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         .opcode         = OP_CB_NOTIFY_LOCK,
 404 };
 405
 406 /*
 407  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 408  * st_{access,deny}_bmap field of the stateid, in order to track not
 409  * only what share bits are currently in force, but also what
 410  * combinations of share bits previous opens have used.  This allows us
 411  * to enforce the recommendation in
 412  * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
 413  * the server return an error if the client attempt to downgrade to a
 414  * combination of share bits not explicable by closing some of its
 415  * previous opens.
 416  *
 417  * This enforcement is arguably incomplete, since we don't keep
 418  * track of access/deny bit combinations; so, e.g., we allow:
 419  *
 420  *      OPEN allow read, deny write
 421  *      OPEN allow both, deny none
 422  *      DOWNGRADE allow read, deny none
 423  *
 424  * which we should reject.
 425  *
 426  * But you could also argue that our current code is already overkill,
 427  * since it only exists to return NFS4ERR_INVAL on incorrect client
 428  * behavior.
 429  */
 430 static unsigned int
 431 bmap_to_share_mode(unsigned long bmap)
 432 {
 433         int i;
 434         unsigned int access = 0;
 435
 436         for (i = 1; i < 4; i++) {
 437                 if (test_bit(i, &bmap))
 438                         access |= i;
 439         }
 440         return access;
 441 }
 442
 443 /* set share access for a given stateid */
 444 static inline void
 445 set_access(u32 access, struct nfs4_ol_stateid *stp)
 446 {
 447         unsigned char mask = 1 << access;
 448
 449         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 450         stp->st_access_bmap |= mask;
 451 }
 452
 453 /* clear share access for a given stateid */
 454 static inline void
 455 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 456 {
 457         unsigned char mask = 1 << access;
 458
 459         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 460         stp->st_access_bmap &= ~mask;
 461 }
 462
 463 /* test whether a given stateid has access */
 464 static inline bool
 465 test_access(u32 access, struct nfs4_ol_stateid *stp)
 466 {
 467         unsigned char mask = 1 << access;
 468
 469         return (bool)(stp->st_access_bmap & mask);
 470 }
 471
 472 /* set share deny for a given stateid */
 473 static inline void
 474 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 475 {
 476         unsigned char mask = 1 << deny;
 477
 478         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 479         stp->st_deny_bmap |= mask;
 480 }
 481
 482 /* clear share deny for a given stateid */
 483 static inline void
 484 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 485 {
 486         unsigned char mask = 1 << deny;
 487
 488         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 489         stp->st_deny_bmap &= ~mask;
 490 }
 491
 492 /* test whether a given stateid is denying specific access */
 493 static inline bool
 494 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 495 {
 496         unsigned char mask = 1 << deny;
 497
 498         return (bool)(stp->st_deny_bmap & mask);
 499 }
 500
 501 static int nfs4_access_to_omode(u32 access)
 502 {
 503         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 504         case NFS4_SHARE_ACCESS_READ:
 505                 return O_RDONLY;
 506         case NFS4_SHARE_ACCESS_WRITE:
 507                 return O_WRONLY;
 508         case NFS4_SHARE_ACCESS_BOTH:
 509                 return O_RDWR;
 510         }
 511         WARN_ON_ONCE(1);
 512         return O_RDONLY;
 513 }
 514
 515 static inline int
 516 access_permit_read(struct nfs4_ol_stateid *stp)
 517 {
 518         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
 519                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
 520                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
 521 }
 522
 523 static inline int
 524 access_permit_write(struct nfs4_ol_stateid *stp)
 525 {
 526         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
 527                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
 528 }
 529
 530 static inline struct nfs4_stateowner *
 531 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 532 {
 533         atomic_inc(&sop->so_count);
 534         return sop;
 535 }
 536
 537 static int
 538 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 539 {
 540         return (sop->so_owner.len == owner->len) &&
 541                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 542 }
 543
 544 static struct nfs4_openowner *
 545 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 546                         struct nfs4_client *clp)
 547 {
 548         struct nfs4_stateowner *so;
 549
 550         lockdep_assert_held(&clp->cl_lock);
 551
 552         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 553                             so_strhash) {
 554                 if (!so->so_is_open_owner)
 555                         continue;
 556                 if (same_owner_str(so, &open->op_owner))
 557                         return openowner(nfs4_get_stateowner(so));
 558         }
 559         return NULL;
 560 }
 561
 562 static inline u32
 563 opaque_hashval(const void *ptr, int nbytes)
 564 {
 565         unsigned char *cptr = (unsigned char *) ptr;
 566
 567         u32 x = 0;
 568         while (nbytes--) {
 569                 x *= 37;
 570                 x += *cptr++;
 571         }
 572         return x;
 573 }
 574
 575 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 576 {
 577         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 578
 579         kmem_cache_free(file_slab, fp);
 580 }
 581
 582 void
 583 put_nfs4_file(struct nfs4_file *fi)
 584 {
 585         if (refcount_dec_and_test(&fi->fi_ref)) {
 586                 nfsd4_file_hash_remove(fi);
 587                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 588                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 589                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 590         }
 591 }
 592
 593 static struct nfsd_file *
 594 find_writeable_file_locked(struct nfs4_file *f)
 595 {
 596         struct nfsd_file *ret;
 597
 598         lockdep_assert_held(&f->fi_lock);
 599
 600         ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 601         if (!ret)
 602                 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 603         return ret;
 604 }
 605
 606 static struct nfsd_file *
 607 find_writeable_file(struct nfs4_file *f)
 608 {
 609         struct nfsd_file *ret;
 610
 611         spin_lock(&f->fi_lock);
 612         ret = find_writeable_file_locked(f);
 613         spin_unlock(&f->fi_lock);
 614
 615         return ret;
 616 }
 617
 618 static struct nfsd_file *
 619 find_readable_file_locked(struct nfs4_file *f)
 620 {
 621         struct nfsd_file *ret;
 622
 623         lockdep_assert_held(&f->fi_lock);
 624
 625         ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 626         if (!ret)
 627                 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 628         return ret;
 629 }
 630
 631 static struct nfsd_file *
 632 find_readable_file(struct nfs4_file *f)
 633 {
 634         struct nfsd_file *ret;
 635
 636         spin_lock(&f->fi_lock);
 637         ret = find_readable_file_locked(f);
 638         spin_unlock(&f->fi_lock);
 639
 640         return ret;
 641 }
 642
 643 static struct nfsd_file *
 644 find_rw_file(struct nfs4_file *f)
 645 {
 646         struct nfsd_file *ret;
 647
 648         spin_lock(&f->fi_lock);
 649         ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 650         spin_unlock(&f->fi_lock);
 651
 652         return ret;
 653 }
 654
 655 struct nfsd_file *
 656 find_any_file(struct nfs4_file *f)
 657 {
 658         struct nfsd_file *ret;
 659
 660         if (!f)
 661                 return NULL;
 662         spin_lock(&f->fi_lock);
 663         ret = nfsd_file_get(f->fi_fds[O_RDWR]);
 664         if (!ret) {
 665                 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
 666                 if (!ret)
 667                         ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
 668         }
 669         spin_unlock(&f->fi_lock);
 670         return ret;
 671 }
 672
 673 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
 674 {
 675         lockdep_assert_held(&f->fi_lock);
 676
 677         if (f->fi_fds[O_RDWR])
 678                 return f->fi_fds[O_RDWR];
 679         if (f->fi_fds[O_WRONLY])
 680                 return f->fi_fds[O_WRONLY];
 681         if (f->fi_fds[O_RDONLY])
 682                 return f->fi_fds[O_RDONLY];
 683         return NULL;
 684 }
 685
 686 static atomic_long_t num_delegations;
 687 unsigned long max_delegations;
 688
 689 /*
 690  * Open owner state (share locks)
 691  */
 692
 693 /* hash tables for lock and open owners */
 694 #define OWNER_HASH_BITS              8
 695 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 696 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 697
 698 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 699 {
 700         unsigned int ret;
 701
 702         ret = opaque_hashval(ownername->data, ownername->len);
 703         return ret & OWNER_HASH_MASK;
 704 }
 705
 706 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
 707
 708 static const struct rhashtable_params nfs4_file_rhash_params = {
 709         .key_len                = sizeof_field(struct nfs4_file, fi_inode),
 710         .key_offset             = offsetof(struct nfs4_file, fi_inode),
 711         .head_offset            = offsetof(struct nfs4_file, fi_rlist),
 712
 713         /*
 714          * Start with a single page hash table to reduce resizing churn
 715          * on light workloads.
 716          */
 717         .min_size               = 256,
 718         .automatic_shrinking    = true,
 719 };
 720
 721 /*
 722  * Check if courtesy clients have conflicting access and resolve it if possible
 723  *
 724  * access:  is op_share_access if share_access is true.
 725  *          Check if access mode, op_share_access, would conflict with
 726  *          the current deny mode of the file 'fp'.
 727  * access:  is op_share_deny if share_access is false.
 728  *          Check if the deny mode, op_share_deny, would conflict with
 729  *          current access of the file 'fp'.
 730  * stp:     skip checking this entry.
 731  * new_stp: normal open, not open upgrade.
 732  *
 733  * Function returns:
 734  *      false - access/deny mode conflict with normal client.
 735  *      true  - no conflict or conflict with courtesy client(s) is resolved.
 736  */
 737 static bool
 738 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
 739                 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
 740 {
 741         struct nfs4_ol_stateid *st;
 742         bool resolvable = true;
 743         unsigned char bmap;
 744         struct nfsd_net *nn;
 745         struct nfs4_client *clp;
 746
 747         lockdep_assert_held(&fp->fi_lock);
 748         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
 749                 /* ignore lock stateid */
 750                 if (st->st_openstp)
 751                         continue;
 752                 if (st == stp && new_stp)
 753                         continue;
 754                 /* check file access against deny mode or vice versa */
 755                 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
 756                 if (!(access & bmap_to_share_mode(bmap)))
 757                         continue;
 758                 clp = st->st_stid.sc_client;
 759                 if (try_to_expire_client(clp))
 760                         continue;
 761                 resolvable = false;
 762                 break;
 763         }
 764         if (resolvable) {
 765                 clp = stp->st_stid.sc_client;
 766                 nn = net_generic(clp->net, nfsd_net_id);
 767                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
 768         }
 769         return resolvable;
 770 }
 771
 772 static void
 773 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 774 {
 775         lockdep_assert_held(&fp->fi_lock);
 776
 777         if (access & NFS4_SHARE_ACCESS_WRITE)
 778                 atomic_inc(&fp->fi_access[O_WRONLY]);
 779         if (access & NFS4_SHARE_ACCESS_READ)
 780                 atomic_inc(&fp->fi_access[O_RDONLY]);
 781 }
 782
 783 static __be32
 784 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 785 {
 786         lockdep_assert_held(&fp->fi_lock);
 787
 788         /* Does this access mode make sense? */
 789         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 790                 return nfserr_inval;
 791
 792         /* Does it conflict with a deny mode already set? */
 793         if ((access & fp->fi_share_deny) != 0)
 794                 return nfserr_share_denied;
 795
 796         __nfs4_file_get_access(fp, access);
 797         return nfs_ok;
 798 }
 799
 800 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 801 {
 802         /* Common case is that there is no deny mode. */
 803         if (deny) {
 804                 /* Does this deny mode make sense? */
 805                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 806                         return nfserr_inval;
 807
 808                 if ((deny & NFS4_SHARE_DENY_READ) &&
 809                     atomic_read(&fp->fi_access[O_RDONLY]))
 810                         return nfserr_share_denied;
 811
 812                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 813                     atomic_read(&fp->fi_access[O_WRONLY]))
 814                         return nfserr_share_denied;
 815         }
 816         return nfs_ok;
 817 }
 818
 819 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 820 {
 821         might_lock(&fp->fi_lock);
 822
 823         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 824                 struct nfsd_file *f1 = NULL;
 825                 struct nfsd_file *f2 = NULL;
 826
 827                 swap(f1, fp->fi_fds[oflag]);
 828                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 829                         swap(f2, fp->fi_fds[O_RDWR]);
 830                 spin_unlock(&fp->fi_lock);
 831                 if (f1)
 832                         nfsd_file_put(f1);
 833                 if (f2)
 834                         nfsd_file_put(f2);
 835         }
 836 }
 837
 838 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 839 {
 840         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 841
 842         if (access & NFS4_SHARE_ACCESS_WRITE)
 843                 __nfs4_file_put_access(fp, O_WRONLY);
 844         if (access & NFS4_SHARE_ACCESS_READ)
 845                 __nfs4_file_put_access(fp, O_RDONLY);
 846 }
 847
 848 /*
 849  * Allocate a new open/delegation state counter. This is needed for
 850  * pNFS for proper return on close semantics.
 851  *
 852  * Note that we only allocate it for pNFS-enabled exports, otherwise
 853  * all pointers to struct nfs4_clnt_odstate are always NULL.
 854  */
 855 static struct nfs4_clnt_odstate *
 856 alloc_clnt_odstate(struct nfs4_client *clp)
 857 {
 858         struct nfs4_clnt_odstate *co;
 859
 860         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 861         if (co) {
 862                 co->co_client = clp;
 863                 refcount_set(&co->co_odcount, 1);
 864         }
 865         return co;
 866 }
 867
 868 static void
 869 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 870 {
 871         struct nfs4_file *fp = co->co_file;
 872
 873         lockdep_assert_held(&fp->fi_lock);
 874         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 875 }
 876
 877 static inline void
 878 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 879 {
 880         if (co)
 881                 refcount_inc(&co->co_odcount);
 882 }
 883
 884 static void
 885 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 886 {
 887         struct nfs4_file *fp;
 888
 889         if (!co)
 890                 return;
 891
 892         fp = co->co_file;
 893         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 894                 list_del(&co->co_perfile);
 895                 spin_unlock(&fp->fi_lock);
 896
 897                 nfsd4_return_all_file_layouts(co->co_client, fp);
 898                 kmem_cache_free(odstate_slab, co);
 899         }
 900 }
 901
 902 static struct nfs4_clnt_odstate *
 903 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 904 {
 905         struct nfs4_clnt_odstate *co;
 906         struct nfs4_client *cl;
 907
 908         if (!new)
 909                 return NULL;
 910
 911         cl = new->co_client;
 912
 913         spin_lock(&fp->fi_lock);
 914         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 915                 if (co->co_client == cl) {
 916                         get_clnt_odstate(co);
 917                         goto out;
 918                 }
 919         }
 920         co = new;
 921         co->co_file = fp;
 922         hash_clnt_odstate_locked(new);
 923 out:
 924         spin_unlock(&fp->fi_lock);
 925         return co;
 926 }
 927
 928 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 929                                   void (*sc_free)(struct nfs4_stid *))
 930 {
 931         struct nfs4_stid *stid;
 932         int new_id;
 933
 934         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 935         if (!stid)
 936                 return NULL;
 937
 938         idr_preload(GFP_KERNEL);
 939         spin_lock(&cl->cl_lock);
 940         /* Reserving 0 for start of file in nfsdfs "states" file: */
 941         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 942         spin_unlock(&cl->cl_lock);
 943         idr_preload_end();
 944         if (new_id < 0)
 945                 goto out_free;
 946
 947         stid->sc_free = sc_free;
 948         stid->sc_client = cl;
 949         stid->sc_stateid.si_opaque.so_id = new_id;
 950         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 951         /* Will be incremented before return to client: */
 952         refcount_set(&stid->sc_count, 1);
 953         spin_lock_init(&stid->sc_lock);
 954         INIT_LIST_HEAD(&stid->sc_cp_list);
 955
 956         /*
 957          * It shouldn't be a problem to reuse an opaque stateid value.
 958          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 959          * example, a stray write retransmission could be accepted by
 960          * the server when it should have been rejected.  Therefore,
 961          * adopt a trick from the sctp code to attempt to maximize the
 962          * amount of time until an id is reused, by ensuring they always
 963          * "increase" (mod INT_MAX):
 964          */
 965         return stid;
 966 out_free:
 967         kmem_cache_free(slab, stid);
 968         return NULL;
 969 }
 970
 971 /*
 972  * Create a unique stateid_t to represent each COPY.
 973  */
 974 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 975                               unsigned char cs_type)
 976 {
 977         int new_id;
 978
 979         stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 980         stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 981
 982         idr_preload(GFP_KERNEL);
 983         spin_lock(&nn->s2s_cp_lock);
 984         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 985         stid->cs_stid.si_opaque.so_id = new_id;
 986         stid->cs_stid.si_generation = 1;
 987         spin_unlock(&nn->s2s_cp_lock);
 988         idr_preload_end();
 989         if (new_id < 0)
 990                 return 0;
 991         stid->cs_type = cs_type;
 992         return 1;
 993 }
 994
 995 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 996 {
 997         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
 998 }
 999
1000 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1001                                                      struct nfs4_stid *p_stid)
1002 {
1003         struct nfs4_cpntf_state *cps;
1004
1005         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1006         if (!cps)
1007                 return NULL;
1008         cps->cpntf_time = ktime_get_boottime_seconds();
1009         refcount_set(&cps->cp_stateid.cs_count, 1);
1010         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1011                 goto out_free;
1012         spin_lock(&nn->s2s_cp_lock);
1013         list_add(&cps->cp_list, &p_stid->sc_cp_list);
1014         spin_unlock(&nn->s2s_cp_lock);
1015         return cps;
1016 out_free:
1017         kfree(cps);
1018         return NULL;
1019 }
1020
1021 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1022 {
1023         struct nfsd_net *nn;
1024
1025         if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1026                 return;
1027         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1028         spin_lock(&nn->s2s_cp_lock);
1029         idr_remove(&nn->s2s_cp_stateids,
1030                    copy->cp_stateid.cs_stid.si_opaque.so_id);
1031         spin_unlock(&nn->s2s_cp_lock);
1032 }
1033
1034 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1035 {
1036         struct nfs4_cpntf_state *cps;
1037         struct nfsd_net *nn;
1038
1039         nn = net_generic(net, nfsd_net_id);
1040         spin_lock(&nn->s2s_cp_lock);
1041         while (!list_empty(&stid->sc_cp_list)) {
1042                 cps = list_first_entry(&stid->sc_cp_list,
1043                                        struct nfs4_cpntf_state, cp_list);
1044                 _free_cpntf_state_locked(nn, cps);
1045         }
1046         spin_unlock(&nn->s2s_cp_lock);
1047 }
1048
1049 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1050 {
1051         struct nfs4_stid *stid;
1052
1053         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1054         if (!stid)
1055                 return NULL;
1056
1057         return openlockstateid(stid);
1058 }
1059
1060 static void nfs4_free_deleg(struct nfs4_stid *stid)
1061 {
1062         struct nfs4_delegation *dp = delegstateid(stid);
1063
1064         WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1065         WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1066         WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1067         WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1068         kmem_cache_free(deleg_slab, stid);
1069         atomic_long_dec(&num_delegations);
1070 }
1071
1072 /*
1073  * When we recall a delegation, we should be careful not to hand it
1074  * out again straight away.
1075  * To ensure this we keep a pair of bloom filters ('new' and 'old')
1076  * in which the filehandles of recalled delegations are "stored".
1077  * If a filehandle appear in either filter, a delegation is blocked.
1078  * When a delegation is recalled, the filehandle is stored in the "new"
1079  * filter.
1080  * Every 30 seconds we swap the filters and clear the "new" one,
1081  * unless both are empty of course.  This results in delegations for a
1082  * given filehandle being blocked for between 30 and 60 seconds.
1083  *
1084  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
1085  * low 3 bytes as hash-table indices.
1086  *
1087  * 'blocked_delegations_lock', which is always taken in block_delegations(),
1088  * is used to manage concurrent access.  Testing does not need the lock
1089  * except when swapping the two filters.
1090  */
1091 static DEFINE_SPINLOCK(blocked_delegations_lock);
1092 static struct bloom_pair {
1093         int     entries, old_entries;
1094         time64_t swap_time;
1095         int     new; /* index into 'set' */
1096         DECLARE_BITMAP(set[2], 256);
1097 } blocked_delegations;
1098
1099 static int delegation_blocked(struct knfsd_fh *fh)
1100 {
1101         u32 hash;
1102         struct bloom_pair *bd = &blocked_delegations;
1103
1104         if (bd->entries == 0)
1105                 return 0;
1106         if (ktime_get_seconds() - bd->swap_time > 30) {
1107                 spin_lock(&blocked_delegations_lock);
1108                 if (ktime_get_seconds() - bd->swap_time > 30) {
1109                         bd->entries -= bd->old_entries;
1110                         bd->old_entries = bd->entries;
1111                         bd->new = 1-bd->new;
1112                         memset(bd->set[bd->new], 0,
1113                                sizeof(bd->set[0]));
1114                         bd->swap_time = ktime_get_seconds();
1115                 }
1116                 spin_unlock(&blocked_delegations_lock);
1117         }
1118         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1119         if (test_bit(hash&255, bd->set[0]) &&
1120             test_bit((hash>>8)&255, bd->set[0]) &&
1121             test_bit((hash>>16)&255, bd->set[0]))
1122                 return 1;
1123
1124         if (test_bit(hash&255, bd->set[1]) &&
1125             test_bit((hash>>8)&255, bd->set[1]) &&
1126             test_bit((hash>>16)&255, bd->set[1]))
1127                 return 1;
1128
1129         return 0;
1130 }
1131
1132 static void block_delegations(struct knfsd_fh *fh)
1133 {
1134         u32 hash;
1135         struct bloom_pair *bd = &blocked_delegations;
1136
1137         hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1138
1139         spin_lock(&blocked_delegations_lock);
1140         __set_bit(hash&255, bd->set[bd->new]);
1141         __set_bit((hash>>8)&255, bd->set[bd->new]);
1142         __set_bit((hash>>16)&255, bd->set[bd->new]);
1143         if (bd->entries == 0)
1144                 bd->swap_time = ktime_get_seconds();
1145         bd->entries += 1;
1146         spin_unlock(&blocked_delegations_lock);
1147 }
1148
1149 static struct nfs4_delegation *
1150 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1151                  struct nfs4_clnt_odstate *odstate, u32 dl_type)
1152 {
1153         struct nfs4_delegation *dp;
1154         struct nfs4_stid *stid;
1155         long n;
1156
1157         dprintk("NFSD alloc_init_deleg\n");
1158         n = atomic_long_inc_return(&num_delegations);
1159         if (n < 0 || n > max_delegations)
1160                 goto out_dec;
1161         if (delegation_blocked(&fp->fi_fhandle))
1162                 goto out_dec;
1163         stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1164         if (stid == NULL)
1165                 goto out_dec;
1166         dp = delegstateid(stid);
1167
1168         /*
1169          * delegation seqid's are never incremented.  The 4.1 special
1170          * meaning of seqid 0 isn't meaningful, really, but let's avoid
1171          * 0 anyway just for consistency and use 1:
1172          */
1173         dp->dl_stid.sc_stateid.si_generation = 1;
1174         INIT_LIST_HEAD(&dp->dl_perfile);
1175         INIT_LIST_HEAD(&dp->dl_perclnt);
1176         INIT_LIST_HEAD(&dp->dl_recall_lru);
1177         dp->dl_clnt_odstate = odstate;
1178         get_clnt_odstate(odstate);
1179         dp->dl_type = dl_type;
1180         dp->dl_retries = 1;
1181         dp->dl_recalled = false;
1182         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1183                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1184         nfsd4_init_cb(&dp->dl_cb_fattr.ncf_getattr, dp->dl_stid.sc_client,
1185                         &nfsd4_cb_getattr_ops, NFSPROC4_CLNT_CB_GETATTR);
1186         dp->dl_cb_fattr.ncf_file_modified = false;
1187         dp->dl_cb_fattr.ncf_cb_bmap[0] = FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE;
1188         get_nfs4_file(fp);
1189         dp->dl_stid.sc_file = fp;
1190         return dp;
1191 out_dec:
1192         atomic_long_dec(&num_delegations);
1193         return NULL;
1194 }
1195
1196 void
1197 nfs4_put_stid(struct nfs4_stid *s)
1198 {
1199         struct nfs4_file *fp = s->sc_file;
1200         struct nfs4_client *clp = s->sc_client;
1201
1202         might_lock(&clp->cl_lock);
1203
1204         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1205                 wake_up_all(&close_wq);
1206                 return;
1207         }
1208         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1209         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1210                 atomic_dec(&s->sc_client->cl_admin_revoked);
1211         nfs4_free_cpntf_statelist(clp->net, s);
1212         spin_unlock(&clp->cl_lock);
1213         s->sc_free(s);
1214         if (fp)
1215                 put_nfs4_file(fp);
1216 }
1217
1218 void
1219 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1220 {
1221         stateid_t *src = &stid->sc_stateid;
1222
1223         spin_lock(&stid->sc_lock);
1224         if (unlikely(++src->si_generation == 0))
1225                 src->si_generation = 1;
1226         memcpy(dst, src, sizeof(*dst));
1227         spin_unlock(&stid->sc_lock);
1228 }
1229
1230 static void put_deleg_file(struct nfs4_file *fp)
1231 {
1232         struct nfsd_file *nf = NULL;
1233
1234         spin_lock(&fp->fi_lock);
1235         if (--fp->fi_delegees == 0)
1236                 swap(nf, fp->fi_deleg_file);
1237         spin_unlock(&fp->fi_lock);
1238
1239         if (nf)
1240                 nfsd_file_put(nf);
1241 }
1242
1243 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1244 {
1245         struct nfs4_file *fp = dp->dl_stid.sc_file;
1246         struct nfsd_file *nf = fp->fi_deleg_file;
1247
1248         WARN_ON_ONCE(!fp->fi_delegees);
1249
1250         kernel_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1251         put_deleg_file(fp);
1252 }
1253
1254 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1255 {
1256         put_clnt_odstate(dp->dl_clnt_odstate);
1257         nfs4_unlock_deleg_lease(dp);
1258         nfs4_put_stid(&dp->dl_stid);
1259 }
1260
1261 /**
1262  * nfs4_delegation_exists - Discover if this delegation already exists
1263  * @clp:     a pointer to the nfs4_client we're granting a delegation to
1264  * @fp:      a pointer to the nfs4_file we're granting a delegation on
1265  *
1266  * Return:
1267  *      On success: true iff an existing delegation is found
1268  */
1269
1270 static bool
1271 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1272 {
1273         struct nfs4_delegation *searchdp = NULL;
1274         struct nfs4_client *searchclp = NULL;
1275
1276         lockdep_assert_held(&state_lock);
1277         lockdep_assert_held(&fp->fi_lock);
1278
1279         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1280                 searchclp = searchdp->dl_stid.sc_client;
1281                 if (clp == searchclp) {
1282                         return true;
1283                 }
1284         }
1285         return false;
1286 }
1287
1288 /**
1289  * hash_delegation_locked - Add a delegation to the appropriate lists
1290  * @dp:     a pointer to the nfs4_delegation we are adding.
1291  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1292  *
1293  * Return:
1294  *      On success: NULL if the delegation was successfully hashed.
1295  *
1296  *      On error: -EAGAIN if one was previously granted to this
1297  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1298  *
1299  */
1300
1301 static int
1302 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1303 {
1304         struct nfs4_client *clp = dp->dl_stid.sc_client;
1305
1306         lockdep_assert_held(&state_lock);
1307         lockdep_assert_held(&fp->fi_lock);
1308         lockdep_assert_held(&clp->cl_lock);
1309
1310         if (nfs4_delegation_exists(clp, fp))
1311                 return -EAGAIN;
1312         refcount_inc(&dp->dl_stid.sc_count);
1313         dp->dl_stid.sc_type = SC_TYPE_DELEG;
1314         list_add(&dp->dl_perfile, &fp->fi_delegations);
1315         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1316         return 0;
1317 }
1318
1319 static bool delegation_hashed(struct nfs4_delegation *dp)
1320 {
1321         return !(list_empty(&dp->dl_perfile));
1322 }
1323
1324 static bool
1325 unhash_delegation_locked(struct nfs4_delegation *dp, unsigned short statusmask)
1326 {
1327         struct nfs4_file *fp = dp->dl_stid.sc_file;
1328
1329         lockdep_assert_held(&state_lock);
1330
1331         if (!delegation_hashed(dp))
1332                 return false;
1333
1334         if (statusmask == SC_STATUS_REVOKED &&
1335             dp->dl_stid.sc_client->cl_minorversion == 0)
1336                 statusmask = SC_STATUS_CLOSED;
1337         dp->dl_stid.sc_status |= statusmask;
1338         if (statusmask & SC_STATUS_ADMIN_REVOKED)
1339                 atomic_inc(&dp->dl_stid.sc_client->cl_admin_revoked);
1340
1341         /* Ensure that deleg break won't try to requeue it */
1342         ++dp->dl_time;
1343         spin_lock(&fp->fi_lock);
1344         list_del_init(&dp->dl_perclnt);
1345         list_del_init(&dp->dl_recall_lru);
1346         list_del_init(&dp->dl_perfile);
1347         spin_unlock(&fp->fi_lock);
1348         return true;
1349 }
1350
1351 static void destroy_delegation(struct nfs4_delegation *dp)
1352 {
1353         bool unhashed;
1354
1355         spin_lock(&state_lock);
1356         unhashed = unhash_delegation_locked(dp, SC_STATUS_CLOSED);
1357         spin_unlock(&state_lock);
1358         if (unhashed)
1359                 destroy_unhashed_deleg(dp);
1360 }
1361
1362 static void revoke_delegation(struct nfs4_delegation *dp)
1363 {
1364         struct nfs4_client *clp = dp->dl_stid.sc_client;
1365
1366         WARN_ON(!list_empty(&dp->dl_recall_lru));
1367
1368         trace_nfsd_stid_revoke(&dp->dl_stid);
1369
1370         if (dp->dl_stid.sc_status &
1371             (SC_STATUS_REVOKED | SC_STATUS_ADMIN_REVOKED)) {
1372                 spin_lock(&clp->cl_lock);
1373                 refcount_inc(&dp->dl_stid.sc_count);
1374                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1375                 spin_unlock(&clp->cl_lock);
1376         }
1377         destroy_unhashed_deleg(dp);
1378 }
1379
1380 /*
1381  * SETCLIENTID state
1382  */
1383
1384 static unsigned int clientid_hashval(u32 id)
1385 {
1386         return id & CLIENT_HASH_MASK;
1387 }
1388
1389 static unsigned int clientstr_hashval(struct xdr_netobj name)
1390 {
1391         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1392 }
1393
1394 /*
1395  * A stateid that had a deny mode associated with it is being released
1396  * or downgraded. Recalculate the deny mode on the file.
1397  */
1398 static void
1399 recalculate_deny_mode(struct nfs4_file *fp)
1400 {
1401         struct nfs4_ol_stateid *stp;
1402         u32 old_deny;
1403
1404         spin_lock(&fp->fi_lock);
1405         old_deny = fp->fi_share_deny;
1406         fp->fi_share_deny = 0;
1407         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1408                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1409                 if (fp->fi_share_deny == old_deny)
1410                         break;
1411         }
1412         spin_unlock(&fp->fi_lock);
1413 }
1414
1415 static void
1416 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1417 {
1418         int i;
1419         bool change = false;
1420
1421         for (i = 1; i < 4; i++) {
1422                 if ((i & deny) != i) {
1423                         change = true;
1424                         clear_deny(i, stp);
1425                 }
1426         }
1427
1428         /* Recalculate per-file deny mode if there was a change */
1429         if (change)
1430                 recalculate_deny_mode(stp->st_stid.sc_file);
1431 }
1432
1433 /* release all access and file references for a given stateid */
1434 static void
1435 release_all_access(struct nfs4_ol_stateid *stp)
1436 {
1437         int i;
1438         struct nfs4_file *fp = stp->st_stid.sc_file;
1439
1440         if (fp && stp->st_deny_bmap != 0)
1441                 recalculate_deny_mode(fp);
1442
1443         for (i = 1; i < 4; i++) {
1444                 if (test_access(i, stp))
1445                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1446                 clear_access(i, stp);
1447         }
1448 }
1449
1450 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1451 {
1452         kfree(sop->so_owner.data);
1453         sop->so_ops->so_free(sop);
1454 }
1455
1456 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1457 {
1458         struct nfs4_client *clp = sop->so_client;
1459
1460         might_lock(&clp->cl_lock);
1461
1462         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1463                 return;
1464         sop->so_ops->so_unhash(sop);
1465         spin_unlock(&clp->cl_lock);
1466         nfs4_free_stateowner(sop);
1467 }
1468
1469 static bool
1470 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1471 {
1472         return list_empty(&stp->st_perfile);
1473 }
1474
1475 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1476 {
1477         struct nfs4_file *fp = stp->st_stid.sc_file;
1478
1479         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1480
1481         if (list_empty(&stp->st_perfile))
1482                 return false;
1483
1484         spin_lock(&fp->fi_lock);
1485         list_del_init(&stp->st_perfile);
1486         spin_unlock(&fp->fi_lock);
1487         list_del(&stp->st_perstateowner);
1488         return true;
1489 }
1490
1491 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1492 {
1493         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1494
1495         put_clnt_odstate(stp->st_clnt_odstate);
1496         release_all_access(stp);
1497         if (stp->st_stateowner)
1498                 nfs4_put_stateowner(stp->st_stateowner);
1499         WARN_ON(!list_empty(&stid->sc_cp_list));
1500         kmem_cache_free(stateid_slab, stid);
1501 }
1502
1503 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1504 {
1505         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1506         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1507         struct nfsd_file *nf;
1508
1509         nf = find_any_file(stp->st_stid.sc_file);
1510         if (nf) {
1511                 get_file(nf->nf_file);
1512                 filp_close(nf->nf_file, (fl_owner_t)lo);
1513                 nfsd_file_put(nf);
1514         }
1515         nfs4_free_ol_stateid(stid);
1516 }
1517
1518 /*
1519  * Put the persistent reference to an already unhashed generic stateid, while
1520  * holding the cl_lock. If it's the last reference, then put it onto the
1521  * reaplist for later destruction.
