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
3046         clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
3047         drop_client(clp);
3048 }
3049
3050 static int
3051 nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3052 {
3053         struct nfs4_cb_fattr *ncf =
3054                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3055
3056         ncf->ncf_cb_status = task->tk_status;
3057         switch (task->tk_status) {
3058         case -NFS4ERR_DELAY:
3059                 rpc_delay(task, 2 * HZ);
3060                 return 0;
3061         default:
3062                 return 1;
3063         }
3064 }
3065
3066 static void
3067 nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3068 {
3069         struct nfs4_cb_fattr *ncf =
3070                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3071         struct nfs4_delegation *dp =
3072                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3073
3074         nfs4_put_stid(&dp->dl_stid);
3075         clear_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags);
3076         wake_up_bit(&ncf->ncf_cb_flags, CB_GETATTR_BUSY);
3077 }
3078
3079 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3080         .done           = nfsd4_cb_recall_any_done,
3081         .release        = nfsd4_cb_recall_any_release,
3082 };
3083
3084 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3085         .done           = nfsd4_cb_getattr_done,
3086         .release        = nfsd4_cb_getattr_release,
3087 };
3088
3089 static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3090 {
3091         struct nfs4_delegation *dp =
3092                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3093
3094         if (test_and_set_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags))
3095                 return;
3096         /* set to proper status when nfsd4_cb_getattr_done runs */
3097         ncf->ncf_cb_status = NFS4ERR_IO;
3098
3099         refcount_inc(&dp->dl_stid.sc_count);
3100         nfsd4_run_cb(&ncf->ncf_getattr);
3101 }
3102
3103 static struct nfs4_client *create_client(struct xdr_netobj name,
3104                 struct svc_rqst *rqstp, nfs4_verifier *verf)
3105 {
3106         struct nfs4_client *clp;
3107         struct sockaddr *sa = svc_addr(rqstp);
3108         int ret;
3109         struct net *net = SVC_NET(rqstp);
3110         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3111         struct dentry *dentries[ARRAY_SIZE(client_files)];
3112
3113         clp = alloc_client(name, nn);
3114         if (clp == NULL)
3115                 return NULL;
3116
3117         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3118         if (ret) {
3119                 free_client(clp);
3120                 return NULL;
3121         }
3122         gen_clid(clp, nn);
3123         kref_init(&clp->cl_nfsdfs.cl_ref);
3124         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3125         clp->cl_time = ktime_get_boottime_seconds();
3126         clear_bit(0, &clp->cl_cb_slot_busy);
3127         copy_verf(clp, verf);
3128         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3129         clp->cl_cb_session = NULL;
3130         clp->net = net;
3131         clp->cl_nfsd_dentry = nfsd_client_mkdir(
3132                 nn, &clp->cl_nfsdfs,
3133                 clp->cl_clientid.cl_id - nn->clientid_base,
3134                 client_files, dentries);
3135         clp->cl_nfsd_info_dentry = dentries[0];
3136         if (!clp->cl_nfsd_dentry) {
3137                 free_client(clp);
3138                 return NULL;
3139         }
3140         clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3141         if (!clp->cl_ra) {
3142                 free_client(clp);
3143                 return NULL;
3144         }
3145         clp->cl_ra_time = 0;
3146         nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3147                         NFSPROC4_CLNT_CB_RECALL_ANY);
3148         return clp;
3149 }
3150
3151 static void
3152 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3153 {
3154         struct rb_node **new = &(root->rb_node), *parent = NULL;
3155         struct nfs4_client *clp;
3156
3157         while (*new) {
3158                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3159                 parent = *new;
3160
3161                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3162                         new = &((*new)->rb_left);
3163                 else
3164                         new = &((*new)->rb_right);
3165         }
3166
3167         rb_link_node(&new_clp->cl_namenode, parent, new);
3168         rb_insert_color(&new_clp->cl_namenode, root);
3169 }
3170
3171 static struct nfs4_client *
3172 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3173 {
3174         int cmp;
3175         struct rb_node *node = root->rb_node;
3176         struct nfs4_client *clp;
3177
3178         while (node) {
3179                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
3180                 cmp = compare_blob(&clp->cl_name, name);
3181                 if (cmp > 0)
3182                         node = node->rb_left;
3183                 else if (cmp < 0)
3184                         node = node->rb_right;
3185                 else
3186                         return clp;
3187         }
3188         return NULL;
3189 }
3190
3191 static void
3192 add_to_unconfirmed(struct nfs4_client *clp)
3193 {
3194         unsigned int idhashval;
3195         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3196
3197         lockdep_assert_held(&nn->client_lock);
3198
3199         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3200         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3201         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3202         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3203         renew_client_locked(clp);
3204 }
3205
3206 static void
3207 move_to_confirmed(struct nfs4_client *clp)
3208 {
3209         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3210         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3211
3212         lockdep_assert_held(&nn->client_lock);
3213
3214         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3215         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3216         add_clp_to_name_tree(clp, &nn->conf_name_tree);
3217         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3218         trace_nfsd_clid_confirmed(&clp->cl_clientid);
3219         renew_client_locked(clp);
3220 }
3221
3222 static struct nfs4_client *
3223 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3224 {
3225         struct nfs4_client *clp;
3226         unsigned int idhashval = clientid_hashval(clid->cl_id);
3227
3228         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3229                 if (same_clid(&clp->cl_clientid, clid)) {
3230                         if ((bool)clp->cl_minorversion != sessions)
3231                                 return NULL;
3232                         renew_client_locked(clp);
3233                         return clp;
3234                 }
3235         }
3236         return NULL;
3237 }
3238
3239 static struct nfs4_client *
3240 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3241 {
3242         struct list_head *tbl = nn->conf_id_hashtbl;
3243
3244         lockdep_assert_held(&nn->client_lock);
3245         return find_client_in_id_table(tbl, clid, sessions);
3246 }
3247
3248 static struct nfs4_client *
3249 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3250 {
3251         struct list_head *tbl = nn->unconf_id_hashtbl;
3252
3253         lockdep_assert_held(&nn->client_lock);
3254         return find_client_in_id_table(tbl, clid, sessions);
3255 }
3256
3257 static bool clp_used_exchangeid(struct nfs4_client *clp)
3258 {
3259         return clp->cl_exchange_flags != 0;
3260 }
3261
3262 static struct nfs4_client *
3263 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3264 {
3265         lockdep_assert_held(&nn->client_lock);
3266         return find_clp_in_name_tree(name, &nn->conf_name_tree);
3267 }
3268
3269 static struct nfs4_client *
3270 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3271 {
3272         lockdep_assert_held(&nn->client_lock);
3273         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3274 }
3275
3276 static void
3277 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3278 {
3279         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3280         struct sockaddr *sa = svc_addr(rqstp);
3281         u32 scopeid = rpc_get_scope_id(sa);
3282         unsigned short expected_family;
3283
3284         /* Currently, we only support tcp and tcp6 for the callback channel */
3285         if (se->se_callback_netid_len == 3 &&
3286             !memcmp(se->se_callback_netid_val, "tcp", 3))
3287                 expected_family = AF_INET;
3288         else if (se->se_callback_netid_len == 4 &&
3289                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
3290                 expected_family = AF_INET6;
3291         else
3292                 goto out_err;
3293
3294         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3295                                             se->se_callback_addr_len,
3296                                             (struct sockaddr *)&conn->cb_addr,
3297                                             sizeof(conn->cb_addr));
3298
3299         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3300                 goto out_err;
3301
3302         if (conn->cb_addr.ss_family == AF_INET6)
3303                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3304
3305         conn->cb_prog = se->se_callback_prog;
3306         conn->cb_ident = se->se_callback_ident;
3307         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3308         trace_nfsd_cb_args(clp, conn);
3309         return;
3310 out_err:
3311         conn->cb_addr.ss_family = AF_UNSPEC;
3312         conn->cb_addrlen = 0;
3313         trace_nfsd_cb_nodelegs(clp);
3314         return;
3315 }
3316
3317 /*
3318  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3319  */
3320 static void
3321 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3322 {
3323         struct xdr_buf *buf = resp->xdr->buf;
3324         struct nfsd4_slot *slot = resp->cstate.slot;
3325         unsigned int base;
3326
3327         dprintk("--> %s slot %p\n", __func__, slot);
3328
3329         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3330         slot->sl_opcnt = resp->opcnt;
3331         slot->sl_status = resp->cstate.status;
3332         free_svc_cred(&slot->sl_cred);
3333         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3334
3335         if (!nfsd4_cache_this(resp)) {
3336                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3337                 return;
3338         }
3339         slot->sl_flags |= NFSD4_SLOT_CACHED;
3340
3341         base = resp->cstate.data_offset;
3342         slot->sl_datalen = buf->len - base;
3343         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3344                 WARN(1, "%s: sessions DRC could not cache compound\n",
3345                      __func__);
3346         return;
3347 }
3348
3349 /*
3350  * Encode the replay sequence operation from the slot values.
3351  * If cachethis is FALSE encode the uncached rep error on the next
3352  * operation which sets resp->p and increments resp->opcnt for
3353  * nfs4svc_encode_compoundres.
3354  *
3355  */
3356 static __be32
3357 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3358                           struct nfsd4_compoundres *resp)
3359 {
3360         struct nfsd4_op *op;
3361         struct nfsd4_slot *slot = resp->cstate.slot;
3362
3363         /* Encode the replayed sequence operation */
3364         op = &args->ops[resp->opcnt - 1];
3365         nfsd4_encode_operation(resp, op);
3366
3367         if (slot->sl_flags & NFSD4_SLOT_CACHED)
3368                 return op->status;
3369         if (args->opcnt == 1) {
3370                 /*
3371                  * The original operation wasn't a solo sequence--we
3372                  * always cache those--so this retry must not match the
3373                  * original:
3374                  */
3375                 op->status = nfserr_seq_false_retry;
3376         } else {
3377                 op = &args->ops[resp->opcnt++];
3378                 op->status = nfserr_retry_uncached_rep;
3379                 nfsd4_encode_operation(resp, op);
3380         }
3381         return op->status;
3382 }
3383
3384 /*
3385  * The sequence operation is not cached because we can use the slot and
3386  * session values.
3387  */
3388 static __be32
3389 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3390                          struct nfsd4_sequence *seq)
3391 {
3392         struct nfsd4_slot *slot = resp->cstate.slot;
3393         struct xdr_stream *xdr = resp->xdr;
3394         __be32 *p;
3395         __be32 status;
3396
3397         dprintk("--> %s slot %p\n", __func__, slot);
3398
3399         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3400         if (status)
3401                 return status;
3402
3403         p = xdr_reserve_space(xdr, slot->sl_datalen);
3404         if (!p) {
3405                 WARN_ON_ONCE(1);
3406                 return nfserr_serverfault;
3407         }
3408         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3409         xdr_commit_encode(xdr);
3410
3411         resp->opcnt = slot->sl_opcnt;
3412         return slot->sl_status;
3413 }
3414
3415 /*
3416  * Set the exchange_id flags returned by the server.
3417  */
3418 static void
3419 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3420 {
3421 #ifdef CONFIG_NFSD_PNFS
3422         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3423 #else
3424         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3425 #endif
3426
3427         /* Referrals are supported, Migration is not. */
3428         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3429
3430         /* set the wire flags to return to client. */
3431         clid->flags = new->cl_exchange_flags;
3432 }
3433
3434 static bool client_has_openowners(struct nfs4_client *clp)
3435 {
3436         struct nfs4_openowner *oo;
3437
3438         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3439                 if (!list_empty(&oo->oo_owner.so_stateids))
3440                         return true;
3441         }
3442         return false;
3443 }
3444
3445 static bool client_has_state(struct nfs4_client *clp)
3446 {
3447         return client_has_openowners(clp)
3448 #ifdef CONFIG_NFSD_PNFS
3449                 || !list_empty(&clp->cl_lo_states)
3450 #endif
3451                 || !list_empty(&clp->cl_delegations)
3452                 || !list_empty(&clp->cl_sessions)
3453                 || !list_empty(&clp->async_copies);
3454 }
3455
3456 static __be32 copy_impl_id(struct nfs4_client *clp,
3457                                 struct nfsd4_exchange_id *exid)
3458 {
3459         if (!exid->nii_domain.data)
3460                 return 0;
3461         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3462         if (!clp->cl_nii_domain.data)
3463                 return nfserr_jukebox;
3464         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3465         if (!clp->cl_nii_name.data)
3466                 return nfserr_jukebox;
3467         clp->cl_nii_time = exid->nii_time;
3468         return 0;
3469 }
3470
3471 __be32
3472 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3473                 union nfsd4_op_u *u)
3474 {
3475         struct nfsd4_exchange_id *exid = &u->exchange_id;
3476         struct nfs4_client *conf, *new;
3477         struct nfs4_client *unconf = NULL;
3478         __be32 status;
3479         char                    addr_str[INET6_ADDRSTRLEN];
3480         nfs4_verifier           verf = exid->verifier;
3481         struct sockaddr         *sa = svc_addr(rqstp);
3482         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3483         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3484
3485         rpc_ntop(sa, addr_str, sizeof(addr_str));
3486         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3487                 "ip_addr=%s flags %x, spa_how %u\n",
3488                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3489                 addr_str, exid->flags, exid->spa_how);
3490
3491         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3492                 return nfserr_inval;
3493
3494         new = create_client(exid->clname, rqstp, &verf);
3495         if (new == NULL)
3496                 return nfserr_jukebox;
3497         status = copy_impl_id(new, exid);
3498         if (status)
3499                 goto out_nolock;
3500
3501         switch (exid->spa_how) {
3502         case SP4_MACH_CRED:
3503                 exid->spo_must_enforce[0] = 0;
3504                 exid->spo_must_enforce[1] = (
3505                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3506                         1 << (OP_EXCHANGE_ID - 32) |
3507                         1 << (OP_CREATE_SESSION - 32) |
3508                         1 << (OP_DESTROY_SESSION - 32) |
3509                         1 << (OP_DESTROY_CLIENTID - 32));
3510
3511                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3512                                         1 << (OP_OPEN_DOWNGRADE) |
3513                                         1 << (OP_LOCKU) |
3514                                         1 << (OP_DELEGRETURN));
3515
3516                 exid->spo_must_allow[1] &= (
3517                                         1 << (OP_TEST_STATEID - 32) |
3518                                         1 << (OP_FREE_STATEID - 32));
3519                 if (!svc_rqst_integrity_protected(rqstp)) {
3520                         status = nfserr_inval;
3521                         goto out_nolock;
3522                 }
3523                 /*
3524                  * Sometimes userspace doesn't give us a principal.
3525                  * Which is a bug, really.  Anyway, we can't enforce
3526                  * MACH_CRED in that case, better to give up now:
3527                  */
3528                 if (!new->cl_cred.cr_principal &&
3529                                         !new->cl_cred.cr_raw_principal) {
3530                         status = nfserr_serverfault;
3531                         goto out_nolock;
3532                 }
3533                 new->cl_mach_cred = true;
3534                 break;
3535         case SP4_NONE:
3536                 break;
3537         default:                                /* checked by xdr code */
3538                 WARN_ON_ONCE(1);
3539                 fallthrough;
3540         case SP4_SSV:
3541                 status = nfserr_encr_alg_unsupp;
3542                 goto out_nolock;
3543         }
3544
3545         /* Cases below refer to rfc 5661 section 18.35.4: */
3546         spin_lock(&nn->client_lock);
3547         conf = find_confirmed_client_by_name(&exid->clname, nn);
3548         if (conf) {
3549                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3550                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3551
3552                 if (update) {
3553                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3554                                 status = nfserr_inval;
3555                                 goto out;
3556                         }
3557                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3558                                 status = nfserr_wrong_cred;
3559                                 goto out;
3560                         }
3561                         if (!creds_match) { /* case 9 */
3562                                 status = nfserr_perm;
3563                                 goto out;
3564                         }
3565                         if (!verfs_match) { /* case 8 */
3566                                 status = nfserr_not_same;
3567                                 goto out;
3568                         }
3569                         /* case 6 */
3570                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3571                         trace_nfsd_clid_confirmed_r(conf);
3572                         goto out_copy;
3573                 }
3574                 if (!creds_match) { /* case 3 */
3575                         if (client_has_state(conf)) {
3576                                 status = nfserr_clid_inuse;
3577                                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3578                                 goto out;
3579                         }
3580                         goto out_new;
3581                 }
3582                 if (verfs_match) { /* case 2 */
3583                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3584                         trace_nfsd_clid_confirmed_r(conf);
3585                         goto out_copy;
3586                 }
3587                 /* case 5, client reboot */
3588                 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3589                 conf = NULL;
3590                 goto out_new;
3591         }
3592
3593         if (update) { /* case 7 */
3594                 status = nfserr_noent;
3595                 goto out;
3596         }
3597
3598         unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3599         if (unconf) /* case 4, possible retry or client restart */
3600                 unhash_client_locked(unconf);
3601
3602         /* case 1, new owner ID */
3603         trace_nfsd_clid_fresh(new);
3604
3605 out_new:
3606         if (conf) {
3607                 status = mark_client_expired_locked(conf);
3608                 if (status)
3609                         goto out;
3610                 trace_nfsd_clid_replaced(&conf->cl_clientid);
3611         }
3612         new->cl_minorversion = cstate->minorversion;
3613         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3614         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3615
3616         /* Contrived initial CREATE_SESSION response */
3617         new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3618
3619         add_to_unconfirmed(new);
3620         swap(new, conf);
3621 out_copy:
3622         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3623         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3624
3625         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3626         nfsd4_set_ex_flags(conf, exid);
3627
3628         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3629                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3630         status = nfs_ok;
3631
3632 out:
3633         spin_unlock(&nn->client_lock);
3634 out_nolock:
3635         if (new)
3636                 expire_client(new);
3637         if (unconf) {
3638                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3639                 expire_client(unconf);
3640         }
3641         return status;
3642 }
3643
3644 static __be32
3645 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3646 {
3647         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3648                 slot_seqid);
3649
3650         /* The slot is in use, and no response has been sent. */
3651         if (slot_inuse) {
3652                 if (seqid == slot_seqid)
3653                         return nfserr_jukebox;
3654                 else
3655                         return nfserr_seq_misordered;
3656         }
3657         /* Note unsigned 32-bit arithmetic handles wraparound: */
3658         if (likely(seqid == slot_seqid + 1))
3659                 return nfs_ok;
3660         if (seqid == slot_seqid)
3661                 return nfserr_replay_cache;
3662         return nfserr_seq_misordered;
3663 }
3664
3665 /*
3666  * Cache the create session result into the create session single DRC
3667  * slot cache by saving the xdr structure. sl_seqid has been set.
3668  * Do this for solo or embedded create session operations.
3669  */
3670 static void
3671 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3672                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3673 {
3674         slot->sl_status = nfserr;
3675         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3676 }
3677
3678 static __be32
3679 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3680                             struct nfsd4_clid_slot *slot)
3681 {
3682         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3683         return slot->sl_status;
3684 }
3685
3686 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3687                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3688                         1 +     /* MIN tag is length with zero, only length */ \
3689                         3 +     /* version, opcount, opcode */ \
3690                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3691                                 /* seqid, slotID, slotID, cache */ \
3692                         4 ) * sizeof(__be32))
3693
3694 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3695                         2 +     /* verifier: AUTH_NULL, length 0 */\
3696                         1 +     /* status */ \
3697                         1 +     /* MIN tag is length with zero, only length */ \
3698                         3 +     /* opcount, opcode, opstatus*/ \
3699                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3700                                 /* seqid, slotID, slotID, slotID, status */ \
3701                         5 ) * sizeof(__be32))
3702
3703 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3704 {
3705         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3706
3707         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3708                 return nfserr_toosmall;
3709         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3710                 return nfserr_toosmall;
3711         ca->headerpadsz = 0;
3712         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3713         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3714         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3715         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3716                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3717         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3718         /*
3719          * Note decreasing slot size below client's request may make it
3720          * difficult for client to function correctly, whereas
3721          * decreasing the number of slots will (just?) affect
3722          * performance.  When short on memory we therefore prefer to
3723          * decrease number of slots instead of their size.  Clients that
3724          * request larger slots than they need will get poor results:
3725          * Note that we always allow at least one slot, because our
3726          * accounting is soft and provides no guarantees either way.
3727          */
3728         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3729
3730         return nfs_ok;
3731 }
3732
3733 /*
3734  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3735  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3736  */
3737 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3738         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3739
3740 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3741         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3742
3743 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3744                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3745 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3746                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3747                                  sizeof(__be32))
3748
3749 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3750 {
3751         ca->headerpadsz = 0;
3752
3753         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3754                 return nfserr_toosmall;
3755         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3756                 return nfserr_toosmall;
3757         ca->maxresp_cached = 0;
3758         if (ca->maxops < 2)
3759                 return nfserr_toosmall;
3760
3761         return nfs_ok;
3762 }
3763
3764 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3765 {
3766         switch (cbs->flavor) {
3767         case RPC_AUTH_NULL:
3768         case RPC_AUTH_UNIX:
3769                 return nfs_ok;
3770         default:
3771                 /*
3772                  * GSS case: the spec doesn't allow us to return this
3773                  * error.  But it also doesn't allow us not to support
3774                  * GSS.
