fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include <linux/jhash.h>
  45 #include <linux/string_helpers.h>
  46 #include "xdr4.h"
  47 #include "xdr4cb.h"
  48 #include "vfs.h"
  49 #include "current_stateid.h"
  50
  51 #include "netns.h"
  52 #include "pnfs.h"
  53 #include "filecache.h"
  54
  55 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  56
  57 #define all_ones {{~0,~0},~0}
  58 static const stateid_t one_stateid = {
  59         .si_generation = ~0,
  60         .si_opaque = all_ones,
  61 };
  62 static const stateid_t zero_stateid = {
  63         /* all fields zero */
  64 };
  65 static const stateid_t currentstateid = {
  66         .si_generation = 1,
  67 };
  68 static const stateid_t close_stateid = {
  69         .si_generation = 0xffffffffU,
  70 };
  71
  72 static u64 current_sessionid = 1;
  73
  74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  75 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  76 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  77 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  78
  79 /* forward declarations */
  80 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  81 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  82 void nfsd4_end_grace(struct nfsd_net *nn);
  83
  84 /* Locking: */
  85
  86 /*
  87  * Currently used for the del_recall_lru and file hash table.  In an
  88  * effort to decrease the scope of the client_mutex, this spinlock may
  89  * eventually cover more:
  90  */
  91 static DEFINE_SPINLOCK(state_lock);
  92
  93 enum nfsd4_st_mutex_lock_subclass {
  94         OPEN_STATEID_MUTEX = 0,
  95         LOCK_STATEID_MUTEX = 1,
  96 };
  97
  98 /*
  99  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
 100  * the refcount on the open stateid to drop.
 101  */
 102 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 103
 104 /*
 105  * A waitqueue where a writer to clients/#/ctl destroying a client can
 106  * wait for cl_rpc_users to drop to 0 and then for the client to be
 107  * unhashed.
 108  */
 109 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
 110
 111 static struct kmem_cache *client_slab;
 112 static struct kmem_cache *openowner_slab;
 113 static struct kmem_cache *lockowner_slab;
 114 static struct kmem_cache *file_slab;
 115 static struct kmem_cache *stateid_slab;
 116 static struct kmem_cache *deleg_slab;
 117 static struct kmem_cache *odstate_slab;
 118
 119 static void free_session(struct nfsd4_session *);
 120
 121 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 122 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 123
 124 static bool is_session_dead(struct nfsd4_session *ses)
 125 {
 126         return ses->se_flags & NFS4_SESSION_DEAD;
 127 }
 128
 129 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 130 {
 131         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 132                 return nfserr_jukebox;
 133         ses->se_flags |= NFS4_SESSION_DEAD;
 134         return nfs_ok;
 135 }
 136
 137 static bool is_client_expired(struct nfs4_client *clp)
 138 {
 139         return clp->cl_time == 0;
 140 }
 141
 142 static __be32 get_client_locked(struct nfs4_client *clp)
 143 {
 144         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 145
 146         lockdep_assert_held(&nn->client_lock);
 147
 148         if (is_client_expired(clp))
 149                 return nfserr_expired;
 150         atomic_inc(&clp->cl_rpc_users);
 151         return nfs_ok;
 152 }
 153
 154 /* must be called under the client_lock */
 155 static inline void
 156 renew_client_locked(struct nfs4_client *clp)
 157 {
 158         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 159
 160         if (is_client_expired(clp)) {
 161                 WARN_ON(1);
 162                 printk("%s: client (clientid %08x/%08x) already expired\n",
 163                         __func__,
 164                         clp->cl_clientid.cl_boot,
 165                         clp->cl_clientid.cl_id);
 166                 return;
 167         }
 168
 169         dprintk("renewing client (clientid %08x/%08x)\n",
 170                         clp->cl_clientid.cl_boot,
 171                         clp->cl_clientid.cl_id);
 172         list_move_tail(&clp->cl_lru, &nn->client_lru);
 173         clp->cl_time = get_seconds();
 174 }
 175
 176 static void put_client_renew_locked(struct nfs4_client *clp)
 177 {
 178         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 179
 180         lockdep_assert_held(&nn->client_lock);
 181
 182         if (!atomic_dec_and_test(&clp->cl_rpc_users))
 183                 return;
 184         if (!is_client_expired(clp))
 185                 renew_client_locked(clp);
 186         else
 187                 wake_up_all(&expiry_wq);
 188 }
 189
 190 static void put_client_renew(struct nfs4_client *clp)
 191 {
 192         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 193
 194         if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
 195                 return;
 196         if (!is_client_expired(clp))
 197                 renew_client_locked(clp);
 198         else
 199                 wake_up_all(&expiry_wq);
 200         spin_unlock(&nn->client_lock);
 201 }
 202
 203 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 204 {
 205         __be32 status;
 206
 207         if (is_session_dead(ses))
 208                 return nfserr_badsession;
 209         status = get_client_locked(ses->se_client);
 210         if (status)
 211                 return status;
 212         atomic_inc(&ses->se_ref);
 213         return nfs_ok;
 214 }
 215
 216 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 217 {
 218         struct nfs4_client *clp = ses->se_client;
 219         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 220
 221         lockdep_assert_held(&nn->client_lock);
 222
 223         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 224                 free_session(ses);
 225         put_client_renew_locked(clp);
 226 }
 227
 228 static void nfsd4_put_session(struct nfsd4_session *ses)
 229 {
 230         struct nfs4_client *clp = ses->se_client;
 231         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 232
 233         spin_lock(&nn->client_lock);
 234         nfsd4_put_session_locked(ses);
 235         spin_unlock(&nn->client_lock);
 236 }
 237
 238 static struct nfsd4_blocked_lock *
 239 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 240                         struct nfsd_net *nn)
 241 {
 242         struct nfsd4_blocked_lock *cur, *found = NULL;
 243
 244         spin_lock(&nn->blocked_locks_lock);
 245         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 246                 if (fh_match(fh, &cur->nbl_fh)) {
 247                         list_del_init(&cur->nbl_list);
 248                         list_del_init(&cur->nbl_lru);
 249                         found = cur;
 250                         break;
 251                 }
 252         }
 253         spin_unlock(&nn->blocked_locks_lock);
 254         if (found)
 255                 locks_delete_block(&found->nbl_lock);
 256         return found;
 257 }
 258
 259 static struct nfsd4_blocked_lock *
 260 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 261                         struct nfsd_net *nn)
 262 {
 263         struct nfsd4_blocked_lock *nbl;
 264
 265         nbl = find_blocked_lock(lo, fh, nn);
 266         if (!nbl) {
 267                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 268                 if (nbl) {
 269                         fh_copy_shallow(&nbl->nbl_fh, fh);
 270                         locks_init_lock(&nbl->nbl_lock);
 271                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 272                                         &nfsd4_cb_notify_lock_ops,
 273                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 274                 }
 275         }
 276         return nbl;
 277 }
 278
 279 static void
 280 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 281 {
 282         locks_delete_block(&nbl->nbl_lock);
 283         locks_release_private(&nbl->nbl_lock);
 284         kfree(nbl);
 285 }
 286
 287 static void
 288 remove_blocked_locks(struct nfs4_lockowner *lo)
 289 {
 290         struct nfs4_client *clp = lo->lo_owner.so_client;
 291         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 292         struct nfsd4_blocked_lock *nbl;
 293         LIST_HEAD(reaplist);
 294
 295         /* Dequeue all blocked locks */
 296         spin_lock(&nn->blocked_locks_lock);
 297         while (!list_empty(&lo->lo_blocked)) {
 298                 nbl = list_first_entry(&lo->lo_blocked,
 299                                         struct nfsd4_blocked_lock,
 300                                         nbl_list);
 301                 list_del_init(&nbl->nbl_list);
 302                 list_move(&nbl->nbl_lru, &reaplist);
 303         }
 304         spin_unlock(&nn->blocked_locks_lock);
 305
 306         /* Now free them */
 307         while (!list_empty(&reaplist)) {
 308                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 309                                         nbl_lru);
 310                 list_del_init(&nbl->nbl_lru);
 311                 free_blocked_lock(nbl);
 312         }
 313 }
 314
 315 static void
 316 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 317 {
 318         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 319                                                 struct nfsd4_blocked_lock, nbl_cb);
 320         locks_delete_block(&nbl->nbl_lock);
 321 }
 322
 323 static int
 324 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 325 {
 326         /*
 327          * Since this is just an optimization, we don't try very hard if it
 328          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 329          * just quit trying on anything else.
 330          */
 331         switch (task->tk_status) {
 332         case -NFS4ERR_DELAY:
 333                 rpc_delay(task, 1 * HZ);
 334                 return 0;
 335         default:
 336                 return 1;
 337         }
 338 }
 339
 340 static void
 341 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 342 {
 343         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 344                                                 struct nfsd4_blocked_lock, nbl_cb);
 345
 346         free_blocked_lock(nbl);
 347 }
 348
 349 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 350         .prepare        = nfsd4_cb_notify_lock_prepare,
 351         .done           = nfsd4_cb_notify_lock_done,
 352         .release        = nfsd4_cb_notify_lock_release,
 353 };
 354
 355 static inline struct nfs4_stateowner *
 356 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 357 {
 358         atomic_inc(&sop->so_count);
 359         return sop;
 360 }
 361
 362 static int
 363 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 364 {
 365         return (sop->so_owner.len == owner->len) &&
 366                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 367 }
 368
 369 static struct nfs4_openowner *
 370 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 371                         struct nfs4_client *clp)
 372 {
 373         struct nfs4_stateowner *so;
 374
 375         lockdep_assert_held(&clp->cl_lock);
 376
 377         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 378                             so_strhash) {
 379                 if (!so->so_is_open_owner)
 380                         continue;
 381                 if (same_owner_str(so, &open->op_owner))
 382                         return openowner(nfs4_get_stateowner(so));
 383         }
 384         return NULL;
 385 }
 386
 387 static struct nfs4_openowner *
 388 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 389                         struct nfs4_client *clp)
 390 {
 391         struct nfs4_openowner *oo;
 392
 393         spin_lock(&clp->cl_lock);
 394         oo = find_openstateowner_str_locked(hashval, open, clp);
 395         spin_unlock(&clp->cl_lock);
 396         return oo;
 397 }
 398
 399 static inline u32
 400 opaque_hashval(const void *ptr, int nbytes)
 401 {
 402         unsigned char *cptr = (unsigned char *) ptr;
 403
 404         u32 x = 0;
 405         while (nbytes--) {
 406                 x *= 37;
 407                 x += *cptr++;
 408         }
 409         return x;
 410 }
 411
 412 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 413 {
 414         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 415
 416         kmem_cache_free(file_slab, fp);
 417 }
 418
 419 void
 420 put_nfs4_file(struct nfs4_file *fi)
 421 {
 422         might_lock(&state_lock);
 423
 424         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 425                 hlist_del_rcu(&fi->fi_hash);
 426                 spin_unlock(&state_lock);
 427                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 428                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 429                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 430         }
 431 }
 432
 433 static struct nfsd_file *
 434 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 435 {
 436         if (f->fi_fds[oflag])
 437                 return nfsd_file_get(f->fi_fds[oflag]);
 438         return NULL;
 439 }
 440
 441 static struct nfsd_file *
 442 find_writeable_file_locked(struct nfs4_file *f)
 443 {
 444         struct nfsd_file *ret;
 445
 446         lockdep_assert_held(&f->fi_lock);
 447
 448         ret = __nfs4_get_fd(f, O_WRONLY);
 449         if (!ret)
 450                 ret = __nfs4_get_fd(f, O_RDWR);
 451         return ret;
 452 }
 453
 454 static struct nfsd_file *
 455 find_writeable_file(struct nfs4_file *f)
 456 {
 457         struct nfsd_file *ret;
 458
 459         spin_lock(&f->fi_lock);
 460         ret = find_writeable_file_locked(f);
 461         spin_unlock(&f->fi_lock);
 462
 463         return ret;
 464 }
 465
 466 static struct nfsd_file *
 467 find_readable_file_locked(struct nfs4_file *f)
 468 {
 469         struct nfsd_file *ret;
 470
 471         lockdep_assert_held(&f->fi_lock);
 472
 473         ret = __nfs4_get_fd(f, O_RDONLY);
 474         if (!ret)
 475                 ret = __nfs4_get_fd(f, O_RDWR);
 476         return ret;
 477 }
 478
 479 static struct nfsd_file *
 480 find_readable_file(struct nfs4_file *f)
 481 {
 482         struct nfsd_file *ret;
 483
 484         spin_lock(&f->fi_lock);
 485         ret = find_readable_file_locked(f);
 486         spin_unlock(&f->fi_lock);
 487
 488         return ret;
 489 }
 490
 491 struct nfsd_file *
 492 find_any_file(struct nfs4_file *f)
 493 {
 494         struct nfsd_file *ret;
 495
 496         spin_lock(&f->fi_lock);
 497         ret = __nfs4_get_fd(f, O_RDWR);
 498         if (!ret) {
 499                 ret = __nfs4_get_fd(f, O_WRONLY);
 500                 if (!ret)
 501                         ret = __nfs4_get_fd(f, O_RDONLY);
 502         }
 503         spin_unlock(&f->fi_lock);
 504         return ret;
 505 }
 506
 507 static atomic_long_t num_delegations;
 508 unsigned long max_delegations;
 509
 510 /*
 511  * Open owner state (share locks)
 512  */
 513
 514 /* hash tables for lock and open owners */
 515 #define OWNER_HASH_BITS              8
 516 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 517 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 518
 519 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 520 {
 521         unsigned int ret;
 522
 523         ret = opaque_hashval(ownername->data, ownername->len);
 524         return ret & OWNER_HASH_MASK;
 525 }
 526
 527 /* hash table for nfs4_file */
 528 #define FILE_HASH_BITS                   8
 529 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 530
 531 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 532 {
 533         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 534 }
 535
 536 static unsigned int file_hashval(struct knfsd_fh *fh)
 537 {
 538         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 539 }
 540
 541 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 542
 543 static void
 544 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 545 {
 546         lockdep_assert_held(&fp->fi_lock);
 547
 548         if (access & NFS4_SHARE_ACCESS_WRITE)
 549                 atomic_inc(&fp->fi_access[O_WRONLY]);
 550         if (access & NFS4_SHARE_ACCESS_READ)
 551                 atomic_inc(&fp->fi_access[O_RDONLY]);
 552 }
 553
 554 static __be32
 555 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 556 {
 557         lockdep_assert_held(&fp->fi_lock);
 558
 559         /* Does this access mode make sense? */
 560         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 561                 return nfserr_inval;
 562
 563         /* Does it conflict with a deny mode already set? */
 564         if ((access & fp->fi_share_deny) != 0)
 565                 return nfserr_share_denied;
 566
 567         __nfs4_file_get_access(fp, access);
 568         return nfs_ok;
 569 }
 570
 571 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 572 {
 573         /* Common case is that there is no deny mode. */
 574         if (deny) {
 575                 /* Does this deny mode make sense? */
 576                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 577                         return nfserr_inval;
 578
 579                 if ((deny & NFS4_SHARE_DENY_READ) &&
 580                     atomic_read(&fp->fi_access[O_RDONLY]))
 581                         return nfserr_share_denied;
 582
 583                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 584                     atomic_read(&fp->fi_access[O_WRONLY]))
 585                         return nfserr_share_denied;
 586         }
 587         return nfs_ok;
 588 }
 589
 590 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 591 {
 592         might_lock(&fp->fi_lock);
 593
 594         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 595                 struct nfsd_file *f1 = NULL;
 596                 struct nfsd_file *f2 = NULL;
 597
 598                 swap(f1, fp->fi_fds[oflag]);
 599                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 600                         swap(f2, fp->fi_fds[O_RDWR]);
 601                 spin_unlock(&fp->fi_lock);
 602                 if (f1)
 603                         nfsd_file_put(f1);
 604                 if (f2)
 605                         nfsd_file_put(f2);
 606         }
 607 }
 608
 609 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 610 {
 611         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 612
 613         if (access & NFS4_SHARE_ACCESS_WRITE)
 614                 __nfs4_file_put_access(fp, O_WRONLY);
 615         if (access & NFS4_SHARE_ACCESS_READ)
 616                 __nfs4_file_put_access(fp, O_RDONLY);
 617 }
 618
 619 /*
 620  * Allocate a new open/delegation state counter. This is needed for
 621  * pNFS for proper return on close semantics.
 622  *
 623  * Note that we only allocate it for pNFS-enabled exports, otherwise
 624  * all pointers to struct nfs4_clnt_odstate are always NULL.
 625  */
 626 static struct nfs4_clnt_odstate *
 627 alloc_clnt_odstate(struct nfs4_client *clp)
 628 {
 629         struct nfs4_clnt_odstate *co;
 630
 631         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 632         if (co) {
 633                 co->co_client = clp;
 634                 refcount_set(&co->co_odcount, 1);
 635         }
 636         return co;
 637 }
 638
 639 static void
 640 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 641 {
 642         struct nfs4_file *fp = co->co_file;
 643
 644         lockdep_assert_held(&fp->fi_lock);
 645         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 646 }
 647
 648 static inline void
 649 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 650 {
 651         if (co)
 652                 refcount_inc(&co->co_odcount);
 653 }
 654
 655 static void
 656 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 657 {
 658         struct nfs4_file *fp;
 659
 660         if (!co)
 661                 return;
 662
 663         fp = co->co_file;
 664         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 665                 list_del(&co->co_perfile);
 666                 spin_unlock(&fp->fi_lock);
 667
 668                 nfsd4_return_all_file_layouts(co->co_client, fp);
 669                 kmem_cache_free(odstate_slab, co);
 670         }
 671 }
 672
 673 static struct nfs4_clnt_odstate *
 674 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 675 {
 676         struct nfs4_clnt_odstate *co;
 677         struct nfs4_client *cl;
 678
 679         if (!new)
 680                 return NULL;
 681
 682         cl = new->co_client;
 683
 684         spin_lock(&fp->fi_lock);
 685         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 686                 if (co->co_client == cl) {
 687                         get_clnt_odstate(co);
 688                         goto out;
 689                 }
 690         }
 691         co = new;
 692         co->co_file = fp;
 693         hash_clnt_odstate_locked(new);
 694 out:
 695         spin_unlock(&fp->fi_lock);
 696         return co;
 697 }
 698
 699 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 700                                   void (*sc_free)(struct nfs4_stid *))
 701 {
 702         struct nfs4_stid *stid;
 703         int new_id;
 704
 705         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 706         if (!stid)
 707                 return NULL;
 708
 709         idr_preload(GFP_KERNEL);
 710         spin_lock(&cl->cl_lock);
 711         /* Reserving 0 for start of file in nfsdfs "states" file: */
 712         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 713         spin_unlock(&cl->cl_lock);
 714         idr_preload_end();
 715         if (new_id < 0)
 716                 goto out_free;
 717
 718         stid->sc_free = sc_free;
 719         stid->sc_client = cl;
 720         stid->sc_stateid.si_opaque.so_id = new_id;
 721         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 722         /* Will be incremented before return to client: */
 723         refcount_set(&stid->sc_count, 1);
 724         spin_lock_init(&stid->sc_lock);
 725
 726         /*
 727          * It shouldn't be a problem to reuse an opaque stateid value.
 728          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 729          * example, a stray write retransmission could be accepted by
 730          * the server when it should have been rejected.  Therefore,
 731          * adopt a trick from the sctp code to attempt to maximize the
 732          * amount of time until an id is reused, by ensuring they always
 733          * "increase" (mod INT_MAX):
 734          */
 735         return stid;
 736 out_free:
 737         kmem_cache_free(slab, stid);
 738         return NULL;
 739 }
 740
 741 /*
 742  * Create a unique stateid_t to represent each COPY.
 743  */
 744 int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 745 {
 746         int new_id;
 747
 748         idr_preload(GFP_KERNEL);
 749         spin_lock(&nn->s2s_cp_lock);
 750         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, copy, 0, 0, GFP_NOWAIT);
 751         spin_unlock(&nn->s2s_cp_lock);
 752         idr_preload_end();
 753         if (new_id < 0)
 754                 return 0;
 755         copy->cp_stateid.si_opaque.so_id = new_id;
 756         copy->cp_stateid.si_opaque.so_clid.cl_boot = nn->boot_time;
 757         copy->cp_stateid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 758         return 1;
 759 }
 760
 761 void nfs4_free_cp_state(struct nfsd4_copy *copy)
 762 {
 763         struct nfsd_net *nn;
 764
 765         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 766         spin_lock(&nn->s2s_cp_lock);
 767         idr_remove(&nn->s2s_cp_stateids, copy->cp_stateid.si_opaque.so_id);
 768         spin_unlock(&nn->s2s_cp_lock);
 769 }
 770
 771 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 772 {
 773         struct nfs4_stid *stid;
 774
 775         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 776         if (!stid)
 777                 return NULL;
 778
 779         return openlockstateid(stid);
 780 }
 781
 782 static void nfs4_free_deleg(struct nfs4_stid *stid)
 783 {
 784         kmem_cache_free(deleg_slab, stid);
 785         atomic_long_dec(&num_delegations);
 786 }
 787
 788 /*
 789  * When we recall a delegation, we should be careful not to hand it
 790  * out again straight away.
 791  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 792  * in which the filehandles of recalled delegations are "stored".
 793  * If a filehandle appear in either filter, a delegation is blocked.
 794  * When a delegation is recalled, the filehandle is stored in the "new"
 795  * filter.
 796  * Every 30 seconds we swap the filters and clear the "new" one,
 797  * unless both are empty of course.
 798  *
 799  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 800  * low 3 bytes as hash-table indices.
 801  *
 802  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 803  * is used to manage concurrent access.  Testing does not need the lock
 804  * except when swapping the two filters.
 805  */
 806 static DEFINE_SPINLOCK(blocked_delegations_lock);
 807 static struct bloom_pair {
 808         int     entries, old_entries;
 809         time_t  swap_time;
 810         int     new; /* index into 'set' */
 811         DECLARE_BITMAP(set[2], 256);
 812 } blocked_delegations;
 813
 814 static int delegation_blocked(struct knfsd_fh *fh)
 815 {
 816         u32 hash;
 817         struct bloom_pair *bd = &blocked_delegations;
 818
 819         if (bd->entries == 0)
 820                 return 0;
 821         if (seconds_since_boot() - bd->swap_time > 30) {
 822                 spin_lock(&blocked_delegations_lock);
 823                 if (seconds_since_boot() - bd->swap_time > 30) {
 824                         bd->entries -= bd->old_entries;
 825                         bd->old_entries = bd->entries;
 826                         memset(bd->set[bd->new], 0,
 827                                sizeof(bd->set[0]));
 828                         bd->new = 1-bd->new;
 829                         bd->swap_time = seconds_since_boot();
 830                 }
 831                 spin_unlock(&blocked_delegations_lock);
 832         }
 833         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 834         if (test_bit(hash&255, bd->set[0]) &&
 835             test_bit((hash>>8)&255, bd->set[0]) &&
 836             test_bit((hash>>16)&255, bd->set[0]))
 837                 return 1;
 838
 839         if (test_bit(hash&255, bd->set[1]) &&
 840             test_bit((hash>>8)&255, bd->set[1]) &&
 841             test_bit((hash>>16)&255, bd->set[1]))
 842                 return 1;
 843
 844         return 0;
 845 }
 846
 847 static void block_delegations(struct knfsd_fh *fh)
 848 {
 849         u32 hash;
 850         struct bloom_pair *bd = &blocked_delegations;
 851
 852         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 853
 854         spin_lock(&blocked_delegations_lock);
 855         __set_bit(hash&255, bd->set[bd->new]);
 856         __set_bit((hash>>8)&255, bd->set[bd->new]);
 857         __set_bit((hash>>16)&255, bd->set[bd->new]);
 858         if (bd->entries == 0)
 859                 bd->swap_time = seconds_since_boot();
 860         bd->entries += 1;
 861         spin_unlock(&blocked_delegations_lock);
 862 }
 863
 864 static struct nfs4_delegation *
 865 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 866                  struct svc_fh *current_fh,
 867                  struct nfs4_clnt_odstate *odstate)
 868 {
 869         struct nfs4_delegation *dp;
 870         long n;
 871
 872         dprintk("NFSD alloc_init_deleg\n");
 873         n = atomic_long_inc_return(&num_delegations);
 874         if (n < 0 || n > max_delegations)
 875                 goto out_dec;
 876         if (delegation_blocked(&current_fh->fh_handle))
 877                 goto out_dec;
 878         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 879         if (dp == NULL)
 880                 goto out_dec;
 881
 882         /*
 883          * delegation seqid's are never incremented.  The 4.1 special
 884          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 885          * 0 anyway just for consistency and use 1:
 886          */
 887         dp->dl_stid.sc_stateid.si_generation = 1;
 888         INIT_LIST_HEAD(&dp->dl_perfile);
 889         INIT_LIST_HEAD(&dp->dl_perclnt);
 890         INIT_LIST_HEAD(&dp->dl_recall_lru);
 891         dp->dl_clnt_odstate = odstate;
 892         get_clnt_odstate(odstate);
 893         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 894         dp->dl_retries = 1;
 895         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 896                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 897         get_nfs4_file(fp);
 898         dp->dl_stid.sc_file = fp;
 899         return dp;
 900 out_dec:
 901         atomic_long_dec(&num_delegations);
 902         return NULL;
 903 }
 904
 905 void
 906 nfs4_put_stid(struct nfs4_stid *s)
 907 {
 908         struct nfs4_file *fp = s->sc_file;
 909         struct nfs4_client *clp = s->sc_client;
 910
 911         might_lock(&clp->cl_lock);
 912
 913         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 914                 wake_up_all(&close_wq);
 915                 return;
 916         }
 917         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 918         spin_unlock(&clp->cl_lock);
 919         s->sc_free(s);
 920         if (fp)
 921                 put_nfs4_file(fp);
 922 }
 923
 924 void
 925 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 926 {
 927         stateid_t *src = &stid->sc_stateid;
 928
 929         spin_lock(&stid->sc_lock);
 930         if (unlikely(++src->si_generation == 0))
 931                 src->si_generation = 1;
 932         memcpy(dst, src, sizeof(*dst));
 933         spin_unlock(&stid->sc_lock);
 934 }
 935
 936 static void put_deleg_file(struct nfs4_file *fp)
 937 {
 938         struct nfsd_file *nf = NULL;
 939
 940         spin_lock(&fp->fi_lock);
 941         if (--fp->fi_delegees == 0)
 942                 swap(nf, fp->fi_deleg_file);
 943         spin_unlock(&fp->fi_lock);
 944
 945         if (nf)
 946                 nfsd_file_put(nf);
 947 }
 948
 949 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 950 {
 951         struct nfs4_file *fp = dp->dl_stid.sc_file;
 952         struct nfsd_file *nf = fp->fi_deleg_file;
 953
 954         WARN_ON_ONCE(!fp->fi_delegees);
 955
 956         vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
 957         put_deleg_file(fp);
 958 }
 959
 960 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 961 {
 962         put_clnt_odstate(dp->dl_clnt_odstate);
 963         nfs4_unlock_deleg_lease(dp);
 964         nfs4_put_stid(&dp->dl_stid);
 965 }
 966
 967 void nfs4_unhash_stid(struct nfs4_stid *s)
 968 {
 969         s->sc_type = 0;
 970 }
 971
 972 /**
 973  * nfs4_delegation_exists - Discover if this delegation already exists
 974  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 975  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 976  *
 977  * Return:
 978  *      On success: true iff an existing delegation is found
 979  */
 980
 981 static bool
 982 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 983 {
 984         struct nfs4_delegation *searchdp = NULL;
 985         struct nfs4_client *searchclp = NULL;
 986
 987         lockdep_assert_held(&state_lock);
 988         lockdep_assert_held(&fp->fi_lock);
 989
 990         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 991                 searchclp = searchdp->dl_stid.sc_client;
 992                 if (clp == searchclp) {
 993                         return true;
 994                 }
 995         }
 996         return false;
 997 }
 998
 999 /**
1000  * hash_delegation_locked - Add a delegation to the appropriate lists
1001  * @dp:     a pointer to the nfs4_delegation we are adding.
1002  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1003  *
1004  * Return:
1005  *      On success: NULL if the delegation was successfully hashed.
