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         switch (status) {
3835         case nfs_ok:
3836                 cs_slot->sl_seqid++;
3837                 cr_ses->seqid = cs_slot->sl_seqid;
3838                 break;
3839         case nfserr_replay_cache:
3840                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3841                 fallthrough;
3842         case nfserr_jukebox:
3843                 /* The server MUST NOT cache NFS4ERR_DELAY */
3844                 goto out_free_conn;
3845         default:
3846                 goto out_cache_error;
3847         }
3848
3849         /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3850         if (conf) {
3851                 status = nfserr_wrong_cred;
3852                 if (!nfsd4_mach_creds_match(conf, rqstp))
3853                         goto out_cache_error;
3854         } else {
3855                 status = nfserr_clid_inuse;
3856                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3857                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3858                         trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3859                         goto out_cache_error;
3860                 }
3861                 status = nfserr_wrong_cred;
3862                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3863                         goto out_cache_error;
3864                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3865                 if (old) {
3866                         status = mark_client_expired_locked(old);
3867                         if (status)
3868                                 goto out_expired_error;
3869                         trace_nfsd_clid_replaced(&old->cl_clientid);
3870                 }
3871                 move_to_confirmed(unconf);
3872                 conf = unconf;
3873         }
3874
3875         /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3876         status = nfs_ok;
3877         /* Persistent sessions are not supported */
3878         cr_ses->flags &= ~SESSION4_PERSIST;
3879         /* Upshifting from TCP to RDMA is not supported */
3880         cr_ses->flags &= ~SESSION4_RDMA;
3881
3882         init_session(rqstp, new, conf, cr_ses);
3883         nfsd4_get_session_locked(new);
3884
3885         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3886                NFS4_MAX_SESSIONID_LEN);
3887
3888         /* cache solo and embedded create sessions under the client_lock */
3889         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3890         spin_unlock(&nn->client_lock);
3891         if (conf == unconf)
3892                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3893         /* init connection and backchannel */
3894         nfsd4_init_conn(rqstp, conn, new);
3895         nfsd4_put_session(new);
3896         if (old)
3897                 expire_client(old);
3898         return status;
3899
3900 out_expired_error:
3901         old = NULL;
3902         /*
3903          * Revert the slot seq_nr change so the server will process
3904          * the client's resend instead of returning a cached response.
3905          */
3906         if (status == nfserr_jukebox) {
3907                 cs_slot->sl_seqid--;
3908                 cr_ses->seqid = cs_slot->sl_seqid;
3909                 goto out_free_conn;
3910         }
3911 out_cache_error:
3912         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3913 out_free_conn:
3914         spin_unlock(&nn->client_lock);
3915         free_conn(conn);
3916         if (old)
3917                 expire_client(old);
3918 out_free_session:
3919         __free_session(new);
3920 out_release_drc_mem:
3921         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3922         return status;
3923 }
3924
3925 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3926 {
3927         switch (*dir) {
3928         case NFS4_CDFC4_FORE:
3929         case NFS4_CDFC4_BACK:
3930                 return nfs_ok;
3931         case NFS4_CDFC4_FORE_OR_BOTH:
3932         case NFS4_CDFC4_BACK_OR_BOTH:
3933                 *dir = NFS4_CDFC4_BOTH;
3934                 return nfs_ok;
3935         }
3936         return nfserr_inval;
3937 }
3938
3939 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3940                 struct nfsd4_compound_state *cstate,
3941                 union nfsd4_op_u *u)
3942 {
3943         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3944         struct nfsd4_session *session = cstate->session;
3945         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3946         __be32 status;
3947
3948         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3949         if (status)
3950                 return status;
3951         spin_lock(&nn->client_lock);
3952         session->se_cb_prog = bc->bc_cb_program;
3953         session->se_cb_sec = bc->bc_cb_sec;
3954         spin_unlock(&nn->client_lock);
3955
3956         nfsd4_probe_callback(session->se_client);
3957
3958         return nfs_ok;
3959 }
3960
3961 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3962 {
3963         struct nfsd4_conn *c;
3964
3965         list_for_each_entry(c, &s->se_conns, cn_persession) {
3966                 if (c->cn_xprt == xpt) {
3967                         return c;
3968                 }
3969         }
3970         return NULL;
3971 }
3972
3973 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3974                 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3975 {
3976         struct nfs4_client *clp = session->se_client;
3977         struct svc_xprt *xpt = rqst->rq_xprt;
3978         struct nfsd4_conn *c;
3979         __be32 status;
3980
3981         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3982         spin_lock(&clp->cl_lock);
3983         c = __nfsd4_find_conn(xpt, session);
3984         if (!c)
3985                 status = nfserr_noent;
3986         else if (req == c->cn_flags)
3987                 status = nfs_ok;
3988         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3989                                 c->cn_flags != NFS4_CDFC4_BACK)
3990                 status = nfs_ok;
3991         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3992                                 c->cn_flags != NFS4_CDFC4_FORE)
3993                 status = nfs_ok;
3994         else
3995                 status = nfserr_inval;
3996         spin_unlock(&clp->cl_lock);
3997         if (status == nfs_ok && conn)
3998                 *conn = c;
3999         return status;
4000 }
4001
4002 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
4003                      struct nfsd4_compound_state *cstate,
4004                      union nfsd4_op_u *u)
4005 {
4006         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
4007         __be32 status;
4008         struct nfsd4_conn *conn;
4009         struct nfsd4_session *session;
4010         struct net *net = SVC_NET(rqstp);
4011         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4012
4013         if (!nfsd4_last_compound_op(rqstp))
4014                 return nfserr_not_only_op;
4015         spin_lock(&nn->client_lock);
4016         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4017         spin_unlock(&nn->client_lock);
4018         if (!session)
4019                 goto out_no_session;
4020         status = nfserr_wrong_cred;
4021         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4022                 goto out;
4023         status = nfsd4_match_existing_connection(rqstp, session,
4024                         bcts->dir, &conn);
4025         if (status == nfs_ok) {
4026                 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4027                                 bcts->dir == NFS4_CDFC4_BACK)
4028                         conn->cn_flags |= NFS4_CDFC4_BACK;
4029                 nfsd4_probe_callback(session->se_client);
4030                 goto out;
4031         }
4032         if (status == nfserr_inval)
4033                 goto out;
4034         status = nfsd4_map_bcts_dir(&bcts->dir);
4035         if (status)
4036                 goto out;
4037         conn = alloc_conn(rqstp, bcts->dir);
4038         status = nfserr_jukebox;
4039         if (!conn)
4040                 goto out;
4041         nfsd4_init_conn(rqstp, conn, session);
4042         status = nfs_ok;
4043 out:
4044         nfsd4_put_session(session);
4045 out_no_session:
4046         return status;
4047 }
4048
4049 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4050 {
4051         if (!cstate->session)
4052                 return false;
4053         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4054 }
4055
4056 __be32
4057 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4058                 union nfsd4_op_u *u)
4059 {
4060         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4061         struct nfsd4_session *ses;
4062         __be32 status;
4063         int ref_held_by_me = 0;
4064         struct net *net = SVC_NET(r);
4065         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4066
4067         status = nfserr_not_only_op;
4068         if (nfsd4_compound_in_session(cstate, sessionid)) {
4069                 if (!nfsd4_last_compound_op(r))
4070                         goto out;
4071                 ref_held_by_me++;
4072         }
4073         dump_sessionid(__func__, sessionid);
4074         spin_lock(&nn->client_lock);
4075         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4076         if (!ses)
4077                 goto out_client_lock;
4078         status = nfserr_wrong_cred;
4079         if (!nfsd4_mach_creds_match(ses->se_client, r))
4080                 goto out_put_session;
4081         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4082         if (status)
4083                 goto out_put_session;
4084         unhash_session(ses);
4085         spin_unlock(&nn->client_lock);
4086
4087         nfsd4_probe_callback_sync(ses->se_client);
4088
4089         spin_lock(&nn->client_lock);
4090         status = nfs_ok;
4091 out_put_session:
4092         nfsd4_put_session_locked(ses);
4093 out_client_lock:
4094         spin_unlock(&nn->client_lock);
4095 out:
4096         return status;
4097 }
4098
4099 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4100 {
4101         struct nfs4_client *clp = ses->se_client;
4102         struct nfsd4_conn *c;
4103         __be32 status = nfs_ok;
4104         int ret;
4105
4106         spin_lock(&clp->cl_lock);
4107         c = __nfsd4_find_conn(new->cn_xprt, ses);
4108         if (c)
4109                 goto out_free;
4110         status = nfserr_conn_not_bound_to_session;
4111         if (clp->cl_mach_cred)
4112                 goto out_free;
4113         __nfsd4_hash_conn(new, ses);
4114         spin_unlock(&clp->cl_lock);
4115         ret = nfsd4_register_conn(new);
4116         if (ret)
4117                 /* oops; xprt is already down: */
4118                 nfsd4_conn_lost(&new->cn_xpt_user);
4119         return nfs_ok;
4120 out_free:
4121         spin_unlock(&clp->cl_lock);
4122         free_conn(new);
4123         return status;
4124 }
4125
4126 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4127 {
4128         struct nfsd4_compoundargs *args = rqstp->rq_argp;
4129
4130         return args->opcnt > session->se_fchannel.maxops;
4131 }
4132
4133 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4134                                   struct nfsd4_session *session)
4135 {
4136         struct xdr_buf *xb = &rqstp->rq_arg;
4137
4138         return xb->len > session->se_fchannel.maxreq_sz;
4139 }
4140
4141 static bool replay_matches_cache(struct svc_rqst *rqstp,
4142                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4143 {
4144         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4145
4146         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4147             (bool)seq->cachethis)
4148                 return false;
4149         /*
4150          * If there's an error then the reply can have fewer ops than
4151          * the call.
4152          */
4153         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4154                 return false;
4155         /*
4156          * But if we cached a reply with *more* ops than the call you're
4157          * sending us now, then this new call is clearly not really a
4158          * replay of the old one:
4159          */
4160         if (slot->sl_opcnt > argp->opcnt)
4161                 return false;
4162         /* This is the only check explicitly called by spec: */
4163         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4164                 return false;
4165         /*
4166          * There may be more comparisons we could actually do, but the
4167          * spec doesn't require us to catch every case where the calls
4168          * don't match (that would require caching the call as well as
4169          * the reply), so we don't bother.
4170          */
4171         return true;
4172 }
4173
4174 __be32
4175 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4176                 union nfsd4_op_u *u)
4177 {
4178         struct nfsd4_sequence *seq = &u->sequence;
4179         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4180         struct xdr_stream *xdr = resp->xdr;
4181         struct nfsd4_session *session;
4182         struct nfs4_client *clp;
4183         struct nfsd4_slot *slot;
4184         struct nfsd4_conn *conn;
4185         __be32 status;
4186         int buflen;
4187         struct net *net = SVC_NET(rqstp);
4188         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4189
4190         if (resp->opcnt != 1)
4191                 return nfserr_sequence_pos;
4192
4193         /*
4194          * Will be either used or freed by nfsd4_sequence_check_conn
4195          * below.
4196          */
4197         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4198         if (!conn)
4199                 return nfserr_jukebox;
4200
4201         spin_lock(&nn->client_lock);
4202         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4203         if (!session)
4204                 goto out_no_session;
4205         clp = session->se_client;
4206
4207         status = nfserr_too_many_ops;
4208         if (nfsd4_session_too_many_ops(rqstp, session))
4209                 goto out_put_session;
4210
4211         status = nfserr_req_too_big;
4212         if (nfsd4_request_too_big(rqstp, session))
4213                 goto out_put_session;
4214
4215         status = nfserr_badslot;
4216         if (seq->slotid >= session->se_fchannel.maxreqs)
4217                 goto out_put_session;
4218
4219         slot = session->se_slots[seq->slotid];
4220         dprintk("%s: slotid %d\n", __func__, seq->slotid);
4221
4222         /* We do not negotiate the number of slots yet, so set the
4223          * maxslots to the session maxreqs which is used to encode
4224          * sr_highest_slotid and the sr_target_slot id to maxslots */
4225         seq->maxslots = session->se_fchannel.maxreqs;
4226
4227         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4228                                         slot->sl_flags & NFSD4_SLOT_INUSE);
4229         if (status == nfserr_replay_cache) {
4230                 status = nfserr_seq_misordered;
4231                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4232                         goto out_put_session;
4233                 status = nfserr_seq_false_retry;
4234                 if (!replay_matches_cache(rqstp, seq, slot))
4235                         goto out_put_session;
4236                 cstate->slot = slot;
4237                 cstate->session = session;
4238                 cstate->clp = clp;
4239                 /* Return the cached reply status and set cstate->status
4240                  * for nfsd4_proc_compound processing */
4241                 status = nfsd4_replay_cache_entry(resp, seq);
4242                 cstate->status = nfserr_replay_cache;
4243                 goto out;
4244         }
4245         if (status)
4246                 goto out_put_session;
4247
4248         status = nfsd4_sequence_check_conn(conn, session);
4249         conn = NULL;
4250         if (status)
4251                 goto out_put_session;
4252
4253         buflen = (seq->cachethis) ?
4254                         session->se_fchannel.maxresp_cached :
4255                         session->se_fchannel.maxresp_sz;
4256         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4257                                     nfserr_rep_too_big;
4258         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4259                 goto out_put_session;
4260         svc_reserve(rqstp, buflen);
4261
4262         status = nfs_ok;
4263         /* Success! bump slot seqid */
4264         slot->sl_seqid = seq->seqid;
4265         slot->sl_flags |= NFSD4_SLOT_INUSE;
4266         if (seq->cachethis)
4267                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4268         else
4269                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4270
4271         cstate->slot = slot;
4272         cstate->session = session;
4273         cstate->clp = clp;
4274
4275 out:
4276         switch (clp->cl_cb_state) {
4277         case NFSD4_CB_DOWN:
4278                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4279                 break;
4280         case NFSD4_CB_FAULT:
4281                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4282                 break;
4283         default:
4284                 seq->status_flags = 0;
4285         }
4286         if (!list_empty(&clp->cl_revoked))
4287                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4288         if (atomic_read(&clp->cl_admin_revoked))
4289                 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4290         trace_nfsd_seq4_status(rqstp, seq);
4291 out_no_session:
4292         if (conn)
4293                 free_conn(conn);
4294         spin_unlock(&nn->client_lock);
4295         return status;
4296 out_put_session:
4297         nfsd4_put_session_locked(session);
4298         goto out_no_session;
4299 }
4300
4301 void
4302 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4303 {
4304         struct nfsd4_compound_state *cs = &resp->cstate;
4305
4306         if (nfsd4_has_session(cs)) {
4307                 if (cs->status != nfserr_replay_cache) {
4308                         nfsd4_store_cache_entry(resp);
4309                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4310                 }
4311                 /* Drop session reference that was taken in nfsd4_sequence() */
4312                 nfsd4_put_session(cs->session);
4313         } else if (cs->clp)
4314                 put_client_renew(cs->clp);
4315 }
4316
4317 __be32
4318 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4319                 struct nfsd4_compound_state *cstate,
4320                 union nfsd4_op_u *u)
4321 {
4322         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4323         struct nfs4_client *conf, *unconf;
4324         struct nfs4_client *clp = NULL;
4325         __be32 status = 0;
4326         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4327
4328         spin_lock(&nn->client_lock);
4329         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4330         conf = find_confirmed_client(&dc->clientid, true, nn);
4331         WARN_ON_ONCE(conf && unconf);
4332
4333         if (conf) {
4334                 if (client_has_state(conf)) {
4335                         status = nfserr_clientid_busy;
4336                         goto out;
4337                 }
4338                 status = mark_client_expired_locked(conf);
4339                 if (status)
4340                         goto out;
4341                 clp = conf;
4342         } else if (unconf)
4343                 clp = unconf;
4344         else {
4345                 status = nfserr_stale_clientid;
4346                 goto out;
4347         }
4348         if (!nfsd4_mach_creds_match(clp, rqstp)) {
4349                 clp = NULL;
4350                 status = nfserr_wrong_cred;
4351                 goto out;
4352         }
4353         trace_nfsd_clid_destroyed(&clp->cl_clientid);
4354         unhash_client_locked(clp);
4355 out:
4356         spin_unlock(&nn->client_lock);
4357         if (clp)
4358                 expire_client(clp);
4359         return status;
4360 }
4361
4362 __be32
4363 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4364                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4365 {
4366         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4367         struct nfs4_client *clp = cstate->clp;
4368         __be32 status = 0;
4369
4370         if (rc->rca_one_fs) {
4371                 if (!cstate->current_fh.fh_dentry)
4372                         return nfserr_nofilehandle;
4373                 /*
4374                  * We don't take advantage of the rca_one_fs case.
4375                  * That's OK, it's optional, we can safely ignore it.
4376                  */
4377                 return nfs_ok;
4378         }
4379
4380         status = nfserr_complete_already;
4381         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4382                 goto out;
4383
4384         status = nfserr_stale_clientid;
4385         if (is_client_expired(clp))
4386                 /*
4387                  * The following error isn't really legal.
4388                  * But we only get here if the client just explicitly
4389                  * destroyed the client.  Surely it no longer cares what
4390                  * error it gets back on an operation for the dead
4391                  * client.
4392                  */
4393                 goto out;
4394
4395         status = nfs_ok;
4396         trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4397         nfsd4_client_record_create(clp);
4398         inc_reclaim_complete(clp);
4399 out:
4400         return status;
4401 }
4402
4403 __be32
4404 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4405                   union nfsd4_op_u *u)
4406 {
4407         struct nfsd4_setclientid *setclid = &u->setclientid;
4408         struct xdr_netobj       clname = setclid->se_name;
4409         nfs4_verifier           clverifier = setclid->se_verf;
4410         struct nfs4_client      *conf, *new;
4411         struct nfs4_client      *unconf = NULL;
4412         __be32                  status;
4413         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4414
4415         new = create_client(clname, rqstp, &clverifier);
4416         if (new == NULL)
4417                 return nfserr_jukebox;
4418         spin_lock(&nn->client_lock);
4419         conf = find_confirmed_client_by_name(&clname, nn);
4420         if (conf && client_has_state(conf)) {
4421                 status = nfserr_clid_inuse;
4422                 if (clp_used_exchangeid(conf))
4423                         goto out;
4424                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4425                         trace_nfsd_clid_cred_mismatch(conf, rqstp);
4426                         goto out;
4427                 }
4428         }
4429         unconf = find_unconfirmed_client_by_name(&clname, nn);
4430         if (unconf)
4431                 unhash_client_locked(unconf);
4432         if (conf) {
4433                 if (same_verf(&conf->cl_verifier, &clverifier)) {
4434                         copy_clid(new, conf);
4435                         gen_confirm(new, nn);
4436                 } else
4437                         trace_nfsd_clid_verf_mismatch(conf, rqstp,
4438                                                       &clverifier);
4439         } else
4440                 trace_nfsd_clid_fresh(new);
4441         new->cl_minorversion = 0;
4442         gen_callback(new, setclid, rqstp);
4443         add_to_unconfirmed(new);
4444         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4445         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4446         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4447         new = NULL;
4448         status = nfs_ok;
4449 out:
4450         spin_unlock(&nn->client_lock);
4451         if (new)
4452                 free_client(new);
4453         if (unconf) {
4454                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4455                 expire_client(unconf);
4456         }
4457         return status;
4458 }
4459
4460 __be32
4461 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4462                         struct nfsd4_compound_state *cstate,
4463                         union nfsd4_op_u *u)
4464 {
4465         struct nfsd4_setclientid_confirm *setclientid_confirm =
4466                         &u->setclientid_confirm;
4467         struct nfs4_client *conf, *unconf;
4468         struct nfs4_client *old = NULL;
4469         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4470         clientid_t * clid = &setclientid_confirm->sc_clientid;
4471         __be32 status;
4472         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4473
4474         if (STALE_CLIENTID(clid, nn))
4475                 return nfserr_stale_clientid;
4476
4477         spin_lock(&nn->client_lock);
4478         conf = find_confirmed_client(clid, false, nn);
4479         unconf = find_unconfirmed_client(clid, false, nn);
4480         /*
4481          * We try hard to give out unique clientid's, so if we get an
4482          * attempt to confirm the same clientid with a different cred,
4483          * the client may be buggy; this should never happen.
