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 "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51 #include "pnfs.h"
  52
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 #define all_ones {{~0,~0},~0}
  56 static const stateid_t one_stateid = {
  57         .si_generation = ~0,
  58         .si_opaque = all_ones,
  59 };
  60 static const stateid_t zero_stateid = {
  61         /* all fields zero */
  62 };
  63 static const stateid_t currentstateid = {
  64         .si_generation = 1,
  65 };
  66
  67 static u64 current_sessionid = 1;
  68
  69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  70 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  72
  73 /* forward declarations */
  74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  76
  77 /* Locking: */
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(state_lock);
  85
  86 /*
  87  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  88  * the refcount on the open stateid to drop.
  89  */
  90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  91
  92 static struct kmem_cache *openowner_slab;
  93 static struct kmem_cache *lockowner_slab;
  94 static struct kmem_cache *file_slab;
  95 static struct kmem_cache *stateid_slab;
  96 static struct kmem_cache *deleg_slab;
  97 static struct kmem_cache *odstate_slab;
  98
  99 static void free_session(struct nfsd4_session *);
 100
 101 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 102 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 103
 104 static bool is_session_dead(struct nfsd4_session *ses)
 105 {
 106         return ses->se_flags & NFS4_SESSION_DEAD;
 107 }
 108
 109 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 110 {
 111         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 112                 return nfserr_jukebox;
 113         ses->se_flags |= NFS4_SESSION_DEAD;
 114         return nfs_ok;
 115 }
 116
 117 static bool is_client_expired(struct nfs4_client *clp)
 118 {
 119         return clp->cl_time == 0;
 120 }
 121
 122 static __be32 get_client_locked(struct nfs4_client *clp)
 123 {
 124         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 125
 126         lockdep_assert_held(&nn->client_lock);
 127
 128         if (is_client_expired(clp))
 129                 return nfserr_expired;
 130         atomic_inc(&clp->cl_refcount);
 131         return nfs_ok;
 132 }
 133
 134 /* must be called under the client_lock */
 135 static inline void
 136 renew_client_locked(struct nfs4_client *clp)
 137 {
 138         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 139
 140         if (is_client_expired(clp)) {
 141                 WARN_ON(1);
 142                 printk("%s: client (clientid %08x/%08x) already expired\n",
 143                         __func__,
 144                         clp->cl_clientid.cl_boot,
 145                         clp->cl_clientid.cl_id);
 146                 return;
 147         }
 148
 149         dprintk("renewing client (clientid %08x/%08x)\n",
 150                         clp->cl_clientid.cl_boot,
 151                         clp->cl_clientid.cl_id);
 152         list_move_tail(&clp->cl_lru, &nn->client_lru);
 153         clp->cl_time = get_seconds();
 154 }
 155
 156 static void put_client_renew_locked(struct nfs4_client *clp)
 157 {
 158         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 159
 160         lockdep_assert_held(&nn->client_lock);
 161
 162         if (!atomic_dec_and_test(&clp->cl_refcount))
 163                 return;
 164         if (!is_client_expired(clp))
 165                 renew_client_locked(clp);
 166 }
 167
 168 static void put_client_renew(struct nfs4_client *clp)
 169 {
 170         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 171
 172         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 173                 return;
 174         if (!is_client_expired(clp))
 175                 renew_client_locked(clp);
 176         spin_unlock(&nn->client_lock);
 177 }
 178
 179 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 180 {
 181         __be32 status;
 182
 183         if (is_session_dead(ses))
 184                 return nfserr_badsession;
 185         status = get_client_locked(ses->se_client);
 186         if (status)
 187                 return status;
 188         atomic_inc(&ses->se_ref);
 189         return nfs_ok;
 190 }
 191
 192 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 193 {
 194         struct nfs4_client *clp = ses->se_client;
 195         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 196
 197         lockdep_assert_held(&nn->client_lock);
 198
 199         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 200                 free_session(ses);
 201         put_client_renew_locked(clp);
 202 }
 203
 204 static void nfsd4_put_session(struct nfsd4_session *ses)
 205 {
 206         struct nfs4_client *clp = ses->se_client;
 207         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 208
 209         spin_lock(&nn->client_lock);
 210         nfsd4_put_session_locked(ses);
 211         spin_unlock(&nn->client_lock);
 212 }
 213
 214 static struct nfsd4_blocked_lock *
 215 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 216                         struct nfsd_net *nn)
 217 {
 218         struct nfsd4_blocked_lock *cur, *found = NULL;
 219
 220         spin_lock(&nn->blocked_locks_lock);
 221         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 222                 if (fh_match(fh, &cur->nbl_fh)) {
 223                         list_del_init(&cur->nbl_list);
 224                         list_del_init(&cur->nbl_lru);
 225                         found = cur;
 226                         break;
 227                 }
 228         }
 229         spin_unlock(&nn->blocked_locks_lock);
 230         if (found)
 231                 posix_unblock_lock(&found->nbl_lock);
 232         return found;
 233 }
 234
 235 static struct nfsd4_blocked_lock *
 236 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 237                         struct nfsd_net *nn)
 238 {
 239         struct nfsd4_blocked_lock *nbl;
 240
 241         nbl = find_blocked_lock(lo, fh, nn);
 242         if (!nbl) {
 243                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 244                 if (nbl) {
 245                         fh_copy_shallow(&nbl->nbl_fh, fh);
 246                         locks_init_lock(&nbl->nbl_lock);
 247                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 248                                         &nfsd4_cb_notify_lock_ops,
 249                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 250                 }
 251         }
 252         return nbl;
 253 }
 254
 255 static void
 256 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 257 {
 258         locks_release_private(&nbl->nbl_lock);
 259         kfree(nbl);
 260 }
 261
 262 static int
 263 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 264 {
 265         /*
 266          * Since this is just an optimization, we don't try very hard if it
 267          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 268          * just quit trying on anything else.
 269          */
 270         switch (task->tk_status) {
 271         case -NFS4ERR_DELAY:
 272                 rpc_delay(task, 1 * HZ);
 273                 return 0;
 274         default:
 275                 return 1;
 276         }
 277 }
 278
 279 static void
 280 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 281 {
 282         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 283                                                 struct nfsd4_blocked_lock, nbl_cb);
 284
 285         free_blocked_lock(nbl);
 286 }
 287
 288 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 289         .done           = nfsd4_cb_notify_lock_done,
 290         .release        = nfsd4_cb_notify_lock_release,
 291 };
 292
 293 static inline struct nfs4_stateowner *
 294 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 295 {
 296         atomic_inc(&sop->so_count);
 297         return sop;
 298 }
 299
 300 static int
 301 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 302 {
 303         return (sop->so_owner.len == owner->len) &&
 304                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 305 }
 306
 307 static struct nfs4_openowner *
 308 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 309                         struct nfs4_client *clp)
 310 {
 311         struct nfs4_stateowner *so;
 312
 313         lockdep_assert_held(&clp->cl_lock);
 314
 315         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 316                             so_strhash) {
 317                 if (!so->so_is_open_owner)
 318                         continue;
 319                 if (same_owner_str(so, &open->op_owner))
 320                         return openowner(nfs4_get_stateowner(so));
 321         }
 322         return NULL;
 323 }
 324
 325 static struct nfs4_openowner *
 326 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 327                         struct nfs4_client *clp)
 328 {
 329         struct nfs4_openowner *oo;
 330
 331         spin_lock(&clp->cl_lock);
 332         oo = find_openstateowner_str_locked(hashval, open, clp);
 333         spin_unlock(&clp->cl_lock);
 334         return oo;
 335 }
 336
 337 static inline u32
 338 opaque_hashval(const void *ptr, int nbytes)
 339 {
 340         unsigned char *cptr = (unsigned char *) ptr;
 341
 342         u32 x = 0;
 343         while (nbytes--) {
 344                 x *= 37;
 345                 x += *cptr++;
 346         }
 347         return x;
 348 }
 349
 350 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 351 {
 352         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 353
 354         kmem_cache_free(file_slab, fp);
 355 }
 356
 357 void
 358 put_nfs4_file(struct nfs4_file *fi)
 359 {
 360         might_lock(&state_lock);
 361
 362         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 363                 hlist_del_rcu(&fi->fi_hash);
 364                 spin_unlock(&state_lock);
 365                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 366                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 367                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 368         }
 369 }
 370
 371 static struct file *
 372 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 373 {
 374         if (f->fi_fds[oflag])
 375                 return get_file(f->fi_fds[oflag]);
 376         return NULL;
 377 }
 378
 379 static struct file *
 380 find_writeable_file_locked(struct nfs4_file *f)
 381 {
 382         struct file *ret;
 383
 384         lockdep_assert_held(&f->fi_lock);
 385
 386         ret = __nfs4_get_fd(f, O_WRONLY);
 387         if (!ret)
 388                 ret = __nfs4_get_fd(f, O_RDWR);
 389         return ret;
 390 }
 391
 392 static struct file *
 393 find_writeable_file(struct nfs4_file *f)
 394 {
 395         struct file *ret;
 396
 397         spin_lock(&f->fi_lock);
 398         ret = find_writeable_file_locked(f);
 399         spin_unlock(&f->fi_lock);
 400
 401         return ret;
 402 }
 403
 404 static struct file *find_readable_file_locked(struct nfs4_file *f)
 405 {
 406         struct file *ret;
 407
 408         lockdep_assert_held(&f->fi_lock);
 409
 410         ret = __nfs4_get_fd(f, O_RDONLY);
 411         if (!ret)
 412                 ret = __nfs4_get_fd(f, O_RDWR);
 413         return ret;
 414 }
 415
 416 static struct file *
 417 find_readable_file(struct nfs4_file *f)
 418 {
 419         struct file *ret;
 420
 421         spin_lock(&f->fi_lock);
 422         ret = find_readable_file_locked(f);
 423         spin_unlock(&f->fi_lock);
 424
 425         return ret;
 426 }
 427
 428 struct file *
 429 find_any_file(struct nfs4_file *f)
 430 {
 431         struct file *ret;
 432
 433         spin_lock(&f->fi_lock);
 434         ret = __nfs4_get_fd(f, O_RDWR);
 435         if (!ret) {
 436                 ret = __nfs4_get_fd(f, O_WRONLY);
 437                 if (!ret)
 438                         ret = __nfs4_get_fd(f, O_RDONLY);
 439         }
 440         spin_unlock(&f->fi_lock);
 441         return ret;
 442 }
 443
 444 static atomic_long_t num_delegations;
 445 unsigned long max_delegations;
 446
 447 /*
 448  * Open owner state (share locks)
 449  */
 450
 451 /* hash tables for lock and open owners */
 452 #define OWNER_HASH_BITS              8
 453 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 454 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 455
 456 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 457 {
 458         unsigned int ret;
 459
 460         ret = opaque_hashval(ownername->data, ownername->len);
 461         return ret & OWNER_HASH_MASK;
 462 }
 463
 464 /* hash table for nfs4_file */
 465 #define FILE_HASH_BITS                   8
 466 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 467
 468 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 469 {
 470         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 471 }
 472
 473 static unsigned int file_hashval(struct knfsd_fh *fh)
 474 {
 475         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 476 }
 477
 478 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 479
 480 static void
 481 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 482 {
 483         lockdep_assert_held(&fp->fi_lock);
 484
 485         if (access & NFS4_SHARE_ACCESS_WRITE)
 486                 atomic_inc(&fp->fi_access[O_WRONLY]);
 487         if (access & NFS4_SHARE_ACCESS_READ)
 488                 atomic_inc(&fp->fi_access[O_RDONLY]);
 489 }
 490
 491 static __be32
 492 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 493 {
 494         lockdep_assert_held(&fp->fi_lock);
 495
 496         /* Does this access mode make sense? */
 497         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 498                 return nfserr_inval;
 499
 500         /* Does it conflict with a deny mode already set? */
 501         if ((access & fp->fi_share_deny) != 0)
 502                 return nfserr_share_denied;
 503
 504         __nfs4_file_get_access(fp, access);
 505         return nfs_ok;
 506 }
 507
 508 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 509 {
 510         /* Common case is that there is no deny mode. */
 511         if (deny) {
 512                 /* Does this deny mode make sense? */
 513                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 514                         return nfserr_inval;
 515
 516                 if ((deny & NFS4_SHARE_DENY_READ) &&
 517                     atomic_read(&fp->fi_access[O_RDONLY]))
 518                         return nfserr_share_denied;
 519
 520                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 521                     atomic_read(&fp->fi_access[O_WRONLY]))
 522                         return nfserr_share_denied;
 523         }
 524         return nfs_ok;
 525 }
 526
 527 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 528 {
 529         might_lock(&fp->fi_lock);
 530
 531         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 532                 struct file *f1 = NULL;
 533                 struct file *f2 = NULL;
 534
 535                 swap(f1, fp->fi_fds[oflag]);
 536                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 537                         swap(f2, fp->fi_fds[O_RDWR]);
 538                 spin_unlock(&fp->fi_lock);
 539                 if (f1)
 540                         fput(f1);
 541                 if (f2)
 542                         fput(f2);
 543         }
 544 }
 545
 546 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 547 {
 548         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 549
 550         if (access & NFS4_SHARE_ACCESS_WRITE)
 551                 __nfs4_file_put_access(fp, O_WRONLY);
 552         if (access & NFS4_SHARE_ACCESS_READ)
 553                 __nfs4_file_put_access(fp, O_RDONLY);
 554 }
 555
 556 /*
 557  * Allocate a new open/delegation state counter. This is needed for
 558  * pNFS for proper return on close semantics.
 559  *
 560  * Note that we only allocate it for pNFS-enabled exports, otherwise
 561  * all pointers to struct nfs4_clnt_odstate are always NULL.
 562  */
 563 static struct nfs4_clnt_odstate *
 564 alloc_clnt_odstate(struct nfs4_client *clp)
 565 {
 566         struct nfs4_clnt_odstate *co;
 567
 568         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 569         if (co) {
 570                 co->co_client = clp;
 571                 refcount_set(&co->co_odcount, 1);
 572         }
 573         return co;
 574 }
 575
 576 static void
 577 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 578 {
 579         struct nfs4_file *fp = co->co_file;
 580
 581         lockdep_assert_held(&fp->fi_lock);
 582         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 583 }
 584
 585 static inline void
 586 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 587 {
 588         if (co)
 589                 refcount_inc(&co->co_odcount);
 590 }
 591
 592 static void
 593 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 594 {
 595         struct nfs4_file *fp;
 596
 597         if (!co)
 598                 return;
 599
 600         fp = co->co_file;
 601         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 602                 list_del(&co->co_perfile);
 603                 spin_unlock(&fp->fi_lock);
 604
 605                 nfsd4_return_all_file_layouts(co->co_client, fp);
 606                 kmem_cache_free(odstate_slab, co);
 607         }
 608 }
 609
 610 static struct nfs4_clnt_odstate *
 611 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 612 {
 613         struct nfs4_clnt_odstate *co;
 614         struct nfs4_client *cl;
 615
 616         if (!new)
 617                 return NULL;
 618
 619         cl = new->co_client;
 620
 621         spin_lock(&fp->fi_lock);
 622         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 623                 if (co->co_client == cl) {
 624                         get_clnt_odstate(co);
 625                         goto out;
 626                 }
 627         }
 628         co = new;
 629         co->co_file = fp;
 630         hash_clnt_odstate_locked(new);
 631 out:
 632         spin_unlock(&fp->fi_lock);
 633         return co;
 634 }
 635
 636 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 637                                   void (*sc_free)(struct nfs4_stid *))
 638 {
 639         struct nfs4_stid *stid;
 640         int new_id;
 641
 642         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 643         if (!stid)
 644                 return NULL;
 645
 646         idr_preload(GFP_KERNEL);
 647         spin_lock(&cl->cl_lock);
 648         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 649         spin_unlock(&cl->cl_lock);
 650         idr_preload_end();
 651         if (new_id < 0)
 652                 goto out_free;
 653
 654         stid->sc_free = sc_free;
 655         stid->sc_client = cl;
 656         stid->sc_stateid.si_opaque.so_id = new_id;
 657         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 658         /* Will be incremented before return to client: */
 659         refcount_set(&stid->sc_count, 1);
 660         spin_lock_init(&stid->sc_lock);
 661
 662         /*
 663          * It shouldn't be a problem to reuse an opaque stateid value.
 664          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 665          * example, a stray write retransmission could be accepted by
 666          * the server when it should have been rejected.  Therefore,
 667          * adopt a trick from the sctp code to attempt to maximize the
 668          * amount of time until an id is reused, by ensuring they always
 669          * "increase" (mod INT_MAX):
 670          */
 671         return stid;
 672 out_free:
 673         kmem_cache_free(slab, stid);
 674         return NULL;
 675 }
 676
 677 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 678 {
 679         struct nfs4_stid *stid;
 680
 681         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 682         if (!stid)
 683                 return NULL;
 684
 685         return openlockstateid(stid);
 686 }
 687
 688 static void nfs4_free_deleg(struct nfs4_stid *stid)
 689 {
 690         kmem_cache_free(deleg_slab, stid);
 691         atomic_long_dec(&num_delegations);
 692 }
 693
 694 /*
 695  * When we recall a delegation, we should be careful not to hand it
 696  * out again straight away.
 697  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 698  * in which the filehandles of recalled delegations are "stored".
 699  * If a filehandle appear in either filter, a delegation is blocked.
 700  * When a delegation is recalled, the filehandle is stored in the "new"
 701  * filter.
 702  * Every 30 seconds we swap the filters and clear the "new" one,
 703  * unless both are empty of course.
 704  *
 705  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 706  * low 3 bytes as hash-table indices.
 707  *
 708  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 709  * is used to manage concurrent access.  Testing does not need the lock
 710  * except when swapping the two filters.
 711  */
 712 static DEFINE_SPINLOCK(blocked_delegations_lock);
 713 static struct bloom_pair {
 714         int     entries, old_entries;
 715         time_t  swap_time;
 716         int     new; /* index into 'set' */
 717         DECLARE_BITMAP(set[2], 256);
 718 } blocked_delegations;
 719
 720 static int delegation_blocked(struct knfsd_fh *fh)
 721 {
 722         u32 hash;
 723         struct bloom_pair *bd = &blocked_delegations;
 724
 725         if (bd->entries == 0)
 726                 return 0;
 727         if (seconds_since_boot() - bd->swap_time > 30) {
 728                 spin_lock(&blocked_delegations_lock);
 729                 if (seconds_since_boot() - bd->swap_time > 30) {
 730                         bd->entries -= bd->old_entries;
 731                         bd->old_entries = bd->entries;
 732                         memset(bd->set[bd->new], 0,
 733                                sizeof(bd->set[0]));
 734                         bd->new = 1-bd->new;
 735                         bd->swap_time = seconds_since_boot();
 736                 }
 737                 spin_unlock(&blocked_delegations_lock);
 738         }
 739         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 740         if (test_bit(hash&255, bd->set[0]) &&
 741             test_bit((hash>>8)&255, bd->set[0]) &&
 742             test_bit((hash>>16)&255, bd->set[0]))
 743                 return 1;
 744
 745         if (test_bit(hash&255, bd->set[1]) &&
 746             test_bit((hash>>8)&255, bd->set[1]) &&
 747             test_bit((hash>>16)&255, bd->set[1]))
 748                 return 1;
 749
 750         return 0;
 751 }
 752
 753 static void block_delegations(struct knfsd_fh *fh)
 754 {
 755         u32 hash;
 756         struct bloom_pair *bd = &blocked_delegations;
 757
 758         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 759
 760         spin_lock(&blocked_delegations_lock);
 761         __set_bit(hash&255, bd->set[bd->new]);
 762         __set_bit((hash>>8)&255, bd->set[bd->new]);
 763         __set_bit((hash>>16)&255, bd->set[bd->new]);
 764         if (bd->entries == 0)
 765                 bd->swap_time = seconds_since_boot();
 766         bd->entries += 1;
 767         spin_unlock(&blocked_delegations_lock);
 768 }
 769
 770 static struct nfs4_delegation *
 771 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
 772                  struct nfs4_clnt_odstate *odstate)
 773 {
 774         struct nfs4_delegation *dp;
 775         long n;
 776
 777         dprintk("NFSD alloc_init_deleg\n");
 778         n = atomic_long_inc_return(&num_delegations);
 779         if (n < 0 || n > max_delegations)
 780                 goto out_dec;
 781         if (delegation_blocked(&current_fh->fh_handle))
 782                 goto out_dec;
 783         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 784         if (dp == NULL)
 785                 goto out_dec;
 786
 787         /*
 788          * delegation seqid's are never incremented.  The 4.1 special
 789          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 790          * 0 anyway just for consistency and use 1:
 791          */
 792         dp->dl_stid.sc_stateid.si_generation = 1;
 793         INIT_LIST_HEAD(&dp->dl_perfile);
 794         INIT_LIST_HEAD(&dp->dl_perclnt);
 795         INIT_LIST_HEAD(&dp->dl_recall_lru);
 796         dp->dl_clnt_odstate = odstate;
 797         get_clnt_odstate(odstate);
 798         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 799         dp->dl_retries = 1;
 800         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 801                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 802         return dp;
 803 out_dec:
 804         atomic_long_dec(&num_delegations);
 805         return NULL;
 806 }
 807
 808 void
 809 nfs4_put_stid(struct nfs4_stid *s)
 810 {
 811         struct nfs4_file *fp = s->sc_file;
 812         struct nfs4_client *clp = s->sc_client;
 813
 814         might_lock(&clp->cl_lock);
 815
 816         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 817                 wake_up_all(&close_wq);
 818                 return;
 819         }
 820         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 821         spin_unlock(&clp->cl_lock);
 822         s->sc_free(s);
 823         if (fp)
 824                 put_nfs4_file(fp);
 825 }
 826
 827 void
 828 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 829 {
 830         stateid_t *src = &stid->sc_stateid;
 831
 832         spin_lock(&stid->sc_lock);
 833         if (unlikely(++src->si_generation == 0))
 834                 src->si_generation = 1;
 835         memcpy(dst, src, sizeof(*dst));
 836         spin_unlock(&stid->sc_lock);
 837 }
 838
 839 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 840 {
 841         struct file *filp = NULL;
 842
 843         spin_lock(&fp->fi_lock);
 844         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 845                 swap(filp, fp->fi_deleg_file);
 846         spin_unlock(&fp->fi_lock);
 847
 848         if (filp) {
 849                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 850                 fput(filp);
 851         }
 852 }
 853
 854 void nfs4_unhash_stid(struct nfs4_stid *s)
 855 {
 856         s->sc_type = 0;
 857 }
 858
 859 /**
 860  * nfs4_get_existing_delegation - Discover if this delegation already exists
 861  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 862  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 863  *
 864  * Return:
 865  *      On success: NULL if an existing delegation was not found.
 866  *
 867  *      On error: -EAGAIN if one was previously granted to this nfs4_client
 868  *                 for this nfs4_file.
 869  *
 870  */
 871
 872 static int
 873 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
 874 {
 875         struct nfs4_delegation *searchdp = NULL;
 876         struct nfs4_client *searchclp = NULL;
 877
 878         lockdep_assert_held(&state_lock);
 879         lockdep_assert_held(&fp->fi_lock);
 880
 881         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 882                 searchclp = searchdp->dl_stid.sc_client;
 883                 if (clp == searchclp) {
 884                         return -EAGAIN;
 885                 }
 886         }
 887         return 0;
 888 }
 889
 890 /**
 891  * hash_delegation_locked - Add a delegation to the appropriate lists
 892  * @dp:     a pointer to the nfs4_delegation we are adding.
 893  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 894  *
 895  * Return:
 896  *      On success: NULL if the delegation was successfully hashed.
 897  *
 898  *      On error: -EAGAIN if one was previously granted to this
 899  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 900  *
 901  */
 902
 903 static int
 904 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 905 {
 906         int status;
 907         struct nfs4_client *clp = dp->dl_stid.sc_client;
 908
 909         lockdep_assert_held(&state_lock);
 910         lockdep_assert_held(&fp->fi_lock);
 911
 912         status = nfs4_get_existing_delegation(clp, fp);
 913         if (status)
 914                 return status;
 915         ++fp->fi_delegees;
 916         refcount_inc(&dp->dl_stid.sc_count);
 917         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 918         list_add(&dp->dl_perfile, &fp->fi_delegations);
 919         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 920         return 0;
 921 }
 922
 923 static bool
 924 unhash_delegation_locked(struct nfs4_delegation *dp)
 925 {
 926         struct nfs4_file *fp = dp->dl_stid.sc_file;
 927
 928         lockdep_assert_held(&state_lock);
 929
 930         if (list_empty(&dp->dl_perfile))
 931                 return false;
 932
 933         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 934         /* Ensure that deleg break won't try to requeue it */
 935         ++dp->dl_time;
 936         spin_lock(&fp->fi_lock);
 937         list_del_init(&dp->dl_perclnt);
 938         list_del_init(&dp->dl_recall_lru);
 939         list_del_init(&dp->dl_perfile);
 940         spin_unlock(&fp->fi_lock);
 941         return true;
 942 }
 943
 944 static void destroy_delegation(struct nfs4_delegation *dp)
 945 {
 946         bool unhashed;
 947
 948         spin_lock(&state_lock);
 949         unhashed = unhash_delegation_locked(dp);
 950         spin_unlock(&state_lock);
 951         if (unhashed) {
 952                 put_clnt_odstate(dp->dl_clnt_odstate);
 953                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 954                 nfs4_put_stid(&dp->dl_stid);
 955         }
 956 }
 957
 958 static void revoke_delegation(struct nfs4_delegation *dp)
 959 {
 960         struct nfs4_client *clp = dp->dl_stid.sc_client;
 961
 962         WARN_ON(!list_empty(&dp->dl_recall_lru));
 963
 964         put_clnt_odstate(dp->dl_clnt_odstate);
 965         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 966
 967         if (clp->cl_minorversion == 0)
 968                 nfs4_put_stid(&dp->dl_stid);
 969         else {
 970                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 971                 spin_lock(&clp->cl_lock);
 972                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 973                 spin_unlock(&clp->cl_lock);
 974         }
 975 }
 976
 977 /*
 978  * SETCLIENTID state
 979  */
 980
 981 static unsigned int clientid_hashval(u32 id)
 982 {
 983         return id & CLIENT_HASH_MASK;
 984 }
 985
 986 static unsigned int clientstr_hashval(const char *name)
 987 {
 988         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 989 }
 990
 991 /*
 992  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 993  * st_{access,deny}_bmap field of the stateid, in order to track not
 994  * only what share bits are currently in force, but also what
 995  * combinations of share bits previous opens have used.  This allows us
 996  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 997  * return an error if the client attempt to downgrade to a combination
 998  * of share bits not explicable by closing some of its previous opens.
