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 "xdr4.h"
  47 #include "xdr4cb.h"
  48 #include "vfs.h"
  49 #include "current_stateid.h"
  50
  51 #include "netns.h"
  52 #include "pnfs.h"
  53 #include "filecache.h"
  54
  55 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  56
  57 #define all_ones {{~0,~0},~0}
  58 static const stateid_t one_stateid = {
  59         .si_generation = ~0,
  60         .si_opaque = all_ones,
  61 };
  62 static const stateid_t zero_stateid = {
  63         /* all fields zero */
  64 };
  65 static const stateid_t currentstateid = {
  66         .si_generation = 1,
  67 };
  68 static const stateid_t close_stateid = {
  69         .si_generation = 0xffffffffU,
  70 };
  71
  72 static u64 current_sessionid = 1;
  73
  74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  75 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  76 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  77 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  78
  79 /* forward declarations */
  80 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  81 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  82 void nfsd4_end_grace(struct nfsd_net *nn);
  83 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
  84
  85 /* Locking: */
  86
  87 /*
  88  * Currently used for the del_recall_lru and file hash table.  In an
  89  * effort to decrease the scope of the client_mutex, this spinlock may
  90  * eventually cover more:
  91  */
  92 static DEFINE_SPINLOCK(state_lock);
  93
  94 enum nfsd4_st_mutex_lock_subclass {
  95         OPEN_STATEID_MUTEX = 0,
  96         LOCK_STATEID_MUTEX = 1,
  97 };
  98
  99 /*
 100  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 101  * the refcount on the open stateid to drop.
 102  */
 103 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 104
 105 /*
 106  * A waitqueue where a writer to clients/#/ctl destroying a client can
 107  * wait for cl_rpc_users to drop to 0 and then for the client to be
 108  * unhashed.
 109  */
 110 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 111
 112 static struct kmem_cache *client_slab;
 113 static struct kmem_cache *openowner_slab;
 114 static struct kmem_cache *lockowner_slab;
 115 static struct kmem_cache *file_slab;
 116 static struct kmem_cache *stateid_slab;
 117 static struct kmem_cache *deleg_slab;
 118 static struct kmem_cache *odstate_slab;
 119
 120 static void free_session(struct nfsd4_session *);
 121
 122 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 123 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 124
 125 static bool is_session_dead(struct nfsd4_session *ses)
 126 {
 127         return ses->se_flags & NFS4_SESSION_DEAD;
 128 }
 129
 130 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 131 {
 132         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 133                 return nfserr_jukebox;
 134         ses->se_flags |= NFS4_SESSION_DEAD;
 135         return nfs_ok;
 136 }
 137
 138 static bool is_client_expired(struct nfs4_client *clp)
 139 {
 140         return clp->cl_time == 0;
 141 }
 142
 143 static __be32 get_client_locked(struct nfs4_client *clp)
 144 {
 145         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 146
 147         lockdep_assert_held(&nn->client_lock);
 148
 149         if (is_client_expired(clp))
 150                 return nfserr_expired;
 151         atomic_inc(&clp->cl_rpc_users);
 152         return nfs_ok;
 153 }
 154
 155 /* must be called under the client_lock */
 156 static inline void
 157 renew_client_locked(struct nfs4_client *clp)
 158 {
 159         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 160
 161         if (is_client_expired(clp)) {
 162                 WARN_ON(1);
 163                 printk("%s: client (clientid %08x/%08x) already expired\n",
 164                         __func__,
 165                         clp->cl_clientid.cl_boot,
 166                         clp->cl_clientid.cl_id);
 167                 return;
 168         }
 169
 170         dprintk("renewing client (clientid %08x/%08x)\n",
 171                         clp->cl_clientid.cl_boot,
 172                         clp->cl_clientid.cl_id);
 173         list_move_tail(&clp->cl_lru, &nn->client_lru);
 174         clp->cl_time = ktime_get_boottime_seconds();
 175 }
 176
 177 static void put_client_renew_locked(struct nfs4_client *clp)
 178 {
 179         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 180
 181         lockdep_assert_held(&nn->client_lock);
 182
 183         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 184                 return;
 185         if (!is_client_expired(clp))
 186                 renew_client_locked(clp);
 187         else
 188                 wake_up_all(&expiry_wq);
 189 }
 190
 191 static void put_client_renew(struct nfs4_client *clp)
 192 {
 193         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 194
 195         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 196                 return;
 197         if (!is_client_expired(clp))
 198                 renew_client_locked(clp);
 199         else
 200                 wake_up_all(&expiry_wq);
 201         spin_unlock(&nn->client_lock);
 202 }
 203
 204 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 205 {
 206         __be32 status;
 207
 208         if (is_session_dead(ses))
 209                 return nfserr_badsession;
 210         status = get_client_locked(ses->se_client);
 211         if (status)
 212                 return status;
 213         atomic_inc(&ses->se_ref);
 214         return nfs_ok;
 215 }
 216
 217 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 218 {
 219         struct nfs4_client *clp = ses->se_client;
 220         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 221
 222         lockdep_assert_held(&nn->client_lock);
 223
 224         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 225                 free_session(ses);
 226         put_client_renew_locked(clp);
 227 }
 228
 229 static void nfsd4_put_session(struct nfsd4_session *ses)
 230 {
 231         struct nfs4_client *clp = ses->se_client;
 232         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 233
 234         spin_lock(&nn->client_lock);
 235         nfsd4_put_session_locked(ses);
 236         spin_unlock(&nn->client_lock);
 237 }
 238
 239 static struct nfsd4_blocked_lock *
 240 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 241                         struct nfsd_net *nn)
 242 {
 243         struct nfsd4_blocked_lock *cur, *found = NULL;
 244
 245         spin_lock(&nn->blocked_locks_lock);
 246         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 247                 if (fh_match(fh, &cur->nbl_fh)) {
 248                         list_del_init(&cur->nbl_list);
 249                         list_del_init(&cur->nbl_lru);
 250                         found = cur;
 251                         break;
 252                 }
 253         }
 254         spin_unlock(&nn->blocked_locks_lock);
 255         if (found)
 256                 locks_delete_block(&found->nbl_lock);
 257         return found;
 258 }
 259
 260 static struct nfsd4_blocked_lock *
 261 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 262                         struct nfsd_net *nn)
 263 {
 264         struct nfsd4_blocked_lock *nbl;
 265
 266         nbl = find_blocked_lock(lo, fh, nn);
 267         if (!nbl) {
 268                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 269                 if (nbl) {
 270                         INIT_LIST_HEAD(&nbl->nbl_list);
 271                         INIT_LIST_HEAD(&nbl->nbl_lru);
 272                         fh_copy_shallow(&nbl->nbl_fh, fh);
 273                         locks_init_lock(&nbl->nbl_lock);
 274                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 275                                         &nfsd4_cb_notify_lock_ops,
 276                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 277                 }
 278         }
 279         return nbl;
 280 }
 281
 282 static void
 283 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 284 {
 285         locks_delete_block(&nbl->nbl_lock);
 286         locks_release_private(&nbl->nbl_lock);
 287         kfree(nbl);
 288 }
 289
 290 static void
 291 remove_blocked_locks(struct nfs4_lockowner *lo)
 292 {
 293         struct nfs4_client *clp = lo->lo_owner.so_client;
 294         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 295         struct nfsd4_blocked_lock *nbl;
 296         LIST_HEAD(reaplist);
 297
 298         /* Dequeue all blocked locks */
 299         spin_lock(&nn->blocked_locks_lock);
 300         while (!list_empty(&lo->lo_blocked)) {
 301                 nbl = list_first_entry(&lo->lo_blocked,
 302                                         struct nfsd4_blocked_lock,
 303                                         nbl_list);
 304                 list_del_init(&nbl->nbl_list);
 305                 list_move(&nbl->nbl_lru, &reaplist);
 306         }
 307         spin_unlock(&nn->blocked_locks_lock);
 308
 309         /* Now free them */
 310         while (!list_empty(&reaplist)) {
 311                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 312                                         nbl_lru);
 313                 list_del_init(&nbl->nbl_lru);
 314                 free_blocked_lock(nbl);
 315         }
 316 }
 317
 318 static void
 319 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 320 {
 321         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 322                                                 struct nfsd4_blocked_lock, nbl_cb);
 323         locks_delete_block(&nbl->nbl_lock);
 324 }
 325
 326 static int
 327 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 328 {
 329         /*
 330          * Since this is just an optimization, we don't try very hard if it
 331          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 332          * just quit trying on anything else.
 333          */
 334         switch (task->tk_status) {
 335         case -NFS4ERR_DELAY:
 336                 rpc_delay(task, 1 * HZ);
 337                 return 0;
 338         default:
 339                 return 1;
 340         }
 341 }
 342
 343 static void
 344 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 345 {
 346         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 347                                                 struct nfsd4_blocked_lock, nbl_cb);
 348
 349         free_blocked_lock(nbl);
 350 }
 351
 352 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 353         .prepare        = nfsd4_cb_notify_lock_prepare,
 354         .done           = nfsd4_cb_notify_lock_done,
 355         .release        = nfsd4_cb_notify_lock_release,
 356 };
 357
 358 static inline struct nfs4_stateowner *
 359 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 360 {
 361         atomic_inc(&sop->so_count);
 362         return sop;
 363 }
 364
 365 static int
 366 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 367 {
 368         return (sop->so_owner.len == owner->len) &&
 369                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 370 }
 371
 372 static struct nfs4_openowner *
 373 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 374                         struct nfs4_client *clp)
 375 {
 376         struct nfs4_stateowner *so;
 377
 378         lockdep_assert_held(&clp->cl_lock);
 379
 380         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 381                             so_strhash) {
 382                 if (!so->so_is_open_owner)
 383                         continue;
 384                 if (same_owner_str(so, &open->op_owner))
 385                         return openowner(nfs4_get_stateowner(so));
 386         }
 387         return NULL;
 388 }
 389
 390 static struct nfs4_openowner *
 391 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 392                         struct nfs4_client *clp)
 393 {
 394         struct nfs4_openowner *oo;
 395
 396         spin_lock(&clp->cl_lock);
 397         oo = find_openstateowner_str_locked(hashval, open, clp);
 398         spin_unlock(&clp->cl_lock);
 399         return oo;
 400 }
 401
 402 static inline u32
 403 opaque_hashval(const void *ptr, int nbytes)
 404 {
 405         unsigned char *cptr = (unsigned char *) ptr;
 406
 407         u32 x = 0;
 408         while (nbytes--) {
 409                 x *= 37;
 410                 x += *cptr++;
 411         }
 412         return x;
 413 }
 414
 415 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 416 {
 417         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 418
 419         kmem_cache_free(file_slab, fp);
 420 }
 421
 422 void
 423 put_nfs4_file(struct nfs4_file *fi)
 424 {
 425         might_lock(&state_lock);
 426
 427         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 428                 hlist_del_rcu(&fi->fi_hash);
 429                 spin_unlock(&state_lock);
 430                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 431                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 432                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 433         }
 434 }
 435
 436 static struct nfsd_file *
 437 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 438 {
 439         if (f->fi_fds[oflag])
 440                 return nfsd_file_get(f->fi_fds[oflag]);
 441         return NULL;
 442 }
 443
 444 static struct nfsd_file *
 445 find_writeable_file_locked(struct nfs4_file *f)
 446 {
 447         struct nfsd_file *ret;
 448
 449         lockdep_assert_held(&f->fi_lock);
 450
 451         ret = __nfs4_get_fd(f, O_WRONLY);
 452         if (!ret)
 453                 ret = __nfs4_get_fd(f, O_RDWR);
 454         return ret;
 455 }
 456
 457 static struct nfsd_file *
 458 find_writeable_file(struct nfs4_file *f)
 459 {
 460         struct nfsd_file *ret;
 461
 462         spin_lock(&f->fi_lock);
 463         ret = find_writeable_file_locked(f);
 464         spin_unlock(&f->fi_lock);
 465
 466         return ret;
 467 }
 468
 469 static struct nfsd_file *
 470 find_readable_file_locked(struct nfs4_file *f)
 471 {
 472         struct nfsd_file *ret;
 473
 474         lockdep_assert_held(&f->fi_lock);
 475
 476         ret = __nfs4_get_fd(f, O_RDONLY);
 477         if (!ret)
 478                 ret = __nfs4_get_fd(f, O_RDWR);
 479         return ret;
 480 }
 481
 482 static struct nfsd_file *
 483 find_readable_file(struct nfs4_file *f)
 484 {
 485         struct nfsd_file *ret;
 486
 487         spin_lock(&f->fi_lock);
 488         ret = find_readable_file_locked(f);
 489         spin_unlock(&f->fi_lock);
 490
 491         return ret;
 492 }
 493
 494 struct nfsd_file *
 495 find_any_file(struct nfs4_file *f)
 496 {
 497         struct nfsd_file *ret;
 498
 499         if (!f)
 500                 return NULL;
 501         spin_lock(&f->fi_lock);
 502         ret = __nfs4_get_fd(f, O_RDWR);
 503         if (!ret) {
 504                 ret = __nfs4_get_fd(f, O_WRONLY);
 505                 if (!ret)
 506                         ret = __nfs4_get_fd(f, O_RDONLY);
 507         }
 508         spin_unlock(&f->fi_lock);
 509         return ret;
 510 }
 511
 512 static atomic_long_t num_delegations;
 513 unsigned long max_delegations;
 514
 515 /*
 516  * Open owner state (share locks)
 517  */
 518
 519 /* hash tables for lock and open owners */
 520 #define OWNER_HASH_BITS              8
 521 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 522 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 523
 524 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 525 {
 526         unsigned int ret;
 527
 528         ret = opaque_hashval(ownername->data, ownername->len);
 529         return ret & OWNER_HASH_MASK;
 530 }
 531
 532 /* hash table for nfs4_file */
 533 #define FILE_HASH_BITS                   8
 534 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 535
 536 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 537 {
 538         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 539 }
 540
 541 static unsigned int file_hashval(struct knfsd_fh *fh)
 542 {
 543         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 544 }
 545
 546 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 547
 548 static void
 549 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 550 {
 551         lockdep_assert_held(&fp->fi_lock);
 552
 553         if (access & NFS4_SHARE_ACCESS_WRITE)
 554                 atomic_inc(&fp->fi_access[O_WRONLY]);
 555         if (access & NFS4_SHARE_ACCESS_READ)
 556                 atomic_inc(&fp->fi_access[O_RDONLY]);
 557 }
 558
 559 static __be32
 560 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 561 {
 562         lockdep_assert_held(&fp->fi_lock);
 563
 564         /* Does this access mode make sense? */
 565         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 566                 return nfserr_inval;
 567
 568         /* Does it conflict with a deny mode already set? */
 569         if ((access & fp->fi_share_deny) != 0)
 570                 return nfserr_share_denied;
 571
 572         __nfs4_file_get_access(fp, access);
 573         return nfs_ok;
 574 }
 575
 576 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 577 {
 578         /* Common case is that there is no deny mode. */
 579         if (deny) {
 580                 /* Does this deny mode make sense? */
 581                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 582                         return nfserr_inval;
 583
 584                 if ((deny & NFS4_SHARE_DENY_READ) &&
 585                     atomic_read(&fp->fi_access[O_RDONLY]))
 586                         return nfserr_share_denied;
 587
 588                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 589                     atomic_read(&fp->fi_access[O_WRONLY]))
 590                         return nfserr_share_denied;
 591         }
 592         return nfs_ok;
 593 }
 594
 595 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 596 {
 597         might_lock(&fp->fi_lock);
 598
 599         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 600                 struct nfsd_file *f1 = NULL;
 601                 struct nfsd_file *f2 = NULL;
 602
 603                 swap(f1, fp->fi_fds[oflag]);
 604                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 605                         swap(f2, fp->fi_fds[O_RDWR]);
 606                 spin_unlock(&fp->fi_lock);
 607                 if (f1)
 608                         nfsd_file_put(f1);
 609                 if (f2)
 610                         nfsd_file_put(f2);
 611         }
 612 }
 613
 614 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 615 {
 616         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 617
 618         if (access & NFS4_SHARE_ACCESS_WRITE)
 619                 __nfs4_file_put_access(fp, O_WRONLY);
 620         if (access & NFS4_SHARE_ACCESS_READ)
 621                 __nfs4_file_put_access(fp, O_RDONLY);
 622 }
 623
 624 /*
 625  * Allocate a new open/delegation state counter. This is needed for
 626  * pNFS for proper return on close semantics.
 627  *
 628  * Note that we only allocate it for pNFS-enabled exports, otherwise
 629  * all pointers to struct nfs4_clnt_odstate are always NULL.
 630  */
 631 static struct nfs4_clnt_odstate *
 632 alloc_clnt_odstate(struct nfs4_client *clp)
 633 {
 634         struct nfs4_clnt_odstate *co;
 635
 636         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 637         if (co) {
 638                 co->co_client = clp;
 639                 refcount_set(&co->co_odcount, 1);
 640         }
 641         return co;
 642 }
 643
 644 static void
 645 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 646 {
 647         struct nfs4_file *fp = co->co_file;
 648
 649         lockdep_assert_held(&fp->fi_lock);
 650         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 651 }
 652
 653 static inline void
 654 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 655 {
 656         if (co)
 657                 refcount_inc(&co->co_odcount);
 658 }
 659
 660 static void
 661 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 662 {
 663         struct nfs4_file *fp;
 664
 665         if (!co)
 666                 return;
 667
 668         fp = co->co_file;
 669         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 670                 list_del(&co->co_perfile);
 671                 spin_unlock(&fp->fi_lock);
 672
 673                 nfsd4_return_all_file_layouts(co->co_client, fp);
 674                 kmem_cache_free(odstate_slab, co);
 675         }
 676 }
 677
 678 static struct nfs4_clnt_odstate *
 679 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 680 {
 681         struct nfs4_clnt_odstate *co;
 682         struct nfs4_client *cl;
 683
 684         if (!new)
 685                 return NULL;
 686
 687         cl = new->co_client;
 688
 689         spin_lock(&fp->fi_lock);
 690         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 691                 if (co->co_client == cl) {
 692                         get_clnt_odstate(co);
 693                         goto out;
 694                 }
 695         }
 696         co = new;
 697         co->co_file = fp;
 698         hash_clnt_odstate_locked(new);
 699 out:
 700         spin_unlock(&fp->fi_lock);
 701         return co;
 702 }
 703
 704 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 705                                   void (*sc_free)(struct nfs4_stid *))
 706 {
 707         struct nfs4_stid *stid;
 708         int new_id;
 709
 710         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 711         if (!stid)
 712                 return NULL;
 713
 714         idr_preload(GFP_KERNEL);
 715         spin_lock(&cl->cl_lock);
 716         /* Reserving 0 for start of file in nfsdfs "states" file: */
 717         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 718         spin_unlock(&cl->cl_lock);
 719         idr_preload_end();
 720         if (new_id < 0)
 721                 goto out_free;
 722
 723         stid->sc_free = sc_free;
 724         stid->sc_client = cl;
 725         stid->sc_stateid.si_opaque.so_id = new_id;
 726         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 727         /* Will be incremented before return to client: */
 728         refcount_set(&stid->sc_count, 1);
 729         spin_lock_init(&stid->sc_lock);
 730         INIT_LIST_HEAD(&stid->sc_cp_list);
 731
 732         /*
 733          * It shouldn't be a problem to reuse an opaque stateid value.
 734          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 735          * example, a stray write retransmission could be accepted by
 736          * the server when it should have been rejected.  Therefore,
 737          * adopt a trick from the sctp code to attempt to maximize the
 738          * amount of time until an id is reused, by ensuring they always
 739          * "increase" (mod INT_MAX):
 740          */
 741         return stid;
 742 out_free:
 743         kmem_cache_free(slab, stid);
 744         return NULL;
 745 }
 746
 747 /*
 748  * Create a unique stateid_t to represent each COPY.
 749  */
 750 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
 751                               unsigned char sc_type)
 752 {
 753         int new_id;
 754
 755         stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
 756         stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 757         stid->sc_type = sc_type;
 758
 759         idr_preload(GFP_KERNEL);
 760         spin_lock(&nn->s2s_cp_lock);
 761         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
 762         stid->stid.si_opaque.so_id = new_id;
 763         spin_unlock(&nn->s2s_cp_lock);
 764         idr_preload_end();
 765         if (new_id < 0)
 766                 return 0;
 767         return 1;
 768 }
 769
 770 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 771 {
 772         return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
 773 }
 774
 775 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
 776                                                      struct nfs4_stid *p_stid)
 777 {
 778         struct nfs4_cpntf_state *cps;
 779
 780         cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
 781         if (!cps)
 782                 return NULL;
 783         cps->cpntf_time = ktime_get_boottime_seconds();
 784         refcount_set(&cps->cp_stateid.sc_count, 1);
 785         if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
 786                 goto out_free;
 787         spin_lock(&nn->s2s_cp_lock);
 788         list_add(&cps->cp_list, &p_stid->sc_cp_list);
 789         spin_unlock(&nn->s2s_cp_lock);
 790         return cps;
 791 out_free:
 792         kfree(cps);
 793         return NULL;
 794 }
 795
 796 void nfs4_free_copy_state(struct nfsd4_copy *copy)
 797 {
 798         struct nfsd_net *nn;
 799
 800         WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
 801         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 802         spin_lock(&nn->s2s_cp_lock);
 803         idr_remove(&nn->s2s_cp_stateids,
 804                    copy->cp_stateid.stid.si_opaque.so_id);
 805         spin_unlock(&nn->s2s_cp_lock);
 806 }
 807
 808 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
 809 {
 810         struct nfs4_cpntf_state *cps;
 811         struct nfsd_net *nn;
 812
 813         nn = net_generic(net, nfsd_net_id);
 814         spin_lock(&nn->s2s_cp_lock);
 815         while (!list_empty(&stid->sc_cp_list)) {
 816                 cps = list_first_entry(&stid->sc_cp_list,
 817                                        struct nfs4_cpntf_state, cp_list);
 818                 _free_cpntf_state_locked(nn, cps);
 819         }
 820         spin_unlock(&nn->s2s_cp_lock);
 821 }
 822
 823 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 824 {
 825         struct nfs4_stid *stid;
 826
 827         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 828         if (!stid)
 829                 return NULL;
 830
 831         return openlockstateid(stid);
 832 }
 833
 834 static void nfs4_free_deleg(struct nfs4_stid *stid)
 835 {
 836         kmem_cache_free(deleg_slab, stid);
 837         atomic_long_dec(&num_delegations);
 838 }
 839
 840 /*
 841  * When we recall a delegation, we should be careful not to hand it
 842  * out again straight away.
 843  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 844  * in which the filehandles of recalled delegations are "stored".
 845  * If a filehandle appear in either filter, a delegation is blocked.
 846  * When a delegation is recalled, the filehandle is stored in the "new"
 847  * filter.
 848  * Every 30 seconds we swap the filters and clear the "new" one,
 849  * unless both are empty of course.
 850  *
 851  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 852  * low 3 bytes as hash-table indices.
 853  *
 854  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 855  * is used to manage concurrent access.  Testing does not need the lock
 856  * except when swapping the two filters.
 857  */
 858 static DEFINE_SPINLOCK(blocked_delegations_lock);
 859 static struct bloom_pair {
 860         int     entries, old_entries;
 861         time64_t swap_time;
 862         int     new; /* index into 'set' */
 863         DECLARE_BITMAP(set[2], 256);
 864 } blocked_delegations;
 865
 866 static int delegation_blocked(struct knfsd_fh *fh)
 867 {
 868         u32 hash;
 869         struct bloom_pair *bd = &blocked_delegations;
 870
 871         if (bd->entries == 0)
 872                 return 0;
 873         if (ktime_get_seconds() - bd->swap_time > 30) {
 874                 spin_lock(&blocked_delegations_lock);
 875                 if (ktime_get_seconds() - bd->swap_time > 30) {
 876                         bd->entries -= bd->old_entries;
 877                         bd->old_entries = bd->entries;
 878                         memset(bd->set[bd->new], 0,
 879                                sizeof(bd->set[0]));
 880                         bd->new = 1-bd->new;
 881                         bd->swap_time = ktime_get_seconds();
 882                 }
 883                 spin_unlock(&blocked_delegations_lock);
 884         }
 885         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 886         if (test_bit(hash&255, bd->set[0]) &&
 887             test_bit((hash>>8)&255, bd->set[0]) &&
 888             test_bit((hash>>16)&255, bd->set[0]))
 889                 return 1;
 890
 891         if (test_bit(hash&255, bd->set[1]) &&
 892             test_bit((hash>>8)&255, bd->set[1]) &&
 893             test_bit((hash>>16)&255, bd->set[1]))
 894                 return 1;
 895
 896         return 0;
 897 }
 898
 899 static void block_delegations(struct knfsd_fh *fh)
 900 {
 901         u32 hash;
 902         struct bloom_pair *bd = &blocked_delegations;
 903
 904         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 905
 906         spin_lock(&blocked_delegations_lock);
 907         __set_bit(hash&255, bd->set[bd->new]);
 908         __set_bit((hash>>8)&255, bd->set[bd->new]);
 909         __set_bit((hash>>16)&255, bd->set[bd->new]);
 910         if (bd->entries == 0)
 911                 bd->swap_time = ktime_get_seconds();
 912         bd->entries += 1;
 913         spin_unlock(&blocked_delegations_lock);
 914 }
 915
 916 static struct nfs4_delegation *
 917 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 918                  struct svc_fh *current_fh,
 919                  struct nfs4_clnt_odstate *odstate)
 920 {
 921         struct nfs4_delegation *dp;
 922         long n;
 923
 924         dprintk("NFSD alloc_init_deleg\n");
 925         n = atomic_long_inc_return(&num_delegations);
 926         if (n < 0 || n > max_delegations)
 927                 goto out_dec;
 928         if (delegation_blocked(&current_fh->fh_handle))
 929                 goto out_dec;
 930         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 931         if (dp == NULL)
 932                 goto out_dec;
 933
 934         /*
 935          * delegation seqid's are never incremented.  The 4.1 special
 936          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 937          * 0 anyway just for consistency and use 1:
 938          */
 939         dp->dl_stid.sc_stateid.si_generation = 1;
 940         INIT_LIST_HEAD(&dp->dl_perfile);
 941         INIT_LIST_HEAD(&dp->dl_perclnt);
 942         INIT_LIST_HEAD(&dp->dl_recall_lru);
 943         dp->dl_clnt_odstate = odstate;
 944         get_clnt_odstate(odstate);
 945         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 946         dp->dl_retries = 1;
 947         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 948                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 949         get_nfs4_file(fp);
 950         dp->dl_stid.sc_file = fp;
 951         return dp;
 952 out_dec:
 953         atomic_long_dec(&num_delegations);
 954         return NULL;
 955 }
 956
 957 void
 958 nfs4_put_stid(struct nfs4_stid *s)
 959 {
 960         struct nfs4_file *fp = s->sc_file;
 961         struct nfs4_client *clp = s->sc_client;
 962
 963         might_lock(&clp->cl_lock);
 964
 965         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 966                 wake_up_all(&close_wq);
 967                 return;
 968         }
 969         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 970         nfs4_free_cpntf_statelist(clp->net, s);
 971         spin_unlock(&clp->cl_lock);
 972         s->sc_free(s);
 973         if (fp)
 974                 put_nfs4_file(fp);
 975 }
 976
 977 void
 978 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 979 {
 980         stateid_t *src = &stid->sc_stateid;
 981
 982         spin_lock(&stid->sc_lock);
 983         if (unlikely(++src->si_generation == 0))
 984                 src->si_generation = 1;
 985         memcpy(dst, src, sizeof(*dst));
 986         spin_unlock(&stid->sc_lock);
 987 }
 988
 989 static void put_deleg_file(struct nfs4_file *fp)
 990 {
 991         struct nfsd_file *nf = NULL;
 992
 993         spin_lock(&fp->fi_lock);
 994         if (--fp->fi_delegees == 0)
 995                 swap(nf, fp->fi_deleg_file);
 996         spin_unlock(&fp->fi_lock);
 997
 998         if (nf)
 999                 nfsd_file_put(nf);
1000 }
1001
1002 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1003 {
1004         struct nfs4_file *fp = dp->dl_stid.sc_file;
1005         struct nfsd_file *nf = fp->fi_deleg_file;
1006
1007         WARN_ON_ONCE(!fp->fi_delegees);
1008
1009         vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1010         put_deleg_file(fp);
1011 }
1012
1013 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1014 {
1015         put_clnt_odstate(dp->dl_clnt_odstate);
1016         nfs4_unlock_deleg_lease(dp);
1017         nfs4_put_stid(&dp->dl_stid);
1018 }
1019
1020 void nfs4_unhash_stid(struct nfs4_stid *s)
1021 {
1022         s->sc_type = 0;
1023 }
1024
1025 /**
1026  * nfs4_delegation_exists - Discover if this delegation already exists
1027  * @clp:     a pointer to the nfs4_client we're granting a delegation to
1028  * @fp:      a pointer to the nfs4_file we're granting a delegation on
1029  *
1030  * Return:
1031  *      On success: true iff an existing delegation is found
1032  */
1033
1034 static bool
1035 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1036 {
1037         struct nfs4_delegation *searchdp = NULL;
1038         struct nfs4_client *searchclp = NULL;
1039
1040         lockdep_assert_held(&state_lock);
1041         lockdep_assert_held(&fp->fi_lock);
1042
1043         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1044                 searchclp = searchdp->dl_stid.sc_client;
1045                 if (clp == searchclp) {
1046                         return true;
1047                 }
1048         }
1049         return false;
1050 }
1051
1052 /**
1053  * hash_delegation_locked - Add a delegation to the appropriate lists
1054  * @dp:     a pointer to the nfs4_delegation we are adding.