1522  */
1523 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1524                                        struct list_head *reaplist)
1525 {
1526         struct nfs4_stid *s = &stp->st_stid;
1527         struct nfs4_client *clp = s->sc_client;
1528
1529         lockdep_assert_held(&clp->cl_lock);
1530
1531         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1532
1533         if (!refcount_dec_and_test(&s->sc_count)) {
1534                 wake_up_all(&close_wq);
1535                 return;
1536         }
1537
1538         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1539         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1540                 atomic_dec(&s->sc_client->cl_admin_revoked);
1541         list_add(&stp->st_locks, reaplist);
1542 }
1543
1544 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1545 {
1546         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1547
1548         if (!unhash_ol_stateid(stp))
1549                 return false;
1550         list_del_init(&stp->st_locks);
1551         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1552         return true;
1553 }
1554
1555 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1556 {
1557         struct nfs4_client *clp = stp->st_stid.sc_client;
1558         bool unhashed;
1559
1560         spin_lock(&clp->cl_lock);
1561         unhashed = unhash_lock_stateid(stp);
1562         spin_unlock(&clp->cl_lock);
1563         if (unhashed)
1564                 nfs4_put_stid(&stp->st_stid);
1565 }
1566
1567 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1568 {
1569         struct nfs4_client *clp = lo->lo_owner.so_client;
1570
1571         lockdep_assert_held(&clp->cl_lock);
1572
1573         list_del_init(&lo->lo_owner.so_strhash);
1574 }
1575
1576 /*
1577  * Free a list of generic stateids that were collected earlier after being
1578  * fully unhashed.
1579  */
1580 static void
1581 free_ol_stateid_reaplist(struct list_head *reaplist)
1582 {
1583         struct nfs4_ol_stateid *stp;
1584         struct nfs4_file *fp;
1585
1586         might_sleep();
1587
1588         while (!list_empty(reaplist)) {
1589                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1590                                        st_locks);
1591                 list_del(&stp->st_locks);
1592                 fp = stp->st_stid.sc_file;
1593                 stp->st_stid.sc_free(&stp->st_stid);
1594                 if (fp)
1595                         put_nfs4_file(fp);
1596         }
1597 }
1598
1599 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1600                                        struct list_head *reaplist)
1601 {
1602         struct nfs4_ol_stateid *stp;
1603
1604         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1605
1606         while (!list_empty(&open_stp->st_locks)) {
1607                 stp = list_entry(open_stp->st_locks.next,
1608                                 struct nfs4_ol_stateid, st_locks);
1609                 unhash_lock_stateid(stp);
1610                 put_ol_stateid_locked(stp, reaplist);
1611         }
1612 }
1613
1614 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1615                                 struct list_head *reaplist)
1616 {
1617         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1618
1619         if (!unhash_ol_stateid(stp))
1620                 return false;
1621         release_open_stateid_locks(stp, reaplist);
1622         return true;
1623 }
1624
1625 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1626 {
1627         LIST_HEAD(reaplist);
1628
1629         spin_lock(&stp->st_stid.sc_client->cl_lock);
1630         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1631         if (unhash_open_stateid(stp, &reaplist))
1632                 put_ol_stateid_locked(stp, &reaplist);
1633         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1634         free_ol_stateid_reaplist(&reaplist);
1635 }
1636
1637 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1638 {
1639         struct nfs4_client *clp = oo->oo_owner.so_client;
1640
1641         lockdep_assert_held(&clp->cl_lock);
1642
1643         list_del_init(&oo->oo_owner.so_strhash);
1644         list_del_init(&oo->oo_perclient);
1645 }
1646
1647 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1648 {
1649         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1650                                           nfsd_net_id);
1651         struct nfs4_ol_stateid *s;
1652
1653         spin_lock(&nn->client_lock);
1654         s = oo->oo_last_closed_stid;
1655         if (s) {
1656                 list_del_init(&oo->oo_close_lru);
1657                 oo->oo_last_closed_stid = NULL;
1658         }
1659         spin_unlock(&nn->client_lock);
1660         if (s)
1661                 nfs4_put_stid(&s->st_stid);
1662 }
1663
1664 static void release_openowner(struct nfs4_openowner *oo)
1665 {
1666         struct nfs4_ol_stateid *stp;
1667         struct nfs4_client *clp = oo->oo_owner.so_client;
1668         LIST_HEAD(reaplist);
1669
1670         spin_lock(&clp->cl_lock);
1671         unhash_openowner_locked(oo);
1672         while (!list_empty(&oo->oo_owner.so_stateids)) {
1673                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1674                                 struct nfs4_ol_stateid, st_perstateowner);
1675                 if (unhash_open_stateid(stp, &reaplist))
1676                         put_ol_stateid_locked(stp, &reaplist);
1677         }
1678         spin_unlock(&clp->cl_lock);
1679         free_ol_stateid_reaplist(&reaplist);
1680         release_last_closed_stateid(oo);
1681         nfs4_put_stateowner(&oo->oo_owner);
1682 }
1683
1684 static struct nfs4_stid *find_one_sb_stid(struct nfs4_client *clp,
1685                                           struct super_block *sb,
1686                                           unsigned int sc_types)
1687 {
1688         unsigned long id, tmp;
1689         struct nfs4_stid *stid;
1690
1691         spin_lock(&clp->cl_lock);
1692         idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
1693                 if ((stid->sc_type & sc_types) &&
1694                     stid->sc_status == 0 &&
1695                     stid->sc_file->fi_inode->i_sb == sb) {
1696                         refcount_inc(&stid->sc_count);
1697                         break;
1698                 }
1699         spin_unlock(&clp->cl_lock);
1700         return stid;
1701 }
1702
1703 /**
1704  * nfsd4_revoke_states - revoke all nfsv4 states associated with given filesystem
1705  * @net:  used to identify instance of nfsd (there is one per net namespace)
1706  * @sb:   super_block used to identify target filesystem
1707  *
1708  * All nfs4 states (open, lock, delegation, layout) held by the server instance
1709  * and associated with a file on the given filesystem will be revoked resulting
1710  * in any files being closed and so all references from nfsd to the filesystem
1711  * being released.  Thus nfsd will no longer prevent the filesystem from being
1712  * unmounted.
1713  *
1714  * The clients which own the states will subsequently being notified that the
1715  * states have been "admin-revoked".
1716  */
1717 void nfsd4_revoke_states(struct net *net, struct super_block *sb)
1718 {
1719         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1720         unsigned int idhashval;
1721         unsigned int sc_types;
1722
1723         sc_types = SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG | SC_TYPE_LAYOUT;
1724
1725         spin_lock(&nn->client_lock);
1726         for (idhashval = 0; idhashval < CLIENT_HASH_MASK; idhashval++) {
1727                 struct list_head *head = &nn->conf_id_hashtbl[idhashval];
1728                 struct nfs4_client *clp;
1729         retry:
1730                 list_for_each_entry(clp, head, cl_idhash) {
1731                         struct nfs4_stid *stid = find_one_sb_stid(clp, sb,
1732                                                                   sc_types);
1733                         if (stid) {
1734                                 struct nfs4_ol_stateid *stp;
1735                                 struct nfs4_delegation *dp;
1736                                 struct nfs4_layout_stateid *ls;
1737
1738                                 spin_unlock(&nn->client_lock);
1739                                 switch (stid->sc_type) {
1740                                 case SC_TYPE_OPEN:
1741                                         stp = openlockstateid(stid);
1742                                         mutex_lock_nested(&stp->st_mutex,
1743                                                           OPEN_STATEID_MUTEX);
1744
1745                                         spin_lock(&clp->cl_lock);
1746                                         if (stid->sc_status == 0) {
1747                                                 stid->sc_status |=
1748                                                         SC_STATUS_ADMIN_REVOKED;
1749                                                 atomic_inc(&clp->cl_admin_revoked);
1750                                                 spin_unlock(&clp->cl_lock);
1751                                                 release_all_access(stp);
1752                                         } else
1753                                                 spin_unlock(&clp->cl_lock);
1754                                         mutex_unlock(&stp->st_mutex);
1755                                         break;
1756                                 case SC_TYPE_LOCK:
1757                                         stp = openlockstateid(stid);
1758                                         mutex_lock_nested(&stp->st_mutex,
1759                                                           LOCK_STATEID_MUTEX);
1760                                         spin_lock(&clp->cl_lock);
1761                                         if (stid->sc_status == 0) {
1762                                                 struct nfs4_lockowner *lo =
1763                                                         lockowner(stp->st_stateowner);
1764                                                 struct nfsd_file *nf;
1765
1766                                                 stid->sc_status |=
1767                                                         SC_STATUS_ADMIN_REVOKED;
1768                                                 atomic_inc(&clp->cl_admin_revoked);
1769                                                 spin_unlock(&clp->cl_lock);
1770                                                 nf = find_any_file(stp->st_stid.sc_file);
1771                                                 if (nf) {
1772                                                         get_file(nf->nf_file);
1773                                                         filp_close(nf->nf_file,
1774                                                                    (fl_owner_t)lo);
1775                                                         nfsd_file_put(nf);
1776                                                 }
1777                                                 release_all_access(stp);
1778                                         } else
1779                                                 spin_unlock(&clp->cl_lock);
1780                                         mutex_unlock(&stp->st_mutex);
1781                                         break;
1782                                 case SC_TYPE_DELEG:
1783                                         dp = delegstateid(stid);
1784                                         spin_lock(&state_lock);
1785                                         if (!unhash_delegation_locked(
1786                                                     dp, SC_STATUS_ADMIN_REVOKED))
1787                                                 dp = NULL;
1788                                         spin_unlock(&state_lock);
1789                                         if (dp)
1790                                                 revoke_delegation(dp);
1791                                         break;
1792                                 case SC_TYPE_LAYOUT:
1793                                         ls = layoutstateid(stid);
1794                                         nfsd4_close_layout(ls);
1795                                         break;
1796                                 }
1797                                 nfs4_put_stid(stid);
1798                                 spin_lock(&nn->client_lock);
1799                                 if (clp->cl_minorversion == 0)
1800                                         /* Allow cleanup after a lease period.
1801                                          * store_release ensures cleanup will
1802                                          * see any newly revoked states if it
1803                                          * sees the time updated.
1804                                          */
1805                                         nn->nfs40_last_revoke =
1806                                                 ktime_get_boottime_seconds();
1807                                 goto retry;
1808                         }
1809                 }
1810         }
1811         spin_unlock(&nn->client_lock);
1812 }
1813
1814 static inline int
1815 hash_sessionid(struct nfs4_sessionid *sessionid)
1816 {
1817         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1818
1819         return sid->sequence % SESSION_HASH_SIZE;
1820 }
1821
1822 #ifdef CONFIG_SUNRPC_DEBUG
1823 static inline void
1824 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1825 {
1826         u32 *ptr = (u32 *)(&sessionid->data[0]);
1827         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1828 }
1829 #else
1830 static inline void
1831 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1832 {
1833 }
1834 #endif
1835
1836 /*
1837  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1838  * won't be used for replay.
1839  */
1840 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1841 {
1842         struct nfs4_stateowner *so = cstate->replay_owner;
1843
1844         if (nfserr == nfserr_replay_me)
1845                 return;
1846
1847         if (!seqid_mutating_err(ntohl(nfserr))) {
1848                 nfsd4_cstate_clear_replay(cstate);
1849                 return;
1850         }
1851         if (!so)
1852                 return;
1853         if (so->so_is_open_owner)
1854                 release_last_closed_stateid(openowner(so));
1855         so->so_seqid++;
1856         return;
1857 }
1858
1859 static void
1860 gen_sessionid(struct nfsd4_session *ses)
1861 {
1862         struct nfs4_client *clp = ses->se_client;
1863         struct nfsd4_sessionid *sid;
1864
1865         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1866         sid->clientid = clp->cl_clientid;
1867         sid->sequence = current_sessionid++;
1868         sid->reserved = 0;
1869 }
1870
1871 /*
1872  * The protocol defines ca_maxresponssize_cached to include the size of
1873  * the rpc header, but all we need to cache is the data starting after
1874  * the end of the initial SEQUENCE operation--the rest we regenerate
1875  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1876  * value that is the number of bytes in our cache plus a few additional
1877  * bytes.  In order to stay on the safe side, and not promise more than
1878  * we can cache, those additional bytes must be the minimum possible: 24
1879  * bytes of rpc header (xid through accept state, with AUTH_NULL
1880  * verifier), 12 for the compound header (with zero-length tag), and 44
1881  * for the SEQUENCE op response:
1882  */
1883 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1884
1885 static void
1886 free_session_slots(struct nfsd4_session *ses)
1887 {
1888         int i;
1889
1890         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1891                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1892                 kfree(ses->se_slots[i]);
1893         }
1894 }
1895
1896 /*
1897  * We don't actually need to cache the rpc and session headers, so we
1898  * can allocate a little less for each slot:
1899  */
1900 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1901 {
1902         u32 size;
1903
1904         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1905                 size = 0;
1906         else
1907                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1908         return size + sizeof(struct nfsd4_slot);
1909 }
1910
1911 /*
1912  * XXX: If we run out of reserved DRC memory we could (up to a point)
1913  * re-negotiate active sessions and reduce their slot usage to make
1914  * room for new connections. For now we just fail the create session.
1915  */
1916 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1917 {
1918         u32 slotsize = slot_bytes(ca);
1919         u32 num = ca->maxreqs;
1920         unsigned long avail, total_avail;
1921         unsigned int scale_factor;
1922
1923         spin_lock(&nfsd_drc_lock);
1924         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1925                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1926         else
1927                 /* We have handed out more space than we chose in
1928                  * set_max_drc() to allow.  That isn't really a
1929                  * problem as long as that doesn't make us think we
1930                  * have lots more due to integer overflow.
1931                  */
1932                 total_avail = 0;
1933         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1934         /*
1935          * Never use more than a fraction of the remaining memory,
1936          * unless it's the only way to give this client a slot.
1937          * The chosen fraction is either 1/8 or 1/number of threads,
1938          * whichever is smaller.  This ensures there are adequate
1939          * slots to support multiple clients per thread.
1940          * Give the client one slot even if that would require
1941          * over-allocation--it is better than failure.
1942          */
1943         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1944
1945         avail = clamp_t(unsigned long, avail, slotsize,
1946                         total_avail/scale_factor);
1947         num = min_t(int, num, avail / slotsize);
1948         num = max_t(int, num, 1);
1949         nfsd_drc_mem_used += num * slotsize;
1950         spin_unlock(&nfsd_drc_lock);
1951
1952         return num;
1953 }
1954
1955 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1956 {
1957         int slotsize = slot_bytes(ca);
1958
1959         spin_lock(&nfsd_drc_lock);
1960         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1961         spin_unlock(&nfsd_drc_lock);
1962 }
1963
1964 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1965                                            struct nfsd4_channel_attrs *battrs)
1966 {
1967         int numslots = fattrs->maxreqs;
1968         int slotsize = slot_bytes(fattrs);
1969         struct nfsd4_session *new;
1970         int i;
1971
1972         BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1973                      > PAGE_SIZE);
1974
1975         new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1976         if (!new)
1977                 return NULL;
1978         /* allocate each struct nfsd4_slot and data cache in one piece */
1979         for (i = 0; i < numslots; i++) {
1980                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1981                 if (!new->se_slots[i])
1982                         goto out_free;
1983         }
1984
1985         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1986         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1987
1988         return new;
1989 out_free:
1990         while (i--)
1991                 kfree(new->se_slots[i]);
1992         kfree(new);
1993         return NULL;
1994 }
1995
1996 static void free_conn(struct nfsd4_conn *c)
1997 {
1998         svc_xprt_put(c->cn_xprt);
1999         kfree(c);
2000 }
2001
2002 static void nfsd4_conn_lost(struct svc_xpt_user *u)
2003 {
2004         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
2005         struct nfs4_client *clp = c->cn_session->se_client;
2006
2007         trace_nfsd_cb_lost(clp);
2008
2009         spin_lock(&clp->cl_lock);
2010         if (!list_empty(&c->cn_persession)) {
2011                 list_del(&c->cn_persession);
2012                 free_conn(c);
2013         }
2014         nfsd4_probe_callback(clp);
2015         spin_unlock(&clp->cl_lock);
2016 }
2017
2018 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
2019 {
2020         struct nfsd4_conn *conn;
2021
2022         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
2023         if (!conn)
2024                 return NULL;
2025         svc_xprt_get(rqstp->rq_xprt);
2026         conn->cn_xprt = rqstp->rq_xprt;
2027         conn->cn_flags = flags;
2028         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
2029         return conn;
2030 }
2031
2032 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2033 {
2034         conn->cn_session = ses;
2035         list_add(&conn->cn_persession, &ses->se_conns);
2036 }
2037
2038 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2039 {
2040         struct nfs4_client *clp = ses->se_client;
2041
2042         spin_lock(&clp->cl_lock);
2043         __nfsd4_hash_conn(conn, ses);
2044         spin_unlock(&clp->cl_lock);
2045 }
2046
2047 static int nfsd4_register_conn(struct nfsd4_conn *conn)
2048 {
2049         conn->cn_xpt_user.callback = nfsd4_conn_lost;
2050         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
2051 }
2052
2053 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
2054 {
2055         int ret;
2056
2057         nfsd4_hash_conn(conn, ses);
2058         ret = nfsd4_register_conn(conn);
2059         if (ret)
2060                 /* oops; xprt is already down: */
2061                 nfsd4_conn_lost(&conn->cn_xpt_user);
2062         /* We may have gained or lost a callback channel: */
2063         nfsd4_probe_callback_sync(ses->se_client);
2064 }
2065
2066 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
2067 {
2068         u32 dir = NFS4_CDFC4_FORE;
2069
2070         if (cses->flags & SESSION4_BACK_CHAN)
2071                 dir |= NFS4_CDFC4_BACK;
2072         return alloc_conn(rqstp, dir);
2073 }
2074
2075 /* must be called under client_lock */
2076 static void nfsd4_del_conns(struct nfsd4_session *s)
2077 {
2078         struct nfs4_client *clp = s->se_client;
2079         struct nfsd4_conn *c;
2080
2081         spin_lock(&clp->cl_lock);
2082         while (!list_empty(&s->se_conns)) {
2083                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
2084                 list_del_init(&c->cn_persession);
2085                 spin_unlock(&clp->cl_lock);
2086
2087                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
2088                 free_conn(c);
2089
2090                 spin_lock(&clp->cl_lock);
2091         }
2092         spin_unlock(&clp->cl_lock);
2093 }
2094
2095 static void __free_session(struct nfsd4_session *ses)
2096 {
2097         free_session_slots(ses);
2098         kfree(ses);
2099 }
2100
2101 static void free_session(struct nfsd4_session *ses)
2102 {
2103         nfsd4_del_conns(ses);
2104         nfsd4_put_drc_mem(&ses->se_fchannel);
2105         __free_session(ses);
2106 }
2107
2108 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
2109 {
2110         int idx;
2111         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2112
2113         new->se_client = clp;
2114         gen_sessionid(new);
2115
2116         INIT_LIST_HEAD(&new->se_conns);
2117
2118         new->se_cb_seq_nr = 1;
2119         new->se_flags = cses->flags;
2120         new->se_cb_prog = cses->callback_prog;
2121         new->se_cb_sec = cses->cb_sec;
2122         atomic_set(&new->se_ref, 0);
2123         idx = hash_sessionid(&new->se_sessionid);
2124         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
2125         spin_lock(&clp->cl_lock);
2126         list_add(&new->se_perclnt, &clp->cl_sessions);
2127         spin_unlock(&clp->cl_lock);
2128
2129         {
2130                 struct sockaddr *sa = svc_addr(rqstp);
2131                 /*
2132                  * This is a little silly; with sessions there's no real
2133                  * use for the callback address.  Use the peer address
2134                  * as a reasonable default for now, but consider fixing
2135                  * the rpc client not to require an address in the
2136                  * future:
2137                  */
2138                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2139                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2140         }
2141 }
2142
2143 /* caller must hold client_lock */
2144 static struct nfsd4_session *
2145 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2146 {
2147         struct nfsd4_session *elem;
2148         int idx;
2149         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2150
2151         lockdep_assert_held(&nn->client_lock);
2152
2153         dump_sessionid(__func__, sessionid);
2154         idx = hash_sessionid(sessionid);
2155         /* Search in the appropriate list */
2156         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2157                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2158                             NFS4_MAX_SESSIONID_LEN)) {
2159                         return elem;
2160                 }
2161         }
2162
2163         dprintk("%s: session not found\n", __func__);
2164         return NULL;
2165 }
2166
2167 static struct nfsd4_session *
2168 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2169                 __be32 *ret)
2170 {
2171         struct nfsd4_session *session;
2172         __be32 status = nfserr_badsession;
2173
2174         session = __find_in_sessionid_hashtbl(sessionid, net);
2175         if (!session)
2176                 goto out;
2177         status = nfsd4_get_session_locked(session);
2178         if (status)
2179                 session = NULL;
2180 out:
2181         *ret = status;
2182         return session;
2183 }
2184
2185 /* caller must hold client_lock */
2186 static void
2187 unhash_session(struct nfsd4_session *ses)
2188 {
2189         struct nfs4_client *clp = ses->se_client;
2190         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2191
2192         lockdep_assert_held(&nn->client_lock);
2193
2194         list_del(&ses->se_hash);
2195         spin_lock(&ses->se_client->cl_lock);
2196         list_del(&ses->se_perclnt);
2197         spin_unlock(&ses->se_client->cl_lock);
2198 }
2199
2200 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2201 static int
2202 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2203 {
2204         /*
2205          * We're assuming the clid was not given out from a boot
2206          * precisely 2^32 (about 136 years) before this one.  That seems
2207          * a safe assumption:
2208          */
2209         if (clid->cl_boot == (u32)nn->boot_time)
2210                 return 0;
2211         trace_nfsd_clid_stale(clid);
2212         return 1;
2213 }
2214
2215 /*
2216  * XXX Should we use a slab cache ?
2217  * This type of memory management is somewhat inefficient, but we use it
2218  * anyway since SETCLIENTID is not a common operation.
2219  */
2220 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2221                                 struct nfsd_net *nn)
2222 {
2223         struct nfs4_client *clp;
2224         int i;
2225
2226         if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2227                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2228                 return NULL;
2229         }
2230         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2231         if (clp == NULL)
2232                 return NULL;
2233         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2234         if (clp->cl_name.data == NULL)
2235                 goto err_no_name;
2236         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2237                                                  sizeof(struct list_head),
2238                                                  GFP_KERNEL);
2239         if (!clp->cl_ownerstr_hashtbl)
2240                 goto err_no_hashtbl;
2241         clp->cl_callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
2242         if (!clp->cl_callback_wq)
2243                 goto err_no_callback_wq;
2244
2245         for (i = 0; i < OWNER_HASH_SIZE; i++)
2246                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2247         INIT_LIST_HEAD(&clp->cl_sessions);
2248         idr_init(&clp->cl_stateids);
2249         atomic_set(&clp->cl_rpc_users, 0);
2250         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2251         clp->cl_state = NFSD4_ACTIVE;
2252         atomic_inc(&nn->nfs4_client_count);
2253         atomic_set(&clp->cl_delegs_in_recall, 0);
2254         INIT_LIST_HEAD(&clp->cl_idhash);
2255         INIT_LIST_HEAD(&clp->cl_openowners);
2256         INIT_LIST_HEAD(&clp->cl_delegations);
2257         INIT_LIST_HEAD(&clp->cl_lru);
2258         INIT_LIST_HEAD(&clp->cl_revoked);
2259 #ifdef CONFIG_NFSD_PNFS
2260         INIT_LIST_HEAD(&clp->cl_lo_states);
2261 #endif
2262         INIT_LIST_HEAD(&clp->async_copies);
2263         spin_lock_init(&clp->async_lock);
2264         spin_lock_init(&clp->cl_lock);
2265         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2266         return clp;
2267 err_no_callback_wq:
2268         kfree(clp->cl_ownerstr_hashtbl);
2269 err_no_hashtbl:
2270         kfree(clp->cl_name.data);
2271 err_no_name:
2272         kmem_cache_free(client_slab, clp);
2273         return NULL;
2274 }
2275
2276 static void __free_client(struct kref *k)
2277 {
2278         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2279         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2280
2281         free_svc_cred(&clp->cl_cred);
2282         destroy_workqueue(clp->cl_callback_wq);
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         int users = atomic_read(&clp->cl_rpc_users);
2356
2357         trace_nfsd_mark_client_expired(clp, users);
2358
2359         if (users)
2360                 return nfserr_jukebox;
2361         unhash_client_locked(clp);
2362         return nfs_ok;
2363 }
2364
2365 static void
2366 __destroy_client(struct nfs4_client *clp)
2367 {
2368         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2369         int i;
2370         struct nfs4_openowner *oo;
2371         struct nfs4_delegation *dp;
2372         LIST_HEAD(reaplist);
2373
2374         spin_lock(&state_lock);
2375         while (!list_empty(&clp->cl_delegations)) {
2376                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2377                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
2378                 list_add(&dp->dl_recall_lru, &reaplist);
2379         }
2380         spin_unlock(&state_lock);
2381         while (!list_empty(&reaplist)) {
2382                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2383                 list_del_init(&dp->dl_recall_lru);
2384                 destroy_unhashed_deleg(dp);
2385         }
2386         while (!list_empty(&clp->cl_revoked)) {
2387                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2388                 list_del_init(&dp->dl_recall_lru);
2389                 nfs4_put_stid(&dp->dl_stid);
2390         }
2391         while (!list_empty(&clp->cl_openowners)) {
2392                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2393                 nfs4_get_stateowner(&oo->oo_owner);
2394                 release_openowner(oo);
2395         }
2396         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2397                 struct nfs4_stateowner *so, *tmp;
2398
2399                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2400                                          so_strhash) {
2401                         /* Should be no openowners at this point */
2402                         WARN_ON_ONCE(so->so_is_open_owner);
2403                         remove_blocked_locks(lockowner(so));
2404                 }
2405         }
2406         nfsd4_return_all_client_layouts(clp);
2407         nfsd4_shutdown_copy(clp);
2408         nfsd4_shutdown_callback(clp);
2409         if (clp->cl_cb_conn.cb_xprt)
2410                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2411         atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2412         nfsd4_dec_courtesy_client_count(nn, clp);
2413         free_client(clp);
2414         wake_up_all(&expiry_wq);
2415 }
2416
2417 static void
2418 destroy_client(struct nfs4_client *clp)
2419 {
2420         unhash_client(clp);
2421         __destroy_client(clp);
2422 }
2423
2424 static void inc_reclaim_complete(struct nfs4_client *clp)
2425 {
2426         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2427
2428         if (!nn->track_reclaim_completes)
2429                 return;
2430         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2431                 return;
2432         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2433                         nn->reclaim_str_hashtbl_size) {
2434                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2435                                 clp->net->ns.inum);
2436                 nfsd4_end_grace(nn);
2437         }
2438 }
2439
2440 static void expire_client(struct nfs4_client *clp)
2441 {
2442         unhash_client(clp);
2443         nfsd4_client_record_remove(clp);
2444         __destroy_client(clp);
2445 }
2446
2447 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2448 {
2449         memcpy(target->cl_verifier.data, source->data,
2450                         sizeof(target->cl_verifier.data));
2451 }
2452
2453 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2454 {
2455         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2456         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2457 }
2458
2459 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2460 {
2461         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2462         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2463                                                                 GFP_KERNEL);
2464         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2465         if ((source->cr_principal && !target->cr_principal) ||
2466             (source->cr_raw_principal && !target->cr_raw_principal) ||
2467             (source->cr_targ_princ && !target->cr_targ_princ))
2468                 return -ENOMEM;
2469
2470         target->cr_flavor = source->cr_flavor;
2471         target->cr_uid = source->cr_uid;
2472         target->cr_gid = source->cr_gid;
2473         target->cr_group_info = source->cr_group_info;
2474         get_group_info(target->cr_group_info);
2475         target->cr_gss_mech = source->cr_gss_mech;
2476         if (source->cr_gss_mech)
2477                 gss_mech_get(source->cr_gss_mech);
2478         return 0;
2479 }
2480
2481 static int
2482 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2483 {
2484         if (o1->len < o2->len)
2485                 return -1;
2486         if (o1->len > o2->len)
2487                 return 1;
2488         return memcmp(o1->data, o2->data, o1->len);
2489 }
2490
2491 static int
2492 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2493 {
2494         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2495 }
2496
2497 static int
2498 same_clid(clientid_t *cl1, clientid_t *cl2)
2499 {
2500         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2501 }
2502
2503 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2504 {
2505         int i;
2506
2507         if (g1->ngroups != g2->ngroups)
2508                 return false;
2509         for (i=0; i<g1->ngroups; i++)
2510                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2511                         return false;
2512         return true;
2513 }
2514
2515 /*
2516  * RFC 3530 language requires clid_inuse be returned when the
2517  * "principal" associated with a requests differs from that previously
2518  * used.  We use uid, gid's, and gss principal string as our best
2519  * approximation.  We also don't want to allow non-gss use of a client
2520  * established using gss: in theory cr_principal should catch that
2521  * change, but in practice cr_principal can be null even in the gss case
2522  * since gssd doesn't always pass down a principal string.
