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
  54 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  55
  56 #define all_ones {{~0,~0},~0}
  57 static const stateid_t one_stateid = {
  58         .si_generation = ~0,
  59         .si_opaque = all_ones,
  60 };
  61 static const stateid_t zero_stateid = {
  62         /* all fields zero */
  63 };
  64 static const stateid_t currentstateid = {
  65         .si_generation = 1,
  66 };
  67 static const stateid_t close_stateid = {
  68         .si_generation = 0xffffffffU,
  69 };
  70
  71 static u64 current_sessionid = 1;
  72
  73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  74 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  75 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  76 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  77
  78 /* forward declarations */
  79 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  80 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  81 void nfsd4_end_grace(struct nfsd_net *nn);
  82
  83 /* Locking: */
  84
  85 /*
  86  * Currently used for the del_recall_lru and file hash table.  In an
  87  * effort to decrease the scope of the client_mutex, this spinlock may
  88  * eventually cover more:
  89  */
  90 static DEFINE_SPINLOCK(state_lock);
  91
  92 enum nfsd4_st_mutex_lock_subclass {
  93         OPEN_STATEID_MUTEX = 0,
  94         LOCK_STATEID_MUTEX = 1,
  95 };
  96
  97 /*
  98  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  99  * the refcount on the open stateid to drop.
 100  */
 101 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 102
 103 /*
 104  * A waitqueue where a writer to clients/#/ctl destroying a client can
 105  * wait for cl_rpc_users to drop to 0 and then for the client to be
 106  * unhashed.
 107  */
 108 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 109
 110 static struct kmem_cache *client_slab;
 111 static struct kmem_cache *openowner_slab;
 112 static struct kmem_cache *lockowner_slab;
 113 static struct kmem_cache *file_slab;
 114 static struct kmem_cache *stateid_slab;
 115 static struct kmem_cache *deleg_slab;
 116 static struct kmem_cache *odstate_slab;
 117
 118 static void free_session(struct nfsd4_session *);
 119
 120 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 121 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 122
 123 static bool is_session_dead(struct nfsd4_session *ses)
 124 {
 125         return ses->se_flags & NFS4_SESSION_DEAD;
 126 }
 127
 128 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 129 {
 130         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 131                 return nfserr_jukebox;
 132         ses->se_flags |= NFS4_SESSION_DEAD;
 133         return nfs_ok;
 134 }
 135
 136 static bool is_client_expired(struct nfs4_client *clp)
 137 {
 138         return clp->cl_time == 0;
 139 }
 140
 141 static __be32 get_client_locked(struct nfs4_client *clp)
 142 {
 143         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 144
 145         lockdep_assert_held(&nn->client_lock);
 146
 147         if (is_client_expired(clp))
 148                 return nfserr_expired;
 149         atomic_inc(&clp->cl_rpc_users);
 150         return nfs_ok;
 151 }
 152
 153 /* must be called under the client_lock */
 154 static inline void
 155 renew_client_locked(struct nfs4_client *clp)
 156 {
 157         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 158
 159         if (is_client_expired(clp)) {
 160                 WARN_ON(1);
 161                 printk("%s: client (clientid %08x/%08x) already expired\n",
 162                         __func__,
 163                         clp->cl_clientid.cl_boot,
 164                         clp->cl_clientid.cl_id);
 165                 return;
 166         }
 167
 168         dprintk("renewing client (clientid %08x/%08x)\n",
 169                         clp->cl_clientid.cl_boot,
 170                         clp->cl_clientid.cl_id);
 171         list_move_tail(&clp->cl_lru, &nn->client_lru);
 172         clp->cl_time = get_seconds();
 173 }
 174
 175 static void put_client_renew_locked(struct nfs4_client *clp)
 176 {
 177         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 178
 179         lockdep_assert_held(&nn->client_lock);
 180
 181         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 182                 return;
 183         if (!is_client_expired(clp))
 184                 renew_client_locked(clp);
 185         else
 186                 wake_up_all(&expiry_wq);
 187 }
 188
 189 static void put_client_renew(struct nfs4_client *clp)
 190 {
 191         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 192
 193         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 194                 return;
 195         if (!is_client_expired(clp))
 196                 renew_client_locked(clp);
 197         else
 198                 wake_up_all(&expiry_wq);
 199         spin_unlock(&nn->client_lock);
 200 }
 201
 202 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 203 {
 204         __be32 status;
 205
 206         if (is_session_dead(ses))
 207                 return nfserr_badsession;
 208         status = get_client_locked(ses->se_client);
 209         if (status)
 210                 return status;
 211         atomic_inc(&ses->se_ref);
 212         return nfs_ok;
 213 }
 214
 215 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 216 {
 217         struct nfs4_client *clp = ses->se_client;
 218         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 219
 220         lockdep_assert_held(&nn->client_lock);
 221
 222         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 223                 free_session(ses);
 224         put_client_renew_locked(clp);
 225 }
 226
 227 static void nfsd4_put_session(struct nfsd4_session *ses)
 228 {
 229         struct nfs4_client *clp = ses->se_client;
 230         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 231
 232         spin_lock(&nn->client_lock);
 233         nfsd4_put_session_locked(ses);
 234         spin_unlock(&nn->client_lock);
 235 }
 236
 237 static struct nfsd4_blocked_lock *
 238 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 239                         struct nfsd_net *nn)
 240 {
 241         struct nfsd4_blocked_lock *cur, *found = NULL;
 242
 243         spin_lock(&nn->blocked_locks_lock);
 244         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 245                 if (fh_match(fh, &cur->nbl_fh)) {
 246                         list_del_init(&cur->nbl_list);
 247                         list_del_init(&cur->nbl_lru);
 248                         found = cur;
 249                         break;
 250                 }
 251         }
 252         spin_unlock(&nn->blocked_locks_lock);
 253         if (found)
 254                 locks_delete_block(&found->nbl_lock);
 255         return found;
 256 }
 257
 258 static struct nfsd4_blocked_lock *
 259 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 260                         struct nfsd_net *nn)
 261 {
 262         struct nfsd4_blocked_lock *nbl;
 263
 264         nbl = find_blocked_lock(lo, fh, nn);
 265         if (!nbl) {
 266                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 267                 if (nbl) {
 268                         fh_copy_shallow(&nbl->nbl_fh, fh);
 269                         locks_init_lock(&nbl->nbl_lock);
 270                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 271                                         &nfsd4_cb_notify_lock_ops,
 272                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 273                 }
 274         }
 275         return nbl;
 276 }
 277
 278 static void
 279 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 280 {
 281         locks_delete_block(&nbl->nbl_lock);
 282         locks_release_private(&nbl->nbl_lock);
 283         kfree(nbl);
 284 }
 285
 286 static void
 287 remove_blocked_locks(struct nfs4_lockowner *lo)
 288 {
 289         struct nfs4_client *clp = lo->lo_owner.so_client;
 290         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 291         struct nfsd4_blocked_lock *nbl;
 292         LIST_HEAD(reaplist);
 293
 294         /* Dequeue all blocked locks */
 295         spin_lock(&nn->blocked_locks_lock);
 296         while (!list_empty(&lo->lo_blocked)) {
 297                 nbl = list_first_entry(&lo->lo_blocked,
 298                                         struct nfsd4_blocked_lock,
 299                                         nbl_list);
 300                 list_del_init(&nbl->nbl_list);
 301                 list_move(&nbl->nbl_lru, &reaplist);
 302         }
 303         spin_unlock(&nn->blocked_locks_lock);
 304
 305         /* Now free them */
 306         while (!list_empty(&reaplist)) {
 307                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 308                                         nbl_lru);
 309                 list_del_init(&nbl->nbl_lru);
 310                 free_blocked_lock(nbl);
 311         }
 312 }
 313
 314 static void
 315 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 316 {
 317         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 318                                                 struct nfsd4_blocked_lock, nbl_cb);
 319         locks_delete_block(&nbl->nbl_lock);
 320 }
 321
 322 static int
 323 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 324 {
 325         /*
 326          * Since this is just an optimization, we don't try very hard if it
 327          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 328          * just quit trying on anything else.
 329          */
 330         switch (task->tk_status) {
 331         case -NFS4ERR_DELAY:
 332                 rpc_delay(task, 1 * HZ);
 333                 return 0;
 334         default:
 335                 return 1;
 336         }
 337 }
 338
 339 static void
 340 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 341 {
 342         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 343                                                 struct nfsd4_blocked_lock, nbl_cb);
 344
 345         free_blocked_lock(nbl);
 346 }
 347
 348 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 349         .prepare        = nfsd4_cb_notify_lock_prepare,
 350         .done           = nfsd4_cb_notify_lock_done,
 351         .release        = nfsd4_cb_notify_lock_release,
 352 };
 353
 354 static inline struct nfs4_stateowner *
 355 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 356 {
 357         atomic_inc(&sop->so_count);
 358         return sop;
 359 }
 360
 361 static int
 362 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 363 {
 364         return (sop->so_owner.len == owner->len) &&
 365                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 366 }
 367
 368 static struct nfs4_openowner *
 369 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 370                         struct nfs4_client *clp)
 371 {
 372         struct nfs4_stateowner *so;
 373
 374         lockdep_assert_held(&clp->cl_lock);
 375
 376         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 377                             so_strhash) {
 378                 if (!so->so_is_open_owner)
 379                         continue;
 380                 if (same_owner_str(so, &open->op_owner))
 381                         return openowner(nfs4_get_stateowner(so));
 382         }
 383         return NULL;
 384 }
 385
 386 static struct nfs4_openowner *
 387 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 388                         struct nfs4_client *clp)
 389 {
 390         struct nfs4_openowner *oo;
 391
 392         spin_lock(&clp->cl_lock);
 393         oo = find_openstateowner_str_locked(hashval, open, clp);
 394         spin_unlock(&clp->cl_lock);
 395         return oo;
 396 }
 397
 398 static inline u32
 399 opaque_hashval(const void *ptr, int nbytes)
 400 {
 401         unsigned char *cptr = (unsigned char *) ptr;
 402
 403         u32 x = 0;
 404         while (nbytes--) {
 405                 x *= 37;
 406                 x += *cptr++;
 407         }
 408         return x;
 409 }
 410
 411 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 412 {
 413         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 414
 415         kmem_cache_free(file_slab, fp);
 416 }
 417
 418 void
 419 put_nfs4_file(struct nfs4_file *fi)
 420 {
 421         might_lock(&state_lock);
 422
 423         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 424                 hlist_del_rcu(&fi->fi_hash);
 425                 spin_unlock(&state_lock);
 426                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 427                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 428                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 429         }
 430 }
 431
 432 static struct file *
 433 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 434 {
 435         if (f->fi_fds[oflag])
 436                 return get_file(f->fi_fds[oflag]);
 437         return NULL;
 438 }
 439
 440 static struct file *
 441 find_writeable_file_locked(struct nfs4_file *f)
 442 {
 443         struct file *ret;
 444
 445         lockdep_assert_held(&f->fi_lock);
 446
 447         ret = __nfs4_get_fd(f, O_WRONLY);
 448         if (!ret)
 449                 ret = __nfs4_get_fd(f, O_RDWR);
 450         return ret;
 451 }
 452
 453 static struct file *
 454 find_writeable_file(struct nfs4_file *f)
 455 {
 456         struct file *ret;
 457
 458         spin_lock(&f->fi_lock);
 459         ret = find_writeable_file_locked(f);
 460         spin_unlock(&f->fi_lock);
 461
 462         return ret;
 463 }
 464
 465 static struct file *find_readable_file_locked(struct nfs4_file *f)
 466 {
 467         struct file *ret;
 468
 469         lockdep_assert_held(&f->fi_lock);
 470
 471         ret = __nfs4_get_fd(f, O_RDONLY);
 472         if (!ret)
 473                 ret = __nfs4_get_fd(f, O_RDWR);
 474         return ret;
 475 }
 476
 477 static struct file *
 478 find_readable_file(struct nfs4_file *f)
 479 {
 480         struct file *ret;
 481
 482         spin_lock(&f->fi_lock);
 483         ret = find_readable_file_locked(f);
 484         spin_unlock(&f->fi_lock);
 485
 486         return ret;
 487 }
 488
 489 struct file *
 490 find_any_file(struct nfs4_file *f)
 491 {
 492         struct file *ret;
 493
 494         spin_lock(&f->fi_lock);
 495         ret = __nfs4_get_fd(f, O_RDWR);
 496         if (!ret) {
 497                 ret = __nfs4_get_fd(f, O_WRONLY);
 498                 if (!ret)
 499                         ret = __nfs4_get_fd(f, O_RDONLY);
 500         }
 501         spin_unlock(&f->fi_lock);
 502         return ret;
 503 }
 504
 505 static atomic_long_t num_delegations;
 506 unsigned long max_delegations;
 507
 508 /*
 509  * Open owner state (share locks)
 510  */
 511
 512 /* hash tables for lock and open owners */
 513 #define OWNER_HASH_BITS              8
 514 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 515 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 516
 517 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 518 {
 519         unsigned int ret;
 520
 521         ret = opaque_hashval(ownername->data, ownername->len);
 522         return ret & OWNER_HASH_MASK;
 523 }
 524
 525 /* hash table for nfs4_file */
 526 #define FILE_HASH_BITS                   8
 527 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 528
 529 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 530 {
 531         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 532 }
 533
 534 static unsigned int file_hashval(struct knfsd_fh *fh)
 535 {
 536         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 537 }
 538
 539 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 540
 541 static void
 542 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 543 {
 544         lockdep_assert_held(&fp->fi_lock);
 545
 546         if (access & NFS4_SHARE_ACCESS_WRITE)
 547                 atomic_inc(&fp->fi_access[O_WRONLY]);
 548         if (access & NFS4_SHARE_ACCESS_READ)
 549                 atomic_inc(&fp->fi_access[O_RDONLY]);
 550 }
 551
 552 static __be32
 553 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 554 {
 555         lockdep_assert_held(&fp->fi_lock);
 556
 557         /* Does this access mode make sense? */
 558         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 559                 return nfserr_inval;
 560
 561         /* Does it conflict with a deny mode already set? */
 562         if ((access & fp->fi_share_deny) != 0)
 563                 return nfserr_share_denied;
 564
 565         __nfs4_file_get_access(fp, access);
 566         return nfs_ok;
 567 }
 568
 569 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 570 {
 571         /* Common case is that there is no deny mode. */
 572         if (deny) {
 573                 /* Does this deny mode make sense? */
 574                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 575                         return nfserr_inval;
 576
 577                 if ((deny & NFS4_SHARE_DENY_READ) &&
 578                     atomic_read(&fp->fi_access[O_RDONLY]))
 579                         return nfserr_share_denied;
 580
 581                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 582                     atomic_read(&fp->fi_access[O_WRONLY]))
 583                         return nfserr_share_denied;
 584         }
 585         return nfs_ok;
 586 }
 587
 588 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 589 {
 590         might_lock(&fp->fi_lock);
 591
 592         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 593                 struct file *f1 = NULL;
 594                 struct file *f2 = NULL;
 595
 596                 swap(f1, fp->fi_fds[oflag]);
 597                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 598                         swap(f2, fp->fi_fds[O_RDWR]);
 599                 spin_unlock(&fp->fi_lock);
 600                 if (f1)
 601                         fput(f1);
 602                 if (f2)
 603                         fput(f2);
 604         }
 605 }
 606
 607 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 608 {
 609         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 610
 611         if (access & NFS4_SHARE_ACCESS_WRITE)
 612                 __nfs4_file_put_access(fp, O_WRONLY);
 613         if (access & NFS4_SHARE_ACCESS_READ)
 614                 __nfs4_file_put_access(fp, O_RDONLY);
 615 }
 616
 617 /*
 618  * Allocate a new open/delegation state counter. This is needed for
 619  * pNFS for proper return on close semantics.
 620  *
 621  * Note that we only allocate it for pNFS-enabled exports, otherwise
 622  * all pointers to struct nfs4_clnt_odstate are always NULL.
 623  */
 624 static struct nfs4_clnt_odstate *
 625 alloc_clnt_odstate(struct nfs4_client *clp)
 626 {
 627         struct nfs4_clnt_odstate *co;
 628
 629         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 630         if (co) {
 631                 co->co_client = clp;
 632                 refcount_set(&co->co_odcount, 1);
 633         }
 634         return co;
 635 }
 636
 637 static void
 638 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 639 {
 640         struct nfs4_file *fp = co->co_file;
 641
 642         lockdep_assert_held(&fp->fi_lock);
 643         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 644 }
 645
 646 static inline void
 647 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 648 {
 649         if (co)
 650                 refcount_inc(&co->co_odcount);
 651 }
 652
 653 static void
 654 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 655 {
 656         struct nfs4_file *fp;
 657
 658         if (!co)
 659                 return;
 660
 661         fp = co->co_file;
 662         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 663                 list_del(&co->co_perfile);
 664                 spin_unlock(&fp->fi_lock);
 665
 666                 nfsd4_return_all_file_layouts(co->co_client, fp);
 667                 kmem_cache_free(odstate_slab, co);
 668         }
 669 }
 670
 671 static struct nfs4_clnt_odstate *
 672 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 673 {
 674         struct nfs4_clnt_odstate *co;
 675         struct nfs4_client *cl;
 676
 677         if (!new)
 678                 return NULL;
 679
 680         cl = new->co_client;
 681
 682         spin_lock(&fp->fi_lock);
 683         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 684                 if (co->co_client == cl) {
 685                         get_clnt_odstate(co);
 686                         goto out;
 687                 }
 688         }
 689         co = new;
 690         co->co_file = fp;
 691         hash_clnt_odstate_locked(new);
 692 out:
 693         spin_unlock(&fp->fi_lock);
 694         return co;
 695 }
 696
 697 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 698                                   void (*sc_free)(struct nfs4_stid *))
 699 {
 700         struct nfs4_stid *stid;
 701         int new_id;
 702
 703         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 704         if (!stid)
 705                 return NULL;
 706
 707         idr_preload(GFP_KERNEL);
 708         spin_lock(&cl->cl_lock);
 709         /* Reserving 0 for start of file in nfsdfs "states" file: */
 710         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 711         spin_unlock(&cl->cl_lock);
 712         idr_preload_end();
 713         if (new_id < 0)
 714                 goto out_free;
 715
 716         stid->sc_free = sc_free;
 717         stid->sc_client = cl;
 718         stid->sc_stateid.si_opaque.so_id = new_id;
 719         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 720         /* Will be incremented before return to client: */
 721         refcount_set(&stid->sc_count, 1);
 722         spin_lock_init(&stid->sc_lock);
 723
 724         /*
 725          * It shouldn't be a problem to reuse an opaque stateid value.
 726          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 727          * example, a stray write retransmission could be accepted by
 728          * the server when it should have been rejected.  Therefore,
 729          * adopt a trick from the sctp code to attempt to maximize the
 730          * amount of time until an id is reused, by ensuring they always
 731          * "increase" (mod INT_MAX):
 732          */
 733         return stid;
 734 out_free:
 735         kmem_cache_free(slab, stid);
 736         return NULL;
 737 }
 738
 739 /*
 740  * Create a unique stateid_t to represent each COPY.
 741  */
 742 int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 743 {
 744         int new_id;
 745
 746         idr_preload(GFP_KERNEL);
 747         spin_lock(&nn->s2s_cp_lock);
 748         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, copy, 0, 0, GFP_NOWAIT);
 749         spin_unlock(&nn->s2s_cp_lock);
 750         idr_preload_end();
 751         if (new_id < 0)
 752                 return 0;
 753         copy->cp_stateid.si_opaque.so_id = new_id;
 754         copy->cp_stateid.si_opaque.so_clid.cl_boot = nn->boot_time;
 755         copy->cp_stateid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 756         return 1;
 757 }
 758
 759 void nfs4_free_cp_state(struct nfsd4_copy *copy)
 760 {
 761         struct nfsd_net *nn;
 762
 763         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 764         spin_lock(&nn->s2s_cp_lock);
 765         idr_remove(&nn->s2s_cp_stateids, copy->cp_stateid.si_opaque.so_id);
 766         spin_unlock(&nn->s2s_cp_lock);
 767 }
 768
 769 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 770 {
 771         struct nfs4_stid *stid;
 772
 773         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 774         if (!stid)
 775                 return NULL;
 776
 777         return openlockstateid(stid);
 778 }
 779
 780 static void nfs4_free_deleg(struct nfs4_stid *stid)
 781 {
 782         kmem_cache_free(deleg_slab, stid);
 783         atomic_long_dec(&num_delegations);
 784 }
 785
 786 /*
 787  * When we recall a delegation, we should be careful not to hand it
 788  * out again straight away.
 789  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 790  * in which the filehandles of recalled delegations are "stored".
 791  * If a filehandle appear in either filter, a delegation is blocked.
 792  * When a delegation is recalled, the filehandle is stored in the "new"
 793  * filter.
 794  * Every 30 seconds we swap the filters and clear the "new" one,
 795  * unless both are empty of course.
 796  *
 797  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 798  * low 3 bytes as hash-table indices.
 799  *
 800  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 801  * is used to manage concurrent access.  Testing does not need the lock
 802  * except when swapping the two filters.
 803  */
 804 static DEFINE_SPINLOCK(blocked_delegations_lock);
 805 static struct bloom_pair {
 806         int     entries, old_entries;
 807         time_t  swap_time;
 808         int     new; /* index into 'set' */
 809         DECLARE_BITMAP(set[2], 256);
 810 } blocked_delegations;
 811
 812 static int delegation_blocked(struct knfsd_fh *fh)
 813 {
 814         u32 hash;
 815         struct bloom_pair *bd = &blocked_delegations;
 816
 817         if (bd->entries == 0)
 818                 return 0;
 819         if (seconds_since_boot() - bd->swap_time > 30) {
 820                 spin_lock(&blocked_delegations_lock);
 821                 if (seconds_since_boot() - bd->swap_time > 30) {
 822                         bd->entries -= bd->old_entries;
 823                         bd->old_entries = bd->entries;
 824                         memset(bd->set[bd->new], 0,
 825                                sizeof(bd->set[0]));
 826                         bd->new = 1-bd->new;
 827                         bd->swap_time = seconds_since_boot();
 828                 }
 829                 spin_unlock(&blocked_delegations_lock);
 830         }
 831         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 832         if (test_bit(hash&255, bd->set[0]) &&
 833             test_bit((hash>>8)&255, bd->set[0]) &&
 834             test_bit((hash>>16)&255, bd->set[0]))
 835                 return 1;
 836
 837         if (test_bit(hash&255, bd->set[1]) &&
 838             test_bit((hash>>8)&255, bd->set[1]) &&
 839             test_bit((hash>>16)&255, bd->set[1]))
 840                 return 1;
 841
 842         return 0;
 843 }
 844
 845 static void block_delegations(struct knfsd_fh *fh)
 846 {
 847         u32 hash;
 848         struct bloom_pair *bd = &blocked_delegations;
 849
 850         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 851
 852         spin_lock(&blocked_delegations_lock);
 853         __set_bit(hash&255, bd->set[bd->new]);
 854         __set_bit((hash>>8)&255, bd->set[bd->new]);
 855         __set_bit((hash>>16)&255, bd->set[bd->new]);
 856         if (bd->entries == 0)
 857                 bd->swap_time = seconds_since_boot();
 858         bd->entries += 1;
 859         spin_unlock(&blocked_delegations_lock);
 860 }
 861
 862 static struct nfs4_delegation *
 863 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 864                  struct svc_fh *current_fh,
 865                  struct nfs4_clnt_odstate *odstate)
 866 {
 867         struct nfs4_delegation *dp;
 868         long n;
 869
 870         dprintk("NFSD alloc_init_deleg\n");
 871         n = atomic_long_inc_return(&num_delegations);
 872         if (n < 0 || n > max_delegations)
 873                 goto out_dec;
 874         if (delegation_blocked(&current_fh->fh_handle))
 875                 goto out_dec;
 876         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 877         if (dp == NULL)
 878                 goto out_dec;
 879
 880         /*
 881          * delegation seqid's are never incremented.  The 4.1 special
 882          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 883          * 0 anyway just for consistency and use 1:
 884          */
 885         dp->dl_stid.sc_stateid.si_generation = 1;
 886         INIT_LIST_HEAD(&dp->dl_perfile);
 887         INIT_LIST_HEAD(&dp->dl_perclnt);
 888         INIT_LIST_HEAD(&dp->dl_recall_lru);
 889         dp->dl_clnt_odstate = odstate;
 890         get_clnt_odstate(odstate);
 891         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 892         dp->dl_retries = 1;
 893         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 894                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 895         get_nfs4_file(fp);
 896         dp->dl_stid.sc_file = fp;
 897         return dp;
 898 out_dec:
 899         atomic_long_dec(&num_delegations);
 900         return NULL;
 901 }
 902
 903 void
 904 nfs4_put_stid(struct nfs4_stid *s)
 905 {
 906         struct nfs4_file *fp = s->sc_file;
 907         struct nfs4_client *clp = s->sc_client;
 908
 909         might_lock(&clp->cl_lock);
 910
 911         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 912                 wake_up_all(&close_wq);
 913                 return;
 914         }
 915         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 916         spin_unlock(&clp->cl_lock);
 917         s->sc_free(s);
 918         if (fp)
 919                 put_nfs4_file(fp);
 920 }
 921
 922 void
 923 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 924 {
 925         stateid_t *src = &stid->sc_stateid;
 926
 927         spin_lock(&stid->sc_lock);
 928         if (unlikely(++src->si_generation == 0))
 929                 src->si_generation = 1;
 930         memcpy(dst, src, sizeof(*dst));
 931         spin_unlock(&stid->sc_lock);
 932 }
 933
 934 static void put_deleg_file(struct nfs4_file *fp)
 935 {
 936         struct file *filp = NULL;
 937
 938         spin_lock(&fp->fi_lock);
 939         if (--fp->fi_delegees == 0)
 940                 swap(filp, fp->fi_deleg_file);
 941         spin_unlock(&fp->fi_lock);
 942
 943         if (filp)
 944                 fput(filp);
 945 }
 946
 947 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 948 {
 949         struct nfs4_file *fp = dp->dl_stid.sc_file;
 950         struct file *filp = fp->fi_deleg_file;
 951
 952         WARN_ON_ONCE(!fp->fi_delegees);
 953
 954         vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
 955         put_deleg_file(fp);
 956 }
 957
 958 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 959 {
 960         put_clnt_odstate(dp->dl_clnt_odstate);
 961         nfs4_unlock_deleg_lease(dp);
 962         nfs4_put_stid(&dp->dl_stid);
 963 }
 964
 965 void nfs4_unhash_stid(struct nfs4_stid *s)
 966 {
 967         s->sc_type = 0;
 968 }
 969
 970 /**
 971  * nfs4_delegation_exists - Discover if this delegation already exists
 972  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 973  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 974  *
 975  * Return:
 976  *      On success: true iff an existing delegation is found
 977  */
 978
 979 static bool
 980 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 981 {
 982         struct nfs4_delegation *searchdp = NULL;
 983         struct nfs4_client *searchclp = NULL;
 984
 985         lockdep_assert_held(&state_lock);
 986         lockdep_assert_held(&fp->fi_lock);
 987
 988         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 989                 searchclp = searchdp->dl_stid.sc_client;
 990                 if (clp == searchclp) {
 991                         return true;
 992                 }
 993         }
 994         return false;
 995 }
 996
 997 /**
 998  * hash_delegation_locked - Add a delegation to the appropriate lists
 999  * @dp:     a pointer to the nfs4_delegation we are adding.
1000  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1001  *
1002  * Return:
1003  *      On success: NULL if the delegation was successfully hashed.