 999  *
1000  * XXX: This enforcement is actually incomplete, since we don't keep
1001  * track of access/deny bit combinations; so, e.g., we allow:
1002  *
1003  *      OPEN allow read, deny write
1004  *      OPEN allow both, deny none
1005  *      DOWNGRADE allow read, deny none
1006  *
1007  * which we should reject.
1008  */
1009 static unsigned int
1010 bmap_to_share_mode(unsigned long bmap) {
1011         int i;
1012         unsigned int access = 0;
1013
1014         for (i = 1; i < 4; i++) {
1015                 if (test_bit(i, &bmap))
1016                         access |= i;
1017         }
1018         return access;
1019 }
1020
1021 /* set share access for a given stateid */
1022 static inline void
1023 set_access(u32 access, struct nfs4_ol_stateid *stp)
1024 {
1025         unsigned char mask = 1 << access;
1026
1027         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1028         stp->st_access_bmap |= mask;
1029 }
1030
1031 /* clear share access for a given stateid */
1032 static inline void
1033 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1034 {
1035         unsigned char mask = 1 << access;
1036
1037         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1038         stp->st_access_bmap &= ~mask;
1039 }
1040
1041 /* test whether a given stateid has access */
1042 static inline bool
1043 test_access(u32 access, struct nfs4_ol_stateid *stp)
1044 {
1045         unsigned char mask = 1 << access;
1046
1047         return (bool)(stp->st_access_bmap & mask);
1048 }
1049
1050 /* set share deny for a given stateid */
1051 static inline void
1052 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1053 {
1054         unsigned char mask = 1 << deny;
1055
1056         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1057         stp->st_deny_bmap |= mask;
1058 }
1059
1060 /* clear share deny for a given stateid */
1061 static inline void
1062 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1063 {
1064         unsigned char mask = 1 << deny;
1065
1066         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1067         stp->st_deny_bmap &= ~mask;
1068 }
1069
1070 /* test whether a given stateid is denying specific access */
1071 static inline bool
1072 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1073 {
1074         unsigned char mask = 1 << deny;
1075
1076         return (bool)(stp->st_deny_bmap & mask);
1077 }
1078
1079 static int nfs4_access_to_omode(u32 access)
1080 {
1081         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1082         case NFS4_SHARE_ACCESS_READ:
1083                 return O_RDONLY;
1084         case NFS4_SHARE_ACCESS_WRITE:
1085                 return O_WRONLY;
1086         case NFS4_SHARE_ACCESS_BOTH:
1087                 return O_RDWR;
1088         }
1089         WARN_ON_ONCE(1);
1090         return O_RDONLY;
1091 }
1092
1093 /*
1094  * A stateid that had a deny mode associated with it is being released
1095  * or downgraded. Recalculate the deny mode on the file.
1096  */
1097 static void
1098 recalculate_deny_mode(struct nfs4_file *fp)
1099 {
1100         struct nfs4_ol_stateid *stp;
1101
1102         spin_lock(&fp->fi_lock);
1103         fp->fi_share_deny = 0;
1104         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1105                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1106         spin_unlock(&fp->fi_lock);
1107 }
1108
1109 static void
1110 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1111 {
1112         int i;
1113         bool change = false;
1114
1115         for (i = 1; i < 4; i++) {
1116                 if ((i & deny) != i) {
1117                         change = true;
1118                         clear_deny(i, stp);
1119                 }
1120         }
1121
1122         /* Recalculate per-file deny mode if there was a change */
1123         if (change)
1124                 recalculate_deny_mode(stp->st_stid.sc_file);
1125 }
1126
1127 /* release all access and file references for a given stateid */
1128 static void
1129 release_all_access(struct nfs4_ol_stateid *stp)
1130 {
1131         int i;
1132         struct nfs4_file *fp = stp->st_stid.sc_file;
1133
1134         if (fp && stp->st_deny_bmap != 0)
1135                 recalculate_deny_mode(fp);
1136
1137         for (i = 1; i < 4; i++) {
1138                 if (test_access(i, stp))
1139                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1140                 clear_access(i, stp);
1141         }
1142 }
1143
1144 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1145 {
1146         kfree(sop->so_owner.data);
1147         sop->so_ops->so_free(sop);
1148 }
1149
1150 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1151 {
1152         struct nfs4_client *clp = sop->so_client;
1153
1154         might_lock(&clp->cl_lock);
1155
1156         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1157                 return;
1158         sop->so_ops->so_unhash(sop);
1159         spin_unlock(&clp->cl_lock);
1160         nfs4_free_stateowner(sop);
1161 }
1162
1163 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1164 {
1165         struct nfs4_file *fp = stp->st_stid.sc_file;
1166
1167         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1168
1169         if (list_empty(&stp->st_perfile))
1170                 return false;
1171
1172         spin_lock(&fp->fi_lock);
1173         list_del_init(&stp->st_perfile);
1174         spin_unlock(&fp->fi_lock);
1175         list_del(&stp->st_perstateowner);
1176         return true;
1177 }
1178
1179 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1180 {
1181         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1182
1183         put_clnt_odstate(stp->st_clnt_odstate);
1184         release_all_access(stp);
1185         if (stp->st_stateowner)
1186                 nfs4_put_stateowner(stp->st_stateowner);
1187         kmem_cache_free(stateid_slab, stid);
1188 }
1189
1190 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1191 {
1192         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1193         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1194         struct file *file;
1195
1196         file = find_any_file(stp->st_stid.sc_file);
1197         if (file)
1198                 filp_close(file, (fl_owner_t)lo);
1199         nfs4_free_ol_stateid(stid);
1200 }
1201
1202 /*
1203  * Put the persistent reference to an already unhashed generic stateid, while
1204  * holding the cl_lock. If it's the last reference, then put it onto the
1205  * reaplist for later destruction.
1206  */
1207 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1208                                        struct list_head *reaplist)
1209 {
1210         struct nfs4_stid *s = &stp->st_stid;
1211         struct nfs4_client *clp = s->sc_client;
1212
1213         lockdep_assert_held(&clp->cl_lock);
1214
1215         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1216
1217         if (!refcount_dec_and_test(&s->sc_count)) {
1218                 wake_up_all(&close_wq);
1219                 return;
1220         }
1221
1222         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1223         list_add(&stp->st_locks, reaplist);
1224 }
1225
1226 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1227 {
1228         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1229
1230         list_del_init(&stp->st_locks);
1231         nfs4_unhash_stid(&stp->st_stid);
1232         return unhash_ol_stateid(stp);
1233 }
1234
1235 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1236 {
1237         struct nfs4_client *clp = stp->st_stid.sc_client;
1238         bool unhashed;
1239
1240         spin_lock(&clp->cl_lock);
1241         unhashed = unhash_lock_stateid(stp);
1242         spin_unlock(&clp->cl_lock);
1243         if (unhashed)
1244                 nfs4_put_stid(&stp->st_stid);
1245 }
1246
1247 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1248 {
1249         struct nfs4_client *clp = lo->lo_owner.so_client;
1250
1251         lockdep_assert_held(&clp->cl_lock);
1252
1253         list_del_init(&lo->lo_owner.so_strhash);
1254 }
1255
1256 /*
1257  * Free a list of generic stateids that were collected earlier after being
1258  * fully unhashed.
1259  */
1260 static void
1261 free_ol_stateid_reaplist(struct list_head *reaplist)
1262 {
1263         struct nfs4_ol_stateid *stp;
1264         struct nfs4_file *fp;
1265
1266         might_sleep();
1267
1268         while (!list_empty(reaplist)) {
1269                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1270                                        st_locks);
1271                 list_del(&stp->st_locks);
1272                 fp = stp->st_stid.sc_file;
1273                 stp->st_stid.sc_free(&stp->st_stid);
1274                 if (fp)
1275                         put_nfs4_file(fp);
1276         }
1277 }
1278
1279 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1280                                        struct list_head *reaplist)
1281 {
1282         struct nfs4_ol_stateid *stp;
1283
1284         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1285
1286         while (!list_empty(&open_stp->st_locks)) {
1287                 stp = list_entry(open_stp->st_locks.next,
1288                                 struct nfs4_ol_stateid, st_locks);
1289                 WARN_ON(!unhash_lock_stateid(stp));
1290                 put_ol_stateid_locked(stp, reaplist);
1291         }
1292 }
1293
1294 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1295                                 struct list_head *reaplist)
1296 {
1297         bool unhashed;
1298
1299         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1300
1301         unhashed = unhash_ol_stateid(stp);
1302         release_open_stateid_locks(stp, reaplist);
1303         return unhashed;
1304 }
1305
1306 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1307 {
1308         LIST_HEAD(reaplist);
1309
1310         spin_lock(&stp->st_stid.sc_client->cl_lock);
1311         if (unhash_open_stateid(stp, &reaplist))
1312                 put_ol_stateid_locked(stp, &reaplist);
1313         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1314         free_ol_stateid_reaplist(&reaplist);
1315 }
1316
1317 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1318 {
1319         struct nfs4_client *clp = oo->oo_owner.so_client;
1320
1321         lockdep_assert_held(&clp->cl_lock);
1322
1323         list_del_init(&oo->oo_owner.so_strhash);
1324         list_del_init(&oo->oo_perclient);
1325 }
1326
1327 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1328 {
1329         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1330                                           nfsd_net_id);
1331         struct nfs4_ol_stateid *s;
1332
1333         spin_lock(&nn->client_lock);
1334         s = oo->oo_last_closed_stid;
1335         if (s) {
1336                 list_del_init(&oo->oo_close_lru);
1337                 oo->oo_last_closed_stid = NULL;
1338         }
1339         spin_unlock(&nn->client_lock);
1340         if (s)
1341                 nfs4_put_stid(&s->st_stid);
1342 }
1343
1344 static void release_openowner(struct nfs4_openowner *oo)
1345 {
1346         struct nfs4_ol_stateid *stp;
1347         struct nfs4_client *clp = oo->oo_owner.so_client;
1348         struct list_head reaplist;
1349
1350         INIT_LIST_HEAD(&reaplist);
1351
1352         spin_lock(&clp->cl_lock);
1353         unhash_openowner_locked(oo);
1354         while (!list_empty(&oo->oo_owner.so_stateids)) {
1355                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1356                                 struct nfs4_ol_stateid, st_perstateowner);
1357                 if (unhash_open_stateid(stp, &reaplist))
1358                         put_ol_stateid_locked(stp, &reaplist);
1359         }
1360         spin_unlock(&clp->cl_lock);
1361         free_ol_stateid_reaplist(&reaplist);
1362         release_last_closed_stateid(oo);
1363         nfs4_put_stateowner(&oo->oo_owner);
1364 }
1365
1366 static inline int
1367 hash_sessionid(struct nfs4_sessionid *sessionid)
1368 {
1369         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1370
1371         return sid->sequence % SESSION_HASH_SIZE;
1372 }
1373
1374 #ifdef CONFIG_SUNRPC_DEBUG
1375 static inline void
1376 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1377 {
1378         u32 *ptr = (u32 *)(&sessionid->data[0]);
1379         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1380 }
1381 #else
1382 static inline void
1383 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1384 {
1385 }
1386 #endif
1387
1388 /*
1389  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1390  * won't be used for replay.
1391  */
1392 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1393 {
1394         struct nfs4_stateowner *so = cstate->replay_owner;
1395
1396         if (nfserr == nfserr_replay_me)
1397                 return;
1398
1399         if (!seqid_mutating_err(ntohl(nfserr))) {
1400                 nfsd4_cstate_clear_replay(cstate);
1401                 return;
1402         }
1403         if (!so)
1404                 return;
1405         if (so->so_is_open_owner)
1406                 release_last_closed_stateid(openowner(so));
1407         so->so_seqid++;
1408         return;
1409 }
1410
1411 static void
1412 gen_sessionid(struct nfsd4_session *ses)
1413 {
1414         struct nfs4_client *clp = ses->se_client;
1415         struct nfsd4_sessionid *sid;
1416
1417         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1418         sid->clientid = clp->cl_clientid;
1419         sid->sequence = current_sessionid++;
1420         sid->reserved = 0;
1421 }
1422
1423 /*
1424  * The protocol defines ca_maxresponssize_cached to include the size of
1425  * the rpc header, but all we need to cache is the data starting after
1426  * the end of the initial SEQUENCE operation--the rest we regenerate
1427  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1428  * value that is the number of bytes in our cache plus a few additional
1429  * bytes.  In order to stay on the safe side, and not promise more than
1430  * we can cache, those additional bytes must be the minimum possible: 24
1431  * bytes of rpc header (xid through accept state, with AUTH_NULL
1432  * verifier), 12 for the compound header (with zero-length tag), and 44
1433  * for the SEQUENCE op response:
1434  */
1435 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1436
1437 static void
1438 free_session_slots(struct nfsd4_session *ses)
1439 {
1440         int i;
1441
1442         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1443                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1444                 kfree(ses->se_slots[i]);
1445         }
1446 }
1447
1448 /*
1449  * We don't actually need to cache the rpc and session headers, so we
1450  * can allocate a little less for each slot:
1451  */
1452 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1453 {
1454         u32 size;
1455
1456         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1457                 size = 0;
1458         else
1459                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1460         return size + sizeof(struct nfsd4_slot);
1461 }
1462
1463 /*
1464  * XXX: If we run out of reserved DRC memory we could (up to a point)
1465  * re-negotiate active sessions and reduce their slot usage to make
1466  * room for new connections. For now we just fail the create session.
1467  */
1468 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1469 {
1470         u32 slotsize = slot_bytes(ca);
1471         u32 num = ca->maxreqs;
1472         int avail;
1473
1474         spin_lock(&nfsd_drc_lock);
1475         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1476                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1477         /*
1478          * Never use more than a third of the remaining memory,
1479          * unless it's the only way to give this client a slot:
1480          */
1481         avail = clamp_t(int, avail, slotsize, avail/3);
1482         num = min_t(int, num, avail / slotsize);
1483         nfsd_drc_mem_used += num * slotsize;
1484         spin_unlock(&nfsd_drc_lock);
1485
1486         return num;
1487 }
1488
1489 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1490 {
1491         int slotsize = slot_bytes(ca);
1492
1493         spin_lock(&nfsd_drc_lock);
1494         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1495         spin_unlock(&nfsd_drc_lock);
1496 }
1497
1498 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1499                                            struct nfsd4_channel_attrs *battrs)
1500 {
1501         int numslots = fattrs->maxreqs;
1502         int slotsize = slot_bytes(fattrs);
1503         struct nfsd4_session *new;
1504         int mem, i;
1505
1506         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1507                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1508         mem = numslots * sizeof(struct nfsd4_slot *);
1509
1510         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1511         if (!new)
1512                 return NULL;
1513         /* allocate each struct nfsd4_slot and data cache in one piece */
1514         for (i = 0; i < numslots; i++) {
1515                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1516                 if (!new->se_slots[i])
1517                         goto out_free;
1518         }
1519
1520         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1521         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1522
1523         return new;
1524 out_free:
1525         while (i--)
1526                 kfree(new->se_slots[i]);
1527         kfree(new);
1528         return NULL;
1529 }
1530
1531 static void free_conn(struct nfsd4_conn *c)
1532 {
1533         svc_xprt_put(c->cn_xprt);
1534         kfree(c);
1535 }
1536
1537 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1538 {
1539         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1540         struct nfs4_client *clp = c->cn_session->se_client;
1541
1542         spin_lock(&clp->cl_lock);
1543         if (!list_empty(&c->cn_persession)) {
1544                 list_del(&c->cn_persession);
1545                 free_conn(c);
1546         }
1547         nfsd4_probe_callback(clp);
1548         spin_unlock(&clp->cl_lock);
1549 }
1550
1551 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1552 {
1553         struct nfsd4_conn *conn;
1554
1555         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1556         if (!conn)
1557                 return NULL;
1558         svc_xprt_get(rqstp->rq_xprt);
1559         conn->cn_xprt = rqstp->rq_xprt;
1560         conn->cn_flags = flags;
1561         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1562         return conn;
1563 }
1564
1565 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1566 {
1567         conn->cn_session = ses;
1568         list_add(&conn->cn_persession, &ses->se_conns);
1569 }
1570
1571 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1572 {
1573         struct nfs4_client *clp = ses->se_client;
1574
1575         spin_lock(&clp->cl_lock);
1576         __nfsd4_hash_conn(conn, ses);
1577         spin_unlock(&clp->cl_lock);
1578 }
1579
1580 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1581 {
1582         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1583         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1584 }
1585
1586 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1587 {
1588         int ret;
1589
1590         nfsd4_hash_conn(conn, ses);
1591         ret = nfsd4_register_conn(conn);
1592         if (ret)
1593                 /* oops; xprt is already down: */
1594                 nfsd4_conn_lost(&conn->cn_xpt_user);
1595         /* We may have gained or lost a callback channel: */
1596         nfsd4_probe_callback_sync(ses->se_client);
1597 }
1598
1599 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1600 {
1601         u32 dir = NFS4_CDFC4_FORE;
1602
1603         if (cses->flags & SESSION4_BACK_CHAN)
1604                 dir |= NFS4_CDFC4_BACK;
1605         return alloc_conn(rqstp, dir);
1606 }
1607
1608 /* must be called under client_lock */
1609 static void nfsd4_del_conns(struct nfsd4_session *s)
1610 {
1611         struct nfs4_client *clp = s->se_client;
1612         struct nfsd4_conn *c;
1613
1614         spin_lock(&clp->cl_lock);
1615         while (!list_empty(&s->se_conns)) {
1616                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1617                 list_del_init(&c->cn_persession);
1618                 spin_unlock(&clp->cl_lock);
1619
1620                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1621                 free_conn(c);
1622
1623                 spin_lock(&clp->cl_lock);
1624         }
1625         spin_unlock(&clp->cl_lock);
1626 }
1627
1628 static void __free_session(struct nfsd4_session *ses)
1629 {
1630         free_session_slots(ses);
1631         kfree(ses);
1632 }
1633
1634 static void free_session(struct nfsd4_session *ses)
1635 {
1636         nfsd4_del_conns(ses);
1637         nfsd4_put_drc_mem(&ses->se_fchannel);
1638         __free_session(ses);
1639 }
1640
1641 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1642 {
1643         int idx;
1644         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1645
1646         new->se_client = clp;
1647         gen_sessionid(new);
1648
1649         INIT_LIST_HEAD(&new->se_conns);
1650
1651         new->se_cb_seq_nr = 1;
1652         new->se_flags = cses->flags;
1653         new->se_cb_prog = cses->callback_prog;
1654         new->se_cb_sec = cses->cb_sec;
1655         atomic_set(&new->se_ref, 0);
1656         idx = hash_sessionid(&new->se_sessionid);
1657         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1658         spin_lock(&clp->cl_lock);
1659         list_add(&new->se_perclnt, &clp->cl_sessions);
1660         spin_unlock(&clp->cl_lock);
1661
1662         {
1663                 struct sockaddr *sa = svc_addr(rqstp);
1664                 /*
1665                  * This is a little silly; with sessions there's no real
1666                  * use for the callback address.  Use the peer address
1667                  * as a reasonable default for now, but consider fixing
1668                  * the rpc client not to require an address in the
1669                  * future:
1670                  */
1671                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1672                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1673         }
1674 }
1675
1676 /* caller must hold client_lock */
1677 static struct nfsd4_session *
1678 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1679 {
1680         struct nfsd4_session *elem;
1681         int idx;
1682         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1683
1684         lockdep_assert_held(&nn->client_lock);
1685
1686         dump_sessionid(__func__, sessionid);
1687         idx = hash_sessionid(sessionid);
1688         /* Search in the appropriate list */
1689         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1690                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1691                             NFS4_MAX_SESSIONID_LEN)) {
1692                         return elem;
1693                 }
1694         }
1695
1696         dprintk("%s: session not found\n", __func__);
1697         return NULL;
1698 }
1699
1700 static struct nfsd4_session *
1701 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1702                 __be32 *ret)
1703 {
1704         struct nfsd4_session *session;
1705         __be32 status = nfserr_badsession;
1706
1707         session = __find_in_sessionid_hashtbl(sessionid, net);
1708         if (!session)
1709                 goto out;
1710         status = nfsd4_get_session_locked(session);
1711         if (status)
1712                 session = NULL;
1713 out:
1714         *ret = status;
1715         return session;
1716 }
1717
1718 /* caller must hold client_lock */
1719 static void
1720 unhash_session(struct nfsd4_session *ses)
1721 {
1722         struct nfs4_client *clp = ses->se_client;
1723         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1724
1725         lockdep_assert_held(&nn->client_lock);
1726
1727         list_del(&ses->se_hash);
1728         spin_lock(&ses->se_client->cl_lock);
1729         list_del(&ses->se_perclnt);
1730         spin_unlock(&ses->se_client->cl_lock);
1731 }
1732
1733 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1734 static int
1735 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1736 {
1737         /*
1738          * We're assuming the clid was not given out from a boot
1739          * precisely 2^32 (about 136 years) before this one.  That seems
1740          * a safe assumption:
1741          */
1742         if (clid->cl_boot == (u32)nn->boot_time)
1743                 return 0;
1744         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1745                 clid->cl_boot, clid->cl_id, nn->boot_time);
1746         return 1;
1747 }
1748
1749 /*
1750  * XXX Should we use a slab cache ?
1751  * This type of memory management is somewhat inefficient, but we use it
1752  * anyway since SETCLIENTID is not a common operation.
1753  */
1754 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1755 {
1756         struct nfs4_client *clp;
1757         int i;
1758
1759         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1760         if (clp == NULL)
1761                 return NULL;
1762         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1763         if (clp->cl_name.data == NULL)
1764                 goto err_no_name;
1765         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1766                         OWNER_HASH_SIZE, GFP_KERNEL);
1767         if (!clp->cl_ownerstr_hashtbl)
1768                 goto err_no_hashtbl;
1769         for (i = 0; i < OWNER_HASH_SIZE; i++)
1770                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1771         clp->cl_name.len = name.len;
1772         INIT_LIST_HEAD(&clp->cl_sessions);
1773         idr_init(&clp->cl_stateids);
1774         atomic_set(&clp->cl_refcount, 0);
1775         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1776         INIT_LIST_HEAD(&clp->cl_idhash);
1777         INIT_LIST_HEAD(&clp->cl_openowners);
1778         INIT_LIST_HEAD(&clp->cl_delegations);
1779         INIT_LIST_HEAD(&clp->cl_lru);
1780         INIT_LIST_HEAD(&clp->cl_revoked);
1781 #ifdef CONFIG_NFSD_PNFS
1782         INIT_LIST_HEAD(&clp->cl_lo_states);
1783 #endif
1784         spin_lock_init(&clp->cl_lock);
1785         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1786         return clp;
1787 err_no_hashtbl:
1788         kfree(clp->cl_name.data);
1789 err_no_name:
1790         kfree(clp);
1791         return NULL;
1792 }
1793
1794 static void
1795 free_client(struct nfs4_client *clp)
1796 {
1797         while (!list_empty(&clp->cl_sessions)) {
1798                 struct nfsd4_session *ses;
1799                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1800                                 se_perclnt);
1801                 list_del(&ses->se_perclnt);
1802                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1803                 free_session(ses);
1804         }
1805         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1806         free_svc_cred(&clp->cl_cred);
1807         kfree(clp->cl_ownerstr_hashtbl);
1808         kfree(clp->cl_name.data);
1809         idr_destroy(&clp->cl_stateids);
1810         kfree(clp);
1811 }
1812
1813 /* must be called under the client_lock */
1814 static void
1815 unhash_client_locked(struct nfs4_client *clp)
1816 {
1817         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1818         struct nfsd4_session *ses;
1819
1820         lockdep_assert_held(&nn->client_lock);
1821
1822         /* Mark the client as expired! */
1823         clp->cl_time = 0;
1824         /* Make it invisible */
1825         if (!list_empty(&clp->cl_idhash)) {
1826                 list_del_init(&clp->cl_idhash);
1827                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1828                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1829                 else
1830                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1831         }
1832         list_del_init(&clp->cl_lru);
1833         spin_lock(&clp->cl_lock);
1834         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1835                 list_del_init(&ses->se_hash);
1836         spin_unlock(&clp->cl_lock);
1837 }
1838
1839 static void
1840 unhash_client(struct nfs4_client *clp)
1841 {
1842         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1843
1844         spin_lock(&nn->client_lock);
1845         unhash_client_locked(clp);
1846         spin_unlock(&nn->client_lock);
1847 }
1848
1849 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1850 {
1851         if (atomic_read(&clp->cl_refcount))
1852                 return nfserr_jukebox;
1853         unhash_client_locked(clp);
1854         return nfs_ok;
1855 }
1856
1857 static void
1858 __destroy_client(struct nfs4_client *clp)
1859 {
1860         struct nfs4_openowner *oo;
1861         struct nfs4_delegation *dp;
1862         struct list_head reaplist;
1863
1864         INIT_LIST_HEAD(&reaplist);
1865         spin_lock(&state_lock);
1866         while (!list_empty(&clp->cl_delegations)) {
1867                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1868                 WARN_ON(!unhash_delegation_locked(dp));
1869                 list_add(&dp->dl_recall_lru, &reaplist);
1870         }
1871         spin_unlock(&state_lock);
1872         while (!list_empty(&reaplist)) {
1873                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1874                 list_del_init(&dp->dl_recall_lru);
1875                 put_clnt_odstate(dp->dl_clnt_odstate);
1876                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1877                 nfs4_put_stid(&dp->dl_stid);
1878         }
1879         while (!list_empty(&clp->cl_revoked)) {
1880                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1881                 list_del_init(&dp->dl_recall_lru);
1882                 nfs4_put_stid(&dp->dl_stid);
1883         }
1884         while (!list_empty(&clp->cl_openowners)) {
1885                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1886                 nfs4_get_stateowner(&oo->oo_owner);
1887                 release_openowner(oo);
1888         }
1889         nfsd4_return_all_client_layouts(clp);
1890         nfsd4_shutdown_callback(clp);
1891         if (clp->cl_cb_conn.cb_xprt)
1892                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1893         free_client(clp);
1894 }
1895
1896 static void
1897 destroy_client(struct nfs4_client *clp)
1898 {
1899         unhash_client(clp);
1900         __destroy_client(clp);
1901 }
1902
1903 static void expire_client(struct nfs4_client *clp)
1904 {
1905         unhash_client(clp);
1906         nfsd4_client_record_remove(clp);
1907         __destroy_client(clp);
1908 }
1909
1910 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1911 {
1912         memcpy(target->cl_verifier.data, source->data,
1913                         sizeof(target->cl_verifier.data));
1914 }
1915
1916 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1917 {
1918         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1919         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1920 }
1921
1922 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1923 {
1924         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1925         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1926                                                                 GFP_KERNEL);
1927         if ((source->cr_principal && ! target->cr_principal) ||
1928             (source->cr_raw_principal && ! target->cr_raw_principal))
1929                 return -ENOMEM;
1930
1931         target->cr_flavor = source->cr_flavor;
1932         target->cr_uid = source->cr_uid;
1933         target->cr_gid = source->cr_gid;
1934         target->cr_group_info = source->cr_group_info;
1935         get_group_info(target->cr_group_info);
1936         target->cr_gss_mech = source->cr_gss_mech;
1937         if (source->cr_gss_mech)
1938                 gss_mech_get(source->cr_gss_mech);
1939         return 0;
1940 }
1941
1942 static int
1943 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1944 {
1945         if (o1->len < o2->len)
1946                 return -1;
1947         if (o1->len > o2->len)
1948                 return 1;
1949         return memcmp(o1->data, o2->data, o1->len);
1950 }
1951
1952 static int same_name(const char *n1, const char *n2)
1953 {
1954         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1955 }
1956
1957 static int
1958 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1959 {
1960         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1961 }
1962
1963 static int
1964 same_clid(clientid_t *cl1, clientid_t *cl2)
1965 {
1966         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1967 }
1968
1969 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1970 {
1971         int i;
1972
1973         if (g1->ngroups != g2->ngroups)
1974                 return false;
1975         for (i=0; i<g1->ngroups; i++)
1976                 if (!gid_eq(g1->gid[i], g2->gid[i]))
1977                         return false;
1978         return true;
1979 }
1980
1981 /*
1982  * RFC 3530 language requires clid_inuse be returned when the
1983  * "principal" associated with a requests differs from that previously
1984  * used.  We use uid, gid's, and gss principal string as our best
1985  * approximation.  We also don't want to allow non-gss use of a client
1986  * established using gss: in theory cr_principal should catch that
1987  * change, but in practice cr_principal can be null even in the gss case
1988  * since gssd doesn't always pass down a principal string.