3775                  * I'd rather this fail hard than return some error the
3776                  * client might think it can already handle:
3777                  */
3778                 return nfserr_encr_alg_unsupp;
3779         }
3780 }
3781
3782 __be32
3783 nfsd4_create_session(struct svc_rqst *rqstp,
3784                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3785 {
3786         struct nfsd4_create_session *cr_ses = &u->create_session;
3787         struct sockaddr *sa = svc_addr(rqstp);
3788         struct nfs4_client *conf, *unconf;
3789         struct nfsd4_clid_slot *cs_slot;
3790         struct nfs4_client *old = NULL;
3791         struct nfsd4_session *new;
3792         struct nfsd4_conn *conn;
3793         __be32 status = 0;
3794         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3795
3796         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3797                 return nfserr_inval;
3798         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3799         if (status)
3800                 return status;
3801         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3802         if (status)
3803                 return status;
3804         status = check_backchannel_attrs(&cr_ses->back_channel);
3805         if (status)
3806                 goto out_release_drc_mem;
3807         status = nfserr_jukebox;
3808         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3809         if (!new)
3810                 goto out_release_drc_mem;
3811         conn = alloc_conn_from_crses(rqstp, cr_ses);
3812         if (!conn)
3813                 goto out_free_session;
3814
3815         spin_lock(&nn->client_lock);
3816
3817         /* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3818         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3819         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3820         if (!conf && !unconf) {
3821                 status = nfserr_stale_clientid;
3822                 goto out_free_conn;
3823         }
3824
3825         /* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
3826         if (conf)
3827                 cs_slot = &conf->cl_cs_slot;
3828         else
3829                 cs_slot = &unconf->cl_cs_slot;
3830         status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3831         switch (status) {
3832         case nfs_ok:
3833                 cs_slot->sl_seqid++;
3834                 cr_ses->seqid = cs_slot->sl_seqid;
3835                 break;
3836         case nfserr_replay_cache:
3837                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3838                 fallthrough;
3839         case nfserr_jukebox:
3840                 /* The server MUST NOT cache NFS4ERR_DELAY */
3841                 goto out_free_conn;
3842         default:
3843                 goto out_cache_error;
3844         }
3845
3846         /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3847         if (conf) {
3848                 status = nfserr_wrong_cred;
3849                 if (!nfsd4_mach_creds_match(conf, rqstp))
3850                         goto out_cache_error;
3851         } else {
3852                 status = nfserr_clid_inuse;
3853                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3854                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3855                         trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3856                         goto out_cache_error;
3857                 }
3858                 status = nfserr_wrong_cred;
3859                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3860                         goto out_cache_error;
3861                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3862                 if (old) {
3863                         status = mark_client_expired_locked(old);
3864                         if (status)
3865                                 goto out_expired_error;
3866                         trace_nfsd_clid_replaced(&old->cl_clientid);
3867                 }
3868                 move_to_confirmed(unconf);
3869                 conf = unconf;
3870         }
3871
3872         /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3873         status = nfs_ok;
3874         /* Persistent sessions are not supported */
3875         cr_ses->flags &= ~SESSION4_PERSIST;
3876         /* Upshifting from TCP to RDMA is not supported */
3877         cr_ses->flags &= ~SESSION4_RDMA;
3878
3879         init_session(rqstp, new, conf, cr_ses);
3880         nfsd4_get_session_locked(new);
3881
3882         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3883                NFS4_MAX_SESSIONID_LEN);
3884
3885         /* cache solo and embedded create sessions under the client_lock */
3886         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3887         spin_unlock(&nn->client_lock);
3888         if (conf == unconf)
3889                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3890         /* init connection and backchannel */
3891         nfsd4_init_conn(rqstp, conn, new);
3892         nfsd4_put_session(new);
3893         if (old)
3894                 expire_client(old);
3895         return status;
3896
3897 out_expired_error:
3898         old = NULL;
3899         /*
3900          * Revert the slot seq_nr change so the server will process
3901          * the client's resend instead of returning a cached response.
3902          */
3903         if (status == nfserr_jukebox) {
3904                 cs_slot->sl_seqid--;
3905                 cr_ses->seqid = cs_slot->sl_seqid;
3906                 goto out_free_conn;
3907         }
3908 out_cache_error:
3909         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3910 out_free_conn:
3911         spin_unlock(&nn->client_lock);
3912         free_conn(conn);
3913         if (old)
3914                 expire_client(old);
3915 out_free_session:
3916         __free_session(new);
3917 out_release_drc_mem:
3918         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3919         return status;
3920 }
3921
3922 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3923 {
3924         switch (*dir) {
3925         case NFS4_CDFC4_FORE:
3926         case NFS4_CDFC4_BACK:
3927                 return nfs_ok;
3928         case NFS4_CDFC4_FORE_OR_BOTH:
3929         case NFS4_CDFC4_BACK_OR_BOTH:
3930                 *dir = NFS4_CDFC4_BOTH;
3931                 return nfs_ok;
3932         }
3933         return nfserr_inval;
3934 }
3935
3936 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3937                 struct nfsd4_compound_state *cstate,
3938                 union nfsd4_op_u *u)
3939 {
3940         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3941         struct nfsd4_session *session = cstate->session;
3942         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3943         __be32 status;
3944
3945         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3946         if (status)
3947                 return status;
3948         spin_lock(&nn->client_lock);
3949         session->se_cb_prog = bc->bc_cb_program;
3950         session->se_cb_sec = bc->bc_cb_sec;
3951         spin_unlock(&nn->client_lock);
3952
3953         nfsd4_probe_callback(session->se_client);
3954
3955         return nfs_ok;
3956 }
3957
3958 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3959 {
3960         struct nfsd4_conn *c;
3961
3962         list_for_each_entry(c, &s->se_conns, cn_persession) {
3963                 if (c->cn_xprt == xpt) {
3964                         return c;
3965                 }
3966         }
3967         return NULL;
3968 }
3969
3970 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3971                 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3972 {
3973         struct nfs4_client *clp = session->se_client;
3974         struct svc_xprt *xpt = rqst->rq_xprt;
3975         struct nfsd4_conn *c;
3976         __be32 status;
3977
3978         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3979         spin_lock(&clp->cl_lock);
3980         c = __nfsd4_find_conn(xpt, session);
3981         if (!c)
3982                 status = nfserr_noent;
3983         else if (req == c->cn_flags)
3984                 status = nfs_ok;
3985         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3986                                 c->cn_flags != NFS4_CDFC4_BACK)
3987                 status = nfs_ok;
3988         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3989                                 c->cn_flags != NFS4_CDFC4_FORE)
3990                 status = nfs_ok;
3991         else
3992                 status = nfserr_inval;
3993         spin_unlock(&clp->cl_lock);
3994         if (status == nfs_ok && conn)
3995                 *conn = c;
3996         return status;
3997 }
3998
3999 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
4000                      struct nfsd4_compound_state *cstate,
4001                      union nfsd4_op_u *u)
4002 {
4003         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
4004         __be32 status;
4005         struct nfsd4_conn *conn;
4006         struct nfsd4_session *session;
4007         struct net *net = SVC_NET(rqstp);
4008         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4009
4010         if (!nfsd4_last_compound_op(rqstp))
4011                 return nfserr_not_only_op;
4012         spin_lock(&nn->client_lock);
4013         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4014         spin_unlock(&nn->client_lock);
4015         if (!session)
4016                 goto out_no_session;
4017         status = nfserr_wrong_cred;
4018         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4019                 goto out;
4020         status = nfsd4_match_existing_connection(rqstp, session,
4021                         bcts->dir, &conn);
4022         if (status == nfs_ok) {
4023                 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4024                                 bcts->dir == NFS4_CDFC4_BACK)
4025                         conn->cn_flags |= NFS4_CDFC4_BACK;
4026                 nfsd4_probe_callback(session->se_client);
4027                 goto out;
4028         }
4029         if (status == nfserr_inval)
4030                 goto out;
4031         status = nfsd4_map_bcts_dir(&bcts->dir);
4032         if (status)
4033                 goto out;
4034         conn = alloc_conn(rqstp, bcts->dir);
4035         status = nfserr_jukebox;
4036         if (!conn)
4037                 goto out;
4038         nfsd4_init_conn(rqstp, conn, session);
4039         status = nfs_ok;
4040 out:
4041         nfsd4_put_session(session);
4042 out_no_session:
4043         return status;
4044 }
4045
4046 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4047 {
4048         if (!cstate->session)
4049                 return false;
4050         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4051 }
4052
4053 __be32
4054 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4055                 union nfsd4_op_u *u)
4056 {
4057         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4058         struct nfsd4_session *ses;
4059         __be32 status;
4060         int ref_held_by_me = 0;
4061         struct net *net = SVC_NET(r);
4062         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4063
4064         status = nfserr_not_only_op;
4065         if (nfsd4_compound_in_session(cstate, sessionid)) {
4066                 if (!nfsd4_last_compound_op(r))
4067                         goto out;
4068                 ref_held_by_me++;
4069         }
4070         dump_sessionid(__func__, sessionid);
4071         spin_lock(&nn->client_lock);
4072         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4073         if (!ses)
4074                 goto out_client_lock;
4075         status = nfserr_wrong_cred;
4076         if (!nfsd4_mach_creds_match(ses->se_client, r))
4077                 goto out_put_session;
4078         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4079         if (status)
4080                 goto out_put_session;
4081         unhash_session(ses);
4082         spin_unlock(&nn->client_lock);
4083
4084         nfsd4_probe_callback_sync(ses->se_client);
4085
4086         spin_lock(&nn->client_lock);
4087         status = nfs_ok;
4088 out_put_session:
4089         nfsd4_put_session_locked(ses);
4090 out_client_lock:
4091         spin_unlock(&nn->client_lock);
4092 out:
4093         return status;
4094 }
4095
4096 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4097 {
4098         struct nfs4_client *clp = ses->se_client;
4099         struct nfsd4_conn *c;
4100         __be32 status = nfs_ok;
4101         int ret;
4102
4103         spin_lock(&clp->cl_lock);
4104         c = __nfsd4_find_conn(new->cn_xprt, ses);
4105         if (c)
4106                 goto out_free;
4107         status = nfserr_conn_not_bound_to_session;
4108         if (clp->cl_mach_cred)
4109                 goto out_free;
4110         __nfsd4_hash_conn(new, ses);
4111         spin_unlock(&clp->cl_lock);
4112         ret = nfsd4_register_conn(new);
4113         if (ret)
4114                 /* oops; xprt is already down: */
4115                 nfsd4_conn_lost(&new->cn_xpt_user);
4116         return nfs_ok;
4117 out_free:
4118         spin_unlock(&clp->cl_lock);
4119         free_conn(new);
4120         return status;
4121 }
4122
4123 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4124 {
4125         struct nfsd4_compoundargs *args = rqstp->rq_argp;
4126
4127         return args->opcnt > session->se_fchannel.maxops;
4128 }
4129
4130 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4131                                   struct nfsd4_session *session)
4132 {
4133         struct xdr_buf *xb = &rqstp->rq_arg;
4134
4135         return xb->len > session->se_fchannel.maxreq_sz;
4136 }
4137
4138 static bool replay_matches_cache(struct svc_rqst *rqstp,
4139                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4140 {
4141         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4142
4143         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4144             (bool)seq->cachethis)
4145                 return false;
4146         /*
4147          * If there's an error then the reply can have fewer ops than
4148          * the call.
4149          */
4150         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4151                 return false;
4152         /*
4153          * But if we cached a reply with *more* ops than the call you're
4154          * sending us now, then this new call is clearly not really a
4155          * replay of the old one:
4156          */
4157         if (slot->sl_opcnt > argp->opcnt)
4158                 return false;
4159         /* This is the only check explicitly called by spec: */
4160         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4161                 return false;
4162         /*
4163          * There may be more comparisons we could actually do, but the
4164          * spec doesn't require us to catch every case where the calls
4165          * don't match (that would require caching the call as well as
4166          * the reply), so we don't bother.
4167          */
4168         return true;
4169 }
4170
4171 __be32
4172 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4173                 union nfsd4_op_u *u)
4174 {
4175         struct nfsd4_sequence *seq = &u->sequence;
4176         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4177         struct xdr_stream *xdr = resp->xdr;
4178         struct nfsd4_session *session;
4179         struct nfs4_client *clp;
4180         struct nfsd4_slot *slot;
4181         struct nfsd4_conn *conn;
4182         __be32 status;
4183         int buflen;
4184         struct net *net = SVC_NET(rqstp);
4185         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4186
4187         if (resp->opcnt != 1)
4188                 return nfserr_sequence_pos;
4189
4190         /*
4191          * Will be either used or freed by nfsd4_sequence_check_conn
4192          * below.
4193          */
4194         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4195         if (!conn)
4196                 return nfserr_jukebox;
4197
4198         spin_lock(&nn->client_lock);
4199         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4200         if (!session)
4201                 goto out_no_session;
4202         clp = session->se_client;
4203
4204         status = nfserr_too_many_ops;
4205         if (nfsd4_session_too_many_ops(rqstp, session))
4206                 goto out_put_session;
4207
4208         status = nfserr_req_too_big;
4209         if (nfsd4_request_too_big(rqstp, session))
4210                 goto out_put_session;
4211
4212         status = nfserr_badslot;
4213         if (seq->slotid >= session->se_fchannel.maxreqs)
4214                 goto out_put_session;
4215
4216         slot = session->se_slots[seq->slotid];
4217         dprintk("%s: slotid %d\n", __func__, seq->slotid);
4218
4219         /* We do not negotiate the number of slots yet, so set the
4220          * maxslots to the session maxreqs which is used to encode
4221          * sr_highest_slotid and the sr_target_slot id to maxslots */
4222         seq->maxslots = session->se_fchannel.maxreqs;
4223
4224         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4225                                         slot->sl_flags & NFSD4_SLOT_INUSE);
4226         if (status == nfserr_replay_cache) {
4227                 status = nfserr_seq_misordered;
4228                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4229                         goto out_put_session;
4230                 status = nfserr_seq_false_retry;
4231                 if (!replay_matches_cache(rqstp, seq, slot))
4232                         goto out_put_session;
4233                 cstate->slot = slot;
4234                 cstate->session = session;
4235                 cstate->clp = clp;
4236                 /* Return the cached reply status and set cstate->status
4237                  * for nfsd4_proc_compound processing */
4238                 status = nfsd4_replay_cache_entry(resp, seq);
4239                 cstate->status = nfserr_replay_cache;
4240                 goto out;
4241         }
4242         if (status)
4243                 goto out_put_session;
4244
4245         status = nfsd4_sequence_check_conn(conn, session);
4246         conn = NULL;
4247         if (status)
4248                 goto out_put_session;
4249
4250         buflen = (seq->cachethis) ?
4251                         session->se_fchannel.maxresp_cached :
4252                         session->se_fchannel.maxresp_sz;
4253         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4254                                     nfserr_rep_too_big;
4255         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4256                 goto out_put_session;
4257         svc_reserve(rqstp, buflen);
4258
4259         status = nfs_ok;
4260         /* Success! bump slot seqid */
4261         slot->sl_seqid = seq->seqid;
4262         slot->sl_flags |= NFSD4_SLOT_INUSE;
4263         if (seq->cachethis)
4264                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4265         else
4266                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4267
4268         cstate->slot = slot;
4269         cstate->session = session;
4270         cstate->clp = clp;
4271
4272 out:
4273         switch (clp->cl_cb_state) {
4274         case NFSD4_CB_DOWN:
4275                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4276                 break;
4277         case NFSD4_CB_FAULT:
4278                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4279                 break;
4280         default:
4281                 seq->status_flags = 0;
4282         }
4283         if (!list_empty(&clp->cl_revoked))
4284                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4285         if (atomic_read(&clp->cl_admin_revoked))
4286                 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4287         trace_nfsd_seq4_status(rqstp, seq);
4288 out_no_session:
4289         if (conn)
4290                 free_conn(conn);
4291         spin_unlock(&nn->client_lock);
4292         return status;
4293 out_put_session:
4294         nfsd4_put_session_locked(session);
4295         goto out_no_session;
4296 }
4297
4298 void
4299 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4300 {
4301         struct nfsd4_compound_state *cs = &resp->cstate;
4302
4303         if (nfsd4_has_session(cs)) {
4304                 if (cs->status != nfserr_replay_cache) {
4305                         nfsd4_store_cache_entry(resp);
4306                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4307                 }
4308                 /* Drop session reference that was taken in nfsd4_sequence() */
4309                 nfsd4_put_session(cs->session);
4310         } else if (cs->clp)
4311                 put_client_renew(cs->clp);
4312 }
4313
4314 __be32
4315 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4316                 struct nfsd4_compound_state *cstate,
4317                 union nfsd4_op_u *u)
4318 {
4319         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4320         struct nfs4_client *conf, *unconf;
4321         struct nfs4_client *clp = NULL;
4322         __be32 status = 0;
4323         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4324
4325         spin_lock(&nn->client_lock);
4326         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4327         conf = find_confirmed_client(&dc->clientid, true, nn);
4328         WARN_ON_ONCE(conf && unconf);
4329
4330         if (conf) {
4331                 if (client_has_state(conf)) {
4332                         status = nfserr_clientid_busy;
4333                         goto out;
4334                 }
4335                 status = mark_client_expired_locked(conf);
4336                 if (status)
4337                         goto out;
4338                 clp = conf;
4339         } else if (unconf)
4340                 clp = unconf;
4341         else {
4342                 status = nfserr_stale_clientid;
4343                 goto out;
4344         }
4345         if (!nfsd4_mach_creds_match(clp, rqstp)) {
4346                 clp = NULL;
4347                 status = nfserr_wrong_cred;
4348                 goto out;
4349         }
4350         trace_nfsd_clid_destroyed(&clp->cl_clientid);
4351         unhash_client_locked(clp);
4352 out:
4353         spin_unlock(&nn->client_lock);
4354         if (clp)
4355                 expire_client(clp);
4356         return status;
4357 }
4358
4359 __be32
4360 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4361                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4362 {
4363         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4364         struct nfs4_client *clp = cstate->clp;
4365         __be32 status = 0;
4366
4367         if (rc->rca_one_fs) {
4368                 if (!cstate->current_fh.fh_dentry)
4369                         return nfserr_nofilehandle;
4370                 /*
4371                  * We don't take advantage of the rca_one_fs case.
4372                  * That's OK, it's optional, we can safely ignore it.
4373                  */
4374                 return nfs_ok;
4375         }
4376
4377         status = nfserr_complete_already;
4378         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4379                 goto out;
4380
4381         status = nfserr_stale_clientid;
4382         if (is_client_expired(clp))
4383                 /*
4384                  * The following error isn't really legal.
4385                  * But we only get here if the client just explicitly
4386                  * destroyed the client.  Surely it no longer cares what
4387                  * error it gets back on an operation for the dead
4388                  * client.
4389                  */
4390                 goto out;
4391
4392         status = nfs_ok;
4393         trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4394         nfsd4_client_record_create(clp);
4395         inc_reclaim_complete(clp);
4396 out:
4397         return status;
4398 }
4399
4400 __be32
4401 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4402                   union nfsd4_op_u *u)
4403 {
4404         struct nfsd4_setclientid *setclid = &u->setclientid;
4405         struct xdr_netobj       clname = setclid->se_name;
4406         nfs4_verifier           clverifier = setclid->se_verf;
4407         struct nfs4_client      *conf, *new;
4408         struct nfs4_client      *unconf = NULL;
4409         __be32                  status;
4410         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4411
4412         new = create_client(clname, rqstp, &clverifier);
4413         if (new == NULL)
4414                 return nfserr_jukebox;
4415         spin_lock(&nn->client_lock);
4416         conf = find_confirmed_client_by_name(&clname, nn);
4417         if (conf && client_has_state(conf)) {
4418                 status = nfserr_clid_inuse;
4419                 if (clp_used_exchangeid(conf))
4420                         goto out;
4421                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4422                         trace_nfsd_clid_cred_mismatch(conf, rqstp);
4423                         goto out;
4424                 }
4425         }
4426         unconf = find_unconfirmed_client_by_name(&clname, nn);
4427         if (unconf)
4428                 unhash_client_locked(unconf);
4429         if (conf) {
4430                 if (same_verf(&conf->cl_verifier, &clverifier)) {
4431                         copy_clid(new, conf);
4432                         gen_confirm(new, nn);
4433                 } else
4434                         trace_nfsd_clid_verf_mismatch(conf, rqstp,
4435                                                       &clverifier);
4436         } else
4437                 trace_nfsd_clid_fresh(new);
4438         new->cl_minorversion = 0;
4439         gen_callback(new, setclid, rqstp);
4440         add_to_unconfirmed(new);
4441         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4442         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4443         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4444         new = NULL;
4445         status = nfs_ok;
4446 out:
4447         spin_unlock(&nn->client_lock);
4448         if (new)
4449                 free_client(new);
4450         if (unconf) {
4451                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4452                 expire_client(unconf);
4453         }
4454         return status;
4455 }
4456
4457 __be32
4458 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4459                         struct nfsd4_compound_state *cstate,
4460                         union nfsd4_op_u *u)
4461 {
4462         struct nfsd4_setclientid_confirm *setclientid_confirm =
4463                         &u->setclientid_confirm;
4464         struct nfs4_client *conf, *unconf;
4465         struct nfs4_client *old = NULL;
4466         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4467         clientid_t * clid = &setclientid_confirm->sc_clientid;
4468         __be32 status;
4469         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4470
4471         if (STALE_CLIENTID(clid, nn))
4472                 return nfserr_stale_clientid;
4473
4474         spin_lock(&nn->client_lock);
4475         conf = find_confirmed_client(clid, false, nn);
4476         unconf = find_unconfirmed_client(clid, false, nn);
4477         /*
4478          * We try hard to give out unique clientid's, so if we get an
4479          * attempt to confirm the same clientid with a different cred,
4480          * the client may be buggy; this should never happen.
