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 static const stateid_t close_stateid = {
  67         .si_generation = 0xffffffffU,
  68 };
  69
  70 static u64 current_sessionid = 1;
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  75 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  76
  77 /* forward declarations */
  78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  80
  81 /* Locking: */
  82
  83 /*
  84  * Currently used for the del_recall_lru and file hash table.  In an
  85  * effort to decrease the scope of the client_mutex, this spinlock may
  86  * eventually cover more:
  87  */
  88 static DEFINE_SPINLOCK(state_lock);
  89
  90 enum nfsd4_st_mutex_lock_subclass {
  91         OPEN_STATEID_MUTEX = 0,
  92         LOCK_STATEID_MUTEX = 1,
  93 };
  94
  95 /*
  96  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  97  * the refcount on the open stateid to drop.
  98  */
  99 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 100
 101 static struct kmem_cache *client_slab;
 102 static struct kmem_cache *openowner_slab;
 103 static struct kmem_cache *lockowner_slab;
 104 static struct kmem_cache *file_slab;
 105 static struct kmem_cache *stateid_slab;
 106 static struct kmem_cache *deleg_slab;
 107 static struct kmem_cache *odstate_slab;
 108
 109 static void free_session(struct nfsd4_session *);
 110
 111 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 112 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 113
 114 static bool is_session_dead(struct nfsd4_session *ses)
 115 {
 116         return ses->se_flags & NFS4_SESSION_DEAD;
 117 }
 118
 119 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 120 {
 121         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 122                 return nfserr_jukebox;
 123         ses->se_flags |= NFS4_SESSION_DEAD;
 124         return nfs_ok;
 125 }
 126
 127 static bool is_client_expired(struct nfs4_client *clp)
 128 {
 129         return clp->cl_time == 0;
 130 }
 131
 132 static __be32 get_client_locked(struct nfs4_client *clp)
 133 {
 134         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 135
 136         lockdep_assert_held(&nn->client_lock);
 137
 138         if (is_client_expired(clp))
 139                 return nfserr_expired;
 140         atomic_inc(&clp->cl_refcount);
 141         return nfs_ok;
 142 }
 143
 144 /* must be called under the client_lock */
 145 static inline void
 146 renew_client_locked(struct nfs4_client *clp)
 147 {
 148         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 149
 150         if (is_client_expired(clp)) {
 151                 WARN_ON(1);
 152                 printk("%s: client (clientid %08x/%08x) already expired\n",
 153                         __func__,
 154                         clp->cl_clientid.cl_boot,
 155                         clp->cl_clientid.cl_id);
 156                 return;
 157         }
 158
 159         dprintk("renewing client (clientid %08x/%08x)\n",
 160                         clp->cl_clientid.cl_boot,
 161                         clp->cl_clientid.cl_id);
 162         list_move_tail(&clp->cl_lru, &nn->client_lru);
 163         clp->cl_time = get_seconds();
 164 }
 165
 166 static void put_client_renew_locked(struct nfs4_client *clp)
 167 {
 168         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 169
 170         lockdep_assert_held(&nn->client_lock);
 171
 172         if (!atomic_dec_and_test(&clp->cl_refcount))
 173                 return;
 174         if (!is_client_expired(clp))
 175                 renew_client_locked(clp);
 176 }
 177
 178 static void put_client_renew(struct nfs4_client *clp)
 179 {
 180         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 181
 182         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 183                 return;
 184         if (!is_client_expired(clp))
 185                 renew_client_locked(clp);
 186         spin_unlock(&nn->client_lock);
 187 }
 188
 189 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 190 {
 191         __be32 status;
 192
 193         if (is_session_dead(ses))
 194                 return nfserr_badsession;
 195         status = get_client_locked(ses->se_client);
 196         if (status)
 197                 return status;
 198         atomic_inc(&ses->se_ref);
 199         return nfs_ok;
 200 }
 201
 202 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 203 {
 204         struct nfs4_client *clp = ses->se_client;
 205         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 206
 207         lockdep_assert_held(&nn->client_lock);
 208
 209         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 210                 free_session(ses);
 211         put_client_renew_locked(clp);
 212 }
 213
 214 static void nfsd4_put_session(struct nfsd4_session *ses)
 215 {
 216         struct nfs4_client *clp = ses->se_client;
 217         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 218
 219         spin_lock(&nn->client_lock);
 220         nfsd4_put_session_locked(ses);
 221         spin_unlock(&nn->client_lock);
 222 }
 223
 224 static struct nfsd4_blocked_lock *
 225 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 226                         struct nfsd_net *nn)
 227 {
 228         struct nfsd4_blocked_lock *cur, *found = NULL;
 229
 230         spin_lock(&nn->blocked_locks_lock);
 231         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 232                 if (fh_match(fh, &cur->nbl_fh)) {
 233                         list_del_init(&cur->nbl_list);
 234                         list_del_init(&cur->nbl_lru);
 235                         found = cur;
 236                         break;
 237                 }
 238         }
 239         spin_unlock(&nn->blocked_locks_lock);
 240         if (found)
 241                 posix_unblock_lock(&found->nbl_lock);
 242         return found;
 243 }
 244
 245 static struct nfsd4_blocked_lock *
 246 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 247                         struct nfsd_net *nn)
 248 {
 249         struct nfsd4_blocked_lock *nbl;
 250
 251         nbl = find_blocked_lock(lo, fh, nn);
 252         if (!nbl) {
 253                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 254                 if (nbl) {
 255                         fh_copy_shallow(&nbl->nbl_fh, fh);
 256                         locks_init_lock(&nbl->nbl_lock);
 257                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 258                                         &nfsd4_cb_notify_lock_ops,
 259                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 260                 }
 261         }
 262         return nbl;
 263 }
 264
 265 static void
 266 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 267 {
 268         locks_release_private(&nbl->nbl_lock);
 269         kfree(nbl);
 270 }
 271
 272 static void
 273 remove_blocked_locks(struct nfs4_lockowner *lo)
 274 {
 275         struct nfs4_client *clp = lo->lo_owner.so_client;
 276         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 277         struct nfsd4_blocked_lock *nbl;
 278         LIST_HEAD(reaplist);
 279
 280         /* Dequeue all blocked locks */
 281         spin_lock(&nn->blocked_locks_lock);
 282         while (!list_empty(&lo->lo_blocked)) {
 283                 nbl = list_first_entry(&lo->lo_blocked,
 284                                         struct nfsd4_blocked_lock,
 285                                         nbl_list);
 286                 list_del_init(&nbl->nbl_list);
 287                 list_move(&nbl->nbl_lru, &reaplist);
 288         }
 289         spin_unlock(&nn->blocked_locks_lock);
 290
 291         /* Now free them */
 292         while (!list_empty(&reaplist)) {
 293                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 294                                         nbl_lru);
 295                 list_del_init(&nbl->nbl_lru);
 296                 posix_unblock_lock(&nbl->nbl_lock);
 297                 free_blocked_lock(nbl);
 298         }
 299 }
 300
 301 static int
 302 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 303 {
 304         /*
 305          * Since this is just an optimization, we don't try very hard if it
 306          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 307          * just quit trying on anything else.