1006  *
1007  *      On error: -EAGAIN if one was previously granted to this
1008  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1009  *
1010  */
1011
1012 static int
1013 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1014 {
1015         struct nfs4_client *clp = dp->dl_stid.sc_client;
1016
1017         lockdep_assert_held(&state_lock);
1018         lockdep_assert_held(&fp->fi_lock);
1019
1020         if (nfs4_delegation_exists(clp, fp))
1021                 return -EAGAIN;
1022         refcount_inc(&dp->dl_stid.sc_count);
1023         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1024         list_add(&dp->dl_perfile, &fp->fi_delegations);
1025         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1026         return 0;
1027 }
1028
1029 static bool
1030 unhash_delegation_locked(struct nfs4_delegation *dp)
1031 {
1032         struct nfs4_file *fp = dp->dl_stid.sc_file;
1033
1034         lockdep_assert_held(&state_lock);
1035
1036         if (list_empty(&dp->dl_perfile))
1037                 return false;
1038
1039         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1040         /* Ensure that deleg break won't try to requeue it */
1041         ++dp->dl_time;
1042         spin_lock(&fp->fi_lock);
1043         list_del_init(&dp->dl_perclnt);
1044         list_del_init(&dp->dl_recall_lru);
1045         list_del_init(&dp->dl_perfile);
1046         spin_unlock(&fp->fi_lock);
1047         return true;
1048 }
1049
1050 static void destroy_delegation(struct nfs4_delegation *dp)
1051 {
1052         bool unhashed;
1053
1054         spin_lock(&state_lock);
1055         unhashed = unhash_delegation_locked(dp);
1056         spin_unlock(&state_lock);
1057         if (unhashed)
1058                 destroy_unhashed_deleg(dp);
1059 }
1060
1061 static void revoke_delegation(struct nfs4_delegation *dp)
1062 {
1063         struct nfs4_client *clp = dp->dl_stid.sc_client;
1064
1065         WARN_ON(!list_empty(&dp->dl_recall_lru));
1066
1067         if (clp->cl_minorversion) {
1068                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1069                 refcount_inc(&dp->dl_stid.sc_count);
1070                 spin_lock(&clp->cl_lock);
1071                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1072                 spin_unlock(&clp->cl_lock);
1073         }
1074         destroy_unhashed_deleg(dp);
1075 }
1076
1077 /*
1078  * SETCLIENTID state
1079  */
1080
1081 static unsigned int clientid_hashval(u32 id)
1082 {
1083         return id & CLIENT_HASH_MASK;
1084 }
1085
1086 static unsigned int clientstr_hashval(struct xdr_netobj name)
1087 {
1088         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1089 }
1090
1091 /*
1092  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1093  * st_{access,deny}_bmap field of the stateid, in order to track not
1094  * only what share bits are currently in force, but also what
1095  * combinations of share bits previous opens have used.  This allows us
1096  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1097  * return an error if the client attempt to downgrade to a combination
1098  * of share bits not explicable by closing some of its previous opens.
1099  *
1100  * XXX: This enforcement is actually incomplete, since we don't keep
1101  * track of access/deny bit combinations; so, e.g., we allow:
1102  *
1103  *      OPEN allow read, deny write
1104  *      OPEN allow both, deny none
1105  *      DOWNGRADE allow read, deny none
1106  *
1107  * which we should reject.
1108  */
1109 static unsigned int
1110 bmap_to_share_mode(unsigned long bmap) {
1111         int i;
1112         unsigned int access = 0;
1113
1114         for (i = 1; i < 4; i++) {
1115                 if (test_bit(i, &bmap))
1116                         access |= i;
1117         }
1118         return access;
1119 }
1120
1121 /* set share access for a given stateid */
1122 static inline void
1123 set_access(u32 access, struct nfs4_ol_stateid *stp)
1124 {
1125         unsigned char mask = 1 << access;
1126
1127         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1128         stp->st_access_bmap |= mask;
1129 }
1130
1131 /* clear share access for a given stateid */
1132 static inline void
1133 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1134 {
1135         unsigned char mask = 1 << access;
1136
1137         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1138         stp->st_access_bmap &= ~mask;
1139 }
1140
1141 /* test whether a given stateid has access */
1142 static inline bool
1143 test_access(u32 access, struct nfs4_ol_stateid *stp)
1144 {
1145         unsigned char mask = 1 << access;
1146
1147         return (bool)(stp->st_access_bmap & mask);
1148 }
1149
1150 /* set share deny for a given stateid */
1151 static inline void
1152 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1153 {
1154         unsigned char mask = 1 << deny;
1155
1156         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1157         stp->st_deny_bmap |= mask;
1158 }
1159
1160 /* clear share deny for a given stateid */
1161 static inline void
1162 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1163 {
1164         unsigned char mask = 1 << deny;
1165
1166         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1167         stp->st_deny_bmap &= ~mask;
1168 }
1169
1170 /* test whether a given stateid is denying specific access */
1171 static inline bool
1172 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1173 {
1174         unsigned char mask = 1 << deny;
1175
1176         return (bool)(stp->st_deny_bmap & mask);
1177 }
1178
1179 static int nfs4_access_to_omode(u32 access)
1180 {
1181         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1182         case NFS4_SHARE_ACCESS_READ:
1183                 return O_RDONLY;
1184         case NFS4_SHARE_ACCESS_WRITE:
1185                 return O_WRONLY;
1186         case NFS4_SHARE_ACCESS_BOTH:
1187                 return O_RDWR;
1188         }
1189         WARN_ON_ONCE(1);
1190         return O_RDONLY;
1191 }
1192
1193 /*
1194  * A stateid that had a deny mode associated with it is being released
1195  * or downgraded. Recalculate the deny mode on the file.
1196  */
1197 static void
1198 recalculate_deny_mode(struct nfs4_file *fp)
1199 {
1200         struct nfs4_ol_stateid *stp;
1201
1202         spin_lock(&fp->fi_lock);
1203         fp->fi_share_deny = 0;
1204         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1205                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1206         spin_unlock(&fp->fi_lock);
1207 }
1208
1209 static void
1210 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1211 {
1212         int i;
1213         bool change = false;
1214
1215         for (i = 1; i < 4; i++) {
1216                 if ((i & deny) != i) {
1217                         change = true;
1218                         clear_deny(i, stp);
1219                 }
1220         }
1221
1222         /* Recalculate per-file deny mode if there was a change */
1223         if (change)
1224                 recalculate_deny_mode(stp->st_stid.sc_file);
1225 }
1226
1227 /* release all access and file references for a given stateid */
1228 static void
1229 release_all_access(struct nfs4_ol_stateid *stp)
1230 {
1231         int i;
1232         struct nfs4_file *fp = stp->st_stid.sc_file;
1233
1234         if (fp && stp->st_deny_bmap != 0)
1235                 recalculate_deny_mode(fp);
1236
1237         for (i = 1; i < 4; i++) {
1238                 if (test_access(i, stp))
1239                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1240                 clear_access(i, stp);
1241         }
1242 }
1243
1244 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1245 {
1246         kfree(sop->so_owner.data);
1247         sop->so_ops->so_free(sop);
1248 }
1249
1250 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1251 {
1252         struct nfs4_client *clp = sop->so_client;
1253
1254         might_lock(&clp->cl_lock);
1255
1256         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1257                 return;
1258         sop->so_ops->so_unhash(sop);
1259         spin_unlock(&clp->cl_lock);
1260         nfs4_free_stateowner(sop);
1261 }
1262
1263 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1264 {
1265         struct nfs4_file *fp = stp->st_stid.sc_file;
1266
1267         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1268
1269         if (list_empty(&stp->st_perfile))
1270                 return false;
1271
1272         spin_lock(&fp->fi_lock);
1273         list_del_init(&stp->st_perfile);
1274         spin_unlock(&fp->fi_lock);
1275         list_del(&stp->st_perstateowner);
1276         return true;
1277 }
1278
1279 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1280 {
1281         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1282
1283         put_clnt_odstate(stp->st_clnt_odstate);
1284         release_all_access(stp);
1285         if (stp->st_stateowner)
1286                 nfs4_put_stateowner(stp->st_stateowner);
1287         kmem_cache_free(stateid_slab, stid);
1288 }
1289
1290 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1291 {
1292         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1293         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1294         struct nfsd_file *nf;
1295
1296         nf = find_any_file(stp->st_stid.sc_file);
1297         if (nf) {
1298                 get_file(nf->nf_file);
1299                 filp_close(nf->nf_file, (fl_owner_t)lo);
1300                 nfsd_file_put(nf);
1301         }
1302         nfs4_free_ol_stateid(stid);
1303 }
1304
1305 /*
1306  * Put the persistent reference to an already unhashed generic stateid, while
1307  * holding the cl_lock. If it's the last reference, then put it onto the
1308  * reaplist for later destruction.
1309  */
1310 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1311                                        struct list_head *reaplist)
1312 {
1313         struct nfs4_stid *s = &stp->st_stid;
1314         struct nfs4_client *clp = s->sc_client;
1315
1316         lockdep_assert_held(&clp->cl_lock);
1317
1318         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1319
1320         if (!refcount_dec_and_test(&s->sc_count)) {
1321                 wake_up_all(&close_wq);
1322                 return;
1323         }
1324
1325         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1326         list_add(&stp->st_locks, reaplist);
1327 }
1328
1329 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1330 {
1331         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1332
1333         list_del_init(&stp->st_locks);
1334         nfs4_unhash_stid(&stp->st_stid);
1335         return unhash_ol_stateid(stp);
1336 }
1337
1338 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1339 {
1340         struct nfs4_client *clp = stp->st_stid.sc_client;
1341         bool unhashed;
1342
1343         spin_lock(&clp->cl_lock);
1344         unhashed = unhash_lock_stateid(stp);
1345         spin_unlock(&clp->cl_lock);
1346         if (unhashed)
1347                 nfs4_put_stid(&stp->st_stid);
1348 }
1349
1350 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1351 {
1352         struct nfs4_client *clp = lo->lo_owner.so_client;
1353
1354         lockdep_assert_held(&clp->cl_lock);
1355
1356         list_del_init(&lo->lo_owner.so_strhash);
1357 }
1358
1359 /*
1360  * Free a list of generic stateids that were collected earlier after being
1361  * fully unhashed.
1362  */
1363 static void
1364 free_ol_stateid_reaplist(struct list_head *reaplist)
1365 {
1366         struct nfs4_ol_stateid *stp;
1367         struct nfs4_file *fp;
1368
1369         might_sleep();
1370
1371         while (!list_empty(reaplist)) {
1372                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1373                                        st_locks);
1374                 list_del(&stp->st_locks);
1375                 fp = stp->st_stid.sc_file;
1376                 stp->st_stid.sc_free(&stp->st_stid);
1377                 if (fp)
1378                         put_nfs4_file(fp);
1379         }
1380 }
1381
1382 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1383                                        struct list_head *reaplist)
1384 {
1385         struct nfs4_ol_stateid *stp;
1386
1387         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1388
1389         while (!list_empty(&open_stp->st_locks)) {
1390                 stp = list_entry(open_stp->st_locks.next,
1391                                 struct nfs4_ol_stateid, st_locks);
1392                 WARN_ON(!unhash_lock_stateid(stp));
1393                 put_ol_stateid_locked(stp, reaplist);
1394         }
1395 }
1396
1397 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1398                                 struct list_head *reaplist)
1399 {
1400         bool unhashed;
1401
1402         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1403
1404         unhashed = unhash_ol_stateid(stp);
1405         release_open_stateid_locks(stp, reaplist);
1406         return unhashed;
1407 }
1408
1409 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1410 {
1411         LIST_HEAD(reaplist);
1412
1413         spin_lock(&stp->st_stid.sc_client->cl_lock);
1414         if (unhash_open_stateid(stp, &reaplist))
1415                 put_ol_stateid_locked(stp, &reaplist);
1416         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1417         free_ol_stateid_reaplist(&reaplist);
1418 }
1419
1420 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1421 {
1422         struct nfs4_client *clp = oo->oo_owner.so_client;
1423
1424         lockdep_assert_held(&clp->cl_lock);
1425
1426         list_del_init(&oo->oo_owner.so_strhash);
1427         list_del_init(&oo->oo_perclient);
1428 }
1429
1430 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1431 {
1432         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1433                                           nfsd_net_id);
1434         struct nfs4_ol_stateid *s;
1435
1436         spin_lock(&nn->client_lock);
1437         s = oo->oo_last_closed_stid;
1438         if (s) {
1439                 list_del_init(&oo->oo_close_lru);
1440                 oo->oo_last_closed_stid = NULL;
1441         }
1442         spin_unlock(&nn->client_lock);
1443         if (s)
1444                 nfs4_put_stid(&s->st_stid);
1445 }
1446
1447 static void release_openowner(struct nfs4_openowner *oo)
1448 {
1449         struct nfs4_ol_stateid *stp;
1450         struct nfs4_client *clp = oo->oo_owner.so_client;
1451         struct list_head reaplist;
1452
1453         INIT_LIST_HEAD(&reaplist);
1454
1455         spin_lock(&clp->cl_lock);
1456         unhash_openowner_locked(oo);
1457         while (!list_empty(&oo->oo_owner.so_stateids)) {
1458                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1459                                 struct nfs4_ol_stateid, st_perstateowner);
1460                 if (unhash_open_stateid(stp, &reaplist))
1461                         put_ol_stateid_locked(stp, &reaplist);
1462         }
1463         spin_unlock(&clp->cl_lock);
1464         free_ol_stateid_reaplist(&reaplist);
1465         release_last_closed_stateid(oo);
1466         nfs4_put_stateowner(&oo->oo_owner);
1467 }
1468
1469 static inline int
1470 hash_sessionid(struct nfs4_sessionid *sessionid)
1471 {
1472         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1473
1474         return sid->sequence % SESSION_HASH_SIZE;
1475 }
1476
1477 #ifdef CONFIG_SUNRPC_DEBUG
1478 static inline void
1479 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1480 {
1481         u32 *ptr = (u32 *)(&sessionid->data[0]);
1482         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1483 }
1484 #else
1485 static inline void
1486 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1487 {
1488 }
1489 #endif
1490
1491 /*
1492  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1493  * won't be used for replay.
1494  */
1495 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1496 {
1497         struct nfs4_stateowner *so = cstate->replay_owner;
1498
1499         if (nfserr == nfserr_replay_me)
1500                 return;
1501
1502         if (!seqid_mutating_err(ntohl(nfserr))) {
1503                 nfsd4_cstate_clear_replay(cstate);
1504                 return;
1505         }
1506         if (!so)
1507                 return;
1508         if (so->so_is_open_owner)
1509                 release_last_closed_stateid(openowner(so));
1510         so->so_seqid++;
1511         return;
1512 }
1513
1514 static void
1515 gen_sessionid(struct nfsd4_session *ses)
1516 {
1517         struct nfs4_client *clp = ses->se_client;
1518         struct nfsd4_sessionid *sid;
1519
1520         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1521         sid->clientid = clp->cl_clientid;
1522         sid->sequence = current_sessionid++;
1523         sid->reserved = 0;
1524 }
1525
1526 /*
1527  * The protocol defines ca_maxresponssize_cached to include the size of
1528  * the rpc header, but all we need to cache is the data starting after
1529  * the end of the initial SEQUENCE operation--the rest we regenerate
1530  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1531  * value that is the number of bytes in our cache plus a few additional
1532  * bytes.  In order to stay on the safe side, and not promise more than
1533  * we can cache, those additional bytes must be the minimum possible: 24
1534  * bytes of rpc header (xid through accept state, with AUTH_NULL
1535  * verifier), 12 for the compound header (with zero-length tag), and 44
1536  * for the SEQUENCE op response:
1537  */
1538 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1539
1540 static void
1541 free_session_slots(struct nfsd4_session *ses)
1542 {
1543         int i;
1544
1545         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1546                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1547                 kfree(ses->se_slots[i]);
1548         }
1549 }
1550
1551 /*
1552  * We don't actually need to cache the rpc and session headers, so we
1553  * can allocate a little less for each slot:
1554  */
1555 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1556 {
1557         u32 size;
1558
1559         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1560                 size = 0;
1561         else
1562                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1563         return size + sizeof(struct nfsd4_slot);
1564 }
1565
1566 /*
1567  * XXX: If we run out of reserved DRC memory we could (up to a point)
1568  * re-negotiate active sessions and reduce their slot usage to make
1569  * room for new connections. For now we just fail the create session.
1570  */
1571 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1572 {
1573         u32 slotsize = slot_bytes(ca);
1574         u32 num = ca->maxreqs;
1575         unsigned long avail, total_avail;
1576         unsigned int scale_factor;
1577
1578         spin_lock(&nfsd_drc_lock);
1579         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1580                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1581         else
1582                 /* We have handed out more space than we chose in
1583                  * set_max_drc() to allow.  That isn't really a
1584                  * problem as long as that doesn't make us think we
1585                  * have lots more due to integer overflow.
1586                  */
1587                 total_avail = 0;
1588         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1589         /*
1590          * Never use more than a fraction of the remaining memory,
1591          * unless it's the only way to give this client a slot.
1592          * The chosen fraction is either 1/8 or 1/number of threads,
1593          * whichever is smaller.  This ensures there are adequate
1594          * slots to support multiple clients per thread.
1595          * Give the client one slot even if that would require
1596          * over-allocation--it is better than failure.
1597          */
1598         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1599
1600         avail = clamp_t(unsigned long, avail, slotsize,
1601                         total_avail/scale_factor);
1602         num = min_t(int, num, avail / slotsize);
1603         num = max_t(int, num, 1);
1604         nfsd_drc_mem_used += num * slotsize;
1605         spin_unlock(&nfsd_drc_lock);
1606
1607         return num;
1608 }
1609
1610 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1611 {
1612         int slotsize = slot_bytes(ca);
1613
1614         spin_lock(&nfsd_drc_lock);
1615         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1616         spin_unlock(&nfsd_drc_lock);
1617 }
1618
1619 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1620                                            struct nfsd4_channel_attrs *battrs)
1621 {
1622         int numslots = fattrs->maxreqs;
1623         int slotsize = slot_bytes(fattrs);
1624         struct nfsd4_session *new;
1625         int mem, i;
1626
1627         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1628                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1629         mem = numslots * sizeof(struct nfsd4_slot *);
1630
1631         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1632         if (!new)
1633                 return NULL;
1634         /* allocate each struct nfsd4_slot and data cache in one piece */
1635         for (i = 0; i < numslots; i++) {
1636                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1637                 if (!new->se_slots[i])
1638                         goto out_free;
1639         }
1640
1641         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1642         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1643
1644         return new;
1645 out_free:
1646         while (i--)
1647                 kfree(new->se_slots[i]);
1648         kfree(new);
1649         return NULL;
1650 }
1651
1652 static void free_conn(struct nfsd4_conn *c)
1653 {
1654         svc_xprt_put(c->cn_xprt);
1655         kfree(c);
1656 }
1657
1658 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1659 {
1660         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1661         struct nfs4_client *clp = c->cn_session->se_client;
1662
1663         spin_lock(&clp->cl_lock);
1664         if (!list_empty(&c->cn_persession)) {
1665                 list_del(&c->cn_persession);
1666                 free_conn(c);
1667         }
1668         nfsd4_probe_callback(clp);
1669         spin_unlock(&clp->cl_lock);
1670 }
1671
1672 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1673 {
1674         struct nfsd4_conn *conn;
1675
1676         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1677         if (!conn)
1678                 return NULL;
1679         svc_xprt_get(rqstp->rq_xprt);
1680         conn->cn_xprt = rqstp->rq_xprt;
1681         conn->cn_flags = flags;
1682         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1683         return conn;
1684 }
1685
1686 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1687 {
1688         conn->cn_session = ses;
1689         list_add(&conn->cn_persession, &ses->se_conns);
1690 }
1691
1692 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1693 {
1694         struct nfs4_client *clp = ses->se_client;
1695
1696         spin_lock(&clp->cl_lock);
1697         __nfsd4_hash_conn(conn, ses);
1698         spin_unlock(&clp->cl_lock);
1699 }
1700
1701 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1702 {
1703         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1704         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1705 }
1706
1707 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1708 {
1709         int ret;
1710
1711         nfsd4_hash_conn(conn, ses);
1712         ret = nfsd4_register_conn(conn);
1713         if (ret)
1714                 /* oops; xprt is already down: */
1715                 nfsd4_conn_lost(&conn->cn_xpt_user);
1716         /* We may have gained or lost a callback channel: */
1717         nfsd4_probe_callback_sync(ses->se_client);
1718 }
1719
1720 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1721 {
1722         u32 dir = NFS4_CDFC4_FORE;
1723
1724         if (cses->flags & SESSION4_BACK_CHAN)
1725                 dir |= NFS4_CDFC4_BACK;
1726         return alloc_conn(rqstp, dir);
1727 }
1728
1729 /* must be called under client_lock */
1730 static void nfsd4_del_conns(struct nfsd4_session *s)
1731 {
1732         struct nfs4_client *clp = s->se_client;
1733         struct nfsd4_conn *c;
1734
1735         spin_lock(&clp->cl_lock);
1736         while (!list_empty(&s->se_conns)) {
1737                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1738                 list_del_init(&c->cn_persession);
1739                 spin_unlock(&clp->cl_lock);
1740
1741                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1742                 free_conn(c);
1743
1744                 spin_lock(&clp->cl_lock);
1745         }
1746         spin_unlock(&clp->cl_lock);
1747 }
1748
1749 static void __free_session(struct nfsd4_session *ses)
1750 {
1751         free_session_slots(ses);
1752         kfree(ses);
1753 }
1754
1755 static void free_session(struct nfsd4_session *ses)
1756 {
1757         nfsd4_del_conns(ses);
1758         nfsd4_put_drc_mem(&ses->se_fchannel);
1759         __free_session(ses);
1760 }
1761
1762 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1763 {
1764         int idx;
1765         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1766
1767         new->se_client = clp;
1768         gen_sessionid(new);
1769
1770         INIT_LIST_HEAD(&new->se_conns);
1771
1772         new->se_cb_seq_nr = 1;
1773         new->se_flags = cses->flags;
1774         new->se_cb_prog = cses->callback_prog;
1775         new->se_cb_sec = cses->cb_sec;
1776         atomic_set(&new->se_ref, 0);
1777         idx = hash_sessionid(&new->se_sessionid);
1778         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1779         spin_lock(&clp->cl_lock);
1780         list_add(&new->se_perclnt, &clp->cl_sessions);
1781         spin_unlock(&clp->cl_lock);
1782
1783         {
1784                 struct sockaddr *sa = svc_addr(rqstp);
1785                 /*
1786                  * This is a little silly; with sessions there's no real
1787                  * use for the callback address.  Use the peer address
1788                  * as a reasonable default for now, but consider fixing
1789                  * the rpc client not to require an address in the
1790                  * future:
1791                  */
1792                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1793                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1794         }
1795 }
1796
1797 /* caller must hold client_lock */
1798 static struct nfsd4_session *
1799 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1800 {
1801         struct nfsd4_session *elem;
1802         int idx;
1803         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1804
1805         lockdep_assert_held(&nn->client_lock);
1806
1807         dump_sessionid(__func__, sessionid);
1808         idx = hash_sessionid(sessionid);
1809         /* Search in the appropriate list */
1810         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1811                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1812                             NFS4_MAX_SESSIONID_LEN)) {
1813                         return elem;
1814                 }
1815         }
1816
1817         dprintk("%s: session not found\n", __func__);
1818         return NULL;
1819 }
1820
1821 static struct nfsd4_session *
1822 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1823                 __be32 *ret)
1824 {
1825         struct nfsd4_session *session;
1826         __be32 status = nfserr_badsession;
1827
1828         session = __find_in_sessionid_hashtbl(sessionid, net);
1829         if (!session)
1830                 goto out;
1831         status = nfsd4_get_session_locked(session);
1832         if (status)
1833                 session = NULL;
1834 out:
1835         *ret = status;
1836         return session;
1837 }
1838
1839 /* caller must hold client_lock */
1840 static void
1841 unhash_session(struct nfsd4_session *ses)
1842 {
1843         struct nfs4_client *clp = ses->se_client;
1844         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1845
1846         lockdep_assert_held(&nn->client_lock);
1847
1848         list_del(&ses->se_hash);
1849         spin_lock(&ses->se_client->cl_lock);
1850         list_del(&ses->se_perclnt);
1851         spin_unlock(&ses->se_client->cl_lock);
1852 }
1853
1854 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1855 static int
1856 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1857 {
1858         /*
1859          * We're assuming the clid was not given out from a boot
1860          * precisely 2^32 (about 136 years) before this one.  That seems
1861          * a safe assumption:
1862          */
1863         if (clid->cl_boot == (u32)nn->boot_time)
1864                 return 0;
1865         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1866                 clid->cl_boot, clid->cl_id, nn->boot_time);
1867         return 1;
1868 }
1869
1870 /*
1871  * XXX Should we use a slab cache ?
1872  * This type of memory management is somewhat inefficient, but we use it
1873  * anyway since SETCLIENTID is not a common operation.
