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                 atomic_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                 atomic_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 (atomic_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         atomic_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 (!atomic_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         atomic_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 (!atomic_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                 kfree(ses->se_slots[i]);
1444 }
1445
1446 /*
1447  * We don't actually need to cache the rpc and session headers, so we
1448  * can allocate a little less for each slot:
1449  */
1450 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1451 {
1452         u32 size;
1453
1454         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1455                 size = 0;
1456         else
1457                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1458         return size + sizeof(struct nfsd4_slot);
1459 }
1460
1461 /*
1462  * XXX: If we run out of reserved DRC memory we could (up to a point)
1463  * re-negotiate active sessions and reduce their slot usage to make
1464  * room for new connections. For now we just fail the create session.
1465  */
1466 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1467 {
1468         u32 slotsize = slot_bytes(ca);
1469         u32 num = ca->maxreqs;
1470         int avail;
1471
1472         spin_lock(&nfsd_drc_lock);
1473         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1474                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1475         num = min_t(int, num, avail / slotsize);
1476         nfsd_drc_mem_used += num * slotsize;
1477         spin_unlock(&nfsd_drc_lock);
1478
1479         return num;
1480 }
1481
1482 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1483 {
1484         int slotsize = slot_bytes(ca);
1485
1486         spin_lock(&nfsd_drc_lock);
1487         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1488         spin_unlock(&nfsd_drc_lock);
1489 }
1490
1491 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1492                                            struct nfsd4_channel_attrs *battrs)
1493 {
1494         int numslots = fattrs->maxreqs;
1495         int slotsize = slot_bytes(fattrs);
1496         struct nfsd4_session *new;
1497         int mem, i;
1498
1499         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1500                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1501         mem = numslots * sizeof(struct nfsd4_slot *);
1502
1503         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1504         if (!new)
1505                 return NULL;
1506         /* allocate each struct nfsd4_slot and data cache in one piece */
1507         for (i = 0; i < numslots; i++) {
1508                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1509                 if (!new->se_slots[i])
1510                         goto out_free;
1511         }
1512
1513         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1514         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1515
1516         return new;
1517 out_free:
1518         while (i--)
1519                 kfree(new->se_slots[i]);
1520         kfree(new);
1521         return NULL;
1522 }
1523
1524 static void free_conn(struct nfsd4_conn *c)
1525 {
1526         svc_xprt_put(c->cn_xprt);
1527         kfree(c);
1528 }
1529
1530 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1531 {
1532         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1533         struct nfs4_client *clp = c->cn_session->se_client;
1534
1535         spin_lock(&clp->cl_lock);
1536         if (!list_empty(&c->cn_persession)) {
1537                 list_del(&c->cn_persession);
1538                 free_conn(c);
1539         }
1540         nfsd4_probe_callback(clp);
1541         spin_unlock(&clp->cl_lock);
1542 }
1543
1544 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1545 {
1546         struct nfsd4_conn *conn;
1547
1548         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1549         if (!conn)
1550                 return NULL;
1551         svc_xprt_get(rqstp->rq_xprt);
1552         conn->cn_xprt = rqstp->rq_xprt;
1553         conn->cn_flags = flags;
1554         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1555         return conn;
1556 }
1557
1558 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1559 {
1560         conn->cn_session = ses;
1561         list_add(&conn->cn_persession, &ses->se_conns);
1562 }
1563
1564 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1565 {
1566         struct nfs4_client *clp = ses->se_client;
1567
1568         spin_lock(&clp->cl_lock);
1569         __nfsd4_hash_conn(conn, ses);
1570         spin_unlock(&clp->cl_lock);
1571 }
1572
1573 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1574 {
1575         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1576         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1577 }
1578
1579 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1580 {
1581         int ret;
1582
1583         nfsd4_hash_conn(conn, ses);
1584         ret = nfsd4_register_conn(conn);
1585         if (ret)
1586                 /* oops; xprt is already down: */
1587                 nfsd4_conn_lost(&conn->cn_xpt_user);
1588         /* We may have gained or lost a callback channel: */
1589         nfsd4_probe_callback_sync(ses->se_client);
1590 }
1591
1592 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1593 {
1594         u32 dir = NFS4_CDFC4_FORE;
1595
1596         if (cses->flags & SESSION4_BACK_CHAN)
1597                 dir |= NFS4_CDFC4_BACK;
1598         return alloc_conn(rqstp, dir);
1599 }
1600
1601 /* must be called under client_lock */
1602 static void nfsd4_del_conns(struct nfsd4_session *s)
1603 {
1604         struct nfs4_client *clp = s->se_client;
1605         struct nfsd4_conn *c;
1606
1607         spin_lock(&clp->cl_lock);
1608         while (!list_empty(&s->se_conns)) {
1609                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1610                 list_del_init(&c->cn_persession);
1611                 spin_unlock(&clp->cl_lock);
1612
1613                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1614                 free_conn(c);
1615
1616                 spin_lock(&clp->cl_lock);
1617         }
1618         spin_unlock(&clp->cl_lock);
1619 }
1620
1621 static void __free_session(struct nfsd4_session *ses)
1622 {
1623         free_session_slots(ses);
1624         kfree(ses);
1625 }
1626
1627 static void free_session(struct nfsd4_session *ses)
1628 {
1629         nfsd4_del_conns(ses);
1630         nfsd4_put_drc_mem(&ses->se_fchannel);
1631         __free_session(ses);
1632 }
1633
1634 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1635 {
1636         int idx;
1637         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1638
1639         new->se_client = clp;
1640         gen_sessionid(new);
1641
1642         INIT_LIST_HEAD(&new->se_conns);
1643
1644         new->se_cb_seq_nr = 1;
1645         new->se_flags = cses->flags;
1646         new->se_cb_prog = cses->callback_prog;
1647         new->se_cb_sec = cses->cb_sec;
1648         atomic_set(&new->se_ref, 0);
1649         idx = hash_sessionid(&new->se_sessionid);
1650         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1651         spin_lock(&clp->cl_lock);
1652         list_add(&new->se_perclnt, &clp->cl_sessions);
1653         spin_unlock(&clp->cl_lock);
1654
1655         {
1656                 struct sockaddr *sa = svc_addr(rqstp);
1657                 /*
1658                  * This is a little silly; with sessions there's no real
1659                  * use for the callback address.  Use the peer address
1660                  * as a reasonable default for now, but consider fixing
1661                  * the rpc client not to require an address in the
1662                  * future:
1663                  */
1664                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1665                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1666         }
1667 }
1668
1669 /* caller must hold client_lock */
1670 static struct nfsd4_session *
1671 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1672 {
1673         struct nfsd4_session *elem;
1674         int idx;
1675         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1676
1677         lockdep_assert_held(&nn->client_lock);
1678
1679         dump_sessionid(__func__, sessionid);
1680         idx = hash_sessionid(sessionid);
1681         /* Search in the appropriate list */
1682         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1683                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1684                             NFS4_MAX_SESSIONID_LEN)) {
1685                         return elem;
1686                 }
1687         }
1688
1689         dprintk("%s: session not found\n", __func__);
1690         return NULL;
1691 }
1692
1693 static struct nfsd4_session *
1694 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1695                 __be32 *ret)
1696 {
1697         struct nfsd4_session *session;
1698         __be32 status = nfserr_badsession;
1699
1700         session = __find_in_sessionid_hashtbl(sessionid, net);
1701         if (!session)
1702                 goto out;
1703         status = nfsd4_get_session_locked(session);
1704         if (status)
1705                 session = NULL;
1706 out:
1707         *ret = status;
1708         return session;
1709 }
1710
1711 /* caller must hold client_lock */
1712 static void
1713 unhash_session(struct nfsd4_session *ses)
1714 {
1715         struct nfs4_client *clp = ses->se_client;
1716         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1717
1718         lockdep_assert_held(&nn->client_lock);
1719
1720         list_del(&ses->se_hash);
1721         spin_lock(&ses->se_client->cl_lock);
1722         list_del(&ses->se_perclnt);
1723         spin_unlock(&ses->se_client->cl_lock);
1724 }
1725
1726 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1727 static int
1728 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1729 {
1730         /*
1731          * We're assuming the clid was not given out from a boot
1732          * precisely 2^32 (about 136 years) before this one.  That seems
1733          * a safe assumption:
1734          */
1735         if (clid->cl_boot == (u32)nn->boot_time)
1736                 return 0;
1737         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1738                 clid->cl_boot, clid->cl_id, nn->boot_time);
1739         return 1;
1740 }
1741
1742 /*
1743  * XXX Should we use a slab cache ?
1744  * This type of memory management is somewhat inefficient, but we use it
1745  * anyway since SETCLIENTID is not a common operation.
1746  */
1747 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1748 {
1749         struct nfs4_client *clp;
1750         int i;
1751
1752         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1753         if (clp == NULL)
1754                 return NULL;
1755         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1756         if (clp->cl_name.data == NULL)
1757                 goto err_no_name;
1758         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1759                         OWNER_HASH_SIZE, GFP_KERNEL);
1760         if (!clp->cl_ownerstr_hashtbl)
1761                 goto err_no_hashtbl;
1762         for (i = 0; i < OWNER_HASH_SIZE; i++)
1763                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1764         clp->cl_name.len = name.len;
1765         INIT_LIST_HEAD(&clp->cl_sessions);
1766         idr_init(&clp->cl_stateids);
1767         atomic_set(&clp->cl_refcount, 0);
1768         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1769         INIT_LIST_HEAD(&clp->cl_idhash);
1770         INIT_LIST_HEAD(&clp->cl_openowners);
1771         INIT_LIST_HEAD(&clp->cl_delegations);
1772         INIT_LIST_HEAD(&clp->cl_lru);
1773         INIT_LIST_HEAD(&clp->cl_revoked);
1774 #ifdef CONFIG_NFSD_PNFS
1775         INIT_LIST_HEAD(&clp->cl_lo_states);
1776 #endif
1777         spin_lock_init(&clp->cl_lock);
1778         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1779         return clp;
1780 err_no_hashtbl:
1781         kfree(clp->cl_name.data);
1782 err_no_name:
1783         kfree(clp);
1784         return NULL;
1785 }
1786
1787 static void
1788 free_client(struct nfs4_client *clp)
1789 {
1790         while (!list_empty(&clp->cl_sessions)) {
1791                 struct nfsd4_session *ses;
1792                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1793                                 se_perclnt);
1794                 list_del(&ses->se_perclnt);
1795                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1796                 free_session(ses);
1797         }
1798         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1799         free_svc_cred(&clp->cl_cred);
1800         kfree(clp->cl_ownerstr_hashtbl);
1801         kfree(clp->cl_name.data);
1802         idr_destroy(&clp->cl_stateids);
1803         kfree(clp);
1804 }
1805
1806 /* must be called under the client_lock */
1807 static void
1808 unhash_client_locked(struct nfs4_client *clp)
1809 {
1810         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1811         struct nfsd4_session *ses;
1812
1813         lockdep_assert_held(&nn->client_lock);
1814
1815         /* Mark the client as expired! */
1816         clp->cl_time = 0;
1817         /* Make it invisible */
1818         if (!list_empty(&clp->cl_idhash)) {
1819                 list_del_init(&clp->cl_idhash);
1820                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1821                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1822                 else
1823                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1824         }
1825         list_del_init(&clp->cl_lru);
1826         spin_lock(&clp->cl_lock);
1827         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1828                 list_del_init(&ses->se_hash);
1829         spin_unlock(&clp->cl_lock);
1830 }
1831
1832 static void
1833 unhash_client(struct nfs4_client *clp)
1834 {
1835         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1836
1837         spin_lock(&nn->client_lock);
1838         unhash_client_locked(clp);
1839         spin_unlock(&nn->client_lock);
1840 }
1841
1842 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1843 {
1844         if (atomic_read(&clp->cl_refcount))
1845                 return nfserr_jukebox;
1846         unhash_client_locked(clp);
1847         return nfs_ok;
1848 }
1849
1850 static void
1851 __destroy_client(struct nfs4_client *clp)
1852 {
1853         struct nfs4_openowner *oo;
1854         struct nfs4_delegation *dp;
1855         struct list_head reaplist;
1856
1857         INIT_LIST_HEAD(&reaplist);
1858         spin_lock(&state_lock);
1859         while (!list_empty(&clp->cl_delegations)) {
1860                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1861                 WARN_ON(!unhash_delegation_locked(dp));
1862                 list_add(&dp->dl_recall_lru, &reaplist);
1863         }
1864         spin_unlock(&state_lock);
1865         while (!list_empty(&reaplist)) {
1866                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1867                 list_del_init(&dp->dl_recall_lru);
1868                 put_clnt_odstate(dp->dl_clnt_odstate);
1869                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1870                 nfs4_put_stid(&dp->dl_stid);
1871         }
1872         while (!list_empty(&clp->cl_revoked)) {
1873                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1874                 list_del_init(&dp->dl_recall_lru);
1875                 nfs4_put_stid(&dp->dl_stid);
1876         }
1877         while (!list_empty(&clp->cl_openowners)) {
1878                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1879                 nfs4_get_stateowner(&oo->oo_owner);
1880                 release_openowner(oo);
1881         }
1882         nfsd4_return_all_client_layouts(clp);
1883         nfsd4_shutdown_callback(clp);
1884         if (clp->cl_cb_conn.cb_xprt)
1885                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1886         free_client(clp);
1887 }
1888
1889 static void
1890 destroy_client(struct nfs4_client *clp)
1891 {
1892         unhash_client(clp);
1893         __destroy_client(clp);
1894 }
1895
1896 static void expire_client(struct nfs4_client *clp)
1897 {
1898         unhash_client(clp);
1899         nfsd4_client_record_remove(clp);
1900         __destroy_client(clp);
1901 }
1902
1903 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1904 {
1905         memcpy(target->cl_verifier.data, source->data,
1906                         sizeof(target->cl_verifier.data));
1907 }
1908
1909 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1910 {
1911         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1912         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1913 }
1914
1915 int strdup_if_nonnull(char **target, char *source)
1916 {
1917         if (source) {
1918                 *target = kstrdup(source, GFP_KERNEL);
1919                 if (!*target)
1920                         return -ENOMEM;
1921         } else
1922                 *target = NULL;
1923         return 0;
1924 }
1925
1926 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1927 {
1928         int ret;
1929
1930         ret = strdup_if_nonnull(&target->cr_principal, source->cr_principal);
1931         if (ret)
1932                 return ret;
1933         ret = strdup_if_nonnull(&target->cr_raw_principal,
1934                                         source->cr_raw_principal);
1935         if (ret)
1936                 return ret;
1937         target->cr_flavor = source->cr_flavor;
1938         target->cr_uid = source->cr_uid;
1939         target->cr_gid = source->cr_gid;
1940         target->cr_group_info = source->cr_group_info;
1941         get_group_info(target->cr_group_info);
1942         target->cr_gss_mech = source->cr_gss_mech;
1943         if (source->cr_gss_mech)
1944                 gss_mech_get(source->cr_gss_mech);
1945         return 0;
1946 }
1947
1948 static int
1949 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1950 {
1951         if (o1->len < o2->len)
1952                 return -1;
1953         if (o1->len > o2->len)
1954                 return 1;
1955         return memcmp(o1->data, o2->data, o1->len);
1956 }
1957
1958 static int same_name(const char *n1, const char *n2)
1959 {
1960         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1961 }
1962
1963 static int
1964 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1965 {
1966         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1967 }
1968
1969 static int
1970 same_clid(clientid_t *cl1, clientid_t *cl2)
1971 {
1972         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1973 }
1974
1975 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1976 {
1977         int i;
1978
1979         if (g1->ngroups != g2->ngroups)
1980                 return false;
1981         for (i=0; i<g1->ngroups; i++)
1982                 if (!gid_eq(g1->gid[i], g2->gid[i]))
1983                         return false;
1984         return true;
1985 }
1986
1987 /*
1988  * RFC 3530 language requires clid_inuse be returned when the
1989  * "principal" associated with a requests differs from that previously
1990  * used.  We use uid, gid's, and gss principal string as our best
1991  * approximation.  We also don't want to allow non-gss use of a client
1992  * established using gss: in theory cr_principal should catch that
1993  * change, but in practice cr_principal can be null even in the gss case
1994  * since gssd doesn't always pass down a principal string.
1995  */
1996 static bool is_gss_cred(struct svc_cred *cr)
1997 {
1998         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1999         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2000 }
2001
2002
2003 static bool
2004 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2005 {
2006         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2007                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2008                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2009                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2010                 return false;
2011         if (cr1->cr_principal == cr2->cr_principal)
2012                 return true;
2013         if (!cr1->cr_principal || !cr2->cr_principal)
2014                 return false;
2015         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2016 }
2017
2018 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2019 {
2020         struct svc_cred *cr = &rqstp->rq_cred;
2021         u32 service;
2022
2023         if (!cr->cr_gss_mech)
2024                 return false;
2025         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2026         return service == RPC_GSS_SVC_INTEGRITY ||
2027                service == RPC_GSS_SVC_PRIVACY;
2028 }
2029
2030 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2031 {
2032         struct svc_cred *cr = &rqstp->rq_cred;
2033
2034         if (!cl->cl_mach_cred)
2035                 return true;
2036         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2037                 return false;
2038         if (!svc_rqst_integrity_protected(rqstp))
2039                 return false;
2040         if (cl->cl_cred.cr_raw_principal)
2041                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2042                                                 cr->cr_raw_principal);
2043         if (!cr->cr_principal)
2044                 return false;
2045         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2046 }
2047
2048 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2049 {
2050         __be32 verf[2];
2051
2052         /*
2053          * This is opaque to client, so no need to byte-swap. Use
2054          * __force to keep sparse happy
2055          */
2056         verf[0] = (__force __be32)get_seconds();
2057         verf[1] = (__force __be32)nn->clverifier_counter++;
2058         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2059 }
2060
2061 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2062 {
2063         clp->cl_clientid.cl_boot = nn->boot_time;
2064         clp->cl_clientid.cl_id = nn->clientid_counter++;
2065         gen_confirm(clp, nn);
2066 }
2067
2068 static struct nfs4_stid *
2069 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2070 {
2071         struct nfs4_stid *ret;
2072
2073         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2074         if (!ret || !ret->sc_type)
2075                 return NULL;
2076         return ret;
2077 }
2078
2079 static struct nfs4_stid *
2080 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2081 {
2082         struct nfs4_stid *s;
2083
2084         spin_lock(&cl->cl_lock);
2085         s = find_stateid_locked(cl, t);
2086         if (s != NULL) {
2087                 if (typemask & s->sc_type)
2088                         atomic_inc(&s->sc_count);
2089                 else
2090                         s = NULL;
2091         }
2092         spin_unlock(&cl->cl_lock);
2093         return s;
2094 }
2095
2096 static struct nfs4_client *create_client(struct xdr_netobj name,
2097                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2098 {
2099         struct nfs4_client *clp;
2100         struct sockaddr *sa = svc_addr(rqstp);
2101         int ret;
2102         struct net *net = SVC_NET(rqstp);
2103
2104         clp = alloc_client(name);
2105         if (clp == NULL)
2106                 return NULL;
2107
2108         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2109         if (ret) {
2110                 free_client(clp);
2111                 return NULL;
2112         }
2113         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2114         clp->cl_time = get_seconds();
2115         clear_bit(0, &clp->cl_cb_slot_busy);
2116         copy_verf(clp, verf);
2117         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2118         clp->cl_cb_session = NULL;
2119         clp->net = net;
2120         return clp;
2121 }
2122
2123 static void
2124 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2125 {
2126         struct rb_node **new = &(root->rb_node), *parent = NULL;
2127         struct nfs4_client *clp;
2128
2129         while (*new) {
2130                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2131                 parent = *new;
2132
2133                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2134                         new = &((*new)->rb_left);
2135                 else
2136                         new = &((*new)->rb_right);
2137         }
2138
2139         rb_link_node(&new_clp->cl_namenode, parent, new);
2140         rb_insert_color(&new_clp->cl_namenode, root);
2141 }
2142
2143 static struct nfs4_client *
2144 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2145 {
2146         int cmp;
2147         struct rb_node *node = root->rb_node;
2148         struct nfs4_client *clp;
2149
2150         while (node) {
2151                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2152                 cmp = compare_blob(&clp->cl_name, name);
2153                 if (cmp > 0)
2154                         node = node->rb_left;
2155                 else if (cmp < 0)
2156                         node = node->rb_right;
2157                 else
2158                         return clp;
2159         }
2160         return NULL;
2161 }
2162
2163 static void
2164 add_to_unconfirmed(struct nfs4_client *clp)
2165 {
2166         unsigned int idhashval;
2167         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2168
2169         lockdep_assert_held(&nn->client_lock);
2170
2171         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2172         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2173         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2174         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2175         renew_client_locked(clp);
2176 }
2177
2178 static void
2179 move_to_confirmed(struct nfs4_client *clp)
2180 {
2181         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2182         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2183
2184         lockdep_assert_held(&nn->client_lock);
2185
2186         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2187         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2188         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2189         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2190         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2191         renew_client_locked(clp);
2192 }
2193
2194 static struct nfs4_client *
2195 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2196 {
2197         struct nfs4_client *clp;
2198         unsigned int idhashval = clientid_hashval(clid->cl_id);
2199
2200         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2201                 if (same_clid(&clp->cl_clientid, clid)) {
2202                         if ((bool)clp->cl_minorversion != sessions)
2203                                 return NULL;
2204                         renew_client_locked(clp);
2205                         return clp;
2206                 }
2207         }
2208         return NULL;
2209 }
2210
2211 static struct nfs4_client *
2212 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2213 {
2214         struct list_head *tbl = nn->conf_id_hashtbl;
2215
2216         lockdep_assert_held(&nn->client_lock);
2217         return find_client_in_id_table(tbl, clid, sessions);
2218 }
2219
2220 static struct nfs4_client *
2221 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2222 {
2223         struct list_head *tbl = nn->unconf_id_hashtbl;
2224
2225         lockdep_assert_held(&nn->client_lock);
2226         return find_client_in_id_table(tbl, clid, sessions);
2227 }
2228
2229 static bool clp_used_exchangeid(struct nfs4_client *clp)
2230 {
2231         return clp->cl_exchange_flags != 0;
2232 }
2233
2234 static struct nfs4_client *
2235 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2236 {
2237         lockdep_assert_held(&nn->client_lock);
2238         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2239 }
2240
2241 static struct nfs4_client *
2242 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2243 {
2244         lockdep_assert_held(&nn->client_lock);
2245         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2246 }
2247
2248 static void
2249 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2250 {
2251         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2252         struct sockaddr *sa = svc_addr(rqstp);
2253         u32 scopeid = rpc_get_scope_id(sa);
2254         unsigned short expected_family;
2255
2256         /* Currently, we only support tcp and tcp6 for the callback channel */
2257         if (se->se_callback_netid_len == 3 &&
2258             !memcmp(se->se_callback_netid_val, "tcp", 3))
2259                 expected_family = AF_INET;
2260         else if (se->se_callback_netid_len == 4 &&
2261                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2262                 expected_family = AF_INET6;
2263         else
2264                 goto out_err;
2265
2266         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2267                                             se->se_callback_addr_len,
2268                                             (struct sockaddr *)&conn->cb_addr,
2269                                             sizeof(conn->cb_addr));
2270
2271         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2272                 goto out_err;
2273
2274         if (conn->cb_addr.ss_family == AF_INET6)
2275                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2276
2277         conn->cb_prog = se->se_callback_prog;
2278         conn->cb_ident = se->se_callback_ident;
2279         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2280         return;
2281 out_err:
2282         conn->cb_addr.ss_family = AF_UNSPEC;
2283         conn->cb_addrlen = 0;
2284         dprintk("NFSD: this client (clientid %08x/%08x) "
2285                 "will not receive delegations\n",
2286                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2287
2288         return;
2289 }
2290
2291 /*
2292  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2293  */
2294 static void
2295 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2296 {
2297         struct xdr_buf *buf = resp->xdr.buf;
2298         struct nfsd4_slot *slot = resp->cstate.slot;
2299         unsigned int base;
2300
2301         dprintk("--> %s slot %p\n", __func__, slot);
2302
2303         slot->sl_opcnt = resp->opcnt;
2304         slot->sl_status = resp->cstate.status;
2305
2306         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2307         if (nfsd4_not_cached(resp)) {
2308                 slot->sl_datalen = 0;
2309                 return;
2310         }
2311         base = resp->cstate.data_offset;
2312         slot->sl_datalen = buf->len - base;
2313         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2314                 WARN(1, "%s: sessions DRC could not cache compound\n",
2315                      __func__);
2316         return;
2317 }
2318
2319 /*
2320  * Encode the replay sequence operation from the slot values.
