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                         INIT_LIST_HEAD(&nbl->nbl_list);
 270                         INIT_LIST_HEAD(&nbl->nbl_lru);
 271                         fh_copy_shallow(&nbl->nbl_fh, fh);
 272                         locks_init_lock(&nbl->nbl_lock);
 273                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 274                                         &nfsd4_cb_notify_lock_ops,
 275                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 276                 }
 277         }
 278         return nbl;
 279 }
 280
 281 static void
 282 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 283 {
 284         locks_delete_block(&nbl->nbl_lock);
 285         locks_release_private(&nbl->nbl_lock);
 286         kfree(nbl);
 287 }
 288
 289 static void
 290 remove_blocked_locks(struct nfs4_lockowner *lo)
 291 {
 292         struct nfs4_client *clp = lo->lo_owner.so_client;
 293         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 294         struct nfsd4_blocked_lock *nbl;
 295         LIST_HEAD(reaplist);
 296
 297         /* Dequeue all blocked locks */
 298         spin_lock(&nn->blocked_locks_lock);
 299         while (!list_empty(&lo->lo_blocked)) {
 300                 nbl = list_first_entry(&lo->lo_blocked,
 301                                         struct nfsd4_blocked_lock,
 302                                         nbl_list);
 303                 list_del_init(&nbl->nbl_list);
 304                 list_move(&nbl->nbl_lru, &reaplist);
 305         }
 306         spin_unlock(&nn->blocked_locks_lock);
 307
 308         /* Now free them */
 309         while (!list_empty(&reaplist)) {
 310                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 311                                         nbl_lru);
 312                 list_del_init(&nbl->nbl_lru);
 313                 free_blocked_lock(nbl);
 314         }
 315 }
 316
 317 static void
 318 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 319 {
 320         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 321                                                 struct nfsd4_blocked_lock, nbl_cb);
 322         locks_delete_block(&nbl->nbl_lock);
 323 }
 324
 325 static int
 326 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 327 {
 328         /*
 329          * Since this is just an optimization, we don't try very hard if it
 330          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 331          * just quit trying on anything else.
 332          */
 333         switch (task->tk_status) {
 334         case -NFS4ERR_DELAY:
 335                 rpc_delay(task, 1 * HZ);
 336                 return 0;
 337         default:
 338                 return 1;
 339         }
 340 }
 341
 342 static void
 343 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 344 {
 345         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 346                                                 struct nfsd4_blocked_lock, nbl_cb);
 347
 348         free_blocked_lock(nbl);
 349 }
 350
 351 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 352         .prepare        = nfsd4_cb_notify_lock_prepare,
 353         .done           = nfsd4_cb_notify_lock_done,
 354         .release        = nfsd4_cb_notify_lock_release,
 355 };
 356
 357 static inline struct nfs4_stateowner *
 358 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 359 {
 360         atomic_inc(&sop->so_count);
 361         return sop;
 362 }
 363
 364 static int
 365 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 366 {
 367         return (sop->so_owner.len == owner->len) &&
 368                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 369 }
 370
 371 static struct nfs4_openowner *
 372 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 373                         struct nfs4_client *clp)
 374 {
 375         struct nfs4_stateowner *so;
 376
 377         lockdep_assert_held(&clp->cl_lock);
 378
 379         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 380                             so_strhash) {
 381                 if (!so->so_is_open_owner)
 382                         continue;
 383                 if (same_owner_str(so, &open->op_owner))
 384                         return openowner(nfs4_get_stateowner(so));
 385         }
 386         return NULL;
 387 }
 388
 389 static struct nfs4_openowner *
 390 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 391                         struct nfs4_client *clp)
 392 {
 393         struct nfs4_openowner *oo;
 394
 395         spin_lock(&clp->cl_lock);
 396         oo = find_openstateowner_str_locked(hashval, open, clp);
 397         spin_unlock(&clp->cl_lock);
 398         return oo;
 399 }
 400
 401 static inline u32
 402 opaque_hashval(const void *ptr, int nbytes)
 403 {
 404         unsigned char *cptr = (unsigned char *) ptr;
 405
 406         u32 x = 0;
 407         while (nbytes--) {
 408                 x *= 37;
 409                 x += *cptr++;
 410         }
 411         return x;
 412 }
 413
 414 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 415 {
 416         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 417
 418         kmem_cache_free(file_slab, fp);
 419 }
 420
 421 void
 422 put_nfs4_file(struct nfs4_file *fi)
 423 {
 424         might_lock(&state_lock);
 425
 426         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 427                 hlist_del_rcu(&fi->fi_hash);
 428                 spin_unlock(&state_lock);
 429                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 430                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 431                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 432         }
 433 }
 434
 435 static struct nfsd_file *
 436 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 437 {
 438         if (f->fi_fds[oflag])
 439                 return nfsd_file_get(f->fi_fds[oflag]);
 440         return NULL;
 441 }
 442
 443 static struct nfsd_file *
 444 find_writeable_file_locked(struct nfs4_file *f)
 445 {
 446         struct nfsd_file *ret;
 447
 448         lockdep_assert_held(&f->fi_lock);
 449
 450         ret = __nfs4_get_fd(f, O_WRONLY);
 451         if (!ret)
 452                 ret = __nfs4_get_fd(f, O_RDWR);
 453         return ret;
 454 }
 455
 456 static struct nfsd_file *
 457 find_writeable_file(struct nfs4_file *f)
 458 {
 459         struct nfsd_file *ret;
 460
 461         spin_lock(&f->fi_lock);
 462         ret = find_writeable_file_locked(f);
 463         spin_unlock(&f->fi_lock);
 464
 465         return ret;
 466 }
 467
 468 static struct nfsd_file *
 469 find_readable_file_locked(struct nfs4_file *f)
 470 {
 471         struct nfsd_file *ret;
 472
 473         lockdep_assert_held(&f->fi_lock);
 474
 475         ret = __nfs4_get_fd(f, O_RDONLY);
 476         if (!ret)
 477                 ret = __nfs4_get_fd(f, O_RDWR);
 478         return ret;
 479 }
 480
 481 static struct nfsd_file *
 482 find_readable_file(struct nfs4_file *f)
 483 {
 484         struct nfsd_file *ret;
 485
 486         spin_lock(&f->fi_lock);
 487         ret = find_readable_file_locked(f);
 488         spin_unlock(&f->fi_lock);
 489
 490         return ret;
 491 }
 492
 493 struct nfsd_file *
 494 find_any_file(struct nfs4_file *f)
 495 {
 496         struct nfsd_file *ret;
 497
 498         spin_lock(&f->fi_lock);
 499         ret = __nfs4_get_fd(f, O_RDWR);
 500         if (!ret) {
 501                 ret = __nfs4_get_fd(f, O_WRONLY);
 502                 if (!ret)
 503                         ret = __nfs4_get_fd(f, O_RDONLY);
 504         }
 505         spin_unlock(&f->fi_lock);
 506         return ret;
 507 }
 508
 509 static atomic_long_t num_delegations;
 510 unsigned long max_delegations;
 511
 512 /*
 513  * Open owner state (share locks)
 514  */
 515
 516 /* hash tables for lock and open owners */
 517 #define OWNER_HASH_BITS              8
 518 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 519 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 520
 521 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 522 {
 523         unsigned int ret;
 524
 525         ret = opaque_hashval(ownername->data, ownername->len);
 526         return ret & OWNER_HASH_MASK;
 527 }
 528
 529 /* hash table for nfs4_file */
 530 #define FILE_HASH_BITS                   8
 531 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 532
 533 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 534 {
 535         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 536 }
 537
 538 static unsigned int file_hashval(struct knfsd_fh *fh)
 539 {
 540         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 541 }
 542
 543 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 544
 545 static void
 546 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 547 {
 548         lockdep_assert_held(&fp->fi_lock);
 549
 550         if (access & NFS4_SHARE_ACCESS_WRITE)
 551                 atomic_inc(&fp->fi_access[O_WRONLY]);
 552         if (access & NFS4_SHARE_ACCESS_READ)
 553                 atomic_inc(&fp->fi_access[O_RDONLY]);
 554 }
 555
 556 static __be32
 557 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 558 {
 559         lockdep_assert_held(&fp->fi_lock);
 560
 561         /* Does this access mode make sense? */
 562         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 563                 return nfserr_inval;
 564
 565         /* Does it conflict with a deny mode already set? */
 566         if ((access & fp->fi_share_deny) != 0)
 567                 return nfserr_share_denied;
 568
 569         __nfs4_file_get_access(fp, access);
 570         return nfs_ok;
 571 }
 572
 573 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 574 {
 575         /* Common case is that there is no deny mode. */
 576         if (deny) {
 577                 /* Does this deny mode make sense? */
 578                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 579                         return nfserr_inval;
 580
 581                 if ((deny & NFS4_SHARE_DENY_READ) &&
 582                     atomic_read(&fp->fi_access[O_RDONLY]))
 583                         return nfserr_share_denied;
 584
 585                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 586                     atomic_read(&fp->fi_access[O_WRONLY]))
 587                         return nfserr_share_denied;
 588         }
 589         return nfs_ok;
 590 }
 591
 592 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 593 {
 594         might_lock(&fp->fi_lock);
 595
 596         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 597                 struct nfsd_file *f1 = NULL;
 598                 struct nfsd_file *f2 = NULL;
 599
 600                 swap(f1, fp->fi_fds[oflag]);
 601                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 602                         swap(f2, fp->fi_fds[O_RDWR]);
 603                 spin_unlock(&fp->fi_lock);
 604                 if (f1)
 605                         nfsd_file_put(f1);
 606                 if (f2)
 607                         nfsd_file_put(f2);
 608         }
 609 }
 610
 611 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 612 {
 613         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 614
 615         if (access & NFS4_SHARE_ACCESS_WRITE)
 616                 __nfs4_file_put_access(fp, O_WRONLY);
 617         if (access & NFS4_SHARE_ACCESS_READ)
 618                 __nfs4_file_put_access(fp, O_RDONLY);
 619 }
 620
 621 /*
 622  * Allocate a new open/delegation state counter. This is needed for
 623  * pNFS for proper return on close semantics.
 624  *
 625  * Note that we only allocate it for pNFS-enabled exports, otherwise
 626  * all pointers to struct nfs4_clnt_odstate are always NULL.
 627  */
 628 static struct nfs4_clnt_odstate *
 629 alloc_clnt_odstate(struct nfs4_client *clp)
 630 {
 631         struct nfs4_clnt_odstate *co;
 632
 633         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 634         if (co) {
 635                 co->co_client = clp;
 636                 refcount_set(&co->co_odcount, 1);
 637         }
 638         return co;
 639 }
 640
 641 static void
 642 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 643 {
 644         struct nfs4_file *fp = co->co_file;
 645
 646         lockdep_assert_held(&fp->fi_lock);
 647         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 648 }
 649
 650 static inline void
 651 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 652 {
 653         if (co)
 654                 refcount_inc(&co->co_odcount);
 655 }
 656
 657 static void
 658 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 659 {
 660         struct nfs4_file *fp;
 661
 662         if (!co)
 663                 return;
 664
 665         fp = co->co_file;
 666         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 667                 list_del(&co->co_perfile);
 668                 spin_unlock(&fp->fi_lock);
 669
 670                 nfsd4_return_all_file_layouts(co->co_client, fp);
 671                 kmem_cache_free(odstate_slab, co);
 672         }
 673 }
 674
 675 static struct nfs4_clnt_odstate *
 676 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 677 {
 678         struct nfs4_clnt_odstate *co;
 679         struct nfs4_client *cl;
 680
 681         if (!new)
 682                 return NULL;
 683
 684         cl = new->co_client;
 685
 686         spin_lock(&fp->fi_lock);
 687         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 688                 if (co->co_client == cl) {
 689                         get_clnt_odstate(co);
 690                         goto out;
 691                 }
 692         }
 693         co = new;
 694         co->co_file = fp;
 695         hash_clnt_odstate_locked(new);
 696 out:
 697         spin_unlock(&fp->fi_lock);
 698         return co;
 699 }
 700
 701 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 702                                   void (*sc_free)(struct nfs4_stid *))
 703 {
 704         struct nfs4_stid *stid;
 705         int new_id;
 706
 707         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 708         if (!stid)
 709                 return NULL;
 710
 711         idr_preload(GFP_KERNEL);
 712         spin_lock(&cl->cl_lock);
 713         /* Reserving 0 for start of file in nfsdfs "states" file: */
 714         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
 715         spin_unlock(&cl->cl_lock);
 716         idr_preload_end();
 717         if (new_id < 0)
 718                 goto out_free;
 719
 720         stid->sc_free = sc_free;
 721         stid->sc_client = cl;
 722         stid->sc_stateid.si_opaque.so_id = new_id;
 723         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 724         /* Will be incremented before return to client: */
 725         refcount_set(&stid->sc_count, 1);
 726         spin_lock_init(&stid->sc_lock);
 727
 728         /*
 729          * It shouldn't be a problem to reuse an opaque stateid value.
 730          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 731          * example, a stray write retransmission could be accepted by
 732          * the server when it should have been rejected.  Therefore,
 733          * adopt a trick from the sctp code to attempt to maximize the
 734          * amount of time until an id is reused, by ensuring they always
 735          * "increase" (mod INT_MAX):
 736          */
 737         return stid;
 738 out_free:
 739         kmem_cache_free(slab, stid);
 740         return NULL;
 741 }
 742
 743 /*
 744  * Create a unique stateid_t to represent each COPY.
 745  */
 746 int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 747 {
 748         int new_id;
 749
 750         idr_preload(GFP_KERNEL);
 751         spin_lock(&nn->s2s_cp_lock);
 752         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, copy, 0, 0, GFP_NOWAIT);
 753         spin_unlock(&nn->s2s_cp_lock);
 754         idr_preload_end();
 755         if (new_id < 0)
 756                 return 0;
 757         copy->cp_stateid.si_opaque.so_id = new_id;
 758         copy->cp_stateid.si_opaque.so_clid.cl_boot = nn->boot_time;
 759         copy->cp_stateid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 760         return 1;
 761 }
 762
 763 void nfs4_free_cp_state(struct nfsd4_copy *copy)
 764 {
 765         struct nfsd_net *nn;
 766
 767         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 768         spin_lock(&nn->s2s_cp_lock);
 769         idr_remove(&nn->s2s_cp_stateids, copy->cp_stateid.si_opaque.so_id);
 770         spin_unlock(&nn->s2s_cp_lock);
 771 }
 772
 773 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 774 {
 775         struct nfs4_stid *stid;
 776
 777         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 778         if (!stid)
 779                 return NULL;
 780
 781         return openlockstateid(stid);
 782 }
 783
 784 static void nfs4_free_deleg(struct nfs4_stid *stid)
 785 {
 786         kmem_cache_free(deleg_slab, stid);
 787         atomic_long_dec(&num_delegations);
 788 }
 789
 790 /*
 791  * When we recall a delegation, we should be careful not to hand it
 792  * out again straight away.
 793  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 794  * in which the filehandles of recalled delegations are "stored".
 795  * If a filehandle appear in either filter, a delegation is blocked.
 796  * When a delegation is recalled, the filehandle is stored in the "new"
 797  * filter.
 798  * Every 30 seconds we swap the filters and clear the "new" one,
 799  * unless both are empty of course.
 800  *
 801  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 802  * low 3 bytes as hash-table indices.
 803  *
 804  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 805  * is used to manage concurrent access.  Testing does not need the lock
 806  * except when swapping the two filters.
 807  */
 808 static DEFINE_SPINLOCK(blocked_delegations_lock);
 809 static struct bloom_pair {
 810         int     entries, old_entries;
 811         time_t  swap_time;
 812         int     new; /* index into 'set' */
 813         DECLARE_BITMAP(set[2], 256);
 814 } blocked_delegations;
 815
 816 static int delegation_blocked(struct knfsd_fh *fh)
 817 {
 818         u32 hash;
 819         struct bloom_pair *bd = &blocked_delegations;
 820
 821         if (bd->entries == 0)
 822                 return 0;
 823         if (seconds_since_boot() - bd->swap_time > 30) {
 824                 spin_lock(&blocked_delegations_lock);
 825                 if (seconds_since_boot() - bd->swap_time > 30) {
 826                         bd->entries -= bd->old_entries;
 827                         bd->old_entries = bd->entries;
 828                         memset(bd->set[bd->new], 0,
 829                                sizeof(bd->set[0]));
 830                         bd->new = 1-bd->new;
 831                         bd->swap_time = seconds_since_boot();
 832                 }
 833                 spin_unlock(&blocked_delegations_lock);
 834         }
 835         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 836         if (test_bit(hash&255, bd->set[0]) &&
 837             test_bit((hash>>8)&255, bd->set[0]) &&
 838             test_bit((hash>>16)&255, bd->set[0]))
 839                 return 1;
 840
 841         if (test_bit(hash&255, bd->set[1]) &&
 842             test_bit((hash>>8)&255, bd->set[1]) &&
 843             test_bit((hash>>16)&255, bd->set[1]))
 844                 return 1;
 845
 846         return 0;
 847 }
 848
 849 static void block_delegations(struct knfsd_fh *fh)
 850 {
 851         u32 hash;
 852         struct bloom_pair *bd = &blocked_delegations;
 853
 854         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 855
 856         spin_lock(&blocked_delegations_lock);
 857         __set_bit(hash&255, bd->set[bd->new]);
 858         __set_bit((hash>>8)&255, bd->set[bd->new]);
 859         __set_bit((hash>>16)&255, bd->set[bd->new]);
 860         if (bd->entries == 0)
 861                 bd->swap_time = seconds_since_boot();
 862         bd->entries += 1;
 863         spin_unlock(&blocked_delegations_lock);
 864 }
 865
 866 static struct nfs4_delegation *
 867 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 868                  struct svc_fh *current_fh,
 869                  struct nfs4_clnt_odstate *odstate)
 870 {
 871         struct nfs4_delegation *dp;
 872         long n;
 873
 874         dprintk("NFSD alloc_init_deleg\n");
 875         n = atomic_long_inc_return(&num_delegations);
 876         if (n < 0 || n > max_delegations)
 877                 goto out_dec;
 878         if (delegation_blocked(&current_fh->fh_handle))
 879                 goto out_dec;
 880         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 881         if (dp == NULL)
 882                 goto out_dec;
 883
 884         /*
 885          * delegation seqid's are never incremented.  The 4.1 special
 886          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 887          * 0 anyway just for consistency and use 1:
 888          */
 889         dp->dl_stid.sc_stateid.si_generation = 1;
 890         INIT_LIST_HEAD(&dp->dl_perfile);
 891         INIT_LIST_HEAD(&dp->dl_perclnt);
 892         INIT_LIST_HEAD(&dp->dl_recall_lru);
 893         dp->dl_clnt_odstate = odstate;
 894         get_clnt_odstate(odstate);
 895         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 896         dp->dl_retries = 1;
 897         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 898                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 899         get_nfs4_file(fp);
 900         dp->dl_stid.sc_file = fp;
 901         return dp;
 902 out_dec:
 903         atomic_long_dec(&num_delegations);
 904         return NULL;
 905 }
 906
 907 void
 908 nfs4_put_stid(struct nfs4_stid *s)
 909 {
 910         struct nfs4_file *fp = s->sc_file;
 911         struct nfs4_client *clp = s->sc_client;
 912
 913         might_lock(&clp->cl_lock);
 914
 915         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 916                 wake_up_all(&close_wq);
 917                 return;
 918         }
 919         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 920         spin_unlock(&clp->cl_lock);
 921         s->sc_free(s);
 922         if (fp)
 923                 put_nfs4_file(fp);
 924 }
 925
 926 void
 927 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 928 {
 929         stateid_t *src = &stid->sc_stateid;
 930
 931         spin_lock(&stid->sc_lock);
 932         if (unlikely(++src->si_generation == 0))
 933                 src->si_generation = 1;
 934         memcpy(dst, src, sizeof(*dst));
 935         spin_unlock(&stid->sc_lock);
 936 }
 937
 938 static void put_deleg_file(struct nfs4_file *fp)
 939 {
 940         struct nfsd_file *nf = NULL;
 941
 942         spin_lock(&fp->fi_lock);
 943         if (--fp->fi_delegees == 0)
 944                 swap(nf, fp->fi_deleg_file);
 945         spin_unlock(&fp->fi_lock);
 946
 947         if (nf)
 948                 nfsd_file_put(nf);
 949 }
 950
 951 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 952 {
 953         struct nfs4_file *fp = dp->dl_stid.sc_file;
 954         struct nfsd_file *nf = fp->fi_deleg_file;
 955
 956         WARN_ON_ONCE(!fp->fi_delegees);
 957
 958         vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
 959         put_deleg_file(fp);
 960 }
 961
 962 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 963 {
 964         put_clnt_odstate(dp->dl_clnt_odstate);
 965         nfs4_unlock_deleg_lease(dp);
 966         nfs4_put_stid(&dp->dl_stid);
 967 }
 968
 969 void nfs4_unhash_stid(struct nfs4_stid *s)
 970 {
 971         s->sc_type = 0;
 972 }
 973
 974 /**
 975  * nfs4_delegation_exists - Discover if this delegation already exists
 976  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 977  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 978  *
 979  * Return:
 980  *      On success: true iff an existing delegation is found
 981  */
 982
 983 static bool
 984 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 985 {
 986         struct nfs4_delegation *searchdp = NULL;
 987         struct nfs4_client *searchclp = NULL;
 988
 989         lockdep_assert_held(&state_lock);
 990         lockdep_assert_held(&fp->fi_lock);
 991
 992         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 993                 searchclp = searchdp->dl_stid.sc_client;
 994                 if (clp == searchclp) {
 995                         return true;
 996                 }
 997         }
 998         return false;
 999 }
1000
1001 /**
1002  * hash_delegation_locked - Add a delegation to the appropriate lists
1003  * @dp:     a pointer to the nfs4_delegation we are adding.
1004  * @fp:     a pointer to the nfs4_file we're granting a delegation on
1005  *
1006  * Return:
1007  *      On success: NULL if the delegation was successfully hashed.
1008  *
1009  *      On error: -EAGAIN if one was previously granted to this
1010  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
1011  *
1012  */
1013
1014 static int
1015 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1016 {
1017         struct nfs4_client *clp = dp->dl_stid.sc_client;
1018
1019         lockdep_assert_held(&state_lock);
1020         lockdep_assert_held(&fp->fi_lock);
1021
1022         if (nfs4_delegation_exists(clp, fp))
1023                 return -EAGAIN;
1024         refcount_inc(&dp->dl_stid.sc_count);
1025         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1026         list_add(&dp->dl_perfile, &fp->fi_delegations);
1027         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1028         return 0;
1029 }
1030
1031 static bool
1032 unhash_delegation_locked(struct nfs4_delegation *dp)
1033 {
1034         struct nfs4_file *fp = dp->dl_stid.sc_file;
1035
1036         lockdep_assert_held(&state_lock);
1037
1038         if (list_empty(&dp->dl_perfile))
1039                 return false;
1040
1041         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1042         /* Ensure that deleg break won't try to requeue it */
1043         ++dp->dl_time;
1044         spin_lock(&fp->fi_lock);
1045         list_del_init(&dp->dl_perclnt);
1046         list_del_init(&dp->dl_recall_lru);
1047         list_del_init(&dp->dl_perfile);
1048         spin_unlock(&fp->fi_lock);
1049         return true;
1050 }
1051
1052 static void destroy_delegation(struct nfs4_delegation *dp)
1053 {
1054         bool unhashed;
1055
1056         spin_lock(&state_lock);
1057         unhashed = unhash_delegation_locked(dp);
1058         spin_unlock(&state_lock);
1059         if (unhashed)
1060                 destroy_unhashed_deleg(dp);
1061 }
1062
1063 static void revoke_delegation(struct nfs4_delegation *dp)
1064 {
1065         struct nfs4_client *clp = dp->dl_stid.sc_client;
1066
1067         WARN_ON(!list_empty(&dp->dl_recall_lru));
1068
1069         if (clp->cl_minorversion) {
1070                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1071                 refcount_inc(&dp->dl_stid.sc_count);
1072                 spin_lock(&clp->cl_lock);
1073                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1074                 spin_unlock(&clp->cl_lock);
1075         }
1076         destroy_unhashed_deleg(dp);
1077 }
1078
1079 /*
1080  * SETCLIENTID state
1081  */
1082
1083 static unsigned int clientid_hashval(u32 id)
1084 {
1085         return id & CLIENT_HASH_MASK;
1086 }
1087
1088 static unsigned int clientstr_hashval(struct xdr_netobj name)
1089 {
1090         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1091 }
1092
1093 /*
1094  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1095  * st_{access,deny}_bmap field of the stateid, in order to track not
1096  * only what share bits are currently in force, but also what
1097  * combinations of share bits previous opens have used.  This allows us
1098  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1099  * return an error if the client attempt to downgrade to a combination
1100  * of share bits not explicable by closing some of its previous opens.
1101  *
1102  * XXX: This enforcement is actually incomplete, since we don't keep
1103  * track of access/deny bit combinations; so, e.g., we allow:
1104  *
1105  *      OPEN allow read, deny write
1106  *      OPEN allow both, deny none
1107  *      DOWNGRADE allow read, deny none
1108  *
1109  * which we should reject.
1110  */
1111 static unsigned int
1112 bmap_to_share_mode(unsigned long bmap) {
1113         int i;
1114         unsigned int access = 0;
1115
1116         for (i = 1; i < 4; i++) {
1117                 if (test_bit(i, &bmap))
1118                         access |= i;
1119         }
1120         return access;
1121 }
1122
1123 /* set share access for a given stateid */
1124 static inline void
1125 set_access(u32 access, struct nfs4_ol_stateid *stp)
1126 {
1127         unsigned char mask = 1 << access;
1128
1129         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1130         stp->st_access_bmap |= mask;
1131 }
1132
1133 /* clear share access for a given stateid */
1134 static inline void
1135 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1136 {
1137         unsigned char mask = 1 << access;
1138
1139         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1140         stp->st_access_bmap &= ~mask;
1141 }
1142
1143 /* test whether a given stateid has access */
1144 static inline bool
1145 test_access(u32 access, struct nfs4_ol_stateid *stp)
1146 {
1147         unsigned char mask = 1 << access;
1148
1149         return (bool)(stp->st_access_bmap & mask);
1150 }
1151
1152 /* set share deny for a given stateid */
1153 static inline void
1154 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1155 {
1156         unsigned char mask = 1 << deny;
1157
1158         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1159         stp->st_deny_bmap |= mask;
1160 }
1161
1162 /* clear share deny for a given stateid */
1163 static inline void
1164 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1165 {
1166         unsigned char mask = 1 << deny;
1167
1168         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1169         stp->st_deny_bmap &= ~mask;
1170 }
1171
1172 /* test whether a given stateid is denying specific access */
1173 static inline bool
1174 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1175 {
1176         unsigned char mask = 1 << deny;
1177
1178         return (bool)(stp->st_deny_bmap & mask);
1179 }
1180
1181 static int nfs4_access_to_omode(u32 access)
1182 {
1183         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1184         case NFS4_SHARE_ACCESS_READ:
1185                 return O_RDONLY;
1186         case NFS4_SHARE_ACCESS_WRITE:
1187                 return O_WRONLY;
1188         case NFS4_SHARE_ACCESS_BOTH:
1189                 return O_RDWR;
1190         }
1191         WARN_ON_ONCE(1);
1192         return O_RDONLY;
1193 }
1194
1195 /*
1196  * A stateid that had a deny mode associated with it is being released
1197  * or downgraded. Recalculate the deny mode on the file.