 308          */
 309         switch (task->tk_status) {
 310         case -NFS4ERR_DELAY:
 311                 rpc_delay(task, 1 * HZ);
 312                 return 0;
 313         default:
 314                 return 1;
 315         }
 316 }
 317
 318 static void
 319 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 320 {
 321         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 322                                                 struct nfsd4_blocked_lock, nbl_cb);
 323
 324         free_blocked_lock(nbl);
 325 }
 326
 327 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 328         .done           = nfsd4_cb_notify_lock_done,
 329         .release        = nfsd4_cb_notify_lock_release,
 330 };
 331
 332 static inline struct nfs4_stateowner *
 333 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 334 {
 335         atomic_inc(&sop->so_count);
 336         return sop;
 337 }
 338
 339 static int
 340 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 341 {
 342         return (sop->so_owner.len == owner->len) &&
 343                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 344 }
 345
 346 static struct nfs4_openowner *
 347 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 348                         struct nfs4_client *clp)
 349 {
 350         struct nfs4_stateowner *so;
 351
 352         lockdep_assert_held(&clp->cl_lock);
 353
 354         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 355                             so_strhash) {
 356                 if (!so->so_is_open_owner)
 357                         continue;
 358                 if (same_owner_str(so, &open->op_owner))
 359                         return openowner(nfs4_get_stateowner(so));
 360         }
 361         return NULL;
 362 }
 363
 364 static struct nfs4_openowner *
 365 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 366                         struct nfs4_client *clp)
 367 {
 368         struct nfs4_openowner *oo;
 369
 370         spin_lock(&clp->cl_lock);
 371         oo = find_openstateowner_str_locked(hashval, open, clp);
 372         spin_unlock(&clp->cl_lock);
 373         return oo;
 374 }
 375
 376 static inline u32
 377 opaque_hashval(const void *ptr, int nbytes)
 378 {
 379         unsigned char *cptr = (unsigned char *) ptr;
 380
 381         u32 x = 0;
 382         while (nbytes--) {
 383                 x *= 37;
 384                 x += *cptr++;
 385         }
 386         return x;
 387 }
 388
 389 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 390 {
 391         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 392
 393         kmem_cache_free(file_slab, fp);
 394 }
 395
 396 void
 397 put_nfs4_file(struct nfs4_file *fi)
 398 {
 399         might_lock(&state_lock);
 400
 401         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 402                 hlist_del_rcu(&fi->fi_hash);
 403                 spin_unlock(&state_lock);
 404                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 405                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 406                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 407         }
 408 }
 409
 410 static struct file *
 411 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 412 {
 413         if (f->fi_fds[oflag])
 414                 return get_file(f->fi_fds[oflag]);
 415         return NULL;
 416 }
 417
 418 static struct file *
 419 find_writeable_file_locked(struct nfs4_file *f)
 420 {
 421         struct file *ret;
 422
 423         lockdep_assert_held(&f->fi_lock);
 424
 425         ret = __nfs4_get_fd(f, O_WRONLY);
 426         if (!ret)
 427                 ret = __nfs4_get_fd(f, O_RDWR);
 428         return ret;
 429 }
 430
 431 static struct file *
 432 find_writeable_file(struct nfs4_file *f)
 433 {
 434         struct file *ret;
 435
 436         spin_lock(&f->fi_lock);
 437         ret = find_writeable_file_locked(f);
 438         spin_unlock(&f->fi_lock);
 439
 440         return ret;
 441 }
 442
 443 static struct file *find_readable_file_locked(struct nfs4_file *f)
 444 {
 445         struct file *ret;
 446
 447         lockdep_assert_held(&f->fi_lock);
 448
 449         ret = __nfs4_get_fd(f, O_RDONLY);
 450         if (!ret)
 451                 ret = __nfs4_get_fd(f, O_RDWR);
 452         return ret;
 453 }
 454
 455 static struct file *
 456 find_readable_file(struct nfs4_file *f)
 457 {
 458         struct file *ret;
 459
 460         spin_lock(&f->fi_lock);
 461         ret = find_readable_file_locked(f);
 462         spin_unlock(&f->fi_lock);
 463
 464         return ret;
 465 }
 466
 467 struct file *
 468 find_any_file(struct nfs4_file *f)
 469 {
 470         struct file *ret;
 471
 472         spin_lock(&f->fi_lock);
 473         ret = __nfs4_get_fd(f, O_RDWR);
 474         if (!ret) {
 475                 ret = __nfs4_get_fd(f, O_WRONLY);
 476                 if (!ret)
 477                         ret = __nfs4_get_fd(f, O_RDONLY);
 478         }
 479         spin_unlock(&f->fi_lock);
 480         return ret;
 481 }
 482
 483 static atomic_long_t num_delegations;
 484 unsigned long max_delegations;
 485
 486 /*
 487  * Open owner state (share locks)
 488  */
 489
 490 /* hash tables for lock and open owners */
 491 #define OWNER_HASH_BITS              8
 492 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 493 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 494
 495 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 496 {
 497         unsigned int ret;
 498
 499         ret = opaque_hashval(ownername->data, ownername->len);
 500         return ret & OWNER_HASH_MASK;
 501 }
 502
 503 /* hash table for nfs4_file */
 504 #define FILE_HASH_BITS                   8
 505 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 506
 507 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 508 {
 509         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 510 }
 511
 512 static unsigned int file_hashval(struct knfsd_fh *fh)
 513 {
 514         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 515 }
 516
 517 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 518
 519 static void
 520 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 521 {
 522         lockdep_assert_held(&fp->fi_lock);
 523
 524         if (access & NFS4_SHARE_ACCESS_WRITE)
 525                 atomic_inc(&fp->fi_access[O_WRONLY]);
 526         if (access & NFS4_SHARE_ACCESS_READ)
 527                 atomic_inc(&fp->fi_access[O_RDONLY]);
 528 }
 529
 530 static __be32
 531 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 532 {
 533         lockdep_assert_held(&fp->fi_lock);
 534
 535         /* Does this access mode make sense? */
 536         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 537                 return nfserr_inval;
 538
 539         /* Does it conflict with a deny mode already set? */
 540         if ((access & fp->fi_share_deny) != 0)
 541                 return nfserr_share_denied;
 542
 543         __nfs4_file_get_access(fp, access);
 544         return nfs_ok;
 545 }
 546
 547 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 548 {
 549         /* Common case is that there is no deny mode. */
 550         if (deny) {
 551                 /* Does this deny mode make sense? */
 552                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 553                         return nfserr_inval;
 554
 555                 if ((deny & NFS4_SHARE_DENY_READ) &&
 556                     atomic_read(&fp->fi_access[O_RDONLY]))
 557                         return nfserr_share_denied;
 558
 559                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 560                     atomic_read(&fp->fi_access[O_WRONLY]))
 561                         return nfserr_share_denied;
 562         }
 563         return nfs_ok;
 564 }
 565
 566 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 567 {
 568         might_lock(&fp->fi_lock);
 569
 570         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 571                 struct file *f1 = NULL;
 572                 struct file *f2 = NULL;
 573
 574                 swap(f1, fp->fi_fds[oflag]);
 575                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 576                         swap(f2, fp->fi_fds[O_RDWR]);
 577                 spin_unlock(&fp->fi_lock);
 578                 if (f1)
 579                         fput(f1);
 580                 if (f2)
 581                         fput(f2);
 582         }
 583 }
 584
 585 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 586 {
 587         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 588
 589         if (access & NFS4_SHARE_ACCESS_WRITE)
 590                 __nfs4_file_put_access(fp, O_WRONLY);
 591         if (access & NFS4_SHARE_ACCESS_READ)
 592                 __nfs4_file_put_access(fp, O_RDONLY);
 593 }
 594
 595 /*
 596  * Allocate a new open/delegation state counter. This is needed for
 597  * pNFS for proper return on close semantics.
 598  *
 599  * Note that we only allocate it for pNFS-enabled exports, otherwise
 600  * all pointers to struct nfs4_clnt_odstate are always NULL.