1055  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1056  *
1057  * Return:
1058  *      On success: NULL if the delegation was successfully hashed.
1059  *
1060  *      On error: -EAGAIN if one was previously granted to this
1061  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1062  *
1063  */
1064
1065 static int
1066 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1067 {
1068         struct nfs4_client *clp = dp->dl_stid.sc_client;
1069
1070         lockdep_assert_held(&state_lock);
1071         lockdep_assert_held(&fp->fi_lock);
1072
1073         if (nfs4_delegation_exists(clp, fp))
1074                 return -EAGAIN;
1075         refcount_inc(&dp->dl_stid.sc_count);
1076         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1077         list_add(&dp->dl_perfile, &fp->fi_delegations);
1078         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1079         return 0;
1080 }
1081
1082 static bool
1083 unhash_delegation_locked(struct nfs4_delegation *dp)
1084 {
1085         struct nfs4_file *fp = dp->dl_stid.sc_file;
1086
1087         lockdep_assert_held(&state_lock);
1088
1089         if (list_empty(&dp->dl_perfile))
1090                 return false;
1091
1092         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1093         /* Ensure that deleg break won't try to requeue it */
1094         ++dp->dl_time;
1095         spin_lock(&fp->fi_lock);
1096         list_del_init(&dp->dl_perclnt);
1097         list_del_init(&dp->dl_recall_lru);
1098         list_del_init(&dp->dl_perfile);
1099         spin_unlock(&fp->fi_lock);
1100         return true;
1101 }
1102
1103 static void destroy_delegation(struct nfs4_delegation *dp)
1104 {
1105         bool unhashed;
1106
1107         spin_lock(&state_lock);
1108         unhashed = unhash_delegation_locked(dp);
1109         spin_unlock(&state_lock);
1110         if (unhashed)
1111                 destroy_unhashed_deleg(dp);
1112 }
1113
1114 static void revoke_delegation(struct nfs4_delegation *dp)
1115 {
1116         struct nfs4_client *clp = dp->dl_stid.sc_client;
1117
1118         WARN_ON(!list_empty(&dp->dl_recall_lru));
1119
1120         if (clp->cl_minorversion) {
1121                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1122                 refcount_inc(&dp->dl_stid.sc_count);
1123                 spin_lock(&clp->cl_lock);
1124                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1125                 spin_unlock(&clp->cl_lock);
1126         }
1127         destroy_unhashed_deleg(dp);
1128 }
1129
1130 /*
1131  * SETCLIENTID state
1132  */
1133
1134 static unsigned int clientid_hashval(u32 id)
1135 {
1136         return id & CLIENT_HASH_MASK;
1137 }
1138
1139 static unsigned int clientstr_hashval(struct xdr_netobj name)
1140 {
1141         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1142 }
1143
1144 /*
1145  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1146  * st_{access,deny}_bmap field of the stateid, in order to track not
1147  * only what share bits are currently in force, but also what
1148  * combinations of share bits previous opens have used.  This allows us
1149  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1150  * return an error if the client attempt to downgrade to a combination
1151  * of share bits not explicable by closing some of its previous opens.
1152  *
1153  * XXX: This enforcement is actually incomplete, since we don't keep
1154  * track of access/deny bit combinations; so, e.g., we allow:
1155  *
1156  *      OPEN allow read, deny write
1157  *      OPEN allow both, deny none
1158  *      DOWNGRADE allow read, deny none
1159  *
1160  * which we should reject.
1161  */
1162 static unsigned int
1163 bmap_to_share_mode(unsigned long bmap) {
1164         int i;
1165         unsigned int access = 0;
1166
1167         for (i = 1; i < 4; i++) {
1168                 if (test_bit(i, &bmap))
1169                         access |= i;
1170         }
1171         return access;
1172 }
1173
1174 /* set share access for a given stateid */
1175 static inline void
1176 set_access(u32 access, struct nfs4_ol_stateid *stp)
1177 {
1178         unsigned char mask = 1 << access;
1179
1180         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1181         stp->st_access_bmap |= mask;
1182 }
1183
1184 /* clear share access for a given stateid */
1185 static inline void
1186 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1187 {
1188         unsigned char mask = 1 << access;
1189
1190         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1191         stp->st_access_bmap &= ~mask;
1192 }
1193
1194 /* test whether a given stateid has access */
1195 static inline bool
1196 test_access(u32 access, struct nfs4_ol_stateid *stp)
1197 {
1198         unsigned char mask = 1 << access;
1199
1200         return (bool)(stp->st_access_bmap & mask);
1201 }
1202
1203 /* set share deny for a given stateid */
1204 static inline void
1205 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1206 {
1207         unsigned char mask = 1 << deny;
1208
1209         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1210         stp->st_deny_bmap |= mask;
1211 }
1212
1213 /* clear share deny for a given stateid */
1214 static inline void
1215 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1216 {
1217         unsigned char mask = 1 << deny;
1218
1219         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1220         stp->st_deny_bmap &= ~mask;
1221 }
1222
1223 /* test whether a given stateid is denying specific access */
1224 static inline bool
1225 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1226 {
1227         unsigned char mask = 1 << deny;
1228
1229         return (bool)(stp->st_deny_bmap & mask);
1230 }
1231
1232 static int nfs4_access_to_omode(u32 access)
1233 {
1234         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1235         case NFS4_SHARE_ACCESS_READ:
1236                 return O_RDONLY;
1237         case NFS4_SHARE_ACCESS_WRITE:
1238                 return O_WRONLY;
1239         case NFS4_SHARE_ACCESS_BOTH:
1240                 return O_RDWR;
1241         }
1242         WARN_ON_ONCE(1);
1243         return O_RDONLY;
1244 }
1245
1246 /*
1247  * A stateid that had a deny mode associated with it is being released
1248  * or downgraded. Recalculate the deny mode on the file.
1249  */
1250 static void
1251 recalculate_deny_mode(struct nfs4_file *fp)
1252 {
1253         struct nfs4_ol_stateid *stp;
1254
1255         spin_lock(&fp->fi_lock);
1256         fp->fi_share_deny = 0;
1257         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1258                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1259         spin_unlock(&fp->fi_lock);
1260 }
1261
1262 static void
1263 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1264 {
1265         int i;
1266         bool change = false;
1267
1268         for (i = 1; i < 4; i++) {
1269                 if ((i & deny) != i) {
1270                         change = true;
1271                         clear_deny(i, stp);
1272                 }
1273         }
1274
1275         /* Recalculate per-file deny mode if there was a change */
1276         if (change)
1277                 recalculate_deny_mode(stp->st_stid.sc_file);
1278 }
1279
1280 /* release all access and file references for a given stateid */
1281 static void
1282 release_all_access(struct nfs4_ol_stateid *stp)
1283 {
1284         int i;
1285         struct nfs4_file *fp = stp->st_stid.sc_file;
1286
1287         if (fp && stp->st_deny_bmap != 0)
1288                 recalculate_deny_mode(fp);
1289
1290         for (i = 1; i < 4; i++) {
1291                 if (test_access(i, stp))
1292                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1293                 clear_access(i, stp);
1294         }
1295 }
1296
1297 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1298 {
1299         kfree(sop->so_owner.data);
1300         sop->so_ops->so_free(sop);
1301 }
1302
1303 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1304 {
1305         struct nfs4_client *clp = sop->so_client;
1306
1307         might_lock(&clp->cl_lock);
1308
1309         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1310                 return;
1311         sop->so_ops->so_unhash(sop);
1312         spin_unlock(&clp->cl_lock);
1313         nfs4_free_stateowner(sop);
1314 }
1315
1316 static bool
1317 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1318 {
1319         return list_empty(&stp->st_perfile);
1320 }
1321
1322 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1323 {
1324         struct nfs4_file *fp = stp->st_stid.sc_file;
1325
1326         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1327
1328         if (list_empty(&stp->st_perfile))
1329                 return false;
1330
1331         spin_lock(&fp->fi_lock);
1332         list_del_init(&stp->st_perfile);
1333         spin_unlock(&fp->fi_lock);
1334         list_del(&stp->st_perstateowner);
1335         return true;
1336 }
1337
1338 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1339 {
1340         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1341
1342         put_clnt_odstate(stp->st_clnt_odstate);
1343         release_all_access(stp);
1344         if (stp->st_stateowner)
1345                 nfs4_put_stateowner(stp->st_stateowner);
1346         kmem_cache_free(stateid_slab, stid);
1347 }
1348
1349 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1350 {
1351         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1352         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1353         struct nfsd_file *nf;
1354
1355         nf = find_any_file(stp->st_stid.sc_file);
1356         if (nf) {
1357                 get_file(nf->nf_file);
1358                 filp_close(nf->nf_file, (fl_owner_t)lo);
1359                 nfsd_file_put(nf);
1360         }
1361         nfs4_free_ol_stateid(stid);
1362 }
1363
1364 /*
1365  * Put the persistent reference to an already unhashed generic stateid, while
1366  * holding the cl_lock. If it's the last reference, then put it onto the
1367  * reaplist for later destruction.
1368  */
1369 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1370                                        struct list_head *reaplist)
1371 {
1372         struct nfs4_stid *s = &stp->st_stid;
1373         struct nfs4_client *clp = s->sc_client;
1374
1375         lockdep_assert_held(&clp->cl_lock);
1376
1377         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1378
1379         if (!refcount_dec_and_test(&s->sc_count)) {
1380                 wake_up_all(&close_wq);
1381                 return;
1382         }
1383
1384         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1385         list_add(&stp->st_locks, reaplist);
1386 }
1387
1388 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1389 {
1390         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1391
1392         if (!unhash_ol_stateid(stp))
1393                 return false;
1394         list_del_init(&stp->st_locks);
1395         nfs4_unhash_stid(&stp->st_stid);
1396         return true;
1397 }
1398
1399 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1400 {
1401         struct nfs4_client *clp = stp->st_stid.sc_client;
1402         bool unhashed;
1403
1404         spin_lock(&clp->cl_lock);
1405         unhashed = unhash_lock_stateid(stp);
1406         spin_unlock(&clp->cl_lock);
1407         if (unhashed)
1408                 nfs4_put_stid(&stp->st_stid);
1409 }
1410
1411 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1412 {
1413         struct nfs4_client *clp = lo->lo_owner.so_client;
1414
1415         lockdep_assert_held(&clp->cl_lock);
1416
1417         list_del_init(&lo->lo_owner.so_strhash);
1418 }
1419
1420 /*
1421  * Free a list of generic stateids that were collected earlier after being
1422  * fully unhashed.
1423  */
1424 static void
1425 free_ol_stateid_reaplist(struct list_head *reaplist)
1426 {
1427         struct nfs4_ol_stateid *stp;
1428         struct nfs4_file *fp;
1429
1430         might_sleep();
1431
1432         while (!list_empty(reaplist)) {
1433                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1434                                        st_locks);
1435                 list_del(&stp->st_locks);
1436                 fp = stp->st_stid.sc_file;
1437                 stp->st_stid.sc_free(&stp->st_stid);
1438                 if (fp)
1439                         put_nfs4_file(fp);
1440         }
1441 }
1442
1443 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1444                                        struct list_head *reaplist)
1445 {
1446         struct nfs4_ol_stateid *stp;
1447
1448         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1449
1450         while (!list_empty(&open_stp->st_locks)) {
1451                 stp = list_entry(open_stp->st_locks.next,
1452                                 struct nfs4_ol_stateid, st_locks);
1453                 WARN_ON(!unhash_lock_stateid(stp));
1454                 put_ol_stateid_locked(stp, reaplist);
1455         }
1456 }
1457
1458 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1459                                 struct list_head *reaplist)
1460 {
1461         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1462
1463         if (!unhash_ol_stateid(stp))
1464                 return false;
1465         release_open_stateid_locks(stp, reaplist);
1466         return true;
1467 }
1468
1469 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1470 {
1471         LIST_HEAD(reaplist);
1472
1473         spin_lock(&stp->st_stid.sc_client->cl_lock);
1474         if (unhash_open_stateid(stp, &reaplist))
1475                 put_ol_stateid_locked(stp, &reaplist);
1476         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1477         free_ol_stateid_reaplist(&reaplist);
1478 }
1479
1480 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1481 {
1482         struct nfs4_client *clp = oo->oo_owner.so_client;
1483
1484         lockdep_assert_held(&clp->cl_lock);
1485
1486         list_del_init(&oo->oo_owner.so_strhash);
1487         list_del_init(&oo->oo_perclient);
1488 }
1489
1490 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1491 {
1492         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1493                                           nfsd_net_id);
1494         struct nfs4_ol_stateid *s;
1495
1496         spin_lock(&nn->client_lock);
1497         s = oo->oo_last_closed_stid;
1498         if (s) {
1499                 list_del_init(&oo->oo_close_lru);
1500                 oo->oo_last_closed_stid = NULL;
1501         }
1502         spin_unlock(&nn->client_lock);
1503         if (s)
1504                 nfs4_put_stid(&s->st_stid);
1505 }
1506
1507 static void release_openowner(struct nfs4_openowner *oo)
1508 {
1509         struct nfs4_ol_stateid *stp;
1510         struct nfs4_client *clp = oo->oo_owner.so_client;
1511         struct list_head reaplist;
1512
1513         INIT_LIST_HEAD(&reaplist);
1514
1515         spin_lock(&clp->cl_lock);
1516         unhash_openowner_locked(oo);
1517         while (!list_empty(&oo->oo_owner.so_stateids)) {
1518                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1519                                 struct nfs4_ol_stateid, st_perstateowner);
1520                 if (unhash_open_stateid(stp, &reaplist))
1521                         put_ol_stateid_locked(stp, &reaplist);
1522         }
1523         spin_unlock(&clp->cl_lock);
1524         free_ol_stateid_reaplist(&reaplist);
1525         release_last_closed_stateid(oo);
1526         nfs4_put_stateowner(&oo->oo_owner);
1527 }
1528
1529 static inline int
1530 hash_sessionid(struct nfs4_sessionid *sessionid)
1531 {
1532         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1533
1534         return sid->sequence % SESSION_HASH_SIZE;
1535 }
1536
1537 #ifdef CONFIG_SUNRPC_DEBUG
1538 static inline void
1539 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1540 {
1541         u32 *ptr = (u32 *)(&sessionid->data[0]);
1542         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1543 }
1544 #else
1545 static inline void
1546 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1547 {
1548 }
1549 #endif
1550
1551 /*
1552  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1553  * won't be used for replay.
1554  */
1555 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1556 {
1557         struct nfs4_stateowner *so = cstate->replay_owner;
1558
1559         if (nfserr == nfserr_replay_me)
1560                 return;
1561
1562         if (!seqid_mutating_err(ntohl(nfserr))) {
1563                 nfsd4_cstate_clear_replay(cstate);
1564                 return;
1565         }
1566         if (!so)
1567                 return;
1568         if (so->so_is_open_owner)
1569                 release_last_closed_stateid(openowner(so));
1570         so->so_seqid++;
1571         return;
1572 }
1573
1574 static void
1575 gen_sessionid(struct nfsd4_session *ses)
1576 {
1577         struct nfs4_client *clp = ses->se_client;
1578         struct nfsd4_sessionid *sid;
1579
1580         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1581         sid->clientid = clp->cl_clientid;
1582         sid->sequence = current_sessionid++;
1583         sid->reserved = 0;
1584 }
1585
1586 /*
1587  * The protocol defines ca_maxresponssize_cached to include the size of
1588  * the rpc header, but all we need to cache is the data starting after
1589  * the end of the initial SEQUENCE operation--the rest we regenerate
1590  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1591  * value that is the number of bytes in our cache plus a few additional
1592  * bytes.  In order to stay on the safe side, and not promise more than
1593  * we can cache, those additional bytes must be the minimum possible: 24
1594  * bytes of rpc header (xid through accept state, with AUTH_NULL
1595  * verifier), 12 for the compound header (with zero-length tag), and 44
1596  * for the SEQUENCE op response:
1597  */
1598 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1599
1600 static void
1601 free_session_slots(struct nfsd4_session *ses)
1602 {
1603         int i;
1604
1605         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1606                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1607                 kfree(ses->se_slots[i]);
1608         }
1609 }
1610
1611 /*
1612  * We don't actually need to cache the rpc and session headers, so we
1613  * can allocate a little less for each slot:
1614  */
1615 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1616 {
1617         u32 size;
1618
1619         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1620                 size = 0;
1621         else
1622                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1623         return size + sizeof(struct nfsd4_slot);
1624 }
1625
1626 /*
1627  * XXX: If we run out of reserved DRC memory we could (up to a point)
1628  * re-negotiate active sessions and reduce their slot usage to make
1629  * room for new connections. For now we just fail the create session.
1630  */
1631 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1632 {
1633         u32 slotsize = slot_bytes(ca);
1634         u32 num = ca->maxreqs;
1635         unsigned long avail, total_avail;
1636         unsigned int scale_factor;
1637
1638         spin_lock(&nfsd_drc_lock);
1639         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1640                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1641         else
1642                 /* We have handed out more space than we chose in
1643                  * set_max_drc() to allow.  That isn't really a
1644                  * problem as long as that doesn't make us think we
1645                  * have lots more due to integer overflow.
1646                  */
1647                 total_avail = 0;
1648         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1649         /*
1650          * Never use more than a fraction of the remaining memory,
1651          * unless it's the only way to give this client a slot.
1652          * The chosen fraction is either 1/8 or 1/number of threads,
1653          * whichever is smaller.  This ensures there are adequate
1654          * slots to support multiple clients per thread.
1655          * Give the client one slot even if that would require
1656          * over-allocation--it is better than failure.
1657          */
1658         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1659
1660         avail = clamp_t(unsigned long, avail, slotsize,
1661                         total_avail/scale_factor);
1662         num = min_t(int, num, avail / slotsize);
1663         num = max_t(int, num, 1);
1664         nfsd_drc_mem_used += num * slotsize;
1665         spin_unlock(&nfsd_drc_lock);
1666
1667         return num;
1668 }
1669
1670 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1671 {
1672         int slotsize = slot_bytes(ca);
1673
1674         spin_lock(&nfsd_drc_lock);
1675         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1676         spin_unlock(&nfsd_drc_lock);
1677 }
1678
1679 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1680                                            struct nfsd4_channel_attrs *battrs)
1681 {
1682         int numslots = fattrs->maxreqs;
1683         int slotsize = slot_bytes(fattrs);
1684         struct nfsd4_session *new;
1685         int mem, i;
1686
1687         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1688                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1689         mem = numslots * sizeof(struct nfsd4_slot *);
1690
1691         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1692         if (!new)
1693                 return NULL;
1694         /* allocate each struct nfsd4_slot and data cache in one piece */
1695         for (i = 0; i < numslots; i++) {
1696                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1697                 if (!new->se_slots[i])
1698                         goto out_free;
1699         }
1700
1701         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1702         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1703
1704         return new;
1705 out_free:
1706         while (i--)
1707                 kfree(new->se_slots[i]);
1708         kfree(new);
1709         return NULL;
1710 }
1711
1712 static void free_conn(struct nfsd4_conn *c)
1713 {
1714         svc_xprt_put(c->cn_xprt);
1715         kfree(c);
1716 }
1717
1718 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1719 {
1720         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1721         struct nfs4_client *clp = c->cn_session->se_client;
1722
1723         spin_lock(&clp->cl_lock);
1724         if (!list_empty(&c->cn_persession)) {
1725                 list_del(&c->cn_persession);
1726                 free_conn(c);
1727         }
1728         nfsd4_probe_callback(clp);
1729         spin_unlock(&clp->cl_lock);
1730 }
1731
1732 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1733 {
1734         struct nfsd4_conn *conn;
1735
1736         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1737         if (!conn)
1738                 return NULL;
1739         svc_xprt_get(rqstp->rq_xprt);
1740         conn->cn_xprt = rqstp->rq_xprt;
1741         conn->cn_flags = flags;
1742         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1743         return conn;
1744 }
1745
1746 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1747 {
1748         conn->cn_session = ses;
1749         list_add(&conn->cn_persession, &ses->se_conns);
1750 }
1751
1752 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1753 {
1754         struct nfs4_client *clp = ses->se_client;
1755
1756         spin_lock(&clp->cl_lock);
1757         __nfsd4_hash_conn(conn, ses);
1758         spin_unlock(&clp->cl_lock);
1759 }
1760
1761 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1762 {
1763         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1764         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1765 }
1766
1767 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1768 {
1769         int ret;
1770
1771         nfsd4_hash_conn(conn, ses);
1772         ret = nfsd4_register_conn(conn);
1773         if (ret)
1774                 /* oops; xprt is already down: */
1775                 nfsd4_conn_lost(&conn->cn_xpt_user);
1776         /* We may have gained or lost a callback channel: */
1777         nfsd4_probe_callback_sync(ses->se_client);
1778 }
1779
1780 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1781 {
1782         u32 dir = NFS4_CDFC4_FORE;
1783
1784         if (cses->flags & SESSION4_BACK_CHAN)
1785                 dir |= NFS4_CDFC4_BACK;
1786         return alloc_conn(rqstp, dir);
1787 }
1788
1789 /* must be called under client_lock */
1790 static void nfsd4_del_conns(struct nfsd4_session *s)
1791 {
1792         struct nfs4_client *clp = s->se_client;
1793         struct nfsd4_conn *c;
1794
1795         spin_lock(&clp->cl_lock);
1796         while (!list_empty(&s->se_conns)) {
1797                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1798                 list_del_init(&c->cn_persession);
1799                 spin_unlock(&clp->cl_lock);
1800
1801                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1802                 free_conn(c);
1803
1804                 spin_lock(&clp->cl_lock);
1805         }
1806         spin_unlock(&clp->cl_lock);
1807 }
1808
1809 static void __free_session(struct nfsd4_session *ses)
1810 {
1811         free_session_slots(ses);
1812         kfree(ses);
1813 }
1814
1815 static void free_session(struct nfsd4_session *ses)
1816 {
1817         nfsd4_del_conns(ses);
1818         nfsd4_put_drc_mem(&ses->se_fchannel);
1819         __free_session(ses);
1820 }
1821
1822 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1823 {
1824         int idx;
1825         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1826
1827         new->se_client = clp;
1828         gen_sessionid(new);
1829
1830         INIT_LIST_HEAD(&new->se_conns);
1831
1832         new->se_cb_seq_nr = 1;
1833         new->se_flags = cses->flags;
1834         new->se_cb_prog = cses->callback_prog;
1835         new->se_cb_sec = cses->cb_sec;
1836         atomic_set(&new->se_ref, 0);
1837         idx = hash_sessionid(&new->se_sessionid);
1838         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1839         spin_lock(&clp->cl_lock);
1840         list_add(&new->se_perclnt, &clp->cl_sessions);
1841         spin_unlock(&clp->cl_lock);
1842
1843         {
1844                 struct sockaddr *sa = svc_addr(rqstp);
1845                 /*
1846                  * This is a little silly; with sessions there's no real
1847                  * use for the callback address.  Use the peer address
1848                  * as a reasonable default for now, but consider fixing
1849                  * the rpc client not to require an address in the
1850                  * future:
1851                  */
1852                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1853                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1854         }
1855 }
1856
1857 /* caller must hold client_lock */
1858 static struct nfsd4_session *
1859 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1860 {
1861         struct nfsd4_session *elem;
1862         int idx;
1863         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1864
1865         lockdep_assert_held(&nn->client_lock);
1866
1867         dump_sessionid(__func__, sessionid);
1868         idx = hash_sessionid(sessionid);
1869         /* Search in the appropriate list */
1870         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1871                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1872                             NFS4_MAX_SESSIONID_LEN)) {
1873                         return elem;
1874                 }
1875         }
1876
1877         dprintk("%s: session not found\n", __func__);
1878         return NULL;
1879 }
1880
1881 static struct nfsd4_session *
1882 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1883                 __be32 *ret)
1884 {
1885         struct nfsd4_session *session;
1886         __be32 status = nfserr_badsession;
1887
1888         session = __find_in_sessionid_hashtbl(sessionid, net);
1889         if (!session)
1890                 goto out;
1891         status = nfsd4_get_session_locked(session);
1892         if (status)
1893                 session = NULL;
1894 out:
1895         *ret = status;
1896         return session;
1897 }
1898
1899 /* caller must hold client_lock */
1900 static void
1901 unhash_session(struct nfsd4_session *ses)
1902 {
1903         struct nfs4_client *clp = ses->se_client;
1904         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1905
1906         lockdep_assert_held(&nn->client_lock);
1907
1908         list_del(&ses->se_hash);
1909         spin_lock(&ses->se_client->cl_lock);
1910         list_del(&ses->se_perclnt);
1911         spin_unlock(&ses->se_client->cl_lock);
1912 }
1913
1914 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1915 static int
1916 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1917 {
1918         /*
1919          * We're assuming the clid was not given out from a boot
1920          * precisely 2^32 (about 136 years) before this one.  That seems
1921          * a safe assumption:
1922          */
1923         if (clid->cl_boot == (u32)nn->boot_time)
1924                 return 0;
1925         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08llx\n",
1926                 clid->cl_boot, clid->cl_id, nn->boot_time);
1927         return 1;
1928 }
1929
1930 /*
1931  * XXX Should we use a slab cache ?
1932  * This type of memory management is somewhat inefficient, but we use it
1933  * anyway since SETCLIENTID is not a common operation.