2321  * If cachethis is FALSE encode the uncached rep error on the next
2322  * operation which sets resp->p and increments resp->opcnt for
2323  * nfs4svc_encode_compoundres.
2324  *
2325  */
2326 static __be32
2327 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2328                           struct nfsd4_compoundres *resp)
2329 {
2330         struct nfsd4_op *op;
2331         struct nfsd4_slot *slot = resp->cstate.slot;
2332
2333         /* Encode the replayed sequence operation */
2334         op = &args->ops[resp->opcnt - 1];
2335         nfsd4_encode_operation(resp, op);
2336
2337         /* Return nfserr_retry_uncached_rep in next operation. */
2338         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2339                 op = &args->ops[resp->opcnt++];
2340                 op->status = nfserr_retry_uncached_rep;
2341                 nfsd4_encode_operation(resp, op);
2342         }
2343         return op->status;
2344 }
2345
2346 /*
2347  * The sequence operation is not cached because we can use the slot and
2348  * session values.
2349  */
2350 static __be32
2351 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2352                          struct nfsd4_sequence *seq)
2353 {
2354         struct nfsd4_slot *slot = resp->cstate.slot;
2355         struct xdr_stream *xdr = &resp->xdr;
2356         __be32 *p;
2357         __be32 status;
2358
2359         dprintk("--> %s slot %p\n", __func__, slot);
2360
2361         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2362         if (status)
2363                 return status;
2364
2365         p = xdr_reserve_space(xdr, slot->sl_datalen);
2366         if (!p) {
2367                 WARN_ON_ONCE(1);
2368                 return nfserr_serverfault;
2369         }
2370         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2371         xdr_commit_encode(xdr);
2372
2373         resp->opcnt = slot->sl_opcnt;
2374         return slot->sl_status;
2375 }
2376
2377 /*
2378  * Set the exchange_id flags returned by the server.
2379  */
2380 static void
2381 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2382 {
2383 #ifdef CONFIG_NFSD_PNFS
2384         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2385 #else
2386         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2387 #endif
2388
2389         /* Referrals are supported, Migration is not. */
2390         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2391
2392         /* set the wire flags to return to client. */
2393         clid->flags = new->cl_exchange_flags;
2394 }
2395
2396 static bool client_has_openowners(struct nfs4_client *clp)
2397 {
2398         struct nfs4_openowner *oo;
2399
2400         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2401                 if (!list_empty(&oo->oo_owner.so_stateids))
2402                         return true;
2403         }
2404         return false;
2405 }
2406
2407 static bool client_has_state(struct nfs4_client *clp)
2408 {
2409         return client_has_openowners(clp)
2410 #ifdef CONFIG_NFSD_PNFS
2411                 || !list_empty(&clp->cl_lo_states)
2412 #endif
2413                 || !list_empty(&clp->cl_delegations)
2414                 || !list_empty(&clp->cl_sessions);
2415 }
2416
2417 __be32
2418 nfsd4_exchange_id(struct svc_rqst *rqstp,
2419                   struct nfsd4_compound_state *cstate,
2420                   struct nfsd4_exchange_id *exid)
2421 {
2422         struct nfs4_client *conf, *new;
2423         struct nfs4_client *unconf = NULL;
2424         __be32 status;
2425         char                    addr_str[INET6_ADDRSTRLEN];
2426         nfs4_verifier           verf = exid->verifier;
2427         struct sockaddr         *sa = svc_addr(rqstp);
2428         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2429         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2430
2431         rpc_ntop(sa, addr_str, sizeof(addr_str));
2432         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2433                 "ip_addr=%s flags %x, spa_how %d\n",
2434                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2435                 addr_str, exid->flags, exid->spa_how);
2436
2437         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2438                 return nfserr_inval;
2439
2440         new = create_client(exid->clname, rqstp, &verf);
2441         if (new == NULL)
2442                 return nfserr_jukebox;
2443
2444         switch (exid->spa_how) {
2445         case SP4_MACH_CRED:
2446                 exid->spo_must_enforce[0] = 0;
2447                 exid->spo_must_enforce[1] = (
2448                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2449                         1 << (OP_EXCHANGE_ID - 32) |
2450                         1 << (OP_CREATE_SESSION - 32) |
2451                         1 << (OP_DESTROY_SESSION - 32) |
2452                         1 << (OP_DESTROY_CLIENTID - 32));
2453
2454                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2455                                         1 << (OP_OPEN_DOWNGRADE) |
2456                                         1 << (OP_LOCKU) |
2457                                         1 << (OP_DELEGRETURN));
2458
2459                 exid->spo_must_allow[1] &= (
2460                                         1 << (OP_TEST_STATEID - 32) |
2461                                         1 << (OP_FREE_STATEID - 32));
2462                 if (!svc_rqst_integrity_protected(rqstp)) {
2463                         status = nfserr_inval;
2464                         goto out_nolock;
2465                 }
2466                 /*
2467                  * Sometimes userspace doesn't give us a principal.
2468                  * Which is a bug, really.  Anyway, we can't enforce
2469                  * MACH_CRED in that case, better to give up now:
2470                  */
2471                 if (!new->cl_cred.cr_principal &&
2472                                         !new->cl_cred.cr_raw_principal) {
2473                         status = nfserr_serverfault;
2474                         goto out_nolock;
2475                 }
2476                 new->cl_mach_cred = true;
2477         case SP4_NONE:
2478                 break;
2479         default:                                /* checked by xdr code */
2480                 WARN_ON_ONCE(1);
2481         case SP4_SSV:
2482                 status = nfserr_encr_alg_unsupp;
2483                 goto out_nolock;
2484         }
2485
2486         /* Cases below refer to rfc 5661 section 18.35.4: */
2487         spin_lock(&nn->client_lock);
2488         conf = find_confirmed_client_by_name(&exid->clname, nn);
2489         if (conf) {
2490                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2491                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2492
2493                 if (update) {
2494                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2495                                 status = nfserr_inval;
2496                                 goto out;
2497                         }
2498                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2499                                 status = nfserr_wrong_cred;
2500                                 goto out;
2501                         }
2502                         if (!creds_match) { /* case 9 */
2503                                 status = nfserr_perm;
2504                                 goto out;
2505                         }
2506                         if (!verfs_match) { /* case 8 */
2507                                 status = nfserr_not_same;
2508                                 goto out;
2509                         }
2510                         /* case 6 */
2511                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2512                         goto out_copy;
2513                 }
2514                 if (!creds_match) { /* case 3 */
2515                         if (client_has_state(conf)) {
2516                                 status = nfserr_clid_inuse;
2517                                 goto out;
2518                         }
2519                         goto out_new;
2520                 }
2521                 if (verfs_match) { /* case 2 */
2522                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2523                         goto out_copy;
2524                 }
2525                 /* case 5, client reboot */
2526                 conf = NULL;
2527                 goto out_new;
2528         }
2529
2530         if (update) { /* case 7 */
2531                 status = nfserr_noent;
2532                 goto out;
2533         }
2534
2535         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2536         if (unconf) /* case 4, possible retry or client restart */
2537                 unhash_client_locked(unconf);
2538
2539         /* case 1 (normal case) */
2540 out_new:
2541         if (conf) {
2542                 status = mark_client_expired_locked(conf);
2543                 if (status)
2544                         goto out;
2545         }
2546         new->cl_minorversion = cstate->minorversion;
2547         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2548         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2549
2550         gen_clid(new, nn);
2551         add_to_unconfirmed(new);
2552         swap(new, conf);
2553 out_copy:
2554         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2555         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2556
2557         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2558         nfsd4_set_ex_flags(conf, exid);
2559
2560         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2561                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2562         status = nfs_ok;
2563
2564 out:
2565         spin_unlock(&nn->client_lock);
2566 out_nolock:
2567         if (new)
2568                 expire_client(new);
2569         if (unconf)
2570                 expire_client(unconf);
2571         return status;
2572 }
2573
2574 static __be32
2575 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2576 {
2577         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2578                 slot_seqid);
2579
2580         /* The slot is in use, and no response has been sent. */
2581         if (slot_inuse) {
2582                 if (seqid == slot_seqid)
2583                         return nfserr_jukebox;
2584                 else
2585                         return nfserr_seq_misordered;
2586         }
2587         /* Note unsigned 32-bit arithmetic handles wraparound: */
2588         if (likely(seqid == slot_seqid + 1))
2589                 return nfs_ok;
2590         if (seqid == slot_seqid)
2591                 return nfserr_replay_cache;
2592         return nfserr_seq_misordered;
2593 }
2594
2595 /*
2596  * Cache the create session result into the create session single DRC
2597  * slot cache by saving the xdr structure. sl_seqid has been set.
2598  * Do this for solo or embedded create session operations.
2599  */
2600 static void
2601 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2602                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2603 {
2604         slot->sl_status = nfserr;
2605         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2606 }
2607
2608 static __be32
2609 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2610                             struct nfsd4_clid_slot *slot)
2611 {
2612         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2613         return slot->sl_status;
2614 }
2615
2616 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2617                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2618                         1 +     /* MIN tag is length with zero, only length */ \
2619                         3 +     /* version, opcount, opcode */ \
2620                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2621                                 /* seqid, slotID, slotID, cache */ \
2622                         4 ) * sizeof(__be32))
2623
2624 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2625                         2 +     /* verifier: AUTH_NULL, length 0 */\
2626                         1 +     /* status */ \
2627                         1 +     /* MIN tag is length with zero, only length */ \
2628                         3 +     /* opcount, opcode, opstatus*/ \
2629                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2630                                 /* seqid, slotID, slotID, slotID, status */ \
2631                         5 ) * sizeof(__be32))
2632
2633 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2634 {
2635         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2636
2637         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2638                 return nfserr_toosmall;
2639         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2640                 return nfserr_toosmall;
2641         ca->headerpadsz = 0;
2642         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2643         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2644         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2645         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2646                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2647         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2648         /*
2649          * Note decreasing slot size below client's request may make it
2650          * difficult for client to function correctly, whereas
2651          * decreasing the number of slots will (just?) affect
2652          * performance.  When short on memory we therefore prefer to
2653          * decrease number of slots instead of their size.  Clients that
2654          * request larger slots than they need will get poor results:
2655          */
2656         ca->maxreqs = nfsd4_get_drc_mem(ca);
2657         if (!ca->maxreqs)
2658                 return nfserr_jukebox;
2659
2660         return nfs_ok;
2661 }
2662
2663 /*
2664  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2665  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2666  */
2667 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2668         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2669
2670 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2671         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2672
2673 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2674                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2675 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2676                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2677                                  sizeof(__be32))
2678
2679 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2680 {
2681         ca->headerpadsz = 0;
2682
2683         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2684                 return nfserr_toosmall;
2685         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2686                 return nfserr_toosmall;
2687         ca->maxresp_cached = 0;
2688         if (ca->maxops < 2)
2689                 return nfserr_toosmall;
2690
2691         return nfs_ok;
2692 }
2693
2694 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2695 {
2696         switch (cbs->flavor) {
2697         case RPC_AUTH_NULL:
2698         case RPC_AUTH_UNIX:
2699                 return nfs_ok;
2700         default:
2701                 /*
2702                  * GSS case: the spec doesn't allow us to return this
2703                  * error.  But it also doesn't allow us not to support
2704                  * GSS.
2705                  * I'd rather this fail hard than return some error the
2706                  * client might think it can already handle:
2707                  */
2708                 return nfserr_encr_alg_unsupp;
2709         }
2710 }
2711
2712 __be32
2713 nfsd4_create_session(struct svc_rqst *rqstp,
2714                      struct nfsd4_compound_state *cstate,
2715                      struct nfsd4_create_session *cr_ses)
2716 {
2717         struct sockaddr *sa = svc_addr(rqstp);
2718         struct nfs4_client *conf, *unconf;
2719         struct nfs4_client *old = NULL;
2720         struct nfsd4_session *new;
2721         struct nfsd4_conn *conn;
2722         struct nfsd4_clid_slot *cs_slot = NULL;
2723         __be32 status = 0;
2724         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2725
2726         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2727                 return nfserr_inval;
2728         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2729         if (status)
2730                 return status;
2731         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2732         if (status)
2733                 return status;
2734         status = check_backchannel_attrs(&cr_ses->back_channel);
2735         if (status)
2736                 goto out_release_drc_mem;
2737         status = nfserr_jukebox;
2738         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2739         if (!new)
2740                 goto out_release_drc_mem;
2741         conn = alloc_conn_from_crses(rqstp, cr_ses);
2742         if (!conn)
2743                 goto out_free_session;
2744
2745         spin_lock(&nn->client_lock);
2746         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2747         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2748         WARN_ON_ONCE(conf && unconf);
2749
2750         if (conf) {
2751                 status = nfserr_wrong_cred;
2752                 if (!nfsd4_mach_creds_match(conf, rqstp))
2753                         goto out_free_conn;
2754                 cs_slot = &conf->cl_cs_slot;
2755                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2756                 if (status) {
2757                         if (status == nfserr_replay_cache)
2758                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2759                         goto out_free_conn;
2760                 }
2761         } else if (unconf) {
2762                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2763                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2764                         status = nfserr_clid_inuse;
2765                         goto out_free_conn;
2766                 }
2767                 status = nfserr_wrong_cred;
2768                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2769                         goto out_free_conn;
2770                 cs_slot = &unconf->cl_cs_slot;
2771                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2772                 if (status) {
2773                         /* an unconfirmed replay returns misordered */
2774                         status = nfserr_seq_misordered;
2775                         goto out_free_conn;
2776                 }
2777                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2778                 if (old) {
2779                         status = mark_client_expired_locked(old);
2780                         if (status) {
2781                                 old = NULL;
2782                                 goto out_free_conn;
2783                         }
2784                 }
2785                 move_to_confirmed(unconf);
2786                 conf = unconf;
2787         } else {
2788                 status = nfserr_stale_clientid;
2789                 goto out_free_conn;
2790         }
2791         status = nfs_ok;
2792         /* Persistent sessions are not supported */
2793         cr_ses->flags &= ~SESSION4_PERSIST;
2794         /* Upshifting from TCP to RDMA is not supported */
2795         cr_ses->flags &= ~SESSION4_RDMA;
2796
2797         init_session(rqstp, new, conf, cr_ses);
2798         nfsd4_get_session_locked(new);
2799
2800         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2801                NFS4_MAX_SESSIONID_LEN);
2802         cs_slot->sl_seqid++;
2803         cr_ses->seqid = cs_slot->sl_seqid;
2804
2805         /* cache solo and embedded create sessions under the client_lock */
2806         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2807         spin_unlock(&nn->client_lock);
2808         /* init connection and backchannel */
2809         nfsd4_init_conn(rqstp, conn, new);
2810         nfsd4_put_session(new);
2811         if (old)
2812                 expire_client(old);
2813         return status;
2814 out_free_conn:
2815         spin_unlock(&nn->client_lock);
2816         free_conn(conn);
2817         if (old)
2818                 expire_client(old);
2819 out_free_session:
2820         __free_session(new);
2821 out_release_drc_mem:
2822         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2823         return status;
2824 }
2825
2826 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2827 {
2828         switch (*dir) {
2829         case NFS4_CDFC4_FORE:
2830         case NFS4_CDFC4_BACK:
2831                 return nfs_ok;
2832         case NFS4_CDFC4_FORE_OR_BOTH:
2833         case NFS4_CDFC4_BACK_OR_BOTH:
2834                 *dir = NFS4_CDFC4_BOTH;
2835                 return nfs_ok;
2836         };
2837         return nfserr_inval;
2838 }
2839
2840 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2841 {
2842         struct nfsd4_session *session = cstate->session;
2843         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2844         __be32 status;
2845
2846         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2847         if (status)
2848                 return status;
2849         spin_lock(&nn->client_lock);
2850         session->se_cb_prog = bc->bc_cb_program;
2851         session->se_cb_sec = bc->bc_cb_sec;
2852         spin_unlock(&nn->client_lock);
2853
2854         nfsd4_probe_callback(session->se_client);
2855
2856         return nfs_ok;
2857 }
2858
2859 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2860                      struct nfsd4_compound_state *cstate,
2861                      struct nfsd4_bind_conn_to_session *bcts)
2862 {
2863         __be32 status;
2864         struct nfsd4_conn *conn;
2865         struct nfsd4_session *session;
2866         struct net *net = SVC_NET(rqstp);
2867         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2868
2869         if (!nfsd4_last_compound_op(rqstp))
2870                 return nfserr_not_only_op;
2871         spin_lock(&nn->client_lock);
2872         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2873         spin_unlock(&nn->client_lock);
2874         if (!session)
2875                 goto out_no_session;
2876         status = nfserr_wrong_cred;
2877         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2878                 goto out;
2879         status = nfsd4_map_bcts_dir(&bcts->dir);
2880         if (status)
2881                 goto out;
2882         conn = alloc_conn(rqstp, bcts->dir);
2883         status = nfserr_jukebox;
2884         if (!conn)
2885                 goto out;
2886         nfsd4_init_conn(rqstp, conn, session);
2887         status = nfs_ok;
2888 out:
2889         nfsd4_put_session(session);
2890 out_no_session:
2891         return status;
2892 }
2893
2894 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2895 {
2896         if (!session)
2897                 return 0;
2898         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2899 }
2900
2901 __be32
2902 nfsd4_destroy_session(struct svc_rqst *r,
2903                       struct nfsd4_compound_state *cstate,
2904                       struct nfsd4_destroy_session *sessionid)
2905 {
2906         struct nfsd4_session *ses;
2907         __be32 status;
2908         int ref_held_by_me = 0;
2909         struct net *net = SVC_NET(r);
2910         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2911
2912         status = nfserr_not_only_op;
2913         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2914                 if (!nfsd4_last_compound_op(r))
2915                         goto out;
2916                 ref_held_by_me++;
2917         }
2918         dump_sessionid(__func__, &sessionid->sessionid);
2919         spin_lock(&nn->client_lock);
2920         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2921         if (!ses)
2922                 goto out_client_lock;
2923         status = nfserr_wrong_cred;
2924         if (!nfsd4_mach_creds_match(ses->se_client, r))
2925                 goto out_put_session;
2926         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2927         if (status)
2928                 goto out_put_session;
2929         unhash_session(ses);
2930         spin_unlock(&nn->client_lock);
2931
2932         nfsd4_probe_callback_sync(ses->se_client);
2933
2934         spin_lock(&nn->client_lock);
2935         status = nfs_ok;
2936 out_put_session:
2937         nfsd4_put_session_locked(ses);
2938 out_client_lock:
2939         spin_unlock(&nn->client_lock);
2940 out:
2941         return status;
2942 }
2943
2944 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2945 {
2946         struct nfsd4_conn *c;
2947
2948         list_for_each_entry(c, &s->se_conns, cn_persession) {
2949                 if (c->cn_xprt == xpt) {
2950                         return c;
2951                 }
2952         }
2953         return NULL;
2954 }
2955
2956 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2957 {
2958         struct nfs4_client *clp = ses->se_client;
2959         struct nfsd4_conn *c;
2960         __be32 status = nfs_ok;
2961         int ret;
2962
2963         spin_lock(&clp->cl_lock);
2964         c = __nfsd4_find_conn(new->cn_xprt, ses);
2965         if (c)
2966                 goto out_free;
2967         status = nfserr_conn_not_bound_to_session;
2968         if (clp->cl_mach_cred)
2969                 goto out_free;
2970         __nfsd4_hash_conn(new, ses);
2971         spin_unlock(&clp->cl_lock);
2972         ret = nfsd4_register_conn(new);
2973         if (ret)
2974                 /* oops; xprt is already down: */
2975                 nfsd4_conn_lost(&new->cn_xpt_user);
2976         return nfs_ok;
2977 out_free:
2978         spin_unlock(&clp->cl_lock);
2979         free_conn(new);
2980         return status;
2981 }
2982
2983 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2984 {
2985         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2986
2987         return args->opcnt > session->se_fchannel.maxops;
2988 }
2989
2990 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2991                                   struct nfsd4_session *session)
2992 {
2993         struct xdr_buf *xb = &rqstp->rq_arg;
2994
2995         return xb->len > session->se_fchannel.maxreq_sz;
2996 }
2997
2998 __be32
2999 nfsd4_sequence(struct svc_rqst *rqstp,
3000                struct nfsd4_compound_state *cstate,
3001                struct nfsd4_sequence *seq)
3002 {
3003         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3004         struct xdr_stream *xdr = &resp->xdr;
3005         struct nfsd4_session *session;
3006         struct nfs4_client *clp;
3007         struct nfsd4_slot *slot;
3008         struct nfsd4_conn *conn;
3009         __be32 status;
3010         int buflen;
3011         struct net *net = SVC_NET(rqstp);
3012         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3013
3014         if (resp->opcnt != 1)
3015                 return nfserr_sequence_pos;
3016
3017         /*
3018          * Will be either used or freed by nfsd4_sequence_check_conn
3019          * below.