1004  *
1005  *      On error: -EAGAIN if one was previously granted to this
1006  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1007  *
1008  */
1009
1010 static int
1011 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1012 {
1013         struct nfs4_client *clp = dp->dl_stid.sc_client;
1014
1015         lockdep_assert_held(&state_lock);
1016         lockdep_assert_held(&fp->fi_lock);
1017
1018         if (nfs4_delegation_exists(clp, fp))
1019                 return -EAGAIN;
1020         refcount_inc(&dp->dl_stid.sc_count);
1021         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1022         list_add(&dp->dl_perfile, &fp->fi_delegations);
1023         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1024         return 0;
1025 }
1026
1027 static bool
1028 unhash_delegation_locked(struct nfs4_delegation *dp)
1029 {
1030         struct nfs4_file *fp = dp->dl_stid.sc_file;
1031
1032         lockdep_assert_held(&state_lock);
1033
1034         if (list_empty(&dp->dl_perfile))
1035                 return false;
1036
1037         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1038         /* Ensure that deleg break won't try to requeue it */
1039         ++dp->dl_time;
1040         spin_lock(&fp->fi_lock);
1041         list_del_init(&dp->dl_perclnt);
1042         list_del_init(&dp->dl_recall_lru);
1043         list_del_init(&dp->dl_perfile);
1044         spin_unlock(&fp->fi_lock);
1045         return true;
1046 }
1047
1048 static void destroy_delegation(struct nfs4_delegation *dp)
1049 {
1050         bool unhashed;
1051
1052         spin_lock(&state_lock);
1053         unhashed = unhash_delegation_locked(dp);
1054         spin_unlock(&state_lock);
1055         if (unhashed)
1056                 destroy_unhashed_deleg(dp);
1057 }
1058
1059 static void revoke_delegation(struct nfs4_delegation *dp)
1060 {
1061         struct nfs4_client *clp = dp->dl_stid.sc_client;
1062
1063         WARN_ON(!list_empty(&dp->dl_recall_lru));
1064
1065         if (clp->cl_minorversion) {
1066                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1067                 refcount_inc(&dp->dl_stid.sc_count);
1068                 spin_lock(&clp->cl_lock);
1069                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1070                 spin_unlock(&clp->cl_lock);
1071         }
1072         destroy_unhashed_deleg(dp);
1073 }
1074
1075 /*
1076  * SETCLIENTID state
1077  */
1078
1079 static unsigned int clientid_hashval(u32 id)
1080 {
1081         return id & CLIENT_HASH_MASK;
1082 }
1083
1084 static unsigned int clientstr_hashval(struct xdr_netobj name)
1085 {
1086         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1087 }
1088
1089 /*
1090  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1091  * st_{access,deny}_bmap field of the stateid, in order to track not
1092  * only what share bits are currently in force, but also what
1093  * combinations of share bits previous opens have used.  This allows us
1094  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1095  * return an error if the client attempt to downgrade to a combination
1096  * of share bits not explicable by closing some of its previous opens.
1097  *
1098  * XXX: This enforcement is actually incomplete, since we don't keep
1099  * track of access/deny bit combinations; so, e.g., we allow:
1100  *
1101  *      OPEN allow read, deny write
1102  *      OPEN allow both, deny none
1103  *      DOWNGRADE allow read, deny none
1104  *
1105  * which we should reject.
1106  */
1107 static unsigned int
1108 bmap_to_share_mode(unsigned long bmap) {
1109         int i;
1110         unsigned int access = 0;
1111
1112         for (i = 1; i < 4; i++) {
1113                 if (test_bit(i, &bmap))
1114                         access |= i;
1115         }
1116         return access;
1117 }
1118
1119 /* set share access for a given stateid */
1120 static inline void
1121 set_access(u32 access, struct nfs4_ol_stateid *stp)
1122 {
1123         unsigned char mask = 1 << access;
1124
1125         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1126         stp->st_access_bmap |= mask;
1127 }
1128
1129 /* clear share access for a given stateid */
1130 static inline void
1131 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1132 {
1133         unsigned char mask = 1 << access;
1134
1135         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1136         stp->st_access_bmap &= ~mask;
1137 }
1138
1139 /* test whether a given stateid has access */
1140 static inline bool
1141 test_access(u32 access, struct nfs4_ol_stateid *stp)
1142 {
1143         unsigned char mask = 1 << access;
1144
1145         return (bool)(stp->st_access_bmap & mask);
1146 }
1147
1148 /* set share deny for a given stateid */
1149 static inline void
1150 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1151 {
1152         unsigned char mask = 1 << deny;
1153
1154         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1155         stp->st_deny_bmap |= mask;
1156 }
1157
1158 /* clear share deny for a given stateid */
1159 static inline void
1160 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1161 {
1162         unsigned char mask = 1 << deny;
1163
1164         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1165         stp->st_deny_bmap &= ~mask;
1166 }
1167
1168 /* test whether a given stateid is denying specific access */
1169 static inline bool
1170 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1171 {
1172         unsigned char mask = 1 << deny;
1173
1174         return (bool)(stp->st_deny_bmap & mask);
1175 }
1176
1177 static int nfs4_access_to_omode(u32 access)
1178 {
1179         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1180         case NFS4_SHARE_ACCESS_READ:
1181                 return O_RDONLY;
1182         case NFS4_SHARE_ACCESS_WRITE:
1183                 return O_WRONLY;
1184         case NFS4_SHARE_ACCESS_BOTH:
1185                 return O_RDWR;
1186         }
1187         WARN_ON_ONCE(1);
1188         return O_RDONLY;
1189 }
1190
1191 /*
1192  * A stateid that had a deny mode associated with it is being released
1193  * or downgraded. Recalculate the deny mode on the file.
1194  */
1195 static void
1196 recalculate_deny_mode(struct nfs4_file *fp)
1197 {
1198         struct nfs4_ol_stateid *stp;
1199
1200         spin_lock(&fp->fi_lock);
1201         fp->fi_share_deny = 0;
1202         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1203                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1204         spin_unlock(&fp->fi_lock);
1205 }
1206
1207 static void
1208 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1209 {
1210         int i;
1211         bool change = false;
1212
1213         for (i = 1; i < 4; i++) {
1214                 if ((i & deny) != i) {
1215                         change = true;
1216                         clear_deny(i, stp);
1217                 }
1218         }
1219
1220         /* Recalculate per-file deny mode if there was a change */
1221         if (change)
1222                 recalculate_deny_mode(stp->st_stid.sc_file);
1223 }
1224
1225 /* release all access and file references for a given stateid */
1226 static void
1227 release_all_access(struct nfs4_ol_stateid *stp)
1228 {
1229         int i;
1230         struct nfs4_file *fp = stp->st_stid.sc_file;
1231
1232         if (fp && stp->st_deny_bmap != 0)
1233                 recalculate_deny_mode(fp);
1234
1235         for (i = 1; i < 4; i++) {
1236                 if (test_access(i, stp))
1237                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1238                 clear_access(i, stp);
1239         }
1240 }
1241
1242 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1243 {
1244         kfree(sop->so_owner.data);
1245         sop->so_ops->so_free(sop);
1246 }
1247
1248 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1249 {
1250         struct nfs4_client *clp = sop->so_client;
1251
1252         might_lock(&clp->cl_lock);
1253
1254         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1255                 return;
1256         sop->so_ops->so_unhash(sop);
1257         spin_unlock(&clp->cl_lock);
1258         nfs4_free_stateowner(sop);
1259 }
1260
1261 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1262 {
1263         struct nfs4_file *fp = stp->st_stid.sc_file;
1264
1265         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1266
1267         if (list_empty(&stp->st_perfile))
1268                 return false;
1269
1270         spin_lock(&fp->fi_lock);
1271         list_del_init(&stp->st_perfile);
1272         spin_unlock(&fp->fi_lock);
1273         list_del(&stp->st_perstateowner);
1274         return true;
1275 }
1276
1277 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1278 {
1279         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1280
1281         put_clnt_odstate(stp->st_clnt_odstate);
1282         release_all_access(stp);
1283         if (stp->st_stateowner)
1284                 nfs4_put_stateowner(stp->st_stateowner);
1285         kmem_cache_free(stateid_slab, stid);
1286 }
1287
1288 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1289 {
1290         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1291         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1292         struct file *file;
1293
1294         file = find_any_file(stp->st_stid.sc_file);
1295         if (file)
1296                 filp_close(file, (fl_owner_t)lo);
1297         nfs4_free_ol_stateid(stid);
1298 }
1299
1300 /*
1301  * Put the persistent reference to an already unhashed generic stateid, while
1302  * holding the cl_lock. If it's the last reference, then put it onto the
1303  * reaplist for later destruction.
1304  */
1305 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1306                                        struct list_head *reaplist)
1307 {
1308         struct nfs4_stid *s = &stp->st_stid;
1309         struct nfs4_client *clp = s->sc_client;
1310
1311         lockdep_assert_held(&clp->cl_lock);
1312
1313         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1314
1315         if (!refcount_dec_and_test(&s->sc_count)) {
1316                 wake_up_all(&close_wq);
1317                 return;
1318         }
1319
1320         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1321         list_add(&stp->st_locks, reaplist);
1322 }
1323
1324 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1325 {
1326         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1327
1328         list_del_init(&stp->st_locks);
1329         nfs4_unhash_stid(&stp->st_stid);
1330         return unhash_ol_stateid(stp);
1331 }
1332
1333 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1334 {
1335         struct nfs4_client *clp = stp->st_stid.sc_client;
1336         bool unhashed;
1337
1338         spin_lock(&clp->cl_lock);
1339         unhashed = unhash_lock_stateid(stp);
1340         spin_unlock(&clp->cl_lock);
1341         if (unhashed)
1342                 nfs4_put_stid(&stp->st_stid);
1343 }
1344
1345 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1346 {
1347         struct nfs4_client *clp = lo->lo_owner.so_client;
1348
1349         lockdep_assert_held(&clp->cl_lock);
1350
1351         list_del_init(&lo->lo_owner.so_strhash);
1352 }
1353
1354 /*
1355  * Free a list of generic stateids that were collected earlier after being
1356  * fully unhashed.
1357  */
1358 static void
1359 free_ol_stateid_reaplist(struct list_head *reaplist)
1360 {
1361         struct nfs4_ol_stateid *stp;
1362         struct nfs4_file *fp;
1363
1364         might_sleep();
1365
1366         while (!list_empty(reaplist)) {
1367                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1368                                        st_locks);
1369                 list_del(&stp->st_locks);
1370                 fp = stp->st_stid.sc_file;
1371                 stp->st_stid.sc_free(&stp->st_stid);
1372                 if (fp)
1373                         put_nfs4_file(fp);
1374         }
1375 }
1376
1377 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1378                                        struct list_head *reaplist)
1379 {
1380         struct nfs4_ol_stateid *stp;
1381
1382         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1383
1384         while (!list_empty(&open_stp->st_locks)) {
1385                 stp = list_entry(open_stp->st_locks.next,
1386                                 struct nfs4_ol_stateid, st_locks);
1387                 WARN_ON(!unhash_lock_stateid(stp));
1388                 put_ol_stateid_locked(stp, reaplist);
1389         }
1390 }
1391
1392 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1393                                 struct list_head *reaplist)
1394 {
1395         bool unhashed;
1396
1397         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1398
1399         unhashed = unhash_ol_stateid(stp);
1400         release_open_stateid_locks(stp, reaplist);
1401         return unhashed;
1402 }
1403
1404 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1405 {
1406         LIST_HEAD(reaplist);
1407
1408         spin_lock(&stp->st_stid.sc_client->cl_lock);
1409         if (unhash_open_stateid(stp, &reaplist))
1410                 put_ol_stateid_locked(stp, &reaplist);
1411         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1412         free_ol_stateid_reaplist(&reaplist);
1413 }
1414
1415 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1416 {
1417         struct nfs4_client *clp = oo->oo_owner.so_client;
1418
1419         lockdep_assert_held(&clp->cl_lock);
1420
1421         list_del_init(&oo->oo_owner.so_strhash);
1422         list_del_init(&oo->oo_perclient);
1423 }
1424
1425 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1426 {
1427         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1428                                           nfsd_net_id);
1429         struct nfs4_ol_stateid *s;
1430
1431         spin_lock(&nn->client_lock);
1432         s = oo->oo_last_closed_stid;
1433         if (s) {
1434                 list_del_init(&oo->oo_close_lru);
1435                 oo->oo_last_closed_stid = NULL;
1436         }
1437         spin_unlock(&nn->client_lock);
1438         if (s)
1439                 nfs4_put_stid(&s->st_stid);
1440 }
1441
1442 static void release_openowner(struct nfs4_openowner *oo)
1443 {
1444         struct nfs4_ol_stateid *stp;
1445         struct nfs4_client *clp = oo->oo_owner.so_client;
1446         struct list_head reaplist;
1447
1448         INIT_LIST_HEAD(&reaplist);
1449
1450         spin_lock(&clp->cl_lock);
1451         unhash_openowner_locked(oo);
1452         while (!list_empty(&oo->oo_owner.so_stateids)) {
1453                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1454                                 struct nfs4_ol_stateid, st_perstateowner);
1455                 if (unhash_open_stateid(stp, &reaplist))
1456                         put_ol_stateid_locked(stp, &reaplist);
1457         }
1458         spin_unlock(&clp->cl_lock);
1459         free_ol_stateid_reaplist(&reaplist);
1460         release_last_closed_stateid(oo);
1461         nfs4_put_stateowner(&oo->oo_owner);
1462 }
1463
1464 static inline int
1465 hash_sessionid(struct nfs4_sessionid *sessionid)
1466 {
1467         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1468
1469         return sid->sequence % SESSION_HASH_SIZE;
1470 }
1471
1472 #ifdef CONFIG_SUNRPC_DEBUG
1473 static inline void
1474 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1475 {
1476         u32 *ptr = (u32 *)(&sessionid->data[0]);
1477         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1478 }
1479 #else
1480 static inline void
1481 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1482 {
1483 }
1484 #endif
1485
1486 /*
1487  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1488  * won't be used for replay.
1489  */
1490 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1491 {
1492         struct nfs4_stateowner *so = cstate->replay_owner;
1493
1494         if (nfserr == nfserr_replay_me)
1495                 return;
1496
1497         if (!seqid_mutating_err(ntohl(nfserr))) {
1498                 nfsd4_cstate_clear_replay(cstate);
1499                 return;
1500         }
1501         if (!so)
1502                 return;
1503         if (so->so_is_open_owner)
1504                 release_last_closed_stateid(openowner(so));
1505         so->so_seqid++;
1506         return;
1507 }
1508
1509 static void
1510 gen_sessionid(struct nfsd4_session *ses)
1511 {
1512         struct nfs4_client *clp = ses->se_client;
1513         struct nfsd4_sessionid *sid;
1514
1515         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1516         sid->clientid = clp->cl_clientid;
1517         sid->sequence = current_sessionid++;
1518         sid->reserved = 0;
1519 }
1520
1521 /*
1522  * The protocol defines ca_maxresponssize_cached to include the size of
1523  * the rpc header, but all we need to cache is the data starting after
1524  * the end of the initial SEQUENCE operation--the rest we regenerate
1525  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1526  * value that is the number of bytes in our cache plus a few additional
1527  * bytes.  In order to stay on the safe side, and not promise more than
1528  * we can cache, those additional bytes must be the minimum possible: 24
1529  * bytes of rpc header (xid through accept state, with AUTH_NULL
1530  * verifier), 12 for the compound header (with zero-length tag), and 44
1531  * for the SEQUENCE op response:
1532  */
1533 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1534
1535 static void
1536 free_session_slots(struct nfsd4_session *ses)
1537 {
1538         int i;
1539
1540         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1541                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1542                 kfree(ses->se_slots[i]);
1543         }
1544 }
1545
1546 /*
1547  * We don't actually need to cache the rpc and session headers, so we
1548  * can allocate a little less for each slot:
1549  */
1550 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1551 {
1552         u32 size;
1553
1554         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1555                 size = 0;
1556         else
1557                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1558         return size + sizeof(struct nfsd4_slot);
1559 }
1560
1561 /*
1562  * XXX: If we run out of reserved DRC memory we could (up to a point)
1563  * re-negotiate active sessions and reduce their slot usage to make
1564  * room for new connections. For now we just fail the create session.
1565  */
1566 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1567 {
1568         u32 slotsize = slot_bytes(ca);
1569         u32 num = ca->maxreqs;
1570         unsigned long avail, total_avail;
1571
1572         spin_lock(&nfsd_drc_lock);
1573         total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1574         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1575         /*
1576          * Never use more than a third of the remaining memory,
1577          * unless it's the only way to give this client a slot:
1578          */
1579         avail = clamp_t(unsigned long, avail, slotsize, total_avail/3);
1580         num = min_t(int, num, avail / slotsize);
1581         nfsd_drc_mem_used += num * slotsize;
1582         spin_unlock(&nfsd_drc_lock);
1583
1584         return num;
1585 }
1586
1587 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1588 {
1589         int slotsize = slot_bytes(ca);
1590
1591         spin_lock(&nfsd_drc_lock);
1592         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1593         spin_unlock(&nfsd_drc_lock);
1594 }
1595
1596 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1597                                            struct nfsd4_channel_attrs *battrs)
1598 {
1599         int numslots = fattrs->maxreqs;
1600         int slotsize = slot_bytes(fattrs);
1601         struct nfsd4_session *new;
1602         int mem, i;
1603
1604         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1605                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1606         mem = numslots * sizeof(struct nfsd4_slot *);
1607
1608         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1609         if (!new)
1610                 return NULL;
1611         /* allocate each struct nfsd4_slot and data cache in one piece */
1612         for (i = 0; i < numslots; i++) {
1613                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1614                 if (!new->se_slots[i])
1615                         goto out_free;
1616         }
1617
1618         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1619         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1620
1621         return new;
1622 out_free:
1623         while (i--)
1624                 kfree(new->se_slots[i]);
1625         kfree(new);
1626         return NULL;
1627 }
1628
1629 static void free_conn(struct nfsd4_conn *c)
1630 {
1631         svc_xprt_put(c->cn_xprt);
1632         kfree(c);
1633 }
1634
1635 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1636 {
1637         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1638         struct nfs4_client *clp = c->cn_session->se_client;
1639
1640         spin_lock(&clp->cl_lock);
1641         if (!list_empty(&c->cn_persession)) {
1642                 list_del(&c->cn_persession);
1643                 free_conn(c);
1644         }
1645         nfsd4_probe_callback(clp);
1646         spin_unlock(&clp->cl_lock);
1647 }
1648
1649 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1650 {
1651         struct nfsd4_conn *conn;
1652
1653         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1654         if (!conn)
1655                 return NULL;
1656         svc_xprt_get(rqstp->rq_xprt);
1657         conn->cn_xprt = rqstp->rq_xprt;
1658         conn->cn_flags = flags;
1659         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1660         return conn;
1661 }
1662
1663 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1664 {
1665         conn->cn_session = ses;
1666         list_add(&conn->cn_persession, &ses->se_conns);
1667 }
1668
1669 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1670 {
1671         struct nfs4_client *clp = ses->se_client;
1672
1673         spin_lock(&clp->cl_lock);
1674         __nfsd4_hash_conn(conn, ses);
1675         spin_unlock(&clp->cl_lock);
1676 }
1677
1678 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1679 {
1680         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1681         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1682 }
1683
1684 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1685 {
1686         int ret;
1687
1688         nfsd4_hash_conn(conn, ses);
1689         ret = nfsd4_register_conn(conn);
1690         if (ret)
1691                 /* oops; xprt is already down: */
1692                 nfsd4_conn_lost(&conn->cn_xpt_user);
1693         /* We may have gained or lost a callback channel: */
1694         nfsd4_probe_callback_sync(ses->se_client);
1695 }
1696
1697 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1698 {
1699         u32 dir = NFS4_CDFC4_FORE;
1700
1701         if (cses->flags & SESSION4_BACK_CHAN)
1702                 dir |= NFS4_CDFC4_BACK;
1703         return alloc_conn(rqstp, dir);
1704 }
1705
1706 /* must be called under client_lock */
1707 static void nfsd4_del_conns(struct nfsd4_session *s)
1708 {
1709         struct nfs4_client *clp = s->se_client;
1710         struct nfsd4_conn *c;
1711
1712         spin_lock(&clp->cl_lock);
1713         while (!list_empty(&s->se_conns)) {
1714                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1715                 list_del_init(&c->cn_persession);
1716                 spin_unlock(&clp->cl_lock);
1717
1718                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1719                 free_conn(c);
1720
1721                 spin_lock(&clp->cl_lock);
1722         }
1723         spin_unlock(&clp->cl_lock);
1724 }
1725
1726 static void __free_session(struct nfsd4_session *ses)
1727 {
1728         free_session_slots(ses);
1729         kfree(ses);
1730 }
1731
1732 static void free_session(struct nfsd4_session *ses)
1733 {
1734         nfsd4_del_conns(ses);
1735         nfsd4_put_drc_mem(&ses->se_fchannel);
1736         __free_session(ses);
1737 }
1738
1739 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1740 {
1741         int idx;
1742         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1743
1744         new->se_client = clp;
1745         gen_sessionid(new);
1746
1747         INIT_LIST_HEAD(&new->se_conns);
1748
1749         new->se_cb_seq_nr = 1;
1750         new->se_flags = cses->flags;
1751         new->se_cb_prog = cses->callback_prog;
1752         new->se_cb_sec = cses->cb_sec;
1753         atomic_set(&new->se_ref, 0);
1754         idx = hash_sessionid(&new->se_sessionid);
1755         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1756         spin_lock(&clp->cl_lock);
1757         list_add(&new->se_perclnt, &clp->cl_sessions);
1758         spin_unlock(&clp->cl_lock);
1759
1760         {
1761                 struct sockaddr *sa = svc_addr(rqstp);
1762                 /*
1763                  * This is a little silly; with sessions there's no real
1764                  * use for the callback address.  Use the peer address
1765                  * as a reasonable default for now, but consider fixing
1766                  * the rpc client not to require an address in the
1767                  * future:
1768                  */
1769                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1770                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1771         }
1772 }
1773
1774 /* caller must hold client_lock */
1775 static struct nfsd4_session *
1776 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1777 {
1778         struct nfsd4_session *elem;
1779         int idx;
1780         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1781
1782         lockdep_assert_held(&nn->client_lock);
1783
1784         dump_sessionid(__func__, sessionid);
1785         idx = hash_sessionid(sessionid);
1786         /* Search in the appropriate list */
1787         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1788                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1789                             NFS4_MAX_SESSIONID_LEN)) {
1790                         return elem;
1791                 }
1792         }
1793
1794         dprintk("%s: session not found\n", __func__);
1795         return NULL;
1796 }
1797
1798 static struct nfsd4_session *
1799 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1800                 __be32 *ret)
1801 {
1802         struct nfsd4_session *session;
1803         __be32 status = nfserr_badsession;
1804
1805         session = __find_in_sessionid_hashtbl(sessionid, net);
1806         if (!session)
1807                 goto out;
1808         status = nfsd4_get_session_locked(session);
1809         if (status)
1810                 session = NULL;
1811 out:
1812         *ret = status;
1813         return session;
1814 }
1815
1816 /* caller must hold client_lock */
1817 static void
1818 unhash_session(struct nfsd4_session *ses)
1819 {
1820         struct nfs4_client *clp = ses->se_client;
1821         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1822
1823         lockdep_assert_held(&nn->client_lock);
1824
1825         list_del(&ses->se_hash);
1826         spin_lock(&ses->se_client->cl_lock);
1827         list_del(&ses->se_perclnt);
1828         spin_unlock(&ses->se_client->cl_lock);
1829 }
1830
1831 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1832 static int
1833 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1834 {
1835         /*
1836          * We're assuming the clid was not given out from a boot
1837          * precisely 2^32 (about 136 years) before this one.  That seems
1838          * a safe assumption:
1839          */
1840         if (clid->cl_boot == (u32)nn->boot_time)
1841                 return 0;
1842         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1843                 clid->cl_boot, clid->cl_id, nn->boot_time);
1844         return 1;
1845 }
1846
1847 /*
1848  * XXX Should we use a slab cache ?
1849  * This type of memory management is somewhat inefficient, but we use it
1850  * anyway since SETCLIENTID is not a common operation.
1851  */
1852 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1853 {
1854         struct nfs4_client *clp;
1855         int i;
1856
1857         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1858         if (clp == NULL)
1859                 return NULL;
1860         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1861         if (clp->cl_name.data == NULL)
1862                 goto err_no_name;
1863         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1864                                                  sizeof(struct list_head),
1865                                                  GFP_KERNEL);
1866         if (!clp->cl_ownerstr_hashtbl)
1867                 goto err_no_hashtbl;
1868         for (i = 0; i < OWNER_HASH_SIZE; i++)
1869                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1870         INIT_LIST_HEAD(&clp->cl_sessions);
1871         idr_init(&clp->cl_stateids);
1872         atomic_set(&clp->cl_rpc_users, 0);
1873         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1874         INIT_LIST_HEAD(&clp->cl_idhash);
1875         INIT_LIST_HEAD(&clp->cl_openowners);
1876         INIT_LIST_HEAD(&clp->cl_delegations);
1877         INIT_LIST_HEAD(&clp->cl_lru);
1878         INIT_LIST_HEAD(&clp->cl_revoked);
1879 #ifdef CONFIG_NFSD_PNFS
1880         INIT_LIST_HEAD(&clp->cl_lo_states);
1881 #endif
1882         INIT_LIST_HEAD(&clp->async_copies);
1883         spin_lock_init(&clp->async_lock);
1884         spin_lock_init(&clp->cl_lock);
1885         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1886         return clp;
1887 err_no_hashtbl:
1888         kfree(clp->cl_name.data);
1889 err_no_name:
1890         kmem_cache_free(client_slab, clp);
1891         return NULL;
1892 }
1893
1894 static void __free_client(struct kref *k)
1895 {
1896         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1897         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1898
1899         free_svc_cred(&clp->cl_cred);
1900         kfree(clp->cl_ownerstr_hashtbl);
1901         kfree(clp->cl_name.data);
1902         kfree(clp->cl_nii_domain.data);
1903         kfree(clp->cl_nii_name.data);
1904         idr_destroy(&clp->cl_stateids);
1905         kmem_cache_free(client_slab, clp);
1906 }
1907
1908 static void drop_client(struct nfs4_client *clp)
1909 {
1910         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1911 }
1912
1913 static void
1914 free_client(struct nfs4_client *clp)
1915 {
1916         while (!list_empty(&clp->cl_sessions)) {
1917                 struct nfsd4_session *ses;
1918                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1919                                 se_perclnt);
1920                 list_del(&ses->se_perclnt);
1921                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1922                 free_session(ses);
1923         }
1924         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1925         if (clp->cl_nfsd_dentry) {
1926                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
1927                 clp->cl_nfsd_dentry = NULL;
1928                 wake_up_all(&expiry_wq);
1929         }
1930         drop_client(clp);
1931 }
1932
1933 /* must be called under the client_lock */
1934 static void
1935 unhash_client_locked(struct nfs4_client *clp)
1936 {
1937         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1938         struct nfsd4_session *ses;
1939
1940         lockdep_assert_held(&nn->client_lock);
1941
1942         /* Mark the client as expired! */
1943         clp->cl_time = 0;
1944         /* Make it invisible */
1945         if (!list_empty(&clp->cl_idhash)) {
1946                 list_del_init(&clp->cl_idhash);
1947                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1948                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1949                 else
1950                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1951         }
1952         list_del_init(&clp->cl_lru);
1953         spin_lock(&clp->cl_lock);
1954         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1955                 list_del_init(&ses->se_hash);
1956         spin_unlock(&clp->cl_lock);
1957 }
1958
1959 static void
1960 unhash_client(struct nfs4_client *clp)
1961 {
1962         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1963
1964         spin_lock(&nn->client_lock);
1965         unhash_client_locked(clp);
1966         spin_unlock(&nn->client_lock);
1967 }
1968
1969 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1970 {
1971         if (atomic_read(&clp->cl_rpc_users))
1972                 return nfserr_jukebox;
1973         unhash_client_locked(clp);
1974         return nfs_ok;
1975 }
1976
1977 static void
1978 __destroy_client(struct nfs4_client *clp)
1979 {
1980         int i;
1981         struct nfs4_openowner *oo;
1982         struct nfs4_delegation *dp;
1983         struct list_head reaplist;
1984
1985         INIT_LIST_HEAD(&reaplist);
1986         spin_lock(&state_lock);
1987         while (!list_empty(&clp->cl_delegations)) {
1988                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1989                 WARN_ON(!unhash_delegation_locked(dp));
1990                 list_add(&dp->dl_recall_lru, &reaplist);
1991         }
1992         spin_unlock(&state_lock);
1993         while (!list_empty(&reaplist)) {
1994                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1995                 list_del_init(&dp->dl_recall_lru);
1996                 destroy_unhashed_deleg(dp);
1997         }
1998         while (!list_empty(&clp->cl_revoked)) {
1999                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2000                 list_del_init(&dp->dl_recall_lru);
2001                 nfs4_put_stid(&dp->dl_stid);
2002         }
2003         while (!list_empty(&clp->cl_openowners)) {
2004                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2005                 nfs4_get_stateowner(&oo->oo_owner);
2006                 release_openowner(oo);
2007         }
2008         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2009                 struct nfs4_stateowner *so, *tmp;
2010
2011                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2012                                          so_strhash) {
2013                         /* Should be no openowners at this point */
2014                         WARN_ON_ONCE(so->so_is_open_owner);
2015                         remove_blocked_locks(lockowner(so));
2016                 }
2017         }
2018         nfsd4_return_all_client_layouts(clp);
2019         nfsd4_shutdown_copy(clp);
2020         nfsd4_shutdown_callback(clp);
2021         if (clp->cl_cb_conn.cb_xprt)
2022                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2023         free_client(clp);
2024         wake_up_all(&expiry_wq);
2025 }
2026
2027 static void
2028 destroy_client(struct nfs4_client *clp)
2029 {
2030         unhash_client(clp);
2031         __destroy_client(clp);
2032 }
2033
2034 static void inc_reclaim_complete(struct nfs4_client *clp)
2035 {
2036         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2037
2038         if (!nn->track_reclaim_completes)
2039                 return;
2040         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2041                 return;
2042         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2043                         nn->reclaim_str_hashtbl_size) {
2044                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2045                                 clp->net->ns.inum);
2046                 nfsd4_end_grace(nn);
2047         }
2048 }
2049
2050 static void expire_client(struct nfs4_client *clp)
2051 {
2052         unhash_client(clp);
2053         nfsd4_client_record_remove(clp);
2054         __destroy_client(clp);
2055 }
2056
2057 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2058 {
2059         memcpy(target->cl_verifier.data, source->data,
2060                         sizeof(target->cl_verifier.data));
2061 }
2062
2063 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2064 {
2065         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2066         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2067 }
2068
2069 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2070 {
2071         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2072         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2073                                                                 GFP_KERNEL);
2074         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2075         if ((source->cr_principal && !target->cr_principal) ||
2076             (source->cr_raw_principal && !target->cr_raw_principal) ||
2077             (source->cr_targ_princ && !target->cr_targ_princ))
2078                 return -ENOMEM;
2079
2080         target->cr_flavor = source->cr_flavor;
2081         target->cr_uid = source->cr_uid;
2082         target->cr_gid = source->cr_gid;
2083         target->cr_group_info = source->cr_group_info;
2084         get_group_info(target->cr_group_info);
2085         target->cr_gss_mech = source->cr_gss_mech;
2086         if (source->cr_gss_mech)
2087                 gss_mech_get(source->cr_gss_mech);
2088         return 0;
2089 }
2090
2091 static int
2092 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2093 {
2094         if (o1->len < o2->len)
2095                 return -1;
2096         if (o1->len > o2->len)
2097                 return 1;
2098         return memcmp(o1->data, o2->data, o1->len);
2099 }
2100
2101 static int
2102 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2103 {
2104         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2105 }
2106
2107 static int
2108 same_clid(clientid_t *cl1, clientid_t *cl2)
2109 {
2110         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2111 }
2112
2113 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2114 {
2115         int i;
2116
2117         if (g1->ngroups != g2->ngroups)
2118                 return false;
2119         for (i=0; i<g1->ngroups; i++)
2120                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2121                         return false;
2122         return true;
2123 }
2124
2125 /*
2126  * RFC 3530 language requires clid_inuse be returned when the
2127  * "principal" associated with a requests differs from that previously
2128  * used.  We use uid, gid's, and gss principal string as our best
2129  * approximation.  We also don't want to allow non-gss use of a client
2130  * established using gss: in theory cr_principal should catch that
2131  * change, but in practice cr_principal can be null even in the gss case
2132  * since gssd doesn't always pass down a principal string.