4484          *
4485          * Nevertheless, RFC 7530 recommends INUSE for this case:
4486          */
4487         status = nfserr_clid_inuse;
4488         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4489                 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4490                 goto out;
4491         }
4492         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4493                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4494                 goto out;
4495         }
4496         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4497                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4498                         status = nfs_ok;
4499                 } else
4500                         status = nfserr_stale_clientid;
4501                 goto out;
4502         }
4503         status = nfs_ok;
4504         if (conf) {
4505                 old = unconf;
4506                 unhash_client_locked(old);
4507                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4508         } else {
4509                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4510                 if (old) {
4511                         status = nfserr_clid_inuse;
4512                         if (client_has_state(old)
4513                                         && !same_creds(&unconf->cl_cred,
4514                                                         &old->cl_cred)) {
4515                                 old = NULL;
4516                                 goto out;
4517                         }
4518                         status = mark_client_expired_locked(old);
4519                         if (status) {
4520                                 old = NULL;
4521                                 goto out;
4522                         }
4523                         trace_nfsd_clid_replaced(&old->cl_clientid);
4524                 }
4525                 move_to_confirmed(unconf);
4526                 conf = unconf;
4527         }
4528         get_client_locked(conf);
4529         spin_unlock(&nn->client_lock);
4530         if (conf == unconf)
4531                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4532         nfsd4_probe_callback(conf);
4533         spin_lock(&nn->client_lock);
4534         put_client_renew_locked(conf);
4535 out:
4536         spin_unlock(&nn->client_lock);
4537         if (old)
4538                 expire_client(old);
4539         return status;
4540 }
4541
4542 static struct nfs4_file *nfsd4_alloc_file(void)
4543 {
4544         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4545 }
4546
4547 /* OPEN Share state helper functions */
4548
4549 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4550 {
4551         refcount_set(&fp->fi_ref, 1);
4552         spin_lock_init(&fp->fi_lock);
4553         INIT_LIST_HEAD(&fp->fi_stateids);
4554         INIT_LIST_HEAD(&fp->fi_delegations);
4555         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4556         fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4557         fp->fi_deleg_file = NULL;
4558         fp->fi_had_conflict = false;
4559         fp->fi_share_deny = 0;
4560         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4561         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4562         fp->fi_aliased = false;
4563         fp->fi_inode = d_inode(fh->fh_dentry);
4564 #ifdef CONFIG_NFSD_PNFS
4565         INIT_LIST_HEAD(&fp->fi_lo_states);
4566         atomic_set(&fp->fi_lo_recalls, 0);
4567 #endif
4568 }
4569
4570 void
4571 nfsd4_free_slabs(void)
4572 {
4573         kmem_cache_destroy(client_slab);
4574         kmem_cache_destroy(openowner_slab);
4575         kmem_cache_destroy(lockowner_slab);
4576         kmem_cache_destroy(file_slab);
4577         kmem_cache_destroy(stateid_slab);
4578         kmem_cache_destroy(deleg_slab);
4579         kmem_cache_destroy(odstate_slab);
4580 }
4581
4582 int
4583 nfsd4_init_slabs(void)
4584 {
4585         client_slab = KMEM_CACHE(nfs4_client, 0);
4586         if (client_slab == NULL)
4587                 goto out;
4588         openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4589         if (openowner_slab == NULL)
4590                 goto out_free_client_slab;
4591         lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4592         if (lockowner_slab == NULL)
4593                 goto out_free_openowner_slab;
4594         file_slab = KMEM_CACHE(nfs4_file, 0);
4595         if (file_slab == NULL)
4596                 goto out_free_lockowner_slab;
4597         stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4598         if (stateid_slab == NULL)
4599                 goto out_free_file_slab;
4600         deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4601         if (deleg_slab == NULL)
4602                 goto out_free_stateid_slab;
4603         odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4604         if (odstate_slab == NULL)
4605                 goto out_free_deleg_slab;
4606         return 0;
4607
4608 out_free_deleg_slab:
4609         kmem_cache_destroy(deleg_slab);
4610 out_free_stateid_slab:
4611         kmem_cache_destroy(stateid_slab);
4612 out_free_file_slab:
4613         kmem_cache_destroy(file_slab);
4614 out_free_lockowner_slab:
4615         kmem_cache_destroy(lockowner_slab);
4616 out_free_openowner_slab:
4617         kmem_cache_destroy(openowner_slab);
4618 out_free_client_slab:
4619         kmem_cache_destroy(client_slab);
4620 out:
4621         return -ENOMEM;
4622 }
4623
4624 static unsigned long
4625 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4626 {
4627         int count;
4628         struct nfsd_net *nn = shrink->private_data;
4629
4630         count = atomic_read(&nn->nfsd_courtesy_clients);
4631         if (!count)
4632                 count = atomic_long_read(&num_delegations);
4633         if (count)
4634                 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4635         return (unsigned long)count;
4636 }
4637
4638 static unsigned long
4639 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4640 {
4641         return SHRINK_STOP;
4642 }
4643
4644 void
4645 nfsd4_init_leases_net(struct nfsd_net *nn)
4646 {
4647         struct sysinfo si;
4648         u64 max_clients;
4649
4650         nn->nfsd4_lease = 90;   /* default lease time */
4651         nn->nfsd4_grace = 90;
4652         nn->somebody_reclaimed = false;
4653         nn->track_reclaim_completes = false;
4654         nn->clverifier_counter = get_random_u32();
4655         nn->clientid_base = get_random_u32();
4656         nn->clientid_counter = nn->clientid_base + 1;
4657         nn->s2s_cp_cl_id = nn->clientid_counter++;
4658
4659         atomic_set(&nn->nfs4_client_count, 0);
4660         si_meminfo(&si);
4661         max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4662         max_clients *= NFS4_CLIENTS_PER_GB;
4663         nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4664
4665         atomic_set(&nn->nfsd_courtesy_clients, 0);
4666 }
4667
4668 static void init_nfs4_replay(struct nfs4_replay *rp)
4669 {
4670         rp->rp_status = nfserr_serverfault;
4671         rp->rp_buflen = 0;
4672         rp->rp_buf = rp->rp_ibuf;
4673         mutex_init(&rp->rp_mutex);
4674 }
4675
4676 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4677                 struct nfs4_stateowner *so)
4678 {
4679         if (!nfsd4_has_session(cstate)) {
4680                 mutex_lock(&so->so_replay.rp_mutex);
4681                 cstate->replay_owner = nfs4_get_stateowner(so);
4682         }
4683 }
4684
4685 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4686 {
4687         struct nfs4_stateowner *so = cstate->replay_owner;
4688
4689         if (so != NULL) {
4690                 cstate->replay_owner = NULL;
4691                 mutex_unlock(&so->so_replay.rp_mutex);
4692                 nfs4_put_stateowner(so);
4693         }
4694 }
4695
4696 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4697 {
4698         struct nfs4_stateowner *sop;
4699
4700         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4701         if (!sop)
4702                 return NULL;
4703
4704         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4705         if (!sop->so_owner.data) {
4706                 kmem_cache_free(slab, sop);
4707                 return NULL;
4708         }
4709
4710         INIT_LIST_HEAD(&sop->so_stateids);
4711         sop->so_client = clp;
4712         init_nfs4_replay(&sop->so_replay);
4713         atomic_set(&sop->so_count, 1);
4714         return sop;
4715 }
4716
4717 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4718 {
4719         lockdep_assert_held(&clp->cl_lock);
4720
4721         list_add(&oo->oo_owner.so_strhash,
4722                  &clp->cl_ownerstr_hashtbl[strhashval]);
4723         list_add(&oo->oo_perclient, &clp->cl_openowners);
4724 }
4725
4726 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4727 {
4728         unhash_openowner_locked(openowner(so));
4729 }
4730
4731 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4732 {
4733         struct nfs4_openowner *oo = openowner(so);
4734
4735         kmem_cache_free(openowner_slab, oo);
4736 }
4737
4738 static const struct nfs4_stateowner_operations openowner_ops = {
4739         .so_unhash =    nfs4_unhash_openowner,
4740         .so_free =      nfs4_free_openowner,
4741 };
4742
4743 static struct nfs4_ol_stateid *
4744 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4745 {
4746         struct nfs4_ol_stateid *local, *ret = NULL;
4747         struct nfs4_openowner *oo = open->op_openowner;
4748
4749         lockdep_assert_held(&fp->fi_lock);
4750
4751         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4752                 /* ignore lock owners */
4753                 if (local->st_stateowner->so_is_open_owner == 0)
4754                         continue;
4755                 if (local->st_stateowner != &oo->oo_owner)
4756                         continue;
4757                 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4758                     !local->st_stid.sc_status) {
4759                         ret = local;
4760                         refcount_inc(&ret->st_stid.sc_count);
4761                         break;
4762                 }
4763         }
4764         return ret;
4765 }
4766
4767 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4768         __releases(&s->sc_client->cl_lock)
4769 {
4770         struct nfs4_client *cl = s->sc_client;
4771         LIST_HEAD(reaplist);
4772         struct nfs4_ol_stateid *stp;
4773         struct nfs4_delegation *dp;
4774         bool unhashed;
4775
4776         switch (s->sc_type) {
4777         case SC_TYPE_OPEN:
4778                 stp = openlockstateid(s);
4779                 if (unhash_open_stateid(stp, &reaplist))
4780                         put_ol_stateid_locked(stp, &reaplist);
4781                 spin_unlock(&cl->cl_lock);
4782                 free_ol_stateid_reaplist(&reaplist);
4783                 break;
4784         case SC_TYPE_LOCK:
4785                 stp = openlockstateid(s);
4786                 unhashed = unhash_lock_stateid(stp);
4787                 spin_unlock(&cl->cl_lock);
4788                 if (unhashed)
4789                         nfs4_put_stid(s);
4790                 break;
4791         case SC_TYPE_DELEG:
4792                 dp = delegstateid(s);
4793                 list_del_init(&dp->dl_recall_lru);
4794                 spin_unlock(&cl->cl_lock);
4795                 nfs4_put_stid(s);
4796                 break;
4797         default:
4798                 spin_unlock(&cl->cl_lock);
4799         }
4800 }
4801
4802 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4803                                     stateid_t *stid)
4804 {
4805         /* NFSv4.0 has no way for the client to tell the server
4806          * that it can forget an admin-revoked stateid.
4807          * So we keep it around until the first time that the
4808          * client uses it, and drop it the first time
4809          * nfserr_admin_revoked is returned.
4810          * For v4.1 and later we wait until explicitly told
4811          * to free the stateid.
4812          */
4813         if (cl->cl_minorversion == 0) {
4814                 struct nfs4_stid *st;
4815
4816                 spin_lock(&cl->cl_lock);
4817                 st = find_stateid_locked(cl, stid);
4818                 if (st)
4819                         nfsd4_drop_revoked_stid(st);
4820                 else
4821                         spin_unlock(&cl->cl_lock);
4822         }
4823 }
4824
4825 static __be32
4826 nfsd4_verify_open_stid(struct nfs4_stid *s)
4827 {
4828         __be32 ret = nfs_ok;
4829
4830         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4831                 ret = nfserr_admin_revoked;
4832         else if (s->sc_status & SC_STATUS_REVOKED)
4833                 ret = nfserr_deleg_revoked;
4834         else if (s->sc_status & SC_STATUS_CLOSED)
4835                 ret = nfserr_bad_stateid;
4836         return ret;
4837 }
4838
4839 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4840 static __be32
4841 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4842 {
4843         __be32 ret;
4844
4845         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4846         ret = nfsd4_verify_open_stid(&stp->st_stid);
4847         if (ret == nfserr_admin_revoked)
4848                 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4849                                         &stp->st_stid.sc_stateid);
4850
4851         if (ret != nfs_ok)
4852                 mutex_unlock(&stp->st_mutex);
4853         return ret;
4854 }
4855
4856 static struct nfs4_ol_stateid *
4857 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4858 {
4859         struct nfs4_ol_stateid *stp;
4860         for (;;) {
4861                 spin_lock(&fp->fi_lock);
4862                 stp = nfsd4_find_existing_open(fp, open);
4863                 spin_unlock(&fp->fi_lock);
4864                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4865                         break;
4866                 nfs4_put_stid(&stp->st_stid);
4867         }
4868         return stp;
4869 }
4870
4871 static struct nfs4_openowner *
4872 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4873                            struct nfsd4_compound_state *cstate)
4874 {
4875         struct nfs4_client *clp = cstate->clp;
4876         struct nfs4_openowner *oo, *ret;
4877
4878         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4879         if (!oo)
4880                 return NULL;
4881         oo->oo_owner.so_ops = &openowner_ops;
4882         oo->oo_owner.so_is_open_owner = 1;
4883         oo->oo_owner.so_seqid = open->op_seqid;
4884         oo->oo_flags = 0;
4885         if (nfsd4_has_session(cstate))
4886                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4887         oo->oo_time = 0;
4888         oo->oo_last_closed_stid = NULL;
4889         INIT_LIST_HEAD(&oo->oo_close_lru);
4890         spin_lock(&clp->cl_lock);
4891         ret = find_openstateowner_str_locked(strhashval, open, clp);
4892         if (ret == NULL) {
4893                 hash_openowner(oo, clp, strhashval);
4894                 ret = oo;
4895         } else
4896                 nfs4_free_stateowner(&oo->oo_owner);
4897
4898         spin_unlock(&clp->cl_lock);
4899         return ret;
4900 }
4901
4902 static struct nfs4_ol_stateid *
4903 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4904 {
4905
4906         struct nfs4_openowner *oo = open->op_openowner;
4907         struct nfs4_ol_stateid *retstp = NULL;
4908         struct nfs4_ol_stateid *stp;
4909
4910         stp = open->op_stp;
4911         /* We are moving these outside of the spinlocks to avoid the warnings */
4912         mutex_init(&stp->st_mutex);
4913         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4914
4915 retry:
4916         spin_lock(&oo->oo_owner.so_client->cl_lock);
4917         spin_lock(&fp->fi_lock);
4918
4919         retstp = nfsd4_find_existing_open(fp, open);
4920         if (retstp)
4921                 goto out_unlock;
4922
4923         open->op_stp = NULL;
4924         refcount_inc(&stp->st_stid.sc_count);
4925         stp->st_stid.sc_type = SC_TYPE_OPEN;
4926         INIT_LIST_HEAD(&stp->st_locks);
4927         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4928         get_nfs4_file(fp);
4929         stp->st_stid.sc_file = fp;
4930         stp->st_access_bmap = 0;
4931         stp->st_deny_bmap = 0;
4932         stp->st_openstp = NULL;
4933         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4934         list_add(&stp->st_perfile, &fp->fi_stateids);
4935
4936 out_unlock:
4937         spin_unlock(&fp->fi_lock);
4938         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4939         if (retstp) {
4940                 /* Handle races with CLOSE */
4941                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4942                         nfs4_put_stid(&retstp->st_stid);
4943                         goto retry;
4944                 }
4945                 /* To keep mutex tracking happy */
4946                 mutex_unlock(&stp->st_mutex);
4947                 stp = retstp;
4948         }
4949         return stp;
4950 }
4951
4952 /*
4953  * In the 4.0 case we need to keep the owners around a little while to handle
4954  * CLOSE replay. We still do need to release any file access that is held by
4955  * them before returning however.
4956  */
4957 static void
4958 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4959 {
4960         struct nfs4_ol_stateid *last;
4961         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4962         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4963                                                 nfsd_net_id);
4964
4965         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4966
4967         /*
4968          * We know that we hold one reference via nfsd4_close, and another
4969          * "persistent" reference for the client. If the refcount is higher
4970          * than 2, then there are still calls in progress that are using this
4971          * stateid. We can't put the sc_file reference until they are finished.
4972          * Wait for the refcount to drop to 2. Since it has been unhashed,
4973          * there should be no danger of the refcount going back up again at
4974          * this point.
4975          */
4976         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4977
4978         release_all_access(s);
4979         if (s->st_stid.sc_file) {
4980                 put_nfs4_file(s->st_stid.sc_file);
4981                 s->st_stid.sc_file = NULL;
4982         }
4983
4984         spin_lock(&nn->client_lock);
4985         last = oo->oo_last_closed_stid;
4986         oo->oo_last_closed_stid = s;
4987         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4988         oo->oo_time = ktime_get_boottime_seconds();
4989         spin_unlock(&nn->client_lock);
4990         if (last)
4991                 nfs4_put_stid(&last->st_stid);
4992 }
4993
4994 static noinline_for_stack struct nfs4_file *
4995 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4996 {
4997         struct inode *inode = d_inode(fhp->fh_dentry);
4998         struct rhlist_head *tmp, *list;
4999         struct nfs4_file *fi;
5000
5001         rcu_read_lock();
5002         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5003                                nfs4_file_rhash_params);
5004         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5005                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5006                         if (refcount_inc_not_zero(&fi->fi_ref)) {
5007                                 rcu_read_unlock();
5008                                 return fi;
5009                         }
5010                 }
5011         }
5012         rcu_read_unlock();
5013         return NULL;
5014 }
5015
5016 /*
5017  * On hash insertion, identify entries with the same inode but
5018  * distinct filehandles. They will all be on the list returned
5019  * by rhltable_lookup().
5020  *
5021  * inode->i_lock prevents racing insertions from adding an entry
5022  * for the same inode/fhp pair twice.
5023  */
5024 static noinline_for_stack struct nfs4_file *
5025 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5026 {
5027         struct inode *inode = d_inode(fhp->fh_dentry);
5028         struct rhlist_head *tmp, *list;
5029         struct nfs4_file *ret = NULL;
5030         bool alias_found = false;
5031         struct nfs4_file *fi;
5032         int err;
5033
5034         rcu_read_lock();
5035         spin_lock(&inode->i_lock);
5036
5037         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5038                                nfs4_file_rhash_params);
5039         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5040                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5041                         if (refcount_inc_not_zero(&fi->fi_ref))
5042                                 ret = fi;
5043                 } else
5044                         fi->fi_aliased = alias_found = true;
5045         }
5046         if (ret)
5047                 goto out_unlock;
5048
5049         nfsd4_file_init(fhp, new);
5050         err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5051                               nfs4_file_rhash_params);
5052         if (err)
5053                 goto out_unlock;
5054
5055         new->fi_aliased = alias_found;
5056         ret = new;
5057
5058 out_unlock:
5059         spin_unlock(&inode->i_lock);
5060         rcu_read_unlock();
5061         return ret;
5062 }
5063
5064 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5065 {
5066         rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5067                         nfs4_file_rhash_params);
5068 }
5069
5070 /*
5071  * Called to check deny when READ with all zero stateid or
5072  * WRITE with all zero or all one stateid
5073  */
5074 static __be32
5075 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5076 {
5077         struct nfs4_file *fp;
5078         __be32 ret = nfs_ok;
5079
5080         fp = nfsd4_file_hash_lookup(current_fh);
5081         if (!fp)
5082                 return ret;
5083
5084         /* Check for conflicting share reservations */
5085         spin_lock(&fp->fi_lock);
5086         if (fp->fi_share_deny & deny_type)
5087                 ret = nfserr_locked;
5088         spin_unlock(&fp->fi_lock);
5089         put_nfs4_file(fp);
5090         return ret;
5091 }
5092
5093 static bool nfsd4_deleg_present(const struct inode *inode)
5094 {
5095         struct file_lock_context *ctx = locks_inode_context(inode);
5096
5097         return ctx && !list_empty_careful(&ctx->flc_lease);
5098 }
5099
5100 /**
5101  * nfsd_wait_for_delegreturn - wait for delegations to be returned
5102  * @rqstp: the RPC transaction being executed
5103  * @inode: in-core inode of the file being waited for
5104  *
5105  * The timeout prevents deadlock if all nfsd threads happen to be
5106  * tied up waiting for returning delegations.