3020          */
3021         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3022         if (!conn)
3023                 return nfserr_jukebox;
3024
3025         spin_lock(&nn->client_lock);
3026         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3027         if (!session)
3028                 goto out_no_session;
3029         clp = session->se_client;
3030
3031         status = nfserr_too_many_ops;
3032         if (nfsd4_session_too_many_ops(rqstp, session))
3033                 goto out_put_session;
3034
3035         status = nfserr_req_too_big;
3036         if (nfsd4_request_too_big(rqstp, session))
3037                 goto out_put_session;
3038
3039         status = nfserr_badslot;
3040         if (seq->slotid >= session->se_fchannel.maxreqs)
3041                 goto out_put_session;
3042
3043         slot = session->se_slots[seq->slotid];
3044         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3045
3046         /* We do not negotiate the number of slots yet, so set the
3047          * maxslots to the session maxreqs which is used to encode
3048          * sr_highest_slotid and the sr_target_slot id to maxslots */
3049         seq->maxslots = session->se_fchannel.maxreqs;
3050
3051         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3052                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3053         if (status == nfserr_replay_cache) {
3054                 status = nfserr_seq_misordered;
3055                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3056                         goto out_put_session;
3057                 cstate->slot = slot;
3058                 cstate->session = session;
3059                 cstate->clp = clp;
3060                 /* Return the cached reply status and set cstate->status
3061                  * for nfsd4_proc_compound processing */
3062                 status = nfsd4_replay_cache_entry(resp, seq);
3063                 cstate->status = nfserr_replay_cache;
3064                 goto out;
3065         }
3066         if (status)
3067                 goto out_put_session;
3068
3069         status = nfsd4_sequence_check_conn(conn, session);
3070         conn = NULL;
3071         if (status)
3072                 goto out_put_session;
3073
3074         buflen = (seq->cachethis) ?
3075                         session->se_fchannel.maxresp_cached :
3076                         session->se_fchannel.maxresp_sz;
3077         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3078                                     nfserr_rep_too_big;
3079         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3080                 goto out_put_session;
3081         svc_reserve(rqstp, buflen);
3082
3083         status = nfs_ok;
3084         /* Success! bump slot seqid */
3085         slot->sl_seqid = seq->seqid;
3086         slot->sl_flags |= NFSD4_SLOT_INUSE;
3087         if (seq->cachethis)
3088                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3089         else
3090                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3091
3092         cstate->slot = slot;
3093         cstate->session = session;
3094         cstate->clp = clp;
3095
3096 out:
3097         switch (clp->cl_cb_state) {
3098         case NFSD4_CB_DOWN:
3099                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3100                 break;
3101         case NFSD4_CB_FAULT:
3102                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3103                 break;
3104         default:
3105                 seq->status_flags = 0;
3106         }
3107         if (!list_empty(&clp->cl_revoked))
3108                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3109 out_no_session:
3110         if (conn)
3111                 free_conn(conn);
3112         spin_unlock(&nn->client_lock);
3113         return status;
3114 out_put_session:
3115         nfsd4_put_session_locked(session);
3116         goto out_no_session;
3117 }
3118
3119 void
3120 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3121 {
3122         struct nfsd4_compound_state *cs = &resp->cstate;
3123
3124         if (nfsd4_has_session(cs)) {
3125                 if (cs->status != nfserr_replay_cache) {
3126                         nfsd4_store_cache_entry(resp);
3127                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3128                 }
3129                 /* Drop session reference that was taken in nfsd4_sequence() */
3130                 nfsd4_put_session(cs->session);
3131         } else if (cs->clp)
3132                 put_client_renew(cs->clp);
3133 }
3134
3135 __be32
3136 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3137 {
3138         struct nfs4_client *conf, *unconf;
3139         struct nfs4_client *clp = NULL;
3140         __be32 status = 0;
3141         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3142
3143         spin_lock(&nn->client_lock);
3144         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3145         conf = find_confirmed_client(&dc->clientid, true, nn);
3146         WARN_ON_ONCE(conf && unconf);
3147
3148         if (conf) {
3149                 if (client_has_state(conf)) {
3150                         status = nfserr_clientid_busy;
3151                         goto out;
3152                 }
3153                 status = mark_client_expired_locked(conf);
3154                 if (status)
3155                         goto out;
3156                 clp = conf;
3157         } else if (unconf)
3158                 clp = unconf;
3159         else {
3160                 status = nfserr_stale_clientid;
3161                 goto out;
3162         }
3163         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3164                 clp = NULL;
3165                 status = nfserr_wrong_cred;
3166                 goto out;
3167         }
3168         unhash_client_locked(clp);
3169 out:
3170         spin_unlock(&nn->client_lock);
3171         if (clp)
3172                 expire_client(clp);
3173         return status;
3174 }
3175
3176 __be32
3177 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3178 {
3179         __be32 status = 0;
3180
3181         if (rc->rca_one_fs) {
3182                 if (!cstate->current_fh.fh_dentry)
3183                         return nfserr_nofilehandle;
3184                 /*
3185                  * We don't take advantage of the rca_one_fs case.
3186                  * That's OK, it's optional, we can safely ignore it.
3187                  */
3188                 return nfs_ok;
3189         }
3190
3191         status = nfserr_complete_already;
3192         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3193                              &cstate->session->se_client->cl_flags))
3194                 goto out;
3195
3196         status = nfserr_stale_clientid;
3197         if (is_client_expired(cstate->session->se_client))
3198                 /*
3199                  * The following error isn't really legal.
3200                  * But we only get here if the client just explicitly
3201                  * destroyed the client.  Surely it no longer cares what
3202                  * error it gets back on an operation for the dead
3203                  * client.
3204                  */
3205                 goto out;
3206
3207         status = nfs_ok;
3208         nfsd4_client_record_create(cstate->session->se_client);
3209 out:
3210         return status;
3211 }
3212
3213 __be32
3214 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3215                   struct nfsd4_setclientid *setclid)
3216 {
3217         struct xdr_netobj       clname = setclid->se_name;
3218         nfs4_verifier           clverifier = setclid->se_verf;
3219         struct nfs4_client      *conf, *new;
3220         struct nfs4_client      *unconf = NULL;
3221         __be32                  status;
3222         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3223
3224         new = create_client(clname, rqstp, &clverifier);
3225         if (new == NULL)
3226                 return nfserr_jukebox;
3227         /* Cases below refer to rfc 3530 section 14.2.33: */
3228         spin_lock(&nn->client_lock);
3229         conf = find_confirmed_client_by_name(&clname, nn);
3230         if (conf && client_has_state(conf)) {
3231                 /* case 0: */
3232                 status = nfserr_clid_inuse;
3233                 if (clp_used_exchangeid(conf))
3234                         goto out;
3235                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3236                         char addr_str[INET6_ADDRSTRLEN];
3237                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3238                                  sizeof(addr_str));
3239                         dprintk("NFSD: setclientid: string in use by client "
3240                                 "at %s\n", addr_str);
3241                         goto out;
3242                 }
3243         }
3244         unconf = find_unconfirmed_client_by_name(&clname, nn);
3245         if (unconf)
3246                 unhash_client_locked(unconf);
3247         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3248                 /* case 1: probable callback update */
3249                 copy_clid(new, conf);
3250                 gen_confirm(new, nn);
3251         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3252                 gen_clid(new, nn);
3253         new->cl_minorversion = 0;
3254         gen_callback(new, setclid, rqstp);
3255         add_to_unconfirmed(new);
3256         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3257         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3258         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3259         new = NULL;
3260         status = nfs_ok;
3261 out:
3262         spin_unlock(&nn->client_lock);
3263         if (new)
3264                 free_client(new);
3265         if (unconf)
3266                 expire_client(unconf);
3267         return status;
3268 }
3269
3270
3271 __be32
3272 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3273                          struct nfsd4_compound_state *cstate,
3274                          struct nfsd4_setclientid_confirm *setclientid_confirm)
3275 {
3276         struct nfs4_client *conf, *unconf;
3277         struct nfs4_client *old = NULL;
3278         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3279         clientid_t * clid = &setclientid_confirm->sc_clientid;
3280         __be32 status;
3281         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3282
3283         if (STALE_CLIENTID(clid, nn))
3284                 return nfserr_stale_clientid;
3285
3286         spin_lock(&nn->client_lock);
3287         conf = find_confirmed_client(clid, false, nn);
3288         unconf = find_unconfirmed_client(clid, false, nn);
3289         /*
3290          * We try hard to give out unique clientid's, so if we get an
3291          * attempt to confirm the same clientid with a different cred,
3292          * the client may be buggy; this should never happen.
3293          *
3294          * Nevertheless, RFC 7530 recommends INUSE for this case:
3295          */
3296         status = nfserr_clid_inuse;
3297         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3298                 goto out;
3299         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3300                 goto out;
3301         /* cases below refer to rfc 3530 section 14.2.34: */
3302         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3303                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3304                         /* case 2: probable retransmit */
3305                         status = nfs_ok;
3306                 } else /* case 4: client hasn't noticed we rebooted yet? */
3307                         status = nfserr_stale_clientid;
3308                 goto out;
3309         }
3310         status = nfs_ok;
3311         if (conf) { /* case 1: callback update */
3312                 old = unconf;
3313                 unhash_client_locked(old);
3314                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3315         } else { /* case 3: normal case; new or rebooted client */
3316                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3317                 if (old) {
3318                         status = nfserr_clid_inuse;
3319                         if (client_has_state(old)
3320                                         && !same_creds(&unconf->cl_cred,
3321                                                         &old->cl_cred))
3322                                 goto out;
3323                         status = mark_client_expired_locked(old);
3324                         if (status) {
3325                                 old = NULL;
3326                                 goto out;
3327                         }
3328                 }
3329                 move_to_confirmed(unconf);
3330                 conf = unconf;
3331         }
3332         get_client_locked(conf);
3333         spin_unlock(&nn->client_lock);
3334         nfsd4_probe_callback(conf);
3335         spin_lock(&nn->client_lock);
3336         put_client_renew_locked(conf);
3337 out:
3338         spin_unlock(&nn->client_lock);
3339         if (old)
3340                 expire_client(old);
3341         return status;
3342 }
3343
3344 static struct nfs4_file *nfsd4_alloc_file(void)
3345 {
3346         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3347 }
3348
3349 /* OPEN Share state helper functions */
3350 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3351                                 struct nfs4_file *fp)
3352 {
3353         lockdep_assert_held(&state_lock);
3354
3355         atomic_set(&fp->fi_ref, 1);
3356         spin_lock_init(&fp->fi_lock);
3357         INIT_LIST_HEAD(&fp->fi_stateids);
3358         INIT_LIST_HEAD(&fp->fi_delegations);
3359         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3360         fh_copy_shallow(&fp->fi_fhandle, fh);
3361         fp->fi_deleg_file = NULL;
3362         fp->fi_had_conflict = false;
3363         fp->fi_share_deny = 0;
3364         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3365         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3366 #ifdef CONFIG_NFSD_PNFS
3367         INIT_LIST_HEAD(&fp->fi_lo_states);
3368         atomic_set(&fp->fi_lo_recalls, 0);
3369 #endif
3370         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3371 }
3372
3373 void
3374 nfsd4_free_slabs(void)
3375 {
3376         kmem_cache_destroy(odstate_slab);
3377         kmem_cache_destroy(openowner_slab);
3378         kmem_cache_destroy(lockowner_slab);
3379         kmem_cache_destroy(file_slab);
3380         kmem_cache_destroy(stateid_slab);
3381         kmem_cache_destroy(deleg_slab);
3382 }
3383
3384 int
3385 nfsd4_init_slabs(void)
3386 {
3387         openowner_slab = kmem_cache_create("nfsd4_openowners",
3388                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3389         if (openowner_slab == NULL)
3390                 goto out;
3391         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3392                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3393         if (lockowner_slab == NULL)
3394                 goto out_free_openowner_slab;
3395         file_slab = kmem_cache_create("nfsd4_files",
3396                         sizeof(struct nfs4_file), 0, 0, NULL);
3397         if (file_slab == NULL)
3398                 goto out_free_lockowner_slab;
3399         stateid_slab = kmem_cache_create("nfsd4_stateids",
3400                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3401         if (stateid_slab == NULL)
3402                 goto out_free_file_slab;
3403         deleg_slab = kmem_cache_create("nfsd4_delegations",
3404                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3405         if (deleg_slab == NULL)
3406                 goto out_free_stateid_slab;
3407         odstate_slab = kmem_cache_create("nfsd4_odstate",
3408                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3409         if (odstate_slab == NULL)
3410                 goto out_free_deleg_slab;
3411         return 0;
3412
3413 out_free_deleg_slab:
3414         kmem_cache_destroy(deleg_slab);
3415 out_free_stateid_slab:
3416         kmem_cache_destroy(stateid_slab);
3417 out_free_file_slab:
3418         kmem_cache_destroy(file_slab);
3419 out_free_lockowner_slab:
3420         kmem_cache_destroy(lockowner_slab);
3421 out_free_openowner_slab:
3422         kmem_cache_destroy(openowner_slab);
3423 out:
3424         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3425         return -ENOMEM;
3426 }
3427
3428 static void init_nfs4_replay(struct nfs4_replay *rp)
3429 {
3430         rp->rp_status = nfserr_serverfault;
3431         rp->rp_buflen = 0;
3432         rp->rp_buf = rp->rp_ibuf;
3433         mutex_init(&rp->rp_mutex);
3434 }
3435
3436 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3437                 struct nfs4_stateowner *so)
3438 {
3439         if (!nfsd4_has_session(cstate)) {
3440                 mutex_lock(&so->so_replay.rp_mutex);
3441                 cstate->replay_owner = nfs4_get_stateowner(so);
3442         }
3443 }
3444
3445 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3446 {
3447         struct nfs4_stateowner *so = cstate->replay_owner;
3448
3449         if (so != NULL) {
3450                 cstate->replay_owner = NULL;
3451                 mutex_unlock(&so->so_replay.rp_mutex);
3452                 nfs4_put_stateowner(so);
3453         }
3454 }
3455
3456 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3457 {
3458         struct nfs4_stateowner *sop;
3459
3460         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3461         if (!sop)
3462                 return NULL;
3463
3464         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3465         if (!sop->so_owner.data) {
3466                 kmem_cache_free(slab, sop);
3467                 return NULL;
3468         }
3469         sop->so_owner.len = owner->len;
3470
3471         INIT_LIST_HEAD(&sop->so_stateids);
3472         sop->so_client = clp;
3473         init_nfs4_replay(&sop->so_replay);
3474         atomic_set(&sop->so_count, 1);
3475         return sop;
3476 }
3477
3478 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3479 {
3480         lockdep_assert_held(&clp->cl_lock);
3481
3482         list_add(&oo->oo_owner.so_strhash,
3483                  &clp->cl_ownerstr_hashtbl[strhashval]);
3484         list_add(&oo->oo_perclient, &clp->cl_openowners);
3485 }
3486
3487 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3488 {
3489         unhash_openowner_locked(openowner(so));
3490 }
3491
3492 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3493 {
3494         struct nfs4_openowner *oo = openowner(so);
3495
3496         kmem_cache_free(openowner_slab, oo);
3497 }
3498
3499 static const struct nfs4_stateowner_operations openowner_ops = {
3500         .so_unhash =    nfs4_unhash_openowner,
3501         .so_free =      nfs4_free_openowner,
3502 };
3503
3504 static struct nfs4_ol_stateid *
3505 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3506 {
3507         struct nfs4_ol_stateid *local, *ret = NULL;
3508         struct nfs4_openowner *oo = open->op_openowner;
3509
3510         lockdep_assert_held(&fp->fi_lock);
3511
3512         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3513                 /* ignore lock owners */
3514                 if (local->st_stateowner->so_is_open_owner == 0)
3515                         continue;
3516                 if (local->st_stateowner == &oo->oo_owner) {
3517                         ret = local;
3518                         atomic_inc(&ret->st_stid.sc_count);
3519                         break;
3520                 }
3521         }
3522         return ret;
3523 }
3524
3525 static struct nfs4_openowner *
3526 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3527                            struct nfsd4_compound_state *cstate)
3528 {
3529         struct nfs4_client *clp = cstate->clp;
3530         struct nfs4_openowner *oo, *ret;
3531
3532         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3533         if (!oo)
3534                 return NULL;
3535         oo->oo_owner.so_ops = &openowner_ops;
3536         oo->oo_owner.so_is_open_owner = 1;
3537         oo->oo_owner.so_seqid = open->op_seqid;
3538         oo->oo_flags = 0;
3539         if (nfsd4_has_session(cstate))
3540                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3541         oo->oo_time = 0;
3542         oo->oo_last_closed_stid = NULL;
3543         INIT_LIST_HEAD(&oo->oo_close_lru);
3544         spin_lock(&clp->cl_lock);
3545         ret = find_openstateowner_str_locked(strhashval, open, clp);
3546         if (ret == NULL) {
3547                 hash_openowner(oo, clp, strhashval);
3548                 ret = oo;
3549         } else
3550                 nfs4_free_stateowner(&oo->oo_owner);
3551
3552         spin_unlock(&clp->cl_lock);
3553         return ret;
3554 }
3555
3556 static struct nfs4_ol_stateid *
3557 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3558 {
3559
3560         struct nfs4_openowner *oo = open->op_openowner;
3561         struct nfs4_ol_stateid *retstp = NULL;
3562         struct nfs4_ol_stateid *stp;
3563
3564         stp = open->op_stp;
3565         /* We are moving these outside of the spinlocks to avoid the warnings */
3566         mutex_init(&stp->st_mutex);
3567         mutex_lock(&stp->st_mutex);
3568
3569         spin_lock(&oo->oo_owner.so_client->cl_lock);
3570         spin_lock(&fp->fi_lock);
3571
3572         retstp = nfsd4_find_existing_open(fp, open);
3573         if (retstp)
3574                 goto out_unlock;
3575
3576         open->op_stp = NULL;
3577         atomic_inc(&stp->st_stid.sc_count);
3578         stp->st_stid.sc_type = NFS4_OPEN_STID;
3579         INIT_LIST_HEAD(&stp->st_locks);
3580         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3581         get_nfs4_file(fp);
3582         stp->st_stid.sc_file = fp;
3583         stp->st_access_bmap = 0;
3584         stp->st_deny_bmap = 0;
3585         stp->st_openstp = NULL;
3586         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3587         list_add(&stp->st_perfile, &fp->fi_stateids);
3588
3589 out_unlock:
3590         spin_unlock(&fp->fi_lock);
3591         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3592         if (retstp) {
3593                 mutex_lock(&retstp->st_mutex);
3594                 /* To keep mutex tracking happy */
3595                 mutex_unlock(&stp->st_mutex);
3596                 stp = retstp;
3597         }
3598         return stp;
3599 }
3600
3601 /*
3602  * In the 4.0 case we need to keep the owners around a little while to handle
3603  * CLOSE replay. We still do need to release any file access that is held by
3604  * them before returning however.