1934  */
1935 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1936 {
1937         struct nfs4_client *clp;
1938         int i;
1939
1940         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1941         if (clp == NULL)
1942                 return NULL;
1943         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1944         if (clp->cl_name.data == NULL)
1945                 goto err_no_name;
1946         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1947                                                  sizeof(struct list_head),
1948                                                  GFP_KERNEL);
1949         if (!clp->cl_ownerstr_hashtbl)
1950                 goto err_no_hashtbl;
1951         for (i = 0; i < OWNER_HASH_SIZE; i++)
1952                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1953         INIT_LIST_HEAD(&clp->cl_sessions);
1954         idr_init(&clp->cl_stateids);
1955         atomic_set(&clp->cl_rpc_users, 0);
1956         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1957         INIT_LIST_HEAD(&clp->cl_idhash);
1958         INIT_LIST_HEAD(&clp->cl_openowners);
1959         INIT_LIST_HEAD(&clp->cl_delegations);
1960         INIT_LIST_HEAD(&clp->cl_lru);
1961         INIT_LIST_HEAD(&clp->cl_revoked);
1962 #ifdef CONFIG_NFSD_PNFS
1963         INIT_LIST_HEAD(&clp->cl_lo_states);
1964 #endif
1965         INIT_LIST_HEAD(&clp->async_copies);
1966         spin_lock_init(&clp->async_lock);
1967         spin_lock_init(&clp->cl_lock);
1968         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1969         return clp;
1970 err_no_hashtbl:
1971         kfree(clp->cl_name.data);
1972 err_no_name:
1973         kmem_cache_free(client_slab, clp);
1974         return NULL;
1975 }
1976
1977 static void __free_client(struct kref *k)
1978 {
1979         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1980         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1981
1982         free_svc_cred(&clp->cl_cred);
1983         kfree(clp->cl_ownerstr_hashtbl);
1984         kfree(clp->cl_name.data);
1985         kfree(clp->cl_nii_domain.data);
1986         kfree(clp->cl_nii_name.data);
1987         idr_destroy(&clp->cl_stateids);
1988         kmem_cache_free(client_slab, clp);
1989 }
1990
1991 static void drop_client(struct nfs4_client *clp)
1992 {
1993         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1994 }
1995
1996 static void
1997 free_client(struct nfs4_client *clp)
1998 {
1999         while (!list_empty(&clp->cl_sessions)) {
2000                 struct nfsd4_session *ses;
2001                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2002                                 se_perclnt);
2003                 list_del(&ses->se_perclnt);
2004                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2005                 free_session(ses);
2006         }
2007         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2008         if (clp->cl_nfsd_dentry) {
2009                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2010                 clp->cl_nfsd_dentry = NULL;
2011                 wake_up_all(&expiry_wq);
2012         }
2013         drop_client(clp);
2014 }
2015
2016 /* must be called under the client_lock */
2017 static void
2018 unhash_client_locked(struct nfs4_client *clp)
2019 {
2020         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2021         struct nfsd4_session *ses;
2022
2023         lockdep_assert_held(&nn->client_lock);
2024
2025         /* Mark the client as expired! */
2026         clp->cl_time = 0;
2027         /* Make it invisible */
2028         if (!list_empty(&clp->cl_idhash)) {
2029                 list_del_init(&clp->cl_idhash);
2030                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2031                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2032                 else
2033                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2034         }
2035         list_del_init(&clp->cl_lru);
2036         spin_lock(&clp->cl_lock);
2037         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2038                 list_del_init(&ses->se_hash);
2039         spin_unlock(&clp->cl_lock);
2040 }
2041
2042 static void
2043 unhash_client(struct nfs4_client *clp)
2044 {
2045         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2046
2047         spin_lock(&nn->client_lock);
2048         unhash_client_locked(clp);
2049         spin_unlock(&nn->client_lock);
2050 }
2051
2052 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2053 {
2054         if (atomic_read(&clp->cl_rpc_users))
2055                 return nfserr_jukebox;
2056         unhash_client_locked(clp);
2057         return nfs_ok;
2058 }
2059
2060 static void
2061 __destroy_client(struct nfs4_client *clp)
2062 {
2063         int i;
2064         struct nfs4_openowner *oo;
2065         struct nfs4_delegation *dp;
2066         struct list_head reaplist;
2067
2068         INIT_LIST_HEAD(&reaplist);
2069         spin_lock(&state_lock);
2070         while (!list_empty(&clp->cl_delegations)) {
2071                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2072                 WARN_ON(!unhash_delegation_locked(dp));
2073                 list_add(&dp->dl_recall_lru, &reaplist);
2074         }
2075         spin_unlock(&state_lock);
2076         while (!list_empty(&reaplist)) {
2077                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2078                 list_del_init(&dp->dl_recall_lru);
2079                 destroy_unhashed_deleg(dp);
2080         }
2081         while (!list_empty(&clp->cl_revoked)) {
2082                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2083                 list_del_init(&dp->dl_recall_lru);
2084                 nfs4_put_stid(&dp->dl_stid);
2085         }
2086         while (!list_empty(&clp->cl_openowners)) {
2087                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2088                 nfs4_get_stateowner(&oo->oo_owner);
2089                 release_openowner(oo);
2090         }
2091         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2092                 struct nfs4_stateowner *so, *tmp;
2093
2094                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2095                                          so_strhash) {
2096                         /* Should be no openowners at this point */
2097                         WARN_ON_ONCE(so->so_is_open_owner);
2098                         remove_blocked_locks(lockowner(so));
2099                 }
2100         }
2101         nfsd4_return_all_client_layouts(clp);
2102         nfsd4_shutdown_copy(clp);
2103         nfsd4_shutdown_callback(clp);
2104         if (clp->cl_cb_conn.cb_xprt)
2105                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2106         free_client(clp);
2107         wake_up_all(&expiry_wq);
2108 }
2109
2110 static void
2111 destroy_client(struct nfs4_client *clp)
2112 {
2113         unhash_client(clp);
2114         __destroy_client(clp);
2115 }
2116
2117 static void inc_reclaim_complete(struct nfs4_client *clp)
2118 {
2119         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2120
2121         if (!nn->track_reclaim_completes)
2122                 return;
2123         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2124                 return;
2125         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2126                         nn->reclaim_str_hashtbl_size) {
2127                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2128                                 clp->net->ns.inum);
2129                 nfsd4_end_grace(nn);
2130         }
2131 }
2132
2133 static void expire_client(struct nfs4_client *clp)
2134 {
2135         unhash_client(clp);
2136         nfsd4_client_record_remove(clp);
2137         __destroy_client(clp);
2138 }
2139
2140 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2141 {
2142         memcpy(target->cl_verifier.data, source->data,
2143                         sizeof(target->cl_verifier.data));
2144 }
2145
2146 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2147 {
2148         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2149         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2150 }
2151
2152 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2153 {
2154         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2155         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2156                                                                 GFP_KERNEL);
2157         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2158         if ((source->cr_principal && !target->cr_principal) ||
2159             (source->cr_raw_principal && !target->cr_raw_principal) ||
2160             (source->cr_targ_princ && !target->cr_targ_princ))
2161                 return -ENOMEM;
2162
2163         target->cr_flavor = source->cr_flavor;
2164         target->cr_uid = source->cr_uid;
2165         target->cr_gid = source->cr_gid;
2166         target->cr_group_info = source->cr_group_info;
2167         get_group_info(target->cr_group_info);
2168         target->cr_gss_mech = source->cr_gss_mech;
2169         if (source->cr_gss_mech)
2170                 gss_mech_get(source->cr_gss_mech);
2171         return 0;
2172 }
2173
2174 static int
2175 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2176 {
2177         if (o1->len < o2->len)
2178                 return -1;
2179         if (o1->len > o2->len)
2180                 return 1;
2181         return memcmp(o1->data, o2->data, o1->len);
2182 }
2183
2184 static int
2185 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2186 {
2187         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2188 }
2189
2190 static int
2191 same_clid(clientid_t *cl1, clientid_t *cl2)
2192 {
2193         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2194 }
2195
2196 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2197 {
2198         int i;
2199
2200         if (g1->ngroups != g2->ngroups)
2201                 return false;
2202         for (i=0; i<g1->ngroups; i++)
2203                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2204                         return false;
2205         return true;
2206 }
2207
2208 /*
2209  * RFC 3530 language requires clid_inuse be returned when the
2210  * "principal" associated with a requests differs from that previously
2211  * used.  We use uid, gid's, and gss principal string as our best
2212  * approximation.  We also don't want to allow non-gss use of a client
2213  * established using gss: in theory cr_principal should catch that
2214  * change, but in practice cr_principal can be null even in the gss case
2215  * since gssd doesn't always pass down a principal string.
2216  */
2217 static bool is_gss_cred(struct svc_cred *cr)
2218 {
2219         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2220         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2221 }
2222
2223
2224 static bool
2225 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2226 {
2227         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2228                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2229                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2230                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2231                 return false;
2232         /* XXX: check that cr_targ_princ fields match ? */
2233         if (cr1->cr_principal == cr2->cr_principal)
2234                 return true;
2235         if (!cr1->cr_principal || !cr2->cr_principal)
2236                 return false;
2237         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2238 }
2239
2240 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2241 {
2242         struct svc_cred *cr = &rqstp->rq_cred;
2243         u32 service;
2244
2245         if (!cr->cr_gss_mech)
2246                 return false;
2247         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2248         return service == RPC_GSS_SVC_INTEGRITY ||
2249                service == RPC_GSS_SVC_PRIVACY;
2250 }
2251
2252 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2253 {
2254         struct svc_cred *cr = &rqstp->rq_cred;
2255
2256         if (!cl->cl_mach_cred)
2257                 return true;
2258         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2259                 return false;
2260         if (!svc_rqst_integrity_protected(rqstp))
2261                 return false;
2262         if (cl->cl_cred.cr_raw_principal)
2263                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2264                                                 cr->cr_raw_principal);
2265         if (!cr->cr_principal)
2266                 return false;
2267         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2268 }
2269
2270 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2271 {
2272         __be32 verf[2];
2273
2274         /*
2275          * This is opaque to client, so no need to byte-swap. Use
2276          * __force to keep sparse happy
2277          */
2278         verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2279         verf[1] = (__force __be32)nn->clverifier_counter++;
2280         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2281 }
2282
2283 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2284 {
2285         clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2286         clp->cl_clientid.cl_id = nn->clientid_counter++;
2287         gen_confirm(clp, nn);
2288 }
2289
2290 static struct nfs4_stid *
2291 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2292 {
2293         struct nfs4_stid *ret;
2294
2295         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2296         if (!ret || !ret->sc_type)
2297                 return NULL;
2298         return ret;
2299 }
2300
2301 static struct nfs4_stid *
2302 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2303 {
2304         struct nfs4_stid *s;
2305
2306         spin_lock(&cl->cl_lock);
2307         s = find_stateid_locked(cl, t);
2308         if (s != NULL) {
2309                 if (typemask & s->sc_type)
2310                         refcount_inc(&s->sc_count);
2311                 else
2312                         s = NULL;
2313         }
2314         spin_unlock(&cl->cl_lock);
2315         return s;
2316 }
2317
2318 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2319 {
2320         struct nfsdfs_client *nc;
2321         nc = get_nfsdfs_client(inode);
2322         if (!nc)
2323                 return NULL;
2324         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2325 }
2326
2327 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2328 {
2329         seq_printf(m, "\"");
2330         seq_escape_mem_ascii(m, data, len);
2331         seq_printf(m, "\"");
2332 }
2333
2334 static int client_info_show(struct seq_file *m, void *v)
2335 {
2336         struct inode *inode = m->private;
2337         struct nfs4_client *clp;
2338         u64 clid;
2339
2340         clp = get_nfsdfs_clp(inode);
2341         if (!clp)
2342                 return -ENXIO;
2343         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2344         seq_printf(m, "clientid: 0x%llx\n", clid);
2345         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2346         seq_printf(m, "name: ");
2347         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2348         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2349         if (clp->cl_nii_domain.data) {
2350                 seq_printf(m, "Implementation domain: ");
2351                 seq_quote_mem(m, clp->cl_nii_domain.data,
2352                                         clp->cl_nii_domain.len);
2353                 seq_printf(m, "\nImplementation name: ");
2354                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2355                 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2356                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2357         }
2358         drop_client(clp);
2359
2360         return 0;
2361 }
2362
2363 static int client_info_open(struct inode *inode, struct file *file)
2364 {
2365         return single_open(file, client_info_show, inode);
2366 }
2367
2368 static const struct file_operations client_info_fops = {
2369         .open           = client_info_open,
2370         .read           = seq_read,
2371         .llseek         = seq_lseek,
2372         .release        = single_release,
2373 };
2374
2375 static void *states_start(struct seq_file *s, loff_t *pos)
2376         __acquires(&clp->cl_lock)
2377 {
2378         struct nfs4_client *clp = s->private;
2379         unsigned long id = *pos;
2380         void *ret;
2381
2382         spin_lock(&clp->cl_lock);
2383         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2384         *pos = id;
2385         return ret;
2386 }
2387
2388 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2389 {
2390         struct nfs4_client *clp = s->private;
2391         unsigned long id = *pos;
2392         void *ret;
2393
2394         id = *pos;
2395         id++;
2396         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2397         *pos = id;
2398         return ret;
2399 }
2400
2401 static void states_stop(struct seq_file *s, void *v)
2402         __releases(&clp->cl_lock)
2403 {
2404         struct nfs4_client *clp = s->private;
2405
2406         spin_unlock(&clp->cl_lock);
2407 }
2408
2409 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2410 {
2411         struct inode *inode = f->nf_inode;
2412
2413         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2414                                         MAJOR(inode->i_sb->s_dev),
2415                                          MINOR(inode->i_sb->s_dev),
2416                                          inode->i_ino);
2417 }
2418
2419 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2420 {
2421         seq_printf(s, "owner: ");
2422         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2423 }
2424
2425 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2426 {
2427         struct nfs4_ol_stateid *ols;
2428         struct nfs4_file *nf;
2429         struct nfsd_file *file;
2430         struct nfs4_stateowner *oo;
2431         unsigned int access, deny;
2432
2433         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2434                 return 0; /* XXX: or SEQ_SKIP? */
2435         ols = openlockstateid(st);
2436         oo = ols->st_stateowner;
2437         nf = st->sc_file;
2438         file = find_any_file(nf);
2439
2440         seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2441
2442         access = bmap_to_share_mode(ols->st_access_bmap);
2443         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2444
2445         seq_printf(s, "access: %s%s, ",
2446                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2447                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2448         seq_printf(s, "deny: %s%s, ",
2449                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2450                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2451
2452         nfs4_show_superblock(s, file);
2453         seq_printf(s, ", ");
2454         nfs4_show_owner(s, oo);
2455         seq_printf(s, " }\n");
2456         nfsd_file_put(file);
2457
2458         return 0;
2459 }
2460
2461 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2462 {
2463         struct nfs4_ol_stateid *ols;
2464         struct nfs4_file *nf;
2465         struct nfsd_file *file;
2466         struct nfs4_stateowner *oo;
2467
2468         ols = openlockstateid(st);
2469         oo = ols->st_stateowner;
2470         nf = st->sc_file;
2471         file = find_any_file(nf);
2472
2473         seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2474
2475         /*
2476          * Note: a lock stateid isn't really the same thing as a lock,
2477          * it's the locking state held by one owner on a file, and there
2478          * may be multiple (or no) lock ranges associated with it.
2479          * (Same for the matter is true of open stateids.)
2480          */
2481
2482         nfs4_show_superblock(s, file);
2483         /* XXX: open stateid? */
2484         seq_printf(s, ", ");
2485         nfs4_show_owner(s, oo);
2486         seq_printf(s, " }\n");
2487         nfsd_file_put(file);
2488
2489         return 0;
2490 }
2491
2492 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2493 {
2494         struct nfs4_delegation *ds;
2495         struct nfs4_file *nf;
2496         struct nfsd_file *file;
2497
2498         ds = delegstateid(st);
2499         nf = st->sc_file;
2500         file = nf->fi_deleg_file;
2501
2502         seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2503
2504         /* Kinda dead code as long as we only support read delegs: */
2505         seq_printf(s, "access: %s, ",
2506                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2507
2508         /* XXX: lease time, whether it's being recalled. */
2509
2510         nfs4_show_superblock(s, file);
2511         seq_printf(s, " }\n");
2512
2513         return 0;
2514 }
2515
2516 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2517 {
2518         struct nfs4_layout_stateid *ls;
2519         struct nfsd_file *file;
2520
2521         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2522         file = ls->ls_file;
2523
2524         seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2525
2526         /* XXX: What else would be useful? */
2527
2528         nfs4_show_superblock(s, file);
2529         seq_printf(s, " }\n");
2530
2531         return 0;
2532 }
2533
2534 static int states_show(struct seq_file *s, void *v)
2535 {
2536         struct nfs4_stid *st = v;
2537
2538         switch (st->sc_type) {
2539         case NFS4_OPEN_STID:
2540                 return nfs4_show_open(s, st);
2541         case NFS4_LOCK_STID:
2542                 return nfs4_show_lock(s, st);
2543         case NFS4_DELEG_STID:
2544                 return nfs4_show_deleg(s, st);
2545         case NFS4_LAYOUT_STID:
2546                 return nfs4_show_layout(s, st);
2547         default:
2548                 return 0; /* XXX: or SEQ_SKIP? */
2549         }
2550         /* XXX: copy stateids? */
2551 }
2552
2553 static struct seq_operations states_seq_ops = {
2554         .start = states_start,
2555         .next = states_next,
2556         .stop = states_stop,
2557         .show = states_show
2558 };
2559
2560 static int client_states_open(struct inode *inode, struct file *file)
2561 {
2562         struct seq_file *s;
2563         struct nfs4_client *clp;
2564         int ret;
2565
2566         clp = get_nfsdfs_clp(inode);
2567         if (!clp)
2568                 return -ENXIO;
2569
2570         ret = seq_open(file, &states_seq_ops);
2571         if (ret)
2572                 return ret;
2573         s = file->private_data;
2574         s->private = clp;
2575         return 0;
2576 }
2577
2578 static int client_opens_release(struct inode *inode, struct file *file)
2579 {
2580         struct seq_file *m = file->private_data;
2581         struct nfs4_client *clp = m->private;
2582
2583         /* XXX: alternatively, we could get/drop in seq start/stop */
2584         drop_client(clp);
2585         return 0;
2586 }
2587
2588 static const struct file_operations client_states_fops = {
2589         .open           = client_states_open,
2590         .read           = seq_read,
2591         .llseek         = seq_lseek,
2592         .release        = client_opens_release,
2593 };
2594
2595 /*
2596  * Normally we refuse to destroy clients that are in use, but here the
2597  * administrator is telling us to just do it.  We also want to wait
2598  * so the caller has a guarantee that the client's locks are gone by
2599  * the time the write returns:
2600  */
2601 static void force_expire_client(struct nfs4_client *clp)
2602 {
2603         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2604         bool already_expired;
2605
2606         spin_lock(&clp->cl_lock);
2607         clp->cl_time = 0;
2608         spin_unlock(&clp->cl_lock);
2609
2610         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2611         spin_lock(&nn->client_lock);
2612         already_expired = list_empty(&clp->cl_lru);
2613         if (!already_expired)
2614                 unhash_client_locked(clp);
2615         spin_unlock(&nn->client_lock);
2616
2617         if (!already_expired)
2618                 expire_client(clp);
2619         else
2620                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2621 }
2622
2623 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2624                                    size_t size, loff_t *pos)
2625 {
2626         char *data;
2627         struct nfs4_client *clp;
2628
2629         data = simple_transaction_get(file, buf, size);
2630         if (IS_ERR(data))
2631                 return PTR_ERR(data);
2632         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2633                 return -EINVAL;
2634         clp = get_nfsdfs_clp(file_inode(file));
2635         if (!clp)
2636                 return -ENXIO;
2637         force_expire_client(clp);
2638         drop_client(clp);
2639         return 7;
2640 }
2641
2642 static const struct file_operations client_ctl_fops = {
2643         .write          = client_ctl_write,
2644         .release        = simple_transaction_release,
2645 };
2646
2647 static const struct tree_descr client_files[] = {
2648         [0] = {"info", &client_info_fops, S_IRUSR},
2649         [1] = {"states", &client_states_fops, S_IRUSR},
2650         [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2651         [3] = {""},
2652 };
2653
2654 static struct nfs4_client *create_client(struct xdr_netobj name,
2655                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2656 {
2657         struct nfs4_client *clp;
2658         struct sockaddr *sa = svc_addr(rqstp);
2659         int ret;
2660         struct net *net = SVC_NET(rqstp);
2661         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2662
2663         clp = alloc_client(name);
2664         if (clp == NULL)
2665                 return NULL;
2666
2667         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2668         if (ret) {
2669                 free_client(clp);
2670                 return NULL;
2671         }
2672         gen_clid(clp, nn);
2673         kref_init(&clp->cl_nfsdfs.cl_ref);
2674         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2675         clp->cl_time = ktime_get_boottime_seconds();
2676         clear_bit(0, &clp->cl_cb_slot_busy);
2677         copy_verf(clp, verf);
2678         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2679         clp->cl_cb_session = NULL;
2680         clp->net = net;
2681         clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2682                         clp->cl_clientid.cl_id - nn->clientid_base,
2683                         client_files);
2684         if (!clp->cl_nfsd_dentry) {
2685                 free_client(clp);
2686                 return NULL;
2687         }
2688         return clp;
2689 }
2690
2691 static void
2692 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2693 {
2694         struct rb_node **new = &(root->rb_node), *parent = NULL;
2695         struct nfs4_client *clp;
2696
2697         while (*new) {
2698                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2699                 parent = *new;
2700
2701                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2702                         new = &((*new)->rb_left);
2703                 else
2704                         new = &((*new)->rb_right);
2705         }
2706
2707         rb_link_node(&new_clp->cl_namenode, parent, new);
2708         rb_insert_color(&new_clp->cl_namenode, root);
2709 }
2710
2711 static struct nfs4_client *
2712 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2713 {
2714         int cmp;
2715         struct rb_node *node = root->rb_node;
2716         struct nfs4_client *clp;
2717
2718         while (node) {
2719                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2720                 cmp = compare_blob(&clp->cl_name, name);
2721                 if (cmp > 0)
2722                         node = node->rb_left;
2723                 else if (cmp < 0)
2724                         node = node->rb_right;
2725                 else
2726                         return clp;
2727         }
2728         return NULL;
2729 }
2730
2731 static void
2732 add_to_unconfirmed(struct nfs4_client *clp)
2733 {
2734         unsigned int idhashval;
2735         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2736
2737         lockdep_assert_held(&nn->client_lock);
2738
2739         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2740         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2741         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2742         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2743         renew_client_locked(clp);
2744 }
2745
2746 static void
2747 move_to_confirmed(struct nfs4_client *clp)
2748 {
2749         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2750         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2751
2752         lockdep_assert_held(&nn->client_lock);
2753
2754         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2755         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2756         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2757         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2758         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2759         renew_client_locked(clp);
2760 }
2761
2762 static struct nfs4_client *
2763 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2764 {
2765         struct nfs4_client *clp;
2766         unsigned int idhashval = clientid_hashval(clid->cl_id);
2767
2768         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2769                 if (same_clid(&clp->cl_clientid, clid)) {
2770                         if ((bool)clp->cl_minorversion != sessions)
2771                                 return NULL;
2772                         renew_client_locked(clp);
2773                         return clp;
2774                 }
2775         }
2776         return NULL;
2777 }
2778
2779 static struct nfs4_client *
2780 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2781 {
2782         struct list_head *tbl = nn->conf_id_hashtbl;
2783
2784         lockdep_assert_held(&nn->client_lock);
2785         return find_client_in_id_table(tbl, clid, sessions);
2786 }
2787
2788 static struct nfs4_client *
2789 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2790 {
2791         struct list_head *tbl = nn->unconf_id_hashtbl;
2792
2793         lockdep_assert_held(&nn->client_lock);
2794         return find_client_in_id_table(tbl, clid, sessions);
2795 }
2796
2797 static bool clp_used_exchangeid(struct nfs4_client *clp)
2798 {
2799         return clp->cl_exchange_flags != 0;
2800 }
2801
2802 static struct nfs4_client *
2803 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2804 {
2805         lockdep_assert_held(&nn->client_lock);
2806         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2807 }
2808
2809 static struct nfs4_client *
2810 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2811 {
2812         lockdep_assert_held(&nn->client_lock);
2813         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2814 }
2815
2816 static void
2817 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2818 {
2819         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2820         struct sockaddr *sa = svc_addr(rqstp);
2821         u32 scopeid = rpc_get_scope_id(sa);
2822         unsigned short expected_family;
2823
2824         /* Currently, we only support tcp and tcp6 for the callback channel */
2825         if (se->se_callback_netid_len == 3 &&
2826             !memcmp(se->se_callback_netid_val, "tcp", 3))
2827                 expected_family = AF_INET;
2828         else if (se->se_callback_netid_len == 4 &&
2829                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2830                 expected_family = AF_INET6;
2831         else
2832                 goto out_err;
2833
2834         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2835                                             se->se_callback_addr_len,
2836                                             (struct sockaddr *)&conn->cb_addr,
2837                                             sizeof(conn->cb_addr));
2838
2839         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2840                 goto out_err;
2841
2842         if (conn->cb_addr.ss_family == AF_INET6)
2843                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2844
2845         conn->cb_prog = se->se_callback_prog;
2846         conn->cb_ident = se->se_callback_ident;
2847         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2848         return;
2849 out_err:
2850         conn->cb_addr.ss_family = AF_UNSPEC;
2851         conn->cb_addrlen = 0;
2852         dprintk("NFSD: this client (clientid %08x/%08x) "
2853                 "will not receive delegations\n",
2854                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2855
2856         return;
2857 }
2858
2859 /*
2860  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2861  */
2862 static void
2863 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2864 {
2865         struct xdr_buf *buf = resp->xdr.buf;
2866         struct nfsd4_slot *slot = resp->cstate.slot;
2867         unsigned int base;
2868
2869         dprintk("--> %s slot %p\n", __func__, slot);
2870
2871         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2872         slot->sl_opcnt = resp->opcnt;
2873         slot->sl_status = resp->cstate.status;
2874         free_svc_cred(&slot->sl_cred);
2875         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2876
2877         if (!nfsd4_cache_this(resp)) {
2878                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2879                 return;
2880         }
2881         slot->sl_flags |= NFSD4_SLOT_CACHED;
2882
2883         base = resp->cstate.data_offset;
2884         slot->sl_datalen = buf->len - base;
2885         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2886                 WARN(1, "%s: sessions DRC could not cache compound\n",
2887                      __func__);
2888         return;
2889 }
2890
2891 /*
2892  * Encode the replay sequence operation from the slot values.
2893  * If cachethis is FALSE encode the uncached rep error on the next
2894  * operation which sets resp->p and increments resp->opcnt for
2895  * nfs4svc_encode_compoundres.
2896  *
2897  */
2898 static __be32
2899 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2900                           struct nfsd4_compoundres *resp)
2901 {
2902         struct nfsd4_op *op;
2903         struct nfsd4_slot *slot = resp->cstate.slot;
2904
2905         /* Encode the replayed sequence operation */
2906         op = &args->ops[resp->opcnt - 1];
2907         nfsd4_encode_operation(resp, op);
2908
2909         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2910                 return op->status;
2911         if (args->opcnt == 1) {
2912                 /*
2913                  * The original operation wasn't a solo sequence--we
2914                  * always cache those--so this retry must not match the
2915                  * original:
2916                  */
2917                 op->status = nfserr_seq_false_retry;
2918         } else {
2919                 op = &args->ops[resp->opcnt++];
2920                 op->status = nfserr_retry_uncached_rep;
2921                 nfsd4_encode_operation(resp, op);
2922         }
2923         return op->status;
2924 }
2925
2926 /*
2927  * The sequence operation is not cached because we can use the slot and
2928  * session values.
2929  */
2930 static __be32
2931 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2932                          struct nfsd4_sequence *seq)
2933 {
2934         struct nfsd4_slot *slot = resp->cstate.slot;
2935         struct xdr_stream *xdr = &resp->xdr;
2936         __be32 *p;
2937         __be32 status;
2938
2939         dprintk("--> %s slot %p\n", __func__, slot);
2940
2941         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2942         if (status)
2943                 return status;
2944
2945         p = xdr_reserve_space(xdr, slot->sl_datalen);
2946         if (!p) {
2947                 WARN_ON_ONCE(1);
2948                 return nfserr_serverfault;
2949         }
2950         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2951         xdr_commit_encode(xdr);
2952
2953         resp->opcnt = slot->sl_opcnt;
2954         return slot->sl_status;
2955 }
2956
2957 /*
2958  * Set the exchange_id flags returned by the server.
2959  */
2960 static void
2961 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2962 {
2963 #ifdef CONFIG_NFSD_PNFS
2964         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2965 #else
2966         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2967 #endif
2968
2969         /* Referrals are supported, Migration is not. */
2970         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2971
2972         /* set the wire flags to return to client. */
2973         clid->flags = new->cl_exchange_flags;
2974 }
2975
2976 static bool client_has_openowners(struct nfs4_client *clp)
2977 {
2978         struct nfs4_openowner *oo;
2979
2980         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2981                 if (!list_empty(&oo->oo_owner.so_stateids))
2982                         return true;
2983         }
2984         return false;
2985 }
2986
2987 static bool client_has_state(struct nfs4_client *clp)
2988 {
2989         return client_has_openowners(clp)
2990 #ifdef CONFIG_NFSD_PNFS
2991                 || !list_empty(&clp->cl_lo_states)
2992 #endif
2993                 || !list_empty(&clp->cl_delegations)
2994                 || !list_empty(&clp->cl_sessions)
2995                 || !list_empty(&clp->async_copies);
2996 }
2997
2998 static __be32 copy_impl_id(struct nfs4_client *clp,
2999                                 struct nfsd4_exchange_id *exid)
3000 {
3001         if (!exid->nii_domain.data)
3002                 return 0;
3003         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3004         if (!clp->cl_nii_domain.data)
3005                 return nfserr_jukebox;
3006         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3007         if (!clp->cl_nii_name.data)
3008                 return nfserr_jukebox;
3009         clp->cl_nii_time = exid->nii_time;
3010         return 0;
3011 }
3012
3013 __be32
3014 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3015                 union nfsd4_op_u *u)
3016 {
3017         struct nfsd4_exchange_id *exid = &u->exchange_id;
3018         struct nfs4_client *conf, *new;
3019         struct nfs4_client *unconf = NULL;
3020         __be32 status;
3021         char                    addr_str[INET6_ADDRSTRLEN];
3022         nfs4_verifier           verf = exid->verifier;
3023         struct sockaddr         *sa = svc_addr(rqstp);
3024         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3025         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3026
3027         rpc_ntop(sa, addr_str, sizeof(addr_str));
3028         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3029                 "ip_addr=%s flags %x, spa_how %d\n",
3030                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3031                 addr_str, exid->flags, exid->spa_how);
3032
3033         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3034                 return nfserr_inval;
3035
3036         new = create_client(exid->clname, rqstp, &verf);
3037         if (new == NULL)
3038                 return nfserr_jukebox;
3039         status = copy_impl_id(new, exid);
3040         if (status)
3041                 goto out_nolock;
3042
3043         switch (exid->spa_how) {
3044         case SP4_MACH_CRED:
3045                 exid->spo_must_enforce[0] = 0;
3046                 exid->spo_must_enforce[1] = (
3047                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
3048                         1 << (OP_EXCHANGE_ID - 32) |
3049                         1 << (OP_CREATE_SESSION - 32) |
3050                         1 << (OP_DESTROY_SESSION - 32) |
3051                         1 << (OP_DESTROY_CLIENTID - 32));
3052
3053                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3054                                         1 << (OP_OPEN_DOWNGRADE) |
3055                                         1 << (OP_LOCKU) |
3056                                         1 << (OP_DELEGRETURN));
3057
3058                 exid->spo_must_allow[1] &= (
3059                                         1 << (OP_TEST_STATEID - 32) |
3060                                         1 << (OP_FREE_STATEID - 32));
3061                 if (!svc_rqst_integrity_protected(rqstp)) {
3062                         status = nfserr_inval;
3063                         goto out_nolock;
3064                 }
3065                 /*
3066                  * Sometimes userspace doesn't give us a principal.