1989  */
1990 static bool is_gss_cred(struct svc_cred *cr)
1991 {
1992         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1993         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1994 }
1995
1996
1997 static bool
1998 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1999 {
2000         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2001                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2002                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2003                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2004                 return false;
2005         if (cr1->cr_principal == cr2->cr_principal)
2006                 return true;
2007         if (!cr1->cr_principal || !cr2->cr_principal)
2008                 return false;
2009         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2010 }
2011
2012 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2013 {
2014         struct svc_cred *cr = &rqstp->rq_cred;
2015         u32 service;
2016
2017         if (!cr->cr_gss_mech)
2018                 return false;
2019         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2020         return service == RPC_GSS_SVC_INTEGRITY ||
2021                service == RPC_GSS_SVC_PRIVACY;
2022 }
2023
2024 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2025 {
2026         struct svc_cred *cr = &rqstp->rq_cred;
2027
2028         if (!cl->cl_mach_cred)
2029                 return true;
2030         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2031                 return false;
2032         if (!svc_rqst_integrity_protected(rqstp))
2033                 return false;
2034         if (cl->cl_cred.cr_raw_principal)
2035                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2036                                                 cr->cr_raw_principal);
2037         if (!cr->cr_principal)
2038                 return false;
2039         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2040 }
2041
2042 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2043 {
2044         __be32 verf[2];
2045
2046         /*
2047          * This is opaque to client, so no need to byte-swap. Use
2048          * __force to keep sparse happy
2049          */
2050         verf[0] = (__force __be32)get_seconds();
2051         verf[1] = (__force __be32)nn->clverifier_counter++;
2052         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2053 }
2054
2055 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2056 {
2057         clp->cl_clientid.cl_boot = nn->boot_time;
2058         clp->cl_clientid.cl_id = nn->clientid_counter++;
2059         gen_confirm(clp, nn);
2060 }
2061
2062 static struct nfs4_stid *
2063 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2064 {
2065         struct nfs4_stid *ret;
2066
2067         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2068         if (!ret || !ret->sc_type)
2069                 return NULL;
2070         return ret;
2071 }
2072
2073 static struct nfs4_stid *
2074 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2075 {
2076         struct nfs4_stid *s;
2077
2078         spin_lock(&cl->cl_lock);
2079         s = find_stateid_locked(cl, t);
2080         if (s != NULL) {
2081                 if (typemask & s->sc_type)
2082                         refcount_inc(&s->sc_count);
2083                 else
2084                         s = NULL;
2085         }
2086         spin_unlock(&cl->cl_lock);
2087         return s;
2088 }
2089
2090 static struct nfs4_client *create_client(struct xdr_netobj name,
2091                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2092 {
2093         struct nfs4_client *clp;
2094         struct sockaddr *sa = svc_addr(rqstp);
2095         int ret;
2096         struct net *net = SVC_NET(rqstp);
2097
2098         clp = alloc_client(name);
2099         if (clp == NULL)
2100                 return NULL;
2101
2102         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2103         if (ret) {
2104                 free_client(clp);
2105                 return NULL;
2106         }
2107         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2108         clp->cl_time = get_seconds();
2109         clear_bit(0, &clp->cl_cb_slot_busy);
2110         copy_verf(clp, verf);
2111         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2112         clp->cl_cb_session = NULL;
2113         clp->net = net;
2114         return clp;
2115 }
2116
2117 static void
2118 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2119 {
2120         struct rb_node **new = &(root->rb_node), *parent = NULL;
2121         struct nfs4_client *clp;
2122
2123         while (*new) {
2124                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2125                 parent = *new;
2126
2127                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2128                         new = &((*new)->rb_left);
2129                 else
2130                         new = &((*new)->rb_right);
2131         }
2132
2133         rb_link_node(&new_clp->cl_namenode, parent, new);
2134         rb_insert_color(&new_clp->cl_namenode, root);
2135 }
2136
2137 static struct nfs4_client *
2138 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2139 {
2140         int cmp;
2141         struct rb_node *node = root->rb_node;
2142         struct nfs4_client *clp;
2143
2144         while (node) {
2145                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2146                 cmp = compare_blob(&clp->cl_name, name);
2147                 if (cmp > 0)
2148                         node = node->rb_left;
2149                 else if (cmp < 0)
2150                         node = node->rb_right;
2151                 else
2152                         return clp;
2153         }
2154         return NULL;
2155 }
2156
2157 static void
2158 add_to_unconfirmed(struct nfs4_client *clp)
2159 {
2160         unsigned int idhashval;
2161         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2162
2163         lockdep_assert_held(&nn->client_lock);
2164
2165         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2166         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2167         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2168         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2169         renew_client_locked(clp);
2170 }
2171
2172 static void
2173 move_to_confirmed(struct nfs4_client *clp)
2174 {
2175         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2176         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2177
2178         lockdep_assert_held(&nn->client_lock);
2179
2180         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2181         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2182         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2183         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2184         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2185         renew_client_locked(clp);
2186 }
2187
2188 static struct nfs4_client *
2189 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2190 {
2191         struct nfs4_client *clp;
2192         unsigned int idhashval = clientid_hashval(clid->cl_id);
2193
2194         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2195                 if (same_clid(&clp->cl_clientid, clid)) {
2196                         if ((bool)clp->cl_minorversion != sessions)
2197                                 return NULL;
2198                         renew_client_locked(clp);
2199                         return clp;
2200                 }
2201         }
2202         return NULL;
2203 }
2204
2205 static struct nfs4_client *
2206 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2207 {
2208         struct list_head *tbl = nn->conf_id_hashtbl;
2209
2210         lockdep_assert_held(&nn->client_lock);
2211         return find_client_in_id_table(tbl, clid, sessions);
2212 }
2213
2214 static struct nfs4_client *
2215 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2216 {
2217         struct list_head *tbl = nn->unconf_id_hashtbl;
2218
2219         lockdep_assert_held(&nn->client_lock);
2220         return find_client_in_id_table(tbl, clid, sessions);
2221 }
2222
2223 static bool clp_used_exchangeid(struct nfs4_client *clp)
2224 {
2225         return clp->cl_exchange_flags != 0;
2226 }
2227
2228 static struct nfs4_client *
2229 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2230 {
2231         lockdep_assert_held(&nn->client_lock);
2232         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2233 }
2234
2235 static struct nfs4_client *
2236 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2237 {
2238         lockdep_assert_held(&nn->client_lock);
2239         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2240 }
2241
2242 static void
2243 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2244 {
2245         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2246         struct sockaddr *sa = svc_addr(rqstp);
2247         u32 scopeid = rpc_get_scope_id(sa);
2248         unsigned short expected_family;
2249
2250         /* Currently, we only support tcp and tcp6 for the callback channel */
2251         if (se->se_callback_netid_len == 3 &&
2252             !memcmp(se->se_callback_netid_val, "tcp", 3))
2253                 expected_family = AF_INET;
2254         else if (se->se_callback_netid_len == 4 &&
2255                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2256                 expected_family = AF_INET6;
2257         else
2258                 goto out_err;
2259
2260         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2261                                             se->se_callback_addr_len,
2262                                             (struct sockaddr *)&conn->cb_addr,
2263                                             sizeof(conn->cb_addr));
2264
2265         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2266                 goto out_err;
2267
2268         if (conn->cb_addr.ss_family == AF_INET6)
2269                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2270
2271         conn->cb_prog = se->se_callback_prog;
2272         conn->cb_ident = se->se_callback_ident;
2273         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2274         return;
2275 out_err:
2276         conn->cb_addr.ss_family = AF_UNSPEC;
2277         conn->cb_addrlen = 0;
2278         dprintk("NFSD: this client (clientid %08x/%08x) "
2279                 "will not receive delegations\n",
2280                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2281
2282         return;
2283 }
2284
2285 /*
2286  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2287  */
2288 static void
2289 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2290 {
2291         struct xdr_buf *buf = resp->xdr.buf;
2292         struct nfsd4_slot *slot = resp->cstate.slot;
2293         unsigned int base;
2294
2295         dprintk("--> %s slot %p\n", __func__, slot);
2296
2297         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2298         slot->sl_opcnt = resp->opcnt;
2299         slot->sl_status = resp->cstate.status;
2300         free_svc_cred(&slot->sl_cred);
2301         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2302
2303         if (!nfsd4_cache_this(resp)) {
2304                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2305                 return;
2306         }
2307         slot->sl_flags |= NFSD4_SLOT_CACHED;
2308
2309         base = resp->cstate.data_offset;
2310         slot->sl_datalen = buf->len - base;
2311         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2312                 WARN(1, "%s: sessions DRC could not cache compound\n",
2313                      __func__);
2314         return;
2315 }
2316
2317 /*
2318  * Encode the replay sequence operation from the slot values.
2319  * If cachethis is FALSE encode the uncached rep error on the next
2320  * operation which sets resp->p and increments resp->opcnt for
2321  * nfs4svc_encode_compoundres.
2322  *
2323  */
2324 static __be32
2325 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2326                           struct nfsd4_compoundres *resp)
2327 {
2328         struct nfsd4_op *op;
2329         struct nfsd4_slot *slot = resp->cstate.slot;
2330
2331         /* Encode the replayed sequence operation */
2332         op = &args->ops[resp->opcnt - 1];
2333         nfsd4_encode_operation(resp, op);
2334
2335         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2336                 return op->status;
2337         if (args->opcnt == 1) {
2338                 /*
2339                  * The original operation wasn't a solo sequence--we
2340                  * always cache those--so this retry must not match the
2341                  * original:
2342                  */
2343                 op->status = nfserr_seq_false_retry;
2344         } else {
2345                 op = &args->ops[resp->opcnt++];
2346                 op->status = nfserr_retry_uncached_rep;
2347                 nfsd4_encode_operation(resp, op);
2348         }
2349         return op->status;
2350 }
2351
2352 /*
2353  * The sequence operation is not cached because we can use the slot and
2354  * session values.
2355  */
2356 static __be32
2357 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2358                          struct nfsd4_sequence *seq)
2359 {
2360         struct nfsd4_slot *slot = resp->cstate.slot;
2361         struct xdr_stream *xdr = &resp->xdr;
2362         __be32 *p;
2363         __be32 status;
2364
2365         dprintk("--> %s slot %p\n", __func__, slot);
2366
2367         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2368         if (status)
2369                 return status;
2370
2371         p = xdr_reserve_space(xdr, slot->sl_datalen);
2372         if (!p) {
2373                 WARN_ON_ONCE(1);
2374                 return nfserr_serverfault;
2375         }
2376         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2377         xdr_commit_encode(xdr);
2378
2379         resp->opcnt = slot->sl_opcnt;
2380         return slot->sl_status;
2381 }
2382
2383 /*
2384  * Set the exchange_id flags returned by the server.
2385  */
2386 static void
2387 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2388 {
2389 #ifdef CONFIG_NFSD_PNFS
2390         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2391 #else
2392         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2393 #endif
2394
2395         /* Referrals are supported, Migration is not. */
2396         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2397
2398         /* set the wire flags to return to client. */
2399         clid->flags = new->cl_exchange_flags;
2400 }
2401
2402 static bool client_has_openowners(struct nfs4_client *clp)
2403 {
2404         struct nfs4_openowner *oo;
2405
2406         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2407                 if (!list_empty(&oo->oo_owner.so_stateids))
2408                         return true;
2409         }
2410         return false;
2411 }
2412
2413 static bool client_has_state(struct nfs4_client *clp)
2414 {
2415         return client_has_openowners(clp)
2416 #ifdef CONFIG_NFSD_PNFS
2417                 || !list_empty(&clp->cl_lo_states)
2418 #endif
2419                 || !list_empty(&clp->cl_delegations)
2420                 || !list_empty(&clp->cl_sessions);
2421 }
2422
2423 __be32
2424 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2425                 union nfsd4_op_u *u)
2426 {
2427         struct nfsd4_exchange_id *exid = &u->exchange_id;
2428         struct nfs4_client *conf, *new;
2429         struct nfs4_client *unconf = NULL;
2430         __be32 status;
2431         char                    addr_str[INET6_ADDRSTRLEN];
2432         nfs4_verifier           verf = exid->verifier;
2433         struct sockaddr         *sa = svc_addr(rqstp);
2434         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2435         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2436
2437         rpc_ntop(sa, addr_str, sizeof(addr_str));
2438         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2439                 "ip_addr=%s flags %x, spa_how %d\n",
2440                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2441                 addr_str, exid->flags, exid->spa_how);
2442
2443         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2444                 return nfserr_inval;
2445
2446         new = create_client(exid->clname, rqstp, &verf);
2447         if (new == NULL)
2448                 return nfserr_jukebox;
2449
2450         switch (exid->spa_how) {
2451         case SP4_MACH_CRED:
2452                 exid->spo_must_enforce[0] = 0;
2453                 exid->spo_must_enforce[1] = (
2454                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2455                         1 << (OP_EXCHANGE_ID - 32) |
2456                         1 << (OP_CREATE_SESSION - 32) |
2457                         1 << (OP_DESTROY_SESSION - 32) |
2458                         1 << (OP_DESTROY_CLIENTID - 32));
2459
2460                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2461                                         1 << (OP_OPEN_DOWNGRADE) |
2462                                         1 << (OP_LOCKU) |
2463                                         1 << (OP_DELEGRETURN));
2464
2465                 exid->spo_must_allow[1] &= (
2466                                         1 << (OP_TEST_STATEID - 32) |
2467                                         1 << (OP_FREE_STATEID - 32));
2468                 if (!svc_rqst_integrity_protected(rqstp)) {
2469                         status = nfserr_inval;
2470                         goto out_nolock;
2471                 }
2472                 /*
2473                  * Sometimes userspace doesn't give us a principal.
2474                  * Which is a bug, really.  Anyway, we can't enforce
2475                  * MACH_CRED in that case, better to give up now:
2476                  */
2477                 if (!new->cl_cred.cr_principal &&
2478                                         !new->cl_cred.cr_raw_principal) {
2479                         status = nfserr_serverfault;
2480                         goto out_nolock;
2481                 }
2482                 new->cl_mach_cred = true;
2483         case SP4_NONE:
2484                 break;
2485         default:                                /* checked by xdr code */
2486                 WARN_ON_ONCE(1);
2487         case SP4_SSV:
2488                 status = nfserr_encr_alg_unsupp;
2489                 goto out_nolock;
2490         }
2491
2492         /* Cases below refer to rfc 5661 section 18.35.4: */
2493         spin_lock(&nn->client_lock);
2494         conf = find_confirmed_client_by_name(&exid->clname, nn);
2495         if (conf) {
2496                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2497                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2498
2499                 if (update) {
2500                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2501                                 status = nfserr_inval;
2502                                 goto out;
2503                         }
2504                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2505                                 status = nfserr_wrong_cred;
2506                                 goto out;
2507                         }
2508                         if (!creds_match) { /* case 9 */
2509                                 status = nfserr_perm;
2510                                 goto out;
2511                         }
2512                         if (!verfs_match) { /* case 8 */
2513                                 status = nfserr_not_same;
2514                                 goto out;
2515                         }
2516                         /* case 6 */
2517                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2518                         goto out_copy;
2519                 }
2520                 if (!creds_match) { /* case 3 */
2521                         if (client_has_state(conf)) {
2522                                 status = nfserr_clid_inuse;
2523                                 goto out;
2524                         }
2525                         goto out_new;
2526                 }
2527                 if (verfs_match) { /* case 2 */
2528                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2529                         goto out_copy;
2530                 }
2531                 /* case 5, client reboot */
2532                 conf = NULL;
2533                 goto out_new;
2534         }
2535
2536         if (update) { /* case 7 */
2537                 status = nfserr_noent;
2538                 goto out;
2539         }
2540
2541         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2542         if (unconf) /* case 4, possible retry or client restart */
2543                 unhash_client_locked(unconf);
2544
2545         /* case 1 (normal case) */
2546 out_new:
2547         if (conf) {
2548                 status = mark_client_expired_locked(conf);
2549                 if (status)
2550                         goto out;
2551         }
2552         new->cl_minorversion = cstate->minorversion;
2553         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2554         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2555
2556         gen_clid(new, nn);
2557         add_to_unconfirmed(new);
2558         swap(new, conf);
2559 out_copy:
2560         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2561         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2562
2563         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2564         nfsd4_set_ex_flags(conf, exid);
2565
2566         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2567                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2568         status = nfs_ok;
2569
2570 out:
2571         spin_unlock(&nn->client_lock);
2572 out_nolock:
2573         if (new)
2574                 expire_client(new);
2575         if (unconf)
2576                 expire_client(unconf);
2577         return status;
2578 }
2579
2580 static __be32
2581 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2582 {
2583         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2584                 slot_seqid);
2585
2586         /* The slot is in use, and no response has been sent. */
2587         if (slot_inuse) {
2588                 if (seqid == slot_seqid)
2589                         return nfserr_jukebox;
2590                 else
2591                         return nfserr_seq_misordered;
2592         }
2593         /* Note unsigned 32-bit arithmetic handles wraparound: */
2594         if (likely(seqid == slot_seqid + 1))
2595                 return nfs_ok;
2596         if (seqid == slot_seqid)
2597                 return nfserr_replay_cache;
2598         return nfserr_seq_misordered;
2599 }
2600
2601 /*
2602  * Cache the create session result into the create session single DRC
2603  * slot cache by saving the xdr structure. sl_seqid has been set.
2604  * Do this for solo or embedded create session operations.
2605  */
2606 static void
2607 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2608                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2609 {
2610         slot->sl_status = nfserr;
2611         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2612 }
2613
2614 static __be32
2615 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2616                             struct nfsd4_clid_slot *slot)
2617 {
2618         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2619         return slot->sl_status;
2620 }
2621
2622 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2623                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2624                         1 +     /* MIN tag is length with zero, only length */ \
2625                         3 +     /* version, opcount, opcode */ \
2626                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2627                                 /* seqid, slotID, slotID, cache */ \
2628                         4 ) * sizeof(__be32))
2629
2630 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2631                         2 +     /* verifier: AUTH_NULL, length 0 */\
2632                         1 +     /* status */ \
2633                         1 +     /* MIN tag is length with zero, only length */ \
2634                         3 +     /* opcount, opcode, opstatus*/ \
2635                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2636                                 /* seqid, slotID, slotID, slotID, status */ \
2637                         5 ) * sizeof(__be32))
2638
2639 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2640 {
2641         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2642
2643         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2644                 return nfserr_toosmall;
2645         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2646                 return nfserr_toosmall;
2647         ca->headerpadsz = 0;
2648         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2649         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2650         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2651         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2652                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2653         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2654         /*
2655          * Note decreasing slot size below client's request may make it
2656          * difficult for client to function correctly, whereas
2657          * decreasing the number of slots will (just?) affect
2658          * performance.  When short on memory we therefore prefer to
2659          * decrease number of slots instead of their size.  Clients that
2660          * request larger slots than they need will get poor results:
2661          */
2662         ca->maxreqs = nfsd4_get_drc_mem(ca);
2663         if (!ca->maxreqs)
2664                 return nfserr_jukebox;
2665
2666         return nfs_ok;
2667 }
2668
2669 /*
2670  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2671  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2672  */
2673 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2674         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2675
2676 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2677         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2678
2679 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2680                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2681 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2682                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2683                                  sizeof(__be32))
2684
2685 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2686 {
2687         ca->headerpadsz = 0;
2688
2689         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2690                 return nfserr_toosmall;
2691         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2692                 return nfserr_toosmall;
2693         ca->maxresp_cached = 0;
2694         if (ca->maxops < 2)
2695                 return nfserr_toosmall;
2696
2697         return nfs_ok;
2698 }
2699
2700 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2701 {
2702         switch (cbs->flavor) {
2703         case RPC_AUTH_NULL:
2704         case RPC_AUTH_UNIX:
2705                 return nfs_ok;
2706         default:
2707                 /*
2708                  * GSS case: the spec doesn't allow us to return this
2709                  * error.  But it also doesn't allow us not to support
2710                  * GSS.