4481          *
4482          * Nevertheless, RFC 7530 recommends INUSE for this case:
4483          */
4484         status = nfserr_clid_inuse;
4485         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4486                 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4487                 goto out;
4488         }
4489         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4490                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4491                 goto out;
4492         }
4493         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4494                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4495                         status = nfs_ok;
4496                 } else
4497                         status = nfserr_stale_clientid;
4498                 goto out;
4499         }
4500         status = nfs_ok;
4501         if (conf) {
4502                 old = unconf;
4503                 unhash_client_locked(old);
4504                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4505         } else {
4506                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4507                 if (old) {
4508                         status = nfserr_clid_inuse;
4509                         if (client_has_state(old)
4510                                         && !same_creds(&unconf->cl_cred,
4511                                                         &old->cl_cred)) {
4512                                 old = NULL;
4513                                 goto out;
4514                         }
4515                         status = mark_client_expired_locked(old);
4516                         if (status) {
4517                                 old = NULL;
4518                                 goto out;
4519                         }
4520                         trace_nfsd_clid_replaced(&old->cl_clientid);
4521                 }
4522                 move_to_confirmed(unconf);
4523                 conf = unconf;
4524         }
4525         get_client_locked(conf);
4526         spin_unlock(&nn->client_lock);
4527         if (conf == unconf)
4528                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4529         nfsd4_probe_callback(conf);
4530         spin_lock(&nn->client_lock);
4531         put_client_renew_locked(conf);
4532 out:
4533         spin_unlock(&nn->client_lock);
4534         if (old)
4535                 expire_client(old);
4536         return status;
4537 }
4538
4539 static struct nfs4_file *nfsd4_alloc_file(void)
4540 {
4541         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4542 }
4543
4544 /* OPEN Share state helper functions */
4545
4546 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4547 {
4548         refcount_set(&fp->fi_ref, 1);
4549         spin_lock_init(&fp->fi_lock);
4550         INIT_LIST_HEAD(&fp->fi_stateids);
4551         INIT_LIST_HEAD(&fp->fi_delegations);
4552         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4553         fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4554         fp->fi_deleg_file = NULL;
4555         fp->fi_had_conflict = false;
4556         fp->fi_share_deny = 0;
4557         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4558         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4559         fp->fi_aliased = false;
4560         fp->fi_inode = d_inode(fh->fh_dentry);
4561 #ifdef CONFIG_NFSD_PNFS
4562         INIT_LIST_HEAD(&fp->fi_lo_states);
4563         atomic_set(&fp->fi_lo_recalls, 0);
4564 #endif
4565 }
4566
4567 void
4568 nfsd4_free_slabs(void)
4569 {
4570         kmem_cache_destroy(client_slab);
4571         kmem_cache_destroy(openowner_slab);
4572         kmem_cache_destroy(lockowner_slab);
4573         kmem_cache_destroy(file_slab);
4574         kmem_cache_destroy(stateid_slab);
4575         kmem_cache_destroy(deleg_slab);
4576         kmem_cache_destroy(odstate_slab);
4577 }
4578
4579 int
4580 nfsd4_init_slabs(void)
4581 {
4582         client_slab = KMEM_CACHE(nfs4_client, 0);
4583         if (client_slab == NULL)
4584                 goto out;
4585         openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4586         if (openowner_slab == NULL)
4587                 goto out_free_client_slab;
4588         lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4589         if (lockowner_slab == NULL)
4590                 goto out_free_openowner_slab;
4591         file_slab = KMEM_CACHE(nfs4_file, 0);
4592         if (file_slab == NULL)
4593                 goto out_free_lockowner_slab;
4594         stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4595         if (stateid_slab == NULL)
4596                 goto out_free_file_slab;
4597         deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4598         if (deleg_slab == NULL)
4599                 goto out_free_stateid_slab;
4600         odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4601         if (odstate_slab == NULL)
4602                 goto out_free_deleg_slab;
4603         return 0;
4604
4605 out_free_deleg_slab:
4606         kmem_cache_destroy(deleg_slab);
4607 out_free_stateid_slab:
4608         kmem_cache_destroy(stateid_slab);
4609 out_free_file_slab:
4610         kmem_cache_destroy(file_slab);
4611 out_free_lockowner_slab:
4612         kmem_cache_destroy(lockowner_slab);
4613 out_free_openowner_slab:
4614         kmem_cache_destroy(openowner_slab);
4615 out_free_client_slab:
4616         kmem_cache_destroy(client_slab);
4617 out:
4618         return -ENOMEM;
4619 }
4620
4621 static unsigned long
4622 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4623 {
4624         int count;
4625         struct nfsd_net *nn = shrink->private_data;
4626
4627         count = atomic_read(&nn->nfsd_courtesy_clients);
4628         if (!count)
4629                 count = atomic_long_read(&num_delegations);
4630         if (count)
4631                 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4632         return (unsigned long)count;
4633 }
4634
4635 static unsigned long
4636 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4637 {
4638         return SHRINK_STOP;
4639 }
4640
4641 void
4642 nfsd4_init_leases_net(struct nfsd_net *nn)
4643 {
4644         struct sysinfo si;
4645         u64 max_clients;
4646
4647         nn->nfsd4_lease = 90;   /* default lease time */
4648         nn->nfsd4_grace = 90;
4649         nn->somebody_reclaimed = false;
4650         nn->track_reclaim_completes = false;
4651         nn->clverifier_counter = get_random_u32();
4652         nn->clientid_base = get_random_u32();
4653         nn->clientid_counter = nn->clientid_base + 1;
4654         nn->s2s_cp_cl_id = nn->clientid_counter++;
4655
4656         atomic_set(&nn->nfs4_client_count, 0);
4657         si_meminfo(&si);
4658         max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4659         max_clients *= NFS4_CLIENTS_PER_GB;
4660         nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4661
4662         atomic_set(&nn->nfsd_courtesy_clients, 0);
4663 }
4664
4665 static void init_nfs4_replay(struct nfs4_replay *rp)
4666 {
4667         rp->rp_status = nfserr_serverfault;
4668         rp->rp_buflen = 0;
4669         rp->rp_buf = rp->rp_ibuf;
4670         mutex_init(&rp->rp_mutex);
4671 }
4672
4673 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4674                 struct nfs4_stateowner *so)
4675 {
4676         if (!nfsd4_has_session(cstate)) {
4677                 mutex_lock(&so->so_replay.rp_mutex);
4678                 cstate->replay_owner = nfs4_get_stateowner(so);
4679         }
4680 }
4681
4682 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4683 {
4684         struct nfs4_stateowner *so = cstate->replay_owner;
4685
4686         if (so != NULL) {
4687                 cstate->replay_owner = NULL;
4688                 mutex_unlock(&so->so_replay.rp_mutex);
4689                 nfs4_put_stateowner(so);
4690         }
4691 }
4692
4693 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4694 {
4695         struct nfs4_stateowner *sop;
4696
4697         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4698         if (!sop)
4699                 return NULL;
4700
4701         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4702         if (!sop->so_owner.data) {
4703                 kmem_cache_free(slab, sop);
4704                 return NULL;
4705         }
4706
4707         INIT_LIST_HEAD(&sop->so_stateids);
4708         sop->so_client = clp;
4709         init_nfs4_replay(&sop->so_replay);
4710         atomic_set(&sop->so_count, 1);
4711         return sop;
4712 }
4713
4714 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4715 {
4716         lockdep_assert_held(&clp->cl_lock);
4717
4718         list_add(&oo->oo_owner.so_strhash,
4719                  &clp->cl_ownerstr_hashtbl[strhashval]);
4720         list_add(&oo->oo_perclient, &clp->cl_openowners);
4721 }
4722
4723 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4724 {
4725         unhash_openowner_locked(openowner(so));
4726 }
4727
4728 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4729 {
4730         struct nfs4_openowner *oo = openowner(so);
4731
4732         kmem_cache_free(openowner_slab, oo);
4733 }
4734
4735 static const struct nfs4_stateowner_operations openowner_ops = {
4736         .so_unhash =    nfs4_unhash_openowner,
4737         .so_free =      nfs4_free_openowner,
4738 };
4739
4740 static struct nfs4_ol_stateid *
4741 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4742 {
4743         struct nfs4_ol_stateid *local, *ret = NULL;
4744         struct nfs4_openowner *oo = open->op_openowner;
4745
4746         lockdep_assert_held(&fp->fi_lock);
4747
4748         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4749                 /* ignore lock owners */
4750                 if (local->st_stateowner->so_is_open_owner == 0)
4751                         continue;
4752                 if (local->st_stateowner != &oo->oo_owner)
4753                         continue;
4754                 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4755                     !local->st_stid.sc_status) {
4756                         ret = local;
4757                         refcount_inc(&ret->st_stid.sc_count);
4758                         break;
4759                 }
4760         }
4761         return ret;
4762 }
4763
4764 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4765         __releases(&s->sc_client->cl_lock)
4766 {
4767         struct nfs4_client *cl = s->sc_client;
4768         LIST_HEAD(reaplist);
4769         struct nfs4_ol_stateid *stp;
4770         struct nfs4_delegation *dp;
4771         bool unhashed;
4772
4773         switch (s->sc_type) {
4774         case SC_TYPE_OPEN:
4775                 stp = openlockstateid(s);
4776                 if (unhash_open_stateid(stp, &reaplist))
4777                         put_ol_stateid_locked(stp, &reaplist);
4778                 spin_unlock(&cl->cl_lock);
4779                 free_ol_stateid_reaplist(&reaplist);
4780                 break;
4781         case SC_TYPE_LOCK:
4782                 stp = openlockstateid(s);
4783                 unhashed = unhash_lock_stateid(stp);
4784                 spin_unlock(&cl->cl_lock);
4785                 if (unhashed)
4786                         nfs4_put_stid(s);
4787                 break;
4788         case SC_TYPE_DELEG:
4789                 dp = delegstateid(s);
4790                 list_del_init(&dp->dl_recall_lru);
4791                 spin_unlock(&cl->cl_lock);
4792                 nfs4_put_stid(s);
4793                 break;
4794         default:
4795                 spin_unlock(&cl->cl_lock);
4796         }
4797 }
4798
4799 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4800                                     stateid_t *stid)
4801 {
4802         /* NFSv4.0 has no way for the client to tell the server
4803          * that it can forget an admin-revoked stateid.
4804          * So we keep it around until the first time that the
4805          * client uses it, and drop it the first time
4806          * nfserr_admin_revoked is returned.
4807          * For v4.1 and later we wait until explicitly told
4808          * to free the stateid.
4809          */
4810         if (cl->cl_minorversion == 0) {
4811                 struct nfs4_stid *st;
4812
4813                 spin_lock(&cl->cl_lock);
4814                 st = find_stateid_locked(cl, stid);
4815                 if (st)
4816                         nfsd4_drop_revoked_stid(st);
4817                 else
4818                         spin_unlock(&cl->cl_lock);
4819         }
4820 }
4821
4822 static __be32
4823 nfsd4_verify_open_stid(struct nfs4_stid *s)
4824 {
4825         __be32 ret = nfs_ok;
4826
4827         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4828                 ret = nfserr_admin_revoked;
4829         else if (s->sc_status & SC_STATUS_REVOKED)
4830                 ret = nfserr_deleg_revoked;
4831         else if (s->sc_status & SC_STATUS_CLOSED)
4832                 ret = nfserr_bad_stateid;
4833         return ret;
4834 }
4835
4836 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4837 static __be32
4838 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4839 {
4840         __be32 ret;
4841
4842         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4843         ret = nfsd4_verify_open_stid(&stp->st_stid);
4844         if (ret == nfserr_admin_revoked)
4845                 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4846                                         &stp->st_stid.sc_stateid);
4847
4848         if (ret != nfs_ok)
4849                 mutex_unlock(&stp->st_mutex);
4850         return ret;
4851 }
4852
4853 static struct nfs4_ol_stateid *
4854 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4855 {
4856         struct nfs4_ol_stateid *stp;
4857         for (;;) {
4858                 spin_lock(&fp->fi_lock);
4859                 stp = nfsd4_find_existing_open(fp, open);
4860                 spin_unlock(&fp->fi_lock);
4861                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4862                         break;
4863                 nfs4_put_stid(&stp->st_stid);
4864         }
4865         return stp;
4866 }
4867
4868 static struct nfs4_openowner *
4869 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4870                            struct nfsd4_compound_state *cstate)
4871 {
4872         struct nfs4_client *clp = cstate->clp;
4873         struct nfs4_openowner *oo, *ret;
4874
4875         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4876         if (!oo)
4877                 return NULL;
4878         oo->oo_owner.so_ops = &openowner_ops;
4879         oo->oo_owner.so_is_open_owner = 1;
4880         oo->oo_owner.so_seqid = open->op_seqid;
4881         oo->oo_flags = 0;
4882         if (nfsd4_has_session(cstate))
4883                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4884         oo->oo_time = 0;
4885         oo->oo_last_closed_stid = NULL;
4886         INIT_LIST_HEAD(&oo->oo_close_lru);
4887         spin_lock(&clp->cl_lock);
4888         ret = find_openstateowner_str_locked(strhashval, open, clp);
4889         if (ret == NULL) {
4890                 hash_openowner(oo, clp, strhashval);
4891                 ret = oo;
4892         } else
4893                 nfs4_free_stateowner(&oo->oo_owner);
4894
4895         spin_unlock(&clp->cl_lock);
4896         return ret;
4897 }
4898
4899 static struct nfs4_ol_stateid *
4900 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4901 {
4902
4903         struct nfs4_openowner *oo = open->op_openowner;
4904         struct nfs4_ol_stateid *retstp = NULL;
4905         struct nfs4_ol_stateid *stp;
4906
4907         stp = open->op_stp;
4908         /* We are moving these outside of the spinlocks to avoid the warnings */
4909         mutex_init(&stp->st_mutex);
4910         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4911
4912 retry:
4913         spin_lock(&oo->oo_owner.so_client->cl_lock);
4914         spin_lock(&fp->fi_lock);
4915
4916         retstp = nfsd4_find_existing_open(fp, open);
4917         if (retstp)
4918                 goto out_unlock;
4919
4920         open->op_stp = NULL;
4921         refcount_inc(&stp->st_stid.sc_count);
4922         stp->st_stid.sc_type = SC_TYPE_OPEN;
4923         INIT_LIST_HEAD(&stp->st_locks);
4924         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4925         get_nfs4_file(fp);
4926         stp->st_stid.sc_file = fp;
4927         stp->st_access_bmap = 0;
4928         stp->st_deny_bmap = 0;
4929         stp->st_openstp = NULL;
4930         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4931         list_add(&stp->st_perfile, &fp->fi_stateids);
4932
4933 out_unlock:
4934         spin_unlock(&fp->fi_lock);
4935         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4936         if (retstp) {
4937                 /* Handle races with CLOSE */
4938                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4939                         nfs4_put_stid(&retstp->st_stid);
4940                         goto retry;
4941                 }
4942                 /* To keep mutex tracking happy */
4943                 mutex_unlock(&stp->st_mutex);
4944                 stp = retstp;
4945         }
4946         return stp;
4947 }
4948
4949 /*
4950  * In the 4.0 case we need to keep the owners around a little while to handle
4951  * CLOSE replay. We still do need to release any file access that is held by
4952  * them before returning however.
4953  */
4954 static void
4955 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4956 {
4957         struct nfs4_ol_stateid *last;
4958         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4959         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4960                                                 nfsd_net_id);
4961
4962         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4963
4964         /*
4965          * We know that we hold one reference via nfsd4_close, and another
4966          * "persistent" reference for the client. If the refcount is higher
4967          * than 2, then there are still calls in progress that are using this
4968          * stateid. We can't put the sc_file reference until they are finished.
4969          * Wait for the refcount to drop to 2. Since it has been unhashed,
4970          * there should be no danger of the refcount going back up again at
4971          * this point.
4972          */
4973         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4974
4975         release_all_access(s);
4976         if (s->st_stid.sc_file) {
4977                 put_nfs4_file(s->st_stid.sc_file);
4978                 s->st_stid.sc_file = NULL;
4979         }
4980
4981         spin_lock(&nn->client_lock);
4982         last = oo->oo_last_closed_stid;
4983         oo->oo_last_closed_stid = s;
4984         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4985         oo->oo_time = ktime_get_boottime_seconds();
4986         spin_unlock(&nn->client_lock);
4987         if (last)
4988                 nfs4_put_stid(&last->st_stid);
4989 }
4990
4991 static noinline_for_stack struct nfs4_file *
4992 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4993 {
4994         struct inode *inode = d_inode(fhp->fh_dentry);
4995         struct rhlist_head *tmp, *list;
4996         struct nfs4_file *fi;
4997
4998         rcu_read_lock();
4999         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5000                                nfs4_file_rhash_params);
5001         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5002                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5003                         if (refcount_inc_not_zero(&fi->fi_ref)) {
5004                                 rcu_read_unlock();
5005                                 return fi;
5006                         }
5007                 }
5008         }
5009         rcu_read_unlock();
5010         return NULL;
5011 }
5012
5013 /*
5014  * On hash insertion, identify entries with the same inode but
5015  * distinct filehandles. They will all be on the list returned
5016  * by rhltable_lookup().
5017  *
5018  * inode->i_lock prevents racing insertions from adding an entry
5019  * for the same inode/fhp pair twice.
5020  */
5021 static noinline_for_stack struct nfs4_file *
5022 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5023 {
5024         struct inode *inode = d_inode(fhp->fh_dentry);
5025         struct rhlist_head *tmp, *list;
5026         struct nfs4_file *ret = NULL;
5027         bool alias_found = false;
5028         struct nfs4_file *fi;
5029         int err;
5030
5031         rcu_read_lock();
5032         spin_lock(&inode->i_lock);
5033
5034         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5035                                nfs4_file_rhash_params);
5036         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5037                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5038                         if (refcount_inc_not_zero(&fi->fi_ref))
5039                                 ret = fi;
5040                 } else
5041                         fi->fi_aliased = alias_found = true;
5042         }
5043         if (ret)
5044                 goto out_unlock;
5045
5046         nfsd4_file_init(fhp, new);
5047         err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5048                               nfs4_file_rhash_params);
5049         if (err)
5050                 goto out_unlock;
5051
5052         new->fi_aliased = alias_found;
5053         ret = new;
5054
5055 out_unlock:
5056         spin_unlock(&inode->i_lock);
5057         rcu_read_unlock();
5058         return ret;
5059 }
5060
5061 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5062 {
5063         rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5064                         nfs4_file_rhash_params);
5065 }
5066
5067 /*
5068  * Called to check deny when READ with all zero stateid or
5069  * WRITE with all zero or all one stateid
5070  */
5071 static __be32
5072 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5073 {
5074         struct nfs4_file *fp;
5075         __be32 ret = nfs_ok;
5076
5077         fp = nfsd4_file_hash_lookup(current_fh);
5078         if (!fp)
5079                 return ret;
5080
5081         /* Check for conflicting share reservations */
5082         spin_lock(&fp->fi_lock);
5083         if (fp->fi_share_deny & deny_type)
5084                 ret = nfserr_locked;
5085         spin_unlock(&fp->fi_lock);
5086         put_nfs4_file(fp);
5087         return ret;
5088 }
5089
5090 static bool nfsd4_deleg_present(const struct inode *inode)
5091 {
5092         struct file_lock_context *ctx = locks_inode_context(inode);
5093
5094         return ctx && !list_empty_careful(&ctx->flc_lease);
5095 }
5096
5097 /**
5098  * nfsd_wait_for_delegreturn - wait for delegations to be returned
5099  * @rqstp: the RPC transaction being executed
5100  * @inode: in-core inode of the file being waited for
5101  *
5102  * The timeout prevents deadlock if all nfsd threads happen to be
5103  * tied up waiting for returning delegations.
5104  *
5105  * Return values:
5106  *   %true: delegation was returned
5107  *   %false: timed out waiting for delegreturn
5108  */
5109 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5110 {
5111         long __maybe_unused timeo;
5112
5113         timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5114                                        NFSD_DELEGRETURN_TIMEOUT);
5115         trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5116         return timeo > 0;
5117 }
5118
5119 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5120 {
5121         struct nfs4_delegation *dp = cb_to_delegation(cb);
5122         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5123                                           nfsd_net_id);
5124
5125         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5126
5127         /*
5128          * We can't do this in nfsd_break_deleg_cb because it is
5129          * already holding inode->i_lock.
5130          *
5131          * If the dl_time != 0, then we know that it has already been
5132          * queued for a lease break. Don't queue it again.
5133          */
5134         spin_lock(&state_lock);
5135         if (delegation_hashed(dp) && dp->dl_time == 0) {
5136                 dp->dl_time = ktime_get_boottime_seconds();
5137                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5138         }
5139         spin_unlock(&state_lock);
5140 }
5141
5142 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5143                 struct rpc_task *task)
5144 {
5145         struct nfs4_delegation *dp = cb_to_delegation(cb);
5146
5147         trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5148
5149         if (dp->dl_stid.sc_status)
5150                 /* CLOSED or REVOKED */
5151                 return 1;
5152
5153         switch (task->tk_status) {
5154         case 0:
5155                 return 1;
5156         case -NFS4ERR_DELAY:
5157                 rpc_delay(task, 2 * HZ);
5158                 return 0;
5159         case -EBADHANDLE:
5160         case -NFS4ERR_BAD_STATEID:
5161                 /*
5162                  * Race: client probably got cb_recall before open reply
5163                  * granting delegation.
5164                  */
5165                 if (dp->dl_retries--) {
5166                         rpc_delay(task, 2 * HZ);
5167                         return 0;
5168                 }
5169                 fallthrough;
5170         default:
5171                 return 1;
5172         }
5173 }
5174
5175 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5176 {
5177         struct nfs4_delegation *dp = cb_to_delegation(cb);
5178
5179         nfs4_put_stid(&dp->dl_stid);
5180 }
5181
5182 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5183         .prepare        = nfsd4_cb_recall_prepare,
5184         .done           = nfsd4_cb_recall_done,
5185         .release        = nfsd4_cb_recall_release,
5186 };
5187
5188 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5189 {
5190         /*
5191          * We're assuming the state code never drops its reference
5192          * without first removing the lease.  Since we're in this lease
5193          * callback (and since the lease code is serialized by the
5194          * flc_lock) we know the server hasn't removed the lease yet, and
5195          * we know it's safe to take a reference.
5196          */
5197         refcount_inc(&dp->dl_stid.sc_count);
5198         WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5199 }
5200
5201 /* Called from break_lease() with flc_lock held. */
5202 static bool
5203 nfsd_break_deleg_cb(struct file_lease *fl)
5204 {
5205         struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5206         struct nfs4_file *fp = dp->dl_stid.sc_file;
5207         struct nfs4_client *clp = dp->dl_stid.sc_client;
5208         struct nfsd_net *nn;
5209
5210         trace_nfsd_cb_recall(&dp->dl_stid);
5211
5212         dp->dl_recalled = true;
5213         atomic_inc(&clp->cl_delegs_in_recall);
5214         if (try_to_expire_client(clp)) {
5215                 nn = net_generic(clp->net, nfsd_net_id);
5216                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5217         }
5218
5219         /*
5220          * We don't want the locks code to timeout the lease for us;
5221          * we'll remove it ourself if a delegation isn't returned
5222          * in time:
5223          */
5224         fl->fl_break_time = 0;
5225
5226         fp->fi_had_conflict = true;
5227         nfsd_break_one_deleg(dp);
5228         return false;
5229 }
5230
5231 /**
5232  * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5233  * @fl: Lock state to check
5234  *
5235  * Return values:
5236  *   %true: Lease conflict was resolved
5237  *   %false: Lease conflict was not resolved.