3067                  * Which is a bug, really.  Anyway, we can't enforce
3068                  * MACH_CRED in that case, better to give up now:
3069                  */
3070                 if (!new->cl_cred.cr_principal &&
3071                                         !new->cl_cred.cr_raw_principal) {
3072                         status = nfserr_serverfault;
3073                         goto out_nolock;
3074                 }
3075                 new->cl_mach_cred = true;
3076         case SP4_NONE:
3077                 break;
3078         default:                                /* checked by xdr code */
3079                 WARN_ON_ONCE(1);
3080                 /* fall through */
3081         case SP4_SSV:
3082                 status = nfserr_encr_alg_unsupp;
3083                 goto out_nolock;
3084         }
3085
3086         /* Cases below refer to rfc 5661 section 18.35.4: */
3087         spin_lock(&nn->client_lock);
3088         conf = find_confirmed_client_by_name(&exid->clname, nn);
3089         if (conf) {
3090                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3091                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3092
3093                 if (update) {
3094                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3095                                 status = nfserr_inval;
3096                                 goto out;
3097                         }
3098                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3099                                 status = nfserr_wrong_cred;
3100                                 goto out;
3101                         }
3102                         if (!creds_match) { /* case 9 */
3103                                 status = nfserr_perm;
3104                                 goto out;
3105                         }
3106                         if (!verfs_match) { /* case 8 */
3107                                 status = nfserr_not_same;
3108                                 goto out;
3109                         }
3110                         /* case 6 */
3111                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3112                         goto out_copy;
3113                 }
3114                 if (!creds_match) { /* case 3 */
3115                         if (client_has_state(conf)) {
3116                                 status = nfserr_clid_inuse;
3117                                 goto out;
3118                         }
3119                         goto out_new;
3120                 }
3121                 if (verfs_match) { /* case 2 */
3122                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3123                         goto out_copy;
3124                 }
3125                 /* case 5, client reboot */
3126                 conf = NULL;
3127                 goto out_new;
3128         }
3129
3130         if (update) { /* case 7 */
3131                 status = nfserr_noent;
3132                 goto out;
3133         }
3134
3135         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
3136         if (unconf) /* case 4, possible retry or client restart */
3137                 unhash_client_locked(unconf);
3138
3139         /* case 1 (normal case) */
3140 out_new:
3141         if (conf) {
3142                 status = mark_client_expired_locked(conf);
3143                 if (status)
3144                         goto out;
3145         }
3146         new->cl_minorversion = cstate->minorversion;
3147         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3148         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3149
3150         add_to_unconfirmed(new);
3151         swap(new, conf);
3152 out_copy:
3153         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3154         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3155
3156         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3157         nfsd4_set_ex_flags(conf, exid);
3158
3159         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3160                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3161         status = nfs_ok;
3162
3163 out:
3164         spin_unlock(&nn->client_lock);
3165 out_nolock:
3166         if (new)
3167                 expire_client(new);
3168         if (unconf)
3169                 expire_client(unconf);
3170         return status;
3171 }
3172
3173 static __be32
3174 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3175 {
3176         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3177                 slot_seqid);
3178
3179         /* The slot is in use, and no response has been sent. */
3180         if (slot_inuse) {
3181                 if (seqid == slot_seqid)
3182                         return nfserr_jukebox;
3183                 else
3184                         return nfserr_seq_misordered;
3185         }
3186         /* Note unsigned 32-bit arithmetic handles wraparound: */
3187         if (likely(seqid == slot_seqid + 1))
3188                 return nfs_ok;
3189         if (seqid == slot_seqid)
3190                 return nfserr_replay_cache;
3191         return nfserr_seq_misordered;
3192 }
3193
3194 /*
3195  * Cache the create session result into the create session single DRC
3196  * slot cache by saving the xdr structure. sl_seqid has been set.
3197  * Do this for solo or embedded create session operations.
3198  */
3199 static void
3200 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3201                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3202 {
3203         slot->sl_status = nfserr;
3204         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3205 }
3206
3207 static __be32
3208 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3209                             struct nfsd4_clid_slot *slot)
3210 {
3211         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3212         return slot->sl_status;
3213 }
3214
3215 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3216                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3217                         1 +     /* MIN tag is length with zero, only length */ \
3218                         3 +     /* version, opcount, opcode */ \
3219                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3220                                 /* seqid, slotID, slotID, cache */ \
3221                         4 ) * sizeof(__be32))
3222
3223 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3224                         2 +     /* verifier: AUTH_NULL, length 0 */\
3225                         1 +     /* status */ \
3226                         1 +     /* MIN tag is length with zero, only length */ \
3227                         3 +     /* opcount, opcode, opstatus*/ \
3228                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3229                                 /* seqid, slotID, slotID, slotID, status */ \
3230                         5 ) * sizeof(__be32))
3231
3232 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3233 {
3234         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3235
3236         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3237                 return nfserr_toosmall;
3238         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3239                 return nfserr_toosmall;
3240         ca->headerpadsz = 0;
3241         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3242         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3243         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3244         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3245                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3246         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3247         /*
3248          * Note decreasing slot size below client's request may make it
3249          * difficult for client to function correctly, whereas
3250          * decreasing the number of slots will (just?) affect
3251          * performance.  When short on memory we therefore prefer to
3252          * decrease number of slots instead of their size.  Clients that
3253          * request larger slots than they need will get poor results:
3254          * Note that we always allow at least one slot, because our
3255          * accounting is soft and provides no guarantees either way.
3256          */
3257         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3258
3259         return nfs_ok;
3260 }
3261
3262 /*
3263  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3264  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3265  */
3266 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3267         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3268
3269 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3270         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3271
3272 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3273                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3274 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3275                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3276                                  sizeof(__be32))
3277
3278 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3279 {
3280         ca->headerpadsz = 0;
3281
3282         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3283                 return nfserr_toosmall;
3284         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3285                 return nfserr_toosmall;
3286         ca->maxresp_cached = 0;
3287         if (ca->maxops < 2)
3288                 return nfserr_toosmall;
3289
3290         return nfs_ok;
3291 }
3292
3293 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3294 {
3295         switch (cbs->flavor) {
3296         case RPC_AUTH_NULL:
3297         case RPC_AUTH_UNIX:
3298                 return nfs_ok;
3299         default:
3300                 /*
3301                  * GSS case: the spec doesn't allow us to return this
3302                  * error.  But it also doesn't allow us not to support
3303                  * GSS.
3304                  * I'd rather this fail hard than return some error the
3305                  * client might think it can already handle:
3306                  */
3307                 return nfserr_encr_alg_unsupp;
3308         }
3309 }
3310
3311 __be32
3312 nfsd4_create_session(struct svc_rqst *rqstp,
3313                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3314 {
3315         struct nfsd4_create_session *cr_ses = &u->create_session;
3316         struct sockaddr *sa = svc_addr(rqstp);
3317         struct nfs4_client *conf, *unconf;
3318         struct nfs4_client *old = NULL;
3319         struct nfsd4_session *new;
3320         struct nfsd4_conn *conn;
3321         struct nfsd4_clid_slot *cs_slot = NULL;
3322         __be32 status = 0;
3323         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3324
3325         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3326                 return nfserr_inval;
3327         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3328         if (status)
3329                 return status;
3330         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3331         if (status)
3332                 return status;
3333         status = check_backchannel_attrs(&cr_ses->back_channel);
3334         if (status)
3335                 goto out_release_drc_mem;
3336         status = nfserr_jukebox;
3337         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3338         if (!new)
3339                 goto out_release_drc_mem;
3340         conn = alloc_conn_from_crses(rqstp, cr_ses);
3341         if (!conn)
3342                 goto out_free_session;
3343
3344         spin_lock(&nn->client_lock);
3345         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3346         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3347         WARN_ON_ONCE(conf && unconf);
3348
3349         if (conf) {
3350                 status = nfserr_wrong_cred;
3351                 if (!nfsd4_mach_creds_match(conf, rqstp))
3352                         goto out_free_conn;
3353                 cs_slot = &conf->cl_cs_slot;
3354                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3355                 if (status) {
3356                         if (status == nfserr_replay_cache)
3357                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3358                         goto out_free_conn;
3359                 }
3360         } else if (unconf) {
3361                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3362                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3363                         status = nfserr_clid_inuse;
3364                         goto out_free_conn;
3365                 }
3366                 status = nfserr_wrong_cred;
3367                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3368                         goto out_free_conn;
3369                 cs_slot = &unconf->cl_cs_slot;
3370                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3371                 if (status) {
3372                         /* an unconfirmed replay returns misordered */
3373                         status = nfserr_seq_misordered;
3374                         goto out_free_conn;
3375                 }
3376                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3377                 if (old) {
3378                         status = mark_client_expired_locked(old);
3379                         if (status) {
3380                                 old = NULL;
3381                                 goto out_free_conn;
3382                         }
3383                 }
3384                 move_to_confirmed(unconf);
3385                 conf = unconf;
3386         } else {
3387                 status = nfserr_stale_clientid;
3388                 goto out_free_conn;
3389         }
3390         status = nfs_ok;
3391         /* Persistent sessions are not supported */
3392         cr_ses->flags &= ~SESSION4_PERSIST;
3393         /* Upshifting from TCP to RDMA is not supported */
3394         cr_ses->flags &= ~SESSION4_RDMA;
3395
3396         init_session(rqstp, new, conf, cr_ses);
3397         nfsd4_get_session_locked(new);
3398
3399         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3400                NFS4_MAX_SESSIONID_LEN);
3401         cs_slot->sl_seqid++;
3402         cr_ses->seqid = cs_slot->sl_seqid;
3403
3404         /* cache solo and embedded create sessions under the client_lock */
3405         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3406         spin_unlock(&nn->client_lock);
3407         /* init connection and backchannel */
3408         nfsd4_init_conn(rqstp, conn, new);
3409         nfsd4_put_session(new);
3410         if (old)
3411                 expire_client(old);
3412         return status;
3413 out_free_conn:
3414         spin_unlock(&nn->client_lock);
3415         free_conn(conn);
3416         if (old)
3417                 expire_client(old);
3418 out_free_session:
3419         __free_session(new);
3420 out_release_drc_mem:
3421         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3422         return status;
3423 }
3424
3425 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3426 {
3427         switch (*dir) {
3428         case NFS4_CDFC4_FORE:
3429         case NFS4_CDFC4_BACK:
3430                 return nfs_ok;
3431         case NFS4_CDFC4_FORE_OR_BOTH:
3432         case NFS4_CDFC4_BACK_OR_BOTH:
3433                 *dir = NFS4_CDFC4_BOTH;
3434                 return nfs_ok;
3435         }
3436         return nfserr_inval;
3437 }
3438
3439 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3440                 struct nfsd4_compound_state *cstate,
3441                 union nfsd4_op_u *u)
3442 {
3443         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3444         struct nfsd4_session *session = cstate->session;
3445         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3446         __be32 status;
3447
3448         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3449         if (status)
3450                 return status;
3451         spin_lock(&nn->client_lock);
3452         session->se_cb_prog = bc->bc_cb_program;
3453         session->se_cb_sec = bc->bc_cb_sec;
3454         spin_unlock(&nn->client_lock);
3455
3456         nfsd4_probe_callback(session->se_client);
3457
3458         return nfs_ok;
3459 }
3460
3461 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3462                      struct nfsd4_compound_state *cstate,
3463                      union nfsd4_op_u *u)
3464 {
3465         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3466         __be32 status;
3467         struct nfsd4_conn *conn;
3468         struct nfsd4_session *session;
3469         struct net *net = SVC_NET(rqstp);
3470         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3471
3472         if (!nfsd4_last_compound_op(rqstp))
3473                 return nfserr_not_only_op;
3474         spin_lock(&nn->client_lock);
3475         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3476         spin_unlock(&nn->client_lock);
3477         if (!session)
3478                 goto out_no_session;
3479         status = nfserr_wrong_cred;
3480         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3481                 goto out;
3482         status = nfsd4_map_bcts_dir(&bcts->dir);
3483         if (status)
3484                 goto out;
3485         conn = alloc_conn(rqstp, bcts->dir);
3486         status = nfserr_jukebox;
3487         if (!conn)
3488                 goto out;
3489         nfsd4_init_conn(rqstp, conn, session);
3490         status = nfs_ok;
3491 out:
3492         nfsd4_put_session(session);
3493 out_no_session:
3494         return status;
3495 }
3496
3497 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3498 {
3499         if (!cstate->session)
3500                 return false;
3501         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3502 }
3503
3504 __be32
3505 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3506                 union nfsd4_op_u *u)
3507 {
3508         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3509         struct nfsd4_session *ses;
3510         __be32 status;
3511         int ref_held_by_me = 0;
3512         struct net *net = SVC_NET(r);
3513         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3514
3515         status = nfserr_not_only_op;
3516         if (nfsd4_compound_in_session(cstate, sessionid)) {
3517                 if (!nfsd4_last_compound_op(r))
3518                         goto out;
3519                 ref_held_by_me++;
3520         }
3521         dump_sessionid(__func__, sessionid);
3522         spin_lock(&nn->client_lock);
3523         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3524         if (!ses)
3525                 goto out_client_lock;
3526         status = nfserr_wrong_cred;
3527         if (!nfsd4_mach_creds_match(ses->se_client, r))
3528                 goto out_put_session;
3529         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3530         if (status)
3531                 goto out_put_session;
3532         unhash_session(ses);
3533         spin_unlock(&nn->client_lock);
3534
3535         nfsd4_probe_callback_sync(ses->se_client);
3536
3537         spin_lock(&nn->client_lock);
3538         status = nfs_ok;
3539 out_put_session:
3540         nfsd4_put_session_locked(ses);
3541 out_client_lock:
3542         spin_unlock(&nn->client_lock);
3543 out:
3544         return status;
3545 }
3546
3547 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3548 {
3549         struct nfsd4_conn *c;
3550
3551         list_for_each_entry(c, &s->se_conns, cn_persession) {
3552                 if (c->cn_xprt == xpt) {
3553                         return c;
3554                 }
3555         }
3556         return NULL;
3557 }
3558
3559 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3560 {
3561         struct nfs4_client *clp = ses->se_client;
3562         struct nfsd4_conn *c;
3563         __be32 status = nfs_ok;
3564         int ret;
3565
3566         spin_lock(&clp->cl_lock);
3567         c = __nfsd4_find_conn(new->cn_xprt, ses);
3568         if (c)
3569                 goto out_free;
3570         status = nfserr_conn_not_bound_to_session;
3571         if (clp->cl_mach_cred)
3572                 goto out_free;
3573         __nfsd4_hash_conn(new, ses);
3574         spin_unlock(&clp->cl_lock);
3575         ret = nfsd4_register_conn(new);
3576         if (ret)
3577                 /* oops; xprt is already down: */
3578                 nfsd4_conn_lost(&new->cn_xpt_user);
3579         return nfs_ok;
3580 out_free:
3581         spin_unlock(&clp->cl_lock);
3582         free_conn(new);
3583         return status;
3584 }
3585
3586 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3587 {
3588         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3589
3590         return args->opcnt > session->se_fchannel.maxops;
3591 }
3592
3593 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3594                                   struct nfsd4_session *session)
3595 {
3596         struct xdr_buf *xb = &rqstp->rq_arg;
3597
3598         return xb->len > session->se_fchannel.maxreq_sz;
3599 }
3600
3601 static bool replay_matches_cache(struct svc_rqst *rqstp,
3602                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3603 {
3604         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3605
3606         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3607             (bool)seq->cachethis)
3608                 return false;
3609         /*
3610          * If there's an error then the reply can have fewer ops than
3611          * the call.
3612          */
3613         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3614                 return false;
3615         /*
3616          * But if we cached a reply with *more* ops than the call you're
3617          * sending us now, then this new call is clearly not really a
3618          * replay of the old one:
3619          */
3620         if (slot->sl_opcnt > argp->opcnt)
3621                 return false;
3622         /* This is the only check explicitly called by spec: */
3623         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3624                 return false;
3625         /*
3626          * There may be more comparisons we could actually do, but the
3627          * spec doesn't require us to catch every case where the calls
3628          * don't match (that would require caching the call as well as
3629          * the reply), so we don't bother.
3630          */
3631         return true;
3632 }
3633
3634 __be32
3635 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3636                 union nfsd4_op_u *u)
3637 {
3638         struct nfsd4_sequence *seq = &u->sequence;
3639         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3640         struct xdr_stream *xdr = &resp->xdr;
3641         struct nfsd4_session *session;
3642         struct nfs4_client *clp;
3643         struct nfsd4_slot *slot;
3644         struct nfsd4_conn *conn;
3645         __be32 status;
3646         int buflen;
3647         struct net *net = SVC_NET(rqstp);
3648         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3649
3650         if (resp->opcnt != 1)
3651                 return nfserr_sequence_pos;
3652
3653         /*
3654          * Will be either used or freed by nfsd4_sequence_check_conn
3655          * below.
3656          */
3657         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3658         if (!conn)
3659                 return nfserr_jukebox;
3660
3661         spin_lock(&nn->client_lock);
3662         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3663         if (!session)
3664                 goto out_no_session;
3665         clp = session->se_client;
3666
3667         status = nfserr_too_many_ops;
3668         if (nfsd4_session_too_many_ops(rqstp, session))
3669                 goto out_put_session;
3670
3671         status = nfserr_req_too_big;
3672         if (nfsd4_request_too_big(rqstp, session))
3673                 goto out_put_session;
3674
3675         status = nfserr_badslot;
3676         if (seq->slotid >= session->se_fchannel.maxreqs)
3677                 goto out_put_session;
3678
3679         slot = session->se_slots[seq->slotid];
3680         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3681
3682         /* We do not negotiate the number of slots yet, so set the
3683          * maxslots to the session maxreqs which is used to encode
3684          * sr_highest_slotid and the sr_target_slot id to maxslots */
3685         seq->maxslots = session->se_fchannel.maxreqs;
3686
3687         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3688                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3689         if (status == nfserr_replay_cache) {
3690                 status = nfserr_seq_misordered;
3691                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3692                         goto out_put_session;
3693                 status = nfserr_seq_false_retry;
3694                 if (!replay_matches_cache(rqstp, seq, slot))
3695                         goto out_put_session;
3696                 cstate->slot = slot;
3697                 cstate->session = session;
3698                 cstate->clp = clp;
3699                 /* Return the cached reply status and set cstate->status
3700                  * for nfsd4_proc_compound processing */
3701                 status = nfsd4_replay_cache_entry(resp, seq);
3702                 cstate->status = nfserr_replay_cache;
3703                 goto out;
3704         }
3705         if (status)
3706                 goto out_put_session;
3707
3708         status = nfsd4_sequence_check_conn(conn, session);
3709         conn = NULL;
3710         if (status)
3711                 goto out_put_session;
3712
3713         buflen = (seq->cachethis) ?
3714                         session->se_fchannel.maxresp_cached :
3715                         session->se_fchannel.maxresp_sz;
3716         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3717                                     nfserr_rep_too_big;
3718         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3719                 goto out_put_session;
3720         svc_reserve(rqstp, buflen);
3721
3722         status = nfs_ok;
3723         /* Success! bump slot seqid */
3724         slot->sl_seqid = seq->seqid;
3725         slot->sl_flags |= NFSD4_SLOT_INUSE;
3726         if (seq->cachethis)
3727                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3728         else
3729                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3730
3731         cstate->slot = slot;
3732         cstate->session = session;
3733         cstate->clp = clp;
3734
3735 out:
3736         switch (clp->cl_cb_state) {
3737         case NFSD4_CB_DOWN:
3738                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3739                 break;
3740         case NFSD4_CB_FAULT:
3741                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3742                 break;
3743         default:
3744                 seq->status_flags = 0;
3745         }
3746         if (!list_empty(&clp->cl_revoked))
3747                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3748 out_no_session:
3749         if (conn)
3750                 free_conn(conn);
3751         spin_unlock(&nn->client_lock);
3752         return status;
3753 out_put_session:
3754         nfsd4_put_session_locked(session);
3755         goto out_no_session;
3756 }
3757
3758 void
3759 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3760 {
3761         struct nfsd4_compound_state *cs = &resp->cstate;
3762
3763         if (nfsd4_has_session(cs)) {
3764                 if (cs->status != nfserr_replay_cache) {
3765                         nfsd4_store_cache_entry(resp);
3766                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3767                 }
3768                 /* Drop session reference that was taken in nfsd4_sequence() */
3769                 nfsd4_put_session(cs->session);
3770         } else if (cs->clp)
3771                 put_client_renew(cs->clp);
3772 }
3773
3774 __be32
3775 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3776                 struct nfsd4_compound_state *cstate,
3777                 union nfsd4_op_u *u)
3778 {
3779         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3780         struct nfs4_client *conf, *unconf;
3781         struct nfs4_client *clp = NULL;
3782         __be32 status = 0;
3783         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3784
3785         spin_lock(&nn->client_lock);
3786         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3787         conf = find_confirmed_client(&dc->clientid, true, nn);
3788         WARN_ON_ONCE(conf && unconf);
3789
3790         if (conf) {
3791                 if (client_has_state(conf)) {
3792                         status = nfserr_clientid_busy;
3793                         goto out;
3794                 }
3795                 status = mark_client_expired_locked(conf);
3796                 if (status)
3797                         goto out;
3798                 clp = conf;
3799         } else if (unconf)
3800                 clp = unconf;
3801         else {
3802                 status = nfserr_stale_clientid;
3803                 goto out;
3804         }
3805         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3806                 clp = NULL;
3807                 status = nfserr_wrong_cred;
3808                 goto out;
3809         }
3810         unhash_client_locked(clp);
3811 out:
3812         spin_unlock(&nn->client_lock);
3813         if (clp)
3814                 expire_client(clp);
3815         return status;
3816 }
3817
3818 __be32
3819 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3820                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3821 {
3822         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3823         __be32 status = 0;
3824
3825         if (rc->rca_one_fs) {
3826                 if (!cstate->current_fh.fh_dentry)
3827                         return nfserr_nofilehandle;
3828                 /*
3829                  * We don't take advantage of the rca_one_fs case.
3830                  * That's OK, it's optional, we can safely ignore it.
3831                  */
3832                 return nfs_ok;
3833         }
3834
3835         status = nfserr_complete_already;
3836         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3837                              &cstate->session->se_client->cl_flags))
3838                 goto out;
3839
3840         status = nfserr_stale_clientid;
3841         if (is_client_expired(cstate->session->se_client))
3842                 /*
3843                  * The following error isn't really legal.
3844                  * But we only get here if the client just explicitly
3845                  * destroyed the client.  Surely it no longer cares what
3846                  * error it gets back on an operation for the dead
3847                  * client.
3848                  */
3849                 goto out;
3850
3851         status = nfs_ok;
3852         nfsd4_client_record_create(cstate->session->se_client);
3853         inc_reclaim_complete(cstate->session->se_client);
3854 out:
3855         return status;
3856 }
3857
3858 __be32
3859 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3860                   union nfsd4_op_u *u)
3861 {
3862         struct nfsd4_setclientid *setclid = &u->setclientid;
3863         struct xdr_netobj       clname = setclid->se_name;
3864         nfs4_verifier           clverifier = setclid->se_verf;
3865         struct nfs4_client      *conf, *new;
3866         struct nfs4_client      *unconf = NULL;
3867         __be32                  status;
3868         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3869
3870         new = create_client(clname, rqstp, &clverifier);
3871         if (new == NULL)
3872                 return nfserr_jukebox;
3873         /* Cases below refer to rfc 3530 section 14.2.33: */
3874         spin_lock(&nn->client_lock);
3875         conf = find_confirmed_client_by_name(&clname, nn);
3876         if (conf && client_has_state(conf)) {
3877                 /* case 0: */
3878                 status = nfserr_clid_inuse;
3879                 if (clp_used_exchangeid(conf))
3880                         goto out;
3881                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3882                         char addr_str[INET6_ADDRSTRLEN];
3883                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3884                                  sizeof(addr_str));
3885                         dprintk("NFSD: setclientid: string in use by client "
3886                                 "at %s\n", addr_str);
3887                         goto out;
3888                 }
3889         }
3890         unconf = find_unconfirmed_client_by_name(&clname, nn);
3891         if (unconf)
3892                 unhash_client_locked(unconf);
3893         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3894                 /* case 1: probable callback update */
3895                 copy_clid(new, conf);
3896                 gen_confirm(new, nn);
3897         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3898                 ;
3899         new->cl_minorversion = 0;
3900         gen_callback(new, setclid, rqstp);
3901         add_to_unconfirmed(new);
3902         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3903         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3904         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3905         new = NULL;
3906         status = nfs_ok;
3907 out:
3908         spin_unlock(&nn->client_lock);
3909         if (new)
3910                 free_client(new);
3911         if (unconf)
3912                 expire_client(unconf);
3913         return status;
3914 }
3915
3916
3917 __be32
3918 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3919                         struct nfsd4_compound_state *cstate,
3920                         union nfsd4_op_u *u)
3921 {
3922         struct nfsd4_setclientid_confirm *setclientid_confirm =
3923                         &u->setclientid_confirm;
3924         struct nfs4_client *conf, *unconf;
3925         struct nfs4_client *old = NULL;
3926         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3927         clientid_t * clid = &setclientid_confirm->sc_clientid;
3928         __be32 status;
3929         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3930
3931         if (STALE_CLIENTID(clid, nn))
3932                 return nfserr_stale_clientid;
3933
3934         spin_lock(&nn->client_lock);
3935         conf = find_confirmed_client(clid, false, nn);
3936         unconf = find_unconfirmed_client(clid, false, nn);
3937         /*
3938          * We try hard to give out unique clientid's, so if we get an
3939          * attempt to confirm the same clientid with a different cred,
3940          * the client may be buggy; this should never happen.