2711                  * I'd rather this fail hard than return some error the
2712                  * client might think it can already handle:
2713                  */
2714                 return nfserr_encr_alg_unsupp;
2715         }
2716 }
2717
2718 __be32
2719 nfsd4_create_session(struct svc_rqst *rqstp,
2720                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2721 {
2722         struct nfsd4_create_session *cr_ses = &u->create_session;
2723         struct sockaddr *sa = svc_addr(rqstp);
2724         struct nfs4_client *conf, *unconf;
2725         struct nfs4_client *old = NULL;
2726         struct nfsd4_session *new;
2727         struct nfsd4_conn *conn;
2728         struct nfsd4_clid_slot *cs_slot = NULL;
2729         __be32 status = 0;
2730         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2731
2732         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2733                 return nfserr_inval;
2734         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2735         if (status)
2736                 return status;
2737         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2738         if (status)
2739                 return status;
2740         status = check_backchannel_attrs(&cr_ses->back_channel);
2741         if (status)
2742                 goto out_release_drc_mem;
2743         status = nfserr_jukebox;
2744         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2745         if (!new)
2746                 goto out_release_drc_mem;
2747         conn = alloc_conn_from_crses(rqstp, cr_ses);
2748         if (!conn)
2749                 goto out_free_session;
2750
2751         spin_lock(&nn->client_lock);
2752         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2753         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2754         WARN_ON_ONCE(conf && unconf);
2755
2756         if (conf) {
2757                 status = nfserr_wrong_cred;
2758                 if (!nfsd4_mach_creds_match(conf, rqstp))
2759                         goto out_free_conn;
2760                 cs_slot = &conf->cl_cs_slot;
2761                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2762                 if (status) {
2763                         if (status == nfserr_replay_cache)
2764                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2765                         goto out_free_conn;
2766                 }
2767         } else if (unconf) {
2768                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2769                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2770                         status = nfserr_clid_inuse;
2771                         goto out_free_conn;
2772                 }
2773                 status = nfserr_wrong_cred;
2774                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2775                         goto out_free_conn;
2776                 cs_slot = &unconf->cl_cs_slot;
2777                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2778                 if (status) {
2779                         /* an unconfirmed replay returns misordered */
2780                         status = nfserr_seq_misordered;
2781                         goto out_free_conn;
2782                 }
2783                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2784                 if (old) {
2785                         status = mark_client_expired_locked(old);
2786                         if (status) {
2787                                 old = NULL;
2788                                 goto out_free_conn;
2789                         }
2790                 }
2791                 move_to_confirmed(unconf);
2792                 conf = unconf;
2793         } else {
2794                 status = nfserr_stale_clientid;
2795                 goto out_free_conn;
2796         }
2797         status = nfs_ok;
2798         /* Persistent sessions are not supported */
2799         cr_ses->flags &= ~SESSION4_PERSIST;
2800         /* Upshifting from TCP to RDMA is not supported */
2801         cr_ses->flags &= ~SESSION4_RDMA;
2802
2803         init_session(rqstp, new, conf, cr_ses);
2804         nfsd4_get_session_locked(new);
2805
2806         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2807                NFS4_MAX_SESSIONID_LEN);
2808         cs_slot->sl_seqid++;
2809         cr_ses->seqid = cs_slot->sl_seqid;
2810
2811         /* cache solo and embedded create sessions under the client_lock */
2812         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2813         spin_unlock(&nn->client_lock);
2814         /* init connection and backchannel */
2815         nfsd4_init_conn(rqstp, conn, new);
2816         nfsd4_put_session(new);
2817         if (old)
2818                 expire_client(old);
2819         return status;
2820 out_free_conn:
2821         spin_unlock(&nn->client_lock);
2822         free_conn(conn);
2823         if (old)
2824                 expire_client(old);
2825 out_free_session:
2826         __free_session(new);
2827 out_release_drc_mem:
2828         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2829         return status;
2830 }
2831
2832 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2833 {
2834         switch (*dir) {
2835         case NFS4_CDFC4_FORE:
2836         case NFS4_CDFC4_BACK:
2837                 return nfs_ok;
2838         case NFS4_CDFC4_FORE_OR_BOTH:
2839         case NFS4_CDFC4_BACK_OR_BOTH:
2840                 *dir = NFS4_CDFC4_BOTH;
2841                 return nfs_ok;
2842         };
2843         return nfserr_inval;
2844 }
2845
2846 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2847                 struct nfsd4_compound_state *cstate,
2848                 union nfsd4_op_u *u)
2849 {
2850         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2851         struct nfsd4_session *session = cstate->session;
2852         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2853         __be32 status;
2854
2855         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2856         if (status)
2857                 return status;
2858         spin_lock(&nn->client_lock);
2859         session->se_cb_prog = bc->bc_cb_program;
2860         session->se_cb_sec = bc->bc_cb_sec;
2861         spin_unlock(&nn->client_lock);
2862
2863         nfsd4_probe_callback(session->se_client);
2864
2865         return nfs_ok;
2866 }
2867
2868 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2869                      struct nfsd4_compound_state *cstate,
2870                      union nfsd4_op_u *u)
2871 {
2872         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2873         __be32 status;
2874         struct nfsd4_conn *conn;
2875         struct nfsd4_session *session;
2876         struct net *net = SVC_NET(rqstp);
2877         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2878
2879         if (!nfsd4_last_compound_op(rqstp))
2880                 return nfserr_not_only_op;
2881         spin_lock(&nn->client_lock);
2882         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2883         spin_unlock(&nn->client_lock);
2884         if (!session)
2885                 goto out_no_session;
2886         status = nfserr_wrong_cred;
2887         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2888                 goto out;
2889         status = nfsd4_map_bcts_dir(&bcts->dir);
2890         if (status)
2891                 goto out;
2892         conn = alloc_conn(rqstp, bcts->dir);
2893         status = nfserr_jukebox;
2894         if (!conn)
2895                 goto out;
2896         nfsd4_init_conn(rqstp, conn, session);
2897         status = nfs_ok;
2898 out:
2899         nfsd4_put_session(session);
2900 out_no_session:
2901         return status;
2902 }
2903
2904 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2905 {
2906         if (!session)
2907                 return 0;
2908         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2909 }
2910
2911 __be32
2912 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2913                 union nfsd4_op_u *u)
2914 {
2915         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2916         struct nfsd4_session *ses;
2917         __be32 status;
2918         int ref_held_by_me = 0;
2919         struct net *net = SVC_NET(r);
2920         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2921
2922         status = nfserr_not_only_op;
2923         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2924                 if (!nfsd4_last_compound_op(r))
2925                         goto out;
2926                 ref_held_by_me++;
2927         }
2928         dump_sessionid(__func__, &sessionid->sessionid);
2929         spin_lock(&nn->client_lock);
2930         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2931         if (!ses)
2932                 goto out_client_lock;
2933         status = nfserr_wrong_cred;
2934         if (!nfsd4_mach_creds_match(ses->se_client, r))
2935                 goto out_put_session;
2936         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2937         if (status)
2938                 goto out_put_session;
2939         unhash_session(ses);
2940         spin_unlock(&nn->client_lock);
2941
2942         nfsd4_probe_callback_sync(ses->se_client);
2943
2944         spin_lock(&nn->client_lock);
2945         status = nfs_ok;
2946 out_put_session:
2947         nfsd4_put_session_locked(ses);
2948 out_client_lock:
2949         spin_unlock(&nn->client_lock);
2950 out:
2951         return status;
2952 }
2953
2954 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2955 {
2956         struct nfsd4_conn *c;
2957
2958         list_for_each_entry(c, &s->se_conns, cn_persession) {
2959                 if (c->cn_xprt == xpt) {
2960                         return c;
2961                 }
2962         }
2963         return NULL;
2964 }
2965
2966 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2967 {
2968         struct nfs4_client *clp = ses->se_client;
2969         struct nfsd4_conn *c;
2970         __be32 status = nfs_ok;
2971         int ret;
2972
2973         spin_lock(&clp->cl_lock);
2974         c = __nfsd4_find_conn(new->cn_xprt, ses);
2975         if (c)
2976                 goto out_free;
2977         status = nfserr_conn_not_bound_to_session;
2978         if (clp->cl_mach_cred)
2979                 goto out_free;
2980         __nfsd4_hash_conn(new, ses);
2981         spin_unlock(&clp->cl_lock);
2982         ret = nfsd4_register_conn(new);
2983         if (ret)
2984                 /* oops; xprt is already down: */
2985                 nfsd4_conn_lost(&new->cn_xpt_user);
2986         return nfs_ok;
2987 out_free:
2988         spin_unlock(&clp->cl_lock);
2989         free_conn(new);
2990         return status;
2991 }
2992
2993 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2994 {
2995         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2996
2997         return args->opcnt > session->se_fchannel.maxops;
2998 }
2999
3000 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3001                                   struct nfsd4_session *session)
3002 {
3003         struct xdr_buf *xb = &rqstp->rq_arg;
3004
3005         return xb->len > session->se_fchannel.maxreq_sz;
3006 }
3007
3008 static bool replay_matches_cache(struct svc_rqst *rqstp,
3009                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3010 {
3011         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3012
3013         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3014             (bool)seq->cachethis)
3015                 return false;
3016         /*
3017          * If there's an error than the reply can have fewer ops than
3018          * the call.  But if we cached a reply with *more* ops than the
3019          * call you're sending us now, then this new call is clearly not
3020          * really a replay of the old one:
3021          */
3022         if (slot->sl_opcnt < argp->opcnt)
3023                 return false;
3024         /* This is the only check explicitly called by spec: */
3025         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3026                 return false;
3027         /*
3028          * There may be more comparisons we could actually do, but the
3029          * spec doesn't require us to catch every case where the calls
3030          * don't match (that would require caching the call as well as
3031          * the reply), so we don't bother.
3032          */
3033         return true;
3034 }
3035
3036 __be32
3037 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3038                 union nfsd4_op_u *u)
3039 {
3040         struct nfsd4_sequence *seq = &u->sequence;
3041         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3042         struct xdr_stream *xdr = &resp->xdr;
3043         struct nfsd4_session *session;
3044         struct nfs4_client *clp;
3045         struct nfsd4_slot *slot;
3046         struct nfsd4_conn *conn;
3047         __be32 status;
3048         int buflen;
3049         struct net *net = SVC_NET(rqstp);
3050         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3051
3052         if (resp->opcnt != 1)
3053                 return nfserr_sequence_pos;
3054
3055         /*
3056          * Will be either used or freed by nfsd4_sequence_check_conn
3057          * below.
3058          */
3059         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3060         if (!conn)
3061                 return nfserr_jukebox;
3062
3063         spin_lock(&nn->client_lock);
3064         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3065         if (!session)
3066                 goto out_no_session;
3067         clp = session->se_client;
3068
3069         status = nfserr_too_many_ops;
3070         if (nfsd4_session_too_many_ops(rqstp, session))
3071                 goto out_put_session;
3072
3073         status = nfserr_req_too_big;
3074         if (nfsd4_request_too_big(rqstp, session))
3075                 goto out_put_session;
3076
3077         status = nfserr_badslot;
3078         if (seq->slotid >= session->se_fchannel.maxreqs)
3079                 goto out_put_session;
3080
3081         slot = session->se_slots[seq->slotid];
3082         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3083
3084         /* We do not negotiate the number of slots yet, so set the
3085          * maxslots to the session maxreqs which is used to encode
3086          * sr_highest_slotid and the sr_target_slot id to maxslots */
3087         seq->maxslots = session->se_fchannel.maxreqs;
3088
3089         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3090                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3091         if (status == nfserr_replay_cache) {
3092                 status = nfserr_seq_misordered;
3093                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3094                         goto out_put_session;
3095                 status = nfserr_seq_false_retry;
3096                 if (!replay_matches_cache(rqstp, seq, slot))
3097                         goto out_put_session;
3098                 cstate->slot = slot;
3099                 cstate->session = session;
3100                 cstate->clp = clp;
3101                 /* Return the cached reply status and set cstate->status
3102                  * for nfsd4_proc_compound processing */
3103                 status = nfsd4_replay_cache_entry(resp, seq);
3104                 cstate->status = nfserr_replay_cache;
3105                 goto out;
3106         }
3107         if (status)
3108                 goto out_put_session;
3109
3110         status = nfsd4_sequence_check_conn(conn, session);
3111         conn = NULL;
3112         if (status)
3113                 goto out_put_session;
3114
3115         buflen = (seq->cachethis) ?
3116                         session->se_fchannel.maxresp_cached :
3117                         session->se_fchannel.maxresp_sz;
3118         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3119                                     nfserr_rep_too_big;
3120         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3121                 goto out_put_session;
3122         svc_reserve(rqstp, buflen);
3123
3124         status = nfs_ok;
3125         /* Success! bump slot seqid */
3126         slot->sl_seqid = seq->seqid;
3127         slot->sl_flags |= NFSD4_SLOT_INUSE;
3128         if (seq->cachethis)
3129                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3130         else
3131                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3132
3133         cstate->slot = slot;
3134         cstate->session = session;
3135         cstate->clp = clp;
3136
3137 out:
3138         switch (clp->cl_cb_state) {
3139         case NFSD4_CB_DOWN:
3140                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3141                 break;
3142         case NFSD4_CB_FAULT:
3143                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3144                 break;
3145         default:
3146                 seq->status_flags = 0;
3147         }
3148         if (!list_empty(&clp->cl_revoked))
3149                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3150 out_no_session:
3151         if (conn)
3152                 free_conn(conn);
3153         spin_unlock(&nn->client_lock);
3154         return status;
3155 out_put_session:
3156         nfsd4_put_session_locked(session);
3157         goto out_no_session;
3158 }
3159
3160 void
3161 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3162 {
3163         struct nfsd4_compound_state *cs = &resp->cstate;
3164
3165         if (nfsd4_has_session(cs)) {
3166                 if (cs->status != nfserr_replay_cache) {
3167                         nfsd4_store_cache_entry(resp);
3168                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3169                 }
3170                 /* Drop session reference that was taken in nfsd4_sequence() */
3171                 nfsd4_put_session(cs->session);
3172         } else if (cs->clp)
3173                 put_client_renew(cs->clp);
3174 }
3175
3176 __be32
3177 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3178                 struct nfsd4_compound_state *cstate,
3179                 union nfsd4_op_u *u)
3180 {
3181         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3182         struct nfs4_client *conf, *unconf;
3183         struct nfs4_client *clp = NULL;
3184         __be32 status = 0;
3185         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3186
3187         spin_lock(&nn->client_lock);
3188         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3189         conf = find_confirmed_client(&dc->clientid, true, nn);
3190         WARN_ON_ONCE(conf && unconf);
3191
3192         if (conf) {
3193                 if (client_has_state(conf)) {
3194                         status = nfserr_clientid_busy;
3195                         goto out;
3196                 }
3197                 status = mark_client_expired_locked(conf);
3198                 if (status)
3199                         goto out;
3200                 clp = conf;
3201         } else if (unconf)
3202                 clp = unconf;
3203         else {
3204                 status = nfserr_stale_clientid;
3205                 goto out;
3206         }
3207         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3208                 clp = NULL;
3209                 status = nfserr_wrong_cred;
3210                 goto out;
3211         }
3212         unhash_client_locked(clp);
3213 out:
3214         spin_unlock(&nn->client_lock);
3215         if (clp)
3216                 expire_client(clp);
3217         return status;
3218 }
3219
3220 __be32
3221 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3222                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3223 {
3224         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3225         __be32 status = 0;
3226
3227         if (rc->rca_one_fs) {
3228                 if (!cstate->current_fh.fh_dentry)
3229                         return nfserr_nofilehandle;
3230                 /*
3231                  * We don't take advantage of the rca_one_fs case.
3232                  * That's OK, it's optional, we can safely ignore it.
3233                  */
3234                 return nfs_ok;
3235         }
3236
3237         status = nfserr_complete_already;
3238         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3239                              &cstate->session->se_client->cl_flags))
3240                 goto out;
3241
3242         status = nfserr_stale_clientid;
3243         if (is_client_expired(cstate->session->se_client))
3244                 /*
3245                  * The following error isn't really legal.
3246                  * But we only get here if the client just explicitly
3247                  * destroyed the client.  Surely it no longer cares what
3248                  * error it gets back on an operation for the dead
3249                  * client.
3250                  */
3251                 goto out;
3252
3253         status = nfs_ok;
3254         nfsd4_client_record_create(cstate->session->se_client);
3255 out:
3256         return status;
3257 }
3258
3259 __be32
3260 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3261                   union nfsd4_op_u *u)
3262 {
3263         struct nfsd4_setclientid *setclid = &u->setclientid;
3264         struct xdr_netobj       clname = setclid->se_name;
3265         nfs4_verifier           clverifier = setclid->se_verf;
3266         struct nfs4_client      *conf, *new;
3267         struct nfs4_client      *unconf = NULL;
3268         __be32                  status;
3269         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3270
3271         new = create_client(clname, rqstp, &clverifier);
3272         if (new == NULL)
3273                 return nfserr_jukebox;
3274         /* Cases below refer to rfc 3530 section 14.2.33: */
3275         spin_lock(&nn->client_lock);
3276         conf = find_confirmed_client_by_name(&clname, nn);
3277         if (conf && client_has_state(conf)) {
3278                 /* case 0: */
3279                 status = nfserr_clid_inuse;
3280                 if (clp_used_exchangeid(conf))
3281                         goto out;
3282                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3283                         char addr_str[INET6_ADDRSTRLEN];
3284                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3285                                  sizeof(addr_str));
3286                         dprintk("NFSD: setclientid: string in use by client "
3287                                 "at %s\n", addr_str);
3288                         goto out;
3289                 }
3290         }
3291         unconf = find_unconfirmed_client_by_name(&clname, nn);
3292         if (unconf)
3293                 unhash_client_locked(unconf);
3294         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3295                 /* case 1: probable callback update */
3296                 copy_clid(new, conf);
3297                 gen_confirm(new, nn);
3298         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3299                 gen_clid(new, nn);
3300         new->cl_minorversion = 0;
3301         gen_callback(new, setclid, rqstp);
3302         add_to_unconfirmed(new);
3303         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3304         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3305         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3306         new = NULL;
3307         status = nfs_ok;
3308 out:
3309         spin_unlock(&nn->client_lock);
3310         if (new)
3311                 free_client(new);
3312         if (unconf)
3313                 expire_client(unconf);
3314         return status;
3315 }
3316
3317
3318 __be32
3319 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3320                         struct nfsd4_compound_state *cstate,
3321                         union nfsd4_op_u *u)
3322 {
3323         struct nfsd4_setclientid_confirm *setclientid_confirm =
3324                         &u->setclientid_confirm;
3325         struct nfs4_client *conf, *unconf;
3326         struct nfs4_client *old = NULL;
3327         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3328         clientid_t * clid = &setclientid_confirm->sc_clientid;
3329         __be32 status;
3330         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3331
3332         if (STALE_CLIENTID(clid, nn))
3333                 return nfserr_stale_clientid;
3334
3335         spin_lock(&nn->client_lock);
3336         conf = find_confirmed_client(clid, false, nn);
3337         unconf = find_unconfirmed_client(clid, false, nn);
3338         /*
3339          * We try hard to give out unique clientid's, so if we get an
3340          * attempt to confirm the same clientid with a different cred,
3341          * the client may be buggy; this should never happen.
3342          *
3343          * Nevertheless, RFC 7530 recommends INUSE for this case:
3344          */
3345         status = nfserr_clid_inuse;
3346         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3347                 goto out;
3348         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3349                 goto out;
3350         /* cases below refer to rfc 3530 section 14.2.34: */
3351         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3352                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3353                         /* case 2: probable retransmit */
3354                         status = nfs_ok;
3355                 } else /* case 4: client hasn't noticed we rebooted yet? */
3356                         status = nfserr_stale_clientid;
3357                 goto out;
3358         }
3359         status = nfs_ok;
3360         if (conf) { /* case 1: callback update */
3361                 old = unconf;
3362                 unhash_client_locked(old);
3363                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3364         } else { /* case 3: normal case; new or rebooted client */
3365                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3366                 if (old) {
3367                         status = nfserr_clid_inuse;
3368                         if (client_has_state(old)
3369                                         && !same_creds(&unconf->cl_cred,
3370                                                         &old->cl_cred))
3371                                 goto out;
3372                         status = mark_client_expired_locked(old);
3373                         if (status) {
3374                                 old = NULL;
3375                                 goto out;
3376                         }
3377                 }
3378                 move_to_confirmed(unconf);
3379                 conf = unconf;
3380         }
3381         get_client_locked(conf);
3382         spin_unlock(&nn->client_lock);
3383         nfsd4_probe_callback(conf);
3384         spin_lock(&nn->client_lock);
3385         put_client_renew_locked(conf);
3386 out:
3387         spin_unlock(&nn->client_lock);
3388         if (old)
3389                 expire_client(old);
3390         return status;
3391 }
3392
3393 static struct nfs4_file *nfsd4_alloc_file(void)
3394 {
3395         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3396 }
3397
3398 /* OPEN Share state helper functions */
3399 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3400                                 struct nfs4_file *fp)
3401 {
3402         lockdep_assert_held(&state_lock);
3403
3404         atomic_set(&fp->fi_ref, 1);
3405         spin_lock_init(&fp->fi_lock);
3406         INIT_LIST_HEAD(&fp->fi_stateids);
3407         INIT_LIST_HEAD(&fp->fi_delegations);
3408         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3409         fh_copy_shallow(&fp->fi_fhandle, fh);
3410         fp->fi_deleg_file = NULL;
3411         fp->fi_had_conflict = false;
3412         fp->fi_share_deny = 0;
3413         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3414         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3415 #ifdef CONFIG_NFSD_PNFS
3416         INIT_LIST_HEAD(&fp->fi_lo_states);
3417         atomic_set(&fp->fi_lo_recalls, 0);
3418 #endif
3419         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3420 }
3421
3422 void
3423 nfsd4_free_slabs(void)
3424 {
3425         kmem_cache_destroy(odstate_slab);
3426         kmem_cache_destroy(openowner_slab);
3427         kmem_cache_destroy(lockowner_slab);
3428         kmem_cache_destroy(file_slab);
3429         kmem_cache_destroy(stateid_slab);
3430         kmem_cache_destroy(deleg_slab);
3431 }
3432
3433 int
3434 nfsd4_init_slabs(void)
3435 {
3436         openowner_slab = kmem_cache_create("nfsd4_openowners",
3437                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3438         if (openowner_slab == NULL)
3439                 goto out;
3440         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3441                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3442         if (lockowner_slab == NULL)
3443                 goto out_free_openowner_slab;
3444         file_slab = kmem_cache_create("nfsd4_files",
3445                         sizeof(struct nfs4_file), 0, 0, NULL);
3446         if (file_slab == NULL)
3447                 goto out_free_lockowner_slab;
3448         stateid_slab = kmem_cache_create("nfsd4_stateids",
3449                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3450         if (stateid_slab == NULL)
3451                 goto out_free_file_slab;
3452         deleg_slab = kmem_cache_create("nfsd4_delegations",
3453                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3454         if (deleg_slab == NULL)
3455                 goto out_free_stateid_slab;
3456         odstate_slab = kmem_cache_create("nfsd4_odstate",
3457                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3458         if (odstate_slab == NULL)
3459                 goto out_free_deleg_slab;
3460         return 0;
3461
3462 out_free_deleg_slab:
3463         kmem_cache_destroy(deleg_slab);
3464 out_free_stateid_slab:
3465         kmem_cache_destroy(stateid_slab);
3466 out_free_file_slab:
3467         kmem_cache_destroy(file_slab);
3468 out_free_lockowner_slab:
3469         kmem_cache_destroy(lockowner_slab);
3470 out_free_openowner_slab:
3471         kmem_cache_destroy(openowner_slab);
3472 out:
3473         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3474         return -ENOMEM;
3475 }
3476
3477 static void init_nfs4_replay(struct nfs4_replay *rp)
3478 {
3479         rp->rp_status = nfserr_serverfault;
3480         rp->rp_buflen = 0;
3481         rp->rp_buf = rp->rp_ibuf;
3482         mutex_init(&rp->rp_mutex);
3483 }
3484
3485 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3486                 struct nfs4_stateowner *so)
3487 {
3488         if (!nfsd4_has_session(cstate)) {
3489                 mutex_lock(&so->so_replay.rp_mutex);
3490                 cstate->replay_owner = nfs4_get_stateowner(so);
3491         }
3492 }
3493
3494 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3495 {
3496         struct nfs4_stateowner *so = cstate->replay_owner;
3497
3498         if (so != NULL) {
3499                 cstate->replay_owner = NULL;
3500                 mutex_unlock(&so->so_replay.rp_mutex);
3501                 nfs4_put_stateowner(so);
3502         }
3503 }
3504
3505 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3506 {
3507         struct nfs4_stateowner *sop;
3508
3509         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3510         if (!sop)
3511                 return NULL;
3512
3513         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3514         if (!sop->so_owner.data) {
3515                 kmem_cache_free(slab, sop);
3516                 return NULL;
3517         }
3518         sop->so_owner.len = owner->len;
3519
3520         INIT_LIST_HEAD(&sop->so_stateids);
3521         sop->so_client = clp;
3522         init_nfs4_replay(&sop->so_replay);
3523         atomic_set(&sop->so_count, 1);
3524         return sop;
3525 }
3526
3527 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3528 {
3529         lockdep_assert_held(&clp->cl_lock);
3530
3531         list_add(&oo->oo_owner.so_strhash,
3532                  &clp->cl_ownerstr_hashtbl[strhashval]);
3533         list_add(&oo->oo_perclient, &clp->cl_openowners);
3534 }
3535
3536 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3537 {
3538         unhash_openowner_locked(openowner(so));
3539 }
3540
3541 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3542 {
3543         struct nfs4_openowner *oo = openowner(so);
3544
3545         kmem_cache_free(openowner_slab, oo);
3546 }
3547
3548 static const struct nfs4_stateowner_operations openowner_ops = {
3549         .so_unhash =    nfs4_unhash_openowner,
3550         .so_free =      nfs4_free_openowner,
3551 };
3552
3553 static struct nfs4_ol_stateid *
3554 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3555 {
3556         struct nfs4_ol_stateid *local, *ret = NULL;
3557         struct nfs4_openowner *oo = open->op_openowner;
3558
3559         lockdep_assert_held(&fp->fi_lock);
3560
3561         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3562                 /* ignore lock owners */
3563                 if (local->st_stateowner->so_is_open_owner == 0)
3564                         continue;
3565                 if (local->st_stateowner == &oo->oo_owner) {
3566                         ret = local;
3567                         refcount_inc(&ret->st_stid.sc_count);
3568                         break;
3569                 }
3570         }
3571         return ret;
3572 }
3573
3574 static struct nfs4_openowner *
3575 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3576                            struct nfsd4_compound_state *cstate)
3577 {
3578         struct nfs4_client *clp = cstate->clp;
3579         struct nfs4_openowner *oo, *ret;
3580
3581         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3582         if (!oo)
3583                 return NULL;
3584         oo->oo_owner.so_ops = &openowner_ops;
3585         oo->oo_owner.so_is_open_owner = 1;
3586         oo->oo_owner.so_seqid = open->op_seqid;
3587         oo->oo_flags = 0;
3588         if (nfsd4_has_session(cstate))
3589                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3590         oo->oo_time = 0;
3591         oo->oo_last_closed_stid = NULL;
3592         INIT_LIST_HEAD(&oo->oo_close_lru);
3593         spin_lock(&clp->cl_lock);
3594         ret = find_openstateowner_str_locked(strhashval, open, clp);
3595         if (ret == NULL) {
3596                 hash_openowner(oo, clp, strhashval);
3597                 ret = oo;
3598         } else
3599                 nfs4_free_stateowner(&oo->oo_owner);
3600
3601         spin_unlock(&clp->cl_lock);
3602         return ret;
3603 }
3604
3605 static struct nfs4_ol_stateid *
3606 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3607 {
3608
3609         struct nfs4_openowner *oo = open->op_openowner;
3610         struct nfs4_ol_stateid *retstp = NULL;
3611         struct nfs4_ol_stateid *stp;
3612
3613         stp = open->op_stp;
3614         /* We are moving these outside of the spinlocks to avoid the warnings */
3615         mutex_init(&stp->st_mutex);
3616         mutex_lock(&stp->st_mutex);
3617
3618         spin_lock(&oo->oo_owner.so_client->cl_lock);
3619         spin_lock(&fp->fi_lock);
3620
3621         retstp = nfsd4_find_existing_open(fp, open);
3622         if (retstp)
3623                 goto out_unlock;
3624
3625         open->op_stp = NULL;
3626         refcount_inc(&stp->st_stid.sc_count);
3627         stp->st_stid.sc_type = NFS4_OPEN_STID;
3628         INIT_LIST_HEAD(&stp->st_locks);
3629         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3630         get_nfs4_file(fp);
3631         stp->st_stid.sc_file = fp;
3632         stp->st_access_bmap = 0;
3633         stp->st_deny_bmap = 0;
3634         stp->st_openstp = NULL;
3635         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3636         list_add(&stp->st_perfile, &fp->fi_stateids);
3637
3638 out_unlock:
3639         spin_unlock(&fp->fi_lock);
3640         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3641         if (retstp) {
3642                 mutex_lock(&retstp->st_mutex);
3643                 /* To keep mutex tracking happy */
3644                 mutex_unlock(&stp->st_mutex);
3645                 stp = retstp;
3646         }
3647         return stp;
3648 }
3649
3650 /*
3651  * In the 4.0 case we need to keep the owners around a little while to handle
3652  * CLOSE replay. We still do need to release any file access that is held by
3653  * them before returning however.
3654  */
3655 static void
3656 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3657 {
3658         struct nfs4_ol_stateid *last;
3659         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3660         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3661                                                 nfsd_net_id);
3662
3663         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3664
3665         /*
3666          * We know that we hold one reference via nfsd4_close, and another
3667          * "persistent" reference for the client. If the refcount is higher
3668          * than 2, then there are still calls in progress that are using this
3669          * stateid. We can't put the sc_file reference until they are finished.