5238  */
5239 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5240 {
5241         struct nfs4_delegation *dl = fl->c.flc_owner;
5242         struct svc_rqst *rqst;
5243         struct nfs4_client *clp;
5244
5245         if (!i_am_nfsd())
5246                 return false;
5247         rqst = kthread_data(current);
5248         /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
5249         if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
5250                 return false;
5251         clp = *(rqst->rq_lease_breaker);
5252         return dl->dl_stid.sc_client == clp;
5253 }
5254
5255 static int
5256 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5257                      struct list_head *dispose)
5258 {
5259         struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5260         struct nfs4_client *clp = dp->dl_stid.sc_client;
5261
5262         if (arg & F_UNLCK) {
5263                 if (dp->dl_recalled)
5264                         atomic_dec(&clp->cl_delegs_in_recall);
5265                 return lease_modify(onlist, arg, dispose);
5266         } else
5267                 return -EAGAIN;
5268 }
5269
5270 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5271         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5272         .lm_break = nfsd_break_deleg_cb,
5273         .lm_change = nfsd_change_deleg_cb,
5274 };
5275
5276 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5277 {
5278         if (nfsd4_has_session(cstate))
5279                 return nfs_ok;
5280         if (seqid == so->so_seqid - 1)
5281                 return nfserr_replay_me;
5282         if (seqid == so->so_seqid)
5283                 return nfs_ok;
5284         return nfserr_bad_seqid;
5285 }
5286
5287 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5288                                                 struct nfsd_net *nn)
5289 {
5290         struct nfs4_client *found;
5291
5292         spin_lock(&nn->client_lock);
5293         found = find_confirmed_client(clid, sessions, nn);
5294         if (found)
5295                 atomic_inc(&found->cl_rpc_users);
5296         spin_unlock(&nn->client_lock);
5297         return found;
5298 }
5299
5300 static __be32 set_client(clientid_t *clid,
5301                 struct nfsd4_compound_state *cstate,
5302                 struct nfsd_net *nn)
5303 {
5304         if (cstate->clp) {
5305                 if (!same_clid(&cstate->clp->cl_clientid, clid))
5306                         return nfserr_stale_clientid;
5307                 return nfs_ok;
5308         }
5309         if (STALE_CLIENTID(clid, nn))
5310                 return nfserr_stale_clientid;
5311         /*
5312          * We're in the 4.0 case (otherwise the SEQUENCE op would have
5313          * set cstate->clp), so session = false:
5314          */
5315         cstate->clp = lookup_clientid(clid, false, nn);
5316         if (!cstate->clp)
5317                 return nfserr_expired;
5318         return nfs_ok;
5319 }
5320
5321 __be32
5322 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5323                     struct nfsd4_open *open, struct nfsd_net *nn)
5324 {
5325         clientid_t *clientid = &open->op_clientid;
5326         struct nfs4_client *clp = NULL;
5327         unsigned int strhashval;
5328         struct nfs4_openowner *oo = NULL;
5329         __be32 status;
5330
5331         /*
5332          * In case we need it later, after we've already created the
5333          * file and don't want to risk a further failure:
5334          */
5335         open->op_file = nfsd4_alloc_file();
5336         if (open->op_file == NULL)
5337                 return nfserr_jukebox;
5338
5339         status = set_client(clientid, cstate, nn);
5340         if (status)
5341                 return status;
5342         clp = cstate->clp;
5343
5344         strhashval = ownerstr_hashval(&open->op_owner);
5345         oo = find_openstateowner_str(strhashval, open, clp);
5346         open->op_openowner = oo;
5347         if (!oo) {
5348                 goto new_owner;
5349         }
5350         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5351                 /* Replace unconfirmed owners without checking for replay. */
5352                 release_openowner(oo);
5353                 open->op_openowner = NULL;
5354                 goto new_owner;
5355         }
5356         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5357         if (status)
5358                 return status;
5359         goto alloc_stateid;
5360 new_owner:
5361         oo = alloc_init_open_stateowner(strhashval, open, cstate);
5362         if (oo == NULL)
5363                 return nfserr_jukebox;
5364         open->op_openowner = oo;
5365 alloc_stateid:
5366         open->op_stp = nfs4_alloc_open_stateid(clp);
5367         if (!open->op_stp)
5368                 return nfserr_jukebox;
5369
5370         if (nfsd4_has_session(cstate) &&
5371             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5372                 open->op_odstate = alloc_clnt_odstate(clp);
5373                 if (!open->op_odstate)
5374                         return nfserr_jukebox;
5375         }
5376
5377         return nfs_ok;
5378 }
5379
5380 static inline __be32
5381 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5382 {
5383         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5384                 return nfserr_openmode;
5385         else
5386                 return nfs_ok;
5387 }
5388
5389 static int share_access_to_flags(u32 share_access)
5390 {
5391         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5392 }
5393
5394 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5395                                                   stateid_t *s)
5396 {
5397         struct nfs4_stid *ret;
5398
5399         ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5400         if (!ret)
5401                 return NULL;
5402         return delegstateid(ret);
5403 }
5404
5405 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5406 {
5407         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5408                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5409 }
5410
5411 static __be32
5412 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5413                 struct nfs4_delegation **dp)
5414 {
5415         int flags;
5416         __be32 status = nfserr_bad_stateid;
5417         struct nfs4_delegation *deleg;
5418
5419         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5420         if (deleg == NULL)
5421                 goto out;
5422         if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5423                 nfs4_put_stid(&deleg->dl_stid);
5424                 status = nfserr_admin_revoked;
5425                 goto out;
5426         }
5427         if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5428                 nfs4_put_stid(&deleg->dl_stid);
5429                 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5430                 status = nfserr_deleg_revoked;
5431                 goto out;
5432         }
5433         flags = share_access_to_flags(open->op_share_access);
5434         status = nfs4_check_delegmode(deleg, flags);
5435         if (status) {
5436                 nfs4_put_stid(&deleg->dl_stid);
5437                 goto out;
5438         }
5439         *dp = deleg;
5440 out:
5441         if (!nfsd4_is_deleg_cur(open))
5442                 return nfs_ok;
5443         if (status)
5444                 return status;
5445         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5446         return nfs_ok;
5447 }
5448
5449 static inline int nfs4_access_to_access(u32 nfs4_access)
5450 {
5451         int flags = 0;
5452
5453         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5454                 flags |= NFSD_MAY_READ;
5455         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5456                 flags |= NFSD_MAY_WRITE;
5457         return flags;
5458 }
5459
5460 static inline __be32
5461 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5462                 struct nfsd4_open *open)
5463 {
5464         struct iattr iattr = {
5465                 .ia_valid = ATTR_SIZE,
5466                 .ia_size = 0,
5467         };
5468         struct nfsd_attrs attrs = {
5469                 .na_iattr       = &iattr,
5470         };
5471         if (!open->op_truncate)
5472                 return 0;
5473         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5474                 return nfserr_inval;
5475         return nfsd_setattr(rqstp, fh, &attrs, NULL);
5476 }
5477
5478 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5479                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5480                 struct nfsd4_open *open, bool new_stp)
5481 {
5482         struct nfsd_file *nf = NULL;
5483         __be32 status;
5484         int oflag = nfs4_access_to_omode(open->op_share_access);
5485         int access = nfs4_access_to_access(open->op_share_access);
5486         unsigned char old_access_bmap, old_deny_bmap;
5487
5488         spin_lock(&fp->fi_lock);
5489
5490         /*
5491          * Are we trying to set a deny mode that would conflict with
5492          * current access?
5493          */
5494         status = nfs4_file_check_deny(fp, open->op_share_deny);
5495         if (status != nfs_ok) {
5496                 if (status != nfserr_share_denied) {
5497                         spin_unlock(&fp->fi_lock);
5498                         goto out;
5499                 }
5500                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5501                                 stp, open->op_share_deny, false))
5502                         status = nfserr_jukebox;
5503                 spin_unlock(&fp->fi_lock);
5504                 goto out;
5505         }
5506
5507         /* set access to the file */
5508         status = nfs4_file_get_access(fp, open->op_share_access);
5509         if (status != nfs_ok) {
5510                 if (status != nfserr_share_denied) {
5511                         spin_unlock(&fp->fi_lock);
5512                         goto out;
5513                 }
5514                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5515                                 stp, open->op_share_access, true))
5516                         status = nfserr_jukebox;
5517                 spin_unlock(&fp->fi_lock);
5518                 goto out;
5519         }
5520
5521         /* Set access bits in stateid */
5522         old_access_bmap = stp->st_access_bmap;
5523         set_access(open->op_share_access, stp);
5524
5525         /* Set new deny mask */
5526         old_deny_bmap = stp->st_deny_bmap;
5527         set_deny(open->op_share_deny, stp);
5528         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5529
5530         if (!fp->fi_fds[oflag]) {
5531                 spin_unlock(&fp->fi_lock);
5532
5533                 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5534                                                   open->op_filp, &nf);
5535                 if (status != nfs_ok)
5536                         goto out_put_access;
5537
5538                 spin_lock(&fp->fi_lock);
5539                 if (!fp->fi_fds[oflag]) {
5540                         fp->fi_fds[oflag] = nf;
5541                         nf = NULL;
5542                 }
5543         }
5544         spin_unlock(&fp->fi_lock);
5545         if (nf)
5546                 nfsd_file_put(nf);
5547
5548         status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5549                                                                 access));
5550         if (status)
5551                 goto out_put_access;
5552
5553         status = nfsd4_truncate(rqstp, cur_fh, open);
5554         if (status)
5555                 goto out_put_access;
5556 out:
5557         return status;
5558 out_put_access:
5559         stp->st_access_bmap = old_access_bmap;
5560         nfs4_file_put_access(fp, open->op_share_access);
5561         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5562         goto out;
5563 }
5564
5565 static __be32
5566 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5567                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5568                 struct nfsd4_open *open)
5569 {
5570         __be32 status;
5571         unsigned char old_deny_bmap = stp->st_deny_bmap;
5572
5573         if (!test_access(open->op_share_access, stp))
5574                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5575
5576         /* test and set deny mode */
5577         spin_lock(&fp->fi_lock);
5578         status = nfs4_file_check_deny(fp, open->op_share_deny);
5579         switch (status) {
5580         case nfs_ok:
5581                 set_deny(open->op_share_deny, stp);
5582                 fp->fi_share_deny |=
5583                         (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5584                 break;
5585         case nfserr_share_denied:
5586                 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5587                                 stp, open->op_share_deny, false))
5588                         status = nfserr_jukebox;
5589                 break;
5590         }
5591         spin_unlock(&fp->fi_lock);
5592
5593         if (status != nfs_ok)
5594                 return status;
5595
5596         status = nfsd4_truncate(rqstp, cur_fh, open);
5597         if (status != nfs_ok)
5598                 reset_union_bmap_deny(old_deny_bmap, stp);
5599         return status;
5600 }
5601
5602 /* Should we give out recallable state?: */
5603 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5604 {
5605         if (clp->cl_cb_state == NFSD4_CB_UP)
5606                 return true;
5607         /*
5608          * In the sessions case, since we don't have to establish a
5609          * separate connection for callbacks, we assume it's OK
5610          * until we hear otherwise:
5611          */
5612         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5613 }
5614
5615 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5616                                                 int flag)
5617 {
5618         struct file_lease *fl;
5619
5620         fl = locks_alloc_lease();
5621         if (!fl)
5622                 return NULL;
5623         fl->fl_lmops = &nfsd_lease_mng_ops;
5624         fl->c.flc_flags = FL_DELEG;
5625         fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5626         fl->c.flc_owner = (fl_owner_t)dp;
5627         fl->c.flc_pid = current->tgid;
5628         fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5629         return fl;
5630 }
5631
5632 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5633                                          struct nfs4_file *fp)
5634 {
5635         struct nfs4_ol_stateid *st;
5636         struct file *f = fp->fi_deleg_file->nf_file;
5637         struct inode *ino = file_inode(f);
5638         int writes;
5639
5640         writes = atomic_read(&ino->i_writecount);
5641         if (!writes)
5642                 return 0;
5643         /*
5644          * There could be multiple filehandles (hence multiple
5645          * nfs4_files) referencing this file, but that's not too
5646          * common; let's just give up in that case rather than
5647          * trying to go look up all the clients using that other
5648          * nfs4_file as well:
5649          */
5650         if (fp->fi_aliased)
5651                 return -EAGAIN;
5652         /*
5653          * If there's a close in progress, make sure that we see it
5654          * clear any fi_fds[] entries before we see it decrement
5655          * i_writecount:
5656          */
5657         smp_mb__after_atomic();
5658
5659         if (fp->fi_fds[O_WRONLY])
5660                 writes--;
5661         if (fp->fi_fds[O_RDWR])
5662                 writes--;
5663         if (writes > 0)
5664                 return -EAGAIN; /* There may be non-NFSv4 writers */
5665         /*
5666          * It's possible there are non-NFSv4 write opens in progress,
5667          * but if they haven't incremented i_writecount yet then they
5668          * also haven't called break lease yet; so, they'll break this
5669          * lease soon enough.  So, all that's left to check for is NFSv4
5670          * opens:
5671          */
5672         spin_lock(&fp->fi_lock);
5673         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5674                 if (st->st_openstp == NULL /* it's an open */ &&
5675                     access_permit_write(st) &&
5676                     st->st_stid.sc_client != clp) {
5677                         spin_unlock(&fp->fi_lock);
5678                         return -EAGAIN;
5679                 }
5680         }
5681         spin_unlock(&fp->fi_lock);
5682         /*
5683          * There's a small chance that we could be racing with another
5684          * NFSv4 open.  However, any open that hasn't added itself to
5685          * the fi_stateids list also hasn't called break_lease yet; so,
5686          * they'll break this lease soon enough.
5687          */
5688         return 0;
5689 }
5690
5691 /*
5692  * It's possible that between opening the dentry and setting the delegation,
5693  * that it has been renamed or unlinked. Redo the lookup to verify that this
5694  * hasn't happened.
5695  */
5696 static int
5697 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5698                           struct svc_fh *parent)
5699 {
5700         struct svc_export *exp;
5701         struct dentry *child;
5702         __be32 err;
5703
5704         err = nfsd_lookup_dentry(open->op_rqstp, parent,
5705                                  open->op_fname, open->op_fnamelen,
5706                                  &exp, &child);
5707
5708         if (err)
5709                 return -EAGAIN;
5710
5711         exp_put(exp);
5712         dput(child);
5713         if (child != file_dentry(fp->fi_deleg_file->nf_file))
5714                 return -EAGAIN;
5715
5716         return 0;
5717 }
5718
5719 /*
5720  * We avoid breaking delegations held by a client due to its own activity, but
5721  * clearing setuid/setgid bits on a write is an implicit activity and the client
5722  * may not notice and continue using the old mode. Avoid giving out a delegation
5723  * on setuid/setgid files when the client is requesting an open for write.
5724  */
5725 static int
5726 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5727 {
5728         struct inode *inode = file_inode(nf->nf_file);
5729
5730         if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5731             (inode->i_mode & (S_ISUID|S_ISGID)))
5732                 return -EAGAIN;
5733         return 0;
5734 }
5735
5736 static struct nfs4_delegation *
5737 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5738                     struct svc_fh *parent)
5739 {
5740         int status = 0;
5741         struct nfs4_client *clp = stp->st_stid.sc_client;
5742         struct nfs4_file *fp = stp->st_stid.sc_file;
5743         struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5744         struct nfs4_delegation *dp;
5745         struct nfsd_file *nf = NULL;
5746         struct file_lease *fl;
5747         u32 dl_type;
5748
5749         /*
5750          * The fi_had_conflict and nfs_get_existing_delegation checks
5751          * here are just optimizations; we'll need to recheck them at
5752          * the end:
5753          */
5754         if (fp->fi_had_conflict)
5755                 return ERR_PTR(-EAGAIN);
5756
5757         /*
5758          * Try for a write delegation first. RFC8881 section 10.4 says:
5759          *
5760          *  "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5761          *   on its own, all opens."
5762          *
5763          * Furthermore the client can use a write delegation for most READ
5764          * operations as well, so we require a O_RDWR file here.
5765          *
5766          * Offer a write delegation in the case of a BOTH open, and ensure
5767          * we get the O_RDWR descriptor.
5768          */
5769         if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5770                 nf = find_rw_file(fp);
5771                 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5772         }
5773
5774         /*
5775          * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5776          * file for some reason, then try for a read delegation instead.
5777          */
5778         if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5779                 nf = find_readable_file(fp);
5780                 dl_type = NFS4_OPEN_DELEGATE_READ;
5781         }
5782
5783         if (!nf)
5784                 return ERR_PTR(-EAGAIN);
5785
5786         spin_lock(&state_lock);
5787         spin_lock(&fp->fi_lock);
5788         if (nfs4_delegation_exists(clp, fp))
5789                 status = -EAGAIN;
5790         else if (nfsd4_verify_setuid_write(open, nf))
5791                 status = -EAGAIN;
5792         else if (!fp->fi_deleg_file) {
5793                 fp->fi_deleg_file = nf;
5794                 /* increment early to prevent fi_deleg_file from being
5795                  * cleared */
5796                 fp->fi_delegees = 1;
5797                 nf = NULL;
5798         } else
5799                 fp->fi_delegees++;
5800         spin_unlock(&fp->fi_lock);
5801         spin_unlock(&state_lock);
5802         if (nf)
5803                 nfsd_file_put(nf);
5804         if (status)
5805                 return ERR_PTR(status);
5806
5807         status = -ENOMEM;
5808         dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5809         if (!dp)
5810                 goto out_delegees;
5811
5812         fl = nfs4_alloc_init_lease(dp, dl_type);
5813         if (!fl)
5814                 goto out_clnt_odstate;
5815
5816         status = kernel_setlease(fp->fi_deleg_file->nf_file,
5817                                       fl->c.flc_type, &fl, NULL);
5818         if (fl)
5819                 locks_free_lease(fl);
5820         if (status)
5821                 goto out_clnt_odstate;
5822
5823         if (parent) {
5824                 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5825                 if (status)
5826                         goto out_unlock;
5827         }
5828
5829         status = nfsd4_check_conflicting_opens(clp, fp);
5830         if (status)
5831                 goto out_unlock;
5832
5833         /*
5834          * Now that the deleg is set, check again to ensure that nothing
5835          * raced in and changed the mode while we weren't lookng.
5836          */
5837         status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5838         if (status)
5839                 goto out_unlock;
5840
5841         status = -EAGAIN;
5842         if (fp->fi_had_conflict)
5843                 goto out_unlock;
5844
5845         spin_lock(&state_lock);
5846         spin_lock(&clp->cl_lock);
5847         spin_lock(&fp->fi_lock);
5848         status = hash_delegation_locked(dp, fp);
5849         spin_unlock(&fp->fi_lock);
5850         spin_unlock(&clp->cl_lock);
5851         spin_unlock(&state_lock);
5852
5853         if (status)
5854                 goto out_unlock;
5855
5856         return dp;
5857 out_unlock:
5858         kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5859 out_clnt_odstate:
5860         put_clnt_odstate(dp->dl_clnt_odstate);
5861         nfs4_put_stid(&dp->dl_stid);
5862 out_delegees:
5863         put_deleg_file(fp);
5864         return ERR_PTR(status);
5865 }
5866
5867 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5868 {
5869         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5870         if (status == -EAGAIN)
5871                 open->op_why_no_deleg = WND4_CONTENTION;
5872         else {
5873                 open->op_why_no_deleg = WND4_RESOURCE;
5874                 switch (open->op_deleg_want) {
5875                 case NFS4_SHARE_WANT_READ_DELEG:
5876                 case NFS4_SHARE_WANT_WRITE_DELEG:
5877                 case NFS4_SHARE_WANT_ANY_DELEG:
5878                         break;
5879                 case NFS4_SHARE_WANT_CANCEL:
5880                         open->op_why_no_deleg = WND4_CANCELLED;
5881                         break;
5882                 case NFS4_SHARE_WANT_NO_DELEG:
5883                         WARN_ON_ONCE(1);
5884                 }
5885         }
5886 }
5887
5888 /*
5889  * The Linux NFS server does not offer write delegations to NFSv4.0
5890  * clients in order to avoid conflicts between write delegations and
5891  * GETATTRs requesting CHANGE or SIZE attributes.
5892  *
5893  * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5894  * begins each COMPOUND contains a client ID. Delegation recall can
5895  * be avoided when the server recognizes the client sending a
5896  * GETATTR also holds write delegation it conflicts with.
5897  *
5898  * However, the NFSv4.0 protocol does not enable a server to
5899  * determine that a GETATTR originated from the client holding the
5900  * conflicting delegation versus coming from some other client. Per
5901  * RFC 7530 Section 16.7.5, the server must recall or send a
5902  * CB_GETATTR even when the GETATTR originates from the client that
5903  * holds the conflicting delegation.
5904  *
5905  * An NFSv4.0 client can trigger a pathological situation if it
5906  * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5907  * the same COMPOUND. COMPOUND execution will always stop at the
5908  * GETATTR and the DELEGRETURN will never get executed. The server
5909  * eventually revokes the delegation, which can result in loss of
5910  * open or lock state.
5911  */
5912 static void
5913 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5914                      struct svc_fh *currentfh)
5915 {
5916         struct nfs4_delegation *dp;
5917         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5918         struct nfs4_client *clp = stp->st_stid.sc_client;
5919         struct svc_fh *parent = NULL;
5920         int cb_up;
5921         int status = 0;
5922         struct kstat stat;
5923         struct path path;
5924
5925         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5926         open->op_recall = false;
5927         switch (open->op_claim_type) {
5928                 case NFS4_OPEN_CLAIM_PREVIOUS:
5929                         if (!cb_up)
5930                                 open->op_recall = true;
5931                         break;
5932                 case NFS4_OPEN_CLAIM_NULL:
5933                         parent = currentfh;
5934                         fallthrough;
5935                 case NFS4_OPEN_CLAIM_FH:
5936                         /*
5937                          * Let's not give out any delegations till everyone's
5938                          * had the chance to reclaim theirs, *and* until
5939                          * NLM locks have all been reclaimed:
5940                          */
5941                         if (locks_in_grace(clp->net))
5942                                 goto out_no_deleg;
5943                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5944                                 goto out_no_deleg;
5945                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5946                                         !clp->cl_minorversion)
5947                                 goto out_no_deleg;
5948                         break;
5949                 default:
5950                         goto out_no_deleg;
5951         }
5952         dp = nfs4_set_delegation(open, stp, parent);
5953         if (IS_ERR(dp))
5954                 goto out_no_deleg;
5955
5956         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5957
5958         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5959                 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5960                 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5961                 path.mnt = currentfh->fh_export->ex_path.mnt;
5962                 path.dentry = currentfh->fh_dentry;
5963                 if (vfs_getattr(&path, &stat,
5964                                 (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5965                                 AT_STATX_SYNC_AS_STAT)) {
5966                         nfs4_put_stid(&dp->dl_stid);
5967                         destroy_delegation(dp);
5968                         goto out_no_deleg;
5969                 }
5970                 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
5971                 dp->dl_cb_fattr.ncf_initial_cinfo =
5972                         nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
5973         } else {
5974                 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5975                 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5976         }
5977         nfs4_put_stid(&dp->dl_stid);
5978         return;
5979 out_no_deleg:
5980         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5981         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5982             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5983                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5984                 open->op_recall = true;
5985         }
5986
5987         /* 4.1 client asking for a delegation? */
5988         if (open->op_deleg_want)
5989                 nfsd4_open_deleg_none_ext(open, status);
5990         return;
5991 }
5992
5993 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5994                                         struct nfs4_delegation *dp)
5995 {
5996         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5997             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5998                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5999                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
6000         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
6001                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6002                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6003                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
6004         }
6005         /* Otherwise the client must be confused wanting a delegation
6006          * it already has, therefore we don't return
6007          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
6008          */
6009 }
6010
6011 /**
6012  * nfsd4_process_open2 - finish open processing
6013  * @rqstp: the RPC transaction being executed
6014  * @current_fh: NFSv4 COMPOUND's current filehandle
6015  * @open: OPEN arguments
6016  *
6017  * If successful, (1) truncate the file if open->op_truncate was
6018  * set, (2) set open->op_stateid, (3) set open->op_delegation.