2523  */
2524 static bool is_gss_cred(struct svc_cred *cr)
2525 {
2526         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2527         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2528 }
2529
2530
2531 static bool
2532 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2533 {
2534         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2535                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2536                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2537                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2538                 return false;
2539         /* XXX: check that cr_targ_princ fields match ? */
2540         if (cr1->cr_principal == cr2->cr_principal)
2541                 return true;
2542         if (!cr1->cr_principal || !cr2->cr_principal)
2543                 return false;
2544         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2545 }
2546
2547 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2548 {
2549         struct svc_cred *cr = &rqstp->rq_cred;
2550         u32 service;
2551
2552         if (!cr->cr_gss_mech)
2553                 return false;
2554         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2555         return service == RPC_GSS_SVC_INTEGRITY ||
2556                service == RPC_GSS_SVC_PRIVACY;
2557 }
2558
2559 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2560 {
2561         struct svc_cred *cr = &rqstp->rq_cred;
2562
2563         if (!cl->cl_mach_cred)
2564                 return true;
2565         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2566                 return false;
2567         if (!svc_rqst_integrity_protected(rqstp))
2568                 return false;
2569         if (cl->cl_cred.cr_raw_principal)
2570                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2571                                                 cr->cr_raw_principal);
2572         if (!cr->cr_principal)
2573                 return false;
2574         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2575 }
2576
2577 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2578 {
2579         __be32 verf[2];
2580
2581         /*
2582          * This is opaque to client, so no need to byte-swap. Use
2583          * __force to keep sparse happy
2584          */
2585         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2586         verf[1] = (__force __be32)nn->clverifier_counter++;
2587         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2588 }
2589
2590 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2591 {
2592         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2593         clp->cl_clientid.cl_id = nn->clientid_counter++;
2594         gen_confirm(clp, nn);
2595 }
2596
2597 static struct nfs4_stid *
2598 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2599 {
2600         struct nfs4_stid *ret;
2601
2602         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2603         if (!ret || !ret->sc_type)
2604                 return NULL;
2605         return ret;
2606 }
2607
2608 static struct nfs4_stid *
2609 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t,
2610                      unsigned short typemask, unsigned short ok_states)
2611 {
2612         struct nfs4_stid *s;
2613
2614         spin_lock(&cl->cl_lock);
2615         s = find_stateid_locked(cl, t);
2616         if (s != NULL) {
2617                 if ((s->sc_status & ~ok_states) == 0 &&
2618                     (typemask & s->sc_type))
2619                         refcount_inc(&s->sc_count);
2620                 else
2621                         s = NULL;
2622         }
2623         spin_unlock(&cl->cl_lock);
2624         return s;
2625 }
2626
2627 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2628 {
2629         struct nfsdfs_client *nc;
2630         nc = get_nfsdfs_client(inode);
2631         if (!nc)
2632                 return NULL;
2633         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2634 }
2635
2636 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2637 {
2638         seq_puts(m, "\"");
2639         seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2640         seq_puts(m, "\"");
2641 }
2642
2643 static const char *cb_state2str(int state)
2644 {
2645         switch (state) {
2646         case NFSD4_CB_UP:
2647                 return "UP";
2648         case NFSD4_CB_UNKNOWN:
2649                 return "UNKNOWN";
2650         case NFSD4_CB_DOWN:
2651                 return "DOWN";
2652         case NFSD4_CB_FAULT:
2653                 return "FAULT";
2654         }
2655         return "UNDEFINED";
2656 }
2657
2658 static int client_info_show(struct seq_file *m, void *v)
2659 {
2660         struct inode *inode = file_inode(m->file);
2661         struct nfs4_client *clp;
2662         u64 clid;
2663
2664         clp = get_nfsdfs_clp(inode);
2665         if (!clp)
2666                 return -ENXIO;
2667         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2668         seq_printf(m, "clientid: 0x%llx\n", clid);
2669         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2670
2671         if (clp->cl_state == NFSD4_COURTESY)
2672                 seq_puts(m, "status: courtesy\n");
2673         else if (clp->cl_state == NFSD4_EXPIRABLE)
2674                 seq_puts(m, "status: expirable\n");
2675         else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2676                 seq_puts(m, "status: confirmed\n");
2677         else
2678                 seq_puts(m, "status: unconfirmed\n");
2679         seq_printf(m, "seconds from last renew: %lld\n",
2680                 ktime_get_boottime_seconds() - clp->cl_time);
2681         seq_puts(m, "name: ");
2682         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2683         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2684         if (clp->cl_nii_domain.data) {
2685                 seq_puts(m, "Implementation domain: ");
2686                 seq_quote_mem(m, clp->cl_nii_domain.data,
2687                                         clp->cl_nii_domain.len);
2688                 seq_puts(m, "\nImplementation name: ");
2689                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2690                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2691                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2692         }
2693         seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2694         seq_printf(m, "callback address: \"%pISpc\"\n", &clp->cl_cb_conn.cb_addr);
2695         seq_printf(m, "admin-revoked states: %d\n",
2696                    atomic_read(&clp->cl_admin_revoked));
2697         drop_client(clp);
2698
2699         return 0;
2700 }
2701
2702 DEFINE_SHOW_ATTRIBUTE(client_info);
2703
2704 static void *states_start(struct seq_file *s, loff_t *pos)
2705         __acquires(&clp->cl_lock)
2706 {
2707         struct nfs4_client *clp = s->private;
2708         unsigned long id = *pos;
2709         void *ret;
2710
2711         spin_lock(&clp->cl_lock);
2712         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2713         *pos = id;
2714         return ret;
2715 }
2716
2717 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2718 {
2719         struct nfs4_client *clp = s->private;
2720         unsigned long id = *pos;
2721         void *ret;
2722
2723         id = *pos;
2724         id++;
2725         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2726         *pos = id;
2727         return ret;
2728 }
2729
2730 static void states_stop(struct seq_file *s, void *v)
2731         __releases(&clp->cl_lock)
2732 {
2733         struct nfs4_client *clp = s->private;
2734
2735         spin_unlock(&clp->cl_lock);
2736 }
2737
2738 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2739 {
2740          seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2741 }
2742
2743 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2744 {
2745         struct inode *inode = file_inode(f->nf_file);
2746
2747         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2748                                         MAJOR(inode->i_sb->s_dev),
2749                                          MINOR(inode->i_sb->s_dev),
2750                                          inode->i_ino);
2751 }
2752
2753 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2754 {
2755         seq_puts(s, "owner: ");
2756         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2757 }
2758
2759 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2760 {
2761         seq_printf(s, "0x%.8x", stid->si_generation);
2762         seq_printf(s, "%12phN", &stid->si_opaque);
2763 }
2764
2765 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2766 {
2767         struct nfs4_ol_stateid *ols;
2768         struct nfs4_file *nf;
2769         struct nfsd_file *file;
2770         struct nfs4_stateowner *oo;
2771         unsigned int access, deny;
2772
2773         ols = openlockstateid(st);
2774         oo = ols->st_stateowner;
2775         nf = st->sc_file;
2776
2777         seq_puts(s, "- ");
2778         nfs4_show_stateid(s, &st->sc_stateid);
2779         seq_puts(s, ": { type: open, ");
2780
2781         access = bmap_to_share_mode(ols->st_access_bmap);
2782         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2783
2784         seq_printf(s, "access: %s%s, ",
2785                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2786                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2787         seq_printf(s, "deny: %s%s, ",
2788                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2789                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2790
2791         if (nf) {
2792                 spin_lock(&nf->fi_lock);
2793                 file = find_any_file_locked(nf);
2794                 if (file) {
2795                         nfs4_show_superblock(s, file);
2796                         seq_puts(s, ", ");
2797                         nfs4_show_fname(s, file);
2798                         seq_puts(s, ", ");
2799                 }
2800                 spin_unlock(&nf->fi_lock);
2801         } else
2802                 seq_puts(s, "closed, ");
2803         nfs4_show_owner(s, oo);
2804         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2805                 seq_puts(s, ", admin-revoked");
2806         seq_puts(s, " }\n");
2807         return 0;
2808 }
2809
2810 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2811 {
2812         struct nfs4_ol_stateid *ols;
2813         struct nfs4_file *nf;
2814         struct nfsd_file *file;
2815         struct nfs4_stateowner *oo;
2816
2817         ols = openlockstateid(st);
2818         oo = ols->st_stateowner;
2819         nf = st->sc_file;
2820
2821         seq_puts(s, "- ");
2822         nfs4_show_stateid(s, &st->sc_stateid);
2823         seq_puts(s, ": { type: lock, ");
2824
2825         spin_lock(&nf->fi_lock);
2826         file = find_any_file_locked(nf);
2827         if (file) {
2828                 /*
2829                  * Note: a lock stateid isn't really the same thing as a lock,
2830                  * it's the locking state held by one owner on a file, and there
2831                  * may be multiple (or no) lock ranges associated with it.
2832                  * (Same for the matter is true of open stateids.)
2833                  */
2834
2835                 nfs4_show_superblock(s, file);
2836                 /* XXX: open stateid? */
2837                 seq_puts(s, ", ");
2838                 nfs4_show_fname(s, file);
2839                 seq_puts(s, ", ");
2840         }
2841         nfs4_show_owner(s, oo);
2842         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2843                 seq_puts(s, ", admin-revoked");
2844         seq_puts(s, " }\n");
2845         spin_unlock(&nf->fi_lock);
2846         return 0;
2847 }
2848
2849 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2850 {
2851         struct nfs4_delegation *ds;
2852         struct nfs4_file *nf;
2853         struct nfsd_file *file;
2854
2855         ds = delegstateid(st);
2856         nf = st->sc_file;
2857
2858         seq_puts(s, "- ");
2859         nfs4_show_stateid(s, &st->sc_stateid);
2860         seq_puts(s, ": { type: deleg, ");
2861
2862         seq_printf(s, "access: %s",
2863                    ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2864
2865         /* XXX: lease time, whether it's being recalled. */
2866
2867         spin_lock(&nf->fi_lock);
2868         file = nf->fi_deleg_file;
2869         if (file) {
2870                 seq_puts(s, ", ");
2871                 nfs4_show_superblock(s, file);
2872                 seq_puts(s, ", ");
2873                 nfs4_show_fname(s, file);
2874         }
2875         spin_unlock(&nf->fi_lock);
2876         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2877                 seq_puts(s, ", admin-revoked");
2878         seq_puts(s, " }\n");
2879         return 0;
2880 }
2881
2882 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2883 {
2884         struct nfs4_layout_stateid *ls;
2885         struct nfsd_file *file;
2886
2887         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2888
2889         seq_puts(s, "- ");
2890         nfs4_show_stateid(s, &st->sc_stateid);
2891         seq_puts(s, ": { type: layout");
2892
2893         /* XXX: What else would be useful? */
2894
2895         spin_lock(&ls->ls_stid.sc_file->fi_lock);
2896         file = ls->ls_file;
2897         if (file) {
2898                 seq_puts(s, ", ");
2899                 nfs4_show_superblock(s, file);
2900                 seq_puts(s, ", ");
2901                 nfs4_show_fname(s, file);
2902         }
2903         spin_unlock(&ls->ls_stid.sc_file->fi_lock);
2904         if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2905                 seq_puts(s, ", admin-revoked");
2906         seq_puts(s, " }\n");
2907
2908         return 0;
2909 }
2910
2911 static int states_show(struct seq_file *s, void *v)
2912 {
2913         struct nfs4_stid *st = v;
2914
2915         switch (st->sc_type) {
2916         case SC_TYPE_OPEN:
2917                 return nfs4_show_open(s, st);
2918         case SC_TYPE_LOCK:
2919                 return nfs4_show_lock(s, st);
2920         case SC_TYPE_DELEG:
2921                 return nfs4_show_deleg(s, st);
2922         case SC_TYPE_LAYOUT:
2923                 return nfs4_show_layout(s, st);
2924         default:
2925                 return 0; /* XXX: or SEQ_SKIP? */
2926         }
2927         /* XXX: copy stateids? */
2928 }
2929
2930 static struct seq_operations states_seq_ops = {
2931         .start = states_start,
2932         .next = states_next,
2933         .stop = states_stop,
2934         .show = states_show
2935 };
2936
2937 static int client_states_open(struct inode *inode, struct file *file)
2938 {
2939         struct seq_file *s;
2940         struct nfs4_client *clp;
2941         int ret;
2942
2943         clp = get_nfsdfs_clp(inode);
2944         if (!clp)
2945                 return -ENXIO;
2946
2947         ret = seq_open(file, &states_seq_ops);
2948         if (ret)
2949                 return ret;
2950         s = file->private_data;
2951         s->private = clp;
2952         return 0;
2953 }
2954
2955 static int client_opens_release(struct inode *inode, struct file *file)
2956 {
2957         struct seq_file *m = file->private_data;
2958         struct nfs4_client *clp = m->private;
2959
2960         /* XXX: alternatively, we could get/drop in seq start/stop */
2961         drop_client(clp);
2962         return seq_release(inode, file);
2963 }
2964
2965 static const struct file_operations client_states_fops = {
2966         .open           = client_states_open,
2967         .read           = seq_read,
2968         .llseek         = seq_lseek,
2969         .release        = client_opens_release,
2970 };
2971
2972 /*
2973  * Normally we refuse to destroy clients that are in use, but here the
2974  * administrator is telling us to just do it.  We also want to wait
2975  * so the caller has a guarantee that the client's locks are gone by
2976  * the time the write returns:
2977  */
2978 static void force_expire_client(struct nfs4_client *clp)
2979 {
2980         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2981         bool already_expired;
2982
2983         trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2984
2985         spin_lock(&nn->client_lock);
2986         clp->cl_time = 0;
2987         spin_unlock(&nn->client_lock);
2988
2989         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2990         spin_lock(&nn->client_lock);
2991         already_expired = list_empty(&clp->cl_lru);
2992         if (!already_expired)
2993                 unhash_client_locked(clp);
2994         spin_unlock(&nn->client_lock);
2995
2996         if (!already_expired)
2997                 expire_client(clp);
2998         else
2999                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
3000 }
3001
3002 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
3003                                    size_t size, loff_t *pos)
3004 {
3005         char *data;
3006         struct nfs4_client *clp;
3007
3008         data = simple_transaction_get(file, buf, size);
3009         if (IS_ERR(data))
3010                 return PTR_ERR(data);
3011         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
3012                 return -EINVAL;
3013         clp = get_nfsdfs_clp(file_inode(file));
3014         if (!clp)
3015                 return -ENXIO;
3016         force_expire_client(clp);
3017         drop_client(clp);
3018         return 7;
3019 }
3020
3021 static const struct file_operations client_ctl_fops = {
3022         .write          = client_ctl_write,
3023         .release        = simple_transaction_release,
3024 };
3025
3026 static const struct tree_descr client_files[] = {
3027         [0] = {"info", &client_info_fops, S_IRUSR},
3028         [1] = {"states", &client_states_fops, S_IRUSR},
3029         [2] = {"ctl", &client_ctl_fops, S_IWUSR},
3030         [3] = {""},
3031 };
3032
3033 static int
3034 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
3035                                 struct rpc_task *task)
3036 {
3037         trace_nfsd_cb_recall_any_done(cb, task);
3038         switch (task->tk_status) {
3039         case -NFS4ERR_DELAY:
3040                 rpc_delay(task, 2 * HZ);
3041                 return 0;
3042         default:
3043                 return 1;
3044         }
3045 }
3046
3047 static void
3048 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
3049 {
3050         struct nfs4_client *clp = cb->cb_clp;
3051
3052         clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
3053         drop_client(clp);
3054 }
3055
3056 static int
3057 nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3058 {
3059         struct nfs4_cb_fattr *ncf =
3060                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3061         struct nfs4_delegation *dp =
3062                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3063
3064         trace_nfsd_cb_getattr_done(&dp->dl_stid.sc_stateid, task);
3065         ncf->ncf_cb_status = task->tk_status;
3066         switch (task->tk_status) {
3067         case -NFS4ERR_DELAY:
3068                 rpc_delay(task, 2 * HZ);
3069                 return 0;
3070         default:
3071                 return 1;
3072         }
3073 }
3074
3075 static void
3076 nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3077 {
3078         struct nfs4_cb_fattr *ncf =
3079                         container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3080         struct nfs4_delegation *dp =
3081                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3082
3083         clear_and_wake_up_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags);
3084         nfs4_put_stid(&dp->dl_stid);
3085 }
3086
3087 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3088         .done           = nfsd4_cb_recall_any_done,
3089         .release        = nfsd4_cb_recall_any_release,
3090         .opcode         = OP_CB_RECALL_ANY,
3091 };
3092
3093 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3094         .done           = nfsd4_cb_getattr_done,
3095         .release        = nfsd4_cb_getattr_release,
3096         .opcode         = OP_CB_GETATTR,
3097 };
3098
3099 static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3100 {
3101         struct nfs4_delegation *dp =
3102                         container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3103
3104         if (test_and_set_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags))
3105                 return;
3106         /* set to proper status when nfsd4_cb_getattr_done runs */
3107         ncf->ncf_cb_status = NFS4ERR_IO;
3108
3109         refcount_inc(&dp->dl_stid.sc_count);
3110         nfsd4_run_cb(&ncf->ncf_getattr);
3111 }
3112
3113 static struct nfs4_client *create_client(struct xdr_netobj name,
3114                 struct svc_rqst *rqstp, nfs4_verifier *verf)
3115 {
3116         struct nfs4_client *clp;
3117         struct sockaddr *sa = svc_addr(rqstp);
3118         int ret;
3119         struct net *net = SVC_NET(rqstp);
3120         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3121         struct dentry *dentries[ARRAY_SIZE(client_files)];
3122
3123         clp = alloc_client(name, nn);
3124         if (clp == NULL)
3125                 return NULL;
3126
3127         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3128         if (ret) {
3129                 free_client(clp);
3130                 return NULL;
3131         }
3132         gen_clid(clp, nn);
3133         kref_init(&clp->cl_nfsdfs.cl_ref);
3134         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3135         clp->cl_time = ktime_get_boottime_seconds();
3136         clear_bit(0, &clp->cl_cb_slot_busy);
3137         copy_verf(clp, verf);
3138         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3139         clp->cl_cb_session = NULL;
3140         clp->net = net;
3141         clp->cl_nfsd_dentry = nfsd_client_mkdir(
3142                 nn, &clp->cl_nfsdfs,
3143                 clp->cl_clientid.cl_id - nn->clientid_base,
3144                 client_files, dentries);
3145         clp->cl_nfsd_info_dentry = dentries[0];
3146         if (!clp->cl_nfsd_dentry) {
3147                 free_client(clp);
3148                 return NULL;
3149         }
3150         clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3151         if (!clp->cl_ra) {
3152                 free_client(clp);
3153                 return NULL;
3154         }
3155         clp->cl_ra_time = 0;
3156         nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3157                         NFSPROC4_CLNT_CB_RECALL_ANY);
3158         return clp;
3159 }
3160
3161 static void
3162 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3163 {
3164         struct rb_node **new = &(root->rb_node), *parent = NULL;
3165         struct nfs4_client *clp;
3166
3167         while (*new) {
3168                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3169                 parent = *new;
3170
3171                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3172                         new = &((*new)->rb_left);
3173                 else
3174                         new = &((*new)->rb_right);
3175         }
3176
3177         rb_link_node(&new_clp->cl_namenode, parent, new);
3178         rb_insert_color(&new_clp->cl_namenode, root);
3179 }
3180
3181 static struct nfs4_client *
3182 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3183 {
3184         int cmp;
3185         struct rb_node *node = root->rb_node;
3186         struct nfs4_client *clp;
3187
3188         while (node) {
3189                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
3190                 cmp = compare_blob(&clp->cl_name, name);
3191                 if (cmp > 0)
3192                         node = node->rb_left;
3193                 else if (cmp < 0)
3194                         node = node->rb_right;
3195                 else
3196                         return clp;
3197         }
3198         return NULL;
3199 }
3200
3201 static void
3202 add_to_unconfirmed(struct nfs4_client *clp)
3203 {
3204         unsigned int idhashval;
3205         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3206
3207         lockdep_assert_held(&nn->client_lock);
3208
3209         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3210         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3211         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3212         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3213         renew_client_locked(clp);
3214 }
3215
3216 static void
3217 move_to_confirmed(struct nfs4_client *clp)
3218 {
3219         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3220         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3221
3222         lockdep_assert_held(&nn->client_lock);
3223
3224         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3225         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3226         add_clp_to_name_tree(clp, &nn->conf_name_tree);
3227         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3228         trace_nfsd_clid_confirmed(&clp->cl_clientid);
3229         renew_client_locked(clp);
3230 }
3231
3232 static struct nfs4_client *
3233 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3234 {
3235         struct nfs4_client *clp;
3236         unsigned int idhashval = clientid_hashval(clid->cl_id);
3237
3238         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3239                 if (same_clid(&clp->cl_clientid, clid)) {
3240                         if ((bool)clp->cl_minorversion != sessions)
3241                                 return NULL;
3242                         renew_client_locked(clp);
3243                         return clp;
3244                 }
3245         }
3246         return NULL;
3247 }
3248
3249 static struct nfs4_client *
3250 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3251 {
3252         struct list_head *tbl = nn->conf_id_hashtbl;
3253
3254         lockdep_assert_held(&nn->client_lock);
3255         return find_client_in_id_table(tbl, clid, sessions);
3256 }
3257
3258 static struct nfs4_client *
3259 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3260 {
3261         struct list_head *tbl = nn->unconf_id_hashtbl;
3262
3263         lockdep_assert_held(&nn->client_lock);
3264         return find_client_in_id_table(tbl, clid, sessions);
3265 }
3266
3267 static bool clp_used_exchangeid(struct nfs4_client *clp)
3268 {
3269         return clp->cl_exchange_flags != 0;
3270 }
3271
3272 static struct nfs4_client *
3273 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3274 {
3275         lockdep_assert_held(&nn->client_lock);
3276         return find_clp_in_name_tree(name, &nn->conf_name_tree);
3277 }
3278
3279 static struct nfs4_client *
3280 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3281 {
3282         lockdep_assert_held(&nn->client_lock);
3283         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3284 }
3285
3286 static void
3287 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3288 {
3289         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3290         struct sockaddr *sa = svc_addr(rqstp);
3291         u32 scopeid = rpc_get_scope_id(sa);
3292         unsigned short expected_family;
3293
3294         /* Currently, we only support tcp and tcp6 for the callback channel */
3295         if (se->se_callback_netid_len == 3 &&
3296             !memcmp(se->se_callback_netid_val, "tcp", 3))
3297                 expected_family = AF_INET;
3298         else if (se->se_callback_netid_len == 4 &&
3299                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
3300                 expected_family = AF_INET6;
3301         else
3302                 goto out_err;
3303
3304         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3305                                             se->se_callback_addr_len,
3306                                             (struct sockaddr *)&conn->cb_addr,
3307                                             sizeof(conn->cb_addr));
3308
3309         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3310                 goto out_err;
3311
3312         if (conn->cb_addr.ss_family == AF_INET6)
3313                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3314
3315         conn->cb_prog = se->se_callback_prog;
3316         conn->cb_ident = se->se_callback_ident;
3317         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3318         trace_nfsd_cb_args(clp, conn);
3319         return;
3320 out_err:
3321         conn->cb_addr.ss_family = AF_UNSPEC;
3322         conn->cb_addrlen = 0;
3323         trace_nfsd_cb_nodelegs(clp);
3324         return;
3325 }
3326
3327 /*
3328  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3329  */
3330 static void
3331 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3332 {
3333         struct xdr_buf *buf = resp->xdr->buf;
3334         struct nfsd4_slot *slot = resp->cstate.slot;
3335         unsigned int base;
3336
3337         dprintk("--> %s slot %p\n", __func__, slot);
3338
3339         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3340         slot->sl_opcnt = resp->opcnt;
3341         slot->sl_status = resp->cstate.status;
3342         free_svc_cred(&slot->sl_cred);
3343         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3344
3345         if (!nfsd4_cache_this(resp)) {
3346                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3347                 return;
3348         }
3349         slot->sl_flags |= NFSD4_SLOT_CACHED;
3350
3351         base = resp->cstate.data_offset;
3352         slot->sl_datalen = buf->len - base;
3353         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3354                 WARN(1, "%s: sessions DRC could not cache compound\n",
3355                      __func__);
3356         return;
3357 }
3358
3359 /*
3360  * Encode the replay sequence operation from the slot values.
3361  * If cachethis is FALSE encode the uncached rep error on the next
3362  * operation which sets resp->p and increments resp->opcnt for
3363  * nfs4svc_encode_compoundres.
3364  *
3365  */
3366 static __be32
3367 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3368                           struct nfsd4_compoundres *resp)
3369 {
3370         struct nfsd4_op *op;
3371         struct nfsd4_slot *slot = resp->cstate.slot;
3372
3373         /* Encode the replayed sequence operation */
3374         op = &args->ops[resp->opcnt - 1];
3375         nfsd4_encode_operation(resp, op);
3376
3377         if (slot->sl_flags & NFSD4_SLOT_CACHED)
3378                 return op->status;
3379         if (args->opcnt == 1) {
3380                 /*
3381                  * The original operation wasn't a solo sequence--we
3382                  * always cache those--so this retry must not match the
3383                  * original:
3384                  */
3385                 op->status = nfserr_seq_false_retry;
3386         } else {
3387                 op = &args->ops[resp->opcnt++];
3388                 op->status = nfserr_retry_uncached_rep;
3389                 nfsd4_encode_operation(resp, op);
3390         }
3391         return op->status;
3392 }
3393
3394 /*
3395  * The sequence operation is not cached because we can use the slot and
3396  * session values.
3397  */
3398 static __be32
3399 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3400                          struct nfsd4_sequence *seq)
3401 {
3402         struct nfsd4_slot *slot = resp->cstate.slot;
3403         struct xdr_stream *xdr = resp->xdr;
3404         __be32 *p;
3405         __be32 status;
3406
3407         dprintk("--> %s slot %p\n", __func__, slot);
3408
3409         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3410         if (status)
3411                 return status;
3412
3413         p = xdr_reserve_space(xdr, slot->sl_datalen);
3414         if (!p) {
3415                 WARN_ON_ONCE(1);
3416                 return nfserr_serverfault;
3417         }
3418         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3419         xdr_commit_encode(xdr);
3420
3421         resp->opcnt = slot->sl_opcnt;
3422         return slot->sl_status;
3423 }
3424
3425 /*
3426  * Set the exchange_id flags returned by the server.
3427  */
3428 static void
3429 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3430 {
3431 #ifdef CONFIG_NFSD_PNFS
3432         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3433 #else
3434         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3435 #endif
3436
3437         /* Referrals are supported, Migration is not. */
3438         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3439
3440         /* set the wire flags to return to client. */
3441         clid->flags = new->cl_exchange_flags;
3442 }
3443
3444 static bool client_has_openowners(struct nfs4_client *clp)
3445 {
3446         struct nfs4_openowner *oo;
3447
3448         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3449                 if (!list_empty(&oo->oo_owner.so_stateids))
3450                         return true;
3451         }
3452         return false;
3453 }
3454
3455 static bool client_has_state(struct nfs4_client *clp)
3456 {
3457         return client_has_openowners(clp)
3458 #ifdef CONFIG_NFSD_PNFS
3459                 || !list_empty(&clp->cl_lo_states)
3460 #endif
3461                 || !list_empty(&clp->cl_delegations)
3462                 || !list_empty(&clp->cl_sessions)
3463                 || !list_empty(&clp->async_copies);
3464 }
3465
3466 static __be32 copy_impl_id(struct nfs4_client *clp,
3467                                 struct nfsd4_exchange_id *exid)
3468 {
3469         if (!exid->nii_domain.data)
3470                 return 0;
3471         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3472         if (!clp->cl_nii_domain.data)
3473                 return nfserr_jukebox;
3474         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3475         if (!clp->cl_nii_name.data)
3476                 return nfserr_jukebox;
3477         clp->cl_nii_time = exid->nii_time;
3478         return 0;
3479 }
3480
3481 __be32
3482 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3483                 union nfsd4_op_u *u)
3484 {
3485         struct nfsd4_exchange_id *exid = &u->exchange_id;
3486         struct nfs4_client *conf, *new;
3487         struct nfs4_client *unconf = NULL;
3488         __be32 status;
3489         char                    addr_str[INET6_ADDRSTRLEN];
3490         nfs4_verifier           verf = exid->verifier;
3491         struct sockaddr         *sa = svc_addr(rqstp);
3492         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3493         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3494
3495         rpc_ntop(sa, addr_str, sizeof(addr_str));
3496         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3497                 "ip_addr=%s flags %x, spa_how %u\n",
3498                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3499                 addr_str, exid->flags, exid->spa_how);
3500
3501         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3502                 return nfserr_inval;
3503
3504         new = create_client(exid->clname, rqstp, &verf);
3505         if (new == NULL)
3506                 return nfserr_jukebox;
3507         status = copy_impl_id(new, exid);
3508         if (status)
3509                 goto out_nolock;
3510
3511         switch (exid->spa_how) {
3512         case SP4_MACH_CRED:
3513                 exid->spo_must_enforce[0] = 0;
3514                 exid->spo_must_enforce[1] = (
3515                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3516                         1 << (OP_EXCHANGE_ID - 32) |
3517                         1 << (OP_CREATE_SESSION - 32) |
3518                         1 << (OP_DESTROY_SESSION - 32) |
3519                         1 << (OP_DESTROY_CLIENTID - 32));
3520
3521                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3522                                         1 << (OP_OPEN_DOWNGRADE) |
3523                                         1 << (OP_LOCKU) |
3524                                         1 << (OP_DELEGRETURN));
3525
3526                 exid->spo_must_allow[1] &= (
3527                                         1 << (OP_TEST_STATEID - 32) |
3528                                         1 << (OP_FREE_STATEID - 32));
3529                 if (!svc_rqst_integrity_protected(rqstp)) {
3530                         status = nfserr_inval;
3531                         goto out_nolock;
3532                 }
3533                 /*
3534                  * Sometimes userspace doesn't give us a principal.
3535                  * Which is a bug, really.  Anyway, we can't enforce
3536                  * MACH_CRED in that case, better to give up now:
3537                  */
3538                 if (!new->cl_cred.cr_principal &&
3539                                         !new->cl_cred.cr_raw_principal) {
3540                         status = nfserr_serverfault;
3541                         goto out_nolock;
3542                 }
3543                 new->cl_mach_cred = true;
3544                 break;
3545         case SP4_NONE:
3546                 break;
3547         default:                                /* checked by xdr code */
3548                 WARN_ON_ONCE(1);
3549                 fallthrough;
3550         case SP4_SSV:
3551                 status = nfserr_encr_alg_unsupp;
3552                 goto out_nolock;
3553         }
3554
3555         /* Cases below refer to rfc 5661 section 18.35.4: */
3556         spin_lock(&nn->client_lock);
3557         conf = find_confirmed_client_by_name(&exid->clname, nn);
3558         if (conf) {
3559                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3560                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3561
3562                 if (update) {
3563                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3564                                 status = nfserr_inval;
3565                                 goto out;
3566                         }
3567                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3568                                 status = nfserr_wrong_cred;
3569                                 goto out;
3570                         }
3571                         if (!creds_match) { /* case 9 */
3572                                 status = nfserr_perm;
3573                                 goto out;
3574                         }
3575                         if (!verfs_match) { /* case 8 */
3576                                 status = nfserr_not_same;
3577                                 goto out;
3578                         }
3579                         /* case 6 */
3580                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3581                         trace_nfsd_clid_confirmed_r(conf);
3582                         goto out_copy;
3583                 }
3584                 if (!creds_match) { /* case 3 */
3585                         if (client_has_state(conf)) {
3586                                 status = nfserr_clid_inuse;
3587                                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3588                                 goto out;
3589                         }
3590                         goto out_new;
3591                 }
3592                 if (verfs_match) { /* case 2 */
3593                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3594                         trace_nfsd_clid_confirmed_r(conf);
3595                         goto out_copy;
3596                 }
3597                 /* case 5, client reboot */
3598                 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3599                 conf = NULL;
3600                 goto out_new;
3601         }
3602
3603         if (update) { /* case 7 */
3604                 status = nfserr_noent;
3605                 goto out;
3606         }
3607
3608         unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3609         if (unconf) /* case 4, possible retry or client restart */
3610                 unhash_client_locked(unconf);
3611
3612         /* case 1, new owner ID */
3613         trace_nfsd_clid_fresh(new);
3614
3615 out_new:
3616         if (conf) {
3617                 status = mark_client_expired_locked(conf);
3618                 if (status)
3619                         goto out;
3620                 trace_nfsd_clid_replaced(&conf->cl_clientid);
3621         }
3622         new->cl_minorversion = cstate->minorversion;
3623         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3624         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3625
3626         /* Contrived initial CREATE_SESSION response */
3627         new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3628
3629         add_to_unconfirmed(new);
3630         swap(new, conf);
3631 out_copy:
3632         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3633         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3634
3635         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3636         nfsd4_set_ex_flags(conf, exid);
3637
3638         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3639                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3640         status = nfs_ok;
3641
3642 out:
3643         spin_unlock(&nn->client_lock);
3644 out_nolock:
3645         if (new)
3646                 expire_client(new);
3647         if (unconf) {
3648                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3649                 expire_client(unconf);
3650         }
3651         return status;
3652 }
3653
3654 static __be32 check_slot_seqid(u32 seqid, u32 slot_seqid, bool slot_inuse)
3655 {
3656         /* The slot is in use, and no response has been sent. */
3657         if (slot_inuse) {
3658                 if (seqid == slot_seqid)
3659                         return nfserr_jukebox;
3660                 else
3661                         return nfserr_seq_misordered;
3662         }
3663         /* Note unsigned 32-bit arithmetic handles wraparound: */
3664         if (likely(seqid == slot_seqid + 1))
3665                 return nfs_ok;
3666         if (seqid == slot_seqid)
3667                 return nfserr_replay_cache;
3668         return nfserr_seq_misordered;
3669 }
3670
3671 /*
3672  * Cache the create session result into the create session single DRC
3673  * slot cache by saving the xdr structure. sl_seqid has been set.
3674  * Do this for solo or embedded create session operations.