3605  */
3606 static void
3607 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3608 {
3609         struct nfs4_ol_stateid *last;
3610         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3611         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3612                                                 nfsd_net_id);
3613
3614         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3615
3616         /*
3617          * We know that we hold one reference via nfsd4_close, and another
3618          * "persistent" reference for the client. If the refcount is higher
3619          * than 2, then there are still calls in progress that are using this
3620          * stateid. We can't put the sc_file reference until they are finished.
3621          * Wait for the refcount to drop to 2. Since it has been unhashed,
3622          * there should be no danger of the refcount going back up again at
3623          * this point.
3624          */
3625         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3626
3627         release_all_access(s);
3628         if (s->st_stid.sc_file) {
3629                 put_nfs4_file(s->st_stid.sc_file);
3630                 s->st_stid.sc_file = NULL;
3631         }
3632
3633         spin_lock(&nn->client_lock);
3634         last = oo->oo_last_closed_stid;
3635         oo->oo_last_closed_stid = s;
3636         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3637         oo->oo_time = get_seconds();
3638         spin_unlock(&nn->client_lock);
3639         if (last)
3640                 nfs4_put_stid(&last->st_stid);
3641 }
3642
3643 /* search file_hashtbl[] for file */
3644 static struct nfs4_file *
3645 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3646 {
3647         struct nfs4_file *fp;
3648
3649         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3650                 if (fh_match(&fp->fi_fhandle, fh)) {
3651                         if (atomic_inc_not_zero(&fp->fi_ref))
3652                                 return fp;
3653                 }
3654         }
3655         return NULL;
3656 }
3657
3658 struct nfs4_file *
3659 find_file(struct knfsd_fh *fh)
3660 {
3661         struct nfs4_file *fp;
3662         unsigned int hashval = file_hashval(fh);
3663
3664         rcu_read_lock();
3665         fp = find_file_locked(fh, hashval);
3666         rcu_read_unlock();
3667         return fp;
3668 }
3669
3670 static struct nfs4_file *
3671 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3672 {
3673         struct nfs4_file *fp;
3674         unsigned int hashval = file_hashval(fh);
3675
3676         rcu_read_lock();
3677         fp = find_file_locked(fh, hashval);
3678         rcu_read_unlock();
3679         if (fp)
3680                 return fp;
3681
3682         spin_lock(&state_lock);
3683         fp = find_file_locked(fh, hashval);
3684         if (likely(fp == NULL)) {
3685                 nfsd4_init_file(fh, hashval, new);
3686                 fp = new;
3687         }
3688         spin_unlock(&state_lock);
3689
3690         return fp;
3691 }
3692
3693 /*
3694  * Called to check deny when READ with all zero stateid or
3695  * WRITE with all zero or all one stateid
3696  */
3697 static __be32
3698 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3699 {
3700         struct nfs4_file *fp;
3701         __be32 ret = nfs_ok;
3702
3703         fp = find_file(&current_fh->fh_handle);
3704         if (!fp)
3705                 return ret;
3706         /* Check for conflicting share reservations */
3707         spin_lock(&fp->fi_lock);
3708         if (fp->fi_share_deny & deny_type)
3709                 ret = nfserr_locked;
3710         spin_unlock(&fp->fi_lock);
3711         put_nfs4_file(fp);
3712         return ret;
3713 }
3714
3715 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3716 {
3717         struct nfs4_delegation *dp = cb_to_delegation(cb);
3718         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3719                                           nfsd_net_id);
3720
3721         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3722
3723         /*
3724          * We can't do this in nfsd_break_deleg_cb because it is
3725          * already holding inode->i_lock.
3726          *
3727          * If the dl_time != 0, then we know that it has already been
3728          * queued for a lease break. Don't queue it again.
3729          */
3730         spin_lock(&state_lock);
3731         if (dp->dl_time == 0) {
3732                 dp->dl_time = get_seconds();
3733                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3734         }
3735         spin_unlock(&state_lock);
3736 }
3737
3738 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3739                 struct rpc_task *task)
3740 {
3741         struct nfs4_delegation *dp = cb_to_delegation(cb);
3742
3743         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3744                 return 1;
3745
3746         switch (task->tk_status) {
3747         case 0:
3748                 return 1;
3749         case -EBADHANDLE:
3750         case -NFS4ERR_BAD_STATEID:
3751                 /*
3752                  * Race: client probably got cb_recall before open reply
3753                  * granting delegation.
3754                  */
3755                 if (dp->dl_retries--) {
3756                         rpc_delay(task, 2 * HZ);
3757                         return 0;
3758                 }
3759                 /*FALLTHRU*/
3760         default:
3761                 return -1;
3762         }
3763 }
3764
3765 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3766 {
3767         struct nfs4_delegation *dp = cb_to_delegation(cb);
3768
3769         nfs4_put_stid(&dp->dl_stid);
3770 }
3771
3772 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3773         .prepare        = nfsd4_cb_recall_prepare,
3774         .done           = nfsd4_cb_recall_done,
3775         .release        = nfsd4_cb_recall_release,
3776 };
3777
3778 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3779 {
3780         /*
3781          * We're assuming the state code never drops its reference
3782          * without first removing the lease.  Since we're in this lease
3783          * callback (and since the lease code is serialized by the kernel
3784          * lock) we know the server hasn't removed the lease yet, we know
3785          * it's safe to take a reference.
3786          */
3787         atomic_inc(&dp->dl_stid.sc_count);
3788         nfsd4_run_cb(&dp->dl_recall);
3789 }
3790
3791 /* Called from break_lease() with i_lock held. */
3792 static bool
3793 nfsd_break_deleg_cb(struct file_lock *fl)
3794 {
3795         bool ret = false;
3796         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3797         struct nfs4_delegation *dp;
3798
3799         if (!fp) {
3800                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3801                 return ret;
3802         }
3803         if (fp->fi_had_conflict) {
3804                 WARN(1, "duplicate break on %p\n", fp);
3805                 return ret;
3806         }
3807         /*
3808          * We don't want the locks code to timeout the lease for us;
3809          * we'll remove it ourself if a delegation isn't returned
3810          * in time:
3811          */
3812         fl->fl_break_time = 0;
3813
3814         spin_lock(&fp->fi_lock);
3815         fp->fi_had_conflict = true;
3816         /*
3817          * If there are no delegations on the list, then return true
3818          * so that the lease code will go ahead and delete it.
3819          */
3820         if (list_empty(&fp->fi_delegations))
3821                 ret = true;
3822         else
3823                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3824                         nfsd_break_one_deleg(dp);
3825         spin_unlock(&fp->fi_lock);
3826         return ret;
3827 }
3828
3829 static int
3830 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3831                      struct list_head *dispose)
3832 {
3833         if (arg & F_UNLCK)
3834                 return lease_modify(onlist, arg, dispose);
3835         else
3836                 return -EAGAIN;
3837 }
3838
3839 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3840         .lm_break = nfsd_break_deleg_cb,
3841         .lm_change = nfsd_change_deleg_cb,
3842 };
3843
3844 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3845 {
3846         if (nfsd4_has_session(cstate))
3847                 return nfs_ok;
3848         if (seqid == so->so_seqid - 1)
3849                 return nfserr_replay_me;
3850         if (seqid == so->so_seqid)
3851                 return nfs_ok;
3852         return nfserr_bad_seqid;
3853 }
3854
3855 static __be32 lookup_clientid(clientid_t *clid,
3856                 struct nfsd4_compound_state *cstate,
3857                 struct nfsd_net *nn)
3858 {
3859         struct nfs4_client *found;
3860
3861         if (cstate->clp) {
3862                 found = cstate->clp;
3863                 if (!same_clid(&found->cl_clientid, clid))
3864                         return nfserr_stale_clientid;
3865                 return nfs_ok;
3866         }
3867
3868         if (STALE_CLIENTID(clid, nn))
3869                 return nfserr_stale_clientid;
3870
3871         /*
3872          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3873          * cached already then we know this is for is for v4.0 and "sessions"
3874          * will be false.
3875          */
3876         WARN_ON_ONCE(cstate->session);
3877         spin_lock(&nn->client_lock);
3878         found = find_confirmed_client(clid, false, nn);
3879         if (!found) {
3880                 spin_unlock(&nn->client_lock);
3881                 return nfserr_expired;
3882         }
3883         atomic_inc(&found->cl_refcount);
3884         spin_unlock(&nn->client_lock);
3885
3886         /* Cache the nfs4_client in cstate! */
3887         cstate->clp = found;
3888         return nfs_ok;
3889 }
3890
3891 __be32
3892 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3893                     struct nfsd4_open *open, struct nfsd_net *nn)
3894 {
3895         clientid_t *clientid = &open->op_clientid;
3896         struct nfs4_client *clp = NULL;
3897         unsigned int strhashval;
3898         struct nfs4_openowner *oo = NULL;
3899         __be32 status;
3900
3901         if (STALE_CLIENTID(&open->op_clientid, nn))
3902                 return nfserr_stale_clientid;
3903         /*
3904          * In case we need it later, after we've already created the
3905          * file and don't want to risk a further failure:
3906          */
3907         open->op_file = nfsd4_alloc_file();
3908         if (open->op_file == NULL)
3909                 return nfserr_jukebox;
3910
3911         status = lookup_clientid(clientid, cstate, nn);
3912         if (status)
3913                 return status;
3914         clp = cstate->clp;
3915
3916         strhashval = ownerstr_hashval(&open->op_owner);
3917         oo = find_openstateowner_str(strhashval, open, clp);
3918         open->op_openowner = oo;
3919         if (!oo) {
3920                 goto new_owner;
3921         }
3922         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3923                 /* Replace unconfirmed owners without checking for replay. */
3924                 release_openowner(oo);
3925                 open->op_openowner = NULL;
3926                 goto new_owner;
3927         }
3928         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3929         if (status)
3930                 return status;
3931         goto alloc_stateid;
3932 new_owner:
3933         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3934         if (oo == NULL)
3935                 return nfserr_jukebox;
3936         open->op_openowner = oo;
3937 alloc_stateid:
3938         open->op_stp = nfs4_alloc_open_stateid(clp);
3939         if (!open->op_stp)
3940                 return nfserr_jukebox;
3941
3942         if (nfsd4_has_session(cstate) &&
3943             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3944                 open->op_odstate = alloc_clnt_odstate(clp);
3945                 if (!open->op_odstate)
3946                         return nfserr_jukebox;
3947         }
3948
3949         return nfs_ok;
3950 }
3951
3952 static inline __be32
3953 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3954 {
3955         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3956                 return nfserr_openmode;
3957         else
3958                 return nfs_ok;
3959 }
3960
3961 static int share_access_to_flags(u32 share_access)
3962 {
3963         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3964 }
3965
3966 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3967 {
3968         struct nfs4_stid *ret;
3969
3970         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3971         if (!ret)
3972                 return NULL;
3973         return delegstateid(ret);
3974 }
3975
3976 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3977 {
3978         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3979                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3980 }
3981
3982 static __be32
3983 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3984                 struct nfs4_delegation **dp)
3985 {
3986         int flags;
3987         __be32 status = nfserr_bad_stateid;
3988         struct nfs4_delegation *deleg;
3989
3990         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3991         if (deleg == NULL)
3992                 goto out;
3993         flags = share_access_to_flags(open->op_share_access);
3994         status = nfs4_check_delegmode(deleg, flags);
3995         if (status) {
3996                 nfs4_put_stid(&deleg->dl_stid);
3997                 goto out;
3998         }
3999         *dp = deleg;
4000 out:
4001         if (!nfsd4_is_deleg_cur(open))
4002                 return nfs_ok;
4003         if (status)
4004                 return status;
4005         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4006         return nfs_ok;
4007 }
4008
4009 static inline int nfs4_access_to_access(u32 nfs4_access)
4010 {
4011         int flags = 0;
4012
4013         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4014                 flags |= NFSD_MAY_READ;
4015         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4016                 flags |= NFSD_MAY_WRITE;
4017         return flags;
4018 }
4019
4020 static inline __be32
4021 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4022                 struct nfsd4_open *open)
4023 {
4024         struct iattr iattr = {
4025                 .ia_valid = ATTR_SIZE,
4026                 .ia_size = 0,
4027         };
4028         if (!open->op_truncate)
4029                 return 0;
4030         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4031                 return nfserr_inval;
4032         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4033 }
4034
4035 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4036                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4037                 struct nfsd4_open *open)
4038 {
4039         struct file *filp = NULL;
4040         __be32 status;
4041         int oflag = nfs4_access_to_omode(open->op_share_access);
4042         int access = nfs4_access_to_access(open->op_share_access);
4043         unsigned char old_access_bmap, old_deny_bmap;
4044
4045         spin_lock(&fp->fi_lock);
4046
4047         /*
4048          * Are we trying to set a deny mode that would conflict with
4049          * current access?
4050          */
4051         status = nfs4_file_check_deny(fp, open->op_share_deny);
4052         if (status != nfs_ok) {
4053                 spin_unlock(&fp->fi_lock);
4054                 goto out;
4055         }
4056
4057         /* set access to the file */
4058         status = nfs4_file_get_access(fp, open->op_share_access);
4059         if (status != nfs_ok) {
4060                 spin_unlock(&fp->fi_lock);
4061                 goto out;
4062         }
4063
4064         /* Set access bits in stateid */
4065         old_access_bmap = stp->st_access_bmap;
4066         set_access(open->op_share_access, stp);
4067
4068         /* Set new deny mask */
4069         old_deny_bmap = stp->st_deny_bmap;
4070         set_deny(open->op_share_deny, stp);
4071         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4072
4073         if (!fp->fi_fds[oflag]) {
4074                 spin_unlock(&fp->fi_lock);
4075                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4076                 if (status)
4077                         goto out_put_access;
4078                 spin_lock(&fp->fi_lock);
4079                 if (!fp->fi_fds[oflag]) {
4080                         fp->fi_fds[oflag] = filp;
4081                         filp = NULL;
4082                 }
4083         }
4084         spin_unlock(&fp->fi_lock);
4085         if (filp)
4086                 fput(filp);
4087
4088         status = nfsd4_truncate(rqstp, cur_fh, open);
4089         if (status)
4090                 goto out_put_access;
4091 out:
4092         return status;
4093 out_put_access:
4094         stp->st_access_bmap = old_access_bmap;
4095         nfs4_file_put_access(fp, open->op_share_access);
4096         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4097         goto out;
4098 }
4099
4100 static __be32
4101 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)
4102 {
4103         __be32 status;
4104         unsigned char old_deny_bmap = stp->st_deny_bmap;
4105
4106         if (!test_access(open->op_share_access, stp))
4107                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4108
4109         /* test and set deny mode */
4110         spin_lock(&fp->fi_lock);
4111         status = nfs4_file_check_deny(fp, open->op_share_deny);
4112         if (status == nfs_ok) {
4113                 set_deny(open->op_share_deny, stp);
4114                 fp->fi_share_deny |=
4115                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4116         }
4117         spin_unlock(&fp->fi_lock);
4118
4119         if (status != nfs_ok)
4120                 return status;
4121
4122         status = nfsd4_truncate(rqstp, cur_fh, open);
4123         if (status != nfs_ok)
4124                 reset_union_bmap_deny(old_deny_bmap, stp);
4125         return status;
4126 }
4127
4128 /* Should we give out recallable state?: */
4129 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4130 {
4131         if (clp->cl_cb_state == NFSD4_CB_UP)
4132                 return true;
4133         /*
4134          * In the sessions case, since we don't have to establish a
4135          * separate connection for callbacks, we assume it's OK
4136          * until we hear otherwise:
4137          */
4138         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4139 }
4140
4141 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4142 {
4143         struct file_lock *fl;
4144
4145         fl = locks_alloc_lock();
4146         if (!fl)
4147                 return NULL;
4148         fl->fl_lmops = &nfsd_lease_mng_ops;
4149         fl->fl_flags = FL_DELEG;
4150         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4151         fl->fl_end = OFFSET_MAX;
4152         fl->fl_owner = (fl_owner_t)fp;
4153         fl->fl_pid = current->tgid;
4154         return fl;
4155 }
4156
4157 /**
4158  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4159  * @dp:   a pointer to the nfs4_delegation we're adding.
4160  *
4161  * Return:
4162  *      On success: Return code will be 0 on success.
4163  *
4164  *      On error: -EAGAIN if there was an existing delegation.
4165  *                 nonzero if there is an error in other cases.
4166  *
4167  */
4168
4169 static int nfs4_setlease(struct nfs4_delegation *dp)
4170 {
4171         struct nfs4_file *fp = dp->dl_stid.sc_file;
4172         struct file_lock *fl;
4173         struct file *filp;
4174         int status = 0;
4175
4176         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4177         if (!fl)
4178                 return -ENOMEM;
4179         filp = find_readable_file(fp);
4180         if (!filp) {
4181                 /* We should always have a readable file here */
4182                 WARN_ON_ONCE(1);
4183                 locks_free_lock(fl);
4184                 return -EBADF;
4185         }
4186         fl->fl_file = filp;
4187         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4188         if (fl)
4189                 locks_free_lock(fl);
4190         if (status)
4191                 goto out_fput;
4192         spin_lock(&state_lock);
4193         spin_lock(&fp->fi_lock);
4194         /* Did the lease get broken before we took the lock? */
4195         status = -EAGAIN;
4196         if (fp->fi_had_conflict)
4197                 goto out_unlock;
4198         /* Race breaker */
4199         if (fp->fi_deleg_file) {
4200                 status = hash_delegation_locked(dp, fp);
4201                 goto out_unlock;
4202         }
4203         fp->fi_deleg_file = filp;
4204         fp->fi_delegees = 0;
4205         status = hash_delegation_locked(dp, fp);
4206         spin_unlock(&fp->fi_lock);
4207         spin_unlock(&state_lock);
4208         if (status) {
4209                 /* Should never happen, this is a new fi_deleg_file  */
4210                 WARN_ON_ONCE(1);
4211                 goto out_fput;
4212         }
4213         return 0;
4214 out_unlock:
4215         spin_unlock(&fp->fi_lock);
4216         spin_unlock(&state_lock);
4217 out_fput:
4218         fput(filp);
4219         return status;
4220 }
4221
4222 static struct nfs4_delegation *
4223 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4224                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4225 {
4226         int status;
4227         struct nfs4_delegation *dp;
4228
4229         if (fp->fi_had_conflict)
4230                 return ERR_PTR(-EAGAIN);
4231
4232         spin_lock(&state_lock);
4233         spin_lock(&fp->fi_lock);
4234         status = nfs4_get_existing_delegation(clp, fp);
4235         spin_unlock(&fp->fi_lock);
4236         spin_unlock(&state_lock);
4237
4238         if (status)
4239                 return ERR_PTR(status);
4240
4241         dp = alloc_init_deleg(clp, fh, odstate);
4242         if (!dp)
4243                 return ERR_PTR(-ENOMEM);
4244
4245         get_nfs4_file(fp);
4246         spin_lock(&state_lock);
4247         spin_lock(&fp->fi_lock);
4248         dp->dl_stid.sc_file = fp;
4249         if (!fp->fi_deleg_file) {
4250                 spin_unlock(&fp->fi_lock);
4251                 spin_unlock(&state_lock);
4252                 status = nfs4_setlease(dp);
4253                 goto out;
4254         }
4255         if (fp->fi_had_conflict) {
4256                 status = -EAGAIN;
4257                 goto out_unlock;
4258         }
4259         status = hash_delegation_locked(dp, fp);
4260 out_unlock:
4261         spin_unlock(&fp->fi_lock);
4262         spin_unlock(&state_lock);
4263 out:
4264         if (status) {
4265                 put_clnt_odstate(dp->dl_clnt_odstate);
4266                 nfs4_put_stid(&dp->dl_stid);
4267                 return ERR_PTR(status);
4268         }
4269         return dp;
4270 }
4271
4272 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4273 {
4274         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4275         if (status == -EAGAIN)
4276                 open->op_why_no_deleg = WND4_CONTENTION;
4277         else {
4278                 open->op_why_no_deleg = WND4_RESOURCE;
4279                 switch (open->op_deleg_want) {
4280                 case NFS4_SHARE_WANT_READ_DELEG:
4281                 case NFS4_SHARE_WANT_WRITE_DELEG:
4282                 case NFS4_SHARE_WANT_ANY_DELEG:
4283                         break;
4284                 case NFS4_SHARE_WANT_CANCEL:
4285                         open->op_why_no_deleg = WND4_CANCELLED;
4286                         break;
4287                 case NFS4_SHARE_WANT_NO_DELEG:
4288                         WARN_ON_ONCE(1);
4289                 }
4290         }
4291 }
4292
4293 /*
4294  * Attempt to hand out a delegation.