6019  *
6020  * Returns %nfs_ok on success; otherwise an nfs4stat value in
6021  * network byte order is returned.
6022  */
6023 __be32
6024 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6025 {
6026         struct nfsd4_compoundres *resp = rqstp->rq_resp;
6027         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6028         struct nfs4_file *fp = NULL;
6029         struct nfs4_ol_stateid *stp = NULL;
6030         struct nfs4_delegation *dp = NULL;
6031         __be32 status;
6032         bool new_stp = false;
6033
6034         /*
6035          * Lookup file; if found, lookup stateid and check open request,
6036          * and check for delegations in the process of being recalled.
6037          * If not found, create the nfs4_file struct
6038          */
6039         fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6040         if (unlikely(!fp))
6041                 return nfserr_jukebox;
6042         if (fp != open->op_file) {
6043                 status = nfs4_check_deleg(cl, open, &dp);
6044                 if (status)
6045                         goto out;
6046                 stp = nfsd4_find_and_lock_existing_open(fp, open);
6047         } else {
6048                 open->op_file = NULL;
6049                 status = nfserr_bad_stateid;
6050                 if (nfsd4_is_deleg_cur(open))
6051                         goto out;
6052         }
6053
6054         if (!stp) {
6055                 stp = init_open_stateid(fp, open);
6056                 if (!open->op_stp)
6057                         new_stp = true;
6058         }
6059
6060         /*
6061          * OPEN the file, or upgrade an existing OPEN.
6062          * If truncate fails, the OPEN fails.
6063          *
6064          * stp is already locked.
6065          */
6066         if (!new_stp) {
6067                 /* Stateid was found, this is an OPEN upgrade */
6068                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6069                 if (status) {
6070                         mutex_unlock(&stp->st_mutex);
6071                         goto out;
6072                 }
6073         } else {
6074                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6075                 if (status) {
6076                         release_open_stateid(stp);
6077                         mutex_unlock(&stp->st_mutex);
6078                         goto out;
6079                 }
6080
6081                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6082                                                         open->op_odstate);
6083                 if (stp->st_clnt_odstate == open->op_odstate)
6084                         open->op_odstate = NULL;
6085         }
6086
6087         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6088         mutex_unlock(&stp->st_mutex);
6089
6090         if (nfsd4_has_session(&resp->cstate)) {
6091                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6092                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6093                         open->op_why_no_deleg = WND4_NOT_WANTED;
6094                         goto nodeleg;
6095                 }
6096         }
6097
6098         /*
6099         * Attempt to hand out a delegation. No error return, because the
6100         * OPEN succeeds even if we fail.
6101         */
6102         nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6103 nodeleg:
6104         status = nfs_ok;
6105         trace_nfsd_open(&stp->st_stid.sc_stateid);
6106 out:
6107         /* 4.1 client trying to upgrade/downgrade delegation? */
6108         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6109             open->op_deleg_want)
6110                 nfsd4_deleg_xgrade_none_ext(open, dp);
6111
6112         if (fp)
6113                 put_nfs4_file(fp);
6114         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6115                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6116         /*
6117         * To finish the open response, we just need to set the rflags.
6118         */
6119         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6120         if (nfsd4_has_session(&resp->cstate))
6121                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6122         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6123                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6124
6125         if (dp)
6126                 nfs4_put_stid(&dp->dl_stid);
6127         if (stp)
6128                 nfs4_put_stid(&stp->st_stid);
6129
6130         return status;
6131 }
6132
6133 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6134                               struct nfsd4_open *open)
6135 {
6136         if (open->op_openowner) {
6137                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
6138
6139                 nfsd4_cstate_assign_replay(cstate, so);
6140                 nfs4_put_stateowner(so);
6141         }
6142         if (open->op_file)
6143                 kmem_cache_free(file_slab, open->op_file);
6144         if (open->op_stp)
6145                 nfs4_put_stid(&open->op_stp->st_stid);
6146         if (open->op_odstate)
6147                 kmem_cache_free(odstate_slab, open->op_odstate);
6148 }
6149
6150 __be32
6151 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6152             union nfsd4_op_u *u)
6153 {
6154         clientid_t *clid = &u->renew;
6155         struct nfs4_client *clp;
6156         __be32 status;
6157         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6158
6159         trace_nfsd_clid_renew(clid);
6160         status = set_client(clid, cstate, nn);
6161         if (status)
6162                 return status;
6163         clp = cstate->clp;
6164         if (!list_empty(&clp->cl_delegations)
6165                         && clp->cl_cb_state != NFSD4_CB_UP)
6166                 return nfserr_cb_path_down;
6167         return nfs_ok;
6168 }
6169
6170 void
6171 nfsd4_end_grace(struct nfsd_net *nn)
6172 {
6173         /* do nothing if grace period already ended */
6174         if (nn->grace_ended)
6175                 return;
6176
6177         trace_nfsd_grace_complete(nn);
6178         nn->grace_ended = true;
6179         /*
6180          * If the server goes down again right now, an NFSv4
6181          * client will still be allowed to reclaim after it comes back up,
6182          * even if it hasn't yet had a chance to reclaim state this time.
6183          *
6184          */
6185         nfsd4_record_grace_done(nn);
6186         /*
6187          * At this point, NFSv4 clients can still reclaim.  But if the
6188          * server crashes, any that have not yet reclaimed will be out
6189          * of luck on the next boot.
6190          *
6191          * (NFSv4.1+ clients are considered to have reclaimed once they
6192          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
6193          * have reclaimed after their first OPEN.)
6194          */
6195         locks_end_grace(&nn->nfsd4_manager);
6196         /*
6197          * At this point, and once lockd and/or any other containers
6198          * exit their grace period, further reclaims will fail and
6199          * regular locking can resume.
6200          */
6201 }
6202
6203 /*
6204  * If we've waited a lease period but there are still clients trying to
6205  * reclaim, wait a little longer to give them a chance to finish.
6206  */
6207 static bool clients_still_reclaiming(struct nfsd_net *nn)
6208 {
6209         time64_t double_grace_period_end = nn->boot_time +
6210                                            2 * nn->nfsd4_lease;
6211
6212         if (nn->track_reclaim_completes &&
6213                         atomic_read(&nn->nr_reclaim_complete) ==
6214                         nn->reclaim_str_hashtbl_size)
6215                 return false;
6216         if (!nn->somebody_reclaimed)
6217                 return false;
6218         nn->somebody_reclaimed = false;
6219         /*
6220          * If we've given them *two* lease times to reclaim, and they're
6221          * still not done, give up:
6222          */
6223         if (ktime_get_boottime_seconds() > double_grace_period_end)
6224                 return false;
6225         return true;
6226 }
6227
6228 struct laundry_time {
6229         time64_t cutoff;
6230         time64_t new_timeo;
6231 };
6232
6233 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6234 {
6235         time64_t time_remaining;
6236
6237         if (last_refresh < lt->cutoff)
6238                 return true;
6239         time_remaining = last_refresh - lt->cutoff;
6240         lt->new_timeo = min(lt->new_timeo, time_remaining);
6241         return false;
6242 }
6243
6244 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6245 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6246 {
6247         spin_lock_init(&nn->nfsd_ssc_lock);
6248         INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6249         init_waitqueue_head(&nn->nfsd_ssc_waitq);
6250 }
6251 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
6252
6253 /*
6254  * This is called when nfsd is being shutdown, after all inter_ssc
6255  * cleanup were done, to destroy the ssc delayed unmount list.
6256  */
6257 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6258 {
6259         struct nfsd4_ssc_umount_item *ni = NULL;
6260         struct nfsd4_ssc_umount_item *tmp;
6261
6262         spin_lock(&nn->nfsd_ssc_lock);
6263         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6264                 list_del(&ni->nsui_list);
6265                 spin_unlock(&nn->nfsd_ssc_lock);
6266                 mntput(ni->nsui_vfsmount);
6267                 kfree(ni);
6268                 spin_lock(&nn->nfsd_ssc_lock);
6269         }
6270         spin_unlock(&nn->nfsd_ssc_lock);
6271 }
6272
6273 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6274 {
6275         bool do_wakeup = false;
6276         struct nfsd4_ssc_umount_item *ni = NULL;
6277         struct nfsd4_ssc_umount_item *tmp;
6278
6279         spin_lock(&nn->nfsd_ssc_lock);
6280         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6281                 if (time_after(jiffies, ni->nsui_expire)) {
6282                         if (refcount_read(&ni->nsui_refcnt) > 1)
6283                                 continue;
6284
6285                         /* mark being unmount */
6286                         ni->nsui_busy = true;
6287                         spin_unlock(&nn->nfsd_ssc_lock);
6288                         mntput(ni->nsui_vfsmount);
6289                         spin_lock(&nn->nfsd_ssc_lock);
6290
6291                         /* waiters need to start from begin of list */
6292                         list_del(&ni->nsui_list);
6293                         kfree(ni);
6294
6295                         /* wakeup ssc_connect waiters */
6296                         do_wakeup = true;
6297                         continue;
6298                 }
6299                 break;
6300         }
6301         if (do_wakeup)
6302                 wake_up_all(&nn->nfsd_ssc_waitq);
6303         spin_unlock(&nn->nfsd_ssc_lock);
6304 }
6305 #endif
6306
6307 /* Check if any lock belonging to this lockowner has any blockers */
6308 static bool
6309 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6310 {
6311         struct file_lock_context *ctx;
6312         struct nfs4_ol_stateid *stp;
6313         struct nfs4_file *nf;
6314
6315         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6316                 nf = stp->st_stid.sc_file;
6317                 ctx = locks_inode_context(nf->fi_inode);
6318                 if (!ctx)
6319                         continue;
6320                 if (locks_owner_has_blockers(ctx, lo))
6321                         return true;
6322         }
6323         return false;
6324 }
6325
6326 static bool
6327 nfs4_anylock_blockers(struct nfs4_client *clp)
6328 {
6329         int i;
6330         struct nfs4_stateowner *so;
6331         struct nfs4_lockowner *lo;
6332
6333         if (atomic_read(&clp->cl_delegs_in_recall))
6334                 return true;
6335         spin_lock(&clp->cl_lock);
6336         for (i = 0; i < OWNER_HASH_SIZE; i++) {
6337                 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6338                                 so_strhash) {
6339                         if (so->so_is_open_owner)
6340                                 continue;
6341                         lo = lockowner(so);
6342                         if (nfs4_lockowner_has_blockers(lo)) {
6343                                 spin_unlock(&clp->cl_lock);
6344                                 return true;
6345                         }
6346                 }
6347         }
6348         spin_unlock(&clp->cl_lock);
6349         return false;
6350 }
6351
6352 static void
6353 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6354                                 struct laundry_time *lt)
6355 {
6356         unsigned int maxreap, reapcnt = 0;
6357         struct list_head *pos, *next;
6358         struct nfs4_client *clp;
6359
6360         maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6361                         NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6362         INIT_LIST_HEAD(reaplist);
6363         spin_lock(&nn->client_lock);
6364         list_for_each_safe(pos, next, &nn->client_lru) {
6365                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6366                 if (clp->cl_state == NFSD4_EXPIRABLE)
6367                         goto exp_client;
6368                 if (!state_expired(lt, clp->cl_time))
6369                         break;
6370                 if (!atomic_read(&clp->cl_rpc_users)) {
6371                         if (clp->cl_state == NFSD4_ACTIVE)
6372                                 atomic_inc(&nn->nfsd_courtesy_clients);
6373                         clp->cl_state = NFSD4_COURTESY;
6374                 }
6375                 if (!client_has_state(clp))
6376                         goto exp_client;
6377                 if (!nfs4_anylock_blockers(clp))
6378                         if (reapcnt >= maxreap)
6379                                 continue;
6380 exp_client:
6381                 if (!mark_client_expired_locked(clp)) {
6382                         list_add(&clp->cl_lru, reaplist);
6383                         reapcnt++;
6384                 }
6385         }
6386         spin_unlock(&nn->client_lock);
6387 }
6388
6389 static void
6390 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6391                                 struct list_head *reaplist)
6392 {
6393         unsigned int maxreap = 0, reapcnt = 0;
6394         struct list_head *pos, *next;
6395         struct nfs4_client *clp;
6396
6397         maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6398         INIT_LIST_HEAD(reaplist);
6399
6400         spin_lock(&nn->client_lock);
6401         list_for_each_safe(pos, next, &nn->client_lru) {
6402                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6403                 if (clp->cl_state == NFSD4_ACTIVE)
6404                         break;
6405                 if (reapcnt >= maxreap)
6406                         break;
6407                 if (!mark_client_expired_locked(clp)) {
6408                         list_add(&clp->cl_lru, reaplist);
6409                         reapcnt++;
6410                 }
6411         }
6412         spin_unlock(&nn->client_lock);
6413 }
6414
6415 static void
6416 nfs4_process_client_reaplist(struct list_head *reaplist)
6417 {
6418         struct list_head *pos, *next;
6419         struct nfs4_client *clp;
6420
6421         list_for_each_safe(pos, next, reaplist) {
6422                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6423                 trace_nfsd_clid_purged(&clp->cl_clientid);
6424                 list_del_init(&clp->cl_lru);
6425                 expire_client(clp);
6426         }
6427 }
6428
6429 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6430                                       struct laundry_time *lt)
6431 {
6432         struct nfs4_client *clp;
6433
6434         spin_lock(&nn->client_lock);
6435         if (nn->nfs40_last_revoke == 0 ||
6436             nn->nfs40_last_revoke > lt->cutoff) {
6437                 spin_unlock(&nn->client_lock);
6438                 return;
6439         }
6440         nn->nfs40_last_revoke = 0;
6441
6442 retry:
6443         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6444                 unsigned long id, tmp;
6445                 struct nfs4_stid *stid;
6446
6447                 if (atomic_read(&clp->cl_admin_revoked) == 0)
6448                         continue;
6449
6450                 spin_lock(&clp->cl_lock);
6451                 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6452                         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6453                                 refcount_inc(&stid->sc_count);
6454                                 spin_unlock(&nn->client_lock);
6455                                 /* this function drops ->cl_lock */
6456                                 nfsd4_drop_revoked_stid(stid);
6457                                 nfs4_put_stid(stid);
6458                                 spin_lock(&nn->client_lock);
6459                                 goto retry;
6460                         }
6461                 spin_unlock(&clp->cl_lock);
6462         }
6463         spin_unlock(&nn->client_lock);
6464 }
6465
6466 static time64_t
6467 nfs4_laundromat(struct nfsd_net *nn)
6468 {
6469         struct nfs4_openowner *oo;
6470         struct nfs4_delegation *dp;
6471         struct nfs4_ol_stateid *stp;
6472         struct nfsd4_blocked_lock *nbl;
6473         struct list_head *pos, *next, reaplist;
6474         struct laundry_time lt = {
6475                 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6476                 .new_timeo = nn->nfsd4_lease
6477         };
6478         struct nfs4_cpntf_state *cps;
6479         copy_stateid_t *cps_t;
6480         int i;
6481
6482         if (clients_still_reclaiming(nn)) {
6483                 lt.new_timeo = 0;
6484                 goto out;
6485         }
6486         nfsd4_end_grace(nn);
6487
6488         spin_lock(&nn->s2s_cp_lock);
6489         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6490                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6491                 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6492                                 state_expired(&lt, cps->cpntf_time))
6493                         _free_cpntf_state_locked(nn, cps);
6494         }
6495         spin_unlock(&nn->s2s_cp_lock);
6496         nfs4_get_client_reaplist(nn, &reaplist, &lt);
6497         nfs4_process_client_reaplist(&reaplist);
6498
6499         nfs40_clean_admin_revoked(nn, &lt);
6500
6501         spin_lock(&state_lock);
6502         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6503                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6504                 if (!state_expired(&lt, dp->dl_time))
6505                         break;
6506                 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6507                 list_add(&dp->dl_recall_lru, &reaplist);
6508         }
6509         spin_unlock(&state_lock);
6510         while (!list_empty(&reaplist)) {
6511                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6512                                         dl_recall_lru);
6513                 list_del_init(&dp->dl_recall_lru);
6514                 revoke_delegation(dp);
6515         }
6516
6517         spin_lock(&nn->client_lock);
6518         while (!list_empty(&nn->close_lru)) {
6519                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6520                                         oo_close_lru);
6521                 if (!state_expired(&lt, oo->oo_time))
6522                         break;
6523                 list_del_init(&oo->oo_close_lru);
6524                 stp = oo->oo_last_closed_stid;
6525                 oo->oo_last_closed_stid = NULL;
6526                 spin_unlock(&nn->client_lock);
6527                 nfs4_put_stid(&stp->st_stid);
6528                 spin_lock(&nn->client_lock);
6529         }
6530         spin_unlock(&nn->client_lock);
6531
6532         /*
6533          * It's possible for a client to try and acquire an already held lock
6534          * that is being held for a long time, and then lose interest in it.
6535          * So, we clean out any un-revisited request after a lease period
6536          * under the assumption that the client is no longer interested.
6537          *
6538          * RFC5661, sec. 9.6 states that the client must not rely on getting
6539          * notifications and must continue to poll for locks, even when the
6540          * server supports them. Thus this shouldn't lead to clients blocking
6541          * indefinitely once the lock does become free.
6542          */
6543         BUG_ON(!list_empty(&reaplist));
6544         spin_lock(&nn->blocked_locks_lock);
6545         while (!list_empty(&nn->blocked_locks_lru)) {
6546                 nbl = list_first_entry(&nn->blocked_locks_lru,
6547                                         struct nfsd4_blocked_lock, nbl_lru);
6548                 if (!state_expired(&lt, nbl->nbl_time))
6549                         break;
6550                 list_move(&nbl->nbl_lru, &reaplist);
6551                 list_del_init(&nbl->nbl_list);
6552         }
6553         spin_unlock(&nn->blocked_locks_lock);
6554
6555         while (!list_empty(&reaplist)) {
6556                 nbl = list_first_entry(&reaplist,
6557                                         struct nfsd4_blocked_lock, nbl_lru);
6558                 list_del_init(&nbl->nbl_lru);
6559                 free_blocked_lock(nbl);
6560         }
6561 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6562         /* service the server-to-server copy delayed unmount list */
6563         nfsd4_ssc_expire_umount(nn);
6564 #endif
6565         if (atomic_long_read(&num_delegations) >= max_delegations)
6566                 deleg_reaper(nn);
6567 out:
6568         return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6569 }
6570
6571 static void laundromat_main(struct work_struct *);
6572
6573 static void
6574 laundromat_main(struct work_struct *laundry)
6575 {
6576         time64_t t;
6577         struct delayed_work *dwork = to_delayed_work(laundry);
6578         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6579                                            laundromat_work);
6580
6581         t = nfs4_laundromat(nn);
6582         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6583 }
6584
6585 static void
6586 courtesy_client_reaper(struct nfsd_net *nn)
6587 {
6588         struct list_head reaplist;
6589
6590         nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6591         nfs4_process_client_reaplist(&reaplist);
6592 }
6593
6594 static void
6595 deleg_reaper(struct nfsd_net *nn)
6596 {
6597         struct list_head *pos, *next;
6598         struct nfs4_client *clp;
6599         struct list_head cblist;
6600
6601         INIT_LIST_HEAD(&cblist);
6602         spin_lock(&nn->client_lock);
6603         list_for_each_safe(pos, next, &nn->client_lru) {
6604                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6605                 if (clp->cl_state != NFSD4_ACTIVE ||
6606                         list_empty(&clp->cl_delegations) ||
6607                         atomic_read(&clp->cl_delegs_in_recall) ||
6608                         test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6609                         (ktime_get_boottime_seconds() -
6610                                 clp->cl_ra_time < 5)) {
6611                         continue;
6612                 }
6613                 list_add(&clp->cl_ra_cblist, &cblist);
6614
6615                 /* release in nfsd4_cb_recall_any_release */
6616                 kref_get(&clp->cl_nfsdfs.cl_ref);
6617                 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6618                 clp->cl_ra_time = ktime_get_boottime_seconds();
6619         }
6620         spin_unlock(&nn->client_lock);
6621
6622         while (!list_empty(&cblist)) {
6623                 clp = list_first_entry(&cblist, struct nfs4_client,
6624                                         cl_ra_cblist);
6625                 list_del_init(&clp->cl_ra_cblist);
6626                 clp->cl_ra->ra_keep = 0;
6627                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6628                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6629                                                 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6630                 trace_nfsd_cb_recall_any(clp->cl_ra);
6631                 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6632         }
6633 }
6634
6635 static void
6636 nfsd4_state_shrinker_worker(struct work_struct *work)
6637 {
6638         struct nfsd_net *nn = container_of(work, struct nfsd_net,
6639                                 nfsd_shrinker_work);
6640
6641         courtesy_client_reaper(nn);
6642         deleg_reaper(nn);
6643 }
6644
6645 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6646 {
6647         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6648                 return nfserr_bad_stateid;
6649         return nfs_ok;
6650 }
6651
6652 static
6653 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6654 {
6655         __be32 status = nfserr_openmode;
6656
6657         /* For lock stateid's, we test the parent open, not the lock: */
6658         if (stp->st_openstp)
6659                 stp = stp->st_openstp;
6660         if ((flags & WR_STATE) && !access_permit_write(stp))
6661                 goto out;
6662         if ((flags & RD_STATE) && !access_permit_read(stp))
6663                 goto out;
6664         status = nfs_ok;
6665 out:
6666         return status;
6667 }
6668
6669 static inline __be32
6670 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6671 {
6672         if (ONE_STATEID(stateid) && (flags & RD_STATE))
6673                 return nfs_ok;
6674         else if (opens_in_grace(net)) {
6675                 /* Answer in remaining cases depends on existence of
6676                  * conflicting state; so we must wait out the grace period. */
6677                 return nfserr_grace;
6678         } else if (flags & WR_STATE)
6679                 return nfs4_share_conflict(current_fh,
6680                                 NFS4_SHARE_DENY_WRITE);
6681         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6682                 return nfs4_share_conflict(current_fh,
6683                                 NFS4_SHARE_DENY_READ);
6684 }
6685
6686 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6687 {
6688         /*
6689          * When sessions are used the stateid generation number is ignored
6690          * when it is zero.