5107  *
5108  * Return values:
5109  *   %true: delegation was returned
5110  *   %false: timed out waiting for delegreturn
5111  */
5112 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5113 {
5114         long __maybe_unused timeo;
5115
5116         timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5117                                        NFSD_DELEGRETURN_TIMEOUT);
5118         trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5119         return timeo > 0;
5120 }
5121
5122 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5123 {
5124         struct nfs4_delegation *dp = cb_to_delegation(cb);
5125         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5126                                           nfsd_net_id);
5127
5128         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5129
5130         /*
5131          * We can't do this in nfsd_break_deleg_cb because it is
5132          * already holding inode->i_lock.
5133          *
5134          * If the dl_time != 0, then we know that it has already been
5135          * queued for a lease break. Don't queue it again.
5136          */
5137         spin_lock(&state_lock);
5138         if (delegation_hashed(dp) && dp->dl_time == 0) {
5139                 dp->dl_time = ktime_get_boottime_seconds();
5140                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5141         }
5142         spin_unlock(&state_lock);
5143 }
5144
5145 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5146                 struct rpc_task *task)
5147 {
5148         struct nfs4_delegation *dp = cb_to_delegation(cb);
5149
5150         trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5151
5152         if (dp->dl_stid.sc_status)
5153                 /* CLOSED or REVOKED */
5154                 return 1;
5155
5156         switch (task->tk_status) {
5157         case 0:
5158                 return 1;
5159         case -NFS4ERR_DELAY:
5160                 rpc_delay(task, 2 * HZ);
5161                 return 0;
5162         case -EBADHANDLE:
5163         case -NFS4ERR_BAD_STATEID:
5164                 /*
5165                  * Race: client probably got cb_recall before open reply
5166                  * granting delegation.
5167                  */
5168                 if (dp->dl_retries--) {
5169                         rpc_delay(task, 2 * HZ);
5170                         return 0;
5171                 }
5172                 fallthrough;
5173         default:
5174                 return 1;
5175         }
5176 }
5177
5178 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5179 {
5180         struct nfs4_delegation *dp = cb_to_delegation(cb);
5181
5182         nfs4_put_stid(&dp->dl_stid);
5183 }
5184
5185 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5186         .prepare        = nfsd4_cb_recall_prepare,
5187         .done           = nfsd4_cb_recall_done,
5188         .release        = nfsd4_cb_recall_release,
5189 };
5190
5191 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5192 {
5193         /*
5194          * We're assuming the state code never drops its reference
5195          * without first removing the lease.  Since we're in this lease
5196          * callback (and since the lease code is serialized by the
5197          * flc_lock) we know the server hasn't removed the lease yet, and
5198          * we know it's safe to take a reference.
5199          */
5200         refcount_inc(&dp->dl_stid.sc_count);
5201         WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5202 }
5203
5204 /* Called from break_lease() with flc_lock held. */
5205 static bool
5206 nfsd_break_deleg_cb(struct file_lease *fl)
5207 {
5208         struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5209         struct nfs4_file *fp = dp->dl_stid.sc_file;
5210         struct nfs4_client *clp = dp->dl_stid.sc_client;
5211         struct nfsd_net *nn;
5212
5213         trace_nfsd_cb_recall(&dp->dl_stid);
5214
5215         dp->dl_recalled = true;
5216         atomic_inc(&clp->cl_delegs_in_recall);
5217         if (try_to_expire_client(clp)) {
5218                 nn = net_generic(clp->net, nfsd_net_id);
5219                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5220         }
5221
5222         /*
5223          * We don't want the locks code to timeout the lease for us;
5224          * we'll remove it ourself if a delegation isn't returned
5225          * in time:
5226          */
5227         fl->fl_break_time = 0;
5228
5229         fp->fi_had_conflict = true;
5230         nfsd_break_one_deleg(dp);
5231         return false;
5232 }
5233
5234 /**
5235  * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5236  * @fl: Lock state to check
5237  *
5238  * Return values:
5239  *   %true: Lease conflict was resolved
5240  *   %false: Lease conflict was not resolved.
5241  */
5242 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5243 {
5244         struct nfs4_delegation *dl = fl->c.flc_owner;
5245         struct svc_rqst *rqst;
5246         struct nfs4_client *clp;
5247
5248         if (!i_am_nfsd())
5249                 return false;
5250         rqst = kthread_data(current);
5251         /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
5252         if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
5253                 return false;
5254         clp = *(rqst->rq_lease_breaker);
5255         return dl->dl_stid.sc_client == clp;
5256 }
5257
5258 static int
5259 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5260                      struct list_head *dispose)
5261 {
5262         struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5263         struct nfs4_client *clp = dp->dl_stid.sc_client;
5264
5265         if (arg & F_UNLCK) {
5266                 if (dp->dl_recalled)
5267                         atomic_dec(&clp->cl_delegs_in_recall);
5268                 return lease_modify(onlist, arg, dispose);
5269         } else
5270                 return -EAGAIN;
5271 }
5272
5273 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5274         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5275         .lm_break = nfsd_break_deleg_cb,
5276         .lm_change = nfsd_change_deleg_cb,
5277 };
5278
5279 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5280 {
5281         if (nfsd4_has_session(cstate))
5282                 return nfs_ok;
5283         if (seqid == so->so_seqid - 1)
5284                 return nfserr_replay_me;
5285         if (seqid == so->so_seqid)
5286                 return nfs_ok;
5287         return nfserr_bad_seqid;
5288 }
5289
5290 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5291                                                 struct nfsd_net *nn)
5292 {
5293         struct nfs4_client *found;
5294
5295         spin_lock(&nn->client_lock);
5296         found = find_confirmed_client(clid, sessions, nn);
5297         if (found)
5298                 atomic_inc(&found->cl_rpc_users);
5299         spin_unlock(&nn->client_lock);
5300         return found;
5301 }
5302
5303 static __be32 set_client(clientid_t *clid,
5304                 struct nfsd4_compound_state *cstate,
5305                 struct nfsd_net *nn)
5306 {
5307         if (cstate->clp) {
5308                 if (!same_clid(&cstate->clp->cl_clientid, clid))
5309                         return nfserr_stale_clientid;
5310                 return nfs_ok;
5311         }
5312         if (STALE_CLIENTID(clid, nn))
5313                 return nfserr_stale_clientid;
5314         /*
5315          * We're in the 4.0 case (otherwise the SEQUENCE op would have
5316          * set cstate->clp), so session = false:
5317          */
5318         cstate->clp = lookup_clientid(clid, false, nn);
5319         if (!cstate->clp)
5320                 return nfserr_expired;
5321         return nfs_ok;
5322 }
5323
5324 __be32
5325 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5326                     struct nfsd4_open *open, struct nfsd_net *nn)
5327 {
5328         clientid_t *clientid = &open->op_clientid;
5329         struct nfs4_client *clp = NULL;
5330         unsigned int strhashval;
5331         struct nfs4_openowner *oo = NULL;
5332         __be32 status;
5333
5334         /*
5335          * In case we need it later, after we've already created the
5336          * file and don't want to risk a further failure:
5337          */
5338         open->op_file = nfsd4_alloc_file();
5339         if (open->op_file == NULL)
5340                 return nfserr_jukebox;
5341
5342         status = set_client(clientid, cstate, nn);
5343         if (status)
5344                 return status;
5345         clp = cstate->clp;
5346
5347         strhashval = ownerstr_hashval(&open->op_owner);
5348         oo = find_openstateowner_str(strhashval, open, clp);
5349         open->op_openowner = oo;
5350         if (!oo) {
5351                 goto new_owner;
5352         }
5353         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5354                 /* Replace unconfirmed owners without checking for replay. */
5355                 release_openowner(oo);
5356                 open->op_openowner = NULL;
5357                 goto new_owner;
5358         }
5359         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5360         if (status)
5361                 return status;
5362         goto alloc_stateid;
5363 new_owner:
5364         oo = alloc_init_open_stateowner(strhashval, open, cstate);
5365         if (oo == NULL)
5366                 return nfserr_jukebox;
5367         open->op_openowner = oo;
5368 alloc_stateid:
5369         open->op_stp = nfs4_alloc_open_stateid(clp);
5370         if (!open->op_stp)
5371                 return nfserr_jukebox;
5372
5373         if (nfsd4_has_session(cstate) &&
5374             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5375                 open->op_odstate = alloc_clnt_odstate(clp);
5376                 if (!open->op_odstate)
5377                         return nfserr_jukebox;
5378         }
5379
5380         return nfs_ok;
5381 }
5382
5383 static inline __be32
5384 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5385 {
5386         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5387                 return nfserr_openmode;
5388         else
5389                 return nfs_ok;
5390 }
5391
5392 static int share_access_to_flags(u32 share_access)
5393 {
5394         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5395 }
5396
5397 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5398                                                   stateid_t *s)
5399 {
5400         struct nfs4_stid *ret;
5401
5402         ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5403         if (!ret)
5404                 return NULL;
5405         return delegstateid(ret);
5406 }
5407
5408 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5409 {
5410         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5411                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5412 }
5413
5414 static __be32
5415 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5416                 struct nfs4_delegation **dp)
5417 {
5418         int flags;
5419         __be32 status = nfserr_bad_stateid;
5420         struct nfs4_delegation *deleg;
5421
5422         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5423         if (deleg == NULL)
5424                 goto out;
5425         if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5426                 nfs4_put_stid(&deleg->dl_stid);
5427                 status = nfserr_admin_revoked;
5428                 goto out;
5429         }
5430         if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5431                 nfs4_put_stid(&deleg->dl_stid);
5432                 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5433                 status = nfserr_deleg_revoked;
5434                 goto out;
5435         }
5436         flags = share_access_to_flags(open->op_share_access);
5437         status = nfs4_check_delegmode(deleg, flags);
5438         if (status) {
5439                 nfs4_put_stid(&deleg->dl_stid);
5440                 goto out;
5441         }
5442         *dp = deleg;
5443 out:
5444         if (!nfsd4_is_deleg_cur(open))
5445                 return nfs_ok;
5446         if (status)
5447                 return status;
5448         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5449         return nfs_ok;
5450 }
5451
5452 static inline int nfs4_access_to_access(u32 nfs4_access)
5453 {
5454         int flags = 0;
5455
5456         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5457                 flags |= NFSD_MAY_READ;
5458         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5459                 flags |= NFSD_MAY_WRITE;
5460         return flags;
5461 }
5462
5463 static inline __be32
5464 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5465                 struct nfsd4_open *open)
5466 {
5467         struct iattr iattr = {
5468                 .ia_valid = ATTR_SIZE,
5469                 .ia_size = 0,
5470         };
5471         struct nfsd_attrs attrs = {
5472                 .na_iattr       = &iattr,
5473         };
5474         if (!open->op_truncate)
5475                 return 0;
5476         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5477                 return nfserr_inval;
5478         return nfsd_setattr(rqstp, fh, &attrs, NULL);
5479 }
5480
5481 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5482                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5483                 struct nfsd4_open *open, bool new_stp)
5484 {
5485         struct nfsd_file *nf = NULL;
5486         __be32 status;
5487         int oflag = nfs4_access_to_omode(open->op_share_access);
5488         int access = nfs4_access_to_access(open->op_share_access);
5489         unsigned char old_access_bmap, old_deny_bmap;
5490
5491         spin_lock(&fp->fi_lock);
5492
5493         /*
5494          * Are we trying to set a deny mode that would conflict with
5495          * current access?
5496          */
5497         status = nfs4_file_check_deny(fp, open->op_share_deny);
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_deny, false))
5505                         status = nfserr_jukebox;
5506                 spin_unlock(&fp->fi_lock);
5507                 goto out;
5508         }
5509
5510         /* set access to the file */
5511         status = nfs4_file_get_access(fp, open->op_share_access);
5512         if (status != nfs_ok) {
5513                 if (status != nfserr_share_denied) {
5514                         spin_unlock(&fp->fi_lock);
5515                         goto out;
5516                 }
5517                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5518                                 stp, open->op_share_access, true))
5519                         status = nfserr_jukebox;
5520                 spin_unlock(&fp->fi_lock);
5521                 goto out;
5522         }
5523
5524         /* Set access bits in stateid */
5525         old_access_bmap = stp->st_access_bmap;
5526         set_access(open->op_share_access, stp);
5527
5528         /* Set new deny mask */
5529         old_deny_bmap = stp->st_deny_bmap;
5530         set_deny(open->op_share_deny, stp);
5531         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5532
5533         if (!fp->fi_fds[oflag]) {
5534                 spin_unlock(&fp->fi_lock);
5535
5536                 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5537                                                   open->op_filp, &nf);
5538                 if (status != nfs_ok)
5539                         goto out_put_access;
5540
5541                 spin_lock(&fp->fi_lock);
5542                 if (!fp->fi_fds[oflag]) {
5543                         fp->fi_fds[oflag] = nf;
5544                         nf = NULL;
5545                 }
5546         }
5547         spin_unlock(&fp->fi_lock);
5548         if (nf)
5549                 nfsd_file_put(nf);
5550
5551         status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5552                                                                 access));
5553         if (status)
5554                 goto out_put_access;
5555
5556         status = nfsd4_truncate(rqstp, cur_fh, open);
5557         if (status)
5558                 goto out_put_access;
5559 out:
5560         return status;
5561 out_put_access:
5562         stp->st_access_bmap = old_access_bmap;
5563         nfs4_file_put_access(fp, open->op_share_access);
5564         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5565         goto out;
5566 }
5567
5568 static __be32
5569 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5570                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5571                 struct nfsd4_open *open)
5572 {
5573         __be32 status;
5574         unsigned char old_deny_bmap = stp->st_deny_bmap;
5575
5576         if (!test_access(open->op_share_access, stp))
5577                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5578
5579         /* test and set deny mode */
5580         spin_lock(&fp->fi_lock);
5581         status = nfs4_file_check_deny(fp, open->op_share_deny);
5582         switch (status) {
5583         case nfs_ok:
5584                 set_deny(open->op_share_deny, stp);
5585                 fp->fi_share_deny |=
5586                         (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5587                 break;
5588         case nfserr_share_denied:
5589                 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5590                                 stp, open->op_share_deny, false))
5591                         status = nfserr_jukebox;
5592                 break;
5593         }
5594         spin_unlock(&fp->fi_lock);
5595
5596         if (status != nfs_ok)
5597                 return status;
5598
5599         status = nfsd4_truncate(rqstp, cur_fh, open);
5600         if (status != nfs_ok)
5601                 reset_union_bmap_deny(old_deny_bmap, stp);
5602         return status;
5603 }
5604
5605 /* Should we give out recallable state?: */
5606 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5607 {
5608         if (clp->cl_cb_state == NFSD4_CB_UP)
5609                 return true;
5610         /*
5611          * In the sessions case, since we don't have to establish a
5612          * separate connection for callbacks, we assume it's OK
5613          * until we hear otherwise:
5614          */
5615         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5616 }
5617
5618 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5619                                                 int flag)
5620 {
5621         struct file_lease *fl;
5622
5623         fl = locks_alloc_lease();
5624         if (!fl)
5625                 return NULL;
5626         fl->fl_lmops = &nfsd_lease_mng_ops;
5627         fl->c.flc_flags = FL_DELEG;
5628         fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5629         fl->c.flc_owner = (fl_owner_t)dp;
5630         fl->c.flc_pid = current->tgid;
5631         fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5632         return fl;
5633 }
5634
5635 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5636                                          struct nfs4_file *fp)
5637 {
5638         struct nfs4_ol_stateid *st;
5639         struct file *f = fp->fi_deleg_file->nf_file;
5640         struct inode *ino = file_inode(f);
5641         int writes;
5642
5643         writes = atomic_read(&ino->i_writecount);
5644         if (!writes)
5645                 return 0;
5646         /*
5647          * There could be multiple filehandles (hence multiple
5648          * nfs4_files) referencing this file, but that's not too
5649          * common; let's just give up in that case rather than
5650          * trying to go look up all the clients using that other
5651          * nfs4_file as well:
5652          */
5653         if (fp->fi_aliased)
5654                 return -EAGAIN;
5655         /*
5656          * If there's a close in progress, make sure that we see it
5657          * clear any fi_fds[] entries before we see it decrement
5658          * i_writecount:
5659          */
5660         smp_mb__after_atomic();
5661
5662         if (fp->fi_fds[O_WRONLY])
5663                 writes--;
5664         if (fp->fi_fds[O_RDWR])
5665                 writes--;
5666         if (writes > 0)
5667                 return -EAGAIN; /* There may be non-NFSv4 writers */
5668         /*
5669          * It's possible there are non-NFSv4 write opens in progress,
5670          * but if they haven't incremented i_writecount yet then they
5671          * also haven't called break lease yet; so, they'll break this
5672          * lease soon enough.  So, all that's left to check for is NFSv4
5673          * opens:
5674          */
5675         spin_lock(&fp->fi_lock);
5676         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5677                 if (st->st_openstp == NULL /* it's an open */ &&
5678                     access_permit_write(st) &&
5679                     st->st_stid.sc_client != clp) {
5680                         spin_unlock(&fp->fi_lock);
5681                         return -EAGAIN;
5682                 }
5683         }
5684         spin_unlock(&fp->fi_lock);
5685         /*
5686          * There's a small chance that we could be racing with another
5687          * NFSv4 open.  However, any open that hasn't added itself to
5688          * the fi_stateids list also hasn't called break_lease yet; so,
5689          * they'll break this lease soon enough.
5690          */
5691         return 0;
5692 }
5693
5694 /*
5695  * It's possible that between opening the dentry and setting the delegation,
5696  * that it has been renamed or unlinked. Redo the lookup to verify that this
5697  * hasn't happened.
5698  */
5699 static int
5700 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5701                           struct svc_fh *parent)
5702 {
5703         struct svc_export *exp;
5704         struct dentry *child;
5705         __be32 err;
5706
5707         err = nfsd_lookup_dentry(open->op_rqstp, parent,
5708                                  open->op_fname, open->op_fnamelen,
5709                                  &exp, &child);
5710
5711         if (err)
5712                 return -EAGAIN;
5713
5714         exp_put(exp);
5715         dput(child);
5716         if (child != file_dentry(fp->fi_deleg_file->nf_file))
5717                 return -EAGAIN;
5718
5719         return 0;
5720 }
5721
5722 /*
5723  * We avoid breaking delegations held by a client due to its own activity, but
5724  * clearing setuid/setgid bits on a write is an implicit activity and the client
5725  * may not notice and continue using the old mode. Avoid giving out a delegation
5726  * on setuid/setgid files when the client is requesting an open for write.