2133  */
2134 static bool is_gss_cred(struct svc_cred *cr)
2135 {
2136         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2137         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2138 }
2139
2140
2141 static bool
2142 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2143 {
2144         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2145                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2146                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2147                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2148                 return false;
2149         /* XXX: check that cr_targ_princ fields match ? */
2150         if (cr1->cr_principal == cr2->cr_principal)
2151                 return true;
2152         if (!cr1->cr_principal || !cr2->cr_principal)
2153                 return false;
2154         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2155 }
2156
2157 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2158 {
2159         struct svc_cred *cr = &rqstp->rq_cred;
2160         u32 service;
2161
2162         if (!cr->cr_gss_mech)
2163                 return false;
2164         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2165         return service == RPC_GSS_SVC_INTEGRITY ||
2166                service == RPC_GSS_SVC_PRIVACY;
2167 }
2168
2169 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2170 {
2171         struct svc_cred *cr = &rqstp->rq_cred;
2172
2173         if (!cl->cl_mach_cred)
2174                 return true;
2175         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2176                 return false;
2177         if (!svc_rqst_integrity_protected(rqstp))
2178                 return false;
2179         if (cl->cl_cred.cr_raw_principal)
2180                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2181                                                 cr->cr_raw_principal);
2182         if (!cr->cr_principal)
2183                 return false;
2184         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2185 }
2186
2187 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2188 {
2189         __be32 verf[2];
2190
2191         /*
2192          * This is opaque to client, so no need to byte-swap. Use
2193          * __force to keep sparse happy
2194          */
2195         verf[0] = (__force __be32)get_seconds();
2196         verf[1] = (__force __be32)nn->clverifier_counter++;
2197         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2198 }
2199
2200 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2201 {
2202         clp->cl_clientid.cl_boot = nn->boot_time;
2203         clp->cl_clientid.cl_id = nn->clientid_counter++;
2204         gen_confirm(clp, nn);
2205 }
2206
2207 static struct nfs4_stid *
2208 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2209 {
2210         struct nfs4_stid *ret;
2211
2212         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2213         if (!ret || !ret->sc_type)
2214                 return NULL;
2215         return ret;
2216 }
2217
2218 static struct nfs4_stid *
2219 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2220 {
2221         struct nfs4_stid *s;
2222
2223         spin_lock(&cl->cl_lock);
2224         s = find_stateid_locked(cl, t);
2225         if (s != NULL) {
2226                 if (typemask & s->sc_type)
2227                         refcount_inc(&s->sc_count);
2228                 else
2229                         s = NULL;
2230         }
2231         spin_unlock(&cl->cl_lock);
2232         return s;
2233 }
2234
2235 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2236 {
2237         struct nfsdfs_client *nc;
2238         nc = get_nfsdfs_client(inode);
2239         if (!nc)
2240                 return NULL;
2241         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2242 }
2243
2244 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2245 {
2246         seq_printf(m, "\"");
2247         seq_escape_mem_ascii(m, data, len);
2248         seq_printf(m, "\"");
2249 }
2250
2251 static int client_info_show(struct seq_file *m, void *v)
2252 {
2253         struct inode *inode = m->private;
2254         struct nfs4_client *clp;
2255         u64 clid;
2256
2257         clp = get_nfsdfs_clp(inode);
2258         if (!clp)
2259                 return -ENXIO;
2260         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2261         seq_printf(m, "clientid: 0x%llx\n", clid);
2262         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2263         seq_printf(m, "name: ");
2264         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2265         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2266         if (clp->cl_nii_domain.data) {
2267                 seq_printf(m, "Implementation domain: ");
2268                 seq_quote_mem(m, clp->cl_nii_domain.data,
2269                                         clp->cl_nii_domain.len);
2270                 seq_printf(m, "\nImplementation name: ");
2271                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2272                 seq_printf(m, "\nImplementation time: [%ld, %ld]\n",
2273                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2274         }
2275         drop_client(clp);
2276
2277         return 0;
2278 }
2279
2280 static int client_info_open(struct inode *inode, struct file *file)
2281 {
2282         return single_open(file, client_info_show, inode);
2283 }
2284
2285 static const struct file_operations client_info_fops = {
2286         .open           = client_info_open,
2287         .read           = seq_read,
2288         .llseek         = seq_lseek,
2289         .release        = single_release,
2290 };
2291
2292 static void *states_start(struct seq_file *s, loff_t *pos)
2293         __acquires(&clp->cl_lock)
2294 {
2295         struct nfs4_client *clp = s->private;
2296         unsigned long id = *pos;
2297         void *ret;
2298
2299         spin_lock(&clp->cl_lock);
2300         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2301         *pos = id;
2302         return ret;
2303 }
2304
2305 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2306 {
2307         struct nfs4_client *clp = s->private;
2308         unsigned long id = *pos;
2309         void *ret;
2310
2311         id = *pos;
2312         id++;
2313         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2314         *pos = id;
2315         return ret;
2316 }
2317
2318 static void states_stop(struct seq_file *s, void *v)
2319         __releases(&clp->cl_lock)
2320 {
2321         struct nfs4_client *clp = s->private;
2322
2323         spin_unlock(&clp->cl_lock);
2324 }
2325
2326 static void nfs4_show_superblock(struct seq_file *s, struct file *f)
2327 {
2328         struct inode *inode = file_inode(f);
2329
2330         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2331                                         MAJOR(inode->i_sb->s_dev),
2332                                          MINOR(inode->i_sb->s_dev),
2333                                          inode->i_ino);
2334 }
2335
2336 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2337 {
2338         seq_printf(s, "owner: ");
2339         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2340 }
2341
2342 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2343 {
2344         struct nfs4_ol_stateid *ols;
2345         struct nfs4_file *nf;
2346         struct file *file;
2347         struct nfs4_stateowner *oo;
2348         unsigned int access, deny;
2349
2350         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2351                 return 0; /* XXX: or SEQ_SKIP? */
2352         ols = openlockstateid(st);
2353         oo = ols->st_stateowner;
2354         nf = st->sc_file;
2355         file = find_any_file(nf);
2356
2357         seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2358
2359         access = bmap_to_share_mode(ols->st_access_bmap);
2360         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2361
2362         seq_printf(s, "access: \%s\%s, ",
2363                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2364                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2365         seq_printf(s, "deny: \%s\%s, ",
2366                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2367                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2368
2369         nfs4_show_superblock(s, file);
2370         seq_printf(s, ", ");
2371         nfs4_show_owner(s, oo);
2372         seq_printf(s, " }\n");
2373         fput(file);
2374
2375         return 0;
2376 }
2377
2378 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2379 {
2380         struct nfs4_ol_stateid *ols;
2381         struct nfs4_file *nf;
2382         struct file *file;
2383         struct nfs4_stateowner *oo;
2384
2385         ols = openlockstateid(st);
2386         oo = ols->st_stateowner;
2387         nf = st->sc_file;
2388         file = find_any_file(nf);
2389
2390         seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2391
2392         /*
2393          * Note: a lock stateid isn't really the same thing as a lock,
2394          * it's the locking state held by one owner on a file, and there
2395          * may be multiple (or no) lock ranges associated with it.
2396          * (Same for the matter is true of open stateids.)
2397          */
2398
2399         nfs4_show_superblock(s, file);
2400         /* XXX: open stateid? */
2401         seq_printf(s, ", ");
2402         nfs4_show_owner(s, oo);
2403         seq_printf(s, " }\n");
2404         fput(file);
2405
2406         return 0;
2407 }
2408
2409 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2410 {
2411         struct nfs4_delegation *ds;
2412         struct nfs4_file *nf;
2413         struct file *file;
2414
2415         ds = delegstateid(st);
2416         nf = st->sc_file;
2417         file = nf->fi_deleg_file;
2418
2419         seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2420
2421         /* Kinda dead code as long as we only support read delegs: */
2422         seq_printf(s, "access: %s, ",
2423                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2424
2425         /* XXX: lease time, whether it's being recalled. */
2426
2427         nfs4_show_superblock(s, file);
2428         seq_printf(s, " }\n");
2429
2430         return 0;
2431 }
2432
2433 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2434 {
2435         struct nfs4_layout_stateid *ls;
2436         struct file *file;
2437
2438         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2439         file = ls->ls_file;
2440
2441         seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2442
2443         /* XXX: What else would be useful? */
2444
2445         nfs4_show_superblock(s, file);
2446         seq_printf(s, " }\n");
2447
2448         return 0;
2449 }
2450
2451 static int states_show(struct seq_file *s, void *v)
2452 {
2453         struct nfs4_stid *st = v;
2454
2455         switch (st->sc_type) {
2456         case NFS4_OPEN_STID:
2457                 return nfs4_show_open(s, st);
2458         case NFS4_LOCK_STID:
2459                 return nfs4_show_lock(s, st);
2460         case NFS4_DELEG_STID:
2461                 return nfs4_show_deleg(s, st);
2462         case NFS4_LAYOUT_STID:
2463                 return nfs4_show_layout(s, st);
2464         default:
2465                 return 0; /* XXX: or SEQ_SKIP? */
2466         }
2467         /* XXX: copy stateids? */
2468 }
2469
2470 static struct seq_operations states_seq_ops = {
2471         .start = states_start,
2472         .next = states_next,
2473         .stop = states_stop,
2474         .show = states_show
2475 };
2476
2477 static int client_states_open(struct inode *inode, struct file *file)
2478 {
2479         struct seq_file *s;
2480         struct nfs4_client *clp;
2481         int ret;
2482
2483         clp = get_nfsdfs_clp(inode);
2484         if (!clp)
2485                 return -ENXIO;
2486
2487         ret = seq_open(file, &states_seq_ops);
2488         if (ret)
2489                 return ret;
2490         s = file->private_data;
2491         s->private = clp;
2492         return 0;
2493 }
2494
2495 static int client_opens_release(struct inode *inode, struct file *file)
2496 {
2497         struct seq_file *m = file->private_data;
2498         struct nfs4_client *clp = m->private;
2499
2500         /* XXX: alternatively, we could get/drop in seq start/stop */
2501         drop_client(clp);
2502         return 0;
2503 }
2504
2505 static const struct file_operations client_states_fops = {
2506         .open           = client_states_open,
2507         .read           = seq_read,
2508         .llseek         = seq_lseek,
2509         .release        = client_opens_release,
2510 };
2511
2512 /*
2513  * Normally we refuse to destroy clients that are in use, but here the
2514  * administrator is telling us to just do it.  We also want to wait
2515  * so the caller has a guarantee that the client's locks are gone by
2516  * the time the write returns:
2517  */
2518 static void force_expire_client(struct nfs4_client *clp)
2519 {
2520         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2521         bool already_expired;
2522
2523         spin_lock(&clp->cl_lock);
2524         clp->cl_time = 0;
2525         spin_unlock(&clp->cl_lock);
2526
2527         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2528         spin_lock(&nn->client_lock);
2529         already_expired = list_empty(&clp->cl_lru);
2530         if (!already_expired)
2531                 unhash_client_locked(clp);
2532         spin_unlock(&nn->client_lock);
2533
2534         if (!already_expired)
2535                 expire_client(clp);
2536         else
2537                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2538 }
2539
2540 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2541                                    size_t size, loff_t *pos)
2542 {
2543         char *data;
2544         struct nfs4_client *clp;
2545
2546         data = simple_transaction_get(file, buf, size);
2547         if (IS_ERR(data))
2548                 return PTR_ERR(data);
2549         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2550                 return -EINVAL;
2551         clp = get_nfsdfs_clp(file_inode(file));
2552         if (!clp)
2553                 return -ENXIO;
2554         force_expire_client(clp);
2555         drop_client(clp);
2556         return 7;
2557 }
2558
2559 static const struct file_operations client_ctl_fops = {
2560         .write          = client_ctl_write,
2561         .release        = simple_transaction_release,
2562 };
2563
2564 static const struct tree_descr client_files[] = {
2565         [0] = {"info", &client_info_fops, S_IRUSR},
2566         [1] = {"states", &client_states_fops, S_IRUSR},
2567         [2] = {"ctl", &client_ctl_fops, S_IRUSR|S_IWUSR},
2568         [3] = {""},
2569 };
2570
2571 static struct nfs4_client *create_client(struct xdr_netobj name,
2572                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2573 {
2574         struct nfs4_client *clp;
2575         struct sockaddr *sa = svc_addr(rqstp);
2576         int ret;
2577         struct net *net = SVC_NET(rqstp);
2578         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2579
2580         clp = alloc_client(name);
2581         if (clp == NULL)
2582                 return NULL;
2583
2584         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2585         if (ret) {
2586                 free_client(clp);
2587                 return NULL;
2588         }
2589         gen_clid(clp, nn);
2590         kref_init(&clp->cl_nfsdfs.cl_ref);
2591         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2592         clp->cl_time = get_seconds();
2593         clear_bit(0, &clp->cl_cb_slot_busy);
2594         copy_verf(clp, verf);
2595         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2596         clp->cl_cb_session = NULL;
2597         clp->net = net;
2598         clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2599                         clp->cl_clientid.cl_id - nn->clientid_base,
2600                         client_files);
2601         if (!clp->cl_nfsd_dentry) {
2602                 free_client(clp);
2603                 return NULL;
2604         }
2605         return clp;
2606 }
2607
2608 static void
2609 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2610 {
2611         struct rb_node **new = &(root->rb_node), *parent = NULL;
2612         struct nfs4_client *clp;
2613
2614         while (*new) {
2615                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2616                 parent = *new;
2617
2618                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2619                         new = &((*new)->rb_left);
2620                 else
2621                         new = &((*new)->rb_right);
2622         }
2623
2624         rb_link_node(&new_clp->cl_namenode, parent, new);
2625         rb_insert_color(&new_clp->cl_namenode, root);
2626 }
2627
2628 static struct nfs4_client *
2629 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2630 {
2631         int cmp;
2632         struct rb_node *node = root->rb_node;
2633         struct nfs4_client *clp;
2634
2635         while (node) {
2636                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2637                 cmp = compare_blob(&clp->cl_name, name);
2638                 if (cmp > 0)
2639                         node = node->rb_left;
2640                 else if (cmp < 0)
2641                         node = node->rb_right;
2642                 else
2643                         return clp;
2644         }
2645         return NULL;
2646 }
2647
2648 static void
2649 add_to_unconfirmed(struct nfs4_client *clp)
2650 {
2651         unsigned int idhashval;
2652         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2653
2654         lockdep_assert_held(&nn->client_lock);
2655
2656         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2657         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2658         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2659         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2660         renew_client_locked(clp);
2661 }
2662
2663 static void
2664 move_to_confirmed(struct nfs4_client *clp)
2665 {
2666         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2667         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2668
2669         lockdep_assert_held(&nn->client_lock);
2670
2671         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2672         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2673         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2674         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2675         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2676         renew_client_locked(clp);
2677 }
2678
2679 static struct nfs4_client *
2680 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2681 {
2682         struct nfs4_client *clp;
2683         unsigned int idhashval = clientid_hashval(clid->cl_id);
2684
2685         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2686                 if (same_clid(&clp->cl_clientid, clid)) {
2687                         if ((bool)clp->cl_minorversion != sessions)
2688                                 return NULL;
2689                         renew_client_locked(clp);
2690                         return clp;
2691                 }
2692         }
2693         return NULL;
2694 }
2695
2696 static struct nfs4_client *
2697 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2698 {
2699         struct list_head *tbl = nn->conf_id_hashtbl;
2700
2701         lockdep_assert_held(&nn->client_lock);
2702         return find_client_in_id_table(tbl, clid, sessions);
2703 }
2704
2705 static struct nfs4_client *
2706 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2707 {
2708         struct list_head *tbl = nn->unconf_id_hashtbl;
2709
2710         lockdep_assert_held(&nn->client_lock);
2711         return find_client_in_id_table(tbl, clid, sessions);
2712 }
2713
2714 static bool clp_used_exchangeid(struct nfs4_client *clp)
2715 {
2716         return clp->cl_exchange_flags != 0;
2717 }
2718
2719 static struct nfs4_client *
2720 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2721 {
2722         lockdep_assert_held(&nn->client_lock);
2723         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2724 }
2725
2726 static struct nfs4_client *
2727 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2728 {
2729         lockdep_assert_held(&nn->client_lock);
2730         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2731 }
2732
2733 static void
2734 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2735 {
2736         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2737         struct sockaddr *sa = svc_addr(rqstp);
2738         u32 scopeid = rpc_get_scope_id(sa);
2739         unsigned short expected_family;
2740
2741         /* Currently, we only support tcp and tcp6 for the callback channel */
2742         if (se->se_callback_netid_len == 3 &&
2743             !memcmp(se->se_callback_netid_val, "tcp", 3))
2744                 expected_family = AF_INET;
2745         else if (se->se_callback_netid_len == 4 &&
2746                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2747                 expected_family = AF_INET6;
2748         else
2749                 goto out_err;
2750
2751         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2752                                             se->se_callback_addr_len,
2753                                             (struct sockaddr *)&conn->cb_addr,
2754                                             sizeof(conn->cb_addr));
2755
2756         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2757                 goto out_err;
2758
2759         if (conn->cb_addr.ss_family == AF_INET6)
2760                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2761
2762         conn->cb_prog = se->se_callback_prog;
2763         conn->cb_ident = se->se_callback_ident;
2764         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2765         return;
2766 out_err:
2767         conn->cb_addr.ss_family = AF_UNSPEC;
2768         conn->cb_addrlen = 0;
2769         dprintk("NFSD: this client (clientid %08x/%08x) "
2770                 "will not receive delegations\n",
2771                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2772
2773         return;
2774 }
2775
2776 /*
2777  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2778  */
2779 static void
2780 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2781 {
2782         struct xdr_buf *buf = resp->xdr.buf;
2783         struct nfsd4_slot *slot = resp->cstate.slot;
2784         unsigned int base;
2785
2786         dprintk("--> %s slot %p\n", __func__, slot);
2787
2788         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2789         slot->sl_opcnt = resp->opcnt;
2790         slot->sl_status = resp->cstate.status;
2791         free_svc_cred(&slot->sl_cred);
2792         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2793
2794         if (!nfsd4_cache_this(resp)) {
2795                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2796                 return;
2797         }
2798         slot->sl_flags |= NFSD4_SLOT_CACHED;
2799
2800         base = resp->cstate.data_offset;
2801         slot->sl_datalen = buf->len - base;
2802         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2803                 WARN(1, "%s: sessions DRC could not cache compound\n",
2804                      __func__);
2805         return;
2806 }
2807
2808 /*
2809  * Encode the replay sequence operation from the slot values.
2810  * If cachethis is FALSE encode the uncached rep error on the next
2811  * operation which sets resp->p and increments resp->opcnt for
2812  * nfs4svc_encode_compoundres.
2813  *
2814  */
2815 static __be32
2816 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2817                           struct nfsd4_compoundres *resp)
2818 {
2819         struct nfsd4_op *op;
2820         struct nfsd4_slot *slot = resp->cstate.slot;
2821
2822         /* Encode the replayed sequence operation */
2823         op = &args->ops[resp->opcnt - 1];
2824         nfsd4_encode_operation(resp, op);
2825
2826         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2827                 return op->status;
2828         if (args->opcnt == 1) {
2829                 /*
2830                  * The original operation wasn't a solo sequence--we
2831                  * always cache those--so this retry must not match the
2832                  * original:
2833                  */
2834                 op->status = nfserr_seq_false_retry;
2835         } else {
2836                 op = &args->ops[resp->opcnt++];
2837                 op->status = nfserr_retry_uncached_rep;
2838                 nfsd4_encode_operation(resp, op);
2839         }
2840         return op->status;
2841 }
2842
2843 /*
2844  * The sequence operation is not cached because we can use the slot and
2845  * session values.
2846  */
2847 static __be32
2848 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2849                          struct nfsd4_sequence *seq)
2850 {
2851         struct nfsd4_slot *slot = resp->cstate.slot;
2852         struct xdr_stream *xdr = &resp->xdr;
2853         __be32 *p;
2854         __be32 status;
2855
2856         dprintk("--> %s slot %p\n", __func__, slot);
2857
2858         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2859         if (status)
2860                 return status;
2861
2862         p = xdr_reserve_space(xdr, slot->sl_datalen);
2863         if (!p) {
2864                 WARN_ON_ONCE(1);
2865                 return nfserr_serverfault;
2866         }
2867         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2868         xdr_commit_encode(xdr);
2869
2870         resp->opcnt = slot->sl_opcnt;
2871         return slot->sl_status;
2872 }
2873
2874 /*
2875  * Set the exchange_id flags returned by the server.
2876  */
2877 static void
2878 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2879 {
2880 #ifdef CONFIG_NFSD_PNFS
2881         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2882 #else
2883         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2884 #endif
2885
2886         /* Referrals are supported, Migration is not. */
2887         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2888
2889         /* set the wire flags to return to client. */
2890         clid->flags = new->cl_exchange_flags;
2891 }
2892
2893 static bool client_has_openowners(struct nfs4_client *clp)
2894 {
2895         struct nfs4_openowner *oo;
2896
2897         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2898                 if (!list_empty(&oo->oo_owner.so_stateids))
2899                         return true;
2900         }
2901         return false;
2902 }
2903
2904 static bool client_has_state(struct nfs4_client *clp)
2905 {
2906         return client_has_openowners(clp)
2907 #ifdef CONFIG_NFSD_PNFS
2908                 || !list_empty(&clp->cl_lo_states)
2909 #endif
2910                 || !list_empty(&clp->cl_delegations)
2911                 || !list_empty(&clp->cl_sessions)
2912                 || !list_empty(&clp->async_copies);
2913 }
2914
2915 static __be32 copy_impl_id(struct nfs4_client *clp,
2916                                 struct nfsd4_exchange_id *exid)
2917 {
2918         if (!exid->nii_domain.data)
2919                 return 0;
2920         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
2921         if (!clp->cl_nii_domain.data)
2922                 return nfserr_jukebox;
2923         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
2924         if (!clp->cl_nii_name.data)
2925                 return nfserr_jukebox;
2926         clp->cl_nii_time.tv_sec = exid->nii_time.tv_sec;
2927         clp->cl_nii_time.tv_nsec = exid->nii_time.tv_nsec;
2928         return 0;
2929 }
2930
2931 __be32
2932 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2933                 union nfsd4_op_u *u)
2934 {
2935         struct nfsd4_exchange_id *exid = &u->exchange_id;
2936         struct nfs4_client *conf, *new;
2937         struct nfs4_client *unconf = NULL;
2938         __be32 status;
2939         char                    addr_str[INET6_ADDRSTRLEN];
2940         nfs4_verifier           verf = exid->verifier;
2941         struct sockaddr         *sa = svc_addr(rqstp);
2942         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2943         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2944
2945         rpc_ntop(sa, addr_str, sizeof(addr_str));
2946         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2947                 "ip_addr=%s flags %x, spa_how %d\n",
2948                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2949                 addr_str, exid->flags, exid->spa_how);
2950
2951         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2952                 return nfserr_inval;
2953
2954         new = create_client(exid->clname, rqstp, &verf);
2955         if (new == NULL)
2956                 return nfserr_jukebox;
2957         status = copy_impl_id(new, exid);
2958         if (status)
2959                 goto out_nolock;
2960
2961         switch (exid->spa_how) {
2962         case SP4_MACH_CRED:
2963                 exid->spo_must_enforce[0] = 0;
2964                 exid->spo_must_enforce[1] = (
2965                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2966                         1 << (OP_EXCHANGE_ID - 32) |
2967                         1 << (OP_CREATE_SESSION - 32) |
2968                         1 << (OP_DESTROY_SESSION - 32) |
2969                         1 << (OP_DESTROY_CLIENTID - 32));
2970
2971                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2972                                         1 << (OP_OPEN_DOWNGRADE) |
2973                                         1 << (OP_LOCKU) |
2974                                         1 << (OP_DELEGRETURN));
2975
2976                 exid->spo_must_allow[1] &= (
2977                                         1 << (OP_TEST_STATEID - 32) |
2978                                         1 << (OP_FREE_STATEID - 32));
2979                 if (!svc_rqst_integrity_protected(rqstp)) {
2980                         status = nfserr_inval;
2981                         goto out_nolock;
2982                 }
2983                 /*
2984                  * Sometimes userspace doesn't give us a principal.