6691          */
6692         if (has_session && in->si_generation == 0)
6693                 return nfs_ok;
6694
6695         if (in->si_generation == ref->si_generation)
6696                 return nfs_ok;
6697
6698         /* If the client sends us a stateid from the future, it's buggy: */
6699         if (nfsd4_stateid_generation_after(in, ref))
6700                 return nfserr_bad_stateid;
6701         /*
6702          * However, we could see a stateid from the past, even from a
6703          * non-buggy client.  For example, if the client sends a lock
6704          * while some IO is outstanding, the lock may bump si_generation
6705          * while the IO is still in flight.  The client could avoid that
6706          * situation by waiting for responses on all the IO requests,
6707          * but better performance may result in retrying IO that
6708          * receives an old_stateid error if requests are rarely
6709          * reordered in flight:
6710          */
6711         return nfserr_old_stateid;
6712 }
6713
6714 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6715 {
6716         __be32 ret;
6717
6718         spin_lock(&s->sc_lock);
6719         ret = nfsd4_verify_open_stid(s);
6720         if (ret == nfs_ok)
6721                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6722         spin_unlock(&s->sc_lock);
6723         if (ret == nfserr_admin_revoked)
6724                 nfsd40_drop_revoked_stid(s->sc_client,
6725                                         &s->sc_stateid);
6726         return ret;
6727 }
6728
6729 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6730 {
6731         if (ols->st_stateowner->so_is_open_owner &&
6732             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6733                 return nfserr_bad_stateid;
6734         return nfs_ok;
6735 }
6736
6737 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6738 {
6739         struct nfs4_stid *s;
6740         __be32 status = nfserr_bad_stateid;
6741
6742         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6743                 CLOSE_STATEID(stateid))
6744                 return status;
6745         spin_lock(&cl->cl_lock);
6746         s = find_stateid_locked(cl, stateid);
6747         if (!s)
6748                 goto out_unlock;
6749         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6750         if (status)
6751                 goto out_unlock;
6752         status = nfsd4_verify_open_stid(s);
6753         if (status)
6754                 goto out_unlock;
6755
6756         switch (s->sc_type) {
6757         case SC_TYPE_DELEG:
6758                 status = nfs_ok;
6759                 break;
6760         case SC_TYPE_OPEN:
6761         case SC_TYPE_LOCK:
6762                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6763                 break;
6764         default:
6765                 printk("unknown stateid type %x\n", s->sc_type);
6766                 status = nfserr_bad_stateid;
6767         }
6768 out_unlock:
6769         spin_unlock(&cl->cl_lock);
6770         if (status == nfserr_admin_revoked)
6771                 nfsd40_drop_revoked_stid(cl, stateid);
6772         return status;
6773 }
6774
6775 __be32
6776 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6777                      stateid_t *stateid,
6778                      unsigned short typemask, unsigned short statusmask,
6779                      struct nfs4_stid **s, struct nfsd_net *nn)
6780 {
6781         __be32 status;
6782         struct nfs4_stid *stid;
6783         bool return_revoked = false;
6784
6785         /*
6786          *  only return revoked delegations if explicitly asked.
6787          *  otherwise we report revoked or bad_stateid status.
6788          */
6789         if (statusmask & SC_STATUS_REVOKED)
6790                 return_revoked = true;
6791         if (typemask & SC_TYPE_DELEG)
6792                 /* Always allow REVOKED for DELEG so we can
6793                  * retturn the appropriate error.
6794                  */
6795                 statusmask |= SC_STATUS_REVOKED;
6796
6797         statusmask |= SC_STATUS_ADMIN_REVOKED;
6798
6799         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6800                 CLOSE_STATEID(stateid))
6801                 return nfserr_bad_stateid;
6802         status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6803         if (status == nfserr_stale_clientid) {
6804                 if (cstate->session)
6805                         return nfserr_bad_stateid;
6806                 return nfserr_stale_stateid;
6807         }
6808         if (status)
6809                 return status;
6810         stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6811         if (!stid)
6812                 return nfserr_bad_stateid;
6813         if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6814                 nfs4_put_stid(stid);
6815                 return nfserr_deleg_revoked;
6816         }
6817         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6818                 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6819                 nfs4_put_stid(stid);
6820                 return nfserr_admin_revoked;
6821         }
6822         *s = stid;
6823         return nfs_ok;
6824 }
6825
6826 static struct nfsd_file *
6827 nfs4_find_file(struct nfs4_stid *s, int flags)
6828 {
6829         struct nfsd_file *ret = NULL;
6830
6831         if (!s || s->sc_status)
6832                 return NULL;
6833
6834         switch (s->sc_type) {
6835         case SC_TYPE_DELEG:
6836                 spin_lock(&s->sc_file->fi_lock);
6837                 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6838                 spin_unlock(&s->sc_file->fi_lock);
6839                 break;
6840         case SC_TYPE_OPEN:
6841         case SC_TYPE_LOCK:
6842                 if (flags & RD_STATE)
6843                         ret = find_readable_file(s->sc_file);
6844                 else
6845                         ret = find_writeable_file(s->sc_file);
6846         }
6847
6848         return ret;
6849 }
6850
6851 static __be32
6852 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6853 {
6854         __be32 status;
6855
6856         status = nfsd4_check_openowner_confirmed(ols);
6857         if (status)
6858                 return status;
6859         return nfs4_check_openmode(ols, flags);
6860 }
6861
6862 static __be32
6863 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6864                 struct nfsd_file **nfp, int flags)
6865 {
6866         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6867         struct nfsd_file *nf;
6868         __be32 status;
6869
6870         nf = nfs4_find_file(s, flags);
6871         if (nf) {
6872                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6873                                 acc | NFSD_MAY_OWNER_OVERRIDE);
6874                 if (status) {
6875                         nfsd_file_put(nf);
6876                         goto out;
6877                 }
6878         } else {
6879                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6880                 if (status)
6881                         return status;
6882         }
6883         *nfp = nf;
6884 out:
6885         return status;
6886 }
6887 static void
6888 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6889 {
6890         WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6891         if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6892                 return;
6893         list_del(&cps->cp_list);
6894         idr_remove(&nn->s2s_cp_stateids,
6895                    cps->cp_stateid.cs_stid.si_opaque.so_id);
6896         kfree(cps);
6897 }
6898 /*
6899  * A READ from an inter server to server COPY will have a
6900  * copy stateid. Look up the copy notify stateid from the
6901  * idr structure and take a reference on it.
6902  */
6903 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6904                           struct nfs4_client *clp,
6905                           struct nfs4_cpntf_state **cps)
6906 {
6907         copy_stateid_t *cps_t;
6908         struct nfs4_cpntf_state *state = NULL;
6909
6910         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6911                 return nfserr_bad_stateid;
6912         spin_lock(&nn->s2s_cp_lock);
6913         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6914         if (cps_t) {
6915                 state = container_of(cps_t, struct nfs4_cpntf_state,
6916                                      cp_stateid);
6917                 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6918                         state = NULL;
6919                         goto unlock;
6920                 }
6921                 if (!clp)
6922                         refcount_inc(&state->cp_stateid.cs_count);
6923                 else
6924                         _free_cpntf_state_locked(nn, state);
6925         }
6926 unlock:
6927         spin_unlock(&nn->s2s_cp_lock);
6928         if (!state)
6929                 return nfserr_bad_stateid;
6930         if (!clp)
6931                 *cps = state;
6932         return 0;
6933 }
6934
6935 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6936                                struct nfs4_stid **stid)
6937 {
6938         __be32 status;
6939         struct nfs4_cpntf_state *cps = NULL;
6940         struct nfs4_client *found;
6941
6942         status = manage_cpntf_state(nn, st, NULL, &cps);
6943         if (status)
6944                 return status;
6945
6946         cps->cpntf_time = ktime_get_boottime_seconds();
6947
6948         status = nfserr_expired;
6949         found = lookup_clientid(&cps->cp_p_clid, true, nn);
6950         if (!found)
6951                 goto out;
6952
6953         *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6954                                      SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6955                                      0);
6956         if (*stid)
6957                 status = nfs_ok;
6958         else
6959                 status = nfserr_bad_stateid;
6960
6961         put_client_renew(found);
6962 out:
6963         nfs4_put_cpntf_state(nn, cps);
6964         return status;
6965 }
6966
6967 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6968 {
6969         spin_lock(&nn->s2s_cp_lock);
6970         _free_cpntf_state_locked(nn, cps);
6971         spin_unlock(&nn->s2s_cp_lock);
6972 }
6973
6974 /**
6975  * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6976  * @rqstp: incoming request from client
6977  * @cstate: current compound state
6978  * @fhp: filehandle associated with requested stateid
6979  * @stateid: stateid (provided by client)
6980  * @flags: flags describing type of operation to be done
6981  * @nfp: optional nfsd_file return pointer (may be NULL)
6982  * @cstid: optional returned nfs4_stid pointer (may be NULL)
6983  *
6984  * Given info from the client, look up a nfs4_stid for the operation. On
6985  * success, it returns a reference to the nfs4_stid and/or the nfsd_file
6986  * associated with it.
6987  */
6988 __be32
6989 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6990                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6991                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6992                 struct nfs4_stid **cstid)
6993 {
6994         struct net *net = SVC_NET(rqstp);
6995         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6996         struct nfs4_stid *s = NULL;
6997         __be32 status;
6998
6999         if (nfp)
7000                 *nfp = NULL;
7001
7002         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
7003                 if (cstid)
7004                         status = nfserr_bad_stateid;
7005                 else
7006                         status = check_special_stateids(net, fhp, stateid,
7007                                                                         flags);
7008                 goto done;
7009         }
7010
7011         status = nfsd4_lookup_stateid(cstate, stateid,
7012                                 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7013                                 0, &s, nn);
7014         if (status == nfserr_bad_stateid)
7015                 status = find_cpntf_state(nn, stateid, &s);
7016         if (status)
7017                 return status;
7018         status = nfsd4_stid_check_stateid_generation(stateid, s,
7019                         nfsd4_has_session(cstate));
7020         if (status)
7021                 goto out;
7022
7023         switch (s->sc_type) {
7024         case SC_TYPE_DELEG:
7025                 status = nfs4_check_delegmode(delegstateid(s), flags);
7026                 break;
7027         case SC_TYPE_OPEN:
7028         case SC_TYPE_LOCK:
7029                 status = nfs4_check_olstateid(openlockstateid(s), flags);
7030                 break;
7031         }
7032         if (status)
7033                 goto out;
7034         status = nfs4_check_fh(fhp, s);
7035
7036 done:
7037         if (status == nfs_ok && nfp)
7038                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7039 out:
7040         if (s) {
7041                 if (!status && cstid)
7042                         *cstid = s;
7043                 else
7044                         nfs4_put_stid(s);
7045         }
7046         return status;
7047 }
7048
7049 /*
7050  * Test if the stateid is valid
7051  */
7052 __be32
7053 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7054                    union nfsd4_op_u *u)
7055 {
7056         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7057         struct nfsd4_test_stateid_id *stateid;
7058         struct nfs4_client *cl = cstate->clp;
7059
7060         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7061                 stateid->ts_id_status =
7062                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7063
7064         return nfs_ok;
7065 }
7066
7067 static __be32
7068 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7069 {
7070         struct nfs4_ol_stateid *stp = openlockstateid(s);
7071         __be32 ret;
7072
7073         ret = nfsd4_lock_ol_stateid(stp);
7074         if (ret)
7075                 goto out_put_stid;
7076
7077         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7078         if (ret)
7079                 goto out;
7080
7081         ret = nfserr_locks_held;
7082         if (check_for_locks(stp->st_stid.sc_file,
7083                             lockowner(stp->st_stateowner)))
7084                 goto out;
7085
7086         release_lock_stateid(stp);
7087         ret = nfs_ok;
7088
7089 out:
7090         mutex_unlock(&stp->st_mutex);
7091 out_put_stid:
7092         nfs4_put_stid(s);
7093         return ret;
7094 }
7095
7096 __be32
7097 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7098                    union nfsd4_op_u *u)
7099 {
7100         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7101         stateid_t *stateid = &free_stateid->fr_stateid;
7102         struct nfs4_stid *s;
7103         struct nfs4_delegation *dp;
7104         struct nfs4_client *cl = cstate->clp;
7105         __be32 ret = nfserr_bad_stateid;
7106
7107         spin_lock(&cl->cl_lock);
7108         s = find_stateid_locked(cl, stateid);
7109         if (!s || s->sc_status & SC_STATUS_CLOSED)
7110                 goto out_unlock;
7111         if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7112                 nfsd4_drop_revoked_stid(s);
7113                 ret = nfs_ok;
7114                 goto out;
7115         }
7116         spin_lock(&s->sc_lock);
7117         switch (s->sc_type) {
7118         case SC_TYPE_DELEG:
7119                 if (s->sc_status & SC_STATUS_REVOKED) {
7120                         spin_unlock(&s->sc_lock);
7121                         dp = delegstateid(s);
7122                         list_del_init(&dp->dl_recall_lru);
7123                         spin_unlock(&cl->cl_lock);
7124                         nfs4_put_stid(s);
7125                         ret = nfs_ok;
7126                         goto out;
7127                 }
7128                 ret = nfserr_locks_held;
7129                 break;
7130         case SC_TYPE_OPEN:
7131                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7132                 if (ret)
7133                         break;
7134                 ret = nfserr_locks_held;
7135                 break;
7136         case SC_TYPE_LOCK:
7137                 spin_unlock(&s->sc_lock);
7138                 refcount_inc(&s->sc_count);
7139                 spin_unlock(&cl->cl_lock);
7140                 ret = nfsd4_free_lock_stateid(stateid, s);
7141                 goto out;
7142         }
7143         spin_unlock(&s->sc_lock);
7144 out_unlock:
7145         spin_unlock(&cl->cl_lock);
7146 out:
7147         return ret;
7148 }
7149
7150 static inline int
7151 setlkflg (int type)
7152 {
7153         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7154                 RD_STATE : WR_STATE;
7155 }
7156
7157 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7158 {
7159         struct svc_fh *current_fh = &cstate->current_fh;
7160         struct nfs4_stateowner *sop = stp->st_stateowner;
7161         __be32 status;
7162
7163         status = nfsd4_check_seqid(cstate, sop, seqid);
7164         if (status)
7165                 return status;
7166         status = nfsd4_lock_ol_stateid(stp);
7167         if (status != nfs_ok)
7168                 return status;
7169         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7170         if (status == nfs_ok)
7171                 status = nfs4_check_fh(current_fh, &stp->st_stid);
7172         if (status != nfs_ok)
7173                 mutex_unlock(&stp->st_mutex);
7174         return status;
7175 }
7176
7177 /**
7178  * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7179  * @cstate: compund state
7180  * @seqid: seqid (provided by client)
7181  * @stateid: stateid (provided by client)
7182  * @typemask: mask of allowable types for this operation
7183  * @statusmask: mask of allowed states: 0 or STID_CLOSED
7184  * @stpp: return pointer for the stateid found
7185  * @nn: net namespace for request
7186  *
7187  * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7188  * return it in @stpp. On a nfs_ok return, the returned stateid will
7189  * have its st_mutex locked.
7190  */
7191 static __be32
7192 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7193                          stateid_t *stateid,
7194                          unsigned short typemask, unsigned short statusmask,
7195                          struct nfs4_ol_stateid **stpp,
7196                          struct nfsd_net *nn)
7197 {
7198         __be32 status;
7199         struct nfs4_stid *s;
7200         struct nfs4_ol_stateid *stp = NULL;
7201
7202         trace_nfsd_preprocess(seqid, stateid);
7203
7204         *stpp = NULL;
7205         status = nfsd4_lookup_stateid(cstate, stateid,
7206                                       typemask, statusmask, &s, nn);
7207         if (status)
7208                 return status;
7209         stp = openlockstateid(s);
7210         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
7211
7212         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7213         if (!status)
7214                 *stpp = stp;
7215         else
7216                 nfs4_put_stid(&stp->st_stid);
7217         return status;
7218 }
7219
7220 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7221                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7222 {
7223         __be32 status;
7224         struct nfs4_openowner *oo;
7225         struct nfs4_ol_stateid *stp;
7226
7227         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7228                                           SC_TYPE_OPEN, 0, &stp, nn);
7229         if (status)
7230                 return status;
7231         oo = openowner(stp->st_stateowner);
7232         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7233                 mutex_unlock(&stp->st_mutex);
7234                 nfs4_put_stid(&stp->st_stid);
7235                 return nfserr_bad_stateid;
7236         }
7237         *stpp = stp;
7238         return nfs_ok;
7239 }
7240
7241 __be32
7242 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7243                    union nfsd4_op_u *u)
7244 {
7245         struct nfsd4_open_confirm *oc = &u->open_confirm;
7246         __be32 status;
7247         struct nfs4_openowner *oo;
7248         struct nfs4_ol_stateid *stp;
7249         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7250
7251         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7252                         cstate->current_fh.fh_dentry);
7253
7254         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7255         if (status)
7256                 return status;
7257
7258         status = nfs4_preprocess_seqid_op(cstate,
7259                                           oc->oc_seqid, &oc->oc_req_stateid,
7260                                           SC_TYPE_OPEN, 0, &stp, nn);
7261         if (status)
7262                 goto out;
7263         oo = openowner(stp->st_stateowner);
7264         status = nfserr_bad_stateid;
7265         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7266                 mutex_unlock(&stp->st_mutex);
7267                 goto put_stateid;
7268         }
7269         oo->oo_flags |= NFS4_OO_CONFIRMED;
7270         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7271         mutex_unlock(&stp->st_mutex);
7272         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7273         nfsd4_client_record_create(oo->oo_owner.so_client);
7274         status = nfs_ok;
7275 put_stateid:
7276         nfs4_put_stid(&stp->st_stid);
7277 out:
7278         nfsd4_bump_seqid(cstate, status);
7279         return status;
7280 }
7281
7282 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7283 {
7284         if (!test_access(access, stp))
7285                 return;
7286         nfs4_file_put_access(stp->st_stid.sc_file, access);
7287         clear_access(access, stp);
7288 }
7289
7290 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7291 {
7292         switch (to_access) {
7293         case NFS4_SHARE_ACCESS_READ:
7294                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7295                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7296                 break;
7297         case NFS4_SHARE_ACCESS_WRITE:
7298                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7299                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7300                 break;
7301         case NFS4_SHARE_ACCESS_BOTH:
7302                 break;
7303         default:
7304                 WARN_ON_ONCE(1);
7305         }
7306 }
7307
7308 __be32
7309 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7310                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7311 {
7312         struct nfsd4_open_downgrade *od = &u->open_downgrade;
7313         __be32 status;
7314         struct nfs4_ol_stateid *stp;
7315         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7316
7317         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7318                         cstate->current_fh.fh_dentry);
7319
7320         /* We don't yet support WANT bits: */
7321         if (od->od_deleg_want)
7322                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7323                         od->od_deleg_want);
7324
7325         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7326                                         &od->od_stateid, &stp, nn);
7327         if (status)
7328                 goto out;
7329         status = nfserr_inval;
7330         if (!test_access(od->od_share_access, stp)) {
7331                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7332                         stp->st_access_bmap, od->od_share_access);
7333                 goto put_stateid;
7334         }
7335         if (!test_deny(od->od_share_deny, stp)) {
7336                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7337                         stp->st_deny_bmap, od->od_share_deny);
7338                 goto put_stateid;
7339         }
7340         nfs4_stateid_downgrade(stp, od->od_share_access);
7341         reset_union_bmap_deny(od->od_share_deny, stp);
7342         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7343         status = nfs_ok;
7344 put_stateid:
7345         mutex_unlock(&stp->st_mutex);
7346         nfs4_put_stid(&stp->st_stid);
7347 out:
7348         nfsd4_bump_seqid(cstate, status);
7349         return status;
7350 }
7351
7352 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7353 {
7354         struct nfs4_client *clp = s->st_stid.sc_client;
7355         bool unhashed;
7356         LIST_HEAD(reaplist);
7357         struct nfs4_ol_stateid *stp;
7358
7359         spin_lock(&clp->cl_lock);
7360         unhashed = unhash_open_stateid(s, &reaplist);
7361
7362         if (clp->cl_minorversion) {
7363                 if (unhashed)
7364                         put_ol_stateid_locked(s, &reaplist);
7365                 spin_unlock(&clp->cl_lock);
7366                 list_for_each_entry(stp, &reaplist, st_locks)
7367                         nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7368                 free_ol_stateid_reaplist(&reaplist);
7369         } else {
7370                 spin_unlock(&clp->cl_lock);
7371                 free_ol_stateid_reaplist(&reaplist);
7372                 if (unhashed)
7373                         move_to_close_lru(s, clp->net);
7374         }
7375 }
7376
7377 /*
7378  * nfs4_unlock_state() called after encode
7379  */
7380 __be32
7381 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7382                 union nfsd4_op_u *u)
7383 {
7384         struct nfsd4_close *close = &u->close;
7385         __be32 status;
7386         struct nfs4_ol_stateid *stp;
7387         struct net *net = SVC_NET(rqstp);
7388         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7389
7390         dprintk("NFSD: nfsd4_close on file %pd\n",
7391                         cstate->current_fh.fh_dentry);
7392
7393         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7394                                           &close->cl_stateid,
7395                                           SC_TYPE_OPEN, SC_STATUS_CLOSED,
7396                                           &stp, nn);
7397         nfsd4_bump_seqid(cstate, status);
7398         if (status)
7399                 goto out;
7400
7401         spin_lock(&stp->st_stid.sc_client->cl_lock);
7402         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7403         spin_unlock(&stp->st_stid.sc_client->cl_lock);
7404
7405         /*
7406          * Technically we don't _really_ have to increment or copy it, since
7407          * it should just be gone after this operation and we clobber the
7408          * copied value below, but we continue to do so here just to ensure
7409          * that racing ops see that there was a state change.
7410          */
7411         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7412
7413         nfsd4_close_open_stateid(stp);
7414         mutex_unlock(&stp->st_mutex);
7415
7416         /* v4.1+ suggests that we send a special stateid in here, since the
7417          * clients should just ignore this anyway. Since this is not useful
7418          * for v4.0 clients either, we set it to the special close_stateid
7419          * universally.