5727  */
5728 static int
5729 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5730 {
5731         struct inode *inode = file_inode(nf->nf_file);
5732
5733         if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5734             (inode->i_mode & (S_ISUID|S_ISGID)))
5735                 return -EAGAIN;
5736         return 0;
5737 }
5738
5739 static struct nfs4_delegation *
5740 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5741                     struct svc_fh *parent)
5742 {
5743         int status = 0;
5744         struct nfs4_client *clp = stp->st_stid.sc_client;
5745         struct nfs4_file *fp = stp->st_stid.sc_file;
5746         struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5747         struct nfs4_delegation *dp;
5748         struct nfsd_file *nf = NULL;
5749         struct file_lease *fl;
5750         u32 dl_type;
5751
5752         /*
5753          * The fi_had_conflict and nfs_get_existing_delegation checks
5754          * here are just optimizations; we'll need to recheck them at
5755          * the end:
5756          */
5757         if (fp->fi_had_conflict)
5758                 return ERR_PTR(-EAGAIN);
5759
5760         /*
5761          * Try for a write delegation first. RFC8881 section 10.4 says:
5762          *
5763          *  "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5764          *   on its own, all opens."
5765          *
5766          * Furthermore the client can use a write delegation for most READ
5767          * operations as well, so we require a O_RDWR file here.
5768          *
5769          * Offer a write delegation in the case of a BOTH open, and ensure
5770          * we get the O_RDWR descriptor.
5771          */
5772         if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5773                 nf = find_rw_file(fp);
5774                 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5775         }
5776
5777         /*
5778          * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5779          * file for some reason, then try for a read delegation instead.
5780          */
5781         if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5782                 nf = find_readable_file(fp);
5783                 dl_type = NFS4_OPEN_DELEGATE_READ;
5784         }
5785
5786         if (!nf)
5787                 return ERR_PTR(-EAGAIN);
5788
5789         spin_lock(&state_lock);
5790         spin_lock(&fp->fi_lock);
5791         if (nfs4_delegation_exists(clp, fp))
5792                 status = -EAGAIN;
5793         else if (nfsd4_verify_setuid_write(open, nf))
5794                 status = -EAGAIN;
5795         else if (!fp->fi_deleg_file) {
5796                 fp->fi_deleg_file = nf;
5797                 /* increment early to prevent fi_deleg_file from being
5798                  * cleared */
5799                 fp->fi_delegees = 1;
5800                 nf = NULL;
5801         } else
5802                 fp->fi_delegees++;
5803         spin_unlock(&fp->fi_lock);
5804         spin_unlock(&state_lock);
5805         if (nf)
5806                 nfsd_file_put(nf);
5807         if (status)
5808                 return ERR_PTR(status);
5809
5810         status = -ENOMEM;
5811         dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5812         if (!dp)
5813                 goto out_delegees;
5814
5815         fl = nfs4_alloc_init_lease(dp, dl_type);
5816         if (!fl)
5817                 goto out_clnt_odstate;
5818
5819         status = kernel_setlease(fp->fi_deleg_file->nf_file,
5820                                       fl->c.flc_type, &fl, NULL);
5821         if (fl)
5822                 locks_free_lease(fl);
5823         if (status)
5824                 goto out_clnt_odstate;
5825
5826         if (parent) {
5827                 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5828                 if (status)
5829                         goto out_unlock;
5830         }
5831
5832         status = nfsd4_check_conflicting_opens(clp, fp);
5833         if (status)
5834                 goto out_unlock;
5835
5836         /*
5837          * Now that the deleg is set, check again to ensure that nothing
5838          * raced in and changed the mode while we weren't lookng.
5839          */
5840         status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5841         if (status)
5842                 goto out_unlock;
5843
5844         status = -EAGAIN;
5845         if (fp->fi_had_conflict)
5846                 goto out_unlock;
5847
5848         spin_lock(&state_lock);
5849         spin_lock(&clp->cl_lock);
5850         spin_lock(&fp->fi_lock);
5851         status = hash_delegation_locked(dp, fp);
5852         spin_unlock(&fp->fi_lock);
5853         spin_unlock(&clp->cl_lock);
5854         spin_unlock(&state_lock);
5855
5856         if (status)
5857                 goto out_unlock;
5858
5859         return dp;
5860 out_unlock:
5861         kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5862 out_clnt_odstate:
5863         put_clnt_odstate(dp->dl_clnt_odstate);
5864         nfs4_put_stid(&dp->dl_stid);
5865 out_delegees:
5866         put_deleg_file(fp);
5867         return ERR_PTR(status);
5868 }
5869
5870 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5871 {
5872         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5873         if (status == -EAGAIN)
5874                 open->op_why_no_deleg = WND4_CONTENTION;
5875         else {
5876                 open->op_why_no_deleg = WND4_RESOURCE;
5877                 switch (open->op_deleg_want) {
5878                 case NFS4_SHARE_WANT_READ_DELEG:
5879                 case NFS4_SHARE_WANT_WRITE_DELEG:
5880                 case NFS4_SHARE_WANT_ANY_DELEG:
5881                         break;
5882                 case NFS4_SHARE_WANT_CANCEL:
5883                         open->op_why_no_deleg = WND4_CANCELLED;
5884                         break;
5885                 case NFS4_SHARE_WANT_NO_DELEG:
5886                         WARN_ON_ONCE(1);
5887                 }
5888         }
5889 }
5890
5891 /*
5892  * The Linux NFS server does not offer write delegations to NFSv4.0
5893  * clients in order to avoid conflicts between write delegations and
5894  * GETATTRs requesting CHANGE or SIZE attributes.
5895  *
5896  * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5897  * begins each COMPOUND contains a client ID. Delegation recall can
5898  * be avoided when the server recognizes the client sending a
5899  * GETATTR also holds write delegation it conflicts with.
5900  *
5901  * However, the NFSv4.0 protocol does not enable a server to
5902  * determine that a GETATTR originated from the client holding the
5903  * conflicting delegation versus coming from some other client. Per
5904  * RFC 7530 Section 16.7.5, the server must recall or send a
5905  * CB_GETATTR even when the GETATTR originates from the client that
5906  * holds the conflicting delegation.
5907  *
5908  * An NFSv4.0 client can trigger a pathological situation if it
5909  * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5910  * the same COMPOUND. COMPOUND execution will always stop at the
5911  * GETATTR and the DELEGRETURN will never get executed. The server
5912  * eventually revokes the delegation, which can result in loss of
5913  * open or lock state.
5914  */
5915 static void
5916 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5917                      struct svc_fh *currentfh)
5918 {
5919         struct nfs4_delegation *dp;
5920         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5921         struct nfs4_client *clp = stp->st_stid.sc_client;
5922         struct svc_fh *parent = NULL;
5923         int cb_up;
5924         int status = 0;
5925         struct kstat stat;
5926         struct path path;
5927
5928         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5929         open->op_recall = false;
5930         switch (open->op_claim_type) {
5931                 case NFS4_OPEN_CLAIM_PREVIOUS:
5932                         if (!cb_up)
5933                                 open->op_recall = true;
5934                         break;
5935                 case NFS4_OPEN_CLAIM_NULL:
5936                         parent = currentfh;
5937                         fallthrough;
5938                 case NFS4_OPEN_CLAIM_FH:
5939                         /*
5940                          * Let's not give out any delegations till everyone's
5941                          * had the chance to reclaim theirs, *and* until
5942                          * NLM locks have all been reclaimed:
5943                          */
5944                         if (locks_in_grace(clp->net))
5945                                 goto out_no_deleg;
5946                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5947                                 goto out_no_deleg;
5948                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5949                                         !clp->cl_minorversion)
5950                                 goto out_no_deleg;
5951                         break;
5952                 default:
5953                         goto out_no_deleg;
5954         }
5955         dp = nfs4_set_delegation(open, stp, parent);
5956         if (IS_ERR(dp))
5957                 goto out_no_deleg;
5958
5959         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5960
5961         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5962                 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5963                 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5964                 path.mnt = currentfh->fh_export->ex_path.mnt;
5965                 path.dentry = currentfh->fh_dentry;
5966                 if (vfs_getattr(&path, &stat,
5967                                 (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5968                                 AT_STATX_SYNC_AS_STAT)) {
5969                         nfs4_put_stid(&dp->dl_stid);
5970                         destroy_delegation(dp);
5971                         goto out_no_deleg;
5972                 }
5973                 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
5974                 dp->dl_cb_fattr.ncf_initial_cinfo =
5975                         nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
5976         } else {
5977                 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5978                 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5979         }
5980         nfs4_put_stid(&dp->dl_stid);
5981         return;
5982 out_no_deleg:
5983         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5984         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5985             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5986                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5987                 open->op_recall = true;
5988         }
5989
5990         /* 4.1 client asking for a delegation? */
5991         if (open->op_deleg_want)
5992                 nfsd4_open_deleg_none_ext(open, status);
5993         return;
5994 }
5995
5996 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5997                                         struct nfs4_delegation *dp)
5998 {
5999         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
6000             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6001                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6002                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
6003         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
6004                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6005                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6006                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
6007         }
6008         /* Otherwise the client must be confused wanting a delegation
6009          * it already has, therefore we don't return
6010          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
6011          */
6012 }
6013
6014 /**
6015  * nfsd4_process_open2 - finish open processing
6016  * @rqstp: the RPC transaction being executed
6017  * @current_fh: NFSv4 COMPOUND's current filehandle
6018  * @open: OPEN arguments
6019  *
6020  * If successful, (1) truncate the file if open->op_truncate was
6021  * set, (2) set open->op_stateid, (3) set open->op_delegation.
6022  *
6023  * Returns %nfs_ok on success; otherwise an nfs4stat value in
6024  * network byte order is returned.
6025  */
6026 __be32
6027 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6028 {
6029         struct nfsd4_compoundres *resp = rqstp->rq_resp;
6030         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6031         struct nfs4_file *fp = NULL;
6032         struct nfs4_ol_stateid *stp = NULL;
6033         struct nfs4_delegation *dp = NULL;
6034         __be32 status;
6035         bool new_stp = false;
6036
6037         /*
6038          * Lookup file; if found, lookup stateid and check open request,
6039          * and check for delegations in the process of being recalled.
6040          * If not found, create the nfs4_file struct
6041          */
6042         fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6043         if (unlikely(!fp))
6044                 return nfserr_jukebox;
6045         if (fp != open->op_file) {
6046                 status = nfs4_check_deleg(cl, open, &dp);
6047                 if (status)
6048                         goto out;
6049                 stp = nfsd4_find_and_lock_existing_open(fp, open);
6050         } else {
6051                 open->op_file = NULL;
6052                 status = nfserr_bad_stateid;
6053                 if (nfsd4_is_deleg_cur(open))
6054                         goto out;
6055         }
6056
6057         if (!stp) {
6058                 stp = init_open_stateid(fp, open);
6059                 if (!open->op_stp)
6060                         new_stp = true;
6061         }
6062
6063         /*
6064          * OPEN the file, or upgrade an existing OPEN.
6065          * If truncate fails, the OPEN fails.
6066          *
6067          * stp is already locked.
6068          */
6069         if (!new_stp) {
6070                 /* Stateid was found, this is an OPEN upgrade */
6071                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6072                 if (status) {
6073                         mutex_unlock(&stp->st_mutex);
6074                         goto out;
6075                 }
6076         } else {
6077                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6078                 if (status) {
6079                         release_open_stateid(stp);
6080                         mutex_unlock(&stp->st_mutex);
6081                         goto out;
6082                 }
6083
6084                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6085                                                         open->op_odstate);
6086                 if (stp->st_clnt_odstate == open->op_odstate)
6087                         open->op_odstate = NULL;
6088         }
6089
6090         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6091         mutex_unlock(&stp->st_mutex);
6092
6093         if (nfsd4_has_session(&resp->cstate)) {
6094                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6095                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6096                         open->op_why_no_deleg = WND4_NOT_WANTED;
6097                         goto nodeleg;
6098                 }
6099         }
6100
6101         /*
6102         * Attempt to hand out a delegation. No error return, because the
6103         * OPEN succeeds even if we fail.
6104         */
6105         nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6106 nodeleg:
6107         status = nfs_ok;
6108         trace_nfsd_open(&stp->st_stid.sc_stateid);
6109 out:
6110         /* 4.1 client trying to upgrade/downgrade delegation? */
6111         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6112             open->op_deleg_want)
6113                 nfsd4_deleg_xgrade_none_ext(open, dp);
6114
6115         if (fp)
6116                 put_nfs4_file(fp);
6117         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6118                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6119         /*
6120         * To finish the open response, we just need to set the rflags.
6121         */
6122         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6123         if (nfsd4_has_session(&resp->cstate))
6124                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6125         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6126                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6127
6128         if (dp)
6129                 nfs4_put_stid(&dp->dl_stid);
6130         if (stp)
6131                 nfs4_put_stid(&stp->st_stid);
6132
6133         return status;
6134 }
6135
6136 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6137                               struct nfsd4_open *open)
6138 {
6139         if (open->op_openowner) {
6140                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
6141
6142                 nfsd4_cstate_assign_replay(cstate, so);
6143                 nfs4_put_stateowner(so);
6144         }
6145         if (open->op_file)
6146                 kmem_cache_free(file_slab, open->op_file);
6147         if (open->op_stp)
6148                 nfs4_put_stid(&open->op_stp->st_stid);
6149         if (open->op_odstate)
6150                 kmem_cache_free(odstate_slab, open->op_odstate);
6151 }
6152
6153 __be32
6154 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6155             union nfsd4_op_u *u)
6156 {
6157         clientid_t *clid = &u->renew;
6158         struct nfs4_client *clp;
6159         __be32 status;
6160         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6161
6162         trace_nfsd_clid_renew(clid);
6163         status = set_client(clid, cstate, nn);
6164         if (status)
6165                 return status;
6166         clp = cstate->clp;
6167         if (!list_empty(&clp->cl_delegations)
6168                         && clp->cl_cb_state != NFSD4_CB_UP)
6169                 return nfserr_cb_path_down;
6170         return nfs_ok;
6171 }
6172
6173 void
6174 nfsd4_end_grace(struct nfsd_net *nn)
6175 {
6176         /* do nothing if grace period already ended */
6177         if (nn->grace_ended)
6178                 return;
6179
6180         trace_nfsd_grace_complete(nn);
6181         nn->grace_ended = true;
6182         /*
6183          * If the server goes down again right now, an NFSv4
6184          * client will still be allowed to reclaim after it comes back up,
6185          * even if it hasn't yet had a chance to reclaim state this time.
6186          *
6187          */
6188         nfsd4_record_grace_done(nn);
6189         /*
6190          * At this point, NFSv4 clients can still reclaim.  But if the
6191          * server crashes, any that have not yet reclaimed will be out
6192          * of luck on the next boot.
6193          *
6194          * (NFSv4.1+ clients are considered to have reclaimed once they
6195          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
6196          * have reclaimed after their first OPEN.)
6197          */
6198         locks_end_grace(&nn->nfsd4_manager);
6199         /*
6200          * At this point, and once lockd and/or any other containers
6201          * exit their grace period, further reclaims will fail and
6202          * regular locking can resume.
6203          */
6204 }
6205
6206 /*
6207  * If we've waited a lease period but there are still clients trying to
6208  * reclaim, wait a little longer to give them a chance to finish.
6209  */
6210 static bool clients_still_reclaiming(struct nfsd_net *nn)
6211 {
6212         time64_t double_grace_period_end = nn->boot_time +
6213                                            2 * nn->nfsd4_lease;
6214
6215         if (nn->track_reclaim_completes &&
6216                         atomic_read(&nn->nr_reclaim_complete) ==
6217                         nn->reclaim_str_hashtbl_size)
6218                 return false;
6219         if (!nn->somebody_reclaimed)
6220                 return false;
6221         nn->somebody_reclaimed = false;
6222         /*
6223          * If we've given them *two* lease times to reclaim, and they're
6224          * still not done, give up:
6225          */
6226         if (ktime_get_boottime_seconds() > double_grace_period_end)
6227                 return false;
6228         return true;
6229 }
6230
6231 struct laundry_time {
6232         time64_t cutoff;
6233         time64_t new_timeo;
6234 };
6235
6236 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6237 {
6238         time64_t time_remaining;
6239
6240         if (last_refresh < lt->cutoff)
6241                 return true;
6242         time_remaining = last_refresh - lt->cutoff;
6243         lt->new_timeo = min(lt->new_timeo, time_remaining);
6244         return false;
6245 }
6246
6247 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6248 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6249 {
6250         spin_lock_init(&nn->nfsd_ssc_lock);
6251         INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6252         init_waitqueue_head(&nn->nfsd_ssc_waitq);
6253 }
6254 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
6255
6256 /*
6257  * This is called when nfsd is being shutdown, after all inter_ssc
6258  * cleanup were done, to destroy the ssc delayed unmount list.