4295  *
4296  * Note we don't support write delegations, and won't until the vfs has
4297  * proper support for them.
4298  */
4299 static void
4300 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4301                         struct nfs4_ol_stateid *stp)
4302 {
4303         struct nfs4_delegation *dp;
4304         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4305         struct nfs4_client *clp = stp->st_stid.sc_client;
4306         int cb_up;
4307         int status = 0;
4308
4309         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4310         open->op_recall = 0;
4311         switch (open->op_claim_type) {
4312                 case NFS4_OPEN_CLAIM_PREVIOUS:
4313                         if (!cb_up)
4314                                 open->op_recall = 1;
4315                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4316                                 goto out_no_deleg;
4317                         break;
4318                 case NFS4_OPEN_CLAIM_NULL:
4319                 case NFS4_OPEN_CLAIM_FH:
4320                         /*
4321                          * Let's not give out any delegations till everyone's
4322                          * had the chance to reclaim theirs, *and* until
4323                          * NLM locks have all been reclaimed:
4324                          */
4325                         if (locks_in_grace(clp->net))
4326                                 goto out_no_deleg;
4327                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4328                                 goto out_no_deleg;
4329                         /*
4330                          * Also, if the file was opened for write or
4331                          * create, there's a good chance the client's
4332                          * about to write to it, resulting in an
4333                          * immediate recall (since we don't support
4334                          * write delegations):
4335                          */
4336                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4337                                 goto out_no_deleg;
4338                         if (open->op_create == NFS4_OPEN_CREATE)
4339                                 goto out_no_deleg;
4340                         break;
4341                 default:
4342                         goto out_no_deleg;
4343         }
4344         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4345         if (IS_ERR(dp))
4346                 goto out_no_deleg;
4347
4348         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4349
4350         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4351                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4352         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4353         nfs4_put_stid(&dp->dl_stid);
4354         return;
4355 out_no_deleg:
4356         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4357         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4358             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4359                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4360                 open->op_recall = 1;
4361         }
4362
4363         /* 4.1 client asking for a delegation? */
4364         if (open->op_deleg_want)
4365                 nfsd4_open_deleg_none_ext(open, status);
4366         return;
4367 }
4368
4369 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4370                                         struct nfs4_delegation *dp)
4371 {
4372         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4373             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4374                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4375                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4376         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4377                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4378                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4379                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4380         }
4381         /* Otherwise the client must be confused wanting a delegation
4382          * it already has, therefore we don't return
4383          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4384          */
4385 }
4386
4387 __be32
4388 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4389 {
4390         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4391         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4392         struct nfs4_file *fp = NULL;
4393         struct nfs4_ol_stateid *stp = NULL;
4394         struct nfs4_delegation *dp = NULL;
4395         __be32 status;
4396
4397         /*
4398          * Lookup file; if found, lookup stateid and check open request,
4399          * and check for delegations in the process of being recalled.
4400          * If not found, create the nfs4_file struct
4401          */
4402         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4403         if (fp != open->op_file) {
4404                 status = nfs4_check_deleg(cl, open, &dp);
4405                 if (status)
4406                         goto out;
4407                 spin_lock(&fp->fi_lock);
4408                 stp = nfsd4_find_existing_open(fp, open);
4409                 spin_unlock(&fp->fi_lock);
4410         } else {
4411                 open->op_file = NULL;
4412                 status = nfserr_bad_stateid;
4413                 if (nfsd4_is_deleg_cur(open))
4414                         goto out;
4415         }
4416
4417         /*
4418          * OPEN the file, or upgrade an existing OPEN.
4419          * If truncate fails, the OPEN fails.
4420          */
4421         if (stp) {
4422                 /* Stateid was found, this is an OPEN upgrade */
4423                 mutex_lock(&stp->st_mutex);
4424                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4425                 if (status) {
4426                         mutex_unlock(&stp->st_mutex);
4427                         goto out;
4428                 }
4429         } else {
4430                 /* stp is returned locked. */
4431                 stp = init_open_stateid(fp, open);
4432                 /* See if we lost the race to some other thread */
4433                 if (stp->st_access_bmap != 0) {
4434                         status = nfs4_upgrade_open(rqstp, fp, current_fh,
4435                                                 stp, open);
4436                         if (status) {
4437                                 mutex_unlock(&stp->st_mutex);
4438                                 goto out;
4439                         }
4440                         goto upgrade_out;
4441                 }
4442                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4443                 if (status) {
4444                         mutex_unlock(&stp->st_mutex);
4445                         release_open_stateid(stp);
4446                         goto out;
4447                 }
4448
4449                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4450                                                         open->op_odstate);
4451                 if (stp->st_clnt_odstate == open->op_odstate)
4452                         open->op_odstate = NULL;
4453         }
4454 upgrade_out:
4455         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4456         mutex_unlock(&stp->st_mutex);
4457
4458         if (nfsd4_has_session(&resp->cstate)) {
4459                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4460                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4461                         open->op_why_no_deleg = WND4_NOT_WANTED;
4462                         goto nodeleg;
4463                 }
4464         }
4465
4466         /*
4467         * Attempt to hand out a delegation. No error return, because the
4468         * OPEN succeeds even if we fail.
4469         */
4470         nfs4_open_delegation(current_fh, open, stp);
4471 nodeleg:
4472         status = nfs_ok;
4473
4474         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4475                 STATEID_VAL(&stp->st_stid.sc_stateid));
4476 out:
4477         /* 4.1 client trying to upgrade/downgrade delegation? */
4478         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4479             open->op_deleg_want)
4480                 nfsd4_deleg_xgrade_none_ext(open, dp);
4481
4482         if (fp)
4483                 put_nfs4_file(fp);
4484         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4485                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4486         /*
4487         * To finish the open response, we just need to set the rflags.
4488         */
4489         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4490         if (nfsd4_has_session(&resp->cstate))
4491                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4492         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4493                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4494
4495         if (dp)
4496                 nfs4_put_stid(&dp->dl_stid);
4497         if (stp)
4498                 nfs4_put_stid(&stp->st_stid);
4499
4500         return status;
4501 }
4502
4503 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4504                               struct nfsd4_open *open)
4505 {
4506         if (open->op_openowner) {
4507                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4508
4509                 nfsd4_cstate_assign_replay(cstate, so);
4510                 nfs4_put_stateowner(so);
4511         }
4512         if (open->op_file)
4513                 kmem_cache_free(file_slab, open->op_file);
4514         if (open->op_stp)
4515                 nfs4_put_stid(&open->op_stp->st_stid);
4516         if (open->op_odstate)
4517                 kmem_cache_free(odstate_slab, open->op_odstate);
4518 }
4519
4520 __be32
4521 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4522             clientid_t *clid)
4523 {
4524         struct nfs4_client *clp;
4525         __be32 status;
4526         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4527
4528         dprintk("process_renew(%08x/%08x): starting\n",
4529                         clid->cl_boot, clid->cl_id);
4530         status = lookup_clientid(clid, cstate, nn);
4531         if (status)
4532                 goto out;
4533         clp = cstate->clp;
4534         status = nfserr_cb_path_down;
4535         if (!list_empty(&clp->cl_delegations)
4536                         && clp->cl_cb_state != NFSD4_CB_UP)
4537                 goto out;
4538         status = nfs_ok;
4539 out:
4540         return status;
4541 }
4542
4543 void
4544 nfsd4_end_grace(struct nfsd_net *nn)
4545 {
4546         /* do nothing if grace period already ended */
4547         if (nn->grace_ended)
4548                 return;
4549
4550         dprintk("NFSD: end of grace period\n");
4551         nn->grace_ended = true;
4552         /*
4553          * If the server goes down again right now, an NFSv4
4554          * client will still be allowed to reclaim after it comes back up,
4555          * even if it hasn't yet had a chance to reclaim state this time.
4556          *
4557          */
4558         nfsd4_record_grace_done(nn);
4559         /*
4560          * At this point, NFSv4 clients can still reclaim.  But if the
4561          * server crashes, any that have not yet reclaimed will be out
4562          * of luck on the next boot.
4563          *
4564          * (NFSv4.1+ clients are considered to have reclaimed once they
4565          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4566          * have reclaimed after their first OPEN.)
4567          */
4568         locks_end_grace(&nn->nfsd4_manager);
4569         /*
4570          * At this point, and once lockd and/or any other containers
4571          * exit their grace period, further reclaims will fail and
4572          * regular locking can resume.
4573          */
4574 }
4575
4576 static time_t
4577 nfs4_laundromat(struct nfsd_net *nn)
4578 {
4579         struct nfs4_client *clp;
4580         struct nfs4_openowner *oo;
4581         struct nfs4_delegation *dp;
4582         struct nfs4_ol_stateid *stp;
4583         struct nfsd4_blocked_lock *nbl;
4584         struct list_head *pos, *next, reaplist;
4585         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4586         time_t t, new_timeo = nn->nfsd4_lease;
4587
4588         dprintk("NFSD: laundromat service - starting\n");
4589         nfsd4_end_grace(nn);
4590         INIT_LIST_HEAD(&reaplist);
4591         spin_lock(&nn->client_lock);
4592         list_for_each_safe(pos, next, &nn->client_lru) {
4593                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4594                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4595                         t = clp->cl_time - cutoff;
4596                         new_timeo = min(new_timeo, t);
4597                         break;
4598                 }
4599                 if (mark_client_expired_locked(clp)) {
4600                         dprintk("NFSD: client in use (clientid %08x)\n",
4601                                 clp->cl_clientid.cl_id);
4602                         continue;
4603                 }
4604                 list_add(&clp->cl_lru, &reaplist);
4605         }
4606         spin_unlock(&nn->client_lock);
4607         list_for_each_safe(pos, next, &reaplist) {
4608                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4609                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4610                         clp->cl_clientid.cl_id);
4611                 list_del_init(&clp->cl_lru);
4612                 expire_client(clp);
4613         }
4614         spin_lock(&state_lock);
4615         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4616                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4617                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4618                         t = dp->dl_time - cutoff;
4619                         new_timeo = min(new_timeo, t);
4620                         break;
4621                 }
4622                 WARN_ON(!unhash_delegation_locked(dp));
4623                 list_add(&dp->dl_recall_lru, &reaplist);
4624         }
4625         spin_unlock(&state_lock);
4626         while (!list_empty(&reaplist)) {
4627                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4628                                         dl_recall_lru);
4629                 list_del_init(&dp->dl_recall_lru);
4630                 revoke_delegation(dp);
4631         }
4632
4633         spin_lock(&nn->client_lock);
4634         while (!list_empty(&nn->close_lru)) {
4635                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4636                                         oo_close_lru);
4637                 if (time_after((unsigned long)oo->oo_time,
4638                                (unsigned long)cutoff)) {
4639                         t = oo->oo_time - cutoff;
4640                         new_timeo = min(new_timeo, t);
4641                         break;
4642                 }
4643                 list_del_init(&oo->oo_close_lru);
4644                 stp = oo->oo_last_closed_stid;
4645                 oo->oo_last_closed_stid = NULL;
4646                 spin_unlock(&nn->client_lock);
4647                 nfs4_put_stid(&stp->st_stid);
4648                 spin_lock(&nn->client_lock);
4649         }
4650         spin_unlock(&nn->client_lock);
4651
4652         /*
4653          * It's possible for a client to try and acquire an already held lock
4654          * that is being held for a long time, and then lose interest in it.
4655          * So, we clean out any un-revisited request after a lease period
4656          * under the assumption that the client is no longer interested.
4657          *
4658          * RFC5661, sec. 9.6 states that the client must not rely on getting
4659          * notifications and must continue to poll for locks, even when the
4660          * server supports them. Thus this shouldn't lead to clients blocking
4661          * indefinitely once the lock does become free.
4662          */
4663         BUG_ON(!list_empty(&reaplist));
4664         spin_lock(&nn->blocked_locks_lock);
4665         while (!list_empty(&nn->blocked_locks_lru)) {
4666                 nbl = list_first_entry(&nn->blocked_locks_lru,
4667                                         struct nfsd4_blocked_lock, nbl_lru);
4668                 if (time_after((unsigned long)nbl->nbl_time,
4669                                (unsigned long)cutoff)) {
4670                         t = nbl->nbl_time - cutoff;
4671                         new_timeo = min(new_timeo, t);
4672                         break;
4673                 }
4674                 list_move(&nbl->nbl_lru, &reaplist);
4675                 list_del_init(&nbl->nbl_list);
4676         }
4677         spin_unlock(&nn->blocked_locks_lock);
4678
4679         while (!list_empty(&reaplist)) {
4680                 nbl = list_first_entry(&nn->blocked_locks_lru,
4681                                         struct nfsd4_blocked_lock, nbl_lru);
4682                 list_del_init(&nbl->nbl_lru);
4683                 posix_unblock_lock(&nbl->nbl_lock);
4684                 free_blocked_lock(nbl);
4685         }
4686
4687         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4688         return new_timeo;
4689 }
4690
4691 static struct workqueue_struct *laundry_wq;
4692 static void laundromat_main(struct work_struct *);
4693
4694 static void
4695 laundromat_main(struct work_struct *laundry)
4696 {
4697         time_t t;
4698         struct delayed_work *dwork = to_delayed_work(laundry);
4699         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4700                                            laundromat_work);
4701
4702         t = nfs4_laundromat(nn);
4703         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4704         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4705 }
4706
4707 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4708 {
4709         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4710                 return nfserr_bad_stateid;
4711         return nfs_ok;
4712 }
4713
4714 static inline int
4715 access_permit_read(struct nfs4_ol_stateid *stp)
4716 {
4717         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4718                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4719                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4720 }
4721
4722 static inline int
4723 access_permit_write(struct nfs4_ol_stateid *stp)
4724 {
4725         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4726                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4727 }
4728
4729 static
4730 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4731 {
4732         __be32 status = nfserr_openmode;
4733
4734         /* For lock stateid's, we test the parent open, not the lock: */
4735         if (stp->st_openstp)
4736                 stp = stp->st_openstp;
4737         if ((flags & WR_STATE) && !access_permit_write(stp))
4738                 goto out;
4739         if ((flags & RD_STATE) && !access_permit_read(stp))
4740                 goto out;
4741         status = nfs_ok;
4742 out:
4743         return status;
4744 }
4745
4746 static inline __be32
4747 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4748 {
4749         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4750                 return nfs_ok;
4751         else if (opens_in_grace(net)) {
4752                 /* Answer in remaining cases depends on existence of
4753                  * conflicting state; so we must wait out the grace period. */
4754                 return nfserr_grace;
4755         } else if (flags & WR_STATE)
4756                 return nfs4_share_conflict(current_fh,
4757                                 NFS4_SHARE_DENY_WRITE);
4758         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4759                 return nfs4_share_conflict(current_fh,
4760                                 NFS4_SHARE_DENY_READ);
4761 }
4762
4763 /*
4764  * Allow READ/WRITE during grace period on recovered state only for files
4765  * that are not able to provide mandatory locking.
4766  */
4767 static inline int
4768 grace_disallows_io(struct net *net, struct inode *inode)
4769 {
4770         return opens_in_grace(net) && mandatory_lock(inode);
4771 }
4772
4773 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4774 {
4775         /*
4776          * When sessions are used the stateid generation number is ignored
4777          * when it is zero.
4778          */
4779         if (has_session && in->si_generation == 0)
4780                 return nfs_ok;
4781
4782         if (in->si_generation == ref->si_generation)
4783                 return nfs_ok;
4784
4785         /* If the client sends us a stateid from the future, it's buggy: */
4786         if (nfsd4_stateid_generation_after(in, ref))
4787                 return nfserr_bad_stateid;
4788         /*
4789          * However, we could see a stateid from the past, even from a
4790          * non-buggy client.  For example, if the client sends a lock
4791          * while some IO is outstanding, the lock may bump si_generation
4792          * while the IO is still in flight.  The client could avoid that
4793          * situation by waiting for responses on all the IO requests,
4794          * but better performance may result in retrying IO that
4795          * receives an old_stateid error if requests are rarely
4796          * reordered in flight:
4797          */
4798         return nfserr_old_stateid;
4799 }
4800
4801 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4802 {
4803         if (ols->st_stateowner->so_is_open_owner &&
4804             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4805                 return nfserr_bad_stateid;
4806         return nfs_ok;
4807 }
4808
4809 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4810 {
4811         struct nfs4_stid *s;
4812         __be32 status = nfserr_bad_stateid;
4813
4814         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4815                 return status;
4816         /* Client debugging aid. */
4817         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4818                 char addr_str[INET6_ADDRSTRLEN];
4819                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4820                                  sizeof(addr_str));
4821                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4822                                         "with incorrect client ID\n", addr_str);
4823                 return status;
4824         }
4825         spin_lock(&cl->cl_lock);
4826         s = find_stateid_locked(cl, stateid);
4827         if (!s)
4828                 goto out_unlock;
4829         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4830         if (status)
4831                 goto out_unlock;
4832         switch (s->sc_type) {
4833         case NFS4_DELEG_STID:
4834                 status = nfs_ok;
4835                 break;
4836         case NFS4_REVOKED_DELEG_STID:
4837                 status = nfserr_deleg_revoked;
4838                 break;
4839         case NFS4_OPEN_STID:
4840         case NFS4_LOCK_STID:
4841                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4842                 break;
4843         default:
4844                 printk("unknown stateid type %x\n", s->sc_type);
4845                 /* Fallthrough */
4846         case NFS4_CLOSED_STID:
4847         case NFS4_CLOSED_DELEG_STID:
4848                 status = nfserr_bad_stateid;
4849         }
4850 out_unlock:
4851         spin_unlock(&cl->cl_lock);
4852         return status;
4853 }
4854
4855 __be32
4856 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4857                      stateid_t *stateid, unsigned char typemask,
4858                      struct nfs4_stid **s, struct nfsd_net *nn)
4859 {
4860         __be32 status;
4861
4862         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4863                 return nfserr_bad_stateid;
4864         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4865         if (status == nfserr_stale_clientid) {
4866                 if (cstate->session)
4867                         return nfserr_bad_stateid;
4868                 return nfserr_stale_stateid;
4869         }
4870         if (status)
4871                 return status;
4872         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4873         if (!*s)
4874                 return nfserr_bad_stateid;
4875         return nfs_ok;
4876 }
4877
4878 static struct file *
4879 nfs4_find_file(struct nfs4_stid *s, int flags)
4880 {
4881         if (!s)
4882                 return NULL;
4883
4884         switch (s->sc_type) {
4885         case NFS4_DELEG_STID:
4886                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4887                         return NULL;
4888                 return get_file(s->sc_file->fi_deleg_file);
4889         case NFS4_OPEN_STID:
4890         case NFS4_LOCK_STID:
4891                 if (flags & RD_STATE)
4892                         return find_readable_file(s->sc_file);
4893                 else
4894                         return find_writeable_file(s->sc_file);
4895                 break;
4896         }
4897
4898         return NULL;
4899 }
4900
4901 static __be32
4902 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4903 {
4904         __be32 status;
4905
4906         status = nfsd4_check_openowner_confirmed(ols);
4907         if (status)
4908                 return status;
4909         return nfs4_check_openmode(ols, flags);
4910 }
4911
4912 static __be32
4913 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4914                 struct file **filpp, bool *tmp_file, int flags)
4915 {
4916         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4917         struct file *file;
4918         __be32 status;
4919
4920         file = nfs4_find_file(s, flags);
4921         if (file) {
4922                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4923                                 acc | NFSD_MAY_OWNER_OVERRIDE);
4924                 if (status) {
4925                         fput(file);
4926                         return status;
4927                 }
4928
4929                 *filpp = file;
4930         } else {
4931                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4932                 if (status)
4933                         return status;
4934
4935                 if (tmp_file)
4936                         *tmp_file = true;
4937         }
4938
4939         return 0;
4940 }
4941
4942 /*
4943  * Checks for stateid operations
4944  */
4945 __be32
4946 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4947                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4948                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4949 {
4950         struct inode *ino = d_inode(fhp->fh_dentry);
4951         struct net *net = SVC_NET(rqstp);
4952         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4953         struct nfs4_stid *s = NULL;
4954         __be32 status;
4955
4956         if (filpp)
4957                 *filpp = NULL;
4958         if (tmp_file)
4959                 *tmp_file = false;
4960
4961         if (grace_disallows_io(net, ino))
4962                 return nfserr_grace;
4963
4964         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4965                 status = check_special_stateids(net, fhp, stateid, flags);
4966                 goto done;
4967         }
4968
4969         status = nfsd4_lookup_stateid(cstate, stateid,
4970                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4971                                 &s, nn);
4972         if (status)
4973                 return status;
4974         status = check_stateid_generation(stateid, &s->sc_stateid,
4975                         nfsd4_has_session(cstate));
4976         if (status)
4977                 goto out;
4978
4979         switch (s->sc_type) {
4980         case NFS4_DELEG_STID:
4981                 status = nfs4_check_delegmode(delegstateid(s), flags);
4982                 break;
4983         case NFS4_OPEN_STID:
4984         case NFS4_LOCK_STID:
4985                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4986                 break;
4987         default:
4988                 status = nfserr_bad_stateid;
4989                 break;
4990         }
4991         if (status)
4992                 goto out;
4993         status = nfs4_check_fh(fhp, s);
4994
4995 done:
4996         if (!status && filpp)
4997                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
4998 out:
4999         if (s)
5000                 nfs4_put_stid(s);
5001         return status;
5002 }
5003
5004 /*
5005  * Test if the stateid is valid
5006  */
5007 __be32
5008 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5009                    struct nfsd4_test_stateid *test_stateid)
5010 {
5011         struct nfsd4_test_stateid_id *stateid;
5012         struct nfs4_client *cl = cstate->session->se_client;
5013
5014         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5015                 stateid->ts_id_status =
5016                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5017
5018         return nfs_ok;
5019 }
5020
5021 static __be32
5022 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5023 {
5024         struct nfs4_ol_stateid *stp = openlockstateid(s);
5025         __be32 ret;
5026
5027         mutex_lock(&stp->st_mutex);
5028
5029         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5030         if (ret)
5031                 goto out;
5032
5033         ret = nfserr_locks_held;
5034         if (check_for_locks(stp->st_stid.sc_file,
5035                             lockowner(stp->st_stateowner)))
5036                 goto out;
5037
5038         release_lock_stateid(stp);
5039         ret = nfs_ok;
5040
5041 out:
5042         mutex_unlock(&stp->st_mutex);
5043         nfs4_put_stid(s);
5044         return ret;
5045 }
5046
5047 __be32
5048 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5049                    struct nfsd4_free_stateid *free_stateid)
5050 {
5051         stateid_t *stateid = &free_stateid->fr_stateid;
5052         struct nfs4_stid *s;
5053         struct nfs4_delegation *dp;
5054         struct nfs4_client *cl = cstate->session->se_client;
5055         __be32 ret = nfserr_bad_stateid;
5056
5057         spin_lock(&cl->cl_lock);
5058         s = find_stateid_locked(cl, stateid);
5059         if (!s)
5060                 goto out_unlock;
5061         switch (s->sc_type) {
5062         case NFS4_DELEG_STID:
5063                 ret = nfserr_locks_held;
5064                 break;
5065         case NFS4_OPEN_STID:
5066                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5067                 if (ret)
5068                         break;
5069                 ret = nfserr_locks_held;
5070                 break;
5071         case NFS4_LOCK_STID:
5072                 atomic_inc(&s->sc_count);
5073                 spin_unlock(&cl->cl_lock);
5074                 ret = nfsd4_free_lock_stateid(stateid, s);
5075                 goto out;
5076         case NFS4_REVOKED_DELEG_STID:
5077                 dp = delegstateid(s);
5078                 list_del_init(&dp->dl_recall_lru);
5079                 spin_unlock(&cl->cl_lock);
5080                 nfs4_put_stid(s);
5081                 ret = nfs_ok;
5082                 goto out;
5083         /* Default falls through and returns nfserr_bad_stateid */
5084         }
5085 out_unlock:
5086         spin_unlock(&cl->cl_lock);
5087 out:
5088         return ret;
5089 }
5090
5091 static inline int
5092 setlkflg (int type)
5093 {
5094         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5095                 RD_STATE : WR_STATE;
5096 }
5097
5098 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5099 {
5100         struct svc_fh *current_fh = &cstate->current_fh;
5101         struct nfs4_stateowner *sop = stp->st_stateowner;
5102         __be32 status;
5103
5104         status = nfsd4_check_seqid(cstate, sop, seqid);
5105         if (status)
5106                 return status;
5107         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
5108                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
5109                 /*
5110                  * "Closed" stateid's exist *only* to return
5111                  * nfserr_replay_me from the previous step, and
5112                  * revoked delegations are kept only for free_stateid.