1198  */
1199 static void
1200 recalculate_deny_mode(struct nfs4_file *fp)
1201 {
1202         struct nfs4_ol_stateid *stp;
1203
1204         spin_lock(&fp->fi_lock);
1205         fp->fi_share_deny = 0;
1206         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1207                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1208         spin_unlock(&fp->fi_lock);
1209 }
1210
1211 static void
1212 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1213 {
1214         int i;
1215         bool change = false;
1216
1217         for (i = 1; i < 4; i++) {
1218                 if ((i & deny) != i) {
1219                         change = true;
1220                         clear_deny(i, stp);
1221                 }
1222         }
1223
1224         /* Recalculate per-file deny mode if there was a change */
1225         if (change)
1226                 recalculate_deny_mode(stp->st_stid.sc_file);
1227 }
1228
1229 /* release all access and file references for a given stateid */
1230 static void
1231 release_all_access(struct nfs4_ol_stateid *stp)
1232 {
1233         int i;
1234         struct nfs4_file *fp = stp->st_stid.sc_file;
1235
1236         if (fp && stp->st_deny_bmap != 0)
1237                 recalculate_deny_mode(fp);
1238
1239         for (i = 1; i < 4; i++) {
1240                 if (test_access(i, stp))
1241                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1242                 clear_access(i, stp);
1243         }
1244 }
1245
1246 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1247 {
1248         kfree(sop->so_owner.data);
1249         sop->so_ops->so_free(sop);
1250 }
1251
1252 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1253 {
1254         struct nfs4_client *clp = sop->so_client;
1255
1256         might_lock(&clp->cl_lock);
1257
1258         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1259                 return;
1260         sop->so_ops->so_unhash(sop);
1261         spin_unlock(&clp->cl_lock);
1262         nfs4_free_stateowner(sop);
1263 }
1264
1265 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1266 {
1267         struct nfs4_file *fp = stp->st_stid.sc_file;
1268
1269         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1270
1271         if (list_empty(&stp->st_perfile))
1272                 return false;
1273
1274         spin_lock(&fp->fi_lock);
1275         list_del_init(&stp->st_perfile);
1276         spin_unlock(&fp->fi_lock);
1277         list_del(&stp->st_perstateowner);
1278         return true;
1279 }
1280
1281 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1282 {
1283         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1284
1285         put_clnt_odstate(stp->st_clnt_odstate);
1286         release_all_access(stp);
1287         if (stp->st_stateowner)
1288                 nfs4_put_stateowner(stp->st_stateowner);
1289         kmem_cache_free(stateid_slab, stid);
1290 }
1291
1292 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1293 {
1294         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1295         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1296         struct nfsd_file *nf;
1297
1298         nf = find_any_file(stp->st_stid.sc_file);
1299         if (nf) {
1300                 get_file(nf->nf_file);
1301                 filp_close(nf->nf_file, (fl_owner_t)lo);
1302                 nfsd_file_put(nf);
1303         }
1304         nfs4_free_ol_stateid(stid);
1305 }
1306
1307 /*
1308  * Put the persistent reference to an already unhashed generic stateid, while
1309  * holding the cl_lock. If it's the last reference, then put it onto the
1310  * reaplist for later destruction.
1311  */
1312 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1313                                        struct list_head *reaplist)
1314 {
1315         struct nfs4_stid *s = &stp->st_stid;
1316         struct nfs4_client *clp = s->sc_client;
1317
1318         lockdep_assert_held(&clp->cl_lock);
1319
1320         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1321
1322         if (!refcount_dec_and_test(&s->sc_count)) {
1323                 wake_up_all(&close_wq);
1324                 return;
1325         }
1326
1327         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1328         list_add(&stp->st_locks, reaplist);
1329 }
1330
1331 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1332 {
1333         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1334
1335         list_del_init(&stp->st_locks);
1336         nfs4_unhash_stid(&stp->st_stid);
1337         return unhash_ol_stateid(stp);
1338 }
1339
1340 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1341 {
1342         struct nfs4_client *clp = stp->st_stid.sc_client;
1343         bool unhashed;
1344
1345         spin_lock(&clp->cl_lock);
1346         unhashed = unhash_lock_stateid(stp);
1347         spin_unlock(&clp->cl_lock);
1348         if (unhashed)
1349                 nfs4_put_stid(&stp->st_stid);
1350 }
1351
1352 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1353 {
1354         struct nfs4_client *clp = lo->lo_owner.so_client;
1355
1356         lockdep_assert_held(&clp->cl_lock);
1357
1358         list_del_init(&lo->lo_owner.so_strhash);
1359 }
1360
1361 /*
1362  * Free a list of generic stateids that were collected earlier after being
1363  * fully unhashed.
1364  */
1365 static void
1366 free_ol_stateid_reaplist(struct list_head *reaplist)
1367 {
1368         struct nfs4_ol_stateid *stp;
1369         struct nfs4_file *fp;
1370
1371         might_sleep();
1372
1373         while (!list_empty(reaplist)) {
1374                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1375                                        st_locks);
1376                 list_del(&stp->st_locks);
1377                 fp = stp->st_stid.sc_file;
1378                 stp->st_stid.sc_free(&stp->st_stid);
1379                 if (fp)
1380                         put_nfs4_file(fp);
1381         }
1382 }
1383
1384 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1385                                        struct list_head *reaplist)
1386 {
1387         struct nfs4_ol_stateid *stp;
1388
1389         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1390
1391         while (!list_empty(&open_stp->st_locks)) {
1392                 stp = list_entry(open_stp->st_locks.next,
1393                                 struct nfs4_ol_stateid, st_locks);
1394                 WARN_ON(!unhash_lock_stateid(stp));
1395                 put_ol_stateid_locked(stp, reaplist);
1396         }
1397 }
1398
1399 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1400                                 struct list_head *reaplist)
1401 {
1402         bool unhashed;
1403
1404         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1405
1406         unhashed = unhash_ol_stateid(stp);
1407         release_open_stateid_locks(stp, reaplist);
1408         return unhashed;
1409 }
1410
1411 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1412 {
1413         LIST_HEAD(reaplist);
1414
1415         spin_lock(&stp->st_stid.sc_client->cl_lock);
1416         if (unhash_open_stateid(stp, &reaplist))
1417                 put_ol_stateid_locked(stp, &reaplist);
1418         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1419         free_ol_stateid_reaplist(&reaplist);
1420 }
1421
1422 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1423 {
1424         struct nfs4_client *clp = oo->oo_owner.so_client;
1425
1426         lockdep_assert_held(&clp->cl_lock);
1427
1428         list_del_init(&oo->oo_owner.so_strhash);
1429         list_del_init(&oo->oo_perclient);
1430 }
1431
1432 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1433 {
1434         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1435                                           nfsd_net_id);
1436         struct nfs4_ol_stateid *s;
1437
1438         spin_lock(&nn->client_lock);
1439         s = oo->oo_last_closed_stid;
1440         if (s) {
1441                 list_del_init(&oo->oo_close_lru);
1442                 oo->oo_last_closed_stid = NULL;
1443         }
1444         spin_unlock(&nn->client_lock);
1445         if (s)
1446                 nfs4_put_stid(&s->st_stid);
1447 }
1448
1449 static void release_openowner(struct nfs4_openowner *oo)
1450 {
1451         struct nfs4_ol_stateid *stp;
1452         struct nfs4_client *clp = oo->oo_owner.so_client;
1453         struct list_head reaplist;
1454
1455         INIT_LIST_HEAD(&reaplist);
1456
1457         spin_lock(&clp->cl_lock);
1458         unhash_openowner_locked(oo);
1459         while (!list_empty(&oo->oo_owner.so_stateids)) {
1460                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1461                                 struct nfs4_ol_stateid, st_perstateowner);
1462                 if (unhash_open_stateid(stp, &reaplist))
1463                         put_ol_stateid_locked(stp, &reaplist);
1464         }
1465         spin_unlock(&clp->cl_lock);
1466         free_ol_stateid_reaplist(&reaplist);
1467         release_last_closed_stateid(oo);
1468         nfs4_put_stateowner(&oo->oo_owner);
1469 }
1470
1471 static inline int
1472 hash_sessionid(struct nfs4_sessionid *sessionid)
1473 {
1474         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1475
1476         return sid->sequence % SESSION_HASH_SIZE;
1477 }
1478
1479 #ifdef CONFIG_SUNRPC_DEBUG
1480 static inline void
1481 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1482 {
1483         u32 *ptr = (u32 *)(&sessionid->data[0]);
1484         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1485 }
1486 #else
1487 static inline void
1488 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1489 {
1490 }
1491 #endif
1492
1493 /*
1494  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1495  * won't be used for replay.
1496  */
1497 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1498 {
1499         struct nfs4_stateowner *so = cstate->replay_owner;
1500
1501         if (nfserr == nfserr_replay_me)
1502                 return;
1503
1504         if (!seqid_mutating_err(ntohl(nfserr))) {
1505                 nfsd4_cstate_clear_replay(cstate);
1506                 return;
1507         }
1508         if (!so)
1509                 return;
1510         if (so->so_is_open_owner)
1511                 release_last_closed_stateid(openowner(so));
1512         so->so_seqid++;
1513         return;
1514 }
1515
1516 static void
1517 gen_sessionid(struct nfsd4_session *ses)
1518 {
1519         struct nfs4_client *clp = ses->se_client;
1520         struct nfsd4_sessionid *sid;
1521
1522         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1523         sid->clientid = clp->cl_clientid;
1524         sid->sequence = current_sessionid++;
1525         sid->reserved = 0;
1526 }
1527
1528 /*
1529  * The protocol defines ca_maxresponssize_cached to include the size of
1530  * the rpc header, but all we need to cache is the data starting after
1531  * the end of the initial SEQUENCE operation--the rest we regenerate
1532  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1533  * value that is the number of bytes in our cache plus a few additional
1534  * bytes.  In order to stay on the safe side, and not promise more than
1535  * we can cache, those additional bytes must be the minimum possible: 24
1536  * bytes of rpc header (xid through accept state, with AUTH_NULL
1537  * verifier), 12 for the compound header (with zero-length tag), and 44
1538  * for the SEQUENCE op response:
1539  */
1540 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1541
1542 static void
1543 free_session_slots(struct nfsd4_session *ses)
1544 {
1545         int i;
1546
1547         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1548                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1549                 kfree(ses->se_slots[i]);
1550         }
1551 }
1552
1553 /*
1554  * We don't actually need to cache the rpc and session headers, so we
1555  * can allocate a little less for each slot:
1556  */
1557 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1558 {
1559         u32 size;
1560
1561         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1562                 size = 0;
1563         else
1564                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1565         return size + sizeof(struct nfsd4_slot);
1566 }
1567
1568 /*
1569  * XXX: If we run out of reserved DRC memory we could (up to a point)
1570  * re-negotiate active sessions and reduce their slot usage to make
1571  * room for new connections. For now we just fail the create session.
1572  */
1573 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1574 {
1575         u32 slotsize = slot_bytes(ca);
1576         u32 num = ca->maxreqs;
1577         unsigned long avail, total_avail;
1578         unsigned int scale_factor;
1579
1580         spin_lock(&nfsd_drc_lock);
1581         if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1582                 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1583         else
1584                 /* We have handed out more space than we chose in
1585                  * set_max_drc() to allow.  That isn't really a
1586                  * problem as long as that doesn't make us think we
1587                  * have lots more due to integer overflow.
1588                  */
1589                 total_avail = 0;
1590         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1591         /*
1592          * Never use more than a fraction of the remaining memory,
1593          * unless it's the only way to give this client a slot.
1594          * The chosen fraction is either 1/8 or 1/number of threads,
1595          * whichever is smaller.  This ensures there are adequate
1596          * slots to support multiple clients per thread.
1597          * Give the client one slot even if that would require
1598          * over-allocation--it is better than failure.
1599          */
1600         scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1601
1602         avail = clamp_t(unsigned long, avail, slotsize,
1603                         total_avail/scale_factor);
1604         num = min_t(int, num, avail / slotsize);
1605         num = max_t(int, num, 1);
1606         nfsd_drc_mem_used += num * slotsize;
1607         spin_unlock(&nfsd_drc_lock);
1608
1609         return num;
1610 }
1611
1612 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1613 {
1614         int slotsize = slot_bytes(ca);
1615
1616         spin_lock(&nfsd_drc_lock);
1617         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1618         spin_unlock(&nfsd_drc_lock);
1619 }
1620
1621 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1622                                            struct nfsd4_channel_attrs *battrs)
1623 {
1624         int numslots = fattrs->maxreqs;
1625         int slotsize = slot_bytes(fattrs);
1626         struct nfsd4_session *new;
1627         int mem, i;
1628
1629         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1630                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1631         mem = numslots * sizeof(struct nfsd4_slot *);
1632
1633         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1634         if (!new)
1635                 return NULL;
1636         /* allocate each struct nfsd4_slot and data cache in one piece */
1637         for (i = 0; i < numslots; i++) {
1638                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1639                 if (!new->se_slots[i])
1640                         goto out_free;
1641         }
1642
1643         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1644         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1645
1646         return new;
1647 out_free:
1648         while (i--)
1649                 kfree(new->se_slots[i]);
1650         kfree(new);
1651         return NULL;
1652 }
1653
1654 static void free_conn(struct nfsd4_conn *c)
1655 {
1656         svc_xprt_put(c->cn_xprt);
1657         kfree(c);
1658 }
1659
1660 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1661 {
1662         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1663         struct nfs4_client *clp = c->cn_session->se_client;
1664
1665         spin_lock(&clp->cl_lock);
1666         if (!list_empty(&c->cn_persession)) {
1667                 list_del(&c->cn_persession);
1668                 free_conn(c);
1669         }
1670         nfsd4_probe_callback(clp);
1671         spin_unlock(&clp->cl_lock);
1672 }
1673
1674 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1675 {
1676         struct nfsd4_conn *conn;
1677
1678         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1679         if (!conn)
1680                 return NULL;
1681         svc_xprt_get(rqstp->rq_xprt);
1682         conn->cn_xprt = rqstp->rq_xprt;
1683         conn->cn_flags = flags;
1684         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1685         return conn;
1686 }
1687
1688 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1689 {
1690         conn->cn_session = ses;
1691         list_add(&conn->cn_persession, &ses->se_conns);
1692 }
1693
1694 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1695 {
1696         struct nfs4_client *clp = ses->se_client;
1697
1698         spin_lock(&clp->cl_lock);
1699         __nfsd4_hash_conn(conn, ses);
1700         spin_unlock(&clp->cl_lock);
1701 }
1702
1703 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1704 {
1705         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1706         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1707 }
1708
1709 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1710 {
1711         int ret;
1712
1713         nfsd4_hash_conn(conn, ses);
1714         ret = nfsd4_register_conn(conn);
1715         if (ret)
1716                 /* oops; xprt is already down: */
1717                 nfsd4_conn_lost(&conn->cn_xpt_user);
1718         /* We may have gained or lost a callback channel: */
1719         nfsd4_probe_callback_sync(ses->se_client);
1720 }
1721
1722 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1723 {
1724         u32 dir = NFS4_CDFC4_FORE;
1725
1726         if (cses->flags & SESSION4_BACK_CHAN)
1727                 dir |= NFS4_CDFC4_BACK;
1728         return alloc_conn(rqstp, dir);
1729 }
1730
1731 /* must be called under client_lock */
1732 static void nfsd4_del_conns(struct nfsd4_session *s)
1733 {
1734         struct nfs4_client *clp = s->se_client;
1735         struct nfsd4_conn *c;
1736
1737         spin_lock(&clp->cl_lock);
1738         while (!list_empty(&s->se_conns)) {
1739                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1740                 list_del_init(&c->cn_persession);
1741                 spin_unlock(&clp->cl_lock);
1742
1743                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1744                 free_conn(c);
1745
1746                 spin_lock(&clp->cl_lock);
1747         }
1748         spin_unlock(&clp->cl_lock);
1749 }
1750
1751 static void __free_session(struct nfsd4_session *ses)
1752 {
1753         free_session_slots(ses);
1754         kfree(ses);
1755 }
1756
1757 static void free_session(struct nfsd4_session *ses)
1758 {
1759         nfsd4_del_conns(ses);
1760         nfsd4_put_drc_mem(&ses->se_fchannel);
1761         __free_session(ses);
1762 }
1763
1764 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1765 {
1766         int idx;
1767         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1768
1769         new->se_client = clp;
1770         gen_sessionid(new);
1771
1772         INIT_LIST_HEAD(&new->se_conns);
1773
1774         new->se_cb_seq_nr = 1;
1775         new->se_flags = cses->flags;
1776         new->se_cb_prog = cses->callback_prog;
1777         new->se_cb_sec = cses->cb_sec;
1778         atomic_set(&new->se_ref, 0);
1779         idx = hash_sessionid(&new->se_sessionid);
1780         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1781         spin_lock(&clp->cl_lock);
1782         list_add(&new->se_perclnt, &clp->cl_sessions);
1783         spin_unlock(&clp->cl_lock);
1784
1785         {
1786                 struct sockaddr *sa = svc_addr(rqstp);
1787                 /*
1788                  * This is a little silly; with sessions there's no real
1789                  * use for the callback address.  Use the peer address
1790                  * as a reasonable default for now, but consider fixing
1791                  * the rpc client not to require an address in the
1792                  * future:
1793                  */
1794                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1795                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1796         }
1797 }
1798
1799 /* caller must hold client_lock */
1800 static struct nfsd4_session *
1801 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1802 {
1803         struct nfsd4_session *elem;
1804         int idx;
1805         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1806
1807         lockdep_assert_held(&nn->client_lock);
1808
1809         dump_sessionid(__func__, sessionid);
1810         idx = hash_sessionid(sessionid);
1811         /* Search in the appropriate list */
1812         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1813                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1814                             NFS4_MAX_SESSIONID_LEN)) {
1815                         return elem;
1816                 }
1817         }
1818
1819         dprintk("%s: session not found\n", __func__);
1820         return NULL;
1821 }
1822
1823 static struct nfsd4_session *
1824 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1825                 __be32 *ret)
1826 {
1827         struct nfsd4_session *session;
1828         __be32 status = nfserr_badsession;
1829
1830         session = __find_in_sessionid_hashtbl(sessionid, net);
1831         if (!session)
1832                 goto out;
1833         status = nfsd4_get_session_locked(session);
1834         if (status)
1835                 session = NULL;
1836 out:
1837         *ret = status;
1838         return session;
1839 }
1840
1841 /* caller must hold client_lock */
1842 static void
1843 unhash_session(struct nfsd4_session *ses)
1844 {
1845         struct nfs4_client *clp = ses->se_client;
1846         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1847
1848         lockdep_assert_held(&nn->client_lock);
1849
1850         list_del(&ses->se_hash);
1851         spin_lock(&ses->se_client->cl_lock);
1852         list_del(&ses->se_perclnt);
1853         spin_unlock(&ses->se_client->cl_lock);
1854 }
1855
1856 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1857 static int
1858 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1859 {
1860         /*
1861          * We're assuming the clid was not given out from a boot
1862          * precisely 2^32 (about 136 years) before this one.  That seems
1863          * a safe assumption:
1864          */
1865         if (clid->cl_boot == (u32)nn->boot_time)
1866                 return 0;
1867         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1868                 clid->cl_boot, clid->cl_id, nn->boot_time);
1869         return 1;
1870 }
1871
1872 /*
1873  * XXX Should we use a slab cache ?
1874  * This type of memory management is somewhat inefficient, but we use it
1875  * anyway since SETCLIENTID is not a common operation.
1876  */
1877 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1878 {
1879         struct nfs4_client *clp;
1880         int i;
1881
1882         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1883         if (clp == NULL)
1884                 return NULL;
1885         xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1886         if (clp->cl_name.data == NULL)
1887                 goto err_no_name;
1888         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1889                                                  sizeof(struct list_head),
1890                                                  GFP_KERNEL);
1891         if (!clp->cl_ownerstr_hashtbl)
1892                 goto err_no_hashtbl;
1893         for (i = 0; i < OWNER_HASH_SIZE; i++)
1894                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1895         INIT_LIST_HEAD(&clp->cl_sessions);
1896         idr_init(&clp->cl_stateids);
1897         atomic_set(&clp->cl_rpc_users, 0);
1898         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1899         INIT_LIST_HEAD(&clp->cl_idhash);
1900         INIT_LIST_HEAD(&clp->cl_openowners);
1901         INIT_LIST_HEAD(&clp->cl_delegations);
1902         INIT_LIST_HEAD(&clp->cl_lru);
1903         INIT_LIST_HEAD(&clp->cl_revoked);
1904 #ifdef CONFIG_NFSD_PNFS
1905         INIT_LIST_HEAD(&clp->cl_lo_states);
1906 #endif
1907         INIT_LIST_HEAD(&clp->async_copies);
1908         spin_lock_init(&clp->async_lock);
1909         spin_lock_init(&clp->cl_lock);
1910         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1911         return clp;
1912 err_no_hashtbl:
1913         kfree(clp->cl_name.data);
1914 err_no_name:
1915         kmem_cache_free(client_slab, clp);
1916         return NULL;
1917 }
1918
1919 static void __free_client(struct kref *k)
1920 {
1921         struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1922         struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1923
1924         free_svc_cred(&clp->cl_cred);
1925         kfree(clp->cl_ownerstr_hashtbl);
1926         kfree(clp->cl_name.data);
1927         kfree(clp->cl_nii_domain.data);
1928         kfree(clp->cl_nii_name.data);
1929         idr_destroy(&clp->cl_stateids);
1930         kmem_cache_free(client_slab, clp);
1931 }
1932
1933 static void drop_client(struct nfs4_client *clp)
1934 {
1935         kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1936 }
1937
1938 static void
1939 free_client(struct nfs4_client *clp)
1940 {
1941         while (!list_empty(&clp->cl_sessions)) {
1942                 struct nfsd4_session *ses;
1943                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1944                                 se_perclnt);
1945                 list_del(&ses->se_perclnt);
1946                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1947                 free_session(ses);
1948         }
1949         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1950         if (clp->cl_nfsd_dentry) {
1951                 nfsd_client_rmdir(clp->cl_nfsd_dentry);
1952                 clp->cl_nfsd_dentry = NULL;
1953                 wake_up_all(&expiry_wq);
1954         }
1955         drop_client(clp);
1956 }
1957
1958 /* must be called under the client_lock */
1959 static void
1960 unhash_client_locked(struct nfs4_client *clp)
1961 {
1962         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1963         struct nfsd4_session *ses;
1964
1965         lockdep_assert_held(&nn->client_lock);
1966
1967         /* Mark the client as expired! */
1968         clp->cl_time = 0;
1969         /* Make it invisible */
1970         if (!list_empty(&clp->cl_idhash)) {
1971                 list_del_init(&clp->cl_idhash);
1972                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1973                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1974                 else
1975                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1976         }
1977         list_del_init(&clp->cl_lru);
1978         spin_lock(&clp->cl_lock);
1979         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1980                 list_del_init(&ses->se_hash);
1981         spin_unlock(&clp->cl_lock);
1982 }
1983
1984 static void
1985 unhash_client(struct nfs4_client *clp)
1986 {
1987         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1988
1989         spin_lock(&nn->client_lock);
1990         unhash_client_locked(clp);
1991         spin_unlock(&nn->client_lock);
1992 }
1993
1994 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1995 {
1996         if (atomic_read(&clp->cl_rpc_users))
1997                 return nfserr_jukebox;
1998         unhash_client_locked(clp);
1999         return nfs_ok;
2000 }
2001
2002 static void
2003 __destroy_client(struct nfs4_client *clp)
2004 {
2005         int i;
2006         struct nfs4_openowner *oo;
2007         struct nfs4_delegation *dp;
2008         struct list_head reaplist;
2009
2010         INIT_LIST_HEAD(&reaplist);
2011         spin_lock(&state_lock);
2012         while (!list_empty(&clp->cl_delegations)) {
2013                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2014                 WARN_ON(!unhash_delegation_locked(dp));
2015                 list_add(&dp->dl_recall_lru, &reaplist);
2016         }
2017         spin_unlock(&state_lock);
2018         while (!list_empty(&reaplist)) {
2019                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2020                 list_del_init(&dp->dl_recall_lru);
2021                 destroy_unhashed_deleg(dp);
2022         }
2023         while (!list_empty(&clp->cl_revoked)) {
2024                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2025                 list_del_init(&dp->dl_recall_lru);
2026                 nfs4_put_stid(&dp->dl_stid);
2027         }
2028         while (!list_empty(&clp->cl_openowners)) {
2029                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2030                 nfs4_get_stateowner(&oo->oo_owner);
2031                 release_openowner(oo);
2032         }
2033         for (i = 0; i < OWNER_HASH_SIZE; i++) {
2034                 struct nfs4_stateowner *so, *tmp;
2035
2036                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2037                                          so_strhash) {
2038                         /* Should be no openowners at this point */
2039                         WARN_ON_ONCE(so->so_is_open_owner);
2040                         remove_blocked_locks(lockowner(so));
2041                 }
2042         }
2043         nfsd4_return_all_client_layouts(clp);
2044         nfsd4_shutdown_copy(clp);
2045         nfsd4_shutdown_callback(clp);
2046         if (clp->cl_cb_conn.cb_xprt)
2047                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2048         free_client(clp);
2049         wake_up_all(&expiry_wq);
2050 }
2051
2052 static void
2053 destroy_client(struct nfs4_client *clp)
2054 {
2055         unhash_client(clp);
2056         __destroy_client(clp);
2057 }
2058
2059 static void inc_reclaim_complete(struct nfs4_client *clp)
2060 {
2061         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2062
2063         if (!nn->track_reclaim_completes)
2064                 return;
2065         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2066                 return;
2067         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2068                         nn->reclaim_str_hashtbl_size) {
2069                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2070                                 clp->net->ns.inum);
2071                 nfsd4_end_grace(nn);
2072         }
2073 }
2074
2075 static void expire_client(struct nfs4_client *clp)
2076 {
2077         unhash_client(clp);
2078         nfsd4_client_record_remove(clp);
2079         __destroy_client(clp);
2080 }
2081
2082 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2083 {
2084         memcpy(target->cl_verifier.data, source->data,
2085                         sizeof(target->cl_verifier.data));
2086 }
2087
2088 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2089 {
2090         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2091         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2092 }
2093
2094 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2095 {
2096         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2097         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2098                                                                 GFP_KERNEL);
2099         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2100         if ((source->cr_principal && !target->cr_principal) ||
2101             (source->cr_raw_principal && !target->cr_raw_principal) ||
2102             (source->cr_targ_princ && !target->cr_targ_princ))
2103                 return -ENOMEM;
2104
2105         target->cr_flavor = source->cr_flavor;
2106         target->cr_uid = source->cr_uid;
2107         target->cr_gid = source->cr_gid;
2108         target->cr_group_info = source->cr_group_info;
2109         get_group_info(target->cr_group_info);
2110         target->cr_gss_mech = source->cr_gss_mech;
2111         if (source->cr_gss_mech)
2112                 gss_mech_get(source->cr_gss_mech);
2113         return 0;
2114 }
2115
2116 static int
2117 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2118 {
2119         if (o1->len < o2->len)
2120                 return -1;
2121         if (o1->len > o2->len)
2122                 return 1;
2123         return memcmp(o1->data, o2->data, o1->len);
2124 }
2125
2126 static int
2127 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2128 {
2129         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2130 }
2131
2132 static int
2133 same_clid(clientid_t *cl1, clientid_t *cl2)
2134 {
2135         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2136 }
2137
2138 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2139 {
2140         int i;
2141
2142         if (g1->ngroups != g2->ngroups)
2143                 return false;
2144         for (i=0; i<g1->ngroups; i++)
2145                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2146                         return false;
2147         return true;
2148 }
2149
2150 /*
2151  * RFC 3530 language requires clid_inuse be returned when the
2152  * "principal" associated with a requests differs from that previously
2153  * used.  We use uid, gid's, and gss principal string as our best
2154  * approximation.  We also don't want to allow non-gss use of a client
2155  * established using gss: in theory cr_principal should catch that
2156  * change, but in practice cr_principal can be null even in the gss case
2157  * since gssd doesn't always pass down a principal string.