1874  */
1875 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1876 {
1877         struct nfs4_client *clp;
1878         int i;
1879
1880         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1881         if (clp == NULL)
1882                 return NULL;
1883         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1884         if (clp->cl_name.data == NULL)
1885                 goto err_no_name;
1886         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1887                                                  sizeof(struct list_head),
1888                                                  GFP_KERNEL);
1889         if (!clp->cl_ownerstr_hashtbl)
1890                 goto err_no_hashtbl;
1891         for (i = 0; i < OWNER_HASH_SIZE; i++)
1892                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1893         INIT_LIST_HEAD(&clp->cl_sessions);
1894         idr_init(&clp->cl_stateids);
1895         atomic_set(&clp->cl_rpc_users, 0);
1896         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1897         INIT_LIST_HEAD(&clp->cl_idhash);
1898         INIT_LIST_HEAD(&clp->cl_openowners);
1899         INIT_LIST_HEAD(&clp->cl_delegations);
1900         INIT_LIST_HEAD(&clp->cl_lru);
1901         INIT_LIST_HEAD(&clp->cl_revoked);
1902 #ifdef CONFIG_NFSD_PNFS
1903         INIT_LIST_HEAD(&clp->cl_lo_states);
1904 #endif
1905         INIT_LIST_HEAD(&clp->async_copies);
1906         spin_lock_init(&clp->async_lock);
1907         spin_lock_init(&clp->cl_lock);
1908         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1909         return clp;
1910 err_no_hashtbl:
1911         kfree(clp->cl_name.data);
1912 err_no_name:
1913         kmem_cache_free(client_slab, clp);
1914         return NULL;
1915 }
1916
1917 static void __free_client(struct kref *k)
1918 {
1919         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1920         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1921
1922         free_svc_cred(&clp->cl_cred);
1923         kfree(clp->cl_ownerstr_hashtbl);
1924         kfree(clp->cl_name.data);
1925         kfree(clp->cl_nii_domain.data);
1926         kfree(clp->cl_nii_name.data);
1927         idr_destroy(&clp->cl_stateids);
1928         kmem_cache_free(client_slab, clp);
1929 }
1930
1931 static void drop_client(struct nfs4_client *clp)
1932 {
1933         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1934 }
1935
1936 static void
1937 free_client(struct nfs4_client *clp)
1938 {
1939         while (!list_empty(&clp->cl_sessions)) {
1940                 struct nfsd4_session *ses;
1941                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1942                                 se_perclnt);
1943                 list_del(&ses->se_perclnt);
1944                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1945                 free_session(ses);
1946         }
1947         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1948         if (clp->cl_nfsd_dentry) {
1949                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
1950                 clp->cl_nfsd_dentry = NULL;
1951                 wake_up_all(&expiry_wq);
1952         }
1953         drop_client(clp);
1954 }
1955
1956 /* must be called under the client_lock */
1957 static void
1958 unhash_client_locked(struct nfs4_client *clp)
1959 {
1960         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1961         struct nfsd4_session *ses;
1962
1963         lockdep_assert_held(&nn->client_lock);
1964
1965         /* Mark the client as expired! */
1966         clp->cl_time = 0;
1967         /* Make it invisible */
1968         if (!list_empty(&clp->cl_idhash)) {
1969                 list_del_init(&clp->cl_idhash);
1970                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1971                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1972                 else
1973                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1974         }
1975         list_del_init(&clp->cl_lru);
1976         spin_lock(&clp->cl_lock);
1977         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1978                 list_del_init(&ses->se_hash);
1979         spin_unlock(&clp->cl_lock);
1980 }
1981
1982 static void
1983 unhash_client(struct nfs4_client *clp)
1984 {
1985         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1986
1987         spin_lock(&nn->client_lock);
1988         unhash_client_locked(clp);
1989         spin_unlock(&nn->client_lock);
1990 }
1991
1992 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1993 {
1994         if (atomic_read(&clp->cl_rpc_users))
1995                 return nfserr_jukebox;
1996         unhash_client_locked(clp);
1997         return nfs_ok;
1998 }
1999
2000 static void
2001 __destroy_client(struct nfs4_client *clp)
2002 {
2003         int i;
2004         struct nfs4_openowner *oo;
2005         struct nfs4_delegation *dp;
2006         struct list_head reaplist;
2007
2008         INIT_LIST_HEAD(&reaplist);
2009         spin_lock(&state_lock);
2010         while (!list_empty(&clp->cl_delegations)) {
2011                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2012                 WARN_ON(!unhash_delegation_locked(dp));
2013                 list_add(&dp->dl_recall_lru, &reaplist);
2014         }
2015         spin_unlock(&state_lock);
2016         while (!list_empty(&reaplist)) {
2017                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2018                 list_del_init(&dp->dl_recall_lru);
2019                 destroy_unhashed_deleg(dp);
2020         }
2021         while (!list_empty(&clp->cl_revoked)) {
2022                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2023                 list_del_init(&dp->dl_recall_lru);
2024                 nfs4_put_stid(&dp->dl_stid);
2025         }
2026         while (!list_empty(&clp->cl_openowners)) {
2027                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2028                 nfs4_get_stateowner(&oo->oo_owner);
2029                 release_openowner(oo);
2030         }
2031         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2032                 struct nfs4_stateowner *so, *tmp;
2033
2034                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2035                                          so_strhash) {
2036                         /* Should be no openowners at this point */
2037                         WARN_ON_ONCE(so->so_is_open_owner);
2038                         remove_blocked_locks(lockowner(so));
2039                 }
2040         }
2041         nfsd4_return_all_client_layouts(clp);
2042         nfsd4_shutdown_copy(clp);
2043         nfsd4_shutdown_callback(clp);
2044         if (clp->cl_cb_conn.cb_xprt)
2045                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2046         free_client(clp);
2047         wake_up_all(&expiry_wq);
2048 }
2049
2050 static void
2051 destroy_client(struct nfs4_client *clp)
2052 {
2053         unhash_client(clp);
2054         __destroy_client(clp);
2055 }
2056
2057 static void inc_reclaim_complete(struct nfs4_client *clp)
2058 {
2059         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2060
2061         if (!nn->track_reclaim_completes)
2062                 return;
2063         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2064                 return;
2065         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2066                         nn->reclaim_str_hashtbl_size) {
2067                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2068                                 clp->net->ns.inum);
2069                 nfsd4_end_grace(nn);
2070         }
2071 }
2072
2073 static void expire_client(struct nfs4_client *clp)
2074 {
2075         unhash_client(clp);
2076         nfsd4_client_record_remove(clp);
2077         __destroy_client(clp);
2078 }
2079
2080 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2081 {
2082         memcpy(target->cl_verifier.data, source->data,
2083                         sizeof(target->cl_verifier.data));
2084 }
2085
2086 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2087 {
2088         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2089         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2090 }
2091
2092 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2093 {
2094         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2095         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2096                                                                 GFP_KERNEL);
2097         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2098         if ((source->cr_principal && !target->cr_principal) ||
2099             (source->cr_raw_principal && !target->cr_raw_principal) ||
2100             (source->cr_targ_princ && !target->cr_targ_princ))
2101                 return -ENOMEM;
2102
2103         target->cr_flavor = source->cr_flavor;
2104         target->cr_uid = source->cr_uid;
2105         target->cr_gid = source->cr_gid;
2106         target->cr_group_info = source->cr_group_info;
2107         get_group_info(target->cr_group_info);
2108         target->cr_gss_mech = source->cr_gss_mech;
2109         if (source->cr_gss_mech)
2110                 gss_mech_get(source->cr_gss_mech);
2111         return 0;
2112 }
2113
2114 static int
2115 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2116 {
2117         if (o1->len < o2->len)
2118                 return -1;
2119         if (o1->len > o2->len)
2120                 return 1;
2121         return memcmp(o1->data, o2->data, o1->len);
2122 }
2123
2124 static int
2125 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2126 {
2127         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2128 }
2129
2130 static int
2131 same_clid(clientid_t *cl1, clientid_t *cl2)
2132 {
2133         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2134 }
2135
2136 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2137 {
2138         int i;
2139
2140         if (g1->ngroups != g2->ngroups)
2141                 return false;
2142         for (i=0; i<g1->ngroups; i++)
2143                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2144                         return false;
2145         return true;
2146 }
2147
2148 /*
2149  * RFC 3530 language requires clid_inuse be returned when the
2150  * "principal" associated with a requests differs from that previously
2151  * used.  We use uid, gid's, and gss principal string as our best
2152  * approximation.  We also don't want to allow non-gss use of a client
2153  * established using gss: in theory cr_principal should catch that
2154  * change, but in practice cr_principal can be null even in the gss case
2155  * since gssd doesn't always pass down a principal string.
2156  */
2157 static bool is_gss_cred(struct svc_cred *cr)
2158 {
2159         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2160         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2161 }
2162
2163
2164 static bool
2165 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2166 {
2167         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2168                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2169                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2170                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2171                 return false;
2172         /* XXX: check that cr_targ_princ fields match ? */
2173         if (cr1->cr_principal == cr2->cr_principal)
2174                 return true;
2175         if (!cr1->cr_principal || !cr2->cr_principal)
2176                 return false;
2177         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2178 }
2179
2180 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2181 {
2182         struct svc_cred *cr = &rqstp->rq_cred;
2183         u32 service;
2184
2185         if (!cr->cr_gss_mech)
2186                 return false;
2187         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2188         return service == RPC_GSS_SVC_INTEGRITY ||
2189                service == RPC_GSS_SVC_PRIVACY;
2190 }
2191
2192 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2193 {
2194         struct svc_cred *cr = &rqstp->rq_cred;
2195
2196         if (!cl->cl_mach_cred)
2197                 return true;
2198         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2199                 return false;
2200         if (!svc_rqst_integrity_protected(rqstp))
2201                 return false;
2202         if (cl->cl_cred.cr_raw_principal)
2203                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2204                                                 cr->cr_raw_principal);
2205         if (!cr->cr_principal)
2206                 return false;
2207         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2208 }
2209
2210 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2211 {
2212         __be32 verf[2];
2213
2214         /*
2215          * This is opaque to client, so no need to byte-swap. Use
2216          * __force to keep sparse happy
2217          */
2218         verf[0] = (__force __be32)get_seconds();
2219         verf[1] = (__force __be32)nn->clverifier_counter++;
2220         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2221 }
2222
2223 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2224 {
2225         clp->cl_clientid.cl_boot = nn->boot_time;
2226         clp->cl_clientid.cl_id = nn->clientid_counter++;
2227         gen_confirm(clp, nn);
2228 }
2229
2230 static struct nfs4_stid *
2231 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2232 {
2233         struct nfs4_stid *ret;
2234
2235         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2236         if (!ret || !ret->sc_type)
2237                 return NULL;
2238         return ret;
2239 }
2240
2241 static struct nfs4_stid *
2242 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2243 {
2244         struct nfs4_stid *s;
2245
2246         spin_lock(&cl->cl_lock);
2247         s = find_stateid_locked(cl, t);
2248         if (s != NULL) {
2249                 if (typemask & s->sc_type)
2250                         refcount_inc(&s->sc_count);
2251                 else
2252                         s = NULL;
2253         }
2254         spin_unlock(&cl->cl_lock);
2255         return s;
2256 }
2257
2258 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2259 {
2260         struct nfsdfs_client *nc;
2261         nc = get_nfsdfs_client(inode);
2262         if (!nc)
2263                 return NULL;
2264         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2265 }
2266
2267 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2268 {
2269         seq_printf(m, "\"");
2270         seq_escape_mem_ascii(m, data, len);
2271         seq_printf(m, "\"");
2272 }
2273
2274 static int client_info_show(struct seq_file *m, void *v)
2275 {
2276         struct inode *inode = m->private;
2277         struct nfs4_client *clp;
2278         u64 clid;
2279
2280         clp = get_nfsdfs_clp(inode);
2281         if (!clp)
2282                 return -ENXIO;
2283         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2284         seq_printf(m, "clientid: 0x%llx\n", clid);
2285         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2286         seq_printf(m, "name: ");
2287         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2288         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2289         if (clp->cl_nii_domain.data) {
2290                 seq_printf(m, "Implementation domain: ");
2291                 seq_quote_mem(m, clp->cl_nii_domain.data,
2292                                         clp->cl_nii_domain.len);
2293                 seq_printf(m, "\nImplementation name: ");
2294                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2295                 seq_printf(m, "\nImplementation time: [%ld, %ld]\n",
2296                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2297         }
2298         drop_client(clp);
2299
2300         return 0;
2301 }
2302
2303 static int client_info_open(struct inode *inode, struct file *file)
2304 {
2305         return single_open(file, client_info_show, inode);
2306 }
2307
2308 static const struct file_operations client_info_fops = {
2309         .open           = client_info_open,
2310         .read           = seq_read,
2311         .llseek         = seq_lseek,
2312         .release        = single_release,
2313 };
2314
2315 static void *states_start(struct seq_file *s, loff_t *pos)
2316         __acquires(&clp->cl_lock)
2317 {
2318         struct nfs4_client *clp = s->private;
2319         unsigned long id = *pos;
2320         void *ret;
2321
2322         spin_lock(&clp->cl_lock);
2323         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2324         *pos = id;
2325         return ret;
2326 }
2327
2328 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2329 {
2330         struct nfs4_client *clp = s->private;
2331         unsigned long id = *pos;
2332         void *ret;
2333
2334         id = *pos;
2335         id++;
2336         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2337         *pos = id;
2338         return ret;
2339 }
2340
2341 static void states_stop(struct seq_file *s, void *v)
2342         __releases(&clp->cl_lock)
2343 {
2344         struct nfs4_client *clp = s->private;
2345
2346         spin_unlock(&clp->cl_lock);
2347 }
2348
2349 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2350 {
2351         struct inode *inode = f->nf_inode;
2352
2353         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2354                                         MAJOR(inode->i_sb->s_dev),
2355                                          MINOR(inode->i_sb->s_dev),
2356                                          inode->i_ino);
2357 }
2358
2359 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2360 {
2361         seq_printf(s, "owner: ");
2362         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2363 }
2364
2365 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2366 {
2367         struct nfs4_ol_stateid *ols;
2368         struct nfs4_file *nf;
2369         struct nfsd_file *file;
2370         struct nfs4_stateowner *oo;
2371         unsigned int access, deny;
2372
2373         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2374                 return 0; /* XXX: or SEQ_SKIP? */
2375         ols = openlockstateid(st);
2376         oo = ols->st_stateowner;
2377         nf = st->sc_file;
2378         file = find_any_file(nf);
2379
2380         seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2381
2382         access = bmap_to_share_mode(ols->st_access_bmap);
2383         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2384
2385         seq_printf(s, "access: \%s\%s, ",
2386                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2387                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2388         seq_printf(s, "deny: \%s\%s, ",
2389                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2390                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2391
2392         nfs4_show_superblock(s, file);
2393         seq_printf(s, ", ");
2394         nfs4_show_owner(s, oo);
2395         seq_printf(s, " }\n");
2396         nfsd_file_put(file);
2397
2398         return 0;
2399 }
2400
2401 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2402 {
2403         struct nfs4_ol_stateid *ols;
2404         struct nfs4_file *nf;
2405         struct nfsd_file *file;
2406         struct nfs4_stateowner *oo;
2407
2408         ols = openlockstateid(st);
2409         oo = ols->st_stateowner;
2410         nf = st->sc_file;
2411         file = find_any_file(nf);
2412
2413         seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2414
2415         /*
2416          * Note: a lock stateid isn't really the same thing as a lock,
2417          * it's the locking state held by one owner on a file, and there
2418          * may be multiple (or no) lock ranges associated with it.
2419          * (Same for the matter is true of open stateids.)
2420          */
2421
2422         nfs4_show_superblock(s, file);
2423         /* XXX: open stateid? */
2424         seq_printf(s, ", ");
2425         nfs4_show_owner(s, oo);
2426         seq_printf(s, " }\n");
2427         nfsd_file_put(file);
2428
2429         return 0;
2430 }
2431
2432 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2433 {
2434         struct nfs4_delegation *ds;
2435         struct nfs4_file *nf;
2436         struct nfsd_file *file;
2437
2438         ds = delegstateid(st);
2439         nf = st->sc_file;
2440         file = nf->fi_deleg_file;
2441
2442         seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2443
2444         /* Kinda dead code as long as we only support read delegs: */
2445         seq_printf(s, "access: %s, ",
2446                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2447
2448         /* XXX: lease time, whether it's being recalled. */
2449
2450         nfs4_show_superblock(s, file);
2451         seq_printf(s, " }\n");
2452
2453         return 0;
2454 }
2455
2456 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2457 {
2458         struct nfs4_layout_stateid *ls;
2459         struct nfsd_file *file;
2460
2461         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2462         file = ls->ls_file;
2463
2464         seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2465
2466         /* XXX: What else would be useful? */
2467
2468         nfs4_show_superblock(s, file);
2469         seq_printf(s, " }\n");
2470
2471         return 0;
2472 }
2473
2474 static int states_show(struct seq_file *s, void *v)
2475 {
2476         struct nfs4_stid *st = v;
2477
2478         switch (st->sc_type) {
2479         case NFS4_OPEN_STID:
2480                 return nfs4_show_open(s, st);
2481         case NFS4_LOCK_STID:
2482                 return nfs4_show_lock(s, st);
2483         case NFS4_DELEG_STID:
2484                 return nfs4_show_deleg(s, st);
2485         case NFS4_LAYOUT_STID:
2486                 return nfs4_show_layout(s, st);
2487         default:
2488                 return 0; /* XXX: or SEQ_SKIP? */
2489         }
2490         /* XXX: copy stateids? */
2491 }
2492
2493 static struct seq_operations states_seq_ops = {
2494         .start = states_start,
2495         .next = states_next,
2496         .stop = states_stop,
2497         .show = states_show
2498 };
2499
2500 static int client_states_open(struct inode *inode, struct file *file)
2501 {
2502         struct seq_file *s;
2503         struct nfs4_client *clp;
2504         int ret;
2505
2506         clp = get_nfsdfs_clp(inode);
2507         if (!clp)
2508                 return -ENXIO;
2509
2510         ret = seq_open(file, &states_seq_ops);
2511         if (ret)
2512                 return ret;
2513         s = file->private_data;
2514         s->private = clp;
2515         return 0;
2516 }
2517
2518 static int client_opens_release(struct inode *inode, struct file *file)
2519 {
2520         struct seq_file *m = file->private_data;
2521         struct nfs4_client *clp = m->private;
2522
2523         /* XXX: alternatively, we could get/drop in seq start/stop */
2524         drop_client(clp);
2525         return 0;
2526 }
2527
2528 static const struct file_operations client_states_fops = {
2529         .open           = client_states_open,
2530         .read           = seq_read,
2531         .llseek         = seq_lseek,
2532         .release        = client_opens_release,
2533 };
2534
2535 /*
2536  * Normally we refuse to destroy clients that are in use, but here the
2537  * administrator is telling us to just do it.  We also want to wait
2538  * so the caller has a guarantee that the client's locks are gone by
2539  * the time the write returns:
2540  */
2541 static void force_expire_client(struct nfs4_client *clp)
2542 {
2543         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2544         bool already_expired;
2545
2546         spin_lock(&clp->cl_lock);
2547         clp->cl_time = 0;
2548         spin_unlock(&clp->cl_lock);
2549
2550         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2551         spin_lock(&nn->client_lock);
2552         already_expired = list_empty(&clp->cl_lru);
2553         if (!already_expired)
2554                 unhash_client_locked(clp);
2555         spin_unlock(&nn->client_lock);
2556
2557         if (!already_expired)
2558                 expire_client(clp);
2559         else
2560                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2561 }
2562
2563 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2564                                    size_t size, loff_t *pos)
2565 {
2566         char *data;
2567         struct nfs4_client *clp;
2568
2569         data = simple_transaction_get(file, buf, size);
2570         if (IS_ERR(data))
2571                 return PTR_ERR(data);
2572         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2573                 return -EINVAL;
2574         clp = get_nfsdfs_clp(file_inode(file));
2575         if (!clp)
2576                 return -ENXIO;
2577         force_expire_client(clp);
2578         drop_client(clp);
2579         return 7;
2580 }
2581
2582 static const struct file_operations client_ctl_fops = {
2583         .write          = client_ctl_write,
2584         .release        = simple_transaction_release,
2585 };
2586
2587 static const struct tree_descr client_files[] = {
2588         [0] = {"info", &client_info_fops, S_IRUSR},
2589         [1] = {"states", &client_states_fops, S_IRUSR},
2590         [2] = {"ctl", &client_ctl_fops, S_IRUSR|S_IWUSR},
2591         [3] = {""},
2592 };
2593
2594 static struct nfs4_client *create_client(struct xdr_netobj name,
2595                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2596 {
2597         struct nfs4_client *clp;
2598         struct sockaddr *sa = svc_addr(rqstp);
2599         int ret;
2600         struct net *net = SVC_NET(rqstp);
2601         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2602
2603         clp = alloc_client(name);
2604         if (clp == NULL)
2605                 return NULL;
2606
2607         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2608         if (ret) {
2609                 free_client(clp);
2610                 return NULL;
2611         }
2612         gen_clid(clp, nn);
2613         kref_init(&clp->cl_nfsdfs.cl_ref);
2614         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2615         clp->cl_time = get_seconds();
2616         clear_bit(0, &clp->cl_cb_slot_busy);
2617         copy_verf(clp, verf);
2618         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2619         clp->cl_cb_session = NULL;
2620         clp->net = net;
2621         clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2622                         clp->cl_clientid.cl_id - nn->clientid_base,
2623                         client_files);
2624         if (!clp->cl_nfsd_dentry) {
2625                 free_client(clp);
2626                 return NULL;
2627         }
2628         return clp;
2629 }
2630
2631 static void
2632 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2633 {
2634         struct rb_node **new = &(root->rb_node), *parent = NULL;
2635         struct nfs4_client *clp;
2636
2637         while (*new) {
2638                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2639                 parent = *new;
2640
2641                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2642                         new = &((*new)->rb_left);
2643                 else
2644                         new = &((*new)->rb_right);
2645         }
2646
2647         rb_link_node(&new_clp->cl_namenode, parent, new);
2648         rb_insert_color(&new_clp->cl_namenode, root);
2649 }
2650
2651 static struct nfs4_client *
2652 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2653 {
2654         int cmp;
2655         struct rb_node *node = root->rb_node;
2656         struct nfs4_client *clp;
2657
2658         while (node) {
2659                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2660                 cmp = compare_blob(&clp->cl_name, name);
2661                 if (cmp > 0)
2662                         node = node->rb_left;
2663                 else if (cmp < 0)
2664                         node = node->rb_right;
2665                 else
2666                         return clp;
2667         }
2668         return NULL;
2669 }
2670
2671 static void
2672 add_to_unconfirmed(struct nfs4_client *clp)
2673 {
2674         unsigned int idhashval;
2675         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2676
2677         lockdep_assert_held(&nn->client_lock);
2678
2679         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2680         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2681         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2682         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2683         renew_client_locked(clp);
2684 }
2685
2686 static void
2687 move_to_confirmed(struct nfs4_client *clp)
2688 {
2689         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2690         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2691
2692         lockdep_assert_held(&nn->client_lock);
2693
2694         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2695         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2696         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2697         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2698         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2699         renew_client_locked(clp);
2700 }
2701
2702 static struct nfs4_client *
2703 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2704 {
2705         struct nfs4_client *clp;
2706         unsigned int idhashval = clientid_hashval(clid->cl_id);
2707
2708         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2709                 if (same_clid(&clp->cl_clientid, clid)) {
2710                         if ((bool)clp->cl_minorversion != sessions)
2711                                 return NULL;
2712                         renew_client_locked(clp);
2713                         return clp;
2714                 }
2715         }
2716         return NULL;
2717 }
2718
2719 static struct nfs4_client *
2720 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2721 {
2722         struct list_head *tbl = nn->conf_id_hashtbl;
2723
2724         lockdep_assert_held(&nn->client_lock);
2725         return find_client_in_id_table(tbl, clid, sessions);
2726 }
2727
2728 static struct nfs4_client *
2729 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2730 {
2731         struct list_head *tbl = nn->unconf_id_hashtbl;
2732
2733         lockdep_assert_held(&nn->client_lock);
2734         return find_client_in_id_table(tbl, clid, sessions);
2735 }
2736
2737 static bool clp_used_exchangeid(struct nfs4_client *clp)
2738 {
2739         return clp->cl_exchange_flags != 0;
2740 }
2741
2742 static struct nfs4_client *
2743 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2744 {
2745         lockdep_assert_held(&nn->client_lock);
2746         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2747 }
2748
2749 static struct nfs4_client *
2750 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2751 {
2752         lockdep_assert_held(&nn->client_lock);
2753         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2754 }
2755
2756 static void
2757 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2758 {
2759         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2760         struct sockaddr *sa = svc_addr(rqstp);
2761         u32 scopeid = rpc_get_scope_id(sa);
2762         unsigned short expected_family;
2763
2764         /* Currently, we only support tcp and tcp6 for the callback channel */
2765         if (se->se_callback_netid_len == 3 &&
2766             !memcmp(se->se_callback_netid_val, "tcp", 3))
2767                 expected_family = AF_INET;
2768         else if (se->se_callback_netid_len == 4 &&
2769                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2770                 expected_family = AF_INET6;
2771         else
2772                 goto out_err;
2773
2774         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2775                                             se->se_callback_addr_len,
2776                                             (struct sockaddr *)&conn->cb_addr,
2777                                             sizeof(conn->cb_addr));
2778
2779         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2780                 goto out_err;
2781
2782         if (conn->cb_addr.ss_family == AF_INET6)
2783                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2784
2785         conn->cb_prog = se->se_callback_prog;
2786         conn->cb_ident = se->se_callback_ident;
2787         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2788         return;
2789 out_err:
2790         conn->cb_addr.ss_family = AF_UNSPEC;
2791         conn->cb_addrlen = 0;
2792         dprintk("NFSD: this client (clientid %08x/%08x) "
2793                 "will not receive delegations\n",
2794                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2795
2796         return;
2797 }
2798
2799 /*
2800  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2801  */
2802 static void
2803 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2804 {
2805         struct xdr_buf *buf = resp->xdr.buf;
2806         struct nfsd4_slot *slot = resp->cstate.slot;
2807         unsigned int base;
2808
2809         dprintk("--> %s slot %p\n", __func__, slot);
2810
2811         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2812         slot->sl_opcnt = resp->opcnt;
2813         slot->sl_status = resp->cstate.status;
2814         free_svc_cred(&slot->sl_cred);
2815         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2816
2817         if (!nfsd4_cache_this(resp)) {
2818                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2819                 return;
2820         }
2821         slot->sl_flags |= NFSD4_SLOT_CACHED;
2822
2823         base = resp->cstate.data_offset;
2824         slot->sl_datalen = buf->len - base;
2825         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2826                 WARN(1, "%s: sessions DRC could not cache compound\n",
2827                      __func__);
2828         return;
2829 }
2830
2831 /*
2832  * Encode the replay sequence operation from the slot values.
2833  * If cachethis is FALSE encode the uncached rep error on the next
2834  * operation which sets resp->p and increments resp->opcnt for
2835  * nfs4svc_encode_compoundres.
2836  *
2837  */
2838 static __be32
2839 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2840                           struct nfsd4_compoundres *resp)
2841 {
2842         struct nfsd4_op *op;
2843         struct nfsd4_slot *slot = resp->cstate.slot;
2844
2845         /* Encode the replayed sequence operation */
2846         op = &args->ops[resp->opcnt - 1];
2847         nfsd4_encode_operation(resp, op);
2848
2849         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2850                 return op->status;
2851         if (args->opcnt == 1) {
2852                 /*
2853                  * The original operation wasn't a solo sequence--we
2854                  * always cache those--so this retry must not match the
2855                  * original:
2856                  */
2857                 op->status = nfserr_seq_false_retry;
2858         } else {
2859                 op = &args->ops[resp->opcnt++];
2860                 op->status = nfserr_retry_uncached_rep;
2861                 nfsd4_encode_operation(resp, op);
2862         }
2863         return op->status;
2864 }
2865
2866 /*
2867  * The sequence operation is not cached because we can use the slot and
2868  * session values.
2869  */
2870 static __be32
2871 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2872                          struct nfsd4_sequence *seq)
2873 {
2874         struct nfsd4_slot *slot = resp->cstate.slot;
2875         struct xdr_stream *xdr = &resp->xdr;
2876         __be32 *p;
2877         __be32 status;
2878
2879         dprintk("--> %s slot %p\n", __func__, slot);
2880
2881         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2882         if (status)
2883                 return status;
2884
2885         p = xdr_reserve_space(xdr, slot->sl_datalen);
2886         if (!p) {
2887                 WARN_ON_ONCE(1);
2888                 return nfserr_serverfault;
2889         }
2890         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2891         xdr_commit_encode(xdr);
2892
2893         resp->opcnt = slot->sl_opcnt;
2894         return slot->sl_status;
2895 }
2896
2897 /*
2898  * Set the exchange_id flags returned by the server.
2899  */
2900 static void
2901 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2902 {
2903 #ifdef CONFIG_NFSD_PNFS
2904         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2905 #else
2906         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2907 #endif
2908
2909         /* Referrals are supported, Migration is not. */
2910         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2911
2912         /* set the wire flags to return to client. */
2913         clid->flags = new->cl_exchange_flags;
2914 }
2915
2916 static bool client_has_openowners(struct nfs4_client *clp)
2917 {
2918         struct nfs4_openowner *oo;
2919
2920         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2921                 if (!list_empty(&oo->oo_owner.so_stateids))
2922                         return true;
2923         }
2924         return false;
2925 }
2926
2927 static bool client_has_state(struct nfs4_client *clp)
2928 {
2929         return client_has_openowners(clp)
2930 #ifdef CONFIG_NFSD_PNFS
2931                 || !list_empty(&clp->cl_lo_states)
2932 #endif
2933                 || !list_empty(&clp->cl_delegations)
2934                 || !list_empty(&clp->cl_sessions)
2935                 || !list_empty(&clp->async_copies);
2936 }
2937
2938 static __be32 copy_impl_id(struct nfs4_client *clp,
2939                                 struct nfsd4_exchange_id *exid)
2940 {
2941         if (!exid->nii_domain.data)
2942                 return 0;
2943         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
2944         if (!clp->cl_nii_domain.data)
2945                 return nfserr_jukebox;
2946         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
2947         if (!clp->cl_nii_name.data)
2948                 return nfserr_jukebox;
2949         clp->cl_nii_time.tv_sec = exid->nii_time.tv_sec;
2950         clp->cl_nii_time.tv_nsec = exid->nii_time.tv_nsec;
2951         return 0;
2952 }
2953
2954 __be32
2955 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2956                 union nfsd4_op_u *u)
2957 {
2958         struct nfsd4_exchange_id *exid = &u->exchange_id;
2959         struct nfs4_client *conf, *new;
2960         struct nfs4_client *unconf = NULL;
2961         __be32 status;
2962         char                    addr_str[INET6_ADDRSTRLEN];
2963         nfs4_verifier           verf = exid->verifier;
2964         struct sockaddr         *sa = svc_addr(rqstp);
2965         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2966         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2967
2968         rpc_ntop(sa, addr_str, sizeof(addr_str));
2969         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2970                 "ip_addr=%s flags %x, spa_how %d\n",
2971                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2972                 addr_str, exid->flags, exid->spa_how);
2973
2974         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2975                 return nfserr_inval;
2976
2977         new = create_client(exid->clname, rqstp, &verf);
2978         if (new == NULL)
2979                 return nfserr_jukebox;
2980         status = copy_impl_id(new, exid);
2981         if (status)
2982                 goto out_nolock;
2983
2984         switch (exid->spa_how) {
2985         case SP4_MACH_CRED:
2986                 exid->spo_must_enforce[0] = 0;
2987                 exid->spo_must_enforce[1] = (
2988                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2989                         1 << (OP_EXCHANGE_ID - 32) |
2990                         1 << (OP_CREATE_SESSION - 32) |
2991                         1 << (OP_DESTROY_SESSION - 32) |
2992                         1 << (OP_DESTROY_CLIENTID - 32));
2993
2994                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2995                                         1 << (OP_OPEN_DOWNGRADE) |
2996                                         1 << (OP_LOCKU) |
2997                                         1 << (OP_DELEGRETURN));
2998
2999                 exid->spo_must_allow[1] &= (
3000                                         1 << (OP_TEST_STATEID - 32) |
3001                                         1 << (OP_FREE_STATEID - 32));
3002                 if (!svc_rqst_integrity_protected(rqstp)) {
3003                         status = nfserr_inval;
3004                         goto out_nolock;
3005                 }
3006                 /*
3007                  * Sometimes userspace doesn't give us a principal.