3941          *
3942          * Nevertheless, RFC 7530 recommends INUSE for this case:
3943          */
3944         status = nfserr_clid_inuse;
3945         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3946                 goto out;
3947         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3948                 goto out;
3949         /* cases below refer to rfc 3530 section 14.2.34: */
3950         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3951                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3952                         /* case 2: probable retransmit */
3953                         status = nfs_ok;
3954                 } else /* case 4: client hasn't noticed we rebooted yet? */
3955                         status = nfserr_stale_clientid;
3956                 goto out;
3957         }
3958         status = nfs_ok;
3959         if (conf) { /* case 1: callback update */
3960                 old = unconf;
3961                 unhash_client_locked(old);
3962                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3963         } else { /* case 3: normal case; new or rebooted client */
3964                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3965                 if (old) {
3966                         status = nfserr_clid_inuse;
3967                         if (client_has_state(old)
3968                                         && !same_creds(&unconf->cl_cred,
3969                                                         &old->cl_cred))
3970                                 goto out;
3971                         status = mark_client_expired_locked(old);
3972                         if (status) {
3973                                 old = NULL;
3974                                 goto out;
3975                         }
3976                 }
3977                 move_to_confirmed(unconf);
3978                 conf = unconf;
3979         }
3980         get_client_locked(conf);
3981         spin_unlock(&nn->client_lock);
3982         nfsd4_probe_callback(conf);
3983         spin_lock(&nn->client_lock);
3984         put_client_renew_locked(conf);
3985 out:
3986         spin_unlock(&nn->client_lock);
3987         if (old)
3988                 expire_client(old);
3989         return status;
3990 }
3991
3992 static struct nfs4_file *nfsd4_alloc_file(void)
3993 {
3994         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3995 }
3996
3997 /* OPEN Share state helper functions */
3998 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3999                                 struct nfs4_file *fp)
4000 {
4001         lockdep_assert_held(&state_lock);
4002
4003         refcount_set(&fp->fi_ref, 1);
4004         spin_lock_init(&fp->fi_lock);
4005         INIT_LIST_HEAD(&fp->fi_stateids);
4006         INIT_LIST_HEAD(&fp->fi_delegations);
4007         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4008         fh_copy_shallow(&fp->fi_fhandle, fh);
4009         fp->fi_deleg_file = NULL;
4010         fp->fi_had_conflict = false;
4011         fp->fi_share_deny = 0;
4012         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4013         memset(fp->fi_access, 0, sizeof(fp->fi_access));
4014 #ifdef CONFIG_NFSD_PNFS
4015         INIT_LIST_HEAD(&fp->fi_lo_states);
4016         atomic_set(&fp->fi_lo_recalls, 0);
4017 #endif
4018         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4019 }
4020
4021 void
4022 nfsd4_free_slabs(void)
4023 {
4024         kmem_cache_destroy(client_slab);
4025         kmem_cache_destroy(openowner_slab);
4026         kmem_cache_destroy(lockowner_slab);
4027         kmem_cache_destroy(file_slab);
4028         kmem_cache_destroy(stateid_slab);
4029         kmem_cache_destroy(deleg_slab);
4030         kmem_cache_destroy(odstate_slab);
4031 }
4032
4033 int
4034 nfsd4_init_slabs(void)
4035 {
4036         client_slab = kmem_cache_create("nfsd4_clients",
4037                         sizeof(struct nfs4_client), 0, 0, NULL);
4038         if (client_slab == NULL)
4039                 goto out;
4040         openowner_slab = kmem_cache_create("nfsd4_openowners",
4041                         sizeof(struct nfs4_openowner), 0, 0, NULL);
4042         if (openowner_slab == NULL)
4043                 goto out_free_client_slab;
4044         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4045                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
4046         if (lockowner_slab == NULL)
4047                 goto out_free_openowner_slab;
4048         file_slab = kmem_cache_create("nfsd4_files",
4049                         sizeof(struct nfs4_file), 0, 0, NULL);
4050         if (file_slab == NULL)
4051                 goto out_free_lockowner_slab;
4052         stateid_slab = kmem_cache_create("nfsd4_stateids",
4053                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4054         if (stateid_slab == NULL)
4055                 goto out_free_file_slab;
4056         deleg_slab = kmem_cache_create("nfsd4_delegations",
4057                         sizeof(struct nfs4_delegation), 0, 0, NULL);
4058         if (deleg_slab == NULL)
4059                 goto out_free_stateid_slab;
4060         odstate_slab = kmem_cache_create("nfsd4_odstate",
4061                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4062         if (odstate_slab == NULL)
4063                 goto out_free_deleg_slab;
4064         return 0;
4065
4066 out_free_deleg_slab:
4067         kmem_cache_destroy(deleg_slab);
4068 out_free_stateid_slab:
4069         kmem_cache_destroy(stateid_slab);
4070 out_free_file_slab:
4071         kmem_cache_destroy(file_slab);
4072 out_free_lockowner_slab:
4073         kmem_cache_destroy(lockowner_slab);
4074 out_free_openowner_slab:
4075         kmem_cache_destroy(openowner_slab);
4076 out_free_client_slab:
4077         kmem_cache_destroy(client_slab);
4078 out:
4079         dprintk("nfsd4: out of memory while initializing nfsv4\n");
4080         return -ENOMEM;
4081 }
4082
4083 static void init_nfs4_replay(struct nfs4_replay *rp)
4084 {
4085         rp->rp_status = nfserr_serverfault;
4086         rp->rp_buflen = 0;
4087         rp->rp_buf = rp->rp_ibuf;
4088         mutex_init(&rp->rp_mutex);
4089 }
4090
4091 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4092                 struct nfs4_stateowner *so)
4093 {
4094         if (!nfsd4_has_session(cstate)) {
4095                 mutex_lock(&so->so_replay.rp_mutex);
4096                 cstate->replay_owner = nfs4_get_stateowner(so);
4097         }
4098 }
4099
4100 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4101 {
4102         struct nfs4_stateowner *so = cstate->replay_owner;
4103
4104         if (so != NULL) {
4105                 cstate->replay_owner = NULL;
4106                 mutex_unlock(&so->so_replay.rp_mutex);
4107                 nfs4_put_stateowner(so);
4108         }
4109 }
4110
4111 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4112 {
4113         struct nfs4_stateowner *sop;
4114
4115         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4116         if (!sop)
4117                 return NULL;
4118
4119         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4120         if (!sop->so_owner.data) {
4121                 kmem_cache_free(slab, sop);
4122                 return NULL;
4123         }
4124
4125         INIT_LIST_HEAD(&sop->so_stateids);
4126         sop->so_client = clp;
4127         init_nfs4_replay(&sop->so_replay);
4128         atomic_set(&sop->so_count, 1);
4129         return sop;
4130 }
4131
4132 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4133 {
4134         lockdep_assert_held(&clp->cl_lock);
4135
4136         list_add(&oo->oo_owner.so_strhash,
4137                  &clp->cl_ownerstr_hashtbl[strhashval]);
4138         list_add(&oo->oo_perclient, &clp->cl_openowners);
4139 }
4140
4141 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4142 {
4143         unhash_openowner_locked(openowner(so));
4144 }
4145
4146 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4147 {
4148         struct nfs4_openowner *oo = openowner(so);
4149
4150         kmem_cache_free(openowner_slab, oo);
4151 }
4152
4153 static const struct nfs4_stateowner_operations openowner_ops = {
4154         .so_unhash =    nfs4_unhash_openowner,
4155         .so_free =      nfs4_free_openowner,
4156 };
4157
4158 static struct nfs4_ol_stateid *
4159 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4160 {
4161         struct nfs4_ol_stateid *local, *ret = NULL;
4162         struct nfs4_openowner *oo = open->op_openowner;
4163
4164         lockdep_assert_held(&fp->fi_lock);
4165
4166         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4167                 /* ignore lock owners */
4168                 if (local->st_stateowner->so_is_open_owner == 0)
4169                         continue;
4170                 if (local->st_stateowner != &oo->oo_owner)
4171                         continue;
4172                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4173                         ret = local;
4174                         refcount_inc(&ret->st_stid.sc_count);
4175                         break;
4176                 }
4177         }
4178         return ret;
4179 }
4180
4181 static __be32
4182 nfsd4_verify_open_stid(struct nfs4_stid *s)
4183 {
4184         __be32 ret = nfs_ok;
4185
4186         switch (s->sc_type) {
4187         default:
4188                 break;
4189         case 0:
4190         case NFS4_CLOSED_STID:
4191         case NFS4_CLOSED_DELEG_STID:
4192                 ret = nfserr_bad_stateid;
4193                 break;
4194         case NFS4_REVOKED_DELEG_STID:
4195                 ret = nfserr_deleg_revoked;
4196         }
4197         return ret;
4198 }
4199
4200 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4201 static __be32
4202 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4203 {
4204         __be32 ret;
4205
4206         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4207         ret = nfsd4_verify_open_stid(&stp->st_stid);
4208         if (ret != nfs_ok)
4209                 mutex_unlock(&stp->st_mutex);
4210         return ret;
4211 }
4212
4213 static struct nfs4_ol_stateid *
4214 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4215 {
4216         struct nfs4_ol_stateid *stp;
4217         for (;;) {
4218                 spin_lock(&fp->fi_lock);
4219                 stp = nfsd4_find_existing_open(fp, open);
4220                 spin_unlock(&fp->fi_lock);
4221                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4222                         break;
4223                 nfs4_put_stid(&stp->st_stid);
4224         }
4225         return stp;
4226 }
4227
4228 static struct nfs4_openowner *
4229 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4230                            struct nfsd4_compound_state *cstate)
4231 {
4232         struct nfs4_client *clp = cstate->clp;
4233         struct nfs4_openowner *oo, *ret;
4234
4235         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4236         if (!oo)
4237                 return NULL;
4238         oo->oo_owner.so_ops = &openowner_ops;
4239         oo->oo_owner.so_is_open_owner = 1;
4240         oo->oo_owner.so_seqid = open->op_seqid;
4241         oo->oo_flags = 0;
4242         if (nfsd4_has_session(cstate))
4243                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4244         oo->oo_time = 0;
4245         oo->oo_last_closed_stid = NULL;
4246         INIT_LIST_HEAD(&oo->oo_close_lru);
4247         spin_lock(&clp->cl_lock);
4248         ret = find_openstateowner_str_locked(strhashval, open, clp);
4249         if (ret == NULL) {
4250                 hash_openowner(oo, clp, strhashval);
4251                 ret = oo;
4252         } else
4253                 nfs4_free_stateowner(&oo->oo_owner);
4254
4255         spin_unlock(&clp->cl_lock);
4256         return ret;
4257 }
4258
4259 static struct nfs4_ol_stateid *
4260 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4261 {
4262
4263         struct nfs4_openowner *oo = open->op_openowner;
4264         struct nfs4_ol_stateid *retstp = NULL;
4265         struct nfs4_ol_stateid *stp;
4266
4267         stp = open->op_stp;
4268         /* We are moving these outside of the spinlocks to avoid the warnings */
4269         mutex_init(&stp->st_mutex);
4270         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4271
4272 retry:
4273         spin_lock(&oo->oo_owner.so_client->cl_lock);
4274         spin_lock(&fp->fi_lock);
4275
4276         retstp = nfsd4_find_existing_open(fp, open);
4277         if (retstp)
4278                 goto out_unlock;
4279
4280         open->op_stp = NULL;
4281         refcount_inc(&stp->st_stid.sc_count);
4282         stp->st_stid.sc_type = NFS4_OPEN_STID;
4283         INIT_LIST_HEAD(&stp->st_locks);
4284         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4285         get_nfs4_file(fp);
4286         stp->st_stid.sc_file = fp;
4287         stp->st_access_bmap = 0;
4288         stp->st_deny_bmap = 0;
4289         stp->st_openstp = NULL;
4290         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4291         list_add(&stp->st_perfile, &fp->fi_stateids);
4292
4293 out_unlock:
4294         spin_unlock(&fp->fi_lock);
4295         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4296         if (retstp) {
4297                 /* Handle races with CLOSE */
4298                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4299                         nfs4_put_stid(&retstp->st_stid);
4300                         goto retry;
4301                 }
4302                 /* To keep mutex tracking happy */
4303                 mutex_unlock(&stp->st_mutex);
4304                 stp = retstp;
4305         }
4306         return stp;
4307 }
4308
4309 /*
4310  * In the 4.0 case we need to keep the owners around a little while to handle
4311  * CLOSE replay. We still do need to release any file access that is held by
4312  * them before returning however.
4313  */
4314 static void
4315 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4316 {
4317         struct nfs4_ol_stateid *last;
4318         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4319         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4320                                                 nfsd_net_id);
4321
4322         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4323
4324         /*
4325          * We know that we hold one reference via nfsd4_close, and another
4326          * "persistent" reference for the client. If the refcount is higher
4327          * than 2, then there are still calls in progress that are using this
4328          * stateid. We can't put the sc_file reference until they are finished.
4329          * Wait for the refcount to drop to 2. Since it has been unhashed,
4330          * there should be no danger of the refcount going back up again at
4331          * this point.
4332          */
4333         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4334
4335         release_all_access(s);
4336         if (s->st_stid.sc_file) {
4337                 put_nfs4_file(s->st_stid.sc_file);
4338                 s->st_stid.sc_file = NULL;
4339         }
4340
4341         spin_lock(&nn->client_lock);
4342         last = oo->oo_last_closed_stid;
4343         oo->oo_last_closed_stid = s;
4344         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4345         oo->oo_time = ktime_get_boottime_seconds();
4346         spin_unlock(&nn->client_lock);
4347         if (last)
4348                 nfs4_put_stid(&last->st_stid);
4349 }
4350
4351 /* search file_hashtbl[] for file */
4352 static struct nfs4_file *
4353 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4354 {
4355         struct nfs4_file *fp;
4356
4357         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4358                                 lockdep_is_held(&state_lock)) {
4359                 if (fh_match(&fp->fi_fhandle, fh)) {
4360                         if (refcount_inc_not_zero(&fp->fi_ref))
4361                                 return fp;
4362                 }
4363         }
4364         return NULL;
4365 }
4366
4367 struct nfs4_file *
4368 find_file(struct knfsd_fh *fh)
4369 {
4370         struct nfs4_file *fp;
4371         unsigned int hashval = file_hashval(fh);
4372
4373         rcu_read_lock();
4374         fp = find_file_locked(fh, hashval);
4375         rcu_read_unlock();
4376         return fp;
4377 }
4378
4379 static struct nfs4_file *
4380 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4381 {
4382         struct nfs4_file *fp;
4383         unsigned int hashval = file_hashval(fh);
4384
4385         rcu_read_lock();
4386         fp = find_file_locked(fh, hashval);
4387         rcu_read_unlock();
4388         if (fp)
4389                 return fp;
4390
4391         spin_lock(&state_lock);
4392         fp = find_file_locked(fh, hashval);
4393         if (likely(fp == NULL)) {
4394                 nfsd4_init_file(fh, hashval, new);
4395                 fp = new;
4396         }
4397         spin_unlock(&state_lock);
4398
4399         return fp;
4400 }
4401
4402 /*
4403  * Called to check deny when READ with all zero stateid or
4404  * WRITE with all zero or all one stateid
4405  */
4406 static __be32
4407 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4408 {
4409         struct nfs4_file *fp;
4410         __be32 ret = nfs_ok;
4411
4412         fp = find_file(&current_fh->fh_handle);
4413         if (!fp)
4414                 return ret;
4415         /* Check for conflicting share reservations */
4416         spin_lock(&fp->fi_lock);
4417         if (fp->fi_share_deny & deny_type)
4418                 ret = nfserr_locked;
4419         spin_unlock(&fp->fi_lock);
4420         put_nfs4_file(fp);
4421         return ret;
4422 }
4423
4424 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4425 {
4426         struct nfs4_delegation *dp = cb_to_delegation(cb);
4427         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4428                                           nfsd_net_id);
4429
4430         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4431
4432         /*
4433          * We can't do this in nfsd_break_deleg_cb because it is
4434          * already holding inode->i_lock.
4435          *
4436          * If the dl_time != 0, then we know that it has already been
4437          * queued for a lease break. Don't queue it again.
4438          */
4439         spin_lock(&state_lock);
4440         if (dp->dl_time == 0) {
4441                 dp->dl_time = ktime_get_boottime_seconds();
4442                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4443         }
4444         spin_unlock(&state_lock);
4445 }
4446
4447 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4448                 struct rpc_task *task)
4449 {
4450         struct nfs4_delegation *dp = cb_to_delegation(cb);
4451
4452         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4453                 return 1;
4454
4455         switch (task->tk_status) {
4456         case 0:
4457                 return 1;
4458         case -NFS4ERR_DELAY:
4459                 rpc_delay(task, 2 * HZ);
4460                 return 0;
4461         case -EBADHANDLE:
4462         case -NFS4ERR_BAD_STATEID:
4463                 /*
4464                  * Race: client probably got cb_recall before open reply
4465                  * granting delegation.
4466                  */
4467                 if (dp->dl_retries--) {
4468                         rpc_delay(task, 2 * HZ);
4469                         return 0;
4470                 }
4471                 /*FALLTHRU*/
4472         default:
4473                 return 1;
4474         }
4475 }
4476
4477 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4478 {
4479         struct nfs4_delegation *dp = cb_to_delegation(cb);
4480
4481         nfs4_put_stid(&dp->dl_stid);
4482 }
4483
4484 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4485         .prepare        = nfsd4_cb_recall_prepare,
4486         .done           = nfsd4_cb_recall_done,
4487         .release        = nfsd4_cb_recall_release,
4488 };
4489
4490 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4491 {
4492         /*
4493          * We're assuming the state code never drops its reference
4494          * without first removing the lease.  Since we're in this lease
4495          * callback (and since the lease code is serialized by the
4496          * i_lock) we know the server hasn't removed the lease yet, and
4497          * we know it's safe to take a reference.
4498          */
4499         refcount_inc(&dp->dl_stid.sc_count);
4500         nfsd4_run_cb(&dp->dl_recall);
4501 }
4502
4503 /* Called from break_lease() with i_lock held. */
4504 static bool
4505 nfsd_break_deleg_cb(struct file_lock *fl)
4506 {
4507         bool ret = false;
4508         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4509         struct nfs4_file *fp = dp->dl_stid.sc_file;
4510
4511         /*
4512          * We don't want the locks code to timeout the lease for us;
4513          * we'll remove it ourself if a delegation isn't returned
4514          * in time:
4515          */
4516         fl->fl_break_time = 0;
4517
4518         spin_lock(&fp->fi_lock);
4519         fp->fi_had_conflict = true;
4520         nfsd_break_one_deleg(dp);
4521         spin_unlock(&fp->fi_lock);
4522         return ret;
4523 }
4524
4525 static int
4526 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4527                      struct list_head *dispose)
4528 {
4529         if (arg & F_UNLCK)
4530                 return lease_modify(onlist, arg, dispose);
4531         else
4532                 return -EAGAIN;
4533 }
4534
4535 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4536         .lm_break = nfsd_break_deleg_cb,
4537         .lm_change = nfsd_change_deleg_cb,
4538 };
4539
4540 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4541 {
4542         if (nfsd4_has_session(cstate))
4543                 return nfs_ok;
4544         if (seqid == so->so_seqid - 1)
4545                 return nfserr_replay_me;
4546         if (seqid == so->so_seqid)
4547                 return nfs_ok;
4548         return nfserr_bad_seqid;
4549 }
4550
4551 static __be32 lookup_clientid(clientid_t *clid,
4552                 struct nfsd4_compound_state *cstate,
4553                 struct nfsd_net *nn,
4554                 bool sessions)
4555 {
4556         struct nfs4_client *found;
4557
4558         if (cstate->clp) {
4559                 found = cstate->clp;
4560                 if (!same_clid(&found->cl_clientid, clid))
4561                         return nfserr_stale_clientid;
4562                 return nfs_ok;
4563         }
4564
4565         if (STALE_CLIENTID(clid, nn))
4566                 return nfserr_stale_clientid;
4567
4568         /*
4569          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4570          * cached already then we know this is for is for v4.0 and "sessions"
4571          * will be false.
4572          */
4573         WARN_ON_ONCE(cstate->session);
4574         spin_lock(&nn->client_lock);
4575         found = find_confirmed_client(clid, sessions, nn);
4576         if (!found) {
4577                 spin_unlock(&nn->client_lock);
4578                 return nfserr_expired;
4579         }
4580         atomic_inc(&found->cl_rpc_users);
4581         spin_unlock(&nn->client_lock);
4582
4583         /* Cache the nfs4_client in cstate! */
4584         cstate->clp = found;
4585         return nfs_ok;
4586 }
4587
4588 __be32
4589 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4590                     struct nfsd4_open *open, struct nfsd_net *nn)
4591 {
4592         clientid_t *clientid = &open->op_clientid;
4593         struct nfs4_client *clp = NULL;
4594         unsigned int strhashval;
4595         struct nfs4_openowner *oo = NULL;
4596         __be32 status;
4597
4598         if (STALE_CLIENTID(&open->op_clientid, nn))
4599                 return nfserr_stale_clientid;
4600         /*
4601          * In case we need it later, after we've already created the
4602          * file and don't want to risk a further failure:
4603          */
4604         open->op_file = nfsd4_alloc_file();
4605         if (open->op_file == NULL)
4606                 return nfserr_jukebox;
4607
4608         status = lookup_clientid(clientid, cstate, nn, false);
4609         if (status)
4610                 return status;
4611         clp = cstate->clp;
4612
4613         strhashval = ownerstr_hashval(&open->op_owner);
4614         oo = find_openstateowner_str(strhashval, open, clp);
4615         open->op_openowner = oo;
4616         if (!oo) {
4617                 goto new_owner;
4618         }
4619         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4620                 /* Replace unconfirmed owners without checking for replay. */
4621                 release_openowner(oo);
4622                 open->op_openowner = NULL;
4623                 goto new_owner;
4624         }
4625         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4626         if (status)
4627                 return status;
4628         goto alloc_stateid;
4629 new_owner:
4630         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4631         if (oo == NULL)
4632                 return nfserr_jukebox;
4633         open->op_openowner = oo;
4634 alloc_stateid:
4635         open->op_stp = nfs4_alloc_open_stateid(clp);
4636         if (!open->op_stp)
4637                 return nfserr_jukebox;
4638
4639         if (nfsd4_has_session(cstate) &&
4640             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4641                 open->op_odstate = alloc_clnt_odstate(clp);
4642                 if (!open->op_odstate)
4643                         return nfserr_jukebox;
4644         }
4645
4646         return nfs_ok;
4647 }
4648
4649 static inline __be32
4650 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4651 {
4652         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4653                 return nfserr_openmode;
4654         else
4655                 return nfs_ok;
4656 }
4657
4658 static int share_access_to_flags(u32 share_access)
4659 {
4660         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4661 }
4662
4663 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4664 {
4665         struct nfs4_stid *ret;
4666
4667         ret = find_stateid_by_type(cl, s,
4668                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4669         if (!ret)
4670                 return NULL;
4671         return delegstateid(ret);
4672 }
4673
4674 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4675 {
4676         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4677                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4678 }
4679
4680 static __be32
4681 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4682                 struct nfs4_delegation **dp)
4683 {
4684         int flags;
4685         __be32 status = nfserr_bad_stateid;
4686         struct nfs4_delegation *deleg;
4687
4688         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4689         if (deleg == NULL)
4690                 goto out;
4691         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4692                 nfs4_put_stid(&deleg->dl_stid);
4693                 if (cl->cl_minorversion)
4694                         status = nfserr_deleg_revoked;
4695                 goto out;
4696         }
4697         flags = share_access_to_flags(open->op_share_access);
4698         status = nfs4_check_delegmode(deleg, flags);
4699         if (status) {
4700                 nfs4_put_stid(&deleg->dl_stid);
4701                 goto out;
4702         }
4703         *dp = deleg;
4704 out:
4705         if (!nfsd4_is_deleg_cur(open))
4706                 return nfs_ok;
4707         if (status)
4708                 return status;
4709         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4710         return nfs_ok;
4711 }
4712
4713 static inline int nfs4_access_to_access(u32 nfs4_access)
4714 {
4715         int flags = 0;
4716
4717         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4718                 flags |= NFSD_MAY_READ;
4719         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4720                 flags |= NFSD_MAY_WRITE;
4721         return flags;
4722 }
4723
4724 static inline __be32
4725 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4726                 struct nfsd4_open *open)
4727 {
4728         struct iattr iattr = {
4729                 .ia_valid = ATTR_SIZE,
4730                 .ia_size = 0,
4731         };
4732         if (!open->op_truncate)
4733                 return 0;
4734         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4735                 return nfserr_inval;
4736         return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4737 }
4738
4739 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4740                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4741                 struct nfsd4_open *open)
4742 {
4743         struct nfsd_file *nf = NULL;
4744         __be32 status;
4745         int oflag = nfs4_access_to_omode(open->op_share_access);
4746         int access = nfs4_access_to_access(open->op_share_access);
4747         unsigned char old_access_bmap, old_deny_bmap;
4748
4749         spin_lock(&fp->fi_lock);
4750
4751         /*
4752          * Are we trying to set a deny mode that would conflict with
4753          * current access?
4754          */
4755         status = nfs4_file_check_deny(fp, open->op_share_deny);
4756         if (status != nfs_ok) {
4757                 spin_unlock(&fp->fi_lock);
4758                 goto out;
4759         }
4760
4761         /* set access to the file */
4762         status = nfs4_file_get_access(fp, open->op_share_access);
4763         if (status != nfs_ok) {
4764                 spin_unlock(&fp->fi_lock);
4765                 goto out;
4766         }
4767
4768         /* Set access bits in stateid */
4769         old_access_bmap = stp->st_access_bmap;
4770         set_access(open->op_share_access, stp);
4771
4772         /* Set new deny mask */
4773         old_deny_bmap = stp->st_deny_bmap;
4774         set_deny(open->op_share_deny, stp);
4775         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4776
4777         if (!fp->fi_fds[oflag]) {
4778                 spin_unlock(&fp->fi_lock);
4779                 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4780                 if (status)
4781                         goto out_put_access;
4782                 spin_lock(&fp->fi_lock);
4783                 if (!fp->fi_fds[oflag]) {
4784                         fp->fi_fds[oflag] = nf;
4785                         nf = NULL;
4786                 }
4787         }
4788         spin_unlock(&fp->fi_lock);
4789         if (nf)
4790                 nfsd_file_put(nf);
4791
4792         status = nfsd4_truncate(rqstp, cur_fh, open);
4793         if (status)
4794                 goto out_put_access;
4795 out:
4796         return status;
4797 out_put_access:
4798         stp->st_access_bmap = old_access_bmap;
4799         nfs4_file_put_access(fp, open->op_share_access);
4800         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4801         goto out;
4802 }
4803
4804 static __be32
4805 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4806 {
4807         __be32 status;
4808         unsigned char old_deny_bmap = stp->st_deny_bmap;
4809
4810         if (!test_access(open->op_share_access, stp))
4811                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4812
4813         /* test and set deny mode */
4814         spin_lock(&fp->fi_lock);
4815         status = nfs4_file_check_deny(fp, open->op_share_deny);
4816         if (status == nfs_ok) {
4817                 set_deny(open->op_share_deny, stp);
4818                 fp->fi_share_deny |=
4819                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4820         }
4821         spin_unlock(&fp->fi_lock);
4822
4823         if (status != nfs_ok)
4824                 return status;
4825
4826         status = nfsd4_truncate(rqstp, cur_fh, open);
4827         if (status != nfs_ok)
4828                 reset_union_bmap_deny(old_deny_bmap, stp);
4829         return status;
4830 }
4831
4832 /* Should we give out recallable state?: */
4833 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4834 {
4835         if (clp->cl_cb_state == NFSD4_CB_UP)
4836                 return true;
4837         /*
4838          * In the sessions case, since we don't have to establish a
4839          * separate connection for callbacks, we assume it's OK
4840          * until we hear otherwise:
4841          */
4842         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4843 }
4844
4845 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4846                                                 int flag)
4847 {
4848         struct file_lock *fl;
4849
4850         fl = locks_alloc_lock();
4851         if (!fl)
4852                 return NULL;
4853         fl->fl_lmops = &nfsd_lease_mng_ops;
4854         fl->fl_flags = FL_DELEG;
4855         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4856         fl->fl_end = OFFSET_MAX;
4857         fl->fl_owner = (fl_owner_t)dp;
4858         fl->fl_pid = current->tgid;
4859         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4860         return fl;
4861 }
4862
4863 static struct nfs4_delegation *
4864 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4865                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4866 {
4867         int status = 0;
4868         struct nfs4_delegation *dp;
4869         struct nfsd_file *nf;
4870         struct file_lock *fl;
4871
4872         /*
4873          * The fi_had_conflict and nfs_get_existing_delegation checks
4874          * here are just optimizations; we'll need to recheck them at
4875          * the end:
4876          */
4877         if (fp->fi_had_conflict)
4878                 return ERR_PTR(-EAGAIN);
4879
4880         nf = find_readable_file(fp);
4881         if (!nf) {
4882                 /* We should always have a readable file here */
4883                 WARN_ON_ONCE(1);
4884                 return ERR_PTR(-EBADF);
4885         }
4886         spin_lock(&state_lock);
4887         spin_lock(&fp->fi_lock);
4888         if (nfs4_delegation_exists(clp, fp))
4889                 status = -EAGAIN;
4890         else if (!fp->fi_deleg_file) {
4891                 fp->fi_deleg_file = nf;
4892                 /* increment early to prevent fi_deleg_file from being
4893                  * cleared */
4894                 fp->fi_delegees = 1;
4895                 nf = NULL;
4896         } else
4897                 fp->fi_delegees++;
4898         spin_unlock(&fp->fi_lock);
4899         spin_unlock(&state_lock);
4900         if (nf)
4901                 nfsd_file_put(nf);
4902         if (status)
4903                 return ERR_PTR(status);
4904
4905         status = -ENOMEM;
4906         dp = alloc_init_deleg(clp, fp, fh, odstate);
4907         if (!dp)
4908                 goto out_delegees;
4909
4910         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4911         if (!fl)
4912                 goto out_clnt_odstate;
4913
4914         status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
4915         if (fl)
4916                 locks_free_lock(fl);
4917         if (status)
4918                 goto out_clnt_odstate;
4919
4920         spin_lock(&state_lock);
4921         spin_lock(&fp->fi_lock);
4922         if (fp->fi_had_conflict)
4923                 status = -EAGAIN;
4924         else
4925                 status = hash_delegation_locked(dp, fp);
4926         spin_unlock(&fp->fi_lock);
4927         spin_unlock(&state_lock);
4928
4929         if (status)
4930                 goto out_unlock;
4931
4932         return dp;
4933 out_unlock:
4934         vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
4935 out_clnt_odstate:
4936         put_clnt_odstate(dp->dl_clnt_odstate);
4937         nfs4_put_stid(&dp->dl_stid);
4938 out_delegees:
4939         put_deleg_file(fp);
4940         return ERR_PTR(status);
4941 }
4942
4943 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4944 {
4945         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4946         if (status == -EAGAIN)
4947                 open->op_why_no_deleg = WND4_CONTENTION;
4948         else {
4949                 open->op_why_no_deleg = WND4_RESOURCE;
4950                 switch (open->op_deleg_want) {
4951                 case NFS4_SHARE_WANT_READ_DELEG:
4952                 case NFS4_SHARE_WANT_WRITE_DELEG:
4953                 case NFS4_SHARE_WANT_ANY_DELEG:
4954                         break;
4955                 case NFS4_SHARE_WANT_CANCEL:
4956                         open->op_why_no_deleg = WND4_CANCELLED;
4957                         break;
4958                 case NFS4_SHARE_WANT_NO_DELEG:
4959                         WARN_ON_ONCE(1);
4960                 }
4961         }
4962 }
4963
4964 /*
4965  * Attempt to hand out a delegation.
4966  *
4967  * Note we don't support write delegations, and won't until the vfs has
4968  * proper support for them.
4969  */
4970 static void
4971 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4972                         struct nfs4_ol_stateid *stp)
4973 {
4974         struct nfs4_delegation *dp;
4975         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4976         struct nfs4_client *clp = stp->st_stid.sc_client;
4977         int cb_up;
4978         int status = 0;
4979
4980         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4981         open->op_recall = 0;
4982         switch (open->op_claim_type) {
4983                 case NFS4_OPEN_CLAIM_PREVIOUS:
4984                         if (!cb_up)
4985                                 open->op_recall = 1;
4986                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4987                                 goto out_no_deleg;
4988                         break;
4989                 case NFS4_OPEN_CLAIM_NULL:
4990                 case NFS4_OPEN_CLAIM_FH:
4991                         /*
4992                          * Let's not give out any delegations till everyone's
4993                          * had the chance to reclaim theirs, *and* until
4994                          * NLM locks have all been reclaimed:
4995                          */
4996                         if (locks_in_grace(clp->net))
4997                                 goto out_no_deleg;
4998                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4999                                 goto out_no_deleg;
5000                         /*
5001                          * Also, if the file was opened for write or
5002                          * create, there's a good chance the client's
5003                          * about to write to it, resulting in an
5004                          * immediate recall (since we don't support
5005                          * write delegations):
5006                          */
5007                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
5008                                 goto out_no_deleg;
5009                         if (open->op_create == NFS4_OPEN_CREATE)
5010                                 goto out_no_deleg;
5011                         break;
5012                 default:
5013                         goto out_no_deleg;
5014         }
5015         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5016         if (IS_ERR(dp))
5017                 goto out_no_deleg;
5018
5019         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5020
5021         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
5022                 STATEID_VAL(&dp->dl_stid.sc_stateid));
5023         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5024         nfs4_put_stid(&dp->dl_stid);
5025         return;
5026 out_no_deleg:
5027         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5028         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5029             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5030                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5031                 open->op_recall = 1;
5032         }
5033
5034         /* 4.1 client asking for a delegation? */
5035         if (open->op_deleg_want)
5036                 nfsd4_open_deleg_none_ext(open, status);
5037         return;
5038 }
5039
5040 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5041                                         struct nfs4_delegation *dp)
5042 {
5043         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5044             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5045                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5046                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5047         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5048                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5049                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5050                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5051         }
5052         /* Otherwise the client must be confused wanting a delegation
5053          * it already has, therefore we don't return
5054          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5055          */
5056 }
5057
5058 __be32
5059 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5060 {
5061         struct nfsd4_compoundres *resp = rqstp->rq_resp;
5062         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5063         struct nfs4_file *fp = NULL;
5064         struct nfs4_ol_stateid *stp = NULL;
5065         struct nfs4_delegation *dp = NULL;
5066         __be32 status;
5067         bool new_stp = false;
5068
5069         /*
5070          * Lookup file; if found, lookup stateid and check open request,
5071          * and check for delegations in the process of being recalled.
