl = NULL;
7838         struct file_lock *file_lock = NULL;
7839         struct file_lock *conflock = NULL;
7840         struct super_block *sb;
7841         __be32 status = 0;
7842         int lkflg;
7843         int err;
7844         bool new = false;
7845         unsigned char type;
7846         unsigned int flags = FL_POSIX;
7847         struct net *net = SVC_NET(rqstp);
7848         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7849
7850         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7851                 (long long) lock->lk_offset,
7852                 (long long) lock->lk_length);
7853
7854         if (check_lock_length(lock->lk_offset, lock->lk_length))
7855                  return nfserr_inval;
7856
7857         if ((status = fh_verify(rqstp, &cstate->current_fh,
7858                                 S_IFREG, NFSD_MAY_LOCK))) {
7859                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7860                 return status;
7861         }
7862         sb = cstate->current_fh.fh_dentry->d_sb;
7863
7864         if (lock->lk_is_new) {
7865                 if (nfsd4_has_session(cstate))
7866                         /* See rfc 5661 18.10.3: given clientid is ignored: */
7867                         memcpy(&lock->lk_new_clientid,
7868                                 &cstate->clp->cl_clientid,
7869                                 sizeof(clientid_t));
7870
7871                 /* validate and update open stateid and open seqid */
7872                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7873                                         lock->lk_new_open_seqid,
7874                                         &lock->lk_new_open_stateid,
7875                                         &open_stp, nn);
7876                 if (status)
7877                         goto out;
7878                 mutex_unlock(&open_stp->st_mutex);
7879                 open_sop = openowner(open_stp->st_stateowner);
7880                 status = nfserr_bad_stateid;
7881                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7882                                                 &lock->lk_new_clientid))
7883                         goto out;
7884                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7885                                                         &lock_stp, &new);
7886         } else {
7887                 status = nfs4_preprocess_seqid_op(cstate,
7888                                                   lock->lk_old_lock_seqid,
7889                                                   &lock->lk_old_lock_stateid,
7890                                                   SC_TYPE_LOCK, 0, &lock_stp,
7891                                                   nn);
7892         }
7893         if (status)
7894                 goto out;
7895         lock_sop = lockowner(lock_stp->st_stateowner);
7896
7897         lkflg = setlkflg(lock->lk_type);
7898         status = nfs4_check_openmode(lock_stp, lkflg);
7899         if (status)
7900                 goto out;
7901
7902         status = nfserr_grace;
7903         if (locks_in_grace(net) && !lock->lk_reclaim)
7904                 goto out;
7905         status = nfserr_no_grace;
7906         if (!locks_in_grace(net) && lock->lk_reclaim)
7907                 goto out;
7908
7909         if (lock->lk_reclaim)
7910                 flags |= FL_RECLAIM;
7911
7912         fp = lock_stp->st_stid.sc_file;
7913         switch (lock->lk_type) {
7914                 case NFS4_READW_LT:
7915                         if (nfsd4_has_session(cstate) ||
7916                             exportfs_lock_op_is_async(sb->s_export_op))
7917                                 flags |= FL_SLEEP;
7918                         fallthrough;
7919                 case NFS4_READ_LT:
7920                         spin_lock(&fp->fi_lock);
7921                         nf = find_readable_file_locked(fp);
7922                         if (nf)
7923                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7924                         spin_unlock(&fp->fi_lock);
7925                         type = F_RDLCK;
7926                         break;
7927                 case NFS4_WRITEW_LT:
7928                         if (nfsd4_has_session(cstate) ||
7929                             exportfs_lock_op_is_async(sb->s_export_op))
7930                                 flags |= FL_SLEEP;
7931                         fallthrough;
7932                 case NFS4_WRITE_LT:
7933                         spin_lock(&fp->fi_lock);
7934                         nf = find_writeable_file_locked(fp);
7935                         if (nf)
7936                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7937                         spin_unlock(&fp->fi_lock);
7938                         type = F_WRLCK;