3670          * Wait for the refcount to drop to 2. Since it has been unhashed,
3671          * there should be no danger of the refcount going back up again at
3672          * this point.
3673          */
3674         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3675
3676         release_all_access(s);
3677         if (s->st_stid.sc_file) {
3678                 put_nfs4_file(s->st_stid.sc_file);
3679                 s->st_stid.sc_file = NULL;
3680         }
3681
3682         spin_lock(&nn->client_lock);
3683         last = oo->oo_last_closed_stid;
3684         oo->oo_last_closed_stid = s;
3685         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3686         oo->oo_time = get_seconds();
3687         spin_unlock(&nn->client_lock);
3688         if (last)
3689                 nfs4_put_stid(&last->st_stid);
3690 }
3691
3692 /* search file_hashtbl[] for file */
3693 static struct nfs4_file *
3694 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3695 {
3696         struct nfs4_file *fp;
3697
3698         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3699                 if (fh_match(&fp->fi_fhandle, fh)) {
3700                         if (atomic_inc_not_zero(&fp->fi_ref))
3701                                 return fp;
3702                 }
3703         }
3704         return NULL;
3705 }
3706
3707 struct nfs4_file *
3708 find_file(struct knfsd_fh *fh)
3709 {
3710         struct nfs4_file *fp;
3711         unsigned int hashval = file_hashval(fh);
3712
3713         rcu_read_lock();
3714         fp = find_file_locked(fh, hashval);
3715         rcu_read_unlock();
3716         return fp;
3717 }
3718
3719 static struct nfs4_file *
3720 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3721 {
3722         struct nfs4_file *fp;
3723         unsigned int hashval = file_hashval(fh);
3724
3725         rcu_read_lock();
3726         fp = find_file_locked(fh, hashval);
3727         rcu_read_unlock();
3728         if (fp)
3729                 return fp;
3730
3731         spin_lock(&state_lock);
3732         fp = find_file_locked(fh, hashval);
3733         if (likely(fp == NULL)) {
3734                 nfsd4_init_file(fh, hashval, new);
3735                 fp = new;
3736         }
3737         spin_unlock(&state_lock);
3738
3739         return fp;
3740 }
3741
3742 /*
3743  * Called to check deny when READ with all zero stateid or
3744  * WRITE with all zero or all one stateid
3745  */
3746 static __be32
3747 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3748 {
3749         struct nfs4_file *fp;
3750         __be32 ret = nfs_ok;
3751
3752         fp = find_file(&current_fh->fh_handle);
3753         if (!fp)
3754                 return ret;
3755         /* Check for conflicting share reservations */
3756         spin_lock(&fp->fi_lock);
3757         if (fp->fi_share_deny & deny_type)
3758                 ret = nfserr_locked;
3759         spin_unlock(&fp->fi_lock);
3760         put_nfs4_file(fp);
3761         return ret;
3762 }
3763
3764 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3765 {
3766         struct nfs4_delegation *dp = cb_to_delegation(cb);
3767         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3768                                           nfsd_net_id);
3769
3770         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3771
3772         /*
3773          * We can't do this in nfsd_break_deleg_cb because it is
3774          * already holding inode->i_lock.
3775          *
3776          * If the dl_time != 0, then we know that it has already been
3777          * queued for a lease break. Don't queue it again.
3778          */
3779         spin_lock(&state_lock);
3780         if (dp->dl_time == 0) {
3781                 dp->dl_time = get_seconds();
3782                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3783         }
3784         spin_unlock(&state_lock);
3785 }
3786
3787 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3788                 struct rpc_task *task)
3789 {
3790         struct nfs4_delegation *dp = cb_to_delegation(cb);
3791
3792         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3793                 return 1;
3794
3795         switch (task->tk_status) {
3796         case 0:
3797                 return 1;
3798         case -EBADHANDLE:
3799         case -NFS4ERR_BAD_STATEID:
3800                 /*
3801                  * Race: client probably got cb_recall before open reply
3802                  * granting delegation.
3803                  */
3804                 if (dp->dl_retries--) {
3805                         rpc_delay(task, 2 * HZ);
3806                         return 0;
3807                 }
3808                 /*FALLTHRU*/
3809         default:
3810                 return -1;
3811         }
3812 }
3813
3814 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3815 {
3816         struct nfs4_delegation *dp = cb_to_delegation(cb);
3817
3818         nfs4_put_stid(&dp->dl_stid);
3819 }
3820
3821 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3822         .prepare        = nfsd4_cb_recall_prepare,
3823         .done           = nfsd4_cb_recall_done,
3824         .release        = nfsd4_cb_recall_release,
3825 };
3826
3827 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3828 {
3829         /*
3830          * We're assuming the state code never drops its reference
3831          * without first removing the lease.  Since we're in this lease
3832          * callback (and since the lease code is serialized by the kernel
3833          * lock) we know the server hasn't removed the lease yet, we know
3834          * it's safe to take a reference.
3835          */
3836         refcount_inc(&dp->dl_stid.sc_count);
3837         nfsd4_run_cb(&dp->dl_recall);
3838 }
3839
3840 /* Called from break_lease() with i_lock held. */
3841 static bool
3842 nfsd_break_deleg_cb(struct file_lock *fl)
3843 {
3844         bool ret = false;
3845         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3846         struct nfs4_delegation *dp;
3847
3848         if (!fp) {
3849                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3850                 return ret;
3851         }
3852         if (fp->fi_had_conflict) {
3853                 WARN(1, "duplicate break on %p\n", fp);
3854                 return ret;
3855         }
3856         /*
3857          * We don't want the locks code to timeout the lease for us;
3858          * we'll remove it ourself if a delegation isn't returned
3859          * in time:
3860          */
3861         fl->fl_break_time = 0;
3862
3863         spin_lock(&fp->fi_lock);
3864         fp->fi_had_conflict = true;
3865         /*
3866          * If there are no delegations on the list, then return true
3867          * so that the lease code will go ahead and delete it.
3868          */
3869         if (list_empty(&fp->fi_delegations))
3870                 ret = true;
3871         else
3872                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3873                         nfsd_break_one_deleg(dp);
3874         spin_unlock(&fp->fi_lock);
3875         return ret;
3876 }
3877
3878 static int
3879 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3880                      struct list_head *dispose)
3881 {
3882         if (arg & F_UNLCK)
3883                 return lease_modify(onlist, arg, dispose);
3884         else
3885                 return -EAGAIN;
3886 }
3887
3888 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3889         .lm_break = nfsd_break_deleg_cb,
3890         .lm_change = nfsd_change_deleg_cb,
3891 };
3892
3893 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3894 {
3895         if (nfsd4_has_session(cstate))
3896                 return nfs_ok;
3897         if (seqid == so->so_seqid - 1)
3898                 return nfserr_replay_me;
3899         if (seqid == so->so_seqid)
3900                 return nfs_ok;
3901         return nfserr_bad_seqid;
3902 }
3903
3904 static __be32 lookup_clientid(clientid_t *clid,
3905                 struct nfsd4_compound_state *cstate,
3906                 struct nfsd_net *nn)
3907 {
3908         struct nfs4_client *found;
3909
3910         if (cstate->clp) {
3911                 found = cstate->clp;
3912                 if (!same_clid(&found->cl_clientid, clid))
3913                         return nfserr_stale_clientid;
3914                 return nfs_ok;
3915         }
3916
3917         if (STALE_CLIENTID(clid, nn))
3918                 return nfserr_stale_clientid;
3919
3920         /*
3921          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3922          * cached already then we know this is for is for v4.0 and "sessions"
3923          * will be false.
3924          */
3925         WARN_ON_ONCE(cstate->session);
3926         spin_lock(&nn->client_lock);
3927         found = find_confirmed_client(clid, false, nn);
3928         if (!found) {
3929                 spin_unlock(&nn->client_lock);
3930                 return nfserr_expired;
3931         }
3932         atomic_inc(&found->cl_refcount);
3933         spin_unlock(&nn->client_lock);
3934
3935         /* Cache the nfs4_client in cstate! */
3936         cstate->clp = found;
3937         return nfs_ok;
3938 }
3939
3940 __be32
3941 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3942                     struct nfsd4_open *open, struct nfsd_net *nn)
3943 {
3944         clientid_t *clientid = &open->op_clientid;
3945         struct nfs4_client *clp = NULL;
3946         unsigned int strhashval;
3947         struct nfs4_openowner *oo = NULL;
3948         __be32 status;
3949
3950         if (STALE_CLIENTID(&open->op_clientid, nn))
3951                 return nfserr_stale_clientid;
3952         /*
3953          * In case we need it later, after we've already created the
3954          * file and don't want to risk a further failure:
3955          */
3956         open->op_file = nfsd4_alloc_file();
3957         if (open->op_file == NULL)
3958                 return nfserr_jukebox;
3959
3960         status = lookup_clientid(clientid, cstate, nn);
3961         if (status)
3962                 return status;
3963         clp = cstate->clp;
3964
3965         strhashval = ownerstr_hashval(&open->op_owner);
3966         oo = find_openstateowner_str(strhashval, open, clp);
3967         open->op_openowner = oo;
3968         if (!oo) {
3969                 goto new_owner;
3970         }
3971         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3972                 /* Replace unconfirmed owners without checking for replay. */
3973                 release_openowner(oo);
3974                 open->op_openowner = NULL;
3975                 goto new_owner;
3976         }
3977         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3978         if (status)
3979                 return status;
3980         goto alloc_stateid;
3981 new_owner:
3982         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3983         if (oo == NULL)
3984                 return nfserr_jukebox;
3985         open->op_openowner = oo;
3986 alloc_stateid:
3987         open->op_stp = nfs4_alloc_open_stateid(clp);
3988         if (!open->op_stp)
3989                 return nfserr_jukebox;
3990
3991         if (nfsd4_has_session(cstate) &&
3992             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3993                 open->op_odstate = alloc_clnt_odstate(clp);
3994                 if (!open->op_odstate)
3995                         return nfserr_jukebox;
3996         }
3997
3998         return nfs_ok;
3999 }
4000
4001 static inline __be32
4002 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4003 {
4004         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4005                 return nfserr_openmode;
4006         else
4007                 return nfs_ok;
4008 }
4009
4010 static int share_access_to_flags(u32 share_access)
4011 {
4012         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4013 }
4014
4015 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4016 {
4017         struct nfs4_stid *ret;
4018
4019         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
4020         if (!ret)
4021                 return NULL;
4022         return delegstateid(ret);
4023 }
4024
4025 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4026 {
4027         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4028                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4029 }
4030
4031 static __be32
4032 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4033                 struct nfs4_delegation **dp)
4034 {
4035         int flags;
4036         __be32 status = nfserr_bad_stateid;
4037         struct nfs4_delegation *deleg;
4038
4039         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4040         if (deleg == NULL)
4041                 goto out;
4042         flags = share_access_to_flags(open->op_share_access);
4043         status = nfs4_check_delegmode(deleg, flags);
4044         if (status) {
4045                 nfs4_put_stid(&deleg->dl_stid);
4046                 goto out;
4047         }
4048         *dp = deleg;
4049 out:
4050         if (!nfsd4_is_deleg_cur(open))
4051                 return nfs_ok;
4052         if (status)
4053                 return status;
4054         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4055         return nfs_ok;
4056 }
4057
4058 static inline int nfs4_access_to_access(u32 nfs4_access)
4059 {
4060         int flags = 0;
4061
4062         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4063                 flags |= NFSD_MAY_READ;
4064         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4065                 flags |= NFSD_MAY_WRITE;
4066         return flags;
4067 }
4068
4069 static inline __be32
4070 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4071                 struct nfsd4_open *open)
4072 {
4073         struct iattr iattr = {
4074                 .ia_valid = ATTR_SIZE,
4075                 .ia_size = 0,
4076         };
4077         if (!open->op_truncate)
4078                 return 0;
4079         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4080                 return nfserr_inval;
4081         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4082 }
4083
4084 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4085                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4086                 struct nfsd4_open *open)
4087 {
4088         struct file *filp = NULL;
4089         __be32 status;
4090         int oflag = nfs4_access_to_omode(open->op_share_access);
4091         int access = nfs4_access_to_access(open->op_share_access);
4092         unsigned char old_access_bmap, old_deny_bmap;
4093
4094         spin_lock(&fp->fi_lock);
4095
4096         /*
4097          * Are we trying to set a deny mode that would conflict with
4098          * current access?
4099          */
4100         status = nfs4_file_check_deny(fp, open->op_share_deny);
4101         if (status != nfs_ok) {
4102                 spin_unlock(&fp->fi_lock);
4103                 goto out;
4104         }
4105
4106         /* set access to the file */
4107         status = nfs4_file_get_access(fp, open->op_share_access);
4108         if (status != nfs_ok) {
4109                 spin_unlock(&fp->fi_lock);
4110                 goto out;
4111         }
4112
4113         /* Set access bits in stateid */
4114         old_access_bmap = stp->st_access_bmap;
4115         set_access(open->op_share_access, stp);
4116
4117         /* Set new deny mask */
4118         old_deny_bmap = stp->st_deny_bmap;
4119         set_deny(open->op_share_deny, stp);
4120         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4121
4122         if (!fp->fi_fds[oflag]) {
4123                 spin_unlock(&fp->fi_lock);
4124                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4125                 if (status)
4126                         goto out_put_access;
4127                 spin_lock(&fp->fi_lock);
4128                 if (!fp->fi_fds[oflag]) {
4129                         fp->fi_fds[oflag] = filp;
4130                         filp = NULL;
4131                 }
4132         }
4133         spin_unlock(&fp->fi_lock);
4134         if (filp)
4135                 fput(filp);
4136
4137         status = nfsd4_truncate(rqstp, cur_fh, open);
4138         if (status)
4139                 goto out_put_access;
4140 out:
4141         return status;
4142 out_put_access:
4143         stp->st_access_bmap = old_access_bmap;
4144         nfs4_file_put_access(fp, open->op_share_access);
4145         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4146         goto out;
4147 }
4148
4149 static __be32
4150 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)
4151 {
4152         __be32 status;
4153         unsigned char old_deny_bmap = stp->st_deny_bmap;
4154
4155         if (!test_access(open->op_share_access, stp))
4156                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4157
4158         /* test and set deny mode */
4159         spin_lock(&fp->fi_lock);
4160         status = nfs4_file_check_deny(fp, open->op_share_deny);
4161         if (status == nfs_ok) {
4162                 set_deny(open->op_share_deny, stp);
4163                 fp->fi_share_deny |=
4164                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4165         }
4166         spin_unlock(&fp->fi_lock);
4167
4168         if (status != nfs_ok)
4169                 return status;
4170
4171         status = nfsd4_truncate(rqstp, cur_fh, open);
4172         if (status != nfs_ok)
4173                 reset_union_bmap_deny(old_deny_bmap, stp);
4174         return status;
4175 }
4176
4177 /* Should we give out recallable state?: */
4178 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4179 {
4180         if (clp->cl_cb_state == NFSD4_CB_UP)
4181                 return true;
4182         /*
4183          * In the sessions case, since we don't have to establish a
4184          * separate connection for callbacks, we assume it's OK
4185          * until we hear otherwise:
4186          */
4187         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4188 }
4189
4190 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4191 {
4192         struct file_lock *fl;
4193
4194         fl = locks_alloc_lock();
4195         if (!fl)
4196                 return NULL;
4197         fl->fl_lmops = &nfsd_lease_mng_ops;
4198         fl->fl_flags = FL_DELEG;
4199         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4200         fl->fl_end = OFFSET_MAX;
4201         fl->fl_owner = (fl_owner_t)fp;
4202         fl->fl_pid = current->tgid;
4203         return fl;
4204 }
4205
4206 /**
4207  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4208  * @dp:   a pointer to the nfs4_delegation we're adding.
4209  *
4210  * Return:
4211  *      On success: Return code will be 0 on success.
4212  *
4213  *      On error: -EAGAIN if there was an existing delegation.
4214  *                 nonzero if there is an error in other cases.
4215  *
4216  */
4217
4218 static int nfs4_setlease(struct nfs4_delegation *dp)
4219 {
4220         struct nfs4_file *fp = dp->dl_stid.sc_file;
4221         struct file_lock *fl;
4222         struct file *filp;
4223         int status = 0;
4224
4225         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4226         if (!fl)
4227                 return -ENOMEM;
4228         filp = find_readable_file(fp);
4229         if (!filp) {
4230                 /* We should always have a readable file here */
4231                 WARN_ON_ONCE(1);
4232                 locks_free_lock(fl);
4233                 return -EBADF;
4234         }
4235         fl->fl_file = filp;
4236         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4237         if (fl)
4238                 locks_free_lock(fl);
4239         if (status)
4240                 goto out_fput;
4241         spin_lock(&state_lock);
4242         spin_lock(&fp->fi_lock);
4243         /* Did the lease get broken before we took the lock? */
4244         status = -EAGAIN;
4245         if (fp->fi_had_conflict)
4246                 goto out_unlock;
4247         /* Race breaker */
4248         if (fp->fi_deleg_file) {
4249                 status = hash_delegation_locked(dp, fp);
4250                 goto out_unlock;
4251         }
4252         fp->fi_deleg_file = filp;
4253         fp->fi_delegees = 0;
4254         status = hash_delegation_locked(dp, fp);
4255         spin_unlock(&fp->fi_lock);
4256         spin_unlock(&state_lock);
4257         if (status) {
4258                 /* Should never happen, this is a new fi_deleg_file  */
4259                 WARN_ON_ONCE(1);
4260                 goto out_fput;
4261         }
4262         return 0;
4263 out_unlock:
4264         spin_unlock(&fp->fi_lock);
4265         spin_unlock(&state_lock);
4266 out_fput:
4267         fput(filp);
4268         return status;
4269 }
4270
4271 static struct nfs4_delegation *
4272 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4273                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4274 {
4275         int status;
4276         struct nfs4_delegation *dp;
4277
4278         if (fp->fi_had_conflict)
4279                 return ERR_PTR(-EAGAIN);
4280
4281         spin_lock(&state_lock);
4282         spin_lock(&fp->fi_lock);
4283         status = nfs4_get_existing_delegation(clp, fp);
4284         spin_unlock(&fp->fi_lock);
4285         spin_unlock(&state_lock);
4286
4287         if (status)
4288                 return ERR_PTR(status);
4289
4290         dp = alloc_init_deleg(clp, fh, odstate);
4291         if (!dp)
4292                 return ERR_PTR(-ENOMEM);
4293
4294         get_nfs4_file(fp);
4295         spin_lock(&state_lock);
4296         spin_lock(&fp->fi_lock);
4297         dp->dl_stid.sc_file = fp;
4298         if (!fp->fi_deleg_file) {
4299                 spin_unlock(&fp->fi_lock);
4300                 spin_unlock(&state_lock);
4301                 status = nfs4_setlease(dp);
4302                 goto out;
4303         }
4304         if (fp->fi_had_conflict) {
4305                 status = -EAGAIN;
4306                 goto out_unlock;
4307         }
4308         status = hash_delegation_locked(dp, fp);
4309 out_unlock:
4310         spin_unlock(&fp->fi_lock);
4311         spin_unlock(&state_lock);
4312 out:
4313         if (status) {
4314                 put_clnt_odstate(dp->dl_clnt_odstate);
4315                 nfs4_put_stid(&dp->dl_stid);
4316                 return ERR_PTR(status);
4317         }
4318         return dp;
4319 }
4320
4321 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4322 {
4323         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4324         if (status == -EAGAIN)
4325                 open->op_why_no_deleg = WND4_CONTENTION;
4326         else {
4327                 open->op_why_no_deleg = WND4_RESOURCE;
4328                 switch (open->op_deleg_want) {
4329                 case NFS4_SHARE_WANT_READ_DELEG:
4330                 case NFS4_SHARE_WANT_WRITE_DELEG:
4331                 case NFS4_SHARE_WANT_ANY_DELEG:
4332                         break;
4333                 case NFS4_SHARE_WANT_CANCEL:
4334                         open->op_why_no_deleg = WND4_CANCELLED;
4335                         break;
4336                 case NFS4_SHARE_WANT_NO_DELEG:
4337                         WARN_ON_ONCE(1);
4338                 }
4339         }
4340 }
4341
4342 /*
4343  * Attempt to hand out a delegation.
4344  *
4345  * Note we don't support write delegations, and won't until the vfs has
4346  * proper support for them.
4347  */
4348 static void
4349 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4350                         struct nfs4_ol_stateid *stp)
4351 {
4352         struct nfs4_delegation *dp;
4353         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4354         struct nfs4_client *clp = stp->st_stid.sc_client;
4355         int cb_up;
4356         int status = 0;
4357
4358         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4359         open->op_recall = 0;
4360         switch (open->op_claim_type) {
4361                 case NFS4_OPEN_CLAIM_PREVIOUS:
4362                         if (!cb_up)
4363                                 open->op_recall = 1;
4364                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4365                                 goto out_no_deleg;
4366                         break;
4367                 case NFS4_OPEN_CLAIM_NULL:
4368                 case NFS4_OPEN_CLAIM_FH:
4369                         /*
4370                          * Let's not give out any delegations till everyone's
4371                          * had the chance to reclaim theirs, *and* until
4372                          * NLM locks have all been reclaimed:
4373                          */
4374                         if (locks_in_grace(clp->net))
4375                                 goto out_no_deleg;
4376                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4377                                 goto out_no_deleg;
4378                         /*
4379                          * Also, if the file was opened for write or
4380                          * create, there's a good chance the client's
4381                          * about to write to it, resulting in an
4382                          * immediate recall (since we don't support
4383                          * write delegations):
4384                          */
4385                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4386                                 goto out_no_deleg;
4387                         if (open->op_create == NFS4_OPEN_CREATE)
4388                                 goto out_no_deleg;
4389                         break;
4390                 default:
4391                         goto out_no_deleg;
4392         }
4393         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4394         if (IS_ERR(dp))
4395                 goto out_no_deleg;
4396
4397         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4398
4399         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4400                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4401         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4402         nfs4_put_stid(&dp->dl_stid);
4403         return;
4404 out_no_deleg:
4405         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4406         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4407             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4408                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4409                 open->op_recall = 1;
4410         }
4411
4412         /* 4.1 client asking for a delegation? */
4413         if (open->op_deleg_want)
4414                 nfsd4_open_deleg_none_ext(open, status);
4415         return;
4416 }
4417
4418 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4419                                         struct nfs4_delegation *dp)
4420 {
4421         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4422             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4423                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4424                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4425         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4426                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4427                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4428                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4429         }
4430         /* Otherwise the client must be confused wanting a delegation
4431          * it already has, therefore we don't return
4432          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4433          */
4434 }
4435
4436 __be32
4437 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4438 {
4439         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4440         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4441         struct nfs4_file *fp = NULL;
4442         struct nfs4_ol_stateid *stp = NULL;
4443         struct nfs4_delegation *dp = NULL;
4444         __be32 status;
4445
4446         /*
4447          * Lookup file; if found, lookup stateid and check open request,
4448          * and check for delegations in the process of being recalled.
4449          * If not found, create the nfs4_file struct
4450          */
4451         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4452         if (fp != open->op_file) {
4453                 status = nfs4_check_deleg(cl, open, &dp);
4454                 if (status)
4455                         goto out;
4456                 spin_lock(&fp->fi_lock);
4457                 stp = nfsd4_find_existing_open(fp, open);
4458                 spin_unlock(&fp->fi_lock);
4459         } else {
4460                 open->op_file = NULL;
4461                 status = nfserr_bad_stateid;
4462                 if (nfsd4_is_deleg_cur(open))
4463                         goto out;
4464         }
4465
4466         /*
4467          * OPEN the file, or upgrade an existing OPEN.
4468          * If truncate fails, the OPEN fails.
4469          */
4470         if (stp) {
4471                 /* Stateid was found, this is an OPEN upgrade */
4472                 mutex_lock(&stp->st_mutex);
4473                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4474                 if (status) {
4475                         mutex_unlock(&stp->st_mutex);
4476                         goto out;
4477                 }
4478         } else {
4479                 /* stp is returned locked. */
4480                 stp = init_open_stateid(fp, open);
4481                 /* See if we lost the race to some other thread */
4482                 if (stp->st_access_bmap != 0) {
4483                         status = nfs4_upgrade_open(rqstp, fp, current_fh,
4484                                                 stp, open);
4485                         if (status) {
4486                                 mutex_unlock(&stp->st_mutex);
4487                                 goto out;
4488                         }
4489                         goto upgrade_out;
4490                 }
4491                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4492                 if (status) {
4493                         mutex_unlock(&stp->st_mutex);
4494                         release_open_stateid(stp);
4495                         goto out;
4496                 }
4497
4498                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4499                                                         open->op_odstate);
4500                 if (stp->st_clnt_odstate == open->op_odstate)
4501                         open->op_odstate = NULL;
4502         }
4503 upgrade_out:
4504         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4505         mutex_unlock(&stp->st_mutex);
4506
4507         if (nfsd4_has_session(&resp->cstate)) {
4508                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4509                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4510                         open->op_why_no_deleg = WND4_NOT_WANTED;
4511                         goto nodeleg;
4512                 }
4513         }
4514
4515         /*
4516         * Attempt to hand out a delegation. No error return, because the
4517         * OPEN succeeds even if we fail.
4518         */
4519         nfs4_open_delegation(current_fh, open, stp);
4520 nodeleg:
4521         status = nfs_ok;
4522
4523         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4524                 STATEID_VAL(&stp->st_stid.sc_stateid));
4525 out:
4526         /* 4.1 client trying to upgrade/downgrade delegation? */
4527         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4528             open->op_deleg_want)
4529                 nfsd4_deleg_xgrade_none_ext(open, dp);
4530
4531         if (fp)
4532                 put_nfs4_file(fp);
4533         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4534                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4535         /*
4536         * To finish the open response, we just need to set the rflags.