2158  */
2159 static bool is_gss_cred(struct svc_cred *cr)
2160 {
2161         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2162         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2163 }
2164
2165
2166 static bool
2167 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2168 {
2169         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2170                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2171                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2172                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2173                 return false;
2174         /* XXX: check that cr_targ_princ fields match ? */
2175         if (cr1->cr_principal == cr2->cr_principal)
2176                 return true;
2177         if (!cr1->cr_principal || !cr2->cr_principal)
2178                 return false;
2179         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2180 }
2181
2182 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2183 {
2184         struct svc_cred *cr = &rqstp->rq_cred;
2185         u32 service;
2186
2187         if (!cr->cr_gss_mech)
2188                 return false;
2189         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2190         return service == RPC_GSS_SVC_INTEGRITY ||
2191                service == RPC_GSS_SVC_PRIVACY;
2192 }
2193
2194 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2195 {
2196         struct svc_cred *cr = &rqstp->rq_cred;
2197
2198         if (!cl->cl_mach_cred)
2199                 return true;
2200         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2201                 return false;
2202         if (!svc_rqst_integrity_protected(rqstp))
2203                 return false;
2204         if (cl->cl_cred.cr_raw_principal)
2205                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2206                                                 cr->cr_raw_principal);
2207         if (!cr->cr_principal)
2208                 return false;
2209         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2210 }
2211
2212 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2213 {
2214         __be32 verf[2];
2215
2216         /*
2217          * This is opaque to client, so no need to byte-swap. Use
2218          * __force to keep sparse happy
2219          */
2220         verf[0] = (__force __be32)get_seconds();
2221         verf[1] = (__force __be32)nn->clverifier_counter++;
2222         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2223 }
2224
2225 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2226 {
2227         clp->cl_clientid.cl_boot = nn->boot_time;
2228         clp->cl_clientid.cl_id = nn->clientid_counter++;
2229         gen_confirm(clp, nn);
2230 }
2231
2232 static struct nfs4_stid *
2233 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2234 {
2235         struct nfs4_stid *ret;
2236
2237         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2238         if (!ret || !ret->sc_type)
2239                 return NULL;
2240         return ret;
2241 }
2242
2243 static struct nfs4_stid *
2244 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2245 {
2246         struct nfs4_stid *s;
2247
2248         spin_lock(&cl->cl_lock);
2249         s = find_stateid_locked(cl, t);
2250         if (s != NULL) {
2251                 if (typemask & s->sc_type)
2252                         refcount_inc(&s->sc_count);
2253                 else
2254                         s = NULL;
2255         }
2256         spin_unlock(&cl->cl_lock);
2257         return s;
2258 }
2259
2260 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2261 {
2262         struct nfsdfs_client *nc;
2263         nc = get_nfsdfs_client(inode);
2264         if (!nc)
2265                 return NULL;
2266         return container_of(nc, struct nfs4_client, cl_nfsdfs);
2267 }
2268
2269 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2270 {
2271         seq_printf(m, "\"");
2272         seq_escape_mem_ascii(m, data, len);
2273         seq_printf(m, "\"");
2274 }
2275
2276 static int client_info_show(struct seq_file *m, void *v)
2277 {
2278         struct inode *inode = m->private;
2279         struct nfs4_client *clp;
2280         u64 clid;
2281
2282         clp = get_nfsdfs_clp(inode);
2283         if (!clp)
2284                 return -ENXIO;
2285         memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2286         seq_printf(m, "clientid: 0x%llx\n", clid);
2287         seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2288         seq_printf(m, "name: ");
2289         seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2290         seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2291         if (clp->cl_nii_domain.data) {
2292                 seq_printf(m, "Implementation domain: ");
2293                 seq_quote_mem(m, clp->cl_nii_domain.data,
2294                                         clp->cl_nii_domain.len);
2295                 seq_printf(m, "\nImplementation name: ");
2296                 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2297                 seq_printf(m, "\nImplementation time: [%ld, %ld]\n",
2298                         clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2299         }
2300         drop_client(clp);
2301
2302         return 0;
2303 }
2304
2305 static int client_info_open(struct inode *inode, struct file *file)
2306 {
2307         return single_open(file, client_info_show, inode);
2308 }
2309
2310 static const struct file_operations client_info_fops = {
2311         .open           = client_info_open,
2312         .read           = seq_read,
2313         .llseek         = seq_lseek,
2314         .release        = single_release,
2315 };
2316
2317 static void *states_start(struct seq_file *s, loff_t *pos)
2318         __acquires(&clp->cl_lock)
2319 {
2320         struct nfs4_client *clp = s->private;
2321         unsigned long id = *pos;
2322         void *ret;
2323
2324         spin_lock(&clp->cl_lock);
2325         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2326         *pos = id;
2327         return ret;
2328 }
2329
2330 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2331 {
2332         struct nfs4_client *clp = s->private;
2333         unsigned long id = *pos;
2334         void *ret;
2335
2336         id = *pos;
2337         id++;
2338         ret = idr_get_next_ul(&clp->cl_stateids, &id);
2339         *pos = id;
2340         return ret;
2341 }
2342
2343 static void states_stop(struct seq_file *s, void *v)
2344         __releases(&clp->cl_lock)
2345 {
2346         struct nfs4_client *clp = s->private;
2347
2348         spin_unlock(&clp->cl_lock);
2349 }
2350
2351 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2352 {
2353         struct inode *inode = f->nf_inode;
2354
2355         seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2356                                         MAJOR(inode->i_sb->s_dev),
2357                                          MINOR(inode->i_sb->s_dev),
2358                                          inode->i_ino);
2359 }
2360
2361 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2362 {
2363         seq_printf(s, "owner: ");
2364         seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2365 }
2366
2367 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2368 {
2369         struct nfs4_ol_stateid *ols;
2370         struct nfs4_file *nf;
2371         struct nfsd_file *file;
2372         struct nfs4_stateowner *oo;
2373         unsigned int access, deny;
2374
2375         if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2376                 return 0; /* XXX: or SEQ_SKIP? */
2377         ols = openlockstateid(st);
2378         oo = ols->st_stateowner;
2379         nf = st->sc_file;
2380         file = find_any_file(nf);
2381
2382         seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2383
2384         access = bmap_to_share_mode(ols->st_access_bmap);
2385         deny   = bmap_to_share_mode(ols->st_deny_bmap);
2386
2387         seq_printf(s, "access: \%s\%s, ",
2388                 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2389                 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2390         seq_printf(s, "deny: \%s\%s, ",
2391                 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2392                 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2393
2394         nfs4_show_superblock(s, file);
2395         seq_printf(s, ", ");
2396         nfs4_show_owner(s, oo);
2397         seq_printf(s, " }\n");
2398         nfsd_file_put(file);
2399
2400         return 0;
2401 }
2402
2403 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2404 {
2405         struct nfs4_ol_stateid *ols;
2406         struct nfs4_file *nf;
2407         struct nfsd_file *file;
2408         struct nfs4_stateowner *oo;
2409
2410         ols = openlockstateid(st);
2411         oo = ols->st_stateowner;
2412         nf = st->sc_file;
2413         file = find_any_file(nf);
2414
2415         seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2416
2417         /*
2418          * Note: a lock stateid isn't really the same thing as a lock,
2419          * it's the locking state held by one owner on a file, and there
2420          * may be multiple (or no) lock ranges associated with it.
2421          * (Same for the matter is true of open stateids.)
2422          */
2423
2424         nfs4_show_superblock(s, file);
2425         /* XXX: open stateid? */
2426         seq_printf(s, ", ");
2427         nfs4_show_owner(s, oo);
2428         seq_printf(s, " }\n");
2429         nfsd_file_put(file);
2430
2431         return 0;
2432 }
2433
2434 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2435 {
2436         struct nfs4_delegation *ds;
2437         struct nfs4_file *nf;
2438         struct nfsd_file *file;
2439
2440         ds = delegstateid(st);
2441         nf = st->sc_file;
2442         file = nf->fi_deleg_file;
2443
2444         seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2445
2446         /* Kinda dead code as long as we only support read delegs: */
2447         seq_printf(s, "access: %s, ",
2448                 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2449
2450         /* XXX: lease time, whether it's being recalled. */
2451
2452         nfs4_show_superblock(s, file);
2453         seq_printf(s, " }\n");
2454
2455         return 0;
2456 }
2457
2458 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2459 {
2460         struct nfs4_layout_stateid *ls;
2461         struct nfsd_file *file;
2462
2463         ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2464         file = ls->ls_file;
2465
2466         seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2467
2468         /* XXX: What else would be useful? */
2469
2470         nfs4_show_superblock(s, file);
2471         seq_printf(s, " }\n");
2472
2473         return 0;
2474 }
2475
2476 static int states_show(struct seq_file *s, void *v)
2477 {
2478         struct nfs4_stid *st = v;
2479
2480         switch (st->sc_type) {
2481         case NFS4_OPEN_STID:
2482                 return nfs4_show_open(s, st);
2483         case NFS4_LOCK_STID:
2484                 return nfs4_show_lock(s, st);
2485         case NFS4_DELEG_STID:
2486                 return nfs4_show_deleg(s, st);
2487         case NFS4_LAYOUT_STID:
2488                 return nfs4_show_layout(s, st);
2489         default:
2490                 return 0; /* XXX: or SEQ_SKIP? */
2491         }
2492         /* XXX: copy stateids? */
2493 }
2494
2495 static struct seq_operations states_seq_ops = {
2496         .start = states_start,
2497         .next = states_next,
2498         .stop = states_stop,
2499         .show = states_show
2500 };
2501
2502 static int client_states_open(struct inode *inode, struct file *file)
2503 {
2504         struct seq_file *s;
2505         struct nfs4_client *clp;
2506         int ret;
2507
2508         clp = get_nfsdfs_clp(inode);
2509         if (!clp)
2510                 return -ENXIO;
2511
2512         ret = seq_open(file, &states_seq_ops);
2513         if (ret)
2514                 return ret;
2515         s = file->private_data;
2516         s->private = clp;
2517         return 0;
2518 }
2519
2520 static int client_opens_release(struct inode *inode, struct file *file)
2521 {
2522         struct seq_file *m = file->private_data;
2523         struct nfs4_client *clp = m->private;
2524
2525         /* XXX: alternatively, we could get/drop in seq start/stop */
2526         drop_client(clp);
2527         return 0;
2528 }
2529
2530 static const struct file_operations client_states_fops = {
2531         .open           = client_states_open,
2532         .read           = seq_read,
2533         .llseek         = seq_lseek,
2534         .release        = client_opens_release,
2535 };
2536
2537 /*
2538  * Normally we refuse to destroy clients that are in use, but here the
2539  * administrator is telling us to just do it.  We also want to wait
2540  * so the caller has a guarantee that the client's locks are gone by
2541  * the time the write returns:
2542  */
2543 static void force_expire_client(struct nfs4_client *clp)
2544 {
2545         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2546         bool already_expired;
2547
2548         spin_lock(&clp->cl_lock);
2549         clp->cl_time = 0;
2550         spin_unlock(&clp->cl_lock);
2551
2552         wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2553         spin_lock(&nn->client_lock);
2554         already_expired = list_empty(&clp->cl_lru);
2555         if (!already_expired)
2556                 unhash_client_locked(clp);
2557         spin_unlock(&nn->client_lock);
2558
2559         if (!already_expired)
2560                 expire_client(clp);
2561         else
2562                 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2563 }
2564
2565 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2566                                    size_t size, loff_t *pos)
2567 {
2568         char *data;
2569         struct nfs4_client *clp;
2570
2571         data = simple_transaction_get(file, buf, size);
2572         if (IS_ERR(data))
2573                 return PTR_ERR(data);
2574         if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2575                 return -EINVAL;
2576         clp = get_nfsdfs_clp(file_inode(file));
2577         if (!clp)
2578                 return -ENXIO;
2579         force_expire_client(clp);
2580         drop_client(clp);
2581         return 7;
2582 }
2583
2584 static const struct file_operations client_ctl_fops = {
2585         .write          = client_ctl_write,
2586         .release        = simple_transaction_release,
2587 };
2588
2589 static const struct tree_descr client_files[] = {
2590         [0] = {"info", &client_info_fops, S_IRUSR},
2591         [1] = {"states", &client_states_fops, S_IRUSR},
2592         [2] = {"ctl", &client_ctl_fops, S_IRUSR|S_IWUSR},
2593         [3] = {""},
2594 };
2595
2596 static struct nfs4_client *create_client(struct xdr_netobj name,
2597                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2598 {
2599         struct nfs4_client *clp;
2600         struct sockaddr *sa = svc_addr(rqstp);
2601         int ret;
2602         struct net *net = SVC_NET(rqstp);
2603         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2604
2605         clp = alloc_client(name);
2606         if (clp == NULL)
2607                 return NULL;
2608
2609         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2610         if (ret) {
2611                 free_client(clp);
2612                 return NULL;
2613         }
2614         gen_clid(clp, nn);
2615         kref_init(&clp->cl_nfsdfs.cl_ref);
2616         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2617         clp->cl_time = get_seconds();
2618         clear_bit(0, &clp->cl_cb_slot_busy);
2619         copy_verf(clp, verf);
2620         memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2621         clp->cl_cb_session = NULL;
2622         clp->net = net;
2623         clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2624                         clp->cl_clientid.cl_id - nn->clientid_base,
2625                         client_files);
2626         if (!clp->cl_nfsd_dentry) {
2627                 free_client(clp);
2628                 return NULL;
2629         }
2630         return clp;
2631 }
2632
2633 static void
2634 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2635 {
2636         struct rb_node **new = &(root->rb_node), *parent = NULL;
2637         struct nfs4_client *clp;
2638
2639         while (*new) {
2640                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2641                 parent = *new;
2642
2643                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2644                         new = &((*new)->rb_left);
2645                 else
2646                         new = &((*new)->rb_right);
2647         }
2648
2649         rb_link_node(&new_clp->cl_namenode, parent, new);
2650         rb_insert_color(&new_clp->cl_namenode, root);
2651 }
2652
2653 static struct nfs4_client *
2654 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2655 {
2656         int cmp;
2657         struct rb_node *node = root->rb_node;
2658         struct nfs4_client *clp;
2659
2660         while (node) {
2661                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2662                 cmp = compare_blob(&clp->cl_name, name);
2663                 if (cmp > 0)
2664                         node = node->rb_left;
2665                 else if (cmp < 0)
2666                         node = node->rb_right;
2667                 else
2668                         return clp;
2669         }
2670         return NULL;
2671 }
2672
2673 static void
2674 add_to_unconfirmed(struct nfs4_client *clp)
2675 {
2676         unsigned int idhashval;
2677         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2678
2679         lockdep_assert_held(&nn->client_lock);
2680
2681         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2682         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2683         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2684         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2685         renew_client_locked(clp);
2686 }
2687
2688 static void
2689 move_to_confirmed(struct nfs4_client *clp)
2690 {
2691         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2692         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2693
2694         lockdep_assert_held(&nn->client_lock);
2695
2696         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2697         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2698         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2699         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2700         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2701         renew_client_locked(clp);
2702 }
2703
2704 static struct nfs4_client *
2705 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2706 {
2707         struct nfs4_client *clp;
2708         unsigned int idhashval = clientid_hashval(clid->cl_id);
2709
2710         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2711                 if (same_clid(&clp->cl_clientid, clid)) {
2712                         if ((bool)clp->cl_minorversion != sessions)
2713                                 return NULL;
2714                         renew_client_locked(clp);
2715                         return clp;
2716                 }
2717         }
2718         return NULL;
2719 }
2720
2721 static struct nfs4_client *
2722 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2723 {
2724         struct list_head *tbl = nn->conf_id_hashtbl;
2725
2726         lockdep_assert_held(&nn->client_lock);
2727         return find_client_in_id_table(tbl, clid, sessions);
2728 }
2729
2730 static struct nfs4_client *
2731 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2732 {
2733         struct list_head *tbl = nn->unconf_id_hashtbl;
2734
2735         lockdep_assert_held(&nn->client_lock);
2736         return find_client_in_id_table(tbl, clid, sessions);
2737 }
2738
2739 static bool clp_used_exchangeid(struct nfs4_client *clp)
2740 {
2741         return clp->cl_exchange_flags != 0;
2742 }
2743
2744 static struct nfs4_client *
2745 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2746 {
2747         lockdep_assert_held(&nn->client_lock);
2748         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2749 }
2750
2751 static struct nfs4_client *
2752 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2753 {
2754         lockdep_assert_held(&nn->client_lock);
2755         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2756 }
2757
2758 static void
2759 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2760 {
2761         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2762         struct sockaddr *sa = svc_addr(rqstp);
2763         u32 scopeid = rpc_get_scope_id(sa);
2764         unsigned short expected_family;
2765
2766         /* Currently, we only support tcp and tcp6 for the callback channel */
2767         if (se->se_callback_netid_len == 3 &&
2768             !memcmp(se->se_callback_netid_val, "tcp", 3))
2769                 expected_family = AF_INET;
2770         else if (se->se_callback_netid_len == 4 &&
2771                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2772                 expected_family = AF_INET6;
2773         else
2774                 goto out_err;
2775
2776         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2777                                             se->se_callback_addr_len,
2778                                             (struct sockaddr *)&conn->cb_addr,
2779                                             sizeof(conn->cb_addr));
2780
2781         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2782                 goto out_err;
2783
2784         if (conn->cb_addr.ss_family == AF_INET6)
2785                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2786
2787         conn->cb_prog = se->se_callback_prog;
2788         conn->cb_ident = se->se_callback_ident;
2789         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2790         return;
2791 out_err:
2792         conn->cb_addr.ss_family = AF_UNSPEC;
2793         conn->cb_addrlen = 0;
2794         dprintk("NFSD: this client (clientid %08x/%08x) "
2795                 "will not receive delegations\n",
2796                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2797
2798         return;
2799 }
2800
2801 /*
2802  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2803  */
2804 static void
2805 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2806 {
2807         struct xdr_buf *buf = resp->xdr.buf;
2808         struct nfsd4_slot *slot = resp->cstate.slot;
2809         unsigned int base;
2810
2811         dprintk("--> %s slot %p\n", __func__, slot);
2812
2813         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2814         slot->sl_opcnt = resp->opcnt;
2815         slot->sl_status = resp->cstate.status;
2816         free_svc_cred(&slot->sl_cred);
2817         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2818
2819         if (!nfsd4_cache_this(resp)) {
2820                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2821                 return;
2822         }
2823         slot->sl_flags |= NFSD4_SLOT_CACHED;
2824
2825         base = resp->cstate.data_offset;
2826         slot->sl_datalen = buf->len - base;
2827         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2828                 WARN(1, "%s: sessions DRC could not cache compound\n",
2829                      __func__);
2830         return;
2831 }
2832
2833 /*
2834  * Encode the replay sequence operation from the slot values.
2835  * If cachethis is FALSE encode the uncached rep error on the next
2836  * operation which sets resp->p and increments resp->opcnt for
2837  * nfs4svc_encode_compoundres.
2838  *
2839  */
2840 static __be32
2841 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2842                           struct nfsd4_compoundres *resp)
2843 {
2844         struct nfsd4_op *op;
2845         struct nfsd4_slot *slot = resp->cstate.slot;
2846
2847         /* Encode the replayed sequence operation */
2848         op = &args->ops[resp->opcnt - 1];
2849         nfsd4_encode_operation(resp, op);
2850
2851         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2852                 return op->status;
2853         if (args->opcnt == 1) {
2854                 /*
2855                  * The original operation wasn't a solo sequence--we
2856                  * always cache those--so this retry must not match the
2857                  * original:
2858                  */
2859                 op->status = nfserr_seq_false_retry;
2860         } else {
2861                 op = &args->ops[resp->opcnt++];
2862                 op->status = nfserr_retry_uncached_rep;
2863                 nfsd4_encode_operation(resp, op);
2864         }
2865         return op->status;
2866 }
2867
2868 /*
2869  * The sequence operation is not cached because we can use the slot and
2870  * session values.
2871  */
2872 static __be32
2873 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2874                          struct nfsd4_sequence *seq)
2875 {
2876         struct nfsd4_slot *slot = resp->cstate.slot;
2877         struct xdr_stream *xdr = &resp->xdr;
2878         __be32 *p;
2879         __be32 status;
2880
2881         dprintk("--> %s slot %p\n", __func__, slot);
2882
2883         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2884         if (status)
2885                 return status;
2886
2887         p = xdr_reserve_space(xdr, slot->sl_datalen);
2888         if (!p) {
2889                 WARN_ON_ONCE(1);
2890                 return nfserr_serverfault;
2891         }
2892         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2893         xdr_commit_encode(xdr);
2894
2895         resp->opcnt = slot->sl_opcnt;
2896         return slot->sl_status;
2897 }
2898
2899 /*
2900  * Set the exchange_id flags returned by the server.
2901  */
2902 static void
2903 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2904 {
2905 #ifdef CONFIG_NFSD_PNFS
2906         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2907 #else
2908         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2909 #endif
2910
2911         /* Referrals are supported, Migration is not. */
2912         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2913
2914         /* set the wire flags to return to client. */
2915         clid->flags = new->cl_exchange_flags;
2916 }
2917
2918 static bool client_has_openowners(struct nfs4_client *clp)
2919 {
2920         struct nfs4_openowner *oo;
2921
2922         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2923                 if (!list_empty(&oo->oo_owner.so_stateids))
2924                         return true;
2925         }
2926         return false;
2927 }
2928
2929 static bool client_has_state(struct nfs4_client *clp)
2930 {
2931         return client_has_openowners(clp)
2932 #ifdef CONFIG_NFSD_PNFS
2933                 || !list_empty(&clp->cl_lo_states)
2934 #endif
2935                 || !list_empty(&clp->cl_delegations)
2936                 || !list_empty(&clp->cl_sessions)
2937                 || !list_empty(&clp->async_copies);
2938 }
2939
2940 static __be32 copy_impl_id(struct nfs4_client *clp,
2941                                 struct nfsd4_exchange_id *exid)
2942 {
2943         if (!exid->nii_domain.data)
2944                 return 0;
2945         xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
2946         if (!clp->cl_nii_domain.data)
2947                 return nfserr_jukebox;
2948         xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
2949         if (!clp->cl_nii_name.data)
2950                 return nfserr_jukebox;
2951         clp->cl_nii_time.tv_sec = exid->nii_time.tv_sec;
2952         clp->cl_nii_time.tv_nsec = exid->nii_time.tv_nsec;
2953         return 0;
2954 }
2955
2956 __be32
2957 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2958                 union nfsd4_op_u *u)
2959 {
2960         struct nfsd4_exchange_id *exid = &u->exchange_id;
2961         struct nfs4_client *conf, *new;
2962         struct nfs4_client *unconf = NULL;
2963         __be32 status;
2964         char                    addr_str[INET6_ADDRSTRLEN];
2965         nfs4_verifier           verf = exid->verifier;
2966         struct sockaddr         *sa = svc_addr(rqstp);
2967         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2968         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2969
2970         rpc_ntop(sa, addr_str, sizeof(addr_str));
2971         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2972                 "ip_addr=%s flags %x, spa_how %d\n",
2973                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2974                 addr_str, exid->flags, exid->spa_how);
2975
2976         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2977                 return nfserr_inval;
2978
2979         new = create_client(exid->clname, rqstp, &verf);
2980         if (new == NULL)
2981                 return nfserr_jukebox;
2982         status = copy_impl_id(new, exid);
2983         if (status)
2984                 goto out_nolock;
2985
2986         switch (exid->spa_how) {
2987         case SP4_MACH_CRED:
2988                 exid->spo_must_enforce[0] = 0;
2989                 exid->spo_must_enforce[1] = (
2990                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2991                         1 << (OP_EXCHANGE_ID - 32) |
2992                         1 << (OP_CREATE_SESSION - 32) |
2993                         1 << (OP_DESTROY_SESSION - 32) |
2994                         1 << (OP_DESTROY_CLIENTID - 32));
2995
2996                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2997                                         1 << (OP_OPEN_DOWNGRADE) |
2998                                         1 << (OP_LOCKU) |
2999                                         1 << (OP_DELEGRETURN));
3000
3001                 exid->spo_must_allow[1] &= (
3002                                         1 << (OP_TEST_STATEID - 32) |
3003                                         1 << (OP_FREE_STATEID - 32));
3004                 if (!svc_rqst_integrity_protected(rqstp)) {
3005                         status = nfserr_inval;
3006                         goto out_nolock;
3007                 }
3008                 /*
3009                  * Sometimes userspace doesn't give us a principal.
3010                  * Which is a bug, really.  Anyway, we can't enforce
3011                  * MACH_CRED in that case, better to give up now:
3012                  */
3013                 if (!new->cl_cred.cr_principal &&
3014                                         !new->cl_cred.cr_raw_principal) {
3015                         status = nfserr_serverfault;
3016                         goto out_nolock;
3017                 }
3018                 new->cl_mach_cred = true;
3019         case SP4_NONE:
3020                 break;
3021         default:                                /* checked by xdr code */
3022                 WARN_ON_ONCE(1);
3023                 /* fall through */
3024         case SP4_SSV:
3025                 status = nfserr_encr_alg_unsupp;
3026                 goto out_nolock;
3027         }
3028
3029         /* Cases below refer to rfc 5661 section 18.35.4: */
3030         spin_lock(&nn->client_lock);
3031         conf = find_confirmed_client_by_name(&exid->clname, nn);
3032         if (conf) {
3033                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3034                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3035
3036                 if (update) {
3037                         if (!clp_used_exchangeid(conf)) { /* buggy client */
3038                                 status = nfserr_inval;
3039                                 goto out;
3040                         }
3041                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
3042                                 status = nfserr_wrong_cred;
3043                                 goto out;
3044                         }
3045                         if (!creds_match) { /* case 9 */
3046                                 status = nfserr_perm;
3047                                 goto out;
3048                         }
3049                         if (!verfs_match) { /* case 8 */
3050                                 status = nfserr_not_same;
3051                                 goto out;
3052                         }
3053                         /* case 6 */
3054                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3055                         goto out_copy;
3056                 }
3057                 if (!creds_match) { /* case 3 */
3058                         if (client_has_state(conf)) {
3059                                 status = nfserr_clid_inuse;
3060                                 goto out;
3061                         }
3062                         goto out_new;
3063                 }
3064                 if (verfs_match) { /* case 2 */
3065                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3066                         goto out_copy;
3067                 }
3068                 /* case 5, client reboot */
3069                 conf = NULL;
3070                 goto out_new;
3071         }
3072
3073         if (update) { /* case 7 */
3074                 status = nfserr_noent;
3075                 goto out;
3076         }
3077
3078         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
3079         if (unconf) /* case 4, possible retry or client restart */
3080                 unhash_client_locked(unconf);
3081
3082         /* case 1 (normal case) */
3083 out_new:
3084         if (conf) {
3085                 status = mark_client_expired_locked(conf);
3086                 if (status)
3087                         goto out;
3088         }
3089         new->cl_minorversion = cstate->minorversion;
3090         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3091         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3092
3093         add_to_unconfirmed(new);
3094         swap(new, conf);
3095 out_copy:
3096         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3097         exid->clientid.cl_id = conf->cl_clientid.cl_id;
3098
3099         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3100         nfsd4_set_ex_flags(conf, exid);
3101
3102         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3103                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3104         status = nfs_ok;
3105
3106 out:
3107         spin_unlock(&nn->client_lock);
3108 out_nolock:
3109         if (new)
3110                 expire_client(new);
3111         if (unconf)
3112                 expire_client(unconf);
3113         return status;
3114 }
3115
3116 static __be32
3117 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3118 {
3119         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3120                 slot_seqid);
3121
3122         /* The slot is in use, and no response has been sent. */
3123         if (slot_inuse) {
3124                 if (seqid == slot_seqid)
3125                         return nfserr_jukebox;
3126                 else
3127                         return nfserr_seq_misordered;
3128         }
3129         /* Note unsigned 32-bit arithmetic handles wraparound: */
3130         if (likely(seqid == slot_seqid + 1))
3131                 return nfs_ok;
3132         if (seqid == slot_seqid)
3133                 return nfserr_replay_cache;
3134         return nfserr_seq_misordered;
3135 }
3136
3137 /*
3138  * Cache the create session result into the create session single DRC
3139  * slot cache by saving the xdr structure. sl_seqid has been set.
3140  * Do this for solo or embedded create session operations.