3675  */
3676 static void
3677 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3678                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3679 {
3680         slot->sl_status = nfserr;
3681         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3682 }
3683
3684 static __be32
3685 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3686                             struct nfsd4_clid_slot *slot)
3687 {
3688         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3689         return slot->sl_status;
3690 }
3691
3692 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3693                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3694                         1 +     /* MIN tag is length with zero, only length */ \
3695                         3 +     /* version, opcount, opcode */ \
3696                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3697                                 /* seqid, slotID, slotID, cache */ \
3698                         4 ) * sizeof(__be32))
3699
3700 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3701                         2 +     /* verifier: AUTH_NULL, length 0 */\
3702                         1 +     /* status */ \
3703                         1 +     /* MIN tag is length with zero, only length */ \
3704                         3 +     /* opcount, opcode, opstatus*/ \
3705                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3706                                 /* seqid, slotID, slotID, slotID, status */ \
3707                         5 ) * sizeof(__be32))
3708
3709 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3710 {
3711         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3712
3713         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3714                 return nfserr_toosmall;
3715         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3716                 return nfserr_toosmall;
3717         ca->headerpadsz = 0;
3718         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3719         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3720         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3721         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3722                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3723         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3724         /*
3725          * Note decreasing slot size below client's request may make it
3726          * difficult for client to function correctly, whereas
3727          * decreasing the number of slots will (just?) affect
3728          * performance.  When short on memory we therefore prefer to
3729          * decrease number of slots instead of their size.  Clients that
3730          * request larger slots than they need will get poor results:
3731          * Note that we always allow at least one slot, because our
3732          * accounting is soft and provides no guarantees either way.
3733          */
3734         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3735
3736         return nfs_ok;
3737 }
3738
3739 /*
3740  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3741  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3742  */
3743 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3744         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3745
3746 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3747         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3748
3749 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3750                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3751 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3752                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3753                                  sizeof(__be32))
3754
3755 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3756 {
3757         ca->headerpadsz = 0;
3758
3759         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3760                 return nfserr_toosmall;
3761         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3762                 return nfserr_toosmall;
3763         ca->maxresp_cached = 0;
3764         if (ca->maxops < 2)
3765                 return nfserr_toosmall;
3766
3767         return nfs_ok;
3768 }
3769
3770 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3771 {
3772         switch (cbs->flavor) {
3773         case RPC_AUTH_NULL:
3774         case RPC_AUTH_UNIX:
3775                 return nfs_ok;
3776         default:
3777                 /*
3778                  * GSS case: the spec doesn't allow us to return this
3779                  * error.  But it also doesn't allow us not to support
3780                  * GSS.
3781                  * I'd rather this fail hard than return some error the
3782                  * client might think it can already handle:
3783                  */
3784                 return nfserr_encr_alg_unsupp;
3785         }
3786 }
3787
3788 __be32
3789 nfsd4_create_session(struct svc_rqst *rqstp,
3790                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3791 {
3792         struct nfsd4_create_session *cr_ses = &u->create_session;
3793         struct sockaddr *sa = svc_addr(rqstp);
3794         struct nfs4_client *conf, *unconf;
3795         struct nfsd4_clid_slot *cs_slot;
3796         struct nfs4_client *old = NULL;
3797         struct nfsd4_session *new;
3798         struct nfsd4_conn *conn;
3799         __be32 status = 0;
3800         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3801
3802         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3803                 return nfserr_inval;
3804         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3805         if (status)
3806                 return status;
3807         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3808         if (status)
3809                 return status;
3810         status = check_backchannel_attrs(&cr_ses->back_channel);
3811         if (status)
3812                 goto out_release_drc_mem;
3813         status = nfserr_jukebox;
3814         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3815         if (!new)
3816                 goto out_release_drc_mem;
3817         conn = alloc_conn_from_crses(rqstp, cr_ses);
3818         if (!conn)
3819                 goto out_free_session;
3820
3821         spin_lock(&nn->client_lock);
3822
3823         /* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3824         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3825         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3826         if (!conf && !unconf) {
3827                 status = nfserr_stale_clientid;
3828                 goto out_free_conn;
3829         }
3830
3831         /* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
3832         if (conf) {
3833                 cs_slot = &conf->cl_cs_slot;
3834                 trace_nfsd_slot_seqid_conf(conf, cr_ses);
3835         } else {
3836                 cs_slot = &unconf->cl_cs_slot;
3837                 trace_nfsd_slot_seqid_unconf(unconf, cr_ses);
3838         }
3839         status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3840         switch (status) {
3841         case nfs_ok:
3842                 cs_slot->sl_seqid++;
3843                 cr_ses->seqid = cs_slot->sl_seqid;
3844                 break;
3845         case nfserr_replay_cache:
3846                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3847                 fallthrough;
3848         case nfserr_jukebox:
3849                 /* The server MUST NOT cache NFS4ERR_DELAY */
3850                 goto out_free_conn;
3851         default:
3852                 goto out_cache_error;
3853         }
3854
3855         /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3856         if (conf) {
3857                 status = nfserr_wrong_cred;
3858                 if (!nfsd4_mach_creds_match(conf, rqstp))
3859                         goto out_cache_error;
3860         } else {
3861                 status = nfserr_clid_inuse;
3862                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3863                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3864                         trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3865                         goto out_cache_error;
3866                 }
3867                 status = nfserr_wrong_cred;
3868                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3869                         goto out_cache_error;
3870                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3871                 if (old) {
3872                         status = mark_client_expired_locked(old);
3873                         if (status)
3874                                 goto out_expired_error;
3875                         trace_nfsd_clid_replaced(&old->cl_clientid);
3876                 }
3877                 move_to_confirmed(unconf);
3878                 conf = unconf;
3879         }
3880
3881         /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3882         status = nfs_ok;
3883         /* Persistent sessions are not supported */
3884         cr_ses->flags &= ~SESSION4_PERSIST;
3885         /* Upshifting from TCP to RDMA is not supported */
3886         cr_ses->flags &= ~SESSION4_RDMA;
3887
3888         init_session(rqstp, new, conf, cr_ses);
3889         nfsd4_get_session_locked(new);
3890
3891         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3892                NFS4_MAX_SESSIONID_LEN);
3893
3894         /* cache solo and embedded create sessions under the client_lock */
3895         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3896         spin_unlock(&nn->client_lock);
3897         if (conf == unconf)
3898                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3899         /* init connection and backchannel */
3900         nfsd4_init_conn(rqstp, conn, new);
3901         nfsd4_put_session(new);
3902         if (old)
3903                 expire_client(old);
3904         return status;
3905
3906 out_expired_error:
3907         old = NULL;
3908         /*
3909          * Revert the slot seq_nr change so the server will process
3910          * the client's resend instead of returning a cached response.
3911          */
3912         if (status == nfserr_jukebox) {
3913                 cs_slot->sl_seqid--;
3914                 cr_ses->seqid = cs_slot->sl_seqid;
3915                 goto out_free_conn;
3916         }
3917 out_cache_error:
3918         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3919 out_free_conn:
3920         spin_unlock(&nn->client_lock);
3921         free_conn(conn);
3922         if (old)
3923                 expire_client(old);
3924 out_free_session:
3925         __free_session(new);
3926 out_release_drc_mem:
3927         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3928         return status;
3929 }
3930
3931 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3932 {
3933         switch (*dir) {
3934         case NFS4_CDFC4_FORE:
3935         case NFS4_CDFC4_BACK:
3936                 return nfs_ok;
3937         case NFS4_CDFC4_FORE_OR_BOTH:
3938         case NFS4_CDFC4_BACK_OR_BOTH:
3939                 *dir = NFS4_CDFC4_BOTH;
3940                 return nfs_ok;
3941         }
3942         return nfserr_inval;
3943 }
3944
3945 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3946                 struct nfsd4_compound_state *cstate,
3947                 union nfsd4_op_u *u)
3948 {
3949         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3950         struct nfsd4_session *session = cstate->session;
3951         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3952         __be32 status;
3953
3954         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3955         if (status)
3956                 return status;
3957         spin_lock(&nn->client_lock);
3958         session->se_cb_prog = bc->bc_cb_program;
3959         session->se_cb_sec = bc->bc_cb_sec;
3960         spin_unlock(&nn->client_lock);
3961
3962         nfsd4_probe_callback(session->se_client);
3963
3964         return nfs_ok;
3965 }
3966
3967 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3968 {
3969         struct nfsd4_conn *c;
3970
3971         list_for_each_entry(c, &s->se_conns, cn_persession) {
3972                 if (c->cn_xprt == xpt) {
3973                         return c;
3974                 }
3975         }
3976         return NULL;
3977 }
3978
3979 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3980                 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3981 {
3982         struct nfs4_client *clp = session->se_client;
3983         struct svc_xprt *xpt = rqst->rq_xprt;
3984         struct nfsd4_conn *c;
3985         __be32 status;
3986
3987         /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3988         spin_lock(&clp->cl_lock);
3989         c = __nfsd4_find_conn(xpt, session);
3990         if (!c)
3991                 status = nfserr_noent;
3992         else if (req == c->cn_flags)
3993                 status = nfs_ok;
3994         else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3995                                 c->cn_flags != NFS4_CDFC4_BACK)
3996                 status = nfs_ok;
3997         else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3998                                 c->cn_flags != NFS4_CDFC4_FORE)
3999                 status = nfs_ok;
4000         else
4001                 status = nfserr_inval;
4002         spin_unlock(&clp->cl_lock);
4003         if (status == nfs_ok && conn)
4004                 *conn = c;
4005         return status;
4006 }
4007
4008 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
4009                      struct nfsd4_compound_state *cstate,
4010                      union nfsd4_op_u *u)
4011 {
4012         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
4013         __be32 status;
4014         struct nfsd4_conn *conn;
4015         struct nfsd4_session *session;
4016         struct net *net = SVC_NET(rqstp);
4017         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4018
4019         if (!nfsd4_last_compound_op(rqstp))
4020                 return nfserr_not_only_op;
4021         spin_lock(&nn->client_lock);
4022         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4023         spin_unlock(&nn->client_lock);
4024         if (!session)
4025                 goto out_no_session;
4026         status = nfserr_wrong_cred;
4027         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4028                 goto out;
4029         status = nfsd4_match_existing_connection(rqstp, session,
4030                         bcts->dir, &conn);
4031         if (status == nfs_ok) {
4032                 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4033                                 bcts->dir == NFS4_CDFC4_BACK)
4034                         conn->cn_flags |= NFS4_CDFC4_BACK;
4035                 nfsd4_probe_callback(session->se_client);
4036                 goto out;
4037         }
4038         if (status == nfserr_inval)
4039                 goto out;
4040         status = nfsd4_map_bcts_dir(&bcts->dir);
4041         if (status)
4042                 goto out;
4043         conn = alloc_conn(rqstp, bcts->dir);
4044         status = nfserr_jukebox;
4045         if (!conn)
4046                 goto out;
4047         nfsd4_init_conn(rqstp, conn, session);
4048         status = nfs_ok;
4049 out:
4050         nfsd4_put_session(session);
4051 out_no_session:
4052         return status;
4053 }
4054
4055 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4056 {
4057         if (!cstate->session)
4058                 return false;
4059         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4060 }
4061
4062 __be32
4063 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4064                 union nfsd4_op_u *u)
4065 {
4066         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4067         struct nfsd4_session *ses;
4068         __be32 status;
4069         int ref_held_by_me = 0;
4070         struct net *net = SVC_NET(r);
4071         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4072
4073         status = nfserr_not_only_op;
4074         if (nfsd4_compound_in_session(cstate, sessionid)) {
4075                 if (!nfsd4_last_compound_op(r))
4076                         goto out;
4077                 ref_held_by_me++;
4078         }
4079         dump_sessionid(__func__, sessionid);
4080         spin_lock(&nn->client_lock);
4081         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4082         if (!ses)
4083                 goto out_client_lock;
4084         status = nfserr_wrong_cred;
4085         if (!nfsd4_mach_creds_match(ses->se_client, r))
4086                 goto out_put_session;
4087         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4088         if (status)
4089                 goto out_put_session;
4090         unhash_session(ses);
4091         spin_unlock(&nn->client_lock);
4092
4093         nfsd4_probe_callback_sync(ses->se_client);
4094
4095         spin_lock(&nn->client_lock);
4096         status = nfs_ok;
4097 out_put_session:
4098         nfsd4_put_session_locked(ses);
4099 out_client_lock:
4100         spin_unlock(&nn->client_lock);
4101 out:
4102         return status;
4103 }
4104
4105 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4106 {
4107         struct nfs4_client *clp = ses->se_client;
4108         struct nfsd4_conn *c;
4109         __be32 status = nfs_ok;
4110         int ret;
4111
4112         spin_lock(&clp->cl_lock);
4113         c = __nfsd4_find_conn(new->cn_xprt, ses);
4114         if (c)
4115                 goto out_free;
4116         status = nfserr_conn_not_bound_to_session;
4117         if (clp->cl_mach_cred)
4118                 goto out_free;
4119         __nfsd4_hash_conn(new, ses);
4120         spin_unlock(&clp->cl_lock);
4121         ret = nfsd4_register_conn(new);
4122         if (ret)
4123                 /* oops; xprt is already down: */
4124                 nfsd4_conn_lost(&new->cn_xpt_user);
4125         return nfs_ok;
4126 out_free:
4127         spin_unlock(&clp->cl_lock);
4128         free_conn(new);
4129         return status;
4130 }
4131
4132 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4133 {
4134         struct nfsd4_compoundargs *args = rqstp->rq_argp;
4135
4136         return args->opcnt > session->se_fchannel.maxops;
4137 }
4138
4139 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4140                                   struct nfsd4_session *session)
4141 {
4142         struct xdr_buf *xb = &rqstp->rq_arg;
4143
4144         return xb->len > session->se_fchannel.maxreq_sz;
4145 }
4146
4147 static bool replay_matches_cache(struct svc_rqst *rqstp,
4148                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4149 {
4150         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4151
4152         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4153             (bool)seq->cachethis)
4154                 return false;
4155         /*
4156          * If there's an error then the reply can have fewer ops than
4157          * the call.
4158          */
4159         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4160                 return false;
4161         /*
4162          * But if we cached a reply with *more* ops than the call you're
4163          * sending us now, then this new call is clearly not really a
4164          * replay of the old one:
4165          */
4166         if (slot->sl_opcnt > argp->opcnt)
4167                 return false;
4168         /* This is the only check explicitly called by spec: */
4169         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4170                 return false;
4171         /*
4172          * There may be more comparisons we could actually do, but the
4173          * spec doesn't require us to catch every case where the calls
4174          * don't match (that would require caching the call as well as
4175          * the reply), so we don't bother.
4176          */
4177         return true;
4178 }
4179
4180 __be32
4181 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4182                 union nfsd4_op_u *u)
4183 {
4184         struct nfsd4_sequence *seq = &u->sequence;
4185         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4186         struct xdr_stream *xdr = resp->xdr;
4187         struct nfsd4_session *session;
4188         struct nfs4_client *clp;
4189         struct nfsd4_slot *slot;
4190         struct nfsd4_conn *conn;
4191         __be32 status;
4192         int buflen;
4193         struct net *net = SVC_NET(rqstp);
4194         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4195
4196         if (resp->opcnt != 1)
4197                 return nfserr_sequence_pos;
4198
4199         /*
4200          * Will be either used or freed by nfsd4_sequence_check_conn
4201          * below.
4202          */
4203         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4204         if (!conn)
4205                 return nfserr_jukebox;
4206
4207         spin_lock(&nn->client_lock);
4208         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4209         if (!session)
4210                 goto out_no_session;
4211         clp = session->se_client;
4212
4213         status = nfserr_too_many_ops;
4214         if (nfsd4_session_too_many_ops(rqstp, session))
4215                 goto out_put_session;
4216
4217         status = nfserr_req_too_big;
4218         if (nfsd4_request_too_big(rqstp, session))
4219                 goto out_put_session;
4220
4221         status = nfserr_badslot;
4222         if (seq->slotid >= session->se_fchannel.maxreqs)
4223                 goto out_put_session;
4224
4225         slot = session->se_slots[seq->slotid];
4226         dprintk("%s: slotid %d\n", __func__, seq->slotid);
4227
4228         /* We do not negotiate the number of slots yet, so set the
4229          * maxslots to the session maxreqs which is used to encode
4230          * sr_highest_slotid and the sr_target_slot id to maxslots */
4231         seq->maxslots = session->se_fchannel.maxreqs;
4232
4233         trace_nfsd_slot_seqid_sequence(clp, seq, slot);
4234         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4235                                         slot->sl_flags & NFSD4_SLOT_INUSE);
4236         if (status == nfserr_replay_cache) {
4237                 status = nfserr_seq_misordered;
4238                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4239                         goto out_put_session;
4240                 status = nfserr_seq_false_retry;
4241                 if (!replay_matches_cache(rqstp, seq, slot))
4242                         goto out_put_session;
4243                 cstate->slot = slot;
4244                 cstate->session = session;
4245                 cstate->clp = clp;
4246                 /* Return the cached reply status and set cstate->status
4247                  * for nfsd4_proc_compound processing */
4248                 status = nfsd4_replay_cache_entry(resp, seq);
4249                 cstate->status = nfserr_replay_cache;
4250                 goto out;
4251         }
4252         if (status)
4253                 goto out_put_session;
4254
4255         status = nfsd4_sequence_check_conn(conn, session);
4256         conn = NULL;
4257         if (status)
4258                 goto out_put_session;
4259
4260         buflen = (seq->cachethis) ?
4261                         session->se_fchannel.maxresp_cached :
4262                         session->se_fchannel.maxresp_sz;
4263         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4264                                     nfserr_rep_too_big;
4265         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4266                 goto out_put_session;
4267         svc_reserve(rqstp, buflen);
4268
4269         status = nfs_ok;
4270         /* Success! bump slot seqid */
4271         slot->sl_seqid = seq->seqid;
4272         slot->sl_flags |= NFSD4_SLOT_INUSE;
4273         if (seq->cachethis)
4274                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4275         else
4276                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4277
4278         cstate->slot = slot;
4279         cstate->session = session;
4280         cstate->clp = clp;
4281
4282 out:
4283         switch (clp->cl_cb_state) {
4284         case NFSD4_CB_DOWN:
4285                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4286                 break;
4287         case NFSD4_CB_FAULT:
4288                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4289                 break;
4290         default:
4291                 seq->status_flags = 0;
4292         }
4293         if (!list_empty(&clp->cl_revoked))
4294                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4295         if (atomic_read(&clp->cl_admin_revoked))
4296                 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4297         trace_nfsd_seq4_status(rqstp, seq);
4298 out_no_session:
4299         if (conn)
4300                 free_conn(conn);
4301         spin_unlock(&nn->client_lock);
4302         return status;
4303 out_put_session:
4304         nfsd4_put_session_locked(session);
4305         goto out_no_session;
4306 }
4307
4308 void
4309 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4310 {
4311         struct nfsd4_compound_state *cs = &resp->cstate;
4312
4313         if (nfsd4_has_session(cs)) {
4314                 if (cs->status != nfserr_replay_cache) {
4315                         nfsd4_store_cache_entry(resp);
4316                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4317                 }
4318                 /* Drop session reference that was taken in nfsd4_sequence() */
4319                 nfsd4_put_session(cs->session);
4320         } else if (cs->clp)
4321                 put_client_renew(cs->clp);
4322 }
4323
4324 __be32
4325 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4326                 struct nfsd4_compound_state *cstate,
4327                 union nfsd4_op_u *u)
4328 {
4329         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4330         struct nfs4_client *conf, *unconf;
4331         struct nfs4_client *clp = NULL;
4332         __be32 status = 0;
4333         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4334
4335         spin_lock(&nn->client_lock);
4336         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4337         conf = find_confirmed_client(&dc->clientid, true, nn);
4338         WARN_ON_ONCE(conf && unconf);
4339
4340         if (conf) {
4341                 if (client_has_state(conf)) {
4342                         status = nfserr_clientid_busy;
4343                         goto out;
4344                 }
4345                 status = mark_client_expired_locked(conf);
4346                 if (status)
4347                         goto out;
4348                 clp = conf;
4349         } else if (unconf)
4350                 clp = unconf;
4351         else {
4352                 status = nfserr_stale_clientid;
4353                 goto out;
4354         }
4355         if (!nfsd4_mach_creds_match(clp, rqstp)) {
4356                 clp = NULL;
4357                 status = nfserr_wrong_cred;
4358                 goto out;
4359         }
4360         trace_nfsd_clid_destroyed(&clp->cl_clientid);
4361         unhash_client_locked(clp);
4362 out:
4363         spin_unlock(&nn->client_lock);
4364         if (clp)
4365                 expire_client(clp);
4366         return status;
4367 }
4368
4369 __be32
4370 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4371                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4372 {
4373         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4374         struct nfs4_client *clp = cstate->clp;
4375         __be32 status = 0;
4376
4377         if (rc->rca_one_fs) {
4378                 if (!cstate->current_fh.fh_dentry)
4379                         return nfserr_nofilehandle;
4380                 /*
4381                  * We don't take advantage of the rca_one_fs case.
4382                  * That's OK, it's optional, we can safely ignore it.
4383                  */
4384                 return nfs_ok;
4385         }
4386
4387         status = nfserr_complete_already;
4388         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4389                 goto out;
4390
4391         status = nfserr_stale_clientid;
4392         if (is_client_expired(clp))
4393                 /*
4394                  * The following error isn't really legal.
4395                  * But we only get here if the client just explicitly
4396                  * destroyed the client.  Surely it no longer cares what
4397                  * error it gets back on an operation for the dead
4398                  * client.
4399                  */
4400                 goto out;
4401
4402         status = nfs_ok;
4403         trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4404         nfsd4_client_record_create(clp);
4405         inc_reclaim_complete(clp);
4406 out:
4407         return status;
4408 }
4409
4410 __be32
4411 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4412                   union nfsd4_op_u *u)
4413 {
4414         struct nfsd4_setclientid *setclid = &u->setclientid;
4415         struct xdr_netobj       clname = setclid->se_name;
4416         nfs4_verifier           clverifier = setclid->se_verf;
4417         struct nfs4_client      *conf, *new;
4418         struct nfs4_client      *unconf = NULL;
4419         __be32                  status;
4420         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4421
4422         new = create_client(clname, rqstp, &clverifier);
4423         if (new == NULL)
4424                 return nfserr_jukebox;
4425         spin_lock(&nn->client_lock);
4426         conf = find_confirmed_client_by_name(&clname, nn);
4427         if (conf && client_has_state(conf)) {
4428                 status = nfserr_clid_inuse;
4429                 if (clp_used_exchangeid(conf))
4430                         goto out;
4431                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4432                         trace_nfsd_clid_cred_mismatch(conf, rqstp);
4433                         goto out;
4434                 }
4435         }
4436         unconf = find_unconfirmed_client_by_name(&clname, nn);
4437         if (unconf)
4438                 unhash_client_locked(unconf);
4439         if (conf) {
4440                 if (same_verf(&conf->cl_verifier, &clverifier)) {
4441                         copy_clid(new, conf);
4442                         gen_confirm(new, nn);
4443                 } else
4444                         trace_nfsd_clid_verf_mismatch(conf, rqstp,
4445                                                       &clverifier);
4446         } else
4447                 trace_nfsd_clid_fresh(new);
4448         new->cl_minorversion = 0;
4449         gen_callback(new, setclid, rqstp);
4450         add_to_unconfirmed(new);
4451         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4452         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4453         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4454         new = NULL;
4455         status = nfs_ok;
4456 out:
4457         spin_unlock(&nn->client_lock);
4458         if (new)
4459                 free_client(new);
4460         if (unconf) {
4461                 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4462                 expire_client(unconf);
4463         }
4464         return status;
4465 }
4466
4467 __be32
4468 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4469                         struct nfsd4_compound_state *cstate,
4470                         union nfsd4_op_u *u)
4471 {
4472         struct nfsd4_setclientid_confirm *setclientid_confirm =
4473                         &u->setclientid_confirm;
4474         struct nfs4_client *conf, *unconf;
4475         struct nfs4_client *old = NULL;
4476         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4477         clientid_t * clid = &setclientid_confirm->sc_clientid;
4478         __be32 status;
4479         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4480
4481         if (STALE_CLIENTID(clid, nn))
4482                 return nfserr_stale_clientid;
4483
4484         spin_lock(&nn->client_lock);
4485         conf = find_confirmed_client(clid, false, nn);
4486         unconf = find_unconfirmed_client(clid, false, nn);
4487         /*
4488          * We try hard to give out unique clientid's, so if we get an
4489          * attempt to confirm the same clientid with a different cred,
4490          * the client may be buggy; this should never happen.
4491          *
4492          * Nevertheless, RFC 7530 recommends INUSE for this case:
4493          */
4494         status = nfserr_clid_inuse;
4495         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4496                 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4497                 goto out;
4498         }
4499         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4500                 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4501                 goto out;
4502         }
4503         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4504                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4505                         status = nfs_ok;
4506                 } else
4507                         status = nfserr_stale_clientid;
4508                 goto out;
4509         }
4510         status = nfs_ok;
4511         if (conf) {
4512                 old = unconf;
4513                 unhash_client_locked(old);
4514                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4515         } else {
4516                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4517                 if (old) {
4518                         status = nfserr_clid_inuse;
4519                         if (client_has_state(old)
4520                                         && !same_creds(&unconf->cl_cred,
4521                                                         &old->cl_cred)) {
4522                                 old = NULL;
4523                                 goto out;
4524                         }
4525                         status = mark_client_expired_locked(old);
4526                         if (status) {
4527                                 old = NULL;
4528                                 goto out;
4529                         }
4530                         trace_nfsd_clid_replaced(&old->cl_clientid);
4531                 }
4532                 move_to_confirmed(unconf);
4533                 conf = unconf;
4534         }
4535         get_client_locked(conf);
4536         spin_unlock(&nn->client_lock);
4537         if (conf == unconf)
4538                 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4539         nfsd4_probe_callback(conf);
4540         spin_lock(&nn->client_lock);
4541         put_client_renew_locked(conf);
4542 out:
4543         spin_unlock(&nn->client_lock);
4544         if (old)
4545                 expire_client(old);
4546         return status;
4547 }
4548
4549 static struct nfs4_file *nfsd4_alloc_file(void)
4550 {
4551         return kmem_cache_alloc(file_slab, GFP_KERNEL);
4552 }
4553
4554 /* OPEN Share state helper functions */
4555
4556 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4557 {
4558         refcount_set(&fp->fi_ref, 1);
4559         spin_lock_init(&fp->fi_lock);
4560         INIT_LIST_HEAD(&fp->fi_stateids);
4561         INIT_LIST_HEAD(&fp->fi_delegations);
4562         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4563         fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4564         fp->fi_deleg_file = NULL;
4565         fp->fi_had_conflict = false;
4566         fp->fi_share_deny = 0;
4567         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4568         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4569         fp->fi_aliased = false;
4570         fp->fi_inode = d_inode(fh->fh_dentry);
4571 #ifdef CONFIG_NFSD_PNFS
4572         INIT_LIST_HEAD(&fp->fi_lo_states);
4573         atomic_set(&fp->fi_lo_recalls, 0);
4574 #endif
4575 }
4576
4577 void
4578 nfsd4_free_slabs(void)
4579 {
4580         kmem_cache_destroy(client_slab);
4581         kmem_cache_destroy(openowner_slab);
4582         kmem_cache_destroy(lockowner_slab);
4583         kmem_cache_destroy(file_slab);
4584         kmem_cache_destroy(stateid_slab);
4585         kmem_cache_destroy(deleg_slab);
4586         kmem_cache_destroy(odstate_slab);
4587 }
4588
4589 int
4590 nfsd4_init_slabs(void)
4591 {
4592         client_slab = KMEM_CACHE(nfs4_client, 0);
4593         if (client_slab == NULL)
4594                 goto out;
4595         openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4596         if (openowner_slab == NULL)
4597                 goto out_free_client_slab;
4598         lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4599         if (lockowner_slab == NULL)
4600                 goto out_free_openowner_slab;
4601         file_slab = KMEM_CACHE(nfs4_file, 0);
4602         if (file_slab == NULL)
4603                 goto out_free_lockowner_slab;
4604         stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4605         if (stateid_slab == NULL)
4606                 goto out_free_file_slab;
4607         deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4608         if (deleg_slab == NULL)
4609                 goto out_free_stateid_slab;
4610         odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4611         if (odstate_slab == NULL)
4612                 goto out_free_deleg_slab;
4613         return 0;
4614
4615 out_free_deleg_slab:
4616         kmem_cache_destroy(deleg_slab);
4617 out_free_stateid_slab:
4618         kmem_cache_destroy(stateid_slab);
4619 out_free_file_slab:
4620         kmem_cache_destroy(file_slab);
4621 out_free_lockowner_slab:
4622         kmem_cache_destroy(lockowner_slab);
4623 out_free_openowner_slab:
4624         kmem_cache_destroy(openowner_slab);
4625 out_free_client_slab:
4626         kmem_cache_destroy(client_slab);
4627 out:
4628         return -ENOMEM;
4629 }
4630
4631 static unsigned long
4632 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4633 {
4634         int count;
4635         struct nfsd_net *nn = shrink->private_data;
4636
4637         count = atomic_read(&nn->nfsd_courtesy_clients);
4638         if (!count)
4639                 count = atomic_long_read(&num_delegations);
4640         if (count)
4641                 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4642         return (unsigned long)count;
4643 }
4644
4645 static unsigned long
4646 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4647 {
4648         return SHRINK_STOP;
4649 }
4650
4651 void
4652 nfsd4_init_leases_net(struct nfsd_net *nn)
4653 {
4654         struct sysinfo si;
4655         u64 max_clients;
4656
4657         nn->nfsd4_lease = 90;   /* default lease time */
4658         nn->nfsd4_grace = 90;
4659         nn->somebody_reclaimed = false;
4660         nn->track_reclaim_completes = false;
4661         nn->clverifier_counter = get_random_u32();
4662         nn->clientid_base = get_random_u32();
4663         nn->clientid_counter = nn->clientid_base + 1;
4664         nn->s2s_cp_cl_id = nn->clientid_counter++;
4665
4666         atomic_set(&nn->nfs4_client_count, 0);
4667         si_meminfo(&si);
4668         max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4669         max_clients *= NFS4_CLIENTS_PER_GB;
4670         nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4671
4672         atomic_set(&nn->nfsd_courtesy_clients, 0);
4673 }
4674
4675 enum rp_lock {
4676         RP_UNLOCKED,
4677         RP_LOCKED,
4678         RP_UNHASHED,
4679 };
4680
4681 static void init_nfs4_replay(struct nfs4_replay *rp)
4682 {
4683         rp->rp_status = nfserr_serverfault;
4684         rp->rp_buflen = 0;
4685         rp->rp_buf = rp->rp_ibuf;
4686         atomic_set(&rp->rp_locked, RP_UNLOCKED);
4687 }
4688
4689 static int nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4690                                       struct nfs4_stateowner *so)
4691 {
4692         if (!nfsd4_has_session(cstate)) {
4693                 wait_var_event(&so->so_replay.rp_locked,
4694                                atomic_cmpxchg(&so->so_replay.rp_locked,
4695                                               RP_UNLOCKED, RP_LOCKED) != RP_LOCKED);
4696                 if (atomic_read(&so->so_replay.rp_locked) == RP_UNHASHED)
4697                         return -EAGAIN;
4698                 cstate->replay_owner = nfs4_get_stateowner(so);
4699         }
4700         return 0;
4701 }
4702
4703 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4704 {
4705         struct nfs4_stateowner *so = cstate->replay_owner;
4706
4707         if (so != NULL) {
4708                 cstate->replay_owner = NULL;
4709                 atomic_set(&so->so_replay.rp_locked, RP_UNLOCKED);
4710                 smp_mb__after_atomic();
4711                 wake_up_var(&so->so_replay.rp_locked);
4712                 nfs4_put_stateowner(so);
4713         }
4714 }
4715
4716 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4717 {
4718         struct nfs4_stateowner *sop;
4719
4720         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4721         if (!sop)
4722                 return NULL;
4723
4724         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4725         if (!sop->so_owner.data) {
4726                 kmem_cache_free(slab, sop);
4727                 return NULL;
4728         }
4729
4730         INIT_LIST_HEAD(&sop->so_stateids);
4731         sop->so_client = clp;
4732         init_nfs4_replay(&sop->so_replay);
4733         atomic_set(&sop->so_count, 1);
4734         return sop;
4735 }
4736
4737 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4738 {
4739         lockdep_assert_held(&clp->cl_lock);
4740
4741         list_add(&oo->oo_owner.so_strhash,
4742                  &clp->cl_ownerstr_hashtbl[strhashval]);
4743         list_add(&oo->oo_perclient, &clp->cl_openowners);
4744 }
4745
4746 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4747 {
4748         unhash_openowner_locked(openowner(so));
4749 }
4750
4751 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4752 {
4753         struct nfs4_openowner *oo = openowner(so);
4754
4755         kmem_cache_free(openowner_slab, oo);
4756 }
4757
4758 static const struct nfs4_stateowner_operations openowner_ops = {
4759         .so_unhash =    nfs4_unhash_openowner,
4760         .so_free =      nfs4_free_openowner,
4761 };
4762
4763 static struct nfs4_ol_stateid *
4764 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4765 {
4766         struct nfs4_ol_stateid *local, *ret = NULL;
4767         struct nfs4_openowner *oo = open->op_openowner;
4768
4769         lockdep_assert_held(&fp->fi_lock);
4770
4771         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4772                 /* ignore lock owners */
4773                 if (local->st_stateowner->so_is_open_owner == 0)
4774                         continue;
4775                 if (local->st_stateowner != &oo->oo_owner)
4776                         continue;
4777                 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4778                     !local->st_stid.sc_status) {
4779                         ret = local;
4780                         refcount_inc(&ret->st_stid.sc_count);
4781                         break;
4782                 }
4783         }
4784         return ret;
4785 }
4786
4787 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4788         __releases(&s->sc_client->cl_lock)
4789 {
4790         struct nfs4_client *cl = s->sc_client;
4791         LIST_HEAD(reaplist);
4792         struct nfs4_ol_stateid *stp;
4793         struct nfs4_delegation *dp;
4794         bool unhashed;
4795
4796         switch (s->sc_type) {
4797         case SC_TYPE_OPEN:
4798                 stp = openlockstateid(s);
4799                 if (unhash_open_stateid(stp, &reaplist))
4800                         put_ol_stateid_locked(stp, &reaplist);
4801                 spin_unlock(&cl->cl_lock);
4802                 free_ol_stateid_reaplist(&reaplist);
4803                 break;
4804         case SC_TYPE_LOCK:
4805                 stp = openlockstateid(s);
4806                 unhashed = unhash_lock_stateid(stp);
4807                 spin_unlock(&cl->cl_lock);
4808                 if (unhashed)
4809                         nfs4_put_stid(s);
4810                 break;
4811         case SC_TYPE_DELEG:
4812                 dp = delegstateid(s);
4813                 list_del_init(&dp->dl_recall_lru);
4814                 spin_unlock(&cl->cl_lock);
4815                 nfs4_put_stid(s);
4816                 break;
4817         default:
4818                 spin_unlock(&cl->cl_lock);
4819         }
4820 }
4821
4822 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4823                                     stateid_t *stid)
4824 {
4825         /* NFSv4.0 has no way for the client to tell the server
4826          * that it can forget an admin-revoked stateid.