3008                  * Which is a bug, really.  Anyway, we can't enforce
3009                  * MACH_CRED in that case, better to give up now:
3010                  */
3011                 if (!new->cl_cred.cr_principal &&
3012                                         !new->cl_cred.cr_raw_principal) {
3013                         status = nfserr_serverfault;
3014                         goto out_nolock;
3015                 }
3016                 new->cl_mach_cred = true;
3017         case SP4_NONE:
3018                 break;
3019         default:                                /* checked by xdr code */
3020                 WARN_ON_ONCE(1);
3021                 /* fall through */
3022         case SP4_SSV:
3023                 status = nfserr_encr_alg_unsupp;
3024                 goto out_nolock;
3025         }
3026
3027         /* Cases below refer to rfc 5661 section 18.35.4: */
3028         spin_lock(&nn->client_lock);
3029         conf = find_confirmed_client_by_name(&exid->clname, nn);
3030         if (conf) {
3031                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3032                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3033
3034                 if (update) {
3035                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3036                                 status = nfserr_inval;
3037                                 goto out;
3038                         }
3039                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3040                                 status = nfserr_wrong_cred;
3041                                 goto out;
3042                         }
3043                         if (!creds_match) { /* case 9 */
3044                                 status = nfserr_perm;
3045                                 goto out;
3046                         }
3047                         if (!verfs_match) { /* case 8 */
3048                                 status = nfserr_not_same;
3049                                 goto out;
3050                         }
3051                         /* case 6 */
3052                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3053                         goto out_copy;
3054                 }
3055                 if (!creds_match) { /* case 3 */
3056                         if (client_has_state(conf)) {
3057                                 status = nfserr_clid_inuse;
3058                                 goto out;
3059                         }
3060                         goto out_new;
3061                 }
3062                 if (verfs_match) { /* case 2 */
3063                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3064                         goto out_copy;
3065                 }
3066                 /* case 5, client reboot */
3067                 conf = NULL;
3068                 goto out_new;
3069         }
3070
3071         if (update) { /* case 7 */
3072                 status = nfserr_noent;
3073                 goto out;
3074         }
3075
3076         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
3077         if (unconf) /* case 4, possible retry or client restart */
3078                 unhash_client_locked(unconf);
3079
3080         /* case 1 (normal case) */
3081 out_new:
3082         if (conf) {
3083                 status = mark_client_expired_locked(conf);
3084                 if (status)
3085                         goto out;
3086         }
3087         new->cl_minorversion = cstate->minorversion;
3088         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3089         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3090
3091         add_to_unconfirmed(new);
3092         swap(new, conf);
3093 out_copy:
3094         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3095         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3096
3097         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3098         nfsd4_set_ex_flags(conf, exid);
3099
3100         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3101                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3102         status = nfs_ok;
3103
3104 out:
3105         spin_unlock(&nn->client_lock);
3106 out_nolock:
3107         if (new)
3108                 expire_client(new);
3109         if (unconf)
3110                 expire_client(unconf);
3111         return status;
3112 }
3113
3114 static __be32
3115 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3116 {
3117         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3118                 slot_seqid);
3119
3120         /* The slot is in use, and no response has been sent. */
3121         if (slot_inuse) {
3122                 if (seqid == slot_seqid)
3123                         return nfserr_jukebox;
3124                 else
3125                         return nfserr_seq_misordered;
3126         }
3127         /* Note unsigned 32-bit arithmetic handles wraparound: */
3128         if (likely(seqid == slot_seqid + 1))
3129                 return nfs_ok;
3130         if (seqid == slot_seqid)
3131                 return nfserr_replay_cache;
3132         return nfserr_seq_misordered;
3133 }
3134
3135 /*
3136  * Cache the create session result into the create session single DRC
3137  * slot cache by saving the xdr structure. sl_seqid has been set.
3138  * Do this for solo or embedded create session operations.
3139  */
3140 static void
3141 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3142                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3143 {
3144         slot->sl_status = nfserr;
3145         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3146 }
3147
3148 static __be32
3149 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3150                             struct nfsd4_clid_slot *slot)
3151 {
3152         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3153         return slot->sl_status;
3154 }
3155
3156 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3157                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3158                         1 +     /* MIN tag is length with zero, only length */ \
3159                         3 +     /* version, opcount, opcode */ \
3160                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3161                                 /* seqid, slotID, slotID, cache */ \
3162                         4 ) * sizeof(__be32))
3163
3164 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3165                         2 +     /* verifier: AUTH_NULL, length 0 */\
3166                         1 +     /* status */ \
3167                         1 +     /* MIN tag is length with zero, only length */ \
3168                         3 +     /* opcount, opcode, opstatus*/ \
3169                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3170                                 /* seqid, slotID, slotID, slotID, status */ \
3171                         5 ) * sizeof(__be32))
3172
3173 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3174 {
3175         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3176
3177         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3178                 return nfserr_toosmall;
3179         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3180                 return nfserr_toosmall;
3181         ca->headerpadsz = 0;
3182         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3183         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3184         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3185         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3186                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3187         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3188         /*
3189          * Note decreasing slot size below client's request may make it
3190          * difficult for client to function correctly, whereas
3191          * decreasing the number of slots will (just?) affect
3192          * performance.  When short on memory we therefore prefer to
3193          * decrease number of slots instead of their size.  Clients that
3194          * request larger slots than they need will get poor results:
3195          * Note that we always allow at least one slot, because our
3196          * accounting is soft and provides no guarantees either way.
3197          */
3198         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3199
3200         return nfs_ok;
3201 }
3202
3203 /*
3204  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3205  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3206  */
3207 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3208         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3209
3210 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3211         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3212
3213 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3214                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3215 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3216                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3217                                  sizeof(__be32))
3218
3219 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3220 {
3221         ca->headerpadsz = 0;
3222
3223         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3224                 return nfserr_toosmall;
3225         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3226                 return nfserr_toosmall;
3227         ca->maxresp_cached = 0;
3228         if (ca->maxops < 2)
3229                 return nfserr_toosmall;
3230
3231         return nfs_ok;
3232 }
3233
3234 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3235 {
3236         switch (cbs->flavor) {
3237         case RPC_AUTH_NULL:
3238         case RPC_AUTH_UNIX:
3239                 return nfs_ok;
3240         default:
3241                 /*
3242                  * GSS case: the spec doesn't allow us to return this
3243                  * error.  But it also doesn't allow us not to support
3244                  * GSS.
3245                  * I'd rather this fail hard than return some error the
3246                  * client might think it can already handle:
3247                  */
3248                 return nfserr_encr_alg_unsupp;
3249         }
3250 }
3251
3252 __be32
3253 nfsd4_create_session(struct svc_rqst *rqstp,
3254                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3255 {
3256         struct nfsd4_create_session *cr_ses = &u->create_session;
3257         struct sockaddr *sa = svc_addr(rqstp);
3258         struct nfs4_client *conf, *unconf;
3259         struct nfs4_client *old = NULL;
3260         struct nfsd4_session *new;
3261         struct nfsd4_conn *conn;
3262         struct nfsd4_clid_slot *cs_slot = NULL;
3263         __be32 status = 0;
3264         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3265
3266         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3267                 return nfserr_inval;
3268         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3269         if (status)
3270                 return status;
3271         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3272         if (status)
3273                 return status;
3274         status = check_backchannel_attrs(&cr_ses->back_channel);
3275         if (status)
3276                 goto out_release_drc_mem;
3277         status = nfserr_jukebox;
3278         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3279         if (!new)
3280                 goto out_release_drc_mem;
3281         conn = alloc_conn_from_crses(rqstp, cr_ses);
3282         if (!conn)
3283                 goto out_free_session;
3284
3285         spin_lock(&nn->client_lock);
3286         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3287         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3288         WARN_ON_ONCE(conf && unconf);
3289
3290         if (conf) {
3291                 status = nfserr_wrong_cred;
3292                 if (!nfsd4_mach_creds_match(conf, rqstp))
3293                         goto out_free_conn;
3294                 cs_slot = &conf->cl_cs_slot;
3295                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3296                 if (status) {
3297                         if (status == nfserr_replay_cache)
3298                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3299                         goto out_free_conn;
3300                 }
3301         } else if (unconf) {
3302                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3303                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3304                         status = nfserr_clid_inuse;
3305                         goto out_free_conn;
3306                 }
3307                 status = nfserr_wrong_cred;
3308                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3309                         goto out_free_conn;
3310                 cs_slot = &unconf->cl_cs_slot;
3311                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3312                 if (status) {
3313                         /* an unconfirmed replay returns misordered */
3314                         status = nfserr_seq_misordered;
3315                         goto out_free_conn;
3316                 }
3317                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3318                 if (old) {
3319                         status = mark_client_expired_locked(old);
3320                         if (status) {
3321                                 old = NULL;
3322                                 goto out_free_conn;
3323                         }
3324                 }
3325                 move_to_confirmed(unconf);
3326                 conf = unconf;
3327         } else {
3328                 status = nfserr_stale_clientid;
3329                 goto out_free_conn;
3330         }
3331         status = nfs_ok;
3332         /* Persistent sessions are not supported */
3333         cr_ses->flags &= ~SESSION4_PERSIST;
3334         /* Upshifting from TCP to RDMA is not supported */
3335         cr_ses->flags &= ~SESSION4_RDMA;
3336
3337         init_session(rqstp, new, conf, cr_ses);
3338         nfsd4_get_session_locked(new);
3339
3340         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3341                NFS4_MAX_SESSIONID_LEN);
3342         cs_slot->sl_seqid++;
3343         cr_ses->seqid = cs_slot->sl_seqid;
3344
3345         /* cache solo and embedded create sessions under the client_lock */
3346         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3347         spin_unlock(&nn->client_lock);
3348         /* init connection and backchannel */
3349         nfsd4_init_conn(rqstp, conn, new);
3350         nfsd4_put_session(new);
3351         if (old)
3352                 expire_client(old);
3353         return status;
3354 out_free_conn:
3355         spin_unlock(&nn->client_lock);
3356         free_conn(conn);
3357         if (old)
3358                 expire_client(old);
3359 out_free_session:
3360         __free_session(new);
3361 out_release_drc_mem:
3362         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3363         return status;
3364 }
3365
3366 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3367 {
3368         switch (*dir) {
3369         case NFS4_CDFC4_FORE:
3370         case NFS4_CDFC4_BACK:
3371                 return nfs_ok;
3372         case NFS4_CDFC4_FORE_OR_BOTH:
3373         case NFS4_CDFC4_BACK_OR_BOTH:
3374                 *dir = NFS4_CDFC4_BOTH;
3375                 return nfs_ok;
3376         };
3377         return nfserr_inval;
3378 }
3379
3380 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3381                 struct nfsd4_compound_state *cstate,
3382                 union nfsd4_op_u *u)
3383 {
3384         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3385         struct nfsd4_session *session = cstate->session;
3386         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3387         __be32 status;
3388
3389         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3390         if (status)
3391                 return status;
3392         spin_lock(&nn->client_lock);
3393         session->se_cb_prog = bc->bc_cb_program;
3394         session->se_cb_sec = bc->bc_cb_sec;
3395         spin_unlock(&nn->client_lock);
3396
3397         nfsd4_probe_callback(session->se_client);
3398
3399         return nfs_ok;
3400 }
3401
3402 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3403                      struct nfsd4_compound_state *cstate,
3404                      union nfsd4_op_u *u)
3405 {
3406         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3407         __be32 status;
3408         struct nfsd4_conn *conn;
3409         struct nfsd4_session *session;
3410         struct net *net = SVC_NET(rqstp);
3411         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3412
3413         if (!nfsd4_last_compound_op(rqstp))
3414                 return nfserr_not_only_op;
3415         spin_lock(&nn->client_lock);
3416         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3417         spin_unlock(&nn->client_lock);
3418         if (!session)
3419                 goto out_no_session;
3420         status = nfserr_wrong_cred;
3421         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3422                 goto out;
3423         status = nfsd4_map_bcts_dir(&bcts->dir);
3424         if (status)
3425                 goto out;
3426         conn = alloc_conn(rqstp, bcts->dir);
3427         status = nfserr_jukebox;
3428         if (!conn)
3429                 goto out;
3430         nfsd4_init_conn(rqstp, conn, session);
3431         status = nfs_ok;
3432 out:
3433         nfsd4_put_session(session);
3434 out_no_session:
3435         return status;
3436 }
3437
3438 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3439 {
3440         if (!cstate->session)
3441                 return false;
3442         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3443 }
3444
3445 __be32
3446 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3447                 union nfsd4_op_u *u)
3448 {
3449         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3450         struct nfsd4_session *ses;
3451         __be32 status;
3452         int ref_held_by_me = 0;
3453         struct net *net = SVC_NET(r);
3454         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3455
3456         status = nfserr_not_only_op;
3457         if (nfsd4_compound_in_session(cstate, sessionid)) {
3458                 if (!nfsd4_last_compound_op(r))
3459                         goto out;
3460                 ref_held_by_me++;
3461         }
3462         dump_sessionid(__func__, sessionid);
3463         spin_lock(&nn->client_lock);
3464         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3465         if (!ses)
3466                 goto out_client_lock;
3467         status = nfserr_wrong_cred;
3468         if (!nfsd4_mach_creds_match(ses->se_client, r))
3469                 goto out_put_session;
3470         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3471         if (status)
3472                 goto out_put_session;
3473         unhash_session(ses);
3474         spin_unlock(&nn->client_lock);
3475
3476         nfsd4_probe_callback_sync(ses->se_client);
3477
3478         spin_lock(&nn->client_lock);
3479         status = nfs_ok;
3480 out_put_session:
3481         nfsd4_put_session_locked(ses);
3482 out_client_lock:
3483         spin_unlock(&nn->client_lock);
3484 out:
3485         return status;
3486 }
3487
3488 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3489 {
3490         struct nfsd4_conn *c;
3491
3492         list_for_each_entry(c, &s->se_conns, cn_persession) {
3493                 if (c->cn_xprt == xpt) {
3494                         return c;
3495                 }
3496         }
3497         return NULL;
3498 }
3499
3500 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3501 {
3502         struct nfs4_client *clp = ses->se_client;
3503         struct nfsd4_conn *c;
3504         __be32 status = nfs_ok;
3505         int ret;
3506
3507         spin_lock(&clp->cl_lock);
3508         c = __nfsd4_find_conn(new->cn_xprt, ses);
3509         if (c)
3510                 goto out_free;
3511         status = nfserr_conn_not_bound_to_session;
3512         if (clp->cl_mach_cred)
3513                 goto out_free;
3514         __nfsd4_hash_conn(new, ses);
3515         spin_unlock(&clp->cl_lock);
3516         ret = nfsd4_register_conn(new);
3517         if (ret)
3518                 /* oops; xprt is already down: */
3519                 nfsd4_conn_lost(&new->cn_xpt_user);
3520         return nfs_ok;
3521 out_free:
3522         spin_unlock(&clp->cl_lock);
3523         free_conn(new);
3524         return status;
3525 }
3526
3527 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3528 {
3529         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3530
3531         return args->opcnt > session->se_fchannel.maxops;
3532 }
3533
3534 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3535                                   struct nfsd4_session *session)
3536 {
3537         struct xdr_buf *xb = &rqstp->rq_arg;
3538
3539         return xb->len > session->se_fchannel.maxreq_sz;
3540 }
3541
3542 static bool replay_matches_cache(struct svc_rqst *rqstp,
3543                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3544 {
3545         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3546
3547         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3548             (bool)seq->cachethis)
3549                 return false;
3550         /*
3551          * If there's an error than the reply can have fewer ops than
3552          * the call.  But if we cached a reply with *more* ops than the
3553          * call you're sending us now, then this new call is clearly not
3554          * really a replay of the old one:
3555          */
3556         if (slot->sl_opcnt < argp->opcnt)
3557                 return false;
3558         /* This is the only check explicitly called by spec: */
3559         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3560                 return false;
3561         /*
3562          * There may be more comparisons we could actually do, but the
3563          * spec doesn't require us to catch every case where the calls
3564          * don't match (that would require caching the call as well as
3565          * the reply), so we don't bother.
3566          */
3567         return true;
3568 }
3569
3570 __be32
3571 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3572                 union nfsd4_op_u *u)
3573 {
3574         struct nfsd4_sequence *seq = &u->sequence;
3575         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3576         struct xdr_stream *xdr = &resp->xdr;
3577         struct nfsd4_session *session;
3578         struct nfs4_client *clp;
3579         struct nfsd4_slot *slot;
3580         struct nfsd4_conn *conn;
3581         __be32 status;
3582         int buflen;
3583         struct net *net = SVC_NET(rqstp);
3584         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3585
3586         if (resp->opcnt != 1)
3587                 return nfserr_sequence_pos;
3588
3589         /*
3590          * Will be either used or freed by nfsd4_sequence_check_conn
3591          * below.
3592          */
3593         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3594         if (!conn)
3595                 return nfserr_jukebox;
3596
3597         spin_lock(&nn->client_lock);
3598         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3599         if (!session)
3600                 goto out_no_session;
3601         clp = session->se_client;
3602
3603         status = nfserr_too_many_ops;
3604         if (nfsd4_session_too_many_ops(rqstp, session))
3605                 goto out_put_session;
3606
3607         status = nfserr_req_too_big;
3608         if (nfsd4_request_too_big(rqstp, session))
3609                 goto out_put_session;
3610
3611         status = nfserr_badslot;
3612         if (seq->slotid >= session->se_fchannel.maxreqs)
3613                 goto out_put_session;
3614
3615         slot = session->se_slots[seq->slotid];
3616         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3617
3618         /* We do not negotiate the number of slots yet, so set the
3619          * maxslots to the session maxreqs which is used to encode
3620          * sr_highest_slotid and the sr_target_slot id to maxslots */
3621         seq->maxslots = session->se_fchannel.maxreqs;
3622
3623         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3624                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3625         if (status == nfserr_replay_cache) {
3626                 status = nfserr_seq_misordered;
3627                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3628                         goto out_put_session;
3629                 status = nfserr_seq_false_retry;
3630                 if (!replay_matches_cache(rqstp, seq, slot))
3631                         goto out_put_session;
3632                 cstate->slot = slot;
3633                 cstate->session = session;
3634                 cstate->clp = clp;
3635                 /* Return the cached reply status and set cstate->status
3636                  * for nfsd4_proc_compound processing */
3637                 status = nfsd4_replay_cache_entry(resp, seq);
3638                 cstate->status = nfserr_replay_cache;
3639                 goto out;
3640         }
3641         if (status)
3642                 goto out_put_session;
3643
3644         status = nfsd4_sequence_check_conn(conn, session);
3645         conn = NULL;
3646         if (status)
3647                 goto out_put_session;
3648
3649         buflen = (seq->cachethis) ?
3650                         session->se_fchannel.maxresp_cached :
3651                         session->se_fchannel.maxresp_sz;
3652         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3653                                     nfserr_rep_too_big;
3654         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3655                 goto out_put_session;
3656         svc_reserve(rqstp, buflen);
3657
3658         status = nfs_ok;
3659         /* Success! bump slot seqid */
3660         slot->sl_seqid = seq->seqid;
3661         slot->sl_flags |= NFSD4_SLOT_INUSE;
3662         if (seq->cachethis)
3663                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3664         else
3665                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3666
3667         cstate->slot = slot;
3668         cstate->session = session;
3669         cstate->clp = clp;
3670
3671 out:
3672         switch (clp->cl_cb_state) {
3673         case NFSD4_CB_DOWN:
3674                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3675                 break;
3676         case NFSD4_CB_FAULT:
3677                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3678                 break;
3679         default:
3680                 seq->status_flags = 0;
3681         }
3682         if (!list_empty(&clp->cl_revoked))
3683                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3684 out_no_session:
3685         if (conn)
3686                 free_conn(conn);
3687         spin_unlock(&nn->client_lock);
3688         return status;
3689 out_put_session:
3690         nfsd4_put_session_locked(session);
3691         goto out_no_session;
3692 }
3693
3694 void
3695 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3696 {
3697         struct nfsd4_compound_state *cs = &resp->cstate;
3698
3699         if (nfsd4_has_session(cs)) {
3700                 if (cs->status != nfserr_replay_cache) {
3701                         nfsd4_store_cache_entry(resp);
3702                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3703                 }
3704                 /* Drop session reference that was taken in nfsd4_sequence() */
3705                 nfsd4_put_session(cs->session);
3706         } else if (cs->clp)
3707                 put_client_renew(cs->clp);
3708 }
3709
3710 __be32
3711 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3712                 struct nfsd4_compound_state *cstate,
3713                 union nfsd4_op_u *u)
3714 {
3715         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3716         struct nfs4_client *conf, *unconf;
3717         struct nfs4_client *clp = NULL;
3718         __be32 status = 0;
3719         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3720
3721         spin_lock(&nn->client_lock);
3722         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3723         conf = find_confirmed_client(&dc->clientid, true, nn);
3724         WARN_ON_ONCE(conf && unconf);
3725
3726         if (conf) {
3727                 if (client_has_state(conf)) {
3728                         status = nfserr_clientid_busy;
3729                         goto out;
3730                 }
3731                 status = mark_client_expired_locked(conf);
3732                 if (status)
3733                         goto out;
3734                 clp = conf;
3735         } else if (unconf)
3736                 clp = unconf;
3737         else {
3738                 status = nfserr_stale_clientid;
3739                 goto out;
3740         }
3741         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3742                 clp = NULL;
3743                 status = nfserr_wrong_cred;
3744                 goto out;
3745         }
3746         unhash_client_locked(clp);
3747 out:
3748         spin_unlock(&nn->client_lock);
3749         if (clp)
3750                 expire_client(clp);
3751         return status;
3752 }
3753
3754 __be32
3755 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3756                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3757 {
3758         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3759         __be32 status = 0;
3760
3761         if (rc->rca_one_fs) {
3762                 if (!cstate->current_fh.fh_dentry)
3763                         return nfserr_nofilehandle;
3764                 /*
3765                  * We don't take advantage of the rca_one_fs case.
3766                  * That's OK, it's optional, we can safely ignore it.
3767                  */
3768                 return nfs_ok;
3769         }
3770
3771         status = nfserr_complete_already;
3772         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3773                              &cstate->session->se_client->cl_flags))
3774                 goto out;
3775
3776         status = nfserr_stale_clientid;
3777         if (is_client_expired(cstate->session->se_client))
3778                 /*
3779                  * The following error isn't really legal.
3780                  * But we only get here if the client just explicitly
3781                  * destroyed the client.  Surely it no longer cares what
3782                  * error it gets back on an operation for the dead
3783                  * client.
3784                  */
3785                 goto out;
3786
3787         status = nfs_ok;
3788         nfsd4_client_record_create(cstate->session->se_client);
3789         inc_reclaim_complete(cstate->session->se_client);
3790 out:
3791         return status;
3792 }
3793
3794 __be32
3795 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3796                   union nfsd4_op_u *u)
3797 {
3798         struct nfsd4_setclientid *setclid = &u->setclientid;
3799         struct xdr_netobj       clname = setclid->se_name;
3800         nfs4_verifier           clverifier = setclid->se_verf;
3801         struct nfs4_client      *conf, *new;
3802         struct nfs4_client      *unconf = NULL;
3803         __be32                  status;
3804         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3805
3806         new = create_client(clname, rqstp, &clverifier);
3807         if (new == NULL)
3808                 return nfserr_jukebox;
3809         /* Cases below refer to rfc 3530 section 14.2.33: */
3810         spin_lock(&nn->client_lock);
3811         conf = find_confirmed_client_by_name(&clname, nn);
3812         if (conf && client_has_state(conf)) {
3813                 /* case 0: */
3814                 status = nfserr_clid_inuse;
3815                 if (clp_used_exchangeid(conf))
3816                         goto out;
3817                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3818                         char addr_str[INET6_ADDRSTRLEN];
3819                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3820                                  sizeof(addr_str));
3821                         dprintk("NFSD: setclientid: string in use by client "
3822                                 "at %s\n", addr_str);
3823                         goto out;
3824                 }
3825         }
3826         unconf = find_unconfirmed_client_by_name(&clname, nn);
3827         if (unconf)
3828                 unhash_client_locked(unconf);
3829         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3830                 /* case 1: probable callback update */
3831                 copy_clid(new, conf);
3832                 gen_confirm(new, nn);
3833         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3834                 ;
3835         new->cl_minorversion = 0;
3836         gen_callback(new, setclid, rqstp);
3837         add_to_unconfirmed(new);
3838         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3839         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3840         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3841         new = NULL;
3842         status = nfs_ok;
3843 out:
3844         spin_unlock(&nn->client_lock);
3845         if (new)
3846                 free_client(new);
3847         if (unconf)
3848                 expire_client(unconf);
3849         return status;
3850 }
3851
3852
3853 __be32
3854 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3855                         struct nfsd4_compound_state *cstate,
3856                         union nfsd4_op_u *u)
3857 {
3858         struct nfsd4_setclientid_confirm *setclientid_confirm =
3859                         &u->setclientid_confirm;
3860         struct nfs4_client *conf, *unconf;
3861         struct nfs4_client *old = NULL;
3862         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3863         clientid_t * clid = &setclientid_confirm->sc_clientid;
3864         __be32 status;
3865         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3866
3867         if (STALE_CLIENTID(clid, nn))
3868                 return nfserr_stale_clientid;
3869
3870         spin_lock(&nn->client_lock);
3871         conf = find_confirmed_client(clid, false, nn);
3872         unconf = find_unconfirmed_client(clid, false, nn);
3873         /*
3874          * We try hard to give out unique clientid's, so if we get an
3875          * attempt to confirm the same clientid with a different cred,
3876          * the client may be buggy; this should never happen.