2985                  * Which is a bug, really.  Anyway, we can't enforce
2986                  * MACH_CRED in that case, better to give up now:
2987                  */
2988                 if (!new->cl_cred.cr_principal &&
2989                                         !new->cl_cred.cr_raw_principal) {
2990                         status = nfserr_serverfault;
2991                         goto out_nolock;
2992                 }
2993                 new->cl_mach_cred = true;
2994         case SP4_NONE:
2995                 break;
2996         default:                                /* checked by xdr code */
2997                 WARN_ON_ONCE(1);
2998                 /* fall through */
2999         case SP4_SSV:
3000                 status = nfserr_encr_alg_unsupp;
3001                 goto out_nolock;
3002         }
3003
3004         /* Cases below refer to rfc 5661 section 18.35.4: */
3005         spin_lock(&nn->client_lock);
3006         conf = find_confirmed_client_by_name(&exid->clname, nn);
3007         if (conf) {
3008                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3009                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3010
3011                 if (update) {
3012                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3013                                 status = nfserr_inval;
3014                                 goto out;
3015                         }
3016                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3017                                 status = nfserr_wrong_cred;
3018                                 goto out;
3019                         }
3020                         if (!creds_match) { /* case 9 */
3021                                 status = nfserr_perm;
3022                                 goto out;
3023                         }
3024                         if (!verfs_match) { /* case 8 */
3025                                 status = nfserr_not_same;
3026                                 goto out;
3027                         }
3028                         /* case 6 */
3029                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3030                         goto out_copy;
3031                 }
3032                 if (!creds_match) { /* case 3 */
3033                         if (client_has_state(conf)) {
3034                                 status = nfserr_clid_inuse;
3035                                 goto out;
3036                         }
3037                         goto out_new;
3038                 }
3039                 if (verfs_match) { /* case 2 */
3040                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3041                         goto out_copy;
3042                 }
3043                 /* case 5, client reboot */
3044                 conf = NULL;
3045                 goto out_new;
3046         }
3047
3048         if (update) { /* case 7 */
3049                 status = nfserr_noent;
3050                 goto out;
3051         }
3052
3053         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
3054         if (unconf) /* case 4, possible retry or client restart */
3055                 unhash_client_locked(unconf);
3056
3057         /* case 1 (normal case) */
3058 out_new:
3059         if (conf) {
3060                 status = mark_client_expired_locked(conf);
3061                 if (status)
3062                         goto out;
3063         }
3064         new->cl_minorversion = cstate->minorversion;
3065         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3066         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3067
3068         add_to_unconfirmed(new);
3069         swap(new, conf);
3070 out_copy:
3071         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3072         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3073
3074         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3075         nfsd4_set_ex_flags(conf, exid);
3076
3077         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3078                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3079         status = nfs_ok;
3080
3081 out:
3082         spin_unlock(&nn->client_lock);
3083 out_nolock:
3084         if (new)
3085                 expire_client(new);
3086         if (unconf)
3087                 expire_client(unconf);
3088         return status;
3089 }
3090
3091 static __be32
3092 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3093 {
3094         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3095                 slot_seqid);
3096
3097         /* The slot is in use, and no response has been sent. */
3098         if (slot_inuse) {
3099                 if (seqid == slot_seqid)
3100                         return nfserr_jukebox;
3101                 else
3102                         return nfserr_seq_misordered;
3103         }
3104         /* Note unsigned 32-bit arithmetic handles wraparound: */
3105         if (likely(seqid == slot_seqid + 1))
3106                 return nfs_ok;
3107         if (seqid == slot_seqid)
3108                 return nfserr_replay_cache;
3109         return nfserr_seq_misordered;
3110 }
3111
3112 /*
3113  * Cache the create session result into the create session single DRC
3114  * slot cache by saving the xdr structure. sl_seqid has been set.
3115  * Do this for solo or embedded create session operations.
3116  */
3117 static void
3118 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3119                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3120 {
3121         slot->sl_status = nfserr;
3122         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3123 }
3124
3125 static __be32
3126 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3127                             struct nfsd4_clid_slot *slot)
3128 {
3129         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3130         return slot->sl_status;
3131 }
3132
3133 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3134                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3135                         1 +     /* MIN tag is length with zero, only length */ \
3136                         3 +     /* version, opcount, opcode */ \
3137                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3138                                 /* seqid, slotID, slotID, cache */ \
3139                         4 ) * sizeof(__be32))
3140
3141 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3142                         2 +     /* verifier: AUTH_NULL, length 0 */\
3143                         1 +     /* status */ \
3144                         1 +     /* MIN tag is length with zero, only length */ \
3145                         3 +     /* opcount, opcode, opstatus*/ \
3146                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3147                                 /* seqid, slotID, slotID, slotID, status */ \
3148                         5 ) * sizeof(__be32))
3149
3150 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3151 {
3152         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3153
3154         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3155                 return nfserr_toosmall;
3156         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3157                 return nfserr_toosmall;
3158         ca->headerpadsz = 0;
3159         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3160         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3161         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3162         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3163                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3164         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3165         /*
3166          * Note decreasing slot size below client's request may make it
3167          * difficult for client to function correctly, whereas
3168          * decreasing the number of slots will (just?) affect
3169          * performance.  When short on memory we therefore prefer to
3170          * decrease number of slots instead of their size.  Clients that
3171          * request larger slots than they need will get poor results:
3172          */
3173         ca->maxreqs = nfsd4_get_drc_mem(ca);
3174         if (!ca->maxreqs)
3175                 return nfserr_jukebox;
3176
3177         return nfs_ok;
3178 }
3179
3180 /*
3181  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3182  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3183  */
3184 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3185         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3186
3187 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3188         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3189
3190 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3191                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3192 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3193                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3194                                  sizeof(__be32))
3195
3196 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3197 {
3198         ca->headerpadsz = 0;
3199
3200         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3201                 return nfserr_toosmall;
3202         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3203                 return nfserr_toosmall;
3204         ca->maxresp_cached = 0;
3205         if (ca->maxops < 2)
3206                 return nfserr_toosmall;
3207
3208         return nfs_ok;
3209 }
3210
3211 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3212 {
3213         switch (cbs->flavor) {
3214         case RPC_AUTH_NULL:
3215         case RPC_AUTH_UNIX:
3216                 return nfs_ok;
3217         default:
3218                 /*
3219                  * GSS case: the spec doesn't allow us to return this
3220                  * error.  But it also doesn't allow us not to support
3221                  * GSS.
3222                  * I'd rather this fail hard than return some error the
3223                  * client might think it can already handle:
3224                  */
3225                 return nfserr_encr_alg_unsupp;
3226         }
3227 }
3228
3229 __be32
3230 nfsd4_create_session(struct svc_rqst *rqstp,
3231                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3232 {
3233         struct nfsd4_create_session *cr_ses = &u->create_session;
3234         struct sockaddr *sa = svc_addr(rqstp);
3235         struct nfs4_client *conf, *unconf;
3236         struct nfs4_client *old = NULL;
3237         struct nfsd4_session *new;
3238         struct nfsd4_conn *conn;
3239         struct nfsd4_clid_slot *cs_slot = NULL;
3240         __be32 status = 0;
3241         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3242
3243         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3244                 return nfserr_inval;
3245         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3246         if (status)
3247                 return status;
3248         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3249         if (status)
3250                 return status;
3251         status = check_backchannel_attrs(&cr_ses->back_channel);
3252         if (status)
3253                 goto out_release_drc_mem;
3254         status = nfserr_jukebox;
3255         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3256         if (!new)
3257                 goto out_release_drc_mem;
3258         conn = alloc_conn_from_crses(rqstp, cr_ses);
3259         if (!conn)
3260                 goto out_free_session;
3261
3262         spin_lock(&nn->client_lock);
3263         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3264         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3265         WARN_ON_ONCE(conf && unconf);
3266
3267         if (conf) {
3268                 status = nfserr_wrong_cred;
3269                 if (!nfsd4_mach_creds_match(conf, rqstp))
3270                         goto out_free_conn;
3271                 cs_slot = &conf->cl_cs_slot;
3272                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3273                 if (status) {
3274                         if (status == nfserr_replay_cache)
3275                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3276                         goto out_free_conn;
3277                 }
3278         } else if (unconf) {
3279                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3280                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3281                         status = nfserr_clid_inuse;
3282                         goto out_free_conn;
3283                 }
3284                 status = nfserr_wrong_cred;
3285                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3286                         goto out_free_conn;
3287                 cs_slot = &unconf->cl_cs_slot;
3288                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3289                 if (status) {
3290                         /* an unconfirmed replay returns misordered */
3291                         status = nfserr_seq_misordered;
3292                         goto out_free_conn;
3293                 }
3294                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3295                 if (old) {
3296                         status = mark_client_expired_locked(old);
3297                         if (status) {
3298                                 old = NULL;
3299                                 goto out_free_conn;
3300                         }
3301                 }
3302                 move_to_confirmed(unconf);
3303                 conf = unconf;
3304         } else {
3305                 status = nfserr_stale_clientid;
3306                 goto out_free_conn;
3307         }
3308         status = nfs_ok;
3309         /* Persistent sessions are not supported */
3310         cr_ses->flags &= ~SESSION4_PERSIST;
3311         /* Upshifting from TCP to RDMA is not supported */
3312         cr_ses->flags &= ~SESSION4_RDMA;
3313
3314         init_session(rqstp, new, conf, cr_ses);
3315         nfsd4_get_session_locked(new);
3316
3317         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3318                NFS4_MAX_SESSIONID_LEN);
3319         cs_slot->sl_seqid++;
3320         cr_ses->seqid = cs_slot->sl_seqid;
3321
3322         /* cache solo and embedded create sessions under the client_lock */
3323         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3324         spin_unlock(&nn->client_lock);
3325         /* init connection and backchannel */
3326         nfsd4_init_conn(rqstp, conn, new);
3327         nfsd4_put_session(new);
3328         if (old)
3329                 expire_client(old);
3330         return status;
3331 out_free_conn:
3332         spin_unlock(&nn->client_lock);
3333         free_conn(conn);
3334         if (old)
3335                 expire_client(old);
3336 out_free_session:
3337         __free_session(new);
3338 out_release_drc_mem:
3339         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3340         return status;
3341 }
3342
3343 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3344 {
3345         switch (*dir) {
3346         case NFS4_CDFC4_FORE:
3347         case NFS4_CDFC4_BACK:
3348                 return nfs_ok;
3349         case NFS4_CDFC4_FORE_OR_BOTH:
3350         case NFS4_CDFC4_BACK_OR_BOTH:
3351                 *dir = NFS4_CDFC4_BOTH;
3352                 return nfs_ok;
3353         };
3354         return nfserr_inval;
3355 }
3356
3357 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3358                 struct nfsd4_compound_state *cstate,
3359                 union nfsd4_op_u *u)
3360 {
3361         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3362         struct nfsd4_session *session = cstate->session;
3363         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3364         __be32 status;
3365
3366         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3367         if (status)
3368                 return status;
3369         spin_lock(&nn->client_lock);
3370         session->se_cb_prog = bc->bc_cb_program;
3371         session->se_cb_sec = bc->bc_cb_sec;
3372         spin_unlock(&nn->client_lock);
3373
3374         nfsd4_probe_callback(session->se_client);
3375
3376         return nfs_ok;
3377 }
3378
3379 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3380                      struct nfsd4_compound_state *cstate,
3381                      union nfsd4_op_u *u)
3382 {
3383         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3384         __be32 status;
3385         struct nfsd4_conn *conn;
3386         struct nfsd4_session *session;
3387         struct net *net = SVC_NET(rqstp);
3388         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3389
3390         if (!nfsd4_last_compound_op(rqstp))
3391                 return nfserr_not_only_op;
3392         spin_lock(&nn->client_lock);
3393         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3394         spin_unlock(&nn->client_lock);
3395         if (!session)
3396                 goto out_no_session;
3397         status = nfserr_wrong_cred;
3398         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3399                 goto out;
3400         status = nfsd4_map_bcts_dir(&bcts->dir);
3401         if (status)
3402                 goto out;
3403         conn = alloc_conn(rqstp, bcts->dir);
3404         status = nfserr_jukebox;
3405         if (!conn)
3406                 goto out;
3407         nfsd4_init_conn(rqstp, conn, session);
3408         status = nfs_ok;
3409 out:
3410         nfsd4_put_session(session);
3411 out_no_session:
3412         return status;
3413 }
3414
3415 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3416 {
3417         if (!cstate->session)
3418                 return false;
3419         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3420 }
3421
3422 __be32
3423 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3424                 union nfsd4_op_u *u)
3425 {
3426         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3427         struct nfsd4_session *ses;
3428         __be32 status;
3429         int ref_held_by_me = 0;
3430         struct net *net = SVC_NET(r);
3431         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3432
3433         status = nfserr_not_only_op;
3434         if (nfsd4_compound_in_session(cstate, sessionid)) {
3435                 if (!nfsd4_last_compound_op(r))
3436                         goto out;
3437                 ref_held_by_me++;
3438         }
3439         dump_sessionid(__func__, sessionid);
3440         spin_lock(&nn->client_lock);
3441         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3442         if (!ses)
3443                 goto out_client_lock;
3444         status = nfserr_wrong_cred;
3445         if (!nfsd4_mach_creds_match(ses->se_client, r))
3446                 goto out_put_session;
3447         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3448         if (status)
3449                 goto out_put_session;
3450         unhash_session(ses);
3451         spin_unlock(&nn->client_lock);
3452
3453         nfsd4_probe_callback_sync(ses->se_client);
3454
3455         spin_lock(&nn->client_lock);
3456         status = nfs_ok;
3457 out_put_session:
3458         nfsd4_put_session_locked(ses);
3459 out_client_lock:
3460         spin_unlock(&nn->client_lock);
3461 out:
3462         return status;
3463 }
3464
3465 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3466 {
3467         struct nfsd4_conn *c;
3468
3469         list_for_each_entry(c, &s->se_conns, cn_persession) {
3470                 if (c->cn_xprt == xpt) {
3471                         return c;
3472                 }
3473         }
3474         return NULL;
3475 }
3476
3477 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3478 {
3479         struct nfs4_client *clp = ses->se_client;
3480         struct nfsd4_conn *c;
3481         __be32 status = nfs_ok;
3482         int ret;
3483
3484         spin_lock(&clp->cl_lock);
3485         c = __nfsd4_find_conn(new->cn_xprt, ses);
3486         if (c)
3487                 goto out_free;
3488         status = nfserr_conn_not_bound_to_session;
3489         if (clp->cl_mach_cred)
3490                 goto out_free;
3491         __nfsd4_hash_conn(new, ses);
3492         spin_unlock(&clp->cl_lock);
3493         ret = nfsd4_register_conn(new);
3494         if (ret)
3495                 /* oops; xprt is already down: */
3496                 nfsd4_conn_lost(&new->cn_xpt_user);
3497         return nfs_ok;
3498 out_free:
3499         spin_unlock(&clp->cl_lock);
3500         free_conn(new);
3501         return status;
3502 }
3503
3504 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3505 {
3506         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3507
3508         return args->opcnt > session->se_fchannel.maxops;
3509 }
3510
3511 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3512                                   struct nfsd4_session *session)
3513 {
3514         struct xdr_buf *xb = &rqstp->rq_arg;
3515
3516         return xb->len > session->se_fchannel.maxreq_sz;
3517 }
3518
3519 static bool replay_matches_cache(struct svc_rqst *rqstp,
3520                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3521 {
3522         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3523
3524         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3525             (bool)seq->cachethis)
3526                 return false;
3527         /*
3528          * If there's an error than the reply can have fewer ops than
3529          * the call.  But if we cached a reply with *more* ops than the
3530          * call you're sending us now, then this new call is clearly not
3531          * really a replay of the old one:
3532          */
3533         if (slot->sl_opcnt < argp->opcnt)
3534                 return false;
3535         /* This is the only check explicitly called by spec: */
3536         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3537                 return false;
3538         /*
3539          * There may be more comparisons we could actually do, but the
3540          * spec doesn't require us to catch every case where the calls
3541          * don't match (that would require caching the call as well as
3542          * the reply), so we don't bother.
3543          */
3544         return true;
3545 }
3546
3547 __be32
3548 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3549                 union nfsd4_op_u *u)
3550 {
3551         struct nfsd4_sequence *seq = &u->sequence;
3552         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3553         struct xdr_stream *xdr = &resp->xdr;
3554         struct nfsd4_session *session;
3555         struct nfs4_client *clp;
3556         struct nfsd4_slot *slot;
3557         struct nfsd4_conn *conn;
3558         __be32 status;
3559         int buflen;
3560         struct net *net = SVC_NET(rqstp);
3561         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3562
3563         if (resp->opcnt != 1)
3564                 return nfserr_sequence_pos;
3565
3566         /*
3567          * Will be either used or freed by nfsd4_sequence_check_conn
3568          * below.
3569          */
3570         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3571         if (!conn)
3572                 return nfserr_jukebox;
3573
3574         spin_lock(&nn->client_lock);
3575         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3576         if (!session)
3577                 goto out_no_session;
3578         clp = session->se_client;
3579
3580         status = nfserr_too_many_ops;
3581         if (nfsd4_session_too_many_ops(rqstp, session))
3582                 goto out_put_session;
3583
3584         status = nfserr_req_too_big;
3585         if (nfsd4_request_too_big(rqstp, session))
3586                 goto out_put_session;
3587
3588         status = nfserr_badslot;
3589         if (seq->slotid >= session->se_fchannel.maxreqs)
3590                 goto out_put_session;
3591
3592         slot = session->se_slots[seq->slotid];
3593         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3594
3595         /* We do not negotiate the number of slots yet, so set the
3596          * maxslots to the session maxreqs which is used to encode
3597          * sr_highest_slotid and the sr_target_slot id to maxslots */
3598         seq->maxslots = session->se_fchannel.maxreqs;
3599
3600         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3601                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3602         if (status == nfserr_replay_cache) {
3603                 status = nfserr_seq_misordered;
3604                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3605                         goto out_put_session;
3606                 status = nfserr_seq_false_retry;
3607                 if (!replay_matches_cache(rqstp, seq, slot))
3608                         goto out_put_session;
3609                 cstate->slot = slot;
3610                 cstate->session = session;
3611                 cstate->clp = clp;
3612                 /* Return the cached reply status and set cstate->status
3613                  * for nfsd4_proc_compound processing */
3614                 status = nfsd4_replay_cache_entry(resp, seq);
3615                 cstate->status = nfserr_replay_cache;
3616                 goto out;
3617         }
3618         if (status)
3619                 goto out_put_session;
3620
3621         status = nfsd4_sequence_check_conn(conn, session);
3622         conn = NULL;
3623         if (status)
3624                 goto out_put_session;
3625
3626         buflen = (seq->cachethis) ?
3627                         session->se_fchannel.maxresp_cached :
3628                         session->se_fchannel.maxresp_sz;
3629         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3630                                     nfserr_rep_too_big;
3631         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3632                 goto out_put_session;
3633         svc_reserve(rqstp, buflen);
3634
3635         status = nfs_ok;
3636         /* Success! bump slot seqid */
3637         slot->sl_seqid = seq->seqid;
3638         slot->sl_flags |= NFSD4_SLOT_INUSE;
3639         if (seq->cachethis)
3640                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3641         else
3642                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3643
3644         cstate->slot = slot;
3645         cstate->session = session;
3646         cstate->clp = clp;
3647
3648 out:
3649         switch (clp->cl_cb_state) {
3650         case NFSD4_CB_DOWN:
3651                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3652                 break;
3653         case NFSD4_CB_FAULT:
3654                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3655                 break;
3656         default:
3657                 seq->status_flags = 0;
3658         }
3659         if (!list_empty(&clp->cl_revoked))
3660                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3661 out_no_session:
3662         if (conn)
3663                 free_conn(conn);
3664         spin_unlock(&nn->client_lock);
3665         return status;
3666 out_put_session:
3667         nfsd4_put_session_locked(session);
3668         goto out_no_session;
3669 }
3670
3671 void
3672 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3673 {
3674         struct nfsd4_compound_state *cs = &resp->cstate;
3675
3676         if (nfsd4_has_session(cs)) {
3677                 if (cs->status != nfserr_replay_cache) {
3678                         nfsd4_store_cache_entry(resp);
3679                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3680                 }
3681                 /* Drop session reference that was taken in nfsd4_sequence() */
3682                 nfsd4_put_session(cs->session);
3683         } else if (cs->clp)
3684                 put_client_renew(cs->clp);
3685 }
3686
3687 __be32
3688 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3689                 struct nfsd4_compound_state *cstate,
3690                 union nfsd4_op_u *u)
3691 {
3692         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3693         struct nfs4_client *conf, *unconf;
3694         struct nfs4_client *clp = NULL;
3695         __be32 status = 0;
3696         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3697
3698         spin_lock(&nn->client_lock);
3699         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3700         conf = find_confirmed_client(&dc->clientid, true, nn);
3701         WARN_ON_ONCE(conf && unconf);
3702
3703         if (conf) {
3704                 if (client_has_state(conf)) {
3705                         status = nfserr_clientid_busy;
3706                         goto out;
3707                 }
3708                 status = mark_client_expired_locked(conf);
3709                 if (status)
3710                         goto out;
3711                 clp = conf;
3712         } else if (unconf)
3713                 clp = unconf;
3714         else {
3715                 status = nfserr_stale_clientid;
3716                 goto out;
3717         }
3718         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3719                 clp = NULL;
3720                 status = nfserr_wrong_cred;
3721                 goto out;
3722         }
3723         unhash_client_locked(clp);
3724 out:
3725         spin_unlock(&nn->client_lock);
3726         if (clp)
3727                 expire_client(clp);
3728         return status;
3729 }
3730
3731 __be32
3732 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3733                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3734 {
3735         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3736         __be32 status = 0;
3737
3738         if (rc->rca_one_fs) {
3739                 if (!cstate->current_fh.fh_dentry)
3740                         return nfserr_nofilehandle;
3741                 /*
3742                  * We don't take advantage of the rca_one_fs case.
3743                  * That's OK, it's optional, we can safely ignore it.
3744                  */
3745                 return nfs_ok;
3746         }
3747
3748         status = nfserr_complete_already;
3749         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3750                              &cstate->session->se_client->cl_flags))
3751                 goto out;
3752
3753         status = nfserr_stale_clientid;
3754         if (is_client_expired(cstate->session->se_client))
3755                 /*
3756                  * The following error isn't really legal.
3757                  * But we only get here if the client just explicitly
3758                  * destroyed the client.  Surely it no longer cares what
3759                  * error it gets back on an operation for the dead
3760                  * client.
3761                  */
3762                 goto out;
3763
3764         status = nfs_ok;
3765         nfsd4_client_record_create(cstate->session->se_client);
3766         inc_reclaim_complete(cstate->session->se_client);
3767 out:
3768         return status;
3769 }
3770
3771 __be32
3772 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3773                   union nfsd4_op_u *u)
3774 {
3775         struct nfsd4_setclientid *setclid = &u->setclientid;
3776         struct xdr_netobj       clname = setclid->se_name;
3777         nfs4_verifier           clverifier = setclid->se_verf;
3778         struct nfs4_client      *conf, *new;
3779         struct nfs4_client      *unconf = NULL;
3780         __be32                  status;
3781         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3782
3783         new = create_client(clname, rqstp, &clverifier);
3784         if (new == NULL)
3785                 return nfserr_jukebox;
3786         /* Cases below refer to rfc 3530 section 14.2.33: */
3787         spin_lock(&nn->client_lock);
3788         conf = find_confirmed_client_by_name(&clname, nn);
3789         if (conf && client_has_state(conf)) {
3790                 /* case 0: */
3791                 status = nfserr_clid_inuse;
3792                 if (clp_used_exchangeid(conf))
3793                         goto out;
3794                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3795                         char addr_str[INET6_ADDRSTRLEN];
3796                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3797                                  sizeof(addr_str));
3798                         dprintk("NFSD: setclientid: string in use by client "
3799                                 "at %s\n", addr_str);
3800                         goto out;
3801                 }
3802         }
3803         unconf = find_unconfirmed_client_by_name(&clname, nn);
3804         if (unconf)
3805                 unhash_client_locked(unconf);
3806         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3807                 /* case 1: probable callback update */
3808                 copy_clid(new, conf);
3809                 gen_confirm(new, nn);
3810         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3811                 ;
3812         new->cl_minorversion = 0;
3813         gen_callback(new, setclid, rqstp);
3814         add_to_unconfirmed(new);
3815         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3816         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3817         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3818         new = NULL;
3819         status = nfs_ok;
3820 out:
3821         spin_unlock(&nn->client_lock);
3822         if (new)
3823                 free_client(new);
3824         if (unconf)
3825                 expire_client(unconf);
3826         return status;
3827 }
3828
3829
3830 __be32
3831 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3832                         struct nfsd4_compound_state *cstate,
3833                         union nfsd4_op_u *u)
3834 {
3835         struct nfsd4_setclientid_confirm *setclientid_confirm =
3836                         &u->setclientid_confirm;
3837         struct nfs4_client *conf, *unconf;
3838         struct nfs4_client *old = NULL;
3839         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3840         clientid_t * clid = &setclientid_confirm->sc_clientid;
3841         __be32 status;
3842         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3843
3844         if (STALE_CLIENTID(clid, nn))
3845                 return nfserr_stale_clientid;
3846
3847         spin_lock(&nn->client_lock);
3848         conf = find_confirmed_client(clid, false, nn);
3849         unconf = find_unconfirmed_client(clid, false, nn);
3850         /*
3851          * We try hard to give out unique clientid's, so if we get an
3852          * attempt to confirm the same clientid with a different cred,
3853          * the client may be buggy; this should never happen.