4537         */
4538         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4539         if (nfsd4_has_session(&resp->cstate))
4540                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4541         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4542                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4543
4544         if (dp)
4545                 nfs4_put_stid(&dp->dl_stid);
4546         if (stp)
4547                 nfs4_put_stid(&stp->st_stid);
4548
4549         return status;
4550 }
4551
4552 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4553                               struct nfsd4_open *open)
4554 {
4555         if (open->op_openowner) {
4556                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4557
4558                 nfsd4_cstate_assign_replay(cstate, so);
4559                 nfs4_put_stateowner(so);
4560         }
4561         if (open->op_file)
4562                 kmem_cache_free(file_slab, open->op_file);
4563         if (open->op_stp)
4564                 nfs4_put_stid(&open->op_stp->st_stid);
4565         if (open->op_odstate)
4566                 kmem_cache_free(odstate_slab, open->op_odstate);
4567 }
4568
4569 __be32
4570 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4571             union nfsd4_op_u *u)
4572 {
4573         clientid_t *clid = &u->renew;
4574         struct nfs4_client *clp;
4575         __be32 status;
4576         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4577
4578         dprintk("process_renew(%08x/%08x): starting\n",
4579                         clid->cl_boot, clid->cl_id);
4580         status = lookup_clientid(clid, cstate, nn);
4581         if (status)
4582                 goto out;
4583         clp = cstate->clp;
4584         status = nfserr_cb_path_down;
4585         if (!list_empty(&clp->cl_delegations)
4586                         && clp->cl_cb_state != NFSD4_CB_UP)
4587                 goto out;
4588         status = nfs_ok;
4589 out:
4590         return status;
4591 }
4592
4593 void
4594 nfsd4_end_grace(struct nfsd_net *nn)
4595 {
4596         /* do nothing if grace period already ended */
4597         if (nn->grace_ended)
4598                 return;
4599
4600         dprintk("NFSD: end of grace period\n");
4601         nn->grace_ended = true;
4602         /*
4603          * If the server goes down again right now, an NFSv4
4604          * client will still be allowed to reclaim after it comes back up,
4605          * even if it hasn't yet had a chance to reclaim state this time.
4606          *
4607          */
4608         nfsd4_record_grace_done(nn);
4609         /*
4610          * At this point, NFSv4 clients can still reclaim.  But if the
4611          * server crashes, any that have not yet reclaimed will be out
4612          * of luck on the next boot.
4613          *
4614          * (NFSv4.1+ clients are considered to have reclaimed once they
4615          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4616          * have reclaimed after their first OPEN.)
4617          */
4618         locks_end_grace(&nn->nfsd4_manager);
4619         /*
4620          * At this point, and once lockd and/or any other containers
4621          * exit their grace period, further reclaims will fail and
4622          * regular locking can resume.
4623          */
4624 }
4625
4626 static time_t
4627 nfs4_laundromat(struct nfsd_net *nn)
4628 {
4629         struct nfs4_client *clp;
4630         struct nfs4_openowner *oo;
4631         struct nfs4_delegation *dp;
4632         struct nfs4_ol_stateid *stp;
4633         struct nfsd4_blocked_lock *nbl;
4634         struct list_head *pos, *next, reaplist;
4635         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4636         time_t t, new_timeo = nn->nfsd4_lease;
4637
4638         dprintk("NFSD: laundromat service - starting\n");
4639         nfsd4_end_grace(nn);
4640         INIT_LIST_HEAD(&reaplist);
4641         spin_lock(&nn->client_lock);
4642         list_for_each_safe(pos, next, &nn->client_lru) {
4643                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4644                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4645                         t = clp->cl_time - cutoff;
4646                         new_timeo = min(new_timeo, t);
4647                         break;
4648                 }
4649                 if (mark_client_expired_locked(clp)) {
4650                         dprintk("NFSD: client in use (clientid %08x)\n",
4651                                 clp->cl_clientid.cl_id);
4652                         continue;
4653                 }
4654                 list_add(&clp->cl_lru, &reaplist);
4655         }
4656         spin_unlock(&nn->client_lock);
4657         list_for_each_safe(pos, next, &reaplist) {
4658                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4659                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4660                         clp->cl_clientid.cl_id);
4661                 list_del_init(&clp->cl_lru);
4662                 expire_client(clp);
4663         }
4664         spin_lock(&state_lock);
4665         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4666                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4667                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4668                         t = dp->dl_time - cutoff;
4669                         new_timeo = min(new_timeo, t);
4670                         break;
4671                 }
4672                 WARN_ON(!unhash_delegation_locked(dp));
4673                 list_add(&dp->dl_recall_lru, &reaplist);
4674         }
4675         spin_unlock(&state_lock);
4676         while (!list_empty(&reaplist)) {
4677                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4678                                         dl_recall_lru);
4679                 list_del_init(&dp->dl_recall_lru);
4680                 revoke_delegation(dp);
4681         }
4682
4683         spin_lock(&nn->client_lock);
4684         while (!list_empty(&nn->close_lru)) {
4685                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4686                                         oo_close_lru);
4687                 if (time_after((unsigned long)oo->oo_time,
4688                                (unsigned long)cutoff)) {
4689                         t = oo->oo_time - cutoff;
4690                         new_timeo = min(new_timeo, t);
4691                         break;
4692                 }
4693                 list_del_init(&oo->oo_close_lru);
4694                 stp = oo->oo_last_closed_stid;
4695                 oo->oo_last_closed_stid = NULL;
4696                 spin_unlock(&nn->client_lock);
4697                 nfs4_put_stid(&stp->st_stid);
4698                 spin_lock(&nn->client_lock);
4699         }
4700         spin_unlock(&nn->client_lock);
4701
4702         /*
4703          * It's possible for a client to try and acquire an already held lock
4704          * that is being held for a long time, and then lose interest in it.
4705          * So, we clean out any un-revisited request after a lease period
4706          * under the assumption that the client is no longer interested.
4707          *
4708          * RFC5661, sec. 9.6 states that the client must not rely on getting
4709          * notifications and must continue to poll for locks, even when the
4710          * server supports them. Thus this shouldn't lead to clients blocking
4711          * indefinitely once the lock does become free.
4712          */
4713         BUG_ON(!list_empty(&reaplist));
4714         spin_lock(&nn->blocked_locks_lock);
4715         while (!list_empty(&nn->blocked_locks_lru)) {
4716                 nbl = list_first_entry(&nn->blocked_locks_lru,
4717                                         struct nfsd4_blocked_lock, nbl_lru);
4718                 if (time_after((unsigned long)nbl->nbl_time,
4719                                (unsigned long)cutoff)) {
4720                         t = nbl->nbl_time - cutoff;
4721                         new_timeo = min(new_timeo, t);
4722                         break;
4723                 }
4724                 list_move(&nbl->nbl_lru, &reaplist);
4725                 list_del_init(&nbl->nbl_list);
4726         }
4727         spin_unlock(&nn->blocked_locks_lock);
4728
4729         while (!list_empty(&reaplist)) {
4730                 nbl = list_first_entry(&nn->blocked_locks_lru,
4731                                         struct nfsd4_blocked_lock, nbl_lru);
4732                 list_del_init(&nbl->nbl_lru);
4733                 posix_unblock_lock(&nbl->nbl_lock);
4734                 free_blocked_lock(nbl);
4735         }
4736
4737         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4738         return new_timeo;
4739 }
4740
4741 static struct workqueue_struct *laundry_wq;
4742 static void laundromat_main(struct work_struct *);
4743
4744 static void
4745 laundromat_main(struct work_struct *laundry)
4746 {
4747         time_t t;
4748         struct delayed_work *dwork = to_delayed_work(laundry);
4749         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4750                                            laundromat_work);
4751
4752         t = nfs4_laundromat(nn);
4753         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4754         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4755 }
4756
4757 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4758 {
4759         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4760                 return nfserr_bad_stateid;
4761         return nfs_ok;
4762 }
4763
4764 static inline int
4765 access_permit_read(struct nfs4_ol_stateid *stp)
4766 {
4767         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4768                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4769                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4770 }
4771
4772 static inline int
4773 access_permit_write(struct nfs4_ol_stateid *stp)
4774 {
4775         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4776                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4777 }
4778
4779 static
4780 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4781 {
4782         __be32 status = nfserr_openmode;
4783
4784         /* For lock stateid's, we test the parent open, not the lock: */
4785         if (stp->st_openstp)
4786                 stp = stp->st_openstp;
4787         if ((flags & WR_STATE) && !access_permit_write(stp))
4788                 goto out;
4789         if ((flags & RD_STATE) && !access_permit_read(stp))
4790                 goto out;
4791         status = nfs_ok;
4792 out:
4793         return status;
4794 }
4795
4796 static inline __be32
4797 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4798 {
4799         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4800                 return nfs_ok;
4801         else if (opens_in_grace(net)) {
4802                 /* Answer in remaining cases depends on existence of
4803                  * conflicting state; so we must wait out the grace period. */
4804                 return nfserr_grace;
4805         } else if (flags & WR_STATE)
4806                 return nfs4_share_conflict(current_fh,
4807                                 NFS4_SHARE_DENY_WRITE);
4808         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4809                 return nfs4_share_conflict(current_fh,
4810                                 NFS4_SHARE_DENY_READ);
4811 }
4812
4813 /*
4814  * Allow READ/WRITE during grace period on recovered state only for files
4815  * that are not able to provide mandatory locking.
4816  */
4817 static inline int
4818 grace_disallows_io(struct net *net, struct inode *inode)
4819 {
4820         return opens_in_grace(net) && mandatory_lock(inode);
4821 }
4822
4823 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4824 {
4825         /*
4826          * When sessions are used the stateid generation number is ignored
4827          * when it is zero.
4828          */
4829         if (has_session && in->si_generation == 0)
4830                 return nfs_ok;
4831
4832         if (in->si_generation == ref->si_generation)
4833                 return nfs_ok;
4834
4835         /* If the client sends us a stateid from the future, it's buggy: */
4836         if (nfsd4_stateid_generation_after(in, ref))
4837                 return nfserr_bad_stateid;
4838         /*
4839          * However, we could see a stateid from the past, even from a
4840          * non-buggy client.  For example, if the client sends a lock
4841          * while some IO is outstanding, the lock may bump si_generation
4842          * while the IO is still in flight.  The client could avoid that
4843          * situation by waiting for responses on all the IO requests,
4844          * but better performance may result in retrying IO that
4845          * receives an old_stateid error if requests are rarely
4846          * reordered in flight:
4847          */
4848         return nfserr_old_stateid;
4849 }
4850
4851 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4852 {
4853         if (ols->st_stateowner->so_is_open_owner &&
4854             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4855                 return nfserr_bad_stateid;
4856         return nfs_ok;
4857 }
4858
4859 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4860 {
4861         struct nfs4_stid *s;
4862         __be32 status = nfserr_bad_stateid;
4863
4864         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4865                 return status;
4866         /* Client debugging aid. */
4867         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4868                 char addr_str[INET6_ADDRSTRLEN];
4869                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4870                                  sizeof(addr_str));
4871                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4872                                         "with incorrect client ID\n", addr_str);
4873                 return status;
4874         }
4875         spin_lock(&cl->cl_lock);
4876         s = find_stateid_locked(cl, stateid);
4877         if (!s)
4878                 goto out_unlock;
4879         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4880         if (status)
4881                 goto out_unlock;
4882         switch (s->sc_type) {
4883         case NFS4_DELEG_STID:
4884                 status = nfs_ok;
4885                 break;
4886         case NFS4_REVOKED_DELEG_STID:
4887                 status = nfserr_deleg_revoked;
4888                 break;
4889         case NFS4_OPEN_STID:
4890         case NFS4_LOCK_STID:
4891                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4892                 break;
4893         default:
4894                 printk("unknown stateid type %x\n", s->sc_type);
4895                 /* Fallthrough */
4896         case NFS4_CLOSED_STID:
4897         case NFS4_CLOSED_DELEG_STID:
4898                 status = nfserr_bad_stateid;
4899         }
4900 out_unlock:
4901         spin_unlock(&cl->cl_lock);
4902         return status;
4903 }
4904
4905 __be32
4906 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4907                      stateid_t *stateid, unsigned char typemask,
4908                      struct nfs4_stid **s, struct nfsd_net *nn)
4909 {
4910         __be32 status;
4911
4912         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4913                 return nfserr_bad_stateid;
4914         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4915         if (status == nfserr_stale_clientid) {
4916                 if (cstate->session)
4917                         return nfserr_bad_stateid;
4918                 return nfserr_stale_stateid;
4919         }
4920         if (status)
4921                 return status;
4922         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4923         if (!*s)
4924                 return nfserr_bad_stateid;
4925         return nfs_ok;
4926 }
4927
4928 static struct file *
4929 nfs4_find_file(struct nfs4_stid *s, int flags)
4930 {
4931         if (!s)
4932                 return NULL;
4933
4934         switch (s->sc_type) {
4935         case NFS4_DELEG_STID:
4936                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4937                         return NULL;
4938                 return get_file(s->sc_file->fi_deleg_file);
4939         case NFS4_OPEN_STID:
4940         case NFS4_LOCK_STID:
4941                 if (flags & RD_STATE)
4942                         return find_readable_file(s->sc_file);
4943                 else
4944                         return find_writeable_file(s->sc_file);
4945                 break;
4946         }
4947
4948         return NULL;
4949 }
4950
4951 static __be32
4952 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4953 {
4954         __be32 status;
4955
4956         status = nfsd4_check_openowner_confirmed(ols);
4957         if (status)
4958                 return status;
4959         return nfs4_check_openmode(ols, flags);
4960 }
4961
4962 static __be32
4963 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4964                 struct file **filpp, bool *tmp_file, int flags)
4965 {
4966         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4967         struct file *file;
4968         __be32 status;
4969
4970         file = nfs4_find_file(s, flags);
4971         if (file) {
4972                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4973                                 acc | NFSD_MAY_OWNER_OVERRIDE);
4974                 if (status) {
4975                         fput(file);
4976                         return status;
4977                 }
4978
4979                 *filpp = file;
4980         } else {
4981                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4982                 if (status)
4983                         return status;
4984
4985                 if (tmp_file)
4986                         *tmp_file = true;
4987         }
4988
4989         return 0;
4990 }
4991
4992 /*
4993  * Checks for stateid operations
4994  */
4995 __be32
4996 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4997                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4998                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4999 {
5000         struct inode *ino = d_inode(fhp->fh_dentry);
5001         struct net *net = SVC_NET(rqstp);
5002         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5003         struct nfs4_stid *s = NULL;
5004         __be32 status;
5005
5006         if (filpp)
5007                 *filpp = NULL;
5008         if (tmp_file)
5009                 *tmp_file = false;
5010
5011         if (grace_disallows_io(net, ino))
5012                 return nfserr_grace;
5013
5014         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5015                 status = check_special_stateids(net, fhp, stateid, flags);
5016                 goto done;
5017         }
5018
5019         status = nfsd4_lookup_stateid(cstate, stateid,
5020                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5021                                 &s, nn);
5022         if (status)
5023                 return status;
5024         status = check_stateid_generation(stateid, &s->sc_stateid,
5025                         nfsd4_has_session(cstate));
5026         if (status)
5027                 goto out;
5028
5029         switch (s->sc_type) {
5030         case NFS4_DELEG_STID:
5031                 status = nfs4_check_delegmode(delegstateid(s), flags);
5032                 break;
5033         case NFS4_OPEN_STID:
5034         case NFS4_LOCK_STID:
5035                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5036                 break;
5037         default:
5038                 status = nfserr_bad_stateid;
5039                 break;
5040         }
5041         if (status)
5042                 goto out;
5043         status = nfs4_check_fh(fhp, s);
5044
5045 done:
5046         if (!status && filpp)
5047                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5048 out:
5049         if (s)
5050                 nfs4_put_stid(s);
5051         return status;
5052 }
5053
5054 /*
5055  * Test if the stateid is valid
5056  */
5057 __be32
5058 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5059                    union nfsd4_op_u *u)
5060 {
5061         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5062         struct nfsd4_test_stateid_id *stateid;
5063         struct nfs4_client *cl = cstate->session->se_client;
5064
5065         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5066                 stateid->ts_id_status =
5067                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5068
5069         return nfs_ok;
5070 }
5071
5072 static __be32
5073 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5074 {
5075         struct nfs4_ol_stateid *stp = openlockstateid(s);
5076         __be32 ret;
5077
5078         mutex_lock(&stp->st_mutex);
5079
5080         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5081         if (ret)
5082                 goto out;
5083
5084         ret = nfserr_locks_held;
5085         if (check_for_locks(stp->st_stid.sc_file,
5086                             lockowner(stp->st_stateowner)))
5087                 goto out;
5088
5089         release_lock_stateid(stp);
5090         ret = nfs_ok;
5091
5092 out:
5093         mutex_unlock(&stp->st_mutex);
5094         nfs4_put_stid(s);
5095         return ret;
5096 }
5097
5098 __be32
5099 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5100                    union nfsd4_op_u *u)
5101 {
5102         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5103         stateid_t *stateid = &free_stateid->fr_stateid;
5104         struct nfs4_stid *s;
5105         struct nfs4_delegation *dp;
5106         struct nfs4_client *cl = cstate->session->se_client;
5107         __be32 ret = nfserr_bad_stateid;
5108
5109         spin_lock(&cl->cl_lock);
5110         s = find_stateid_locked(cl, stateid);
5111         if (!s)
5112                 goto out_unlock;
5113         switch (s->sc_type) {
5114         case NFS4_DELEG_STID:
5115                 ret = nfserr_locks_held;
5116                 break;
5117         case NFS4_OPEN_STID:
5118                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5119                 if (ret)
5120                         break;
5121                 ret = nfserr_locks_held;
5122                 break;
5123         case NFS4_LOCK_STID:
5124                 refcount_inc(&s->sc_count);
5125                 spin_unlock(&cl->cl_lock);
5126                 ret = nfsd4_free_lock_stateid(stateid, s);
5127                 goto out;
5128         case NFS4_REVOKED_DELEG_STID:
5129                 dp = delegstateid(s);
5130                 list_del_init(&dp->dl_recall_lru);
5131                 spin_unlock(&cl->cl_lock);
5132                 nfs4_put_stid(s);
5133                 ret = nfs_ok;
5134                 goto out;
5135         /* Default falls through and returns nfserr_bad_stateid */
5136         }
5137 out_unlock:
5138         spin_unlock(&cl->cl_lock);
5139 out:
5140         return ret;
5141 }
5142
5143 static inline int
5144 setlkflg (int type)
5145 {
5146         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5147                 RD_STATE : WR_STATE;
5148 }
5149
5150 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5151 {
5152         struct svc_fh *current_fh = &cstate->current_fh;
5153         struct nfs4_stateowner *sop = stp->st_stateowner;
5154         __be32 status;
5155
5156         status = nfsd4_check_seqid(cstate, sop, seqid);
5157         if (status)
5158                 return status;
5159         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
5160                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
5161                 /*
5162                  * "Closed" stateid's exist *only* to return
5163                  * nfserr_replay_me from the previous step, and
5164                  * revoked delegations are kept only for free_stateid.
5165                  */
5166                 return nfserr_bad_stateid;
5167         mutex_lock(&stp->st_mutex);
5168         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5169         if (status == nfs_ok)
5170                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5171         if (status != nfs_ok)
5172                 mutex_unlock(&stp->st_mutex);
5173         return status;
5174 }
5175
5176 /*
5177  * Checks for sequence id mutating operations.
5178  */
5179 static __be32
5180 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5181                          stateid_t *stateid, char typemask,
5182                          struct nfs4_ol_stateid **stpp,
5183                          struct nfsd_net *nn)
5184 {
5185         __be32 status;
5186         struct nfs4_stid *s;
5187         struct nfs4_ol_stateid *stp = NULL;
5188
5189         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5190                 seqid, STATEID_VAL(stateid));
5191
5192         *stpp = NULL;
5193         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5194         if (status)
5195                 return status;
5196         stp = openlockstateid(s);
5197         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5198
5199         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5200         if (!status)
5201                 *stpp = stp;
5202         else
5203                 nfs4_put_stid(&stp->st_stid);
5204         return status;
5205 }
5206
5207 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5208                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5209 {
5210         __be32 status;
5211         struct nfs4_openowner *oo;
5212         struct nfs4_ol_stateid *stp;
5213
5214         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5215                                                 NFS4_OPEN_STID, &stp, nn);
5216         if (status)
5217                 return status;
5218         oo = openowner(stp->st_stateowner);
5219         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5220                 mutex_unlock(&stp->st_mutex);
5221                 nfs4_put_stid(&stp->st_stid);
5222                 return nfserr_bad_stateid;
5223         }
5224         *stpp = stp;
5225         return nfs_ok;
5226 }
5227
5228 __be32
5229 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5230                    union nfsd4_op_u *u)
5231 {
5232         struct nfsd4_open_confirm *oc = &u->open_confirm;
5233         __be32 status;
5234         struct nfs4_openowner *oo;
5235         struct nfs4_ol_stateid *stp;
5236         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5237
5238         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5239                         cstate->current_fh.fh_dentry);
5240
5241         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5242         if (status)
5243                 return status;
5244
5245         status = nfs4_preprocess_seqid_op(cstate,
5246                                         oc->oc_seqid, &oc->oc_req_stateid,
5247                                         NFS4_OPEN_STID, &stp, nn);
5248         if (status)
5249                 goto out;
5250         oo = openowner(stp->st_stateowner);
5251         status = nfserr_bad_stateid;
5252         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5253                 mutex_unlock(&stp->st_mutex);
5254                 goto put_stateid;
5255         }
5256         oo->oo_flags |= NFS4_OO_CONFIRMED;
5257         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5258         mutex_unlock(&stp->st_mutex);
5259         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5260                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5261
5262         nfsd4_client_record_create(oo->oo_owner.so_client);
5263         status = nfs_ok;
5264 put_stateid:
5265         nfs4_put_stid(&stp->st_stid);
5266 out:
5267         nfsd4_bump_seqid(cstate, status);
5268         return status;
5269 }
5270
5271 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5272 {
5273         if (!test_access(access, stp))
5274                 return;
5275         nfs4_file_put_access(stp->st_stid.sc_file, access);
5276         clear_access(access, stp);
5277 }
5278
5279 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5280 {
5281         switch (to_access) {
5282         case NFS4_SHARE_ACCESS_READ:
5283                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5284                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5285                 break;
5286         case NFS4_SHARE_ACCESS_WRITE:
5287                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5288                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5289                 break;
5290         case NFS4_SHARE_ACCESS_BOTH:
5291                 break;
5292         default:
5293                 WARN_ON_ONCE(1);
5294         }
5295 }
5296
5297 __be32
5298 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5299                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5300 {
5301         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5302         __be32 status;
5303         struct nfs4_ol_stateid *stp;
5304         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5305
5306         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5307                         cstate->current_fh.fh_dentry);
5308
5309         /* We don't yet support WANT bits: */
5310         if (od->od_deleg_want)
5311                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5312                         od->od_deleg_want);
5313
5314         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5315                                         &od->od_stateid, &stp, nn);
5316         if (status)
5317                 goto out;
5318         status = nfserr_inval;
5319         if (!test_access(od->od_share_access, stp)) {
5320                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5321                         stp->st_access_bmap, od->od_share_access);
5322                 goto put_stateid;
5323         }
5324         if (!test_deny(od->od_share_deny, stp)) {
5325                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5326                         stp->st_deny_bmap, od->od_share_deny);
5327                 goto put_stateid;
5328         }
5329         nfs4_stateid_downgrade(stp, od->od_share_access);
5330         reset_union_bmap_deny(od->od_share_deny, stp);
5331         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5332         status = nfs_ok;
5333 put_stateid:
5334         mutex_unlock(&stp->st_mutex);
5335         nfs4_put_stid(&stp->st_stid);
5336 out:
5337         nfsd4_bump_seqid(cstate, status);
5338         return status;
5339 }
5340
5341 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5342 {
5343         struct nfs4_client *clp = s->st_stid.sc_client;
5344         bool unhashed;
5345         LIST_HEAD(reaplist);
5346
5347         s->st_stid.sc_type = NFS4_CLOSED_STID;
5348         spin_lock(&clp->cl_lock);
5349         unhashed = unhash_open_stateid(s, &reaplist);
5350
5351         if (clp->cl_minorversion) {
5352                 if (unhashed)
5353                         put_ol_stateid_locked(s, &reaplist);
5354                 spin_unlock(&clp->cl_lock);
5355                 free_ol_stateid_reaplist(&reaplist);
5356         } else {
5357                 spin_unlock(&clp->cl_lock);
5358                 free_ol_stateid_reaplist(&reaplist);
5359                 if (unhashed)
5360                         move_to_close_lru(s, clp->net);
5361         }
5362 }
5363
5364 /*
5365  * nfs4_unlock_state() called after encode
5366  */
5367 __be32
5368 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5369                 union nfsd4_op_u *u)
5370 {
5371         struct nfsd4_close *close = &u->close;
5372         __be32 status;
5373         struct nfs4_ol_stateid *stp;
5374         struct net *net = SVC_NET(rqstp);
5375         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5376
5377         dprintk("NFSD: nfsd4_close on file %pd\n",
5378                         cstate->current_fh.fh_dentry);
5379
5380         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5381                                         &close->cl_stateid,
5382                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5383                                         &stp, nn);
5384         nfsd4_bump_seqid(cstate, status);
5385         if (status)
5386                 goto out;
5387         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5388         mutex_unlock(&stp->st_mutex);
5389
5390         nfsd4_close_open_stateid(stp);
5391
5392         /* put reference from nfs4_preprocess_seqid_op */
5393         nfs4_put_stid(&stp->st_stid);
5394 out:
5395         return status;
5396 }
5397
5398 __be32
5399 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5400                   union nfsd4_op_u *u)
5401 {
5402         struct nfsd4_delegreturn *dr = &u->delegreturn;
5403         struct nfs4_delegation *dp;
5404         stateid_t *stateid = &dr->dr_stateid;
5405         struct nfs4_stid *s;
5406         __be32 status;
5407         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5408
5409         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5410                 return status;
5411
5412         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5413         if (status)
5414                 goto out;
5415         dp = delegstateid(s);
5416         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5417         if (status)
5418                 goto put_stateid;
5419
5420         destroy_delegation(dp);
5421 put_stateid:
5422         nfs4_put_stid(&dp->dl_stid);
5423 out:
5424         return status;
5425 }
5426
5427 static inline u64
5428 end_offset(u64 start, u64 len)
5429 {
5430         u64 end;
5431
5432         end = start + len;
5433         return end >= start ? end: NFS4_MAX_UINT64;
5434 }
5435
5436 /* last octet in a range */
5437 static inline u64
5438 last_byte_offset(u64 start, u64 len)
5439 {
5440         u64 end;
5441
5442         WARN_ON_ONCE(!len);
5443         end = start + len;
5444         return end > start ? end - 1: NFS4_MAX_UINT64;
5445 }
5446
5447 /*
5448  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5449  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5450  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5451  * locking, this prevents us from being completely protocol-compliant.  The
5452  * real solution to this problem is to start using unsigned file offsets in
5453  * the VFS, but this is a very deep change!