 601  */
 602 static struct nfs4_clnt_odstate *
 603 alloc_clnt_odstate(struct nfs4_client *clp)
 604 {
 605         struct nfs4_clnt_odstate *co;
 606
 607         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 608         if (co) {
 609                 co->co_client = clp;
 610                 refcount_set(&co->co_odcount, 1);
 611         }
 612         return co;
 613 }
 614
 615 static void
 616 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 617 {
 618         struct nfs4_file *fp = co->co_file;
 619
 620         lockdep_assert_held(&fp->fi_lock);
 621         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 622 }
 623
 624 static inline void
 625 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 626 {
 627         if (co)
 628                 refcount_inc(&co->co_odcount);
 629 }
 630
 631 static void
 632 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 633 {
 634         struct nfs4_file *fp;
 635
 636         if (!co)
 637                 return;
 638
 639         fp = co->co_file;
 640         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 641                 list_del(&co->co_perfile);
 642                 spin_unlock(&fp->fi_lock);
 643
 644                 nfsd4_return_all_file_layouts(co->co_client, fp);
 645                 kmem_cache_free(odstate_slab, co);
 646         }
 647 }
 648
 649 static struct nfs4_clnt_odstate *
 650 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 651 {
 652         struct nfs4_clnt_odstate *co;
 653         struct nfs4_client *cl;
 654
 655         if (!new)
 656                 return NULL;
 657
 658         cl = new->co_client;
 659
 660         spin_lock(&fp->fi_lock);
 661         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 662                 if (co->co_client == cl) {
 663                         get_clnt_odstate(co);
 664                         goto out;
 665                 }
 666         }
 667         co = new;
 668         co->co_file = fp;
 669         hash_clnt_odstate_locked(new);
 670 out:
 671         spin_unlock(&fp->fi_lock);
 672         return co;
 673 }
 674
 675 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 676                                   void (*sc_free)(struct nfs4_stid *))
 677 {
 678         struct nfs4_stid *stid;
 679         int new_id;
 680
 681         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 682         if (!stid)
 683                 return NULL;
 684
 685         idr_preload(GFP_KERNEL);
 686         spin_lock(&cl->cl_lock);
 687         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 688         spin_unlock(&cl->cl_lock);
 689         idr_preload_end();
 690         if (new_id < 0)
 691                 goto out_free;
 692
 693         stid->sc_free = sc_free;
 694         stid->sc_client = cl;
 695         stid->sc_stateid.si_opaque.so_id = new_id;
 696         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 697         /* Will be incremented before return to client: */
 698         refcount_set(&stid->sc_count, 1);
 699         spin_lock_init(&stid->sc_lock);
 700
 701         /*
 702          * It shouldn't be a problem to reuse an opaque stateid value.
 703          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 704          * example, a stray write retransmission could be accepted by
 705          * the server when it should have been rejected.  Therefore,
 706          * adopt a trick from the sctp code to attempt to maximize the
 707          * amount of time until an id is reused, by ensuring they always
 708          * "increase" (mod INT_MAX):
 709          */
 710         return stid;
 711 out_free:
 712         kmem_cache_free(slab, stid);
 713         return NULL;
 714 }
 715
 716 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 717 {
 718         struct nfs4_stid *stid;
 719
 720         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 721         if (!stid)
 722                 return NULL;
 723
 724         return openlockstateid(stid);
 725 }
 726
 727 static void nfs4_free_deleg(struct nfs4_stid *stid)
 728 {
 729         kmem_cache_free(deleg_slab, stid);
 730         atomic_long_dec(&num_delegations);
 731 }
 732
 733 /*
 734  * When we recall a delegation, we should be careful not to hand it
 735  * out again straight away.
 736  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 737  * in which the filehandles of recalled delegations are "stored".
 738  * If a filehandle appear in either filter, a delegation is blocked.
 739  * When a delegation is recalled, the filehandle is stored in the "new"
 740  * filter.
 741  * Every 30 seconds we swap the filters and clear the "new" one,
 742  * unless both are empty of course.
 743  *
 744  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 745  * low 3 bytes as hash-table indices.
 746  *
 747  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 748  * is used to manage concurrent access.  Testing does not need the lock
 749  * except when swapping the two filters.
 750  */
 751 static DEFINE_SPINLOCK(blocked_delegations_lock);
 752 static struct bloom_pair {
 753         int     entries, old_entries;
 754         time_t  swap_time;
 755         int     new; /* index into 'set' */
 756         DECLARE_BITMAP(set[2], 256);
 757 } blocked_delegations;
 758
 759 static int delegation_blocked(struct knfsd_fh *fh)
 760 {
 761         u32 hash;
 762         struct bloom_pair *bd = &blocked_delegations;
 763
 764         if (bd->entries == 0)
 765                 return 0;
 766         if (seconds_since_boot() - bd->swap_time > 30) {
 767                 spin_lock(&blocked_delegations_lock);
 768                 if (seconds_since_boot() - bd->swap_time > 30) {
 769                         bd->entries -= bd->old_entries;
 770                         bd->old_entries = bd->entries;
 771                         memset(bd->set[bd->new], 0,
 772                                sizeof(bd->set[0]));
 773                         bd->new = 1-bd->new;
 774                         bd->swap_time = seconds_since_boot();
 775                 }
 776                 spin_unlock(&blocked_delegations_lock);
 777         }
 778         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 779         if (test_bit(hash&255, bd->set[0]) &&
 780             test_bit((hash>>8)&255, bd->set[0]) &&
 781             test_bit((hash>>16)&255, bd->set[0]))
 782                 return 1;
 783
 784         if (test_bit(hash&255, bd->set[1]) &&
 785             test_bit((hash>>8)&255, bd->set[1]) &&
 786             test_bit((hash>>16)&255, bd->set[1]))
 787                 return 1;
 788
 789         return 0;
 790 }
 791
 792 static void block_delegations(struct knfsd_fh *fh)
 793 {
 794         u32 hash;
 795         struct bloom_pair *bd = &blocked_delegations;
 796
 797         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 798
 799         spin_lock(&blocked_delegations_lock);
 800         __set_bit(hash&255, bd->set[bd->new]);
 801         __set_bit((hash>>8)&255, bd->set[bd->new]);
 802         __set_bit((hash>>16)&255, bd->set[bd->new]);
 803         if (bd->entries == 0)
 804                 bd->swap_time = seconds_since_boot();
 805         bd->entries += 1;
 806         spin_unlock(&blocked_delegations_lock);
 807 }
 808
 809 static struct nfs4_delegation *
 810 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 811                  struct svc_fh *current_fh,
 812                  struct nfs4_clnt_odstate *odstate)
 813 {
 814         struct nfs4_delegation *dp;
 815         long n;
 816
 817         dprintk("NFSD alloc_init_deleg\n");
 818         n = atomic_long_inc_return(&num_delegations);
 819         if (n < 0 || n > max_delegations)
 820                 goto out_dec;
 821         if (delegation_blocked(&current_fh->fh_handle))
 822                 goto out_dec;
 823         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 824         if (dp == NULL)
 825                 goto out_dec;
 826
 827         /*
 828          * delegation seqid's are never incremented.  The 4.1 special
 829          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 830          * 0 anyway just for consistency and use 1:
 831          */
 832         dp->dl_stid.sc_stateid.si_generation = 1;
 833         INIT_LIST_HEAD(&dp->dl_perfile);
 834         INIT_LIST_HEAD(&dp->dl_perclnt);
 835         INIT_LIST_HEAD(&dp->dl_recall_lru);
 836         dp->dl_clnt_odstate = odstate;
 837         get_clnt_odstate(odstate);
 838         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 839         dp->dl_retries = 1;
 840         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 841                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 842         get_nfs4_file(fp);
 843         dp->dl_stid.sc_file = fp;
 844         return dp;
 845 out_dec:
 846         atomic_long_dec(&num_delegations);
 847         return NULL;
 848 }
 849
 850 void
 851 nfs4_put_stid(struct nfs4_stid *s)
 852 {
 853         struct nfs4_file *fp = s->sc_file;
 854         struct nfs4_client *clp = s->sc_client;
 855
 856         might_lock(&clp->cl_lock);