3877          *
3878          * Nevertheless, RFC 7530 recommends INUSE for this case:
3879          */
3880         status = nfserr_clid_inuse;
3881         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3882                 goto out;
3883         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3884                 goto out;
3885         /* cases below refer to rfc 3530 section 14.2.34: */
3886         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3887                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3888                         /* case 2: probable retransmit */
3889                         status = nfs_ok;
3890                 } else /* case 4: client hasn't noticed we rebooted yet? */
3891                         status = nfserr_stale_clientid;
3892                 goto out;
3893         }
3894         status = nfs_ok;
3895         if (conf) { /* case 1: callback update */
3896                 old = unconf;
3897                 unhash_client_locked(old);
3898                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3899         } else { /* case 3: normal case; new or rebooted client */
3900                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3901                 if (old) {
3902                         status = nfserr_clid_inuse;
3903                         if (client_has_state(old)
3904                                         && !same_creds(&unconf->cl_cred,
3905                                                         &old->cl_cred))
3906                                 goto out;
3907                         status = mark_client_expired_locked(old);
3908                         if (status) {
3909                                 old = NULL;
3910                                 goto out;
3911                         }
3912                 }
3913                 move_to_confirmed(unconf);
3914                 conf = unconf;
3915         }
3916         get_client_locked(conf);
3917         spin_unlock(&nn->client_lock);
3918         nfsd4_probe_callback(conf);
3919         spin_lock(&nn->client_lock);
3920         put_client_renew_locked(conf);
3921 out:
3922         spin_unlock(&nn->client_lock);
3923         if (old)
3924                 expire_client(old);
3925         return status;
3926 }
3927
3928 static struct nfs4_file *nfsd4_alloc_file(void)
3929 {
3930         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3931 }
3932
3933 /* OPEN Share state helper functions */
3934 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3935                                 struct nfs4_file *fp)
3936 {
3937         lockdep_assert_held(&state_lock);
3938
3939         refcount_set(&fp->fi_ref, 1);
3940         spin_lock_init(&fp->fi_lock);
3941         INIT_LIST_HEAD(&fp->fi_stateids);
3942         INIT_LIST_HEAD(&fp->fi_delegations);
3943         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3944         fh_copy_shallow(&fp->fi_fhandle, fh);
3945         fp->fi_deleg_file = NULL;
3946         fp->fi_had_conflict = false;
3947         fp->fi_share_deny = 0;
3948         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3949         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3950 #ifdef CONFIG_NFSD_PNFS
3951         INIT_LIST_HEAD(&fp->fi_lo_states);
3952         atomic_set(&fp->fi_lo_recalls, 0);
3953 #endif
3954         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3955 }
3956
3957 void
3958 nfsd4_free_slabs(void)
3959 {
3960         kmem_cache_destroy(client_slab);
3961         kmem_cache_destroy(openowner_slab);
3962         kmem_cache_destroy(lockowner_slab);
3963         kmem_cache_destroy(file_slab);
3964         kmem_cache_destroy(stateid_slab);
3965         kmem_cache_destroy(deleg_slab);
3966         kmem_cache_destroy(odstate_slab);
3967 }
3968
3969 int
3970 nfsd4_init_slabs(void)
3971 {
3972         client_slab = kmem_cache_create("nfsd4_clients",
3973                         sizeof(struct nfs4_client), 0, 0, NULL);
3974         if (client_slab == NULL)
3975                 goto out;
3976         openowner_slab = kmem_cache_create("nfsd4_openowners",
3977                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3978         if (openowner_slab == NULL)
3979                 goto out_free_client_slab;
3980         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3981                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3982         if (lockowner_slab == NULL)
3983                 goto out_free_openowner_slab;
3984         file_slab = kmem_cache_create("nfsd4_files",
3985                         sizeof(struct nfs4_file), 0, 0, NULL);
3986         if (file_slab == NULL)
3987                 goto out_free_lockowner_slab;
3988         stateid_slab = kmem_cache_create("nfsd4_stateids",
3989                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3990         if (stateid_slab == NULL)
3991                 goto out_free_file_slab;
3992         deleg_slab = kmem_cache_create("nfsd4_delegations",
3993                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3994         if (deleg_slab == NULL)
3995                 goto out_free_stateid_slab;
3996         odstate_slab = kmem_cache_create("nfsd4_odstate",
3997                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3998         if (odstate_slab == NULL)
3999                 goto out_free_deleg_slab;
4000         return 0;
4001
4002 out_free_deleg_slab:
4003         kmem_cache_destroy(deleg_slab);
4004 out_free_stateid_slab:
4005         kmem_cache_destroy(stateid_slab);
4006 out_free_file_slab:
4007         kmem_cache_destroy(file_slab);
4008 out_free_lockowner_slab:
4009         kmem_cache_destroy(lockowner_slab);
4010 out_free_openowner_slab:
4011         kmem_cache_destroy(openowner_slab);
4012 out_free_client_slab:
4013         kmem_cache_destroy(client_slab);
4014 out:
4015         dprintk("nfsd4: out of memory while initializing nfsv4\n");
4016         return -ENOMEM;
4017 }
4018
4019 static void init_nfs4_replay(struct nfs4_replay *rp)
4020 {
4021         rp->rp_status = nfserr_serverfault;
4022         rp->rp_buflen = 0;
4023         rp->rp_buf = rp->rp_ibuf;
4024         mutex_init(&rp->rp_mutex);
4025 }
4026
4027 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4028                 struct nfs4_stateowner *so)
4029 {
4030         if (!nfsd4_has_session(cstate)) {
4031                 mutex_lock(&so->so_replay.rp_mutex);
4032                 cstate->replay_owner = nfs4_get_stateowner(so);
4033         }
4034 }
4035
4036 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4037 {
4038         struct nfs4_stateowner *so = cstate->replay_owner;
4039
4040         if (so != NULL) {
4041                 cstate->replay_owner = NULL;
4042                 mutex_unlock(&so->so_replay.rp_mutex);
4043                 nfs4_put_stateowner(so);
4044         }
4045 }
4046
4047 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4048 {
4049         struct nfs4_stateowner *sop;
4050
4051         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4052         if (!sop)
4053                 return NULL;
4054
4055         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4056         if (!sop->so_owner.data) {
4057                 kmem_cache_free(slab, sop);
4058                 return NULL;
4059         }
4060
4061         INIT_LIST_HEAD(&sop->so_stateids);
4062         sop->so_client = clp;
4063         init_nfs4_replay(&sop->so_replay);
4064         atomic_set(&sop->so_count, 1);
4065         return sop;
4066 }
4067
4068 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4069 {
4070         lockdep_assert_held(&clp->cl_lock);
4071
4072         list_add(&oo->oo_owner.so_strhash,
4073                  &clp->cl_ownerstr_hashtbl[strhashval]);
4074         list_add(&oo->oo_perclient, &clp->cl_openowners);
4075 }
4076
4077 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4078 {
4079         unhash_openowner_locked(openowner(so));
4080 }
4081
4082 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4083 {
4084         struct nfs4_openowner *oo = openowner(so);
4085
4086         kmem_cache_free(openowner_slab, oo);
4087 }
4088
4089 static const struct nfs4_stateowner_operations openowner_ops = {
4090         .so_unhash =    nfs4_unhash_openowner,
4091         .so_free =      nfs4_free_openowner,
4092 };
4093
4094 static struct nfs4_ol_stateid *
4095 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4096 {
4097         struct nfs4_ol_stateid *local, *ret = NULL;
4098         struct nfs4_openowner *oo = open->op_openowner;
4099
4100         lockdep_assert_held(&fp->fi_lock);
4101
4102         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4103                 /* ignore lock owners */
4104                 if (local->st_stateowner->so_is_open_owner == 0)
4105                         continue;
4106                 if (local->st_stateowner != &oo->oo_owner)
4107                         continue;
4108                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4109                         ret = local;
4110                         refcount_inc(&ret->st_stid.sc_count);
4111                         break;
4112                 }
4113         }
4114         return ret;
4115 }
4116
4117 static __be32
4118 nfsd4_verify_open_stid(struct nfs4_stid *s)
4119 {
4120         __be32 ret = nfs_ok;
4121
4122         switch (s->sc_type) {
4123         default:
4124                 break;
4125         case 0:
4126         case NFS4_CLOSED_STID:
4127         case NFS4_CLOSED_DELEG_STID:
4128                 ret = nfserr_bad_stateid;
4129                 break;
4130         case NFS4_REVOKED_DELEG_STID:
4131                 ret = nfserr_deleg_revoked;
4132         }
4133         return ret;
4134 }
4135
4136 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4137 static __be32
4138 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4139 {
4140         __be32 ret;
4141
4142         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4143         ret = nfsd4_verify_open_stid(&stp->st_stid);
4144         if (ret != nfs_ok)
4145                 mutex_unlock(&stp->st_mutex);
4146         return ret;
4147 }
4148
4149 static struct nfs4_ol_stateid *
4150 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4151 {
4152         struct nfs4_ol_stateid *stp;
4153         for (;;) {
4154                 spin_lock(&fp->fi_lock);
4155                 stp = nfsd4_find_existing_open(fp, open);
4156                 spin_unlock(&fp->fi_lock);
4157                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4158                         break;
4159                 nfs4_put_stid(&stp->st_stid);
4160         }
4161         return stp;
4162 }
4163
4164 static struct nfs4_openowner *
4165 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4166                            struct nfsd4_compound_state *cstate)
4167 {
4168         struct nfs4_client *clp = cstate->clp;
4169         struct nfs4_openowner *oo, *ret;
4170
4171         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4172         if (!oo)
4173                 return NULL;
4174         oo->oo_owner.so_ops = &openowner_ops;
4175         oo->oo_owner.so_is_open_owner = 1;
4176         oo->oo_owner.so_seqid = open->op_seqid;
4177         oo->oo_flags = 0;
4178         if (nfsd4_has_session(cstate))
4179                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4180         oo->oo_time = 0;
4181         oo->oo_last_closed_stid = NULL;
4182         INIT_LIST_HEAD(&oo->oo_close_lru);
4183         spin_lock(&clp->cl_lock);
4184         ret = find_openstateowner_str_locked(strhashval, open, clp);
4185         if (ret == NULL) {
4186                 hash_openowner(oo, clp, strhashval);
4187                 ret = oo;
4188         } else
4189                 nfs4_free_stateowner(&oo->oo_owner);
4190
4191         spin_unlock(&clp->cl_lock);
4192         return ret;
4193 }
4194
4195 static struct nfs4_ol_stateid *
4196 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4197 {
4198
4199         struct nfs4_openowner *oo = open->op_openowner;
4200         struct nfs4_ol_stateid *retstp = NULL;
4201         struct nfs4_ol_stateid *stp;
4202
4203         stp = open->op_stp;
4204         /* We are moving these outside of the spinlocks to avoid the warnings */
4205         mutex_init(&stp->st_mutex);
4206         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4207
4208 retry:
4209         spin_lock(&oo->oo_owner.so_client->cl_lock);
4210         spin_lock(&fp->fi_lock);
4211
4212         retstp = nfsd4_find_existing_open(fp, open);
4213         if (retstp)
4214                 goto out_unlock;
4215
4216         open->op_stp = NULL;
4217         refcount_inc(&stp->st_stid.sc_count);
4218         stp->st_stid.sc_type = NFS4_OPEN_STID;
4219         INIT_LIST_HEAD(&stp->st_locks);
4220         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4221         get_nfs4_file(fp);
4222         stp->st_stid.sc_file = fp;
4223         stp->st_access_bmap = 0;
4224         stp->st_deny_bmap = 0;
4225         stp->st_openstp = NULL;
4226         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4227         list_add(&stp->st_perfile, &fp->fi_stateids);
4228
4229 out_unlock:
4230         spin_unlock(&fp->fi_lock);
4231         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4232         if (retstp) {
4233                 /* Handle races with CLOSE */
4234                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4235                         nfs4_put_stid(&retstp->st_stid);
4236                         goto retry;
4237                 }
4238                 /* To keep mutex tracking happy */
4239                 mutex_unlock(&stp->st_mutex);
4240                 stp = retstp;
4241         }
4242         return stp;
4243 }
4244
4245 /*
4246  * In the 4.0 case we need to keep the owners around a little while to handle
4247  * CLOSE replay. We still do need to release any file access that is held by
4248  * them before returning however.
4249  */
4250 static void
4251 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4252 {
4253         struct nfs4_ol_stateid *last;
4254         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4255         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4256                                                 nfsd_net_id);
4257
4258         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4259
4260         /*
4261          * We know that we hold one reference via nfsd4_close, and another
4262          * "persistent" reference for the client. If the refcount is higher
4263          * than 2, then there are still calls in progress that are using this
4264          * stateid. We can't put the sc_file reference until they are finished.
4265          * Wait for the refcount to drop to 2. Since it has been unhashed,
4266          * there should be no danger of the refcount going back up again at
4267          * this point.
4268          */
4269         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4270
4271         release_all_access(s);
4272         if (s->st_stid.sc_file) {
4273                 put_nfs4_file(s->st_stid.sc_file);
4274                 s->st_stid.sc_file = NULL;
4275         }
4276
4277         spin_lock(&nn->client_lock);
4278         last = oo->oo_last_closed_stid;
4279         oo->oo_last_closed_stid = s;
4280         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4281         oo->oo_time = get_seconds();
4282         spin_unlock(&nn->client_lock);
4283         if (last)
4284                 nfs4_put_stid(&last->st_stid);
4285 }
4286
4287 /* search file_hashtbl[] for file */
4288 static struct nfs4_file *
4289 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4290 {
4291         struct nfs4_file *fp;
4292
4293         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
4294                 if (fh_match(&fp->fi_fhandle, fh)) {
4295                         if (refcount_inc_not_zero(&fp->fi_ref))
4296                                 return fp;
4297                 }
4298         }
4299         return NULL;
4300 }
4301
4302 struct nfs4_file *
4303 find_file(struct knfsd_fh *fh)
4304 {
4305         struct nfs4_file *fp;
4306         unsigned int hashval = file_hashval(fh);
4307
4308         rcu_read_lock();
4309         fp = find_file_locked(fh, hashval);
4310         rcu_read_unlock();
4311         return fp;
4312 }
4313
4314 static struct nfs4_file *
4315 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4316 {
4317         struct nfs4_file *fp;
4318         unsigned int hashval = file_hashval(fh);
4319
4320         rcu_read_lock();
4321         fp = find_file_locked(fh, hashval);
4322         rcu_read_unlock();
4323         if (fp)
4324                 return fp;
4325
4326         spin_lock(&state_lock);
4327         fp = find_file_locked(fh, hashval);
4328         if (likely(fp == NULL)) {
4329                 nfsd4_init_file(fh, hashval, new);
4330                 fp = new;
4331         }
4332         spin_unlock(&state_lock);
4333
4334         return fp;
4335 }
4336
4337 /*
4338  * Called to check deny when READ with all zero stateid or
4339  * WRITE with all zero or all one stateid
4340  */
4341 static __be32
4342 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4343 {
4344         struct nfs4_file *fp;
4345         __be32 ret = nfs_ok;
4346
4347         fp = find_file(&current_fh->fh_handle);
4348         if (!fp)
4349                 return ret;
4350         /* Check for conflicting share reservations */
4351         spin_lock(&fp->fi_lock);
4352         if (fp->fi_share_deny & deny_type)
4353                 ret = nfserr_locked;
4354         spin_unlock(&fp->fi_lock);
4355         put_nfs4_file(fp);
4356         return ret;
4357 }
4358
4359 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4360 {
4361         struct nfs4_delegation *dp = cb_to_delegation(cb);
4362         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4363                                           nfsd_net_id);
4364
4365         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4366
4367         /*
4368          * We can't do this in nfsd_break_deleg_cb because it is
4369          * already holding inode->i_lock.
4370          *
4371          * If the dl_time != 0, then we know that it has already been
4372          * queued for a lease break. Don't queue it again.
4373          */
4374         spin_lock(&state_lock);
4375         if (dp->dl_time == 0) {
4376                 dp->dl_time = get_seconds();
4377                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4378         }
4379         spin_unlock(&state_lock);
4380 }
4381
4382 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4383                 struct rpc_task *task)
4384 {
4385         struct nfs4_delegation *dp = cb_to_delegation(cb);
4386
4387         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4388                 return 1;
4389
4390         switch (task->tk_status) {
4391         case 0:
4392                 return 1;
4393         case -NFS4ERR_DELAY:
4394                 rpc_delay(task, 2 * HZ);
4395                 return 0;
4396         case -EBADHANDLE:
4397         case -NFS4ERR_BAD_STATEID:
4398                 /*
4399                  * Race: client probably got cb_recall before open reply
4400                  * granting delegation.
4401                  */
4402                 if (dp->dl_retries--) {
4403                         rpc_delay(task, 2 * HZ);
4404                         return 0;
4405                 }
4406                 /*FALLTHRU*/
4407         default:
4408                 return 1;
4409         }
4410 }
4411
4412 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4413 {
4414         struct nfs4_delegation *dp = cb_to_delegation(cb);
4415
4416         nfs4_put_stid(&dp->dl_stid);
4417 }
4418
4419 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4420         .prepare        = nfsd4_cb_recall_prepare,
4421         .done           = nfsd4_cb_recall_done,
4422         .release        = nfsd4_cb_recall_release,
4423 };
4424
4425 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4426 {
4427         /*
4428          * We're assuming the state code never drops its reference
4429          * without first removing the lease.  Since we're in this lease
4430          * callback (and since the lease code is serialized by the
4431          * i_lock) we know the server hasn't removed the lease yet, and
4432          * we know it's safe to take a reference.
4433          */
4434         refcount_inc(&dp->dl_stid.sc_count);
4435         nfsd4_run_cb(&dp->dl_recall);
4436 }
4437
4438 /* Called from break_lease() with i_lock held. */
4439 static bool
4440 nfsd_break_deleg_cb(struct file_lock *fl)
4441 {
4442         bool ret = false;
4443         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4444         struct nfs4_file *fp = dp->dl_stid.sc_file;
4445
4446         /*
4447          * We don't want the locks code to timeout the lease for us;
4448          * we'll remove it ourself if a delegation isn't returned
4449          * in time:
4450          */
4451         fl->fl_break_time = 0;
4452
4453         spin_lock(&fp->fi_lock);
4454         fp->fi_had_conflict = true;
4455         nfsd_break_one_deleg(dp);
4456         spin_unlock(&fp->fi_lock);
4457         return ret;
4458 }
4459
4460 static int
4461 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4462                      struct list_head *dispose)
4463 {
4464         if (arg & F_UNLCK)
4465                 return lease_modify(onlist, arg, dispose);
4466         else
4467                 return -EAGAIN;
4468 }
4469
4470 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4471         .lm_break = nfsd_break_deleg_cb,
4472         .lm_change = nfsd_change_deleg_cb,
4473 };
4474
4475 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4476 {
4477         if (nfsd4_has_session(cstate))
4478                 return nfs_ok;
4479         if (seqid == so->so_seqid - 1)
4480                 return nfserr_replay_me;
4481         if (seqid == so->so_seqid)
4482                 return nfs_ok;
4483         return nfserr_bad_seqid;
4484 }
4485
4486 static __be32 lookup_clientid(clientid_t *clid,
4487                 struct nfsd4_compound_state *cstate,
4488                 struct nfsd_net *nn)
4489 {
4490         struct nfs4_client *found;
4491
4492         if (cstate->clp) {
4493                 found = cstate->clp;
4494                 if (!same_clid(&found->cl_clientid, clid))
4495                         return nfserr_stale_clientid;
4496                 return nfs_ok;
4497         }
4498
4499         if (STALE_CLIENTID(clid, nn))
4500                 return nfserr_stale_clientid;
4501
4502         /*
4503          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4504          * cached already then we know this is for is for v4.0 and "sessions"
4505          * will be false.
4506          */
4507         WARN_ON_ONCE(cstate->session);
4508         spin_lock(&nn->client_lock);
4509         found = find_confirmed_client(clid, false, nn);
4510         if (!found) {
4511                 spin_unlock(&nn->client_lock);
4512                 return nfserr_expired;
4513         }
4514         atomic_inc(&found->cl_rpc_users);
4515         spin_unlock(&nn->client_lock);
4516
4517         /* Cache the nfs4_client in cstate! */
4518         cstate->clp = found;
4519         return nfs_ok;
4520 }
4521
4522 __be32
4523 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4524                     struct nfsd4_open *open, struct nfsd_net *nn)
4525 {
4526         clientid_t *clientid = &open->op_clientid;
4527         struct nfs4_client *clp = NULL;
4528         unsigned int strhashval;
4529         struct nfs4_openowner *oo = NULL;
4530         __be32 status;
4531
4532         if (STALE_CLIENTID(&open->op_clientid, nn))
4533                 return nfserr_stale_clientid;
4534         /*
4535          * In case we need it later, after we've already created the
4536          * file and don't want to risk a further failure:
4537          */
4538         open->op_file = nfsd4_alloc_file();
4539         if (open->op_file == NULL)
4540                 return nfserr_jukebox;
4541
4542         status = lookup_clientid(clientid, cstate, nn);
4543         if (status)
4544                 return status;
4545         clp = cstate->clp;
4546
4547         strhashval = ownerstr_hashval(&open->op_owner);
4548         oo = find_openstateowner_str(strhashval, open, clp);
4549         open->op_openowner = oo;
4550         if (!oo) {
4551                 goto new_owner;
4552         }
4553         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4554                 /* Replace unconfirmed owners without checking for replay. */
4555                 release_openowner(oo);
4556                 open->op_openowner = NULL;
4557                 goto new_owner;
4558         }
4559         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4560         if (status)
4561                 return status;
4562         goto alloc_stateid;
4563 new_owner:
4564         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4565         if (oo == NULL)
4566                 return nfserr_jukebox;
4567         open->op_openowner = oo;
4568 alloc_stateid:
4569         open->op_stp = nfs4_alloc_open_stateid(clp);
4570         if (!open->op_stp)
4571                 return nfserr_jukebox;
4572
4573         if (nfsd4_has_session(cstate) &&
4574             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4575                 open->op_odstate = alloc_clnt_odstate(clp);
4576                 if (!open->op_odstate)
4577                         return nfserr_jukebox;
4578         }
4579
4580         return nfs_ok;
4581 }
4582
4583 static inline __be32
4584 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4585 {
4586         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4587                 return nfserr_openmode;
4588         else
4589                 return nfs_ok;
4590 }
4591
4592 static int share_access_to_flags(u32 share_access)
4593 {
4594         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4595 }
4596
4597 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4598 {
4599         struct nfs4_stid *ret;
4600
4601         ret = find_stateid_by_type(cl, s,
4602                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4603         if (!ret)
4604                 return NULL;
4605         return delegstateid(ret);
4606 }
4607
4608 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4609 {
4610         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4611                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4612 }
4613
4614 static __be32
4615 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4616                 struct nfs4_delegation **dp)
4617 {
4618         int flags;
4619         __be32 status = nfserr_bad_stateid;
4620         struct nfs4_delegation *deleg;
4621
4622         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4623         if (deleg == NULL)
4624                 goto out;
4625         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4626                 nfs4_put_stid(&deleg->dl_stid);
4627                 if (cl->cl_minorversion)
4628                         status = nfserr_deleg_revoked;
4629                 goto out;
4630         }
4631         flags = share_access_to_flags(open->op_share_access);
4632         status = nfs4_check_delegmode(deleg, flags);
4633         if (status) {
4634                 nfs4_put_stid(&deleg->dl_stid);
4635                 goto out;
4636         }
4637         *dp = deleg;
4638 out:
4639         if (!nfsd4_is_deleg_cur(open))
4640                 return nfs_ok;
4641         if (status)
4642                 return status;
4643         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4644         return nfs_ok;
4645 }
4646
4647 static inline int nfs4_access_to_access(u32 nfs4_access)
4648 {
4649         int flags = 0;
4650
4651         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4652                 flags |= NFSD_MAY_READ;
4653         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4654                 flags |= NFSD_MAY_WRITE;
4655         return flags;
4656 }
4657
4658 static inline __be32
4659 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4660                 struct nfsd4_open *open)
4661 {
4662         struct iattr iattr = {
4663                 .ia_valid = ATTR_SIZE,
4664                 .ia_size = 0,
4665         };
4666         if (!open->op_truncate)
4667                 return 0;
4668         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4669                 return nfserr_inval;
4670         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4671 }
4672
4673 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4674                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4675                 struct nfsd4_open *open)
4676 {
4677         struct nfsd_file *nf = NULL;
4678         __be32 status;
4679         int oflag = nfs4_access_to_omode(open->op_share_access);
4680         int access = nfs4_access_to_access(open->op_share_access);
4681         unsigned char old_access_bmap, old_deny_bmap;
4682
4683         spin_lock(&fp->fi_lock);
4684
4685         /*
4686          * Are we trying to set a deny mode that would conflict with
4687          * current access?
4688          */
4689         status = nfs4_file_check_deny(fp, open->op_share_deny);
4690         if (status != nfs_ok) {
4691                 spin_unlock(&fp->fi_lock);
4692                 goto out;
4693         }
4694
4695         /* set access to the file */
4696         status = nfs4_file_get_access(fp, open->op_share_access);
4697         if (status != nfs_ok) {
4698                 spin_unlock(&fp->fi_lock);
4699                 goto out;
4700         }
4701
4702         /* Set access bits in stateid */
4703         old_access_bmap = stp->st_access_bmap;
4704         set_access(open->op_share_access, stp);
4705
4706         /* Set new deny mask */
4707         old_deny_bmap = stp->st_deny_bmap;
4708         set_deny(open->op_share_deny, stp);
4709         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4710
4711         if (!fp->fi_fds[oflag]) {
4712                 spin_unlock(&fp->fi_lock);
4713                 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4714                 if (status)
4715                         goto out_put_access;
4716                 spin_lock(&fp->fi_lock);
4717                 if (!fp->fi_fds[oflag]) {
4718                         fp->fi_fds[oflag] = nf;
4719                         nf = NULL;
4720                 }
4721         }
4722         spin_unlock(&fp->fi_lock);
4723         if (nf)
4724                 nfsd_file_put(nf);
4725
4726         status = nfsd4_truncate(rqstp, cur_fh, open);
4727         if (status)
4728                 goto out_put_access;
4729 out:
4730         return status;
4731 out_put_access:
4732         stp->st_access_bmap = old_access_bmap;
4733         nfs4_file_put_access(fp, open->op_share_access);
4734         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4735         goto out;
4736 }
4737
4738 static __be32
4739 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)
4740 {
4741         __be32 status;
4742         unsigned char old_deny_bmap = stp->st_deny_bmap;
4743
4744         if (!test_access(open->op_share_access, stp))
4745                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4746
4747         /* test and set deny mode */
4748         spin_lock(&fp->fi_lock);
4749         status = nfs4_file_check_deny(fp, open->op_share_deny);
4750         if (status == nfs_ok) {
4751                 set_deny(open->op_share_deny, stp);
4752                 fp->fi_share_deny |=
4753                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4754         }
4755         spin_unlock(&fp->fi_lock);
4756
4757         if (status != nfs_ok)
4758                 return status;
4759
4760         status = nfsd4_truncate(rqstp, cur_fh, open);
4761         if (status != nfs_ok)
4762                 reset_union_bmap_deny(old_deny_bmap, stp);
4763         return status;
4764 }
4765
4766 /* Should we give out recallable state?: */
4767 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4768 {
4769         if (clp->cl_cb_state == NFSD4_CB_UP)
4770                 return true;
4771         /*
4772          * In the sessions case, since we don't have to establish a
4773          * separate connection for callbacks, we assume it's OK
4774          * until we hear otherwise:
4775          */
4776         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4777 }
4778
4779 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4780                                                 int flag)
4781 {
4782         struct file_lock *fl;
4783
4784         fl = locks_alloc_lock();
4785         if (!fl)
4786                 return NULL;
4787         fl->fl_lmops = &nfsd_lease_mng_ops;
4788         fl->fl_flags = FL_DELEG;
4789         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4790         fl->fl_end = OFFSET_MAX;
4791         fl->fl_owner = (fl_owner_t)dp;
4792         fl->fl_pid = current->tgid;
4793         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4794         return fl;
4795 }
4796
4797 static struct nfs4_delegation *
4798 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4799                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4800 {
4801         int status = 0;
4802         struct nfs4_delegation *dp;
4803         struct nfsd_file *nf;
4804         struct file_lock *fl;
4805
4806         /*
4807          * The fi_had_conflict and nfs_get_existing_delegation checks
4808          * here are just optimizations; we'll need to recheck them at
4809          * the end:
4810          */
4811         if (fp->fi_had_conflict)
4812                 return ERR_PTR(-EAGAIN);
4813
4814         nf = find_readable_file(fp);
4815         if (!nf) {
4816                 /* We should always have a readable file here */
4817                 WARN_ON_ONCE(1);
4818                 return ERR_PTR(-EBADF);
4819         }
4820         spin_lock(&state_lock);
4821         spin_lock(&fp->fi_lock);
4822         if (nfs4_delegation_exists(clp, fp))
4823                 status = -EAGAIN;
4824         else if (!fp->fi_deleg_file) {
4825                 fp->fi_deleg_file = nf;
4826                 /* increment early to prevent fi_deleg_file from being
4827                  * cleared */
4828                 fp->fi_delegees = 1;
4829                 nf = NULL;
4830         } else
4831                 fp->fi_delegees++;
4832         spin_unlock(&fp->fi_lock);
4833         spin_unlock(&state_lock);
4834         if (nf)
4835                 nfsd_file_put(nf);
4836         if (status)
4837                 return ERR_PTR(status);
4838
4839         status = -ENOMEM;
4840         dp = alloc_init_deleg(clp, fp, fh, odstate);
4841         if (!dp)
4842                 goto out_delegees;
4843
4844         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4845         if (!fl)
4846                 goto out_clnt_odstate;
4847
4848         status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
4849         if (fl)
4850                 locks_free_lock(fl);
4851         if (status)
4852                 goto out_clnt_odstate;
4853
4854         spin_lock(&state_lock);
4855         spin_lock(&fp->fi_lock);
4856         if (fp->fi_had_conflict)
4857                 status = -EAGAIN;
4858         else
4859                 status = hash_delegation_locked(dp, fp);
4860         spin_unlock(&fp->fi_lock);
4861         spin_unlock(&state_lock);
4862
4863         if (status)
4864                 goto out_unlock;
4865
4866         return dp;
4867 out_unlock:
4868         vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
4869 out_clnt_odstate:
4870         put_clnt_odstate(dp->dl_clnt_odstate);
4871         nfs4_put_stid(&dp->dl_stid);
4872 out_delegees:
4873         put_deleg_file(fp);
4874         return ERR_PTR(status);
4875 }
4876
4877 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4878 {
4879         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4880         if (status == -EAGAIN)
4881                 open->op_why_no_deleg = WND4_CONTENTION;
4882         else {
4883                 open->op_why_no_deleg = WND4_RESOURCE;
4884                 switch (open->op_deleg_want) {
4885                 case NFS4_SHARE_WANT_READ_DELEG:
4886                 case NFS4_SHARE_WANT_WRITE_DELEG:
4887                 case NFS4_SHARE_WANT_ANY_DELEG:
4888                         break;
4889                 case NFS4_SHARE_WANT_CANCEL:
4890                         open->op_why_no_deleg = WND4_CANCELLED;
4891                         break;
4892                 case NFS4_SHARE_WANT_NO_DELEG:
4893                         WARN_ON_ONCE(1);
4894                 }
4895         }
4896 }
4897
4898 /*
4899  * Attempt to hand out a delegation.
4900  *
4901  * Note we don't support write delegations, and won't until the vfs has
4902  * proper support for them.