5072          * If not found, create the nfs4_file struct
5073          */
5074         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
5075         if (fp != open->op_file) {
5076                 status = nfs4_check_deleg(cl, open, &dp);
5077                 if (status)
5078                         goto out;
5079                 stp = nfsd4_find_and_lock_existing_open(fp, open);
5080         } else {
5081                 open->op_file = NULL;
5082                 status = nfserr_bad_stateid;
5083                 if (nfsd4_is_deleg_cur(open))
5084                         goto out;
5085         }
5086
5087         if (!stp) {
5088                 stp = init_open_stateid(fp, open);
5089                 if (!open->op_stp)
5090                         new_stp = true;
5091         }
5092
5093         /*
5094          * OPEN the file, or upgrade an existing OPEN.
5095          * If truncate fails, the OPEN fails.
5096          *
5097          * stp is already locked.
5098          */
5099         if (!new_stp) {
5100                 /* Stateid was found, this is an OPEN upgrade */
5101                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5102                 if (status) {
5103                         mutex_unlock(&stp->st_mutex);
5104                         goto out;
5105                 }
5106         } else {
5107                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5108                 if (status) {
5109                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5110                         release_open_stateid(stp);
5111                         mutex_unlock(&stp->st_mutex);
5112                         goto out;
5113                 }
5114
5115                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5116                                                         open->op_odstate);
5117                 if (stp->st_clnt_odstate == open->op_odstate)
5118                         open->op_odstate = NULL;
5119         }
5120
5121         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5122         mutex_unlock(&stp->st_mutex);
5123
5124         if (nfsd4_has_session(&resp->cstate)) {
5125                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5126                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5127                         open->op_why_no_deleg = WND4_NOT_WANTED;
5128                         goto nodeleg;
5129                 }
5130         }
5131
5132         /*
5133         * Attempt to hand out a delegation. No error return, because the
5134         * OPEN succeeds even if we fail.
5135         */
5136         nfs4_open_delegation(current_fh, open, stp);
5137 nodeleg:
5138         status = nfs_ok;
5139
5140         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5141                 STATEID_VAL(&stp->st_stid.sc_stateid));
5142 out:
5143         /* 4.1 client trying to upgrade/downgrade delegation? */
5144         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5145             open->op_deleg_want)
5146                 nfsd4_deleg_xgrade_none_ext(open, dp);
5147
5148         if (fp)
5149                 put_nfs4_file(fp);
5150         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5151                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5152         /*
5153         * To finish the open response, we just need to set the rflags.
5154         */
5155         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5156         if (nfsd4_has_session(&resp->cstate))
5157                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5158         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5159                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5160
5161         if (dp)
5162                 nfs4_put_stid(&dp->dl_stid);
5163         if (stp)
5164                 nfs4_put_stid(&stp->st_stid);
5165
5166         return status;
5167 }
5168
5169 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5170                               struct nfsd4_open *open)
5171 {
5172         if (open->op_openowner) {
5173                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5174
5175                 nfsd4_cstate_assign_replay(cstate, so);
5176                 nfs4_put_stateowner(so);
5177         }
5178         if (open->op_file)
5179                 kmem_cache_free(file_slab, open->op_file);
5180         if (open->op_stp)
5181                 nfs4_put_stid(&open->op_stp->st_stid);
5182         if (open->op_odstate)
5183                 kmem_cache_free(odstate_slab, open->op_odstate);
5184 }
5185
5186 __be32
5187 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5188             union nfsd4_op_u *u)
5189 {
5190         clientid_t *clid = &u->renew;
5191         struct nfs4_client *clp;
5192         __be32 status;
5193         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5194
5195         dprintk("process_renew(%08x/%08x): starting\n",
5196                         clid->cl_boot, clid->cl_id);
5197         status = lookup_clientid(clid, cstate, nn, false);
5198         if (status)
5199                 goto out;
5200         clp = cstate->clp;
5201         status = nfserr_cb_path_down;
5202         if (!list_empty(&clp->cl_delegations)
5203                         && clp->cl_cb_state != NFSD4_CB_UP)
5204                 goto out;
5205         status = nfs_ok;
5206 out:
5207         return status;
5208 }
5209
5210 void
5211 nfsd4_end_grace(struct nfsd_net *nn)
5212 {
5213         /* do nothing if grace period already ended */
5214         if (nn->grace_ended)
5215                 return;
5216
5217         nn->grace_ended = true;
5218         /*
5219          * If the server goes down again right now, an NFSv4
5220          * client will still be allowed to reclaim after it comes back up,
5221          * even if it hasn't yet had a chance to reclaim state this time.
5222          *
5223          */
5224         nfsd4_record_grace_done(nn);
5225         /*
5226          * At this point, NFSv4 clients can still reclaim.  But if the
5227          * server crashes, any that have not yet reclaimed will be out
5228          * of luck on the next boot.
5229          *
5230          * (NFSv4.1+ clients are considered to have reclaimed once they
5231          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5232          * have reclaimed after their first OPEN.)
5233          */
5234         locks_end_grace(&nn->nfsd4_manager);
5235         /*
5236          * At this point, and once lockd and/or any other containers
5237          * exit their grace period, further reclaims will fail and
5238          * regular locking can resume.
5239          */
5240 }
5241
5242 /*
5243  * If we've waited a lease period but there are still clients trying to
5244  * reclaim, wait a little longer to give them a chance to finish.
5245  */
5246 static bool clients_still_reclaiming(struct nfsd_net *nn)
5247 {
5248         time64_t double_grace_period_end = nn->boot_time +
5249                                            2 * nn->nfsd4_lease;
5250
5251         if (nn->track_reclaim_completes &&
5252                         atomic_read(&nn->nr_reclaim_complete) ==
5253                         nn->reclaim_str_hashtbl_size)
5254                 return false;
5255         if (!nn->somebody_reclaimed)
5256                 return false;
5257         nn->somebody_reclaimed = false;
5258         /*
5259          * If we've given them *two* lease times to reclaim, and they're
5260          * still not done, give up:
5261          */
5262         if (ktime_get_boottime_seconds() > double_grace_period_end)
5263                 return false;
5264         return true;
5265 }
5266
5267 static time64_t
5268 nfs4_laundromat(struct nfsd_net *nn)
5269 {
5270         struct nfs4_client *clp;
5271         struct nfs4_openowner *oo;
5272         struct nfs4_delegation *dp;
5273         struct nfs4_ol_stateid *stp;
5274         struct nfsd4_blocked_lock *nbl;
5275         struct list_head *pos, *next, reaplist;
5276         time64_t cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease;
5277         time64_t t, new_timeo = nn->nfsd4_lease;
5278         struct nfs4_cpntf_state *cps;
5279         copy_stateid_t *cps_t;
5280         int i;
5281
5282         dprintk("NFSD: laundromat service - starting\n");
5283
5284         if (clients_still_reclaiming(nn)) {
5285                 new_timeo = 0;
5286                 goto out;
5287         }
5288         dprintk("NFSD: end of grace period\n");
5289         nfsd4_end_grace(nn);
5290         INIT_LIST_HEAD(&reaplist);
5291
5292         spin_lock(&nn->s2s_cp_lock);
5293         idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5294                 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5295                 if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5296                                 cps->cpntf_time > cutoff)
5297                         _free_cpntf_state_locked(nn, cps);
5298         }
5299         spin_unlock(&nn->s2s_cp_lock);
5300
5301         spin_lock(&nn->client_lock);
5302         list_for_each_safe(pos, next, &nn->client_lru) {
5303                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5304                 if (clp->cl_time > cutoff) {
5305                         t = clp->cl_time - cutoff;
5306                         new_timeo = min(new_timeo, t);
5307                         break;
5308                 }
5309                 if (mark_client_expired_locked(clp)) {
5310                         dprintk("NFSD: client in use (clientid %08x)\n",
5311                                 clp->cl_clientid.cl_id);
5312                         continue;
5313                 }
5314                 list_add(&clp->cl_lru, &reaplist);
5315         }
5316         spin_unlock(&nn->client_lock);
5317         list_for_each_safe(pos, next, &reaplist) {
5318                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5319                 dprintk("NFSD: purging unused client (clientid %08x)\n",
5320                         clp->cl_clientid.cl_id);
5321                 list_del_init(&clp->cl_lru);
5322                 expire_client(clp);
5323         }
5324         spin_lock(&state_lock);
5325         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5326                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5327                 if (dp->dl_time > cutoff) {
5328                         t = dp->dl_time - cutoff;
5329                         new_timeo = min(new_timeo, t);
5330                         break;
5331                 }
5332                 WARN_ON(!unhash_delegation_locked(dp));
5333                 list_add(&dp->dl_recall_lru, &reaplist);
5334         }
5335         spin_unlock(&state_lock);
5336         while (!list_empty(&reaplist)) {
5337                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5338                                         dl_recall_lru);
5339                 list_del_init(&dp->dl_recall_lru);
5340                 revoke_delegation(dp);
5341         }
5342
5343         spin_lock(&nn->client_lock);
5344         while (!list_empty(&nn->close_lru)) {
5345                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5346                                         oo_close_lru);
5347                 if (oo->oo_time > cutoff) {
5348                         t = oo->oo_time - cutoff;
5349                         new_timeo = min(new_timeo, t);
5350                         break;
5351                 }
5352                 list_del_init(&oo->oo_close_lru);
5353                 stp = oo->oo_last_closed_stid;
5354                 oo->oo_last_closed_stid = NULL;
5355                 spin_unlock(&nn->client_lock);
5356                 nfs4_put_stid(&stp->st_stid);
5357                 spin_lock(&nn->client_lock);
5358         }
5359         spin_unlock(&nn->client_lock);
5360
5361         /*
5362          * It's possible for a client to try and acquire an already held lock
5363          * that is being held for a long time, and then lose interest in it.
5364          * So, we clean out any un-revisited request after a lease period
5365          * under the assumption that the client is no longer interested.
5366          *
5367          * RFC5661, sec. 9.6 states that the client must not rely on getting
5368          * notifications and must continue to poll for locks, even when the
5369          * server supports them. Thus this shouldn't lead to clients blocking
5370          * indefinitely once the lock does become free.
5371          */
5372         BUG_ON(!list_empty(&reaplist));
5373         spin_lock(&nn->blocked_locks_lock);
5374         while (!list_empty(&nn->blocked_locks_lru)) {
5375                 nbl = list_first_entry(&nn->blocked_locks_lru,
5376                                         struct nfsd4_blocked_lock, nbl_lru);
5377                 if (nbl->nbl_time > cutoff) {
5378                         t = nbl->nbl_time - cutoff;
5379                         new_timeo = min(new_timeo, t);
5380                         break;
5381                 }
5382                 list_move(&nbl->nbl_lru, &reaplist);
5383                 list_del_init(&nbl->nbl_list);
5384         }
5385         spin_unlock(&nn->blocked_locks_lock);
5386
5387         while (!list_empty(&reaplist)) {
5388                 nbl = list_first_entry(&reaplist,
5389                                         struct nfsd4_blocked_lock, nbl_lru);
5390                 list_del_init(&nbl->nbl_lru);
5391                 free_blocked_lock(nbl);
5392         }
5393 out:
5394         new_timeo = max_t(time64_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5395         return new_timeo;
5396 }
5397
5398 static struct workqueue_struct *laundry_wq;
5399 static void laundromat_main(struct work_struct *);
5400
5401 static void
5402 laundromat_main(struct work_struct *laundry)
5403 {
5404         time64_t t;
5405         struct delayed_work *dwork = to_delayed_work(laundry);
5406         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5407                                            laundromat_work);
5408
5409         t = nfs4_laundromat(nn);
5410         dprintk("NFSD: laundromat_main - sleeping for %lld seconds\n", t);
5411         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5412 }
5413
5414 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5415 {
5416         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5417                 return nfserr_bad_stateid;
5418         return nfs_ok;
5419 }
5420
5421 static inline int
5422 access_permit_read(struct nfs4_ol_stateid *stp)
5423 {
5424         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5425                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5426                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5427 }
5428
5429 static inline int
5430 access_permit_write(struct nfs4_ol_stateid *stp)
5431 {
5432         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5433                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5434 }
5435
5436 static
5437 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5438 {
5439         __be32 status = nfserr_openmode;
5440
5441         /* For lock stateid's, we test the parent open, not the lock: */
5442         if (stp->st_openstp)
5443                 stp = stp->st_openstp;
5444         if ((flags & WR_STATE) && !access_permit_write(stp))
5445                 goto out;
5446         if ((flags & RD_STATE) && !access_permit_read(stp))
5447                 goto out;
5448         status = nfs_ok;
5449 out:
5450         return status;
5451 }
5452
5453 static inline __be32
5454 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5455 {
5456         if (ONE_STATEID(stateid) && (flags & RD_STATE))
5457                 return nfs_ok;
5458         else if (opens_in_grace(net)) {
5459                 /* Answer in remaining cases depends on existence of
5460                  * conflicting state; so we must wait out the grace period. */
5461                 return nfserr_grace;
5462         } else if (flags & WR_STATE)
5463                 return nfs4_share_conflict(current_fh,
5464                                 NFS4_SHARE_DENY_WRITE);
5465         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5466                 return nfs4_share_conflict(current_fh,
5467                                 NFS4_SHARE_DENY_READ);
5468 }
5469
5470 /*
5471  * Allow READ/WRITE during grace period on recovered state only for files
5472  * that are not able to provide mandatory locking.
5473  */
5474 static inline int
5475 grace_disallows_io(struct net *net, struct inode *inode)
5476 {
5477         return opens_in_grace(net) && mandatory_lock(inode);
5478 }
5479
5480 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5481 {
5482         /*
5483          * When sessions are used the stateid generation number is ignored
5484          * when it is zero.
5485          */
5486         if (has_session && in->si_generation == 0)
5487                 return nfs_ok;
5488
5489         if (in->si_generation == ref->si_generation)
5490                 return nfs_ok;
5491
5492         /* If the client sends us a stateid from the future, it's buggy: */
5493         if (nfsd4_stateid_generation_after(in, ref))
5494                 return nfserr_bad_stateid;
5495         /*
5496          * However, we could see a stateid from the past, even from a
5497          * non-buggy client.  For example, if the client sends a lock
5498          * while some IO is outstanding, the lock may bump si_generation
5499          * while the IO is still in flight.  The client could avoid that
5500          * situation by waiting for responses on all the IO requests,
5501          * but better performance may result in retrying IO that
5502          * receives an old_stateid error if requests are rarely
5503          * reordered in flight:
5504          */
5505         return nfserr_old_stateid;
5506 }
5507
5508 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5509 {
5510         __be32 ret;
5511
5512         spin_lock(&s->sc_lock);
5513         ret = nfsd4_verify_open_stid(s);
5514         if (ret == nfs_ok)
5515                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5516         spin_unlock(&s->sc_lock);
5517         return ret;
5518 }
5519
5520 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5521 {
5522         if (ols->st_stateowner->so_is_open_owner &&
5523             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5524                 return nfserr_bad_stateid;
5525         return nfs_ok;
5526 }
5527
5528 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5529 {
5530         struct nfs4_stid *s;
5531         __be32 status = nfserr_bad_stateid;
5532
5533         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5534                 CLOSE_STATEID(stateid))
5535                 return status;
5536         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
5537                 return status;
5538         spin_lock(&cl->cl_lock);
5539         s = find_stateid_locked(cl, stateid);
5540         if (!s)
5541                 goto out_unlock;
5542         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5543         if (status)
5544                 goto out_unlock;
5545         switch (s->sc_type) {
5546         case NFS4_DELEG_STID:
5547                 status = nfs_ok;
5548                 break;
5549         case NFS4_REVOKED_DELEG_STID:
5550                 status = nfserr_deleg_revoked;
5551                 break;
5552         case NFS4_OPEN_STID:
5553         case NFS4_LOCK_STID:
5554                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5555                 break;
5556         default:
5557                 printk("unknown stateid type %x\n", s->sc_type);
5558                 /* Fallthrough */
5559         case NFS4_CLOSED_STID:
5560         case NFS4_CLOSED_DELEG_STID:
5561                 status = nfserr_bad_stateid;
5562         }
5563 out_unlock:
5564         spin_unlock(&cl->cl_lock);
5565         return status;
5566 }
5567
5568 __be32
5569 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5570                      stateid_t *stateid, unsigned char typemask,
5571                      struct nfs4_stid **s, struct nfsd_net *nn)
5572 {
5573         __be32 status;
5574         bool return_revoked = false;
5575
5576         /*
5577          *  only return revoked delegations if explicitly asked.
5578          *  otherwise we report revoked or bad_stateid status.
5579          */
5580         if (typemask & NFS4_REVOKED_DELEG_STID)
5581                 return_revoked = true;
5582         else if (typemask & NFS4_DELEG_STID)
5583                 typemask |= NFS4_REVOKED_DELEG_STID;
5584
5585         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5586                 CLOSE_STATEID(stateid))
5587                 return nfserr_bad_stateid;
5588         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn,
5589                                  false);
5590         if (status == nfserr_stale_clientid) {
5591                 if (cstate->session)
5592                         return nfserr_bad_stateid;
5593                 return nfserr_stale_stateid;
5594         }
5595         if (status)
5596                 return status;
5597         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5598         if (!*s)
5599                 return nfserr_bad_stateid;
5600         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5601                 nfs4_put_stid(*s);
5602                 if (cstate->minorversion)
5603                         return nfserr_deleg_revoked;
5604                 return nfserr_bad_stateid;
5605         }
5606         return nfs_ok;
5607 }
5608
5609 static struct nfsd_file *
5610 nfs4_find_file(struct nfs4_stid *s, int flags)
5611 {
5612         if (!s)
5613                 return NULL;
5614
5615         switch (s->sc_type) {
5616         case NFS4_DELEG_STID:
5617                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5618                         return NULL;
5619                 return nfsd_file_get(s->sc_file->fi_deleg_file);
5620         case NFS4_OPEN_STID:
5621         case NFS4_LOCK_STID:
5622                 if (flags & RD_STATE)
5623                         return find_readable_file(s->sc_file);
5624                 else
5625                         return find_writeable_file(s->sc_file);
5626                 break;
5627         }
5628
5629         return NULL;
5630 }
5631
5632 static __be32
5633 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5634 {
5635         __be32 status;
5636
5637         status = nfsd4_check_openowner_confirmed(ols);
5638         if (status)
5639                 return status;
5640         return nfs4_check_openmode(ols, flags);
5641 }
5642
5643 static __be32
5644 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5645                 struct nfsd_file **nfp, int flags)
5646 {
5647         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5648         struct nfsd_file *nf;
5649         __be32 status;
5650
5651         nf = nfs4_find_file(s, flags);
5652         if (nf) {
5653                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5654                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5655                 if (status) {
5656                         nfsd_file_put(nf);
5657                         goto out;
5658                 }
5659         } else {
5660                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5661                 if (status)
5662                         return status;
5663         }
5664         *nfp = nf;
5665 out:
5666         return status;
5667 }
5668 static void
5669 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5670 {
5671         WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
5672         if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
5673                 return;
5674         list_del(&cps->cp_list);
5675         idr_remove(&nn->s2s_cp_stateids,
5676                    cps->cp_stateid.stid.si_opaque.so_id);
5677         kfree(cps);
5678 }
5679 /*
5680  * A READ from an inter server to server COPY will have a
5681  * copy stateid. Look up the copy notify stateid from the
5682  * idr structure and take a reference on it.
5683  */
5684 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5685                           struct nfs4_client *clp,
5686                           struct nfs4_cpntf_state **cps)
5687 {
5688         copy_stateid_t *cps_t;
5689         struct nfs4_cpntf_state *state = NULL;
5690
5691         if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
5692                 return nfserr_bad_stateid;
5693         spin_lock(&nn->s2s_cp_lock);
5694         cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
5695         if (cps_t) {
5696                 state = container_of(cps_t, struct nfs4_cpntf_state,
5697                                      cp_stateid);
5698                 if (state->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID) {
5699                         state = NULL;
5700                         goto unlock;
5701                 }
5702                 if (!clp)
5703                         refcount_inc(&state->cp_stateid.sc_count);
5704                 else
5705                         _free_cpntf_state_locked(nn, state);
5706         }
5707 unlock:
5708         spin_unlock(&nn->s2s_cp_lock);
5709         if (!state)
5710                 return nfserr_bad_stateid;
5711         if (!clp && state)
5712                 *cps = state;
5713         return 0;
5714 }
5715
5716 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5717                                struct nfs4_stid **stid)
5718 {
5719         __be32 status;
5720         struct nfs4_cpntf_state *cps = NULL;
5721         struct nfsd4_compound_state cstate;
5722
5723         status = manage_cpntf_state(nn, st, NULL, &cps);
5724         if (status)
5725                 return status;
5726
5727         cps->cpntf_time = ktime_get_boottime_seconds();
5728         memset(&cstate, 0, sizeof(cstate));
5729         status = lookup_clientid(&cps->cp_p_clid, &cstate, nn, true);
5730         if (status)
5731                 goto out;
5732         status = nfsd4_lookup_stateid(&cstate, &cps->cp_p_stateid,
5733                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5734                                 stid, nn);
5735         put_client_renew(cstate.clp);
5736 out:
5737         nfs4_put_cpntf_state(nn, cps);
5738         return status;
5739 }
5740
5741 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5742 {
5743         spin_lock(&nn->s2s_cp_lock);
5744         _free_cpntf_state_locked(nn, cps);
5745         spin_unlock(&nn->s2s_cp_lock);
5746 }
5747
5748 /*
5749  * Checks for stateid operations
5750  */
5751 __be32
5752 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5753                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5754                 stateid_t *stateid, int flags, struct nfsd_file **nfp,
5755                 struct nfs4_stid **cstid)
5756 {
5757         struct inode *ino = d_inode(fhp->fh_dentry);
5758         struct net *net = SVC_NET(rqstp);
5759         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5760         struct nfs4_stid *s = NULL;
5761         __be32 status;
5762
5763         if (nfp)
5764                 *nfp = NULL;
5765
5766         if (grace_disallows_io(net, ino))
5767                 return nfserr_grace;
5768
5769         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5770                 status = check_special_stateids(net, fhp, stateid, flags);
5771                 goto done;
5772         }
5773
5774         status = nfsd4_lookup_stateid(cstate, stateid,
5775                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5776                                 &s, nn);
5777         if (status == nfserr_bad_stateid)
5778                 status = find_cpntf_state(nn, stateid, &s);
5779         if (status)
5780                 return status;
5781         status = nfsd4_stid_check_stateid_generation(stateid, s,
5782                         nfsd4_has_session(cstate));
5783         if (status)
5784                 goto out;
5785
5786         switch (s->sc_type) {
5787         case NFS4_DELEG_STID:
5788                 status = nfs4_check_delegmode(delegstateid(s), flags);
5789                 break;
5790         case NFS4_OPEN_STID:
5791         case NFS4_LOCK_STID:
5792                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5793                 break;
5794         default:
5795                 status = nfserr_bad_stateid;
5796                 break;
5797         }
5798         if (status)
5799                 goto out;
5800         status = nfs4_check_fh(fhp, s);
5801
5802 done:
5803         if (status == nfs_ok && nfp)
5804                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5805 out:
5806         if (s) {
5807                 if (!status && cstid)
5808                         *cstid = s;
5809                 else
5810                         nfs4_put_stid(s);
5811         }
5812         return status;
5813 }
5814
5815 /*
5816  * Test if the stateid is valid
5817  */
5818 __be32
5819 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5820                    union nfsd4_op_u *u)
5821 {
5822         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5823         struct nfsd4_test_stateid_id *stateid;
5824         struct nfs4_client *cl = cstate->session->se_client;
5825
5826         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5827                 stateid->ts_id_status =
5828                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5829
5830         return nfs_ok;
5831 }
5832
5833 static __be32
5834 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5835 {
5836         struct nfs4_ol_stateid *stp = openlockstateid(s);
5837         __be32 ret;
5838
5839         ret = nfsd4_lock_ol_stateid(stp);
5840         if (ret)
5841                 goto out_put_stid;
5842
5843         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5844         if (ret)
5845                 goto out;
5846
5847         ret = nfserr_locks_held;
5848         if (check_for_locks(stp->st_stid.sc_file,
5849                             lockowner(stp->st_stateowner)))
5850                 goto out;
5851
5852         release_lock_stateid(stp);
5853         ret = nfs_ok;
5854
5855 out:
5856         mutex_unlock(&stp->st_mutex);
5857 out_put_stid:
5858         nfs4_put_stid(s);
5859         return ret;
5860 }
5861
5862 __be32
5863 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5864                    union nfsd4_op_u *u)
5865 {
5866         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5867         stateid_t *stateid = &free_stateid->fr_stateid;
5868         struct nfs4_stid *s;
5869         struct nfs4_delegation *dp;
5870         struct nfs4_client *cl = cstate->session->se_client;
5871         __be32 ret = nfserr_bad_stateid;
5872
5873         spin_lock(&cl->cl_lock);
5874         s = find_stateid_locked(cl, stateid);
5875         if (!s)
5876                 goto out_unlock;
5877         spin_lock(&s->sc_lock);
5878         switch (s->sc_type) {
5879         case NFS4_DELEG_STID:
5880                 ret = nfserr_locks_held;
5881                 break;
5882         case NFS4_OPEN_STID:
5883                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5884                 if (ret)
5885                         break;
5886                 ret = nfserr_locks_held;
5887                 break;
5888         case NFS4_LOCK_STID:
5889                 spin_unlock(&s->sc_lock);
5890                 refcount_inc(&s->sc_count);
5891                 spin_unlock(&cl->cl_lock);
5892                 ret = nfsd4_free_lock_stateid(stateid, s);
5893                 goto out;
5894         case NFS4_REVOKED_DELEG_STID:
5895                 spin_unlock(&s->sc_lock);
5896                 dp = delegstateid(s);
5897                 list_del_init(&dp->dl_recall_lru);
5898                 spin_unlock(&cl->cl_lock);
5899                 nfs4_put_stid(s);
5900                 ret = nfs_ok;
5901                 goto out;
5902         /* Default falls through and returns nfserr_bad_stateid */
5903         }
5904         spin_unlock(&s->sc_lock);
5905 out_unlock:
5906         spin_unlock(&cl->cl_lock);
5907 out:
5908         return ret;
5909 }
5910
5911 static inline int
5912 setlkflg (int type)
5913 {
5914         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5915                 RD_STATE : WR_STATE;
5916 }
5917
5918 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5919 {
5920         struct svc_fh *current_fh = &cstate->current_fh;
5921         struct nfs4_stateowner *sop = stp->st_stateowner;
5922         __be32 status;
5923
5924         status = nfsd4_check_seqid(cstate, sop, seqid);
5925         if (status)
5926                 return status;
5927         status = nfsd4_lock_ol_stateid(stp);
5928         if (status != nfs_ok)
5929                 return status;
5930         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5931         if (status == nfs_ok)
5932                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5933         if (status != nfs_ok)
5934                 mutex_unlock(&stp->st_mutex);
5935         return status;
5936 }
5937
5938 /*
5939  * Checks for sequence id mutating operations.