5454  */
5455 static inline void
5456 nfs4_transform_lock_offset(struct file_lock *lock)
5457 {
5458         if (lock->fl_start < 0)
5459                 lock->fl_start = OFFSET_MAX;
5460         if (lock->fl_end < 0)
5461                 lock->fl_end = OFFSET_MAX;
5462 }
5463
5464 static fl_owner_t
5465 nfsd4_fl_get_owner(fl_owner_t owner)
5466 {
5467         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5468
5469         nfs4_get_stateowner(&lo->lo_owner);
5470         return owner;
5471 }
5472
5473 static void
5474 nfsd4_fl_put_owner(fl_owner_t owner)
5475 {
5476         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5477
5478         if (lo)
5479                 nfs4_put_stateowner(&lo->lo_owner);
5480 }
5481
5482 static void
5483 nfsd4_lm_notify(struct file_lock *fl)
5484 {
5485         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5486         struct net                      *net = lo->lo_owner.so_client->net;
5487         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5488         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5489                                                 struct nfsd4_blocked_lock, nbl_lock);
5490         bool queue = false;
5491
5492         /* An empty list means that something else is going to be using it */
5493         spin_lock(&nn->blocked_locks_lock);
5494         if (!list_empty(&nbl->nbl_list)) {
5495                 list_del_init(&nbl->nbl_list);
5496                 list_del_init(&nbl->nbl_lru);
5497                 queue = true;
5498         }
5499         spin_unlock(&nn->blocked_locks_lock);
5500
5501         if (queue)
5502                 nfsd4_run_cb(&nbl->nbl_cb);
5503 }
5504
5505 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5506         .lm_notify = nfsd4_lm_notify,
5507         .lm_get_owner = nfsd4_fl_get_owner,
5508         .lm_put_owner = nfsd4_fl_put_owner,
5509 };
5510
5511 static inline void
5512 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5513 {
5514         struct nfs4_lockowner *lo;
5515
5516         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5517                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5518                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5519                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5520                 if (!deny->ld_owner.data)
5521                         /* We just don't care that much */
5522                         goto nevermind;
5523                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5524                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5525         } else {
5526 nevermind:
5527                 deny->ld_owner.len = 0;
5528                 deny->ld_owner.data = NULL;
5529                 deny->ld_clientid.cl_boot = 0;
5530                 deny->ld_clientid.cl_id = 0;
5531         }
5532         deny->ld_start = fl->fl_start;
5533         deny->ld_length = NFS4_MAX_UINT64;
5534         if (fl->fl_end != NFS4_MAX_UINT64)
5535                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5536         deny->ld_type = NFS4_READ_LT;
5537         if (fl->fl_type != F_RDLCK)
5538                 deny->ld_type = NFS4_WRITE_LT;
5539 }
5540
5541 static struct nfs4_lockowner *
5542 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5543 {
5544         unsigned int strhashval = ownerstr_hashval(owner);
5545         struct nfs4_stateowner *so;
5546
5547         lockdep_assert_held(&clp->cl_lock);
5548
5549         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5550                             so_strhash) {
5551                 if (so->so_is_open_owner)
5552                         continue;
5553                 if (same_owner_str(so, owner))
5554                         return lockowner(nfs4_get_stateowner(so));
5555         }
5556         return NULL;
5557 }
5558
5559 static struct nfs4_lockowner *
5560 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5561 {
5562         struct nfs4_lockowner *lo;
5563
5564         spin_lock(&clp->cl_lock);
5565         lo = find_lockowner_str_locked(clp, owner);
5566         spin_unlock(&clp->cl_lock);
5567         return lo;
5568 }
5569
5570 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5571 {
5572         unhash_lockowner_locked(lockowner(sop));
5573 }
5574
5575 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5576 {
5577         struct nfs4_lockowner *lo = lockowner(sop);
5578
5579         kmem_cache_free(lockowner_slab, lo);
5580 }
5581
5582 static const struct nfs4_stateowner_operations lockowner_ops = {
5583         .so_unhash =    nfs4_unhash_lockowner,
5584         .so_free =      nfs4_free_lockowner,
5585 };
5586
5587 /*
5588  * Alloc a lock owner structure.
5589  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5590  * occurred.
5591  *
5592  * strhashval = ownerstr_hashval
5593  */
5594 static struct nfs4_lockowner *
5595 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5596                            struct nfs4_ol_stateid *open_stp,
5597                            struct nfsd4_lock *lock)
5598 {
5599         struct nfs4_lockowner *lo, *ret;
5600
5601         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5602         if (!lo)
5603                 return NULL;
5604         INIT_LIST_HEAD(&lo->lo_blocked);
5605         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5606         lo->lo_owner.so_is_open_owner = 0;
5607         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5608         lo->lo_owner.so_ops = &lockowner_ops;
5609         spin_lock(&clp->cl_lock);
5610         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5611         if (ret == NULL) {
5612                 list_add(&lo->lo_owner.so_strhash,
5613                          &clp->cl_ownerstr_hashtbl[strhashval]);
5614                 ret = lo;
5615         } else
5616                 nfs4_free_stateowner(&lo->lo_owner);
5617
5618         spin_unlock(&clp->cl_lock);
5619         return ret;
5620 }
5621
5622 static void
5623 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5624                   struct nfs4_file *fp, struct inode *inode,
5625                   struct nfs4_ol_stateid *open_stp)
5626 {
5627         struct nfs4_client *clp = lo->lo_owner.so_client;
5628
5629         lockdep_assert_held(&clp->cl_lock);
5630
5631         refcount_inc(&stp->st_stid.sc_count);
5632         stp->st_stid.sc_type = NFS4_LOCK_STID;
5633         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5634         get_nfs4_file(fp);
5635         stp->st_stid.sc_file = fp;
5636         stp->st_access_bmap = 0;
5637         stp->st_deny_bmap = open_stp->st_deny_bmap;
5638         stp->st_openstp = open_stp;
5639         mutex_init(&stp->st_mutex);
5640         list_add(&stp->st_locks, &open_stp->st_locks);
5641         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5642         spin_lock(&fp->fi_lock);
5643         list_add(&stp->st_perfile, &fp->fi_stateids);
5644         spin_unlock(&fp->fi_lock);
5645 }
5646
5647 static struct nfs4_ol_stateid *
5648 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5649 {
5650         struct nfs4_ol_stateid *lst;
5651         struct nfs4_client *clp = lo->lo_owner.so_client;
5652
5653         lockdep_assert_held(&clp->cl_lock);
5654
5655         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5656                 if (lst->st_stid.sc_file == fp) {
5657                         refcount_inc(&lst->st_stid.sc_count);
5658                         return lst;
5659                 }
5660         }
5661         return NULL;
5662 }
5663
5664 static struct nfs4_ol_stateid *
5665 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5666                             struct inode *inode, struct nfs4_ol_stateid *ost,
5667                             bool *new)
5668 {
5669         struct nfs4_stid *ns = NULL;
5670         struct nfs4_ol_stateid *lst;
5671         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5672         struct nfs4_client *clp = oo->oo_owner.so_client;
5673
5674         spin_lock(&clp->cl_lock);
5675         lst = find_lock_stateid(lo, fi);
5676         if (lst == NULL) {
5677                 spin_unlock(&clp->cl_lock);
5678                 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5679                 if (ns == NULL)
5680                         return NULL;
5681
5682                 spin_lock(&clp->cl_lock);
5683                 lst = find_lock_stateid(lo, fi);
5684                 if (likely(!lst)) {
5685                         lst = openlockstateid(ns);
5686                         init_lock_stateid(lst, lo, fi, inode, ost);
5687                         ns = NULL;
5688                         *new = true;
5689                 }
5690         }
5691         spin_unlock(&clp->cl_lock);
5692         if (ns)
5693                 nfs4_put_stid(ns);
5694         return lst;
5695 }
5696
5697 static int
5698 check_lock_length(u64 offset, u64 length)
5699 {
5700         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5701                 (length > ~offset)));
5702 }
5703
5704 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5705 {
5706         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5707
5708         lockdep_assert_held(&fp->fi_lock);
5709
5710         if (test_access(access, lock_stp))
5711                 return;
5712         __nfs4_file_get_access(fp, access);
5713         set_access(access, lock_stp);
5714 }
5715
5716 static __be32
5717 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5718                             struct nfs4_ol_stateid *ost,
5719                             struct nfsd4_lock *lock,
5720                             struct nfs4_ol_stateid **plst, bool *new)
5721 {
5722         __be32 status;
5723         struct nfs4_file *fi = ost->st_stid.sc_file;
5724         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5725         struct nfs4_client *cl = oo->oo_owner.so_client;
5726         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5727         struct nfs4_lockowner *lo;
5728         struct nfs4_ol_stateid *lst;
5729         unsigned int strhashval;
5730         bool hashed;
5731
5732         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5733         if (!lo) {
5734                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5735                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5736                 if (lo == NULL)
5737                         return nfserr_jukebox;
5738         } else {
5739                 /* with an existing lockowner, seqids must be the same */
5740                 status = nfserr_bad_seqid;
5741                 if (!cstate->minorversion &&
5742                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5743                         goto out;
5744         }
5745
5746 retry:
5747         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5748         if (lst == NULL) {
5749                 status = nfserr_jukebox;
5750                 goto out;
5751         }
5752
5753         mutex_lock(&lst->st_mutex);
5754
5755         /* See if it's still hashed to avoid race with FREE_STATEID */
5756         spin_lock(&cl->cl_lock);
5757         hashed = !list_empty(&lst->st_perfile);
5758         spin_unlock(&cl->cl_lock);
5759
5760         if (!hashed) {
5761                 mutex_unlock(&lst->st_mutex);
5762                 nfs4_put_stid(&lst->st_stid);
5763                 goto retry;
5764         }
5765         status = nfs_ok;
5766         *plst = lst;
5767 out:
5768         nfs4_put_stateowner(&lo->lo_owner);
5769         return status;
5770 }
5771
5772 /*
5773  *  LOCK operation
5774  */
5775 __be32
5776 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5777            union nfsd4_op_u *u)
5778 {
5779         struct nfsd4_lock *lock = &u->lock;
5780         struct nfs4_openowner *open_sop = NULL;
5781         struct nfs4_lockowner *lock_sop = NULL;
5782         struct nfs4_ol_stateid *lock_stp = NULL;
5783         struct nfs4_ol_stateid *open_stp = NULL;
5784         struct nfs4_file *fp;
5785         struct file *filp = NULL;
5786         struct nfsd4_blocked_lock *nbl = NULL;
5787         struct file_lock *file_lock = NULL;
5788         struct file_lock *conflock = NULL;
5789         __be32 status = 0;
5790         int lkflg;
5791         int err;
5792         bool new = false;
5793         unsigned char fl_type;
5794         unsigned int fl_flags = FL_POSIX;
5795         struct net *net = SVC_NET(rqstp);
5796         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5797
5798         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5799                 (long long) lock->lk_offset,
5800                 (long long) lock->lk_length);
5801
5802         if (check_lock_length(lock->lk_offset, lock->lk_length))
5803                  return nfserr_inval;
5804
5805         if ((status = fh_verify(rqstp, &cstate->current_fh,
5806                                 S_IFREG, NFSD_MAY_LOCK))) {
5807                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5808                 return status;
5809         }
5810
5811         if (lock->lk_is_new) {
5812                 if (nfsd4_has_session(cstate))
5813                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5814                         memcpy(&lock->lk_new_clientid,
5815                                 &cstate->session->se_client->cl_clientid,
5816                                 sizeof(clientid_t));
5817
5818                 status = nfserr_stale_clientid;
5819                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5820                         goto out;
5821
5822                 /* validate and update open stateid and open seqid */
5823                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5824                                         lock->lk_new_open_seqid,
5825                                         &lock->lk_new_open_stateid,
5826                                         &open_stp, nn);
5827                 if (status)
5828                         goto out;
5829                 mutex_unlock(&open_stp->st_mutex);
5830                 open_sop = openowner(open_stp->st_stateowner);
5831                 status = nfserr_bad_stateid;
5832                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5833                                                 &lock->lk_new_clientid))
5834                         goto out;
5835                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5836                                                         &lock_stp, &new);
5837         } else {
5838                 status = nfs4_preprocess_seqid_op(cstate,
5839                                        lock->lk_old_lock_seqid,
5840                                        &lock->lk_old_lock_stateid,
5841                                        NFS4_LOCK_STID, &lock_stp, nn);
5842         }
5843         if (status)
5844                 goto out;
5845         lock_sop = lockowner(lock_stp->st_stateowner);
5846
5847         lkflg = setlkflg(lock->lk_type);
5848         status = nfs4_check_openmode(lock_stp, lkflg);
5849         if (status)
5850                 goto out;
5851
5852         status = nfserr_grace;
5853         if (locks_in_grace(net) && !lock->lk_reclaim)
5854                 goto out;
5855         status = nfserr_no_grace;
5856         if (!locks_in_grace(net) && lock->lk_reclaim)
5857                 goto out;
5858
5859         fp = lock_stp->st_stid.sc_file;
5860         switch (lock->lk_type) {
5861                 case NFS4_READW_LT:
5862                         if (nfsd4_has_session(cstate))
5863                                 fl_flags |= FL_SLEEP;
5864                         /* Fallthrough */
5865                 case NFS4_READ_LT:
5866                         spin_lock(&fp->fi_lock);
5867                         filp = find_readable_file_locked(fp);
5868                         if (filp)
5869                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5870                         spin_unlock(&fp->fi_lock);
5871                         fl_type = F_RDLCK;
5872                         break;
5873                 case NFS4_WRITEW_LT:
5874                         if (nfsd4_has_session(cstate))
5875                                 fl_flags |= FL_SLEEP;
5876                         /* Fallthrough */
5877                 case NFS4_WRITE_LT:
5878                         spin_lock(&fp->fi_lock);
5879                         filp = find_writeable_file_locked(fp);
5880                         if (filp)
5881                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5882                         spin_unlock(&fp->fi_lock);
5883                         fl_type = F_WRLCK;
5884                         break;
5885                 default:
5886                         status = nfserr_inval;
5887                 goto out;
5888         }
5889
5890         if (!filp) {
5891                 status = nfserr_openmode;
5892                 goto out;
5893         }
5894
5895         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
5896         if (!nbl) {
5897                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
5898                 status = nfserr_jukebox;
5899                 goto out;
5900         }
5901
5902         file_lock = &nbl->nbl_lock;
5903         file_lock->fl_type = fl_type;
5904         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5905         file_lock->fl_pid = current->tgid;
5906         file_lock->fl_file = filp;
5907         file_lock->fl_flags = fl_flags;
5908         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5909         file_lock->fl_start = lock->lk_offset;
5910         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5911         nfs4_transform_lock_offset(file_lock);
5912
5913         conflock = locks_alloc_lock();
5914         if (!conflock) {
5915                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5916                 status = nfserr_jukebox;
5917                 goto out;
5918         }
5919
5920         if (fl_flags & FL_SLEEP) {
5921                 nbl->nbl_time = jiffies;
5922                 spin_lock(&nn->blocked_locks_lock);
5923                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
5924                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
5925                 spin_unlock(&nn->blocked_locks_lock);
5926         }
5927
5928         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5929         switch (err) {
5930         case 0: /* success! */
5931                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5932                 status = 0;
5933                 break;
5934         case FILE_LOCK_DEFERRED:
5935                 nbl = NULL;
5936                 /* Fallthrough */
5937         case -EAGAIN:           /* conflock holds conflicting lock */
5938                 status = nfserr_denied;
5939                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5940                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5941                 break;
5942         case -EDEADLK:
5943                 status = nfserr_deadlock;
5944                 break;
5945         default:
5946                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5947                 status = nfserrno(err);
5948                 break;
5949         }
5950 out:
5951         if (nbl) {
5952                 /* dequeue it if we queued it before */
5953                 if (fl_flags & FL_SLEEP) {
5954                         spin_lock(&nn->blocked_locks_lock);
5955                         list_del_init(&nbl->nbl_list);
5956                         list_del_init(&nbl->nbl_lru);
5957                         spin_unlock(&nn->blocked_locks_lock);
5958                 }
5959                 free_blocked_lock(nbl);
5960         }
5961         if (filp)
5962                 fput(filp);
5963         if (lock_stp) {
5964                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5965                 if (cstate->replay_owner &&
5966                     cstate->replay_owner != &lock_sop->lo_owner &&
5967                     seqid_mutating_err(ntohl(status)))
5968                         lock_sop->lo_owner.so_seqid++;
5969
5970                 mutex_unlock(&lock_stp->st_mutex);
5971
5972                 /*
5973                  * If this is a new, never-before-used stateid, and we are
5974                  * returning an error, then just go ahead and release it.
5975                  */
5976                 if (status && new)
5977                         release_lock_stateid(lock_stp);
5978
5979                 nfs4_put_stid(&lock_stp->st_stid);
5980         }
5981         if (open_stp)
5982                 nfs4_put_stid(&open_stp->st_stid);
5983         nfsd4_bump_seqid(cstate, status);
5984         if (conflock)
5985                 locks_free_lock(conflock);
5986         return status;
5987 }
5988
5989 /*
5990  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5991  * so we do a temporary open here just to get an open file to pass to
5992  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5993  * inode operation.)
5994  */
5995 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5996 {
5997         struct file *file;
5998         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5999         if (!err) {
6000                 err = nfserrno(vfs_test_lock(file, lock));
6001                 fput(file);
6002         }
6003         return err;
6004 }
6005
6006 /*
6007  * LOCKT operation
6008  */
6009 __be32
6010 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6011             union nfsd4_op_u *u)
6012 {
6013         struct nfsd4_lockt *lockt = &u->lockt;
6014         struct file_lock *file_lock = NULL;
6015         struct nfs4_lockowner *lo = NULL;
6016         __be32 status;
6017         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6018
6019         if (locks_in_grace(SVC_NET(rqstp)))
6020                 return nfserr_grace;
6021
6022         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6023                  return nfserr_inval;
6024
6025         if (!nfsd4_has_session(cstate)) {
6026                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6027                 if (status)
6028                         goto out;
6029         }
6030
6031         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6032                 goto out;
6033
6034         file_lock = locks_alloc_lock();
6035         if (!file_lock) {
6036                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6037                 status = nfserr_jukebox;
6038                 goto out;
6039         }
6040
6041         switch (lockt->lt_type) {
6042                 case NFS4_READ_LT:
6043                 case NFS4_READW_LT:
6044                         file_lock->fl_type = F_RDLCK;
6045                 break;
6046                 case NFS4_WRITE_LT:
6047                 case NFS4_WRITEW_LT:
6048                         file_lock->fl_type = F_WRLCK;
6049                 break;
6050                 default:
6051                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6052                         status = nfserr_inval;
6053                 goto out;
6054         }
6055
6056         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6057         if (lo)
6058                 file_lock->fl_owner = (fl_owner_t)lo;
6059         file_lock->fl_pid = current->tgid;
6060         file_lock->fl_flags = FL_POSIX;
6061
6062         file_lock->fl_start = lockt->lt_offset;
6063         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6064
6065         nfs4_transform_lock_offset(file_lock);
6066
6067         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6068         if (status)
6069                 goto out;
6070
6071         if (file_lock->fl_type != F_UNLCK) {
6072                 status = nfserr_denied;
6073                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6074         }
6075 out:
6076         if (lo)
6077                 nfs4_put_stateowner(&lo->lo_owner);
6078         if (file_lock)
6079                 locks_free_lock(file_lock);
6080         return status;
6081 }
6082
6083 __be32
6084 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6085             union nfsd4_op_u *u)
6086 {
6087         struct nfsd4_locku *locku = &u->locku;
6088         struct nfs4_ol_stateid *stp;
6089         struct file *filp = NULL;
6090         struct file_lock *file_lock = NULL;
6091         __be32 status;
6092         int err;
6093         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6094
6095         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6096                 (long long) locku->lu_offset,
6097                 (long long) locku->lu_length);
6098
6099         if (check_lock_length(locku->lu_offset, locku->lu_length))
6100                  return nfserr_inval;
6101
6102         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6103                                         &locku->lu_stateid, NFS4_LOCK_STID,
6104                                         &stp, nn);
6105         if (status)
6106                 goto out;
6107         filp = find_any_file(stp->st_stid.sc_file);
6108         if (!filp) {
6109                 status = nfserr_lock_range;
6110                 goto put_stateid;
6111         }
6112         file_lock = locks_alloc_lock();
6113         if (!file_lock) {
6114                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6115                 status = nfserr_jukebox;
6116                 goto fput;
6117         }
6118
6119         file_lock->fl_type = F_UNLCK;
6120         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6121         file_lock->fl_pid = current->tgid;
6122         file_lock->fl_file = filp;
6123         file_lock->fl_flags = FL_POSIX;
6124         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6125         file_lock->fl_start = locku->lu_offset;
6126
6127         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6128                                                 locku->lu_length);
6129         nfs4_transform_lock_offset(file_lock);
6130
6131         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6132         if (err) {
6133                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6134                 goto out_nfserr;
6135         }
6136         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6137 fput:
6138         fput(filp);
6139 put_stateid:
6140         mutex_unlock(&stp->st_mutex);
6141         nfs4_put_stid(&stp->st_stid);
6142 out:
6143         nfsd4_bump_seqid(cstate, status);
6144         if (file_lock)
6145                 locks_free_lock(file_lock);
6146         return status;
6147
6148 out_nfserr:
6149         status = nfserrno(err);
6150         goto fput;
6151 }
6152
6153 /*
6154  * returns
6155  *      true:  locks held by lockowner
6156  *      false: no locks held by lockowner
6157  */
6158 static bool
6159 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6160 {
6161         struct file_lock *fl;
6162         int status = false;
6163         struct file *filp = find_any_file(fp);
6164         struct inode *inode;
6165         struct file_lock_context *flctx;
6166
6167         if (!filp) {
6168                 /* Any valid lock stateid should have some sort of access */
6169                 WARN_ON_ONCE(1);
6170                 return status;
6171         }
6172
6173         inode = file_inode(filp);
6174         flctx = inode->i_flctx;
6175
6176         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6177                 spin_lock(&flctx->flc_lock);
6178                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6179                         if (fl->fl_owner == (fl_owner_t)lowner) {
6180                                 status = true;
6181                                 break;
6182                         }
6183                 }
6184                 spin_unlock(&flctx->flc_lock);
6185         }
6186         fput(filp);
6187         return status;
6188 }
6189
6190 __be32
6191 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6192                         struct nfsd4_compound_state *cstate,
6193                         union nfsd4_op_u *u)
6194 {
6195         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6196         clientid_t *clid = &rlockowner->rl_clientid;
6197         struct nfs4_stateowner *sop;
6198         struct nfs4_lockowner *lo = NULL;
6199         struct nfs4_ol_stateid *stp;
6200         struct xdr_netobj *owner = &rlockowner->rl_owner;
6201         unsigned int hashval = ownerstr_hashval(owner);
6202         __be32 status;
6203         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6204         struct nfs4_client *clp;
6205         LIST_HEAD (reaplist);
6206
6207         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6208                 clid->cl_boot, clid->cl_id);
6209
6210         status = lookup_clientid(clid, cstate, nn);
6211         if (status)
6212                 return status;
6213
6214         clp = cstate->clp;
6215         /* Find the matching lock stateowner */
6216         spin_lock(&clp->cl_lock);
6217         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6218                             so_strhash) {
6219
6220                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6221                         continue;
6222
6223                 /* see if there are still any locks associated with it */
6224                 lo = lockowner(sop);
6225                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6226                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6227                                 status = nfserr_locks_held;
6228                                 spin_unlock(&clp->cl_lock);
6229                                 return status;
6230                         }
6231                 }
6232
6233                 nfs4_get_stateowner(sop);
6234                 break;
6235         }
6236         if (!lo) {
6237                 spin_unlock(&clp->cl_lock);
6238                 return status;
6239         }
6240
6241         unhash_lockowner_locked(lo);
6242         while (!list_empty(&lo->lo_owner.so_stateids)) {
6243                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6244                                        struct nfs4_ol_stateid,
6245                                        st_perstateowner);
6246                 WARN_ON(!unhash_lock_stateid(stp));
6247                 put_ol_stateid_locked(stp, &reaplist);
6248         }
6249         spin_unlock(&clp->cl_lock);
6250         free_ol_stateid_reaplist(&reaplist);
6251         nfs4_put_stateowner(&lo->lo_owner);
6252
6253         return status;
6254 }
6255
6256 static inline struct nfs4_client_reclaim *
6257 alloc_reclaim(void)
6258 {
6259         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6260 }
6261
6262 bool
6263 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6264 {
6265         struct nfs4_client_reclaim *crp;
6266
6267         crp = nfsd4_find_reclaim_client(name, nn);
6268         return (crp && crp->cr_clp);
6269 }
6270
6271 /*
6272  * failure => all reset bets are off, nfserr_no_grace...
6273  */
6274 struct nfs4_client_reclaim *
6275 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6276 {
6277         unsigned int strhashval;
6278         struct nfs4_client_reclaim *crp;
6279
6280         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6281         crp = alloc_reclaim();
6282         if (crp) {
6283                 strhashval = clientstr_hashval(name);
6284                 INIT_LIST_HEAD(&crp->cr_strhash);
6285                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6286                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6287                 crp->cr_clp = NULL;
6288                 nn->reclaim_str_hashtbl_size++;
6289         }
6290         return crp;
6291 }
6292
6293 void
6294 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6295 {
6296         list_del(&crp->cr_strhash);
6297         kfree(crp);
6298         nn->reclaim_str_hashtbl_size--;
6299 }
6300
6301 void
6302 nfs4_release_reclaim(struct nfsd_net *nn)
6303 {
6304         struct nfs4_client_reclaim *crp = NULL;
6305         int i;
6306
6307         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6308                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6309                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6310                                         struct nfs4_client_reclaim, cr_strhash);
6311                         nfs4_remove_reclaim_record(crp, nn);
6312                 }
6313         }
6314         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6315 }
6316
6317 /*
6318  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6319 struct nfs4_client_reclaim *
6320 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6321 {
6322         unsigned int strhashval;
6323         struct nfs4_client_reclaim *crp = NULL;
6324
6325         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6326
6327         strhashval = clientstr_hashval(recdir);
6328         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6329                 if (same_name(crp->cr_recdir, recdir)) {
6330                         return crp;
6331                 }
6332         }
6333         return NULL;
6334 }
6335
6336 /*
6337 * Called from OPEN. Look for clientid in reclaim list.