3854          *
3855          * Nevertheless, RFC 7530 recommends INUSE for this case:
3856          */
3857         status = nfserr_clid_inuse;
3858         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3859                 goto out;
3860         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3861                 goto out;
3862         /* cases below refer to rfc 3530 section 14.2.34: */
3863         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3864                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3865                         /* case 2: probable retransmit */
3866                         status = nfs_ok;
3867                 } else /* case 4: client hasn't noticed we rebooted yet? */
3868                         status = nfserr_stale_clientid;
3869                 goto out;
3870         }
3871         status = nfs_ok;
3872         if (conf) { /* case 1: callback update */
3873                 old = unconf;
3874                 unhash_client_locked(old);
3875                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3876         } else { /* case 3: normal case; new or rebooted client */
3877                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3878                 if (old) {
3879                         status = nfserr_clid_inuse;
3880                         if (client_has_state(old)
3881                                         && !same_creds(&unconf->cl_cred,
3882                                                         &old->cl_cred))
3883                                 goto out;
3884                         status = mark_client_expired_locked(old);
3885                         if (status) {
3886                                 old = NULL;
3887                                 goto out;
3888                         }
3889                 }
3890                 move_to_confirmed(unconf);
3891                 conf = unconf;
3892         }
3893         get_client_locked(conf);
3894         spin_unlock(&nn->client_lock);
3895         nfsd4_probe_callback(conf);
3896         spin_lock(&nn->client_lock);
3897         put_client_renew_locked(conf);
3898 out:
3899         spin_unlock(&nn->client_lock);
3900         if (old)
3901                 expire_client(old);
3902         return status;
3903 }
3904
3905 static struct nfs4_file *nfsd4_alloc_file(void)
3906 {
3907         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3908 }
3909
3910 /* OPEN Share state helper functions */
3911 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3912                                 struct nfs4_file *fp)
3913 {
3914         lockdep_assert_held(&state_lock);
3915
3916         refcount_set(&fp->fi_ref, 1);
3917         spin_lock_init(&fp->fi_lock);
3918         INIT_LIST_HEAD(&fp->fi_stateids);
3919         INIT_LIST_HEAD(&fp->fi_delegations);
3920         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3921         fh_copy_shallow(&fp->fi_fhandle, fh);
3922         fp->fi_deleg_file = NULL;
3923         fp->fi_had_conflict = false;
3924         fp->fi_share_deny = 0;
3925         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3926         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3927 #ifdef CONFIG_NFSD_PNFS
3928         INIT_LIST_HEAD(&fp->fi_lo_states);
3929         atomic_set(&fp->fi_lo_recalls, 0);
3930 #endif
3931         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3932 }
3933
3934 void
3935 nfsd4_free_slabs(void)
3936 {
3937         kmem_cache_destroy(client_slab);
3938         kmem_cache_destroy(openowner_slab);
3939         kmem_cache_destroy(lockowner_slab);
3940         kmem_cache_destroy(file_slab);
3941         kmem_cache_destroy(stateid_slab);
3942         kmem_cache_destroy(deleg_slab);
3943         kmem_cache_destroy(odstate_slab);
3944 }
3945
3946 int
3947 nfsd4_init_slabs(void)
3948 {
3949         client_slab = kmem_cache_create("nfsd4_clients",
3950                         sizeof(struct nfs4_client), 0, 0, NULL);
3951         if (client_slab == NULL)
3952                 goto out;
3953         openowner_slab = kmem_cache_create("nfsd4_openowners",
3954                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3955         if (openowner_slab == NULL)
3956                 goto out_free_client_slab;
3957         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3958                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3959         if (lockowner_slab == NULL)
3960                 goto out_free_openowner_slab;
3961         file_slab = kmem_cache_create("nfsd4_files",
3962                         sizeof(struct nfs4_file), 0, 0, NULL);
3963         if (file_slab == NULL)
3964                 goto out_free_lockowner_slab;
3965         stateid_slab = kmem_cache_create("nfsd4_stateids",
3966                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3967         if (stateid_slab == NULL)
3968                 goto out_free_file_slab;
3969         deleg_slab = kmem_cache_create("nfsd4_delegations",
3970                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3971         if (deleg_slab == NULL)
3972                 goto out_free_stateid_slab;
3973         odstate_slab = kmem_cache_create("nfsd4_odstate",
3974                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3975         if (odstate_slab == NULL)
3976                 goto out_free_deleg_slab;
3977         return 0;
3978
3979 out_free_deleg_slab:
3980         kmem_cache_destroy(deleg_slab);
3981 out_free_stateid_slab:
3982         kmem_cache_destroy(stateid_slab);
3983 out_free_file_slab:
3984         kmem_cache_destroy(file_slab);
3985 out_free_lockowner_slab:
3986         kmem_cache_destroy(lockowner_slab);
3987 out_free_openowner_slab:
3988         kmem_cache_destroy(openowner_slab);
3989 out_free_client_slab:
3990         kmem_cache_destroy(client_slab);
3991 out:
3992         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3993         return -ENOMEM;
3994 }
3995
3996 static void init_nfs4_replay(struct nfs4_replay *rp)
3997 {
3998         rp->rp_status = nfserr_serverfault;
3999         rp->rp_buflen = 0;
4000         rp->rp_buf = rp->rp_ibuf;
4001         mutex_init(&rp->rp_mutex);
4002 }
4003
4004 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4005                 struct nfs4_stateowner *so)
4006 {
4007         if (!nfsd4_has_session(cstate)) {
4008                 mutex_lock(&so->so_replay.rp_mutex);
4009                 cstate->replay_owner = nfs4_get_stateowner(so);
4010         }
4011 }
4012
4013 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4014 {
4015         struct nfs4_stateowner *so = cstate->replay_owner;
4016
4017         if (so != NULL) {
4018                 cstate->replay_owner = NULL;
4019                 mutex_unlock(&so->so_replay.rp_mutex);
4020                 nfs4_put_stateowner(so);
4021         }
4022 }
4023
4024 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4025 {
4026         struct nfs4_stateowner *sop;
4027
4028         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4029         if (!sop)
4030                 return NULL;
4031
4032         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4033         if (!sop->so_owner.data) {
4034                 kmem_cache_free(slab, sop);
4035                 return NULL;
4036         }
4037
4038         INIT_LIST_HEAD(&sop->so_stateids);
4039         sop->so_client = clp;
4040         init_nfs4_replay(&sop->so_replay);
4041         atomic_set(&sop->so_count, 1);
4042         return sop;
4043 }
4044
4045 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4046 {
4047         lockdep_assert_held(&clp->cl_lock);
4048
4049         list_add(&oo->oo_owner.so_strhash,
4050                  &clp->cl_ownerstr_hashtbl[strhashval]);
4051         list_add(&oo->oo_perclient, &clp->cl_openowners);
4052 }
4053
4054 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4055 {
4056         unhash_openowner_locked(openowner(so));
4057 }
4058
4059 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4060 {
4061         struct nfs4_openowner *oo = openowner(so);
4062
4063         kmem_cache_free(openowner_slab, oo);
4064 }
4065
4066 static const struct nfs4_stateowner_operations openowner_ops = {
4067         .so_unhash =    nfs4_unhash_openowner,
4068         .so_free =      nfs4_free_openowner,
4069 };
4070
4071 static struct nfs4_ol_stateid *
4072 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4073 {
4074         struct nfs4_ol_stateid *local, *ret = NULL;
4075         struct nfs4_openowner *oo = open->op_openowner;
4076
4077         lockdep_assert_held(&fp->fi_lock);
4078
4079         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4080                 /* ignore lock owners */
4081                 if (local->st_stateowner->so_is_open_owner == 0)
4082                         continue;
4083                 if (local->st_stateowner != &oo->oo_owner)
4084                         continue;
4085                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4086                         ret = local;
4087                         refcount_inc(&ret->st_stid.sc_count);
4088                         break;
4089                 }
4090         }
4091         return ret;
4092 }
4093
4094 static __be32
4095 nfsd4_verify_open_stid(struct nfs4_stid *s)
4096 {
4097         __be32 ret = nfs_ok;
4098
4099         switch (s->sc_type) {
4100         default:
4101                 break;
4102         case 0:
4103         case NFS4_CLOSED_STID:
4104         case NFS4_CLOSED_DELEG_STID:
4105                 ret = nfserr_bad_stateid;
4106                 break;
4107         case NFS4_REVOKED_DELEG_STID:
4108                 ret = nfserr_deleg_revoked;
4109         }
4110         return ret;
4111 }
4112
4113 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4114 static __be32
4115 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4116 {
4117         __be32 ret;
4118
4119         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4120         ret = nfsd4_verify_open_stid(&stp->st_stid);
4121         if (ret != nfs_ok)
4122                 mutex_unlock(&stp->st_mutex);
4123         return ret;
4124 }
4125
4126 static struct nfs4_ol_stateid *
4127 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4128 {
4129         struct nfs4_ol_stateid *stp;
4130         for (;;) {
4131                 spin_lock(&fp->fi_lock);
4132                 stp = nfsd4_find_existing_open(fp, open);
4133                 spin_unlock(&fp->fi_lock);
4134                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4135                         break;
4136                 nfs4_put_stid(&stp->st_stid);
4137         }
4138         return stp;
4139 }
4140
4141 static struct nfs4_openowner *
4142 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4143                            struct nfsd4_compound_state *cstate)
4144 {
4145         struct nfs4_client *clp = cstate->clp;
4146         struct nfs4_openowner *oo, *ret;
4147
4148         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4149         if (!oo)
4150                 return NULL;
4151         oo->oo_owner.so_ops = &openowner_ops;
4152         oo->oo_owner.so_is_open_owner = 1;
4153         oo->oo_owner.so_seqid = open->op_seqid;
4154         oo->oo_flags = 0;
4155         if (nfsd4_has_session(cstate))
4156                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4157         oo->oo_time = 0;
4158         oo->oo_last_closed_stid = NULL;
4159         INIT_LIST_HEAD(&oo->oo_close_lru);
4160         spin_lock(&clp->cl_lock);
4161         ret = find_openstateowner_str_locked(strhashval, open, clp);
4162         if (ret == NULL) {
4163                 hash_openowner(oo, clp, strhashval);
4164                 ret = oo;
4165         } else
4166                 nfs4_free_stateowner(&oo->oo_owner);
4167
4168         spin_unlock(&clp->cl_lock);
4169         return ret;
4170 }
4171
4172 static struct nfs4_ol_stateid *
4173 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4174 {
4175
4176         struct nfs4_openowner *oo = open->op_openowner;
4177         struct nfs4_ol_stateid *retstp = NULL;
4178         struct nfs4_ol_stateid *stp;
4179
4180         stp = open->op_stp;
4181         /* We are moving these outside of the spinlocks to avoid the warnings */
4182         mutex_init(&stp->st_mutex);
4183         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4184
4185 retry:
4186         spin_lock(&oo->oo_owner.so_client->cl_lock);
4187         spin_lock(&fp->fi_lock);
4188
4189         retstp = nfsd4_find_existing_open(fp, open);
4190         if (retstp)
4191                 goto out_unlock;
4192
4193         open->op_stp = NULL;
4194         refcount_inc(&stp->st_stid.sc_count);
4195         stp->st_stid.sc_type = NFS4_OPEN_STID;
4196         INIT_LIST_HEAD(&stp->st_locks);
4197         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4198         get_nfs4_file(fp);
4199         stp->st_stid.sc_file = fp;
4200         stp->st_access_bmap = 0;
4201         stp->st_deny_bmap = 0;
4202         stp->st_openstp = NULL;
4203         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4204         list_add(&stp->st_perfile, &fp->fi_stateids);
4205
4206 out_unlock:
4207         spin_unlock(&fp->fi_lock);
4208         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4209         if (retstp) {
4210                 /* Handle races with CLOSE */
4211                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4212                         nfs4_put_stid(&retstp->st_stid);
4213                         goto retry;
4214                 }
4215                 /* To keep mutex tracking happy */
4216                 mutex_unlock(&stp->st_mutex);
4217                 stp = retstp;
4218         }
4219         return stp;
4220 }
4221
4222 /*
4223  * In the 4.0 case we need to keep the owners around a little while to handle
4224  * CLOSE replay. We still do need to release any file access that is held by
4225  * them before returning however.
4226  */
4227 static void
4228 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4229 {
4230         struct nfs4_ol_stateid *last;
4231         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4232         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4233                                                 nfsd_net_id);
4234
4235         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4236
4237         /*
4238          * We know that we hold one reference via nfsd4_close, and another
4239          * "persistent" reference for the client. If the refcount is higher
4240          * than 2, then there are still calls in progress that are using this
4241          * stateid. We can't put the sc_file reference until they are finished.
4242          * Wait for the refcount to drop to 2. Since it has been unhashed,
4243          * there should be no danger of the refcount going back up again at
4244          * this point.
4245          */
4246         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4247
4248         release_all_access(s);
4249         if (s->st_stid.sc_file) {
4250                 put_nfs4_file(s->st_stid.sc_file);
4251                 s->st_stid.sc_file = NULL;
4252         }
4253
4254         spin_lock(&nn->client_lock);
4255         last = oo->oo_last_closed_stid;
4256         oo->oo_last_closed_stid = s;
4257         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4258         oo->oo_time = get_seconds();
4259         spin_unlock(&nn->client_lock);
4260         if (last)
4261                 nfs4_put_stid(&last->st_stid);
4262 }
4263
4264 /* search file_hashtbl[] for file */
4265 static struct nfs4_file *
4266 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4267 {
4268         struct nfs4_file *fp;
4269
4270         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
4271                 if (fh_match(&fp->fi_fhandle, fh)) {
4272                         if (refcount_inc_not_zero(&fp->fi_ref))
4273                                 return fp;
4274                 }
4275         }
4276         return NULL;
4277 }
4278
4279 struct nfs4_file *
4280 find_file(struct knfsd_fh *fh)
4281 {
4282         struct nfs4_file *fp;
4283         unsigned int hashval = file_hashval(fh);
4284
4285         rcu_read_lock();
4286         fp = find_file_locked(fh, hashval);
4287         rcu_read_unlock();
4288         return fp;
4289 }
4290
4291 static struct nfs4_file *
4292 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4293 {
4294         struct nfs4_file *fp;
4295         unsigned int hashval = file_hashval(fh);
4296
4297         rcu_read_lock();
4298         fp = find_file_locked(fh, hashval);
4299         rcu_read_unlock();
4300         if (fp)
4301                 return fp;
4302
4303         spin_lock(&state_lock);
4304         fp = find_file_locked(fh, hashval);
4305         if (likely(fp == NULL)) {
4306                 nfsd4_init_file(fh, hashval, new);
4307                 fp = new;
4308         }
4309         spin_unlock(&state_lock);
4310
4311         return fp;
4312 }
4313
4314 /*
4315  * Called to check deny when READ with all zero stateid or
4316  * WRITE with all zero or all one stateid
4317  */
4318 static __be32
4319 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4320 {
4321         struct nfs4_file *fp;
4322         __be32 ret = nfs_ok;
4323
4324         fp = find_file(&current_fh->fh_handle);
4325         if (!fp)
4326                 return ret;
4327         /* Check for conflicting share reservations */
4328         spin_lock(&fp->fi_lock);
4329         if (fp->fi_share_deny & deny_type)
4330                 ret = nfserr_locked;
4331         spin_unlock(&fp->fi_lock);
4332         put_nfs4_file(fp);
4333         return ret;
4334 }
4335
4336 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4337 {
4338         struct nfs4_delegation *dp = cb_to_delegation(cb);
4339         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4340                                           nfsd_net_id);
4341
4342         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4343
4344         /*
4345          * We can't do this in nfsd_break_deleg_cb because it is
4346          * already holding inode->i_lock.
4347          *
4348          * If the dl_time != 0, then we know that it has already been
4349          * queued for a lease break. Don't queue it again.
4350          */
4351         spin_lock(&state_lock);
4352         if (dp->dl_time == 0) {
4353                 dp->dl_time = get_seconds();
4354                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4355         }
4356         spin_unlock(&state_lock);
4357 }
4358
4359 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4360                 struct rpc_task *task)
4361 {
4362         struct nfs4_delegation *dp = cb_to_delegation(cb);
4363
4364         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4365                 return 1;
4366
4367         switch (task->tk_status) {
4368         case 0:
4369                 return 1;
4370         case -NFS4ERR_DELAY:
4371                 rpc_delay(task, 2 * HZ);
4372                 return 0;
4373         case -EBADHANDLE:
4374         case -NFS4ERR_BAD_STATEID:
4375                 /*
4376                  * Race: client probably got cb_recall before open reply
4377                  * granting delegation.
4378                  */
4379                 if (dp->dl_retries--) {
4380                         rpc_delay(task, 2 * HZ);
4381                         return 0;
4382                 }
4383                 /*FALLTHRU*/
4384         default:
4385                 return 1;
4386         }
4387 }
4388
4389 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4390 {
4391         struct nfs4_delegation *dp = cb_to_delegation(cb);
4392
4393         nfs4_put_stid(&dp->dl_stid);
4394 }
4395
4396 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4397         .prepare        = nfsd4_cb_recall_prepare,
4398         .done           = nfsd4_cb_recall_done,
4399         .release        = nfsd4_cb_recall_release,
4400 };
4401
4402 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4403 {
4404         /*
4405          * We're assuming the state code never drops its reference
4406          * without first removing the lease.  Since we're in this lease
4407          * callback (and since the lease code is serialized by the
4408          * i_lock) we know the server hasn't removed the lease yet, and
4409          * we know it's safe to take a reference.
4410          */
4411         refcount_inc(&dp->dl_stid.sc_count);
4412         nfsd4_run_cb(&dp->dl_recall);
4413 }
4414
4415 /* Called from break_lease() with i_lock held. */
4416 static bool
4417 nfsd_break_deleg_cb(struct file_lock *fl)
4418 {
4419         bool ret = false;
4420         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4421         struct nfs4_file *fp = dp->dl_stid.sc_file;
4422
4423         /*
4424          * We don't want the locks code to timeout the lease for us;
4425          * we'll remove it ourself if a delegation isn't returned
4426          * in time:
4427          */
4428         fl->fl_break_time = 0;
4429
4430         spin_lock(&fp->fi_lock);
4431         fp->fi_had_conflict = true;
4432         nfsd_break_one_deleg(dp);
4433         spin_unlock(&fp->fi_lock);
4434         return ret;
4435 }
4436
4437 static int
4438 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4439                      struct list_head *dispose)
4440 {
4441         if (arg & F_UNLCK)
4442                 return lease_modify(onlist, arg, dispose);
4443         else
4444                 return -EAGAIN;
4445 }
4446
4447 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4448         .lm_break = nfsd_break_deleg_cb,
4449         .lm_change = nfsd_change_deleg_cb,
4450 };
4451
4452 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4453 {
4454         if (nfsd4_has_session(cstate))
4455                 return nfs_ok;
4456         if (seqid == so->so_seqid - 1)
4457                 return nfserr_replay_me;
4458         if (seqid == so->so_seqid)
4459                 return nfs_ok;
4460         return nfserr_bad_seqid;
4461 }
4462
4463 static __be32 lookup_clientid(clientid_t *clid,
4464                 struct nfsd4_compound_state *cstate,
4465                 struct nfsd_net *nn)
4466 {
4467         struct nfs4_client *found;
4468
4469         if (cstate->clp) {
4470                 found = cstate->clp;
4471                 if (!same_clid(&found->cl_clientid, clid))
4472                         return nfserr_stale_clientid;
4473                 return nfs_ok;
4474         }
4475
4476         if (STALE_CLIENTID(clid, nn))
4477                 return nfserr_stale_clientid;
4478
4479         /*
4480          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4481          * cached already then we know this is for is for v4.0 and "sessions"
4482          * will be false.
4483          */
4484         WARN_ON_ONCE(cstate->session);
4485         spin_lock(&nn->client_lock);
4486         found = find_confirmed_client(clid, false, nn);
4487         if (!found) {
4488                 spin_unlock(&nn->client_lock);
4489                 return nfserr_expired;
4490         }
4491         atomic_inc(&found->cl_rpc_users);
4492         spin_unlock(&nn->client_lock);
4493
4494         /* Cache the nfs4_client in cstate! */
4495         cstate->clp = found;
4496         return nfs_ok;
4497 }
4498
4499 __be32
4500 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4501                     struct nfsd4_open *open, struct nfsd_net *nn)
4502 {
4503         clientid_t *clientid = &open->op_clientid;
4504         struct nfs4_client *clp = NULL;
4505         unsigned int strhashval;
4506         struct nfs4_openowner *oo = NULL;
4507         __be32 status;
4508
4509         if (STALE_CLIENTID(&open->op_clientid, nn))
4510                 return nfserr_stale_clientid;
4511         /*
4512          * In case we need it later, after we've already created the
4513          * file and don't want to risk a further failure:
4514          */
4515         open->op_file = nfsd4_alloc_file();
4516         if (open->op_file == NULL)
4517                 return nfserr_jukebox;
4518
4519         status = lookup_clientid(clientid, cstate, nn);
4520         if (status)
4521                 return status;
4522         clp = cstate->clp;
4523
4524         strhashval = ownerstr_hashval(&open->op_owner);
4525         oo = find_openstateowner_str(strhashval, open, clp);
4526         open->op_openowner = oo;
4527         if (!oo) {
4528                 goto new_owner;
4529         }
4530         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4531                 /* Replace unconfirmed owners without checking for replay. */
4532                 release_openowner(oo);
4533                 open->op_openowner = NULL;
4534                 goto new_owner;
4535         }
4536         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4537         if (status)
4538                 return status;
4539         goto alloc_stateid;
4540 new_owner:
4541         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4542         if (oo == NULL)
4543                 return nfserr_jukebox;
4544         open->op_openowner = oo;
4545 alloc_stateid:
4546         open->op_stp = nfs4_alloc_open_stateid(clp);
4547         if (!open->op_stp)
4548                 return nfserr_jukebox;
4549
4550         if (nfsd4_has_session(cstate) &&
4551             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4552                 open->op_odstate = alloc_clnt_odstate(clp);
4553                 if (!open->op_odstate)
4554                         return nfserr_jukebox;
4555         }
4556
4557         return nfs_ok;
4558 }
4559
4560 static inline __be32
4561 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4562 {
4563         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4564                 return nfserr_openmode;
4565         else
4566                 return nfs_ok;
4567 }
4568
4569 static int share_access_to_flags(u32 share_access)
4570 {
4571         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4572 }
4573
4574 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4575 {
4576         struct nfs4_stid *ret;
4577
4578         ret = find_stateid_by_type(cl, s,
4579                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4580         if (!ret)
4581                 return NULL;
4582         return delegstateid(ret);
4583 }
4584
4585 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4586 {
4587         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4588                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4589 }
4590
4591 static __be32
4592 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4593                 struct nfs4_delegation **dp)
4594 {
4595         int flags;
4596         __be32 status = nfserr_bad_stateid;
4597         struct nfs4_delegation *deleg;
4598
4599         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4600         if (deleg == NULL)
4601                 goto out;
4602         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4603                 nfs4_put_stid(&deleg->dl_stid);
4604                 if (cl->cl_minorversion)
4605                         status = nfserr_deleg_revoked;
4606                 goto out;
4607         }
4608         flags = share_access_to_flags(open->op_share_access);
4609         status = nfs4_check_delegmode(deleg, flags);
4610         if (status) {
4611                 nfs4_put_stid(&deleg->dl_stid);
4612                 goto out;
4613         }
4614         *dp = deleg;
4615 out:
4616         if (!nfsd4_is_deleg_cur(open))
4617                 return nfs_ok;
4618         if (status)
4619                 return status;
4620         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4621         return nfs_ok;
4622 }
4623
4624 static inline int nfs4_access_to_access(u32 nfs4_access)
4625 {
4626         int flags = 0;
4627
4628         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4629                 flags |= NFSD_MAY_READ;
4630         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4631                 flags |= NFSD_MAY_WRITE;
4632         return flags;
4633 }
4634
4635 static inline __be32
4636 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4637                 struct nfsd4_open *open)
4638 {
4639         struct iattr iattr = {
4640                 .ia_valid = ATTR_SIZE,
4641                 .ia_size = 0,
4642         };
4643         if (!open->op_truncate)
4644                 return 0;
4645         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4646                 return nfserr_inval;
4647         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4648 }
4649
4650 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4651                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4652                 struct nfsd4_open *open)
4653 {
4654         struct file *filp = NULL;
4655         __be32 status;
4656         int oflag = nfs4_access_to_omode(open->op_share_access);
4657         int access = nfs4_access_to_access(open->op_share_access);
4658         unsigned char old_access_bmap, old_deny_bmap;
4659
4660         spin_lock(&fp->fi_lock);
4661
4662         /*
4663          * Are we trying to set a deny mode that would conflict with
4664          * current access?
4665          */
4666         status = nfs4_file_check_deny(fp, open->op_share_deny);
4667         if (status != nfs_ok) {
4668                 spin_unlock(&fp->fi_lock);
4669                 goto out;
4670         }
4671
4672         /* set access to the file */
4673         status = nfs4_file_get_access(fp, open->op_share_access);
4674         if (status != nfs_ok) {
4675                 spin_unlock(&fp->fi_lock);
4676                 goto out;
4677         }
4678
4679         /* Set access bits in stateid */
4680         old_access_bmap = stp->st_access_bmap;
4681         set_access(open->op_share_access, stp);
4682
4683         /* Set new deny mask */
4684         old_deny_bmap = stp->st_deny_bmap;
4685         set_deny(open->op_share_deny, stp);
4686         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4687
4688         if (!fp->fi_fds[oflag]) {
4689                 spin_unlock(&fp->fi_lock);
4690                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4691                 if (status)
4692                         goto out_put_access;
4693                 spin_lock(&fp->fi_lock);
4694                 if (!fp->fi_fds[oflag]) {
4695                         fp->fi_fds[oflag] = filp;
4696                         filp = NULL;
4697                 }
4698         }
4699         spin_unlock(&fp->fi_lock);
4700         if (filp)
4701                 fput(filp);
4702
4703         status = nfsd4_truncate(rqstp, cur_fh, open);
4704         if (status)
4705                 goto out_put_access;
4706 out:
4707         return status;
4708 out_put_access:
4709         stp->st_access_bmap = old_access_bmap;
4710         nfs4_file_put_access(fp, open->op_share_access);
4711         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4712         goto out;
4713 }
4714
4715 static __be32
4716 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)
4717 {
4718         __be32 status;
4719         unsigned char old_deny_bmap = stp->st_deny_bmap;
4720
4721         if (!test_access(open->op_share_access, stp))
4722                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4723
4724         /* test and set deny mode */
4725         spin_lock(&fp->fi_lock);
4726         status = nfs4_file_check_deny(fp, open->op_share_deny);
4727         if (status == nfs_ok) {
4728                 set_deny(open->op_share_deny, stp);
4729                 fp->fi_share_deny |=
4730                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4731         }
4732         spin_unlock(&fp->fi_lock);
4733
4734         if (status != nfs_ok)
4735                 return status;
4736
4737         status = nfsd4_truncate(rqstp, cur_fh, open);
4738         if (status != nfs_ok)
4739                 reset_union_bmap_deny(old_deny_bmap, stp);
4740         return status;
4741 }
4742
4743 /* Should we give out recallable state?: */
4744 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4745 {
4746         if (clp->cl_cb_state == NFSD4_CB_UP)
4747                 return true;
4748         /*
4749          * In the sessions case, since we don't have to establish a
4750          * separate connection for callbacks, we assume it's OK
4751          * until we hear otherwise:
4752          */
4753         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4754 }
4755
4756 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4757                                                 int flag)
4758 {
4759         struct file_lock *fl;
4760
4761         fl = locks_alloc_lock();
4762         if (!fl)
4763                 return NULL;
4764         fl->fl_lmops = &nfsd_lease_mng_ops;
4765         fl->fl_flags = FL_DELEG;
4766         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4767         fl->fl_end = OFFSET_MAX;
4768         fl->fl_owner = (fl_owner_t)dp;
4769         fl->fl_pid = current->tgid;
4770         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4771         return fl;
4772 }
4773
4774 static struct nfs4_delegation *
4775 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4776                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4777 {
4778         int status = 0;
4779         struct nfs4_delegation *dp;
4780         struct file *filp;
4781         struct file_lock *fl;
4782
4783         /*
4784          * The fi_had_conflict and nfs_get_existing_delegation checks
4785          * here are just optimizations; we'll need to recheck them at
4786          * the end:
4787          */
4788         if (fp->fi_had_conflict)
4789                 return ERR_PTR(-EAGAIN);
4790
4791         filp = find_readable_file(fp);
4792         if (!filp) {
4793                 /* We should always have a readable file here */
4794                 WARN_ON_ONCE(1);
4795                 return ERR_PTR(-EBADF);
4796         }
4797         spin_lock(&state_lock);
4798         spin_lock(&fp->fi_lock);
4799         if (nfs4_delegation_exists(clp, fp))
4800                 status = -EAGAIN;
4801         else if (!fp->fi_deleg_file) {
4802                 fp->fi_deleg_file = filp;
4803                 /* increment early to prevent fi_deleg_file from being
4804                  * cleared */
4805                 fp->fi_delegees = 1;
4806                 filp = NULL;
4807         } else
4808                 fp->fi_delegees++;
4809         spin_unlock(&fp->fi_lock);
4810         spin_unlock(&state_lock);
4811         if (filp)
4812                 fput(filp);
4813         if (status)
4814                 return ERR_PTR(status);
4815
4816         status = -ENOMEM;
4817         dp = alloc_init_deleg(clp, fp, fh, odstate);
4818         if (!dp)
4819                 goto out_delegees;
4820
4821         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4822         if (!fl)
4823                 goto out_clnt_odstate;
4824
4825         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4826         if (fl)
4827                 locks_free_lock(fl);
4828         if (status)
4829                 goto out_clnt_odstate;
4830
4831         spin_lock(&state_lock);
4832         spin_lock(&fp->fi_lock);
4833         if (fp->fi_had_conflict)
4834                 status = -EAGAIN;
4835         else
4836                 status = hash_delegation_locked(dp, fp);
4837         spin_unlock(&fp->fi_lock);
4838         spin_unlock(&state_lock);
4839
4840         if (status)
4841                 goto out_unlock;
4842
4843         return dp;
4844 out_unlock:
4845         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4846 out_clnt_odstate:
4847         put_clnt_odstate(dp->dl_clnt_odstate);
4848         nfs4_put_stid(&dp->dl_stid);
4849 out_delegees:
4850         put_deleg_file(fp);
4851         return ERR_PTR(status);
4852 }
4853
4854 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4855 {
4856         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4857         if (status == -EAGAIN)
4858                 open->op_why_no_deleg = WND4_CONTENTION;
4859         else {
4860                 open->op_why_no_deleg = WND4_RESOURCE;
4861                 switch (open->op_deleg_want) {
4862                 case NFS4_SHARE_WANT_READ_DELEG:
4863                 case NFS4_SHARE_WANT_WRITE_DELEG:
4864                 case NFS4_SHARE_WANT_ANY_DELEG:
4865                         break;
4866                 case NFS4_SHARE_WANT_CANCEL:
4867                         open->op_why_no_deleg = WND4_CANCELLED;
4868                         break;
4869                 case NFS4_SHARE_WANT_NO_DELEG:
4870                         WARN_ON_ONCE(1);
4871                 }
4872         }
4873 }
4874
4875 /*
4876  * Attempt to hand out a delegation.
4877  *
4878  * Note we don't support write delegations, and won't until the vfs has
4879  * proper support for them.