5940  */
5941 static __be32
5942 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5943                          stateid_t *stateid, char typemask,
5944                          struct nfs4_ol_stateid **stpp,
5945                          struct nfsd_net *nn)
5946 {
5947         __be32 status;
5948         struct nfs4_stid *s;
5949         struct nfs4_ol_stateid *stp = NULL;
5950
5951         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5952                 seqid, STATEID_VAL(stateid));
5953
5954         *stpp = NULL;
5955         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5956         if (status)
5957                 return status;
5958         stp = openlockstateid(s);
5959         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5960
5961         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5962         if (!status)
5963                 *stpp = stp;
5964         else
5965                 nfs4_put_stid(&stp->st_stid);
5966         return status;
5967 }
5968
5969 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5970                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5971 {
5972         __be32 status;
5973         struct nfs4_openowner *oo;
5974         struct nfs4_ol_stateid *stp;
5975
5976         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5977                                                 NFS4_OPEN_STID, &stp, nn);
5978         if (status)
5979                 return status;
5980         oo = openowner(stp->st_stateowner);
5981         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5982                 mutex_unlock(&stp->st_mutex);
5983                 nfs4_put_stid(&stp->st_stid);
5984                 return nfserr_bad_stateid;
5985         }
5986         *stpp = stp;
5987         return nfs_ok;
5988 }
5989
5990 __be32
5991 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5992                    union nfsd4_op_u *u)
5993 {
5994         struct nfsd4_open_confirm *oc = &u->open_confirm;
5995         __be32 status;
5996         struct nfs4_openowner *oo;
5997         struct nfs4_ol_stateid *stp;
5998         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5999
6000         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6001                         cstate->current_fh.fh_dentry);
6002
6003         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6004         if (status)
6005                 return status;
6006
6007         status = nfs4_preprocess_seqid_op(cstate,
6008                                         oc->oc_seqid, &oc->oc_req_stateid,
6009                                         NFS4_OPEN_STID, &stp, nn);
6010         if (status)
6011                 goto out;
6012         oo = openowner(stp->st_stateowner);
6013         status = nfserr_bad_stateid;
6014         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6015                 mutex_unlock(&stp->st_mutex);
6016                 goto put_stateid;
6017         }
6018         oo->oo_flags |= NFS4_OO_CONFIRMED;
6019         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6020         mutex_unlock(&stp->st_mutex);
6021         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
6022                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
6023
6024         nfsd4_client_record_create(oo->oo_owner.so_client);
6025         status = nfs_ok;
6026 put_stateid:
6027         nfs4_put_stid(&stp->st_stid);
6028 out:
6029         nfsd4_bump_seqid(cstate, status);
6030         return status;
6031 }
6032
6033 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6034 {
6035         if (!test_access(access, stp))
6036                 return;
6037         nfs4_file_put_access(stp->st_stid.sc_file, access);
6038         clear_access(access, stp);
6039 }
6040
6041 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6042 {
6043         switch (to_access) {
6044         case NFS4_SHARE_ACCESS_READ:
6045                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6046                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6047                 break;
6048         case NFS4_SHARE_ACCESS_WRITE:
6049                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6050                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6051                 break;
6052         case NFS4_SHARE_ACCESS_BOTH:
6053                 break;
6054         default:
6055                 WARN_ON_ONCE(1);
6056         }
6057 }
6058
6059 __be32
6060 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6061                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6062 {
6063         struct nfsd4_open_downgrade *od = &u->open_downgrade;
6064         __be32 status;
6065         struct nfs4_ol_stateid *stp;
6066         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6067
6068         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6069                         cstate->current_fh.fh_dentry);
6070
6071         /* We don't yet support WANT bits: */
6072         if (od->od_deleg_want)
6073                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6074                         od->od_deleg_want);
6075
6076         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6077                                         &od->od_stateid, &stp, nn);
6078         if (status)
6079                 goto out;
6080         status = nfserr_inval;
6081         if (!test_access(od->od_share_access, stp)) {
6082                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6083                         stp->st_access_bmap, od->od_share_access);
6084                 goto put_stateid;
6085         }
6086         if (!test_deny(od->od_share_deny, stp)) {
6087                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6088                         stp->st_deny_bmap, od->od_share_deny);
6089                 goto put_stateid;
6090         }
6091         nfs4_stateid_downgrade(stp, od->od_share_access);
6092         reset_union_bmap_deny(od->od_share_deny, stp);
6093         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6094         status = nfs_ok;
6095 put_stateid:
6096         mutex_unlock(&stp->st_mutex);
6097         nfs4_put_stid(&stp->st_stid);
6098 out:
6099         nfsd4_bump_seqid(cstate, status);
6100         return status;
6101 }
6102
6103 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6104 {
6105         struct nfs4_client *clp = s->st_stid.sc_client;
6106         bool unhashed;
6107         LIST_HEAD(reaplist);
6108
6109         spin_lock(&clp->cl_lock);
6110         unhashed = unhash_open_stateid(s, &reaplist);
6111
6112         if (clp->cl_minorversion) {
6113                 if (unhashed)
6114                         put_ol_stateid_locked(s, &reaplist);
6115                 spin_unlock(&clp->cl_lock);
6116                 free_ol_stateid_reaplist(&reaplist);
6117         } else {
6118                 spin_unlock(&clp->cl_lock);
6119                 free_ol_stateid_reaplist(&reaplist);
6120                 if (unhashed)
6121                         move_to_close_lru(s, clp->net);
6122         }
6123 }
6124
6125 /*
6126  * nfs4_unlock_state() called after encode
6127  */
6128 __be32
6129 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6130                 union nfsd4_op_u *u)
6131 {
6132         struct nfsd4_close *close = &u->close;
6133         __be32 status;
6134         struct nfs4_ol_stateid *stp;
6135         struct net *net = SVC_NET(rqstp);
6136         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6137
6138         dprintk("NFSD: nfsd4_close on file %pd\n",
6139                         cstate->current_fh.fh_dentry);
6140
6141         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6142                                         &close->cl_stateid,
6143                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
6144                                         &stp, nn);
6145         nfsd4_bump_seqid(cstate, status);
6146         if (status)
6147                 goto out;
6148
6149         stp->st_stid.sc_type = NFS4_CLOSED_STID;
6150
6151         /*
6152          * Technically we don't _really_ have to increment or copy it, since
6153          * it should just be gone after this operation and we clobber the
6154          * copied value below, but we continue to do so here just to ensure
6155          * that racing ops see that there was a state change.
6156          */
6157         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6158
6159         nfsd4_close_open_stateid(stp);
6160         mutex_unlock(&stp->st_mutex);
6161
6162         /* v4.1+ suggests that we send a special stateid in here, since the
6163          * clients should just ignore this anyway. Since this is not useful
6164          * for v4.0 clients either, we set it to the special close_stateid
6165          * universally.
6166          *
6167          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6168          */
6169         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6170
6171         /* put reference from nfs4_preprocess_seqid_op */
6172         nfs4_put_stid(&stp->st_stid);
6173 out:
6174         return status;
6175 }
6176
6177 __be32
6178 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6179                   union nfsd4_op_u *u)
6180 {
6181         struct nfsd4_delegreturn *dr = &u->delegreturn;
6182         struct nfs4_delegation *dp;
6183         stateid_t *stateid = &dr->dr_stateid;
6184         struct nfs4_stid *s;
6185         __be32 status;
6186         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6187
6188         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6189                 return status;
6190
6191         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6192         if (status)
6193                 goto out;
6194         dp = delegstateid(s);
6195         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6196         if (status)
6197                 goto put_stateid;
6198
6199         destroy_delegation(dp);
6200 put_stateid:
6201         nfs4_put_stid(&dp->dl_stid);
6202 out:
6203         return status;
6204 }
6205
6206 static inline u64
6207 end_offset(u64 start, u64 len)
6208 {
6209         u64 end;
6210
6211         end = start + len;
6212         return end >= start ? end: NFS4_MAX_UINT64;
6213 }
6214
6215 /* last octet in a range */
6216 static inline u64
6217 last_byte_offset(u64 start, u64 len)
6218 {
6219         u64 end;
6220
6221         WARN_ON_ONCE(!len);
6222         end = start + len;
6223         return end > start ? end - 1: NFS4_MAX_UINT64;
6224 }
6225
6226 /*
6227  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6228  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6229  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
6230  * locking, this prevents us from being completely protocol-compliant.  The
6231  * real solution to this problem is to start using unsigned file offsets in
6232  * the VFS, but this is a very deep change!
6233  */
6234 static inline void
6235 nfs4_transform_lock_offset(struct file_lock *lock)
6236 {
6237         if (lock->fl_start < 0)
6238                 lock->fl_start = OFFSET_MAX;
6239         if (lock->fl_end < 0)
6240                 lock->fl_end = OFFSET_MAX;
6241 }
6242
6243 static fl_owner_t
6244 nfsd4_fl_get_owner(fl_owner_t owner)
6245 {
6246         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6247
6248         nfs4_get_stateowner(&lo->lo_owner);
6249         return owner;
6250 }
6251
6252 static void
6253 nfsd4_fl_put_owner(fl_owner_t owner)
6254 {
6255         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6256
6257         if (lo)
6258                 nfs4_put_stateowner(&lo->lo_owner);
6259 }
6260
6261 static void
6262 nfsd4_lm_notify(struct file_lock *fl)
6263 {
6264         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
6265         struct net                      *net = lo->lo_owner.so_client->net;
6266         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
6267         struct nfsd4_blocked_lock       *nbl = container_of(fl,
6268                                                 struct nfsd4_blocked_lock, nbl_lock);
6269         bool queue = false;
6270
6271         /* An empty list means that something else is going to be using it */
6272         spin_lock(&nn->blocked_locks_lock);
6273         if (!list_empty(&nbl->nbl_list)) {
6274                 list_del_init(&nbl->nbl_list);
6275                 list_del_init(&nbl->nbl_lru);
6276                 queue = true;
6277         }
6278         spin_unlock(&nn->blocked_locks_lock);
6279
6280         if (queue)
6281                 nfsd4_run_cb(&nbl->nbl_cb);
6282 }
6283
6284 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
6285         .lm_notify = nfsd4_lm_notify,
6286         .lm_get_owner = nfsd4_fl_get_owner,
6287         .lm_put_owner = nfsd4_fl_put_owner,
6288 };
6289
6290 static inline void
6291 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6292 {
6293         struct nfs4_lockowner *lo;
6294
6295         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6296                 lo = (struct nfs4_lockowner *) fl->fl_owner;
6297                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6298                                                 GFP_KERNEL);
6299                 if (!deny->ld_owner.data)
6300                         /* We just don't care that much */
6301                         goto nevermind;
6302                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6303         } else {
6304 nevermind:
6305                 deny->ld_owner.len = 0;
6306                 deny->ld_owner.data = NULL;
6307                 deny->ld_clientid.cl_boot = 0;
6308                 deny->ld_clientid.cl_id = 0;
6309         }
6310         deny->ld_start = fl->fl_start;
6311         deny->ld_length = NFS4_MAX_UINT64;
6312         if (fl->fl_end != NFS4_MAX_UINT64)
6313                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6314         deny->ld_type = NFS4_READ_LT;
6315         if (fl->fl_type != F_RDLCK)
6316                 deny->ld_type = NFS4_WRITE_LT;
6317 }
6318
6319 static struct nfs4_lockowner *
6320 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6321 {
6322         unsigned int strhashval = ownerstr_hashval(owner);
6323         struct nfs4_stateowner *so;
6324
6325         lockdep_assert_held(&clp->cl_lock);
6326
6327         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6328                             so_strhash) {
6329                 if (so->so_is_open_owner)
6330                         continue;
6331                 if (same_owner_str(so, owner))
6332                         return lockowner(nfs4_get_stateowner(so));
6333         }
6334         return NULL;
6335 }
6336
6337 static struct nfs4_lockowner *
6338 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6339 {
6340         struct nfs4_lockowner *lo;
6341
6342         spin_lock(&clp->cl_lock);
6343         lo = find_lockowner_str_locked(clp, owner);
6344         spin_unlock(&clp->cl_lock);
6345         return lo;
6346 }
6347
6348 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6349 {
6350         unhash_lockowner_locked(lockowner(sop));
6351 }
6352
6353 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6354 {
6355         struct nfs4_lockowner *lo = lockowner(sop);
6356
6357         kmem_cache_free(lockowner_slab, lo);
6358 }
6359
6360 static const struct nfs4_stateowner_operations lockowner_ops = {
6361         .so_unhash =    nfs4_unhash_lockowner,
6362         .so_free =      nfs4_free_lockowner,
6363 };
6364
6365 /*
6366  * Alloc a lock owner structure.
6367  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6368  * occurred.
6369  *
6370  * strhashval = ownerstr_hashval
6371  */
6372 static struct nfs4_lockowner *
6373 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6374                            struct nfs4_ol_stateid *open_stp,
6375                            struct nfsd4_lock *lock)
6376 {
6377         struct nfs4_lockowner *lo, *ret;
6378
6379         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6380         if (!lo)
6381                 return NULL;
6382         INIT_LIST_HEAD(&lo->lo_blocked);
6383         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6384         lo->lo_owner.so_is_open_owner = 0;
6385         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6386         lo->lo_owner.so_ops = &lockowner_ops;
6387         spin_lock(&clp->cl_lock);
6388         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6389         if (ret == NULL) {
6390                 list_add(&lo->lo_owner.so_strhash,
6391                          &clp->cl_ownerstr_hashtbl[strhashval]);
6392                 ret = lo;
6393         } else
6394                 nfs4_free_stateowner(&lo->lo_owner);
6395
6396         spin_unlock(&clp->cl_lock);
6397         return ret;
6398 }
6399
6400 static struct nfs4_ol_stateid *
6401 find_lock_stateid(const struct nfs4_lockowner *lo,
6402                   const struct nfs4_ol_stateid *ost)
6403 {
6404         struct nfs4_ol_stateid *lst;
6405
6406         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
6407
6408         /* If ost is not hashed, ost->st_locks will not be valid */
6409         if (!nfs4_ol_stateid_unhashed(ost))
6410                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
6411                         if (lst->st_stateowner == &lo->lo_owner) {
6412                                 refcount_inc(&lst->st_stid.sc_count);
6413                                 return lst;
6414                         }
6415                 }
6416         return NULL;
6417 }
6418
6419 static struct nfs4_ol_stateid *
6420 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6421                   struct nfs4_file *fp, struct inode *inode,
6422                   struct nfs4_ol_stateid *open_stp)
6423 {
6424         struct nfs4_client *clp = lo->lo_owner.so_client;
6425         struct nfs4_ol_stateid *retstp;
6426
6427         mutex_init(&stp->st_mutex);
6428         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6429 retry:
6430         spin_lock(&clp->cl_lock);
6431         if (nfs4_ol_stateid_unhashed(open_stp))
6432                 goto out_close;
6433         retstp = find_lock_stateid(lo, open_stp);
6434         if (retstp)
6435                 goto out_found;
6436         refcount_inc(&stp->st_stid.sc_count);
6437         stp->st_stid.sc_type = NFS4_LOCK_STID;
6438         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6439         get_nfs4_file(fp);
6440         stp->st_stid.sc_file = fp;
6441         stp->st_access_bmap = 0;
6442         stp->st_deny_bmap = open_stp->st_deny_bmap;
6443         stp->st_openstp = open_stp;
6444         spin_lock(&fp->fi_lock);
6445         list_add(&stp->st_locks, &open_stp->st_locks);
6446         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6447         list_add(&stp->st_perfile, &fp->fi_stateids);
6448         spin_unlock(&fp->fi_lock);
6449         spin_unlock(&clp->cl_lock);
6450         return stp;
6451 out_found:
6452         spin_unlock(&clp->cl_lock);
6453         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6454                 nfs4_put_stid(&retstp->st_stid);
6455                 goto retry;
6456         }
6457         /* To keep mutex tracking happy */
6458         mutex_unlock(&stp->st_mutex);
6459         return retstp;
6460 out_close:
6461         spin_unlock(&clp->cl_lock);
6462         mutex_unlock(&stp->st_mutex);
6463         return NULL;
6464 }
6465
6466 static struct nfs4_ol_stateid *
6467 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6468                             struct inode *inode, struct nfs4_ol_stateid *ost,
6469                             bool *new)
6470 {
6471         struct nfs4_stid *ns = NULL;
6472         struct nfs4_ol_stateid *lst;
6473         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6474         struct nfs4_client *clp = oo->oo_owner.so_client;
6475
6476         *new = false;
6477         spin_lock(&clp->cl_lock);
6478         lst = find_lock_stateid(lo, ost);
6479         spin_unlock(&clp->cl_lock);
6480         if (lst != NULL) {
6481                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6482                         goto out;
6483                 nfs4_put_stid(&lst->st_stid);
6484         }
6485         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6486         if (ns == NULL)
6487                 return NULL;
6488
6489         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6490         if (lst == openlockstateid(ns))
6491                 *new = true;
6492         else
6493                 nfs4_put_stid(ns);
6494 out:
6495         return lst;
6496 }
6497
6498 static int
6499 check_lock_length(u64 offset, u64 length)
6500 {
6501         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6502                 (length > ~offset)));
6503 }
6504
6505 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6506 {
6507         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6508
6509         lockdep_assert_held(&fp->fi_lock);
6510
6511         if (test_access(access, lock_stp))
6512                 return;
6513         __nfs4_file_get_access(fp, access);
6514         set_access(access, lock_stp);
6515 }
6516
6517 static __be32
6518 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6519                             struct nfs4_ol_stateid *ost,
6520                             struct nfsd4_lock *lock,
6521                             struct nfs4_ol_stateid **plst, bool *new)
6522 {
6523         __be32 status;
6524         struct nfs4_file *fi = ost->st_stid.sc_file;
6525         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6526         struct nfs4_client *cl = oo->oo_owner.so_client;
6527         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6528         struct nfs4_lockowner *lo;
6529         struct nfs4_ol_stateid *lst;
6530         unsigned int strhashval;
6531
6532         lo = find_lockowner_str(cl, &lock->lk_new_owner);
6533         if (!lo) {
6534                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6535                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6536                 if (lo == NULL)
6537                         return nfserr_jukebox;
6538         } else {
6539                 /* with an existing lockowner, seqids must be the same */
6540                 status = nfserr_bad_seqid;
6541                 if (!cstate->minorversion &&
6542                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6543                         goto out;
6544         }
6545
6546         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6547         if (lst == NULL) {
6548                 status = nfserr_jukebox;
6549                 goto out;
6550         }
6551
6552         status = nfs_ok;
6553         *plst = lst;
6554 out:
6555         nfs4_put_stateowner(&lo->lo_owner);
6556         return status;
6557 }
6558
6559 /*
6560  *  LOCK operation
6561  */
6562 __be32
6563 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6564            union nfsd4_op_u *u)
6565 {
6566         struct nfsd4_lock *lock = &u->lock;
6567         struct nfs4_openowner *open_sop = NULL;
6568         struct nfs4_lockowner *lock_sop = NULL;
6569         struct nfs4_ol_stateid *lock_stp = NULL;
6570         struct nfs4_ol_stateid *open_stp = NULL;
6571         struct nfs4_file *fp;
6572         struct nfsd_file *nf = NULL;
6573         struct nfsd4_blocked_lock *nbl = NULL;
6574         struct file_lock *file_lock = NULL;
6575         struct file_lock *conflock = NULL;
6576         __be32 status = 0;
6577         int lkflg;
6578         int err;
6579         bool new = false;
6580         unsigned char fl_type;
6581         unsigned int fl_flags = FL_POSIX;
6582         struct net *net = SVC_NET(rqstp);
6583         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6584
6585         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6586                 (long long) lock->lk_offset,
6587                 (long long) lock->lk_length);
6588
6589         if (check_lock_length(lock->lk_offset, lock->lk_length))
6590                  return nfserr_inval;
6591
6592         if ((status = fh_verify(rqstp, &cstate->current_fh,
6593                                 S_IFREG, NFSD_MAY_LOCK))) {
6594                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6595                 return status;
6596         }
6597
6598         if (lock->lk_is_new) {
6599                 if (nfsd4_has_session(cstate))
6600                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6601                         memcpy(&lock->lk_new_clientid,
6602                                 &cstate->session->se_client->cl_clientid,
6603                                 sizeof(clientid_t));
6604
6605                 status = nfserr_stale_clientid;
6606                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6607                         goto out;
6608
6609                 /* validate and update open stateid and open seqid */
6610                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6611                                         lock->lk_new_open_seqid,
6612                                         &lock->lk_new_open_stateid,
6613                                         &open_stp, nn);
6614                 if (status)
6615                         goto out;
6616                 mutex_unlock(&open_stp->st_mutex);
6617                 open_sop = openowner(open_stp->st_stateowner);
6618                 status = nfserr_bad_stateid;
6619                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6620                                                 &lock->lk_new_clientid))
6621                         goto out;
6622                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6623                                                         &lock_stp, &new);
6624         } else {
6625                 status = nfs4_preprocess_seqid_op(cstate,
6626                                        lock->lk_old_lock_seqid,
6627                                        &lock->lk_old_lock_stateid,
6628                                        NFS4_LOCK_STID, &lock_stp, nn);
6629         }
6630         if (status)
6631                 goto out;
6632         lock_sop = lockowner(lock_stp->st_stateowner);
6633
6634         lkflg = setlkflg(lock->lk_type);
6635         status = nfs4_check_openmode(lock_stp, lkflg);
6636         if (status)
6637                 goto out;
6638
6639         status = nfserr_grace;
6640         if (locks_in_grace(net) && !lock->lk_reclaim)
6641                 goto out;
6642         status = nfserr_no_grace;
6643         if (!locks_in_grace(net) && lock->lk_reclaim)
6644                 goto out;
6645
6646         fp = lock_stp->st_stid.sc_file;
6647         switch (lock->lk_type) {
6648                 case NFS4_READW_LT:
6649                         if (nfsd4_has_session(cstate))
6650                                 fl_flags |= FL_SLEEP;
6651                         /* Fallthrough */
6652                 case NFS4_READ_LT:
6653                         spin_lock(&fp->fi_lock);
6654                         nf = find_readable_file_locked(fp);
6655                         if (nf)
6656                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6657                         spin_unlock(&fp->fi_lock);
6658                         fl_type = F_RDLCK;
6659                         break;
6660                 case NFS4_WRITEW_LT:
6661                         if (nfsd4_has_session(cstate))
6662                                 fl_flags |= FL_SLEEP;
6663                         /* Fallthrough */
6664                 case NFS4_WRITE_LT:
6665                         spin_lock(&fp->fi_lock);
6666                         nf = find_writeable_file_locked(fp);
6667                         if (nf)
6668                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6669                         spin_unlock(&fp->fi_lock);
6670                         fl_type = F_WRLCK;
6671                         break;
6672                 default:
6673                         status = nfserr_inval;
6674                 goto out;
6675         }
6676
6677         if (!nf) {
6678                 status = nfserr_openmode;
6679                 goto out;
6680         }
6681
6682         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6683         if (!nbl) {
6684                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6685                 status = nfserr_jukebox;
6686                 goto out;
6687         }
6688
6689         file_lock = &nbl->nbl_lock;
6690         file_lock->fl_type = fl_type;
6691         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6692         file_lock->fl_pid = current->tgid;
6693         file_lock->fl_file = nf->nf_file;
6694         file_lock->fl_flags = fl_flags;
6695         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6696         file_lock->fl_start = lock->lk_offset;
6697         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6698         nfs4_transform_lock_offset(file_lock);
6699
6700         conflock = locks_alloc_lock();
6701         if (!conflock) {
6702                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6703                 status = nfserr_jukebox;
6704                 goto out;
6705         }
6706
6707         if (fl_flags & FL_SLEEP) {
6708                 nbl->nbl_time = ktime_get_boottime_seconds();
6709                 spin_lock(&nn->blocked_locks_lock);
6710                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6711                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6712                 spin_unlock(&nn->blocked_locks_lock);
6713         }
6714
6715         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6716         switch (err) {
6717         case 0: /* success! */
6718                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6719                 status = 0;
6720                 if (lock->lk_reclaim)
6721                         nn->somebody_reclaimed = true;
6722                 break;
6723         case FILE_LOCK_DEFERRED:
6724                 nbl = NULL;
6725                 /* Fallthrough */
6726         case -EAGAIN:           /* conflock holds conflicting lock */
6727                 status = nfserr_denied;
6728                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6729                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6730                 break;
6731         case -EDEADLK:
6732                 status = nfserr_deadlock;
6733                 break;
6734         default:
6735                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6736                 status = nfserrno(err);
6737                 break;
6738         }
6739 out:
6740         if (nbl) {
6741                 /* dequeue it if we queued it before */
6742                 if (fl_flags & FL_SLEEP) {
6743                         spin_lock(&nn->blocked_locks_lock);
6744                         list_del_init(&nbl->nbl_list);
6745                         list_del_init(&nbl->nbl_lru);
6746                         spin_unlock(&nn->blocked_locks_lock);
6747                 }
6748                 free_blocked_lock(nbl);
6749         }
6750         if (nf)
6751                 nfsd_file_put(nf);
6752         if (lock_stp) {
6753                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6754                 if (cstate->replay_owner &&
6755                     cstate->replay_owner != &lock_sop->lo_owner &&
6756                     seqid_mutating_err(ntohl(status)))
6757                         lock_sop->lo_owner.so_seqid++;
6758
6759                 /*
6760                  * If this is a new, never-before-used stateid, and we are
6761                  * returning an error, then just go ahead and release it.
6762                  */
6763                 if (status && new)
6764                         release_lock_stateid(lock_stp);
6765
6766                 mutex_unlock(&lock_stp->st_mutex);
6767
6768                 nfs4_put_stid(&lock_stp->st_stid);
6769         }
6770         if (open_stp)
6771                 nfs4_put_stid(&open_stp->st_stid);
6772         nfsd4_bump_seqid(cstate, status);
6773         if (conflock)
6774                 locks_free_lock(conflock);
6775         return status;
6776 }
6777
6778 /*
6779  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6780  * so we do a temporary open here just to get an open file to pass to
6781  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6782  * inode operation.)
6783  */
6784 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6785 {
6786         struct nfsd_file *nf;
6787         __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6788         if (!err) {
6789                 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6790                 nfsd_file_put(nf);
6791         }
6792         return err;
6793 }
6794
6795 /*
6796  * LOCKT operation
6797  */
6798 __be32
6799 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6800             union nfsd4_op_u *u)
6801 {
6802         struct nfsd4_lockt *lockt = &u->lockt;
6803         struct file_lock *file_lock = NULL;
6804         struct nfs4_lockowner *lo = NULL;
6805         __be32 status;
6806         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6807
6808         if (locks_in_grace(SVC_NET(rqstp)))
6809                 return nfserr_grace;
6810
6811         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6812                  return nfserr_inval;
6813
6814         if (!nfsd4_has_session(cstate)) {
6815                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn,
6816                                          false);
6817                 if (status)
6818                         goto out;
6819         }
6820
6821         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6822                 goto out;
6823
6824         file_lock = locks_alloc_lock();
6825         if (!file_lock) {
6826                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6827                 status = nfserr_jukebox;
6828                 goto out;
6829         }
6830
6831         switch (lockt->lt_type) {
6832                 case NFS4_READ_LT:
6833                 case NFS4_READW_LT:
6834                         file_lock->fl_type = F_RDLCK;
6835                         break;
6836                 case NFS4_WRITE_LT:
6837                 case NFS4_WRITEW_LT:
6838                         file_lock->fl_type = F_WRLCK;
6839                         break;
6840                 default:
6841                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6842                         status = nfserr_inval;
6843                         goto out;
6844         }
6845
6846         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6847         if (lo)
6848                 file_lock->fl_owner = (fl_owner_t)lo;
6849         file_lock->fl_pid = current->tgid;
6850         file_lock->fl_flags = FL_POSIX;
6851
6852         file_lock->fl_start = lockt->lt_offset;
6853         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6854
6855         nfs4_transform_lock_offset(file_lock);
6856
6857         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6858         if (status)
6859                 goto out;
6860
6861         if (file_lock->fl_type != F_UNLCK) {
6862                 status = nfserr_denied;
6863                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6864         }
6865 out:
6866         if (lo)
6867                 nfs4_put_stateowner(&lo->lo_owner);
6868         if (file_lock)
6869                 locks_free_lock(file_lock);
6870         return status;
6871 }
6872
6873 __be32
6874 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6875             union nfsd4_op_u *u)
6876 {
6877         struct nfsd4_locku *locku = &u->locku;
6878         struct nfs4_ol_stateid *stp;
6879         struct nfsd_file *nf = NULL;
6880         struct file_lock *file_lock = NULL;
6881         __be32 status;
6882         int err;
6883         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6884
6885         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6886                 (long long) locku->lu_offset,
6887                 (long long) locku->lu_length);
6888
6889         if (check_lock_length(locku->lu_offset, locku->lu_length))
6890                  return nfserr_inval;
6891
6892         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6893                                         &locku->lu_stateid, NFS4_LOCK_STID,
6894                                         &stp, nn);
6895         if (status)
6896                 goto out;
6897         nf = find_any_file(stp->st_stid.sc_file);
6898         if (!nf) {
6899                 status = nfserr_lock_range;
6900                 goto put_stateid;
6901         }
6902         file_lock = locks_alloc_lock();
6903         if (!file_lock) {
6904                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6905                 status = nfserr_jukebox;
6906                 goto put_file;
6907         }
6908
6909         file_lock->fl_type = F_UNLCK;
6910         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6911         file_lock->fl_pid = current->tgid;
6912         file_lock->fl_file = nf->nf_file;
6913         file_lock->fl_flags = FL_POSIX;
6914         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6915         file_lock->fl_start = locku->lu_offset;
6916
6917         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6918                                                 locku->lu_length);
6919         nfs4_transform_lock_offset(file_lock);
6920
6921         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
6922         if (err) {
6923                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6924                 goto out_nfserr;
6925         }
6926         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6927 put_file:
6928         nfsd_file_put(nf);
6929 put_stateid:
6930         mutex_unlock(&stp->st_mutex);
6931         nfs4_put_stid(&stp->st_stid);
6932 out:
6933         nfsd4_bump_seqid(cstate, status);
6934         if (file_lock)
6935                 locks_free_lock(file_lock);
6936         return status;
6937
6938 out_nfserr:
6939         status = nfserrno(err);
6940         goto put_file;
6941 }
6942
6943 /*
6944  * returns
6945  *      true:  locks held by lockowner
6946  *      false: no locks held by lockowner
6947  */
6948 static bool
6949 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6950 {
6951         struct file_lock *fl;
6952         int status = false;
6953         struct nfsd_file *nf = find_any_file(fp);
6954         struct inode *inode;
6955         struct file_lock_context *flctx;
6956
6957         if (!nf) {
6958                 /* Any valid lock stateid should have some sort of access */
6959                 WARN_ON_ONCE(1);
6960                 return status;
6961         }
6962
6963         inode = locks_inode(nf->nf_file);
6964         flctx = inode->i_flctx;
6965
6966         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6967                 spin_lock(&flctx->flc_lock);
6968                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6969                         if (fl->fl_owner == (fl_owner_t)lowner) {
6970                                 status = true;
6971                                 break;
6972                         }
6973                 }
6974                 spin_unlock(&flctx->flc_lock);
6975         }
6976         nfsd_file_put(nf);
6977         return status;
6978 }
6979
6980 __be32
6981 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6982                         struct nfsd4_compound_state *cstate,
6983                         union nfsd4_op_u *u)
6984 {
6985         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6986         clientid_t *clid = &rlockowner->rl_clientid;
6987         struct nfs4_stateowner *sop;
6988         struct nfs4_lockowner *lo = NULL;
6989         struct nfs4_ol_stateid *stp;
6990         struct xdr_netobj *owner = &rlockowner->rl_owner;
6991         unsigned int hashval = ownerstr_hashval(owner);
6992         __be32 status;
6993         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6994         struct nfs4_client *clp;
6995         LIST_HEAD (reaplist);
6996
6997         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6998                 clid->cl_boot, clid->cl_id);
6999
7000         status = lookup_clientid(clid, cstate, nn, false);
7001         if (status)
7002                 return status;
7003
7004         clp = cstate->clp;
7005         /* Find the matching lock stateowner */
7006         spin_lock(&clp->cl_lock);
7007         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
7008                             so_strhash) {
7009
7010                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
7011                         continue;
7012
7013                 /* see if there are still any locks associated with it */
7014                 lo = lockowner(sop);
7015                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
7016                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
7017                                 status = nfserr_locks_held;
7018                                 spin_unlock(&clp->cl_lock);
7019                                 return status;
7020                         }
7021                 }
7022
7023                 nfs4_get_stateowner(sop);
7024                 break;
7025         }
7026         if (!lo) {
7027                 spin_unlock(&clp->cl_lock);
7028                 return status;
7029         }
7030
7031         unhash_lockowner_locked(lo);
7032         while (!list_empty(&lo->lo_owner.so_stateids)) {
7033                 stp = list_first_entry(&lo->lo_owner.so_stateids,
7034                                        struct nfs4_ol_stateid,
7035                                        st_perstateowner);
7036                 WARN_ON(!unhash_lock_stateid(stp));
7037                 put_ol_stateid_locked(stp, &reaplist);
7038         }
7039         spin_unlock(&clp->cl_lock);
7040         free_ol_stateid_reaplist(&reaplist);
7041         remove_blocked_locks(lo);
7042         nfs4_put_stateowner(&lo->lo_owner);
7043
7044         return status;
7045 }
7046
7047 static inline struct nfs4_client_reclaim *
7048 alloc_reclaim(void)
7049 {
7050         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7051 }
7052
7053 bool
7054 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7055 {
7056         struct nfs4_client_reclaim *crp;
7057
7058         crp = nfsd4_find_reclaim_client(name, nn);
7059         return (crp && crp->cr_clp);
7060 }
7061
7062 /*
7063  * failure => all reset bets are off, nfserr_no_grace...