7420          *
7421          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7422          */
7423         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7424
7425         /* put reference from nfs4_preprocess_seqid_op */
7426         nfs4_put_stid(&stp->st_stid);
7427 out:
7428         return status;
7429 }
7430
7431 __be32
7432 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7433                   union nfsd4_op_u *u)
7434 {
7435         struct nfsd4_delegreturn *dr = &u->delegreturn;
7436         struct nfs4_delegation *dp;
7437         stateid_t *stateid = &dr->dr_stateid;
7438         struct nfs4_stid *s;
7439         __be32 status;
7440         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7441
7442         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7443                 return status;
7444
7445         status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7446         if (status)
7447                 goto out;
7448         dp = delegstateid(s);
7449         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7450         if (status)
7451                 goto put_stateid;
7452
7453         trace_nfsd_deleg_return(stateid);
7454         wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7455         destroy_delegation(dp);
7456 put_stateid:
7457         nfs4_put_stid(&dp->dl_stid);
7458 out:
7459         return status;
7460 }
7461
7462 /* last octet in a range */
7463 static inline u64
7464 last_byte_offset(u64 start, u64 len)
7465 {
7466         u64 end;
7467
7468         WARN_ON_ONCE(!len);
7469         end = start + len;
7470         return end > start ? end - 1: NFS4_MAX_UINT64;
7471 }
7472
7473 /*
7474  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7475  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7476  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
7477  * locking, this prevents us from being completely protocol-compliant.  The
7478  * real solution to this problem is to start using unsigned file offsets in
7479  * the VFS, but this is a very deep change!
7480  */
7481 static inline void
7482 nfs4_transform_lock_offset(struct file_lock *lock)
7483 {
7484         if (lock->fl_start < 0)
7485                 lock->fl_start = OFFSET_MAX;
7486         if (lock->fl_end < 0)
7487                 lock->fl_end = OFFSET_MAX;
7488 }
7489
7490 static fl_owner_t
7491 nfsd4_lm_get_owner(fl_owner_t owner)
7492 {
7493         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7494
7495         nfs4_get_stateowner(&lo->lo_owner);
7496         return owner;
7497 }
7498
7499 static void
7500 nfsd4_lm_put_owner(fl_owner_t owner)
7501 {
7502         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7503
7504         if (lo)
7505                 nfs4_put_stateowner(&lo->lo_owner);
7506 }
7507
7508 /* return pointer to struct nfs4_client if client is expirable */
7509 static bool
7510 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7511 {
7512         struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7513         struct nfs4_client *clp = lo->lo_owner.so_client;
7514         struct nfsd_net *nn;
7515
7516         if (try_to_expire_client(clp)) {
7517                 nn = net_generic(clp->net, nfsd_net_id);
7518                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7519                 return true;
7520         }
7521         return false;
7522 }
7523
7524 /* schedule laundromat to run immediately and wait for it to complete */
7525 static void
7526 nfsd4_lm_expire_lock(void)
7527 {
7528         flush_workqueue(laundry_wq);
7529 }
7530
7531 static void
7532 nfsd4_lm_notify(struct file_lock *fl)
7533 {
7534         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7535         struct net                      *net = lo->lo_owner.so_client->net;
7536         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
7537         struct nfsd4_blocked_lock       *nbl = container_of(fl,
7538                                                 struct nfsd4_blocked_lock, nbl_lock);
7539         bool queue = false;
7540
7541         /* An empty list means that something else is going to be using it */
7542         spin_lock(&nn->blocked_locks_lock);
7543         if (!list_empty(&nbl->nbl_list)) {
7544                 list_del_init(&nbl->nbl_list);
7545                 list_del_init(&nbl->nbl_lru);
7546                 queue = true;
7547         }
7548         spin_unlock(&nn->blocked_locks_lock);
7549
7550         if (queue) {
7551                 trace_nfsd_cb_notify_lock(lo, nbl);
7552                 nfsd4_run_cb(&nbl->nbl_cb);
7553         }
7554 }
7555
7556 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
7557         .lm_mod_owner = THIS_MODULE,
7558         .lm_notify = nfsd4_lm_notify,
7559         .lm_get_owner = nfsd4_lm_get_owner,
7560         .lm_put_owner = nfsd4_lm_put_owner,
7561         .lm_lock_expirable = nfsd4_lm_lock_expirable,
7562         .lm_expire_lock = nfsd4_lm_expire_lock,
7563 };
7564
7565 static inline void
7566 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7567 {
7568         struct nfs4_lockowner *lo;
7569
7570         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7571                 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7572                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7573                                                 GFP_KERNEL);
7574                 if (!deny->ld_owner.data)
7575                         /* We just don't care that much */
7576                         goto nevermind;
7577                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7578         } else {
7579 nevermind:
7580                 deny->ld_owner.len = 0;
7581                 deny->ld_owner.data = NULL;
7582                 deny->ld_clientid.cl_boot = 0;
7583                 deny->ld_clientid.cl_id = 0;
7584         }
7585         deny->ld_start = fl->fl_start;
7586         deny->ld_length = NFS4_MAX_UINT64;
7587         if (fl->fl_end != NFS4_MAX_UINT64)
7588                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7589         deny->ld_type = NFS4_READ_LT;
7590         if (fl->c.flc_type != F_RDLCK)
7591                 deny->ld_type = NFS4_WRITE_LT;
7592 }
7593
7594 static struct nfs4_lockowner *
7595 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7596 {
7597         unsigned int strhashval = ownerstr_hashval(owner);
7598         struct nfs4_stateowner *so;
7599
7600         lockdep_assert_held(&clp->cl_lock);
7601
7602         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7603                             so_strhash) {
7604                 if (so->so_is_open_owner)
7605                         continue;
7606                 if (same_owner_str(so, owner))
7607                         return lockowner(nfs4_get_stateowner(so));
7608         }
7609         return NULL;
7610 }
7611
7612 static struct nfs4_lockowner *
7613 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7614 {
7615         struct nfs4_lockowner *lo;
7616
7617         spin_lock(&clp->cl_lock);
7618         lo = find_lockowner_str_locked(clp, owner);
7619         spin_unlock(&clp->cl_lock);
7620         return lo;
7621 }
7622
7623 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7624 {
7625         unhash_lockowner_locked(lockowner(sop));
7626 }
7627
7628 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7629 {
7630         struct nfs4_lockowner *lo = lockowner(sop);
7631
7632         kmem_cache_free(lockowner_slab, lo);
7633 }
7634
7635 static const struct nfs4_stateowner_operations lockowner_ops = {
7636         .so_unhash =    nfs4_unhash_lockowner,
7637         .so_free =      nfs4_free_lockowner,
7638 };
7639
7640 /*
7641  * Alloc a lock owner structure.
7642  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7643  * occurred.
7644  *
7645  * strhashval = ownerstr_hashval
7646  */
7647 static struct nfs4_lockowner *
7648 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7649                            struct nfs4_ol_stateid *open_stp,
7650                            struct nfsd4_lock *lock)
7651 {
7652         struct nfs4_lockowner *lo, *ret;
7653
7654         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7655         if (!lo)
7656                 return NULL;
7657         INIT_LIST_HEAD(&lo->lo_blocked);
7658         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7659         lo->lo_owner.so_is_open_owner = 0;
7660         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7661         lo->lo_owner.so_ops = &lockowner_ops;
7662         spin_lock(&clp->cl_lock);
7663         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7664         if (ret == NULL) {
7665                 list_add(&lo->lo_owner.so_strhash,
7666                          &clp->cl_ownerstr_hashtbl[strhashval]);
7667                 ret = lo;
7668         } else
7669                 nfs4_free_stateowner(&lo->lo_owner);
7670
7671         spin_unlock(&clp->cl_lock);
7672         return ret;
7673 }
7674
7675 static struct nfs4_ol_stateid *
7676 find_lock_stateid(const struct nfs4_lockowner *lo,
7677                   const struct nfs4_ol_stateid *ost)
7678 {
7679         struct nfs4_ol_stateid *lst;
7680
7681         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7682
7683         /* If ost is not hashed, ost->st_locks will not be valid */
7684         if (!nfs4_ol_stateid_unhashed(ost))
7685                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7686                         if (lst->st_stateowner == &lo->lo_owner) {
7687                                 refcount_inc(&lst->st_stid.sc_count);
7688                                 return lst;
7689                         }
7690                 }
7691         return NULL;
7692 }
7693
7694 static struct nfs4_ol_stateid *
7695 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7696                   struct nfs4_file *fp, struct inode *inode,
7697                   struct nfs4_ol_stateid *open_stp)
7698 {
7699         struct nfs4_client *clp = lo->lo_owner.so_client;
7700         struct nfs4_ol_stateid *retstp;
7701
7702         mutex_init(&stp->st_mutex);
7703         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7704 retry:
7705         spin_lock(&clp->cl_lock);
7706         if (nfs4_ol_stateid_unhashed(open_stp))
7707                 goto out_close;
7708         retstp = find_lock_stateid(lo, open_stp);
7709         if (retstp)
7710                 goto out_found;
7711         refcount_inc(&stp->st_stid.sc_count);
7712         stp->st_stid.sc_type = SC_TYPE_LOCK;
7713         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7714         get_nfs4_file(fp);
7715         stp->st_stid.sc_file = fp;
7716         stp->st_access_bmap = 0;
7717         stp->st_deny_bmap = open_stp->st_deny_bmap;
7718         stp->st_openstp = open_stp;
7719         spin_lock(&fp->fi_lock);
7720         list_add(&stp->st_locks, &open_stp->st_locks);
7721         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7722         list_add(&stp->st_perfile, &fp->fi_stateids);
7723         spin_unlock(&fp->fi_lock);
7724         spin_unlock(&clp->cl_lock);
7725         return stp;
7726 out_found:
7727         spin_unlock(&clp->cl_lock);
7728         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7729                 nfs4_put_stid(&retstp->st_stid);
7730                 goto retry;
7731         }
7732         /* To keep mutex tracking happy */
7733         mutex_unlock(&stp->st_mutex);
7734         return retstp;
7735 out_close:
7736         spin_unlock(&clp->cl_lock);
7737         mutex_unlock(&stp->st_mutex);
7738         return NULL;
7739 }
7740
7741 static struct nfs4_ol_stateid *
7742 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7743                             struct inode *inode, struct nfs4_ol_stateid *ost,
7744                             bool *new)
7745 {
7746         struct nfs4_stid *ns = NULL;
7747         struct nfs4_ol_stateid *lst;
7748         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7749         struct nfs4_client *clp = oo->oo_owner.so_client;
7750
7751         *new = false;
7752         spin_lock(&clp->cl_lock);
7753         lst = find_lock_stateid(lo, ost);
7754         spin_unlock(&clp->cl_lock);
7755         if (lst != NULL) {
7756                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7757                         goto out;
7758                 nfs4_put_stid(&lst->st_stid);
7759         }
7760         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7761         if (ns == NULL)
7762                 return NULL;
7763
7764         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7765         if (lst == openlockstateid(ns))
7766                 *new = true;
7767         else
7768                 nfs4_put_stid(ns);
7769 out:
7770         return lst;
7771 }
7772
7773 static int
7774 check_lock_length(u64 offset, u64 length)
7775 {
7776         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7777                 (length > ~offset)));
7778 }
7779
7780 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7781 {
7782         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7783
7784         lockdep_assert_held(&fp->fi_lock);
7785
7786         if (test_access(access, lock_stp))
7787                 return;
7788         __nfs4_file_get_access(fp, access);
7789         set_access(access, lock_stp);
7790 }
7791
7792 static __be32
7793 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7794                             struct nfs4_ol_stateid *ost,
7795                             struct nfsd4_lock *lock,
7796                             struct nfs4_ol_stateid **plst, bool *new)
7797 {
7798         __be32 status;
7799         struct nfs4_file *fi = ost->st_stid.sc_file;
7800         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7801         struct nfs4_client *cl = oo->oo_owner.so_client;
7802         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7803         struct nfs4_lockowner *lo;
7804         struct nfs4_ol_stateid *lst;
7805         unsigned int strhashval;
7806
7807         lo = find_lockowner_str(cl, &lock->lk_new_owner);
7808         if (!lo) {
7809                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7810                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7811                 if (lo == NULL)
7812                         return nfserr_jukebox;
7813         } else {
7814                 /* with an existing lockowner, seqids must be the same */
7815                 status = nfserr_bad_seqid;
7816                 if (!cstate->minorversion &&
7817                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7818                         goto out;
7819         }
7820
7821         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7822         if (lst == NULL) {
7823                 status = nfserr_jukebox;
7824                 goto out;
7825         }
7826
7827         status = nfs_ok;
7828         *plst = lst;
7829 out:
7830         nfs4_put_stateowner(&lo->lo_owner);
7831         return status;
7832 }
7833
7834 /*
7835  *  LOCK operation
7836  */
7837 __be32
7838 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7839            union nfsd4_op_u *u)
7840 {
7841         struct nfsd4_lock *lock = &u->lock;
7842         struct nfs4_openowner *open_sop = NULL;
7843         struct nfs4_lockowner *lock_sop = NULL;
7844         struct nfs4_ol_stateid *lock_stp = NULL;
7845         struct nfs4_ol_stateid *open_stp = NULL;
7846         struct nfs4_file *fp;
7847         struct nfsd_file *nf = NULL;
7848         struct nfsd4_blocked_lock *nbl = NULL;
7849         struct file_lock *file_lock = NULL;
7850         struct file_lock *conflock = NULL;
7851         struct super_block *sb;
7852         __be32 status = 0;
7853         int lkflg;
7854         int err;
7855         bool new = false;
7856         unsigned char type;
7857         unsigned int flags = FL_POSIX;
7858         struct net *net = SVC_NET(rqstp);
7859         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7860
7861         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7862                 (long long) lock->lk_offset,
7863                 (long long) lock->lk_length);
7864
7865         if (check_lock_length(lock->lk_offset, lock->lk_length))
7866                  return nfserr_inval;
7867
7868         if ((status = fh_verify(rqstp, &cstate->current_fh,
7869                                 S_IFREG, NFSD_MAY_LOCK))) {
7870                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7871                 return status;
7872         }
7873         sb = cstate->current_fh.fh_dentry->d_sb;
7874
7875         if (lock->lk_is_new) {
7876                 if (nfsd4_has_session(cstate))
7877                         /* See rfc 5661 18.10.3: given clientid is ignored: */
7878                         memcpy(&lock->lk_new_clientid,
7879                                 &cstate->clp->cl_clientid,
7880                                 sizeof(clientid_t));
7881
7882                 /* validate and update open stateid and open seqid */
7883                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7884                                         lock->lk_new_open_seqid,
7885                                         &lock->lk_new_open_stateid,
7886                                         &open_stp, nn);
7887                 if (status)
7888                         goto out;
7889                 mutex_unlock(&open_stp->st_mutex);
7890                 open_sop = openowner(open_stp->st_stateowner);
7891                 status = nfserr_bad_stateid;
7892                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7893                                                 &lock->lk_new_clientid))
7894                         goto out;
7895                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7896                                                         &lock_stp, &new);
7897         } else {
7898                 status = nfs4_preprocess_seqid_op(cstate,
7899                                                   lock->lk_old_lock_seqid,
7900                                                   &lock->lk_old_lock_stateid,
7901                                                   SC_TYPE_LOCK, 0, &lock_stp,
7902                                                   nn);
7903         }
7904         if (status)
7905                 goto out;
7906         lock_sop = lockowner(lock_stp->st_stateowner);
7907
7908         lkflg = setlkflg(lock->lk_type);
7909         status = nfs4_check_openmode(lock_stp, lkflg);
7910         if (status)
7911                 goto out;
7912
7913         status = nfserr_grace;
7914         if (locks_in_grace(net) && !lock->lk_reclaim)
7915                 goto out;
7916         status = nfserr_no_grace;
7917         if (!locks_in_grace(net) && lock->lk_reclaim)
7918                 goto out;
7919
7920         if (lock->lk_reclaim)
7921                 flags |= FL_RECLAIM;
7922
7923         fp = lock_stp->st_stid.sc_file;
7924         switch (lock->lk_type) {
7925                 case NFS4_READW_LT:
7926                         if (nfsd4_has_session(cstate) ||
7927                             exportfs_lock_op_is_async(sb->s_export_op))
7928                                 flags |= FL_SLEEP;
7929                         fallthrough;
7930                 case NFS4_READ_LT:
7931                         spin_lock(&fp->fi_lock);
7932                         nf = find_readable_file_locked(fp);
7933                         if (nf)
7934                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7935                         spin_unlock(&fp->fi_lock);
7936                         type = F_RDLCK;
7937                         break;
7938                 case NFS4_WRITEW_LT:
7939                         if (nfsd4_has_session(cstate) ||
7940                             exportfs_lock_op_is_async(sb->s_export_op))
7941                                 flags |= FL_SLEEP;
7942                         fallthrough;
7943                 case NFS4_WRITE_LT:
7944                         spin_lock(&fp->fi_lock);
7945                         nf = find_writeable_file_locked(fp);
7946                         if (nf)
7947                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7948                         spin_unlock(&fp->fi_lock);
7949                         type = F_WRLCK;
7950                         break;
7951                 default:
7952                         status = nfserr_inval;
7953                 goto out;
7954         }
7955
7956         if (!nf) {
7957                 status = nfserr_openmode;
7958                 goto out;
7959         }
7960
7961         /*
7962          * Most filesystems with their own ->lock operations will block
7963          * the nfsd thread waiting to acquire the lock.  That leads to
7964          * deadlocks (we don't want every nfsd thread tied up waiting
7965          * for file locks), so don't attempt blocking lock notifications
7966          * on those filesystems:
7967          */
7968         if (!exportfs_lock_op_is_async(sb->s_export_op))
7969                 flags &= ~FL_SLEEP;
7970
7971         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7972         if (!nbl) {
7973                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7974                 status = nfserr_jukebox;
7975                 goto out;
7976         }
7977
7978         file_lock = &nbl->nbl_lock;
7979         file_lock->c.flc_type = type;
7980         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7981         file_lock->c.flc_pid = current->tgid;
7982         file_lock->c.flc_file = nf->nf_file;
7983         file_lock->c.flc_flags = flags;
7984         file_lock->fl_lmops = &nfsd_posix_mng_ops;
7985         file_lock->fl_start = lock->lk_offset;
7986         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7987         nfs4_transform_lock_offset(file_lock);
7988
7989         conflock = locks_alloc_lock();
7990         if (!conflock) {
7991                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7992                 status = nfserr_jukebox;
7993                 goto out;
7994         }
7995
7996         if (flags & FL_SLEEP) {
7997                 nbl->nbl_time = ktime_get_boottime_seconds();
7998                 spin_lock(&nn->blocked_locks_lock);
7999                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
8000                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
8001                 kref_get(&nbl->nbl_kref);
8002                 spin_unlock(&nn->blocked_locks_lock);
8003         }
8004
8005         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
8006         switch (err) {
8007         case 0: /* success! */
8008                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
8009                 status = 0;
8010                 if (lock->lk_reclaim)
8011                         nn->somebody_reclaimed = true;
8012                 break;
8013         case FILE_LOCK_DEFERRED:
8014                 kref_put(&nbl->nbl_kref, free_nbl);
8015                 nbl = NULL;
8016                 fallthrough;
8017         case -EAGAIN:           /* conflock holds conflicting lock */
8018                 status = nfserr_denied;
8019                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8020                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8021                 break;
8022         case -EDEADLK:
8023                 status = nfserr_deadlock;
8024                 break;
8025         default:
8026                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8027                 status = nfserrno(err);
8028                 break;
8029         }
8030 out:
8031         if (nbl) {
8032                 /* dequeue it if we queued it before */
8033                 if (flags & FL_SLEEP) {
8034                         spin_lock(&nn->blocked_locks_lock);
8035                         if (!list_empty(&nbl->nbl_list) &&
8036                             !list_empty(&nbl->nbl_lru)) {
8037                                 list_del_init(&nbl->nbl_list);
8038                                 list_del_init(&nbl->nbl_lru);
8039                                 kref_put(&nbl->nbl_kref, free_nbl);
8040                         }
8041                         /* nbl can use one of lists to be linked to reaplist */
8042                         spin_unlock(&nn->blocked_locks_lock);
8043                 }
8044                 free_blocked_lock(nbl);
8045         }
8046         if (nf)
8047                 nfsd_file_put(nf);
8048         if (lock_stp) {
8049                 /* Bump seqid manually if the 4.0 replay owner is openowner */
8050                 if (cstate->replay_owner &&
8051                     cstate->replay_owner != &lock_sop->lo_owner &&
8052                     seqid_mutating_err(ntohl(status)))
8053                         lock_sop->lo_owner.so_seqid++;
8054
8055                 /*
8056                  * If this is a new, never-before-used stateid, and we are
8057                  * returning an error, then just go ahead and release it.
8058                  */
8059                 if (status && new)
8060                         release_lock_stateid(lock_stp);
8061
8062                 mutex_unlock(&lock_stp->st_mutex);
8063
8064                 nfs4_put_stid(&lock_stp->st_stid);
8065         }
8066         if (open_stp)
8067                 nfs4_put_stid(&open_stp->st_stid);
8068         nfsd4_bump_seqid(cstate, status);
8069         if (conflock)
8070                 locks_free_lock(conflock);
8071         return status;
8072 }
8073
8074 void nfsd4_lock_release(union nfsd4_op_u *u)
8075 {
8076         struct nfsd4_lock *lock = &u->lock;
8077         struct nfsd4_lock_denied *deny = &lock->lk_denied;
8078
8079         kfree(deny->ld_owner.data);
8080 }
8081
8082 /*
8083  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8084  * so we do a temporary open here just to get an open file to pass to
8085  * vfs_test_lock.