7939                         break;
7940                 default:
7941                         status = nfserr_inval;
7942                 goto out;
7943         }
7944
7945         if (!nf) {
7946                 status = nfserr_openmode;
7947                 goto out;
7948         }
7949
7950         /*
7951          * Most filesystems with their own ->lock operations will block
7952          * the nfsd thread waiting to acquire the lock.  That leads to
7953          * deadlocks (we don't want every nfsd thread tied up waiting
7954          * for file locks), so don't attempt blocking lock notifications
7955          * on those filesystems:
7956          */
7957         if (!exportfs_lock_op_is_async(sb->s_export_op))
7958                 flags &= ~FL_SLEEP;
7959
7960         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7961         if (!nbl) {
7962                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7963                 status = nfserr_jukebox;
7964                 goto out;
7965         }
7966
7967         file_lock = &nbl->nbl_lock;
7968         file_lock->c.flc_type = type;
7969         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7970         file_lock->c.flc_pid = current->tgid;
7971         file_lock->c.flc_file = nf->nf_file;
7972         file_lock->c.flc_flags = flags;
7973         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7974         file_lock->fl_start = lock->lk_offset;
7975         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7976         nfs4_transform_lock_offset(file_lock);
7977
7978         conflock = locks_alloc_lock();
7979         if (!conflock) {
7980                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7981                 status = nfserr_jukebox;
7982                 goto out;
7983         }
7984
7985         if (flags & FL_SLEEP) {
7986                 nbl->nbl_time = ktime_get_boottime_seconds();
7987                 spin_lock(&nn->blocked_locks_lock);
7988                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7989                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7990                 kref_get(&nbl->nbl_kref);
7991                 spin_unlock(&nn->blocked_locks_lock);
7992         }
7993
7994         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7995         switch (err) {
7996         case 0: /* success! */
7997                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7998                 status = 0;
7999                 if (lock->lk_reclaim)
8000                         nn->somebody_reclaimed = true;
8001                 break;
8002         case FILE_LOCK_DEFERRED:
8003                 kref_put(&nbl->nbl_kref, free_nbl);
8004                 nbl = NULL;
8005                 fallthrough;
8006         case -EAGAIN:           /* conflock holds conflicting lock */
8007                 status = nfserr_denied;
8008                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8009                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8010                 break;
8011         case -EDEADLK:
8012                 status = nfserr_deadlock;
8013                 break;
8014         default:
8015                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8016                 status = nfserrno(err);
8017                 break;
8018         }
8019 out:
8020         if (nbl) {
8021                 /* dequeue it if we queued it before */
8022                 if (flags & FL_SLEEP) {
8023                         spin_lock(&nn->blocked_locks_lock);
8024                         if (!list_empty(&nbl->nbl_list) &&
8025                             !list_empty(&nbl->nbl_lru)) {
8026                                 list_del_init(&nbl->nbl_list);
8027                                 list_del_init(&nbl->nbl_lru);
8028                                 kref_put(&nbl->nbl_kref, free_nbl);
8029                         }
8030                         /* nbl can use one of lists to be linked to reaplist */
8031                         spin_unlock(&nn->blocked_locks_lock);
8032                 }
8033                 free_blocked_lock(nbl);
8034         }
8035         if (nf)
8036                 nfsd_file_put(nf);
8037         if (lock_stp) {
8038                 /* Bump seqid manually if the 4.0 replay owner is openowner */
8039                 if (cstate->replay_owner &&
8040                     cstate->replay_owner != &lock_sop->lo_owner &&
8041                     seqid_mutating_err(ntohl(status)))
8042                         lock_sop->lo_owner.so_seqid++;
8043
8044                 /*
8045                  * If this is a new, never-before-used stateid, and we are
8046                  * returning an error, then just go ahead and release it.