6259  */
6260 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6261 {
6262         struct nfsd4_ssc_umount_item *ni = NULL;
6263         struct nfsd4_ssc_umount_item *tmp;
6264
6265         spin_lock(&nn->nfsd_ssc_lock);
6266         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6267                 list_del(&ni->nsui_list);
6268                 spin_unlock(&nn->nfsd_ssc_lock);
6269                 mntput(ni->nsui_vfsmount);
6270                 kfree(ni);
6271                 spin_lock(&nn->nfsd_ssc_lock);
6272         }
6273         spin_unlock(&nn->nfsd_ssc_lock);
6274 }
6275
6276 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6277 {
6278         bool do_wakeup = false;
6279         struct nfsd4_ssc_umount_item *ni = NULL;
6280         struct nfsd4_ssc_umount_item *tmp;
6281
6282         spin_lock(&nn->nfsd_ssc_lock);
6283         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6284                 if (time_after(jiffies, ni->nsui_expire)) {
6285                         if (refcount_read(&ni->nsui_refcnt) > 1)
6286                                 continue;
6287
6288                         /* mark being unmount */
6289                         ni->nsui_busy = true;
6290                         spin_unlock(&nn->nfsd_ssc_lock);
6291                         mntput(ni->nsui_vfsmount);
6292                         spin_lock(&nn->nfsd_ssc_lock);
6293
6294                         /* waiters need to start from begin of list */
6295                         list_del(&ni->nsui_list);
6296                         kfree(ni);
6297
6298                         /* wakeup ssc_connect waiters */
6299                         do_wakeup = true;
6300                         continue;
6301                 }
6302                 break;
6303         }
6304         if (do_wakeup)
6305                 wake_up_all(&nn->nfsd_ssc_waitq);
6306         spin_unlock(&nn->nfsd_ssc_lock);
6307 }
6308 #endif
6309
6310 /* Check if any lock belonging to this lockowner has any blockers */
6311 static bool
6312 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6313 {
6314         struct file_lock_context *ctx;
6315         struct nfs4_ol_stateid *stp;
6316         struct nfs4_file *nf;
6317
6318         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6319                 nf = stp->st_stid.sc_file;
6320                 ctx = locks_inode_context(nf->fi_inode);
6321                 if (!ctx)
6322                         continue;
6323                 if (locks_owner_has_blockers(ctx, lo))
6324                         return true;
6325         }
6326         return false;
6327 }
6328
6329 static bool
6330 nfs4_anylock_blockers(struct nfs4_client *clp)
6331 {
6332         int i;
6333         struct nfs4_stateowner *so;
6334         struct nfs4_lockowner *lo;
6335
6336         if (atomic_read(&clp->cl_delegs_in_recall))
6337                 return true;
6338         spin_lock(&clp->cl_lock);
6339         for (i = 0; i < OWNER_HASH_SIZE; i++) {
6340                 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6341                                 so_strhash) {
6342                         if (so->so_is_open_owner)
6343                                 continue;
6344                         lo = lockowner(so);
6345                         if (nfs4_lockowner_has_blockers(lo)) {
6346                                 spin_unlock(&clp->cl_lock);
6347                                 return true;
6348                         }
6349                 }
6350         }
6351         spin_unlock(&clp->cl_lock);
6352         return false;
6353 }
6354
6355 static void
6356 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6357                                 struct laundry_time *lt)
6358 {
6359         unsigned int maxreap, reapcnt = 0;
6360         struct list_head *pos, *next;
6361         struct nfs4_client *clp;
6362
6363         maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6364                         NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6365         INIT_LIST_HEAD(reaplist);
6366         spin_lock(&nn->client_lock);
6367         list_for_each_safe(pos, next, &nn->client_lru) {
6368                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6369                 if (clp->cl_state == NFSD4_EXPIRABLE)
6370                         goto exp_client;
6371                 if (!state_expired(lt, clp->cl_time))
6372                         break;
6373                 if (!atomic_read(&clp->cl_rpc_users)) {
6374                         if (clp->cl_state == NFSD4_ACTIVE)
6375                                 atomic_inc(&nn->nfsd_courtesy_clients);
6376                         clp->cl_state = NFSD4_COURTESY;
6377                 }
6378                 if (!client_has_state(clp))
6379                         goto exp_client;
6380                 if (!nfs4_anylock_blockers(clp))
6381                         if (reapcnt >= maxreap)
6382                                 continue;
6383 exp_client:
6384                 if (!mark_client_expired_locked(clp)) {
6385                         list_add(&clp->cl_lru, reaplist);
6386                         reapcnt++;
6387                 }
6388         }
6389         spin_unlock(&nn->client_lock);
6390 }
6391
6392 static void
6393 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6394                                 struct list_head *reaplist)
6395 {
6396         unsigned int maxreap = 0, reapcnt = 0;
6397         struct list_head *pos, *next;
6398         struct nfs4_client *clp;
6399
6400         maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6401         INIT_LIST_HEAD(reaplist);
6402
6403         spin_lock(&nn->client_lock);
6404         list_for_each_safe(pos, next, &nn->client_lru) {
6405                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6406                 if (clp->cl_state == NFSD4_ACTIVE)
6407                         break;
6408                 if (reapcnt >= maxreap)
6409                         break;
6410                 if (!mark_client_expired_locked(clp)) {
6411                         list_add(&clp->cl_lru, reaplist);
6412                         reapcnt++;
6413                 }
6414         }
6415         spin_unlock(&nn->client_lock);
6416 }
6417
6418 static void
6419 nfs4_process_client_reaplist(struct list_head *reaplist)
6420 {
6421         struct list_head *pos, *next;
6422         struct nfs4_client *clp;
6423
6424         list_for_each_safe(pos, next, reaplist) {
6425                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6426                 trace_nfsd_clid_purged(&clp->cl_clientid);
6427                 list_del_init(&clp->cl_lru);
6428                 expire_client(clp);
6429         }
6430 }
6431
6432 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6433                                       struct laundry_time *lt)
6434 {
6435         struct nfs4_client *clp;
6436
6437         spin_lock(&nn->client_lock);
6438         if (nn->nfs40_last_revoke == 0 ||
6439             nn->nfs40_last_revoke > lt->cutoff) {
6440                 spin_unlock(&nn->client_lock);
6441                 return;
6442         }
6443         nn->nfs40_last_revoke = 0;
6444
6445 retry:
6446         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6447                 unsigned long id, tmp;
6448                 struct nfs4_stid *stid;
6449
6450                 if (atomic_read(&clp->cl_admin_revoked) == 0)
6451                         continue;
6452
6453                 spin_lock(&clp->cl_lock);
6454                 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6455                         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6456                                 refcount_inc(&stid->sc_count);
6457                                 spin_unlock(&nn->client_lock);
6458                                 /* this function drops ->cl_lock */
6459                                 nfsd4_drop_revoked_stid(stid);
6460                                 nfs4_put_stid(stid);
6461                                 spin_lock(&nn->client_lock);
6462                                 goto retry;
6463                         }
6464                 spin_unlock(&clp->cl_lock);
6465         }
6466         spin_unlock(&nn->client_lock);
6467 }
6468
6469 static time64_t
6470 nfs4_laundromat(struct nfsd_net *nn)
6471 {
6472         struct nfs4_openowner *oo;
6473         struct nfs4_delegation *dp;
6474         struct nfs4_ol_stateid *stp;
6475         struct nfsd4_blocked_lock *nbl;
6476         struct list_head *pos, *next, reaplist;
6477         struct laundry_time lt = {
6478                 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6479                 .new_timeo = nn->nfsd4_lease
6480         };
6481         struct nfs4_cpntf_state *cps;
6482         copy_stateid_t *cps_t;
6483         int i;
6484
6485         if (clients_still_reclaiming(nn)) {
6486                 lt.new_timeo = 0;
6487                 goto out;
6488         }
6489         nfsd4_end_grace(nn);
6490
6491         spin_lock(&nn->s2s_cp_lock);
6492         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6493                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6494                 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6495                                 state_expired(&lt, cps->cpntf_time))
6496                         _free_cpntf_state_locked(nn, cps);
6497         }
6498         spin_unlock(&nn->s2s_cp_lock);
6499         nfs4_get_client_reaplist(nn, &reaplist, &lt);
6500         nfs4_process_client_reaplist(&reaplist);
6501
6502         nfs40_clean_admin_revoked(nn, &lt);
6503
6504         spin_lock(&state_lock);
6505         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6506                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6507                 if (!state_expired(&lt, dp->dl_time))
6508                         break;
6509                 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6510                 list_add(&dp->dl_recall_lru, &reaplist);
6511         }
6512         spin_unlock(&state_lock);
6513         while (!list_empty(&reaplist)) {
6514                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6515                                         dl_recall_lru);
6516                 list_del_init(&dp->dl_recall_lru);
6517                 revoke_delegation(dp);
6518         }
6519
6520         spin_lock(&nn->client_lock);
6521         while (!list_empty(&nn->close_lru)) {
6522                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6523                                         oo_close_lru);
6524                 if (!state_expired(&lt, oo->oo_time))
6525                         break;
6526                 list_del_init(&oo->oo_close_lru);
6527                 stp = oo->oo_last_closed_stid;
6528                 oo->oo_last_closed_stid = NULL;
6529                 spin_unlock(&nn->client_lock);
6530                 nfs4_put_stid(&stp->st_stid);
6531                 spin_lock(&nn->client_lock);
6532         }
6533         spin_unlock(&nn->client_lock);
6534
6535         /*
6536          * It's possible for a client to try and acquire an already held lock
6537          * that is being held for a long time, and then lose interest in it.
6538          * So, we clean out any un-revisited request after a lease period
6539          * under the assumption that the client is no longer interested.
6540          *
6541          * RFC5661, sec. 9.6 states that the client must not rely on getting
6542          * notifications and must continue to poll for locks, even when the
6543          * server supports them. Thus this shouldn't lead to clients blocking
6544          * indefinitely once the lock does become free.
6545          */
6546         BUG_ON(!list_empty(&reaplist));
6547         spin_lock(&nn->blocked_locks_lock);
6548         while (!list_empty(&nn->blocked_locks_lru)) {
6549                 nbl = list_first_entry(&nn->blocked_locks_lru,
6550                                         struct nfsd4_blocked_lock, nbl_lru);
6551                 if (!state_expired(&lt, nbl->nbl_time))
6552                         break;
6553                 list_move(&nbl->nbl_lru, &reaplist);
6554                 list_del_init(&nbl->nbl_list);
6555         }
6556         spin_unlock(&nn->blocked_locks_lock);
6557
6558         while (!list_empty(&reaplist)) {
6559                 nbl = list_first_entry(&reaplist,
6560                                         struct nfsd4_blocked_lock, nbl_lru);
6561                 list_del_init(&nbl->nbl_lru);
6562                 free_blocked_lock(nbl);
6563         }
6564 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6565         /* service the server-to-server copy delayed unmount list */
6566         nfsd4_ssc_expire_umount(nn);
6567 #endif
6568         if (atomic_long_read(&num_delegations) >= max_delegations)
6569                 deleg_reaper(nn);
6570 out:
6571         return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6572 }
6573
6574 static void laundromat_main(struct work_struct *);
6575
6576 static void
6577 laundromat_main(struct work_struct *laundry)
6578 {
6579         time64_t t;
6580         struct delayed_work *dwork = to_delayed_work(laundry);
6581         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6582                                            laundromat_work);
6583
6584         t = nfs4_laundromat(nn);
6585         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6586 }
6587
6588 static void
6589 courtesy_client_reaper(struct nfsd_net *nn)
6590 {
6591         struct list_head reaplist;
6592
6593         nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6594         nfs4_process_client_reaplist(&reaplist);
6595 }
6596
6597 static void
6598 deleg_reaper(struct nfsd_net *nn)
6599 {
6600         struct list_head *pos, *next;
6601         struct nfs4_client *clp;
6602         struct list_head cblist;
6603
6604         INIT_LIST_HEAD(&cblist);
6605         spin_lock(&nn->client_lock);
6606         list_for_each_safe(pos, next, &nn->client_lru) {
6607                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6608                 if (clp->cl_state != NFSD4_ACTIVE ||
6609                         list_empty(&clp->cl_delegations) ||
6610                         atomic_read(&clp->cl_delegs_in_recall) ||
6611                         test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6612                         (ktime_get_boottime_seconds() -
6613                                 clp->cl_ra_time < 5)) {
6614                         continue;
6615                 }
6616                 list_add(&clp->cl_ra_cblist, &cblist);
6617
6618                 /* release in nfsd4_cb_recall_any_release */
6619                 atomic_inc(&clp->cl_rpc_users);
6620                 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6621                 clp->cl_ra_time = ktime_get_boottime_seconds();
6622         }
6623         spin_unlock(&nn->client_lock);
6624
6625         while (!list_empty(&cblist)) {
6626                 clp = list_first_entry(&cblist, struct nfs4_client,
6627                                         cl_ra_cblist);
6628                 list_del_init(&clp->cl_ra_cblist);
6629                 clp->cl_ra->ra_keep = 0;
6630                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6631                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6632                                                 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6633                 trace_nfsd_cb_recall_any(clp->cl_ra);
6634                 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6635         }
6636 }
6637
6638 static void
6639 nfsd4_state_shrinker_worker(struct work_struct *work)
6640 {
6641         struct nfsd_net *nn = container_of(work, struct nfsd_net,
6642                                 nfsd_shrinker_work);
6643
6644         courtesy_client_reaper(nn);
6645         deleg_reaper(nn);
6646 }
6647
6648 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6649 {
6650         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6651                 return nfserr_bad_stateid;
6652         return nfs_ok;
6653 }
6654
6655 static
6656 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6657 {
6658         __be32 status = nfserr_openmode;
6659
6660         /* For lock stateid's, we test the parent open, not the lock: */
6661         if (stp->st_openstp)
6662                 stp = stp->st_openstp;
6663         if ((flags & WR_STATE) && !access_permit_write(stp))
6664                 goto out;
6665         if ((flags & RD_STATE) && !access_permit_read(stp))
6666                 goto out;
6667         status = nfs_ok;
6668 out:
6669         return status;
6670 }
6671
6672 static inline __be32
6673 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6674 {
6675         if (ONE_STATEID(stateid) && (flags & RD_STATE))
6676                 return nfs_ok;
6677         else if (opens_in_grace(net)) {
6678                 /* Answer in remaining cases depends on existence of
6679                  * conflicting state; so we must wait out the grace period. */
6680                 return nfserr_grace;
6681         } else if (flags & WR_STATE)
6682                 return nfs4_share_conflict(current_fh,
6683                                 NFS4_SHARE_DENY_WRITE);
6684         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6685                 return nfs4_share_conflict(current_fh,
6686                                 NFS4_SHARE_DENY_READ);
6687 }
6688
6689 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6690 {
6691         /*
6692          * When sessions are used the stateid generation number is ignored
6693          * when it is zero.
6694          */
6695         if (has_session && in->si_generation == 0)
6696                 return nfs_ok;
6697
6698         if (in->si_generation == ref->si_generation)
6699                 return nfs_ok;
6700
6701         /* If the client sends us a stateid from the future, it's buggy: */
6702         if (nfsd4_stateid_generation_after(in, ref))
6703                 return nfserr_bad_stateid;
6704         /*
6705          * However, we could see a stateid from the past, even from a
6706          * non-buggy client.  For example, if the client sends a lock
6707          * while some IO is outstanding, the lock may bump si_generation
6708          * while the IO is still in flight.  The client could avoid that
6709          * situation by waiting for responses on all the IO requests,
6710          * but better performance may result in retrying IO that
6711          * receives an old_stateid error if requests are rarely
6712          * reordered in flight:
6713          */
6714         return nfserr_old_stateid;
6715 }
6716
6717 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6718 {
6719         __be32 ret;
6720
6721         spin_lock(&s->sc_lock);
6722         ret = nfsd4_verify_open_stid(s);
6723         if (ret == nfs_ok)
6724                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6725         spin_unlock(&s->sc_lock);
6726         if (ret == nfserr_admin_revoked)
6727                 nfsd40_drop_revoked_stid(s->sc_client,
6728                                         &s->sc_stateid);
6729         return ret;
6730 }
6731
6732 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6733 {
6734         if (ols->st_stateowner->so_is_open_owner &&
6735             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6736                 return nfserr_bad_stateid;
6737         return nfs_ok;
6738 }
6739
6740 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6741 {
6742         struct nfs4_stid *s;
6743         __be32 status = nfserr_bad_stateid;
6744
6745         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6746                 CLOSE_STATEID(stateid))
6747                 return status;
6748         spin_lock(&cl->cl_lock);
6749         s = find_stateid_locked(cl, stateid);
6750         if (!s)
6751                 goto out_unlock;
6752         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6753         if (status)
6754                 goto out_unlock;
6755         status = nfsd4_verify_open_stid(s);
6756         if (status)
6757                 goto out_unlock;
6758
6759         switch (s->sc_type) {
6760         case SC_TYPE_DELEG:
6761                 status = nfs_ok;
6762                 break;
6763         case SC_TYPE_OPEN:
6764         case SC_TYPE_LOCK:
6765                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6766                 break;
6767         default:
6768                 printk("unknown stateid type %x\n", s->sc_type);
6769                 status = nfserr_bad_stateid;
6770         }
6771 out_unlock:
6772         spin_unlock(&cl->cl_lock);
6773         if (status == nfserr_admin_revoked)
6774                 nfsd40_drop_revoked_stid(cl, stateid);
6775         return status;
6776 }
6777
6778 __be32
6779 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6780                      stateid_t *stateid,
6781                      unsigned short typemask, unsigned short statusmask,
6782                      struct nfs4_stid **s, struct nfsd_net *nn)
6783 {
6784         __be32 status;
6785         struct nfs4_stid *stid;
6786         bool return_revoked = false;
6787
6788         /*
6789          *  only return revoked delegations if explicitly asked.
6790          *  otherwise we report revoked or bad_stateid status.
6791          */
6792         if (statusmask & SC_STATUS_REVOKED)
6793                 return_revoked = true;
6794         if (typemask & SC_TYPE_DELEG)
6795                 /* Always allow REVOKED for DELEG so we can
6796                  * retturn the appropriate error.
6797                  */
6798                 statusmask |= SC_STATUS_REVOKED;
6799
6800         statusmask |= SC_STATUS_ADMIN_REVOKED;
6801
6802         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6803                 CLOSE_STATEID(stateid))
6804                 return nfserr_bad_stateid;
6805         status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6806         if (status == nfserr_stale_clientid) {
6807                 if (cstate->session)
6808                         return nfserr_bad_stateid;
6809                 return nfserr_stale_stateid;
6810         }
6811         if (status)
6812                 return status;
6813         stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6814         if (!stid)
6815                 return nfserr_bad_stateid;
6816         if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6817                 nfs4_put_stid(stid);
6818                 return nfserr_deleg_revoked;
6819         }
6820         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6821                 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6822                 nfs4_put_stid(stid);
6823                 return nfserr_admin_revoked;
6824         }
6825         *s = stid;
6826         return nfs_ok;
6827 }
6828
6829 static struct nfsd_file *
6830 nfs4_find_file(struct nfs4_stid *s, int flags)
6831 {
6832         struct nfsd_file *ret = NULL;
6833
6834         if (!s || s->sc_status)
6835                 return NULL;
6836
6837         switch (s->sc_type) {
6838         case SC_TYPE_DELEG:
6839                 spin_lock(&s->sc_file->fi_lock);
6840                 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6841                 spin_unlock(&s->sc_file->fi_lock);
6842                 break;
6843         case SC_TYPE_OPEN:
6844         case SC_TYPE_LOCK:
6845                 if (flags & RD_STATE)
6846                         ret = find_readable_file(s->sc_file);
6847                 else
6848                         ret = find_writeable_file(s->sc_file);
6849         }
6850
6851         return ret;
6852 }
6853
6854 static __be32
6855 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6856 {
6857         __be32 status;
6858
6859         status = nfsd4_check_openowner_confirmed(ols);
6860         if (status)
6861                 return status;
6862         return nfs4_check_openmode(ols, flags);
6863 }
6864
6865 static __be32
6866 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6867                 struct nfsd_file **nfp, int flags)
6868 {
6869         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6870         struct nfsd_file *nf;
6871         __be32 status;
6872
6873         nf = nfs4_find_file(s, flags);
6874         if (nf) {
6875                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6876                                 acc | NFSD_MAY_OWNER_OVERRIDE);
6877                 if (status) {
6878                         nfsd_file_put(nf);
6879                         goto out;
6880                 }
6881         } else {
6882                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6883                 if (status)
6884                         return status;
6885         }
6886         *nfp = nf;
6887 out:
6888         return status;
6889 }
6890 static void
6891 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6892 {
6893         WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6894         if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6895                 return;
6896         list_del(&cps->cp_list);
6897         idr_remove(&nn->s2s_cp_stateids,
6898                    cps->cp_stateid.cs_stid.si_opaque.so_id);
6899         kfree(cps);
6900 }
6901 /*
6902  * A READ from an inter server to server COPY will have a
6903  * copy stateid. Look up the copy notify stateid from the
6904  * idr structure and take a reference on it.
6905  */
6906 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6907                           struct nfs4_client *clp,
6908                           struct nfs4_cpntf_state **cps)
6909 {
6910         copy_stateid_t *cps_t;
6911         struct nfs4_cpntf_state *state = NULL;
6912
6913         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6914                 return nfserr_bad_stateid;
6915         spin_lock(&nn->s2s_cp_lock);
6916         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6917         if (cps_t) {
6918                 state = container_of(cps_t, struct nfs4_cpntf_state,
6919                                      cp_stateid);
6920                 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6921                         state = NULL;
6922                         goto unlock;
6923                 }
6924                 if (!clp)
6925                         refcount_inc(&state->cp_stateid.cs_count);
6926                 else
6927                         _free_cpntf_state_locked(nn, state);
6928         }
6929 unlock:
6930         spin_unlock(&nn->s2s_cp_lock);
6931         if (!state)
6932                 return nfserr_bad_stateid;
6933         if (!clp)
6934                 *cps = state;
6935         return 0;
6936 }
6937
6938 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6939                                struct nfs4_stid **stid)
6940 {
6941         __be32 status;
6942         struct nfs4_cpntf_state *cps = NULL;
6943         struct nfs4_client *found;
6944
6945         status = manage_cpntf_state(nn, st, NULL, &cps);
6946         if (status)
6947                 return status;
6948
6949         cps->cpntf_time = ktime_get_boottime_seconds();
6950
6951         status = nfserr_expired;
6952         found = lookup_clientid(&cps->cp_p_clid, true, nn);
6953         if (!found)
6954                 goto out;
6955
6956         *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6957                                      SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6958                                      0);
6959         if (*stid)
6960                 status = nfs_ok;
6961         else
6962                 status = nfserr_bad_stateid;
6963
6964         put_client_renew(found);
6965 out:
6966         nfs4_put_cpntf_state(nn, cps);
6967         return status;
6968 }
6969
6970 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6971 {
6972         spin_lock(&nn->s2s_cp_lock);
6973         _free_cpntf_state_locked(nn, cps);
6974         spin_unlock(&nn->s2s_cp_lock);
6975 }
6976
6977 /**
6978  * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6979  * @rqstp: incoming request from client
6980  * @cstate: current compound state
6981  * @fhp: filehandle associated with requested stateid
6982  * @stateid: stateid (provided by client)
6983  * @flags: flags describing type of operation to be done
6984  * @nfp: optional nfsd_file return pointer (may be NULL)
6985  * @cstid: optional returned nfs4_stid pointer (may be NULL)
6986  *
6987  * Given info from the client, look up a nfs4_stid for the operation. On
6988  * success, it returns a reference to the nfs4_stid and/or the nfsd_file
6989  * associated with it.