8086  */
8087 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8088 {
8089         struct nfsd_file *nf;
8090         struct inode *inode;
8091         __be32 err;
8092
8093         err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8094         if (err)
8095                 return err;
8096         inode = fhp->fh_dentry->d_inode;
8097         inode_lock(inode); /* to block new leases till after test_lock: */
8098         err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8099         if (err)
8100                 goto out;
8101         lock->c.flc_file = nf->nf_file;
8102         err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8103         lock->c.flc_file = NULL;
8104 out:
8105         inode_unlock(inode);
8106         nfsd_file_put(nf);
8107         return err;
8108 }
8109
8110 /*
8111  * LOCKT operation
8112  */
8113 __be32
8114 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8115             union nfsd4_op_u *u)
8116 {
8117         struct nfsd4_lockt *lockt = &u->lockt;
8118         struct file_lock *file_lock = NULL;
8119         struct nfs4_lockowner *lo = NULL;
8120         __be32 status;
8121         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8122
8123         if (locks_in_grace(SVC_NET(rqstp)))
8124                 return nfserr_grace;
8125
8126         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8127                  return nfserr_inval;
8128
8129         if (!nfsd4_has_session(cstate)) {
8130                 status = set_client(&lockt->lt_clientid, cstate, nn);
8131                 if (status)
8132                         goto out;
8133         }
8134
8135         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8136                 goto out;
8137
8138         file_lock = locks_alloc_lock();
8139         if (!file_lock) {
8140                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8141                 status = nfserr_jukebox;
8142                 goto out;
8143         }
8144
8145         switch (lockt->lt_type) {
8146                 case NFS4_READ_LT:
8147                 case NFS4_READW_LT:
8148                         file_lock->c.flc_type = F_RDLCK;
8149                         break;
8150                 case NFS4_WRITE_LT:
8151                 case NFS4_WRITEW_LT:
8152                         file_lock->c.flc_type = F_WRLCK;
8153                         break;
8154                 default:
8155                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8156                         status = nfserr_inval;
8157                         goto out;
8158         }
8159
8160         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8161         if (lo)
8162                 file_lock->c.flc_owner = (fl_owner_t)lo;
8163         file_lock->c.flc_pid = current->tgid;
8164         file_lock->c.flc_flags = FL_POSIX;
8165
8166         file_lock->fl_start = lockt->lt_offset;
8167         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8168
8169         nfs4_transform_lock_offset(file_lock);
8170
8171         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8172         if (status)
8173                 goto out;
8174
8175         if (file_lock->c.flc_type != F_UNLCK) {
8176                 status = nfserr_denied;
8177                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8178         }
8179 out:
8180         if (lo)
8181                 nfs4_put_stateowner(&lo->lo_owner);
8182         if (file_lock)
8183                 locks_free_lock(file_lock);
8184         return status;
8185 }
8186
8187 void nfsd4_lockt_release(union nfsd4_op_u *u)
8188 {
8189         struct nfsd4_lockt *lockt = &u->lockt;
8190         struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8191
8192         kfree(deny->ld_owner.data);
8193 }
8194
8195 __be32
8196 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8197             union nfsd4_op_u *u)
8198 {
8199         struct nfsd4_locku *locku = &u->locku;
8200         struct nfs4_ol_stateid *stp;
8201         struct nfsd_file *nf = NULL;
8202         struct file_lock *file_lock = NULL;
8203         __be32 status;
8204         int err;
8205         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8206
8207         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8208                 (long long) locku->lu_offset,
8209                 (long long) locku->lu_length);
8210
8211         if (check_lock_length(locku->lu_offset, locku->lu_length))
8212                  return nfserr_inval;
8213
8214         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8215                                           &locku->lu_stateid, SC_TYPE_LOCK, 0,
8216                                           &stp, nn);
8217         if (status)
8218                 goto out;
8219         nf = find_any_file(stp->st_stid.sc_file);
8220         if (!nf) {
8221                 status = nfserr_lock_range;
8222                 goto put_stateid;
8223         }
8224         file_lock = locks_alloc_lock();
8225         if (!file_lock) {
8226                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8227                 status = nfserr_jukebox;
8228                 goto put_file;
8229         }
8230
8231         file_lock->c.flc_type = F_UNLCK;
8232         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8233         file_lock->c.flc_pid = current->tgid;
8234         file_lock->c.flc_file = nf->nf_file;
8235         file_lock->c.flc_flags = FL_POSIX;
8236         file_lock->fl_lmops = &nfsd_posix_mng_ops;
8237         file_lock->fl_start = locku->lu_offset;
8238
8239         file_lock->fl_end = last_byte_offset(locku->lu_offset,
8240                                                 locku->lu_length);
8241         nfs4_transform_lock_offset(file_lock);
8242
8243         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8244         if (err) {
8245                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8246                 goto out_nfserr;
8247         }
8248         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8249 put_file:
8250         nfsd_file_put(nf);
8251 put_stateid:
8252         mutex_unlock(&stp->st_mutex);
8253         nfs4_put_stid(&stp->st_stid);
8254 out:
8255         nfsd4_bump_seqid(cstate, status);
8256         if (file_lock)
8257                 locks_free_lock(file_lock);
8258         return status;
8259
8260 out_nfserr:
8261         status = nfserrno(err);
8262         goto put_file;
8263 }
8264
8265 /*
8266  * returns
8267  *      true:  locks held by lockowner
8268  *      false: no locks held by lockowner
8269  */
8270 static bool
8271 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8272 {
8273         struct file_lock *fl;
8274         int status = false;
8275         struct nfsd_file *nf;
8276         struct inode *inode;
8277         struct file_lock_context *flctx;
8278
8279         spin_lock(&fp->fi_lock);
8280         nf = find_any_file_locked(fp);
8281         if (!nf) {
8282                 /* Any valid lock stateid should have some sort of access */
8283                 WARN_ON_ONCE(1);
8284                 goto out;
8285         }
8286
8287         inode = file_inode(nf->nf_file);
8288         flctx = locks_inode_context(inode);
8289
8290         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8291                 spin_lock(&flctx->flc_lock);
8292                 for_each_file_lock(fl, &flctx->flc_posix) {
8293                         if (fl->c.flc_owner == (fl_owner_t)lowner) {
8294                                 status = true;
8295                                 break;
8296                         }
8297                 }
8298                 spin_unlock(&flctx->flc_lock);
8299         }
8300 out:
8301         spin_unlock(&fp->fi_lock);
8302         return status;
8303 }
8304
8305 /**
8306  * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8307  * @rqstp: RPC transaction
8308  * @cstate: NFSv4 COMPOUND state
8309  * @u: RELEASE_LOCKOWNER arguments
8310  *
8311  * Check if theree are any locks still held and if not - free the lockowner
8312  * and any lock state that is owned.
8313  *
8314  * Return values:
8315  *   %nfs_ok: lockowner released or not found
8316  *   %nfserr_locks_held: lockowner still in use
8317  *   %nfserr_stale_clientid: clientid no longer active
8318  *   %nfserr_expired: clientid not recognized
8319  */
8320 __be32
8321 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8322                         struct nfsd4_compound_state *cstate,
8323                         union nfsd4_op_u *u)
8324 {
8325         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8326         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8327         clientid_t *clid = &rlockowner->rl_clientid;
8328         struct nfs4_ol_stateid *stp;
8329         struct nfs4_lockowner *lo;
8330         struct nfs4_client *clp;
8331         LIST_HEAD(reaplist);
8332         __be32 status;
8333
8334         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8335                 clid->cl_boot, clid->cl_id);
8336
8337         status = set_client(clid, cstate, nn);
8338         if (status)
8339                 return status;
8340         clp = cstate->clp;
8341
8342         spin_lock(&clp->cl_lock);
8343         lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8344         if (!lo) {
8345                 spin_unlock(&clp->cl_lock);
8346                 return nfs_ok;
8347         }
8348
8349         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8350                 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8351                         spin_unlock(&clp->cl_lock);
8352                         nfs4_put_stateowner(&lo->lo_owner);
8353                         return nfserr_locks_held;
8354                 }
8355         }
8356         unhash_lockowner_locked(lo);
8357         while (!list_empty(&lo->lo_owner.so_stateids)) {
8358                 stp = list_first_entry(&lo->lo_owner.so_stateids,
8359                                        struct nfs4_ol_stateid,
8360                                        st_perstateowner);
8361                 unhash_lock_stateid(stp);
8362                 put_ol_stateid_locked(stp, &reaplist);
8363         }
8364         spin_unlock(&clp->cl_lock);
8365
8366         free_ol_stateid_reaplist(&reaplist);
8367         remove_blocked_locks(lo);
8368         nfs4_put_stateowner(&lo->lo_owner);
8369         return nfs_ok;
8370 }
8371
8372 static inline struct nfs4_client_reclaim *
8373 alloc_reclaim(void)
8374 {
8375         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8376 }
8377
8378 bool
8379 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8380 {
8381         struct nfs4_client_reclaim *crp;
8382
8383         crp = nfsd4_find_reclaim_client(name, nn);
8384         return (crp && crp->cr_clp);
8385 }
8386
8387 /*
8388  * failure => all reset bets are off, nfserr_no_grace...
8389  *
8390  * The caller is responsible for freeing name.data if NULL is returned (it
8391  * will be freed in nfs4_remove_reclaim_record in the normal case).
8392  */
8393 struct nfs4_client_reclaim *
8394 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8395                 struct nfsd_net *nn)
8396 {
8397         unsigned int strhashval;
8398         struct nfs4_client_reclaim *crp;
8399
8400         crp = alloc_reclaim();
8401         if (crp) {
8402                 strhashval = clientstr_hashval(name);
8403                 INIT_LIST_HEAD(&crp->cr_strhash);
8404                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8405                 crp->cr_name.data = name.data;
8406                 crp->cr_name.len = name.len;
8407                 crp->cr_princhash.data = princhash.data;
8408                 crp->cr_princhash.len = princhash.len;
8409                 crp->cr_clp = NULL;
8410                 nn->reclaim_str_hashtbl_size++;
8411         }
8412         return crp;
8413 }
8414
8415 void
8416 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8417 {
8418         list_del(&crp->cr_strhash);
8419         kfree(crp->cr_name.data);
8420         kfree(crp->cr_princhash.data);
8421         kfree(crp);
8422         nn->reclaim_str_hashtbl_size--;
8423 }
8424
8425 void
8426 nfs4_release_reclaim(struct nfsd_net *nn)
8427 {
8428         struct nfs4_client_reclaim *crp = NULL;
8429         int i;
8430
8431         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8432                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8433                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8434                                         struct nfs4_client_reclaim, cr_strhash);
8435                         nfs4_remove_reclaim_record(crp, nn);
8436                 }
8437         }
8438         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8439 }
8440
8441 /*
8442  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8443 struct nfs4_client_reclaim *
8444 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8445 {
8446         unsigned int strhashval;
8447         struct nfs4_client_reclaim *crp = NULL;
8448
8449         strhashval = clientstr_hashval(name);
8450         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8451                 if (compare_blob(&crp->cr_name, &name) == 0) {
8452                         return crp;
8453                 }
8454         }
8455         return NULL;
8456 }
8457
8458 __be32
8459 nfs4_check_open_reclaim(struct nfs4_client *clp)
8460 {
8461         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8462                 return nfserr_no_grace;
8463
8464         if (nfsd4_client_record_check(clp))
8465                 return nfserr_reclaim_bad;
8466
8467         return nfs_ok;
8468 }
8469
8470 /*
8471  * Since the lifetime of a delegation isn't limited to that of an open, a
8472  * client may quite reasonably hang on to a delegation as long as it has
8473  * the inode cached.  This becomes an obvious problem the first time a
8474  * client's inode cache approaches the size of the server's total memory.
8475  *
8476  * For now we avoid this problem by imposing a hard limit on the number
8477  * of delegations, which varies according to the server's memory size.
8478  */
8479 static void
8480 set_max_delegations(void)
8481 {
8482         /*
8483          * Allow at most 4 delegations per megabyte of RAM.  Quick
8484          * estimates suggest that in the worst case (where every delegation
8485          * is for a different inode), a delegation could take about 1.5K,
8486          * giving a worst case usage of about 6% of memory.
8487          */
8488         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8489 }
8490
8491 static int nfs4_state_create_net(struct net *net)
8492 {
8493         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8494         int i;
8495
8496         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8497                                             sizeof(struct list_head),
8498                                             GFP_KERNEL);
8499         if (!nn->conf_id_hashtbl)
8500                 goto err;
8501         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8502                                               sizeof(struct list_head),
8503                                               GFP_KERNEL);
8504         if (!nn->unconf_id_hashtbl)
8505                 goto err_unconf_id;
8506         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8507                                               sizeof(struct list_head),
8508                                               GFP_KERNEL);
8509         if (!nn->sessionid_hashtbl)
8510                 goto err_sessionid;
8511
8512         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8513                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8514                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8515         }
8516         for (i = 0; i < SESSION_HASH_SIZE; i++)
8517                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8518         nn->conf_name_tree = RB_ROOT;
8519         nn->unconf_name_tree = RB_ROOT;
8520         nn->boot_time = ktime_get_real_seconds();
8521         nn->grace_ended = false;
8522         nn->nfsd4_manager.block_opens = true;
8523         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8524         INIT_LIST_HEAD(&nn->client_lru);
8525         INIT_LIST_HEAD(&nn->close_lru);
8526         INIT_LIST_HEAD(&nn->del_recall_lru);
8527         spin_lock_init(&nn->client_lock);
8528         spin_lock_init(&nn->s2s_cp_lock);
8529         idr_init(&nn->s2s_cp_stateids);
8530
8531         spin_lock_init(&nn->blocked_locks_lock);
8532         INIT_LIST_HEAD(&nn->blocked_locks_lru);
8533
8534         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8535         INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8536         get_net(net);
8537
8538         nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8539         if (!nn->nfsd_client_shrinker)
8540                 goto err_shrinker;
8541
8542         nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8543         nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8544         nn->nfsd_client_shrinker->private_data = nn;
8545
8546         shrinker_register(nn->nfsd_client_shrinker);
8547
8548         return 0;
8549
8550 err_shrinker:
8551         put_net(net);
8552         kfree(nn->sessionid_hashtbl);
8553 err_sessionid:
8554         kfree(nn->unconf_id_hashtbl);
8555 err_unconf_id:
8556         kfree(nn->conf_id_hashtbl);
8557 err:
8558         return -ENOMEM;
8559 }
8560
8561 static void
8562 nfs4_state_destroy_net(struct net *net)
8563 {
8564         int i;
8565         struct nfs4_client *clp = NULL;
8566         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8567
8568         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8569                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8570                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8571                         destroy_client(clp);
8572                 }
8573         }
8574
8575         WARN_ON(!list_empty(&nn->blocked_locks_lru));
8576
8577         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8578                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8579                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8580                         destroy_client(clp);
8581                 }
8582         }
8583
8584         kfree(nn->sessionid_hashtbl);
8585         kfree(nn->unconf_id_hashtbl);
8586         kfree(nn->conf_id_hashtbl);
8587         put_net(net);
8588 }
8589
8590 int
8591 nfs4_state_start_net(struct net *net)
8592 {
8593         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8594         int ret;
8595
8596         ret = nfs4_state_create_net(net);
8597         if (ret)
8598                 return ret;
8599         locks_start_grace(net, &nn->nfsd4_manager);
8600         nfsd4_client_tracking_init(net);
8601         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8602                 goto skip_grace;
8603         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8604                nn->nfsd4_grace, net->ns.inum);
8605         trace_nfsd_grace_start(nn);
8606         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8607         return 0;
8608
8609 skip_grace:
8610         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8611                         net->ns.inum);
8612         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8613         nfsd4_end_grace(nn);
8614         return 0;
8615 }
8616
8617 /* initialization to perform when the nfsd service is started: */
8618
8619 int
8620 nfs4_state_start(void)
8621 {
8622         int ret;
8623
8624         ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8625         if (ret)
8626                 return ret;
8627
8628         ret = nfsd4_create_callback_queue();
8629         if (ret) {
8630                 rhltable_destroy(&nfs4_file_rhltable);
8631                 return ret;
8632         }
8633
8634         set_max_delegations();
8635         return 0;
8636 }
8637
8638 void
8639 nfs4_state_shutdown_net(struct net *net)
8640 {
8641         struct nfs4_delegation *dp = NULL;
8642         struct list_head *pos, *next, reaplist;
8643         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8644
8645         shrinker_free(nn->nfsd_client_shrinker);
8646         cancel_work(&nn->nfsd_shrinker_work);
8647         cancel_delayed_work_sync(&nn->laundromat_work);
8648         locks_end_grace(&nn->nfsd4_manager);
8649
8650         INIT_LIST_HEAD(&reaplist);
8651         spin_lock(&state_lock);
8652         list_for_each_safe(pos, next, &nn->del_recall_lru) {
8653                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8654                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8655                 list_add(&dp->dl_recall_lru, &reaplist);
8656         }
8657         spin_unlock(&state_lock);
8658         list_for_each_safe(pos, next, &reaplist) {
8659                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8660                 list_del_init(&dp->dl_recall_lru);
8661                 destroy_unhashed_deleg(dp);
8662         }
8663
8664         nfsd4_client_tracking_exit(net);
8665         nfs4_state_destroy_net(net);
8666 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8667         nfsd4_ssc_shutdown_umount(nn);
8668 #endif
8669 }
8670
8671 void
8672 nfs4_state_shutdown(void)
8673 {
8674         nfsd4_destroy_callback_queue();
8675         rhltable_destroy(&nfs4_file_rhltable);
8676 }
8677
8678 static void
8679 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8680 {
8681         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8682             CURRENT_STATEID(stateid))
8683                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8684 }
8685
8686 static void
8687 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8688 {
8689         if (cstate->minorversion) {
8690                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8691                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8692         }
8693 }
8694
8695 void
8696 clear_current_stateid(struct nfsd4_compound_state *cstate)
8697 {
8698         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8699 }
8700
8701 /*
8702  * functions to set current state id
8703  */
8704 void
8705 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8706                 union nfsd4_op_u *u)
8707 {
8708         put_stateid(cstate, &u->open_downgrade.od_stateid);
8709 }
8710
8711 void
8712 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8713                 union nfsd4_op_u *u)
8714 {
8715         put_stateid(cstate, &u->open.op_stateid);
8716 }
8717
8718 void
8719 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8720                 union nfsd4_op_u *u)
8721 {
8722         put_stateid(cstate, &u->close.cl_stateid);
8723 }
8724
8725 void
8726 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8727                 union nfsd4_op_u *u)
8728 {
8729         put_stateid(cstate, &u->lock.lk_resp_stateid);
8730 }
8731
8732 /*
8733  * functions to consume current state id
8734  */
8735
8736 void
8737 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8738                 union nfsd4_op_u *u)
8739 {
8740         get_stateid(cstate, &u->open_downgrade.od_stateid);
8741 }
8742
8743 void
8744 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8745                 union nfsd4_op_u *u)
8746 {
8747         get_stateid(cstate, &u->delegreturn.dr_stateid);
8748 }
8749
8750 void
8751 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8752                 union nfsd4_op_u *u)
8753 {
8754         get_stateid(cstate, &u->free_stateid.fr_stateid);
8755 }
8756
8757 void
8758 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8759                 union nfsd4_op_u *u)
8760 {
8761         get_stateid(cstate, &u->setattr.sa_stateid);
8762 }
8763
8764 void
8765 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8766                 union nfsd4_op_u *u)
8767 {
8768         get_stateid(cstate, &u->close.cl_stateid);
8769 }
8770
8771 void
8772 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8773                 union nfsd4_op_u *u)
8774 {
8775         get_stateid(cstate, &u->locku.lu_stateid);
8776 }
8777
8778 void
8779 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8780                 union nfsd4_op_u *u)
8781 {
8782         get_stateid(cstate, &u->read.rd_stateid);
8783 }
8784
8785 void
8786 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8787                 union nfsd4_op_u *u)
8788 {
8789         get_stateid(cstate, &u->write.wr_stateid);
8790 }
8791
8792 /**
8793  * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8794  * @rqstp: RPC transaction context
8795  * @inode: file to be checked for a conflict
8796  * @modified: return true if file was modified
8797  * @size: new size of file if modified is true
8798  *
8799  * This function is called when there is a conflict between a write
8800  * delegation and a change/size GETATTR from another client. The server
8801  * must either use the CB_GETATTR to get the current values of the
8802  * attributes from the client that holds the delegation or recall the
8803  * delegation before replying to the GETATTR. See RFC 8881 section
8804  * 18.7.4.
8805  *
8806  * Returns 0 if there is no conflict; otherwise an nfs_stat
8807  * code is returned.
8808  */
8809 __be32
8810 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode,
8811                                 bool *modified, u64 *size)
8812 {
8813         __be32 status;
8814         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8815         struct file_lock_context *ctx;
8816         struct file_lease *fl;
8817         struct nfs4_delegation *dp;
8818         struct iattr attrs;
8819         struct nfs4_cb_fattr *ncf;
8820
8821         *modified = false;
8822         ctx = locks_inode_context(inode);
8823         if (!ctx)
8824                 return 0;
8825         spin_lock(&ctx->flc_lock);
8826         for_each_file_lock(fl, &ctx->flc_lease) {
8827                 unsigned char type = fl->c.flc_type;
8828
8829                 if (fl->c.flc_flags == FL_LAYOUT)
8830                         continue;
8831                 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8832                         /*
8833                          * non-nfs lease, if it's a lease with F_RDLCK then
8834                          * we are done; there isn't any write delegation
8835                          * on this inode
8836                          */
8837                         if (type == F_RDLCK)
8838                                 break;
8839                         goto break_lease;
8840                 }
8841                 if (type == F_WRLCK) {
8842                         dp = fl->c.flc_owner;
8843                         if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8844                                 spin_unlock(&ctx->flc_lock);
8845                                 return 0;
8846                         }
8847 break_lease:
8848                         nfsd_stats_wdeleg_getattr_inc(nn);
8849                         dp = fl->c.flc_owner;
8850                         ncf = &dp->dl_cb_fattr;
8851                         nfs4_cb_getattr(&dp->dl_cb_fattr);
8852                         spin_unlock(&ctx->flc_lock);
8853                         wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8854                                         TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8855                         if (ncf->ncf_cb_status) {
8856                                 /* Recall delegation only if client didn't respond */
8857                                 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8858                                 if (status != nfserr_jukebox ||
8859                                                 !nfsd_wait_for_delegreturn(rqstp, inode))
8860                                         return status;
8861                         }
8862                         if (!ncf->ncf_file_modified &&
8863                                         (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8864                                         ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8865                                 ncf->ncf_file_modified = true;
8866                         if (ncf->ncf_file_modified) {
8867                                 /*
8868                                  * Per section 10.4.3 of RFC 8881, the server would
8869                                  * not update the file's metadata with the client's
8870                                  * modified size
8871                                  */
8872                                 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8873                                 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME;
8874                                 setattr_copy(&nop_mnt_idmap, inode, &attrs);
8875                                 mark_inode_dirty(inode);
8876                                 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8877                                 *size = ncf->ncf_cur_fsize;
8878                                 *modified = true;
8879                         }
8880                         return 0;
8881                 }
8882                 break;
8883         }
8884         spin_unlock(&ctx->flc_lock);
8885         return 0;
8886 }