 857
 858         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 859                 wake_up_all(&close_wq);
 860                 return;
 861         }
 862         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 863         spin_unlock(&clp->cl_lock);
 864         s->sc_free(s);
 865         if (fp)
 866                 put_nfs4_file(fp);
 867 }
 868
 869 void
 870 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 871 {
 872         stateid_t *src = &stid->sc_stateid;
 873
 874         spin_lock(&stid->sc_lock);
 875         if (unlikely(++src->si_generation == 0))
 876                 src->si_generation = 1;
 877         memcpy(dst, src, sizeof(*dst));
 878         spin_unlock(&stid->sc_lock);
 879 }
 880
 881 static void put_deleg_file(struct nfs4_file *fp)
 882 {
 883         struct file *filp = NULL;
 884
 885         spin_lock(&fp->fi_lock);
 886         if (--fp->fi_delegees == 0)
 887                 swap(filp, fp->fi_deleg_file);
 888         spin_unlock(&fp->fi_lock);
 889
 890         if (filp)
 891                 fput(filp);
 892 }
 893
 894 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 895 {
 896         struct nfs4_file *fp = dp->dl_stid.sc_file;
 897         struct file *filp = fp->fi_deleg_file;
 898
 899         WARN_ON_ONCE(!fp->fi_delegees);
 900
 901         vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
 902         put_deleg_file(fp);
 903 }
 904
 905 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 906 {
 907         put_clnt_odstate(dp->dl_clnt_odstate);
 908         nfs4_unlock_deleg_lease(dp);
 909         nfs4_put_stid(&dp->dl_stid);
 910 }
 911
 912 void nfs4_unhash_stid(struct nfs4_stid *s)
 913 {
 914         s->sc_type = 0;
 915 }
 916
 917 /**
 918  * nfs4_delegation_exists - Discover if this delegation already exists
 919  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 920  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 921  *
 922  * Return:
 923  *      On success: true iff an existing delegation is found
 924  */
 925
 926 static bool
 927 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 928 {
 929         struct nfs4_delegation *searchdp = NULL;
 930         struct nfs4_client *searchclp = NULL;
 931
 932         lockdep_assert_held(&state_lock);
 933         lockdep_assert_held(&fp->fi_lock);
 934
 935         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 936                 searchclp = searchdp->dl_stid.sc_client;
 937                 if (clp == searchclp) {
 938                         return true;
 939                 }
 940         }
 941         return false;
 942 }
 943
 944 /**
 945  * hash_delegation_locked - Add a delegation to the appropriate lists
 946  * @dp:     a pointer to the nfs4_delegation we are adding.
 947  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 948  *
 949  * Return:
 950  *      On success: NULL if the delegation was successfully hashed.
 951  *
 952  *      On error: -EAGAIN if one was previously granted to this
 953  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 954  *
 955  */
 956
 957 static int
 958 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 959 {
 960         struct nfs4_client *clp = dp->dl_stid.sc_client;
 961
 962         lockdep_assert_held(&state_lock);
 963         lockdep_assert_held(&fp->fi_lock);
 964
 965         if (nfs4_delegation_exists(clp, fp))
 966                 return -EAGAIN;
 967         refcount_inc(&dp->dl_stid.sc_count);
 968         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 969         list_add(&dp->dl_perfile, &fp->fi_delegations);
 970         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 971         return 0;
 972 }
 973
 974 static bool
 975 unhash_delegation_locked(struct nfs4_delegation *dp)
 976 {
 977         struct nfs4_file *fp = dp->dl_stid.sc_file;
 978
 979         lockdep_assert_held(&state_lock);
 980
 981         if (list_empty(&dp->dl_perfile))
 982                 return false;
 983
 984         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 985         /* Ensure that deleg break won't try to requeue it */
 986         ++dp->dl_time;
 987         spin_lock(&fp->fi_lock);
 988         list_del_init(&dp->dl_perclnt);
 989         list_del_init(&dp->dl_recall_lru);
 990         list_del_init(&dp->dl_perfile);
 991         spin_unlock(&fp->fi_lock);
 992         return true;
 993 }
 994
 995 static void destroy_delegation(struct nfs4_delegation *dp)
 996 {
 997         bool unhashed;
 998
 999         spin_lock(&state_lock);
1000         unhashed = unhash_delegation_locked(dp);
1001         spin_unlock(&state_lock);
1002         if (unhashed)
1003                 destroy_unhashed_deleg(dp);
1004 }
1005
1006 static void revoke_delegation(struct nfs4_delegation *dp)
1007 {
1008         struct nfs4_client *clp = dp->dl_stid.sc_client;
1009
1010         WARN_ON(!list_empty(&dp->dl_recall_lru));
1011
1012         if (clp->cl_minorversion) {
1013                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1014                 refcount_inc(&dp->dl_stid.sc_count);
1015                 spin_lock(&clp->cl_lock);
1016                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1017                 spin_unlock(&clp->cl_lock);
1018         }
1019         destroy_unhashed_deleg(dp);
1020 }
1021
1022 /*
1023  * SETCLIENTID state
1024  */
1025
1026 static unsigned int clientid_hashval(u32 id)
1027 {
1028         return id & CLIENT_HASH_MASK;
1029 }
1030
1031 static unsigned int clientstr_hashval(const char *name)
1032 {
1033         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1034 }
1035
1036 /*
1037  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1038  * st_{access,deny}_bmap field of the stateid, in order to track not
1039  * only what share bits are currently in force, but also what
1040  * combinations of share bits previous opens have used.  This allows us
1041  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1042  * return an error if the client attempt to downgrade to a combination
1043  * of share bits not explicable by closing some of its previous opens.
1044  *
1045  * XXX: This enforcement is actually incomplete, since we don't keep
1046  * track of access/deny bit combinations; so, e.g., we allow:
1047  *
1048  *      OPEN allow read, deny write
1049  *      OPEN allow both, deny none
1050  *      DOWNGRADE allow read, deny none
1051  *
1052  * which we should reject.
1053  */
1054 static unsigned int
1055 bmap_to_share_mode(unsigned long bmap) {
1056         int i;
1057         unsigned int access = 0;
1058
1059         for (i = 1; i < 4; i++) {
1060                 if (test_bit(i, &bmap))
1061                         access |= i;
1062         }
1063         return access;
1064 }
1065
1066 /* set share access for a given stateid */
1067 static inline void
1068 set_access(u32 access, struct nfs4_ol_stateid *stp)
1069 {
1070         unsigned char mask = 1 << access;
1071
1072         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1073         stp->st_access_bmap |= mask;
1074 }
1075
1076 /* clear share access for a given stateid */
1077 static inline void
1078 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1079 {
1080         unsigned char mask = 1 << access;
1081
1082         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1083         stp->st_access_bmap &= ~mask;
1084 }
1085
1086 /* test whether a given stateid has access */
1087 static inline bool
1088 test_access(u32 access, struct nfs4_ol_stateid *stp)
1089 {
1090         unsigned char mask = 1 << access;
1091
1092         return (bool)(stp->st_access_bmap & mask);
1093 }
1094
1095 /* set share deny for a given stateid */
1096 static inline void
1097 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1098 {
1099         unsigned char mask = 1 << deny;
1100
1101         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1102         stp->st_deny_bmap |= mask;
1103 }
1104
1105 /* clear share deny for a given stateid */
1106 static inline void
1107 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1108 {
1109         unsigned char mask = 1 << deny;
1110
1111         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1112         stp->st_deny_bmap &= ~mask;
1113 }
1114
1115 /* test whether a given stateid is denying specific access */
1116 static inline bool
1117 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1118 {
1119         unsigned char mask = 1 << deny;
1120
1121         return (bool)(stp->st_deny_bmap & mask);
1122 }
1123
1124 static int nfs4_access_to_omode(u32 access)
1125 {
1126         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1127         case NFS4_SHARE_ACCESS_READ:
1128                 return O_RDONLY;
1129         case NFS4_SHARE_ACCESS_WRITE:
1130                 return O_WRONLY;
1131         case NFS4_SHARE_ACCESS_BOTH:
1132                 return O_RDWR;
1133         }
1134         WARN_ON_ONCE(1);
1135         return O_RDONLY;
1136 }
1137
1138 /*
1139  * A stateid that had a deny mode associated with it is being released
1140  * or downgraded. Recalculate the deny mode on the file.