3141  */
3142 static void
3143 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3144                            struct nfsd4_clid_slot *slot, __be32 nfserr)
3145 {
3146         slot->sl_status = nfserr;
3147         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3148 }
3149
3150 static __be32
3151 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3152                             struct nfsd4_clid_slot *slot)
3153 {
3154         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3155         return slot->sl_status;
3156 }
3157
3158 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3159                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3160                         1 +     /* MIN tag is length with zero, only length */ \
3161                         3 +     /* version, opcount, opcode */ \
3162                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3163                                 /* seqid, slotID, slotID, cache */ \
3164                         4 ) * sizeof(__be32))
3165
3166 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3167                         2 +     /* verifier: AUTH_NULL, length 0 */\
3168                         1 +     /* status */ \
3169                         1 +     /* MIN tag is length with zero, only length */ \
3170                         3 +     /* opcount, opcode, opstatus*/ \
3171                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3172                                 /* seqid, slotID, slotID, slotID, status */ \
3173                         5 ) * sizeof(__be32))
3174
3175 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3176 {
3177         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3178
3179         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3180                 return nfserr_toosmall;
3181         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3182                 return nfserr_toosmall;
3183         ca->headerpadsz = 0;
3184         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3185         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3186         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3187         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3188                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3189         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3190         /*
3191          * Note decreasing slot size below client's request may make it
3192          * difficult for client to function correctly, whereas
3193          * decreasing the number of slots will (just?) affect
3194          * performance.  When short on memory we therefore prefer to
3195          * decrease number of slots instead of their size.  Clients that
3196          * request larger slots than they need will get poor results:
3197          * Note that we always allow at least one slot, because our
3198          * accounting is soft and provides no guarantees either way.
3199          */
3200         ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3201
3202         return nfs_ok;
3203 }
3204
3205 /*
3206  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3207  * These are based on similar macros in linux/sunrpc/msg_prot.h .
3208  */
3209 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3210         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3211
3212 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3213         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3214
3215 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
3216                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3217 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
3218                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3219                                  sizeof(__be32))
3220
3221 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3222 {
3223         ca->headerpadsz = 0;
3224
3225         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3226                 return nfserr_toosmall;
3227         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3228                 return nfserr_toosmall;
3229         ca->maxresp_cached = 0;
3230         if (ca->maxops < 2)
3231                 return nfserr_toosmall;
3232
3233         return nfs_ok;
3234 }
3235
3236 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3237 {
3238         switch (cbs->flavor) {
3239         case RPC_AUTH_NULL:
3240         case RPC_AUTH_UNIX:
3241                 return nfs_ok;
3242         default:
3243                 /*
3244                  * GSS case: the spec doesn't allow us to return this
3245                  * error.  But it also doesn't allow us not to support
3246                  * GSS.
3247                  * I'd rather this fail hard than return some error the
3248                  * client might think it can already handle:
3249                  */
3250                 return nfserr_encr_alg_unsupp;
3251         }
3252 }
3253
3254 __be32
3255 nfsd4_create_session(struct svc_rqst *rqstp,
3256                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3257 {
3258         struct nfsd4_create_session *cr_ses = &u->create_session;
3259         struct sockaddr *sa = svc_addr(rqstp);
3260         struct nfs4_client *conf, *unconf;
3261         struct nfs4_client *old = NULL;
3262         struct nfsd4_session *new;
3263         struct nfsd4_conn *conn;
3264         struct nfsd4_clid_slot *cs_slot = NULL;
3265         __be32 status = 0;
3266         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3267
3268         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3269                 return nfserr_inval;
3270         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3271         if (status)
3272                 return status;
3273         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3274         if (status)
3275                 return status;
3276         status = check_backchannel_attrs(&cr_ses->back_channel);
3277         if (status)
3278                 goto out_release_drc_mem;
3279         status = nfserr_jukebox;
3280         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3281         if (!new)
3282                 goto out_release_drc_mem;
3283         conn = alloc_conn_from_crses(rqstp, cr_ses);
3284         if (!conn)
3285                 goto out_free_session;
3286
3287         spin_lock(&nn->client_lock);
3288         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3289         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3290         WARN_ON_ONCE(conf && unconf);
3291
3292         if (conf) {
3293                 status = nfserr_wrong_cred;
3294                 if (!nfsd4_mach_creds_match(conf, rqstp))
3295                         goto out_free_conn;
3296                 cs_slot = &conf->cl_cs_slot;
3297                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3298                 if (status) {
3299                         if (status == nfserr_replay_cache)
3300                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3301                         goto out_free_conn;
3302                 }
3303         } else if (unconf) {
3304                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3305                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3306                         status = nfserr_clid_inuse;
3307                         goto out_free_conn;
3308                 }
3309                 status = nfserr_wrong_cred;
3310                 if (!nfsd4_mach_creds_match(unconf, rqstp))
3311                         goto out_free_conn;
3312                 cs_slot = &unconf->cl_cs_slot;
3313                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3314                 if (status) {
3315                         /* an unconfirmed replay returns misordered */
3316                         status = nfserr_seq_misordered;
3317                         goto out_free_conn;
3318                 }
3319                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3320                 if (old) {
3321                         status = mark_client_expired_locked(old);
3322                         if (status) {
3323                                 old = NULL;
3324                                 goto out_free_conn;
3325                         }
3326                 }
3327                 move_to_confirmed(unconf);
3328                 conf = unconf;
3329         } else {
3330                 status = nfserr_stale_clientid;
3331                 goto out_free_conn;
3332         }
3333         status = nfs_ok;
3334         /* Persistent sessions are not supported */
3335         cr_ses->flags &= ~SESSION4_PERSIST;
3336         /* Upshifting from TCP to RDMA is not supported */
3337         cr_ses->flags &= ~SESSION4_RDMA;
3338
3339         init_session(rqstp, new, conf, cr_ses);
3340         nfsd4_get_session_locked(new);
3341
3342         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3343                NFS4_MAX_SESSIONID_LEN);
3344         cs_slot->sl_seqid++;
3345         cr_ses->seqid = cs_slot->sl_seqid;
3346
3347         /* cache solo and embedded create sessions under the client_lock */
3348         nfsd4_cache_create_session(cr_ses, cs_slot, status);
3349         spin_unlock(&nn->client_lock);
3350         /* init connection and backchannel */
3351         nfsd4_init_conn(rqstp, conn, new);
3352         nfsd4_put_session(new);
3353         if (old)
3354                 expire_client(old);
3355         return status;
3356 out_free_conn:
3357         spin_unlock(&nn->client_lock);
3358         free_conn(conn);
3359         if (old)
3360                 expire_client(old);
3361 out_free_session:
3362         __free_session(new);
3363 out_release_drc_mem:
3364         nfsd4_put_drc_mem(&cr_ses->fore_channel);
3365         return status;
3366 }
3367
3368 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3369 {
3370         switch (*dir) {
3371         case NFS4_CDFC4_FORE:
3372         case NFS4_CDFC4_BACK:
3373                 return nfs_ok;
3374         case NFS4_CDFC4_FORE_OR_BOTH:
3375         case NFS4_CDFC4_BACK_OR_BOTH:
3376                 *dir = NFS4_CDFC4_BOTH;
3377                 return nfs_ok;
3378         };
3379         return nfserr_inval;
3380 }
3381
3382 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3383                 struct nfsd4_compound_state *cstate,
3384                 union nfsd4_op_u *u)
3385 {
3386         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3387         struct nfsd4_session *session = cstate->session;
3388         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3389         __be32 status;
3390
3391         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3392         if (status)
3393                 return status;
3394         spin_lock(&nn->client_lock);
3395         session->se_cb_prog = bc->bc_cb_program;
3396         session->se_cb_sec = bc->bc_cb_sec;
3397         spin_unlock(&nn->client_lock);
3398
3399         nfsd4_probe_callback(session->se_client);
3400
3401         return nfs_ok;
3402 }
3403
3404 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3405                      struct nfsd4_compound_state *cstate,
3406                      union nfsd4_op_u *u)
3407 {
3408         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3409         __be32 status;
3410         struct nfsd4_conn *conn;
3411         struct nfsd4_session *session;
3412         struct net *net = SVC_NET(rqstp);
3413         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3414
3415         if (!nfsd4_last_compound_op(rqstp))
3416                 return nfserr_not_only_op;
3417         spin_lock(&nn->client_lock);
3418         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3419         spin_unlock(&nn->client_lock);
3420         if (!session)
3421                 goto out_no_session;
3422         status = nfserr_wrong_cred;
3423         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3424                 goto out;
3425         status = nfsd4_map_bcts_dir(&bcts->dir);
3426         if (status)
3427                 goto out;
3428         conn = alloc_conn(rqstp, bcts->dir);
3429         status = nfserr_jukebox;
3430         if (!conn)
3431                 goto out;
3432         nfsd4_init_conn(rqstp, conn, session);
3433         status = nfs_ok;
3434 out:
3435         nfsd4_put_session(session);
3436 out_no_session:
3437         return status;
3438 }
3439
3440 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3441 {
3442         if (!cstate->session)
3443                 return false;
3444         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3445 }
3446
3447 __be32
3448 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3449                 union nfsd4_op_u *u)
3450 {
3451         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3452         struct nfsd4_session *ses;
3453         __be32 status;
3454         int ref_held_by_me = 0;
3455         struct net *net = SVC_NET(r);
3456         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3457
3458         status = nfserr_not_only_op;
3459         if (nfsd4_compound_in_session(cstate, sessionid)) {
3460                 if (!nfsd4_last_compound_op(r))
3461                         goto out;
3462                 ref_held_by_me++;
3463         }
3464         dump_sessionid(__func__, sessionid);
3465         spin_lock(&nn->client_lock);
3466         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3467         if (!ses)
3468                 goto out_client_lock;
3469         status = nfserr_wrong_cred;
3470         if (!nfsd4_mach_creds_match(ses->se_client, r))
3471                 goto out_put_session;
3472         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3473         if (status)
3474                 goto out_put_session;
3475         unhash_session(ses);
3476         spin_unlock(&nn->client_lock);
3477
3478         nfsd4_probe_callback_sync(ses->se_client);
3479
3480         spin_lock(&nn->client_lock);
3481         status = nfs_ok;
3482 out_put_session:
3483         nfsd4_put_session_locked(ses);
3484 out_client_lock:
3485         spin_unlock(&nn->client_lock);
3486 out:
3487         return status;
3488 }
3489
3490 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3491 {
3492         struct nfsd4_conn *c;
3493
3494         list_for_each_entry(c, &s->se_conns, cn_persession) {
3495                 if (c->cn_xprt == xpt) {
3496                         return c;
3497                 }
3498         }
3499         return NULL;
3500 }
3501
3502 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3503 {
3504         struct nfs4_client *clp = ses->se_client;
3505         struct nfsd4_conn *c;
3506         __be32 status = nfs_ok;
3507         int ret;
3508
3509         spin_lock(&clp->cl_lock);
3510         c = __nfsd4_find_conn(new->cn_xprt, ses);
3511         if (c)
3512                 goto out_free;
3513         status = nfserr_conn_not_bound_to_session;
3514         if (clp->cl_mach_cred)
3515                 goto out_free;
3516         __nfsd4_hash_conn(new, ses);
3517         spin_unlock(&clp->cl_lock);
3518         ret = nfsd4_register_conn(new);
3519         if (ret)
3520                 /* oops; xprt is already down: */
3521                 nfsd4_conn_lost(&new->cn_xpt_user);
3522         return nfs_ok;
3523 out_free:
3524         spin_unlock(&clp->cl_lock);
3525         free_conn(new);
3526         return status;
3527 }
3528
3529 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3530 {
3531         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3532
3533         return args->opcnt > session->se_fchannel.maxops;
3534 }
3535
3536 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3537                                   struct nfsd4_session *session)
3538 {
3539         struct xdr_buf *xb = &rqstp->rq_arg;
3540
3541         return xb->len > session->se_fchannel.maxreq_sz;
3542 }
3543
3544 static bool replay_matches_cache(struct svc_rqst *rqstp,
3545                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3546 {
3547         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3548
3549         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3550             (bool)seq->cachethis)
3551                 return false;
3552         /*
3553          * If there's an error then the reply can have fewer ops than
3554          * the call.
3555          */
3556         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3557                 return false;
3558         /*
3559          * But if we cached a reply with *more* ops than the call you're
3560          * sending us now, then this new call is clearly not really a
3561          * replay of the old one:
3562          */
3563         if (slot->sl_opcnt > argp->opcnt)
3564                 return false;
3565         /* This is the only check explicitly called by spec: */
3566         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3567                 return false;
3568         /*
3569          * There may be more comparisons we could actually do, but the
3570          * spec doesn't require us to catch every case where the calls
3571          * don't match (that would require caching the call as well as
3572          * the reply), so we don't bother.
3573          */
3574         return true;
3575 }
3576
3577 __be32
3578 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3579                 union nfsd4_op_u *u)
3580 {
3581         struct nfsd4_sequence *seq = &u->sequence;
3582         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3583         struct xdr_stream *xdr = &resp->xdr;
3584         struct nfsd4_session *session;
3585         struct nfs4_client *clp;
3586         struct nfsd4_slot *slot;
3587         struct nfsd4_conn *conn;
3588         __be32 status;
3589         int buflen;
3590         struct net *net = SVC_NET(rqstp);
3591         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3592
3593         if (resp->opcnt != 1)
3594                 return nfserr_sequence_pos;
3595
3596         /*
3597          * Will be either used or freed by nfsd4_sequence_check_conn
3598          * below.
3599          */
3600         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3601         if (!conn)
3602                 return nfserr_jukebox;
3603
3604         spin_lock(&nn->client_lock);
3605         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3606         if (!session)
3607                 goto out_no_session;
3608         clp = session->se_client;
3609
3610         status = nfserr_too_many_ops;
3611         if (nfsd4_session_too_many_ops(rqstp, session))
3612                 goto out_put_session;
3613
3614         status = nfserr_req_too_big;
3615         if (nfsd4_request_too_big(rqstp, session))
3616                 goto out_put_session;
3617
3618         status = nfserr_badslot;
3619         if (seq->slotid >= session->se_fchannel.maxreqs)
3620                 goto out_put_session;
3621
3622         slot = session->se_slots[seq->slotid];
3623         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3624
3625         /* We do not negotiate the number of slots yet, so set the
3626          * maxslots to the session maxreqs which is used to encode
3627          * sr_highest_slotid and the sr_target_slot id to maxslots */
3628         seq->maxslots = session->se_fchannel.maxreqs;
3629
3630         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3631                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3632         if (status == nfserr_replay_cache) {
3633                 status = nfserr_seq_misordered;
3634                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3635                         goto out_put_session;
3636                 status = nfserr_seq_false_retry;
3637                 if (!replay_matches_cache(rqstp, seq, slot))
3638                         goto out_put_session;
3639                 cstate->slot = slot;
3640                 cstate->session = session;
3641                 cstate->clp = clp;
3642                 /* Return the cached reply status and set cstate->status
3643                  * for nfsd4_proc_compound processing */
3644                 status = nfsd4_replay_cache_entry(resp, seq);
3645                 cstate->status = nfserr_replay_cache;
3646                 goto out;
3647         }
3648         if (status)
3649                 goto out_put_session;
3650
3651         status = nfsd4_sequence_check_conn(conn, session);
3652         conn = NULL;
3653         if (status)
3654                 goto out_put_session;
3655
3656         buflen = (seq->cachethis) ?
3657                         session->se_fchannel.maxresp_cached :
3658                         session->se_fchannel.maxresp_sz;
3659         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3660                                     nfserr_rep_too_big;
3661         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3662                 goto out_put_session;
3663         svc_reserve(rqstp, buflen);
3664
3665         status = nfs_ok;
3666         /* Success! bump slot seqid */
3667         slot->sl_seqid = seq->seqid;
3668         slot->sl_flags |= NFSD4_SLOT_INUSE;
3669         if (seq->cachethis)
3670                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3671         else
3672                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3673
3674         cstate->slot = slot;
3675         cstate->session = session;
3676         cstate->clp = clp;
3677
3678 out:
3679         switch (clp->cl_cb_state) {
3680         case NFSD4_CB_DOWN:
3681                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3682                 break;
3683         case NFSD4_CB_FAULT:
3684                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3685                 break;
3686         default:
3687                 seq->status_flags = 0;
3688         }
3689         if (!list_empty(&clp->cl_revoked))
3690                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3691 out_no_session:
3692         if (conn)
3693                 free_conn(conn);
3694         spin_unlock(&nn->client_lock);
3695         return status;
3696 out_put_session:
3697         nfsd4_put_session_locked(session);
3698         goto out_no_session;
3699 }
3700
3701 void
3702 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3703 {
3704         struct nfsd4_compound_state *cs = &resp->cstate;
3705
3706         if (nfsd4_has_session(cs)) {
3707                 if (cs->status != nfserr_replay_cache) {
3708                         nfsd4_store_cache_entry(resp);
3709                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3710                 }
3711                 /* Drop session reference that was taken in nfsd4_sequence() */
3712                 nfsd4_put_session(cs->session);
3713         } else if (cs->clp)
3714                 put_client_renew(cs->clp);
3715 }
3716
3717 __be32
3718 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3719                 struct nfsd4_compound_state *cstate,
3720                 union nfsd4_op_u *u)
3721 {
3722         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3723         struct nfs4_client *conf, *unconf;
3724         struct nfs4_client *clp = NULL;
3725         __be32 status = 0;
3726         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3727
3728         spin_lock(&nn->client_lock);
3729         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3730         conf = find_confirmed_client(&dc->clientid, true, nn);
3731         WARN_ON_ONCE(conf && unconf);
3732
3733         if (conf) {
3734                 if (client_has_state(conf)) {
3735                         status = nfserr_clientid_busy;
3736                         goto out;
3737                 }
3738                 status = mark_client_expired_locked(conf);
3739                 if (status)
3740                         goto out;
3741                 clp = conf;
3742         } else if (unconf)
3743                 clp = unconf;
3744         else {
3745                 status = nfserr_stale_clientid;
3746                 goto out;
3747         }
3748         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3749                 clp = NULL;
3750                 status = nfserr_wrong_cred;
3751                 goto out;
3752         }
3753         unhash_client_locked(clp);
3754 out:
3755         spin_unlock(&nn->client_lock);
3756         if (clp)
3757                 expire_client(clp);
3758         return status;
3759 }
3760
3761 __be32
3762 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3763                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3764 {
3765         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3766         __be32 status = 0;
3767
3768         if (rc->rca_one_fs) {
3769                 if (!cstate->current_fh.fh_dentry)
3770                         return nfserr_nofilehandle;
3771                 /*
3772                  * We don't take advantage of the rca_one_fs case.
3773                  * That's OK, it's optional, we can safely ignore it.
3774                  */
3775                 return nfs_ok;
3776         }
3777
3778         status = nfserr_complete_already;
3779         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3780                              &cstate->session->se_client->cl_flags))
3781                 goto out;
3782
3783         status = nfserr_stale_clientid;
3784         if (is_client_expired(cstate->session->se_client))
3785                 /*
3786                  * The following error isn't really legal.
3787                  * But we only get here if the client just explicitly
3788                  * destroyed the client.  Surely it no longer cares what
3789                  * error it gets back on an operation for the dead
3790                  * client.
3791                  */
3792                 goto out;
3793
3794         status = nfs_ok;
3795         nfsd4_client_record_create(cstate->session->se_client);
3796         inc_reclaim_complete(cstate->session->se_client);
3797 out:
3798         return status;
3799 }
3800
3801 __be32
3802 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3803                   union nfsd4_op_u *u)
3804 {
3805         struct nfsd4_setclientid *setclid = &u->setclientid;
3806         struct xdr_netobj       clname = setclid->se_name;
3807         nfs4_verifier           clverifier = setclid->se_verf;
3808         struct nfs4_client      *conf, *new;
3809         struct nfs4_client      *unconf = NULL;
3810         __be32                  status;
3811         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3812
3813         new = create_client(clname, rqstp, &clverifier);
3814         if (new == NULL)
3815                 return nfserr_jukebox;
3816         /* Cases below refer to rfc 3530 section 14.2.33: */
3817         spin_lock(&nn->client_lock);
3818         conf = find_confirmed_client_by_name(&clname, nn);
3819         if (conf && client_has_state(conf)) {
3820                 /* case 0: */
3821                 status = nfserr_clid_inuse;
3822                 if (clp_used_exchangeid(conf))
3823                         goto out;
3824                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3825                         char addr_str[INET6_ADDRSTRLEN];
3826                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3827                                  sizeof(addr_str));
3828                         dprintk("NFSD: setclientid: string in use by client "
3829                                 "at %s\n", addr_str);
3830                         goto out;
3831                 }
3832         }
3833         unconf = find_unconfirmed_client_by_name(&clname, nn);
3834         if (unconf)
3835                 unhash_client_locked(unconf);
3836         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3837                 /* case 1: probable callback update */
3838                 copy_clid(new, conf);
3839                 gen_confirm(new, nn);
3840         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3841                 ;
3842         new->cl_minorversion = 0;
3843         gen_callback(new, setclid, rqstp);
3844         add_to_unconfirmed(new);
3845         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3846         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3847         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3848         new = NULL;
3849         status = nfs_ok;
3850 out:
3851         spin_unlock(&nn->client_lock);
3852         if (new)
3853                 free_client(new);
3854         if (unconf)
3855                 expire_client(unconf);
3856         return status;
3857 }
3858
3859
3860 __be32
3861 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3862                         struct nfsd4_compound_state *cstate,
3863                         union nfsd4_op_u *u)
3864 {
3865         struct nfsd4_setclientid_confirm *setclientid_confirm =
3866                         &u->setclientid_confirm;
3867         struct nfs4_client *conf, *unconf;
3868         struct nfs4_client *old = NULL;
3869         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3870         clientid_t * clid = &setclientid_confirm->sc_clientid;
3871         __be32 status;
3872         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3873
3874         if (STALE_CLIENTID(clid, nn))
3875                 return nfserr_stale_clientid;
3876
3877         spin_lock(&nn->client_lock);
3878         conf = find_confirmed_client(clid, false, nn);
3879         unconf = find_unconfirmed_client(clid, false, nn);
3880         /*
3881          * We try hard to give out unique clientid's, so if we get an
3882          * attempt to confirm the same clientid with a different cred,
3883          * the client may be buggy; this should never happen.
3884          *
3885          * Nevertheless, RFC 7530 recommends INUSE for this case:
3886          */
3887         status = nfserr_clid_inuse;
3888         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3889                 goto out;
3890         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3891                 goto out;
3892         /* cases below refer to rfc 3530 section 14.2.34: */
3893         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3894                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3895                         /* case 2: probable retransmit */
3896                         status = nfs_ok;
3897                 } else /* case 4: client hasn't noticed we rebooted yet? */
3898                         status = nfserr_stale_clientid;
3899                 goto out;
3900         }
3901         status = nfs_ok;
3902         if (conf) { /* case 1: callback update */
3903                 old = unconf;
3904                 unhash_client_locked(old);
3905                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3906         } else { /* case 3: normal case; new or rebooted client */
3907                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3908                 if (old) {
3909                         status = nfserr_clid_inuse;
3910                         if (client_has_state(old)
3911                                         && !same_creds(&unconf->cl_cred,
3912                                                         &old->cl_cred))
3913                                 goto out;
3914                         status = mark_client_expired_locked(old);
3915                         if (status) {
3916                                 old = NULL;
3917                                 goto out;
3918                         }
3919                 }
3920                 move_to_confirmed(unconf);
3921                 conf = unconf;
3922         }
3923         get_client_locked(conf);
3924         spin_unlock(&nn->client_lock);
3925         nfsd4_probe_callback(conf);
3926         spin_lock(&nn->client_lock);
3927         put_client_renew_locked(conf);
3928 out:
3929         spin_unlock(&nn->client_lock);
3930         if (old)
3931                 expire_client(old);
3932         return status;
3933 }
3934
3935 static struct nfs4_file *nfsd4_alloc_file(void)
3936 {
3937         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3938 }
3939
3940 /* OPEN Share state helper functions */
3941 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3942                                 struct nfs4_file *fp)
3943 {
3944         lockdep_assert_held(&state_lock);
3945
3946         refcount_set(&fp->fi_ref, 1);
3947         spin_lock_init(&fp->fi_lock);
3948         INIT_LIST_HEAD(&fp->fi_stateids);
3949         INIT_LIST_HEAD(&fp->fi_delegations);
3950         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3951         fh_copy_shallow(&fp->fi_fhandle, fh);
3952         fp->fi_deleg_file = NULL;
3953         fp->fi_had_conflict = false;
3954         fp->fi_share_deny = 0;
3955         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3956         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3957 #ifdef CONFIG_NFSD_PNFS
3958         INIT_LIST_HEAD(&fp->fi_lo_states);
3959         atomic_set(&fp->fi_lo_recalls, 0);
3960 #endif
3961         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3962 }
3963
3964 void
3965 nfsd4_free_slabs(void)
3966 {
3967         kmem_cache_destroy(client_slab);
3968         kmem_cache_destroy(openowner_slab);
3969         kmem_cache_destroy(lockowner_slab);
3970         kmem_cache_destroy(file_slab);
3971         kmem_cache_destroy(stateid_slab);
3972         kmem_cache_destroy(deleg_slab);
3973         kmem_cache_destroy(odstate_slab);
3974 }
3975
3976 int
3977 nfsd4_init_slabs(void)
3978 {
3979         client_slab = kmem_cache_create("nfsd4_clients",
3980                         sizeof(struct nfs4_client), 0, 0, NULL);
3981         if (client_slab == NULL)
3982                 goto out;
3983         openowner_slab = kmem_cache_create("nfsd4_openowners",
3984                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3985         if (openowner_slab == NULL)
3986                 goto out_free_client_slab;
3987         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3988                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3989         if (lockowner_slab == NULL)
3990                 goto out_free_openowner_slab;
3991         file_slab = kmem_cache_create("nfsd4_files",
3992                         sizeof(struct nfs4_file), 0, 0, NULL);
3993         if (file_slab == NULL)
3994                 goto out_free_lockowner_slab;
3995         stateid_slab = kmem_cache_create("nfsd4_stateids",
3996                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3997         if (stateid_slab == NULL)
3998                 goto out_free_file_slab;
3999         deleg_slab = kmem_cache_create("nfsd4_delegations",
4000                         sizeof(struct nfs4_delegation), 0, 0, NULL);
4001         if (deleg_slab == NULL)
4002                 goto out_free_stateid_slab;
4003         odstate_slab = kmem_cache_create("nfsd4_odstate",
4004                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4005         if (odstate_slab == NULL)
4006                 goto out_free_deleg_slab;
4007         return 0;
4008
4009 out_free_deleg_slab:
4010         kmem_cache_destroy(deleg_slab);
4011 out_free_stateid_slab:
4012         kmem_cache_destroy(stateid_slab);
4013 out_free_file_slab:
4014         kmem_cache_destroy(file_slab);
4015 out_free_lockowner_slab:
4016         kmem_cache_destroy(lockowner_slab);
4017 out_free_openowner_slab:
4018         kmem_cache_destroy(openowner_slab);
4019 out_free_client_slab:
4020         kmem_cache_destroy(client_slab);
4021 out:
4022         dprintk("nfsd4: out of memory while initializing nfsv4\n");
4023         return -ENOMEM;
4024 }
4025
4026 static void init_nfs4_replay(struct nfs4_replay *rp)
4027 {
4028         rp->rp_status = nfserr_serverfault;
4029         rp->rp_buflen = 0;
4030         rp->rp_buf = rp->rp_ibuf;
4031         mutex_init(&rp->rp_mutex);
4032 }
4033
4034 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4035                 struct nfs4_stateowner *so)
4036 {
4037         if (!nfsd4_has_session(cstate)) {
4038                 mutex_lock(&so->so_replay.rp_mutex);
4039                 cstate->replay_owner = nfs4_get_stateowner(so);
4040         }
4041 }
4042
4043 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4044 {
4045         struct nfs4_stateowner *so = cstate->replay_owner;
4046
4047         if (so != NULL) {
4048                 cstate->replay_owner = NULL;
4049                 mutex_unlock(&so->so_replay.rp_mutex);
4050                 nfs4_put_stateowner(so);
4051         }
4052 }
4053
4054 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4055 {
4056         struct nfs4_stateowner *sop;
4057
4058         sop = kmem_cache_alloc(slab, GFP_KERNEL);
4059         if (!sop)
4060                 return NULL;
4061
4062         xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4063         if (!sop->so_owner.data) {
4064                 kmem_cache_free(slab, sop);
4065                 return NULL;
4066         }
4067
4068         INIT_LIST_HEAD(&sop->so_stateids);
4069         sop->so_client = clp;
4070         init_nfs4_replay(&sop->so_replay);
4071         atomic_set(&sop->so_count, 1);
4072         return sop;
4073 }
4074
4075 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4076 {
4077         lockdep_assert_held(&clp->cl_lock);
4078
4079         list_add(&oo->oo_owner.so_strhash,
4080                  &clp->cl_ownerstr_hashtbl[strhashval]);
4081         list_add(&oo->oo_perclient, &clp->cl_openowners);
4082 }
4083
4084 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4085 {
4086         unhash_openowner_locked(openowner(so));
4087 }
4088
4089 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4090 {
4091         struct nfs4_openowner *oo = openowner(so);
4092
4093         kmem_cache_free(openowner_slab, oo);
4094 }
4095
4096 static const struct nfs4_stateowner_operations openowner_ops = {
4097         .so_unhash =    nfs4_unhash_openowner,
4098         .so_free =      nfs4_free_openowner,
4099 };
4100
4101 static struct nfs4_ol_stateid *
4102 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4103 {
4104         struct nfs4_ol_stateid *local, *ret = NULL;
4105         struct nfs4_openowner *oo = open->op_openowner;
4106
4107         lockdep_assert_held(&fp->fi_lock);
4108
4109         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4110                 /* ignore lock owners */
4111                 if (local->st_stateowner->so_is_open_owner == 0)
4112                         continue;
4113                 if (local->st_stateowner != &oo->oo_owner)
4114                         continue;
4115                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4116                         ret = local;
4117                         refcount_inc(&ret->st_stid.sc_count);
4118                         break;
4119                 }
4120         }
4121         return ret;
4122 }
4123
4124 static __be32
4125 nfsd4_verify_open_stid(struct nfs4_stid *s)
4126 {
4127         __be32 ret = nfs_ok;
4128
4129         switch (s->sc_type) {
4130         default:
4131                 break;
4132         case 0:
4133         case NFS4_CLOSED_STID:
4134         case NFS4_CLOSED_DELEG_STID:
4135                 ret = nfserr_bad_stateid;
4136                 break;
4137         case NFS4_REVOKED_DELEG_STID:
4138                 ret = nfserr_deleg_revoked;
4139         }
4140         return ret;
4141 }
4142
4143 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4144 static __be32
4145 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4146 {
4147         __be32 ret;
4148
4149         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4150         ret = nfsd4_verify_open_stid(&stp->st_stid);
4151         if (ret != nfs_ok)
4152                 mutex_unlock(&stp->st_mutex);
4153         return ret;
4154 }
4155
4156 static struct nfs4_ol_stateid *
4157 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4158 {
4159         struct nfs4_ol_stateid *stp;
4160         for (;;) {
4161                 spin_lock(&fp->fi_lock);
4162                 stp = nfsd4_find_existing_open(fp, open);
4163                 spin_unlock(&fp->fi_lock);
4164                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4165                         break;
4166                 nfs4_put_stid(&stp->st_stid);
4167         }
4168         return stp;
4169 }
4170
4171 static struct nfs4_openowner *
4172 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4173                            struct nfsd4_compound_state *cstate)
4174 {
4175         struct nfs4_client *clp = cstate->clp;
4176         struct nfs4_openowner *oo, *ret;
4177
4178         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4179         if (!oo)
4180                 return NULL;
4181         oo->oo_owner.so_ops = &openowner_ops;
4182         oo->oo_owner.so_is_open_owner = 1;
4183         oo->oo_owner.so_seqid = open->op_seqid;
4184         oo->oo_flags = 0;
4185         if (nfsd4_has_session(cstate))
4186                 oo->oo_flags |= NFS4_OO_CONFIRMED;
4187         oo->oo_time = 0;
4188         oo->oo_last_closed_stid = NULL;
4189         INIT_LIST_HEAD(&oo->oo_close_lru);
4190         spin_lock(&clp->cl_lock);
4191         ret = find_openstateowner_str_locked(strhashval, open, clp);
4192         if (ret == NULL) {
4193                 hash_openowner(oo, clp, strhashval);
4194                 ret = oo;
4195         } else
4196                 nfs4_free_stateowner(&oo->oo_owner);
4197
4198         spin_unlock(&clp->cl_lock);
4199         return ret;
4200 }
4201
4202 static struct nfs4_ol_stateid *
4203 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4204 {
4205
4206         struct nfs4_openowner *oo = open->op_openowner;
4207         struct nfs4_ol_stateid *retstp = NULL;
4208         struct nfs4_ol_stateid *stp;
4209
4210         stp = open->op_stp;
4211         /* We are moving these outside of the spinlocks to avoid the warnings */
4212         mutex_init(&stp->st_mutex);
4213         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4214
4215 retry:
4216         spin_lock(&oo->oo_owner.so_client->cl_lock);
4217         spin_lock(&fp->fi_lock);
4218
4219         retstp = nfsd4_find_existing_open(fp, open);
4220         if (retstp)
4221                 goto out_unlock;
4222
4223         open->op_stp = NULL;
4224         refcount_inc(&stp->st_stid.sc_count);
4225         stp->st_stid.sc_type = NFS4_OPEN_STID;
4226         INIT_LIST_HEAD(&stp->st_locks);
4227         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4228         get_nfs4_file(fp);
4229         stp->st_stid.sc_file = fp;
4230         stp->st_access_bmap = 0;
4231         stp->st_deny_bmap = 0;
4232         stp->st_openstp = NULL;
4233         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4234         list_add(&stp->st_perfile, &fp->fi_stateids);
4235
4236 out_unlock:
4237         spin_unlock(&fp->fi_lock);
4238         spin_unlock(&oo->oo_owner.so_client->cl_lock);
4239         if (retstp) {
4240                 /* Handle races with CLOSE */
4241                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4242                         nfs4_put_stid(&retstp->st_stid);
4243                         goto retry;
4244                 }
4245                 /* To keep mutex tracking happy */
4246                 mutex_unlock(&stp->st_mutex);
4247                 stp = retstp;
4248         }
4249         return stp;
4250 }
4251
4252 /*
4253  * In the 4.0 case we need to keep the owners around a little while to handle
4254  * CLOSE replay. We still do need to release any file access that is held by
4255  * them before returning however.