8047                  */
8048                 if (status && new)
8049                         release_lock_stateid(lock_stp);
8050
8051                 mutex_unlock(&lock_stp->st_mutex);
8052
8053                 nfs4_put_stid(&lock_stp->st_stid);
8054         }
8055         if (open_stp)
8056                 nfs4_put_stid(&open_stp->st_stid);
8057         nfsd4_bump_seqid(cstate, status);
8058         if (conflock)
8059                 locks_free_lock(conflock);
8060         return status;
8061 }
8062
8063 void nfsd4_lock_release(union nfsd4_op_u *u)
8064 {
8065         struct nfsd4_lock *lock = &u->lock;
8066         struct nfsd4_lock_denied *deny = &lock->lk_denied;
8067
8068         kfree(deny->ld_owner.data);
8069 }
8070
8071 /*
8072  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8073  * so we do a temporary open here just to get an open file to pass to
8074  * vfs_test_lock.
8075  */
8076 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8077 {
8078         struct nfsd_file *nf;
8079         struct inode *inode;
8080         __be32 err;
8081
8082         err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8083         if (err)
8084                 return err;
8085         inode = fhp->fh_dentry->d_inode;
8086         inode_lock(inode); /* to block new leases till after test_lock: */
8087         err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8088         if (err)
8089                 goto out;
8090         lock->c.flc_file = nf->nf_file;
8091         err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8092         lock->c.flc_file = NULL;
8093 out:
8094         inode_unlock(inode);
8095         nfsd_file_put(nf);
8096         return err;
8097 }
8098
8099 /*
8100  * LOCKT operation
8101  */
8102 __be32
8103 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8104             union nfsd4_op_u *u)
8105 {
8106         struct nfsd4_lockt *lockt = &u->lockt;
8107         struct file_lock *file_lock = NULL;
8108         struct nfs4_lockowner *lo = NULL;
8109         __be32 status;
8110         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8111
8112         if (locks_in_grace(SVC_NET(rqstp)))
8113                 return nfserr_grace;
8114
8115         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8116                  return nfserr_inval;
8117
8118         if (!nfsd4_has_session(cstate)) {
8119                 status = set_client(&lockt->lt_clientid, cstate, nn);
8120                 if (status)
8121                         goto out;
8122         }
8123
8124         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8125                 goto out;
8126
8127         file_lock = locks_alloc_lock();
8128         if (!file_lock) {
8129                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8130                 status = nfserr_jukebox;
8131                 goto out;
8132         }
8133
8134         switch (lockt->lt_type) {
8135                 case NFS4_READ_LT:
8136                 case NFS4_READW_LT:
8137                         file_lock->c.flc_type = F_RDLCK;
8138                         break;
8139                 case NFS4_WRITE_LT:
8140                 case NFS4_WRITEW_LT:
8141                         file_lock->c.flc_type = F_WRLCK;
8142                         break;
8143                 default:
8144                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8145                         status = nfserr_inval;
8146                         goto out;
8147         }
8148
8149         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8150         if (lo)
8151                 file_lock->c.flc_owner = (fl_owner_t)lo;
8152         file_lock->c.flc_pid = current->tgid;
8153         file_lock->c.flc_flags = FL_POSIX;
8154
8155         file_lock->fl_start = lockt->lt_offset;
8156         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8157
8158         nfs4_transform_lock_offset(file_lock);
8159
8160         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8161         if (status)
8162                 goto out;
8163
8164         if (file_lock->c.flc_type != F_UNLCK) {
8165                 status = nfserr_denied;
8166                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8167         }
8168 out:
8169         if (lo)
8170                 nfs4_put_stateowner(&lo->lo_owner);
8171         if (file_lock)
8172                 locks_free_lock(file_lock);
8173         return status;
8174 }
8175
8176 void nfsd4_lockt_release(union nfsd4_op_u *u)