1141  */
1142 static void
1143 recalculate_deny_mode(struct nfs4_file *fp)
1144 {
1145         struct nfs4_ol_stateid *stp;
1146
1147         spin_lock(&fp->fi_lock);
1148         fp->fi_share_deny = 0;
1149         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1150                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1151         spin_unlock(&fp->fi_lock);
1152 }
1153
1154 static void
1155 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1156 {
1157         int i;
1158         bool change = false;
1159
1160         for (i = 1; i < 4; i++) {
1161                 if ((i & deny) != i) {
1162                         change = true;
1163                         clear_deny(i, stp);
1164                 }
1165         }
1166
1167         /* Recalculate per-file deny mode if there was a change */
1168         if (change)
1169                 recalculate_deny_mode(stp->st_stid.sc_file);
1170 }
1171
1172 /* release all access and file references for a given stateid */
1173 static void
1174 release_all_access(struct nfs4_ol_stateid *stp)
1175 {
1176         int i;
1177         struct nfs4_file *fp = stp->st_stid.sc_file;
1178
1179         if (fp && stp->st_deny_bmap != 0)
1180                 recalculate_deny_mode(fp);
1181
1182         for (i = 1; i < 4; i++) {
1183                 if (test_access(i, stp))
1184                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1185                 clear_access(i, stp);
1186         }
1187 }
1188
1189 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1190 {
1191         kfree(sop->so_owner.data);
1192         sop->so_ops->so_free(sop);
1193 }
1194
1195 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1196 {
1197         struct nfs4_client *clp = sop->so_client;
1198
1199         might_lock(&clp->cl_lock);
1200
1201         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1202                 return;
1203         sop->so_ops->so_unhash(sop);
1204         spin_unlock(&clp->cl_lock);
1205         nfs4_free_stateowner(sop);
1206 }
1207
1208 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1209 {
1210         struct nfs4_file *fp = stp->st_stid.sc_file;
1211
1212         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1213
1214         if (list_empty(&stp->st_perfile))
1215                 return false;
1216
1217         spin_lock(&fp->fi_lock);
1218         list_del_init(&stp->st_perfile);
1219         spin_unlock(&fp->fi_lock);
1220         list_del(&stp->st_perstateowner);
1221         return true;
1222 }
1223
1224 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1225 {
1226         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1227
1228         put_clnt_odstate(stp->st_clnt_odstate);
1229         release_all_access(stp);
1230         if (stp->st_stateowner)
1231                 nfs4_put_stateowner(stp->st_stateowner);
1232         kmem_cache_free(stateid_slab, stid);
1233 }
1234
1235 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1236 {
1237         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1238         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1239         struct file *file;
1240
1241         file = find_any_file(stp->st_stid.sc_file);
1242         if (file)
1243                 filp_close(file, (fl_owner_t)lo);
1244         nfs4_free_ol_stateid(stid);
1245 }
1246
1247 /*
1248  * Put the persistent reference to an already unhashed generic stateid, while
1249  * holding the cl_lock. If it's the last reference, then put it onto the
1250  * reaplist for later destruction.
1251  */
1252 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1253                                        struct list_head *reaplist)
1254 {
1255         struct nfs4_stid *s = &stp->st_stid;
1256         struct nfs4_client *clp = s->sc_client;
1257
1258         lockdep_assert_held(&clp->cl_lock);
1259
1260         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1261
1262         if (!refcount_dec_and_test(&s->sc_count)) {
1263                 wake_up_all(&close_wq);
1264                 return;
1265         }
1266
1267         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1268         list_add(&stp->st_locks, reaplist);
1269 }
1270
1271 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1272 {
1273         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1274
1275         list_del_init(&stp->st_locks);
1276         nfs4_unhash_stid(&stp->st_stid);
1277         return unhash_ol_stateid(stp);
1278 }
1279
1280 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1281 {
1282         struct nfs4_client *clp = stp->st_stid.sc_client;
1283         bool unhashed;
1284
1285         spin_lock(&clp->cl_lock);
1286         unhashed = unhash_lock_stateid(stp);
1287         spin_unlock(&clp->cl_lock);
1288         if (unhashed)
1289                 nfs4_put_stid(&stp->st_stid);
1290 }
1291
1292 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1293 {
1294         struct nfs4_client *clp = lo->lo_owner.so_client;
1295
1296         lockdep_assert_held(&clp->cl_lock);
1297
1298         list_del_init(&lo->lo_owner.so_strhash);
1299 }
1300
1301 /*
1302  * Free a list of generic stateids that were collected earlier after being
1303  * fully unhashed.
1304  */
1305 static void
1306 free_ol_stateid_reaplist(struct list_head *reaplist)
1307 {
1308         struct nfs4_ol_stateid *stp;
1309         struct nfs4_file *fp;
1310
1311         might_sleep();
1312
1313         while (!list_empty(reaplist)) {
1314                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1315                                        st_locks);
1316                 list_del(&stp->st_locks);
1317                 fp = stp->st_stid.sc_file;
1318                 stp->st_stid.sc_free(&stp->st_stid);
1319                 if (fp)
1320                         put_nfs4_file(fp);
1321         }
1322 }
1323
1324 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1325                                        struct list_head *reaplist)
1326 {
1327         struct nfs4_ol_stateid *stp;
1328
1329         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1330
1331         while (!list_empty(&open_stp->st_locks)) {
1332                 stp = list_entry(open_stp->st_locks.next,
1333                                 struct nfs4_ol_stateid, st_locks);
1334                 WARN_ON(!unhash_lock_stateid(stp));
1335                 put_ol_stateid_locked(stp, reaplist);
1336         }
1337 }
1338
1339 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1340                                 struct list_head *reaplist)
1341 {
1342         bool unhashed;
1343
1344         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1345
1346         unhashed = unhash_ol_stateid(stp);
1347         release_open_stateid_locks(stp, reaplist);
1348         return unhashed;
1349 }
1350
1351 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1352 {
1353         LIST_HEAD(reaplist);
1354
1355         spin_lock(&stp->st_stid.sc_client->cl_lock);
1356         if (unhash_open_stateid(stp, &reaplist))
1357                 put_ol_stateid_locked(stp, &reaplist);
1358         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1359         free_ol_stateid_reaplist(&reaplist);
1360 }
1361
1362 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1363 {
1364         struct nfs4_client *clp = oo->oo_owner.so_client;
1365
1366         lockdep_assert_held(&clp->cl_lock);
1367
1368         list_del_init(&oo->oo_owner.so_strhash);
1369         list_del_init(&oo->oo_perclient);
1370 }
1371
1372 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1373 {
1374         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1375                                           nfsd_net_id);
1376         struct nfs4_ol_stateid *s;
1377
1378         spin_lock(&nn->client_lock);
1379         s = oo->oo_last_closed_stid;
1380         if (s) {
1381                 list_del_init(&oo->oo_close_lru);
1382                 oo->oo_last_closed_stid = NULL;
1383         }
1384         spin_unlock(&nn->client_lock);
1385         if (s)
1386                 nfs4_put_stid(&s->st_stid);
1387 }
1388
1389 static void release_openowner(struct nfs4_openowner *oo)
1390 {
1391         struct nfs4_ol_stateid *stp;
1392         struct nfs4_client *clp = oo->oo_owner.so_client;
1393         struct list_head reaplist;
1394
1395         INIT_LIST_HEAD(&reaplist);
1396
1397         spin_lock(&clp->cl_lock);
1398         unhash_openowner_locked(oo);
1399         while (!list_empty(&oo->oo_owner.so_stateids)) {
1400                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1401                                 struct nfs4_ol_stateid, st_perstateowner);
1402                 if (unhash_open_stateid(stp, &reaplist))
1403                         put_ol_stateid_locked(stp, &reaplist);
1404         }
1405         spin_unlock(&clp->cl_lock);
1406         free_ol_stateid_reaplist(&reaplist);
1407         release_last_closed_stateid(oo);
1408         nfs4_put_stateowner(&oo->oo_owner);
1409 }
1410
1411 static inline int
1412 hash_sessionid(struct nfs4_sessionid *sessionid)
1413 {
1414         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1415
1416         return sid->sequence % SESSION_HASH_SIZE;
1417 }
1418
1419 #ifdef CONFIG_SUNRPC_DEBUG
1420 static inline void
1421 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1422 {
1423         u32 *ptr = (u32 *)(&sessionid->data[0]);
1424         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1425 }
1426 #else
1427 static inline void
1428 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1429 {
1430 }
1431 #endif
1432
1433 /*
1434  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1435  * won't be used for replay.