4827          * So we keep it around until the first time that the
4828          * client uses it, and drop it the first time
4829          * nfserr_admin_revoked is returned.
4830          * For v4.1 and later we wait until explicitly told
4831          * to free the stateid.
4832          */
4833         if (cl->cl_minorversion == 0) {
4834                 struct nfs4_stid *st;
4835
4836                 spin_lock(&cl->cl_lock);
4837                 st = find_stateid_locked(cl, stid);
4838                 if (st)
4839                         nfsd4_drop_revoked_stid(st);
4840                 else
4841                         spin_unlock(&cl->cl_lock);
4842         }
4843 }
4844
4845 static __be32
4846 nfsd4_verify_open_stid(struct nfs4_stid *s)
4847 {
4848         __be32 ret = nfs_ok;
4849
4850         if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4851                 ret = nfserr_admin_revoked;
4852         else if (s->sc_status & SC_STATUS_REVOKED)
4853                 ret = nfserr_deleg_revoked;
4854         else if (s->sc_status & SC_STATUS_CLOSED)
4855                 ret = nfserr_bad_stateid;
4856         return ret;
4857 }
4858
4859 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4860 static __be32
4861 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4862 {
4863         __be32 ret;
4864
4865         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4866         ret = nfsd4_verify_open_stid(&stp->st_stid);
4867         if (ret == nfserr_admin_revoked)
4868                 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4869                                         &stp->st_stid.sc_stateid);
4870
4871         if (ret != nfs_ok)
4872                 mutex_unlock(&stp->st_mutex);
4873         return ret;
4874 }
4875
4876 static struct nfs4_ol_stateid *
4877 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4878 {
4879         struct nfs4_ol_stateid *stp;
4880         for (;;) {
4881                 spin_lock(&fp->fi_lock);
4882                 stp = nfsd4_find_existing_open(fp, open);
4883                 spin_unlock(&fp->fi_lock);
4884                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4885                         break;
4886                 nfs4_put_stid(&stp->st_stid);
4887         }
4888         return stp;
4889 }
4890
4891 static struct nfs4_openowner *
4892 find_or_alloc_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4893                               struct nfsd4_compound_state *cstate)
4894 {
4895         struct nfs4_client *clp = cstate->clp;
4896         struct nfs4_openowner *oo, *new = NULL;
4897
4898 retry:
4899         spin_lock(&clp->cl_lock);
4900         oo = find_openstateowner_str(strhashval, open, clp);
4901         if (!oo && new) {
4902                 hash_openowner(new, clp, strhashval);
4903                 spin_unlock(&clp->cl_lock);
4904                 return new;
4905         }
4906         spin_unlock(&clp->cl_lock);
4907
4908         if (oo && !(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4909                 /* Replace unconfirmed owners without checking for replay. */
4910                 release_openowner(oo);
4911                 oo = NULL;
4912         }
4913         if (oo) {
4914                 if (new)
4915                         nfs4_free_stateowner(&new->oo_owner);
4916                 return oo;
4917         }
4918
4919         new = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4920         if (!new)
4921                 return NULL;
4922         new->oo_owner.so_ops = &openowner_ops;
4923         new->oo_owner.so_is_open_owner = 1;
4924         new->oo_owner.so_seqid = open->op_seqid;
4925         new->oo_flags = 0;
4926         if (nfsd4_has_session(cstate))
4927                 new->oo_flags |= NFS4_OO_CONFIRMED;
4928         new->oo_time = 0;
4929         new->oo_last_closed_stid = NULL;
4930         INIT_LIST_HEAD(&new->oo_close_lru);
4931         goto retry;
4932 }
4933
4934 static struct nfs4_ol_stateid *
4935 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4936 {
4937
4938         struct nfs4_openowner *oo = open->op_openowner;
4939         struct nfs4_ol_stateid *retstp = NULL;
4940         struct nfs4_ol_stateid *stp;
4941
4942         stp = open->op_stp;
4943         /* We are moving these outside of the spinlocks to avoid the warnings */
4944         mutex_init(&stp->st_mutex);
4945         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4946
4947 retry:
4948         spin_lock(&oo->oo_owner.so_client->cl_lock);
4949         spin_lock(&fp->fi_lock);
4950
4951         retstp = nfsd4_find_existing_open(fp, open);
4952         if (retstp)
4953                 goto out_unlock;
4954
4955         open->op_stp = NULL;
4956         refcount_inc(&stp->st_stid.sc_count);
4957         stp->st_stid.sc_type = SC_TYPE_OPEN;
4958         INIT_LIST_HEAD(&stp->st_locks);
4959         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4960         get_nfs4_file(fp);
4961         stp->st_stid.sc_file = fp;
4962         stp->st_access_bmap = 0;
4963         stp->st_deny_bmap = 0;
4964         stp->st_openstp = NULL;
4965         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4966         list_add(&stp->st_perfile, &fp->fi_stateids);
4967
4968 out_unlock:
4969         spin_unlock(&fp->fi_lock);
4970         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4971         if (retstp) {
4972                 /* Handle races with CLOSE */
4973                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4974                         nfs4_put_stid(&retstp->st_stid);
4975                         goto retry;
4976                 }
4977                 /* To keep mutex tracking happy */
4978                 mutex_unlock(&stp->st_mutex);
4979                 stp = retstp;
4980         }
4981         return stp;
4982 }
4983
4984 /*
4985  * In the 4.0 case we need to keep the owners around a little while to handle
4986  * CLOSE replay. We still do need to release any file access that is held by
4987  * them before returning however.
4988  */
4989 static void
4990 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4991 {
4992         struct nfs4_ol_stateid *last;
4993         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4994         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4995                                                 nfsd_net_id);
4996
4997         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4998
4999         /*
5000          * We know that we hold one reference via nfsd4_close, and another
5001          * "persistent" reference for the client. If the refcount is higher
5002          * than 2, then there are still calls in progress that are using this
5003          * stateid. We can't put the sc_file reference until they are finished.
5004          * Wait for the refcount to drop to 2. Since it has been unhashed,
5005          * there should be no danger of the refcount going back up again at
5006          * this point.
5007          * Some threads with a reference might be waiting for rp_locked,
5008          * so tell them to stop waiting.
5009          */
5010         atomic_set(&oo->oo_owner.so_replay.rp_locked, RP_UNHASHED);
5011         smp_mb__after_atomic();
5012         wake_up_var(&oo->oo_owner.so_replay.rp_locked);
5013         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
5014
5015         release_all_access(s);
5016         if (s->st_stid.sc_file) {
5017                 put_nfs4_file(s->st_stid.sc_file);
5018                 s->st_stid.sc_file = NULL;
5019         }
5020
5021         spin_lock(&nn->client_lock);
5022         last = oo->oo_last_closed_stid;
5023         oo->oo_last_closed_stid = s;
5024         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
5025         oo->oo_time = ktime_get_boottime_seconds();
5026         spin_unlock(&nn->client_lock);
5027         if (last)
5028                 nfs4_put_stid(&last->st_stid);
5029 }
5030
5031 static noinline_for_stack struct nfs4_file *
5032 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
5033 {
5034         struct inode *inode = d_inode(fhp->fh_dentry);
5035         struct rhlist_head *tmp, *list;
5036         struct nfs4_file *fi;
5037
5038         rcu_read_lock();
5039         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5040                                nfs4_file_rhash_params);
5041         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5042                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5043                         if (refcount_inc_not_zero(&fi->fi_ref)) {
5044                                 rcu_read_unlock();
5045                                 return fi;
5046                         }
5047                 }
5048         }
5049         rcu_read_unlock();
5050         return NULL;
5051 }
5052
5053 /*
5054  * On hash insertion, identify entries with the same inode but
5055  * distinct filehandles. They will all be on the list returned
5056  * by rhltable_lookup().
5057  *
5058  * inode->i_lock prevents racing insertions from adding an entry
5059  * for the same inode/fhp pair twice.
5060  */
5061 static noinline_for_stack struct nfs4_file *
5062 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5063 {
5064         struct inode *inode = d_inode(fhp->fh_dentry);
5065         struct rhlist_head *tmp, *list;
5066         struct nfs4_file *ret = NULL;
5067         bool alias_found = false;
5068         struct nfs4_file *fi;
5069         int err;
5070
5071         rcu_read_lock();
5072         spin_lock(&inode->i_lock);
5073
5074         list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5075                                nfs4_file_rhash_params);
5076         rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5077                 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5078                         if (refcount_inc_not_zero(&fi->fi_ref))
5079                                 ret = fi;
5080                 } else
5081                         fi->fi_aliased = alias_found = true;
5082         }
5083         if (ret)
5084                 goto out_unlock;
5085
5086         nfsd4_file_init(fhp, new);
5087         err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5088                               nfs4_file_rhash_params);
5089         if (err)
5090                 goto out_unlock;
5091
5092         new->fi_aliased = alias_found;
5093         ret = new;
5094
5095 out_unlock:
5096         spin_unlock(&inode->i_lock);
5097         rcu_read_unlock();
5098         return ret;
5099 }
5100
5101 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5102 {
5103         rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5104                         nfs4_file_rhash_params);
5105 }
5106
5107 /*
5108  * Called to check deny when READ with all zero stateid or
5109  * WRITE with all zero or all one stateid
5110  */
5111 static __be32
5112 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5113 {
5114         struct nfs4_file *fp;
5115         __be32 ret = nfs_ok;
5116
5117         fp = nfsd4_file_hash_lookup(current_fh);
5118         if (!fp)
5119                 return ret;
5120
5121         /* Check for conflicting share reservations */
5122         spin_lock(&fp->fi_lock);
5123         if (fp->fi_share_deny & deny_type)
5124                 ret = nfserr_locked;
5125         spin_unlock(&fp->fi_lock);
5126         put_nfs4_file(fp);
5127         return ret;
5128 }
5129
5130 static bool nfsd4_deleg_present(const struct inode *inode)
5131 {
5132         struct file_lock_context *ctx = locks_inode_context(inode);
5133
5134         return ctx && !list_empty_careful(&ctx->flc_lease);
5135 }
5136
5137 /**
5138  * nfsd_wait_for_delegreturn - wait for delegations to be returned
5139  * @rqstp: the RPC transaction being executed
5140  * @inode: in-core inode of the file being waited for
5141  *
5142  * The timeout prevents deadlock if all nfsd threads happen to be
5143  * tied up waiting for returning delegations.
5144  *
5145  * Return values:
5146  *   %true: delegation was returned
5147  *   %false: timed out waiting for delegreturn
5148  */
5149 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5150 {
5151         long __maybe_unused timeo;
5152
5153         timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5154                                        NFSD_DELEGRETURN_TIMEOUT);
5155         trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5156         return timeo > 0;
5157 }
5158
5159 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5160 {
5161         struct nfs4_delegation *dp = cb_to_delegation(cb);
5162         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5163                                           nfsd_net_id);
5164
5165         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5166
5167         /*
5168          * We can't do this in nfsd_break_deleg_cb because it is
5169          * already holding inode->i_lock.
5170          *
5171          * If the dl_time != 0, then we know that it has already been
5172          * queued for a lease break. Don't queue it again.
5173          */
5174         spin_lock(&state_lock);
5175         if (delegation_hashed(dp) && dp->dl_time == 0) {
5176                 dp->dl_time = ktime_get_boottime_seconds();
5177                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5178         }
5179         spin_unlock(&state_lock);
5180 }
5181
5182 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5183                 struct rpc_task *task)
5184 {
5185         struct nfs4_delegation *dp = cb_to_delegation(cb);
5186
5187         trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5188
5189         if (dp->dl_stid.sc_status)
5190                 /* CLOSED or REVOKED */
5191                 return 1;
5192
5193         switch (task->tk_status) {
5194         case 0:
5195                 return 1;
5196         case -NFS4ERR_DELAY:
5197                 rpc_delay(task, 2 * HZ);
5198                 return 0;
5199         case -EBADHANDLE:
5200         case -NFS4ERR_BAD_STATEID:
5201                 /*
5202                  * Race: client probably got cb_recall before open reply
5203                  * granting delegation.
5204                  */
5205                 if (dp->dl_retries--) {
5206                         rpc_delay(task, 2 * HZ);
5207                         return 0;
5208                 }
5209                 fallthrough;
5210         default:
5211                 return 1;
5212         }
5213 }
5214
5215 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5216 {
5217         struct nfs4_delegation *dp = cb_to_delegation(cb);
5218
5219         nfs4_put_stid(&dp->dl_stid);
5220 }
5221
5222 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5223         .prepare        = nfsd4_cb_recall_prepare,
5224         .done           = nfsd4_cb_recall_done,
5225         .release        = nfsd4_cb_recall_release,
5226         .opcode         = OP_CB_RECALL,
5227 };
5228
5229 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5230 {
5231         /*
5232          * We're assuming the state code never drops its reference
5233          * without first removing the lease.  Since we're in this lease
5234          * callback (and since the lease code is serialized by the
5235          * flc_lock) we know the server hasn't removed the lease yet, and
5236          * we know it's safe to take a reference.
5237          */
5238         refcount_inc(&dp->dl_stid.sc_count);
5239         WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5240 }
5241
5242 /* Called from break_lease() with flc_lock held. */
5243 static bool
5244 nfsd_break_deleg_cb(struct file_lease *fl)
5245 {
5246         struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5247         struct nfs4_file *fp = dp->dl_stid.sc_file;
5248         struct nfs4_client *clp = dp->dl_stid.sc_client;
5249         struct nfsd_net *nn;
5250
5251         trace_nfsd_cb_recall(&dp->dl_stid);
5252
5253         dp->dl_recalled = true;
5254         atomic_inc(&clp->cl_delegs_in_recall);
5255         if (try_to_expire_client(clp)) {
5256                 nn = net_generic(clp->net, nfsd_net_id);
5257                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5258         }
5259
5260         /*
5261          * We don't want the locks code to timeout the lease for us;
5262          * we'll remove it ourself if a delegation isn't returned
5263          * in time:
5264          */
5265         fl->fl_break_time = 0;
5266
5267         fp->fi_had_conflict = true;
5268         nfsd_break_one_deleg(dp);
5269         return false;
5270 }
5271
5272 /**
5273  * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5274  * @fl: Lock state to check
5275  *
5276  * Return values:
5277  *   %true: Lease conflict was resolved
5278  *   %false: Lease conflict was not resolved.
5279  */
5280 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5281 {
5282         struct nfs4_delegation *dl = fl->c.flc_owner;
5283         struct svc_rqst *rqst;
5284         struct nfs4_client *clp;
5285
5286         rqst = nfsd_current_rqst();
5287         if (!nfsd_v4client(rqst))
5288                 return false;
5289         clp = *(rqst->rq_lease_breaker);
5290         return dl->dl_stid.sc_client == clp;
5291 }
5292
5293 static int
5294 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5295                      struct list_head *dispose)
5296 {
5297         struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5298         struct nfs4_client *clp = dp->dl_stid.sc_client;
5299
5300         if (arg & F_UNLCK) {
5301                 if (dp->dl_recalled)
5302                         atomic_dec(&clp->cl_delegs_in_recall);
5303                 return lease_modify(onlist, arg, dispose);
5304         } else
5305                 return -EAGAIN;
5306 }
5307
5308 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5309         .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5310         .lm_break = nfsd_break_deleg_cb,
5311         .lm_change = nfsd_change_deleg_cb,
5312 };
5313
5314 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5315 {
5316         if (nfsd4_has_session(cstate))
5317                 return nfs_ok;
5318         if (seqid == so->so_seqid - 1)
5319                 return nfserr_replay_me;
5320         if (seqid == so->so_seqid)
5321                 return nfs_ok;
5322         return nfserr_bad_seqid;
5323 }
5324
5325 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5326                                                 struct nfsd_net *nn)
5327 {
5328         struct nfs4_client *found;
5329
5330         spin_lock(&nn->client_lock);
5331         found = find_confirmed_client(clid, sessions, nn);
5332         if (found)
5333                 atomic_inc(&found->cl_rpc_users);
5334         spin_unlock(&nn->client_lock);
5335         return found;
5336 }
5337
5338 static __be32 set_client(clientid_t *clid,
5339                 struct nfsd4_compound_state *cstate,
5340                 struct nfsd_net *nn)
5341 {
5342         if (cstate->clp) {
5343                 if (!same_clid(&cstate->clp->cl_clientid, clid))
5344                         return nfserr_stale_clientid;
5345                 return nfs_ok;
5346         }
5347         if (STALE_CLIENTID(clid, nn))
5348                 return nfserr_stale_clientid;
5349         /*
5350          * We're in the 4.0 case (otherwise the SEQUENCE op would have
5351          * set cstate->clp), so session = false:
5352          */
5353         cstate->clp = lookup_clientid(clid, false, nn);
5354         if (!cstate->clp)
5355                 return nfserr_expired;
5356         return nfs_ok;
5357 }
5358
5359 __be32
5360 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5361                     struct nfsd4_open *open, struct nfsd_net *nn)
5362 {
5363         clientid_t *clientid = &open->op_clientid;
5364         struct nfs4_client *clp = NULL;
5365         unsigned int strhashval;
5366         struct nfs4_openowner *oo = NULL;
5367         __be32 status;
5368
5369         /*
5370          * In case we need it later, after we've already created the
5371          * file and don't want to risk a further failure:
5372          */
5373         open->op_file = nfsd4_alloc_file();
5374         if (open->op_file == NULL)
5375                 return nfserr_jukebox;
5376
5377         status = set_client(clientid, cstate, nn);
5378         if (status)
5379                 return status;
5380         clp = cstate->clp;
5381
5382         strhashval = ownerstr_hashval(&open->op_owner);
5383 retry:
5384         oo = find_or_alloc_open_stateowner(strhashval, open, cstate);
5385         open->op_openowner = oo;
5386         if (!oo)
5387                 return nfserr_jukebox;
5388         if (nfsd4_cstate_assign_replay(cstate, &oo->oo_owner) == -EAGAIN) {
5389                 nfs4_put_stateowner(&oo->oo_owner);
5390                 goto retry;
5391         }
5392         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5393         if (status)
5394                 return status;
5395
5396         open->op_stp = nfs4_alloc_open_stateid(clp);
5397         if (!open->op_stp)
5398                 return nfserr_jukebox;
5399
5400         if (nfsd4_has_session(cstate) &&
5401             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5402                 open->op_odstate = alloc_clnt_odstate(clp);
5403                 if (!open->op_odstate)
5404                         return nfserr_jukebox;
5405         }
5406
5407         return nfs_ok;
5408 }
5409
5410 static inline __be32
5411 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5412 {
5413         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5414                 return nfserr_openmode;
5415         else
5416                 return nfs_ok;
5417 }
5418
5419 static int share_access_to_flags(u32 share_access)
5420 {
5421         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5422 }
5423
5424 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5425                                                   stateid_t *s)
5426 {
5427         struct nfs4_stid *ret;
5428
5429         ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5430         if (!ret)
5431                 return NULL;
5432         return delegstateid(ret);
5433 }
5434
5435 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5436 {
5437         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5438                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5439 }
5440
5441 static __be32
5442 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5443                 struct nfs4_delegation **dp)
5444 {
5445         int flags;
5446         __be32 status = nfserr_bad_stateid;
5447         struct nfs4_delegation *deleg;
5448
5449         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5450         if (deleg == NULL)
5451                 goto out;
5452         if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5453                 nfs4_put_stid(&deleg->dl_stid);
5454                 status = nfserr_admin_revoked;
5455                 goto out;
5456         }
5457         if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5458                 nfs4_put_stid(&deleg->dl_stid);
5459                 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5460                 status = nfserr_deleg_revoked;
5461                 goto out;
5462         }
5463         flags = share_access_to_flags(open->op_share_access);
5464         status = nfs4_check_delegmode(deleg, flags);
5465         if (status) {
5466                 nfs4_put_stid(&deleg->dl_stid);
5467                 goto out;
5468         }
5469         *dp = deleg;
5470 out:
5471         if (!nfsd4_is_deleg_cur(open))
5472                 return nfs_ok;
5473         if (status)
5474                 return status;
5475         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5476         return nfs_ok;
5477 }
5478
5479 static inline int nfs4_access_to_access(u32 nfs4_access)
5480 {
5481         int flags = 0;
5482
5483         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5484                 flags |= NFSD_MAY_READ;
5485         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5486                 flags |= NFSD_MAY_WRITE;
5487         return flags;
5488 }
5489
5490 static inline __be32
5491 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5492                 struct nfsd4_open *open)
5493 {
5494         struct iattr iattr = {
5495                 .ia_valid = ATTR_SIZE,
5496                 .ia_size = 0,
5497         };
5498         struct nfsd_attrs attrs = {
5499                 .na_iattr       = &iattr,
5500         };
5501         if (!open->op_truncate)
5502                 return 0;
5503         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5504                 return nfserr_inval;
5505         return nfsd_setattr(rqstp, fh, &attrs, NULL);
5506 }
5507
5508 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5509                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5510                 struct nfsd4_open *open, bool new_stp)
5511 {
5512         struct nfsd_file *nf = NULL;
5513         __be32 status;
5514         int oflag = nfs4_access_to_omode(open->op_share_access);
5515         int access = nfs4_access_to_access(open->op_share_access);
5516         unsigned char old_access_bmap, old_deny_bmap;
5517
5518         spin_lock(&fp->fi_lock);
5519
5520         /*
5521          * Are we trying to set a deny mode that would conflict with
5522          * current access?
5523          */
5524         status = nfs4_file_check_deny(fp, open->op_share_deny);
5525         if (status != nfs_ok) {
5526                 if (status != nfserr_share_denied) {
5527                         spin_unlock(&fp->fi_lock);
5528                         goto out;
5529                 }
5530                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5531                                 stp, open->op_share_deny, false))
5532                         status = nfserr_jukebox;
5533                 spin_unlock(&fp->fi_lock);
5534                 goto out;
5535         }
5536
5537         /* set access to the file */
5538         status = nfs4_file_get_access(fp, open->op_share_access);
5539         if (status != nfs_ok) {
5540                 if (status != nfserr_share_denied) {
5541                         spin_unlock(&fp->fi_lock);
5542                         goto out;
5543                 }
5544                 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5545                                 stp, open->op_share_access, true))
5546                         status = nfserr_jukebox;
5547                 spin_unlock(&fp->fi_lock);
5548                 goto out;
5549         }
5550
5551         /* Set access bits in stateid */
5552         old_access_bmap = stp->st_access_bmap;
5553         set_access(open->op_share_access, stp);
5554
5555         /* Set new deny mask */
5556         old_deny_bmap = stp->st_deny_bmap;
5557         set_deny(open->op_share_deny, stp);
5558         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5559
5560         if (!fp->fi_fds[oflag]) {
5561                 spin_unlock(&fp->fi_lock);
5562
5563                 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5564                                                   open->op_filp, &nf);
5565                 if (status != nfs_ok)
5566                         goto out_put_access;
5567
5568                 spin_lock(&fp->fi_lock);
5569                 if (!fp->fi_fds[oflag]) {
5570                         fp->fi_fds[oflag] = nf;
5571                         nf = NULL;
5572                 }
5573         }
5574         spin_unlock(&fp->fi_lock);
5575         if (nf)
5576                 nfsd_file_put(nf);
5577
5578         status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5579                                                                 access));
5580         if (status)
5581                 goto out_put_access;
5582
5583         status = nfsd4_truncate(rqstp, cur_fh, open);
5584         if (status)
5585                 goto out_put_access;
5586 out:
5587         return status;
5588 out_put_access:
5589         stp->st_access_bmap = old_access_bmap;
5590         nfs4_file_put_access(fp, open->op_share_access);
5591         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5592         goto out;
5593 }
5594
5595 static __be32
5596 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5597                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5598                 struct nfsd4_open *open)
5599 {
5600         __be32 status;
5601         unsigned char old_deny_bmap = stp->st_deny_bmap;
5602
5603         if (!test_access(open->op_share_access, stp))
5604                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5605
5606         /* test and set deny mode */
5607         spin_lock(&fp->fi_lock);
5608         status = nfs4_file_check_deny(fp, open->op_share_deny);
5609         switch (status) {
5610         case nfs_ok:
5611                 set_deny(open->op_share_deny, stp);
5612                 fp->fi_share_deny |=
5613                         (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5614                 break;
5615         case nfserr_share_denied:
5616                 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5617                                 stp, open->op_share_deny, false))
5618                         status = nfserr_jukebox;
5619                 break;
5620         }
5621         spin_unlock(&fp->fi_lock);
5622
5623         if (status != nfs_ok)
5624                 return status;
5625
5626         status = nfsd4_truncate(rqstp, cur_fh, open);
5627         if (status != nfs_ok)
5628                 reset_union_bmap_deny(old_deny_bmap, stp);
5629         return status;
5630 }
5631
5632 /* Should we give out recallable state?: */
5633 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5634 {
5635         if (clp->cl_cb_state == NFSD4_CB_UP)
5636                 return true;
5637         /*
5638          * In the sessions case, since we don't have to establish a
5639          * separate connection for callbacks, we assume it's OK
5640          * until we hear otherwise:
5641          */
5642         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5643 }
5644
5645 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5646                                                 int flag)
5647 {
5648         struct file_lease *fl;
5649
5650         fl = locks_alloc_lease();
5651         if (!fl)
5652                 return NULL;
5653         fl->fl_lmops = &nfsd_lease_mng_ops;
5654         fl->c.flc_flags = FL_DELEG;
5655         fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5656         fl->c.flc_owner = (fl_owner_t)dp;
5657         fl->c.flc_pid = current->tgid;
5658         fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5659         return fl;
5660 }
5661
5662 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5663                                          struct nfs4_file *fp)
5664 {
5665         struct nfs4_ol_stateid *st;
5666         struct file *f = fp->fi_deleg_file->nf_file;
5667         struct inode *ino = file_inode(f);
5668         int writes;
5669
5670         writes = atomic_read(&ino->i_writecount);
5671         if (!writes)
5672                 return 0;
5673         /*
5674          * There could be multiple filehandles (hence multiple
5675          * nfs4_files) referencing this file, but that's not too
5676          * common; let's just give up in that case rather than
5677          * trying to go look up all the clients using that other
5678          * nfs4_file as well:
5679          */
5680         if (fp->fi_aliased)
5681                 return -EAGAIN;
5682         /*
5683          * If there's a close in progress, make sure that we see it
5684          * clear any fi_fds[] entries before we see it decrement
5685          * i_writecount:
5686          */
5687         smp_mb__after_atomic();
5688
5689         if (fp->fi_fds[O_WRONLY])
5690                 writes--;
5691         if (fp->fi_fds[O_RDWR])
5692                 writes--;
5693         if (writes > 0)
5694                 return -EAGAIN; /* There may be non-NFSv4 writers */
5695         /*
5696          * It's possible there are non-NFSv4 write opens in progress,
5697          * but if they haven't incremented i_writecount yet then they
5698          * also haven't called break lease yet; so, they'll break this
5699          * lease soon enough.  So, all that's left to check for is NFSv4
5700          * opens:
5701          */
5702         spin_lock(&fp->fi_lock);
5703         list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5704                 if (st->st_openstp == NULL /* it's an open */ &&
5705                     access_permit_write(st) &&
5706                     st->st_stid.sc_client != clp) {
5707                         spin_unlock(&fp->fi_lock);
5708                         return -EAGAIN;
5709                 }
5710         }
5711         spin_unlock(&fp->fi_lock);
5712         /*
5713          * There's a small chance that we could be racing with another
5714          * NFSv4 open.  However, any open that hasn't added itself to
5715          * the fi_stateids list also hasn't called break_lease yet; so,
5716          * they'll break this lease soon enough.
5717          */
5718         return 0;
5719 }
5720
5721 /*
5722  * It's possible that between opening the dentry and setting the delegation,
5723  * that it has been renamed or unlinked. Redo the lookup to verify that this
5724  * hasn't happened.
5725  */
5726 static int
5727 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5728                           struct svc_fh *parent)
5729 {
5730         struct svc_export *exp;
5731         struct dentry *child;
5732         __be32 err;
5733
5734         err = nfsd_lookup_dentry(open->op_rqstp, parent,
5735                                  open->op_fname, open->op_fnamelen,
5736                                  &exp, &child);
5737
5738         if (err)
5739                 return -EAGAIN;
5740
5741         exp_put(exp);
5742         dput(child);
5743         if (child != file_dentry(fp->fi_deleg_file->nf_file))
5744                 return -EAGAIN;
5745
5746         return 0;
5747 }
5748
5749 /*
5750  * We avoid breaking delegations held by a client due to its own activity, but
5751  * clearing setuid/setgid bits on a write is an implicit activity and the client
5752  * may not notice and continue using the old mode. Avoid giving out a delegation
5753  * on setuid/setgid files when the client is requesting an open for write.
5754  */
5755 static int
5756 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5757 {
5758         struct inode *inode = file_inode(nf->nf_file);
5759
5760         if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5761             (inode->i_mode & (S_ISUID|S_ISGID)))
5762                 return -EAGAIN;
5763         return 0;
5764 }
5765
5766 static struct nfs4_delegation *
5767 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5768                     struct svc_fh *parent)
5769 {
5770         int status = 0;
5771         struct nfs4_client *clp = stp->st_stid.sc_client;
5772         struct nfs4_file *fp = stp->st_stid.sc_file;
5773         struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5774         struct nfs4_delegation *dp;
5775         struct nfsd_file *nf = NULL;
5776         struct file_lease *fl;
5777         u32 dl_type;
5778
5779         /*
5780          * The fi_had_conflict and nfs_get_existing_delegation checks
5781          * here are just optimizations; we'll need to recheck them at
5782          * the end:
5783          */
5784         if (fp->fi_had_conflict)
5785                 return ERR_PTR(-EAGAIN);
5786
5787         /*
5788          * Try for a write delegation first. RFC8881 section 10.4 says:
5789          *
5790          *  "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5791          *   on its own, all opens."
5792          *
5793          * Furthermore the client can use a write delegation for most READ
5794          * operations as well, so we require a O_RDWR file here.
5795          *
5796          * Offer a write delegation in the case of a BOTH open, and ensure
5797          * we get the O_RDWR descriptor.
5798          */
5799         if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5800                 nf = find_rw_file(fp);
5801                 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5802         }
5803
5804         /*
5805          * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5806          * file for some reason, then try for a read delegation instead.