8177 {
8178         struct nfsd4_lockt *lockt = &u->lockt;
8179         struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8180
8181         kfree(deny->ld_owner.data);
8182 }
8183
8184 __be32
8185 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8186             union nfsd4_op_u *u)
8187 {
8188         struct nfsd4_locku *locku = &u->locku;
8189         struct nfs4_ol_stateid *stp;
8190         struct nfsd_file *nf = NULL;
8191         struct file_lock *file_lock = NULL;
8192         __be32 status;
8193         int err;
8194         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8195
8196         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8197                 (long long) locku->lu_offset,
8198                 (long long) locku->lu_length);
8199
8200         if (check_lock_length(locku->lu_offset, locku->lu_length))
8201                  return nfserr_inval;
8202
8203         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8204                                           &locku->lu_stateid, SC_TYPE_LOCK, 0,
8205                                           &stp, nn);
8206         if (status)
8207                 goto out;
8208         nf = find_any_file(stp->st_stid.sc_file);
8209         if (!nf) {
8210                 status = nfserr_lock_range;
8211                 goto put_stateid;
8212         }
8213         file_lock = locks_alloc_lock();
8214         if (!file_lock) {
8215                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8216                 status = nfserr_jukebox;
8217                 goto put_file;
8218         }
8219
8220         file_lock->c.flc_type = F_UNLCK;
8221         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8222         file_lock->c.flc_pid = current->tgid;
8223         file_lock->c.flc_file = nf->nf_file;
8224         file_lock->c.flc_flags = FL_POSIX;
8225         file_lock->fl_lmops = &nfsd_posix_mng_ops;
8226         file_lock->fl_start = locku->lu_offset;
8227
8228         file_lock->fl_end = last_byte_offset(locku->lu_offset,
8229                                                 locku->lu_length);
8230         nfs4_transform_lock_offset(file_lock);
8231
8232         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8233         if (err) {
8234                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8235                 goto out_nfserr;
8236         }
8237         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8238 put_file:
8239         nfsd_file_put(nf);
8240 put_stateid:
8241         mutex_unlock(&stp->st_mutex);
8242         nfs4_put_stid(&stp->st_stid);
8243 out:
8244         nfsd4_bump_seqid(cstate, status);
8245         if (file_lock)
8246                 locks_free_lock(file_lock);
8247         return status;
8248
8249 out_nfserr:
8250         status = nfserrno(err);
8251         goto put_file;
8252 }
8253
8254 /*
8255  * returns
8256  *      true:  locks held by lockowner
8257  *      false: no locks held by lockowner
8258  */
8259 static bool
8260 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8261 {
8262         struct file_lock *fl;
8263         int status = false;
8264         struct nfsd_file *nf;
8265         struct inode *inode;
8266         struct file_lock_context *flctx;
8267
8268         spin_lock(&fp->fi_lock);
8269         nf = find_any_file_locked(fp);
8270         if (!nf) {
8271                 /* Any valid lock stateid should have some sort of access */
8272                 WARN_ON_ONCE(1);
8273                 goto out;
8274         }
8275
8276         inode = file_inode(nf->nf_file);
8277         flctx = locks_inode_context(inode);
8278
8279         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8280                 spin_lock(&flctx->flc_lock);
8281                 for_each_file_lock(fl, &flctx->flc_posix) {
8282                         if (fl->c.flc_owner == (fl_owner_t)lowner) {
8283                                 status = true;
8284                                 break;
8285                         }
8286                 }
8287                 spin_unlock(&flctx->flc_lock);
8288         }
8289 out:
8290         spin_unlock(&fp->fi_lock);
8291         return status;
8292 }
8293
8294 /**
8295  * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8296  * @rqstp: RPC transaction
8297  * @cstate: NFSv4 COMPOUND state
8298  * @u: RELEASE_LOCKOWNER arguments
8299  *
8300  * Check if theree are any locks still held and if not - free the lockowner
8301  * and any lock state that is owned.