1436  */
1437 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1438 {
1439         struct nfs4_stateowner *so = cstate->replay_owner;
1440
1441         if (nfserr == nfserr_replay_me)
1442                 return;
1443
1444         if (!seqid_mutating_err(ntohl(nfserr))) {
1445                 nfsd4_cstate_clear_replay(cstate);
1446                 return;
1447         }
1448         if (!so)
1449                 return;
1450         if (so->so_is_open_owner)
1451                 release_last_closed_stateid(openowner(so));
1452         so->so_seqid++;
1453         return;
1454 }
1455
1456 static void
1457 gen_sessionid(struct nfsd4_session *ses)
1458 {
1459         struct nfs4_client *clp = ses->se_client;
1460         struct nfsd4_sessionid *sid;
1461
1462         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1463         sid->clientid = clp->cl_clientid;
1464         sid->sequence = current_sessionid++;
1465         sid->reserved = 0;
1466 }
1467
1468 /*
1469  * The protocol defines ca_maxresponssize_cached to include the size of
1470  * the rpc header, but all we need to cache is the data starting after
1471  * the end of the initial SEQUENCE operation--the rest we regenerate
1472  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1473  * value that is the number of bytes in our cache plus a few additional
1474  * bytes.  In order to stay on the safe side, and not promise more than
1475  * we can cache, those additional bytes must be the minimum possible: 24
1476  * bytes of rpc header (xid through accept state, with AUTH_NULL
1477  * verifier), 12 for the compound header (with zero-length tag), and 44
1478  * for the SEQUENCE op response:
1479  */
1480 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1481
1482 static void
1483 free_session_slots(struct nfsd4_session *ses)
1484 {
1485         int i;
1486
1487         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1488                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1489                 kfree(ses->se_slots[i]);
1490         }
1491 }
1492
1493 /*
1494  * We don't actually need to cache the rpc and session headers, so we
1495  * can allocate a little less for each slot:
1496  */
1497 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1498 {
1499         u32 size;
1500
1501         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1502                 size = 0;
1503         else
1504                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1505         return size + sizeof(struct nfsd4_slot);
1506 }
1507
1508 /*
1509  * XXX: If we run out of reserved DRC memory we could (up to a point)
1510  * re-negotiate active sessions and reduce their slot usage to make
1511  * room for new connections. For now we just fail the create session.
1512  */
1513 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1514 {
1515         u32 slotsize = slot_bytes(ca);
1516         u32 num = ca->maxreqs;
1517         int avail;
1518
1519         spin_lock(&nfsd_drc_lock);
1520         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1521                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1522         /*
1523          * Never use more than a third of the remaining memory,
1524          * unless it's the only way to give this client a slot:
1525          */
1526         avail = clamp_t(int, avail, slotsize, avail/3);
1527         num = min_t(int, num, avail / slotsize);
1528         nfsd_drc_mem_used += num * slotsize;
1529         spin_unlock(&nfsd_drc_lock);
1530
1531         return num;
1532 }
1533
1534 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1535 {
1536         int slotsize = slot_bytes(ca);
1537
1538         spin_lock(&nfsd_drc_lock);
1539         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1540         spin_unlock(&nfsd_drc_lock);
1541 }
1542
1543 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1544                                            struct nfsd4_channel_attrs *battrs)
1545 {
1546         int numslots = fattrs->maxreqs;
1547         int slotsize = slot_bytes(fattrs);
1548         struct nfsd4_session *new;
1549         int mem, i;
1550
1551         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1552                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1553         mem = numslots * sizeof(struct nfsd4_slot *);
1554
1555         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1556         if (!new)
1557                 return NULL;
1558         /* allocate each struct nfsd4_slot and data cache in one piece */
1559         for (i = 0; i < numslots; i++) {
1560                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1561                 if (!new->se_slots[i])
1562                         goto out_free;
1563         }
1564
1565         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1566         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1567
1568         return new;
1569 out_free:
1570         while (i--)
1571                 kfree(new->se_slots[i]);
1572         kfree(new);
1573         return NULL;
1574 }
1575
1576 static void free_conn(struct nfsd4_conn *c)
1577 {
1578         svc_xprt_put(c->cn_xprt);
1579         kfree(c);
1580 }
1581
1582 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1583 {
1584         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1585         struct nfs4_client *clp = c->cn_session->se_client;
1586
1587         spin_lock(&clp->cl_lock);
1588         if (!list_empty(&c->cn_persession)) {
1589                 list_del(&c->cn_persession);
1590                 free_conn(c);
1591         }
1592         nfsd4_probe_callback(clp);
1593         spin_unlock(&clp->cl_lock);
1594 }
1595
1596 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1597 {
1598         struct nfsd4_conn *conn;
1599
1600         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1601         if (!conn)
1602                 return NULL;
1603         svc_xprt_get(rqstp->rq_xprt);
1604         conn->cn_xprt = rqstp->rq_xprt;
1605         conn->cn_flags = flags;
1606         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1607         return conn;
1608 }
1609
1610 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1611 {
1612         conn->cn_session = ses;
1613         list_add(&conn->cn_persession, &ses->se_conns);
1614 }
1615
1616 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1617 {
1618         struct nfs4_client *clp = ses->se_client;
1619
1620         spin_lock(&clp->cl_lock);
1621         __nfsd4_hash_conn(conn, ses);
1622         spin_unlock(&clp->cl_lock);
1623 }
1624
1625 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1626 {
1627         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1628         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1629 }
1630
1631 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1632 {
1633         int ret;
1634
1635         nfsd4_hash_conn(conn, ses);
1636         ret = nfsd4_register_conn(conn);
1637         if (ret)
1638                 /* oops; xprt is already down: */
1639                 nfsd4_conn_lost(&conn->cn_xpt_user);
1640         /* We may have gained or lost a callback channel: */
1641         nfsd4_probe_callback_sync(ses->se_client);
1642 }
1643
1644 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1645 {
1646         u32 dir = NFS4_CDFC4_FORE;
1647
1648         if (cses->flags & SESSION4_BACK_CHAN)
1649                 dir |= NFS4_CDFC4_BACK;
1650         return alloc_conn(rqstp, dir);
1651 }
1652
1653 /* must be called under client_lock */
1654 static void nfsd4_del_conns(struct nfsd4_session *s)
1655 {
1656         struct nfs4_client *clp = s->se_client;
1657         struct nfsd4_conn *c;
1658
1659         spin_lock(&clp->cl_lock);
1660         while (!list_empty(&s->se_conns)) {
1661                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1662                 list_del_init(&c->cn_persession);
1663                 spin_unlock(&clp->cl_lock);
1664
1665                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1666                 free_conn(c);
1667
1668                 spin_lock(&clp->cl_lock);
1669         }
1670         spin_unlock(&clp->cl_lock);
1671 }
1672
1673 static void __free_session(struct nfsd4_session *ses)
1674 {
1675         free_session_slots(ses);
1676         kfree(ses);
1677 }
1678
1679 static void free_session(struct nfsd4_session *ses)
1680 {
1681         nfsd4_del_conns(ses);
1682         nfsd4_put_drc_mem(&ses->se_fchannel);
1683         __free_session(ses);
1684 }
1685
1686 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1687 {
1688         int idx;
1689         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1690
1691         new->se_client = clp;
1692         gen_sessionid(new);
1693
1694         INIT_LIST_HEAD(&new->se_conns);