4880  */
4881 static void
4882 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4883                         struct nfs4_ol_stateid *stp)
4884 {
4885         struct nfs4_delegation *dp;
4886         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4887         struct nfs4_client *clp = stp->st_stid.sc_client;
4888         int cb_up;
4889         int status = 0;
4890
4891         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4892         open->op_recall = 0;
4893         switch (open->op_claim_type) {
4894                 case NFS4_OPEN_CLAIM_PREVIOUS:
4895                         if (!cb_up)
4896                                 open->op_recall = 1;
4897                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4898                                 goto out_no_deleg;
4899                         break;
4900                 case NFS4_OPEN_CLAIM_NULL:
4901                 case NFS4_OPEN_CLAIM_FH:
4902                         /*
4903                          * Let's not give out any delegations till everyone's
4904                          * had the chance to reclaim theirs, *and* until
4905                          * NLM locks have all been reclaimed:
4906                          */
4907                         if (locks_in_grace(clp->net))
4908                                 goto out_no_deleg;
4909                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4910                                 goto out_no_deleg;
4911                         /*
4912                          * Also, if the file was opened for write or
4913                          * create, there's a good chance the client's
4914                          * about to write to it, resulting in an
4915                          * immediate recall (since we don't support
4916                          * write delegations):
4917                          */
4918                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4919                                 goto out_no_deleg;
4920                         if (open->op_create == NFS4_OPEN_CREATE)
4921                                 goto out_no_deleg;
4922                         break;
4923                 default:
4924                         goto out_no_deleg;
4925         }
4926         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4927         if (IS_ERR(dp))
4928                 goto out_no_deleg;
4929
4930         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4931
4932         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4933                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4934         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4935         nfs4_put_stid(&dp->dl_stid);
4936         return;
4937 out_no_deleg:
4938         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4939         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4940             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4941                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4942                 open->op_recall = 1;
4943         }
4944
4945         /* 4.1 client asking for a delegation? */
4946         if (open->op_deleg_want)
4947                 nfsd4_open_deleg_none_ext(open, status);
4948         return;
4949 }
4950
4951 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4952                                         struct nfs4_delegation *dp)
4953 {
4954         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4955             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4956                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4957                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4958         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4959                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4960                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4961                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4962         }
4963         /* Otherwise the client must be confused wanting a delegation
4964          * it already has, therefore we don't return
4965          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4966          */
4967 }
4968
4969 __be32
4970 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4971 {
4972         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4973         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4974         struct nfs4_file *fp = NULL;
4975         struct nfs4_ol_stateid *stp = NULL;
4976         struct nfs4_delegation *dp = NULL;
4977         __be32 status;
4978         bool new_stp = false;
4979
4980         /*
4981          * Lookup file; if found, lookup stateid and check open request,
4982          * and check for delegations in the process of being recalled.
4983          * If not found, create the nfs4_file struct
4984          */
4985         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4986         if (fp != open->op_file) {
4987                 status = nfs4_check_deleg(cl, open, &dp);
4988                 if (status)
4989                         goto out;
4990                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4991         } else {
4992                 open->op_file = NULL;
4993                 status = nfserr_bad_stateid;
4994                 if (nfsd4_is_deleg_cur(open))
4995                         goto out;
4996         }
4997
4998         if (!stp) {
4999                 stp = init_open_stateid(fp, open);
5000                 if (!open->op_stp)
5001                         new_stp = true;
5002         }
5003
5004         /*
5005          * OPEN the file, or upgrade an existing OPEN.
5006          * If truncate fails, the OPEN fails.
5007          *
5008          * stp is already locked.
5009          */
5010         if (!new_stp) {
5011                 /* Stateid was found, this is an OPEN upgrade */
5012                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5013                 if (status) {
5014                         mutex_unlock(&stp->st_mutex);
5015                         goto out;
5016                 }
5017         } else {
5018                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5019                 if (status) {
5020                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5021                         release_open_stateid(stp);
5022                         mutex_unlock(&stp->st_mutex);
5023                         goto out;
5024                 }
5025
5026                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5027                                                         open->op_odstate);
5028                 if (stp->st_clnt_odstate == open->op_odstate)
5029                         open->op_odstate = NULL;
5030         }
5031
5032         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5033         mutex_unlock(&stp->st_mutex);
5034
5035         if (nfsd4_has_session(&resp->cstate)) {
5036                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5037                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5038                         open->op_why_no_deleg = WND4_NOT_WANTED;
5039                         goto nodeleg;
5040                 }
5041         }
5042
5043         /*
5044         * Attempt to hand out a delegation. No error return, because the
5045         * OPEN succeeds even if we fail.
5046         */
5047         nfs4_open_delegation(current_fh, open, stp);
5048 nodeleg:
5049         status = nfs_ok;
5050
5051         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5052                 STATEID_VAL(&stp->st_stid.sc_stateid));
5053 out:
5054         /* 4.1 client trying to upgrade/downgrade delegation? */
5055         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5056             open->op_deleg_want)
5057                 nfsd4_deleg_xgrade_none_ext(open, dp);
5058
5059         if (fp)
5060                 put_nfs4_file(fp);
5061         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5062                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5063         /*
5064         * To finish the open response, we just need to set the rflags.
5065         */
5066         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5067         if (nfsd4_has_session(&resp->cstate))
5068                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5069         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5070                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5071
5072         if (dp)
5073                 nfs4_put_stid(&dp->dl_stid);
5074         if (stp)
5075                 nfs4_put_stid(&stp->st_stid);
5076
5077         return status;
5078 }
5079
5080 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5081                               struct nfsd4_open *open)
5082 {
5083         if (open->op_openowner) {
5084                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5085
5086                 nfsd4_cstate_assign_replay(cstate, so);
5087                 nfs4_put_stateowner(so);
5088         }
5089         if (open->op_file)
5090                 kmem_cache_free(file_slab, open->op_file);
5091         if (open->op_stp)
5092                 nfs4_put_stid(&open->op_stp->st_stid);
5093         if (open->op_odstate)
5094                 kmem_cache_free(odstate_slab, open->op_odstate);
5095 }
5096
5097 __be32
5098 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5099             union nfsd4_op_u *u)
5100 {
5101         clientid_t *clid = &u->renew;
5102         struct nfs4_client *clp;
5103         __be32 status;
5104         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5105
5106         dprintk("process_renew(%08x/%08x): starting\n",
5107                         clid->cl_boot, clid->cl_id);
5108         status = lookup_clientid(clid, cstate, nn);
5109         if (status)
5110                 goto out;
5111         clp = cstate->clp;
5112         status = nfserr_cb_path_down;
5113         if (!list_empty(&clp->cl_delegations)
5114                         && clp->cl_cb_state != NFSD4_CB_UP)
5115                 goto out;
5116         status = nfs_ok;
5117 out:
5118         return status;
5119 }
5120
5121 void
5122 nfsd4_end_grace(struct nfsd_net *nn)
5123 {
5124         /* do nothing if grace period already ended */
5125         if (nn->grace_ended)
5126                 return;
5127
5128         nn->grace_ended = true;
5129         /*
5130          * If the server goes down again right now, an NFSv4
5131          * client will still be allowed to reclaim after it comes back up,
5132          * even if it hasn't yet had a chance to reclaim state this time.
5133          *
5134          */
5135         nfsd4_record_grace_done(nn);
5136         /*
5137          * At this point, NFSv4 clients can still reclaim.  But if the
5138          * server crashes, any that have not yet reclaimed will be out
5139          * of luck on the next boot.
5140          *
5141          * (NFSv4.1+ clients are considered to have reclaimed once they
5142          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5143          * have reclaimed after their first OPEN.)
5144          */
5145         locks_end_grace(&nn->nfsd4_manager);
5146         /*
5147          * At this point, and once lockd and/or any other containers
5148          * exit their grace period, further reclaims will fail and
5149          * regular locking can resume.
5150          */
5151 }
5152
5153 /*
5154  * If we've waited a lease period but there are still clients trying to
5155  * reclaim, wait a little longer to give them a chance to finish.
5156  */
5157 static bool clients_still_reclaiming(struct nfsd_net *nn)
5158 {
5159         unsigned long now = get_seconds();
5160         unsigned long double_grace_period_end = nn->boot_time +
5161                                                 2 * nn->nfsd4_lease;
5162
5163         if (nn->track_reclaim_completes &&
5164                         atomic_read(&nn->nr_reclaim_complete) ==
5165                         nn->reclaim_str_hashtbl_size)
5166                 return false;
5167         if (!nn->somebody_reclaimed)
5168                 return false;
5169         nn->somebody_reclaimed = false;
5170         /*
5171          * If we've given them *two* lease times to reclaim, and they're
5172          * still not done, give up:
5173          */
5174         if (time_after(now, double_grace_period_end))
5175                 return false;
5176         return true;
5177 }
5178
5179 static time_t
5180 nfs4_laundromat(struct nfsd_net *nn)
5181 {
5182         struct nfs4_client *clp;
5183         struct nfs4_openowner *oo;
5184         struct nfs4_delegation *dp;
5185         struct nfs4_ol_stateid *stp;
5186         struct nfsd4_blocked_lock *nbl;
5187         struct list_head *pos, *next, reaplist;
5188         time_t cutoff = get_seconds() - nn->nfsd4_lease;
5189         time_t t, new_timeo = nn->nfsd4_lease;
5190
5191         dprintk("NFSD: laundromat service - starting\n");
5192
5193         if (clients_still_reclaiming(nn)) {
5194                 new_timeo = 0;
5195                 goto out;
5196         }
5197         dprintk("NFSD: end of grace period\n");
5198         nfsd4_end_grace(nn);
5199         INIT_LIST_HEAD(&reaplist);
5200         spin_lock(&nn->client_lock);
5201         list_for_each_safe(pos, next, &nn->client_lru) {
5202                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5203                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
5204                         t = clp->cl_time - cutoff;
5205                         new_timeo = min(new_timeo, t);
5206                         break;
5207                 }
5208                 if (mark_client_expired_locked(clp)) {
5209                         dprintk("NFSD: client in use (clientid %08x)\n",
5210                                 clp->cl_clientid.cl_id);
5211                         continue;
5212                 }
5213                 list_add(&clp->cl_lru, &reaplist);
5214         }
5215         spin_unlock(&nn->client_lock);
5216         list_for_each_safe(pos, next, &reaplist) {
5217                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5218                 dprintk("NFSD: purging unused client (clientid %08x)\n",
5219                         clp->cl_clientid.cl_id);
5220                 list_del_init(&clp->cl_lru);
5221                 expire_client(clp);
5222         }
5223         spin_lock(&state_lock);
5224         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5225                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5226                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
5227                         t = dp->dl_time - cutoff;
5228                         new_timeo = min(new_timeo, t);
5229                         break;
5230                 }
5231                 WARN_ON(!unhash_delegation_locked(dp));
5232                 list_add(&dp->dl_recall_lru, &reaplist);
5233         }
5234         spin_unlock(&state_lock);
5235         while (!list_empty(&reaplist)) {
5236                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5237                                         dl_recall_lru);
5238                 list_del_init(&dp->dl_recall_lru);
5239                 revoke_delegation(dp);
5240         }
5241
5242         spin_lock(&nn->client_lock);
5243         while (!list_empty(&nn->close_lru)) {
5244                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5245                                         oo_close_lru);
5246                 if (time_after((unsigned long)oo->oo_time,
5247                                (unsigned long)cutoff)) {
5248                         t = oo->oo_time - cutoff;
5249                         new_timeo = min(new_timeo, t);
5250                         break;
5251                 }
5252                 list_del_init(&oo->oo_close_lru);
5253                 stp = oo->oo_last_closed_stid;
5254                 oo->oo_last_closed_stid = NULL;
5255                 spin_unlock(&nn->client_lock);
5256                 nfs4_put_stid(&stp->st_stid);
5257                 spin_lock(&nn->client_lock);
5258         }
5259         spin_unlock(&nn->client_lock);
5260
5261         /*
5262          * It's possible for a client to try and acquire an already held lock
5263          * that is being held for a long time, and then lose interest in it.
5264          * So, we clean out any un-revisited request after a lease period
5265          * under the assumption that the client is no longer interested.
5266          *
5267          * RFC5661, sec. 9.6 states that the client must not rely on getting
5268          * notifications and must continue to poll for locks, even when the
5269          * server supports them. Thus this shouldn't lead to clients blocking
5270          * indefinitely once the lock does become free.
5271          */
5272         BUG_ON(!list_empty(&reaplist));
5273         spin_lock(&nn->blocked_locks_lock);
5274         while (!list_empty(&nn->blocked_locks_lru)) {
5275                 nbl = list_first_entry(&nn->blocked_locks_lru,
5276                                         struct nfsd4_blocked_lock, nbl_lru);
5277                 if (time_after((unsigned long)nbl->nbl_time,
5278                                (unsigned long)cutoff)) {
5279                         t = nbl->nbl_time - cutoff;
5280                         new_timeo = min(new_timeo, t);
5281                         break;
5282                 }
5283                 list_move(&nbl->nbl_lru, &reaplist);
5284                 list_del_init(&nbl->nbl_list);
5285         }
5286         spin_unlock(&nn->blocked_locks_lock);
5287
5288         while (!list_empty(&reaplist)) {
5289                 nbl = list_first_entry(&reaplist,
5290                                         struct nfsd4_blocked_lock, nbl_lru);
5291                 list_del_init(&nbl->nbl_lru);
5292                 free_blocked_lock(nbl);
5293         }
5294 out:
5295         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5296         return new_timeo;
5297 }
5298
5299 static struct workqueue_struct *laundry_wq;
5300 static void laundromat_main(struct work_struct *);
5301
5302 static void
5303 laundromat_main(struct work_struct *laundry)
5304 {
5305         time_t t;
5306         struct delayed_work *dwork = to_delayed_work(laundry);
5307         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5308                                            laundromat_work);
5309
5310         t = nfs4_laundromat(nn);
5311         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
5312         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5313 }
5314
5315 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5316 {
5317         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5318                 return nfserr_bad_stateid;
5319         return nfs_ok;
5320 }
5321
5322 static inline int
5323 access_permit_read(struct nfs4_ol_stateid *stp)
5324 {
5325         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5326                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5327                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5328 }
5329
5330 static inline int
5331 access_permit_write(struct nfs4_ol_stateid *stp)
5332 {
5333         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5334                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5335 }
5336
5337 static
5338 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5339 {
5340         __be32 status = nfserr_openmode;
5341
5342         /* For lock stateid's, we test the parent open, not the lock: */
5343         if (stp->st_openstp)
5344                 stp = stp->st_openstp;
5345         if ((flags & WR_STATE) && !access_permit_write(stp))
5346                 goto out;
5347         if ((flags & RD_STATE) && !access_permit_read(stp))
5348                 goto out;
5349         status = nfs_ok;
5350 out:
5351         return status;
5352 }
5353
5354 static inline __be32
5355 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5356 {
5357         if (ONE_STATEID(stateid) && (flags & RD_STATE))
5358                 return nfs_ok;
5359         else if (opens_in_grace(net)) {
5360                 /* Answer in remaining cases depends on existence of
5361                  * conflicting state; so we must wait out the grace period. */
5362                 return nfserr_grace;
5363         } else if (flags & WR_STATE)
5364                 return nfs4_share_conflict(current_fh,
5365                                 NFS4_SHARE_DENY_WRITE);
5366         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5367                 return nfs4_share_conflict(current_fh,
5368                                 NFS4_SHARE_DENY_READ);
5369 }
5370
5371 /*
5372  * Allow READ/WRITE during grace period on recovered state only for files
5373  * that are not able to provide mandatory locking.
5374  */
5375 static inline int
5376 grace_disallows_io(struct net *net, struct inode *inode)
5377 {
5378         return opens_in_grace(net) && mandatory_lock(inode);
5379 }
5380
5381 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5382 {
5383         /*
5384          * When sessions are used the stateid generation number is ignored
5385          * when it is zero.
5386          */
5387         if (has_session && in->si_generation == 0)
5388                 return nfs_ok;
5389
5390         if (in->si_generation == ref->si_generation)
5391                 return nfs_ok;
5392
5393         /* If the client sends us a stateid from the future, it's buggy: */
5394         if (nfsd4_stateid_generation_after(in, ref))
5395                 return nfserr_bad_stateid;
5396         /*
5397          * However, we could see a stateid from the past, even from a
5398          * non-buggy client.  For example, if the client sends a lock
5399          * while some IO is outstanding, the lock may bump si_generation
5400          * while the IO is still in flight.  The client could avoid that
5401          * situation by waiting for responses on all the IO requests,
5402          * but better performance may result in retrying IO that
5403          * receives an old_stateid error if requests are rarely
5404          * reordered in flight:
5405          */
5406         return nfserr_old_stateid;
5407 }
5408
5409 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5410 {
5411         __be32 ret;
5412
5413         spin_lock(&s->sc_lock);
5414         ret = nfsd4_verify_open_stid(s);
5415         if (ret == nfs_ok)
5416                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5417         spin_unlock(&s->sc_lock);
5418         return ret;
5419 }
5420
5421 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5422 {
5423         if (ols->st_stateowner->so_is_open_owner &&
5424             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5425                 return nfserr_bad_stateid;
5426         return nfs_ok;
5427 }
5428
5429 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5430 {
5431         struct nfs4_stid *s;
5432         __be32 status = nfserr_bad_stateid;
5433
5434         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5435                 CLOSE_STATEID(stateid))
5436                 return status;
5437         /* Client debugging aid. */
5438         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5439                 char addr_str[INET6_ADDRSTRLEN];
5440                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5441                                  sizeof(addr_str));
5442                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5443                                         "with incorrect client ID\n", addr_str);
5444                 return status;
5445         }
5446         spin_lock(&cl->cl_lock);
5447         s = find_stateid_locked(cl, stateid);
5448         if (!s)
5449                 goto out_unlock;
5450         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5451         if (status)
5452                 goto out_unlock;
5453         switch (s->sc_type) {
5454         case NFS4_DELEG_STID:
5455                 status = nfs_ok;
5456                 break;
5457         case NFS4_REVOKED_DELEG_STID:
5458                 status = nfserr_deleg_revoked;
5459                 break;
5460         case NFS4_OPEN_STID:
5461         case NFS4_LOCK_STID:
5462                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5463                 break;
5464         default:
5465                 printk("unknown stateid type %x\n", s->sc_type);
5466                 /* Fallthrough */
5467         case NFS4_CLOSED_STID:
5468         case NFS4_CLOSED_DELEG_STID:
5469                 status = nfserr_bad_stateid;
5470         }
5471 out_unlock:
5472         spin_unlock(&cl->cl_lock);
5473         return status;
5474 }
5475
5476 __be32
5477 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5478                      stateid_t *stateid, unsigned char typemask,
5479                      struct nfs4_stid **s, struct nfsd_net *nn)
5480 {
5481         __be32 status;
5482         bool return_revoked = false;
5483
5484         /*
5485          *  only return revoked delegations if explicitly asked.
5486          *  otherwise we report revoked or bad_stateid status.
5487          */
5488         if (typemask & NFS4_REVOKED_DELEG_STID)
5489                 return_revoked = true;
5490         else if (typemask & NFS4_DELEG_STID)
5491                 typemask |= NFS4_REVOKED_DELEG_STID;
5492
5493         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5494                 CLOSE_STATEID(stateid))
5495                 return nfserr_bad_stateid;
5496         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5497         if (status == nfserr_stale_clientid) {
5498                 if (cstate->session)
5499                         return nfserr_bad_stateid;
5500                 return nfserr_stale_stateid;
5501         }
5502         if (status)
5503                 return status;
5504         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5505         if (!*s)
5506                 return nfserr_bad_stateid;
5507         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5508                 nfs4_put_stid(*s);
5509                 if (cstate->minorversion)
5510                         return nfserr_deleg_revoked;
5511                 return nfserr_bad_stateid;
5512         }
5513         return nfs_ok;
5514 }
5515
5516 static struct file *
5517 nfs4_find_file(struct nfs4_stid *s, int flags)
5518 {
5519         if (!s)
5520                 return NULL;
5521
5522         switch (s->sc_type) {
5523         case NFS4_DELEG_STID:
5524                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5525                         return NULL;
5526                 return get_file(s->sc_file->fi_deleg_file);
5527         case NFS4_OPEN_STID:
5528         case NFS4_LOCK_STID:
5529                 if (flags & RD_STATE)
5530                         return find_readable_file(s->sc_file);
5531                 else
5532                         return find_writeable_file(s->sc_file);
5533                 break;
5534         }
5535
5536         return NULL;
5537 }
5538
5539 static __be32
5540 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5541 {
5542         __be32 status;
5543
5544         status = nfsd4_check_openowner_confirmed(ols);
5545         if (status)
5546                 return status;
5547         return nfs4_check_openmode(ols, flags);
5548 }
5549
5550 static __be32
5551 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5552                 struct file **filpp, bool *tmp_file, int flags)
5553 {
5554         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5555         struct file *file;
5556         __be32 status;
5557
5558         file = nfs4_find_file(s, flags);
5559         if (file) {
5560                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5561                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5562                 if (status) {
5563                         fput(file);
5564                         return status;
5565                 }
5566
5567                 *filpp = file;
5568         } else {
5569                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5570                 if (status)
5571                         return status;
5572
5573                 if (tmp_file)
5574                         *tmp_file = true;
5575         }
5576
5577         return 0;
5578 }
5579
5580 /*
5581  * Checks for stateid operations
5582  */
5583 __be32
5584 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5585                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5586                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5587 {
5588         struct inode *ino = d_inode(fhp->fh_dentry);
5589         struct net *net = SVC_NET(rqstp);
5590         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5591         struct nfs4_stid *s = NULL;
5592         __be32 status;
5593
5594         if (filpp)
5595                 *filpp = NULL;
5596         if (tmp_file)
5597                 *tmp_file = false;
5598
5599         if (grace_disallows_io(net, ino))
5600                 return nfserr_grace;
5601
5602         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5603                 status = check_special_stateids(net, fhp, stateid, flags);
5604                 goto done;
5605         }
5606
5607         status = nfsd4_lookup_stateid(cstate, stateid,
5608                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5609                                 &s, nn);
5610         if (status)
5611                 return status;
5612         status = nfsd4_stid_check_stateid_generation(stateid, s,
5613                         nfsd4_has_session(cstate));
5614         if (status)
5615                 goto out;
5616
5617         switch (s->sc_type) {
5618         case NFS4_DELEG_STID:
5619                 status = nfs4_check_delegmode(delegstateid(s), flags);
5620                 break;
5621         case NFS4_OPEN_STID:
5622         case NFS4_LOCK_STID:
5623                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5624                 break;
5625         default:
5626                 status = nfserr_bad_stateid;
5627                 break;
5628         }
5629         if (status)
5630                 goto out;
5631         status = nfs4_check_fh(fhp, s);
5632
5633 done:
5634         if (!status && filpp)
5635                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5636 out:
5637         if (s)
5638                 nfs4_put_stid(s);
5639         return status;
5640 }
5641
5642 /*
5643  * Test if the stateid is valid
5644  */
5645 __be32
5646 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5647                    union nfsd4_op_u *u)
5648 {
5649         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5650         struct nfsd4_test_stateid_id *stateid;
5651         struct nfs4_client *cl = cstate->session->se_client;
5652
5653         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5654                 stateid->ts_id_status =
5655                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5656
5657         return nfs_ok;
5658 }
5659
5660 static __be32
5661 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5662 {
5663         struct nfs4_ol_stateid *stp = openlockstateid(s);
5664         __be32 ret;
5665
5666         ret = nfsd4_lock_ol_stateid(stp);
5667         if (ret)
5668                 goto out_put_stid;
5669
5670         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5671         if (ret)
5672                 goto out;
5673
5674         ret = nfserr_locks_held;
5675         if (check_for_locks(stp->st_stid.sc_file,
5676                             lockowner(stp->st_stateowner)))
5677                 goto out;
5678
5679         release_lock_stateid(stp);
5680         ret = nfs_ok;
5681
5682 out:
5683         mutex_unlock(&stp->st_mutex);
5684 out_put_stid:
5685         nfs4_put_stid(s);
5686         return ret;
5687 }
5688
5689 __be32
5690 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5691                    union nfsd4_op_u *u)
5692 {
5693         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5694         stateid_t *stateid = &free_stateid->fr_stateid;
5695         struct nfs4_stid *s;
5696         struct nfs4_delegation *dp;
5697         struct nfs4_client *cl = cstate->session->se_client;
5698         __be32 ret = nfserr_bad_stateid;
5699
5700         spin_lock(&cl->cl_lock);
5701         s = find_stateid_locked(cl, stateid);
5702         if (!s)
5703                 goto out_unlock;
5704         spin_lock(&s->sc_lock);
5705         switch (s->sc_type) {
5706         case NFS4_DELEG_STID:
5707                 ret = nfserr_locks_held;
5708                 break;
5709         case NFS4_OPEN_STID:
5710                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5711                 if (ret)
5712                         break;
5713                 ret = nfserr_locks_held;
5714                 break;
5715         case NFS4_LOCK_STID:
5716                 spin_unlock(&s->sc_lock);
5717                 refcount_inc(&s->sc_count);
5718                 spin_unlock(&cl->cl_lock);
5719                 ret = nfsd4_free_lock_stateid(stateid, s);
5720                 goto out;
5721         case NFS4_REVOKED_DELEG_STID:
5722                 spin_unlock(&s->sc_lock);
5723                 dp = delegstateid(s);
5724                 list_del_init(&dp->dl_recall_lru);
5725                 spin_unlock(&cl->cl_lock);
5726                 nfs4_put_stid(s);
5727                 ret = nfs_ok;
5728                 goto out;
5729         /* Default falls through and returns nfserr_bad_stateid */
5730         }
5731         spin_unlock(&s->sc_lock);
5732 out_unlock:
5733         spin_unlock(&cl->cl_lock);
5734 out:
5735         return ret;
5736 }
5737
5738 static inline int
5739 setlkflg (int type)
5740 {
5741         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5742                 RD_STATE : WR_STATE;
5743 }
5744
5745 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5746 {
5747         struct svc_fh *current_fh = &cstate->current_fh;
5748         struct nfs4_stateowner *sop = stp->st_stateowner;
5749         __be32 status;
5750
5751         status = nfsd4_check_seqid(cstate, sop, seqid);
5752         if (status)
5753                 return status;
5754         status = nfsd4_lock_ol_stateid(stp);
5755         if (status != nfs_ok)
5756                 return status;
5757         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5758         if (status == nfs_ok)
5759                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5760         if (status != nfs_ok)
5761                 mutex_unlock(&stp->st_mutex);
5762         return status;
5763 }
5764
5765 /*
5766  * Checks for sequence id mutating operations.