5113                  */
5114                 return nfserr_bad_stateid;
5115         mutex_lock(&stp->st_mutex);
5116         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5117         if (status == nfs_ok)
5118                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5119         if (status != nfs_ok)
5120                 mutex_unlock(&stp->st_mutex);
5121         return status;
5122 }
5123
5124 /*
5125  * Checks for sequence id mutating operations.
5126  */
5127 static __be32
5128 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5129                          stateid_t *stateid, char typemask,
5130                          struct nfs4_ol_stateid **stpp,
5131                          struct nfsd_net *nn)
5132 {
5133         __be32 status;
5134         struct nfs4_stid *s;
5135         struct nfs4_ol_stateid *stp = NULL;
5136
5137         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5138                 seqid, STATEID_VAL(stateid));
5139
5140         *stpp = NULL;
5141         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5142         if (status)
5143                 return status;
5144         stp = openlockstateid(s);
5145         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5146
5147         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5148         if (!status)
5149                 *stpp = stp;
5150         else
5151                 nfs4_put_stid(&stp->st_stid);
5152         return status;
5153 }
5154
5155 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5156                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5157 {
5158         __be32 status;
5159         struct nfs4_openowner *oo;
5160         struct nfs4_ol_stateid *stp;
5161
5162         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5163                                                 NFS4_OPEN_STID, &stp, nn);
5164         if (status)
5165                 return status;
5166         oo = openowner(stp->st_stateowner);
5167         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5168                 mutex_unlock(&stp->st_mutex);
5169                 nfs4_put_stid(&stp->st_stid);
5170                 return nfserr_bad_stateid;
5171         }
5172         *stpp = stp;
5173         return nfs_ok;
5174 }
5175
5176 __be32
5177 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5178                    struct nfsd4_open_confirm *oc)
5179 {
5180         __be32 status;
5181         struct nfs4_openowner *oo;
5182         struct nfs4_ol_stateid *stp;
5183         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5184
5185         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5186                         cstate->current_fh.fh_dentry);
5187
5188         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5189         if (status)
5190                 return status;
5191
5192         status = nfs4_preprocess_seqid_op(cstate,
5193                                         oc->oc_seqid, &oc->oc_req_stateid,
5194                                         NFS4_OPEN_STID, &stp, nn);
5195         if (status)
5196                 goto out;
5197         oo = openowner(stp->st_stateowner);
5198         status = nfserr_bad_stateid;
5199         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5200                 mutex_unlock(&stp->st_mutex);
5201                 goto put_stateid;
5202         }
5203         oo->oo_flags |= NFS4_OO_CONFIRMED;
5204         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5205         mutex_unlock(&stp->st_mutex);
5206         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5207                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5208
5209         nfsd4_client_record_create(oo->oo_owner.so_client);
5210         status = nfs_ok;
5211 put_stateid:
5212         nfs4_put_stid(&stp->st_stid);
5213 out:
5214         nfsd4_bump_seqid(cstate, status);
5215         return status;
5216 }
5217
5218 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5219 {
5220         if (!test_access(access, stp))
5221                 return;
5222         nfs4_file_put_access(stp->st_stid.sc_file, access);
5223         clear_access(access, stp);
5224 }
5225
5226 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5227 {
5228         switch (to_access) {
5229         case NFS4_SHARE_ACCESS_READ:
5230                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5231                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5232                 break;
5233         case NFS4_SHARE_ACCESS_WRITE:
5234                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5235                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5236                 break;
5237         case NFS4_SHARE_ACCESS_BOTH:
5238                 break;
5239         default:
5240                 WARN_ON_ONCE(1);
5241         }
5242 }
5243
5244 __be32
5245 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5246                      struct nfsd4_compound_state *cstate,
5247                      struct nfsd4_open_downgrade *od)
5248 {
5249         __be32 status;
5250         struct nfs4_ol_stateid *stp;
5251         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5252
5253         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5254                         cstate->current_fh.fh_dentry);
5255
5256         /* We don't yet support WANT bits: */
5257         if (od->od_deleg_want)
5258                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5259                         od->od_deleg_want);
5260
5261         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5262                                         &od->od_stateid, &stp, nn);
5263         if (status)
5264                 goto out;
5265         status = nfserr_inval;
5266         if (!test_access(od->od_share_access, stp)) {
5267                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5268                         stp->st_access_bmap, od->od_share_access);
5269                 goto put_stateid;
5270         }
5271         if (!test_deny(od->od_share_deny, stp)) {
5272                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5273                         stp->st_deny_bmap, od->od_share_deny);
5274                 goto put_stateid;
5275         }
5276         nfs4_stateid_downgrade(stp, od->od_share_access);
5277         reset_union_bmap_deny(od->od_share_deny, stp);
5278         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5279         status = nfs_ok;
5280 put_stateid:
5281         mutex_unlock(&stp->st_mutex);
5282         nfs4_put_stid(&stp->st_stid);
5283 out:
5284         nfsd4_bump_seqid(cstate, status);
5285         return status;
5286 }
5287
5288 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5289 {
5290         struct nfs4_client *clp = s->st_stid.sc_client;
5291         bool unhashed;
5292         LIST_HEAD(reaplist);
5293
5294         s->st_stid.sc_type = NFS4_CLOSED_STID;
5295         spin_lock(&clp->cl_lock);
5296         unhashed = unhash_open_stateid(s, &reaplist);
5297
5298         if (clp->cl_minorversion) {
5299                 if (unhashed)
5300                         put_ol_stateid_locked(s, &reaplist);
5301                 spin_unlock(&clp->cl_lock);
5302                 free_ol_stateid_reaplist(&reaplist);
5303         } else {
5304                 spin_unlock(&clp->cl_lock);
5305                 free_ol_stateid_reaplist(&reaplist);
5306                 if (unhashed)
5307                         move_to_close_lru(s, clp->net);
5308         }
5309 }
5310
5311 /*
5312  * nfs4_unlock_state() called after encode
5313  */
5314 __be32
5315 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5316             struct nfsd4_close *close)
5317 {
5318         __be32 status;
5319         struct nfs4_ol_stateid *stp;
5320         struct net *net = SVC_NET(rqstp);
5321         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5322
5323         dprintk("NFSD: nfsd4_close on file %pd\n",
5324                         cstate->current_fh.fh_dentry);
5325
5326         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5327                                         &close->cl_stateid,
5328                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5329                                         &stp, nn);
5330         nfsd4_bump_seqid(cstate, status);
5331         if (status)
5332                 goto out;
5333         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5334         mutex_unlock(&stp->st_mutex);
5335
5336         nfsd4_close_open_stateid(stp);
5337
5338         /* put reference from nfs4_preprocess_seqid_op */
5339         nfs4_put_stid(&stp->st_stid);
5340 out:
5341         return status;
5342 }
5343
5344 __be32
5345 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5346                   struct nfsd4_delegreturn *dr)
5347 {
5348         struct nfs4_delegation *dp;
5349         stateid_t *stateid = &dr->dr_stateid;
5350         struct nfs4_stid *s;
5351         __be32 status;
5352         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5353
5354         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5355                 return status;
5356
5357         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5358         if (status)
5359                 goto out;
5360         dp = delegstateid(s);
5361         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5362         if (status)
5363                 goto put_stateid;
5364
5365         destroy_delegation(dp);
5366 put_stateid:
5367         nfs4_put_stid(&dp->dl_stid);
5368 out:
5369         return status;
5370 }
5371
5372 static inline u64
5373 end_offset(u64 start, u64 len)
5374 {
5375         u64 end;
5376
5377         end = start + len;
5378         return end >= start ? end: NFS4_MAX_UINT64;
5379 }
5380
5381 /* last octet in a range */
5382 static inline u64
5383 last_byte_offset(u64 start, u64 len)
5384 {
5385         u64 end;
5386
5387         WARN_ON_ONCE(!len);
5388         end = start + len;
5389         return end > start ? end - 1: NFS4_MAX_UINT64;
5390 }
5391
5392 /*
5393  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5394  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5395  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5396  * locking, this prevents us from being completely protocol-compliant.  The
5397  * real solution to this problem is to start using unsigned file offsets in
5398  * the VFS, but this is a very deep change!
5399  */
5400 static inline void
5401 nfs4_transform_lock_offset(struct file_lock *lock)
5402 {
5403         if (lock->fl_start < 0)
5404                 lock->fl_start = OFFSET_MAX;
5405         if (lock->fl_end < 0)
5406                 lock->fl_end = OFFSET_MAX;
5407 }
5408
5409 static fl_owner_t
5410 nfsd4_fl_get_owner(fl_owner_t owner)
5411 {
5412         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5413
5414         nfs4_get_stateowner(&lo->lo_owner);
5415         return owner;
5416 }
5417
5418 static void
5419 nfsd4_fl_put_owner(fl_owner_t owner)
5420 {
5421         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5422
5423         if (lo)
5424                 nfs4_put_stateowner(&lo->lo_owner);
5425 }
5426
5427 static void
5428 nfsd4_lm_notify(struct file_lock *fl)
5429 {
5430         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5431         struct net                      *net = lo->lo_owner.so_client->net;
5432         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5433         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5434                                                 struct nfsd4_blocked_lock, nbl_lock);
5435         bool queue = false;
5436
5437         /* An empty list means that something else is going to be using it */
5438         spin_lock(&nn->blocked_locks_lock);
5439         if (!list_empty(&nbl->nbl_list)) {
5440                 list_del_init(&nbl->nbl_list);
5441                 list_del_init(&nbl->nbl_lru);
5442                 queue = true;
5443         }
5444         spin_unlock(&nn->blocked_locks_lock);
5445
5446         if (queue)
5447                 nfsd4_run_cb(&nbl->nbl_cb);
5448 }
5449
5450 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5451         .lm_notify = nfsd4_lm_notify,
5452         .lm_get_owner = nfsd4_fl_get_owner,
5453         .lm_put_owner = nfsd4_fl_put_owner,
5454 };
5455
5456 static inline void
5457 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5458 {
5459         struct nfs4_lockowner *lo;
5460
5461         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5462                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5463                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5464                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5465                 if (!deny->ld_owner.data)
5466                         /* We just don't care that much */
5467                         goto nevermind;
5468                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5469                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5470         } else {
5471 nevermind:
5472                 deny->ld_owner.len = 0;
5473                 deny->ld_owner.data = NULL;
5474                 deny->ld_clientid.cl_boot = 0;
5475                 deny->ld_clientid.cl_id = 0;
5476         }
5477         deny->ld_start = fl->fl_start;
5478         deny->ld_length = NFS4_MAX_UINT64;
5479         if (fl->fl_end != NFS4_MAX_UINT64)
5480                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5481         deny->ld_type = NFS4_READ_LT;
5482         if (fl->fl_type != F_RDLCK)
5483                 deny->ld_type = NFS4_WRITE_LT;
5484 }
5485
5486 static struct nfs4_lockowner *
5487 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5488 {
5489         unsigned int strhashval = ownerstr_hashval(owner);
5490         struct nfs4_stateowner *so;
5491
5492         lockdep_assert_held(&clp->cl_lock);
5493
5494         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5495                             so_strhash) {
5496                 if (so->so_is_open_owner)
5497                         continue;
5498                 if (same_owner_str(so, owner))
5499                         return lockowner(nfs4_get_stateowner(so));
5500         }
5501         return NULL;
5502 }
5503
5504 static struct nfs4_lockowner *
5505 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5506 {
5507         struct nfs4_lockowner *lo;
5508
5509         spin_lock(&clp->cl_lock);
5510         lo = find_lockowner_str_locked(clp, owner);
5511         spin_unlock(&clp->cl_lock);
5512         return lo;
5513 }
5514
5515 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5516 {
5517         unhash_lockowner_locked(lockowner(sop));
5518 }
5519
5520 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5521 {
5522         struct nfs4_lockowner *lo = lockowner(sop);
5523
5524         kmem_cache_free(lockowner_slab, lo);
5525 }
5526
5527 static const struct nfs4_stateowner_operations lockowner_ops = {
5528         .so_unhash =    nfs4_unhash_lockowner,
5529         .so_free =      nfs4_free_lockowner,
5530 };
5531
5532 /*
5533  * Alloc a lock owner structure.
5534  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5535  * occurred.