6990  */
6991 __be32
6992 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6993                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6994                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6995                 struct nfs4_stid **cstid)
6996 {
6997         struct net *net = SVC_NET(rqstp);
6998         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6999         struct nfs4_stid *s = NULL;
7000         __be32 status;
7001
7002         if (nfp)
7003                 *nfp = NULL;
7004
7005         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
7006                 if (cstid)
7007                         status = nfserr_bad_stateid;
7008                 else
7009                         status = check_special_stateids(net, fhp, stateid,
7010                                                                         flags);
7011                 goto done;
7012         }
7013
7014         status = nfsd4_lookup_stateid(cstate, stateid,
7015                                 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7016                                 0, &s, nn);
7017         if (status == nfserr_bad_stateid)
7018                 status = find_cpntf_state(nn, stateid, &s);
7019         if (status)
7020                 return status;
7021         status = nfsd4_stid_check_stateid_generation(stateid, s,
7022                         nfsd4_has_session(cstate));
7023         if (status)
7024                 goto out;
7025
7026         switch (s->sc_type) {
7027         case SC_TYPE_DELEG:
7028                 status = nfs4_check_delegmode(delegstateid(s), flags);
7029                 break;
7030         case SC_TYPE_OPEN:
7031         case SC_TYPE_LOCK:
7032                 status = nfs4_check_olstateid(openlockstateid(s), flags);
7033                 break;
7034         }
7035         if (status)
7036                 goto out;
7037         status = nfs4_check_fh(fhp, s);
7038
7039 done:
7040         if (status == nfs_ok && nfp)
7041                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7042 out:
7043         if (s) {
7044                 if (!status && cstid)
7045                         *cstid = s;
7046                 else
7047                         nfs4_put_stid(s);
7048         }
7049         return status;
7050 }
7051
7052 /*
7053  * Test if the stateid is valid
7054  */
7055 __be32
7056 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7057                    union nfsd4_op_u *u)
7058 {
7059         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7060         struct nfsd4_test_stateid_id *stateid;
7061         struct nfs4_client *cl = cstate->clp;
7062
7063         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7064                 stateid->ts_id_status =
7065                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7066
7067         return nfs_ok;
7068 }
7069
7070 static __be32
7071 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7072 {
7073         struct nfs4_ol_stateid *stp = openlockstateid(s);
7074         __be32 ret;
7075
7076         ret = nfsd4_lock_ol_stateid(stp);
7077         if (ret)
7078                 goto out_put_stid;
7079
7080         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7081         if (ret)
7082                 goto out;
7083
7084         ret = nfserr_locks_held;
7085         if (check_for_locks(stp->st_stid.sc_file,
7086                             lockowner(stp->st_stateowner)))
7087                 goto out;
7088
7089         release_lock_stateid(stp);
7090         ret = nfs_ok;
7091
7092 out:
7093         mutex_unlock(&stp->st_mutex);
7094 out_put_stid:
7095         nfs4_put_stid(s);
7096         return ret;
7097 }
7098
7099 __be32
7100 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7101                    union nfsd4_op_u *u)
7102 {
7103         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7104         stateid_t *stateid = &free_stateid->fr_stateid;
7105         struct nfs4_stid *s;
7106         struct nfs4_delegation *dp;
7107         struct nfs4_client *cl = cstate->clp;
7108         __be32 ret = nfserr_bad_stateid;
7109
7110         spin_lock(&cl->cl_lock);
7111         s = find_stateid_locked(cl, stateid);
7112         if (!s || s->sc_status & SC_STATUS_CLOSED)
7113                 goto out_unlock;
7114         if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7115                 nfsd4_drop_revoked_stid(s);
7116                 ret = nfs_ok;
7117                 goto out;
7118         }
7119         spin_lock(&s->sc_lock);
7120         switch (s->sc_type) {
7121         case SC_TYPE_DELEG:
7122                 if (s->sc_status & SC_STATUS_REVOKED) {
7123                         spin_unlock(&s->sc_lock);
7124                         dp = delegstateid(s);
7125                         list_del_init(&dp->dl_recall_lru);
7126                         spin_unlock(&cl->cl_lock);
7127                         nfs4_put_stid(s);
7128                         ret = nfs_ok;
7129                         goto out;
7130                 }
7131                 ret = nfserr_locks_held;
7132                 break;
7133         case SC_TYPE_OPEN:
7134                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7135                 if (ret)
7136                         break;
7137                 ret = nfserr_locks_held;
7138                 break;
7139         case SC_TYPE_LOCK:
7140                 spin_unlock(&s->sc_lock);
7141                 refcount_inc(&s->sc_count);
7142                 spin_unlock(&cl->cl_lock);
7143                 ret = nfsd4_free_lock_stateid(stateid, s);
7144                 goto out;
7145         }
7146         spin_unlock(&s->sc_lock);
7147 out_unlock:
7148         spin_unlock(&cl->cl_lock);
7149 out:
7150         return ret;
7151 }
7152
7153 static inline int
7154 setlkflg (int type)
7155 {
7156         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7157                 RD_STATE : WR_STATE;
7158 }
7159
7160 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7161 {
7162         struct svc_fh *current_fh = &cstate->current_fh;
7163         struct nfs4_stateowner *sop = stp->st_stateowner;
7164         __be32 status;
7165
7166         status = nfsd4_check_seqid(cstate, sop, seqid);
7167         if (status)
7168                 return status;
7169         status = nfsd4_lock_ol_stateid(stp);
7170         if (status != nfs_ok)
7171                 return status;
7172         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7173         if (status == nfs_ok)
7174                 status = nfs4_check_fh(current_fh, &stp->st_stid);
7175         if (status != nfs_ok)
7176                 mutex_unlock(&stp->st_mutex);
7177         return status;
7178 }
7179
7180 /**
7181  * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7182  * @cstate: compund state
7183  * @seqid: seqid (provided by client)
7184  * @stateid: stateid (provided by client)
7185  * @typemask: mask of allowable types for this operation
7186  * @statusmask: mask of allowed states: 0 or STID_CLOSED
7187  * @stpp: return pointer for the stateid found
7188  * @nn: net namespace for request
7189  *
7190  * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7191  * return it in @stpp. On a nfs_ok return, the returned stateid will
7192  * have its st_mutex locked.
7193  */
7194 static __be32
7195 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7196                          stateid_t *stateid,
7197                          unsigned short typemask, unsigned short statusmask,
7198                          struct nfs4_ol_stateid **stpp,
7199                          struct nfsd_net *nn)
7200 {
7201         __be32 status;
7202         struct nfs4_stid *s;
7203         struct nfs4_ol_stateid *stp = NULL;
7204
7205         trace_nfsd_preprocess(seqid, stateid);
7206
7207         *stpp = NULL;
7208         status = nfsd4_lookup_stateid(cstate, stateid,
7209                                       typemask, statusmask, &s, nn);
7210         if (status)
7211                 return status;
7212         stp = openlockstateid(s);
7213         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
7214
7215         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7216         if (!status)
7217                 *stpp = stp;
7218         else
7219                 nfs4_put_stid(&stp->st_stid);
7220         return status;
7221 }
7222
7223 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7224                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7225 {
7226         __be32 status;
7227         struct nfs4_openowner *oo;
7228         struct nfs4_ol_stateid *stp;
7229
7230         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7231                                           SC_TYPE_OPEN, 0, &stp, nn);
7232         if (status)
7233                 return status;
7234         oo = openowner(stp->st_stateowner);
7235         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7236                 mutex_unlock(&stp->st_mutex);
7237                 nfs4_put_stid(&stp->st_stid);
7238                 return nfserr_bad_stateid;
7239         }
7240         *stpp = stp;
7241         return nfs_ok;
7242 }
7243
7244 __be32
7245 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7246                    union nfsd4_op_u *u)
7247 {
7248         struct nfsd4_open_confirm *oc = &u->open_confirm;
7249         __be32 status;
7250         struct nfs4_openowner *oo;
7251         struct nfs4_ol_stateid *stp;
7252         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7253
7254         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7255                         cstate->current_fh.fh_dentry);
7256
7257         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7258         if (status)
7259                 return status;
7260
7261         status = nfs4_preprocess_seqid_op(cstate,
7262                                           oc->oc_seqid, &oc->oc_req_stateid,
7263                                           SC_TYPE_OPEN, 0, &stp, nn);
7264         if (status)
7265                 goto out;
7266         oo = openowner(stp->st_stateowner);
7267         status = nfserr_bad_stateid;
7268         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7269                 mutex_unlock(&stp->st_mutex);
7270                 goto put_stateid;
7271         }
7272         oo->oo_flags |= NFS4_OO_CONFIRMED;
7273         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7274         mutex_unlock(&stp->st_mutex);
7275         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7276         nfsd4_client_record_create(oo->oo_owner.so_client);
7277         status = nfs_ok;
7278 put_stateid:
7279         nfs4_put_stid(&stp->st_stid);
7280 out:
7281         nfsd4_bump_seqid(cstate, status);
7282         return status;
7283 }
7284
7285 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7286 {
7287         if (!test_access(access, stp))
7288                 return;
7289         nfs4_file_put_access(stp->st_stid.sc_file, access);
7290         clear_access(access, stp);
7291 }
7292
7293 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7294 {
7295         switch (to_access) {
7296         case NFS4_SHARE_ACCESS_READ:
7297                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7298                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7299                 break;
7300         case NFS4_SHARE_ACCESS_WRITE:
7301                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7302                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7303                 break;
7304         case NFS4_SHARE_ACCESS_BOTH:
7305                 break;
7306         default:
7307                 WARN_ON_ONCE(1);
7308         }
7309 }
7310
7311 __be32
7312 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7313                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7314 {
7315         struct nfsd4_open_downgrade *od = &u->open_downgrade;
7316         __be32 status;
7317         struct nfs4_ol_stateid *stp;
7318         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7319
7320         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7321                         cstate->current_fh.fh_dentry);
7322
7323         /* We don't yet support WANT bits: */
7324         if (od->od_deleg_want)
7325                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7326                         od->od_deleg_want);
7327
7328         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7329                                         &od->od_stateid, &stp, nn);
7330         if (status)
7331                 goto out;
7332         status = nfserr_inval;
7333         if (!test_access(od->od_share_access, stp)) {
7334                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7335                         stp->st_access_bmap, od->od_share_access);
7336                 goto put_stateid;
7337         }
7338         if (!test_deny(od->od_share_deny, stp)) {
7339                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7340                         stp->st_deny_bmap, od->od_share_deny);
7341                 goto put_stateid;
7342         }
7343         nfs4_stateid_downgrade(stp, od->od_share_access);
7344         reset_union_bmap_deny(od->od_share_deny, stp);
7345         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7346         status = nfs_ok;
7347 put_stateid:
7348         mutex_unlock(&stp->st_mutex);
7349         nfs4_put_stid(&stp->st_stid);
7350 out:
7351         nfsd4_bump_seqid(cstate, status);
7352         return status;
7353 }
7354
7355 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7356 {
7357         struct nfs4_client *clp = s->st_stid.sc_client;
7358         bool unhashed;
7359         LIST_HEAD(reaplist);
7360         struct nfs4_ol_stateid *stp;
7361
7362         spin_lock(&clp->cl_lock);
7363         unhashed = unhash_open_stateid(s, &reaplist);
7364
7365         if (clp->cl_minorversion) {
7366                 if (unhashed)
7367                         put_ol_stateid_locked(s, &reaplist);
7368                 spin_unlock(&clp->cl_lock);
7369                 list_for_each_entry(stp, &reaplist, st_locks)
7370                         nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7371                 free_ol_stateid_reaplist(&reaplist);
7372         } else {
7373                 spin_unlock(&clp->cl_lock);
7374                 free_ol_stateid_reaplist(&reaplist);
7375                 if (unhashed)
7376                         move_to_close_lru(s, clp->net);
7377         }
7378 }
7379
7380 /*
7381  * nfs4_unlock_state() called after encode
7382  */
7383 __be32
7384 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7385                 union nfsd4_op_u *u)
7386 {
7387         struct nfsd4_close *close = &u->close;
7388         __be32 status;
7389         struct nfs4_ol_stateid *stp;
7390         struct net *net = SVC_NET(rqstp);
7391         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7392
7393         dprintk("NFSD: nfsd4_close on file %pd\n",
7394                         cstate->current_fh.fh_dentry);
7395
7396         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7397                                           &close->cl_stateid,
7398                                           SC_TYPE_OPEN, SC_STATUS_CLOSED,
7399                                           &stp, nn);
7400         nfsd4_bump_seqid(cstate, status);
7401         if (status)
7402                 goto out;
7403
7404         spin_lock(&stp->st_stid.sc_client->cl_lock);
7405         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7406         spin_unlock(&stp->st_stid.sc_client->cl_lock);
7407
7408         /*
7409          * Technically we don't _really_ have to increment or copy it, since
7410          * it should just be gone after this operation and we clobber the
7411          * copied value below, but we continue to do so here just to ensure
7412          * that racing ops see that there was a state change.
7413          */
7414         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7415
7416         nfsd4_close_open_stateid(stp);
7417         mutex_unlock(&stp->st_mutex);
7418
7419         /* v4.1+ suggests that we send a special stateid in here, since the
7420          * clients should just ignore this anyway. Since this is not useful
7421          * for v4.0 clients either, we set it to the special close_stateid
7422          * universally.
7423          *
7424          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7425          */
7426         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7427
7428         /* put reference from nfs4_preprocess_seqid_op */
7429         nfs4_put_stid(&stp->st_stid);
7430 out:
7431         return status;
7432 }
7433
7434 __be32
7435 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7436                   union nfsd4_op_u *u)
7437 {
7438         struct nfsd4_delegreturn *dr = &u->delegreturn;
7439         struct nfs4_delegation *dp;
7440         stateid_t *stateid = &dr->dr_stateid;
7441         struct nfs4_stid *s;
7442         __be32 status;
7443         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7444
7445         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7446                 return status;
7447
7448         status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7449         if (status)
7450                 goto out;
7451         dp = delegstateid(s);
7452         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7453         if (status)
7454                 goto put_stateid;
7455
7456         trace_nfsd_deleg_return(stateid);
7457         wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7458         destroy_delegation(dp);
7459 put_stateid:
7460         nfs4_put_stid(&dp->dl_stid);
7461 out:
7462         return status;
7463 }
7464
7465 /* last octet in a range */
7466 static inline u64
7467 last_byte_offset(u64 start, u64 len)
7468 {
7469         u64 end;
7470
7471         WARN_ON_ONCE(!len);
7472         end = start + len;
7473         return end > start ? end - 1: NFS4_MAX_UINT64;
7474 }
7475
7476 /*
7477  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7478  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7479  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
7480  * locking, this prevents us from being completely protocol-compliant.  The
7481  * real solution to this problem is to start using unsigned file offsets in
7482  * the VFS, but this is a very deep change!
7483  */
7484 static inline void
7485 nfs4_transform_lock_offset(struct file_lock *lock)
7486 {
7487         if (lock->fl_start < 0)
7488                 lock->fl_start = OFFSET_MAX;
7489         if (lock->fl_end < 0)
7490                 lock->fl_end = OFFSET_MAX;
7491 }
7492
7493 static fl_owner_t
7494 nfsd4_lm_get_owner(fl_owner_t owner)
7495 {
7496         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7497
7498         nfs4_get_stateowner(&lo->lo_owner);
7499         return owner;
7500 }
7501
7502 static void
7503 nfsd4_lm_put_owner(fl_owner_t owner)
7504 {
7505         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7506
7507         if (lo)
7508                 nfs4_put_stateowner(&lo->lo_owner);
7509 }
7510
7511 /* return pointer to struct nfs4_client if client is expirable */
7512 static bool
7513 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7514 {
7515         struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7516         struct nfs4_client *clp = lo->lo_owner.so_client;
7517         struct nfsd_net *nn;
7518
7519         if (try_to_expire_client(clp)) {
7520                 nn = net_generic(clp->net, nfsd_net_id);
7521                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7522                 return true;
7523         }
7524         return false;
7525 }
7526
7527 /* schedule laundromat to run immediately and wait for it to complete */
7528 static void
7529 nfsd4_lm_expire_lock(void)
7530 {
7531         flush_workqueue(laundry_wq);
7532 }
7533
7534 static void
7535 nfsd4_lm_notify(struct file_lock *fl)
7536 {
7537         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7538         struct net                      *net = lo->lo_owner.so_client->net;
7539         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
7540         struct nfsd4_blocked_lock       *nbl = container_of(fl,
7541                                                 struct nfsd4_blocked_lock, nbl_lock);
7542         bool queue = false;
7543
7544         /* An empty list means that something else is going to be using it */
7545         spin_lock(&nn->blocked_locks_lock);
7546         if (!list_empty(&nbl->nbl_list)) {
7547                 list_del_init(&nbl->nbl_list);
7548                 list_del_init(&nbl->nbl_lru);
7549                 queue = true;
7550         }
7551         spin_unlock(&nn->blocked_locks_lock);
7552
7553         if (queue) {
7554                 trace_nfsd_cb_notify_lock(lo, nbl);
7555                 nfsd4_run_cb(&nbl->nbl_cb);
7556         }
7557 }
7558
7559 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
7560         .lm_mod_owner = THIS_MODULE,
7561         .lm_notify = nfsd4_lm_notify,
7562         .lm_get_owner = nfsd4_lm_get_owner,
7563         .lm_put_owner = nfsd4_lm_put_owner,
7564         .lm_lock_expirable = nfsd4_lm_lock_expirable,
7565         .lm_expire_lock = nfsd4_lm_expire_lock,
7566 };
7567
7568 static inline void
7569 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7570 {
7571         struct nfs4_lockowner *lo;
7572
7573         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7574                 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7575                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7576                                                 GFP_KERNEL);
7577                 if (!deny->ld_owner.data)
7578                         /* We just don't care that much */
7579                         goto nevermind;
7580                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7581         } else {
7582 nevermind:
7583                 deny->ld_owner.len = 0;
7584                 deny->ld_owner.data = NULL;
7585                 deny->ld_clientid.cl_boot = 0;
7586                 deny->ld_clientid.cl_id = 0;
7587         }
7588         deny->ld_start = fl->fl_start;
7589         deny->ld_length = NFS4_MAX_UINT64;
7590         if (fl->fl_end != NFS4_MAX_UINT64)
7591                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7592         deny->ld_type = NFS4_READ_LT;
7593         if (fl->c.flc_type != F_RDLCK)
7594                 deny->ld_type = NFS4_WRITE_LT;
7595 }
7596
7597 static struct nfs4_lockowner *
7598 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7599 {
7600         unsigned int strhashval = ownerstr_hashval(owner);
7601         struct nfs4_stateowner *so;
7602
7603         lockdep_assert_held(&clp->cl_lock);
7604
7605         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7606                             so_strhash) {
7607                 if (so->so_is_open_owner)
7608                         continue;
7609                 if (same_owner_str(so, owner))
7610                         return lockowner(nfs4_get_stateowner(so));
7611         }
7612         return NULL;
7613 }
7614
7615 static struct nfs4_lockowner *
7616 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7617 {
7618         struct nfs4_lockowner *lo;
7619
7620         spin_lock(&clp->cl_lock);
7621         lo = find_lockowner_str_locked(clp, owner);
7622         spin_unlock(&clp->cl_lock);
7623         return lo;
7624 }
7625
7626 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7627 {
7628         unhash_lockowner_locked(lockowner(sop));
7629 }
7630
7631 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7632 {
7633         struct nfs4_lockowner *lo = lockowner(sop);
7634
7635         kmem_cache_free(lockowner_slab, lo);
7636 }
7637
7638 static const struct nfs4_stateowner_operations lockowner_ops = {
7639         .so_unhash =    nfs4_unhash_lockowner,
7640         .so_free =      nfs4_free_lockowner,
7641 };
7642
7643 /*
7644  * Alloc a lock owner structure.