5767  */
5768 static __be32
5769 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5770                          stateid_t *stateid, char typemask,
5771                          struct nfs4_ol_stateid **stpp,
5772                          struct nfsd_net *nn)
5773 {
5774         __be32 status;
5775         struct nfs4_stid *s;
5776         struct nfs4_ol_stateid *stp = NULL;
5777
5778         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5779                 seqid, STATEID_VAL(stateid));
5780
5781         *stpp = NULL;
5782         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5783         if (status)
5784                 return status;
5785         stp = openlockstateid(s);
5786         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5787
5788         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5789         if (!status)
5790                 *stpp = stp;
5791         else
5792                 nfs4_put_stid(&stp->st_stid);
5793         return status;
5794 }
5795
5796 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5797                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5798 {
5799         __be32 status;
5800         struct nfs4_openowner *oo;
5801         struct nfs4_ol_stateid *stp;
5802
5803         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5804                                                 NFS4_OPEN_STID, &stp, nn);
5805         if (status)
5806                 return status;
5807         oo = openowner(stp->st_stateowner);
5808         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5809                 mutex_unlock(&stp->st_mutex);
5810                 nfs4_put_stid(&stp->st_stid);
5811                 return nfserr_bad_stateid;
5812         }
5813         *stpp = stp;
5814         return nfs_ok;
5815 }
5816
5817 __be32
5818 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5819                    union nfsd4_op_u *u)
5820 {
5821         struct nfsd4_open_confirm *oc = &u->open_confirm;
5822         __be32 status;
5823         struct nfs4_openowner *oo;
5824         struct nfs4_ol_stateid *stp;
5825         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5826
5827         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5828                         cstate->current_fh.fh_dentry);
5829
5830         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5831         if (status)
5832                 return status;
5833
5834         status = nfs4_preprocess_seqid_op(cstate,
5835                                         oc->oc_seqid, &oc->oc_req_stateid,
5836                                         NFS4_OPEN_STID, &stp, nn);
5837         if (status)
5838                 goto out;
5839         oo = openowner(stp->st_stateowner);
5840         status = nfserr_bad_stateid;
5841         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5842                 mutex_unlock(&stp->st_mutex);
5843                 goto put_stateid;
5844         }
5845         oo->oo_flags |= NFS4_OO_CONFIRMED;
5846         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5847         mutex_unlock(&stp->st_mutex);
5848         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5849                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5850
5851         nfsd4_client_record_create(oo->oo_owner.so_client);
5852         status = nfs_ok;
5853 put_stateid:
5854         nfs4_put_stid(&stp->st_stid);
5855 out:
5856         nfsd4_bump_seqid(cstate, status);
5857         return status;
5858 }
5859
5860 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5861 {
5862         if (!test_access(access, stp))
5863                 return;
5864         nfs4_file_put_access(stp->st_stid.sc_file, access);
5865         clear_access(access, stp);
5866 }
5867
5868 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5869 {
5870         switch (to_access) {
5871         case NFS4_SHARE_ACCESS_READ:
5872                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5873                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5874                 break;
5875         case NFS4_SHARE_ACCESS_WRITE:
5876                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5877                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5878                 break;
5879         case NFS4_SHARE_ACCESS_BOTH:
5880                 break;
5881         default:
5882                 WARN_ON_ONCE(1);
5883         }
5884 }
5885
5886 __be32
5887 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5888                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5889 {
5890         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5891         __be32 status;
5892         struct nfs4_ol_stateid *stp;
5893         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5894
5895         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5896                         cstate->current_fh.fh_dentry);
5897
5898         /* We don't yet support WANT bits: */
5899         if (od->od_deleg_want)
5900                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5901                         od->od_deleg_want);
5902
5903         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5904                                         &od->od_stateid, &stp, nn);
5905         if (status)
5906                 goto out;
5907         status = nfserr_inval;
5908         if (!test_access(od->od_share_access, stp)) {
5909                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5910                         stp->st_access_bmap, od->od_share_access);
5911                 goto put_stateid;
5912         }
5913         if (!test_deny(od->od_share_deny, stp)) {
5914                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5915                         stp->st_deny_bmap, od->od_share_deny);
5916                 goto put_stateid;
5917         }
5918         nfs4_stateid_downgrade(stp, od->od_share_access);
5919         reset_union_bmap_deny(od->od_share_deny, stp);
5920         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5921         status = nfs_ok;
5922 put_stateid:
5923         mutex_unlock(&stp->st_mutex);
5924         nfs4_put_stid(&stp->st_stid);
5925 out:
5926         nfsd4_bump_seqid(cstate, status);
5927         return status;
5928 }
5929
5930 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5931 {
5932         struct nfs4_client *clp = s->st_stid.sc_client;
5933         bool unhashed;
5934         LIST_HEAD(reaplist);
5935
5936         spin_lock(&clp->cl_lock);
5937         unhashed = unhash_open_stateid(s, &reaplist);
5938
5939         if (clp->cl_minorversion) {
5940                 if (unhashed)
5941                         put_ol_stateid_locked(s, &reaplist);
5942                 spin_unlock(&clp->cl_lock);
5943                 free_ol_stateid_reaplist(&reaplist);
5944         } else {
5945                 spin_unlock(&clp->cl_lock);
5946                 free_ol_stateid_reaplist(&reaplist);
5947                 if (unhashed)
5948                         move_to_close_lru(s, clp->net);
5949         }
5950 }
5951
5952 /*
5953  * nfs4_unlock_state() called after encode
5954  */
5955 __be32
5956 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5957                 union nfsd4_op_u *u)
5958 {
5959         struct nfsd4_close *close = &u->close;
5960         __be32 status;
5961         struct nfs4_ol_stateid *stp;
5962         struct net *net = SVC_NET(rqstp);
5963         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5964
5965         dprintk("NFSD: nfsd4_close on file %pd\n",
5966                         cstate->current_fh.fh_dentry);
5967
5968         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5969                                         &close->cl_stateid,
5970                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5971                                         &stp, nn);
5972         nfsd4_bump_seqid(cstate, status);
5973         if (status)
5974                 goto out;
5975
5976         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5977
5978         /*
5979          * Technically we don't _really_ have to increment or copy it, since
5980          * it should just be gone after this operation and we clobber the
5981          * copied value below, but we continue to do so here just to ensure
5982          * that racing ops see that there was a state change.
5983          */
5984         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5985
5986         nfsd4_close_open_stateid(stp);
5987         mutex_unlock(&stp->st_mutex);
5988
5989         /* v4.1+ suggests that we send a special stateid in here, since the
5990          * clients should just ignore this anyway. Since this is not useful
5991          * for v4.0 clients either, we set it to the special close_stateid
5992          * universally.
5993          *
5994          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5995          */
5996         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5997
5998         /* put reference from nfs4_preprocess_seqid_op */
5999         nfs4_put_stid(&stp->st_stid);
6000 out:
6001         return status;
6002 }
6003
6004 __be32
6005 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6006                   union nfsd4_op_u *u)
6007 {
6008         struct nfsd4_delegreturn *dr = &u->delegreturn;
6009         struct nfs4_delegation *dp;
6010         stateid_t *stateid = &dr->dr_stateid;
6011         struct nfs4_stid *s;
6012         __be32 status;
6013         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6014
6015         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6016                 return status;
6017
6018         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6019         if (status)
6020                 goto out;
6021         dp = delegstateid(s);
6022         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6023         if (status)
6024                 goto put_stateid;
6025
6026         destroy_delegation(dp);
6027 put_stateid:
6028         nfs4_put_stid(&dp->dl_stid);
6029 out:
6030         return status;
6031 }
6032
6033 static inline u64
6034 end_offset(u64 start, u64 len)
6035 {
6036         u64 end;
6037
6038         end = start + len;
6039         return end >= start ? end: NFS4_MAX_UINT64;
6040 }
6041
6042 /* last octet in a range */
6043 static inline u64
6044 last_byte_offset(u64 start, u64 len)
6045 {
6046         u64 end;
6047
6048         WARN_ON_ONCE(!len);
6049         end = start + len;
6050         return end > start ? end - 1: NFS4_MAX_UINT64;
6051 }
6052
6053 /*
6054  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6055  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6056  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
6057  * locking, this prevents us from being completely protocol-compliant.  The
6058  * real solution to this problem is to start using unsigned file offsets in
6059  * the VFS, but this is a very deep change!
6060  */
6061 static inline void
6062 nfs4_transform_lock_offset(struct file_lock *lock)
6063 {
6064         if (lock->fl_start < 0)
6065                 lock->fl_start = OFFSET_MAX;
6066         if (lock->fl_end < 0)
6067                 lock->fl_end = OFFSET_MAX;
6068 }
6069
6070 static fl_owner_t
6071 nfsd4_fl_get_owner(fl_owner_t owner)
6072 {
6073         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6074
6075         nfs4_get_stateowner(&lo->lo_owner);
6076         return owner;
6077 }
6078
6079 static void
6080 nfsd4_fl_put_owner(fl_owner_t owner)
6081 {
6082         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6083
6084         if (lo)
6085                 nfs4_put_stateowner(&lo->lo_owner);
6086 }
6087
6088 static void
6089 nfsd4_lm_notify(struct file_lock *fl)
6090 {
6091         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
6092         struct net                      *net = lo->lo_owner.so_client->net;
6093         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
6094         struct nfsd4_blocked_lock       *nbl = container_of(fl,
6095                                                 struct nfsd4_blocked_lock, nbl_lock);
6096         bool queue = false;
6097
6098         /* An empty list means that something else is going to be using it */
6099         spin_lock(&nn->blocked_locks_lock);
6100         if (!list_empty(&nbl->nbl_list)) {
6101                 list_del_init(&nbl->nbl_list);
6102                 list_del_init(&nbl->nbl_lru);
6103                 queue = true;
6104         }
6105         spin_unlock(&nn->blocked_locks_lock);
6106
6107         if (queue)
6108                 nfsd4_run_cb(&nbl->nbl_cb);
6109 }
6110
6111 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
6112         .lm_notify = nfsd4_lm_notify,
6113         .lm_get_owner = nfsd4_fl_get_owner,
6114         .lm_put_owner = nfsd4_fl_put_owner,
6115 };
6116
6117 static inline void
6118 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6119 {
6120         struct nfs4_lockowner *lo;
6121
6122         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6123                 lo = (struct nfs4_lockowner *) fl->fl_owner;
6124                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6125                                                 GFP_KERNEL);
6126                 if (!deny->ld_owner.data)
6127                         /* We just don't care that much */
6128                         goto nevermind;
6129                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6130         } else {
6131 nevermind:
6132                 deny->ld_owner.len = 0;
6133                 deny->ld_owner.data = NULL;
6134                 deny->ld_clientid.cl_boot = 0;
6135                 deny->ld_clientid.cl_id = 0;
6136         }
6137         deny->ld_start = fl->fl_start;
6138         deny->ld_length = NFS4_MAX_UINT64;
6139         if (fl->fl_end != NFS4_MAX_UINT64)
6140                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6141         deny->ld_type = NFS4_READ_LT;
6142         if (fl->fl_type != F_RDLCK)
6143                 deny->ld_type = NFS4_WRITE_LT;
6144 }
6145
6146 static struct nfs4_lockowner *
6147 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6148 {
6149         unsigned int strhashval = ownerstr_hashval(owner);
6150         struct nfs4_stateowner *so;
6151
6152         lockdep_assert_held(&clp->cl_lock);
6153
6154         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6155                             so_strhash) {
6156                 if (so->so_is_open_owner)
6157                         continue;
6158                 if (same_owner_str(so, owner))
6159                         return lockowner(nfs4_get_stateowner(so));
6160         }
6161         return NULL;
6162 }
6163
6164 static struct nfs4_lockowner *
6165 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6166 {
6167         struct nfs4_lockowner *lo;
6168
6169         spin_lock(&clp->cl_lock);
6170         lo = find_lockowner_str_locked(clp, owner);
6171         spin_unlock(&clp->cl_lock);
6172         return lo;
6173 }
6174
6175 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6176 {
6177         unhash_lockowner_locked(lockowner(sop));
6178 }
6179
6180 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6181 {
6182         struct nfs4_lockowner *lo = lockowner(sop);
6183
6184         kmem_cache_free(lockowner_slab, lo);
6185 }
6186
6187 static const struct nfs4_stateowner_operations lockowner_ops = {
6188         .so_unhash =    nfs4_unhash_lockowner,
6189         .so_free =      nfs4_free_lockowner,
6190 };
6191
6192 /*
6193  * Alloc a lock owner structure.
6194  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6195  * occurred.
6196  *
6197  * strhashval = ownerstr_hashval
6198  */
6199 static struct nfs4_lockowner *
6200 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6201                            struct nfs4_ol_stateid *open_stp,
6202                            struct nfsd4_lock *lock)
6203 {
6204         struct nfs4_lockowner *lo, *ret;
6205
6206         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6207         if (!lo)
6208                 return NULL;
6209         INIT_LIST_HEAD(&lo->lo_blocked);
6210         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6211         lo->lo_owner.so_is_open_owner = 0;
6212         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6213         lo->lo_owner.so_ops = &lockowner_ops;
6214         spin_lock(&clp->cl_lock);
6215         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6216         if (ret == NULL) {
6217                 list_add(&lo->lo_owner.so_strhash,
6218                          &clp->cl_ownerstr_hashtbl[strhashval]);
6219                 ret = lo;
6220         } else
6221                 nfs4_free_stateowner(&lo->lo_owner);
6222
6223         spin_unlock(&clp->cl_lock);
6224         return ret;
6225 }
6226
6227 static struct nfs4_ol_stateid *
6228 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
6229 {
6230         struct nfs4_ol_stateid *lst;
6231         struct nfs4_client *clp = lo->lo_owner.so_client;
6232
6233         lockdep_assert_held(&clp->cl_lock);
6234
6235         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
6236                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
6237                         continue;
6238                 if (lst->st_stid.sc_file == fp) {
6239                         refcount_inc(&lst->st_stid.sc_count);
6240                         return lst;
6241                 }
6242         }
6243         return NULL;
6244 }
6245
6246 static struct nfs4_ol_stateid *
6247 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6248                   struct nfs4_file *fp, struct inode *inode,
6249                   struct nfs4_ol_stateid *open_stp)
6250 {
6251         struct nfs4_client *clp = lo->lo_owner.so_client;
6252         struct nfs4_ol_stateid *retstp;
6253
6254         mutex_init(&stp->st_mutex);
6255         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6256 retry:
6257         spin_lock(&clp->cl_lock);
6258         spin_lock(&fp->fi_lock);
6259         retstp = find_lock_stateid(lo, fp);
6260         if (retstp)
6261                 goto out_unlock;
6262
6263         refcount_inc(&stp->st_stid.sc_count);
6264         stp->st_stid.sc_type = NFS4_LOCK_STID;
6265         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6266         get_nfs4_file(fp);
6267         stp->st_stid.sc_file = fp;
6268         stp->st_access_bmap = 0;
6269         stp->st_deny_bmap = open_stp->st_deny_bmap;
6270         stp->st_openstp = open_stp;
6271         list_add(&stp->st_locks, &open_stp->st_locks);
6272         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6273         list_add(&stp->st_perfile, &fp->fi_stateids);
6274 out_unlock:
6275         spin_unlock(&fp->fi_lock);
6276         spin_unlock(&clp->cl_lock);
6277         if (retstp) {
6278                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6279                         nfs4_put_stid(&retstp->st_stid);
6280                         goto retry;
6281                 }
6282                 /* To keep mutex tracking happy */
6283                 mutex_unlock(&stp->st_mutex);
6284                 stp = retstp;
6285         }
6286         return stp;
6287 }
6288
6289 static struct nfs4_ol_stateid *
6290 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6291                             struct inode *inode, struct nfs4_ol_stateid *ost,
6292                             bool *new)
6293 {
6294         struct nfs4_stid *ns = NULL;
6295         struct nfs4_ol_stateid *lst;
6296         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6297         struct nfs4_client *clp = oo->oo_owner.so_client;
6298
6299         *new = false;
6300         spin_lock(&clp->cl_lock);
6301         lst = find_lock_stateid(lo, fi);
6302         spin_unlock(&clp->cl_lock);
6303         if (lst != NULL) {
6304                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6305                         goto out;
6306                 nfs4_put_stid(&lst->st_stid);
6307         }
6308         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6309         if (ns == NULL)
6310                 return NULL;
6311
6312         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6313         if (lst == openlockstateid(ns))
6314                 *new = true;
6315         else
6316                 nfs4_put_stid(ns);
6317 out:
6318         return lst;
6319 }
6320
6321 static int
6322 check_lock_length(u64 offset, u64 length)
6323 {
6324         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6325                 (length > ~offset)));
6326 }
6327
6328 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6329 {
6330         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6331
6332         lockdep_assert_held(&fp->fi_lock);
6333
6334         if (test_access(access, lock_stp))
6335                 return;
6336         __nfs4_file_get_access(fp, access);
6337         set_access(access, lock_stp);
6338 }
6339
6340 static __be32
6341 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6342                             struct nfs4_ol_stateid *ost,
6343                             struct nfsd4_lock *lock,
6344                             struct nfs4_ol_stateid **plst, bool *new)
6345 {
6346         __be32 status;
6347         struct nfs4_file *fi = ost->st_stid.sc_file;
6348         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6349         struct nfs4_client *cl = oo->oo_owner.so_client;
6350         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6351         struct nfs4_lockowner *lo;
6352         struct nfs4_ol_stateid *lst;
6353         unsigned int strhashval;
6354
6355         lo = find_lockowner_str(cl, &lock->lk_new_owner);
6356         if (!lo) {
6357                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6358                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6359                 if (lo == NULL)
6360                         return nfserr_jukebox;
6361         } else {
6362                 /* with an existing lockowner, seqids must be the same */
6363                 status = nfserr_bad_seqid;
6364                 if (!cstate->minorversion &&
6365                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6366                         goto out;
6367         }
6368
6369         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6370         if (lst == NULL) {
6371                 status = nfserr_jukebox;
6372                 goto out;
6373         }
6374
6375         status = nfs_ok;
6376         *plst = lst;
6377 out:
6378         nfs4_put_stateowner(&lo->lo_owner);
6379         return status;
6380 }
6381
6382 /*
6383  *  LOCK operation
6384  */
6385 __be32
6386 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6387            union nfsd4_op_u *u)
6388 {
6389         struct nfsd4_lock *lock = &u->lock;
6390         struct nfs4_openowner *open_sop = NULL;
6391         struct nfs4_lockowner *lock_sop = NULL;
6392         struct nfs4_ol_stateid *lock_stp = NULL;
6393         struct nfs4_ol_stateid *open_stp = NULL;
6394         struct nfs4_file *fp;
6395         struct file *filp = NULL;
6396         struct nfsd4_blocked_lock *nbl = NULL;
6397         struct file_lock *file_lock = NULL;
6398         struct file_lock *conflock = NULL;
6399         __be32 status = 0;
6400         int lkflg;
6401         int err;
6402         bool new = false;
6403         unsigned char fl_type;
6404         unsigned int fl_flags = FL_POSIX;
6405         struct net *net = SVC_NET(rqstp);
6406         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6407
6408         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6409                 (long long) lock->lk_offset,
6410                 (long long) lock->lk_length);
6411
6412         if (check_lock_length(lock->lk_offset, lock->lk_length))
6413                  return nfserr_inval;
6414
6415         if ((status = fh_verify(rqstp, &cstate->current_fh,
6416                                 S_IFREG, NFSD_MAY_LOCK))) {
6417                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6418                 return status;
6419         }
6420
6421         if (lock->lk_is_new) {
6422                 if (nfsd4_has_session(cstate))
6423                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6424                         memcpy(&lock->lk_new_clientid,
6425                                 &cstate->session->se_client->cl_clientid,
6426                                 sizeof(clientid_t));
6427
6428                 status = nfserr_stale_clientid;
6429                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6430                         goto out;
6431
6432                 /* validate and update open stateid and open seqid */
6433                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6434                                         lock->lk_new_open_seqid,
6435                                         &lock->lk_new_open_stateid,
6436                                         &open_stp, nn);
6437                 if (status)
6438                         goto out;
6439                 mutex_unlock(&open_stp->st_mutex);
6440                 open_sop = openowner(open_stp->st_stateowner);
6441                 status = nfserr_bad_stateid;
6442                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6443                                                 &lock->lk_new_clientid))
6444                         goto out;
6445                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6446                                                         &lock_stp, &new);
6447         } else {
6448                 status = nfs4_preprocess_seqid_op(cstate,
6449                                        lock->lk_old_lock_seqid,
6450                                        &lock->lk_old_lock_stateid,
6451                                        NFS4_LOCK_STID, &lock_stp, nn);
6452         }
6453         if (status)
6454                 goto out;
6455         lock_sop = lockowner(lock_stp->st_stateowner);
6456
6457         lkflg = setlkflg(lock->lk_type);
6458         status = nfs4_check_openmode(lock_stp, lkflg);
6459         if (status)
6460                 goto out;
6461
6462         status = nfserr_grace;
6463         if (locks_in_grace(net) && !lock->lk_reclaim)
6464                 goto out;
6465         status = nfserr_no_grace;
6466         if (!locks_in_grace(net) && lock->lk_reclaim)
6467                 goto out;
6468
6469         fp = lock_stp->st_stid.sc_file;
6470         switch (lock->lk_type) {
6471                 case NFS4_READW_LT:
6472                         if (nfsd4_has_session(cstate))
6473                                 fl_flags |= FL_SLEEP;
6474                         /* Fallthrough */
6475                 case NFS4_READ_LT:
6476                         spin_lock(&fp->fi_lock);
6477                         filp = find_readable_file_locked(fp);
6478                         if (filp)
6479                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6480                         spin_unlock(&fp->fi_lock);
6481                         fl_type = F_RDLCK;
6482                         break;
6483                 case NFS4_WRITEW_LT:
6484                         if (nfsd4_has_session(cstate))
6485                                 fl_flags |= FL_SLEEP;
6486                         /* Fallthrough */
6487                 case NFS4_WRITE_LT:
6488                         spin_lock(&fp->fi_lock);
6489                         filp = find_writeable_file_locked(fp);
6490                         if (filp)
6491                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6492                         spin_unlock(&fp->fi_lock);
6493                         fl_type = F_WRLCK;
6494                         break;
6495                 default:
6496                         status = nfserr_inval;
6497                 goto out;
6498         }
6499
6500         if (!filp) {
6501                 status = nfserr_openmode;
6502                 goto out;
6503         }
6504
6505         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6506         if (!nbl) {
6507                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6508                 status = nfserr_jukebox;
6509                 goto out;
6510         }
6511
6512         file_lock = &nbl->nbl_lock;
6513         file_lock->fl_type = fl_type;
6514         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6515         file_lock->fl_pid = current->tgid;
6516         file_lock->fl_file = filp;
6517         file_lock->fl_flags = fl_flags;
6518         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6519         file_lock->fl_start = lock->lk_offset;
6520         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6521         nfs4_transform_lock_offset(file_lock);
6522
6523         conflock = locks_alloc_lock();
6524         if (!conflock) {
6525                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6526                 status = nfserr_jukebox;
6527                 goto out;
6528         }
6529
6530         if (fl_flags & FL_SLEEP) {
6531                 nbl->nbl_time = jiffies;
6532                 spin_lock(&nn->blocked_locks_lock);
6533                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6534                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6535                 spin_unlock(&nn->blocked_locks_lock);
6536         }
6537
6538         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6539         switch (err) {
6540         case 0: /* success! */
6541                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6542                 status = 0;
6543                 if (lock->lk_reclaim)
6544                         nn->somebody_reclaimed = true;
6545                 break;
6546         case FILE_LOCK_DEFERRED:
6547                 nbl = NULL;
6548                 /* Fallthrough */
6549         case -EAGAIN:           /* conflock holds conflicting lock */
6550                 status = nfserr_denied;
6551                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6552                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6553                 break;
6554         case -EDEADLK:
6555                 status = nfserr_deadlock;
6556                 break;
6557         default:
6558                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6559                 status = nfserrno(err);
6560                 break;
6561         }
6562 out:
6563         if (nbl) {
6564                 /* dequeue it if we queued it before */
6565                 if (fl_flags & FL_SLEEP) {
6566                         spin_lock(&nn->blocked_locks_lock);
6567                         list_del_init(&nbl->nbl_list);
6568                         list_del_init(&nbl->nbl_lru);
6569                         spin_unlock(&nn->blocked_locks_lock);
6570                 }
6571                 free_blocked_lock(nbl);
6572         }
6573         if (filp)
6574                 fput(filp);
6575         if (lock_stp) {
6576                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6577                 if (cstate->replay_owner &&
6578                     cstate->replay_owner != &lock_sop->lo_owner &&
6579                     seqid_mutating_err(ntohl(status)))
6580                         lock_sop->lo_owner.so_seqid++;
6581
6582                 /*
6583                  * If this is a new, never-before-used stateid, and we are
6584                  * returning an error, then just go ahead and release it.
6585                  */
6586                 if (status && new)
6587                         release_lock_stateid(lock_stp);
6588
6589                 mutex_unlock(&lock_stp->st_mutex);
6590
6591                 nfs4_put_stid(&lock_stp->st_stid);
6592         }
6593         if (open_stp)
6594                 nfs4_put_stid(&open_stp->st_stid);
6595         nfsd4_bump_seqid(cstate, status);
6596         if (conflock)
6597                 locks_free_lock(conflock);
6598         return status;
6599 }
6600
6601 /*
6602  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6603  * so we do a temporary open here just to get an open file to pass to
6604  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6605  * inode operation.)
6606  */
6607 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6608 {
6609         struct file *file;
6610         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6611         if (!err) {
6612                 err = nfserrno(vfs_test_lock(file, lock));
6613                 fput(file);
6614         }
6615         return err;
6616 }
6617
6618 /*
6619  * LOCKT operation
6620  */
6621 __be32
6622 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6623             union nfsd4_op_u *u)
6624 {
6625         struct nfsd4_lockt *lockt = &u->lockt;
6626         struct file_lock *file_lock = NULL;
6627         struct nfs4_lockowner *lo = NULL;
6628         __be32 status;
6629         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6630
6631         if (locks_in_grace(SVC_NET(rqstp)))
6632                 return nfserr_grace;
6633
6634         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6635                  return nfserr_inval;
6636
6637         if (!nfsd4_has_session(cstate)) {
6638                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6639                 if (status)
6640                         goto out;
6641         }
6642
6643         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6644                 goto out;
6645
6646         file_lock = locks_alloc_lock();
6647         if (!file_lock) {
6648                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6649                 status = nfserr_jukebox;
6650                 goto out;
6651         }
6652
6653         switch (lockt->lt_type) {
6654                 case NFS4_READ_LT:
6655                 case NFS4_READW_LT:
6656                         file_lock->fl_type = F_RDLCK;
6657                         break;
6658                 case NFS4_WRITE_LT:
6659                 case NFS4_WRITEW_LT:
6660                         file_lock->fl_type = F_WRLCK;
6661                         break;
6662                 default:
6663                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6664                         status = nfserr_inval;
6665                         goto out;
6666         }
6667
6668         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6669         if (lo)
6670                 file_lock->fl_owner = (fl_owner_t)lo;
6671         file_lock->fl_pid = current->tgid;
6672         file_lock->fl_flags = FL_POSIX;
6673
6674         file_lock->fl_start = lockt->lt_offset;
6675         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6676
6677         nfs4_transform_lock_offset(file_lock);
6678
6679         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6680         if (status)
6681                 goto out;
6682
6683         if (file_lock->fl_type != F_UNLCK) {
6684                 status = nfserr_denied;
6685                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6686         }
6687 out:
6688         if (lo)
6689                 nfs4_put_stateowner(&lo->lo_owner);
6690         if (file_lock)
6691                 locks_free_lock(file_lock);
6692         return status;
6693 }
6694
6695 __be32
6696 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6697             union nfsd4_op_u *u)
6698 {
6699         struct nfsd4_locku *locku = &u->locku;
6700         struct nfs4_ol_stateid *stp;
6701         struct file *filp = NULL;
6702         struct file_lock *file_lock = NULL;
6703         __be32 status;
6704         int err;
6705         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6706
6707         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6708                 (long long) locku->lu_offset,
6709                 (long long) locku->lu_length);
6710
6711         if (check_lock_length(locku->lu_offset, locku->lu_length))
6712                  return nfserr_inval;
6713
6714         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6715                                         &locku->lu_stateid, NFS4_LOCK_STID,
6716                                         &stp, nn);
6717         if (status)
6718                 goto out;
6719         filp = find_any_file(stp->st_stid.sc_file);
6720         if (!filp) {
6721                 status = nfserr_lock_range;
6722                 goto put_stateid;
6723         }
6724         file_lock = locks_alloc_lock();
6725         if (!file_lock) {
6726                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6727                 status = nfserr_jukebox;
6728                 goto fput;
6729         }
6730
6731         file_lock->fl_type = F_UNLCK;
6732         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6733         file_lock->fl_pid = current->tgid;
6734         file_lock->fl_file = filp;
6735         file_lock->fl_flags = FL_POSIX;
6736         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6737         file_lock->fl_start = locku->lu_offset;
6738
6739         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6740                                                 locku->lu_length);
6741         nfs4_transform_lock_offset(file_lock);
6742
6743         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6744         if (err) {
6745                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6746                 goto out_nfserr;
6747         }
6748         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6749 fput:
6750         fput(filp);
6751 put_stateid:
6752         mutex_unlock(&stp->st_mutex);
6753         nfs4_put_stid(&stp->st_stid);
6754 out:
6755         nfsd4_bump_seqid(cstate, status);
6756         if (file_lock)
6757                 locks_free_lock(file_lock);
6758         return status;
6759
6760 out_nfserr:
6761         status = nfserrno(err);
6762         goto fput;
6763 }
6764
6765 /*
6766  * returns
6767  *      true:  locks held by lockowner
6768  *      false: no locks held by lockowner
6769  */
6770 static bool
6771 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6772 {
6773         struct file_lock *fl;
6774         int status = false;
6775         struct file *filp = find_any_file(fp);
6776         struct inode *inode;
6777         struct file_lock_context *flctx;
6778
6779         if (!filp) {
6780                 /* Any valid lock stateid should have some sort of access */
6781                 WARN_ON_ONCE(1);
6782                 return status;
6783         }
6784
6785         inode = locks_inode(filp);
6786         flctx = inode->i_flctx;
6787
6788         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6789                 spin_lock(&flctx->flc_lock);
6790                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6791                         if (fl->fl_owner == (fl_owner_t)lowner) {
6792                                 status = true;
6793                                 break;
6794                         }
6795                 }
6796                 spin_unlock(&flctx->flc_lock);
6797         }
6798         fput(filp);
6799         return status;
6800 }
6801
6802 __be32
6803 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6804                         struct nfsd4_compound_state *cstate,
6805                         union nfsd4_op_u *u)
6806 {
6807         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6808         clientid_t *clid = &rlockowner->rl_clientid;
6809         struct nfs4_stateowner *sop;
6810         struct nfs4_lockowner *lo = NULL;
6811         struct nfs4_ol_stateid *stp;
6812         struct xdr_netobj *owner = &rlockowner->rl_owner;
6813         unsigned int hashval = ownerstr_hashval(owner);
6814         __be32 status;
6815         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6816         struct nfs4_client *clp;
6817         LIST_HEAD (reaplist);
6818
6819         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6820                 clid->cl_boot, clid->cl_id);
6821
6822         status = lookup_clientid(clid, cstate, nn);
6823         if (status)
6824                 return status;
6825
6826         clp = cstate->clp;
6827         /* Find the matching lock stateowner */
6828         spin_lock(&clp->cl_lock);
6829         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6830                             so_strhash) {
6831
6832                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6833                         continue;
6834
6835                 /* see if there are still any locks associated with it */
6836                 lo = lockowner(sop);
6837                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6838                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6839                                 status = nfserr_locks_held;
6840                                 spin_unlock(&clp->cl_lock);
6841                                 return status;
6842                         }
6843                 }
6844
6845                 nfs4_get_stateowner(sop);
6846                 break;
6847         }
6848         if (!lo) {
6849                 spin_unlock(&clp->cl_lock);
6850                 return status;
6851         }
6852
6853         unhash_lockowner_locked(lo);
6854         while (!list_empty(&lo->lo_owner.so_stateids)) {
6855                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6856                                        struct nfs4_ol_stateid,
6857                                        st_perstateowner);
6858                 WARN_ON(!unhash_lock_stateid(stp));
6859                 put_ol_stateid_locked(stp, &reaplist);
6860         }
6861         spin_unlock(&clp->cl_lock);
6862         free_ol_stateid_reaplist(&reaplist);
6863         remove_blocked_locks(lo);
6864         nfs4_put_stateowner(&lo->lo_owner);
6865
6866         return status;
6867 }
6868
6869 static inline struct nfs4_client_reclaim *
6870 alloc_reclaim(void)
6871 {
6872         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6873 }
6874
6875 bool
6876 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
6877 {
6878         struct nfs4_client_reclaim *crp;
6879
6880         crp = nfsd4_find_reclaim_client(name, nn);
6881         return (crp && crp->cr_clp);
6882 }
6883
6884 /*
6885  * failure => all reset bets are off, nfserr_no_grace...