4903  */
4904 static void
4905 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4906                         struct nfs4_ol_stateid *stp)
4907 {
4908         struct nfs4_delegation *dp;
4909         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4910         struct nfs4_client *clp = stp->st_stid.sc_client;
4911         int cb_up;
4912         int status = 0;
4913
4914         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4915         open->op_recall = 0;
4916         switch (open->op_claim_type) {
4917                 case NFS4_OPEN_CLAIM_PREVIOUS:
4918                         if (!cb_up)
4919                                 open->op_recall = 1;
4920                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4921                                 goto out_no_deleg;
4922                         break;
4923                 case NFS4_OPEN_CLAIM_NULL:
4924                 case NFS4_OPEN_CLAIM_FH:
4925                         /*
4926                          * Let's not give out any delegations till everyone's
4927                          * had the chance to reclaim theirs, *and* until
4928                          * NLM locks have all been reclaimed:
4929                          */
4930                         if (locks_in_grace(clp->net))
4931                                 goto out_no_deleg;
4932                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4933                                 goto out_no_deleg;
4934                         /*
4935                          * Also, if the file was opened for write or
4936                          * create, there's a good chance the client's
4937                          * about to write to it, resulting in an
4938                          * immediate recall (since we don't support
4939                          * write delegations):
4940                          */
4941                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4942                                 goto out_no_deleg;
4943                         if (open->op_create == NFS4_OPEN_CREATE)
4944                                 goto out_no_deleg;
4945                         break;
4946                 default:
4947                         goto out_no_deleg;
4948         }
4949         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4950         if (IS_ERR(dp))
4951                 goto out_no_deleg;
4952
4953         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4954
4955         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4956                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4957         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4958         nfs4_put_stid(&dp->dl_stid);
4959         return;
4960 out_no_deleg:
4961         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4962         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4963             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4964                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4965                 open->op_recall = 1;
4966         }
4967
4968         /* 4.1 client asking for a delegation? */
4969         if (open->op_deleg_want)
4970                 nfsd4_open_deleg_none_ext(open, status);
4971         return;
4972 }
4973
4974 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4975                                         struct nfs4_delegation *dp)
4976 {
4977         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4978             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4979                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4980                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4981         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4982                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4983                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4984                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4985         }
4986         /* Otherwise the client must be confused wanting a delegation
4987          * it already has, therefore we don't return
4988          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4989          */
4990 }
4991
4992 __be32
4993 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4994 {
4995         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4996         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4997         struct nfs4_file *fp = NULL;
4998         struct nfs4_ol_stateid *stp = NULL;
4999         struct nfs4_delegation *dp = NULL;
5000         __be32 status;
5001         bool new_stp = false;
5002
5003         /*
5004          * Lookup file; if found, lookup stateid and check open request,
5005          * and check for delegations in the process of being recalled.
5006          * If not found, create the nfs4_file struct
5007          */
5008         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
5009         if (fp != open->op_file) {
5010                 status = nfs4_check_deleg(cl, open, &dp);
5011                 if (status)
5012                         goto out;
5013                 stp = nfsd4_find_and_lock_existing_open(fp, open);
5014         } else {
5015                 open->op_file = NULL;
5016                 status = nfserr_bad_stateid;
5017                 if (nfsd4_is_deleg_cur(open))
5018                         goto out;
5019         }
5020
5021         if (!stp) {
5022                 stp = init_open_stateid(fp, open);
5023                 if (!open->op_stp)
5024                         new_stp = true;
5025         }
5026
5027         /*
5028          * OPEN the file, or upgrade an existing OPEN.
5029          * If truncate fails, the OPEN fails.
5030          *
5031          * stp is already locked.
5032          */
5033         if (!new_stp) {
5034                 /* Stateid was found, this is an OPEN upgrade */
5035                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5036                 if (status) {
5037                         mutex_unlock(&stp->st_mutex);
5038                         goto out;
5039                 }
5040         } else {
5041                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5042                 if (status) {
5043                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5044                         release_open_stateid(stp);
5045                         mutex_unlock(&stp->st_mutex);
5046                         goto out;
5047                 }
5048
5049                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5050                                                         open->op_odstate);
5051                 if (stp->st_clnt_odstate == open->op_odstate)
5052                         open->op_odstate = NULL;
5053         }
5054
5055         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5056         mutex_unlock(&stp->st_mutex);
5057
5058         if (nfsd4_has_session(&resp->cstate)) {
5059                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5060                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5061                         open->op_why_no_deleg = WND4_NOT_WANTED;
5062                         goto nodeleg;
5063                 }
5064         }
5065
5066         /*
5067         * Attempt to hand out a delegation. No error return, because the
5068         * OPEN succeeds even if we fail.
5069         */
5070         nfs4_open_delegation(current_fh, open, stp);
5071 nodeleg:
5072         status = nfs_ok;
5073
5074         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5075                 STATEID_VAL(&stp->st_stid.sc_stateid));
5076 out:
5077         /* 4.1 client trying to upgrade/downgrade delegation? */
5078         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5079             open->op_deleg_want)
5080                 nfsd4_deleg_xgrade_none_ext(open, dp);
5081
5082         if (fp)
5083                 put_nfs4_file(fp);
5084         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5085                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5086         /*
5087         * To finish the open response, we just need to set the rflags.
5088         */
5089         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5090         if (nfsd4_has_session(&resp->cstate))
5091                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5092         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5093                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5094
5095         if (dp)
5096                 nfs4_put_stid(&dp->dl_stid);
5097         if (stp)
5098                 nfs4_put_stid(&stp->st_stid);
5099
5100         return status;
5101 }
5102
5103 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5104                               struct nfsd4_open *open)
5105 {
5106         if (open->op_openowner) {
5107                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5108
5109                 nfsd4_cstate_assign_replay(cstate, so);
5110                 nfs4_put_stateowner(so);
5111         }
5112         if (open->op_file)
5113                 kmem_cache_free(file_slab, open->op_file);
5114         if (open->op_stp)
5115                 nfs4_put_stid(&open->op_stp->st_stid);
5116         if (open->op_odstate)
5117                 kmem_cache_free(odstate_slab, open->op_odstate);
5118 }
5119
5120 __be32
5121 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5122             union nfsd4_op_u *u)
5123 {
5124         clientid_t *clid = &u->renew;
5125         struct nfs4_client *clp;
5126         __be32 status;
5127         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5128
5129         dprintk("process_renew(%08x/%08x): starting\n",
5130                         clid->cl_boot, clid->cl_id);
5131         status = lookup_clientid(clid, cstate, nn);
5132         if (status)
5133                 goto out;
5134         clp = cstate->clp;
5135         status = nfserr_cb_path_down;
5136         if (!list_empty(&clp->cl_delegations)
5137                         && clp->cl_cb_state != NFSD4_CB_UP)
5138                 goto out;
5139         status = nfs_ok;
5140 out:
5141         return status;
5142 }
5143
5144 void
5145 nfsd4_end_grace(struct nfsd_net *nn)
5146 {
5147         /* do nothing if grace period already ended */
5148         if (nn->grace_ended)
5149                 return;
5150
5151         nn->grace_ended = true;
5152         /*
5153          * If the server goes down again right now, an NFSv4
5154          * client will still be allowed to reclaim after it comes back up,
5155          * even if it hasn't yet had a chance to reclaim state this time.
5156          *
5157          */
5158         nfsd4_record_grace_done(nn);
5159         /*
5160          * At this point, NFSv4 clients can still reclaim.  But if the
5161          * server crashes, any that have not yet reclaimed will be out
5162          * of luck on the next boot.
5163          *
5164          * (NFSv4.1+ clients are considered to have reclaimed once they
5165          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5166          * have reclaimed after their first OPEN.)
5167          */
5168         locks_end_grace(&nn->nfsd4_manager);
5169         /*
5170          * At this point, and once lockd and/or any other containers
5171          * exit their grace period, further reclaims will fail and
5172          * regular locking can resume.
5173          */
5174 }
5175
5176 /*
5177  * If we've waited a lease period but there are still clients trying to
5178  * reclaim, wait a little longer to give them a chance to finish.
5179  */
5180 static bool clients_still_reclaiming(struct nfsd_net *nn)
5181 {
5182         unsigned long now = get_seconds();
5183         unsigned long double_grace_period_end = nn->boot_time +
5184                                                 2 * nn->nfsd4_lease;
5185
5186         if (nn->track_reclaim_completes &&
5187                         atomic_read(&nn->nr_reclaim_complete) ==
5188                         nn->reclaim_str_hashtbl_size)
5189                 return false;
5190         if (!nn->somebody_reclaimed)
5191                 return false;
5192         nn->somebody_reclaimed = false;
5193         /*
5194          * If we've given them *two* lease times to reclaim, and they're
5195          * still not done, give up:
5196          */
5197         if (time_after(now, double_grace_period_end))
5198                 return false;
5199         return true;
5200 }
5201
5202 static time_t
5203 nfs4_laundromat(struct nfsd_net *nn)
5204 {
5205         struct nfs4_client *clp;
5206         struct nfs4_openowner *oo;
5207         struct nfs4_delegation *dp;
5208         struct nfs4_ol_stateid *stp;
5209         struct nfsd4_blocked_lock *nbl;
5210         struct list_head *pos, *next, reaplist;
5211         time_t cutoff = get_seconds() - nn->nfsd4_lease;
5212         time_t t, new_timeo = nn->nfsd4_lease;
5213
5214         dprintk("NFSD: laundromat service - starting\n");
5215
5216         if (clients_still_reclaiming(nn)) {
5217                 new_timeo = 0;
5218                 goto out;
5219         }
5220         dprintk("NFSD: end of grace period\n");
5221         nfsd4_end_grace(nn);
5222         INIT_LIST_HEAD(&reaplist);
5223         spin_lock(&nn->client_lock);
5224         list_for_each_safe(pos, next, &nn->client_lru) {
5225                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5226                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
5227                         t = clp->cl_time - cutoff;
5228                         new_timeo = min(new_timeo, t);
5229                         break;
5230                 }
5231                 if (mark_client_expired_locked(clp)) {
5232                         dprintk("NFSD: client in use (clientid %08x)\n",
5233                                 clp->cl_clientid.cl_id);
5234                         continue;
5235                 }
5236                 list_add(&clp->cl_lru, &reaplist);
5237         }
5238         spin_unlock(&nn->client_lock);
5239         list_for_each_safe(pos, next, &reaplist) {
5240                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5241                 dprintk("NFSD: purging unused client (clientid %08x)\n",
5242                         clp->cl_clientid.cl_id);
5243                 list_del_init(&clp->cl_lru);
5244                 expire_client(clp);
5245         }
5246         spin_lock(&state_lock);
5247         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5248                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5249                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
5250                         t = dp->dl_time - cutoff;
5251                         new_timeo = min(new_timeo, t);
5252                         break;
5253                 }
5254                 WARN_ON(!unhash_delegation_locked(dp));
5255                 list_add(&dp->dl_recall_lru, &reaplist);
5256         }
5257         spin_unlock(&state_lock);
5258         while (!list_empty(&reaplist)) {
5259                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5260                                         dl_recall_lru);
5261                 list_del_init(&dp->dl_recall_lru);
5262                 revoke_delegation(dp);
5263         }
5264
5265         spin_lock(&nn->client_lock);
5266         while (!list_empty(&nn->close_lru)) {
5267                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5268                                         oo_close_lru);
5269                 if (time_after((unsigned long)oo->oo_time,
5270                                (unsigned long)cutoff)) {
5271                         t = oo->oo_time - cutoff;
5272                         new_timeo = min(new_timeo, t);
5273                         break;
5274                 }
5275                 list_del_init(&oo->oo_close_lru);
5276                 stp = oo->oo_last_closed_stid;
5277                 oo->oo_last_closed_stid = NULL;
5278                 spin_unlock(&nn->client_lock);
5279                 nfs4_put_stid(&stp->st_stid);
5280                 spin_lock(&nn->client_lock);
5281         }
5282         spin_unlock(&nn->client_lock);
5283
5284         /*
5285          * It's possible for a client to try and acquire an already held lock
5286          * that is being held for a long time, and then lose interest in it.
5287          * So, we clean out any un-revisited request after a lease period
5288          * under the assumption that the client is no longer interested.
5289          *
5290          * RFC5661, sec. 9.6 states that the client must not rely on getting
5291          * notifications and must continue to poll for locks, even when the
5292          * server supports them. Thus this shouldn't lead to clients blocking
5293          * indefinitely once the lock does become free.
5294          */
5295         BUG_ON(!list_empty(&reaplist));
5296         spin_lock(&nn->blocked_locks_lock);
5297         while (!list_empty(&nn->blocked_locks_lru)) {
5298                 nbl = list_first_entry(&nn->blocked_locks_lru,
5299                                         struct nfsd4_blocked_lock, nbl_lru);
5300                 if (time_after((unsigned long)nbl->nbl_time,
5301                                (unsigned long)cutoff)) {
5302                         t = nbl->nbl_time - cutoff;
5303                         new_timeo = min(new_timeo, t);
5304                         break;
5305                 }
5306                 list_move(&nbl->nbl_lru, &reaplist);
5307                 list_del_init(&nbl->nbl_list);
5308         }
5309         spin_unlock(&nn->blocked_locks_lock);
5310
5311         while (!list_empty(&reaplist)) {
5312                 nbl = list_first_entry(&reaplist,
5313                                         struct nfsd4_blocked_lock, nbl_lru);
5314                 list_del_init(&nbl->nbl_lru);
5315                 free_blocked_lock(nbl);
5316         }
5317 out:
5318         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5319         return new_timeo;
5320 }
5321
5322 static struct workqueue_struct *laundry_wq;
5323 static void laundromat_main(struct work_struct *);
5324
5325 static void
5326 laundromat_main(struct work_struct *laundry)
5327 {
5328         time_t t;
5329         struct delayed_work *dwork = to_delayed_work(laundry);
5330         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5331                                            laundromat_work);
5332
5333         t = nfs4_laundromat(nn);
5334         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
5335         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5336 }
5337
5338 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5339 {
5340         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5341                 return nfserr_bad_stateid;
5342         return nfs_ok;
5343 }
5344
5345 static inline int
5346 access_permit_read(struct nfs4_ol_stateid *stp)
5347 {
5348         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5349                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5350                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5351 }
5352
5353 static inline int
5354 access_permit_write(struct nfs4_ol_stateid *stp)
5355 {
5356         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5357                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5358 }
5359
5360 static
5361 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5362 {
5363         __be32 status = nfserr_openmode;
5364
5365         /* For lock stateid's, we test the parent open, not the lock: */
5366         if (stp->st_openstp)
5367                 stp = stp->st_openstp;
5368         if ((flags & WR_STATE) && !access_permit_write(stp))
5369                 goto out;
5370         if ((flags & RD_STATE) && !access_permit_read(stp))
5371                 goto out;
5372         status = nfs_ok;
5373 out:
5374         return status;
5375 }
5376
5377 static inline __be32
5378 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5379 {
5380         if (ONE_STATEID(stateid) && (flags & RD_STATE))
5381                 return nfs_ok;
5382         else if (opens_in_grace(net)) {
5383                 /* Answer in remaining cases depends on existence of
5384                  * conflicting state; so we must wait out the grace period. */
5385                 return nfserr_grace;
5386         } else if (flags & WR_STATE)
5387                 return nfs4_share_conflict(current_fh,
5388                                 NFS4_SHARE_DENY_WRITE);
5389         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5390                 return nfs4_share_conflict(current_fh,
5391                                 NFS4_SHARE_DENY_READ);
5392 }
5393
5394 /*
5395  * Allow READ/WRITE during grace period on recovered state only for files
5396  * that are not able to provide mandatory locking.
5397  */
5398 static inline int
5399 grace_disallows_io(struct net *net, struct inode *inode)
5400 {
5401         return opens_in_grace(net) && mandatory_lock(inode);
5402 }
5403
5404 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5405 {
5406         /*
5407          * When sessions are used the stateid generation number is ignored
5408          * when it is zero.
5409          */
5410         if (has_session && in->si_generation == 0)
5411                 return nfs_ok;
5412
5413         if (in->si_generation == ref->si_generation)
5414                 return nfs_ok;
5415
5416         /* If the client sends us a stateid from the future, it's buggy: */
5417         if (nfsd4_stateid_generation_after(in, ref))
5418                 return nfserr_bad_stateid;
5419         /*
5420          * However, we could see a stateid from the past, even from a
5421          * non-buggy client.  For example, if the client sends a lock
5422          * while some IO is outstanding, the lock may bump si_generation
5423          * while the IO is still in flight.  The client could avoid that
5424          * situation by waiting for responses on all the IO requests,
5425          * but better performance may result in retrying IO that
5426          * receives an old_stateid error if requests are rarely
5427          * reordered in flight:
5428          */
5429         return nfserr_old_stateid;
5430 }
5431
5432 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5433 {
5434         __be32 ret;
5435
5436         spin_lock(&s->sc_lock);
5437         ret = nfsd4_verify_open_stid(s);
5438         if (ret == nfs_ok)
5439                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5440         spin_unlock(&s->sc_lock);
5441         return ret;
5442 }
5443
5444 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5445 {
5446         if (ols->st_stateowner->so_is_open_owner &&
5447             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5448                 return nfserr_bad_stateid;
5449         return nfs_ok;
5450 }
5451
5452 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5453 {
5454         struct nfs4_stid *s;
5455         __be32 status = nfserr_bad_stateid;
5456
5457         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5458                 CLOSE_STATEID(stateid))
5459                 return status;
5460         /* Client debugging aid. */
5461         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5462                 char addr_str[INET6_ADDRSTRLEN];
5463                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5464                                  sizeof(addr_str));
5465                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5466                                         "with incorrect client ID\n", addr_str);
5467                 return status;
5468         }
5469         spin_lock(&cl->cl_lock);
5470         s = find_stateid_locked(cl, stateid);
5471         if (!s)
5472                 goto out_unlock;
5473         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5474         if (status)
5475                 goto out_unlock;
5476         switch (s->sc_type) {
5477         case NFS4_DELEG_STID:
5478                 status = nfs_ok;
5479                 break;
5480         case NFS4_REVOKED_DELEG_STID:
5481                 status = nfserr_deleg_revoked;
5482                 break;
5483         case NFS4_OPEN_STID:
5484         case NFS4_LOCK_STID:
5485                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5486                 break;
5487         default:
5488                 printk("unknown stateid type %x\n", s->sc_type);
5489                 /* Fallthrough */
5490         case NFS4_CLOSED_STID:
5491         case NFS4_CLOSED_DELEG_STID:
5492                 status = nfserr_bad_stateid;
5493         }
5494 out_unlock:
5495         spin_unlock(&cl->cl_lock);
5496         return status;
5497 }
5498
5499 __be32
5500 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5501                      stateid_t *stateid, unsigned char typemask,
5502                      struct nfs4_stid **s, struct nfsd_net *nn)
5503 {
5504         __be32 status;
5505         bool return_revoked = false;
5506
5507         /*
5508          *  only return revoked delegations if explicitly asked.
5509          *  otherwise we report revoked or bad_stateid status.
5510          */
5511         if (typemask & NFS4_REVOKED_DELEG_STID)
5512                 return_revoked = true;
5513         else if (typemask & NFS4_DELEG_STID)
5514                 typemask |= NFS4_REVOKED_DELEG_STID;
5515
5516         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5517                 CLOSE_STATEID(stateid))
5518                 return nfserr_bad_stateid;
5519         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5520         if (status == nfserr_stale_clientid) {
5521                 if (cstate->session)
5522                         return nfserr_bad_stateid;
5523                 return nfserr_stale_stateid;
5524         }
5525         if (status)
5526                 return status;
5527         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5528         if (!*s)
5529                 return nfserr_bad_stateid;
5530         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5531                 nfs4_put_stid(*s);
5532                 if (cstate->minorversion)
5533                         return nfserr_deleg_revoked;
5534                 return nfserr_bad_stateid;
5535         }
5536         return nfs_ok;
5537 }
5538
5539 static struct nfsd_file *
5540 nfs4_find_file(struct nfs4_stid *s, int flags)
5541 {
5542         if (!s)
5543                 return NULL;
5544
5545         switch (s->sc_type) {
5546         case NFS4_DELEG_STID:
5547                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5548                         return NULL;
5549                 return nfsd_file_get(s->sc_file->fi_deleg_file);
5550         case NFS4_OPEN_STID:
5551         case NFS4_LOCK_STID:
5552                 if (flags & RD_STATE)
5553                         return find_readable_file(s->sc_file);
5554                 else
5555                         return find_writeable_file(s->sc_file);
5556                 break;
5557         }
5558
5559         return NULL;
5560 }
5561
5562 static __be32
5563 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5564 {
5565         __be32 status;
5566
5567         status = nfsd4_check_openowner_confirmed(ols);
5568         if (status)
5569                 return status;
5570         return nfs4_check_openmode(ols, flags);
5571 }
5572
5573 static __be32
5574 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5575                 struct nfsd_file **nfp, int flags)
5576 {
5577         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5578         struct nfsd_file *nf;
5579         __be32 status;
5580
5581         nf = nfs4_find_file(s, flags);
5582         if (nf) {
5583                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5584                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5585                 if (status) {
5586                         nfsd_file_put(nf);
5587                         goto out;
5588                 }
5589         } else {
5590                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5591                 if (status)
5592                         return status;
5593         }
5594         *nfp = nf;
5595 out:
5596         return status;
5597 }
5598
5599 /*
5600  * Checks for stateid operations
5601  */
5602 __be32
5603 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5604                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5605                 stateid_t *stateid, int flags, struct nfsd_file **nfp)
5606 {
5607         struct inode *ino = d_inode(fhp->fh_dentry);
5608         struct net *net = SVC_NET(rqstp);
5609         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5610         struct nfs4_stid *s = NULL;
5611         __be32 status;
5612
5613         if (nfp)
5614                 *nfp = NULL;
5615
5616         if (grace_disallows_io(net, ino))
5617                 return nfserr_grace;
5618
5619         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5620                 status = check_special_stateids(net, fhp, stateid, flags);
5621                 goto done;
5622         }
5623
5624         status = nfsd4_lookup_stateid(cstate, stateid,
5625                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5626                                 &s, nn);
5627         if (status)
5628                 return status;
5629         status = nfsd4_stid_check_stateid_generation(stateid, s,
5630                         nfsd4_has_session(cstate));
5631         if (status)
5632                 goto out;
5633
5634         switch (s->sc_type) {
5635         case NFS4_DELEG_STID:
5636                 status = nfs4_check_delegmode(delegstateid(s), flags);
5637                 break;
5638         case NFS4_OPEN_STID:
5639         case NFS4_LOCK_STID:
5640                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5641                 break;
5642         default:
5643                 status = nfserr_bad_stateid;
5644                 break;
5645         }
5646         if (status)
5647                 goto out;
5648         status = nfs4_check_fh(fhp, s);
5649
5650 done:
5651         if (status == nfs_ok && nfp)
5652                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5653 out:
5654         if (s)
5655                 nfs4_put_stid(s);
5656         return status;
5657 }
5658
5659 /*
5660  * Test if the stateid is valid
5661  */
5662 __be32
5663 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5664                    union nfsd4_op_u *u)
5665 {
5666         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5667         struct nfsd4_test_stateid_id *stateid;
5668         struct nfs4_client *cl = cstate->session->se_client;
5669
5670         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5671                 stateid->ts_id_status =
5672                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5673
5674         return nfs_ok;
5675 }
5676
5677 static __be32
5678 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5679 {
5680         struct nfs4_ol_stateid *stp = openlockstateid(s);
5681         __be32 ret;
5682
5683         ret = nfsd4_lock_ol_stateid(stp);
5684         if (ret)
5685                 goto out_put_stid;
5686
5687         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5688         if (ret)
5689                 goto out;
5690
5691         ret = nfserr_locks_held;
5692         if (check_for_locks(stp->st_stid.sc_file,
5693                             lockowner(stp->st_stateowner)))
5694                 goto out;
5695
5696         release_lock_stateid(stp);
5697         ret = nfs_ok;
5698
5699 out:
5700         mutex_unlock(&stp->st_mutex);
5701 out_put_stid:
5702         nfs4_put_stid(s);
5703         return ret;
5704 }
5705
5706 __be32
5707 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5708                    union nfsd4_op_u *u)
5709 {
5710         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5711         stateid_t *stateid = &free_stateid->fr_stateid;
5712         struct nfs4_stid *s;
5713         struct nfs4_delegation *dp;
5714         struct nfs4_client *cl = cstate->session->se_client;
5715         __be32 ret = nfserr_bad_stateid;
5716
5717         spin_lock(&cl->cl_lock);
5718         s = find_stateid_locked(cl, stateid);
5719         if (!s)
5720                 goto out_unlock;
5721         spin_lock(&s->sc_lock);
5722         switch (s->sc_type) {
5723         case NFS4_DELEG_STID:
5724                 ret = nfserr_locks_held;
5725                 break;
5726         case NFS4_OPEN_STID:
5727                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5728                 if (ret)
5729                         break;
5730                 ret = nfserr_locks_held;
5731                 break;
5732         case NFS4_LOCK_STID:
5733                 spin_unlock(&s->sc_lock);
5734                 refcount_inc(&s->sc_count);
5735                 spin_unlock(&cl->cl_lock);
5736                 ret = nfsd4_free_lock_stateid(stateid, s);
5737                 goto out;
5738         case NFS4_REVOKED_DELEG_STID:
5739                 spin_unlock(&s->sc_lock);
5740                 dp = delegstateid(s);
5741                 list_del_init(&dp->dl_recall_lru);
5742                 spin_unlock(&cl->cl_lock);
5743                 nfs4_put_stid(s);
5744                 ret = nfs_ok;
5745                 goto out;
5746         /* Default falls through and returns nfserr_bad_stateid */
5747         }
5748         spin_unlock(&s->sc_lock);
5749 out_unlock:
5750         spin_unlock(&cl->cl_lock);
5751 out:
5752         return ret;
5753 }
5754
5755 static inline int
5756 setlkflg (int type)
5757 {
5758         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5759                 RD_STATE : WR_STATE;
5760 }
5761
5762 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5763 {
5764         struct svc_fh *current_fh = &cstate->current_fh;
5765         struct nfs4_stateowner *sop = stp->st_stateowner;
5766         __be32 status;
5767
5768         status = nfsd4_check_seqid(cstate, sop, seqid);
5769         if (status)
5770                 return status;
5771         status = nfsd4_lock_ol_stateid(stp);
5772         if (status != nfs_ok)
5773                 return status;
5774         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5775         if (status == nfs_ok)
5776                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5777         if (status != nfs_ok)
5778                 mutex_unlock(&stp->st_mutex);
5779         return status;
5780 }
5781
5782 /*
5783  * Checks for sequence id mutating operations.