7645  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7646  * occurred.
7647  *
7648  * strhashval = ownerstr_hashval
7649  */
7650 static struct nfs4_lockowner *
7651 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7652                            struct nfs4_ol_stateid *open_stp,
7653                            struct nfsd4_lock *lock)
7654 {
7655         struct nfs4_lockowner *lo, *ret;
7656
7657         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7658         if (!lo)
7659                 return NULL;
7660         INIT_LIST_HEAD(&lo->lo_blocked);
7661         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7662         lo->lo_owner.so_is_open_owner = 0;
7663         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7664         lo->lo_owner.so_ops = &lockowner_ops;
7665         spin_lock(&clp->cl_lock);
7666         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7667         if (ret == NULL) {
7668                 list_add(&lo->lo_owner.so_strhash,
7669                          &clp->cl_ownerstr_hashtbl[strhashval]);
7670                 ret = lo;
7671         } else
7672                 nfs4_free_stateowner(&lo->lo_owner);
7673
7674         spin_unlock(&clp->cl_lock);
7675         return ret;
7676 }
7677
7678 static struct nfs4_ol_stateid *
7679 find_lock_stateid(const struct nfs4_lockowner *lo,
7680                   const struct nfs4_ol_stateid *ost)
7681 {
7682         struct nfs4_ol_stateid *lst;
7683
7684         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7685
7686         /* If ost is not hashed, ost->st_locks will not be valid */
7687         if (!nfs4_ol_stateid_unhashed(ost))
7688                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7689                         if (lst->st_stateowner == &lo->lo_owner) {
7690                                 refcount_inc(&lst->st_stid.sc_count);
7691                                 return lst;
7692                         }
7693                 }
7694         return NULL;
7695 }
7696
7697 static struct nfs4_ol_stateid *
7698 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7699                   struct nfs4_file *fp, struct inode *inode,
7700                   struct nfs4_ol_stateid *open_stp)
7701 {
7702         struct nfs4_client *clp = lo->lo_owner.so_client;
7703         struct nfs4_ol_stateid *retstp;
7704
7705         mutex_init(&stp->st_mutex);
7706         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7707 retry:
7708         spin_lock(&clp->cl_lock);
7709         if (nfs4_ol_stateid_unhashed(open_stp))
7710                 goto out_close;
7711         retstp = find_lock_stateid(lo, open_stp);
7712         if (retstp)
7713                 goto out_found;
7714         refcount_inc(&stp->st_stid.sc_count);
7715         stp->st_stid.sc_type = SC_TYPE_LOCK;
7716         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7717         get_nfs4_file(fp);
7718         stp->st_stid.sc_file = fp;
7719         stp->st_access_bmap = 0;
7720         stp->st_deny_bmap = open_stp->st_deny_bmap;
7721         stp->st_openstp = open_stp;
7722         spin_lock(&fp->fi_lock);
7723         list_add(&stp->st_locks, &open_stp->st_locks);
7724         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7725         list_add(&stp->st_perfile, &fp->fi_stateids);
7726         spin_unlock(&fp->fi_lock);
7727         spin_unlock(&clp->cl_lock);
7728         return stp;
7729 out_found:
7730         spin_unlock(&clp->cl_lock);
7731         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7732                 nfs4_put_stid(&retstp->st_stid);
7733                 goto retry;
7734         }
7735         /* To keep mutex tracking happy */
7736         mutex_unlock(&stp->st_mutex);
7737         return retstp;
7738 out_close:
7739         spin_unlock(&clp->cl_lock);
7740         mutex_unlock(&stp->st_mutex);
7741         return NULL;
7742 }
7743
7744 static struct nfs4_ol_stateid *
7745 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7746                             struct inode *inode, struct nfs4_ol_stateid *ost,
7747                             bool *new)
7748 {
7749         struct nfs4_stid *ns = NULL;
7750         struct nfs4_ol_stateid *lst;
7751         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7752         struct nfs4_client *clp = oo->oo_owner.so_client;
7753
7754         *new = false;
7755         spin_lock(&clp->cl_lock);
7756         lst = find_lock_stateid(lo, ost);
7757         spin_unlock(&clp->cl_lock);
7758         if (lst != NULL) {
7759                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7760                         goto out;
7761                 nfs4_put_stid(&lst->st_stid);
7762         }
7763         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7764         if (ns == NULL)
7765                 return NULL;
7766
7767         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7768         if (lst == openlockstateid(ns))
7769                 *new = true;
7770         else
7771                 nfs4_put_stid(ns);
7772 out:
7773         return lst;
7774 }
7775
7776 static int
7777 check_lock_length(u64 offset, u64 length)
7778 {
7779         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7780                 (length > ~offset)));
7781 }
7782
7783 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7784 {
7785         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7786
7787         lockdep_assert_held(&fp->fi_lock);
7788
7789         if (test_access(access, lock_stp))
7790                 return;
7791         __nfs4_file_get_access(fp, access);
7792         set_access(access, lock_stp);
7793 }
7794
7795 static __be32
7796 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7797                             struct nfs4_ol_stateid *ost,
7798                             struct nfsd4_lock *lock,
7799                             struct nfs4_ol_stateid **plst, bool *new)
7800 {
7801         __be32 status;
7802         struct nfs4_file *fi = ost->st_stid.sc_file;
7803         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7804         struct nfs4_client *cl = oo->oo_owner.so_client;
7805         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7806         struct nfs4_lockowner *lo;
7807         struct nfs4_ol_stateid *lst;
7808         unsigned int strhashval;
7809
7810         lo = find_lockowner_str(cl, &lock->lk_new_owner);
7811         if (!lo) {
7812                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7813                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7814                 if (lo == NULL)
7815                         return nfserr_jukebox;
7816         } else {
7817                 /* with an existing lockowner, seqids must be the same */
7818                 status = nfserr_bad_seqid;
7819                 if (!cstate->minorversion &&
7820                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7821                         goto out;
7822         }
7823
7824         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7825         if (lst == NULL) {
7826                 status = nfserr_jukebox;
7827                 goto out;
7828         }
7829
7830         status = nfs_ok;
7831         *plst = lst;
7832 out:
7833         nfs4_put_stateowner(&lo->lo_owner);
7834         return status;
7835 }
7836
7837 /*
7838  *  LOCK operation
7839  */
7840 __be32
7841 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7842            union nfsd4_op_u *u)
7843 {
7844         struct nfsd4_lock *lock = &u->lock;
7845         struct nfs4_openowner *open_sop = NULL;
7846         struct nfs4_lockowner *lock_sop = NULL;
7847         struct nfs4_ol_stateid *lock_stp = NULL;
7848         struct nfs4_ol_stateid *open_stp = NULL;
7849         struct nfs4_file *fp;
7850         struct nfsd_file *nf = NULL;
7851         struct nfsd4_blocked_lock *nbl = NULL;
7852         struct file_lock *file_lock = NULL;
7853         struct file_lock *conflock = NULL;
7854         struct super_block *sb;
7855         __be32 status = 0;
7856         int lkflg;
7857         int err;
7858         bool new = false;
7859         unsigned char type;
7860         unsigned int flags = FL_POSIX;
7861         struct net *net = SVC_NET(rqstp);
7862         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7863
7864         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7865                 (long long) lock->lk_offset,
7866                 (long long) lock->lk_length);
7867
7868         if (check_lock_length(lock->lk_offset, lock->lk_length))
7869                  return nfserr_inval;
7870
7871         if ((status = fh_verify(rqstp, &cstate->current_fh,
7872                                 S_IFREG, NFSD_MAY_LOCK))) {
7873                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7874                 return status;
7875         }
7876         sb = cstate->current_fh.fh_dentry->d_sb;
7877
7878         if (lock->lk_is_new) {
7879                 if (nfsd4_has_session(cstate))
7880                         /* See rfc 5661 18.10.3: given clientid is ignored: */
7881                         memcpy(&lock->lk_new_clientid,
7882                                 &cstate->clp->cl_clientid,
7883                                 sizeof(clientid_t));
7884
7885                 /* validate and update open stateid and open seqid */
7886                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7887                                         lock->lk_new_open_seqid,
7888                                         &lock->lk_new_open_stateid,
7889                                         &open_stp, nn);
7890                 if (status)
7891                         goto out;
7892                 mutex_unlock(&open_stp->st_mutex);
7893                 open_sop = openowner(open_stp->st_stateowner);
7894                 status = nfserr_bad_stateid;
7895                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7896                                                 &lock->lk_new_clientid))
7897                         goto out;
7898                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7899                                                         &lock_stp, &new);
7900         } else {
7901                 status = nfs4_preprocess_seqid_op(cstate,
7902                                                   lock->lk_old_lock_seqid,
7903                                                   &lock->lk_old_lock_stateid,
7904                                                   SC_TYPE_LOCK, 0, &lock_stp,
7905                                                   nn);
7906         }
7907         if (status)
7908                 goto out;
7909         lock_sop = lockowner(lock_stp->st_stateowner);
7910
7911         lkflg = setlkflg(lock->lk_type);
7912         status = nfs4_check_openmode(lock_stp, lkflg);
7913         if (status)
7914                 goto out;
7915
7916         status = nfserr_grace;
7917         if (locks_in_grace(net) && !lock->lk_reclaim)
7918                 goto out;
7919         status = nfserr_no_grace;
7920         if (!locks_in_grace(net) && lock->lk_reclaim)
7921                 goto out;
7922
7923         if (lock->lk_reclaim)
7924                 flags |= FL_RECLAIM;
7925
7926         fp = lock_stp->st_stid.sc_file;
7927         switch (lock->lk_type) {
7928                 case NFS4_READW_LT:
7929                         if (nfsd4_has_session(cstate) ||
7930                             exportfs_lock_op_is_async(sb->s_export_op))
7931                                 flags |= FL_SLEEP;
7932                         fallthrough;
7933                 case NFS4_READ_LT:
7934                         spin_lock(&fp->fi_lock);
7935                         nf = find_readable_file_locked(fp);
7936                         if (nf)
7937                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7938                         spin_unlock(&fp->fi_lock);
7939                         type = F_RDLCK;
7940                         break;
7941                 case NFS4_WRITEW_LT:
7942                         if (nfsd4_has_session(cstate) ||
7943                             exportfs_lock_op_is_async(sb->s_export_op))
7944                                 flags |= FL_SLEEP;
7945                         fallthrough;
7946                 case NFS4_WRITE_LT:
7947                         spin_lock(&fp->fi_lock);
7948                         nf = find_writeable_file_locked(fp);
7949                         if (nf)
7950                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7951                         spin_unlock(&fp->fi_lock);
7952                         type = F_WRLCK;
7953                         break;
7954                 default:
7955                         status = nfserr_inval;
7956                 goto out;
7957         }
7958
7959         if (!nf) {
7960                 status = nfserr_openmode;
7961                 goto out;
7962         }
7963
7964         /*
7965          * Most filesystems with their own ->lock operations will block
7966          * the nfsd thread waiting to acquire the lock.  That leads to
7967          * deadlocks (we don't want every nfsd thread tied up waiting
7968          * for file locks), so don't attempt blocking lock notifications
7969          * on those filesystems:
7970          */
7971         if (!exportfs_lock_op_is_async(sb->s_export_op))
7972                 flags &= ~FL_SLEEP;
7973
7974         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7975         if (!nbl) {
7976                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7977                 status = nfserr_jukebox;
7978                 goto out;
7979         }
7980
7981         file_lock = &nbl->nbl_lock;
7982         file_lock->c.flc_type = type;
7983         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7984         file_lock->c.flc_pid = current->tgid;
7985         file_lock->c.flc_file = nf->nf_file;
7986         file_lock->c.flc_flags = flags;
7987         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7988         file_lock->fl_start = lock->lk_offset;
7989         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7990         nfs4_transform_lock_offset(file_lock);
7991
7992         conflock = locks_alloc_lock();
7993         if (!conflock) {
7994                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7995                 status = nfserr_jukebox;
7996                 goto out;
7997         }
7998
7999         if (flags & FL_SLEEP) {
8000                 nbl->nbl_time = ktime_get_boottime_seconds();
8001                 spin_lock(&nn->blocked_locks_lock);
8002                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
8003                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
8004                 kref_get(&nbl->nbl_kref);
8005                 spin_unlock(&nn->blocked_locks_lock);
8006         }
8007
8008         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
8009         switch (err) {
8010         case 0: /* success! */
8011                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
8012                 status = 0;
8013                 if (lock->lk_reclaim)
8014                         nn->somebody_reclaimed = true;
8015                 break;
8016         case FILE_LOCK_DEFERRED:
8017                 kref_put(&nbl->nbl_kref, free_nbl);
8018                 nbl = NULL;
8019                 fallthrough;
8020         case -EAGAIN:           /* conflock holds conflicting lock */
8021                 status = nfserr_denied;
8022                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8023                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8024                 break;
8025         case -EDEADLK:
8026                 status = nfserr_deadlock;
8027                 break;
8028         default:
8029                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8030                 status = nfserrno(err);
8031                 break;
8032         }
8033 out:
8034         if (nbl) {
8035                 /* dequeue it if we queued it before */
8036                 if (flags & FL_SLEEP) {
8037                         spin_lock(&nn->blocked_locks_lock);
8038                         if (!list_empty(&nbl->nbl_list) &&
8039                             !list_empty(&nbl->nbl_lru)) {
8040                                 list_del_init(&nbl->nbl_list);
8041                                 list_del_init(&nbl->nbl_lru);
8042                                 kref_put(&nbl->nbl_kref, free_nbl);
8043                         }
8044                         /* nbl can use one of lists to be linked to reaplist */
8045                         spin_unlock(&nn->blocked_locks_lock);
8046                 }
8047                 free_blocked_lock(nbl);
8048         }
8049         if (nf)
8050                 nfsd_file_put(nf);
8051         if (lock_stp) {
8052                 /* Bump seqid manually if the 4.0 replay owner is openowner */
8053                 if (cstate->replay_owner &&
8054                     cstate->replay_owner != &lock_sop->lo_owner &&
8055                     seqid_mutating_err(ntohl(status)))
8056                         lock_sop->lo_owner.so_seqid++;
8057
8058                 /*
8059                  * If this is a new, never-before-used stateid, and we are
8060                  * returning an error, then just go ahead and release it.
8061                  */
8062                 if (status && new)
8063                         release_lock_stateid(lock_stp);
8064
8065                 mutex_unlock(&lock_stp->st_mutex);
8066
8067                 nfs4_put_stid(&lock_stp->st_stid);
8068         }
8069         if (open_stp)
8070                 nfs4_put_stid(&open_stp->st_stid);
8071         nfsd4_bump_seqid(cstate, status);
8072         if (conflock)
8073                 locks_free_lock(conflock);
8074         return status;
8075 }
8076
8077 void nfsd4_lock_release(union nfsd4_op_u *u)
8078 {
8079         struct nfsd4_lock *lock = &u->lock;
8080         struct nfsd4_lock_denied *deny = &lock->lk_denied;
8081
8082         kfree(deny->ld_owner.data);
8083 }
8084
8085 /*
8086  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8087  * so we do a temporary open here just to get an open file to pass to
8088  * vfs_test_lock.
8089  */
8090 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8091 {
8092         struct nfsd_file *nf;
8093         struct inode *inode;
8094         __be32 err;
8095
8096         err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8097         if (err)
8098                 return err;
8099         inode = fhp->fh_dentry->d_inode;
8100         inode_lock(inode); /* to block new leases till after test_lock: */
8101         err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8102         if (err)
8103                 goto out;
8104         lock->c.flc_file = nf->nf_file;
8105         err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8106         lock->c.flc_file = NULL;
8107 out:
8108         inode_unlock(inode);
8109         nfsd_file_put(nf);
8110         return err;
8111 }
8112
8113 /*
8114  * LOCKT operation
8115  */
8116 __be32
8117 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8118             union nfsd4_op_u *u)
8119 {
8120         struct nfsd4_lockt *lockt = &u->lockt;
8121         struct file_lock *file_lock = NULL;
8122         struct nfs4_lockowner *lo = NULL;
8123         __be32 status;
8124         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8125
8126         if (locks_in_grace(SVC_NET(rqstp)))
8127                 return nfserr_grace;
8128
8129         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8130                  return nfserr_inval;
8131
8132         if (!nfsd4_has_session(cstate)) {
8133                 status = set_client(&lockt->lt_clientid, cstate, nn);
8134                 if (status)
8135                         goto out;
8136         }
8137
8138         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8139                 goto out;
8140
8141         file_lock = locks_alloc_lock();
8142         if (!file_lock) {
8143                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8144                 status = nfserr_jukebox;
8145                 goto out;
8146         }
8147
8148         switch (lockt->lt_type) {
8149                 case NFS4_READ_LT:
8150                 case NFS4_READW_LT:
8151                         file_lock->c.flc_type = F_RDLCK;
8152                         break;
8153                 case NFS4_WRITE_LT:
8154                 case NFS4_WRITEW_LT:
8155                         file_lock->c.flc_type = F_WRLCK;
8156                         break;
8157                 default:
8158                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8159                         status = nfserr_inval;
8160                         goto out;
8161         }
8162
8163         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8164         if (lo)
8165                 file_lock->c.flc_owner = (fl_owner_t)lo;
8166         file_lock->c.flc_pid = current->tgid;
8167         file_lock->c.flc_flags = FL_POSIX;
8168
8169         file_lock->fl_start = lockt->lt_offset;
8170         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8171
8172         nfs4_transform_lock_offset(file_lock);
8173
8174         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8175         if (status)
8176                 goto out;
8177
8178         if (file_lock->c.flc_type != F_UNLCK) {
8179                 status = nfserr_denied;
8180                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8181         }
8182 out:
8183         if (lo)
8184                 nfs4_put_stateowner(&lo->lo_owner);
8185         if (file_lock)
8186                 locks_free_lock(file_lock);
8187         return status;
8188 }
8189
8190 void nfsd4_lockt_release(union nfsd4_op_u *u)
8191 {
8192         struct nfsd4_lockt *lockt = &u->lockt;
8193         struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8194
8195         kfree(deny->ld_owner.data);
8196 }
8197
8198 __be32
8199 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8200             union nfsd4_op_u *u)
8201 {
8202         struct nfsd4_locku *locku = &u->locku;
8203         struct nfs4_ol_stateid *stp;
8204         struct nfsd_file *nf = NULL;
8205         struct file_lock *file_lock = NULL;
8206         __be32 status;
8207         int err;
8208         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8209
8210         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8211                 (long long) locku->lu_offset,
8212                 (long long) locku->lu_length);
8213
8214         if (check_lock_length(locku->lu_offset, locku->lu_length))
8215                  return nfserr_inval;
8216
8217         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8218                                           &locku->lu_stateid, SC_TYPE_LOCK, 0,
8219                                           &stp, nn);
8220         if (status)
8221                 goto out;
8222         nf = find_any_file(stp->st_stid.sc_file);
8223         if (!nf) {
8224                 status = nfserr_lock_range;
8225                 goto put_stateid;
8226         }
8227         file_lock = locks_alloc_lock();
8228         if (!file_lock) {
8229                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8230                 status = nfserr_jukebox;
8231                 goto put_file;
8232         }
8233
8234         file_lock->c.flc_type = F_UNLCK;
8235         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8236         file_lock->c.flc_pid = current->tgid;
8237         file_lock->c.flc_file = nf->nf_file;
8238         file_lock->c.flc_flags = FL_POSIX;
8239         file_lock->fl_lmops = &nfsd_posix_mng_ops;
8240         file_lock->fl_start = locku->lu_offset;
8241
8242         file_lock->fl_end = last_byte_offset(locku->lu_offset,
8243                                                 locku->lu_length);
8244         nfs4_transform_lock_offset(file_lock);
8245
8246         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8247         if (err) {
8248                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8249                 goto out_nfserr;
8250         }
8251         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8252 put_file:
8253         nfsd_file_put(nf);
8254 put_stateid:
8255         mutex_unlock(&stp->st_mutex);
8256         nfs4_put_stid(&stp->st_stid);
8257 out:
8258         nfsd4_bump_seqid(cstate, status);
8259         if (file_lock)
8260                 locks_free_lock(file_lock);
8261         return status;
8262
8263 out_nfserr:
8264         status = nfserrno(err);
8265         goto put_file;
8266 }
8267
8268 /*
8269  * returns
8270  *      true:  locks held by lockowner
8271  *      false: no locks held by lockowner
8272  */
8273 static bool
8274 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8275 {
8276         struct file_lock *fl;
8277         int status = false;
8278         struct nfsd_file *nf;
8279         struct inode *inode;
8280         struct file_lock_context *flctx;
8281
8282         spin_lock(&fp->fi_lock);
8283         nf = find_any_file_locked(fp);
8284         if (!nf) {
8285                 /* Any valid lock stateid should have some sort of access */
8286                 WARN_ON_ONCE(1);
8287                 goto out;
8288         }
8289
8290         inode = file_inode(nf->nf_file);
8291         flctx = locks_inode_context(inode);
8292
8293         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8294                 spin_lock(&flctx->flc_lock);
8295                 for_each_file_lock(fl, &flctx->flc_posix) {
8296                         if (fl->c.flc_owner == (fl_owner_t)lowner) {
8297                                 status = true;
8298                                 break;
8299                         }
8300                 }
8301                 spin_unlock(&flctx->flc_lock);
8302         }
8303 out:
8304         spin_unlock(&fp->fi_lock);
8305         return status;
8306 }
8307
8308 /**
8309  * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8310  * @rqstp: RPC transaction
8311  * @cstate: NFSv4 COMPOUND state
8312  * @u: RELEASE_LOCKOWNER arguments
8313  *
8314  * Check if theree are any locks still held and if not - free the lockowner
8315  * and any lock state that is owned.