5807          */
5808         if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5809                 nf = find_readable_file(fp);
5810                 dl_type = NFS4_OPEN_DELEGATE_READ;
5811         }
5812
5813         if (!nf)
5814                 return ERR_PTR(-EAGAIN);
5815
5816         spin_lock(&state_lock);
5817         spin_lock(&fp->fi_lock);
5818         if (nfs4_delegation_exists(clp, fp))
5819                 status = -EAGAIN;
5820         else if (nfsd4_verify_setuid_write(open, nf))
5821                 status = -EAGAIN;
5822         else if (!fp->fi_deleg_file) {
5823                 fp->fi_deleg_file = nf;
5824                 /* increment early to prevent fi_deleg_file from being
5825                  * cleared */
5826                 fp->fi_delegees = 1;
5827                 nf = NULL;
5828         } else
5829                 fp->fi_delegees++;
5830         spin_unlock(&fp->fi_lock);
5831         spin_unlock(&state_lock);
5832         if (nf)
5833                 nfsd_file_put(nf);
5834         if (status)
5835                 return ERR_PTR(status);
5836
5837         status = -ENOMEM;
5838         dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5839         if (!dp)
5840                 goto out_delegees;
5841
5842         fl = nfs4_alloc_init_lease(dp, dl_type);
5843         if (!fl)
5844                 goto out_clnt_odstate;
5845
5846         status = kernel_setlease(fp->fi_deleg_file->nf_file,
5847                                       fl->c.flc_type, &fl, NULL);
5848         if (fl)
5849                 locks_free_lease(fl);
5850         if (status)
5851                 goto out_clnt_odstate;
5852
5853         if (parent) {
5854                 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5855                 if (status)
5856                         goto out_unlock;
5857         }
5858
5859         status = nfsd4_check_conflicting_opens(clp, fp);
5860         if (status)
5861                 goto out_unlock;
5862
5863         /*
5864          * Now that the deleg is set, check again to ensure that nothing
5865          * raced in and changed the mode while we weren't looking.
5866          */
5867         status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5868         if (status)
5869                 goto out_unlock;
5870
5871         status = -EAGAIN;
5872         if (fp->fi_had_conflict)
5873                 goto out_unlock;
5874
5875         spin_lock(&state_lock);
5876         spin_lock(&clp->cl_lock);
5877         spin_lock(&fp->fi_lock);
5878         status = hash_delegation_locked(dp, fp);
5879         spin_unlock(&fp->fi_lock);
5880         spin_unlock(&clp->cl_lock);
5881         spin_unlock(&state_lock);
5882
5883         if (status)
5884                 goto out_unlock;
5885
5886         return dp;
5887 out_unlock:
5888         kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5889 out_clnt_odstate:
5890         put_clnt_odstate(dp->dl_clnt_odstate);
5891         nfs4_put_stid(&dp->dl_stid);
5892 out_delegees:
5893         put_deleg_file(fp);
5894         return ERR_PTR(status);
5895 }
5896
5897 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5898 {
5899         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5900         if (status == -EAGAIN)
5901                 open->op_why_no_deleg = WND4_CONTENTION;
5902         else {
5903                 open->op_why_no_deleg = WND4_RESOURCE;
5904                 switch (open->op_deleg_want) {
5905                 case NFS4_SHARE_WANT_READ_DELEG:
5906                 case NFS4_SHARE_WANT_WRITE_DELEG:
5907                 case NFS4_SHARE_WANT_ANY_DELEG:
5908                         break;
5909                 case NFS4_SHARE_WANT_CANCEL:
5910                         open->op_why_no_deleg = WND4_CANCELLED;
5911                         break;
5912                 case NFS4_SHARE_WANT_NO_DELEG:
5913                         WARN_ON_ONCE(1);
5914                 }
5915         }
5916 }
5917
5918 static bool
5919 nfs4_delegation_stat(struct nfs4_delegation *dp, struct svc_fh *currentfh,
5920                      struct kstat *stat)
5921 {
5922         struct nfsd_file *nf = find_rw_file(dp->dl_stid.sc_file);
5923         struct path path;
5924         int rc;
5925
5926         if (!nf)
5927                 return false;
5928
5929         path.mnt = currentfh->fh_export->ex_path.mnt;
5930         path.dentry = file_dentry(nf->nf_file);
5931
5932         rc = vfs_getattr(&path, stat,
5933                          (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5934                          AT_STATX_SYNC_AS_STAT);
5935
5936         nfsd_file_put(nf);
5937         return rc == 0;
5938 }
5939
5940 /*
5941  * The Linux NFS server does not offer write delegations to NFSv4.0
5942  * clients in order to avoid conflicts between write delegations and
5943  * GETATTRs requesting CHANGE or SIZE attributes.
5944  *
5945  * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5946  * begins each COMPOUND contains a client ID. Delegation recall can
5947  * be avoided when the server recognizes the client sending a
5948  * GETATTR also holds write delegation it conflicts with.
5949  *
5950  * However, the NFSv4.0 protocol does not enable a server to
5951  * determine that a GETATTR originated from the client holding the
5952  * conflicting delegation versus coming from some other client. Per
5953  * RFC 7530 Section 16.7.5, the server must recall or send a
5954  * CB_GETATTR even when the GETATTR originates from the client that
5955  * holds the conflicting delegation.
5956  *
5957  * An NFSv4.0 client can trigger a pathological situation if it
5958  * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5959  * the same COMPOUND. COMPOUND execution will always stop at the
5960  * GETATTR and the DELEGRETURN will never get executed. The server
5961  * eventually revokes the delegation, which can result in loss of
5962  * open or lock state.
5963  */
5964 static void
5965 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5966                      struct svc_fh *currentfh)
5967 {
5968         struct nfs4_delegation *dp;
5969         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5970         struct nfs4_client *clp = stp->st_stid.sc_client;
5971         struct svc_fh *parent = NULL;
5972         int cb_up;
5973         int status = 0;
5974         struct kstat stat;
5975
5976         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5977         open->op_recall = false;
5978         switch (open->op_claim_type) {
5979                 case NFS4_OPEN_CLAIM_PREVIOUS:
5980                         if (!cb_up)
5981                                 open->op_recall = true;
5982                         break;
5983                 case NFS4_OPEN_CLAIM_NULL:
5984                         parent = currentfh;
5985                         fallthrough;
5986                 case NFS4_OPEN_CLAIM_FH:
5987                         /*
5988                          * Let's not give out any delegations till everyone's
5989                          * had the chance to reclaim theirs, *and* until
5990                          * NLM locks have all been reclaimed:
5991                          */
5992                         if (locks_in_grace(clp->net))
5993                                 goto out_no_deleg;
5994                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5995                                 goto out_no_deleg;
5996                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5997                                         !clp->cl_minorversion)
5998                                 goto out_no_deleg;
5999                         break;
6000                 default:
6001                         goto out_no_deleg;
6002         }
6003         dp = nfs4_set_delegation(open, stp, parent);
6004         if (IS_ERR(dp))
6005                 goto out_no_deleg;
6006
6007         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
6008
6009         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
6010                 if (!nfs4_delegation_stat(dp, currentfh, &stat)) {
6011                         nfs4_put_stid(&dp->dl_stid);
6012                         destroy_delegation(dp);
6013                         goto out_no_deleg;
6014                 }
6015                 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
6016                 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
6017                 dp->dl_cb_fattr.ncf_initial_cinfo =
6018                         nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
6019                 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
6020         } else {
6021                 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
6022                 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
6023         }
6024         nfs4_put_stid(&dp->dl_stid);
6025         return;
6026 out_no_deleg:
6027         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
6028         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
6029             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
6030                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
6031                 open->op_recall = true;
6032         }
6033
6034         /* 4.1 client asking for a delegation? */
6035         if (open->op_deleg_want)
6036                 nfsd4_open_deleg_none_ext(open, status);
6037         return;
6038 }
6039
6040 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
6041                                         struct nfs4_delegation *dp)
6042 {
6043         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
6044             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6045                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6046                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
6047         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
6048                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6049                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6050                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
6051         }
6052         /* Otherwise the client must be confused wanting a delegation
6053          * it already has, therefore we don't return
6054          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
6055          */
6056 }
6057
6058 /**
6059  * nfsd4_process_open2 - finish open processing
6060  * @rqstp: the RPC transaction being executed
6061  * @current_fh: NFSv4 COMPOUND's current filehandle
6062  * @open: OPEN arguments
6063  *
6064  * If successful, (1) truncate the file if open->op_truncate was
6065  * set, (2) set open->op_stateid, (3) set open->op_delegation.
6066  *
6067  * Returns %nfs_ok on success; otherwise an nfs4stat value in
6068  * network byte order is returned.
6069  */
6070 __be32
6071 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6072 {
6073         struct nfsd4_compoundres *resp = rqstp->rq_resp;
6074         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6075         struct nfs4_file *fp = NULL;
6076         struct nfs4_ol_stateid *stp = NULL;
6077         struct nfs4_delegation *dp = NULL;
6078         __be32 status;
6079         bool new_stp = false;
6080
6081         /*
6082          * Lookup file; if found, lookup stateid and check open request,
6083          * and check for delegations in the process of being recalled.
6084          * If not found, create the nfs4_file struct
6085          */
6086         fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6087         if (unlikely(!fp))
6088                 return nfserr_jukebox;
6089         if (fp != open->op_file) {
6090                 status = nfs4_check_deleg(cl, open, &dp);
6091                 if (status)
6092                         goto out;
6093                 stp = nfsd4_find_and_lock_existing_open(fp, open);
6094         } else {
6095                 open->op_file = NULL;
6096                 status = nfserr_bad_stateid;
6097                 if (nfsd4_is_deleg_cur(open))
6098                         goto out;
6099         }
6100
6101         if (!stp) {
6102                 stp = init_open_stateid(fp, open);
6103                 if (!open->op_stp)
6104                         new_stp = true;
6105         }
6106
6107         /*
6108          * OPEN the file, or upgrade an existing OPEN.
6109          * If truncate fails, the OPEN fails.
6110          *
6111          * stp is already locked.
6112          */
6113         if (!new_stp) {
6114                 /* Stateid was found, this is an OPEN upgrade */
6115                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6116                 if (status) {
6117                         mutex_unlock(&stp->st_mutex);
6118                         goto out;
6119                 }
6120         } else {
6121                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6122                 if (status) {
6123                         release_open_stateid(stp);
6124                         mutex_unlock(&stp->st_mutex);
6125                         goto out;
6126                 }
6127
6128                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6129                                                         open->op_odstate);
6130                 if (stp->st_clnt_odstate == open->op_odstate)
6131                         open->op_odstate = NULL;
6132         }
6133
6134         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6135         mutex_unlock(&stp->st_mutex);
6136
6137         if (nfsd4_has_session(&resp->cstate)) {
6138                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6139                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6140                         open->op_why_no_deleg = WND4_NOT_WANTED;
6141                         goto nodeleg;
6142                 }
6143         }
6144
6145         /*
6146         * Attempt to hand out a delegation. No error return, because the
6147         * OPEN succeeds even if we fail.
6148         */
6149         nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6150 nodeleg:
6151         status = nfs_ok;
6152         trace_nfsd_open(&stp->st_stid.sc_stateid);
6153 out:
6154         /* 4.1 client trying to upgrade/downgrade delegation? */
6155         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6156             open->op_deleg_want)
6157                 nfsd4_deleg_xgrade_none_ext(open, dp);
6158
6159         if (fp)
6160                 put_nfs4_file(fp);
6161         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6162                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6163         /*
6164         * To finish the open response, we just need to set the rflags.
6165         */
6166         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6167         if (nfsd4_has_session(&resp->cstate))
6168                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6169         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6170                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6171
6172         if (dp)
6173                 nfs4_put_stid(&dp->dl_stid);
6174         if (stp)
6175                 nfs4_put_stid(&stp->st_stid);
6176
6177         return status;
6178 }
6179
6180 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6181                               struct nfsd4_open *open)
6182 {
6183         if (open->op_openowner)
6184                 nfs4_put_stateowner(&open->op_openowner->oo_owner);
6185         if (open->op_file)
6186                 kmem_cache_free(file_slab, open->op_file);
6187         if (open->op_stp)
6188                 nfs4_put_stid(&open->op_stp->st_stid);
6189         if (open->op_odstate)
6190                 kmem_cache_free(odstate_slab, open->op_odstate);
6191 }
6192
6193 __be32
6194 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6195             union nfsd4_op_u *u)
6196 {
6197         clientid_t *clid = &u->renew;
6198         struct nfs4_client *clp;
6199         __be32 status;
6200         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6201
6202         trace_nfsd_clid_renew(clid);
6203         status = set_client(clid, cstate, nn);
6204         if (status)
6205                 return status;
6206         clp = cstate->clp;
6207         if (!list_empty(&clp->cl_delegations)
6208                         && clp->cl_cb_state != NFSD4_CB_UP)
6209                 return nfserr_cb_path_down;
6210         return nfs_ok;
6211 }
6212
6213 void
6214 nfsd4_end_grace(struct nfsd_net *nn)
6215 {
6216         /* do nothing if grace period already ended */
6217         if (nn->grace_ended)
6218                 return;
6219
6220         trace_nfsd_grace_complete(nn);
6221         nn->grace_ended = true;
6222         /*
6223          * If the server goes down again right now, an NFSv4
6224          * client will still be allowed to reclaim after it comes back up,
6225          * even if it hasn't yet had a chance to reclaim state this time.
6226          *
6227          */
6228         nfsd4_record_grace_done(nn);
6229         /*
6230          * At this point, NFSv4 clients can still reclaim.  But if the
6231          * server crashes, any that have not yet reclaimed will be out
6232          * of luck on the next boot.
6233          *
6234          * (NFSv4.1+ clients are considered to have reclaimed once they
6235          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
6236          * have reclaimed after their first OPEN.)
6237          */
6238         locks_end_grace(&nn->nfsd4_manager);
6239         /*
6240          * At this point, and once lockd and/or any other containers
6241          * exit their grace period, further reclaims will fail and
6242          * regular locking can resume.
6243          */
6244 }
6245
6246 /*
6247  * If we've waited a lease period but there are still clients trying to
6248  * reclaim, wait a little longer to give them a chance to finish.
6249  */
6250 static bool clients_still_reclaiming(struct nfsd_net *nn)
6251 {
6252         time64_t double_grace_period_end = nn->boot_time +
6253                                            2 * nn->nfsd4_lease;
6254
6255         if (nn->track_reclaim_completes &&
6256                         atomic_read(&nn->nr_reclaim_complete) ==
6257                         nn->reclaim_str_hashtbl_size)
6258                 return false;
6259         if (!nn->somebody_reclaimed)
6260                 return false;
6261         nn->somebody_reclaimed = false;
6262         /*
6263          * If we've given them *two* lease times to reclaim, and they're
6264          * still not done, give up:
6265          */
6266         if (ktime_get_boottime_seconds() > double_grace_period_end)
6267                 return false;
6268         return true;
6269 }
6270
6271 struct laundry_time {
6272         time64_t cutoff;
6273         time64_t new_timeo;
6274 };
6275
6276 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6277 {
6278         time64_t time_remaining;
6279
6280         if (last_refresh < lt->cutoff)
6281                 return true;
6282         time_remaining = last_refresh - lt->cutoff;
6283         lt->new_timeo = min(lt->new_timeo, time_remaining);
6284         return false;
6285 }
6286
6287 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6288 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6289 {
6290         spin_lock_init(&nn->nfsd_ssc_lock);
6291         INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6292         init_waitqueue_head(&nn->nfsd_ssc_waitq);
6293 }
6294
6295 /*
6296  * This is called when nfsd is being shutdown, after all inter_ssc
6297  * cleanup were done, to destroy the ssc delayed unmount list.
6298  */
6299 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6300 {
6301         struct nfsd4_ssc_umount_item *ni = NULL;
6302         struct nfsd4_ssc_umount_item *tmp;
6303
6304         spin_lock(&nn->nfsd_ssc_lock);
6305         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6306                 list_del(&ni->nsui_list);
6307                 spin_unlock(&nn->nfsd_ssc_lock);
6308                 mntput(ni->nsui_vfsmount);
6309                 kfree(ni);
6310                 spin_lock(&nn->nfsd_ssc_lock);
6311         }
6312         spin_unlock(&nn->nfsd_ssc_lock);
6313 }
6314
6315 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6316 {
6317         bool do_wakeup = false;
6318         struct nfsd4_ssc_umount_item *ni = NULL;
6319         struct nfsd4_ssc_umount_item *tmp;
6320
6321         spin_lock(&nn->nfsd_ssc_lock);
6322         list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6323                 if (time_after(jiffies, ni->nsui_expire)) {
6324                         if (refcount_read(&ni->nsui_refcnt) > 1)
6325                                 continue;
6326
6327                         /* mark being unmount */
6328                         ni->nsui_busy = true;
6329                         spin_unlock(&nn->nfsd_ssc_lock);
6330                         mntput(ni->nsui_vfsmount);
6331                         spin_lock(&nn->nfsd_ssc_lock);
6332
6333                         /* waiters need to start from begin of list */
6334                         list_del(&ni->nsui_list);
6335                         kfree(ni);
6336
6337                         /* wakeup ssc_connect waiters */
6338                         do_wakeup = true;
6339                         continue;
6340                 }
6341                 break;
6342         }
6343         if (do_wakeup)
6344                 wake_up_all(&nn->nfsd_ssc_waitq);
6345         spin_unlock(&nn->nfsd_ssc_lock);
6346 }
6347 #endif
6348
6349 /* Check if any lock belonging to this lockowner has any blockers */
6350 static bool
6351 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6352 {
6353         struct file_lock_context *ctx;
6354         struct nfs4_ol_stateid *stp;
6355         struct nfs4_file *nf;
6356
6357         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6358                 nf = stp->st_stid.sc_file;
6359                 ctx = locks_inode_context(nf->fi_inode);
6360                 if (!ctx)
6361                         continue;
6362                 if (locks_owner_has_blockers(ctx, lo))
6363                         return true;
6364         }
6365         return false;
6366 }
6367
6368 static bool
6369 nfs4_anylock_blockers(struct nfs4_client *clp)
6370 {
6371         int i;
6372         struct nfs4_stateowner *so;
6373         struct nfs4_lockowner *lo;
6374
6375         if (atomic_read(&clp->cl_delegs_in_recall))
6376                 return true;
6377         spin_lock(&clp->cl_lock);
6378         for (i = 0; i < OWNER_HASH_SIZE; i++) {
6379                 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6380                                 so_strhash) {
6381                         if (so->so_is_open_owner)
6382                                 continue;
6383                         lo = lockowner(so);
6384                         if (nfs4_lockowner_has_blockers(lo)) {
6385                                 spin_unlock(&clp->cl_lock);
6386                                 return true;
6387                         }
6388                 }
6389         }
6390         spin_unlock(&clp->cl_lock);
6391         return false;
6392 }
6393
6394 static void
6395 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6396                                 struct laundry_time *lt)
6397 {
6398         unsigned int maxreap, reapcnt = 0;
6399         struct list_head *pos, *next;
6400         struct nfs4_client *clp;
6401
6402         maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6403                         NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6404         INIT_LIST_HEAD(reaplist);
6405         spin_lock(&nn->client_lock);
6406         list_for_each_safe(pos, next, &nn->client_lru) {
6407                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6408                 if (clp->cl_state == NFSD4_EXPIRABLE)
6409                         goto exp_client;
6410                 if (!state_expired(lt, clp->cl_time))
6411                         break;
6412                 if (!atomic_read(&clp->cl_rpc_users)) {
6413                         if (clp->cl_state == NFSD4_ACTIVE)
6414                                 atomic_inc(&nn->nfsd_courtesy_clients);
6415                         clp->cl_state = NFSD4_COURTESY;
6416                 }
6417                 if (!client_has_state(clp))
6418                         goto exp_client;
6419                 if (!nfs4_anylock_blockers(clp))
6420                         if (reapcnt >= maxreap)
6421                                 continue;
6422 exp_client:
6423                 if (!mark_client_expired_locked(clp)) {
6424                         list_add(&clp->cl_lru, reaplist);
6425                         reapcnt++;
6426                 }
6427         }
6428         spin_unlock(&nn->client_lock);
6429 }
6430
6431 static void
6432 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6433                                 struct list_head *reaplist)
6434 {
6435         unsigned int maxreap = 0, reapcnt = 0;
6436         struct list_head *pos, *next;
6437         struct nfs4_client *clp;
6438
6439         maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6440         INIT_LIST_HEAD(reaplist);
6441
6442         spin_lock(&nn->client_lock);
6443         list_for_each_safe(pos, next, &nn->client_lru) {
6444                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6445                 if (clp->cl_state == NFSD4_ACTIVE)
6446                         break;
6447                 if (reapcnt >= maxreap)
6448                         break;
6449                 if (!mark_client_expired_locked(clp)) {
6450                         list_add(&clp->cl_lru, reaplist);
6451                         reapcnt++;
6452                 }
6453         }
6454         spin_unlock(&nn->client_lock);
6455 }
6456
6457 static void
6458 nfs4_process_client_reaplist(struct list_head *reaplist)
6459 {
6460         struct list_head *pos, *next;
6461         struct nfs4_client *clp;
6462
6463         list_for_each_safe(pos, next, reaplist) {
6464                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6465                 trace_nfsd_clid_purged(&clp->cl_clientid);
6466                 list_del_init(&clp->cl_lru);
6467                 expire_client(clp);
6468         }
6469 }
6470
6471 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6472                                       struct laundry_time *lt)
6473 {
6474         struct nfs4_client *clp;
6475
6476         spin_lock(&nn->client_lock);
6477         if (nn->nfs40_last_revoke == 0 ||
6478             nn->nfs40_last_revoke > lt->cutoff) {
6479                 spin_unlock(&nn->client_lock);
6480                 return;
6481         }
6482         nn->nfs40_last_revoke = 0;
6483
6484 retry:
6485         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6486                 unsigned long id, tmp;
6487                 struct nfs4_stid *stid;
6488
6489                 if (atomic_read(&clp->cl_admin_revoked) == 0)
6490                         continue;
6491
6492                 spin_lock(&clp->cl_lock);
6493                 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6494                         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6495                                 refcount_inc(&stid->sc_count);
6496                                 spin_unlock(&nn->client_lock);
6497                                 /* this function drops ->cl_lock */
6498                                 nfsd4_drop_revoked_stid(stid);
6499                                 nfs4_put_stid(stid);
6500                                 spin_lock(&nn->client_lock);
6501                                 goto retry;
6502                         }
6503                 spin_unlock(&clp->cl_lock);
6504         }
6505         spin_unlock(&nn->client_lock);
6506 }
6507
6508 static time64_t
6509 nfs4_laundromat(struct nfsd_net *nn)
6510 {
6511         struct nfs4_openowner *oo;
6512         struct nfs4_delegation *dp;
6513         struct nfs4_ol_stateid *stp;
6514         struct nfsd4_blocked_lock *nbl;
6515         struct list_head *pos, *next, reaplist;
6516         struct laundry_time lt = {
6517                 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6518                 .new_timeo = nn->nfsd4_lease
6519         };
6520         struct nfs4_cpntf_state *cps;
6521         copy_stateid_t *cps_t;
6522         int i;
6523
6524         if (clients_still_reclaiming(nn)) {
6525                 lt.new_timeo = 0;
6526                 goto out;
6527         }
6528         nfsd4_end_grace(nn);
6529
6530         spin_lock(&nn->s2s_cp_lock);
6531         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6532                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6533                 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6534                                 state_expired(&lt, cps->cpntf_time))
6535                         _free_cpntf_state_locked(nn, cps);
6536         }
6537         spin_unlock(&nn->s2s_cp_lock);
6538         nfs4_get_client_reaplist(nn, &reaplist, &lt);
6539         nfs4_process_client_reaplist(&reaplist);
6540
6541         nfs40_clean_admin_revoked(nn, &lt);
6542
6543         spin_lock(&state_lock);
6544         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6545                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6546                 if (!state_expired(&lt, dp->dl_time))
6547                         break;
6548                 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6549                 list_add(&dp->dl_recall_lru, &reaplist);
6550         }
6551         spin_unlock(&state_lock);
6552         while (!list_empty(&reaplist)) {
6553                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6554                                         dl_recall_lru);
6555                 list_del_init(&dp->dl_recall_lru);
6556                 revoke_delegation(dp);
6557         }
6558
6559         spin_lock(&nn->client_lock);
6560         while (!list_empty(&nn->close_lru)) {
6561                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6562                                         oo_close_lru);
6563                 if (!state_expired(&lt, oo->oo_time))
6564                         break;
6565                 list_del_init(&oo->oo_close_lru);
6566                 stp = oo->oo_last_closed_stid;
6567                 oo->oo_last_closed_stid = NULL;
6568                 spin_unlock(&nn->client_lock);
6569                 nfs4_put_stid(&stp->st_stid);
6570                 spin_lock(&nn->client_lock);
6571         }
6572         spin_unlock(&nn->client_lock);
6573
6574         /*
6575          * It's possible for a client to try and acquire an already held lock
6576          * that is being held for a long time, and then lose interest in it.
6577          * So, we clean out any un-revisited request after a lease period
6578          * under the assumption that the client is no longer interested.
6579          *
6580          * RFC5661, sec. 9.6 states that the client must not rely on getting
6581          * notifications and must continue to poll for locks, even when the
6582          * server supports them. Thus this shouldn't lead to clients blocking
6583          * indefinitely once the lock does become free.
6584          */
6585         BUG_ON(!list_empty(&reaplist));
6586         spin_lock(&nn->blocked_locks_lock);
6587         while (!list_empty(&nn->blocked_locks_lru)) {
6588                 nbl = list_first_entry(&nn->blocked_locks_lru,
6589                                         struct nfsd4_blocked_lock, nbl_lru);
6590                 if (!state_expired(&lt, nbl->nbl_time))
6591                         break;
6592                 list_move(&nbl->nbl_lru, &reaplist);
6593                 list_del_init(&nbl->nbl_list);
6594         }
6595         spin_unlock(&nn->blocked_locks_lock);
6596
6597         while (!list_empty(&reaplist)) {
6598                 nbl = list_first_entry(&reaplist,
6599                                         struct nfsd4_blocked_lock, nbl_lru);
6600                 list_del_init(&nbl->nbl_lru);
6601                 free_blocked_lock(nbl);
6602         }
6603 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6604         /* service the server-to-server copy delayed unmount list */
6605         nfsd4_ssc_expire_umount(nn);
6606 #endif
6607         if (atomic_long_read(&num_delegations) >= max_delegations)
6608                 deleg_reaper(nn);
6609 out:
6610         return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6611 }
6612
6613 static void laundromat_main(struct work_struct *);
6614
6615 static void
6616 laundromat_main(struct work_struct *laundry)
6617 {
6618         time64_t t;
6619         struct delayed_work *dwork = to_delayed_work(laundry);
6620         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6621                                            laundromat_work);
6622
6623         t = nfs4_laundromat(nn);
6624         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6625 }
6626
6627 static void
6628 courtesy_client_reaper(struct nfsd_net *nn)
6629 {
6630         struct list_head reaplist;
6631
6632         nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6633         nfs4_process_client_reaplist(&reaplist);
6634 }
6635
6636 static void
6637 deleg_reaper(struct nfsd_net *nn)
6638 {
6639         struct list_head *pos, *next;
6640         struct nfs4_client *clp;
6641         LIST_HEAD(cblist);
6642
6643         spin_lock(&nn->client_lock);
6644         list_for_each_safe(pos, next, &nn->client_lru) {
6645                 clp = list_entry(pos, struct nfs4_client, cl_lru);
6646                 if (clp->cl_state != NFSD4_ACTIVE ||
6647                         list_empty(&clp->cl_delegations) ||
6648                         atomic_read(&clp->cl_delegs_in_recall) ||
6649                         test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6650                         (ktime_get_boottime_seconds() -
6651                                 clp->cl_ra_time < 5)) {
6652                         continue;
6653                 }
6654                 list_add(&clp->cl_ra_cblist, &cblist);
6655
6656                 /* release in nfsd4_cb_recall_any_release */
6657                 kref_get(&clp->cl_nfsdfs.cl_ref);
6658                 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6659                 clp->cl_ra_time = ktime_get_boottime_seconds();
6660         }
6661         spin_unlock(&nn->client_lock);
6662
6663         while (!list_empty(&cblist)) {
6664                 clp = list_first_entry(&cblist, struct nfs4_client,
6665                                         cl_ra_cblist);
6666                 list_del_init(&clp->cl_ra_cblist);
6667                 clp->cl_ra->ra_keep = 0;
6668                 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6669                                                 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6670                 trace_nfsd_cb_recall_any(clp->cl_ra);
6671                 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6672         }
6673 }
6674
6675 static void
6676 nfsd4_state_shrinker_worker(struct work_struct *work)
6677 {
6678         struct nfsd_net *nn = container_of(work, struct nfsd_net,
6679                                 nfsd_shrinker_work);
6680
6681         courtesy_client_reaper(nn);
6682         deleg_reaper(nn);
6683 }
6684
6685 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6686 {
6687         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6688                 return nfserr_bad_stateid;
6689         return nfs_ok;
6690 }
6691
6692 static
6693 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6694 {
6695         __be32 status = nfserr_openmode;
6696
6697         /* For lock stateid's, we test the parent open, not the lock: */
6698         if (stp->st_openstp)
6699                 stp = stp->st_openstp;
6700         if ((flags & WR_STATE) && !access_permit_write(stp))
6701                 goto out;
6702         if ((flags & RD_STATE) && !access_permit_read(stp))
6703                 goto out;
6704         status = nfs_ok;
6705 out:
6706         return status;
6707 }
6708
6709 static inline __be32
6710 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6711 {
6712         if (ONE_STATEID(stateid) && (flags & RD_STATE))
6713                 return nfs_ok;
6714         else if (opens_in_grace(net)) {
6715                 /* Answer in remaining cases depends on existence of
6716                  * conflicting state; so we must wait out the grace period. */
6717                 return nfserr_grace;
6718         } else if (flags & WR_STATE)
6719                 return nfs4_share_conflict(current_fh,
6720                                 NFS4_SHARE_DENY_WRITE);
6721         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6722                 return nfs4_share_conflict(current_fh,
6723                                 NFS4_SHARE_DENY_READ);
6724 }
6725
6726 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6727 {
6728         /*
6729          * When sessions are used the stateid generation number is ignored
6730          * when it is zero.
6731          */
6732         if (has_session && in->si_generation == 0)
6733                 return nfs_ok;
6734
6735         if (in->si_generation == ref->si_generation)
6736                 return nfs_ok;
6737
6738         /* If the client sends us a stateid from the future, it's buggy: */
6739         if (nfsd4_stateid_generation_after(in, ref))
6740                 return nfserr_bad_stateid;
6741         /*
6742          * However, we could see a stateid from the past, even from a
6743          * non-buggy client.  For example, if the client sends a lock
6744          * while some IO is outstanding, the lock may bump si_generation
6745          * while the IO is still in flight.  The client could avoid that
6746          * situation by waiting for responses on all the IO requests,
6747          * but better performance may result in retrying IO that
6748          * receives an old_stateid error if requests are rarely
6749          * reordered in flight:
6750          */
6751         return nfserr_old_stateid;
6752 }
6753
6754 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6755 {
6756         __be32 ret;
6757
6758         spin_lock(&s->sc_lock);
6759         ret = nfsd4_verify_open_stid(s);
6760         if (ret == nfs_ok)
6761                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6762         spin_unlock(&s->sc_lock);
6763         if (ret == nfserr_admin_revoked)
6764                 nfsd40_drop_revoked_stid(s->sc_client,
6765                                         &s->sc_stateid);
6766         return ret;
6767 }
6768
6769 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6770 {
6771         if (ols->st_stateowner->so_is_open_owner &&
6772             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6773                 return nfserr_bad_stateid;
6774         return nfs_ok;
6775 }
6776
6777 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6778 {
6779         struct nfs4_stid *s;
6780         __be32 status = nfserr_bad_stateid;
6781
6782         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6783                 CLOSE_STATEID(stateid))
6784                 return status;
6785         spin_lock(&cl->cl_lock);
6786         s = find_stateid_locked(cl, stateid);
6787         if (!s)
6788                 goto out_unlock;
6789         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6790         if (status)
6791                 goto out_unlock;
6792         status = nfsd4_verify_open_stid(s);
6793         if (status)
6794                 goto out_unlock;
6795
6796         switch (s->sc_type) {
6797         case SC_TYPE_DELEG:
6798                 status = nfs_ok;
6799                 break;
6800         case SC_TYPE_OPEN:
6801         case SC_TYPE_LOCK:
6802                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6803                 break;
6804         default:
6805                 printk("unknown stateid type %x\n", s->sc_type);
6806                 status = nfserr_bad_stateid;
6807         }
6808 out_unlock:
6809         spin_unlock(&cl->cl_lock);
6810         if (status == nfserr_admin_revoked)
6811                 nfsd40_drop_revoked_stid(cl, stateid);
6812         return status;
6813 }
6814
6815 __be32
6816 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6817                      stateid_t *stateid,
6818                      unsigned short typemask, unsigned short statusmask,
6819                      struct nfs4_stid **s, struct nfsd_net *nn)
6820 {
6821         __be32 status;
6822         struct nfs4_stid *stid;
6823         bool return_revoked = false;
6824
6825         /*
6826          *  only return revoked delegations if explicitly asked.