7064  *
7065  * The caller is responsible for freeing name.data if NULL is returned (it
7066  * will be freed in nfs4_remove_reclaim_record in the normal case).
7067  */
7068 struct nfs4_client_reclaim *
7069 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7070                 struct nfsd_net *nn)
7071 {
7072         unsigned int strhashval;
7073         struct nfs4_client_reclaim *crp;
7074
7075         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
7076         crp = alloc_reclaim();
7077         if (crp) {
7078                 strhashval = clientstr_hashval(name);
7079                 INIT_LIST_HEAD(&crp->cr_strhash);
7080                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7081                 crp->cr_name.data = name.data;
7082                 crp->cr_name.len = name.len;
7083                 crp->cr_princhash.data = princhash.data;
7084                 crp->cr_princhash.len = princhash.len;
7085                 crp->cr_clp = NULL;
7086                 nn->reclaim_str_hashtbl_size++;
7087         }
7088         return crp;
7089 }
7090
7091 void
7092 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7093 {
7094         list_del(&crp->cr_strhash);
7095         kfree(crp->cr_name.data);
7096         kfree(crp->cr_princhash.data);
7097         kfree(crp);
7098         nn->reclaim_str_hashtbl_size--;
7099 }
7100
7101 void
7102 nfs4_release_reclaim(struct nfsd_net *nn)
7103 {
7104         struct nfs4_client_reclaim *crp = NULL;
7105         int i;
7106
7107         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7108                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7109                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7110                                         struct nfs4_client_reclaim, cr_strhash);
7111                         nfs4_remove_reclaim_record(crp, nn);
7112                 }
7113         }
7114         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7115 }
7116
7117 /*
7118  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7119 struct nfs4_client_reclaim *
7120 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7121 {
7122         unsigned int strhashval;
7123         struct nfs4_client_reclaim *crp = NULL;
7124
7125         dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
7126
7127         strhashval = clientstr_hashval(name);
7128         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7129                 if (compare_blob(&crp->cr_name, &name) == 0) {
7130                         return crp;
7131                 }
7132         }
7133         return NULL;
7134 }
7135
7136 /*
7137 * Called from OPEN. Look for clientid in reclaim list.
7138 */
7139 __be32
7140 nfs4_check_open_reclaim(clientid_t *clid,
7141                 struct nfsd4_compound_state *cstate,
7142                 struct nfsd_net *nn)
7143 {
7144         __be32 status;
7145
7146         /* find clientid in conf_id_hashtbl */
7147         status = lookup_clientid(clid, cstate, nn, false);
7148         if (status)
7149                 return nfserr_reclaim_bad;
7150
7151         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
7152                 return nfserr_no_grace;
7153
7154         if (nfsd4_client_record_check(cstate->clp))
7155                 return nfserr_reclaim_bad;
7156
7157         return nfs_ok;
7158 }
7159
7160 #ifdef CONFIG_NFSD_FAULT_INJECTION
7161 static inline void
7162 put_client(struct nfs4_client *clp)
7163 {
7164         atomic_dec(&clp->cl_rpc_users);
7165 }
7166
7167 static struct nfs4_client *
7168 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
7169 {
7170         struct nfs4_client *clp;
7171         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7172                                           nfsd_net_id);
7173
7174         if (!nfsd_netns_ready(nn))
7175                 return NULL;
7176
7177         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7178                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
7179                         return clp;
7180         }
7181         return NULL;
7182 }
7183
7184 u64
7185 nfsd_inject_print_clients(void)
7186 {
7187         struct nfs4_client *clp;
7188         u64 count = 0;
7189         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7190                                           nfsd_net_id);
7191         char buf[INET6_ADDRSTRLEN];
7192
7193         if (!nfsd_netns_ready(nn))
7194                 return 0;
7195
7196         spin_lock(&nn->client_lock);
7197         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7198                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7199                 pr_info("NFS Client: %s\n", buf);
7200                 ++count;
7201         }
7202         spin_unlock(&nn->client_lock);
7203
7204         return count;
7205 }
7206
7207 u64
7208 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7209 {
7210         u64 count = 0;
7211         struct nfs4_client *clp;
7212         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7213                                           nfsd_net_id);
7214
7215         if (!nfsd_netns_ready(nn))
7216                 return count;
7217
7218         spin_lock(&nn->client_lock);
7219         clp = nfsd_find_client(addr, addr_size);
7220         if (clp) {
7221                 if (mark_client_expired_locked(clp) == nfs_ok)
7222                         ++count;
7223                 else
7224                         clp = NULL;
7225         }
7226         spin_unlock(&nn->client_lock);
7227
7228         if (clp)
7229                 expire_client(clp);
7230
7231         return count;
7232 }
7233
7234 u64
7235 nfsd_inject_forget_clients(u64 max)
7236 {
7237         u64 count = 0;
7238         struct nfs4_client *clp, *next;
7239         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7240                                                 nfsd_net_id);
7241         LIST_HEAD(reaplist);
7242
7243         if (!nfsd_netns_ready(nn))
7244                 return count;
7245
7246         spin_lock(&nn->client_lock);
7247         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7248                 if (mark_client_expired_locked(clp) == nfs_ok) {
7249                         list_add(&clp->cl_lru, &reaplist);
7250                         if (max != 0 && ++count >= max)
7251                                 break;
7252                 }
7253         }
7254         spin_unlock(&nn->client_lock);
7255
7256         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7257                 expire_client(clp);
7258
7259         return count;
7260 }
7261
7262 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7263                              const char *type)
7264 {
7265         char buf[INET6_ADDRSTRLEN];
7266         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7267         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7268 }
7269
7270 static void
7271 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7272                              struct list_head *collect)
7273 {
7274         struct nfs4_client *clp = lst->st_stid.sc_client;
7275         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7276                                           nfsd_net_id);
7277
7278         if (!collect)
7279                 return;
7280
7281         lockdep_assert_held(&nn->client_lock);
7282         atomic_inc(&clp->cl_rpc_users);
7283         list_add(&lst->st_locks, collect);
7284 }
7285
7286 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7287                                     struct list_head *collect,
7288                                     bool (*func)(struct nfs4_ol_stateid *))
7289 {
7290         struct nfs4_openowner *oop;
7291         struct nfs4_ol_stateid *stp, *st_next;
7292         struct nfs4_ol_stateid *lst, *lst_next;
7293         u64 count = 0;
7294
7295         spin_lock(&clp->cl_lock);
7296         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7297                 list_for_each_entry_safe(stp, st_next,
7298                                 &oop->oo_owner.so_stateids, st_perstateowner) {
7299                         list_for_each_entry_safe(lst, lst_next,
7300                                         &stp->st_locks, st_locks) {
7301                                 if (func) {
7302                                         if (func(lst))
7303                                                 nfsd_inject_add_lock_to_list(lst,
7304                                                                         collect);
7305                                 }
7306                                 ++count;
7307                                 /*
7308                                  * Despite the fact that these functions deal
7309                                  * with 64-bit integers for "count", we must
7310                                  * ensure that it doesn't blow up the
7311                                  * clp->cl_rpc_users. Throw a warning if we
7312                                  * start to approach INT_MAX here.
7313                                  */
7314                                 WARN_ON_ONCE(count == (INT_MAX / 2));
7315                                 if (count == max)
7316                                         goto out;
7317                         }
7318                 }
7319         }
7320 out:
7321         spin_unlock(&clp->cl_lock);
7322
7323         return count;
7324 }
7325
7326 static u64
7327 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7328                           u64 max)
7329 {
7330         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7331 }
7332
7333 static u64
7334 nfsd_print_client_locks(struct nfs4_client *clp)
7335 {
7336         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7337         nfsd_print_count(clp, count, "locked files");
7338         return count;
7339 }
7340
7341 u64
7342 nfsd_inject_print_locks(void)
7343 {
7344         struct nfs4_client *clp;
7345         u64 count = 0;
7346         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7347                                                 nfsd_net_id);
7348
7349         if (!nfsd_netns_ready(nn))
7350                 return 0;
7351
7352         spin_lock(&nn->client_lock);
7353         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7354                 count += nfsd_print_client_locks(clp);
7355         spin_unlock(&nn->client_lock);
7356
7357         return count;
7358 }
7359
7360 static void
7361 nfsd_reap_locks(struct list_head *reaplist)
7362 {
7363         struct nfs4_client *clp;
7364         struct nfs4_ol_stateid *stp, *next;
7365
7366         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7367                 list_del_init(&stp->st_locks);
7368                 clp = stp->st_stid.sc_client;
7369                 nfs4_put_stid(&stp->st_stid);
7370                 put_client(clp);
7371         }
7372 }
7373
7374 u64
7375 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7376 {
7377         unsigned int count = 0;
7378         struct nfs4_client *clp;
7379         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7380                                                 nfsd_net_id);
7381         LIST_HEAD(reaplist);
7382
7383         if (!nfsd_netns_ready(nn))
7384                 return count;
7385
7386         spin_lock(&nn->client_lock);
7387         clp = nfsd_find_client(addr, addr_size);
7388         if (clp)
7389                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7390         spin_unlock(&nn->client_lock);
7391         nfsd_reap_locks(&reaplist);
7392         return count;
7393 }
7394
7395 u64
7396 nfsd_inject_forget_locks(u64 max)
7397 {
7398         u64 count = 0;
7399         struct nfs4_client *clp;
7400         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7401                                                 nfsd_net_id);
7402         LIST_HEAD(reaplist);
7403
7404         if (!nfsd_netns_ready(nn))
7405                 return count;
7406
7407         spin_lock(&nn->client_lock);
7408         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7409                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7410                 if (max != 0 && count >= max)
7411                         break;
7412         }
7413         spin_unlock(&nn->client_lock);
7414         nfsd_reap_locks(&reaplist);
7415         return count;
7416 }
7417
7418 static u64
7419 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7420                               struct list_head *collect,
7421                               void (*func)(struct nfs4_openowner *))
7422 {
7423         struct nfs4_openowner *oop, *next;
7424         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7425                                                 nfsd_net_id);
7426         u64 count = 0;
7427
7428         lockdep_assert_held(&nn->client_lock);
7429
7430         spin_lock(&clp->cl_lock);
7431         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7432                 if (func) {
7433                         func(oop);
7434                         if (collect) {
7435                                 atomic_inc(&clp->cl_rpc_users);
7436                                 list_add(&oop->oo_perclient, collect);
7437                         }
7438                 }
7439                 ++count;
7440                 /*
7441                  * Despite the fact that these functions deal with
7442                  * 64-bit integers for "count", we must ensure that
7443                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7444                  * warning if we start to approach INT_MAX here.
7445                  */
7446                 WARN_ON_ONCE(count == (INT_MAX / 2));
7447                 if (count == max)
7448                         break;
7449         }
7450         spin_unlock(&clp->cl_lock);
7451
7452         return count;
7453 }
7454
7455 static u64
7456 nfsd_print_client_openowners(struct nfs4_client *clp)
7457 {
7458         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7459
7460         nfsd_print_count(clp, count, "openowners");
7461         return count;
7462 }
7463
7464 static u64
7465 nfsd_collect_client_openowners(struct nfs4_client *clp,
7466                                struct list_head *collect, u64 max)
7467 {
7468         return nfsd_foreach_client_openowner(clp, max, collect,
7469                                                 unhash_openowner_locked);
7470 }
7471
7472 u64
7473 nfsd_inject_print_openowners(void)
7474 {
7475         struct nfs4_client *clp;
7476         u64 count = 0;
7477         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7478                                                 nfsd_net_id);
7479
7480         if (!nfsd_netns_ready(nn))
7481                 return 0;
7482
7483         spin_lock(&nn->client_lock);
7484         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7485                 count += nfsd_print_client_openowners(clp);
7486         spin_unlock(&nn->client_lock);
7487
7488         return count;
7489 }
7490
7491 static void
7492 nfsd_reap_openowners(struct list_head *reaplist)
7493 {
7494         struct nfs4_client *clp;
7495         struct nfs4_openowner *oop, *next;
7496
7497         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7498                 list_del_init(&oop->oo_perclient);
7499                 clp = oop->oo_owner.so_client;
7500                 release_openowner(oop);
7501                 put_client(clp);
7502         }
7503 }
7504
7505 u64
7506 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7507                                      size_t addr_size)
7508 {
7509         unsigned int count = 0;
7510         struct nfs4_client *clp;
7511         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7512                                                 nfsd_net_id);
7513         LIST_HEAD(reaplist);
7514
7515         if (!nfsd_netns_ready(nn))
7516                 return count;
7517
7518         spin_lock(&nn->client_lock);
7519         clp = nfsd_find_client(addr, addr_size);
7520         if (clp)
7521                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7522         spin_unlock(&nn->client_lock);
7523         nfsd_reap_openowners(&reaplist);
7524         return count;
7525 }
7526
7527 u64
7528 nfsd_inject_forget_openowners(u64 max)
7529 {
7530         u64 count = 0;
7531         struct nfs4_client *clp;
7532         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7533                                                 nfsd_net_id);
7534         LIST_HEAD(reaplist);
7535
7536         if (!nfsd_netns_ready(nn))
7537                 return count;
7538
7539         spin_lock(&nn->client_lock);
7540         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7541                 count += nfsd_collect_client_openowners(clp, &reaplist,
7542                                                         max - count);
7543                 if (max != 0 && count >= max)
7544                         break;
7545         }
7546         spin_unlock(&nn->client_lock);
7547         nfsd_reap_openowners(&reaplist);
7548         return count;
7549 }
7550
7551 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7552                                      struct list_head *victims)
7553 {
7554         struct nfs4_delegation *dp, *next;
7555         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7556                                                 nfsd_net_id);
7557         u64 count = 0;
7558
7559         lockdep_assert_held(&nn->client_lock);
7560
7561         spin_lock(&state_lock);
7562         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7563                 if (victims) {
7564                         /*
7565                          * It's not safe to mess with delegations that have a
7566                          * non-zero dl_time. They might have already been broken
7567                          * and could be processed by the laundromat outside of
7568                          * the state_lock. Just leave them be.
7569                          */
7570                         if (dp->dl_time != 0)
7571                                 continue;
7572
7573                         atomic_inc(&clp->cl_rpc_users);
7574                         WARN_ON(!unhash_delegation_locked(dp));
7575                         list_add(&dp->dl_recall_lru, victims);
7576                 }
7577                 ++count;
7578                 /*
7579                  * Despite the fact that these functions deal with
7580                  * 64-bit integers for "count", we must ensure that
7581                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7582                  * warning if we start to approach INT_MAX here.
7583                  */
7584                 WARN_ON_ONCE(count == (INT_MAX / 2));
7585                 if (count == max)
7586                         break;
7587         }
7588         spin_unlock(&state_lock);
7589         return count;
7590 }
7591
7592 static u64
7593 nfsd_print_client_delegations(struct nfs4_client *clp)
7594 {
7595         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7596
7597         nfsd_print_count(clp, count, "delegations");
7598         return count;
7599 }
7600
7601 u64
7602 nfsd_inject_print_delegations(void)
7603 {
7604         struct nfs4_client *clp;
7605         u64 count = 0;
7606         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7607                                                 nfsd_net_id);
7608
7609         if (!nfsd_netns_ready(nn))
7610                 return 0;
7611
7612         spin_lock(&nn->client_lock);
7613         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7614                 count += nfsd_print_client_delegations(clp);
7615         spin_unlock(&nn->client_lock);
7616
7617         return count;
7618 }
7619
7620 static void
7621 nfsd_forget_delegations(struct list_head *reaplist)
7622 {
7623         struct nfs4_client *clp;
7624         struct nfs4_delegation *dp, *next;
7625
7626         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7627                 list_del_init(&dp->dl_recall_lru);
7628                 clp = dp->dl_stid.sc_client;
7629                 revoke_delegation(dp);
7630                 put_client(clp);
7631         }
7632 }
7633
7634 u64
7635 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7636                                       size_t addr_size)
7637 {
7638         u64 count = 0;
7639         struct nfs4_client *clp;
7640         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7641                                                 nfsd_net_id);
7642         LIST_HEAD(reaplist);
7643
7644         if (!nfsd_netns_ready(nn))
7645                 return count;
7646
7647         spin_lock(&nn->client_lock);
7648         clp = nfsd_find_client(addr, addr_size);
7649         if (clp)
7650                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7651         spin_unlock(&nn->client_lock);
7652
7653         nfsd_forget_delegations(&reaplist);
7654         return count;
7655 }
7656
7657 u64
7658 nfsd_inject_forget_delegations(u64 max)
7659 {
7660         u64 count = 0;
7661         struct nfs4_client *clp;
7662         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7663                                                 nfsd_net_id);
7664         LIST_HEAD(reaplist);
7665
7666         if (!nfsd_netns_ready(nn))
7667                 return count;
7668
7669         spin_lock(&nn->client_lock);
7670         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7671                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7672                 if (max != 0 && count >= max)
7673                         break;
7674         }
7675         spin_unlock(&nn->client_lock);
7676         nfsd_forget_delegations(&reaplist);
7677         return count;
7678 }
7679
7680 static void
7681 nfsd_recall_delegations(struct list_head *reaplist)
7682 {
7683         struct nfs4_client *clp;
7684         struct nfs4_delegation *dp, *next;
7685
7686         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7687                 list_del_init(&dp->dl_recall_lru);
7688                 clp = dp->dl_stid.sc_client;
7689                 /*
7690                  * We skipped all entries that had a zero dl_time before,
7691                  * so we can now reset the dl_time back to 0. If a delegation
7692                  * break comes in now, then it won't make any difference since
7693                  * we're recalling it either way.
7694                  */
7695                 spin_lock(&state_lock);
7696                 dp->dl_time = 0;
7697                 spin_unlock(&state_lock);
7698                 nfsd_break_one_deleg(dp);
7699                 put_client(clp);
7700         }
7701 }
7702
7703 u64
7704 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7705                                       size_t addr_size)
7706 {
7707         u64 count = 0;
7708         struct nfs4_client *clp;
7709         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7710                                                 nfsd_net_id);
7711         LIST_HEAD(reaplist);
7712
7713         if (!nfsd_netns_ready(nn))
7714                 return count;
7715
7716         spin_lock(&nn->client_lock);
7717         clp = nfsd_find_client(addr, addr_size);
7718         if (clp)
7719                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7720         spin_unlock(&nn->client_lock);
7721
7722         nfsd_recall_delegations(&reaplist);
7723         return count;
7724 }
7725
7726 u64
7727 nfsd_inject_recall_delegations(u64 max)
7728 {
7729         u64 count = 0;
7730         struct nfs4_client *clp, *next;
7731         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7732                                                 nfsd_net_id);
7733         LIST_HEAD(reaplist);
7734
7735         if (!nfsd_netns_ready(nn))
7736                 return count;
7737
7738         spin_lock(&nn->client_lock);
7739         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7740                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7741                 if (max != 0 && ++count >= max)
7742                         break;
7743         }
7744         spin_unlock(&nn->client_lock);
7745         nfsd_recall_delegations(&reaplist);
7746         return count;
7747 }
7748 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7749
7750 /*
7751  * Since the lifetime of a delegation isn't limited to that of an open, a
7752  * client may quite reasonably hang on to a delegation as long as it has
7753  * the inode cached.  This becomes an obvious problem the first time a
7754  * client's inode cache approaches the size of the server's total memory.
7755  *
7756  * For now we avoid this problem by imposing a hard limit on the number
7757  * of delegations, which varies according to the server's memory size.
7758  */
7759 static void
7760 set_max_delegations(void)
7761 {
7762         /*
7763          * Allow at most 4 delegations per megabyte of RAM.  Quick
7764          * estimates suggest that in the worst case (where every delegation
7765          * is for a different inode), a delegation could take about 1.5K,
7766          * giving a worst case usage of about 6% of memory.
7767          */
7768         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7769 }
7770
7771 static int nfs4_state_create_net(struct net *net)
7772 {
7773         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7774         int i;
7775
7776         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7777                                             sizeof(struct list_head),
7778                                             GFP_KERNEL);
7779         if (!nn->conf_id_hashtbl)
7780                 goto err;
7781         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7782                                               sizeof(struct list_head),
7783                                               GFP_KERNEL);
7784         if (!nn->unconf_id_hashtbl)
7785                 goto err_unconf_id;
7786         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7787                                               sizeof(struct list_head),
7788                                               GFP_KERNEL);
7789         if (!nn->sessionid_hashtbl)
7790                 goto err_sessionid;
7791
7792         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7793                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7794                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7795         }
7796         for (i = 0; i < SESSION_HASH_SIZE; i++)
7797                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7798         nn->conf_name_tree = RB_ROOT;
7799         nn->unconf_name_tree = RB_ROOT;
7800         nn->boot_time = ktime_get_real_seconds();
7801         nn->grace_ended = false;
7802         nn->nfsd4_manager.block_opens = true;
7803         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7804         INIT_LIST_HEAD(&nn->client_lru);
7805         INIT_LIST_HEAD(&nn->close_lru);
7806         INIT_LIST_HEAD(&nn->del_recall_lru);
7807         spin_lock_init(&nn->client_lock);
7808         spin_lock_init(&nn->s2s_cp_lock);
7809         idr_init(&nn->s2s_cp_stateids);
7810
7811         spin_lock_init(&nn->blocked_locks_lock);
7812         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7813
7814         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7815         get_net(net);
7816
7817         return 0;
7818
7819 err_sessionid:
7820         kfree(nn->unconf_id_hashtbl);
7821 err_unconf_id:
7822         kfree(nn->conf_id_hashtbl);
7823 err:
7824         return -ENOMEM;
7825 }
7826
7827 static void
7828 nfs4_state_destroy_net(struct net *net)
7829 {
7830         int i;
7831         struct nfs4_client *clp = NULL;
7832         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7833
7834         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7835                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7836                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7837                         destroy_client(clp);
7838                 }
7839         }
7840
7841         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7842
7843         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7844                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7845                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7846                         destroy_client(clp);
7847                 }
7848         }
7849
7850         kfree(nn->sessionid_hashtbl);
7851         kfree(nn->unconf_id_hashtbl);
7852         kfree(nn->conf_id_hashtbl);
7853         put_net(net);
7854 }
7855
7856 int
7857 nfs4_state_start_net(struct net *net)
7858 {
7859         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7860         int ret;
7861
7862         ret = nfs4_state_create_net(net);
7863         if (ret)
7864                 return ret;
7865         locks_start_grace(net, &nn->nfsd4_manager);
7866         nfsd4_client_tracking_init(net);
7867         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7868                 goto skip_grace;
7869         printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
7870                nn->nfsd4_grace, net->ns.inum);
7871         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7872         return 0;
7873
7874 skip_grace:
7875         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7876                         net->ns.inum);
7877         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7878         nfsd4_end_grace(nn);
7879         return 0;
7880 }
7881
7882 /* initialization to perform when the nfsd service is started: */
7883
7884 int
7885 nfs4_state_start(void)
7886 {
7887         int ret;
7888
7889         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7890         if (laundry_wq == NULL) {
7891                 ret = -ENOMEM;
7892                 goto out;
7893         }
7894         ret = nfsd4_create_callback_queue();
7895         if (ret)
7896                 goto out_free_laundry;
7897
7898         set_max_delegations();
7899         return 0;
7900
7901 out_free_laundry:
7902         destroy_workqueue(laundry_wq);
7903 out:
7904         return ret;
7905 }
7906
7907 void
7908 nfs4_state_shutdown_net(struct net *net)
7909 {
7910         struct nfs4_delegation *dp = NULL;
7911         struct list_head *pos, *next, reaplist;
7912         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7913
7914         cancel_delayed_work_sync(&nn->laundromat_work);
7915         locks_end_grace(&nn->nfsd4_manager);
7916
7917         INIT_LIST_HEAD(&reaplist);
7918         spin_lock(&state_lock);
7919         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7920                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7921                 WARN_ON(!unhash_delegation_locked(dp));
7922                 list_add(&dp->dl_recall_lru, &reaplist);
7923         }
7924         spin_unlock(&state_lock);
7925         list_for_each_safe(pos, next, &reaplist) {
7926                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7927                 list_del_init(&dp->dl_recall_lru);
7928                 destroy_unhashed_deleg(dp);
7929         }
7930
7931         nfsd4_client_tracking_exit(net);
7932         nfs4_state_destroy_net(net);
7933 }
7934
7935 void
7936 nfs4_state_shutdown(void)
7937 {
7938         destroy_workqueue(laundry_wq);
7939         nfsd4_destroy_callback_queue();
7940 }
7941
7942 static void
7943 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7944 {
7945         if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
7946             CURRENT_STATEID(stateid))
7947                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7948 }
7949
7950 static void
7951 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7952 {
7953         if (cstate->minorversion) {
7954                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7955                 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7956         }
7957 }
7958
7959 void
7960 clear_current_stateid(struct nfsd4_compound_state *cstate)
7961 {
7962         CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7963 }
7964
7965 /*
7966  * functions to set current state id
7967  */
7968 void
7969 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7970                 union nfsd4_op_u *u)
7971 {
7972         put_stateid(cstate, &u->open_downgrade.od_stateid);
7973 }
7974
7975 void
7976 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7977                 union nfsd4_op_u *u)
7978 {
7979         put_stateid(cstate, &u->open.op_stateid);
7980 }
7981
7982 void
7983 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7984                 union nfsd4_op_u *u)
7985 {
7986         put_stateid(cstate, &u->close.cl_stateid);
7987 }
7988
7989 void
7990 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7991                 union nfsd4_op_u *u)
7992 {
7993         put_stateid(cstate, &u->lock.lk_resp_stateid);
7994 }
7995
7996 /*
7997  * functions to consume current state id
7998  */
7999
8000 void
8001 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8002                 union nfsd4_op_u *u)
8003 {
8004         get_stateid(cstate, &u->open_downgrade.od_stateid);
8005 }
8006
8007 void
8008 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8009                 union nfsd4_op_u *u)
8010 {
8011         get_stateid(cstate, &u->delegreturn.dr_stateid);
8012 }
8013
8014 void
8015 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8016                 union nfsd4_op_u *u)
8017 {
8018         get_stateid(cstate, &u->free_stateid.fr_stateid);
8019 }
8020
8021 void
8022 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8023                 union nfsd4_op_u *u)
8024 {
8025         get_stateid(cstate, &u->setattr.sa_stateid);
8026 }
8027
8028 void
8029 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8030                 union nfsd4_op_u *u)
8031 {
8032         get_stateid(cstate, &u->close.cl_stateid);
8033 }
8034
8035 void
8036 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8037                 union nfsd4_op_u *u)
8038 {
8039         get_stateid(cstate, &u->locku.lu_stateid);
8040 }
8041
8042 void
8043 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8044                 union nfsd4_op_u *u)
8045 {
8046         get_stateid(cstate, &u->read.rd_stateid);
8047 }
8048
8049 void
8050 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8051                 union nfsd4_op_u *u)
8052 {
8053         get_stateid(cstate, &u->write.wr_stateid);
8054 }