8316  *
8317  * Return values:
8318  *   %nfs_ok: lockowner released or not found
8319  *   %nfserr_locks_held: lockowner still in use
8320  *   %nfserr_stale_clientid: clientid no longer active
8321  *   %nfserr_expired: clientid not recognized
8322  */
8323 __be32
8324 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8325                         struct nfsd4_compound_state *cstate,
8326                         union nfsd4_op_u *u)
8327 {
8328         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8329         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8330         clientid_t *clid = &rlockowner->rl_clientid;
8331         struct nfs4_ol_stateid *stp;
8332         struct nfs4_lockowner *lo;
8333         struct nfs4_client *clp;
8334         LIST_HEAD(reaplist);
8335         __be32 status;
8336
8337         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8338                 clid->cl_boot, clid->cl_id);
8339
8340         status = set_client(clid, cstate, nn);
8341         if (status)
8342                 return status;
8343         clp = cstate->clp;
8344
8345         spin_lock(&clp->cl_lock);
8346         lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8347         if (!lo) {
8348                 spin_unlock(&clp->cl_lock);
8349                 return nfs_ok;
8350         }
8351
8352         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8353                 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8354                         spin_unlock(&clp->cl_lock);
8355                         nfs4_put_stateowner(&lo->lo_owner);
8356                         return nfserr_locks_held;
8357                 }
8358         }
8359         unhash_lockowner_locked(lo);
8360         while (!list_empty(&lo->lo_owner.so_stateids)) {
8361                 stp = list_first_entry(&lo->lo_owner.so_stateids,
8362                                        struct nfs4_ol_stateid,
8363                                        st_perstateowner);
8364                 unhash_lock_stateid(stp);
8365                 put_ol_stateid_locked(stp, &reaplist);
8366         }
8367         spin_unlock(&clp->cl_lock);
8368
8369         free_ol_stateid_reaplist(&reaplist);
8370         remove_blocked_locks(lo);
8371         nfs4_put_stateowner(&lo->lo_owner);
8372         return nfs_ok;
8373 }
8374
8375 static inline struct nfs4_client_reclaim *
8376 alloc_reclaim(void)
8377 {
8378         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8379 }
8380
8381 bool
8382 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8383 {
8384         struct nfs4_client_reclaim *crp;
8385
8386         crp = nfsd4_find_reclaim_client(name, nn);
8387         return (crp && crp->cr_clp);
8388 }
8389
8390 /*
8391  * failure => all reset bets are off, nfserr_no_grace...
8392  *
8393  * The caller is responsible for freeing name.data if NULL is returned (it
8394  * will be freed in nfs4_remove_reclaim_record in the normal case).
8395  */
8396 struct nfs4_client_reclaim *
8397 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8398                 struct nfsd_net *nn)
8399 {
8400         unsigned int strhashval;
8401         struct nfs4_client_reclaim *crp;
8402
8403         crp = alloc_reclaim();
8404         if (crp) {
8405                 strhashval = clientstr_hashval(name);
8406                 INIT_LIST_HEAD(&crp->cr_strhash);
8407                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8408                 crp->cr_name.data = name.data;
8409                 crp->cr_name.len = name.len;
8410                 crp->cr_princhash.data = princhash.data;
8411                 crp->cr_princhash.len = princhash.len;
8412                 crp->cr_clp = NULL;
8413                 nn->reclaim_str_hashtbl_size++;
8414         }
8415         return crp;
8416 }
8417
8418 void
8419 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8420 {
8421         list_del(&crp->cr_strhash);
8422         kfree(crp->cr_name.data);
8423         kfree(crp->cr_princhash.data);
8424         kfree(crp);
8425         nn->reclaim_str_hashtbl_size--;
8426 }
8427
8428 void
8429 nfs4_release_reclaim(struct nfsd_net *nn)
8430 {
8431         struct nfs4_client_reclaim *crp = NULL;
8432         int i;
8433
8434         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8435                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8436                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8437                                         struct nfs4_client_reclaim, cr_strhash);
8438                         nfs4_remove_reclaim_record(crp, nn);
8439                 }
8440         }
8441         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8442 }
8443
8444 /*
8445  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8446 struct nfs4_client_reclaim *
8447 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8448 {
8449         unsigned int strhashval;
8450         struct nfs4_client_reclaim *crp = NULL;
8451
8452         strhashval = clientstr_hashval(name);
8453         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8454                 if (compare_blob(&crp->cr_name, &name) == 0) {
8455                         return crp;
8456                 }
8457         }
8458         return NULL;
8459 }
8460
8461 __be32
8462 nfs4_check_open_reclaim(struct nfs4_client *clp)
8463 {
8464         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8465                 return nfserr_no_grace;
8466
8467         if (nfsd4_client_record_check(clp))
8468                 return nfserr_reclaim_bad;
8469
8470         return nfs_ok;
8471 }
8472
8473 /*
8474  * Since the lifetime of a delegation isn't limited to that of an open, a
8475  * client may quite reasonably hang on to a delegation as long as it has
8476  * the inode cached.  This becomes an obvious problem the first time a
8477  * client's inode cache approaches the size of the server's total memory.
8478  *
8479  * For now we avoid this problem by imposing a hard limit on the number
8480  * of delegations, which varies according to the server's memory size.
8481  */
8482 static void
8483 set_max_delegations(void)
8484 {
8485         /*
8486          * Allow at most 4 delegations per megabyte of RAM.  Quick
8487          * estimates suggest that in the worst case (where every delegation
8488          * is for a different inode), a delegation could take about 1.5K,
8489          * giving a worst case usage of about 6% of memory.
8490          */
8491         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8492 }
8493
8494 static int nfs4_state_create_net(struct net *net)
8495 {
8496         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8497         int i;
8498
8499         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8500                                             sizeof(struct list_head),
8501                                             GFP_KERNEL);
8502         if (!nn->conf_id_hashtbl)
8503                 goto err;
8504         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8505                                               sizeof(struct list_head),
8506                                               GFP_KERNEL);
8507         if (!nn->unconf_id_hashtbl)
8508                 goto err_unconf_id;
8509         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8510                                               sizeof(struct list_head),
8511                                               GFP_KERNEL);
8512         if (!nn->sessionid_hashtbl)
8513                 goto err_sessionid;
8514
8515         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8516                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8517                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8518         }
8519         for (i = 0; i < SESSION_HASH_SIZE; i++)
8520                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8521         nn->conf_name_tree = RB_ROOT;
8522         nn->unconf_name_tree = RB_ROOT;
8523         nn->boot_time = ktime_get_real_seconds();
8524         nn->grace_ended = false;
8525         nn->nfsd4_manager.block_opens = true;
8526         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8527         INIT_LIST_HEAD(&nn->client_lru);
8528         INIT_LIST_HEAD(&nn->close_lru);
8529         INIT_LIST_HEAD(&nn->del_recall_lru);
8530         spin_lock_init(&nn->client_lock);
8531         spin_lock_init(&nn->s2s_cp_lock);
8532         idr_init(&nn->s2s_cp_stateids);
8533
8534         spin_lock_init(&nn->blocked_locks_lock);
8535         INIT_LIST_HEAD(&nn->blocked_locks_lru);
8536
8537         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8538         INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8539         get_net(net);
8540
8541         nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8542         if (!nn->nfsd_client_shrinker)
8543                 goto err_shrinker;
8544
8545         nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8546         nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8547         nn->nfsd_client_shrinker->private_data = nn;
8548
8549         shrinker_register(nn->nfsd_client_shrinker);
8550
8551         return 0;
8552
8553 err_shrinker:
8554         put_net(net);
8555         kfree(nn->sessionid_hashtbl);
8556 err_sessionid:
8557         kfree(nn->unconf_id_hashtbl);
8558 err_unconf_id:
8559         kfree(nn->conf_id_hashtbl);
8560 err:
8561         return -ENOMEM;
8562 }
8563
8564 static void
8565 nfs4_state_destroy_net(struct net *net)
8566 {
8567         int i;
8568         struct nfs4_client *clp = NULL;
8569         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8570
8571         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8572                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8573                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8574                         destroy_client(clp);
8575                 }
8576         }
8577
8578         WARN_ON(!list_empty(&nn->blocked_locks_lru));
8579
8580         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8581                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8582                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8583                         destroy_client(clp);
8584                 }
8585         }
8586
8587         kfree(nn->sessionid_hashtbl);
8588         kfree(nn->unconf_id_hashtbl);
8589         kfree(nn->conf_id_hashtbl);
8590         put_net(net);
8591 }
8592
8593 int
8594 nfs4_state_start_net(struct net *net)
8595 {
8596         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8597         int ret;
8598
8599         ret = nfs4_state_create_net(net);
8600         if (ret)
8601                 return ret;
8602         locks_start_grace(net, &nn->nfsd4_manager);
8603         nfsd4_client_tracking_init(net);
8604         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8605                 goto skip_grace;
8606         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8607                nn->nfsd4_grace, net->ns.inum);
8608         trace_nfsd_grace_start(nn);
8609         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8610         return 0;
8611
8612 skip_grace:
8613         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8614                         net->ns.inum);
8615         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8616         nfsd4_end_grace(nn);
8617         return 0;
8618 }
8619
8620 /* initialization to perform when the nfsd service is started: */
8621
8622 int
8623 nfs4_state_start(void)
8624 {
8625         int ret;
8626
8627         ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8628         if (ret)
8629                 return ret;
8630
8631         ret = nfsd4_create_callback_queue();
8632         if (ret) {
8633                 rhltable_destroy(&nfs4_file_rhltable);
8634                 return ret;
8635         }
8636
8637         set_max_delegations();
8638         return 0;
8639 }
8640
8641 void
8642 nfs4_state_shutdown_net(struct net *net)
8643 {
8644         struct nfs4_delegation *dp = NULL;
8645         struct list_head *pos, *next, reaplist;
8646         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8647
8648         shrinker_free(nn->nfsd_client_shrinker);
8649         cancel_work(&nn->nfsd_shrinker_work);
8650         cancel_delayed_work_sync(&nn->laundromat_work);
8651         locks_end_grace(&nn->nfsd4_manager);
8652
8653         INIT_LIST_HEAD(&reaplist);
8654         spin_lock(&state_lock);
8655         list_for_each_safe(pos, next, &nn->del_recall_lru) {
8656                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8657                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8658                 list_add(&dp->dl_recall_lru, &reaplist);
8659         }
8660         spin_unlock(&state_lock);
8661         list_for_each_safe(pos, next, &reaplist) {
8662                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8663                 list_del_init(&dp->dl_recall_lru);
8664                 destroy_unhashed_deleg(dp);
8665         }
8666
8667         nfsd4_client_tracking_exit(net);
8668         nfs4_state_destroy_net(net);
8669 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8670         nfsd4_ssc_shutdown_umount(nn);
8671 #endif
8672 }
8673
8674 void
8675 nfs4_state_shutdown(void)
8676 {
8677         nfsd4_destroy_callback_queue();
8678         rhltable_destroy(&nfs4_file_rhltable);
8679 }
8680
8681 static void
8682 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8683 {
8684         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8685             CURRENT_STATEID(stateid))
8686                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8687 }
8688
8689 static void
8690 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8691 {
8692         if (cstate->minorversion) {
8693                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8694                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8695         }
8696 }
8697
8698 void
8699 clear_current_stateid(struct nfsd4_compound_state *cstate)
8700 {
8701         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8702 }
8703
8704 /*
8705  * functions to set current state id
8706  */
8707 void
8708 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8709                 union nfsd4_op_u *u)
8710 {
8711         put_stateid(cstate, &u->open_downgrade.od_stateid);
8712 }
8713
8714 void
8715 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8716                 union nfsd4_op_u *u)
8717 {
8718         put_stateid(cstate, &u->open.op_stateid);
8719 }
8720
8721 void
8722 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8723                 union nfsd4_op_u *u)
8724 {
8725         put_stateid(cstate, &u->close.cl_stateid);
8726 }
8727
8728 void
8729 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8730                 union nfsd4_op_u *u)
8731 {
8732         put_stateid(cstate, &u->lock.lk_resp_stateid);
8733 }
8734
8735 /*
8736  * functions to consume current state id
8737  */
8738
8739 void
8740 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8741                 union nfsd4_op_u *u)
8742 {
8743         get_stateid(cstate, &u->open_downgrade.od_stateid);
8744 }
8745
8746 void
8747 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8748                 union nfsd4_op_u *u)
8749 {
8750         get_stateid(cstate, &u->delegreturn.dr_stateid);
8751 }
8752
8753 void
8754 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8755                 union nfsd4_op_u *u)
8756 {
8757         get_stateid(cstate, &u->free_stateid.fr_stateid);
8758 }
8759
8760 void
8761 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8762                 union nfsd4_op_u *u)
8763 {
8764         get_stateid(cstate, &u->setattr.sa_stateid);
8765 }
8766
8767 void
8768 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8769                 union nfsd4_op_u *u)
8770 {
8771         get_stateid(cstate, &u->close.cl_stateid);
8772 }
8773
8774 void
8775 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8776                 union nfsd4_op_u *u)
8777 {
8778         get_stateid(cstate, &u->locku.lu_stateid);
8779 }
8780
8781 void
8782 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8783                 union nfsd4_op_u *u)
8784 {
8785         get_stateid(cstate, &u->read.rd_stateid);
8786 }
8787
8788 void
8789 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8790                 union nfsd4_op_u *u)
8791 {
8792         get_stateid(cstate, &u->write.wr_stateid);
8793 }
8794
8795 /**
8796  * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8797  * @rqstp: RPC transaction context
8798  * @inode: file to be checked for a conflict
8799  * @modified: return true if file was modified
8800  * @size: new size of file if modified is true
8801  *
8802  * This function is called when there is a conflict between a write
8803  * delegation and a change/size GETATTR from another client. The server
8804  * must either use the CB_GETATTR to get the current values of the
8805  * attributes from the client that holds the delegation or recall the
8806  * delegation before replying to the GETATTR. See RFC 8881 section
8807  * 18.7.4.
8808  *
8809  * Returns 0 if there is no conflict; otherwise an nfs_stat
8810  * code is returned.
8811  */
8812 __be32
8813 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode,
8814                                 bool *modified, u64 *size)
8815 {
8816         __be32 status;
8817         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8818         struct file_lock_context *ctx;
8819         struct file_lease *fl;
8820         struct nfs4_delegation *dp;
8821         struct iattr attrs;
8822         struct nfs4_cb_fattr *ncf;
8823
8824         *modified = false;
8825         ctx = locks_inode_context(inode);
8826         if (!ctx)
8827                 return 0;
8828         spin_lock(&ctx->flc_lock);
8829         for_each_file_lock(fl, &ctx->flc_lease) {
8830                 unsigned char type = fl->c.flc_type;
8831
8832                 if (fl->c.flc_flags == FL_LAYOUT)
8833                         continue;
8834                 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8835                         /*
8836                          * non-nfs lease, if it's a lease with F_RDLCK then
8837                          * we are done; there isn't any write delegation
8838                          * on this inode
8839                          */
8840                         if (type == F_RDLCK)
8841                                 break;
8842                         goto break_lease;
8843                 }
8844                 if (type == F_WRLCK) {
8845                         dp = fl->c.flc_owner;
8846                         if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8847                                 spin_unlock(&ctx->flc_lock);
8848                                 return 0;
8849                         }
8850 break_lease:
8851                         nfsd_stats_wdeleg_getattr_inc(nn);
8852                         dp = fl->c.flc_owner;
8853                         ncf = &dp->dl_cb_fattr;
8854                         nfs4_cb_getattr(&dp->dl_cb_fattr);
8855                         spin_unlock(&ctx->flc_lock);
8856                         wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8857                                         TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8858                         if (ncf->ncf_cb_status) {
8859                                 /* Recall delegation only if client didn't respond */
8860                                 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8861                                 if (status != nfserr_jukebox ||
8862                                                 !nfsd_wait_for_delegreturn(rqstp, inode))
8863                                         return status;
8864                         }
8865                         if (!ncf->ncf_file_modified &&
8866                                         (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8867                                         ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8868                                 ncf->ncf_file_modified = true;
8869                         if (ncf->ncf_file_modified) {
8870                                 /*
8871                                  * Per section 10.4.3 of RFC 8881, the server would
8872                                  * not update the file's metadata with the client's
8873                                  * modified size
8874                                  */
8875                                 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8876                                 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME;
8877                                 setattr_copy(&nop_mnt_idmap, inode, &attrs);
8878                                 mark_inode_dirty(inode);
8879                                 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8880                                 *size = ncf->ncf_cur_fsize;
8881                                 *modified = true;
8882                         }
8883                         return 0;
8884                 }
8885                 break;
8886         }
8887         spin_unlock(&ctx->flc_lock);
8888         return 0;
8889 }