6827          *  otherwise we report revoked or bad_stateid status.
6828          */
6829         if (statusmask & SC_STATUS_REVOKED)
6830                 return_revoked = true;
6831         if (typemask & SC_TYPE_DELEG)
6832                 /* Always allow REVOKED for DELEG so we can
6833                  * retturn the appropriate error.
6834                  */
6835                 statusmask |= SC_STATUS_REVOKED;
6836
6837         statusmask |= SC_STATUS_ADMIN_REVOKED;
6838
6839         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6840                 CLOSE_STATEID(stateid))
6841                 return nfserr_bad_stateid;
6842         status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6843         if (status == nfserr_stale_clientid) {
6844                 if (cstate->session)
6845                         return nfserr_bad_stateid;
6846                 return nfserr_stale_stateid;
6847         }
6848         if (status)
6849                 return status;
6850         stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6851         if (!stid)
6852                 return nfserr_bad_stateid;
6853         if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6854                 nfs4_put_stid(stid);
6855                 return nfserr_deleg_revoked;
6856         }
6857         if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6858                 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6859                 nfs4_put_stid(stid);
6860                 return nfserr_admin_revoked;
6861         }
6862         *s = stid;
6863         return nfs_ok;
6864 }
6865
6866 static struct nfsd_file *
6867 nfs4_find_file(struct nfs4_stid *s, int flags)
6868 {
6869         struct nfsd_file *ret = NULL;
6870
6871         if (!s || s->sc_status)
6872                 return NULL;
6873
6874         switch (s->sc_type) {
6875         case SC_TYPE_DELEG:
6876                 spin_lock(&s->sc_file->fi_lock);
6877                 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6878                 spin_unlock(&s->sc_file->fi_lock);
6879                 break;
6880         case SC_TYPE_OPEN:
6881         case SC_TYPE_LOCK:
6882                 if (flags & RD_STATE)
6883                         ret = find_readable_file(s->sc_file);
6884                 else
6885                         ret = find_writeable_file(s->sc_file);
6886         }
6887
6888         return ret;
6889 }
6890
6891 static __be32
6892 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6893 {
6894         __be32 status;
6895
6896         status = nfsd4_check_openowner_confirmed(ols);
6897         if (status)
6898                 return status;
6899         return nfs4_check_openmode(ols, flags);
6900 }
6901
6902 static __be32
6903 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6904                 struct nfsd_file **nfp, int flags)
6905 {
6906         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6907         struct nfsd_file *nf;
6908         __be32 status;
6909
6910         nf = nfs4_find_file(s, flags);
6911         if (nf) {
6912                 status = nfsd_permission(&rqstp->rq_cred,
6913                                          fhp->fh_export, fhp->fh_dentry,
6914                                 acc | NFSD_MAY_OWNER_OVERRIDE);
6915                 if (status) {
6916                         nfsd_file_put(nf);
6917                         goto out;
6918                 }
6919         } else {
6920                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6921                 if (status)
6922                         return status;
6923         }
6924         *nfp = nf;
6925 out:
6926         return status;
6927 }
6928 static void
6929 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6930 {
6931         WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6932         if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6933                 return;
6934         list_del(&cps->cp_list);
6935         idr_remove(&nn->s2s_cp_stateids,
6936                    cps->cp_stateid.cs_stid.si_opaque.so_id);
6937         kfree(cps);
6938 }
6939 /*
6940  * A READ from an inter server to server COPY will have a
6941  * copy stateid. Look up the copy notify stateid from the
6942  * idr structure and take a reference on it.
6943  */
6944 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6945                           struct nfs4_client *clp,
6946                           struct nfs4_cpntf_state **cps)
6947 {
6948         copy_stateid_t *cps_t;
6949         struct nfs4_cpntf_state *state = NULL;
6950
6951         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6952                 return nfserr_bad_stateid;
6953         spin_lock(&nn->s2s_cp_lock);
6954         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6955         if (cps_t) {
6956                 state = container_of(cps_t, struct nfs4_cpntf_state,
6957                                      cp_stateid);
6958                 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6959                         state = NULL;
6960                         goto unlock;
6961                 }
6962                 if (!clp)
6963                         refcount_inc(&state->cp_stateid.cs_count);
6964                 else
6965                         _free_cpntf_state_locked(nn, state);
6966         }
6967 unlock:
6968         spin_unlock(&nn->s2s_cp_lock);
6969         if (!state)
6970                 return nfserr_bad_stateid;
6971         if (!clp)
6972                 *cps = state;
6973         return 0;
6974 }
6975
6976 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6977                                struct nfs4_stid **stid)
6978 {
6979         __be32 status;
6980         struct nfs4_cpntf_state *cps = NULL;
6981         struct nfs4_client *found;
6982
6983         status = manage_cpntf_state(nn, st, NULL, &cps);
6984         if (status)
6985                 return status;
6986
6987         cps->cpntf_time = ktime_get_boottime_seconds();
6988
6989         status = nfserr_expired;
6990         found = lookup_clientid(&cps->cp_p_clid, true, nn);
6991         if (!found)
6992                 goto out;
6993
6994         *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6995                                      SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6996                                      0);
6997         if (*stid)
6998                 status = nfs_ok;
6999         else
7000                 status = nfserr_bad_stateid;
7001
7002         put_client_renew(found);
7003 out:
7004         nfs4_put_cpntf_state(nn, cps);
7005         return status;
7006 }
7007
7008 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
7009 {
7010         spin_lock(&nn->s2s_cp_lock);
7011         _free_cpntf_state_locked(nn, cps);
7012         spin_unlock(&nn->s2s_cp_lock);
7013 }
7014
7015 /**
7016  * nfs4_preprocess_stateid_op - find and prep stateid for an operation
7017  * @rqstp: incoming request from client
7018  * @cstate: current compound state
7019  * @fhp: filehandle associated with requested stateid
7020  * @stateid: stateid (provided by client)
7021  * @flags: flags describing type of operation to be done
7022  * @nfp: optional nfsd_file return pointer (may be NULL)
7023  * @cstid: optional returned nfs4_stid pointer (may be NULL)
7024  *
7025  * Given info from the client, look up a nfs4_stid for the operation. On
7026  * success, it returns a reference to the nfs4_stid and/or the nfsd_file
7027  * associated with it.
7028  */
7029 __be32
7030 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
7031                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
7032                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
7033                 struct nfs4_stid **cstid)
7034 {
7035         struct net *net = SVC_NET(rqstp);
7036         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7037         struct nfs4_stid *s = NULL;
7038         __be32 status;
7039
7040         if (nfp)
7041                 *nfp = NULL;
7042
7043         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
7044                 status = check_special_stateids(net, fhp, stateid, flags);
7045                 goto done;
7046         }
7047
7048         status = nfsd4_lookup_stateid(cstate, stateid,
7049                                 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7050                                 0, &s, nn);
7051         if (status == nfserr_bad_stateid)
7052                 status = find_cpntf_state(nn, stateid, &s);
7053         if (status)
7054                 return status;
7055         status = nfsd4_stid_check_stateid_generation(stateid, s,
7056                         nfsd4_has_session(cstate));
7057         if (status)
7058                 goto out;
7059
7060         switch (s->sc_type) {
7061         case SC_TYPE_DELEG:
7062                 status = nfs4_check_delegmode(delegstateid(s), flags);
7063                 break;
7064         case SC_TYPE_OPEN:
7065         case SC_TYPE_LOCK:
7066                 status = nfs4_check_olstateid(openlockstateid(s), flags);
7067                 break;
7068         }
7069         if (status)
7070                 goto out;
7071         status = nfs4_check_fh(fhp, s);
7072
7073 done:
7074         if (status == nfs_ok && nfp)
7075                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7076 out:
7077         if (s) {
7078                 if (!status && cstid)
7079                         *cstid = s;
7080                 else
7081                         nfs4_put_stid(s);
7082         }
7083         return status;
7084 }
7085
7086 /*
7087  * Test if the stateid is valid
7088  */
7089 __be32
7090 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7091                    union nfsd4_op_u *u)
7092 {
7093         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7094         struct nfsd4_test_stateid_id *stateid;
7095         struct nfs4_client *cl = cstate->clp;
7096
7097         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7098                 stateid->ts_id_status =
7099                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7100
7101         return nfs_ok;
7102 }
7103
7104 static __be32
7105 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7106 {
7107         struct nfs4_ol_stateid *stp = openlockstateid(s);
7108         __be32 ret;
7109
7110         ret = nfsd4_lock_ol_stateid(stp);
7111         if (ret)
7112                 goto out_put_stid;
7113
7114         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7115         if (ret)
7116                 goto out;
7117
7118         ret = nfserr_locks_held;
7119         if (check_for_locks(stp->st_stid.sc_file,
7120                             lockowner(stp->st_stateowner)))
7121                 goto out;
7122
7123         release_lock_stateid(stp);
7124         ret = nfs_ok;
7125
7126 out:
7127         mutex_unlock(&stp->st_mutex);
7128 out_put_stid:
7129         nfs4_put_stid(s);
7130         return ret;
7131 }
7132
7133 __be32
7134 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7135                    union nfsd4_op_u *u)
7136 {
7137         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7138         stateid_t *stateid = &free_stateid->fr_stateid;
7139         struct nfs4_stid *s;
7140         struct nfs4_delegation *dp;
7141         struct nfs4_client *cl = cstate->clp;
7142         __be32 ret = nfserr_bad_stateid;
7143
7144         spin_lock(&cl->cl_lock);
7145         s = find_stateid_locked(cl, stateid);
7146         if (!s || s->sc_status & SC_STATUS_CLOSED)
7147                 goto out_unlock;
7148         if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7149                 nfsd4_drop_revoked_stid(s);
7150                 ret = nfs_ok;
7151                 goto out;
7152         }
7153         spin_lock(&s->sc_lock);
7154         switch (s->sc_type) {
7155         case SC_TYPE_DELEG:
7156                 if (s->sc_status & SC_STATUS_REVOKED) {
7157                         spin_unlock(&s->sc_lock);
7158                         dp = delegstateid(s);
7159                         list_del_init(&dp->dl_recall_lru);
7160                         spin_unlock(&cl->cl_lock);
7161                         nfs4_put_stid(s);
7162                         ret = nfs_ok;
7163                         goto out;
7164                 }
7165                 ret = nfserr_locks_held;
7166                 break;
7167         case SC_TYPE_OPEN:
7168                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7169                 if (ret)
7170                         break;
7171                 ret = nfserr_locks_held;
7172                 break;
7173         case SC_TYPE_LOCK:
7174                 spin_unlock(&s->sc_lock);
7175                 refcount_inc(&s->sc_count);
7176                 spin_unlock(&cl->cl_lock);
7177                 ret = nfsd4_free_lock_stateid(stateid, s);
7178                 goto out;
7179         }
7180         spin_unlock(&s->sc_lock);
7181 out_unlock:
7182         spin_unlock(&cl->cl_lock);
7183 out:
7184         return ret;
7185 }
7186
7187 static inline int
7188 setlkflg (int type)
7189 {
7190         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7191                 RD_STATE : WR_STATE;
7192 }
7193
7194 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7195 {
7196         struct svc_fh *current_fh = &cstate->current_fh;
7197         struct nfs4_stateowner *sop = stp->st_stateowner;
7198         __be32 status;
7199
7200         status = nfsd4_check_seqid(cstate, sop, seqid);
7201         if (status)
7202                 return status;
7203         status = nfsd4_lock_ol_stateid(stp);
7204         if (status != nfs_ok)
7205                 return status;
7206         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7207         if (status == nfs_ok)
7208                 status = nfs4_check_fh(current_fh, &stp->st_stid);
7209         if (status != nfs_ok)
7210                 mutex_unlock(&stp->st_mutex);
7211         return status;
7212 }
7213
7214 /**
7215  * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7216  * @cstate: compund state
7217  * @seqid: seqid (provided by client)
7218  * @stateid: stateid (provided by client)
7219  * @typemask: mask of allowable types for this operation
7220  * @statusmask: mask of allowed states: 0 or STID_CLOSED
7221  * @stpp: return pointer for the stateid found
7222  * @nn: net namespace for request
7223  *
7224  * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7225  * return it in @stpp. On a nfs_ok return, the returned stateid will
7226  * have its st_mutex locked.
7227  */
7228 static __be32
7229 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7230                          stateid_t *stateid,
7231                          unsigned short typemask, unsigned short statusmask,
7232                          struct nfs4_ol_stateid **stpp,
7233                          struct nfsd_net *nn)
7234 {
7235         __be32 status;
7236         struct nfs4_stid *s;
7237         struct nfs4_ol_stateid *stp = NULL;
7238
7239         trace_nfsd_preprocess(seqid, stateid);
7240
7241         *stpp = NULL;
7242 retry:
7243         status = nfsd4_lookup_stateid(cstate, stateid,
7244                                       typemask, statusmask, &s, nn);
7245         if (status)
7246                 return status;
7247         stp = openlockstateid(s);
7248         if (nfsd4_cstate_assign_replay(cstate, stp->st_stateowner) == -EAGAIN) {
7249                 nfs4_put_stateowner(stp->st_stateowner);
7250                 goto retry;
7251         }
7252
7253         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7254         if (!status)
7255                 *stpp = stp;
7256         else
7257                 nfs4_put_stid(&stp->st_stid);
7258         return status;
7259 }
7260
7261 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7262                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7263 {
7264         __be32 status;
7265         struct nfs4_openowner *oo;
7266         struct nfs4_ol_stateid *stp;
7267
7268         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7269                                           SC_TYPE_OPEN, 0, &stp, nn);
7270         if (status)
7271                 return status;
7272         oo = openowner(stp->st_stateowner);
7273         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7274                 mutex_unlock(&stp->st_mutex);
7275                 nfs4_put_stid(&stp->st_stid);
7276                 return nfserr_bad_stateid;
7277         }
7278         *stpp = stp;
7279         return nfs_ok;
7280 }
7281
7282 __be32
7283 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7284                    union nfsd4_op_u *u)
7285 {
7286         struct nfsd4_open_confirm *oc = &u->open_confirm;
7287         __be32 status;
7288         struct nfs4_openowner *oo;
7289         struct nfs4_ol_stateid *stp;
7290         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7291
7292         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7293                         cstate->current_fh.fh_dentry);
7294
7295         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7296         if (status)
7297                 return status;
7298
7299         status = nfs4_preprocess_seqid_op(cstate,
7300                                           oc->oc_seqid, &oc->oc_req_stateid,
7301                                           SC_TYPE_OPEN, 0, &stp, nn);
7302         if (status)
7303                 goto out;
7304         oo = openowner(stp->st_stateowner);
7305         status = nfserr_bad_stateid;
7306         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7307                 mutex_unlock(&stp->st_mutex);
7308                 goto put_stateid;
7309         }
7310         oo->oo_flags |= NFS4_OO_CONFIRMED;
7311         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7312         mutex_unlock(&stp->st_mutex);
7313         trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7314         nfsd4_client_record_create(oo->oo_owner.so_client);
7315         status = nfs_ok;
7316 put_stateid:
7317         nfs4_put_stid(&stp->st_stid);
7318 out:
7319         nfsd4_bump_seqid(cstate, status);
7320         return status;
7321 }
7322
7323 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7324 {
7325         if (!test_access(access, stp))
7326                 return;
7327         nfs4_file_put_access(stp->st_stid.sc_file, access);
7328         clear_access(access, stp);
7329 }
7330
7331 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7332 {
7333         switch (to_access) {
7334         case NFS4_SHARE_ACCESS_READ:
7335                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7336                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7337                 break;
7338         case NFS4_SHARE_ACCESS_WRITE:
7339                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7340                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7341                 break;
7342         case NFS4_SHARE_ACCESS_BOTH:
7343                 break;
7344         default:
7345                 WARN_ON_ONCE(1);
7346         }
7347 }
7348
7349 __be32
7350 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7351                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7352 {
7353         struct nfsd4_open_downgrade *od = &u->open_downgrade;
7354         __be32 status;
7355         struct nfs4_ol_stateid *stp;
7356         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7357
7358         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7359                         cstate->current_fh.fh_dentry);
7360
7361         /* We don't yet support WANT bits: */
7362         if (od->od_deleg_want)
7363                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7364                         od->od_deleg_want);
7365
7366         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7367                                         &od->od_stateid, &stp, nn);
7368         if (status)
7369                 goto out;
7370         status = nfserr_inval;
7371         if (!test_access(od->od_share_access, stp)) {
7372                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7373                         stp->st_access_bmap, od->od_share_access);
7374                 goto put_stateid;
7375         }
7376         if (!test_deny(od->od_share_deny, stp)) {
7377                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7378                         stp->st_deny_bmap, od->od_share_deny);
7379                 goto put_stateid;
7380         }
7381         nfs4_stateid_downgrade(stp, od->od_share_access);
7382         reset_union_bmap_deny(od->od_share_deny, stp);
7383         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7384         status = nfs_ok;
7385 put_stateid:
7386         mutex_unlock(&stp->st_mutex);
7387         nfs4_put_stid(&stp->st_stid);
7388 out:
7389         nfsd4_bump_seqid(cstate, status);
7390         return status;
7391 }
7392
7393 static bool nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7394 {
7395         struct nfs4_client *clp = s->st_stid.sc_client;
7396         bool unhashed;
7397         LIST_HEAD(reaplist);
7398         struct nfs4_ol_stateid *stp;
7399
7400         spin_lock(&clp->cl_lock);
7401         unhashed = unhash_open_stateid(s, &reaplist);
7402
7403         if (clp->cl_minorversion) {
7404                 if (unhashed)
7405                         put_ol_stateid_locked(s, &reaplist);
7406                 spin_unlock(&clp->cl_lock);
7407                 list_for_each_entry(stp, &reaplist, st_locks)
7408                         nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7409                 free_ol_stateid_reaplist(&reaplist);
7410                 return false;
7411         } else {
7412                 spin_unlock(&clp->cl_lock);
7413                 free_ol_stateid_reaplist(&reaplist);
7414                 return unhashed;
7415         }
7416 }
7417
7418 /*
7419  * nfs4_unlock_state() called after encode
7420  */
7421 __be32
7422 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7423                 union nfsd4_op_u *u)
7424 {
7425         struct nfsd4_close *close = &u->close;
7426         __be32 status;
7427         struct nfs4_ol_stateid *stp;
7428         struct net *net = SVC_NET(rqstp);
7429         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7430         bool need_move_to_close_list;
7431
7432         dprintk("NFSD: nfsd4_close on file %pd\n",
7433                         cstate->current_fh.fh_dentry);
7434
7435         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7436                                           &close->cl_stateid,
7437                                           SC_TYPE_OPEN, SC_STATUS_CLOSED,
7438                                           &stp, nn);
7439         nfsd4_bump_seqid(cstate, status);
7440         if (status)
7441                 goto out;
7442
7443         spin_lock(&stp->st_stid.sc_client->cl_lock);
7444         stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7445         spin_unlock(&stp->st_stid.sc_client->cl_lock);
7446
7447         /*
7448          * Technically we don't _really_ have to increment or copy it, since
7449          * it should just be gone after this operation and we clobber the
7450          * copied value below, but we continue to do so here just to ensure
7451          * that racing ops see that there was a state change.
7452          */
7453         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7454
7455         need_move_to_close_list = nfsd4_close_open_stateid(stp);
7456         mutex_unlock(&stp->st_mutex);
7457         if (need_move_to_close_list)
7458                 move_to_close_lru(stp, net);
7459
7460         /* v4.1+ suggests that we send a special stateid in here, since the
7461          * clients should just ignore this anyway. Since this is not useful
7462          * for v4.0 clients either, we set it to the special close_stateid
7463          * universally.
7464          *
7465          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7466          */
7467         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7468
7469         /* put reference from nfs4_preprocess_seqid_op */
7470         nfs4_put_stid(&stp->st_stid);
7471 out:
7472         return status;
7473 }
7474
7475 __be32
7476 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7477                   union nfsd4_op_u *u)
7478 {
7479         struct nfsd4_delegreturn *dr = &u->delegreturn;
7480         struct nfs4_delegation *dp;
7481         stateid_t *stateid = &dr->dr_stateid;
7482         struct nfs4_stid *s;
7483         __be32 status;
7484         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7485
7486         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7487                 return status;
7488
7489         status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7490         if (status)
7491                 goto out;
7492         dp = delegstateid(s);
7493         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7494         if (status)
7495                 goto put_stateid;
7496
7497         trace_nfsd_deleg_return(stateid);
7498         destroy_delegation(dp);
7499         smp_mb__after_atomic();
7500         wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7501 put_stateid:
7502         nfs4_put_stid(&dp->dl_stid);
7503 out:
7504         return status;
7505 }
7506
7507 /* last octet in a range */
7508 static inline u64
7509 last_byte_offset(u64 start, u64 len)
7510 {
7511         u64 end;
7512
7513         WARN_ON_ONCE(!len);
7514         end = start + len;
7515         return end > start ? end - 1: NFS4_MAX_UINT64;
7516 }
7517
7518 /*
7519  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7520  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7521  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
7522  * locking, this prevents us from being completely protocol-compliant.  The
7523  * real solution to this problem is to start using unsigned file offsets in
7524  * the VFS, but this is a very deep change!
7525  */
7526 static inline void
7527 nfs4_transform_lock_offset(struct file_lock *lock)
7528 {
7529         if (lock->fl_start < 0)
7530                 lock->fl_start = OFFSET_MAX;
7531         if (lock->fl_end < 0)
7532                 lock->fl_end = OFFSET_MAX;
7533 }
7534
7535 static fl_owner_t
7536 nfsd4_lm_get_owner(fl_owner_t owner)
7537 {
7538         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7539
7540         nfs4_get_stateowner(&lo->lo_owner);
7541         return owner;
7542 }
7543
7544 static void
7545 nfsd4_lm_put_owner(fl_owner_t owner)
7546 {
7547         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7548
7549         if (lo)
7550                 nfs4_put_stateowner(&lo->lo_owner);
7551 }
7552
7553 /* return pointer to struct nfs4_client if client is expirable */
7554 static bool
7555 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7556 {
7557         struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7558         struct nfs4_client *clp = lo->lo_owner.so_client;
7559         struct nfsd_net *nn;
7560
7561         if (try_to_expire_client(clp)) {
7562                 nn = net_generic(clp->net, nfsd_net_id);
7563                 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7564                 return true;
7565         }
7566         return false;
7567 }
7568
7569 /* schedule laundromat to run immediately and wait for it to complete */
7570 static void
7571 nfsd4_lm_expire_lock(void)
7572 {
7573         flush_workqueue(laundry_wq);
7574 }
7575
7576 static void
7577 nfsd4_lm_notify(struct file_lock *fl)
7578 {
7579         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7580         struct net                      *net = lo->lo_owner.so_client->net;
7581         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
7582         struct nfsd4_blocked_lock       *nbl = container_of(fl,
7583                                                 struct nfsd4_blocked_lock, nbl_lock);
7584         bool queue = false;
7585
7586         /* An empty list means that something else is going to be using it */
7587         spin_lock(&nn->blocked_locks_lock);
7588         if (!list_empty(&nbl->nbl_list)) {
7589                 list_del_init(&nbl->nbl_list);
7590                 list_del_init(&nbl->nbl_lru);
7591                 queue = true;
7592         }
7593         spin_unlock(&nn->blocked_locks_lock);
7594
7595         if (queue) {
7596                 trace_nfsd_cb_notify_lock(lo, nbl);
7597                 nfsd4_run_cb(&nbl->nbl_cb);
7598         }
7599 }
7600
7601 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
7602         .lm_mod_owner = THIS_MODULE,
7603         .lm_notify = nfsd4_lm_notify,
7604         .lm_get_owner = nfsd4_lm_get_owner,
7605         .lm_put_owner = nfsd4_lm_put_owner,
7606         .lm_lock_expirable = nfsd4_lm_lock_expirable,
7607         .lm_expire_lock = nfsd4_lm_expire_lock,
7608 };
7609
7610 static inline void
7611 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7612 {
7613         struct nfs4_lockowner *lo;
7614
7615         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7616                 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7617                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7618                                                 GFP_KERNEL);
7619                 if (!deny->ld_owner.data)
7620                         /* We just don't care that much */
7621                         goto nevermind;
7622                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7623         } else {
7624 nevermind:
7625                 deny->ld_owner.len = 0;
7626                 deny->ld_owner.data = NULL;
7627                 deny->ld_clientid.cl_boot = 0;
7628                 deny->ld_clientid.cl_id = 0;
7629         }
7630         deny->ld_start = fl->fl_start;
7631         deny->ld_length = NFS4_MAX_UINT64;
7632         if (fl->fl_end != NFS4_MAX_UINT64)
7633                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7634         deny->ld_type = NFS4_READ_LT;
7635         if (fl->c.flc_type != F_RDLCK)
7636                 deny->ld_type = NFS4_WRITE_LT;
7637 }
7638
7639 static struct nfs4_lockowner *
7640 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7641 {
7642         unsigned int strhashval = ownerstr_hashval(owner);
7643         struct nfs4_stateowner *so;
7644
7645         lockdep_assert_held(&clp->cl_lock);
7646
7647         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7648                             so_strhash) {
7649                 if (so->so_is_open_owner)
7650                         continue;
7651                 if (same_owner_str(so, owner))
7652                         return lockowner(nfs4_get_stateowner(so));
7653         }
7654         return NULL;
7655 }
7656
7657 static struct nfs4_lockowner *
7658 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7659 {
7660         struct nfs4_lockowner *lo;
7661
7662         spin_lock(&clp->cl_lock);
7663         lo = find_lockowner_str_locked(clp, owner);
7664         spin_unlock(&clp->cl_lock);
7665         return lo;
7666 }
7667
7668 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7669 {
7670         unhash_lockowner_locked(lockowner(sop));
7671 }
7672
7673 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7674 {
7675         struct nfs4_lockowner *lo = lockowner(sop);
7676
7677         kmem_cache_free(lockowner_slab, lo);
7678 }
7679
7680 static const struct nfs4_stateowner_operations lockowner_ops = {
7681         .so_unhash =    nfs4_unhash_lockowner,
7682         .so_free =      nfs4_free_lockowner,
7683 };
7684
7685 /*
7686  * Alloc a lock owner structure.
7687  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7688  * occurred.