1695
1696         new->se_cb_seq_nr = 1;
1697         new->se_flags = cses->flags;
1698         new->se_cb_prog = cses->callback_prog;
1699         new->se_cb_sec = cses->cb_sec;
1700         atomic_set(&new->se_ref, 0);
1701         idx = hash_sessionid(&new->se_sessionid);
1702         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1703         spin_lock(&clp->cl_lock);
1704         list_add(&new->se_perclnt, &clp->cl_sessions);
1705         spin_unlock(&clp->cl_lock);
1706
1707         {
1708                 struct sockaddr *sa = svc_addr(rqstp);
1709                 /*
1710                  * This is a little silly; with sessions there's no real
1711                  * use for the callback address.  Use the peer address
1712                  * as a reasonable default for now, but consider fixing
1713                  * the rpc client not to require an address in the
1714                  * future:
1715                  */
1716                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1717                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1718         }
1719 }
1720
1721 /* caller must hold client_lock */
1722 static struct nfsd4_session *
1723 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1724 {
1725         struct nfsd4_session *elem;
1726         int idx;
1727         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1728
1729         lockdep_assert_held(&nn->client_lock);
1730
1731         dump_sessionid(__func__, sessionid);
1732         idx = hash_sessionid(sessionid);
1733         /* Search in the appropriate list */
1734         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1735                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1736                             NFS4_MAX_SESSIONID_LEN)) {
1737                         return elem;
1738                 }
1739         }
1740
1741         dprintk("%s: session not found\n", __func__);
1742         return NULL;
1743 }
1744
1745 static struct nfsd4_session *
1746 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1747                 __be32 *ret)
1748 {
1749         struct nfsd4_session *session;
1750         __be32 status = nfserr_badsession;
1751
1752         session = __find_in_sessionid_hashtbl(sessionid, net);
1753         if (!session)
1754                 goto out;
1755         status = nfsd4_get_session_locked(session);
1756         if (status)
1757                 session = NULL;
1758 out:
1759         *ret = status;
1760         return session;
1761 }
1762
1763 /* caller must hold client_lock */
1764 static void
1765 unhash_session(struct nfsd4_session *ses)
1766 {
1767         struct nfs4_client *clp = ses->se_client;
1768         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1769
1770         lockdep_assert_held(&nn->client_lock);
1771
1772         list_del(&ses->se_hash);
1773         spin_lock(&ses->se_client->cl_lock);
1774         list_del(&ses->se_perclnt);
1775         spin_unlock(&ses->se_client->cl_lock);
1776 }
1777
1778 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1779 static int
1780 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1781 {
1782         /*
1783          * We're assuming the clid was not given out from a boot
1784          * precisely 2^32 (about 136 years) before this one.  That seems
1785          * a safe assumption:
1786          */
1787         if (clid->cl_boot == (u32)nn->boot_time)
1788                 return 0;
1789         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1790                 clid->cl_boot, clid->cl_id, nn->boot_time);
1791         return 1;
1792 }
1793
1794 /*
1795  * XXX Should we use a slab cache ?
1796  * This type of memory management is somewhat inefficient, but we use it
1797  * anyway since SETCLIENTID is not a common operation.
1798  */
1799 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1800 {
1801         struct nfs4_client *clp;
1802         int i;
1803
1804         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1805         if (clp == NULL)
1806                 return NULL;
1807         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1808         if (clp->cl_name.data == NULL)
1809                 goto err_no_name;
1810         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1811                                                  sizeof(struct list_head),
1812                                                  GFP_KERNEL);
1813         if (!clp->cl_ownerstr_hashtbl)
1814                 goto err_no_hashtbl;
1815         for (i = 0; i < OWNER_HASH_SIZE; i++)
1816                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1817         clp->cl_name.len = name.len;
1818         INIT_LIST_HEAD(&clp->cl_sessions);
1819         idr_init(&clp->cl_stateids);
1820         atomic_set(&clp->cl_refcount, 0);
1821         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1822         INIT_LIST_HEAD(&clp->cl_idhash);
1823         INIT_LIST_HEAD(&clp->cl_openowners);
1824         INIT_LIST_HEAD(&clp->cl_delegations);
1825         INIT_LIST_HEAD(&clp->cl_lru);
1826         INIT_LIST_HEAD(&clp->cl_revoked);
1827 #ifdef CONFIG_NFSD_PNFS
1828         INIT_LIST_HEAD(&clp->cl_lo_states);
1829 #endif
1830         spin_lock_init(&clp->cl_lock);
1831         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1832         return clp;
1833 err_no_hashtbl:
1834         kfree(clp->cl_name.data);
1835 err_no_name:
1836         kmem_cache_free(client_slab, clp);
1837         return NULL;
1838 }
1839
1840 static void
1841 free_client(struct nfs4_client *clp)
1842 {
1843         while (!list_empty(&clp->cl_sessions)) {
1844                 struct nfsd4_session *ses;
1845                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1846                                 se_perclnt);
1847                 list_del(&ses->se_perclnt);
1848                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1849                 free_session(ses);
1850         }
1851         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1852         free_svc_cred(&clp->cl_cred);
1853         kfree(clp->cl_ownerstr_hashtbl);
1854         kfree(clp->cl_name.data);
1855         idr_destroy(&clp->cl_stateids);
1856         kmem_cache_free(client_slab, clp);
1857 }
1858
1859 /* must be called under the client_lock */
1860 static void
1861 unhash_client_locked(struct nfs4_client *clp)
1862 {
1863         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1864         struct nfsd4_session *ses;
1865
1866         lockdep_assert_held(&nn->client_lock);
1867
1868         /* Mark the client as expired! */
1869         clp->cl_time = 0;
1870         /* Make it invisible */
1871         if (!list_empty(&clp->cl_idhash)) {
1872                 list_del_init(&clp->cl_idhash);
1873                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1874                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1875                 else
1876                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1877         }
1878         list_del_init(&clp->cl_lru);
1879         spin_lock(&clp->cl_lock);
1880         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1881                 list_del_init(&ses->se_hash);
1882         spin_unlock(&clp->cl_lock);
1883 }
1884
1885 static void
1886 unhash_client(struct nfs4_client *clp)
1887 {
1888         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1889
1890         spin_lock(&nn->client_lock);
1891         unhash_client_locked(clp);
1892         spin_unlock(&nn->client_lock);
1893 }
1894
1895 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1896 {
1897         if (atomic_read(&clp->cl_refcount))
1898                 return nfserr_jukebox;
1899         unhash_client_locked(clp);
1900         return nfs_ok;
1901 }
1902
1903 static void
1904 __destroy_client(struct nfs4_client *clp)
1905 {
1906         int i;
1907         struct nfs4_openowner *oo;
1908         struct nfs4_delegation *dp;
1909         struct list_head reaplist;
1910
1911         INIT_LIST_HEAD(&reaplist);
1912         spin_lock(&state_lock);
1913         while (!list_empty(&clp->cl_delegations)) {
1914                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1915                 WARN_ON(!unhash_delegation_locked(dp));
1916                 list_add(&dp->dl_recall_lru, &reaplist);
1917         }
1918         spin_unlock(&state_lock);
1919         while (!list_empty(&reaplist)) {
1920                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1921                 list_del_init(&dp->dl_recall_lru);
1922                 destroy_unhashed_deleg(dp);
1923         }
1924         while (!list_empty(&clp->cl_revoked)) {
1925                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1926                 list_del_init(&dp->dl_recall_lru);