8302  *
8303  * Return values:
8304  *   %nfs_ok: lockowner released or not found
8305  *   %nfserr_locks_held: lockowner still in use
8306  *   %nfserr_stale_clientid: clientid no longer active
8307  *   %nfserr_expired: clientid not recognized
8308  */
8309 __be32
8310 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8311                         struct nfsd4_compound_state *cstate,
8312                         union nfsd4_op_u *u)
8313 {
8314         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8315         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8316         clientid_t *clid = &rlockowner->rl_clientid;
8317         struct nfs4_ol_stateid *stp;
8318         struct nfs4_lockowner *lo;
8319         struct nfs4_client *clp;
8320         LIST_HEAD(reaplist);
8321         __be32 status;
8322
8323         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8324                 clid->cl_boot, clid->cl_id);
8325
8326         status = set_client(clid, cstate, nn);
8327         if (status)
8328                 return status;
8329         clp = cstate->clp;
8330
8331         spin_lock(&clp->cl_lock);
8332         lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8333         if (!lo) {
8334                 spin_unlock(&clp->cl_lock);
8335                 return nfs_ok;
8336         }
8337
8338         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8339                 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8340                         spin_unlock(&clp->cl_lock);
8341                         nfs4_put_stateowner(&lo->lo_owner);
8342                         return nfserr_locks_held;
8343                 }
8344         }
8345         unhash_lockowner_locked(lo);
8346         while (!list_empty(&lo->lo_owner.so_stateids)) {
8347                 stp = list_first_entry(&lo->lo_owner.so_stateids,
8348                                        struct nfs4_ol_stateid,
8349                                        st_perstateowner);
8350                 unhash_lock_stateid(stp);
8351                 put_ol_stateid_locked(stp, &reaplist);
8352         }
8353         spin_unlock(&clp->cl_lock);
8354
8355         free_ol_stateid_reaplist(&reaplist);
8356         remove_blocked_locks(lo);
8357         nfs4_put_stateowner(&lo->lo_owner);
8358         return nfs_ok;
8359 }
8360
8361 static inline struct nfs4_client_reclaim *
8362 alloc_reclaim(void)
8363 {
8364         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8365 }
8366
8367 bool
8368 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8369 {
8370         struct nfs4_client_reclaim *crp;
8371
8372         crp = nfsd4_find_reclaim_client(name, nn);
8373         return (crp && crp->cr_clp);
8374 }
8375
8376 /*
8377  * failure => all reset bets are off, nfserr_no_grace...
8378  *
8379  * The caller is responsible for freeing name.data if NULL is returned (it
8380  * will be freed in nfs4_remove_reclaim_record in the normal case).
8381  */
8382 struct nfs4_client_reclaim *
8383 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8384                 struct nfsd_net *nn)
8385 {
8386         unsigned int strhashval;
8387         struct nfs4_client_reclaim *crp;
8388
8389         crp = alloc_reclaim();
8390         if (crp) {
8391                 strhashval = clientstr_hashval(name);
8392                 INIT_LIST_HEAD(&crp->cr_strhash);
8393                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8394                 crp->cr_name.data = name.data;
8395                 crp->cr_name.len = name.len;
8396                 crp->cr_princhash.data = princhash.data;
8397                 crp->cr_princhash.len = princhash.len;
8398                 crp->cr_clp = NULL;
8399                 nn->reclaim_str_hashtbl_size++;
8400         }
8401         return crp;
8402 }
8403
8404 void
8405 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8406 {
8407         list_del(&crp->cr_strhash);
8408         kfree(crp->cr_name.data);
8409         kfree(crp->cr_princhash.data);
8410         kfree(crp);
8411         nn->reclaim_str_hashtbl_size--;
8412 }
8413
8414 void
8415 nfs4_release_reclaim(struct nfsd_net *nn)
8416 {
8417         struct nfs4_client_reclaim *crp = NULL;
8418         int i;
8419
8420         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8421                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8422                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8423                                         struct nfs4_client_reclaim, cr_strhash);
8424                         nfs4_remove_reclaim_record(crp, nn);
8425                 }
8426         }
8427         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8428 }
8429
8430 /*
8431  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8432 struct nfs4_client_reclaim *
8433 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8434 {
8435         unsigned int strhashval;
8436         struct nfs4_client_reclaim *crp = NULL;
8437
8438         strhashval = clientstr_hashval(name);
8439         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8440                 if (compare_blob(&crp->cr_name, &name) == 0) {
8441                         return crp;
8442                 }
8443         }
8444         return NULL;
8445 }
8446
8447 __be32
8448 nfs4_check_open_reclaim(struct nfs4_client *clp)
8449 {
8450         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8451                 return nfserr_no_grace;
8452
8453         if (nfsd4_client_record_check(clp))
8454                 return nfserr_reclaim_bad;
8455
8456         return nfs_ok;
8457 }
8458
8459 /*
8460  * Since the lifetime of a delegation isn't limited to that of an open, a
8461  * client may quite reasonably hang on to a delegation as long as it has
8462  * the inode cached.  This becomes an obvious problem the first time a
8463  * client's inode cache approaches the size of the server's total memory.