7689  *
7690  * strhashval = ownerstr_hashval
7691  */
7692 static struct nfs4_lockowner *
7693 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7694                            struct nfs4_ol_stateid *open_stp,
7695                            struct nfsd4_lock *lock)
7696 {
7697         struct nfs4_lockowner *lo, *ret;
7698
7699         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7700         if (!lo)
7701                 return NULL;
7702         INIT_LIST_HEAD(&lo->lo_blocked);
7703         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7704         lo->lo_owner.so_is_open_owner = 0;
7705         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7706         lo->lo_owner.so_ops = &lockowner_ops;
7707         spin_lock(&clp->cl_lock);
7708         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7709         if (ret == NULL) {
7710                 list_add(&lo->lo_owner.so_strhash,
7711                          &clp->cl_ownerstr_hashtbl[strhashval]);
7712                 ret = lo;
7713         } else
7714                 nfs4_free_stateowner(&lo->lo_owner);
7715
7716         spin_unlock(&clp->cl_lock);
7717         return ret;
7718 }
7719
7720 static struct nfs4_ol_stateid *
7721 find_lock_stateid(const struct nfs4_lockowner *lo,
7722                   const struct nfs4_ol_stateid *ost)
7723 {
7724         struct nfs4_ol_stateid *lst;
7725
7726         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7727
7728         /* If ost is not hashed, ost->st_locks will not be valid */
7729         if (!nfs4_ol_stateid_unhashed(ost))
7730                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7731                         if (lst->st_stateowner == &lo->lo_owner) {
7732                                 refcount_inc(&lst->st_stid.sc_count);
7733                                 return lst;
7734                         }
7735                 }
7736         return NULL;
7737 }
7738
7739 static struct nfs4_ol_stateid *
7740 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7741                   struct nfs4_file *fp, struct inode *inode,
7742                   struct nfs4_ol_stateid *open_stp)
7743 {
7744         struct nfs4_client *clp = lo->lo_owner.so_client;
7745         struct nfs4_ol_stateid *retstp;
7746
7747         mutex_init(&stp->st_mutex);
7748         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7749 retry:
7750         spin_lock(&clp->cl_lock);
7751         if (nfs4_ol_stateid_unhashed(open_stp))
7752                 goto out_close;
7753         retstp = find_lock_stateid(lo, open_stp);
7754         if (retstp)
7755                 goto out_found;
7756         refcount_inc(&stp->st_stid.sc_count);
7757         stp->st_stid.sc_type = SC_TYPE_LOCK;
7758         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7759         get_nfs4_file(fp);
7760         stp->st_stid.sc_file = fp;
7761         stp->st_access_bmap = 0;
7762         stp->st_deny_bmap = open_stp->st_deny_bmap;
7763         stp->st_openstp = open_stp;
7764         spin_lock(&fp->fi_lock);
7765         list_add(&stp->st_locks, &open_stp->st_locks);
7766         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7767         list_add(&stp->st_perfile, &fp->fi_stateids);
7768         spin_unlock(&fp->fi_lock);
7769         spin_unlock(&clp->cl_lock);
7770         return stp;
7771 out_found:
7772         spin_unlock(&clp->cl_lock);
7773         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7774                 nfs4_put_stid(&retstp->st_stid);
7775                 goto retry;
7776         }
7777         /* To keep mutex tracking happy */
7778         mutex_unlock(&stp->st_mutex);
7779         return retstp;
7780 out_close:
7781         spin_unlock(&clp->cl_lock);
7782         mutex_unlock(&stp->st_mutex);
7783         return NULL;
7784 }
7785
7786 static struct nfs4_ol_stateid *
7787 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7788                             struct inode *inode, struct nfs4_ol_stateid *ost,
7789                             bool *new)
7790 {
7791         struct nfs4_stid *ns = NULL;
7792         struct nfs4_ol_stateid *lst;
7793         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7794         struct nfs4_client *clp = oo->oo_owner.so_client;
7795
7796         *new = false;
7797         spin_lock(&clp->cl_lock);
7798         lst = find_lock_stateid(lo, ost);
7799         spin_unlock(&clp->cl_lock);
7800         if (lst != NULL) {
7801                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7802                         goto out;
7803                 nfs4_put_stid(&lst->st_stid);
7804         }
7805         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7806         if (ns == NULL)
7807                 return NULL;
7808
7809         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7810         if (lst == openlockstateid(ns))
7811                 *new = true;
7812         else
7813                 nfs4_put_stid(ns);
7814 out:
7815         return lst;
7816 }
7817
7818 static int
7819 check_lock_length(u64 offset, u64 length)
7820 {
7821         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7822                 (length > ~offset)));
7823 }
7824
7825 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7826 {
7827         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7828
7829         lockdep_assert_held(&fp->fi_lock);
7830
7831         if (test_access(access, lock_stp))
7832                 return;
7833         __nfs4_file_get_access(fp, access);
7834         set_access(access, lock_stp);
7835 }
7836
7837 static __be32
7838 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7839                             struct nfs4_ol_stateid *ost,
7840                             struct nfsd4_lock *lock,
7841                             struct nfs4_ol_stateid **plst, bool *new)
7842 {
7843         __be32 status;
7844         struct nfs4_file *fi = ost->st_stid.sc_file;
7845         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7846         struct nfs4_client *cl = oo->oo_owner.so_client;
7847         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7848         struct nfs4_lockowner *lo;
7849         struct nfs4_ol_stateid *lst;
7850         unsigned int strhashval;
7851
7852         lo = find_lockowner_str(cl, &lock->lk_new_owner);
7853         if (!lo) {
7854                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7855                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7856                 if (lo == NULL)
7857                         return nfserr_jukebox;
7858         } else {
7859                 /* with an existing lockowner, seqids must be the same */
7860                 status = nfserr_bad_seqid;
7861                 if (!cstate->minorversion &&
7862                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7863                         goto out;
7864         }
7865
7866         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7867         if (lst == NULL) {
7868                 status = nfserr_jukebox;
7869                 goto out;
7870         }
7871
7872         status = nfs_ok;
7873         *plst = lst;
7874 out:
7875         nfs4_put_stateowner(&lo->lo_owner);
7876         return status;
7877 }
7878
7879 /*
7880  *  LOCK operation
7881  */
7882 __be32
7883 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7884            union nfsd4_op_u *u)
7885 {
7886         struct nfsd4_lock *lock = &u->lock;
7887         struct nfs4_openowner *open_sop = NULL;
7888         struct nfs4_lockowner *lock_sop = NULL;
7889         struct nfs4_ol_stateid *lock_stp = NULL;
7890         struct nfs4_ol_stateid *open_stp = NULL;
7891         struct nfs4_file *fp;
7892         struct nfsd_file *nf = NULL;
7893         struct nfsd4_blocked_lock *nbl = NULL;
7894         struct file_lock *file_lock = NULL;
7895         struct file_lock *conflock = NULL;
7896         struct super_block *sb;
7897         __be32 status = 0;
7898         int lkflg;
7899         int err;
7900         bool new = false;
7901         unsigned char type;
7902         unsigned int flags = FL_POSIX;
7903         struct net *net = SVC_NET(rqstp);
7904         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7905
7906         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7907                 (long long) lock->lk_offset,
7908                 (long long) lock->lk_length);
7909
7910         if (check_lock_length(lock->lk_offset, lock->lk_length))
7911                  return nfserr_inval;
7912
7913         if ((status = fh_verify(rqstp, &cstate->current_fh,
7914                                 S_IFREG, NFSD_MAY_LOCK))) {
7915                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7916                 return status;
7917         }
7918         sb = cstate->current_fh.fh_dentry->d_sb;
7919
7920         if (lock->lk_is_new) {
7921                 if (nfsd4_has_session(cstate))
7922                         /* See rfc 5661 18.10.3: given clientid is ignored: */
7923                         memcpy(&lock->lk_new_clientid,
7924                                 &cstate->clp->cl_clientid,
7925                                 sizeof(clientid_t));
7926
7927                 /* validate and update open stateid and open seqid */
7928                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7929                                         lock->lk_new_open_seqid,
7930                                         &lock->lk_new_open_stateid,
7931                                         &open_stp, nn);
7932                 if (status)
7933                         goto out;
7934                 mutex_unlock(&open_stp->st_mutex);
7935                 open_sop = openowner(open_stp->st_stateowner);
7936                 status = nfserr_bad_stateid;
7937                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7938                                                 &lock->lk_new_clientid))
7939                         goto out;
7940                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7941                                                         &lock_stp, &new);
7942         } else {
7943                 status = nfs4_preprocess_seqid_op(cstate,
7944                                                   lock->lk_old_lock_seqid,
7945                                                   &lock->lk_old_lock_stateid,
7946                                                   SC_TYPE_LOCK, 0, &lock_stp,
7947                                                   nn);
7948         }
7949         if (status)
7950                 goto out;
7951         lock_sop = lockowner(lock_stp->st_stateowner);
7952
7953         lkflg = setlkflg(lock->lk_type);
7954         status = nfs4_check_openmode(lock_stp, lkflg);
7955         if (status)
7956                 goto out;
7957
7958         status = nfserr_grace;
7959         if (locks_in_grace(net) && !lock->lk_reclaim)
7960                 goto out;
7961         status = nfserr_no_grace;
7962         if (!locks_in_grace(net) && lock->lk_reclaim)
7963                 goto out;
7964
7965         if (lock->lk_reclaim)
7966                 flags |= FL_RECLAIM;
7967
7968         fp = lock_stp->st_stid.sc_file;
7969         switch (lock->lk_type) {
7970                 case NFS4_READW_LT:
7971                         if (nfsd4_has_session(cstate) ||
7972                             exportfs_lock_op_is_async(sb->s_export_op))
7973                                 flags |= FL_SLEEP;
7974                         fallthrough;
7975                 case NFS4_READ_LT:
7976                         spin_lock(&fp->fi_lock);
7977                         nf = find_readable_file_locked(fp);
7978                         if (nf)
7979                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7980                         spin_unlock(&fp->fi_lock);
7981                         type = F_RDLCK;
7982                         break;
7983                 case NFS4_WRITEW_LT:
7984                         if (nfsd4_has_session(cstate) ||
7985                             exportfs_lock_op_is_async(sb->s_export_op))
7986                                 flags |= FL_SLEEP;
7987                         fallthrough;
7988                 case NFS4_WRITE_LT:
7989                         spin_lock(&fp->fi_lock);
7990                         nf = find_writeable_file_locked(fp);
7991                         if (nf)
7992                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7993                         spin_unlock(&fp->fi_lock);
7994                         type = F_WRLCK;
7995                         break;
7996                 default:
7997                         status = nfserr_inval;
7998                 goto out;
7999         }
8000
8001         if (!nf) {
8002                 status = nfserr_openmode;
8003                 goto out;
8004         }
8005
8006         /*
8007          * Most filesystems with their own ->lock operations will block
8008          * the nfsd thread waiting to acquire the lock.  That leads to
8009          * deadlocks (we don't want every nfsd thread tied up waiting
8010          * for file locks), so don't attempt blocking lock notifications
8011          * on those filesystems:
8012          */
8013         if (!exportfs_lock_op_is_async(sb->s_export_op))
8014                 flags &= ~FL_SLEEP;
8015
8016         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
8017         if (!nbl) {
8018                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
8019                 status = nfserr_jukebox;
8020                 goto out;
8021         }
8022
8023         file_lock = &nbl->nbl_lock;
8024         file_lock->c.flc_type = type;
8025         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
8026         file_lock->c.flc_pid = current->tgid;
8027         file_lock->c.flc_file = nf->nf_file;
8028         file_lock->c.flc_flags = flags;
8029         file_lock->fl_lmops = &nfsd_posix_mng_ops;
8030         file_lock->fl_start = lock->lk_offset;
8031         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
8032         nfs4_transform_lock_offset(file_lock);
8033
8034         conflock = locks_alloc_lock();
8035         if (!conflock) {
8036                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8037                 status = nfserr_jukebox;
8038                 goto out;
8039         }
8040
8041         if (flags & FL_SLEEP) {
8042                 nbl->nbl_time = ktime_get_boottime_seconds();
8043                 spin_lock(&nn->blocked_locks_lock);
8044                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
8045                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
8046                 kref_get(&nbl->nbl_kref);
8047                 spin_unlock(&nn->blocked_locks_lock);
8048         }
8049
8050         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
8051         switch (err) {
8052         case 0: /* success! */
8053                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
8054                 status = 0;
8055                 if (lock->lk_reclaim)
8056                         nn->somebody_reclaimed = true;
8057                 break;
8058         case FILE_LOCK_DEFERRED:
8059                 kref_put(&nbl->nbl_kref, free_nbl);
8060                 nbl = NULL;
8061                 fallthrough;
8062         case -EAGAIN:           /* conflock holds conflicting lock */
8063                 status = nfserr_denied;
8064                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8065                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8066                 break;
8067         case -EDEADLK:
8068                 status = nfserr_deadlock;
8069                 break;
8070         default:
8071                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8072                 status = nfserrno(err);
8073                 break;
8074         }
8075 out:
8076         if (nbl) {
8077                 /* dequeue it if we queued it before */
8078                 if (flags & FL_SLEEP) {
8079                         spin_lock(&nn->blocked_locks_lock);
8080                         if (!list_empty(&nbl->nbl_list) &&
8081                             !list_empty(&nbl->nbl_lru)) {
8082                                 list_del_init(&nbl->nbl_list);
8083                                 list_del_init(&nbl->nbl_lru);
8084                                 kref_put(&nbl->nbl_kref, free_nbl);
8085                         }
8086                         /* nbl can use one of lists to be linked to reaplist */
8087                         spin_unlock(&nn->blocked_locks_lock);
8088                 }
8089                 free_blocked_lock(nbl);
8090         }
8091         if (nf)
8092                 nfsd_file_put(nf);
8093         if (lock_stp) {
8094                 /* Bump seqid manually if the 4.0 replay owner is openowner */
8095                 if (cstate->replay_owner &&
8096                     cstate->replay_owner != &lock_sop->lo_owner &&
8097                     seqid_mutating_err(ntohl(status)))
8098                         lock_sop->lo_owner.so_seqid++;
8099
8100                 /*
8101                  * If this is a new, never-before-used stateid, and we are
8102                  * returning an error, then just go ahead and release it.
8103                  */
8104                 if (status && new)
8105                         release_lock_stateid(lock_stp);
8106
8107                 mutex_unlock(&lock_stp->st_mutex);
8108
8109                 nfs4_put_stid(&lock_stp->st_stid);
8110         }
8111         if (open_stp)
8112                 nfs4_put_stid(&open_stp->st_stid);
8113         nfsd4_bump_seqid(cstate, status);
8114         if (conflock)
8115                 locks_free_lock(conflock);
8116         return status;
8117 }
8118
8119 void nfsd4_lock_release(union nfsd4_op_u *u)
8120 {
8121         struct nfsd4_lock *lock = &u->lock;
8122         struct nfsd4_lock_denied *deny = &lock->lk_denied;
8123
8124         kfree(deny->ld_owner.data);
8125 }
8126
8127 /*
8128  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8129  * so we do a temporary open here just to get an open file to pass to
8130  * vfs_test_lock.
8131  */
8132 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8133 {
8134         struct nfsd_file *nf;
8135         struct inode *inode;
8136         __be32 err;
8137
8138         err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8139         if (err)
8140                 return err;
8141         inode = fhp->fh_dentry->d_inode;
8142         inode_lock(inode); /* to block new leases till after test_lock: */
8143         err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8144         if (err)
8145                 goto out;
8146         lock->c.flc_file = nf->nf_file;
8147         err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8148         lock->c.flc_file = NULL;
8149 out:
8150         inode_unlock(inode);
8151         nfsd_file_put(nf);
8152         return err;
8153 }
8154
8155 /*
8156  * LOCKT operation
8157  */
8158 __be32
8159 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8160             union nfsd4_op_u *u)
8161 {
8162         struct nfsd4_lockt *lockt = &u->lockt;
8163         struct file_lock *file_lock = NULL;
8164         struct nfs4_lockowner *lo = NULL;
8165         __be32 status;
8166         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8167
8168         if (locks_in_grace(SVC_NET(rqstp)))
8169                 return nfserr_grace;
8170
8171         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8172                  return nfserr_inval;
8173
8174         if (!nfsd4_has_session(cstate)) {
8175                 status = set_client(&lockt->lt_clientid, cstate, nn);
8176                 if (status)
8177                         goto out;
8178         }
8179
8180         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8181                 goto out;
8182
8183         file_lock = locks_alloc_lock();
8184         if (!file_lock) {
8185                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8186                 status = nfserr_jukebox;
8187                 goto out;
8188         }
8189
8190         switch (lockt->lt_type) {
8191                 case NFS4_READ_LT:
8192                 case NFS4_READW_LT:
8193                         file_lock->c.flc_type = F_RDLCK;
8194                         break;
8195                 case NFS4_WRITE_LT:
8196                 case NFS4_WRITEW_LT:
8197                         file_lock->c.flc_type = F_WRLCK;
8198                         break;
8199                 default:
8200                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8201                         status = nfserr_inval;
8202                         goto out;
8203         }
8204
8205         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8206         if (lo)
8207                 file_lock->c.flc_owner = (fl_owner_t)lo;
8208         file_lock->c.flc_pid = current->tgid;
8209         file_lock->c.flc_flags = FL_POSIX;
8210
8211         file_lock->fl_start = lockt->lt_offset;
8212         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8213
8214         nfs4_transform_lock_offset(file_lock);
8215
8216         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8217         if (status)
8218                 goto out;
8219
8220         if (file_lock->c.flc_type != F_UNLCK) {
8221                 status = nfserr_denied;
8222                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8223         }
8224 out:
8225         if (lo)
8226                 nfs4_put_stateowner(&lo->lo_owner);
8227         if (file_lock)
8228                 locks_free_lock(file_lock);
8229         return status;
8230 }
8231
8232 void nfsd4_lockt_release(union nfsd4_op_u *u)
8233 {
8234         struct nfsd4_lockt *lockt = &u->lockt;
8235         struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8236
8237         kfree(deny->ld_owner.data);
8238 }
8239
8240 __be32
8241 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8242             union nfsd4_op_u *u)
8243 {
8244         struct nfsd4_locku *locku = &u->locku;
8245         struct nfs4_ol_stateid *stp;
8246         struct nfsd_file *nf = NULL;
8247         struct file_lock *file_lock = NULL;
8248         __be32 status;
8249         int err;
8250         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8251
8252         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8253                 (long long) locku->lu_offset,
8254                 (long long) locku->lu_length);
8255
8256         if (check_lock_length(locku->lu_offset, locku->lu_length))
8257                  return nfserr_inval;
8258
8259         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8260                                           &locku->lu_stateid, SC_TYPE_LOCK, 0,
8261                                           &stp, nn);
8262         if (status)
8263                 goto out;
8264         nf = find_any_file(stp->st_stid.sc_file);
8265         if (!nf) {
8266                 status = nfserr_lock_range;
8267                 goto put_stateid;
8268         }
8269         file_lock = locks_alloc_lock();
8270         if (!file_lock) {
8271                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8272                 status = nfserr_jukebox;
8273                 goto put_file;
8274         }
8275
8276         file_lock->c.flc_type = F_UNLCK;
8277         file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8278         file_lock->c.flc_pid = current->tgid;
8279         file_lock->c.flc_file = nf->nf_file;
8280         file_lock->c.flc_flags = FL_POSIX;
8281         file_lock->fl_lmops = &nfsd_posix_mng_ops;
8282         file_lock->fl_start = locku->lu_offset;
8283
8284         file_lock->fl_end = last_byte_offset(locku->lu_offset,
8285                                                 locku->lu_length);
8286         nfs4_transform_lock_offset(file_lock);
8287
8288         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8289         if (err) {
8290                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8291                 goto out_nfserr;
8292         }
8293         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8294 put_file:
8295         nfsd_file_put(nf);
8296 put_stateid:
8297         mutex_unlock(&stp->st_mutex);
8298         nfs4_put_stid(&stp->st_stid);
8299 out:
8300         nfsd4_bump_seqid(cstate, status);
8301         if (file_lock)
8302                 locks_free_lock(file_lock);
8303         return status;
8304
8305 out_nfserr:
8306         status = nfserrno(err);
8307         goto put_file;
8308 }
8309
8310 /*
8311  * returns
8312  *      true:  locks held by lockowner
8313  *      false: no locks held by lockowner
8314  */
8315 static bool
8316 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8317 {
8318         struct file_lock *fl;
8319         int status = false;
8320         struct nfsd_file *nf;
8321         struct inode *inode;
8322         struct file_lock_context *flctx;
8323
8324         spin_lock(&fp->fi_lock);
8325         nf = find_any_file_locked(fp);
8326         if (!nf) {
8327                 /* Any valid lock stateid should have some sort of access */
8328                 WARN_ON_ONCE(1);
8329                 goto out;
8330         }
8331
8332         inode = file_inode(nf->nf_file);
8333         flctx = locks_inode_context(inode);
8334
8335         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8336                 spin_lock(&flctx->flc_lock);
8337                 for_each_file_lock(fl, &flctx->flc_posix) {
8338                         if (fl->c.flc_owner == (fl_owner_t)lowner) {
8339                                 status = true;
8340                                 break;
8341                         }
8342                 }
8343                 spin_unlock(&flctx->flc_lock);
8344         }
8345 out:
8346         spin_unlock(&fp->fi_lock);
8347         return status;
8348 }
8349
8350 /**
8351  * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8352  * @rqstp: RPC transaction
8353  * @cstate: NFSv4 COMPOUND state
8354  * @u: RELEASE_LOCKOWNER arguments
8355  *
8356  * Check if there are any locks still held and if not, free the lockowner
8357  * and any lock state that is owned.
8358  *
8359  * Return values:
8360  *   %nfs_ok: lockowner released or not found
8361  *   %nfserr_locks_held: lockowner still in use
8362  *   %nfserr_stale_clientid: clientid no longer active
8363  *   %nfserr_expired: clientid not recognized
8364  */
8365 __be32
8366 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8367                         struct nfsd4_compound_state *cstate,
8368                         union nfsd4_op_u *u)
8369 {
8370         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8371         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8372         clientid_t *clid = &rlockowner->rl_clientid;
8373         struct nfs4_ol_stateid *stp;
8374         struct nfs4_lockowner *lo;
8375         struct nfs4_client *clp;
8376         LIST_HEAD(reaplist);
8377         __be32 status;
8378
8379         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8380                 clid->cl_boot, clid->cl_id);
8381
8382         status = set_client(clid, cstate, nn);
8383         if (status)
8384                 return status;
8385         clp = cstate->clp;
8386
8387         spin_lock(&clp->cl_lock);
8388         lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8389         if (!lo) {
8390                 spin_unlock(&clp->cl_lock);
8391                 return nfs_ok;
8392         }
8393
8394         list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8395                 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8396                         spin_unlock(&clp->cl_lock);
8397                         nfs4_put_stateowner(&lo->lo_owner);
8398                         return nfserr_locks_held;
8399                 }
8400         }
8401         unhash_lockowner_locked(lo);
8402         while (!list_empty(&lo->lo_owner.so_stateids)) {
8403                 stp = list_first_entry(&lo->lo_owner.so_stateids,
8404                                        struct nfs4_ol_stateid,
8405                                        st_perstateowner);
8406                 unhash_lock_stateid(stp);
8407                 put_ol_stateid_locked(stp, &reaplist);
8408         }
8409         spin_unlock(&clp->cl_lock);
8410
8411         free_ol_stateid_reaplist(&reaplist);
8412         remove_blocked_locks(lo);
8413         nfs4_put_stateowner(&lo->lo_owner);
8414         return nfs_ok;
8415 }
8416
8417 static inline struct nfs4_client_reclaim *
8418 alloc_reclaim(void)
8419 {
8420         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8421 }
8422
8423 bool
8424 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8425 {
8426         struct nfs4_client_reclaim *crp;
8427
8428         crp = nfsd4_find_reclaim_client(name, nn);
8429         return (crp && crp->cr_clp);
8430 }
8431
8432 /*
8433  * failure => all reset bets are off, nfserr_no_grace...
8434  *
8435  * The caller is responsible for freeing name.data if NULL is returned (it
8436  * will be freed in nfs4_remove_reclaim_record in the normal case).
8437  */
8438 struct nfs4_client_reclaim *
8439 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8440                 struct nfsd_net *nn)
8441 {
8442         unsigned int strhashval;
8443         struct nfs4_client_reclaim *crp;
8444
8445         crp = alloc_reclaim();
8446         if (crp) {
8447                 strhashval = clientstr_hashval(name);
8448                 INIT_LIST_HEAD(&crp->cr_strhash);
8449                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8450                 crp->cr_name.data = name.data;
8451                 crp->cr_name.len = name.len;
8452                 crp->cr_princhash.data = princhash.data;
8453                 crp->cr_princhash.len = princhash.len;
8454                 crp->cr_clp = NULL;
8455                 nn->reclaim_str_hashtbl_size++;
8456         }
8457         return crp;
8458 }
8459
8460 void
8461 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8462 {
8463         list_del(&crp->cr_strhash);
8464         kfree(crp->cr_name.data);
8465         kfree(crp->cr_princhash.data);
8466         kfree(crp);
8467         nn->reclaim_str_hashtbl_size--;
8468 }
8469
8470 void
8471 nfs4_release_reclaim(struct nfsd_net *nn)
8472 {
8473         struct nfs4_client_reclaim *crp = NULL;
8474         int i;
8475
8476         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8477                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8478                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8479                                         struct nfs4_client_reclaim, cr_strhash);
8480                         nfs4_remove_reclaim_record(crp, nn);
8481                 }
8482         }
8483         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8484 }
8485
8486 /*
8487  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8488 struct nfs4_client_reclaim *
8489 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8490 {
8491         unsigned int strhashval;
8492         struct nfs4_client_reclaim *crp = NULL;
8493
8494         strhashval = clientstr_hashval(name);
8495         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8496                 if (compare_blob(&crp->cr_name, &name) == 0) {
8497                         return crp;
8498                 }
8499         }
8500         return NULL;
8501 }
8502
8503 __be32
8504 nfs4_check_open_reclaim(struct nfs4_client *clp)
8505 {
8506         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8507                 return nfserr_no_grace;
8508
8509         if (nfsd4_client_record_check(clp))
8510                 return nfserr_reclaim_bad;
8511
8512         return nfs_ok;
8513 }
8514
8515 /*
8516  * Since the lifetime of a delegation isn't limited to that of an open, a
8517  * client may quite reasonably hang on to a delegation as long as it has
8518  * the inode cached.  This becomes an obvious problem the first time a
8519  * client's inode cache approaches the size of the server's total memory.
8520  *
8521  * For now we avoid this problem by imposing a hard limit on the number
8522  * of delegations, which varies according to the server's memory size.
8523  */
8524 static void
8525 set_max_delegations(void)
8526 {
8527         /*
8528          * Allow at most 4 delegations per megabyte of RAM.  Quick
8529          * estimates suggest that in the worst case (where every delegation
8530          * is for a different inode), a delegation could take about 1.5K,
8531          * giving a worst case usage of about 6% of memory.
8532          */
8533         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8534 }
8535
8536 static int nfs4_state_create_net(struct net *net)
8537 {
8538         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8539         int i;
8540
8541         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8542                                             sizeof(struct list_head),
8543                                             GFP_KERNEL);
8544         if (!nn->conf_id_hashtbl)
8545                 goto err;
8546         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8547                                               sizeof(struct list_head),
8548                                               GFP_KERNEL);
8549         if (!nn->unconf_id_hashtbl)
8550                 goto err_unconf_id;
8551         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8552                                               sizeof(struct list_head),
8553                                               GFP_KERNEL);
8554         if (!nn->sessionid_hashtbl)
8555                 goto err_sessionid;
8556
8557         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8558                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8559                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8560         }
8561         for (i = 0; i < SESSION_HASH_SIZE; i++)
8562                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8563         nn->conf_name_tree = RB_ROOT;
8564         nn->unconf_name_tree = RB_ROOT;
8565         nn->boot_time = ktime_get_real_seconds();
8566         nn->grace_ended = false;
8567         nn->nfsd4_manager.block_opens = true;
8568         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8569         INIT_LIST_HEAD(&nn->client_lru);
8570         INIT_LIST_HEAD(&nn->close_lru);
8571         INIT_LIST_HEAD(&nn->del_recall_lru);
8572         spin_lock_init(&nn->client_lock);
8573         spin_lock_init(&nn->s2s_cp_lock);
8574         idr_init(&nn->s2s_cp_stateids);
8575         atomic_set(&nn->pending_async_copies, 0);
8576
8577         spin_lock_init(&nn->blocked_locks_lock);
8578         INIT_LIST_HEAD(&nn->blocked_locks_lru);
8579
8580         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8581         INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8582         get_net(net);
8583
8584         nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8585         if (!nn->nfsd_client_shrinker)
8586                 goto err_shrinker;
8587
8588         nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8589         nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8590         nn->nfsd_client_shrinker->private_data = nn;
8591
8592         shrinker_register(nn->nfsd_client_shrinker);
8593
8594         return 0;
8595
8596 err_shrinker:
8597         put_net(net);
8598         kfree(nn->sessionid_hashtbl);
8599 err_sessionid:
8600         kfree(nn->unconf_id_hashtbl);
8601 err_unconf_id:
8602         kfree(nn->conf_id_hashtbl);
8603 err:
8604         return -ENOMEM;
8605 }
8606
8607 static void
8608 nfs4_state_destroy_net(struct net *net)
8609 {
8610         int i;
8611         struct nfs4_client *clp = NULL;
8612         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8613
8614         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8615                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8616                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8617                         destroy_client(clp);
8618                 }
8619         }
8620
8621         WARN_ON(!list_empty(&nn->blocked_locks_lru));
8622
8623         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8624                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8625                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8626                         destroy_client(clp);
8627                 }
8628         }
8629
8630         kfree(nn->sessionid_hashtbl);
8631         kfree(nn->unconf_id_hashtbl);
8632         kfree(nn->conf_id_hashtbl);
8633         put_net(net);
8634 }
8635
8636 int
8637 nfs4_state_start_net(struct net *net)
8638 {
8639         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8640         int ret;
8641
8642         ret = nfs4_state_create_net(net);
8643         if (ret)
8644                 return ret;
8645         locks_start_grace(net, &nn->nfsd4_manager);
8646         nfsd4_client_tracking_init(net);
8647         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8648                 goto skip_grace;
8649         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8650                nn->nfsd4_grace, net->ns.inum);
8651         trace_nfsd_grace_start(nn);
8652         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8653         return 0;
8654
8655 skip_grace:
8656         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8657                         net->ns.inum);
8658         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8659         nfsd4_end_grace(nn);
8660         return 0;
8661 }
8662
8663 /* initialization to perform when the nfsd service is started: */
8664
8665 int
8666 nfs4_state_start(void)
8667 {
8668         int ret;
8669
8670         ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8671         if (ret)
8672                 return ret;
8673
8674         set_max_delegations();
8675         return 0;
8676 }
8677
8678 void
8679 nfs4_state_shutdown_net(struct net *net)
8680 {
8681         struct nfs4_delegation *dp = NULL;
8682         struct list_head *pos, *next, reaplist;
8683         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8684
8685         shrinker_free(nn->nfsd_client_shrinker);
8686         cancel_work(&nn->nfsd_shrinker_work);
8687         cancel_delayed_work_sync(&nn->laundromat_work);
8688         locks_end_grace(&nn->nfsd4_manager);
8689
8690         INIT_LIST_HEAD(&reaplist);
8691         spin_lock(&state_lock);
8692         list_for_each_safe(pos, next, &nn->del_recall_lru) {
8693                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8694                 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8695                 list_add(&dp->dl_recall_lru, &reaplist);
8696         }
8697         spin_unlock(&state_lock);
8698         list_for_each_safe(pos, next, &reaplist) {
8699                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8700                 list_del_init(&dp->dl_recall_lru);
8701                 destroy_unhashed_deleg(dp);
8702         }
8703
8704         nfsd4_client_tracking_exit(net);
8705         nfs4_state_destroy_net(net);
8706 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8707         nfsd4_ssc_shutdown_umount(nn);
8708 #endif
8709 }
8710
8711 void
8712 nfs4_state_shutdown(void)
8713 {
8714         rhltable_destroy(&nfs4_file_rhltable);
8715 }
8716
8717 static void
8718 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8719 {
8720         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8721             CURRENT_STATEID(stateid))
8722                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8723 }
8724
8725 static void
8726 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8727 {
8728         if (cstate->minorversion) {
8729                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8730                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8731         }
8732 }
8733
8734 void
8735 clear_current_stateid(struct nfsd4_compound_state *cstate)
8736 {
8737         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8738 }
8739
8740 /*
8741  * functions to set current state id
8742  */
8743 void
8744 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8745                 union nfsd4_op_u *u)
8746 {
8747         put_stateid(cstate, &u->open_downgrade.od_stateid);
8748 }
8749
8750 void
8751 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8752                 union nfsd4_op_u *u)
8753 {
8754         put_stateid(cstate, &u->open.op_stateid);
8755 }
8756
8757 void
8758 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8759                 union nfsd4_op_u *u)
8760 {
8761         put_stateid(cstate, &u->close.cl_stateid);
8762 }
8763
8764 void
8765 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8766                 union nfsd4_op_u *u)
8767 {
8768         put_stateid(cstate, &u->lock.lk_resp_stateid);
8769 }
8770
8771 /*
8772  * functions to consume current state id
8773  */
8774
8775 void
8776 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8777                 union nfsd4_op_u *u)
8778 {
8779         get_stateid(cstate, &u->open_downgrade.od_stateid);
8780 }
8781
8782 void
8783 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8784                 union nfsd4_op_u *u)
8785 {
8786         get_stateid(cstate, &u->delegreturn.dr_stateid);
8787 }
8788
8789 void
8790 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8791                 union nfsd4_op_u *u)
8792 {
8793         get_stateid(cstate, &u->free_stateid.fr_stateid);
8794 }
8795
8796 void
8797 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8798                 union nfsd4_op_u *u)
8799 {
8800         get_stateid(cstate, &u->setattr.sa_stateid);
8801 }
8802
8803 void
8804 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8805                 union nfsd4_op_u *u)
8806 {
8807         get_stateid(cstate, &u->close.cl_stateid);
8808 }
8809
8810 void
8811 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8812                 union nfsd4_op_u *u)
8813 {
8814         get_stateid(cstate, &u->locku.lu_stateid);
8815 }
8816
8817 void
8818 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8819                 union nfsd4_op_u *u)
8820 {
8821         get_stateid(cstate, &u->read.rd_stateid);
8822 }
8823
8824 void
8825 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8826                 union nfsd4_op_u *u)
8827 {
8828         get_stateid(cstate, &u->write.wr_stateid);
8829 }
8830
8831 /**
8832  * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8833  * @rqstp: RPC transaction context
8834  * @dentry: dentry of inode to be checked for a conflict
8835  * @modified: return true if file was modified
8836  * @size: new size of file if modified is true
8837  *
8838  * This function is called when there is a conflict between a write
8839  * delegation and a change/size GETATTR from another client. The server
8840  * must either use the CB_GETATTR to get the current values of the
8841  * attributes from the client that holds the delegation or recall the
8842  * delegation before replying to the GETATTR. See RFC 8881 section
8843  * 18.7.4.
8844  *
8845  * Returns 0 if there is no conflict; otherwise an nfs_stat
8846  * code is returned.
8847  */
8848 __be32
8849 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct dentry *dentry,
8850                                 bool *modified, u64 *size)
8851 {
8852         __be32 status;
8853         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8854         struct file_lock_context *ctx;
8855         struct nfs4_delegation *dp = NULL;
8856         struct file_lease *fl;
8857         struct iattr attrs;
8858         struct nfs4_cb_fattr *ncf;
8859         struct inode *inode = d_inode(dentry);
8860
8861         *modified = false;
8862         ctx = locks_inode_context(inode);
8863         if (!ctx)
8864                 return 0;
8865
8866 #define NON_NFSD_LEASE ((void *)1)
8867
8868         spin_lock(&ctx->flc_lock);
8869         for_each_file_lock(fl, &ctx->flc_lease) {
8870                 if (fl->c.flc_flags == FL_LAYOUT)
8871                         continue;
8872                 if (fl->c.flc_type == F_WRLCK) {
8873                         if (fl->fl_lmops == &nfsd_lease_mng_ops)
8874                                 dp = fl->c.flc_owner;
8875                         else
8876                                 dp = NON_NFSD_LEASE;
8877                 }
8878                 break;
8879         }
8880         if (dp == NULL || dp == NON_NFSD_LEASE ||
8881             dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8882                 spin_unlock(&ctx->flc_lock);
8883                 if (dp == NON_NFSD_LEASE) {
8884                         status = nfserrno(nfsd_open_break_lease(inode,
8885                                                                 NFSD_MAY_READ));
8886                         if (status != nfserr_jukebox ||
8887                             !nfsd_wait_for_delegreturn(rqstp, inode))
8888                                 return status;
8889                 }
8890                 return 0;
8891         }
8892
8893         nfsd_stats_wdeleg_getattr_inc(nn);
8894         refcount_inc(&dp->dl_stid.sc_count);
8895         ncf = &dp->dl_cb_fattr;
8896         nfs4_cb_getattr(&dp->dl_cb_fattr);
8897         spin_unlock(&ctx->flc_lock);
8898
8899         wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8900                             TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8901         if (ncf->ncf_cb_status) {
8902                 /* Recall delegation only if client didn't respond */
8903                 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8904                 if (status != nfserr_jukebox ||
8905                     !nfsd_wait_for_delegreturn(rqstp, inode))
8906                         goto out_status;
8907         }
8908         if (!ncf->ncf_file_modified &&
8909             (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8910              ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8911                 ncf->ncf_file_modified = true;
8912         if (ncf->ncf_file_modified) {
8913                 int err;
8914
8915                 /*
8916                  * Per section 10.4.3 of RFC 8881, the server would
8917                  * not update the file's metadata with the client's
8918                  * modified size
8919                  */
8920                 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8921                 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME | ATTR_DELEG;
8922                 inode_lock(inode);
8923                 err = notify_change(&nop_mnt_idmap, dentry, &attrs, NULL);
8924                 inode_unlock(inode);
8925                 if (err) {
8926                         status = nfserrno(err);
8927                         goto out_status;
8928                 }
8929                 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8930                 *size = ncf->ncf_cur_fsize;
8931                 *modified = true;
8932         }
8933         status = 0;
8934 out_status:
8935         nfs4_put_stid(&dp->dl_stid);
8936         return status;
8937 }