5536  *
5537  * strhashval = ownerstr_hashval
5538  */
5539 static struct nfs4_lockowner *
5540 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5541                            struct nfs4_ol_stateid *open_stp,
5542                            struct nfsd4_lock *lock)
5543 {
5544         struct nfs4_lockowner *lo, *ret;
5545
5546         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5547         if (!lo)
5548                 return NULL;
5549         INIT_LIST_HEAD(&lo->lo_blocked);
5550         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5551         lo->lo_owner.so_is_open_owner = 0;
5552         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5553         lo->lo_owner.so_ops = &lockowner_ops;
5554         spin_lock(&clp->cl_lock);
5555         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5556         if (ret == NULL) {
5557                 list_add(&lo->lo_owner.so_strhash,
5558                          &clp->cl_ownerstr_hashtbl[strhashval]);
5559                 ret = lo;
5560         } else
5561                 nfs4_free_stateowner(&lo->lo_owner);
5562
5563         spin_unlock(&clp->cl_lock);
5564         return ret;
5565 }
5566
5567 static void
5568 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5569                   struct nfs4_file *fp, struct inode *inode,
5570                   struct nfs4_ol_stateid *open_stp)
5571 {
5572         struct nfs4_client *clp = lo->lo_owner.so_client;
5573
5574         lockdep_assert_held(&clp->cl_lock);
5575
5576         atomic_inc(&stp->st_stid.sc_count);
5577         stp->st_stid.sc_type = NFS4_LOCK_STID;
5578         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5579         get_nfs4_file(fp);
5580         stp->st_stid.sc_file = fp;
5581         stp->st_access_bmap = 0;
5582         stp->st_deny_bmap = open_stp->st_deny_bmap;
5583         stp->st_openstp = open_stp;
5584         mutex_init(&stp->st_mutex);
5585         list_add(&stp->st_locks, &open_stp->st_locks);
5586         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5587         spin_lock(&fp->fi_lock);
5588         list_add(&stp->st_perfile, &fp->fi_stateids);
5589         spin_unlock(&fp->fi_lock);
5590 }
5591
5592 static struct nfs4_ol_stateid *
5593 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5594 {
5595         struct nfs4_ol_stateid *lst;
5596         struct nfs4_client *clp = lo->lo_owner.so_client;
5597
5598         lockdep_assert_held(&clp->cl_lock);
5599
5600         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5601                 if (lst->st_stid.sc_file == fp) {
5602                         atomic_inc(&lst->st_stid.sc_count);
5603                         return lst;
5604                 }
5605         }
5606         return NULL;
5607 }
5608
5609 static struct nfs4_ol_stateid *
5610 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5611                             struct inode *inode, struct nfs4_ol_stateid *ost,
5612                             bool *new)
5613 {
5614         struct nfs4_stid *ns = NULL;
5615         struct nfs4_ol_stateid *lst;
5616         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5617         struct nfs4_client *clp = oo->oo_owner.so_client;
5618
5619         spin_lock(&clp->cl_lock);
5620         lst = find_lock_stateid(lo, fi);
5621         if (lst == NULL) {
5622                 spin_unlock(&clp->cl_lock);
5623                 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5624                 if (ns == NULL)
5625                         return NULL;
5626
5627                 spin_lock(&clp->cl_lock);
5628                 lst = find_lock_stateid(lo, fi);
5629                 if (likely(!lst)) {
5630                         lst = openlockstateid(ns);
5631                         init_lock_stateid(lst, lo, fi, inode, ost);
5632                         ns = NULL;
5633                         *new = true;
5634                 }
5635         }
5636         spin_unlock(&clp->cl_lock);
5637         if (ns)
5638                 nfs4_put_stid(ns);
5639         return lst;
5640 }
5641
5642 static int
5643 check_lock_length(u64 offset, u64 length)
5644 {
5645         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5646                 (length > ~offset)));
5647 }
5648
5649 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5650 {
5651         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5652
5653         lockdep_assert_held(&fp->fi_lock);
5654
5655         if (test_access(access, lock_stp))
5656                 return;
5657         __nfs4_file_get_access(fp, access);
5658         set_access(access, lock_stp);
5659 }
5660
5661 static __be32
5662 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5663                             struct nfs4_ol_stateid *ost,
5664                             struct nfsd4_lock *lock,
5665                             struct nfs4_ol_stateid **plst, bool *new)
5666 {
5667         __be32 status;
5668         struct nfs4_file *fi = ost->st_stid.sc_file;
5669         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5670         struct nfs4_client *cl = oo->oo_owner.so_client;
5671         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5672         struct nfs4_lockowner *lo;
5673         struct nfs4_ol_stateid *lst;
5674         unsigned int strhashval;
5675         bool hashed;
5676
5677         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5678         if (!lo) {
5679                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5680                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5681                 if (lo == NULL)
5682                         return nfserr_jukebox;
5683         } else {
5684                 /* with an existing lockowner, seqids must be the same */
5685                 status = nfserr_bad_seqid;
5686                 if (!cstate->minorversion &&
5687                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5688                         goto out;
5689         }
5690
5691 retry:
5692         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5693         if (lst == NULL) {
5694                 status = nfserr_jukebox;
5695                 goto out;
5696         }
5697
5698         mutex_lock(&lst->st_mutex);
5699
5700         /* See if it's still hashed to avoid race with FREE_STATEID */
5701         spin_lock(&cl->cl_lock);
5702         hashed = !list_empty(&lst->st_perfile);
5703         spin_unlock(&cl->cl_lock);
5704
5705         if (!hashed) {
5706                 mutex_unlock(&lst->st_mutex);
5707                 nfs4_put_stid(&lst->st_stid);
5708                 goto retry;
5709         }
5710         status = nfs_ok;
5711         *plst = lst;
5712 out:
5713         nfs4_put_stateowner(&lo->lo_owner);
5714         return status;
5715 }
5716
5717 /*
5718  *  LOCK operation
5719  */
5720 __be32
5721 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5722            struct nfsd4_lock *lock)
5723 {
5724         struct nfs4_openowner *open_sop = NULL;
5725         struct nfs4_lockowner *lock_sop = NULL;
5726         struct nfs4_ol_stateid *lock_stp = NULL;
5727         struct nfs4_ol_stateid *open_stp = NULL;
5728         struct nfs4_file *fp;
5729         struct file *filp = NULL;
5730         struct nfsd4_blocked_lock *nbl = NULL;
5731         struct file_lock *file_lock = NULL;
5732         struct file_lock *conflock = NULL;
5733         __be32 status = 0;
5734         int lkflg;
5735         int err;
5736         bool new = false;
5737         unsigned char fl_type;
5738         unsigned int fl_flags = FL_POSIX;
5739         struct net *net = SVC_NET(rqstp);
5740         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5741
5742         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5743                 (long long) lock->lk_offset,
5744                 (long long) lock->lk_length);
5745
5746         if (check_lock_length(lock->lk_offset, lock->lk_length))
5747                  return nfserr_inval;
5748
5749         if ((status = fh_verify(rqstp, &cstate->current_fh,
5750                                 S_IFREG, NFSD_MAY_LOCK))) {
5751                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5752                 return status;
5753         }
5754
5755         if (lock->lk_is_new) {
5756                 if (nfsd4_has_session(cstate))
5757                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5758                         memcpy(&lock->lk_new_clientid,
5759                                 &cstate->session->se_client->cl_clientid,
5760                                 sizeof(clientid_t));
5761
5762                 status = nfserr_stale_clientid;
5763                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5764                         goto out;
5765
5766                 /* validate and update open stateid and open seqid */
5767                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5768                                         lock->lk_new_open_seqid,
5769                                         &lock->lk_new_open_stateid,
5770                                         &open_stp, nn);
5771                 if (status)
5772                         goto out;
5773                 mutex_unlock(&open_stp->st_mutex);
5774                 open_sop = openowner(open_stp->st_stateowner);
5775                 status = nfserr_bad_stateid;
5776                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5777                                                 &lock->lk_new_clientid))
5778                         goto out;
5779                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5780                                                         &lock_stp, &new);
5781         } else {
5782                 status = nfs4_preprocess_seqid_op(cstate,
5783                                        lock->lk_old_lock_seqid,
5784                                        &lock->lk_old_lock_stateid,
5785                                        NFS4_LOCK_STID, &lock_stp, nn);
5786         }
5787         if (status)
5788                 goto out;
5789         lock_sop = lockowner(lock_stp->st_stateowner);
5790
5791         lkflg = setlkflg(lock->lk_type);
5792         status = nfs4_check_openmode(lock_stp, lkflg);
5793         if (status)
5794                 goto out;
5795
5796         status = nfserr_grace;
5797         if (locks_in_grace(net) && !lock->lk_reclaim)
5798                 goto out;
5799         status = nfserr_no_grace;
5800         if (!locks_in_grace(net) && lock->lk_reclaim)
5801                 goto out;
5802
5803         fp = lock_stp->st_stid.sc_file;
5804         switch (lock->lk_type) {
5805                 case NFS4_READW_LT:
5806                         if (nfsd4_has_session(cstate))
5807                                 fl_flags |= FL_SLEEP;
5808                         /* Fallthrough */
5809                 case NFS4_READ_LT:
5810                         spin_lock(&fp->fi_lock);
5811                         filp = find_readable_file_locked(fp);
5812                         if (filp)
5813                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5814                         spin_unlock(&fp->fi_lock);
5815                         fl_type = F_RDLCK;
5816                         break;
5817                 case NFS4_WRITEW_LT:
5818                         if (nfsd4_has_session(cstate))
5819                                 fl_flags |= FL_SLEEP;
5820                         /* Fallthrough */
5821                 case NFS4_WRITE_LT:
5822                         spin_lock(&fp->fi_lock);
5823                         filp = find_writeable_file_locked(fp);
5824                         if (filp)
5825                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5826                         spin_unlock(&fp->fi_lock);
5827                         fl_type = F_WRLCK;
5828                         break;
5829                 default:
5830                         status = nfserr_inval;
5831                 goto out;
5832         }
5833
5834         if (!filp) {
5835                 status = nfserr_openmode;
5836                 goto out;
5837         }
5838
5839         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
5840         if (!nbl) {
5841                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
5842                 status = nfserr_jukebox;
5843                 goto out;
5844         }
5845
5846         file_lock = &nbl->nbl_lock;
5847         file_lock->fl_type = fl_type;
5848         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5849         file_lock->fl_pid = current->tgid;
5850         file_lock->fl_file = filp;
5851         file_lock->fl_flags = fl_flags;
5852         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5853         file_lock->fl_start = lock->lk_offset;
5854         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5855         nfs4_transform_lock_offset(file_lock);
5856
5857         conflock = locks_alloc_lock();
5858         if (!conflock) {
5859                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5860                 status = nfserr_jukebox;
5861                 goto out;
5862         }
5863
5864         if (fl_flags & FL_SLEEP) {
5865                 nbl->nbl_time = jiffies;
5866                 spin_lock(&nn->blocked_locks_lock);
5867                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
5868                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
5869                 spin_unlock(&nn->blocked_locks_lock);
5870         }
5871
5872         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5873         switch (err) {
5874         case 0: /* success! */
5875                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5876                 status = 0;
5877                 break;
5878         case FILE_LOCK_DEFERRED:
5879                 nbl = NULL;
5880                 /* Fallthrough */
5881         case -EAGAIN:           /* conflock holds conflicting lock */
5882                 status = nfserr_denied;
5883                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5884                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5885                 break;
5886         case -EDEADLK:
5887                 status = nfserr_deadlock;
5888                 break;
5889         default:
5890                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5891                 status = nfserrno(err);
5892                 break;
5893         }
5894 out:
5895         if (nbl) {
5896                 /* dequeue it if we queued it before */
5897                 if (fl_flags & FL_SLEEP) {
5898                         spin_lock(&nn->blocked_locks_lock);
5899                         list_del_init(&nbl->nbl_list);
5900                         list_del_init(&nbl->nbl_lru);
5901                         spin_unlock(&nn->blocked_locks_lock);
5902                 }
5903                 free_blocked_lock(nbl);
5904         }
5905         if (filp)
5906                 fput(filp);
5907         if (lock_stp) {
5908                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5909                 if (cstate->replay_owner &&
5910                     cstate->replay_owner != &lock_sop->lo_owner &&
5911                     seqid_mutating_err(ntohl(status)))
5912                         lock_sop->lo_owner.so_seqid++;
5913
5914                 mutex_unlock(&lock_stp->st_mutex);
5915
5916                 /*
5917                  * If this is a new, never-before-used stateid, and we are
5918                  * returning an error, then just go ahead and release it.
5919                  */
5920                 if (status && new)
5921                         release_lock_stateid(lock_stp);
5922
5923                 nfs4_put_stid(&lock_stp->st_stid);
5924         }
5925         if (open_stp)
5926                 nfs4_put_stid(&open_stp->st_stid);
5927         nfsd4_bump_seqid(cstate, status);
5928         if (conflock)
5929                 locks_free_lock(conflock);
5930         return status;
5931 }
5932
5933 /*
5934  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5935  * so we do a temporary open here just to get an open file to pass to
5936  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5937  * inode operation.)
5938  */
5939 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5940 {
5941         struct file *file;
5942         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5943         if (!err) {
5944                 err = nfserrno(vfs_test_lock(file, lock));
5945                 fput(file);
5946         }
5947         return err;
5948 }
5949
5950 /*
5951  * LOCKT operation
5952  */
5953 __be32
5954 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5955             struct nfsd4_lockt *lockt)
5956 {
5957         struct file_lock *file_lock = NULL;
5958         struct nfs4_lockowner *lo = NULL;
5959         __be32 status;
5960         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5961
5962         if (locks_in_grace(SVC_NET(rqstp)))
5963                 return nfserr_grace;
5964
5965         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5966                  return nfserr_inval;
5967
5968         if (!nfsd4_has_session(cstate)) {
5969                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5970                 if (status)
5971                         goto out;
5972         }
5973
5974         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5975                 goto out;
5976
5977         file_lock = locks_alloc_lock();
5978         if (!file_lock) {
5979                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5980                 status = nfserr_jukebox;
5981                 goto out;
5982         }
5983
5984         switch (lockt->lt_type) {
5985                 case NFS4_READ_LT:
5986                 case NFS4_READW_LT:
5987                         file_lock->fl_type = F_RDLCK;
5988                 break;
5989                 case NFS4_WRITE_LT:
5990                 case NFS4_WRITEW_LT:
5991                         file_lock->fl_type = F_WRLCK;
5992                 break;
5993                 default:
5994                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5995                         status = nfserr_inval;
5996                 goto out;
5997         }
5998
5999         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6000         if (lo)
6001                 file_lock->fl_owner = (fl_owner_t)lo;
6002         file_lock->fl_pid = current->tgid;
6003         file_lock->fl_flags = FL_POSIX;
6004
6005         file_lock->fl_start = lockt->lt_offset;
6006         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6007
6008         nfs4_transform_lock_offset(file_lock);
6009
6010         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6011         if (status)
6012                 goto out;
6013
6014         if (file_lock->fl_type != F_UNLCK) {
6015                 status = nfserr_denied;
6016                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6017         }
6018 out:
6019         if (lo)
6020                 nfs4_put_stateowner(&lo->lo_owner);
6021         if (file_lock)
6022                 locks_free_lock(file_lock);
6023         return status;
6024 }
6025
6026 __be32
6027 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6028             struct nfsd4_locku *locku)
6029 {
6030         struct nfs4_ol_stateid *stp;
6031         struct file *filp = NULL;
6032         struct file_lock *file_lock = NULL;
6033         __be32 status;
6034         int err;
6035         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6036
6037         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6038                 (long long) locku->lu_offset,
6039                 (long long) locku->lu_length);
6040
6041         if (check_lock_length(locku->lu_offset, locku->lu_length))
6042                  return nfserr_inval;
6043
6044         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6045                                         &locku->lu_stateid, NFS4_LOCK_STID,
6046                                         &stp, nn);
6047         if (status)
6048                 goto out;
6049         filp = find_any_file(stp->st_stid.sc_file);
6050         if (!filp) {
6051                 status = nfserr_lock_range;
6052                 goto put_stateid;
6053         }
6054         file_lock = locks_alloc_lock();
6055         if (!file_lock) {
6056                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6057                 status = nfserr_jukebox;
6058                 goto fput;
6059         }
6060
6061         file_lock->fl_type = F_UNLCK;
6062         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6063         file_lock->fl_pid = current->tgid;
6064         file_lock->fl_file = filp;
6065         file_lock->fl_flags = FL_POSIX;
6066         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6067         file_lock->fl_start = locku->lu_offset;
6068
6069         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6070                                                 locku->lu_length);
6071         nfs4_transform_lock_offset(file_lock);
6072
6073         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6074         if (err) {
6075                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6076                 goto out_nfserr;
6077         }
6078         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6079 fput:
6080         fput(filp);
6081 put_stateid:
6082         mutex_unlock(&stp->st_mutex);
6083         nfs4_put_stid(&stp->st_stid);
6084 out:
6085         nfsd4_bump_seqid(cstate, status);
6086         if (file_lock)
6087                 locks_free_lock(file_lock);
6088         return status;
6089
6090 out_nfserr:
6091         status = nfserrno(err);
6092         goto fput;
6093 }
6094
6095 /*
6096  * returns
6097  *      true:  locks held by lockowner
6098  *      false: no locks held by lockowner
6099  */
6100 static bool
6101 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6102 {
6103         struct file_lock *fl;
6104         int status = false;
6105         struct file *filp = find_any_file(fp);
6106         struct inode *inode;
6107         struct file_lock_context *flctx;
6108
6109         if (!filp) {
6110                 /* Any valid lock stateid should have some sort of access */
6111                 WARN_ON_ONCE(1);
6112                 return status;
6113         }
6114
6115         inode = file_inode(filp);
6116         flctx = inode->i_flctx;
6117
6118         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6119                 spin_lock(&flctx->flc_lock);
6120                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6121                         if (fl->fl_owner == (fl_owner_t)lowner) {
6122                                 status = true;
6123                                 break;
6124                         }
6125                 }
6126                 spin_unlock(&flctx->flc_lock);
6127         }
6128         fput(filp);
6129         return status;
6130 }
6131
6132 __be32
6133 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6134                         struct nfsd4_compound_state *cstate,
6135                         struct nfsd4_release_lockowner *rlockowner)
6136 {
6137         clientid_t *clid = &rlockowner->rl_clientid;
6138         struct nfs4_stateowner *sop;
6139         struct nfs4_lockowner *lo = NULL;
6140         struct nfs4_ol_stateid *stp;
6141         struct xdr_netobj *owner = &rlockowner->rl_owner;
6142         unsigned int hashval = ownerstr_hashval(owner);
6143         __be32 status;
6144         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6145         struct nfs4_client *clp;
6146         LIST_HEAD (reaplist);
6147
6148         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6149                 clid->cl_boot, clid->cl_id);
6150
6151         status = lookup_clientid(clid, cstate, nn);
6152         if (status)
6153                 return status;
6154
6155         clp = cstate->clp;
6156         /* Find the matching lock stateowner */
6157         spin_lock(&clp->cl_lock);
6158         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6159                             so_strhash) {
6160
6161                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6162                         continue;
6163
6164                 /* see if there are still any locks associated with it */
6165                 lo = lockowner(sop);
6166                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6167                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6168                                 status = nfserr_locks_held;
6169                                 spin_unlock(&clp->cl_lock);
6170                                 return status;
6171                         }
6172                 }
6173
6174                 nfs4_get_stateowner(sop);
6175                 break;
6176         }
6177         if (!lo) {
6178                 spin_unlock(&clp->cl_lock);
6179                 return status;
6180         }
6181
6182         unhash_lockowner_locked(lo);
6183         while (!list_empty(&lo->lo_owner.so_stateids)) {
6184                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6185                                        struct nfs4_ol_stateid,
6186                                        st_perstateowner);
6187                 WARN_ON(!unhash_lock_stateid(stp));
6188                 put_ol_stateid_locked(stp, &reaplist);
6189         }
6190         spin_unlock(&clp->cl_lock);
6191         free_ol_stateid_reaplist(&reaplist);
6192         nfs4_put_stateowner(&lo->lo_owner);
6193
6194         return status;
6195 }
6196
6197 static inline struct nfs4_client_reclaim *
6198 alloc_reclaim(void)
6199 {
6200         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6201 }
6202
6203 bool
6204 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6205 {
6206         struct nfs4_client_reclaim *crp;
6207
6208         crp = nfsd4_find_reclaim_client(name, nn);
6209         return (crp && crp->cr_clp);
6210 }
6211
6212 /*
6213  * failure => all reset bets are off, nfserr_no_grace...
6214  */
6215 struct nfs4_client_reclaim *
6216 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6217 {
6218         unsigned int strhashval;
6219         struct nfs4_client_reclaim *crp;
6220
6221         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6222         crp = alloc_reclaim();
6223         if (crp) {
6224                 strhashval = clientstr_hashval(name);
6225                 INIT_LIST_HEAD(&crp->cr_strhash);
6226                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6227                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6228                 crp->cr_clp = NULL;
6229                 nn->reclaim_str_hashtbl_size++;
6230         }
6231         return crp;
6232 }
6233
6234 void
6235 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6236 {
6237         list_del(&crp->cr_strhash);
6238         kfree(crp);
6239         nn->reclaim_str_hashtbl_size--;
6240 }
6241
6242 void
6243 nfs4_release_reclaim(struct nfsd_net *nn)
6244 {
6245         struct nfs4_client_reclaim *crp = NULL;
6246         int i;
6247
6248         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6249                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6250                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6251                                         struct nfs4_client_reclaim, cr_strhash);
6252                         nfs4_remove_reclaim_record(crp, nn);
6253                 }
6254         }
6255         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6256 }
6257
6258 /*
6259  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6260 struct nfs4_client_reclaim *
6261 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6262 {
6263         unsigned int strhashval;
6264         struct nfs4_client_reclaim *crp = NULL;
6265
6266         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6267
6268         strhashval = clientstr_hashval(recdir);
6269         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6270                 if (same_name(crp->cr_recdir, recdir)) {
6271                         return crp;
6272                 }
6273         }
6274         return NULL;
6275 }
6276
6277 /*
6278 * Called from OPEN. Look for clientid in reclaim list.
6279 */
6280 __be32
6281 nfs4_check_open_reclaim(clientid_t *clid,
6282                 struct nfsd4_compound_state *cstate,
6283                 struct nfsd_net *nn)
6284 {
6285         __be32 status;
6286
6287         /* find clientid in conf_id_hashtbl */
6288         status = lookup_clientid(clid, cstate, nn);
6289         if (status)
6290                 return nfserr_reclaim_bad;
6291
6292         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6293                 return nfserr_no_grace;
6294
6295         if (nfsd4_client_record_check(cstate->clp))
6296                 return nfserr_reclaim_bad;
6297
6298         return nfs_ok;
6299 }
6300
6301 #ifdef CONFIG_NFSD_FAULT_INJECTION
6302 static inline void
6303 put_client(struct nfs4_client *clp)
6304 {
6305         atomic_dec(&clp->cl_refcount);
6306 }
6307
6308 static struct nfs4_client *
6309 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6310 {
6311         struct nfs4_client *clp;
6312         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6313                                           nfsd_net_id);
6314
6315         if (!nfsd_netns_ready(nn))
6316                 return NULL;
6317
6318         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6319                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6320                         return clp;
6321         }
6322         return NULL;
6323 }
6324
6325 u64
6326 nfsd_inject_print_clients(void)
6327 {
6328         struct nfs4_client *clp;
6329         u64 count = 0;
6330         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6331                                           nfsd_net_id);
6332         char buf[INET6_ADDRSTRLEN];
6333
6334         if (!nfsd_netns_ready(nn))
6335                 return 0;
6336
6337         spin_lock(&nn->client_lock);
6338         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6339                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6340                 pr_info("NFS Client: %s\n", buf);
6341                 ++count;
6342         }
6343         spin_unlock(&nn->client_lock);
6344
6345         return count;
6346 }
6347
6348 u64
6349 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6350 {
6351         u64 count = 0;
6352         struct nfs4_client *clp;
6353         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6354                                           nfsd_net_id);
6355
6356         if (!nfsd_netns_ready(nn))
6357                 return count;
6358
6359         spin_lock(&nn->client_lock);
6360         clp = nfsd_find_client(addr, addr_size);
6361         if (clp) {
6362                 if (mark_client_expired_locked(clp) == nfs_ok)
6363                         ++count;
6364                 else
6365                         clp = NULL;
6366         }
6367         spin_unlock(&nn->client_lock);
6368
6369         if (clp)
6370                 expire_client(clp);
6371
6372         return count;
6373 }
6374
6375 u64
6376 nfsd_inject_forget_clients(u64 max)
6377 {
6378         u64 count = 0;
6379         struct nfs4_client *clp, *next;
6380         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6381                                                 nfsd_net_id);
6382         LIST_HEAD(reaplist);
6383
6384         if (!nfsd_netns_ready(nn))
6385                 return count;
6386
6387         spin_lock(&nn->client_lock);
6388         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6389                 if (mark_client_expired_locked(clp) == nfs_ok) {
6390                         list_add(&clp->cl_lru, &reaplist);
6391                         if (max != 0 && ++count >= max)
6392                                 break;
6393                 }
6394         }
6395         spin_unlock(&nn->client_lock);
6396
6397         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6398                 expire_client(clp);
6399
6400         return count;
6401 }
6402
6403 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6404                              const char *type)
6405 {
6406         char buf[INET6_ADDRSTRLEN];
6407         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6408         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6409 }
6410
6411 static void
6412 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6413                              struct list_head *collect)
6414 {
6415         struct nfs4_client *clp = lst->st_stid.sc_client;
6416         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6417                                           nfsd_net_id);
6418
6419         if (!collect)
6420                 return;
6421
6422         lockdep_assert_held(&nn->client_lock);
6423         atomic_inc(&clp->cl_refcount);
6424         list_add(&lst->st_locks, collect);
6425 }
6426
6427 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6428                                     struct list_head *collect,
6429                                     bool (*func)(struct nfs4_ol_stateid *))
6430 {
6431         struct nfs4_openowner *oop;
6432         struct nfs4_ol_stateid *stp, *st_next;
6433         struct nfs4_ol_stateid *lst, *lst_next;
6434         u64 count = 0;
6435
6436         spin_lock(&clp->cl_lock);
6437         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6438                 list_for_each_entry_safe(stp, st_next,
6439                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6440                         list_for_each_entry_safe(lst, lst_next,
6441                                         &stp->st_locks, st_locks) {
6442                                 if (func) {
6443                                         if (func(lst))
6444                                                 nfsd_inject_add_lock_to_list(lst,
6445                                                                         collect);
6446                                 }
6447                                 ++count;
6448                                 /*
6449                                  * Despite the fact that these functions deal
6450                                  * with 64-bit integers for "count", we must
6451                                  * ensure that it doesn't blow up the
6452                                  * clp->cl_refcount. Throw a warning if we
6453                                  * start to approach INT_MAX here.