8464  *
8465  * For now we avoid this problem by imposing a hard limit on the number
8466  * of delegations, which varies according to the server's memory size.
8467  */
8468 static void
8469 set_max_delegations(void)
8470 {
8471         /*
8472          * Allow at most 4 delegations per megabyte of RAM.  Quick
8473          * estimates suggest that in the worst case (where every delegation
8474          * is for a different inode), a delegation could take about 1.5K,
8475          * giving a worst case usage of about 6% of memory.
8476          */
8477         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8478 }
8479
8480 static int nfs4_state_create_net(struct net *net)
8481 {
8482         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8483         int i;
8484
8485         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8486                                             sizeof(struct list_head),
8487                                             GFP_KERNEL);
8488         if (!nn->conf_id_hashtbl)
8489                 goto err;
8490         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8491                                               sizeof(struct list_head),
8492                                               GFP_KERNEL);
8493         if (!nn->unconf_id_hashtbl)
8494                 goto err_unconf_id;
8495         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8496                                               sizeof(struct list_head),
8497                                               GFP_KERNEL);
8498         if (!nn->sessionid_hashtbl)
8499                 goto err_sessionid;
8500
8501         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8502                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8503                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8504         }
8505         for (i = 0; i < SESSION_HASH_SIZE; i++)
8506                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8507         nn->conf_name_tree = RB_ROOT;
8508         nn->unconf_name_tree = RB_ROOT;
8509         nn->boot_time = ktime_get_real_seconds();
8510         nn->grace_ended = false;
8511         nn->nfsd4_manager.block_opens = true;
8512         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8513         INIT_LIST_HEAD(&nn->client_lru);
8514         INIT_LIST_HEAD(&nn->close_lru);
8515         INIT_LIST_HEAD(&nn->del_recall_lru);
8516         spin_lock_init(&nn->client_lock);
8517         spin_lock_init(&nn->s2s_cp_lock);
8518         idr_init(&nn->s2s_cp_stateids);
8519
8520         spin_lock_init(&nn->blocked_locks_lock);
8521         INIT_LIST_HEAD(&nn->blocked_locks_lru);
8522
8523         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8524         INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8525         get_net(net);
8526
8527         nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8528         if (!nn->nfsd_client_shrinker)
8529                 goto err_shrinker;
8530
8531         nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8532         nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8533         nn->nfsd_client_shrinker->private_data = nn;
8534
8535         shrinker_register(nn->nfsd_client_shrinker);
8536
8537         return 0;
8538
8539 err_shrinker:
8540         put_net(net);
8541         kfree(nn->sessionid_hashtbl);
8542 err_sessionid:
8543         kfree(nn->unconf_id_hashtbl);
8544 err_unconf_id:
8545         kfree(nn->conf_id_hashtbl);
8546 err:
8547         return -ENOMEM;
8548 }
8549
8550 static void
8551 nfs4_state_destroy_net(struct net *net)
8552 {
8553         int i;
8554         struct nfs4_client *clp = NULL;
8555         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8556
8557         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8558                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8559                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8560                         destroy_client(clp);
8561                 }
8562         }
8563
8564         WARN_ON(!list_empty(&nn->blocked_locks_lru));
8565
8566         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8567                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8568                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8569                         destroy_client(clp);
8570                 }
8571         }
8572
8573         kfree(nn->sessionid_hashtbl);
8574         kfree(nn->unconf_id_hashtbl);
8575         kfree(nn->conf_id_hashtbl);
8576         put_net(net);
8577 }
8578
8579 int
8580 nfs4_state_start_net(struct net *net)
8581 {
8582         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8583         int ret;
8584
8585         ret = nfs4_state_create_net(net);
8586         if (ret)
8587                 return ret;
8588         locks_start_grace(net, &nn->nfsd4_manager);
8589         nfsd4_client_tracking_init(net);
8590         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8591                 goto skip_grace;
8592         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8593                nn->nfsd4_grace, net->ns.inum);
8594         trace_nfsd_grace_start(nn);
8595         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8596         return 0;
8597
8598 skip_grace:
8599         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8600                         net->ns.inum);
8601         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8602         nfsd4_end_grace(nn);
8603         return 0;
8604 }
8605
8606 /* initialization to perform when the nfsd service is started: */
8607
8608 int
8609 nfs4_state_start(void)