4256  */
4257 static void
4258 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4259 {
4260         struct nfs4_ol_stateid *last;
4261         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4262         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4263                                                 nfsd_net_id);
4264
4265         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4266
4267         /*
4268          * We know that we hold one reference via nfsd4_close, and another
4269          * "persistent" reference for the client. If the refcount is higher
4270          * than 2, then there are still calls in progress that are using this
4271          * stateid. We can't put the sc_file reference until they are finished.
4272          * Wait for the refcount to drop to 2. Since it has been unhashed,
4273          * there should be no danger of the refcount going back up again at
4274          * this point.
4275          */
4276         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4277
4278         release_all_access(s);
4279         if (s->st_stid.sc_file) {
4280                 put_nfs4_file(s->st_stid.sc_file);
4281                 s->st_stid.sc_file = NULL;
4282         }
4283
4284         spin_lock(&nn->client_lock);
4285         last = oo->oo_last_closed_stid;
4286         oo->oo_last_closed_stid = s;
4287         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4288         oo->oo_time = get_seconds();
4289         spin_unlock(&nn->client_lock);
4290         if (last)
4291                 nfs4_put_stid(&last->st_stid);
4292 }
4293
4294 /* search file_hashtbl[] for file */
4295 static struct nfs4_file *
4296 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4297 {
4298         struct nfs4_file *fp;
4299
4300         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
4301                 if (fh_match(&fp->fi_fhandle, fh)) {
4302                         if (refcount_inc_not_zero(&fp->fi_ref))
4303                                 return fp;
4304                 }
4305         }
4306         return NULL;
4307 }
4308
4309 struct nfs4_file *
4310 find_file(struct knfsd_fh *fh)
4311 {
4312         struct nfs4_file *fp;
4313         unsigned int hashval = file_hashval(fh);
4314
4315         rcu_read_lock();
4316         fp = find_file_locked(fh, hashval);
4317         rcu_read_unlock();
4318         return fp;
4319 }
4320
4321 static struct nfs4_file *
4322 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4323 {
4324         struct nfs4_file *fp;
4325         unsigned int hashval = file_hashval(fh);
4326
4327         rcu_read_lock();
4328         fp = find_file_locked(fh, hashval);
4329         rcu_read_unlock();
4330         if (fp)
4331                 return fp;
4332
4333         spin_lock(&state_lock);
4334         fp = find_file_locked(fh, hashval);
4335         if (likely(fp == NULL)) {
4336                 nfsd4_init_file(fh, hashval, new);
4337                 fp = new;
4338         }
4339         spin_unlock(&state_lock);
4340
4341         return fp;
4342 }
4343
4344 /*
4345  * Called to check deny when READ with all zero stateid or
4346  * WRITE with all zero or all one stateid
4347  */
4348 static __be32
4349 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4350 {
4351         struct nfs4_file *fp;
4352         __be32 ret = nfs_ok;
4353
4354         fp = find_file(&current_fh->fh_handle);
4355         if (!fp)
4356                 return ret;
4357         /* Check for conflicting share reservations */
4358         spin_lock(&fp->fi_lock);
4359         if (fp->fi_share_deny & deny_type)
4360                 ret = nfserr_locked;
4361         spin_unlock(&fp->fi_lock);
4362         put_nfs4_file(fp);
4363         return ret;
4364 }
4365
4366 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4367 {
4368         struct nfs4_delegation *dp = cb_to_delegation(cb);
4369         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4370                                           nfsd_net_id);
4371
4372         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4373
4374         /*
4375          * We can't do this in nfsd_break_deleg_cb because it is
4376          * already holding inode->i_lock.
4377          *
4378          * If the dl_time != 0, then we know that it has already been
4379          * queued for a lease break. Don't queue it again.
4380          */
4381         spin_lock(&state_lock);
4382         if (dp->dl_time == 0) {
4383                 dp->dl_time = get_seconds();
4384                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4385         }
4386         spin_unlock(&state_lock);
4387 }
4388
4389 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4390                 struct rpc_task *task)
4391 {
4392         struct nfs4_delegation *dp = cb_to_delegation(cb);
4393
4394         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4395                 return 1;
4396
4397         switch (task->tk_status) {
4398         case 0:
4399                 return 1;
4400         case -NFS4ERR_DELAY:
4401                 rpc_delay(task, 2 * HZ);
4402                 return 0;
4403         case -EBADHANDLE:
4404         case -NFS4ERR_BAD_STATEID:
4405                 /*
4406                  * Race: client probably got cb_recall before open reply
4407                  * granting delegation.
4408                  */
4409                 if (dp->dl_retries--) {
4410                         rpc_delay(task, 2 * HZ);
4411                         return 0;
4412                 }
4413                 /*FALLTHRU*/
4414         default:
4415                 return 1;
4416         }
4417 }
4418
4419 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4420 {
4421         struct nfs4_delegation *dp = cb_to_delegation(cb);
4422
4423         nfs4_put_stid(&dp->dl_stid);
4424 }
4425
4426 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4427         .prepare        = nfsd4_cb_recall_prepare,
4428         .done           = nfsd4_cb_recall_done,
4429         .release        = nfsd4_cb_recall_release,
4430 };
4431
4432 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4433 {
4434         /*
4435          * We're assuming the state code never drops its reference
4436          * without first removing the lease.  Since we're in this lease
4437          * callback (and since the lease code is serialized by the
4438          * i_lock) we know the server hasn't removed the lease yet, and
4439          * we know it's safe to take a reference.
4440          */
4441         refcount_inc(&dp->dl_stid.sc_count);
4442         nfsd4_run_cb(&dp->dl_recall);
4443 }
4444
4445 /* Called from break_lease() with i_lock held. */
4446 static bool
4447 nfsd_break_deleg_cb(struct file_lock *fl)
4448 {
4449         bool ret = false;
4450         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4451         struct nfs4_file *fp = dp->dl_stid.sc_file;
4452
4453         /*
4454          * We don't want the locks code to timeout the lease for us;
4455          * we'll remove it ourself if a delegation isn't returned
4456          * in time:
4457          */
4458         fl->fl_break_time = 0;
4459
4460         spin_lock(&fp->fi_lock);
4461         fp->fi_had_conflict = true;
4462         nfsd_break_one_deleg(dp);
4463         spin_unlock(&fp->fi_lock);
4464         return ret;
4465 }
4466
4467 static int
4468 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4469                      struct list_head *dispose)
4470 {
4471         if (arg & F_UNLCK)
4472                 return lease_modify(onlist, arg, dispose);
4473         else
4474                 return -EAGAIN;
4475 }
4476
4477 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4478         .lm_break = nfsd_break_deleg_cb,
4479         .lm_change = nfsd_change_deleg_cb,
4480 };
4481
4482 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4483 {
4484         if (nfsd4_has_session(cstate))
4485                 return nfs_ok;
4486         if (seqid == so->so_seqid - 1)
4487                 return nfserr_replay_me;
4488         if (seqid == so->so_seqid)
4489                 return nfs_ok;
4490         return nfserr_bad_seqid;
4491 }
4492
4493 static __be32 lookup_clientid(clientid_t *clid,
4494                 struct nfsd4_compound_state *cstate,
4495                 struct nfsd_net *nn)
4496 {
4497         struct nfs4_client *found;
4498
4499         if (cstate->clp) {
4500                 found = cstate->clp;
4501                 if (!same_clid(&found->cl_clientid, clid))
4502                         return nfserr_stale_clientid;
4503                 return nfs_ok;
4504         }
4505
4506         if (STALE_CLIENTID(clid, nn))
4507                 return nfserr_stale_clientid;
4508
4509         /*
4510          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4511          * cached already then we know this is for is for v4.0 and "sessions"
4512          * will be false.
4513          */
4514         WARN_ON_ONCE(cstate->session);
4515         spin_lock(&nn->client_lock);
4516         found = find_confirmed_client(clid, false, nn);
4517         if (!found) {
4518                 spin_unlock(&nn->client_lock);
4519                 return nfserr_expired;
4520         }
4521         atomic_inc(&found->cl_rpc_users);
4522         spin_unlock(&nn->client_lock);
4523
4524         /* Cache the nfs4_client in cstate! */
4525         cstate->clp = found;
4526         return nfs_ok;
4527 }
4528
4529 __be32
4530 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4531                     struct nfsd4_open *open, struct nfsd_net *nn)
4532 {
4533         clientid_t *clientid = &open->op_clientid;
4534         struct nfs4_client *clp = NULL;
4535         unsigned int strhashval;
4536         struct nfs4_openowner *oo = NULL;
4537         __be32 status;
4538
4539         if (STALE_CLIENTID(&open->op_clientid, nn))
4540                 return nfserr_stale_clientid;
4541         /*
4542          * In case we need it later, after we've already created the
4543          * file and don't want to risk a further failure:
4544          */
4545         open->op_file = nfsd4_alloc_file();
4546         if (open->op_file == NULL)
4547                 return nfserr_jukebox;
4548
4549         status = lookup_clientid(clientid, cstate, nn);
4550         if (status)
4551                 return status;
4552         clp = cstate->clp;
4553
4554         strhashval = ownerstr_hashval(&open->op_owner);
4555         oo = find_openstateowner_str(strhashval, open, clp);
4556         open->op_openowner = oo;
4557         if (!oo) {
4558                 goto new_owner;
4559         }
4560         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4561                 /* Replace unconfirmed owners without checking for replay. */
4562                 release_openowner(oo);
4563                 open->op_openowner = NULL;
4564                 goto new_owner;
4565         }
4566         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4567         if (status)
4568                 return status;
4569         goto alloc_stateid;
4570 new_owner:
4571         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4572         if (oo == NULL)
4573                 return nfserr_jukebox;
4574         open->op_openowner = oo;
4575 alloc_stateid:
4576         open->op_stp = nfs4_alloc_open_stateid(clp);
4577         if (!open->op_stp)
4578                 return nfserr_jukebox;
4579
4580         if (nfsd4_has_session(cstate) &&
4581             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4582                 open->op_odstate = alloc_clnt_odstate(clp);
4583                 if (!open->op_odstate)
4584                         return nfserr_jukebox;
4585         }
4586
4587         return nfs_ok;
4588 }
4589
4590 static inline __be32
4591 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4592 {
4593         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4594                 return nfserr_openmode;
4595         else
4596                 return nfs_ok;
4597 }
4598
4599 static int share_access_to_flags(u32 share_access)
4600 {
4601         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4602 }
4603
4604 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4605 {
4606         struct nfs4_stid *ret;
4607
4608         ret = find_stateid_by_type(cl, s,
4609                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4610         if (!ret)
4611                 return NULL;
4612         return delegstateid(ret);
4613 }
4614
4615 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4616 {
4617         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4618                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4619 }
4620
4621 static __be32
4622 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4623                 struct nfs4_delegation **dp)
4624 {
4625         int flags;
4626         __be32 status = nfserr_bad_stateid;
4627         struct nfs4_delegation *deleg;
4628
4629         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4630         if (deleg == NULL)
4631                 goto out;
4632         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4633                 nfs4_put_stid(&deleg->dl_stid);
4634                 if (cl->cl_minorversion)
4635                         status = nfserr_deleg_revoked;
4636                 goto out;
4637         }
4638         flags = share_access_to_flags(open->op_share_access);
4639         status = nfs4_check_delegmode(deleg, flags);
4640         if (status) {
4641                 nfs4_put_stid(&deleg->dl_stid);
4642                 goto out;
4643         }
4644         *dp = deleg;
4645 out:
4646         if (!nfsd4_is_deleg_cur(open))
4647                 return nfs_ok;
4648         if (status)
4649                 return status;
4650         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4651         return nfs_ok;
4652 }
4653
4654 static inline int nfs4_access_to_access(u32 nfs4_access)
4655 {
4656         int flags = 0;
4657
4658         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4659                 flags |= NFSD_MAY_READ;
4660         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4661                 flags |= NFSD_MAY_WRITE;
4662         return flags;
4663 }
4664
4665 static inline __be32
4666 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4667                 struct nfsd4_open *open)
4668 {
4669         struct iattr iattr = {
4670                 .ia_valid = ATTR_SIZE,
4671                 .ia_size = 0,
4672         };
4673         if (!open->op_truncate)
4674                 return 0;
4675         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4676                 return nfserr_inval;
4677         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4678 }
4679
4680 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4681                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4682                 struct nfsd4_open *open)
4683 {
4684         struct nfsd_file *nf = NULL;
4685         __be32 status;
4686         int oflag = nfs4_access_to_omode(open->op_share_access);
4687         int access = nfs4_access_to_access(open->op_share_access);
4688         unsigned char old_access_bmap, old_deny_bmap;
4689
4690         spin_lock(&fp->fi_lock);
4691
4692         /*
4693          * Are we trying to set a deny mode that would conflict with
4694          * current access?
4695          */
4696         status = nfs4_file_check_deny(fp, open->op_share_deny);
4697         if (status != nfs_ok) {
4698                 spin_unlock(&fp->fi_lock);
4699                 goto out;
4700         }
4701
4702         /* set access to the file */
4703         status = nfs4_file_get_access(fp, open->op_share_access);
4704         if (status != nfs_ok) {
4705                 spin_unlock(&fp->fi_lock);
4706                 goto out;
4707         }
4708
4709         /* Set access bits in stateid */
4710         old_access_bmap = stp->st_access_bmap;
4711         set_access(open->op_share_access, stp);
4712
4713         /* Set new deny mask */
4714         old_deny_bmap = stp->st_deny_bmap;
4715         set_deny(open->op_share_deny, stp);
4716         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4717
4718         if (!fp->fi_fds[oflag]) {
4719                 spin_unlock(&fp->fi_lock);
4720                 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4721                 if (status)
4722                         goto out_put_access;
4723                 spin_lock(&fp->fi_lock);
4724                 if (!fp->fi_fds[oflag]) {
4725                         fp->fi_fds[oflag] = nf;
4726                         nf = NULL;
4727                 }
4728         }
4729         spin_unlock(&fp->fi_lock);
4730         if (nf)
4731                 nfsd_file_put(nf);
4732
4733         status = nfsd4_truncate(rqstp, cur_fh, open);
4734         if (status)
4735                 goto out_put_access;
4736 out:
4737         return status;
4738 out_put_access:
4739         stp->st_access_bmap = old_access_bmap;
4740         nfs4_file_put_access(fp, open->op_share_access);
4741         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4742         goto out;
4743 }
4744
4745 static __be32
4746 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)
4747 {
4748         __be32 status;
4749         unsigned char old_deny_bmap = stp->st_deny_bmap;
4750
4751         if (!test_access(open->op_share_access, stp))
4752                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4753
4754         /* test and set deny mode */
4755         spin_lock(&fp->fi_lock);
4756         status = nfs4_file_check_deny(fp, open->op_share_deny);
4757         if (status == nfs_ok) {
4758                 set_deny(open->op_share_deny, stp);
4759                 fp->fi_share_deny |=
4760                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4761         }
4762         spin_unlock(&fp->fi_lock);
4763
4764         if (status != nfs_ok)
4765                 return status;
4766
4767         status = nfsd4_truncate(rqstp, cur_fh, open);
4768         if (status != nfs_ok)
4769                 reset_union_bmap_deny(old_deny_bmap, stp);
4770         return status;
4771 }
4772
4773 /* Should we give out recallable state?: */
4774 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4775 {
4776         if (clp->cl_cb_state == NFSD4_CB_UP)
4777                 return true;
4778         /*
4779          * In the sessions case, since we don't have to establish a
4780          * separate connection for callbacks, we assume it's OK
4781          * until we hear otherwise:
4782          */
4783         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4784 }
4785
4786 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4787                                                 int flag)
4788 {
4789         struct file_lock *fl;
4790
4791         fl = locks_alloc_lock();
4792         if (!fl)
4793                 return NULL;
4794         fl->fl_lmops = &nfsd_lease_mng_ops;
4795         fl->fl_flags = FL_DELEG;
4796         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4797         fl->fl_end = OFFSET_MAX;
4798         fl->fl_owner = (fl_owner_t)dp;
4799         fl->fl_pid = current->tgid;
4800         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4801         return fl;
4802 }
4803
4804 static struct nfs4_delegation *
4805 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4806                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4807 {
4808         int status = 0;
4809         struct nfs4_delegation *dp;
4810         struct nfsd_file *nf;
4811         struct file_lock *fl;
4812
4813         /*
4814          * The fi_had_conflict and nfs_get_existing_delegation checks
4815          * here are just optimizations; we'll need to recheck them at
4816          * the end:
4817          */
4818         if (fp->fi_had_conflict)
4819                 return ERR_PTR(-EAGAIN);
4820
4821         nf = find_readable_file(fp);
4822         if (!nf) {
4823                 /* We should always have a readable file here */
4824                 WARN_ON_ONCE(1);
4825                 return ERR_PTR(-EBADF);
4826         }
4827         spin_lock(&state_lock);
4828         spin_lock(&fp->fi_lock);
4829         if (nfs4_delegation_exists(clp, fp))
4830                 status = -EAGAIN;
4831         else if (!fp->fi_deleg_file) {
4832                 fp->fi_deleg_file = nf;
4833                 /* increment early to prevent fi_deleg_file from being
4834                  * cleared */
4835                 fp->fi_delegees = 1;
4836                 nf = NULL;
4837         } else
4838                 fp->fi_delegees++;
4839         spin_unlock(&fp->fi_lock);
4840         spin_unlock(&state_lock);
4841         if (nf)
4842                 nfsd_file_put(nf);
4843         if (status)
4844                 return ERR_PTR(status);
4845
4846         status = -ENOMEM;
4847         dp = alloc_init_deleg(clp, fp, fh, odstate);
4848         if (!dp)
4849                 goto out_delegees;
4850
4851         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4852         if (!fl)
4853                 goto out_clnt_odstate;
4854
4855         status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
4856         if (fl)
4857                 locks_free_lock(fl);
4858         if (status)
4859                 goto out_clnt_odstate;
4860
4861         spin_lock(&state_lock);
4862         spin_lock(&fp->fi_lock);
4863         if (fp->fi_had_conflict)
4864                 status = -EAGAIN;
4865         else
4866                 status = hash_delegation_locked(dp, fp);
4867         spin_unlock(&fp->fi_lock);
4868         spin_unlock(&state_lock);
4869
4870         if (status)
4871                 goto out_unlock;
4872
4873         return dp;
4874 out_unlock:
4875         vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
4876 out_clnt_odstate:
4877         put_clnt_odstate(dp->dl_clnt_odstate);
4878         nfs4_put_stid(&dp->dl_stid);
4879 out_delegees:
4880         put_deleg_file(fp);
4881         return ERR_PTR(status);
4882 }
4883
4884 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4885 {
4886         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4887         if (status == -EAGAIN)
4888                 open->op_why_no_deleg = WND4_CONTENTION;
4889         else {
4890                 open->op_why_no_deleg = WND4_RESOURCE;
4891                 switch (open->op_deleg_want) {
4892                 case NFS4_SHARE_WANT_READ_DELEG:
4893                 case NFS4_SHARE_WANT_WRITE_DELEG:
4894                 case NFS4_SHARE_WANT_ANY_DELEG:
4895                         break;
4896                 case NFS4_SHARE_WANT_CANCEL:
4897                         open->op_why_no_deleg = WND4_CANCELLED;
4898                         break;
4899                 case NFS4_SHARE_WANT_NO_DELEG:
4900                         WARN_ON_ONCE(1);
4901                 }
4902         }
4903 }
4904
4905 /*
4906  * Attempt to hand out a delegation.
4907  *
4908  * Note we don't support write delegations, and won't until the vfs has
4909  * proper support for them.
4910  */
4911 static void
4912 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4913                         struct nfs4_ol_stateid *stp)
4914 {
4915         struct nfs4_delegation *dp;
4916         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4917         struct nfs4_client *clp = stp->st_stid.sc_client;
4918         int cb_up;
4919         int status = 0;
4920
4921         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4922         open->op_recall = 0;
4923         switch (open->op_claim_type) {
4924                 case NFS4_OPEN_CLAIM_PREVIOUS:
4925                         if (!cb_up)
4926                                 open->op_recall = 1;
4927                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4928                                 goto out_no_deleg;
4929                         break;
4930                 case NFS4_OPEN_CLAIM_NULL:
4931                 case NFS4_OPEN_CLAIM_FH:
4932                         /*
4933                          * Let's not give out any delegations till everyone's
4934                          * had the chance to reclaim theirs, *and* until
4935                          * NLM locks have all been reclaimed:
4936                          */
4937                         if (locks_in_grace(clp->net))
4938                                 goto out_no_deleg;
4939                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4940                                 goto out_no_deleg;
4941                         /*
4942                          * Also, if the file was opened for write or
4943                          * create, there's a good chance the client's
4944                          * about to write to it, resulting in an
4945                          * immediate recall (since we don't support
4946                          * write delegations):
4947                          */
4948                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4949                                 goto out_no_deleg;
4950                         if (open->op_create == NFS4_OPEN_CREATE)
4951                                 goto out_no_deleg;
4952                         break;
4953                 default:
4954                         goto out_no_deleg;
4955         }
4956         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4957         if (IS_ERR(dp))
4958                 goto out_no_deleg;
4959
4960         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4961
4962         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4963                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4964         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4965         nfs4_put_stid(&dp->dl_stid);
4966         return;
4967 out_no_deleg:
4968         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4969         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4970             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4971                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4972                 open->op_recall = 1;
4973         }
4974
4975         /* 4.1 client asking for a delegation? */
4976         if (open->op_deleg_want)
4977                 nfsd4_open_deleg_none_ext(open, status);
4978         return;
4979 }
4980
4981 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4982                                         struct nfs4_delegation *dp)
4983 {
4984         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4985             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4986                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4987                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4988         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4989                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4990                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4991                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4992         }
4993         /* Otherwise the client must be confused wanting a delegation
4994          * it already has, therefore we don't return
4995          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4996          */
4997 }
4998
4999 __be32
5000 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5001 {
5002         struct nfsd4_compoundres *resp = rqstp->rq_resp;
5003         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5004         struct nfs4_file *fp = NULL;
5005         struct nfs4_ol_stateid *stp = NULL;
5006         struct nfs4_delegation *dp = NULL;
5007         __be32 status;
5008         bool new_stp = false;
5009
5010         /*
5011          * Lookup file; if found, lookup stateid and check open request,
5012          * and check for delegations in the process of being recalled.