6454                                  */
6455                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6456                                 if (count == max)
6457                                         goto out;
6458                         }
6459                 }
6460         }
6461 out:
6462         spin_unlock(&clp->cl_lock);
6463
6464         return count;
6465 }
6466
6467 static u64
6468 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6469                           u64 max)
6470 {
6471         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6472 }
6473
6474 static u64
6475 nfsd_print_client_locks(struct nfs4_client *clp)
6476 {
6477         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6478         nfsd_print_count(clp, count, "locked files");
6479         return count;
6480 }
6481
6482 u64
6483 nfsd_inject_print_locks(void)
6484 {
6485         struct nfs4_client *clp;
6486         u64 count = 0;
6487         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6488                                                 nfsd_net_id);
6489
6490         if (!nfsd_netns_ready(nn))
6491                 return 0;
6492
6493         spin_lock(&nn->client_lock);
6494         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6495                 count += nfsd_print_client_locks(clp);
6496         spin_unlock(&nn->client_lock);
6497
6498         return count;
6499 }
6500
6501 static void
6502 nfsd_reap_locks(struct list_head *reaplist)
6503 {
6504         struct nfs4_client *clp;
6505         struct nfs4_ol_stateid *stp, *next;
6506
6507         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6508                 list_del_init(&stp->st_locks);
6509                 clp = stp->st_stid.sc_client;
6510                 nfs4_put_stid(&stp->st_stid);
6511                 put_client(clp);
6512         }
6513 }
6514
6515 u64
6516 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6517 {
6518         unsigned int count = 0;
6519         struct nfs4_client *clp;
6520         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6521                                                 nfsd_net_id);
6522         LIST_HEAD(reaplist);
6523
6524         if (!nfsd_netns_ready(nn))
6525                 return count;
6526
6527         spin_lock(&nn->client_lock);
6528         clp = nfsd_find_client(addr, addr_size);
6529         if (clp)
6530                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6531         spin_unlock(&nn->client_lock);
6532         nfsd_reap_locks(&reaplist);
6533         return count;
6534 }
6535
6536 u64
6537 nfsd_inject_forget_locks(u64 max)
6538 {
6539         u64 count = 0;
6540         struct nfs4_client *clp;
6541         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6542                                                 nfsd_net_id);
6543         LIST_HEAD(reaplist);
6544
6545         if (!nfsd_netns_ready(nn))
6546                 return count;
6547
6548         spin_lock(&nn->client_lock);
6549         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6550                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6551                 if (max != 0 && count >= max)
6552                         break;
6553         }
6554         spin_unlock(&nn->client_lock);
6555         nfsd_reap_locks(&reaplist);
6556         return count;
6557 }
6558
6559 static u64
6560 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6561                               struct list_head *collect,
6562                               void (*func)(struct nfs4_openowner *))
6563 {
6564         struct nfs4_openowner *oop, *next;
6565         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6566                                                 nfsd_net_id);
6567         u64 count = 0;
6568
6569         lockdep_assert_held(&nn->client_lock);
6570
6571         spin_lock(&clp->cl_lock);
6572         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6573                 if (func) {
6574                         func(oop);
6575                         if (collect) {
6576                                 atomic_inc(&clp->cl_refcount);
6577                                 list_add(&oop->oo_perclient, collect);
6578                         }
6579                 }
6580                 ++count;
6581                 /*
6582                  * Despite the fact that these functions deal with
6583                  * 64-bit integers for "count", we must ensure that
6584                  * it doesn't blow up the clp->cl_refcount. Throw a
6585                  * warning if we start to approach INT_MAX here.
6586                  */
6587                 WARN_ON_ONCE(count == (INT_MAX / 2));
6588                 if (count == max)
6589                         break;
6590         }
6591         spin_unlock(&clp->cl_lock);
6592
6593         return count;
6594 }
6595
6596 static u64
6597 nfsd_print_client_openowners(struct nfs4_client *clp)
6598 {
6599         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6600
6601         nfsd_print_count(clp, count, "openowners");
6602         return count;
6603 }
6604
6605 static u64
6606 nfsd_collect_client_openowners(struct nfs4_client *clp,
6607                                struct list_head *collect, u64 max)
6608 {
6609         return nfsd_foreach_client_openowner(clp, max, collect,
6610                                                 unhash_openowner_locked);
6611 }
6612
6613 u64
6614 nfsd_inject_print_openowners(void)
6615 {
6616         struct nfs4_client *clp;
6617         u64 count = 0;
6618         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6619                                                 nfsd_net_id);
6620
6621         if (!nfsd_netns_ready(nn))
6622                 return 0;
6623
6624         spin_lock(&nn->client_lock);
6625         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6626                 count += nfsd_print_client_openowners(clp);
6627         spin_unlock(&nn->client_lock);
6628
6629         return count;
6630 }
6631
6632 static void
6633 nfsd_reap_openowners(struct list_head *reaplist)
6634 {
6635         struct nfs4_client *clp;
6636         struct nfs4_openowner *oop, *next;
6637
6638         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6639                 list_del_init(&oop->oo_perclient);
6640                 clp = oop->oo_owner.so_client;
6641                 release_openowner(oop);
6642                 put_client(clp);
6643         }
6644 }
6645
6646 u64
6647 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6648                                      size_t addr_size)
6649 {
6650         unsigned int count = 0;
6651         struct nfs4_client *clp;
6652         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6653                                                 nfsd_net_id);
6654         LIST_HEAD(reaplist);
6655
6656         if (!nfsd_netns_ready(nn))
6657                 return count;
6658
6659         spin_lock(&nn->client_lock);
6660         clp = nfsd_find_client(addr, addr_size);
6661         if (clp)
6662                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6663         spin_unlock(&nn->client_lock);
6664         nfsd_reap_openowners(&reaplist);
6665         return count;
6666 }
6667
6668 u64
6669 nfsd_inject_forget_openowners(u64 max)
6670 {
6671         u64 count = 0;
6672         struct nfs4_client *clp;
6673         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6674                                                 nfsd_net_id);
6675         LIST_HEAD(reaplist);
6676
6677         if (!nfsd_netns_ready(nn))
6678                 return count;
6679
6680         spin_lock(&nn->client_lock);
6681         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6682                 count += nfsd_collect_client_openowners(clp, &reaplist,
6683                                                         max - count);
6684                 if (max != 0 && count >= max)
6685                         break;
6686         }
6687         spin_unlock(&nn->client_lock);
6688         nfsd_reap_openowners(&reaplist);
6689         return count;
6690 }
6691
6692 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6693                                      struct list_head *victims)
6694 {
6695         struct nfs4_delegation *dp, *next;
6696         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6697                                                 nfsd_net_id);
6698         u64 count = 0;
6699
6700         lockdep_assert_held(&nn->client_lock);
6701
6702         spin_lock(&state_lock);
6703         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6704                 if (victims) {
6705                         /*
6706                          * It's not safe to mess with delegations that have a
6707                          * non-zero dl_time. They might have already been broken
6708                          * and could be processed by the laundromat outside of
6709                          * the state_lock. Just leave them be.
6710                          */
6711                         if (dp->dl_time != 0)
6712                                 continue;
6713
6714                         atomic_inc(&clp->cl_refcount);
6715                         WARN_ON(!unhash_delegation_locked(dp));
6716                         list_add(&dp->dl_recall_lru, victims);
6717                 }
6718                 ++count;
6719                 /*
6720                  * Despite the fact that these functions deal with
6721                  * 64-bit integers for "count", we must ensure that
6722                  * it doesn't blow up the clp->cl_refcount. Throw a
6723                  * warning if we start to approach INT_MAX here.
6724                  */
6725                 WARN_ON_ONCE(count == (INT_MAX / 2));
6726                 if (count == max)
6727                         break;
6728         }
6729         spin_unlock(&state_lock);
6730         return count;
6731 }
6732
6733 static u64
6734 nfsd_print_client_delegations(struct nfs4_client *clp)
6735 {
6736         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6737
6738         nfsd_print_count(clp, count, "delegations");
6739         return count;
6740 }
6741
6742 u64
6743 nfsd_inject_print_delegations(void)
6744 {
6745         struct nfs4_client *clp;
6746         u64 count = 0;
6747         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6748                                                 nfsd_net_id);
6749
6750         if (!nfsd_netns_ready(nn))
6751                 return 0;
6752
6753         spin_lock(&nn->client_lock);
6754         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6755                 count += nfsd_print_client_delegations(clp);
6756         spin_unlock(&nn->client_lock);
6757
6758         return count;
6759 }
6760
6761 static void
6762 nfsd_forget_delegations(struct list_head *reaplist)
6763 {
6764         struct nfs4_client *clp;
6765         struct nfs4_delegation *dp, *next;
6766
6767         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6768                 list_del_init(&dp->dl_recall_lru);
6769                 clp = dp->dl_stid.sc_client;
6770                 revoke_delegation(dp);
6771                 put_client(clp);
6772         }
6773 }
6774
6775 u64
6776 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6777                                       size_t addr_size)
6778 {
6779         u64 count = 0;
6780         struct nfs4_client *clp;
6781         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6782                                                 nfsd_net_id);
6783         LIST_HEAD(reaplist);
6784
6785         if (!nfsd_netns_ready(nn))
6786                 return count;
6787
6788         spin_lock(&nn->client_lock);
6789         clp = nfsd_find_client(addr, addr_size);
6790         if (clp)
6791                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6792         spin_unlock(&nn->client_lock);
6793
6794         nfsd_forget_delegations(&reaplist);
6795         return count;
6796 }
6797
6798 u64
6799 nfsd_inject_forget_delegations(u64 max)
6800 {
6801         u64 count = 0;
6802         struct nfs4_client *clp;
6803         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6804                                                 nfsd_net_id);
6805         LIST_HEAD(reaplist);
6806
6807         if (!nfsd_netns_ready(nn))
6808                 return count;
6809
6810         spin_lock(&nn->client_lock);
6811         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6812                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6813                 if (max != 0 && count >= max)
6814                         break;
6815         }
6816         spin_unlock(&nn->client_lock);
6817         nfsd_forget_delegations(&reaplist);
6818         return count;
6819 }
6820
6821 static void
6822 nfsd_recall_delegations(struct list_head *reaplist)
6823 {
6824         struct nfs4_client *clp;
6825         struct nfs4_delegation *dp, *next;
6826
6827         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6828                 list_del_init(&dp->dl_recall_lru);
6829                 clp = dp->dl_stid.sc_client;
6830                 /*
6831                  * We skipped all entries that had a zero dl_time before,
6832                  * so we can now reset the dl_time back to 0. If a delegation
6833                  * break comes in now, then it won't make any difference since
6834                  * we're recalling it either way.
6835                  */
6836                 spin_lock(&state_lock);
6837                 dp->dl_time = 0;
6838                 spin_unlock(&state_lock);
6839                 nfsd_break_one_deleg(dp);
6840                 put_client(clp);
6841         }
6842 }
6843
6844 u64
6845 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6846                                       size_t addr_size)
6847 {
6848         u64 count = 0;
6849         struct nfs4_client *clp;
6850         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6851                                                 nfsd_net_id);
6852         LIST_HEAD(reaplist);
6853
6854         if (!nfsd_netns_ready(nn))
6855                 return count;
6856
6857         spin_lock(&nn->client_lock);
6858         clp = nfsd_find_client(addr, addr_size);
6859         if (clp)
6860                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6861         spin_unlock(&nn->client_lock);
6862
6863         nfsd_recall_delegations(&reaplist);
6864         return count;
6865 }
6866
6867 u64
6868 nfsd_inject_recall_delegations(u64 max)
6869 {
6870         u64 count = 0;
6871         struct nfs4_client *clp, *next;
6872         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6873                                                 nfsd_net_id);
6874         LIST_HEAD(reaplist);
6875
6876         if (!nfsd_netns_ready(nn))
6877                 return count;
6878
6879         spin_lock(&nn->client_lock);
6880         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6881                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6882                 if (max != 0 && ++count >= max)
6883                         break;
6884         }
6885         spin_unlock(&nn->client_lock);
6886         nfsd_recall_delegations(&reaplist);
6887         return count;
6888 }
6889 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6890
6891 /*
6892  * Since the lifetime of a delegation isn't limited to that of an open, a
6893  * client may quite reasonably hang on to a delegation as long as it has
6894  * the inode cached.  This becomes an obvious problem the first time a
6895  * client's inode cache approaches the size of the server's total memory.
6896  *
6897  * For now we avoid this problem by imposing a hard limit on the number
6898  * of delegations, which varies according to the server's memory size.
6899  */
6900 static void
6901 set_max_delegations(void)
6902 {
6903         /*
6904          * Allow at most 4 delegations per megabyte of RAM.  Quick
6905          * estimates suggest that in the worst case (where every delegation
6906          * is for a different inode), a delegation could take about 1.5K,
6907          * giving a worst case usage of about 6% of memory.
6908          */
6909         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6910 }
6911
6912 static int nfs4_state_create_net(struct net *net)
6913 {
6914         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6915         int i;
6916
6917         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6918                         CLIENT_HASH_SIZE, GFP_KERNEL);
6919         if (!nn->conf_id_hashtbl)
6920                 goto err;
6921         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6922                         CLIENT_HASH_SIZE, GFP_KERNEL);
6923         if (!nn->unconf_id_hashtbl)
6924                 goto err_unconf_id;
6925         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6926                         SESSION_HASH_SIZE, GFP_KERNEL);
6927         if (!nn->sessionid_hashtbl)
6928                 goto err_sessionid;
6929
6930         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6931                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6932                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6933         }
6934         for (i = 0; i < SESSION_HASH_SIZE; i++)
6935                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6936         nn->conf_name_tree = RB_ROOT;
6937         nn->unconf_name_tree = RB_ROOT;
6938         INIT_LIST_HEAD(&nn->client_lru);
6939         INIT_LIST_HEAD(&nn->close_lru);
6940         INIT_LIST_HEAD(&nn->del_recall_lru);
6941         spin_lock_init(&nn->client_lock);
6942
6943         spin_lock_init(&nn->blocked_locks_lock);
6944         INIT_LIST_HEAD(&nn->blocked_locks_lru);
6945
6946         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6947         get_net(net);
6948
6949         return 0;
6950
6951 err_sessionid:
6952         kfree(nn->unconf_id_hashtbl);
6953 err_unconf_id:
6954         kfree(nn->conf_id_hashtbl);
6955 err:
6956         return -ENOMEM;
6957 }
6958
6959 static void
6960 nfs4_state_destroy_net(struct net *net)
6961 {
6962         int i;
6963         struct nfs4_client *clp = NULL;
6964         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6965
6966         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6967                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6968                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6969                         destroy_client(clp);
6970                 }
6971         }
6972
6973         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6974                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6975                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6976                         destroy_client(clp);
6977                 }
6978         }
6979
6980         kfree(nn->sessionid_hashtbl);
6981         kfree(nn->unconf_id_hashtbl);
6982         kfree(nn->conf_id_hashtbl);
6983         put_net(net);
6984 }
6985
6986 int
6987 nfs4_state_start_net(struct net *net)
6988 {
6989         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6990         int ret;
6991
6992         ret = nfs4_state_create_net(net);
6993         if (ret)
6994                 return ret;
6995         nn->boot_time = get_seconds();
6996         nn->grace_ended = false;
6997         nn->nfsd4_manager.block_opens = true;
6998         locks_start_grace(net, &nn->nfsd4_manager);
6999         nfsd4_client_tracking_init(net);
7000         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
7001                nn->nfsd4_grace, net);
7002         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7003         return 0;
7004 }
7005
7006 /* initialization to perform when the nfsd service is started: */
7007
7008 int
7009 nfs4_state_start(void)
7010 {
7011         int ret;
7012
7013         ret = set_callback_cred();
7014         if (ret)
7015                 return ret;
7016
7017         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7018         if (laundry_wq == NULL) {
7019                 ret = -ENOMEM;
7020                 goto out_cleanup_cred;
7021         }
7022         ret = nfsd4_create_callback_queue();
7023         if (ret)
7024                 goto out_free_laundry;
7025
7026         set_max_delegations();
7027         return 0;
7028
7029 out_free_laundry:
7030         destroy_workqueue(laundry_wq);
7031 out_cleanup_cred:
7032         cleanup_callback_cred();
7033         return ret;
7034 }
7035
7036 void
7037 nfs4_state_shutdown_net(struct net *net)
7038 {
7039         struct nfs4_delegation *dp = NULL;
7040         struct list_head *pos, *next, reaplist;
7041         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7042         struct nfsd4_blocked_lock *nbl;
7043
7044         cancel_delayed_work_sync(&nn->laundromat_work);
7045         locks_end_grace(&nn->nfsd4_manager);
7046
7047         INIT_LIST_HEAD(&reaplist);
7048         spin_lock(&state_lock);
7049         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7050                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7051                 WARN_ON(!unhash_delegation_locked(dp));
7052                 list_add(&dp->dl_recall_lru, &reaplist);
7053         }
7054         spin_unlock(&state_lock);
7055         list_for_each_safe(pos, next, &reaplist) {
7056                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7057                 list_del_init(&dp->dl_recall_lru);
7058                 put_clnt_odstate(dp->dl_clnt_odstate);
7059                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7060                 nfs4_put_stid(&dp->dl_stid);
7061         }
7062
7063         BUG_ON(!list_empty(&reaplist));
7064         spin_lock(&nn->blocked_locks_lock);
7065         while (!list_empty(&nn->blocked_locks_lru)) {
7066                 nbl = list_first_entry(&nn->blocked_locks_lru,
7067                                         struct nfsd4_blocked_lock, nbl_lru);
7068                 list_move(&nbl->nbl_lru, &reaplist);
7069                 list_del_init(&nbl->nbl_list);
7070         }
7071         spin_unlock(&nn->blocked_locks_lock);
7072
7073         while (!list_empty(&reaplist)) {
7074                 nbl = list_first_entry(&nn->blocked_locks_lru,
7075                                         struct nfsd4_blocked_lock, nbl_lru);
7076                 list_del_init(&nbl->nbl_lru);
7077                 posix_unblock_lock(&nbl->nbl_lock);
7078                 free_blocked_lock(nbl);
7079         }
7080
7081         nfsd4_client_tracking_exit(net);
7082         nfs4_state_destroy_net(net);
7083 }
7084
7085 void
7086 nfs4_state_shutdown(void)
7087 {
7088         destroy_workqueue(laundry_wq);
7089         nfsd4_destroy_callback_queue();
7090         cleanup_callback_cred();
7091 }
7092
7093 static void
7094 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7095 {
7096         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7097                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7098 }
7099
7100 static void
7101 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7102 {
7103         if (cstate->minorversion) {
7104                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7105                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7106         }
7107 }
7108
7109 void
7110 clear_current_stateid(struct nfsd4_compound_state *cstate)
7111 {
7112         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7113 }
7114
7115 /*
7116  * functions to set current state id
7117  */
7118 void
7119 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7120 {
7121         put_stateid(cstate, &odp->od_stateid);
7122 }
7123
7124 void
7125 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
7126 {
7127         put_stateid(cstate, &open->op_stateid);
7128 }
7129
7130 void
7131 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7132 {
7133         put_stateid(cstate, &close->cl_stateid);
7134 }
7135
7136 void
7137 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
7138 {
7139         put_stateid(cstate, &lock->lk_resp_stateid);
7140 }
7141
7142 /*
7143  * functions to consume current state id
7144  */
7145
7146 void
7147 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7148 {
7149         get_stateid(cstate, &odp->od_stateid);
7150 }
7151
7152 void
7153 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
7154 {
7155         get_stateid(cstate, &drp->dr_stateid);
7156 }
7157
7158 void
7159 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
7160 {
7161         get_stateid(cstate, &fsp->fr_stateid);
7162 }
7163
7164 void
7165 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
7166 {
7167         get_stateid(cstate, &setattr->sa_stateid);
7168 }
7169
7170 void
7171 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7172 {
7173         get_stateid(cstate, &close->cl_stateid);
7174 }
7175
7176 void
7177 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
7178 {
7179         get_stateid(cstate, &locku->lu_stateid);
7180 }
7181
7182 void
7183 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
7184 {
7185         get_stateid(cstate, &read->rd_stateid);
7186 }
7187
7188 void
7189 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
7190 {
7191         get_stateid(cstate, &write->wr_stateid);
7192 }