6886  *
6887  * The caller is responsible for freeing name.data if NULL is returned (it
6888  * will be freed in nfs4_remove_reclaim_record in the normal case).
6889  */
6890 struct nfs4_client_reclaim *
6891 nfs4_client_to_reclaim(struct xdr_netobj name, struct nfsd_net *nn)
6892 {
6893         unsigned int strhashval;
6894         struct nfs4_client_reclaim *crp;
6895
6896         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
6897         crp = alloc_reclaim();
6898         if (crp) {
6899                 strhashval = clientstr_hashval(name);
6900                 INIT_LIST_HEAD(&crp->cr_strhash);
6901                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6902                 crp->cr_name.data = name.data;
6903                 crp->cr_name.len = name.len;
6904                 crp->cr_clp = NULL;
6905                 nn->reclaim_str_hashtbl_size++;
6906         }
6907         return crp;
6908 }
6909
6910 void
6911 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6912 {
6913         list_del(&crp->cr_strhash);
6914         kfree(crp->cr_name.data);
6915         kfree(crp);
6916         nn->reclaim_str_hashtbl_size--;
6917 }
6918
6919 void
6920 nfs4_release_reclaim(struct nfsd_net *nn)
6921 {
6922         struct nfs4_client_reclaim *crp = NULL;
6923         int i;
6924
6925         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6926                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6927                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6928                                         struct nfs4_client_reclaim, cr_strhash);
6929                         nfs4_remove_reclaim_record(crp, nn);
6930                 }
6931         }
6932         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6933 }
6934
6935 /*
6936  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6937 struct nfs4_client_reclaim *
6938 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
6939 {
6940         unsigned int strhashval;
6941         struct nfs4_client_reclaim *crp = NULL;
6942
6943         dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
6944
6945         strhashval = clientstr_hashval(name);
6946         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6947                 if (compare_blob(&crp->cr_name, &name) == 0) {
6948                         return crp;
6949                 }
6950         }
6951         return NULL;
6952 }
6953
6954 /*
6955 * Called from OPEN. Look for clientid in reclaim list.
6956 */
6957 __be32
6958 nfs4_check_open_reclaim(clientid_t *clid,
6959                 struct nfsd4_compound_state *cstate,
6960                 struct nfsd_net *nn)
6961 {
6962         __be32 status;
6963
6964         /* find clientid in conf_id_hashtbl */
6965         status = lookup_clientid(clid, cstate, nn);
6966         if (status)
6967                 return nfserr_reclaim_bad;
6968
6969         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6970                 return nfserr_no_grace;
6971
6972         if (nfsd4_client_record_check(cstate->clp))
6973                 return nfserr_reclaim_bad;
6974
6975         return nfs_ok;
6976 }
6977
6978 #ifdef CONFIG_NFSD_FAULT_INJECTION
6979 static inline void
6980 put_client(struct nfs4_client *clp)
6981 {
6982         atomic_dec(&clp->cl_rpc_users);
6983 }
6984
6985 static struct nfs4_client *
6986 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6987 {
6988         struct nfs4_client *clp;
6989         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6990                                           nfsd_net_id);
6991
6992         if (!nfsd_netns_ready(nn))
6993                 return NULL;
6994
6995         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6996                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6997                         return clp;
6998         }
6999         return NULL;
7000 }
7001
7002 u64
7003 nfsd_inject_print_clients(void)
7004 {
7005         struct nfs4_client *clp;
7006         u64 count = 0;
7007         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7008                                           nfsd_net_id);
7009         char buf[INET6_ADDRSTRLEN];
7010
7011         if (!nfsd_netns_ready(nn))
7012                 return 0;
7013
7014         spin_lock(&nn->client_lock);
7015         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7016                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7017                 pr_info("NFS Client: %s\n", buf);
7018                 ++count;
7019         }
7020         spin_unlock(&nn->client_lock);
7021
7022         return count;
7023 }
7024
7025 u64
7026 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7027 {
7028         u64 count = 0;
7029         struct nfs4_client *clp;
7030         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7031                                           nfsd_net_id);
7032
7033         if (!nfsd_netns_ready(nn))
7034                 return count;
7035
7036         spin_lock(&nn->client_lock);
7037         clp = nfsd_find_client(addr, addr_size);
7038         if (clp) {
7039                 if (mark_client_expired_locked(clp) == nfs_ok)
7040                         ++count;
7041                 else
7042                         clp = NULL;
7043         }
7044         spin_unlock(&nn->client_lock);
7045
7046         if (clp)
7047                 expire_client(clp);
7048
7049         return count;
7050 }
7051
7052 u64
7053 nfsd_inject_forget_clients(u64 max)
7054 {
7055         u64 count = 0;
7056         struct nfs4_client *clp, *next;
7057         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7058                                                 nfsd_net_id);
7059         LIST_HEAD(reaplist);
7060
7061         if (!nfsd_netns_ready(nn))
7062                 return count;
7063
7064         spin_lock(&nn->client_lock);
7065         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7066                 if (mark_client_expired_locked(clp) == nfs_ok) {
7067                         list_add(&clp->cl_lru, &reaplist);
7068                         if (max != 0 && ++count >= max)
7069                                 break;
7070                 }
7071         }
7072         spin_unlock(&nn->client_lock);
7073
7074         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7075                 expire_client(clp);
7076
7077         return count;
7078 }
7079
7080 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7081                              const char *type)
7082 {
7083         char buf[INET6_ADDRSTRLEN];
7084         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7085         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7086 }
7087
7088 static void
7089 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7090                              struct list_head *collect)
7091 {
7092         struct nfs4_client *clp = lst->st_stid.sc_client;
7093         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7094                                           nfsd_net_id);
7095
7096         if (!collect)
7097                 return;
7098
7099         lockdep_assert_held(&nn->client_lock);
7100         atomic_inc(&clp->cl_rpc_users);
7101         list_add(&lst->st_locks, collect);
7102 }
7103
7104 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7105                                     struct list_head *collect,
7106                                     bool (*func)(struct nfs4_ol_stateid *))
7107 {
7108         struct nfs4_openowner *oop;
7109         struct nfs4_ol_stateid *stp, *st_next;
7110         struct nfs4_ol_stateid *lst, *lst_next;
7111         u64 count = 0;
7112
7113         spin_lock(&clp->cl_lock);
7114         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7115                 list_for_each_entry_safe(stp, st_next,
7116                                 &oop->oo_owner.so_stateids, st_perstateowner) {
7117                         list_for_each_entry_safe(lst, lst_next,
7118                                         &stp->st_locks, st_locks) {
7119                                 if (func) {
7120                                         if (func(lst))
7121                                                 nfsd_inject_add_lock_to_list(lst,
7122                                                                         collect);
7123                                 }
7124                                 ++count;
7125                                 /*
7126                                  * Despite the fact that these functions deal
7127                                  * with 64-bit integers for "count", we must
7128                                  * ensure that it doesn't blow up the
7129                                  * clp->cl_rpc_users. Throw a warning if we
7130                                  * start to approach INT_MAX here.
7131                                  */
7132                                 WARN_ON_ONCE(count == (INT_MAX / 2));
7133                                 if (count == max)
7134                                         goto out;
7135                         }
7136                 }
7137         }
7138 out:
7139         spin_unlock(&clp->cl_lock);
7140
7141         return count;
7142 }
7143
7144 static u64
7145 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7146                           u64 max)
7147 {
7148         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7149 }
7150
7151 static u64
7152 nfsd_print_client_locks(struct nfs4_client *clp)
7153 {
7154         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7155         nfsd_print_count(clp, count, "locked files");
7156         return count;
7157 }
7158
7159 u64
7160 nfsd_inject_print_locks(void)
7161 {
7162         struct nfs4_client *clp;
7163         u64 count = 0;
7164         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7165                                                 nfsd_net_id);
7166
7167         if (!nfsd_netns_ready(nn))
7168                 return 0;
7169
7170         spin_lock(&nn->client_lock);
7171         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7172                 count += nfsd_print_client_locks(clp);
7173         spin_unlock(&nn->client_lock);
7174
7175         return count;
7176 }
7177
7178 static void
7179 nfsd_reap_locks(struct list_head *reaplist)
7180 {
7181         struct nfs4_client *clp;
7182         struct nfs4_ol_stateid *stp, *next;
7183
7184         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7185                 list_del_init(&stp->st_locks);
7186                 clp = stp->st_stid.sc_client;
7187                 nfs4_put_stid(&stp->st_stid);
7188                 put_client(clp);
7189         }
7190 }
7191
7192 u64
7193 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7194 {
7195         unsigned int count = 0;
7196         struct nfs4_client *clp;
7197         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7198                                                 nfsd_net_id);
7199         LIST_HEAD(reaplist);
7200
7201         if (!nfsd_netns_ready(nn))
7202                 return count;
7203
7204         spin_lock(&nn->client_lock);
7205         clp = nfsd_find_client(addr, addr_size);
7206         if (clp)
7207                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7208         spin_unlock(&nn->client_lock);
7209         nfsd_reap_locks(&reaplist);
7210         return count;
7211 }
7212
7213 u64
7214 nfsd_inject_forget_locks(u64 max)
7215 {
7216         u64 count = 0;
7217         struct nfs4_client *clp;
7218         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7219                                                 nfsd_net_id);
7220         LIST_HEAD(reaplist);
7221
7222         if (!nfsd_netns_ready(nn))
7223                 return count;
7224
7225         spin_lock(&nn->client_lock);
7226         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7227                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7228                 if (max != 0 && count >= max)
7229                         break;
7230         }
7231         spin_unlock(&nn->client_lock);
7232         nfsd_reap_locks(&reaplist);
7233         return count;
7234 }
7235
7236 static u64
7237 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7238                               struct list_head *collect,
7239                               void (*func)(struct nfs4_openowner *))
7240 {
7241         struct nfs4_openowner *oop, *next;
7242         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7243                                                 nfsd_net_id);
7244         u64 count = 0;
7245
7246         lockdep_assert_held(&nn->client_lock);
7247
7248         spin_lock(&clp->cl_lock);
7249         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7250                 if (func) {
7251                         func(oop);
7252                         if (collect) {
7253                                 atomic_inc(&clp->cl_rpc_users);
7254                                 list_add(&oop->oo_perclient, collect);
7255                         }
7256                 }
7257                 ++count;
7258                 /*
7259                  * Despite the fact that these functions deal with
7260                  * 64-bit integers for "count", we must ensure that
7261                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7262                  * warning if we start to approach INT_MAX here.
7263                  */
7264                 WARN_ON_ONCE(count == (INT_MAX / 2));
7265                 if (count == max)
7266                         break;
7267         }
7268         spin_unlock(&clp->cl_lock);
7269
7270         return count;
7271 }
7272
7273 static u64
7274 nfsd_print_client_openowners(struct nfs4_client *clp)
7275 {
7276         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7277
7278         nfsd_print_count(clp, count, "openowners");
7279         return count;
7280 }
7281
7282 static u64
7283 nfsd_collect_client_openowners(struct nfs4_client *clp,
7284                                struct list_head *collect, u64 max)
7285 {
7286         return nfsd_foreach_client_openowner(clp, max, collect,
7287                                                 unhash_openowner_locked);
7288 }
7289
7290 u64
7291 nfsd_inject_print_openowners(void)
7292 {
7293         struct nfs4_client *clp;
7294         u64 count = 0;
7295         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7296                                                 nfsd_net_id);
7297
7298         if (!nfsd_netns_ready(nn))
7299                 return 0;
7300
7301         spin_lock(&nn->client_lock);
7302         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7303                 count += nfsd_print_client_openowners(clp);
7304         spin_unlock(&nn->client_lock);
7305
7306         return count;
7307 }
7308
7309 static void
7310 nfsd_reap_openowners(struct list_head *reaplist)
7311 {
7312         struct nfs4_client *clp;
7313         struct nfs4_openowner *oop, *next;
7314
7315         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7316                 list_del_init(&oop->oo_perclient);
7317                 clp = oop->oo_owner.so_client;
7318                 release_openowner(oop);
7319                 put_client(clp);
7320         }
7321 }
7322
7323 u64
7324 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7325                                      size_t addr_size)
7326 {
7327         unsigned int count = 0;
7328         struct nfs4_client *clp;
7329         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7330                                                 nfsd_net_id);
7331         LIST_HEAD(reaplist);
7332
7333         if (!nfsd_netns_ready(nn))
7334                 return count;
7335
7336         spin_lock(&nn->client_lock);
7337         clp = nfsd_find_client(addr, addr_size);
7338         if (clp)
7339                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7340         spin_unlock(&nn->client_lock);
7341         nfsd_reap_openowners(&reaplist);
7342         return count;
7343 }
7344
7345 u64
7346 nfsd_inject_forget_openowners(u64 max)
7347 {
7348         u64 count = 0;
7349         struct nfs4_client *clp;
7350         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7351                                                 nfsd_net_id);
7352         LIST_HEAD(reaplist);
7353
7354         if (!nfsd_netns_ready(nn))
7355                 return count;
7356
7357         spin_lock(&nn->client_lock);
7358         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7359                 count += nfsd_collect_client_openowners(clp, &reaplist,
7360                                                         max - count);
7361                 if (max != 0 && count >= max)
7362                         break;
7363         }
7364         spin_unlock(&nn->client_lock);
7365         nfsd_reap_openowners(&reaplist);
7366         return count;
7367 }
7368
7369 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7370                                      struct list_head *victims)
7371 {
7372         struct nfs4_delegation *dp, *next;
7373         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7374                                                 nfsd_net_id);
7375         u64 count = 0;
7376
7377         lockdep_assert_held(&nn->client_lock);
7378
7379         spin_lock(&state_lock);
7380         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7381                 if (victims) {
7382                         /*
7383                          * It's not safe to mess with delegations that have a
7384                          * non-zero dl_time. They might have already been broken
7385                          * and could be processed by the laundromat outside of
7386                          * the state_lock. Just leave them be.
7387                          */
7388                         if (dp->dl_time != 0)
7389                                 continue;
7390
7391                         atomic_inc(&clp->cl_rpc_users);
7392                         WARN_ON(!unhash_delegation_locked(dp));
7393                         list_add(&dp->dl_recall_lru, victims);
7394                 }
7395                 ++count;
7396                 /*
7397                  * Despite the fact that these functions deal with
7398                  * 64-bit integers for "count", we must ensure that
7399                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7400                  * warning if we start to approach INT_MAX here.
7401                  */
7402                 WARN_ON_ONCE(count == (INT_MAX / 2));
7403                 if (count == max)
7404                         break;
7405         }
7406         spin_unlock(&state_lock);
7407         return count;
7408 }
7409
7410 static u64
7411 nfsd_print_client_delegations(struct nfs4_client *clp)
7412 {
7413         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7414
7415         nfsd_print_count(clp, count, "delegations");
7416         return count;
7417 }
7418
7419 u64
7420 nfsd_inject_print_delegations(void)
7421 {
7422         struct nfs4_client *clp;
7423         u64 count = 0;
7424         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7425                                                 nfsd_net_id);
7426
7427         if (!nfsd_netns_ready(nn))
7428                 return 0;
7429
7430         spin_lock(&nn->client_lock);
7431         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7432                 count += nfsd_print_client_delegations(clp);
7433         spin_unlock(&nn->client_lock);
7434
7435         return count;
7436 }
7437
7438 static void
7439 nfsd_forget_delegations(struct list_head *reaplist)
7440 {
7441         struct nfs4_client *clp;
7442         struct nfs4_delegation *dp, *next;
7443
7444         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7445                 list_del_init(&dp->dl_recall_lru);
7446                 clp = dp->dl_stid.sc_client;
7447                 revoke_delegation(dp);
7448                 put_client(clp);
7449         }
7450 }
7451
7452 u64
7453 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7454                                       size_t addr_size)
7455 {
7456         u64 count = 0;
7457         struct nfs4_client *clp;
7458         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7459                                                 nfsd_net_id);
7460         LIST_HEAD(reaplist);
7461
7462         if (!nfsd_netns_ready(nn))
7463                 return count;
7464
7465         spin_lock(&nn->client_lock);
7466         clp = nfsd_find_client(addr, addr_size);
7467         if (clp)
7468                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7469         spin_unlock(&nn->client_lock);
7470
7471         nfsd_forget_delegations(&reaplist);
7472         return count;
7473 }
7474
7475 u64
7476 nfsd_inject_forget_delegations(u64 max)
7477 {
7478         u64 count = 0;
7479         struct nfs4_client *clp;
7480         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7481                                                 nfsd_net_id);
7482         LIST_HEAD(reaplist);
7483
7484         if (!nfsd_netns_ready(nn))
7485                 return count;
7486
7487         spin_lock(&nn->client_lock);
7488         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7489                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7490                 if (max != 0 && count >= max)
7491                         break;
7492         }
7493         spin_unlock(&nn->client_lock);
7494         nfsd_forget_delegations(&reaplist);
7495         return count;
7496 }
7497
7498 static void
7499 nfsd_recall_delegations(struct list_head *reaplist)
7500 {
7501         struct nfs4_client *clp;
7502         struct nfs4_delegation *dp, *next;
7503
7504         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7505                 list_del_init(&dp->dl_recall_lru);
7506                 clp = dp->dl_stid.sc_client;
7507                 /*
7508                  * We skipped all entries that had a zero dl_time before,
7509                  * so we can now reset the dl_time back to 0. If a delegation
7510                  * break comes in now, then it won't make any difference since
7511                  * we're recalling it either way.
7512                  */
7513                 spin_lock(&state_lock);
7514                 dp->dl_time = 0;
7515                 spin_unlock(&state_lock);
7516                 nfsd_break_one_deleg(dp);
7517                 put_client(clp);
7518         }
7519 }
7520
7521 u64
7522 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7523                                       size_t addr_size)
7524 {
7525         u64 count = 0;
7526         struct nfs4_client *clp;
7527         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7528                                                 nfsd_net_id);
7529         LIST_HEAD(reaplist);
7530
7531         if (!nfsd_netns_ready(nn))
7532                 return count;
7533
7534         spin_lock(&nn->client_lock);
7535         clp = nfsd_find_client(addr, addr_size);
7536         if (clp)
7537                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7538         spin_unlock(&nn->client_lock);
7539
7540         nfsd_recall_delegations(&reaplist);
7541         return count;
7542 }
7543
7544 u64
7545 nfsd_inject_recall_delegations(u64 max)
7546 {
7547         u64 count = 0;
7548         struct nfs4_client *clp, *next;
7549         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7550                                                 nfsd_net_id);
7551         LIST_HEAD(reaplist);
7552
7553         if (!nfsd_netns_ready(nn))
7554                 return count;
7555
7556         spin_lock(&nn->client_lock);
7557         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7558                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7559                 if (max != 0 && ++count >= max)
7560                         break;
7561         }
7562         spin_unlock(&nn->client_lock);
7563         nfsd_recall_delegations(&reaplist);
7564         return count;
7565 }
7566 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7567
7568 /*
7569  * Since the lifetime of a delegation isn't limited to that of an open, a
7570  * client may quite reasonably hang on to a delegation as long as it has
7571  * the inode cached.  This becomes an obvious problem the first time a
7572  * client's inode cache approaches the size of the server's total memory.
7573  *
7574  * For now we avoid this problem by imposing a hard limit on the number
7575  * of delegations, which varies according to the server's memory size.
7576  */
7577 static void
7578 set_max_delegations(void)
7579 {
7580         /*
7581          * Allow at most 4 delegations per megabyte of RAM.  Quick
7582          * estimates suggest that in the worst case (where every delegation
7583          * is for a different inode), a delegation could take about 1.5K,
7584          * giving a worst case usage of about 6% of memory.
7585          */
7586         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7587 }
7588
7589 static int nfs4_state_create_net(struct net *net)
7590 {
7591         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7592         int i;
7593
7594         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7595                                             sizeof(struct list_head),
7596                                             GFP_KERNEL);
7597         if (!nn->conf_id_hashtbl)
7598                 goto err;
7599         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7600                                               sizeof(struct list_head),
7601                                               GFP_KERNEL);
7602         if (!nn->unconf_id_hashtbl)
7603                 goto err_unconf_id;
7604         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7605                                               sizeof(struct list_head),
7606                                               GFP_KERNEL);
7607         if (!nn->sessionid_hashtbl)
7608                 goto err_sessionid;
7609
7610         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7611                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7612                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7613         }
7614         for (i = 0; i < SESSION_HASH_SIZE; i++)
7615                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7616         nn->conf_name_tree = RB_ROOT;
7617         nn->unconf_name_tree = RB_ROOT;
7618         nn->boot_time = get_seconds();
7619         nn->grace_ended = false;
7620         nn->nfsd4_manager.block_opens = true;
7621         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7622         INIT_LIST_HEAD(&nn->client_lru);
7623         INIT_LIST_HEAD(&nn->close_lru);
7624         INIT_LIST_HEAD(&nn->del_recall_lru);
7625         spin_lock_init(&nn->client_lock);
7626         spin_lock_init(&nn->s2s_cp_lock);
7627         idr_init(&nn->s2s_cp_stateids);
7628
7629         spin_lock_init(&nn->blocked_locks_lock);
7630         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7631
7632         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7633         get_net(net);
7634
7635         return 0;
7636
7637 err_sessionid:
7638         kfree(nn->unconf_id_hashtbl);
7639 err_unconf_id:
7640         kfree(nn->conf_id_hashtbl);
7641 err:
7642         return -ENOMEM;
7643 }
7644
7645 static void
7646 nfs4_state_destroy_net(struct net *net)
7647 {
7648         int i;
7649         struct nfs4_client *clp = NULL;
7650         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7651
7652         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7653                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7654                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7655                         destroy_client(clp);
7656                 }
7657         }
7658
7659         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7660
7661         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7662                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7663                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7664                         destroy_client(clp);
7665                 }
7666         }
7667
7668         kfree(nn->sessionid_hashtbl);
7669         kfree(nn->unconf_id_hashtbl);
7670         kfree(nn->conf_id_hashtbl);
7671         put_net(net);
7672 }
7673
7674 int
7675 nfs4_state_start_net(struct net *net)
7676 {
7677         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7678         int ret;
7679
7680         ret = nfs4_state_create_net(net);
7681         if (ret)
7682                 return ret;
7683         locks_start_grace(net, &nn->nfsd4_manager);
7684         nfsd4_client_tracking_init(net);
7685         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7686                 goto skip_grace;
7687         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7688                nn->nfsd4_grace, net->ns.inum);
7689         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7690         return 0;
7691
7692 skip_grace:
7693         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7694                         net->ns.inum);
7695         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7696         nfsd4_end_grace(nn);
7697         return 0;
7698 }
7699
7700 /* initialization to perform when the nfsd service is started: */
7701
7702 int
7703 nfs4_state_start(void)
7704 {
7705         int ret;
7706
7707         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7708         if (laundry_wq == NULL) {
7709                 ret = -ENOMEM;
7710                 goto out;
7711         }
7712         ret = nfsd4_create_callback_queue();
7713         if (ret)
7714                 goto out_free_laundry;
7715
7716         set_max_delegations();
7717         return 0;
7718
7719 out_free_laundry:
7720         destroy_workqueue(laundry_wq);
7721 out:
7722         return ret;
7723 }
7724
7725 void
7726 nfs4_state_shutdown_net(struct net *net)
7727 {
7728         struct nfs4_delegation *dp = NULL;
7729         struct list_head *pos, *next, reaplist;
7730         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7731
7732         cancel_delayed_work_sync(&nn->laundromat_work);
7733         locks_end_grace(&nn->nfsd4_manager);
7734
7735         INIT_LIST_HEAD(&reaplist);
7736         spin_lock(&state_lock);
7737         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7738                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7739                 WARN_ON(!unhash_delegation_locked(dp));
7740                 list_add(&dp->dl_recall_lru, &reaplist);
7741         }
7742         spin_unlock(&state_lock);
7743         list_for_each_safe(pos, next, &reaplist) {
7744                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7745                 list_del_init(&dp->dl_recall_lru);
7746                 destroy_unhashed_deleg(dp);
7747         }
7748
7749         nfsd4_client_tracking_exit(net);
7750         nfs4_state_destroy_net(net);
7751 }
7752
7753 void
7754 nfs4_state_shutdown(void)
7755 {
7756         destroy_workqueue(laundry_wq);
7757         nfsd4_destroy_callback_queue();
7758 }
7759
7760 static void
7761 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7762 {
7763         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7764                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7765 }
7766
7767 static void
7768 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7769 {
7770         if (cstate->minorversion) {
7771                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7772                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7773         }
7774 }
7775
7776 void
7777 clear_current_stateid(struct nfsd4_compound_state *cstate)
7778 {
7779         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7780 }
7781
7782 /*
7783  * functions to set current state id
7784  */
7785 void
7786 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7787                 union nfsd4_op_u *u)
7788 {
7789         put_stateid(cstate, &u->open_downgrade.od_stateid);
7790 }
7791
7792 void
7793 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7794                 union nfsd4_op_u *u)
7795 {
7796         put_stateid(cstate, &u->open.op_stateid);
7797 }
7798
7799 void
7800 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7801                 union nfsd4_op_u *u)
7802 {
7803         put_stateid(cstate, &u->close.cl_stateid);
7804 }
7805
7806 void
7807 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7808                 union nfsd4_op_u *u)
7809 {
7810         put_stateid(cstate, &u->lock.lk_resp_stateid);
7811 }
7812
7813 /*
7814  * functions to consume current state id
7815  */
7816
7817 void
7818 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7819                 union nfsd4_op_u *u)
7820 {
7821         get_stateid(cstate, &u->open_downgrade.od_stateid);
7822 }
7823
7824 void
7825 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7826                 union nfsd4_op_u *u)
7827 {
7828         get_stateid(cstate, &u->delegreturn.dr_stateid);
7829 }
7830
7831 void
7832 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7833                 union nfsd4_op_u *u)
7834 {
7835         get_stateid(cstate, &u->free_stateid.fr_stateid);
7836 }
7837
7838 void
7839 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7840                 union nfsd4_op_u *u)
7841 {
7842         get_stateid(cstate, &u->setattr.sa_stateid);
7843 }
7844
7845 void
7846 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7847                 union nfsd4_op_u *u)
7848 {
7849         get_stateid(cstate, &u->close.cl_stateid);
7850 }
7851
7852 void
7853 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7854                 union nfsd4_op_u *u)
7855 {
7856         get_stateid(cstate, &u->locku.lu_stateid);
7857 }
7858
7859 void
7860 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7861                 union nfsd4_op_u *u)
7862 {
7863         get_stateid(cstate, &u->read.rd_stateid);
7864 }
7865
7866 void
7867 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7868                 union nfsd4_op_u *u)
7869 {
7870         get_stateid(cstate, &u->write.wr_stateid);
7871 }