5013          * If not found, create the nfs4_file struct
5014          */
5015         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
5016         if (fp != open->op_file) {
5017                 status = nfs4_check_deleg(cl, open, &dp);
5018                 if (status)
5019                         goto out;
5020                 stp = nfsd4_find_and_lock_existing_open(fp, open);
5021         } else {
5022                 open->op_file = NULL;
5023                 status = nfserr_bad_stateid;
5024                 if (nfsd4_is_deleg_cur(open))
5025                         goto out;
5026         }
5027
5028         if (!stp) {
5029                 stp = init_open_stateid(fp, open);
5030                 if (!open->op_stp)
5031                         new_stp = true;
5032         }
5033
5034         /*
5035          * OPEN the file, or upgrade an existing OPEN.
5036          * If truncate fails, the OPEN fails.
5037          *
5038          * stp is already locked.
5039          */
5040         if (!new_stp) {
5041                 /* Stateid was found, this is an OPEN upgrade */
5042                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5043                 if (status) {
5044                         mutex_unlock(&stp->st_mutex);
5045                         goto out;
5046                 }
5047         } else {
5048                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5049                 if (status) {
5050                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5051                         release_open_stateid(stp);
5052                         mutex_unlock(&stp->st_mutex);
5053                         goto out;
5054                 }
5055
5056                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5057                                                         open->op_odstate);
5058                 if (stp->st_clnt_odstate == open->op_odstate)
5059                         open->op_odstate = NULL;
5060         }
5061
5062         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5063         mutex_unlock(&stp->st_mutex);
5064
5065         if (nfsd4_has_session(&resp->cstate)) {
5066                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5067                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5068                         open->op_why_no_deleg = WND4_NOT_WANTED;
5069                         goto nodeleg;
5070                 }
5071         }
5072
5073         /*
5074         * Attempt to hand out a delegation. No error return, because the
5075         * OPEN succeeds even if we fail.
5076         */
5077         nfs4_open_delegation(current_fh, open, stp);
5078 nodeleg:
5079         status = nfs_ok;
5080
5081         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5082                 STATEID_VAL(&stp->st_stid.sc_stateid));
5083 out:
5084         /* 4.1 client trying to upgrade/downgrade delegation? */
5085         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5086             open->op_deleg_want)
5087                 nfsd4_deleg_xgrade_none_ext(open, dp);
5088
5089         if (fp)
5090                 put_nfs4_file(fp);
5091         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5092                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5093         /*
5094         * To finish the open response, we just need to set the rflags.
5095         */
5096         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5097         if (nfsd4_has_session(&resp->cstate))
5098                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5099         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5100                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5101
5102         if (dp)
5103                 nfs4_put_stid(&dp->dl_stid);
5104         if (stp)
5105                 nfs4_put_stid(&stp->st_stid);
5106
5107         return status;
5108 }
5109
5110 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5111                               struct nfsd4_open *open)
5112 {
5113         if (open->op_openowner) {
5114                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5115
5116                 nfsd4_cstate_assign_replay(cstate, so);
5117                 nfs4_put_stateowner(so);
5118         }
5119         if (open->op_file)
5120                 kmem_cache_free(file_slab, open->op_file);
5121         if (open->op_stp)
5122                 nfs4_put_stid(&open->op_stp->st_stid);
5123         if (open->op_odstate)
5124                 kmem_cache_free(odstate_slab, open->op_odstate);
5125 }
5126
5127 __be32
5128 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5129             union nfsd4_op_u *u)
5130 {
5131         clientid_t *clid = &u->renew;
5132         struct nfs4_client *clp;
5133         __be32 status;
5134         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5135
5136         dprintk("process_renew(%08x/%08x): starting\n",
5137                         clid->cl_boot, clid->cl_id);
5138         status = lookup_clientid(clid, cstate, nn);
5139         if (status)
5140                 goto out;
5141         clp = cstate->clp;
5142         status = nfserr_cb_path_down;
5143         if (!list_empty(&clp->cl_delegations)
5144                         && clp->cl_cb_state != NFSD4_CB_UP)
5145                 goto out;
5146         status = nfs_ok;
5147 out:
5148         return status;
5149 }
5150
5151 void
5152 nfsd4_end_grace(struct nfsd_net *nn)
5153 {
5154         /* do nothing if grace period already ended */
5155         if (nn->grace_ended)
5156                 return;
5157
5158         nn->grace_ended = true;
5159         /*
5160          * If the server goes down again right now, an NFSv4
5161          * client will still be allowed to reclaim after it comes back up,
5162          * even if it hasn't yet had a chance to reclaim state this time.
5163          *
5164          */
5165         nfsd4_record_grace_done(nn);
5166         /*
5167          * At this point, NFSv4 clients can still reclaim.  But if the
5168          * server crashes, any that have not yet reclaimed will be out
5169          * of luck on the next boot.
5170          *
5171          * (NFSv4.1+ clients are considered to have reclaimed once they
5172          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
5173          * have reclaimed after their first OPEN.)
5174          */
5175         locks_end_grace(&nn->nfsd4_manager);
5176         /*
5177          * At this point, and once lockd and/or any other containers
5178          * exit their grace period, further reclaims will fail and
5179          * regular locking can resume.
5180          */
5181 }
5182
5183 /*
5184  * If we've waited a lease period but there are still clients trying to
5185  * reclaim, wait a little longer to give them a chance to finish.
5186  */
5187 static bool clients_still_reclaiming(struct nfsd_net *nn)
5188 {
5189         unsigned long now = get_seconds();
5190         unsigned long double_grace_period_end = nn->boot_time +
5191                                                 2 * nn->nfsd4_lease;
5192
5193         if (nn->track_reclaim_completes &&
5194                         atomic_read(&nn->nr_reclaim_complete) ==
5195                         nn->reclaim_str_hashtbl_size)
5196                 return false;
5197         if (!nn->somebody_reclaimed)
5198                 return false;
5199         nn->somebody_reclaimed = false;
5200         /*
5201          * If we've given them *two* lease times to reclaim, and they're
5202          * still not done, give up:
5203          */
5204         if (time_after(now, double_grace_period_end))
5205                 return false;
5206         return true;
5207 }
5208
5209 static time_t
5210 nfs4_laundromat(struct nfsd_net *nn)
5211 {
5212         struct nfs4_client *clp;
5213         struct nfs4_openowner *oo;
5214         struct nfs4_delegation *dp;
5215         struct nfs4_ol_stateid *stp;
5216         struct nfsd4_blocked_lock *nbl;
5217         struct list_head *pos, *next, reaplist;
5218         time_t cutoff = get_seconds() - nn->nfsd4_lease;
5219         time_t t, new_timeo = nn->nfsd4_lease;
5220
5221         dprintk("NFSD: laundromat service - starting\n");
5222
5223         if (clients_still_reclaiming(nn)) {
5224                 new_timeo = 0;
5225                 goto out;
5226         }
5227         dprintk("NFSD: end of grace period\n");
5228         nfsd4_end_grace(nn);
5229         INIT_LIST_HEAD(&reaplist);
5230         spin_lock(&nn->client_lock);
5231         list_for_each_safe(pos, next, &nn->client_lru) {
5232                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5233                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
5234                         t = clp->cl_time - cutoff;
5235                         new_timeo = min(new_timeo, t);
5236                         break;
5237                 }
5238                 if (mark_client_expired_locked(clp)) {
5239                         dprintk("NFSD: client in use (clientid %08x)\n",
5240                                 clp->cl_clientid.cl_id);
5241                         continue;
5242                 }
5243                 list_add(&clp->cl_lru, &reaplist);
5244         }
5245         spin_unlock(&nn->client_lock);
5246         list_for_each_safe(pos, next, &reaplist) {
5247                 clp = list_entry(pos, struct nfs4_client, cl_lru);
5248                 dprintk("NFSD: purging unused client (clientid %08x)\n",
5249                         clp->cl_clientid.cl_id);
5250                 list_del_init(&clp->cl_lru);
5251                 expire_client(clp);
5252         }
5253         spin_lock(&state_lock);
5254         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5255                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5256                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
5257                         t = dp->dl_time - cutoff;
5258                         new_timeo = min(new_timeo, t);
5259                         break;
5260                 }
5261                 WARN_ON(!unhash_delegation_locked(dp));
5262                 list_add(&dp->dl_recall_lru, &reaplist);
5263         }
5264         spin_unlock(&state_lock);
5265         while (!list_empty(&reaplist)) {
5266                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5267                                         dl_recall_lru);
5268                 list_del_init(&dp->dl_recall_lru);
5269                 revoke_delegation(dp);
5270         }
5271
5272         spin_lock(&nn->client_lock);
5273         while (!list_empty(&nn->close_lru)) {
5274                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5275                                         oo_close_lru);
5276                 if (time_after((unsigned long)oo->oo_time,
5277                                (unsigned long)cutoff)) {
5278                         t = oo->oo_time - cutoff;
5279                         new_timeo = min(new_timeo, t);
5280                         break;
5281                 }
5282                 list_del_init(&oo->oo_close_lru);
5283                 stp = oo->oo_last_closed_stid;
5284                 oo->oo_last_closed_stid = NULL;
5285                 spin_unlock(&nn->client_lock);
5286                 nfs4_put_stid(&stp->st_stid);
5287                 spin_lock(&nn->client_lock);
5288         }
5289         spin_unlock(&nn->client_lock);
5290
5291         /*
5292          * It's possible for a client to try and acquire an already held lock
5293          * that is being held for a long time, and then lose interest in it.
5294          * So, we clean out any un-revisited request after a lease period
5295          * under the assumption that the client is no longer interested.
5296          *
5297          * RFC5661, sec. 9.6 states that the client must not rely on getting
5298          * notifications and must continue to poll for locks, even when the
5299          * server supports them. Thus this shouldn't lead to clients blocking
5300          * indefinitely once the lock does become free.
5301          */
5302         BUG_ON(!list_empty(&reaplist));
5303         spin_lock(&nn->blocked_locks_lock);
5304         while (!list_empty(&nn->blocked_locks_lru)) {
5305                 nbl = list_first_entry(&nn->blocked_locks_lru,
5306                                         struct nfsd4_blocked_lock, nbl_lru);
5307                 if (time_after((unsigned long)nbl->nbl_time,
5308                                (unsigned long)cutoff)) {
5309                         t = nbl->nbl_time - cutoff;
5310                         new_timeo = min(new_timeo, t);
5311                         break;
5312                 }
5313                 list_move(&nbl->nbl_lru, &reaplist);
5314                 list_del_init(&nbl->nbl_list);
5315         }
5316         spin_unlock(&nn->blocked_locks_lock);
5317
5318         while (!list_empty(&reaplist)) {
5319                 nbl = list_first_entry(&reaplist,
5320                                         struct nfsd4_blocked_lock, nbl_lru);
5321                 list_del_init(&nbl->nbl_lru);
5322                 free_blocked_lock(nbl);
5323         }
5324 out:
5325         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5326         return new_timeo;
5327 }
5328
5329 static struct workqueue_struct *laundry_wq;
5330 static void laundromat_main(struct work_struct *);
5331
5332 static void
5333 laundromat_main(struct work_struct *laundry)
5334 {
5335         time_t t;
5336         struct delayed_work *dwork = to_delayed_work(laundry);
5337         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5338                                            laundromat_work);
5339
5340         t = nfs4_laundromat(nn);
5341         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
5342         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5343 }
5344
5345 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5346 {
5347         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5348                 return nfserr_bad_stateid;
5349         return nfs_ok;
5350 }
5351
5352 static inline int
5353 access_permit_read(struct nfs4_ol_stateid *stp)
5354 {
5355         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5356                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5357                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5358 }
5359
5360 static inline int
5361 access_permit_write(struct nfs4_ol_stateid *stp)
5362 {
5363         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5364                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5365 }
5366
5367 static
5368 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5369 {
5370         __be32 status = nfserr_openmode;
5371
5372         /* For lock stateid's, we test the parent open, not the lock: */
5373         if (stp->st_openstp)
5374                 stp = stp->st_openstp;
5375         if ((flags & WR_STATE) && !access_permit_write(stp))
5376                 goto out;
5377         if ((flags & RD_STATE) && !access_permit_read(stp))
5378                 goto out;
5379         status = nfs_ok;
5380 out:
5381         return status;
5382 }
5383
5384 static inline __be32
5385 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5386 {
5387         if (ONE_STATEID(stateid) && (flags & RD_STATE))
5388                 return nfs_ok;
5389         else if (opens_in_grace(net)) {
5390                 /* Answer in remaining cases depends on existence of
5391                  * conflicting state; so we must wait out the grace period. */
5392                 return nfserr_grace;
5393         } else if (flags & WR_STATE)
5394                 return nfs4_share_conflict(current_fh,
5395                                 NFS4_SHARE_DENY_WRITE);
5396         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5397                 return nfs4_share_conflict(current_fh,
5398                                 NFS4_SHARE_DENY_READ);
5399 }
5400
5401 /*
5402  * Allow READ/WRITE during grace period on recovered state only for files
5403  * that are not able to provide mandatory locking.
5404  */
5405 static inline int
5406 grace_disallows_io(struct net *net, struct inode *inode)
5407 {
5408         return opens_in_grace(net) && mandatory_lock(inode);
5409 }
5410
5411 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5412 {
5413         /*
5414          * When sessions are used the stateid generation number is ignored
5415          * when it is zero.
5416          */
5417         if (has_session && in->si_generation == 0)
5418                 return nfs_ok;
5419
5420         if (in->si_generation == ref->si_generation)
5421                 return nfs_ok;
5422
5423         /* If the client sends us a stateid from the future, it's buggy: */
5424         if (nfsd4_stateid_generation_after(in, ref))
5425                 return nfserr_bad_stateid;
5426         /*
5427          * However, we could see a stateid from the past, even from a
5428          * non-buggy client.  For example, if the client sends a lock
5429          * while some IO is outstanding, the lock may bump si_generation
5430          * while the IO is still in flight.  The client could avoid that
5431          * situation by waiting for responses on all the IO requests,
5432          * but better performance may result in retrying IO that
5433          * receives an old_stateid error if requests are rarely
5434          * reordered in flight:
5435          */
5436         return nfserr_old_stateid;
5437 }
5438
5439 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5440 {
5441         __be32 ret;
5442
5443         spin_lock(&s->sc_lock);
5444         ret = nfsd4_verify_open_stid(s);
5445         if (ret == nfs_ok)
5446                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5447         spin_unlock(&s->sc_lock);
5448         return ret;
5449 }
5450
5451 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5452 {
5453         if (ols->st_stateowner->so_is_open_owner &&
5454             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5455                 return nfserr_bad_stateid;
5456         return nfs_ok;
5457 }
5458
5459 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5460 {
5461         struct nfs4_stid *s;
5462         __be32 status = nfserr_bad_stateid;
5463
5464         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5465                 CLOSE_STATEID(stateid))
5466                 return status;
5467         /* Client debugging aid. */
5468         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5469                 char addr_str[INET6_ADDRSTRLEN];
5470                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5471                                  sizeof(addr_str));
5472                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5473                                         "with incorrect client ID\n", addr_str);
5474                 return status;
5475         }
5476         spin_lock(&cl->cl_lock);
5477         s = find_stateid_locked(cl, stateid);
5478         if (!s)
5479                 goto out_unlock;
5480         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5481         if (status)
5482                 goto out_unlock;
5483         switch (s->sc_type) {
5484         case NFS4_DELEG_STID:
5485                 status = nfs_ok;
5486                 break;
5487         case NFS4_REVOKED_DELEG_STID:
5488                 status = nfserr_deleg_revoked;
5489                 break;
5490         case NFS4_OPEN_STID:
5491         case NFS4_LOCK_STID:
5492                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5493                 break;
5494         default:
5495                 printk("unknown stateid type %x\n", s->sc_type);
5496                 /* Fallthrough */
5497         case NFS4_CLOSED_STID:
5498         case NFS4_CLOSED_DELEG_STID:
5499                 status = nfserr_bad_stateid;
5500         }
5501 out_unlock:
5502         spin_unlock(&cl->cl_lock);
5503         return status;
5504 }
5505
5506 __be32
5507 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5508                      stateid_t *stateid, unsigned char typemask,
5509                      struct nfs4_stid **s, struct nfsd_net *nn)
5510 {
5511         __be32 status;
5512         bool return_revoked = false;
5513
5514         /*
5515          *  only return revoked delegations if explicitly asked.
5516          *  otherwise we report revoked or bad_stateid status.
5517          */
5518         if (typemask & NFS4_REVOKED_DELEG_STID)
5519                 return_revoked = true;
5520         else if (typemask & NFS4_DELEG_STID)
5521                 typemask |= NFS4_REVOKED_DELEG_STID;
5522
5523         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5524                 CLOSE_STATEID(stateid))
5525                 return nfserr_bad_stateid;
5526         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5527         if (status == nfserr_stale_clientid) {
5528                 if (cstate->session)
5529                         return nfserr_bad_stateid;
5530                 return nfserr_stale_stateid;
5531         }
5532         if (status)
5533                 return status;
5534         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5535         if (!*s)
5536                 return nfserr_bad_stateid;
5537         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5538                 nfs4_put_stid(*s);
5539                 if (cstate->minorversion)
5540                         return nfserr_deleg_revoked;
5541                 return nfserr_bad_stateid;
5542         }
5543         return nfs_ok;
5544 }
5545
5546 static struct nfsd_file *
5547 nfs4_find_file(struct nfs4_stid *s, int flags)
5548 {
5549         if (!s)
5550                 return NULL;
5551
5552         switch (s->sc_type) {
5553         case NFS4_DELEG_STID:
5554                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5555                         return NULL;
5556                 return nfsd_file_get(s->sc_file->fi_deleg_file);
5557         case NFS4_OPEN_STID:
5558         case NFS4_LOCK_STID:
5559                 if (flags & RD_STATE)
5560                         return find_readable_file(s->sc_file);
5561                 else
5562                         return find_writeable_file(s->sc_file);
5563                 break;
5564         }
5565
5566         return NULL;
5567 }
5568
5569 static __be32
5570 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5571 {
5572         __be32 status;
5573
5574         status = nfsd4_check_openowner_confirmed(ols);
5575         if (status)
5576                 return status;
5577         return nfs4_check_openmode(ols, flags);
5578 }
5579
5580 static __be32
5581 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5582                 struct nfsd_file **nfp, int flags)
5583 {
5584         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5585         struct nfsd_file *nf;
5586         __be32 status;
5587
5588         nf = nfs4_find_file(s, flags);
5589         if (nf) {
5590                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5591                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5592                 if (status) {
5593                         nfsd_file_put(nf);
5594                         goto out;
5595                 }
5596         } else {
5597                 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5598                 if (status)
5599                         return status;
5600         }
5601         *nfp = nf;
5602 out:
5603         return status;
5604 }
5605
5606 /*
5607  * Checks for stateid operations
5608  */
5609 __be32
5610 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5611                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5612                 stateid_t *stateid, int flags, struct nfsd_file **nfp)
5613 {
5614         struct inode *ino = d_inode(fhp->fh_dentry);
5615         struct net *net = SVC_NET(rqstp);
5616         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5617         struct nfs4_stid *s = NULL;
5618         __be32 status;
5619
5620         if (nfp)
5621                 *nfp = NULL;
5622
5623         if (grace_disallows_io(net, ino))
5624                 return nfserr_grace;
5625
5626         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5627                 status = check_special_stateids(net, fhp, stateid, flags);
5628                 goto done;
5629         }
5630
5631         status = nfsd4_lookup_stateid(cstate, stateid,
5632                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5633                                 &s, nn);
5634         if (status)
5635                 return status;
5636         status = nfsd4_stid_check_stateid_generation(stateid, s,
5637                         nfsd4_has_session(cstate));
5638         if (status)
5639                 goto out;
5640
5641         switch (s->sc_type) {
5642         case NFS4_DELEG_STID:
5643                 status = nfs4_check_delegmode(delegstateid(s), flags);
5644                 break;
5645         case NFS4_OPEN_STID:
5646         case NFS4_LOCK_STID:
5647                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5648                 break;
5649         default:
5650                 status = nfserr_bad_stateid;
5651                 break;
5652         }
5653         if (status)
5654                 goto out;
5655         status = nfs4_check_fh(fhp, s);
5656
5657 done:
5658         if (status == nfs_ok && nfp)
5659                 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5660 out:
5661         if (s)
5662                 nfs4_put_stid(s);
5663         return status;
5664 }
5665
5666 /*
5667  * Test if the stateid is valid
5668  */
5669 __be32
5670 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5671                    union nfsd4_op_u *u)
5672 {
5673         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5674         struct nfsd4_test_stateid_id *stateid;
5675         struct nfs4_client *cl = cstate->session->se_client;
5676
5677         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5678                 stateid->ts_id_status =
5679                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5680
5681         return nfs_ok;
5682 }
5683
5684 static __be32
5685 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5686 {
5687         struct nfs4_ol_stateid *stp = openlockstateid(s);
5688         __be32 ret;
5689
5690         ret = nfsd4_lock_ol_stateid(stp);
5691         if (ret)
5692                 goto out_put_stid;
5693
5694         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5695         if (ret)
5696                 goto out;
5697
5698         ret = nfserr_locks_held;
5699         if (check_for_locks(stp->st_stid.sc_file,
5700                             lockowner(stp->st_stateowner)))
5701                 goto out;
5702
5703         release_lock_stateid(stp);
5704         ret = nfs_ok;
5705
5706 out:
5707         mutex_unlock(&stp->st_mutex);
5708 out_put_stid:
5709         nfs4_put_stid(s);
5710         return ret;
5711 }
5712
5713 __be32
5714 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5715                    union nfsd4_op_u *u)
5716 {
5717         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5718         stateid_t *stateid = &free_stateid->fr_stateid;
5719         struct nfs4_stid *s;
5720         struct nfs4_delegation *dp;
5721         struct nfs4_client *cl = cstate->session->se_client;
5722         __be32 ret = nfserr_bad_stateid;
5723
5724         spin_lock(&cl->cl_lock);
5725         s = find_stateid_locked(cl, stateid);
5726         if (!s)
5727                 goto out_unlock;
5728         spin_lock(&s->sc_lock);
5729         switch (s->sc_type) {
5730         case NFS4_DELEG_STID:
5731                 ret = nfserr_locks_held;
5732                 break;
5733         case NFS4_OPEN_STID:
5734                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5735                 if (ret)
5736                         break;
5737                 ret = nfserr_locks_held;
5738                 break;
5739         case NFS4_LOCK_STID:
5740                 spin_unlock(&s->sc_lock);
5741                 refcount_inc(&s->sc_count);
5742                 spin_unlock(&cl->cl_lock);
5743                 ret = nfsd4_free_lock_stateid(stateid, s);
5744                 goto out;
5745         case NFS4_REVOKED_DELEG_STID:
5746                 spin_unlock(&s->sc_lock);
5747                 dp = delegstateid(s);
5748                 list_del_init(&dp->dl_recall_lru);
5749                 spin_unlock(&cl->cl_lock);
5750                 nfs4_put_stid(s);
5751                 ret = nfs_ok;
5752                 goto out;
5753         /* Default falls through and returns nfserr_bad_stateid */
5754         }
5755         spin_unlock(&s->sc_lock);
5756 out_unlock:
5757         spin_unlock(&cl->cl_lock);
5758 out:
5759         return ret;
5760 }
5761
5762 static inline int
5763 setlkflg (int type)
5764 {
5765         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5766                 RD_STATE : WR_STATE;
5767 }
5768
5769 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5770 {
5771         struct svc_fh *current_fh = &cstate->current_fh;
5772         struct nfs4_stateowner *sop = stp->st_stateowner;
5773         __be32 status;
5774
5775         status = nfsd4_check_seqid(cstate, sop, seqid);
5776         if (status)
5777                 return status;
5778         status = nfsd4_lock_ol_stateid(stp);
5779         if (status != nfs_ok)
5780                 return status;
5781         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5782         if (status == nfs_ok)
5783                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5784         if (status != nfs_ok)
5785                 mutex_unlock(&stp->st_mutex);
5786         return status;
5787 }
5788
5789 /*
5790  * Checks for sequence id mutating operations.
5791  */
5792 static __be32
5793 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5794                          stateid_t *stateid, char typemask,
5795                          struct nfs4_ol_stateid **stpp,
5796                          struct nfsd_net *nn)
5797 {
5798         __be32 status;
5799         struct nfs4_stid *s;
5800         struct nfs4_ol_stateid *stp = NULL;
5801
5802         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5803                 seqid, STATEID_VAL(stateid));
5804
5805         *stpp = NULL;
5806         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5807         if (status)
5808                 return status;
5809         stp = openlockstateid(s);
5810         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5811
5812         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5813         if (!status)
5814                 *stpp = stp;
5815         else
5816                 nfs4_put_stid(&stp->st_stid);
5817         return status;
5818 }
5819
5820 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5821                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5822 {
5823         __be32 status;
5824         struct nfs4_openowner *oo;
5825         struct nfs4_ol_stateid *stp;
5826
5827         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5828                                                 NFS4_OPEN_STID, &stp, nn);
5829         if (status)
5830                 return status;
5831         oo = openowner(stp->st_stateowner);
5832         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5833                 mutex_unlock(&stp->st_mutex);
5834                 nfs4_put_stid(&stp->st_stid);
5835                 return nfserr_bad_stateid;
5836         }
5837         *stpp = stp;
5838         return nfs_ok;
5839 }
5840
5841 __be32
5842 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5843                    union nfsd4_op_u *u)
5844 {
5845         struct nfsd4_open_confirm *oc = &u->open_confirm;
5846         __be32 status;
5847         struct nfs4_openowner *oo;
5848         struct nfs4_ol_stateid *stp;
5849         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5850
5851         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5852                         cstate->current_fh.fh_dentry);
5853
5854         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5855         if (status)
5856                 return status;
5857
5858         status = nfs4_preprocess_seqid_op(cstate,
5859                                         oc->oc_seqid, &oc->oc_req_stateid,
5860                                         NFS4_OPEN_STID, &stp, nn);
5861         if (status)
5862                 goto out;
5863         oo = openowner(stp->st_stateowner);
5864         status = nfserr_bad_stateid;
5865         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5866                 mutex_unlock(&stp->st_mutex);
5867                 goto put_stateid;
5868         }
5869         oo->oo_flags |= NFS4_OO_CONFIRMED;
5870         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5871         mutex_unlock(&stp->st_mutex);
5872         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5873                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5874
5875         nfsd4_client_record_create(oo->oo_owner.so_client);
5876         status = nfs_ok;
5877 put_stateid:
5878         nfs4_put_stid(&stp->st_stid);
5879 out:
5880         nfsd4_bump_seqid(cstate, status);
5881         return status;
5882 }
5883
5884 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5885 {
5886         if (!test_access(access, stp))
5887                 return;
5888         nfs4_file_put_access(stp->st_stid.sc_file, access);
5889         clear_access(access, stp);
5890 }
5891
5892 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5893 {
5894         switch (to_access) {
5895         case NFS4_SHARE_ACCESS_READ:
5896                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5897                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5898                 break;
5899         case NFS4_SHARE_ACCESS_WRITE:
5900                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5901                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5902                 break;
5903         case NFS4_SHARE_ACCESS_BOTH:
5904                 break;
5905         default:
5906                 WARN_ON_ONCE(1);
5907         }
5908 }
5909
5910 __be32
5911 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5912                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5913 {
5914         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5915         __be32 status;
5916         struct nfs4_ol_stateid *stp;
5917         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5918
5919         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5920                         cstate->current_fh.fh_dentry);
5921
5922         /* We don't yet support WANT bits: */
5923         if (od->od_deleg_want)
5924                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5925                         od->od_deleg_want);
5926
5927         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5928                                         &od->od_stateid, &stp, nn);
5929         if (status)
5930                 goto out;
5931         status = nfserr_inval;
5932         if (!test_access(od->od_share_access, stp)) {
5933                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5934                         stp->st_access_bmap, od->od_share_access);
5935                 goto put_stateid;
5936         }
5937         if (!test_deny(od->od_share_deny, stp)) {
5938                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5939                         stp->st_deny_bmap, od->od_share_deny);
5940                 goto put_stateid;
5941         }
5942         nfs4_stateid_downgrade(stp, od->od_share_access);
5943         reset_union_bmap_deny(od->od_share_deny, stp);
5944         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5945         status = nfs_ok;
5946 put_stateid:
5947         mutex_unlock(&stp->st_mutex);
5948         nfs4_put_stid(&stp->st_stid);
5949 out:
5950         nfsd4_bump_seqid(cstate, status);
5951         return status;
5952 }
5953
5954 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5955 {
5956         struct nfs4_client *clp = s->st_stid.sc_client;
5957         bool unhashed;
5958         LIST_HEAD(reaplist);
5959
5960         spin_lock(&clp->cl_lock);
5961         unhashed = unhash_open_stateid(s, &reaplist);
5962
5963         if (clp->cl_minorversion) {
5964                 if (unhashed)
5965                         put_ol_stateid_locked(s, &reaplist);
5966                 spin_unlock(&clp->cl_lock);
5967                 free_ol_stateid_reaplist(&reaplist);
5968         } else {
5969                 spin_unlock(&clp->cl_lock);
5970                 free_ol_stateid_reaplist(&reaplist);
5971                 if (unhashed)
5972                         move_to_close_lru(s, clp->net);
5973         }
5974 }
5975
5976 /*
5977  * nfs4_unlock_state() called after encode
5978  */
5979 __be32
5980 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5981                 union nfsd4_op_u *u)
5982 {
5983         struct nfsd4_close *close = &u->close;
5984         __be32 status;
5985         struct nfs4_ol_stateid *stp;
5986         struct net *net = SVC_NET(rqstp);
5987         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5988
5989         dprintk("NFSD: nfsd4_close on file %pd\n",
5990                         cstate->current_fh.fh_dentry);
5991
5992         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5993                                         &close->cl_stateid,
5994                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5995                                         &stp, nn);
5996         nfsd4_bump_seqid(cstate, status);
5997         if (status)
5998                 goto out;
5999
6000         stp->st_stid.sc_type = NFS4_CLOSED_STID;
6001
6002         /*
6003          * Technically we don't _really_ have to increment or copy it, since
6004          * it should just be gone after this operation and we clobber the
6005          * copied value below, but we continue to do so here just to ensure
6006          * that racing ops see that there was a state change.