6338 */
6339 __be32
6340 nfs4_check_open_reclaim(clientid_t *clid,
6341                 struct nfsd4_compound_state *cstate,
6342                 struct nfsd_net *nn)
6343 {
6344         __be32 status;
6345
6346         /* find clientid in conf_id_hashtbl */
6347         status = lookup_clientid(clid, cstate, nn);
6348         if (status)
6349                 return nfserr_reclaim_bad;
6350
6351         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6352                 return nfserr_no_grace;
6353
6354         if (nfsd4_client_record_check(cstate->clp))
6355                 return nfserr_reclaim_bad;
6356
6357         return nfs_ok;
6358 }
6359
6360 #ifdef CONFIG_NFSD_FAULT_INJECTION
6361 static inline void
6362 put_client(struct nfs4_client *clp)
6363 {
6364         atomic_dec(&clp->cl_refcount);
6365 }
6366
6367 static struct nfs4_client *
6368 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6369 {
6370         struct nfs4_client *clp;
6371         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6372                                           nfsd_net_id);
6373
6374         if (!nfsd_netns_ready(nn))
6375                 return NULL;
6376
6377         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6378                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6379                         return clp;
6380         }
6381         return NULL;
6382 }
6383
6384 u64
6385 nfsd_inject_print_clients(void)
6386 {
6387         struct nfs4_client *clp;
6388         u64 count = 0;
6389         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6390                                           nfsd_net_id);
6391         char buf[INET6_ADDRSTRLEN];
6392
6393         if (!nfsd_netns_ready(nn))
6394                 return 0;
6395
6396         spin_lock(&nn->client_lock);
6397         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6398                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6399                 pr_info("NFS Client: %s\n", buf);
6400                 ++count;
6401         }
6402         spin_unlock(&nn->client_lock);
6403
6404         return count;
6405 }
6406
6407 u64
6408 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6409 {
6410         u64 count = 0;
6411         struct nfs4_client *clp;
6412         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6413                                           nfsd_net_id);
6414
6415         if (!nfsd_netns_ready(nn))
6416                 return count;
6417
6418         spin_lock(&nn->client_lock);
6419         clp = nfsd_find_client(addr, addr_size);
6420         if (clp) {
6421                 if (mark_client_expired_locked(clp) == nfs_ok)
6422                         ++count;
6423                 else
6424                         clp = NULL;
6425         }
6426         spin_unlock(&nn->client_lock);
6427
6428         if (clp)
6429                 expire_client(clp);
6430
6431         return count;
6432 }
6433
6434 u64
6435 nfsd_inject_forget_clients(u64 max)
6436 {
6437         u64 count = 0;
6438         struct nfs4_client *clp, *next;
6439         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6440                                                 nfsd_net_id);
6441         LIST_HEAD(reaplist);
6442
6443         if (!nfsd_netns_ready(nn))
6444                 return count;
6445
6446         spin_lock(&nn->client_lock);
6447         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6448                 if (mark_client_expired_locked(clp) == nfs_ok) {
6449                         list_add(&clp->cl_lru, &reaplist);
6450                         if (max != 0 && ++count >= max)
6451                                 break;
6452                 }
6453         }
6454         spin_unlock(&nn->client_lock);
6455
6456         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6457                 expire_client(clp);
6458
6459         return count;
6460 }
6461
6462 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6463                              const char *type)
6464 {
6465         char buf[INET6_ADDRSTRLEN];
6466         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6467         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6468 }
6469
6470 static void
6471 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6472                              struct list_head *collect)
6473 {
6474         struct nfs4_client *clp = lst->st_stid.sc_client;
6475         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6476                                           nfsd_net_id);
6477
6478         if (!collect)
6479                 return;
6480
6481         lockdep_assert_held(&nn->client_lock);
6482         atomic_inc(&clp->cl_refcount);
6483         list_add(&lst->st_locks, collect);
6484 }
6485
6486 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6487                                     struct list_head *collect,
6488                                     bool (*func)(struct nfs4_ol_stateid *))
6489 {
6490         struct nfs4_openowner *oop;
6491         struct nfs4_ol_stateid *stp, *st_next;
6492         struct nfs4_ol_stateid *lst, *lst_next;
6493         u64 count = 0;
6494
6495         spin_lock(&clp->cl_lock);
6496         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6497                 list_for_each_entry_safe(stp, st_next,
6498                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6499                         list_for_each_entry_safe(lst, lst_next,
6500                                         &stp->st_locks, st_locks) {
6501                                 if (func) {
6502                                         if (func(lst))
6503                                                 nfsd_inject_add_lock_to_list(lst,
6504                                                                         collect);
6505                                 }
6506                                 ++count;
6507                                 /*
6508                                  * Despite the fact that these functions deal
6509                                  * with 64-bit integers for "count", we must
6510                                  * ensure that it doesn't blow up the
6511                                  * clp->cl_refcount. Throw a warning if we
6512                                  * start to approach INT_MAX here.
6513                                  */
6514                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6515                                 if (count == max)
6516                                         goto out;
6517                         }
6518                 }
6519         }
6520 out:
6521         spin_unlock(&clp->cl_lock);
6522
6523         return count;
6524 }
6525
6526 static u64
6527 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6528                           u64 max)
6529 {
6530         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6531 }
6532
6533 static u64
6534 nfsd_print_client_locks(struct nfs4_client *clp)
6535 {
6536         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6537         nfsd_print_count(clp, count, "locked files");
6538         return count;
6539 }
6540
6541 u64
6542 nfsd_inject_print_locks(void)
6543 {
6544         struct nfs4_client *clp;
6545         u64 count = 0;
6546         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6547                                                 nfsd_net_id);
6548
6549         if (!nfsd_netns_ready(nn))
6550                 return 0;
6551
6552         spin_lock(&nn->client_lock);
6553         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6554                 count += nfsd_print_client_locks(clp);
6555         spin_unlock(&nn->client_lock);
6556
6557         return count;
6558 }
6559
6560 static void
6561 nfsd_reap_locks(struct list_head *reaplist)
6562 {
6563         struct nfs4_client *clp;
6564         struct nfs4_ol_stateid *stp, *next;
6565
6566         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6567                 list_del_init(&stp->st_locks);
6568                 clp = stp->st_stid.sc_client;
6569                 nfs4_put_stid(&stp->st_stid);
6570                 put_client(clp);
6571         }
6572 }
6573
6574 u64
6575 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6576 {
6577         unsigned int count = 0;
6578         struct nfs4_client *clp;
6579         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6580                                                 nfsd_net_id);
6581         LIST_HEAD(reaplist);
6582
6583         if (!nfsd_netns_ready(nn))
6584                 return count;
6585
6586         spin_lock(&nn->client_lock);
6587         clp = nfsd_find_client(addr, addr_size);
6588         if (clp)
6589                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6590         spin_unlock(&nn->client_lock);
6591         nfsd_reap_locks(&reaplist);
6592         return count;
6593 }
6594
6595 u64
6596 nfsd_inject_forget_locks(u64 max)
6597 {
6598         u64 count = 0;
6599         struct nfs4_client *clp;
6600         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6601                                                 nfsd_net_id);
6602         LIST_HEAD(reaplist);
6603
6604         if (!nfsd_netns_ready(nn))
6605                 return count;
6606
6607         spin_lock(&nn->client_lock);
6608         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6609                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6610                 if (max != 0 && count >= max)
6611                         break;
6612         }
6613         spin_unlock(&nn->client_lock);
6614         nfsd_reap_locks(&reaplist);
6615         return count;
6616 }
6617
6618 static u64
6619 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6620                               struct list_head *collect,
6621                               void (*func)(struct nfs4_openowner *))
6622 {
6623         struct nfs4_openowner *oop, *next;
6624         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6625                                                 nfsd_net_id);
6626         u64 count = 0;
6627
6628         lockdep_assert_held(&nn->client_lock);
6629
6630         spin_lock(&clp->cl_lock);
6631         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6632                 if (func) {
6633                         func(oop);
6634                         if (collect) {
6635                                 atomic_inc(&clp->cl_refcount);
6636                                 list_add(&oop->oo_perclient, collect);
6637                         }
6638                 }
6639                 ++count;
6640                 /*
6641                  * Despite the fact that these functions deal with
6642                  * 64-bit integers for "count", we must ensure that
6643                  * it doesn't blow up the clp->cl_refcount. Throw a
6644                  * warning if we start to approach INT_MAX here.
6645                  */
6646                 WARN_ON_ONCE(count == (INT_MAX / 2));
6647                 if (count == max)
6648                         break;
6649         }
6650         spin_unlock(&clp->cl_lock);
6651
6652         return count;
6653 }
6654
6655 static u64
6656 nfsd_print_client_openowners(struct nfs4_client *clp)
6657 {
6658         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6659
6660         nfsd_print_count(clp, count, "openowners");
6661         return count;
6662 }
6663
6664 static u64
6665 nfsd_collect_client_openowners(struct nfs4_client *clp,
6666                                struct list_head *collect, u64 max)
6667 {
6668         return nfsd_foreach_client_openowner(clp, max, collect,
6669                                                 unhash_openowner_locked);
6670 }
6671
6672 u64
6673 nfsd_inject_print_openowners(void)
6674 {
6675         struct nfs4_client *clp;
6676         u64 count = 0;
6677         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6678                                                 nfsd_net_id);
6679
6680         if (!nfsd_netns_ready(nn))
6681                 return 0;
6682
6683         spin_lock(&nn->client_lock);
6684         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6685                 count += nfsd_print_client_openowners(clp);
6686         spin_unlock(&nn->client_lock);
6687
6688         return count;
6689 }
6690
6691 static void
6692 nfsd_reap_openowners(struct list_head *reaplist)
6693 {
6694         struct nfs4_client *clp;
6695         struct nfs4_openowner *oop, *next;
6696
6697         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6698                 list_del_init(&oop->oo_perclient);
6699                 clp = oop->oo_owner.so_client;
6700                 release_openowner(oop);
6701                 put_client(clp);
6702         }
6703 }
6704
6705 u64
6706 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6707                                      size_t addr_size)
6708 {
6709         unsigned int count = 0;
6710         struct nfs4_client *clp;
6711         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6712                                                 nfsd_net_id);
6713         LIST_HEAD(reaplist);
6714
6715         if (!nfsd_netns_ready(nn))
6716                 return count;
6717
6718         spin_lock(&nn->client_lock);
6719         clp = nfsd_find_client(addr, addr_size);
6720         if (clp)
6721                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6722         spin_unlock(&nn->client_lock);
6723         nfsd_reap_openowners(&reaplist);
6724         return count;
6725 }
6726
6727 u64
6728 nfsd_inject_forget_openowners(u64 max)
6729 {
6730         u64 count = 0;
6731         struct nfs4_client *clp;
6732         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6733                                                 nfsd_net_id);
6734         LIST_HEAD(reaplist);
6735
6736         if (!nfsd_netns_ready(nn))
6737                 return count;
6738
6739         spin_lock(&nn->client_lock);
6740         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6741                 count += nfsd_collect_client_openowners(clp, &reaplist,
6742                                                         max - count);
6743                 if (max != 0 && count >= max)
6744                         break;
6745         }
6746         spin_unlock(&nn->client_lock);
6747         nfsd_reap_openowners(&reaplist);
6748         return count;
6749 }
6750
6751 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6752                                      struct list_head *victims)
6753 {
6754         struct nfs4_delegation *dp, *next;
6755         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6756                                                 nfsd_net_id);
6757         u64 count = 0;
6758
6759         lockdep_assert_held(&nn->client_lock);
6760
6761         spin_lock(&state_lock);
6762         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6763                 if (victims) {
6764                         /*
6765                          * It's not safe to mess with delegations that have a
6766                          * non-zero dl_time. They might have already been broken
6767                          * and could be processed by the laundromat outside of
6768                          * the state_lock. Just leave them be.
6769                          */
6770                         if (dp->dl_time != 0)
6771                                 continue;
6772
6773                         atomic_inc(&clp->cl_refcount);
6774                         WARN_ON(!unhash_delegation_locked(dp));
6775                         list_add(&dp->dl_recall_lru, victims);
6776                 }
6777                 ++count;
6778                 /*
6779                  * Despite the fact that these functions deal with
6780                  * 64-bit integers for "count", we must ensure that
6781                  * it doesn't blow up the clp->cl_refcount. Throw a
6782                  * warning if we start to approach INT_MAX here.
6783                  */
6784                 WARN_ON_ONCE(count == (INT_MAX / 2));
6785                 if (count == max)
6786                         break;
6787         }
6788         spin_unlock(&state_lock);
6789         return count;
6790 }
6791
6792 static u64
6793 nfsd_print_client_delegations(struct nfs4_client *clp)
6794 {
6795         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6796
6797         nfsd_print_count(clp, count, "delegations");
6798         return count;
6799 }
6800
6801 u64
6802 nfsd_inject_print_delegations(void)
6803 {
6804         struct nfs4_client *clp;
6805         u64 count = 0;
6806         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6807                                                 nfsd_net_id);
6808
6809         if (!nfsd_netns_ready(nn))
6810                 return 0;
6811
6812         spin_lock(&nn->client_lock);
6813         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6814                 count += nfsd_print_client_delegations(clp);
6815         spin_unlock(&nn->client_lock);
6816
6817         return count;
6818 }
6819
6820 static void
6821 nfsd_forget_delegations(struct list_head *reaplist)
6822 {
6823         struct nfs4_client *clp;
6824         struct nfs4_delegation *dp, *next;
6825
6826         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6827                 list_del_init(&dp->dl_recall_lru);
6828                 clp = dp->dl_stid.sc_client;
6829                 revoke_delegation(dp);
6830                 put_client(clp);
6831         }
6832 }
6833
6834 u64
6835 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6836                                       size_t addr_size)
6837 {
6838         u64 count = 0;
6839         struct nfs4_client *clp;
6840         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6841                                                 nfsd_net_id);
6842         LIST_HEAD(reaplist);
6843
6844         if (!nfsd_netns_ready(nn))
6845                 return count;
6846
6847         spin_lock(&nn->client_lock);
6848         clp = nfsd_find_client(addr, addr_size);
6849         if (clp)
6850                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6851         spin_unlock(&nn->client_lock);
6852
6853         nfsd_forget_delegations(&reaplist);
6854         return count;
6855 }
6856
6857 u64
6858 nfsd_inject_forget_delegations(u64 max)
6859 {
6860         u64 count = 0;
6861         struct nfs4_client *clp;
6862         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6863                                                 nfsd_net_id);
6864         LIST_HEAD(reaplist);
6865
6866         if (!nfsd_netns_ready(nn))
6867                 return count;
6868
6869         spin_lock(&nn->client_lock);
6870         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6871                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6872                 if (max != 0 && count >= max)
6873                         break;
6874         }
6875         spin_unlock(&nn->client_lock);
6876         nfsd_forget_delegations(&reaplist);
6877         return count;
6878 }
6879
6880 static void
6881 nfsd_recall_delegations(struct list_head *reaplist)
6882 {
6883         struct nfs4_client *clp;
6884         struct nfs4_delegation *dp, *next;
6885
6886         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6887                 list_del_init(&dp->dl_recall_lru);
6888                 clp = dp->dl_stid.sc_client;
6889                 /*
6890                  * We skipped all entries that had a zero dl_time before,
6891                  * so we can now reset the dl_time back to 0. If a delegation
6892                  * break comes in now, then it won't make any difference since
6893                  * we're recalling it either way.
6894                  */
6895                 spin_lock(&state_lock);
6896                 dp->dl_time = 0;
6897                 spin_unlock(&state_lock);
6898                 nfsd_break_one_deleg(dp);
6899                 put_client(clp);
6900         }
6901 }
6902
6903 u64
6904 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6905                                       size_t addr_size)
6906 {
6907         u64 count = 0;
6908         struct nfs4_client *clp;
6909         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6910                                                 nfsd_net_id);
6911         LIST_HEAD(reaplist);
6912
6913         if (!nfsd_netns_ready(nn))
6914                 return count;
6915
6916         spin_lock(&nn->client_lock);
6917         clp = nfsd_find_client(addr, addr_size);
6918         if (clp)
6919                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6920         spin_unlock(&nn->client_lock);
6921
6922         nfsd_recall_delegations(&reaplist);
6923         return count;
6924 }
6925
6926 u64
6927 nfsd_inject_recall_delegations(u64 max)
6928 {
6929         u64 count = 0;
6930         struct nfs4_client *clp, *next;
6931         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6932                                                 nfsd_net_id);
6933         LIST_HEAD(reaplist);
6934
6935         if (!nfsd_netns_ready(nn))
6936                 return count;
6937
6938         spin_lock(&nn->client_lock);
6939         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6940                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6941                 if (max != 0 && ++count >= max)
6942                         break;
6943         }
6944         spin_unlock(&nn->client_lock);
6945         nfsd_recall_delegations(&reaplist);
6946         return count;
6947 }
6948 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6949
6950 /*
6951  * Since the lifetime of a delegation isn't limited to that of an open, a
6952  * client may quite reasonably hang on to a delegation as long as it has
6953  * the inode cached.  This becomes an obvious problem the first time a
6954  * client's inode cache approaches the size of the server's total memory.
6955  *
6956  * For now we avoid this problem by imposing a hard limit on the number
6957  * of delegations, which varies according to the server's memory size.
6958  */
6959 static void
6960 set_max_delegations(void)
6961 {
6962         /*
6963          * Allow at most 4 delegations per megabyte of RAM.  Quick
6964          * estimates suggest that in the worst case (where every delegation
6965          * is for a different inode), a delegation could take about 1.5K,
6966          * giving a worst case usage of about 6% of memory.
6967          */
6968         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6969 }
6970
6971 static int nfs4_state_create_net(struct net *net)
6972 {
6973         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6974         int i;
6975
6976         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6977                         CLIENT_HASH_SIZE, GFP_KERNEL);
6978         if (!nn->conf_id_hashtbl)
6979                 goto err;
6980         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6981                         CLIENT_HASH_SIZE, GFP_KERNEL);
6982         if (!nn->unconf_id_hashtbl)
6983                 goto err_unconf_id;
6984         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6985                         SESSION_HASH_SIZE, GFP_KERNEL);
6986         if (!nn->sessionid_hashtbl)
6987                 goto err_sessionid;
6988
6989         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6990                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6991                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6992         }
6993         for (i = 0; i < SESSION_HASH_SIZE; i++)
6994                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6995         nn->conf_name_tree = RB_ROOT;
6996         nn->unconf_name_tree = RB_ROOT;
6997         INIT_LIST_HEAD(&nn->client_lru);
6998         INIT_LIST_HEAD(&nn->close_lru);
6999         INIT_LIST_HEAD(&nn->del_recall_lru);
7000         spin_lock_init(&nn->client_lock);
7001
7002         spin_lock_init(&nn->blocked_locks_lock);
7003         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7004
7005         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7006         get_net(net);
7007
7008         return 0;
7009
7010 err_sessionid:
7011         kfree(nn->unconf_id_hashtbl);
7012 err_unconf_id:
7013         kfree(nn->conf_id_hashtbl);
7014 err:
7015         return -ENOMEM;
7016 }
7017
7018 static void
7019 nfs4_state_destroy_net(struct net *net)
7020 {
7021         int i;
7022         struct nfs4_client *clp = NULL;
7023         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7024
7025         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7026                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7027                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7028                         destroy_client(clp);
7029                 }
7030         }
7031
7032         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7033                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7034                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7035                         destroy_client(clp);
7036                 }
7037         }
7038
7039         kfree(nn->sessionid_hashtbl);
7040         kfree(nn->unconf_id_hashtbl);
7041         kfree(nn->conf_id_hashtbl);
7042         put_net(net);
7043 }
7044
7045 int
7046 nfs4_state_start_net(struct net *net)
7047 {
7048         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7049         int ret;
7050
7051         ret = nfs4_state_create_net(net);
7052         if (ret)
7053                 return ret;
7054         nn->boot_time = get_seconds();
7055         nn->grace_ended = false;
7056         nn->nfsd4_manager.block_opens = true;
7057         locks_start_grace(net, &nn->nfsd4_manager);
7058         nfsd4_client_tracking_init(net);
7059         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
7060                nn->nfsd4_grace, net);
7061         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7062         return 0;
7063 }
7064
7065 /* initialization to perform when the nfsd service is started: */
7066
7067 int
7068 nfs4_state_start(void)
7069 {
7070         int ret;
7071
7072         ret = set_callback_cred();
7073         if (ret)
7074                 return ret;
7075
7076         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7077         if (laundry_wq == NULL) {
7078                 ret = -ENOMEM;
7079                 goto out_cleanup_cred;
7080         }
7081         ret = nfsd4_create_callback_queue();
7082         if (ret)
7083                 goto out_free_laundry;
7084
7085         set_max_delegations();
7086         return 0;
7087
7088 out_free_laundry:
7089         destroy_workqueue(laundry_wq);
7090 out_cleanup_cred:
7091         cleanup_callback_cred();
7092         return ret;
7093 }
7094
7095 void
7096 nfs4_state_shutdown_net(struct net *net)
7097 {
7098         struct nfs4_delegation *dp = NULL;
7099         struct list_head *pos, *next, reaplist;
7100         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7101         struct nfsd4_blocked_lock *nbl;
7102
7103         cancel_delayed_work_sync(&nn->laundromat_work);
7104         locks_end_grace(&nn->nfsd4_manager);
7105
7106         INIT_LIST_HEAD(&reaplist);
7107         spin_lock(&state_lock);
7108         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7109                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7110                 WARN_ON(!unhash_delegation_locked(dp));
7111                 list_add(&dp->dl_recall_lru, &reaplist);
7112         }
7113         spin_unlock(&state_lock);
7114         list_for_each_safe(pos, next, &reaplist) {
7115                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7116                 list_del_init(&dp->dl_recall_lru);
7117                 put_clnt_odstate(dp->dl_clnt_odstate);
7118                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7119                 nfs4_put_stid(&dp->dl_stid);
7120         }
7121
7122         BUG_ON(!list_empty(&reaplist));
7123         spin_lock(&nn->blocked_locks_lock);
7124         while (!list_empty(&nn->blocked_locks_lru)) {
7125                 nbl = list_first_entry(&nn->blocked_locks_lru,
7126                                         struct nfsd4_blocked_lock, nbl_lru);
7127                 list_move(&nbl->nbl_lru, &reaplist);
7128                 list_del_init(&nbl->nbl_list);
7129         }
7130         spin_unlock(&nn->blocked_locks_lock);
7131
7132         while (!list_empty(&reaplist)) {
7133                 nbl = list_first_entry(&nn->blocked_locks_lru,
7134                                         struct nfsd4_blocked_lock, nbl_lru);
7135                 list_del_init(&nbl->nbl_lru);
7136                 posix_unblock_lock(&nbl->nbl_lock);
7137                 free_blocked_lock(nbl);
7138         }
7139
7140         nfsd4_client_tracking_exit(net);
7141         nfs4_state_destroy_net(net);
7142 }
7143
7144 void
7145 nfs4_state_shutdown(void)
7146 {
7147         destroy_workqueue(laundry_wq);
7148         nfsd4_destroy_callback_queue();
7149         cleanup_callback_cred();
7150 }
7151
7152 static void
7153 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7154 {
7155         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7156                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7157 }
7158
7159 static void
7160 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7161 {
7162         if (cstate->minorversion) {
7163                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7164                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7165         }
7166 }
7167
7168 void
7169 clear_current_stateid(struct nfsd4_compound_state *cstate)
7170 {
7171         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7172 }
7173
7174 /*
7175  * functions to set current state id
7176  */
7177 void
7178 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7179                 union nfsd4_op_u *u)
7180 {
7181         put_stateid(cstate, &u->open_downgrade.od_stateid);
7182 }
7183
7184 void
7185 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7186                 union nfsd4_op_u *u)
7187 {
7188         put_stateid(cstate, &u->open.op_stateid);
7189 }
7190
7191 void
7192 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7193                 union nfsd4_op_u *u)
7194 {
7195         put_stateid(cstate, &u->close.cl_stateid);
7196 }
7197
7198 void
7199 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7200                 union nfsd4_op_u *u)
7201 {
7202         put_stateid(cstate, &u->lock.lk_resp_stateid);
7203 }
7204
7205 /*
7206  * functions to consume current state id
7207  */
7208
7209 void
7210 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7211                 union nfsd4_op_u *u)
7212 {
7213         get_stateid(cstate, &u->open_downgrade.od_stateid);
7214 }
7215
7216 void
7217 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7218                 union nfsd4_op_u *u)
7219 {
7220         get_stateid(cstate, &u->delegreturn.dr_stateid);
7221 }
7222
7223 void
7224 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7225                 union nfsd4_op_u *u)
7226 {
7227         get_stateid(cstate, &u->free_stateid.fr_stateid);
7228 }
7229
7230 void
7231 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7232                 union nfsd4_op_u *u)
7233 {
7234         get_stateid(cstate, &u->setattr.sa_stateid);
7235 }
7236
7237 void
7238 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7239                 union nfsd4_op_u *u)
7240 {
7241         get_stateid(cstate, &u->close.cl_stateid);
7242 }
7243
7244 void
7245 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7246                 union nfsd4_op_u *u)
7247 {
7248         get_stateid(cstate, &u->locku.lu_stateid);
7249 }
7250
7251 void
7252 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7253                 union nfsd4_op_u *u)
7254 {
7255         get_stateid(cstate, &u->read.rd_stateid);
7256 }
7257
7258 void
7259 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7260                 union nfsd4_op_u *u)
7261 {
7262         get_stateid(cstate, &u->write.wr_stateid);
7263 }