8610 {
8611         int ret;
8612
8613         ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8614         if (ret)
8615                 return ret;
8616
8617         ret = nfsd4_create_callback_queue();
8618         if (ret) {
8619                 rhltable_destroy(&nfs4_file_rhltable);
8620                 return ret;
8621         }
8622
8623         set_max_delegations();
8624         return 0;
8625 }
8626
8627 void
8628 nfs4_state_shutdown_net(struct net *net)
8629 {
8630         struct nfs4_delegation *dp = NULL;
8631         struct list_head *pos, *next, reaplist;
8632         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8633
8634         shrinker_free(nn->nfsd_client_shrinker);
8635         cancel_work(&nn->nfsd_shrinker_work);
8636         cancel_delayed_work_sync(&nn->laundromat_work);
8637         locks_end_grace(&nn->nfsd4_manager);
8638
8639         INIT_LIST_HEAD(&reaplist);
8640         spin_lock(&state_lock);
8641         list_for_each_safe(pos, next, &nn->del_recall_lru) {
8642                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8643                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8644                 list_add(&dp->dl_recall_lru, &reaplist);
8645         }
8646         spin_unlock(&state_lock);
8647         list_for_each_safe(pos, next, &reaplist) {
8648                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8649                 list_del_init(&dp->dl_recall_lru);
8650                 destroy_unhashed_deleg(dp);
8651         }
8652
8653         nfsd4_client_tracking_exit(net);
8654         nfs4_state_destroy_net(net);
8655 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8656         nfsd4_ssc_shutdown_umount(nn);
8657 #endif
8658 }
8659
8660 void
8661 nfs4_state_shutdown(void)
8662 {
8663         nfsd4_destroy_callback_queue();
8664         rhltable_destroy(&nfs4_file_rhltable);
8665 }
8666
8667 static void
8668 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8669 {
8670         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8671             CURRENT_STATEID(stateid))
8672                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8673 }
8674
8675 static void
8676 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8677 {
8678         if (cstate->minorversion) {
8679                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8680                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8681         }
8682 }
8683
8684 void
8685 clear_current_stateid(struct nfsd4_compound_state *cstate)
8686 {
8687         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8688 }
8689
8690 /*
8691  * functions to set current state id
8692  */
8693 void
8694 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8695                 union nfsd4_op_u *u)
8696 {
8697         put_stateid(cstate, &u->open_downgrade.od_stateid);
8698 }
8699
8700 void
8701 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8702                 union nfsd4_op_u *u)
8703 {
8704         put_stateid(cstate, &u->open.op_stateid);
8705 }
8706
8707 void
8708 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8709                 union nfsd4_op_u *u)
8710 {
8711         put_stateid(cstate, &u->close.cl_stateid);
8712 }
8713
8714 void
8715 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8716                 union nfsd4_op_u *u)
8717 {
8718         put_stateid(cstate, &u->lock.lk_resp_stateid);
8719 }
8720
8721 /*
8722  * functions to consume current state id
8723  */
8724
8725 void
8726 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8727                 union nfsd4_op_u *u)
8728 {
8729         get_stateid(cstate, &u->open_downgrade.od_stateid);
8730 }
8731
8732 void
8733 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8734                 union nfsd4_op_u *u)
8735 {
8736         get_stateid(cstate, &u->delegreturn.dr_stateid);
8737 }
8738
8739 void
8740 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8741                 union nfsd4_op_u *u)
8742 {
8743         get_stateid(cstate, &u->free_stateid.fr_stateid);
8744 }
8745
8746 void
8747 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8748                 union nfsd4_op_u *u)
8749 {
8750         get_stateid(cstate, &u->setattr.sa_stateid);
8751 }
8752
8753 void
8754 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8755                 union nfsd4_op_u *u)
8756 {
8757         get_stateid(cstate, &u->close.cl_stateid);
8758 }
8759
8760 void
8761 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8762                 union nfsd4_op_u *u)
8763 {
8764         get_stateid(cstate, &u->locku.lu_stateid);
8765 }
8766
8767 void
8768 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8769                 union nfsd4_op_u *u)
8770 {
8771         get_stateid(cstate, &u->read.rd_stateid);
8772 }
8773
8774 void
8775 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8776                 union nfsd4_op_u *u)
8777 {
8778         get_stateid(cstate, &u->write.wr_stateid);
8779 }
8780
8781 /**
8782  * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8783  * @rqstp: RPC transaction context
8784  * @inode: file to be checked for a conflict
8785  * @modified: return true if file was modified
8786  * @size: new size of file if modified is true
8787  *
8788  * This function is called when there is a conflict between a write
8789  * delegation and a change/size GETATTR from another client. The server
8790  * must either use the CB_GETATTR to get the current values of the
8791  * attributes from the client that holds the delegation or recall the
8792  * delegation before replying to the GETATTR. See RFC 8881 section
8793  * 18.7.4.