6007          */
6008         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6009
6010         nfsd4_close_open_stateid(stp);
6011         mutex_unlock(&stp->st_mutex);
6012
6013         /* v4.1+ suggests that we send a special stateid in here, since the
6014          * clients should just ignore this anyway. Since this is not useful
6015          * for v4.0 clients either, we set it to the special close_stateid
6016          * universally.
6017          *
6018          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6019          */
6020         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6021
6022         /* put reference from nfs4_preprocess_seqid_op */
6023         nfs4_put_stid(&stp->st_stid);
6024 out:
6025         return status;
6026 }
6027
6028 __be32
6029 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6030                   union nfsd4_op_u *u)
6031 {
6032         struct nfsd4_delegreturn *dr = &u->delegreturn;
6033         struct nfs4_delegation *dp;
6034         stateid_t *stateid = &dr->dr_stateid;
6035         struct nfs4_stid *s;
6036         __be32 status;
6037         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6038
6039         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6040                 return status;
6041
6042         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6043         if (status)
6044                 goto out;
6045         dp = delegstateid(s);
6046         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6047         if (status)
6048                 goto put_stateid;
6049
6050         destroy_delegation(dp);
6051 put_stateid:
6052         nfs4_put_stid(&dp->dl_stid);
6053 out:
6054         return status;
6055 }
6056
6057 static inline u64
6058 end_offset(u64 start, u64 len)
6059 {
6060         u64 end;
6061
6062         end = start + len;
6063         return end >= start ? end: NFS4_MAX_UINT64;
6064 }
6065
6066 /* last octet in a range */
6067 static inline u64
6068 last_byte_offset(u64 start, u64 len)
6069 {
6070         u64 end;
6071
6072         WARN_ON_ONCE(!len);
6073         end = start + len;
6074         return end > start ? end - 1: NFS4_MAX_UINT64;
6075 }
6076
6077 /*
6078  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6079  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6080  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
6081  * locking, this prevents us from being completely protocol-compliant.  The
6082  * real solution to this problem is to start using unsigned file offsets in
6083  * the VFS, but this is a very deep change!
6084  */
6085 static inline void
6086 nfs4_transform_lock_offset(struct file_lock *lock)
6087 {
6088         if (lock->fl_start < 0)
6089                 lock->fl_start = OFFSET_MAX;
6090         if (lock->fl_end < 0)
6091                 lock->fl_end = OFFSET_MAX;
6092 }
6093
6094 static fl_owner_t
6095 nfsd4_fl_get_owner(fl_owner_t owner)
6096 {
6097         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6098
6099         nfs4_get_stateowner(&lo->lo_owner);
6100         return owner;
6101 }
6102
6103 static void
6104 nfsd4_fl_put_owner(fl_owner_t owner)
6105 {
6106         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6107
6108         if (lo)
6109                 nfs4_put_stateowner(&lo->lo_owner);
6110 }
6111
6112 static void
6113 nfsd4_lm_notify(struct file_lock *fl)
6114 {
6115         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
6116         struct net                      *net = lo->lo_owner.so_client->net;
6117         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
6118         struct nfsd4_blocked_lock       *nbl = container_of(fl,
6119                                                 struct nfsd4_blocked_lock, nbl_lock);
6120         bool queue = false;
6121
6122         /* An empty list means that something else is going to be using it */
6123         spin_lock(&nn->blocked_locks_lock);
6124         if (!list_empty(&nbl->nbl_list)) {
6125                 list_del_init(&nbl->nbl_list);
6126                 list_del_init(&nbl->nbl_lru);
6127                 queue = true;
6128         }
6129         spin_unlock(&nn->blocked_locks_lock);
6130
6131         if (queue)
6132                 nfsd4_run_cb(&nbl->nbl_cb);
6133 }
6134
6135 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
6136         .lm_notify = nfsd4_lm_notify,
6137         .lm_get_owner = nfsd4_fl_get_owner,
6138         .lm_put_owner = nfsd4_fl_put_owner,
6139 };
6140
6141 static inline void
6142 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6143 {
6144         struct nfs4_lockowner *lo;
6145
6146         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6147                 lo = (struct nfs4_lockowner *) fl->fl_owner;
6148                 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6149                                                 GFP_KERNEL);
6150                 if (!deny->ld_owner.data)
6151                         /* We just don't care that much */
6152                         goto nevermind;
6153                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6154         } else {
6155 nevermind:
6156                 deny->ld_owner.len = 0;
6157                 deny->ld_owner.data = NULL;
6158                 deny->ld_clientid.cl_boot = 0;
6159                 deny->ld_clientid.cl_id = 0;
6160         }
6161         deny->ld_start = fl->fl_start;
6162         deny->ld_length = NFS4_MAX_UINT64;
6163         if (fl->fl_end != NFS4_MAX_UINT64)
6164                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6165         deny->ld_type = NFS4_READ_LT;
6166         if (fl->fl_type != F_RDLCK)
6167                 deny->ld_type = NFS4_WRITE_LT;
6168 }
6169
6170 static struct nfs4_lockowner *
6171 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6172 {
6173         unsigned int strhashval = ownerstr_hashval(owner);
6174         struct nfs4_stateowner *so;
6175
6176         lockdep_assert_held(&clp->cl_lock);
6177
6178         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6179                             so_strhash) {
6180                 if (so->so_is_open_owner)
6181                         continue;
6182                 if (same_owner_str(so, owner))
6183                         return lockowner(nfs4_get_stateowner(so));
6184         }
6185         return NULL;
6186 }
6187
6188 static struct nfs4_lockowner *
6189 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6190 {
6191         struct nfs4_lockowner *lo;
6192
6193         spin_lock(&clp->cl_lock);
6194         lo = find_lockowner_str_locked(clp, owner);
6195         spin_unlock(&clp->cl_lock);
6196         return lo;
6197 }
6198
6199 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6200 {
6201         unhash_lockowner_locked(lockowner(sop));
6202 }
6203
6204 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6205 {
6206         struct nfs4_lockowner *lo = lockowner(sop);
6207
6208         kmem_cache_free(lockowner_slab, lo);
6209 }
6210
6211 static const struct nfs4_stateowner_operations lockowner_ops = {
6212         .so_unhash =    nfs4_unhash_lockowner,
6213         .so_free =      nfs4_free_lockowner,
6214 };
6215
6216 /*
6217  * Alloc a lock owner structure.
6218  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6219  * occurred.
6220  *
6221  * strhashval = ownerstr_hashval
6222  */
6223 static struct nfs4_lockowner *
6224 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6225                            struct nfs4_ol_stateid *open_stp,
6226                            struct nfsd4_lock *lock)
6227 {
6228         struct nfs4_lockowner *lo, *ret;
6229
6230         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6231         if (!lo)
6232                 return NULL;
6233         INIT_LIST_HEAD(&lo->lo_blocked);
6234         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6235         lo->lo_owner.so_is_open_owner = 0;
6236         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6237         lo->lo_owner.so_ops = &lockowner_ops;
6238         spin_lock(&clp->cl_lock);
6239         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6240         if (ret == NULL) {
6241                 list_add(&lo->lo_owner.so_strhash,
6242                          &clp->cl_ownerstr_hashtbl[strhashval]);
6243                 ret = lo;
6244         } else
6245                 nfs4_free_stateowner(&lo->lo_owner);
6246
6247         spin_unlock(&clp->cl_lock);
6248         return ret;
6249 }
6250
6251 static struct nfs4_ol_stateid *
6252 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
6253 {
6254         struct nfs4_ol_stateid *lst;
6255         struct nfs4_client *clp = lo->lo_owner.so_client;
6256
6257         lockdep_assert_held(&clp->cl_lock);
6258
6259         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
6260                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
6261                         continue;
6262                 if (lst->st_stid.sc_file == fp) {
6263                         refcount_inc(&lst->st_stid.sc_count);
6264                         return lst;
6265                 }
6266         }
6267         return NULL;
6268 }
6269
6270 static struct nfs4_ol_stateid *
6271 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6272                   struct nfs4_file *fp, struct inode *inode,
6273                   struct nfs4_ol_stateid *open_stp)
6274 {
6275         struct nfs4_client *clp = lo->lo_owner.so_client;
6276         struct nfs4_ol_stateid *retstp;
6277
6278         mutex_init(&stp->st_mutex);
6279         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6280 retry:
6281         spin_lock(&clp->cl_lock);
6282         spin_lock(&fp->fi_lock);
6283         retstp = find_lock_stateid(lo, fp);
6284         if (retstp)
6285                 goto out_unlock;
6286
6287         refcount_inc(&stp->st_stid.sc_count);
6288         stp->st_stid.sc_type = NFS4_LOCK_STID;
6289         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6290         get_nfs4_file(fp);
6291         stp->st_stid.sc_file = fp;
6292         stp->st_access_bmap = 0;
6293         stp->st_deny_bmap = open_stp->st_deny_bmap;
6294         stp->st_openstp = open_stp;
6295         list_add(&stp->st_locks, &open_stp->st_locks);
6296         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6297         list_add(&stp->st_perfile, &fp->fi_stateids);
6298 out_unlock:
6299         spin_unlock(&fp->fi_lock);
6300         spin_unlock(&clp->cl_lock);
6301         if (retstp) {
6302                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6303                         nfs4_put_stid(&retstp->st_stid);
6304                         goto retry;
6305                 }
6306                 /* To keep mutex tracking happy */
6307                 mutex_unlock(&stp->st_mutex);
6308                 stp = retstp;
6309         }
6310         return stp;
6311 }
6312
6313 static struct nfs4_ol_stateid *
6314 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6315                             struct inode *inode, struct nfs4_ol_stateid *ost,
6316                             bool *new)
6317 {
6318         struct nfs4_stid *ns = NULL;
6319         struct nfs4_ol_stateid *lst;
6320         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6321         struct nfs4_client *clp = oo->oo_owner.so_client;
6322
6323         *new = false;
6324         spin_lock(&clp->cl_lock);
6325         lst = find_lock_stateid(lo, fi);
6326         spin_unlock(&clp->cl_lock);
6327         if (lst != NULL) {
6328                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6329                         goto out;
6330                 nfs4_put_stid(&lst->st_stid);
6331         }
6332         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6333         if (ns == NULL)
6334                 return NULL;
6335
6336         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6337         if (lst == openlockstateid(ns))
6338                 *new = true;
6339         else
6340                 nfs4_put_stid(ns);
6341 out:
6342         return lst;
6343 }
6344
6345 static int
6346 check_lock_length(u64 offset, u64 length)
6347 {
6348         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6349                 (length > ~offset)));
6350 }
6351
6352 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6353 {
6354         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6355
6356         lockdep_assert_held(&fp->fi_lock);
6357
6358         if (test_access(access, lock_stp))
6359                 return;
6360         __nfs4_file_get_access(fp, access);
6361         set_access(access, lock_stp);
6362 }
6363
6364 static __be32
6365 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6366                             struct nfs4_ol_stateid *ost,
6367                             struct nfsd4_lock *lock,
6368                             struct nfs4_ol_stateid **plst, bool *new)
6369 {
6370         __be32 status;
6371         struct nfs4_file *fi = ost->st_stid.sc_file;
6372         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6373         struct nfs4_client *cl = oo->oo_owner.so_client;
6374         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6375         struct nfs4_lockowner *lo;
6376         struct nfs4_ol_stateid *lst;
6377         unsigned int strhashval;
6378
6379         lo = find_lockowner_str(cl, &lock->lk_new_owner);
6380         if (!lo) {
6381                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6382                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6383                 if (lo == NULL)
6384                         return nfserr_jukebox;
6385         } else {
6386                 /* with an existing lockowner, seqids must be the same */
6387                 status = nfserr_bad_seqid;
6388                 if (!cstate->minorversion &&
6389                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6390                         goto out;
6391         }
6392
6393         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6394         if (lst == NULL) {
6395                 status = nfserr_jukebox;
6396                 goto out;
6397         }
6398
6399         status = nfs_ok;
6400         *plst = lst;
6401 out:
6402         nfs4_put_stateowner(&lo->lo_owner);
6403         return status;
6404 }
6405
6406 /*
6407  *  LOCK operation
6408  */
6409 __be32
6410 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6411            union nfsd4_op_u *u)
6412 {
6413         struct nfsd4_lock *lock = &u->lock;
6414         struct nfs4_openowner *open_sop = NULL;
6415         struct nfs4_lockowner *lock_sop = NULL;
6416         struct nfs4_ol_stateid *lock_stp = NULL;
6417         struct nfs4_ol_stateid *open_stp = NULL;
6418         struct nfs4_file *fp;
6419         struct nfsd_file *nf = NULL;
6420         struct nfsd4_blocked_lock *nbl = NULL;
6421         struct file_lock *file_lock = NULL;
6422         struct file_lock *conflock = NULL;
6423         __be32 status = 0;
6424         int lkflg;
6425         int err;
6426         bool new = false;
6427         unsigned char fl_type;
6428         unsigned int fl_flags = FL_POSIX;
6429         struct net *net = SVC_NET(rqstp);
6430         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6431
6432         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6433                 (long long) lock->lk_offset,
6434                 (long long) lock->lk_length);
6435
6436         if (check_lock_length(lock->lk_offset, lock->lk_length))
6437                  return nfserr_inval;
6438
6439         if ((status = fh_verify(rqstp, &cstate->current_fh,
6440                                 S_IFREG, NFSD_MAY_LOCK))) {
6441                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6442                 return status;
6443         }
6444
6445         if (lock->lk_is_new) {
6446                 if (nfsd4_has_session(cstate))
6447                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6448                         memcpy(&lock->lk_new_clientid,
6449                                 &cstate->session->se_client->cl_clientid,
6450                                 sizeof(clientid_t));
6451
6452                 status = nfserr_stale_clientid;
6453                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6454                         goto out;
6455
6456                 /* validate and update open stateid and open seqid */
6457                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6458                                         lock->lk_new_open_seqid,
6459                                         &lock->lk_new_open_stateid,
6460                                         &open_stp, nn);
6461                 if (status)
6462                         goto out;
6463                 mutex_unlock(&open_stp->st_mutex);
6464                 open_sop = openowner(open_stp->st_stateowner);
6465                 status = nfserr_bad_stateid;
6466                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6467                                                 &lock->lk_new_clientid))
6468                         goto out;
6469                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6470                                                         &lock_stp, &new);
6471         } else {
6472                 status = nfs4_preprocess_seqid_op(cstate,
6473                                        lock->lk_old_lock_seqid,
6474                                        &lock->lk_old_lock_stateid,
6475                                        NFS4_LOCK_STID, &lock_stp, nn);
6476         }
6477         if (status)
6478                 goto out;
6479         lock_sop = lockowner(lock_stp->st_stateowner);
6480
6481         lkflg = setlkflg(lock->lk_type);
6482         status = nfs4_check_openmode(lock_stp, lkflg);
6483         if (status)
6484                 goto out;
6485
6486         status = nfserr_grace;
6487         if (locks_in_grace(net) && !lock->lk_reclaim)
6488                 goto out;
6489         status = nfserr_no_grace;
6490         if (!locks_in_grace(net) && lock->lk_reclaim)
6491                 goto out;
6492
6493         fp = lock_stp->st_stid.sc_file;
6494         switch (lock->lk_type) {
6495                 case NFS4_READW_LT:
6496                         if (nfsd4_has_session(cstate))
6497                                 fl_flags |= FL_SLEEP;
6498                         /* Fallthrough */
6499                 case NFS4_READ_LT:
6500                         spin_lock(&fp->fi_lock);
6501                         nf = find_readable_file_locked(fp);
6502                         if (nf)
6503                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6504                         spin_unlock(&fp->fi_lock);
6505                         fl_type = F_RDLCK;
6506                         break;
6507                 case NFS4_WRITEW_LT:
6508                         if (nfsd4_has_session(cstate))
6509                                 fl_flags |= FL_SLEEP;
6510                         /* Fallthrough */
6511                 case NFS4_WRITE_LT:
6512                         spin_lock(&fp->fi_lock);
6513                         nf = find_writeable_file_locked(fp);
6514                         if (nf)
6515                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6516                         spin_unlock(&fp->fi_lock);
6517                         fl_type = F_WRLCK;
6518                         break;
6519                 default:
6520                         status = nfserr_inval;
6521                 goto out;
6522         }
6523
6524         if (!nf) {
6525                 status = nfserr_openmode;
6526                 goto out;
6527         }
6528
6529         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6530         if (!nbl) {
6531                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6532                 status = nfserr_jukebox;
6533                 goto out;
6534         }
6535
6536         file_lock = &nbl->nbl_lock;
6537         file_lock->fl_type = fl_type;
6538         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6539         file_lock->fl_pid = current->tgid;
6540         file_lock->fl_file = nf->nf_file;
6541         file_lock->fl_flags = fl_flags;
6542         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6543         file_lock->fl_start = lock->lk_offset;
6544         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6545         nfs4_transform_lock_offset(file_lock);
6546
6547         conflock = locks_alloc_lock();
6548         if (!conflock) {
6549                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6550                 status = nfserr_jukebox;
6551                 goto out;
6552         }
6553
6554         if (fl_flags & FL_SLEEP) {
6555                 nbl->nbl_time = get_seconds();
6556                 spin_lock(&nn->blocked_locks_lock);
6557                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6558                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6559                 spin_unlock(&nn->blocked_locks_lock);
6560         }
6561
6562         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6563         switch (err) {
6564         case 0: /* success! */
6565                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6566                 status = 0;
6567                 if (lock->lk_reclaim)
6568                         nn->somebody_reclaimed = true;
6569                 break;
6570         case FILE_LOCK_DEFERRED:
6571                 nbl = NULL;
6572                 /* Fallthrough */
6573         case -EAGAIN:           /* conflock holds conflicting lock */
6574                 status = nfserr_denied;
6575                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6576                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6577                 break;
6578         case -EDEADLK:
6579                 status = nfserr_deadlock;
6580                 break;
6581         default:
6582                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6583                 status = nfserrno(err);
6584                 break;
6585         }
6586 out:
6587         if (nbl) {
6588                 /* dequeue it if we queued it before */
6589                 if (fl_flags & FL_SLEEP) {
6590                         spin_lock(&nn->blocked_locks_lock);
6591                         list_del_init(&nbl->nbl_list);
6592                         list_del_init(&nbl->nbl_lru);
6593                         spin_unlock(&nn->blocked_locks_lock);
6594                 }
6595                 free_blocked_lock(nbl);
6596         }
6597         if (nf)
6598                 nfsd_file_put(nf);
6599         if (lock_stp) {
6600                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6601                 if (cstate->replay_owner &&
6602                     cstate->replay_owner != &lock_sop->lo_owner &&
6603                     seqid_mutating_err(ntohl(status)))
6604                         lock_sop->lo_owner.so_seqid++;
6605
6606                 /*
6607                  * If this is a new, never-before-used stateid, and we are
6608                  * returning an error, then just go ahead and release it.
6609                  */
6610                 if (status && new)
6611                         release_lock_stateid(lock_stp);
6612
6613                 mutex_unlock(&lock_stp->st_mutex);
6614
6615                 nfs4_put_stid(&lock_stp->st_stid);
6616         }
6617         if (open_stp)
6618                 nfs4_put_stid(&open_stp->st_stid);
6619         nfsd4_bump_seqid(cstate, status);
6620         if (conflock)
6621                 locks_free_lock(conflock);
6622         return status;
6623 }
6624
6625 /*
6626  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6627  * so we do a temporary open here just to get an open file to pass to
6628  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6629  * inode operation.)
6630  */
6631 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6632 {
6633         struct nfsd_file *nf;
6634         __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6635         if (!err) {
6636                 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6637                 nfsd_file_put(nf);
6638         }
6639         return err;
6640 }
6641
6642 /*
6643  * LOCKT operation
6644  */
6645 __be32
6646 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6647             union nfsd4_op_u *u)
6648 {
6649         struct nfsd4_lockt *lockt = &u->lockt;
6650         struct file_lock *file_lock = NULL;
6651         struct nfs4_lockowner *lo = NULL;
6652         __be32 status;
6653         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6654
6655         if (locks_in_grace(SVC_NET(rqstp)))
6656                 return nfserr_grace;
6657
6658         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6659                  return nfserr_inval;
6660
6661         if (!nfsd4_has_session(cstate)) {
6662                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6663                 if (status)
6664                         goto out;
6665         }
6666
6667         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6668                 goto out;
6669
6670         file_lock = locks_alloc_lock();
6671         if (!file_lock) {
6672                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6673                 status = nfserr_jukebox;
6674                 goto out;
6675         }
6676
6677         switch (lockt->lt_type) {
6678                 case NFS4_READ_LT:
6679                 case NFS4_READW_LT:
6680                         file_lock->fl_type = F_RDLCK;
6681                         break;
6682                 case NFS4_WRITE_LT:
6683                 case NFS4_WRITEW_LT:
6684                         file_lock->fl_type = F_WRLCK;
6685                         break;
6686                 default:
6687                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6688                         status = nfserr_inval;
6689                         goto out;
6690         }
6691
6692         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6693         if (lo)
6694                 file_lock->fl_owner = (fl_owner_t)lo;
6695         file_lock->fl_pid = current->tgid;
6696         file_lock->fl_flags = FL_POSIX;
6697
6698         file_lock->fl_start = lockt->lt_offset;
6699         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6700
6701         nfs4_transform_lock_offset(file_lock);
6702
6703         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6704         if (status)
6705                 goto out;
6706
6707         if (file_lock->fl_type != F_UNLCK) {
6708                 status = nfserr_denied;
6709                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6710         }
6711 out:
6712         if (lo)
6713                 nfs4_put_stateowner(&lo->lo_owner);
6714         if (file_lock)
6715                 locks_free_lock(file_lock);
6716         return status;
6717 }
6718
6719 __be32
6720 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6721             union nfsd4_op_u *u)
6722 {
6723         struct nfsd4_locku *locku = &u->locku;
6724         struct nfs4_ol_stateid *stp;
6725         struct nfsd_file *nf = NULL;
6726         struct file_lock *file_lock = NULL;
6727         __be32 status;
6728         int err;
6729         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6730
6731         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6732                 (long long) locku->lu_offset,
6733                 (long long) locku->lu_length);
6734
6735         if (check_lock_length(locku->lu_offset, locku->lu_length))
6736                  return nfserr_inval;
6737
6738         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6739                                         &locku->lu_stateid, NFS4_LOCK_STID,
6740                                         &stp, nn);
6741         if (status)
6742                 goto out;
6743         nf = find_any_file(stp->st_stid.sc_file);
6744         if (!nf) {
6745                 status = nfserr_lock_range;
6746                 goto put_stateid;
6747         }
6748         file_lock = locks_alloc_lock();
6749         if (!file_lock) {
6750                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6751                 status = nfserr_jukebox;
6752                 goto put_file;
6753         }
6754
6755         file_lock->fl_type = F_UNLCK;
6756         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6757         file_lock->fl_pid = current->tgid;
6758         file_lock->fl_file = nf->nf_file;
6759         file_lock->fl_flags = FL_POSIX;
6760         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6761         file_lock->fl_start = locku->lu_offset;
6762
6763         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6764                                                 locku->lu_length);
6765         nfs4_transform_lock_offset(file_lock);
6766
6767         err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
6768         if (err) {
6769                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6770                 goto out_nfserr;
6771         }
6772         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6773 put_file:
6774         nfsd_file_put(nf);
6775 put_stateid:
6776         mutex_unlock(&stp->st_mutex);
6777         nfs4_put_stid(&stp->st_stid);
6778 out:
6779         nfsd4_bump_seqid(cstate, status);
6780         if (file_lock)
6781                 locks_free_lock(file_lock);
6782         return status;
6783
6784 out_nfserr:
6785         status = nfserrno(err);
6786         goto put_file;
6787 }
6788
6789 /*
6790  * returns
6791  *      true:  locks held by lockowner
6792  *      false: no locks held by lockowner
6793  */
6794 static bool
6795 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6796 {
6797         struct file_lock *fl;
6798         int status = false;
6799         struct nfsd_file *nf = find_any_file(fp);
6800         struct inode *inode;
6801         struct file_lock_context *flctx;
6802
6803         if (!nf) {
6804                 /* Any valid lock stateid should have some sort of access */
6805                 WARN_ON_ONCE(1);
6806                 return status;
6807         }
6808
6809         inode = locks_inode(nf->nf_file);
6810         flctx = inode->i_flctx;
6811
6812         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6813                 spin_lock(&flctx->flc_lock);
6814                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6815                         if (fl->fl_owner == (fl_owner_t)lowner) {
6816                                 status = true;
6817                                 break;
6818                         }
6819                 }
6820                 spin_unlock(&flctx->flc_lock);
6821         }
6822         nfsd_file_put(nf);
6823         return status;
6824 }
6825
6826 __be32
6827 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6828                         struct nfsd4_compound_state *cstate,
6829                         union nfsd4_op_u *u)
6830 {
6831         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6832         clientid_t *clid = &rlockowner->rl_clientid;
6833         struct nfs4_stateowner *sop;
6834         struct nfs4_lockowner *lo = NULL;
6835         struct nfs4_ol_stateid *stp;
6836         struct xdr_netobj *owner = &rlockowner->rl_owner;
6837         unsigned int hashval = ownerstr_hashval(owner);
6838         __be32 status;
6839         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6840         struct nfs4_client *clp;
6841         LIST_HEAD (reaplist);
6842
6843         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6844                 clid->cl_boot, clid->cl_id);
6845
6846         status = lookup_clientid(clid, cstate, nn);
6847         if (status)
6848                 return status;
6849
6850         clp = cstate->clp;
6851         /* Find the matching lock stateowner */
6852         spin_lock(&clp->cl_lock);
6853         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6854                             so_strhash) {
6855
6856                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6857                         continue;
6858
6859                 /* see if there are still any locks associated with it */
6860                 lo = lockowner(sop);
6861                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6862                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6863                                 status = nfserr_locks_held;
6864                                 spin_unlock(&clp->cl_lock);
6865                                 return status;
6866                         }
6867                 }
6868
6869                 nfs4_get_stateowner(sop);
6870                 break;
6871         }
6872         if (!lo) {
6873                 spin_unlock(&clp->cl_lock);
6874                 return status;
6875         }
6876
6877         unhash_lockowner_locked(lo);
6878         while (!list_empty(&lo->lo_owner.so_stateids)) {
6879                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6880                                        struct nfs4_ol_stateid,
6881                                        st_perstateowner);
6882                 WARN_ON(!unhash_lock_stateid(stp));
6883                 put_ol_stateid_locked(stp, &reaplist);
6884         }
6885         spin_unlock(&clp->cl_lock);
6886         free_ol_stateid_reaplist(&reaplist);
6887         remove_blocked_locks(lo);
6888         nfs4_put_stateowner(&lo->lo_owner);
6889
6890         return status;
6891 }
6892
6893 static inline struct nfs4_client_reclaim *
6894 alloc_reclaim(void)
6895 {
6896         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6897 }
6898
6899 bool
6900 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
6901 {
6902         struct nfs4_client_reclaim *crp;
6903
6904         crp = nfsd4_find_reclaim_client(name, nn);
6905         return (crp && crp->cr_clp);
6906 }
6907
6908 /*
6909  * failure => all reset bets are off, nfserr_no_grace...