5784  */
5785 static __be32
5786 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5787                          stateid_t *stateid, char typemask,
5788                          struct nfs4_ol_stateid **stpp,
5789                          struct nfsd_net *nn)
5790 {
5791         __be32 status;
5792         struct nfs4_stid *s;
5793         struct nfs4_ol_stateid *stp = NULL;
5794
5795         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5796                 seqid, STATEID_VAL(stateid));
5797
5798         *stpp = NULL;
5799         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5800         if (status)
5801                 return status;
5802         stp = openlockstateid(s);
5803         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5804
5805         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5806         if (!status)
5807                 *stpp = stp;
5808         else
5809                 nfs4_put_stid(&stp->st_stid);
5810         return status;
5811 }
5812
5813 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5814                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5815 {
5816         __be32 status;
5817         struct nfs4_openowner *oo;
5818         struct nfs4_ol_stateid *stp;
5819
5820         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5821                                                 NFS4_OPEN_STID, &stp, nn);
5822         if (status)
5823                 return status;
5824         oo = openowner(stp->st_stateowner);
5825         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5826                 mutex_unlock(&stp->st_mutex);
5827                 nfs4_put_stid(&stp->st_stid);
5828                 return nfserr_bad_stateid;
5829         }
5830         *stpp = stp;
5831         return nfs_ok;
5832 }
5833
5834 __be32
5835 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5836                    union nfsd4_op_u *u)
5837 {
5838         struct nfsd4_open_confirm *oc = &u->open_confirm;
5839         __be32 status;
5840         struct nfs4_openowner *oo;
5841         struct nfs4_ol_stateid *stp;
5842         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5843
5844         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5845                         cstate->current_fh.fh_dentry);
5846
5847         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5848         if (status)
5849                 return status;
5850
5851         status = nfs4_preprocess_seqid_op(cstate,
5852                                         oc->oc_seqid, &oc->oc_req_stateid,
5853                                         NFS4_OPEN_STID, &stp, nn);
5854         if (status)
5855                 goto out;
5856         oo = openowner(stp->st_stateowner);
5857         status = nfserr_bad_stateid;
5858         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5859                 mutex_unlock(&stp->st_mutex);
5860                 goto put_stateid;
5861         }
5862         oo->oo_flags |= NFS4_OO_CONFIRMED;
5863         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5864         mutex_unlock(&stp->st_mutex);
5865         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5866                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5867
5868         nfsd4_client_record_create(oo->oo_owner.so_client);
5869         status = nfs_ok;
5870 put_stateid:
5871         nfs4_put_stid(&stp->st_stid);
5872 out:
5873         nfsd4_bump_seqid(cstate, status);
5874         return status;
5875 }
5876
5877 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5878 {
5879         if (!test_access(access, stp))
5880                 return;
5881         nfs4_file_put_access(stp->st_stid.sc_file, access);
5882         clear_access(access, stp);
5883 }
5884
5885 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5886 {
5887         switch (to_access) {
5888         case NFS4_SHARE_ACCESS_READ:
5889                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5890                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5891                 break;
5892         case NFS4_SHARE_ACCESS_WRITE:
5893                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5894                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5895                 break;
5896         case NFS4_SHARE_ACCESS_BOTH:
5897                 break;
5898         default:
5899                 WARN_ON_ONCE(1);
5900         }
5901 }
5902
5903 __be32
5904 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5905                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5906 {
5907         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5908         __be32 status;
5909         struct nfs4_ol_stateid *stp;
5910         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5911
5912         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5913                         cstate->current_fh.fh_dentry);
5914
5915         /* We don't yet support WANT bits: */
5916         if (od->od_deleg_want)
5917                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5918                         od->od_deleg_want);
5919
5920         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5921                                         &od->od_stateid, &stp, nn);
5922         if (status)
5923                 goto out;
5924         status = nfserr_inval;
5925         if (!test_access(od->od_share_access, stp)) {
5926                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5927                         stp->st_access_bmap, od->od_share_access);
5928                 goto put_stateid;
5929         }
5930         if (!test_deny(od->od_share_deny, stp)) {
5931                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5932                         stp->st_deny_bmap, od->od_share_deny);
5933                 goto put_stateid;
5934         }
5935         nfs4_stateid_downgrade(stp, od->od_share_access);
5936         reset_union_bmap_deny(od->od_share_deny, stp);
5937         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5938         status = nfs_ok;
5939 put_stateid:
5940         mutex_unlock(&stp->st_mutex);
5941         nfs4_put_stid(&stp->st_stid);
5942 out:
5943         nfsd4_bump_seqid(cstate, status);
5944         return status;
5945 }
5946
5947 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5948 {
5949         struct nfs4_client *clp = s->st_stid.sc_client;
5950         bool unhashed;
5951         LIST_HEAD(reaplist);
5952
5953         spin_lock(&clp->cl_lock);
5954         unhashed = unhash_open_stateid(s, &reaplist);
5955
5956         if (clp->cl_minorversion) {
5957                 if (unhashed)
5958                         put_ol_stateid_locked(s, &reaplist);
5959                 spin_unlock(&clp->cl_lock);
5960                 free_ol_stateid_reaplist(&reaplist);
5961         } else {
5962                 spin_unlock(&clp->cl_lock);
5963                 free_ol_stateid_reaplist(&reaplist);
5964                 if (unhashed)
5965                         move_to_close_lru(s, clp->net);
5966         }
5967 }
5968
5969 /*
5970  * nfs4_unlock_state() called after encode
5971  */
5972 __be32
5973 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5974                 union nfsd4_op_u *u)
5975 {
5976         struct nfsd4_close *close = &u->close;
5977         __be32 status;
5978         struct nfs4_ol_stateid *stp;
5979         struct net *net = SVC_NET(rqstp);
5980         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5981
5982         dprintk("NFSD: nfsd4_close on file %pd\n",
5983                         cstate->current_fh.fh_dentry);
5984
5985         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5986                                         &close->cl_stateid,
5987                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5988                                         &stp, nn);
5989         nfsd4_bump_seqid(cstate, status);
5990         if (status)
5991                 goto out;
5992
5993         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5994
5995         /*
5996          * Technically we don't _really_ have to increment or copy it, since
5997          * it should just be gone after this operation and we clobber the
5998          * copied value below, but we continue to do so here just to ensure
5999          * that racing ops see that there was a state change.
6000          */
6001         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6002
6003         nfsd4_close_open_stateid(stp);
6004         mutex_unlock(&stp->st_mutex);
6005
6006         /* v4.1+ suggests that we send a special stateid in here, since the
6007          * clients should just ignore this anyway. Since this is not useful
6008          * for v4.0 clients either, we set it to the special close_stateid
6009          * universally.
6010          *
6011          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6012          */
6013         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6014
6015         /* put reference from nfs4_preprocess_seqid_op */
6016         nfs4_put_stid(&stp->st_stid);
6017 out:
6018         return status;
6019 }
6020
6021 __be32
6022 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6023                   union nfsd4_op_u *u)
6024 {
6025         struct nfsd4_delegreturn *dr = &u->delegreturn;
6026         struct nfs4_delegation *dp;
6027         stateid_t *stateid = &dr->dr_stateid;
6028         struct nfs4_stid *s;
6029         __be32 status;
6030         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6031
6032         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6033                 return status;
6034
6035         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6036         if (status)
6037                 goto out;
6038         dp = delegstateid(s);
6039         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6040         if (status)
6041                 goto put_stateid;
6042
6043         destroy_delegation(dp);
6044 put_stateid:
6045         nfs4_put_stid(&dp->dl_stid);
6046 out:
6047         return status;
6048 }
6049
6050 static inline u64
6051 end_offset(u64 start, u64 len)
6052 {
6053         u64 end;
6054
6055         end = start + len;
6056         return end >= start ? end: NFS4_MAX_UINT64;
6057 }
6058
6059 /* last octet in a range */
6060 static inline u64
6061 last_byte_offset(u64 start, u64 len)
6062 {
6063         u64 end;
6064
6065         WARN_ON_ONCE(!len);
6066         end = start + len;
6067         return end > start ? end - 1: NFS4_MAX_UINT64;
6068 }
6069
6070 /*
6071  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6072  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6073  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
6074  * locking, this prevents us from being completely protocol-compliant.  The
6075  * real solution to this problem is to start using unsigned file offsets in
6076  * the VFS, but this is a very deep change!
6077  */
6078 static inline void
6079 nfs4_transform_lock_offset(struct file_lock *lock)
6080 {
6081         if (lock->fl_start < 0)
6082                 lock->fl_start = OFFSET_MAX;
6083         if (lock->fl_end < 0)
6084                 lock->fl_end = OFFSET_MAX;
6085 }
6086
6087 static fl_owner_t
6088 nfsd4_fl_get_owner(fl_owner_t owner)
6089 {
6090         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6091
6092         nfs4_get_stateowner(&lo->lo_owner);
6093         return owner;
6094 }
6095
6096 static void
6097 nfsd4_fl_put_owner(fl_owner_t owner)
6098 {
6099         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6100
6101         if (lo)
6102                 nfs4_put_stateowner(&lo->lo_owner);
6103 }
6104
6105 static void
6106 nfsd4_lm_notify(struct file_lock *fl)
6107 {
6108         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
6109         struct net                      *net = lo->lo_owner.so_client->net;
6110         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
6111         struct nfsd4_blocked_lock       *nbl = container_of(fl,
6112                                                 struct nfsd4_blocked_lock, nbl_lock);
6113         bool queue = false;
6114
6115         /* An empty list means that something else is going to be using it */
6116         spin_lock(&nn->blocked_locks_lock);
6117         if (!list_empty(&nbl->nbl_list)) {
6118                 list_del_init(&nbl->nbl_list);
6119                 list_del_init(&nbl->nbl_lru);
6120                 queue = true;
6121         }
6122         spin_unlock(&nn->blocked_locks_lock);
6123
6124         if (queue)
6125                 nfsd4_run_cb(&nbl->nbl_cb);
6126 }
6127
6128 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
6129         .lm_notify = nfsd4_lm_notify,
6130         .lm_get_owner = nfsd4_fl_get_owner,
6131         .lm_put_owner = nfsd4_fl_put_owner,
6132 };
6133
6134 static inline void
6135 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6136 {
6137         struct nfs4_lockowner *lo;
6138
6139         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6140                 lo = (struct nfs4_lockowner *) fl->fl_owner;
6141                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6142                                                 GFP_KERNEL);
6143                 if (!deny->ld_owner.data)
6144                         /* We just don't care that much */
6145                         goto nevermind;
6146                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6147         } else {
6148 nevermind:
6149                 deny->ld_owner.len = 0;
6150                 deny->ld_owner.data = NULL;
6151                 deny->ld_clientid.cl_boot = 0;
6152                 deny->ld_clientid.cl_id = 0;
6153         }
6154         deny->ld_start = fl->fl_start;
6155         deny->ld_length = NFS4_MAX_UINT64;
6156         if (fl->fl_end != NFS4_MAX_UINT64)
6157                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6158         deny->ld_type = NFS4_READ_LT;
6159         if (fl->fl_type != F_RDLCK)
6160                 deny->ld_type = NFS4_WRITE_LT;
6161 }
6162
6163 static struct nfs4_lockowner *
6164 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6165 {
6166         unsigned int strhashval = ownerstr_hashval(owner);
6167         struct nfs4_stateowner *so;
6168
6169         lockdep_assert_held(&clp->cl_lock);
6170
6171         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6172                             so_strhash) {
6173                 if (so->so_is_open_owner)
6174                         continue;
6175                 if (same_owner_str(so, owner))
6176                         return lockowner(nfs4_get_stateowner(so));
6177         }
6178         return NULL;
6179 }
6180
6181 static struct nfs4_lockowner *
6182 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6183 {
6184         struct nfs4_lockowner *lo;
6185
6186         spin_lock(&clp->cl_lock);
6187         lo = find_lockowner_str_locked(clp, owner);
6188         spin_unlock(&clp->cl_lock);
6189         return lo;
6190 }
6191
6192 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6193 {
6194         unhash_lockowner_locked(lockowner(sop));
6195 }
6196
6197 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6198 {
6199         struct nfs4_lockowner *lo = lockowner(sop);
6200
6201         kmem_cache_free(lockowner_slab, lo);
6202 }
6203
6204 static const struct nfs4_stateowner_operations lockowner_ops = {
6205         .so_unhash =    nfs4_unhash_lockowner,
6206         .so_free =      nfs4_free_lockowner,
6207 };
6208
6209 /*
6210  * Alloc a lock owner structure.
6211  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6212  * occurred.
6213  *
6214  * strhashval = ownerstr_hashval
6215  */
6216 static struct nfs4_lockowner *
6217 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6218                            struct nfs4_ol_stateid *open_stp,
6219                            struct nfsd4_lock *lock)
6220 {
6221         struct nfs4_lockowner *lo, *ret;
6222
6223         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6224         if (!lo)
6225                 return NULL;
6226         INIT_LIST_HEAD(&lo->lo_blocked);
6227         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6228         lo->lo_owner.so_is_open_owner = 0;
6229         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6230         lo->lo_owner.so_ops = &lockowner_ops;
6231         spin_lock(&clp->cl_lock);
6232         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6233         if (ret == NULL) {
6234                 list_add(&lo->lo_owner.so_strhash,
6235                          &clp->cl_ownerstr_hashtbl[strhashval]);
6236                 ret = lo;
6237         } else
6238                 nfs4_free_stateowner(&lo->lo_owner);
6239
6240         spin_unlock(&clp->cl_lock);
6241         return ret;
6242 }
6243
6244 static struct nfs4_ol_stateid *
6245 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
6246 {
6247         struct nfs4_ol_stateid *lst;
6248         struct nfs4_client *clp = lo->lo_owner.so_client;
6249
6250         lockdep_assert_held(&clp->cl_lock);
6251
6252         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
6253                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
6254                         continue;
6255                 if (lst->st_stid.sc_file == fp) {
6256                         refcount_inc(&lst->st_stid.sc_count);
6257                         return lst;
6258                 }
6259         }
6260         return NULL;
6261 }
6262
6263 static struct nfs4_ol_stateid *
6264 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6265                   struct nfs4_file *fp, struct inode *inode,
6266                   struct nfs4_ol_stateid *open_stp)
6267 {
6268         struct nfs4_client *clp = lo->lo_owner.so_client;
6269         struct nfs4_ol_stateid *retstp;
6270
6271         mutex_init(&stp->st_mutex);
6272         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6273 retry:
6274         spin_lock(&clp->cl_lock);
6275         spin_lock(&fp->fi_lock);
6276         retstp = find_lock_stateid(lo, fp);
6277         if (retstp)
6278                 goto out_unlock;
6279
6280         refcount_inc(&stp->st_stid.sc_count);
6281         stp->st_stid.sc_type = NFS4_LOCK_STID;
6282         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6283         get_nfs4_file(fp);
6284         stp->st_stid.sc_file = fp;
6285         stp->st_access_bmap = 0;
6286         stp->st_deny_bmap = open_stp->st_deny_bmap;
6287         stp->st_openstp = open_stp;
6288         list_add(&stp->st_locks, &open_stp->st_locks);
6289         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6290         list_add(&stp->st_perfile, &fp->fi_stateids);
6291 out_unlock:
6292         spin_unlock(&fp->fi_lock);
6293         spin_unlock(&clp->cl_lock);
6294         if (retstp) {
6295                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6296                         nfs4_put_stid(&retstp->st_stid);
6297                         goto retry;
6298                 }
6299                 /* To keep mutex tracking happy */
6300                 mutex_unlock(&stp->st_mutex);
6301                 stp = retstp;
6302         }
6303         return stp;
6304 }
6305
6306 static struct nfs4_ol_stateid *
6307 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6308                             struct inode *inode, struct nfs4_ol_stateid *ost,
6309                             bool *new)
6310 {
6311         struct nfs4_stid *ns = NULL;
6312         struct nfs4_ol_stateid *lst;
6313         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6314         struct nfs4_client *clp = oo->oo_owner.so_client;
6315
6316         *new = false;
6317         spin_lock(&clp->cl_lock);
6318         lst = find_lock_stateid(lo, fi);
6319         spin_unlock(&clp->cl_lock);
6320         if (lst != NULL) {
6321                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6322                         goto out;
6323                 nfs4_put_stid(&lst->st_stid);
6324         }
6325         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6326         if (ns == NULL)
6327                 return NULL;
6328
6329         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6330         if (lst == openlockstateid(ns))
6331                 *new = true;
6332         else
6333                 nfs4_put_stid(ns);
6334 out:
6335         return lst;
6336 }
6337
6338 static int
6339 check_lock_length(u64 offset, u64 length)
6340 {
6341         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6342                 (length > ~offset)));
6343 }
6344
6345 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6346 {
6347         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6348
6349         lockdep_assert_held(&fp->fi_lock);
6350
6351         if (test_access(access, lock_stp))
6352                 return;
6353         __nfs4_file_get_access(fp, access);
6354         set_access(access, lock_stp);
6355 }
6356
6357 static __be32
6358 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6359                             struct nfs4_ol_stateid *ost,
6360                             struct nfsd4_lock *lock,
6361                             struct nfs4_ol_stateid **plst, bool *new)
6362 {
6363         __be32 status;
6364         struct nfs4_file *fi = ost->st_stid.sc_file;
6365         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6366         struct nfs4_client *cl = oo->oo_owner.so_client;
6367         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6368         struct nfs4_lockowner *lo;
6369         struct nfs4_ol_stateid *lst;
6370         unsigned int strhashval;
6371
6372         lo = find_lockowner_str(cl, &lock->lk_new_owner);
6373         if (!lo) {
6374                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6375                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6376                 if (lo == NULL)
6377                         return nfserr_jukebox;
6378         } else {
6379                 /* with an existing lockowner, seqids must be the same */
6380                 status = nfserr_bad_seqid;
6381                 if (!cstate->minorversion &&
6382                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6383                         goto out;
6384         }
6385
6386         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6387         if (lst == NULL) {
6388                 status = nfserr_jukebox;
6389                 goto out;
6390         }
6391
6392         status = nfs_ok;
6393         *plst = lst;
6394 out:
6395         nfs4_put_stateowner(&lo->lo_owner);
6396         return status;
6397 }
6398
6399 /*
6400  *  LOCK operation
6401  */
6402 __be32
6403 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6404            union nfsd4_op_u *u)
6405 {
6406         struct nfsd4_lock *lock = &u->lock;
6407         struct nfs4_openowner *open_sop = NULL;
6408         struct nfs4_lockowner *lock_sop = NULL;
6409         struct nfs4_ol_stateid *lock_stp = NULL;
6410         struct nfs4_ol_stateid *open_stp = NULL;
6411         struct nfs4_file *fp;
6412         struct nfsd_file *nf = NULL;
6413         struct nfsd4_blocked_lock *nbl = NULL;
6414         struct file_lock *file_lock = NULL;
6415         struct file_lock *conflock = NULL;
6416         __be32 status = 0;
6417         int lkflg;
6418         int err;
6419         bool new = false;
6420         unsigned char fl_type;
6421         unsigned int fl_flags = FL_POSIX;
6422         struct net *net = SVC_NET(rqstp);
6423         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6424
6425         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6426                 (long long) lock->lk_offset,
6427                 (long long) lock->lk_length);
6428
6429         if (check_lock_length(lock->lk_offset, lock->lk_length))
6430                  return nfserr_inval;
6431
6432         if ((status = fh_verify(rqstp, &cstate->current_fh,
6433                                 S_IFREG, NFSD_MAY_LOCK))) {
6434                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6435                 return status;
6436         }
6437
6438         if (lock->lk_is_new) {
6439                 if (nfsd4_has_session(cstate))
6440                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6441                         memcpy(&lock->lk_new_clientid,
6442                                 &cstate->session->se_client->cl_clientid,
6443                                 sizeof(clientid_t));
6444
6445                 status = nfserr_stale_clientid;
6446                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6447                         goto out;
6448
6449                 /* validate and update open stateid and open seqid */
6450                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6451                                         lock->lk_new_open_seqid,
6452                                         &lock->lk_new_open_stateid,
6453                                         &open_stp, nn);
6454                 if (status)
6455                         goto out;
6456                 mutex_unlock(&open_stp->st_mutex);
6457                 open_sop = openowner(open_stp->st_stateowner);
6458                 status = nfserr_bad_stateid;
6459                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6460                                                 &lock->lk_new_clientid))
6461                         goto out;
6462                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6463                                                         &lock_stp, &new);
6464         } else {
6465                 status = nfs4_preprocess_seqid_op(cstate,
6466                                        lock->lk_old_lock_seqid,
6467                                        &lock->lk_old_lock_stateid,
6468                                        NFS4_LOCK_STID, &lock_stp, nn);
6469         }
6470         if (status)
6471                 goto out;
6472         lock_sop = lockowner(lock_stp->st_stateowner);
6473
6474         lkflg = setlkflg(lock->lk_type);
6475         status = nfs4_check_openmode(lock_stp, lkflg);
6476         if (status)
6477                 goto out;
6478
6479         status = nfserr_grace;
6480         if (locks_in_grace(net) && !lock->lk_reclaim)
6481                 goto out;
6482         status = nfserr_no_grace;
6483         if (!locks_in_grace(net) && lock->lk_reclaim)
6484                 goto out;
6485
6486         fp = lock_stp->st_stid.sc_file;
6487         switch (lock->lk_type) {
6488                 case NFS4_READW_LT:
6489                         if (nfsd4_has_session(cstate))
6490                                 fl_flags |= FL_SLEEP;
6491                         /* Fallthrough */
6492                 case NFS4_READ_LT:
6493                         spin_lock(&fp->fi_lock);
6494                         nf = find_readable_file_locked(fp);
6495                         if (nf)
6496                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6497                         spin_unlock(&fp->fi_lock);
6498                         fl_type = F_RDLCK;
6499                         break;
6500                 case NFS4_WRITEW_LT:
6501                         if (nfsd4_has_session(cstate))
6502                                 fl_flags |= FL_SLEEP;
6503                         /* Fallthrough */
6504                 case NFS4_WRITE_LT:
6505                         spin_lock(&fp->fi_lock);
6506                         nf = find_writeable_file_locked(fp);
6507                         if (nf)
6508                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6509                         spin_unlock(&fp->fi_lock);
6510                         fl_type = F_WRLCK;
6511                         break;
6512                 default:
6513                         status = nfserr_inval;
6514                 goto out;
6515         }
6516
6517         if (!nf) {
6518                 status = nfserr_openmode;
6519                 goto out;
6520         }
6521
6522         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6523         if (!nbl) {
6524                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6525                 status = nfserr_jukebox;
6526                 goto out;
6527         }
6528
6529         file_lock = &nbl->nbl_lock;
6530         file_lock->fl_type = fl_type;
6531         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6532         file_lock->fl_pid = current->tgid;
6533         file_lock->fl_file = nf->nf_file;
6534         file_lock->fl_flags = fl_flags;
6535         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6536         file_lock->fl_start = lock->lk_offset;
6537         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6538         nfs4_transform_lock_offset(file_lock);
6539
6540         conflock = locks_alloc_lock();
6541         if (!conflock) {
6542                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6543                 status = nfserr_jukebox;
6544                 goto out;
6545         }
6546
6547         if (fl_flags & FL_SLEEP) {
6548                 nbl->nbl_time = jiffies;
6549                 spin_lock(&nn->blocked_locks_lock);
6550                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6551                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6552                 spin_unlock(&nn->blocked_locks_lock);
6553         }
6554
6555         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6556         switch (err) {
6557         case 0: /* success! */
6558                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6559                 status = 0;
6560                 if (lock->lk_reclaim)
6561                         nn->somebody_reclaimed = true;
6562                 break;
6563         case FILE_LOCK_DEFERRED:
6564                 nbl = NULL;
6565                 /* Fallthrough */
6566         case -EAGAIN:           /* conflock holds conflicting lock */
6567                 status = nfserr_denied;
6568                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6569                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6570                 break;
6571         case -EDEADLK:
6572                 status = nfserr_deadlock;
6573                 break;
6574         default:
6575                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6576                 status = nfserrno(err);
6577                 break;
6578         }
6579 out:
6580         if (nbl) {
6581                 /* dequeue it if we queued it before */
6582                 if (fl_flags & FL_SLEEP) {
6583                         spin_lock(&nn->blocked_locks_lock);
6584                         list_del_init(&nbl->nbl_list);
6585                         list_del_init(&nbl->nbl_lru);
6586                         spin_unlock(&nn->blocked_locks_lock);
6587                 }
6588                 free_blocked_lock(nbl);
6589         }
6590         if (nf)
6591                 nfsd_file_put(nf);
6592         if (lock_stp) {
6593                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6594                 if (cstate->replay_owner &&
6595                     cstate->replay_owner != &lock_sop->lo_owner &&
6596                     seqid_mutating_err(ntohl(status)))
6597                         lock_sop->lo_owner.so_seqid++;
6598
6599                 /*
6600                  * If this is a new, never-before-used stateid, and we are
6601                  * returning an error, then just go ahead and release it.
6602                  */
6603                 if (status && new)
6604                         release_lock_stateid(lock_stp);
6605
6606                 mutex_unlock(&lock_stp->st_mutex);
6607
6608                 nfs4_put_stid(&lock_stp->st_stid);
6609         }
6610         if (open_stp)
6611                 nfs4_put_stid(&open_stp->st_stid);
6612         nfsd4_bump_seqid(cstate, status);
6613         if (conflock)
6614                 locks_free_lock(conflock);
6615         return status;
6616 }
6617
6618 /*
6619  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6620  * so we do a temporary open here just to get an open file to pass to
6621  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6622  * inode operation.)
6623  */
6624 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6625 {
6626         struct nfsd_file *nf;
6627         __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6628         if (!err) {
6629                 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6630                 nfsd_file_put(nf);
6631         }
6632         return err;
6633 }
6634
6635 /*
6636  * LOCKT operation
6637  */
6638 __be32
6639 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6640             union nfsd4_op_u *u)
6641 {
6642         struct nfsd4_lockt *lockt = &u->lockt;
6643         struct file_lock *file_lock = NULL;
6644         struct nfs4_lockowner *lo = NULL;
6645         __be32 status;
6646         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6647
6648         if (locks_in_grace(SVC_NET(rqstp)))
6649                 return nfserr_grace;
6650
6651         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6652                  return nfserr_inval;
6653
6654         if (!nfsd4_has_session(cstate)) {
6655                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6656                 if (status)
6657                         goto out;
6658         }
6659
6660         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6661                 goto out;
6662
6663         file_lock = locks_alloc_lock();
6664         if (!file_lock) {
6665                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6666                 status = nfserr_jukebox;
6667                 goto out;
6668         }
6669
6670         switch (lockt->lt_type) {
6671                 case NFS4_READ_LT:
6672                 case NFS4_READW_LT:
6673                         file_lock->fl_type = F_RDLCK;
6674                         break;
6675                 case NFS4_WRITE_LT:
6676                 case NFS4_WRITEW_LT:
6677                         file_lock->fl_type = F_WRLCK;
6678                         break;
6679                 default:
6680                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6681                         status = nfserr_inval;
6682                         goto out;
6683         }
6684
6685         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6686         if (lo)
6687                 file_lock->fl_owner = (fl_owner_t)lo;
6688         file_lock->fl_pid = current->tgid;
6689         file_lock->fl_flags = FL_POSIX;
6690
6691         file_lock->fl_start = lockt->lt_offset;
6692         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6693
6694         nfs4_transform_lock_offset(file_lock);
6695
6696         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6697         if (status)
6698                 goto out;
6699
6700         if (file_lock->fl_type != F_UNLCK) {
6701                 status = nfserr_denied;
6702                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6703         }
6704 out:
6705         if (lo)
6706                 nfs4_put_stateowner(&lo->lo_owner);
6707         if (file_lock)
6708                 locks_free_lock(file_lock);
6709         return status;
6710 }
6711
6712 __be32
6713 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6714             union nfsd4_op_u *u)
6715 {
6716         struct nfsd4_locku *locku = &u->locku;
6717         struct nfs4_ol_stateid *stp;
6718         struct nfsd_file *nf = NULL;
6719         struct file_lock *file_lock = NULL;
6720         __be32 status;
6721         int err;
6722         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6723
6724         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6725                 (long long) locku->lu_offset,
6726                 (long long) locku->lu_length);
6727
6728         if (check_lock_length(locku->lu_offset, locku->lu_length))
6729                  return nfserr_inval;
6730
6731         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6732                                         &locku->lu_stateid, NFS4_LOCK_STID,
6733                                         &stp, nn);
6734         if (status)
6735                 goto out;
6736         nf = find_any_file(stp->st_stid.sc_file);
6737         if (!nf) {
6738                 status = nfserr_lock_range;
6739                 goto put_stateid;
6740         }
6741         file_lock = locks_alloc_lock();
6742         if (!file_lock) {
6743                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6744                 status = nfserr_jukebox;
6745                 goto put_file;
6746         }
6747
6748         file_lock->fl_type = F_UNLCK;
6749         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6750         file_lock->fl_pid = current->tgid;
6751         file_lock->fl_file = nf->nf_file;
6752         file_lock->fl_flags = FL_POSIX;
6753         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6754         file_lock->fl_start = locku->lu_offset;
6755
6756         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6757                                                 locku->lu_length);
6758         nfs4_transform_lock_offset(file_lock);
6759
6760         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
6761         if (err) {
6762                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6763                 goto out_nfserr;
6764         }
6765         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6766 put_file:
6767         nfsd_file_put(nf);
6768 put_stateid:
6769         mutex_unlock(&stp->st_mutex);
6770         nfs4_put_stid(&stp->st_stid);
6771 out:
6772         nfsd4_bump_seqid(cstate, status);
6773         if (file_lock)
6774                 locks_free_lock(file_lock);
6775         return status;
6776
6777 out_nfserr:
6778         status = nfserrno(err);
6779         goto put_file;
6780 }
6781
6782 /*
6783  * returns
6784  *      true:  locks held by lockowner
6785  *      false: no locks held by lockowner
6786  */
6787 static bool
6788 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6789 {
6790         struct file_lock *fl;
6791         int status = false;
6792         struct nfsd_file *nf = find_any_file(fp);
6793         struct inode *inode;
6794         struct file_lock_context *flctx;
6795
6796         if (!nf) {
6797                 /* Any valid lock stateid should have some sort of access */
6798                 WARN_ON_ONCE(1);
6799                 return status;
6800         }
6801
6802         inode = locks_inode(nf->nf_file);
6803         flctx = inode->i_flctx;
6804
6805         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6806                 spin_lock(&flctx->flc_lock);
6807                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6808                         if (fl->fl_owner == (fl_owner_t)lowner) {
6809                                 status = true;
6810                                 break;
6811                         }
6812                 }
6813                 spin_unlock(&flctx->flc_lock);
6814         }
6815         nfsd_file_put(nf);
6816         return status;
6817 }
6818
6819 __be32
6820 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6821                         struct nfsd4_compound_state *cstate,
6822                         union nfsd4_op_u *u)
6823 {
6824         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6825         clientid_t *clid = &rlockowner->rl_clientid;
6826         struct nfs4_stateowner *sop;
6827         struct nfs4_lockowner *lo = NULL;
6828         struct nfs4_ol_stateid *stp;
6829         struct xdr_netobj *owner = &rlockowner->rl_owner;
6830         unsigned int hashval = ownerstr_hashval(owner);
6831         __be32 status;
6832         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6833         struct nfs4_client *clp;
6834         LIST_HEAD (reaplist);
6835
6836         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6837                 clid->cl_boot, clid->cl_id);
6838
6839         status = lookup_clientid(clid, cstate, nn);
6840         if (status)
6841                 return status;
6842
6843         clp = cstate->clp;
6844         /* Find the matching lock stateowner */
6845         spin_lock(&clp->cl_lock);
6846         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6847                             so_strhash) {
6848
6849                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6850                         continue;
6851
6852                 /* see if there are still any locks associated with it */
6853                 lo = lockowner(sop);
6854                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6855                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6856                                 status = nfserr_locks_held;
6857                                 spin_unlock(&clp->cl_lock);
6858                                 return status;
6859                         }
6860                 }
6861
6862                 nfs4_get_stateowner(sop);
6863                 break;
6864         }
6865         if (!lo) {
6866                 spin_unlock(&clp->cl_lock);
6867                 return status;
6868         }
6869
6870         unhash_lockowner_locked(lo);
6871         while (!list_empty(&lo->lo_owner.so_stateids)) {
6872                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6873                                        struct nfs4_ol_stateid,
6874                                        st_perstateowner);
6875                 WARN_ON(!unhash_lock_stateid(stp));
6876                 put_ol_stateid_locked(stp, &reaplist);
6877         }
6878         spin_unlock(&clp->cl_lock);
6879         free_ol_stateid_reaplist(&reaplist);
6880         remove_blocked_locks(lo);
6881         nfs4_put_stateowner(&lo->lo_owner);
6882
6883         return status;
6884 }
6885
6886 static inline struct nfs4_client_reclaim *
6887 alloc_reclaim(void)
6888 {
6889         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6890 }
6891
6892 bool
6893 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
6894 {
6895         struct nfs4_client_reclaim *crp;
6896
6897         crp = nfsd4_find_reclaim_client(name, nn);
6898         return (crp && crp->cr_clp);
6899 }
6900
6901 /*
6902  * failure => all reset bets are off, nfserr_no_grace...