8794  *
8795  * Returns 0 if there is no conflict; otherwise an nfs_stat
8796  * code is returned.
8797  */
8798 __be32
8799 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode,
8800                                 bool *modified, u64 *size)
8801 {
8802         __be32 status;
8803         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8804         struct file_lock_context *ctx;
8805         struct file_lease *fl;
8806         struct nfs4_delegation *dp;
8807         struct iattr attrs;
8808         struct nfs4_cb_fattr *ncf;
8809
8810         *modified = false;
8811         ctx = locks_inode_context(inode);
8812         if (!ctx)
8813                 return 0;
8814         spin_lock(&ctx->flc_lock);
8815         for_each_file_lock(fl, &ctx->flc_lease) {
8816                 unsigned char type = fl->c.flc_type;
8817
8818                 if (fl->c.flc_flags == FL_LAYOUT)
8819                         continue;
8820                 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8821                         /*
8822                          * non-nfs lease, if it's a lease with F_RDLCK then
8823                          * we are done; there isn't any write delegation
8824                          * on this inode
8825                          */
8826                         if (type == F_RDLCK)
8827                                 break;
8828                         goto break_lease;
8829                 }
8830                 if (type == F_WRLCK) {
8831                         dp = fl->c.flc_owner;
8832                         if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8833                                 spin_unlock(&ctx->flc_lock);
8834                                 return 0;
8835                         }
8836 break_lease:
8837                         nfsd_stats_wdeleg_getattr_inc(nn);
8838                         dp = fl->c.flc_owner;
8839                         ncf = &dp->dl_cb_fattr;
8840                         nfs4_cb_getattr(&dp->dl_cb_fattr);
8841                         spin_unlock(&ctx->flc_lock);
8842                         wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8843                                         TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8844                         if (ncf->ncf_cb_status) {
8845                                 /* Recall delegation only if client didn't respond */
8846                                 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8847                                 if (status != nfserr_jukebox ||
8848                                                 !nfsd_wait_for_delegreturn(rqstp, inode))
8849                                         return status;
8850                         }
8851                         if (!ncf->ncf_file_modified &&
8852                                         (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8853                                         ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8854                                 ncf->ncf_file_modified = true;
8855                         if (ncf->ncf_file_modified) {
8856                                 /*
8857                                  * Per section 10.4.3 of RFC 8881, the server would
8858                                  * not update the file's metadata with the client's
8859                                  * modified size
8860                                  */
8861                                 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8862                                 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME;
8863                                 setattr_copy(&nop_mnt_idmap, inode, &attrs);
8864                                 mark_inode_dirty(inode);
8865                                 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8866                                 *size = ncf->ncf_cur_fsize;
8867                                 *modified = true;
8868                         }
8869                         return 0;
8870                 }
8871                 break;
8872         }
8873         spin_unlock(&ctx->flc_lock);
8874         return 0;
8875 }