6910  *
6911  * The caller is responsible for freeing name.data if NULL is returned (it
6912  * will be freed in nfs4_remove_reclaim_record in the normal case).
6913  */
6914 struct nfs4_client_reclaim *
6915 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
6916                 struct nfsd_net *nn)
6917 {
6918         unsigned int strhashval;
6919         struct nfs4_client_reclaim *crp;
6920
6921         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
6922         crp = alloc_reclaim();
6923         if (crp) {
6924                 strhashval = clientstr_hashval(name);
6925                 INIT_LIST_HEAD(&crp->cr_strhash);
6926                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6927                 crp->cr_name.data = name.data;
6928                 crp->cr_name.len = name.len;
6929                 crp->cr_princhash.data = princhash.data;
6930                 crp->cr_princhash.len = princhash.len;
6931                 crp->cr_clp = NULL;
6932                 nn->reclaim_str_hashtbl_size++;
6933         }
6934         return crp;
6935 }
6936
6937 void
6938 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6939 {
6940         list_del(&crp->cr_strhash);
6941         kfree(crp->cr_name.data);
6942         kfree(crp->cr_princhash.data);
6943         kfree(crp);
6944         nn->reclaim_str_hashtbl_size--;
6945 }
6946
6947 void
6948 nfs4_release_reclaim(struct nfsd_net *nn)
6949 {
6950         struct nfs4_client_reclaim *crp = NULL;
6951         int i;
6952
6953         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6954                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6955                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6956                                         struct nfs4_client_reclaim, cr_strhash);
6957                         nfs4_remove_reclaim_record(crp, nn);
6958                 }
6959         }
6960         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6961 }
6962
6963 /*
6964  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6965 struct nfs4_client_reclaim *
6966 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
6967 {
6968         unsigned int strhashval;
6969         struct nfs4_client_reclaim *crp = NULL;
6970
6971         dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
6972
6973         strhashval = clientstr_hashval(name);
6974         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6975                 if (compare_blob(&crp->cr_name, &name) == 0) {
6976                         return crp;
6977                 }
6978         }
6979         return NULL;
6980 }
6981
6982 /*
6983 * Called from OPEN. Look for clientid in reclaim list.
6984 */
6985 __be32
6986 nfs4_check_open_reclaim(clientid_t *clid,
6987                 struct nfsd4_compound_state *cstate,
6988                 struct nfsd_net *nn)
6989 {
6990         __be32 status;
6991
6992         /* find clientid in conf_id_hashtbl */
6993         status = lookup_clientid(clid, cstate, nn);
6994         if (status)
6995                 return nfserr_reclaim_bad;
6996
6997         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6998                 return nfserr_no_grace;
6999
7000         if (nfsd4_client_record_check(cstate->clp))
7001                 return nfserr_reclaim_bad;
7002
7003         return nfs_ok;
7004 }
7005
7006 #ifdef CONFIG_NFSD_FAULT_INJECTION
7007 static inline void
7008 put_client(struct nfs4_client *clp)
7009 {
7010         atomic_dec(&clp->cl_rpc_users);
7011 }
7012
7013 static struct nfs4_client *
7014 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
7015 {
7016         struct nfs4_client *clp;
7017         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7018                                           nfsd_net_id);
7019
7020         if (!nfsd_netns_ready(nn))
7021                 return NULL;
7022
7023         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7024                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
7025                         return clp;
7026         }
7027         return NULL;
7028 }
7029
7030 u64
7031 nfsd_inject_print_clients(void)
7032 {
7033         struct nfs4_client *clp;
7034         u64 count = 0;
7035         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7036                                           nfsd_net_id);
7037         char buf[INET6_ADDRSTRLEN];
7038
7039         if (!nfsd_netns_ready(nn))
7040                 return 0;
7041
7042         spin_lock(&nn->client_lock);
7043         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7044                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7045                 pr_info("NFS Client: %s\n", buf);
7046                 ++count;
7047         }
7048         spin_unlock(&nn->client_lock);
7049
7050         return count;
7051 }
7052
7053 u64
7054 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7055 {
7056         u64 count = 0;
7057         struct nfs4_client *clp;
7058         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7059                                           nfsd_net_id);
7060
7061         if (!nfsd_netns_ready(nn))
7062                 return count;
7063
7064         spin_lock(&nn->client_lock);
7065         clp = nfsd_find_client(addr, addr_size);
7066         if (clp) {
7067                 if (mark_client_expired_locked(clp) == nfs_ok)
7068                         ++count;
7069                 else
7070                         clp = NULL;
7071         }
7072         spin_unlock(&nn->client_lock);
7073
7074         if (clp)
7075                 expire_client(clp);
7076
7077         return count;
7078 }
7079
7080 u64
7081 nfsd_inject_forget_clients(u64 max)
7082 {
7083         u64 count = 0;
7084         struct nfs4_client *clp, *next;
7085         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7086                                                 nfsd_net_id);
7087         LIST_HEAD(reaplist);
7088
7089         if (!nfsd_netns_ready(nn))
7090                 return count;
7091
7092         spin_lock(&nn->client_lock);
7093         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7094                 if (mark_client_expired_locked(clp) == nfs_ok) {
7095                         list_add(&clp->cl_lru, &reaplist);
7096                         if (max != 0 && ++count >= max)
7097                                 break;
7098                 }
7099         }
7100         spin_unlock(&nn->client_lock);
7101
7102         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7103                 expire_client(clp);
7104
7105         return count;
7106 }
7107
7108 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7109                              const char *type)
7110 {
7111         char buf[INET6_ADDRSTRLEN];
7112         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7113         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7114 }
7115
7116 static void
7117 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7118                              struct list_head *collect)
7119 {
7120         struct nfs4_client *clp = lst->st_stid.sc_client;
7121         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7122                                           nfsd_net_id);
7123
7124         if (!collect)
7125                 return;
7126
7127         lockdep_assert_held(&nn->client_lock);
7128         atomic_inc(&clp->cl_rpc_users);
7129         list_add(&lst->st_locks, collect);
7130 }
7131
7132 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7133                                     struct list_head *collect,
7134                                     bool (*func)(struct nfs4_ol_stateid *))
7135 {
7136         struct nfs4_openowner *oop;
7137         struct nfs4_ol_stateid *stp, *st_next;
7138         struct nfs4_ol_stateid *lst, *lst_next;
7139         u64 count = 0;
7140
7141         spin_lock(&clp->cl_lock);
7142         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7143                 list_for_each_entry_safe(stp, st_next,
7144                                 &oop->oo_owner.so_stateids, st_perstateowner) {
7145                         list_for_each_entry_safe(lst, lst_next,
7146                                         &stp->st_locks, st_locks) {
7147                                 if (func) {
7148                                         if (func(lst))
7149                                                 nfsd_inject_add_lock_to_list(lst,
7150                                                                         collect);
7151                                 }
7152                                 ++count;
7153                                 /*
7154                                  * Despite the fact that these functions deal
7155                                  * with 64-bit integers for "count", we must
7156                                  * ensure that it doesn't blow up the
7157                                  * clp->cl_rpc_users. Throw a warning if we
7158                                  * start to approach INT_MAX here.
7159                                  */
7160                                 WARN_ON_ONCE(count == (INT_MAX / 2));
7161                                 if (count == max)
7162                                         goto out;
7163                         }
7164                 }
7165         }
7166 out:
7167         spin_unlock(&clp->cl_lock);
7168
7169         return count;
7170 }
7171
7172 static u64
7173 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7174                           u64 max)
7175 {
7176         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7177 }
7178
7179 static u64
7180 nfsd_print_client_locks(struct nfs4_client *clp)
7181 {
7182         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7183         nfsd_print_count(clp, count, "locked files");
7184         return count;
7185 }
7186
7187 u64
7188 nfsd_inject_print_locks(void)
7189 {
7190         struct nfs4_client *clp;
7191         u64 count = 0;
7192         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7193                                                 nfsd_net_id);
7194
7195         if (!nfsd_netns_ready(nn))
7196                 return 0;
7197
7198         spin_lock(&nn->client_lock);
7199         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7200                 count += nfsd_print_client_locks(clp);
7201         spin_unlock(&nn->client_lock);
7202
7203         return count;
7204 }
7205
7206 static void
7207 nfsd_reap_locks(struct list_head *reaplist)
7208 {
7209         struct nfs4_client *clp;
7210         struct nfs4_ol_stateid *stp, *next;
7211
7212         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7213                 list_del_init(&stp->st_locks);
7214                 clp = stp->st_stid.sc_client;
7215                 nfs4_put_stid(&stp->st_stid);
7216                 put_client(clp);
7217         }
7218 }
7219
7220 u64
7221 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7222 {
7223         unsigned int count = 0;
7224         struct nfs4_client *clp;
7225         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7226                                                 nfsd_net_id);
7227         LIST_HEAD(reaplist);
7228
7229         if (!nfsd_netns_ready(nn))
7230                 return count;
7231
7232         spin_lock(&nn->client_lock);
7233         clp = nfsd_find_client(addr, addr_size);
7234         if (clp)
7235                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7236         spin_unlock(&nn->client_lock);
7237         nfsd_reap_locks(&reaplist);
7238         return count;
7239 }
7240
7241 u64
7242 nfsd_inject_forget_locks(u64 max)
7243 {
7244         u64 count = 0;
7245         struct nfs4_client *clp;
7246         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7247                                                 nfsd_net_id);
7248         LIST_HEAD(reaplist);
7249
7250         if (!nfsd_netns_ready(nn))
7251                 return count;
7252
7253         spin_lock(&nn->client_lock);
7254         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7255                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7256                 if (max != 0 && count >= max)
7257                         break;
7258         }
7259         spin_unlock(&nn->client_lock);
7260         nfsd_reap_locks(&reaplist);
7261         return count;
7262 }
7263
7264 static u64
7265 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7266                               struct list_head *collect,
7267                               void (*func)(struct nfs4_openowner *))
7268 {
7269         struct nfs4_openowner *oop, *next;
7270         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7271                                                 nfsd_net_id);
7272         u64 count = 0;
7273
7274         lockdep_assert_held(&nn->client_lock);
7275
7276         spin_lock(&clp->cl_lock);
7277         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7278                 if (func) {
7279                         func(oop);
7280                         if (collect) {
7281                                 atomic_inc(&clp->cl_rpc_users);
7282                                 list_add(&oop->oo_perclient, collect);
7283                         }
7284                 }
7285                 ++count;
7286                 /*
7287                  * Despite the fact that these functions deal with
7288                  * 64-bit integers for "count", we must ensure that
7289                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7290                  * warning if we start to approach INT_MAX here.
7291                  */
7292                 WARN_ON_ONCE(count == (INT_MAX / 2));
7293                 if (count == max)
7294                         break;
7295         }
7296         spin_unlock(&clp->cl_lock);
7297
7298         return count;
7299 }
7300
7301 static u64
7302 nfsd_print_client_openowners(struct nfs4_client *clp)
7303 {
7304         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7305
7306         nfsd_print_count(clp, count, "openowners");
7307         return count;
7308 }
7309
7310 static u64
7311 nfsd_collect_client_openowners(struct nfs4_client *clp,
7312                                struct list_head *collect, u64 max)
7313 {
7314         return nfsd_foreach_client_openowner(clp, max, collect,
7315                                                 unhash_openowner_locked);
7316 }
7317
7318 u64
7319 nfsd_inject_print_openowners(void)
7320 {
7321         struct nfs4_client *clp;
7322         u64 count = 0;
7323         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7324                                                 nfsd_net_id);
7325
7326         if (!nfsd_netns_ready(nn))
7327                 return 0;
7328
7329         spin_lock(&nn->client_lock);
7330         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7331                 count += nfsd_print_client_openowners(clp);
7332         spin_unlock(&nn->client_lock);
7333
7334         return count;
7335 }
7336
7337 static void
7338 nfsd_reap_openowners(struct list_head *reaplist)
7339 {
7340         struct nfs4_client *clp;
7341         struct nfs4_openowner *oop, *next;
7342
7343         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7344                 list_del_init(&oop->oo_perclient);
7345                 clp = oop->oo_owner.so_client;
7346                 release_openowner(oop);
7347                 put_client(clp);
7348         }
7349 }
7350
7351 u64
7352 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7353                                      size_t addr_size)
7354 {
7355         unsigned int count = 0;
7356         struct nfs4_client *clp;
7357         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7358                                                 nfsd_net_id);
7359         LIST_HEAD(reaplist);
7360
7361         if (!nfsd_netns_ready(nn))
7362                 return count;
7363
7364         spin_lock(&nn->client_lock);
7365         clp = nfsd_find_client(addr, addr_size);
7366         if (clp)
7367                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7368         spin_unlock(&nn->client_lock);
7369         nfsd_reap_openowners(&reaplist);
7370         return count;
7371 }
7372
7373 u64
7374 nfsd_inject_forget_openowners(u64 max)
7375 {
7376         u64 count = 0;
7377         struct nfs4_client *clp;
7378         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7379                                                 nfsd_net_id);
7380         LIST_HEAD(reaplist);
7381
7382         if (!nfsd_netns_ready(nn))
7383                 return count;
7384
7385         spin_lock(&nn->client_lock);
7386         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7387                 count += nfsd_collect_client_openowners(clp, &reaplist,
7388                                                         max - count);
7389                 if (max != 0 && count >= max)
7390                         break;
7391         }
7392         spin_unlock(&nn->client_lock);
7393         nfsd_reap_openowners(&reaplist);
7394         return count;
7395 }
7396
7397 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7398                                      struct list_head *victims)
7399 {
7400         struct nfs4_delegation *dp, *next;
7401         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7402                                                 nfsd_net_id);
7403         u64 count = 0;
7404
7405         lockdep_assert_held(&nn->client_lock);
7406
7407         spin_lock(&state_lock);
7408         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7409                 if (victims) {
7410                         /*
7411                          * It's not safe to mess with delegations that have a
7412                          * non-zero dl_time. They might have already been broken
7413                          * and could be processed by the laundromat outside of
7414                          * the state_lock. Just leave them be.
7415                          */
7416                         if (dp->dl_time != 0)
7417                                 continue;
7418
7419                         atomic_inc(&clp->cl_rpc_users);
7420                         WARN_ON(!unhash_delegation_locked(dp));
7421                         list_add(&dp->dl_recall_lru, victims);
7422                 }
7423                 ++count;
7424                 /*
7425                  * Despite the fact that these functions deal with
7426                  * 64-bit integers for "count", we must ensure that
7427                  * it doesn't blow up the clp->cl_rpc_users. Throw a
7428                  * warning if we start to approach INT_MAX here.
7429                  */
7430                 WARN_ON_ONCE(count == (INT_MAX / 2));
7431                 if (count == max)
7432                         break;
7433         }
7434         spin_unlock(&state_lock);
7435         return count;
7436 }
7437
7438 static u64
7439 nfsd_print_client_delegations(struct nfs4_client *clp)
7440 {
7441         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7442
7443         nfsd_print_count(clp, count, "delegations");
7444         return count;
7445 }
7446
7447 u64
7448 nfsd_inject_print_delegations(void)
7449 {
7450         struct nfs4_client *clp;
7451         u64 count = 0;
7452         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7453                                                 nfsd_net_id);
7454
7455         if (!nfsd_netns_ready(nn))
7456                 return 0;
7457
7458         spin_lock(&nn->client_lock);
7459         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7460                 count += nfsd_print_client_delegations(clp);
7461         spin_unlock(&nn->client_lock);
7462
7463         return count;
7464 }
7465
7466 static void
7467 nfsd_forget_delegations(struct list_head *reaplist)
7468 {
7469         struct nfs4_client *clp;
7470         struct nfs4_delegation *dp, *next;
7471
7472         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7473                 list_del_init(&dp->dl_recall_lru);
7474                 clp = dp->dl_stid.sc_client;
7475                 revoke_delegation(dp);
7476                 put_client(clp);
7477         }
7478 }
7479
7480 u64
7481 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7482                                       size_t addr_size)
7483 {
7484         u64 count = 0;
7485         struct nfs4_client *clp;
7486         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7487                                                 nfsd_net_id);
7488         LIST_HEAD(reaplist);
7489
7490         if (!nfsd_netns_ready(nn))
7491                 return count;
7492
7493         spin_lock(&nn->client_lock);
7494         clp = nfsd_find_client(addr, addr_size);
7495         if (clp)
7496                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7497         spin_unlock(&nn->client_lock);
7498
7499         nfsd_forget_delegations(&reaplist);
7500         return count;
7501 }
7502
7503 u64
7504 nfsd_inject_forget_delegations(u64 max)
7505 {
7506         u64 count = 0;
7507         struct nfs4_client *clp;
7508         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7509                                                 nfsd_net_id);
7510         LIST_HEAD(reaplist);
7511
7512         if (!nfsd_netns_ready(nn))
7513                 return count;
7514
7515         spin_lock(&nn->client_lock);
7516         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7517                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7518                 if (max != 0 && count >= max)
7519                         break;
7520         }
7521         spin_unlock(&nn->client_lock);
7522         nfsd_forget_delegations(&reaplist);
7523         return count;
7524 }
7525
7526 static void
7527 nfsd_recall_delegations(struct list_head *reaplist)
7528 {
7529         struct nfs4_client *clp;
7530         struct nfs4_delegation *dp, *next;
7531
7532         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7533                 list_del_init(&dp->dl_recall_lru);
7534                 clp = dp->dl_stid.sc_client;
7535                 /*
7536                  * We skipped all entries that had a zero dl_time before,
7537                  * so we can now reset the dl_time back to 0. If a delegation
7538                  * break comes in now, then it won't make any difference since
7539                  * we're recalling it either way.
7540                  */
7541                 spin_lock(&state_lock);
7542                 dp->dl_time = 0;
7543                 spin_unlock(&state_lock);
7544                 nfsd_break_one_deleg(dp);
7545                 put_client(clp);
7546         }
7547 }
7548
7549 u64
7550 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7551                                       size_t addr_size)
7552 {
7553         u64 count = 0;
7554         struct nfs4_client *clp;
7555         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7556                                                 nfsd_net_id);
7557         LIST_HEAD(reaplist);
7558
7559         if (!nfsd_netns_ready(nn))
7560                 return count;
7561
7562         spin_lock(&nn->client_lock);
7563         clp = nfsd_find_client(addr, addr_size);
7564         if (clp)
7565                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7566         spin_unlock(&nn->client_lock);
7567
7568         nfsd_recall_delegations(&reaplist);
7569         return count;
7570 }
7571
7572 u64
7573 nfsd_inject_recall_delegations(u64 max)
7574 {
7575         u64 count = 0;
7576         struct nfs4_client *clp, *next;
7577         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7578                                                 nfsd_net_id);
7579         LIST_HEAD(reaplist);
7580
7581         if (!nfsd_netns_ready(nn))
7582                 return count;
7583
7584         spin_lock(&nn->client_lock);
7585         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7586                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7587                 if (max != 0 && ++count >= max)
7588                         break;
7589         }
7590         spin_unlock(&nn->client_lock);
7591         nfsd_recall_delegations(&reaplist);
7592         return count;
7593 }
7594 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7595
7596 /*
7597  * Since the lifetime of a delegation isn't limited to that of an open, a
7598  * client may quite reasonably hang on to a delegation as long as it has
7599  * the inode cached.  This becomes an obvious problem the first time a
7600  * client's inode cache approaches the size of the server's total memory.
7601  *
7602  * For now we avoid this problem by imposing a hard limit on the number
7603  * of delegations, which varies according to the server's memory size.
7604  */
7605 static void
7606 set_max_delegations(void)
7607 {
7608         /*
7609          * Allow at most 4 delegations per megabyte of RAM.  Quick
7610          * estimates suggest that in the worst case (where every delegation
7611          * is for a different inode), a delegation could take about 1.5K,
7612          * giving a worst case usage of about 6% of memory.
7613          */
7614         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7615 }
7616
7617 static int nfs4_state_create_net(struct net *net)
7618 {
7619         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7620         int i;
7621
7622         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7623                                             sizeof(struct list_head),
7624                                             GFP_KERNEL);
7625         if (!nn->conf_id_hashtbl)
7626                 goto err;
7627         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7628                                               sizeof(struct list_head),
7629                                               GFP_KERNEL);
7630         if (!nn->unconf_id_hashtbl)
7631                 goto err_unconf_id;
7632         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7633                                               sizeof(struct list_head),
7634                                               GFP_KERNEL);
7635         if (!nn->sessionid_hashtbl)
7636                 goto err_sessionid;
7637
7638         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7639                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7640                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7641         }
7642         for (i = 0; i < SESSION_HASH_SIZE; i++)
7643                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7644         nn->conf_name_tree = RB_ROOT;
7645         nn->unconf_name_tree = RB_ROOT;
7646         nn->boot_time = get_seconds();
7647         nn->grace_ended = false;
7648         nn->nfsd4_manager.block_opens = true;
7649         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7650         INIT_LIST_HEAD(&nn->client_lru);
7651         INIT_LIST_HEAD(&nn->close_lru);
7652         INIT_LIST_HEAD(&nn->del_recall_lru);
7653         spin_lock_init(&nn->client_lock);
7654         spin_lock_init(&nn->s2s_cp_lock);
7655         idr_init(&nn->s2s_cp_stateids);
7656
7657         spin_lock_init(&nn->blocked_locks_lock);
7658         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7659
7660         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7661         get_net(net);
7662
7663         return 0;
7664
7665 err_sessionid:
7666         kfree(nn->unconf_id_hashtbl);
7667 err_unconf_id:
7668         kfree(nn->conf_id_hashtbl);
7669 err:
7670         return -ENOMEM;
7671 }
7672
7673 static void
7674 nfs4_state_destroy_net(struct net *net)
7675 {
7676         int i;
7677         struct nfs4_client *clp = NULL;
7678         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7679
7680         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7681                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7682                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7683                         destroy_client(clp);
7684                 }
7685         }
7686
7687         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7688
7689         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7690                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7691                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7692                         destroy_client(clp);
7693                 }
7694         }
7695
7696         kfree(nn->sessionid_hashtbl);
7697         kfree(nn->unconf_id_hashtbl);
7698         kfree(nn->conf_id_hashtbl);
7699         put_net(net);
7700 }
7701
7702 int
7703 nfs4_state_start_net(struct net *net)
7704 {
7705         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7706         int ret;
7707
7708         ret = nfs4_state_create_net(net);
7709         if (ret)
7710                 return ret;
7711         locks_start_grace(net, &nn->nfsd4_manager);
7712         nfsd4_client_tracking_init(net);
7713         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7714                 goto skip_grace;
7715         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7716                nn->nfsd4_grace, net->ns.inum);
7717         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7718         return 0;
7719
7720 skip_grace:
7721         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7722                         net->ns.inum);
7723         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7724         nfsd4_end_grace(nn);
7725         return 0;
7726 }
7727
7728 /* initialization to perform when the nfsd service is started: */
7729
7730 int
7731 nfs4_state_start(void)
7732 {
7733         int ret;
7734
7735         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7736         if (laundry_wq == NULL) {
7737                 ret = -ENOMEM;
7738                 goto out;
7739         }
7740         ret = nfsd4_create_callback_queue();
7741         if (ret)
7742                 goto out_free_laundry;
7743
7744         set_max_delegations();
7745         return 0;
7746
7747 out_free_laundry:
7748         destroy_workqueue(laundry_wq);
7749 out:
7750         return ret;
7751 }
7752
7753 void
7754 nfs4_state_shutdown_net(struct net *net)
7755 {
7756         struct nfs4_delegation *dp = NULL;
7757         struct list_head *pos, *next, reaplist;
7758         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7759
7760         cancel_delayed_work_sync(&nn->laundromat_work);
7761         locks_end_grace(&nn->nfsd4_manager);
7762
7763         INIT_LIST_HEAD(&reaplist);
7764         spin_lock(&state_lock);
7765         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7766                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7767                 WARN_ON(!unhash_delegation_locked(dp));
7768                 list_add(&dp->dl_recall_lru, &reaplist);
7769         }
7770         spin_unlock(&state_lock);
7771         list_for_each_safe(pos, next, &reaplist) {
7772                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7773                 list_del_init(&dp->dl_recall_lru);
7774                 destroy_unhashed_deleg(dp);
7775         }
7776
7777         nfsd4_client_tracking_exit(net);
7778         nfs4_state_destroy_net(net);
7779 }
7780
7781 void
7782 nfs4_state_shutdown(void)
7783 {
7784         destroy_workqueue(laundry_wq);
7785         nfsd4_destroy_callback_queue();
7786 }
7787
7788 static void
7789 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7790 {
7791         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7792                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7793 }
7794
7795 static void
7796 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7797 {
7798         if (cstate->minorversion) {
7799                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7800                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7801         }
7802 }
7803
7804 void
7805 clear_current_stateid(struct nfsd4_compound_state *cstate)
7806 {
7807         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7808 }
7809
7810 /*
7811  * functions to set current state id
7812  */
7813 void
7814 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7815                 union nfsd4_op_u *u)
7816 {
7817         put_stateid(cstate, &u->open_downgrade.od_stateid);
7818 }
7819
7820 void
7821 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7822                 union nfsd4_op_u *u)
7823 {
7824         put_stateid(cstate, &u->open.op_stateid);
7825 }
7826
7827 void
7828 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7829                 union nfsd4_op_u *u)
7830 {
7831         put_stateid(cstate, &u->close.cl_stateid);
7832 }
7833
7834 void
7835 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7836                 union nfsd4_op_u *u)
7837 {
7838         put_stateid(cstate, &u->lock.lk_resp_stateid);
7839 }
7840
7841 /*
7842  * functions to consume current state id
7843  */
7844
7845 void
7846 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7847                 union nfsd4_op_u *u)
7848 {
7849         get_stateid(cstate, &u->open_downgrade.od_stateid);
7850 }
7851
7852 void
7853 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7854                 union nfsd4_op_u *u)
7855 {
7856         get_stateid(cstate, &u->delegreturn.dr_stateid);
7857 }
7858
7859 void
7860 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7861                 union nfsd4_op_u *u)
7862 {
7863         get_stateid(cstate, &u->free_stateid.fr_stateid);
7864 }
7865
7866 void
7867 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7868                 union nfsd4_op_u *u)
7869 {
7870         get_stateid(cstate, &u->setattr.sa_stateid);
7871 }
7872
7873 void
7874 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7875                 union nfsd4_op_u *u)
7876 {
7877         get_stateid(cstate, &u->close.cl_stateid);
7878 }
7879
7880 void
7881 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7882                 union nfsd4_op_u *u)
7883 {
7884         get_stateid(cstate, &u->locku.lu_stateid);
7885 }
7886
7887 void
7888 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7889                 union nfsd4_op_u *u)
7890 {
7891         get_stateid(cstate, &u->read.rd_stateid);
7892 }
7893
7894 void
7895 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7896                 union nfsd4_op_u *u)
7897 {
7898         get_stateid(cstate, &u->write.wr_stateid);
7899 }