6903  *
6904  * The caller is responsible for freeing name.data if NULL is returned (it
6905  * will be freed in nfs4_remove_reclaim_record in the normal case).
6906  */
6907 struct nfs4_client_reclaim *
6908 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
6909                 struct nfsd_net *nn)
6910 {
6911         unsigned int strhashval;
6912         struct nfs4_client_reclaim *crp;
6913
6914         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
6915         crp = alloc_reclaim();
6916         if (crp) {
6917                 strhashval = clientstr_hashval(name);
6918                 INIT_LIST_HEAD(&crp->cr_strhash);
6919                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6920                 crp->cr_name.data = name.data;
6921                 crp->cr_name.len = name.len;
6922                 crp->cr_princhash.data = princhash.data;
6923                 crp->cr_princhash.len = princhash.len;
6924                 crp->cr_clp = NULL;
6925                 nn->reclaim_str_hashtbl_size++;
6926         }
6927         return crp;
6928 }
6929
6930 void
6931 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6932 {
6933         list_del(&crp->cr_strhash);
6934         kfree(crp->cr_name.data);
6935         kfree(crp->cr_princhash.data);
6936         kfree(crp);
6937         nn->reclaim_str_hashtbl_size--;
6938 }
6939
6940 void
6941 nfs4_release_reclaim(struct nfsd_net *nn)
6942 {
6943         struct nfs4_client_reclaim *crp = NULL;
6944         int i;
6945
6946         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6947                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6948                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6949                                         struct nfs4_client_reclaim, cr_strhash);
6950                         nfs4_remove_reclaim_record(crp, nn);
6951                 }
6952         }
6953         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6954 }
6955
6956 /*
6957  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6958 struct nfs4_client_reclaim *
6959 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
6960 {
6961         unsigned int strhashval;
6962         struct nfs4_client_reclaim *crp = NULL;
6963
6964         dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
6965
6966         strhashval = clientstr_hashval(name);
6967         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6968                 if (compare_blob(&crp->cr_name, &name) == 0) {
6969                         return crp;
6970                 }
6971         }
6972         return NULL;
6973 }
6974
6975 /*
6976 * Called from OPEN. Look for clientid in reclaim list.
6977 */
6978 __be32
6979 nfs4_check_open_reclaim(clientid_t *clid,
6980                 struct nfsd4_compound_state *cstate,
6981                 struct nfsd_net *nn)
6982 {
6983         __be32 status;
6984
6985         /* find clientid in conf_id_hashtbl */
6986         status = lookup_clientid(clid, cstate, nn);
6987         if (status)
6988                 return nfserr_reclaim_bad;
6989
6990         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6991                 return nfserr_no_grace;
6992
6993         if (nfsd4_client_record_check(cstate->clp))
6994                 return nfserr_reclaim_bad;
6995
6996         return nfs_ok;
6997 }
6998
6999 #ifdef CONFIG_NFSD_FAULT_INJECTION
7000 static inline void
7001 put_client(struct nfs4_client *clp)
7002 {
7003         atomic_dec(&clp->cl_rpc_users);
7004 }
7005
7006 static struct nfs4_client *
7007 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
7008 {
7009         struct nfs4_client *clp;
7010         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7011                                           nfsd_net_id);
7012
7013         if (!nfsd_netns_ready(nn))
7014                 return NULL;
7015
7016         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7017                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
7018                         return clp;
7019         }
7020         return NULL;
7021 }
7022
7023 u64
7024 nfsd_inject_print_clients(void)
7025 {
7026         struct nfs4_client *clp;
7027         u64 count = 0;
7028         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7029                                           nfsd_net_id);
7030         char buf[INET6_ADDRSTRLEN];
7031
7032         if (!nfsd_netns_ready(nn))
7033                 return 0;
7034
7035         spin_lock(&nn->client_lock);
7036         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7037                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7038                 pr_info("NFS Client: %s\n", buf);
7039                 ++count;
7040         }
7041         spin_unlock(&nn->client_lock);
7042
7043         return count;
7044 }
7045
7046 u64
7047 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7048 {
7049         u64 count = 0;
7050         struct nfs4_client *clp;
7051         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7052                                           nfsd_net_id);
7053
7054         if (!nfsd_netns_ready(nn))
7055                 return count;
7056
7057         spin_lock(&nn->client_lock);
7058         clp = nfsd_find_client(addr, addr_size);
7059         if (clp) {
7060                 if (mark_client_expired_locked(clp) == nfs_ok)
7061                         ++count;
7062                 else
7063                         clp = NULL;
7064         }
7065         spin_unlock(&nn->client_lock);
7066
7067         if (clp)
7068                 expire_client(clp);
7069
7070         return count;
7071 }
7072
7073 u64
7074 nfsd_inject_forget_clients(u64 max)
7075 {
7076         u64 count = 0;
7077         struct nfs4_client *clp, *next;
7078         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7079                                                 nfsd_net_id);
7080         LIST_HEAD(reaplist);
7081
7082         if (!nfsd_netns_ready(nn))
7083                 return count;
7084
7085         spin_lock(&nn->client_lock);
7086         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7087                 if (mark_client_expired_locked(clp) == nfs_ok) {
7088                         list_add(&clp->cl_lru, &reaplist);
7089                         if (max != 0 && ++count >= max)
7090                                 break;
7091                 }
7092         }
7093         spin_unlock(&nn->client_lock);
7094
7095         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7096                 expire_client(clp);
7097
7098         return count;
7099 }
7100
7101 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7102                              const char *type)
7103 {
7104         char buf[INET6_ADDRSTRLEN];
7105         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7106         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7107 }
7108
7109 static void
7110 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7111                              struct list_head *collect)
7112 {
7113         struct nfs4_client *clp = lst->st_stid.sc_client;
7114         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7115                                           nfsd_net_id);
7116
7117         if (!collect)
7118                 return;
7119
7120         lockdep_assert_held(&nn->client_lock);
7121         atomic_inc(&clp->cl_rpc_users);
7122         list_add(&lst->st_locks, collect);
7123 }
7124
7125 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7126                                     struct list_head *collect,
7127                                     bool (*func)(struct nfs4_ol_stateid *))
7128 {
7129         struct nfs4_openowner *oop;
7130         struct nfs4_ol_stateid *stp, *st_next;
7131         struct nfs4_ol_stateid *lst, *lst_next;
7132         u64 count = 0;
7133
7134         spin_lock(&clp->cl_lock);
7135         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7136                 list_for_each_entry_safe(stp, st_next,
7137                                 &oop->oo_owner.so_stateids, st_perstateowner) {
7138                         list_for_each_entry_safe(lst, lst_next,
7139                                         &stp->st_locks, st_locks) {
7140                                 if (func) {
7141                                         if (func(lst))
7142                                                 nfsd_inject_add_lock_to_list(lst,
7143                                                                         collect);
7144                                 }
7145                                 ++count;
7146                                 /*
7147                                  * Despite the fact that these functions deal
7148                                  * with 64-bit integers for "count", we must
7149                                  * ensure that it doesn't blow up the
7150                                  * clp->cl_rpc_users. Throw a warning if we
7151                                  * start to approach INT_MAX here.
7152                                  */
7153                                 WARN_ON_ONCE(count == (INT_MAX / 2));
7154                                 if (count == max)
7155                                         goto out;
7156                         }
7157                 }
7158         }
7159 out:
7160         spin_unlock(&clp->cl_lock);
7161
7162         return count;
7163 }
7164
7165 static u64
7166 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7167                           u64 max)
7168 {
7169         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7170 }
7171
7172 static u64
7173 nfsd_print_client_locks(struct nfs4_client *clp)
7174 {
7175         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7176         nfsd_print_count(clp, count, "locked files");
7177         return count;
7178 }
7179
7180 u64
7181 nfsd_inject_print_locks(void)
7182 {
7183         struct nfs4_client *clp;
7184         u64 count = 0;
7185         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7186                                                 nfsd_net_id);
7187
7188         if (!nfsd_netns_ready(nn))
7189                 return 0;
7190
7191         spin_lock(&nn->client_lock);
7192         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7193                 count += nfsd_print_client_locks(clp);
7194         spin_unlock(&nn->client_lock);
7195
7196         return count;
7197 }
7198
7199 static void
7200 nfsd_reap_locks(struct list_head *reaplist)
7201 {
7202         struct nfs4_client *clp;
7203         struct nfs4_ol_stateid *stp, *next;
7204
7205         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7206                 list_del_init(&stp->st_locks);
7207                 clp = stp->st_stid.sc_client;
7208                 nfs4_put_stid(&stp->st_stid);
7209                 put_client(clp);
7210         }
7211 }
7212
7213 u64
7214 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7215 {
7216         unsigned int 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         clp = nfsd_find_client(addr, addr_size);
7227         if (clp)
7228                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7229         spin_unlock(&nn->client_lock);
7230         nfsd_reap_locks(&reaplist);
7231         return count;
7232 }
7233
7234 u64
7235 nfsd_inject_forget_locks(u64 max)
7236 {
7237         u64 count = 0;
7238         struct nfs4_client *clp;
7239         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7240                                                 nfsd_net_id);
7241         LIST_HEAD(reaplist);
7242
7243         if (!nfsd_netns_ready(nn))
7244                 return count;
7245
7246         spin_lock(&nn->client_lock);
7247         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7248                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7249                 if (max != 0 && count >= max)
7250                         break;
7251         }
7252         spin_unlock(&nn->client_lock);
7253         nfsd_reap_locks(&reaplist);
7254         return count;
7255 }
7256
7257 static u64
7258 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7259                               struct list_head *collect,
7260                               void (*func)(struct nfs4_openowner *))
7261 {
7262         struct nfs4_openowner *oop, *next;
7263         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7264                                                 nfsd_net_id);
7265         u64 count = 0;
7266
7267         lockdep_assert_held(&nn->client_lock);
7268
7269         spin_lock(&clp->cl_lock);
7270         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7271                 if (func) {
7272                         func(oop);
7273                         if (collect) {
7274                                 atomic_inc(&clp->cl_rpc_users);
7275                                 list_add(&oop->oo_perclient, collect);
7276                         }
7277                 }
7278                 ++count;
7279                 /*
7280                  * Despite the fact that these functions deal with
7281                  * 64-bit integers for "count", we must ensure that
7282                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7283                  * warning if we start to approach INT_MAX here.
7284                  */
7285                 WARN_ON_ONCE(count == (INT_MAX / 2));
7286                 if (count == max)
7287                         break;
7288         }
7289         spin_unlock(&clp->cl_lock);
7290
7291         return count;
7292 }
7293
7294 static u64
7295 nfsd_print_client_openowners(struct nfs4_client *clp)
7296 {
7297         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7298
7299         nfsd_print_count(clp, count, "openowners");
7300         return count;
7301 }
7302
7303 static u64
7304 nfsd_collect_client_openowners(struct nfs4_client *clp,
7305                                struct list_head *collect, u64 max)
7306 {
7307         return nfsd_foreach_client_openowner(clp, max, collect,
7308                                                 unhash_openowner_locked);
7309 }
7310
7311 u64
7312 nfsd_inject_print_openowners(void)
7313 {
7314         struct nfs4_client *clp;
7315         u64 count = 0;
7316         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7317                                                 nfsd_net_id);
7318
7319         if (!nfsd_netns_ready(nn))
7320                 return 0;
7321
7322         spin_lock(&nn->client_lock);
7323         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7324                 count += nfsd_print_client_openowners(clp);
7325         spin_unlock(&nn->client_lock);
7326
7327         return count;
7328 }
7329
7330 static void
7331 nfsd_reap_openowners(struct list_head *reaplist)
7332 {
7333         struct nfs4_client *clp;
7334         struct nfs4_openowner *oop, *next;
7335
7336         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7337                 list_del_init(&oop->oo_perclient);
7338                 clp = oop->oo_owner.so_client;
7339                 release_openowner(oop);
7340                 put_client(clp);
7341         }
7342 }
7343
7344 u64
7345 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7346                                      size_t addr_size)
7347 {
7348         unsigned int 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         clp = nfsd_find_client(addr, addr_size);
7359         if (clp)
7360                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7361         spin_unlock(&nn->client_lock);
7362         nfsd_reap_openowners(&reaplist);
7363         return count;
7364 }
7365
7366 u64
7367 nfsd_inject_forget_openowners(u64 max)
7368 {
7369         u64 count = 0;
7370         struct nfs4_client *clp;
7371         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7372                                                 nfsd_net_id);
7373         LIST_HEAD(reaplist);
7374
7375         if (!nfsd_netns_ready(nn))
7376                 return count;
7377
7378         spin_lock(&nn->client_lock);
7379         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7380                 count += nfsd_collect_client_openowners(clp, &reaplist,
7381                                                         max - count);
7382                 if (max != 0 && count >= max)
7383                         break;
7384         }
7385         spin_unlock(&nn->client_lock);
7386         nfsd_reap_openowners(&reaplist);
7387         return count;
7388 }
7389
7390 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7391                                      struct list_head *victims)
7392 {
7393         struct nfs4_delegation *dp, *next;
7394         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7395                                                 nfsd_net_id);
7396         u64 count = 0;
7397
7398         lockdep_assert_held(&nn->client_lock);
7399
7400         spin_lock(&state_lock);
7401         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7402                 if (victims) {
7403                         /*
7404                          * It's not safe to mess with delegations that have a
7405                          * non-zero dl_time. They might have already been broken
7406                          * and could be processed by the laundromat outside of
7407                          * the state_lock. Just leave them be.
7408                          */
7409                         if (dp->dl_time != 0)
7410                                 continue;
7411
7412                         atomic_inc(&clp->cl_rpc_users);
7413                         WARN_ON(!unhash_delegation_locked(dp));
7414                         list_add(&dp->dl_recall_lru, victims);
7415                 }
7416                 ++count;
7417                 /*
7418                  * Despite the fact that these functions deal with
7419                  * 64-bit integers for "count", we must ensure that
7420                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7421                  * warning if we start to approach INT_MAX here.
7422                  */
7423                 WARN_ON_ONCE(count == (INT_MAX / 2));
7424                 if (count == max)
7425                         break;
7426         }
7427         spin_unlock(&state_lock);
7428         return count;
7429 }
7430
7431 static u64
7432 nfsd_print_client_delegations(struct nfs4_client *clp)
7433 {
7434         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7435
7436         nfsd_print_count(clp, count, "delegations");
7437         return count;
7438 }
7439
7440 u64
7441 nfsd_inject_print_delegations(void)
7442 {
7443         struct nfs4_client *clp;
7444         u64 count = 0;
7445         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7446                                                 nfsd_net_id);
7447
7448         if (!nfsd_netns_ready(nn))
7449                 return 0;
7450
7451         spin_lock(&nn->client_lock);
7452         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7453                 count += nfsd_print_client_delegations(clp);
7454         spin_unlock(&nn->client_lock);
7455
7456         return count;
7457 }
7458
7459 static void
7460 nfsd_forget_delegations(struct list_head *reaplist)
7461 {
7462         struct nfs4_client *clp;
7463         struct nfs4_delegation *dp, *next;
7464
7465         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7466                 list_del_init(&dp->dl_recall_lru);
7467                 clp = dp->dl_stid.sc_client;
7468                 revoke_delegation(dp);
7469                 put_client(clp);
7470         }
7471 }
7472
7473 u64
7474 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7475                                       size_t addr_size)
7476 {
7477         u64 count = 0;
7478         struct nfs4_client *clp;
7479         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7480                                                 nfsd_net_id);
7481         LIST_HEAD(reaplist);
7482
7483         if (!nfsd_netns_ready(nn))
7484                 return count;
7485
7486         spin_lock(&nn->client_lock);
7487         clp = nfsd_find_client(addr, addr_size);
7488         if (clp)
7489                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7490         spin_unlock(&nn->client_lock);
7491
7492         nfsd_forget_delegations(&reaplist);
7493         return count;
7494 }
7495
7496 u64
7497 nfsd_inject_forget_delegations(u64 max)
7498 {
7499         u64 count = 0;
7500         struct nfs4_client *clp;
7501         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7502                                                 nfsd_net_id);
7503         LIST_HEAD(reaplist);
7504
7505         if (!nfsd_netns_ready(nn))
7506                 return count;
7507
7508         spin_lock(&nn->client_lock);
7509         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7510                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7511                 if (max != 0 && count >= max)
7512                         break;
7513         }
7514         spin_unlock(&nn->client_lock);
7515         nfsd_forget_delegations(&reaplist);
7516         return count;
7517 }
7518
7519 static void
7520 nfsd_recall_delegations(struct list_head *reaplist)
7521 {
7522         struct nfs4_client *clp;
7523         struct nfs4_delegation *dp, *next;
7524
7525         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7526                 list_del_init(&dp->dl_recall_lru);
7527                 clp = dp->dl_stid.sc_client;
7528                 /*
7529                  * We skipped all entries that had a zero dl_time before,
7530                  * so we can now reset the dl_time back to 0. If a delegation
7531                  * break comes in now, then it won't make any difference since
7532                  * we're recalling it either way.
7533                  */
7534                 spin_lock(&state_lock);
7535                 dp->dl_time = 0;
7536                 spin_unlock(&state_lock);
7537                 nfsd_break_one_deleg(dp);
7538                 put_client(clp);
7539         }
7540 }
7541
7542 u64
7543 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7544                                       size_t addr_size)
7545 {
7546         u64 count = 0;
7547         struct nfs4_client *clp;
7548         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7549                                                 nfsd_net_id);
7550         LIST_HEAD(reaplist);
7551
7552         if (!nfsd_netns_ready(nn))
7553                 return count;
7554
7555         spin_lock(&nn->client_lock);
7556         clp = nfsd_find_client(addr, addr_size);
7557         if (clp)
7558                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7559         spin_unlock(&nn->client_lock);
7560
7561         nfsd_recall_delegations(&reaplist);
7562         return count;
7563 }
7564
7565 u64
7566 nfsd_inject_recall_delegations(u64 max)
7567 {
7568         u64 count = 0;
7569         struct nfs4_client *clp, *next;
7570         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7571                                                 nfsd_net_id);
7572         LIST_HEAD(reaplist);
7573
7574         if (!nfsd_netns_ready(nn))
7575                 return count;
7576
7577         spin_lock(&nn->client_lock);
7578         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7579                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7580                 if (max != 0 && ++count >= max)
7581                         break;
7582         }
7583         spin_unlock(&nn->client_lock);
7584         nfsd_recall_delegations(&reaplist);
7585         return count;
7586 }
7587 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7588
7589 /*
7590  * Since the lifetime of a delegation isn't limited to that of an open, a
7591  * client may quite reasonably hang on to a delegation as long as it has
7592  * the inode cached.  This becomes an obvious problem the first time a
7593  * client's inode cache approaches the size of the server's total memory.
7594  *
7595  * For now we avoid this problem by imposing a hard limit on the number
7596  * of delegations, which varies according to the server's memory size.
7597  */
7598 static void
7599 set_max_delegations(void)
7600 {
7601         /*
7602          * Allow at most 4 delegations per megabyte of RAM.  Quick
7603          * estimates suggest that in the worst case (where every delegation
7604          * is for a different inode), a delegation could take about 1.5K,
7605          * giving a worst case usage of about 6% of memory.
7606          */
7607         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7608 }
7609
7610 static int nfs4_state_create_net(struct net *net)
7611 {
7612         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7613         int i;
7614
7615         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7616                                             sizeof(struct list_head),
7617                                             GFP_KERNEL);
7618         if (!nn->conf_id_hashtbl)
7619                 goto err;
7620         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7621                                               sizeof(struct list_head),
7622                                               GFP_KERNEL);
7623         if (!nn->unconf_id_hashtbl)
7624                 goto err_unconf_id;
7625         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7626                                               sizeof(struct list_head),
7627                                               GFP_KERNEL);
7628         if (!nn->sessionid_hashtbl)
7629                 goto err_sessionid;
7630
7631         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7632                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7633                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7634         }
7635         for (i = 0; i < SESSION_HASH_SIZE; i++)
7636                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7637         nn->conf_name_tree = RB_ROOT;
7638         nn->unconf_name_tree = RB_ROOT;
7639         nn->boot_time = get_seconds();
7640         nn->grace_ended = false;
7641         nn->nfsd4_manager.block_opens = true;
7642         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7643         INIT_LIST_HEAD(&nn->client_lru);
7644         INIT_LIST_HEAD(&nn->close_lru);
7645         INIT_LIST_HEAD(&nn->del_recall_lru);
7646         spin_lock_init(&nn->client_lock);
7647         spin_lock_init(&nn->s2s_cp_lock);
7648         idr_init(&nn->s2s_cp_stateids);
7649
7650         spin_lock_init(&nn->blocked_locks_lock);
7651         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7652
7653         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7654         get_net(net);
7655
7656         return 0;
7657
7658 err_sessionid:
7659         kfree(nn->unconf_id_hashtbl);
7660 err_unconf_id:
7661         kfree(nn->conf_id_hashtbl);
7662 err:
7663         return -ENOMEM;
7664 }
7665
7666 static void
7667 nfs4_state_destroy_net(struct net *net)
7668 {
7669         int i;
7670         struct nfs4_client *clp = NULL;
7671         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7672
7673         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7674                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7675                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7676                         destroy_client(clp);
7677                 }
7678         }
7679
7680         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7681
7682         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7683                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7684                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7685                         destroy_client(clp);
7686                 }
7687         }
7688
7689         kfree(nn->sessionid_hashtbl);
7690         kfree(nn->unconf_id_hashtbl);
7691         kfree(nn->conf_id_hashtbl);
7692         put_net(net);
7693 }
7694
7695 int
7696 nfs4_state_start_net(struct net *net)
7697 {
7698         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7699         int ret;
7700
7701         ret = nfs4_state_create_net(net);
7702         if (ret)
7703                 return ret;
7704         locks_start_grace(net, &nn->nfsd4_manager);
7705         nfsd4_client_tracking_init(net);
7706         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7707                 goto skip_grace;
7708         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7709                nn->nfsd4_grace, net->ns.inum);
7710         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7711         return 0;
7712
7713 skip_grace:
7714         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7715                         net->ns.inum);
7716         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7717         nfsd4_end_grace(nn);
7718         return 0;
7719 }
7720
7721 /* initialization to perform when the nfsd service is started: */
7722
7723 int
7724 nfs4_state_start(void)
7725 {
7726         int ret;
7727
7728         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7729         if (laundry_wq == NULL) {
7730                 ret = -ENOMEM;
7731                 goto out;
7732         }
7733         ret = nfsd4_create_callback_queue();
7734         if (ret)
7735                 goto out_free_laundry;
7736
7737         set_max_delegations();
7738         return 0;
7739
7740 out_free_laundry:
7741         destroy_workqueue(laundry_wq);
7742 out:
7743         return ret;
7744 }
7745
7746 void
7747 nfs4_state_shutdown_net(struct net *net)
7748 {
7749         struct nfs4_delegation *dp = NULL;
7750         struct list_head *pos, *next, reaplist;
7751         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7752
7753         cancel_delayed_work_sync(&nn->laundromat_work);
7754         locks_end_grace(&nn->nfsd4_manager);
7755
7756         INIT_LIST_HEAD(&reaplist);
7757         spin_lock(&state_lock);
7758         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7759                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7760                 WARN_ON(!unhash_delegation_locked(dp));
7761                 list_add(&dp->dl_recall_lru, &reaplist);
7762         }
7763         spin_unlock(&state_lock);
7764         list_for_each_safe(pos, next, &reaplist) {
7765                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7766                 list_del_init(&dp->dl_recall_lru);
7767                 destroy_unhashed_deleg(dp);
7768         }
7769
7770         nfsd4_client_tracking_exit(net);
7771         nfs4_state_destroy_net(net);
7772 }
7773
7774 void
7775 nfs4_state_shutdown(void)
7776 {
7777         destroy_workqueue(laundry_wq);
7778         nfsd4_destroy_callback_queue();
7779 }
7780
7781 static void
7782 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7783 {
7784         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7785                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7786 }
7787
7788 static void
7789 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7790 {
7791         if (cstate->minorversion) {
7792                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7793                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7794         }
7795 }
7796
7797 void
7798 clear_current_stateid(struct nfsd4_compound_state *cstate)
7799 {
7800         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7801 }
7802
7803 /*
7804  * functions to set current state id
7805  */
7806 void
7807 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7808                 union nfsd4_op_u *u)
7809 {
7810         put_stateid(cstate, &u->open_downgrade.od_stateid);
7811 }
7812
7813 void
7814 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7815                 union nfsd4_op_u *u)
7816 {
7817         put_stateid(cstate, &u->open.op_stateid);
7818 }
7819
7820 void
7821 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7822                 union nfsd4_op_u *u)
7823 {
7824         put_stateid(cstate, &u->close.cl_stateid);
7825 }
7826
7827 void
7828 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7829                 union nfsd4_op_u *u)
7830 {
7831         put_stateid(cstate, &u->lock.lk_resp_stateid);
7832 }
7833
7834 /*
7835  * functions to consume current state id
7836  */
7837
7838 void
7839 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7840                 union nfsd4_op_u *u)
7841 {
7842         get_stateid(cstate, &u->open_downgrade.od_stateid);
7843 }
7844
7845 void
7846 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7847                 union nfsd4_op_u *u)
7848 {
7849         get_stateid(cstate, &u->delegreturn.dr_stateid);
7850 }
7851
7852 void
7853 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7854                 union nfsd4_op_u *u)
7855 {
7856         get_stateid(cstate, &u->free_stateid.fr_stateid);
7857 }
7858
7859 void
7860 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7861                 union nfsd4_op_u *u)
7862 {
7863         get_stateid(cstate, &u->setattr.sa_stateid);
7864 }
7865
7866 void
7867 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7868                 union nfsd4_op_u *u)
7869 {
7870         get_stateid(cstate, &u->close.cl_stateid);
7871 }
7872
7873 void
7874 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7875                 union nfsd4_op_u *u)
7876 {
7877         get_stateid(cstate, &u->locku.lu_stateid);
7878 }
7879
7880 void
7881 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7882                 union nfsd4_op_u *u)
7883 {
7884         get_stateid(cstate, &u->read.rd_stateid);
7885 }
7886
7887 void
7888 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7889                 union nfsd4_op_u *u)
7890 {
7891         get_stateid(cstate, &u->write.wr_stateid);
7892 }