1927                 nfs4_put_stid(&dp->dl_stid);
1928         }
1929         while (!list_empty(&clp->cl_openowners)) {
1930                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1931                 nfs4_get_stateowner(&oo->oo_owner);
1932                 release_openowner(oo);
1933         }
1934         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1935                 struct nfs4_stateowner *so, *tmp;
1936
1937                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1938                                          so_strhash) {
1939                         /* Should be no openowners at this point */
1940                         WARN_ON_ONCE(so->so_is_open_owner);
1941                         remove_blocked_locks(lockowner(so));
1942                 }
1943         }
1944         nfsd4_return_all_client_layouts(clp);
1945         nfsd4_shutdown_callback(clp);
1946         if (clp->cl_cb_conn.cb_xprt)
1947                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1948         free_client(clp);
1949 }
1950
1951 static void
1952 destroy_client(struct nfs4_client *clp)
1953 {
1954         unhash_client(clp);
1955         __destroy_client(clp);
1956 }
1957
1958 static void expire_client(struct nfs4_client *clp)
1959 {
1960         unhash_client(clp);
1961         nfsd4_client_record_remove(clp);
1962         __destroy_client(clp);
1963 }
1964
1965 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1966 {
1967         memcpy(target->cl_verifier.data, source->data,
1968                         sizeof(target->cl_verifier.data));
1969 }
1970
1971 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1972 {
1973         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1974         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1975 }
1976
1977 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1978 {
1979         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1980         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1981                                                                 GFP_KERNEL);
1982         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
1983         if ((source->cr_principal && !target->cr_principal) ||
1984             (source->cr_raw_principal && !target->cr_raw_principal) ||
1985             (source->cr_targ_princ && !target->cr_targ_princ))
1986                 return -ENOMEM;
1987
1988         target->cr_flavor = source->cr_flavor;
1989         target->cr_uid = source->cr_uid;
1990         target->cr_gid = source->cr_gid;
1991         target->cr_group_info = source->cr_group_info;
1992         get_group_info(target->cr_group_info);
1993         target->cr_gss_mech = source->cr_gss_mech;
1994         if (source->cr_gss_mech)
1995                 gss_mech_get(source->cr_gss_mech);
1996         return 0;
1997 }
1998
1999 static int
2000 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2001 {
2002         if (o1->len < o2->len)
2003                 return -1;
2004         if (o1->len > o2->len)
2005                 return 1;
2006         return memcmp(o1->data, o2->data, o1->len);
2007 }
2008
2009 static int same_name(const char *n1, const char *n2)
2010 {
2011         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2012 }
2013
2014 static int
2015 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2016 {
2017         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2018 }
2019
2020 static int
2021 same_clid(clientid_t *cl1, clientid_t *cl2)
2022 {
2023         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2024 }
2025
2026 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2027 {
2028         int i;
2029
2030         if (g1->ngroups != g2->ngroups)
2031                 return false;
2032         for (i=0; i<g1->ngroups; i++)
2033                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2034                         return false;
2035         return true;
2036 }
2037
2038 /*
2039  * RFC 3530 language requires clid_inuse be returned when the
2040  * "principal" associated with a requests differs from that previously
2041  * used.  We use uid, gid's, and gss principal string as our best
2042  * approximation.  We also don't want to allow non-gss use of a client
2043  * established using gss: in theory cr_principal should catch that
2044  * change, but in practice cr_principal can be null even in the gss case
2045  * since gssd doesn't always pass down a principal string.
2046  */
2047 static bool is_gss_cred(struct svc_cred *cr)
2048 {
2049         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2050         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2051 }
2052
2053
2054 static bool
2055 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2056 {
2057         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2058                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2059                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2060                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2061                 return false;
2062         /* XXX: check that cr_targ_princ fields match ? */
2063         if (cr1->cr_principal == cr2->cr_principal)
2064                 return true;
2065         if (!cr1->cr_principal || !cr2->cr_principal)
2066                 return false;
2067         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2068 }
2069
2070 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2071 {
2072         struct svc_cred *cr = &rqstp->rq_cred;
2073         u32 service;
2074
2075         if (!cr->cr_gss_mech)
2076                 return false;
2077         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2078         return service == RPC_GSS_SVC_INTEGRITY ||
2079                service == RPC_GSS_SVC_PRIVACY;
2080 }
2081
2082 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2083 {
2084         struct svc_cred *cr = &rqstp->rq_cred;
2085
2086         if (!cl->cl_mach_cred)
2087                 return true;
2088         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2089                 return false;
2090         if (!svc_rqst_integrity_protected(rqstp))
2091                 return false;
2092         if (cl->cl_cred.cr_raw_principal)
2093                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2094                                                 cr->cr_raw_principal);
2095         if (!cr->cr_principal)
2096                 return false;
2097         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2098 }
2099
2100 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2101 {
2102         __be32 verf[2];
2103
2104         /*
2105          * This is opaque to client, so no need to byte-swap. Use
2106          * __force to keep sparse happy
2107          */
2108         verf[0] = (__force __be32)get_seconds();
2109         verf[1] = (__force __be32)nn->clverifier_counter++;
2110         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2111 }
2112
2113 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2114 {
2115         clp->cl_clientid.cl_boot = nn->boot_time;
2116         clp->cl_clientid.cl_id = nn->clientid_counter++;
2117         gen_confirm(clp, nn);
2118 }
2119
2120 static struct nfs4_stid *
2121 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2122 {
2123         struct nfs4_stid *ret;
2124
2125         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2126         if (!ret || !ret->sc_type)
2127                 return NULL;
2128         return ret;
2129 }
2130
2131 static struct nfs4_stid *
2132 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2133 {
2134         struct nfs4_stid *s;
2135
2136         spin_lock(&cl->cl_lock);
2137         s = find_stateid_locked(cl, t);
2138         if (s != NULL) {
2139                 if (typemask & s->sc_type)
2140                         refcount_inc(&s->sc_count);
2141                 else
2142                         s = NULL;
2143         }
2144         spin_unlock(&cl->cl_lock);
2145         return s;
2146 }
2147
2148 static struct nfs4_client *create_client(struct xdr_netobj name,
2149                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2150 {
2151         struct nfs4_client *clp;
2152         struct sockaddr *sa = svc_addr(rqstp);
2153         int ret;
2154         struct net *net = SVC_NET(rqstp);
2155
2156         clp = alloc_client(name);
2157         if (clp == NULL)
2158                 return NULL;
2159
2160         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2161         if (ret) {
2162                 free_client(clp);
2163                 return NULL;
2164         }
2165         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2166         clp->cl_time = get_seconds();
2167         clear_bit(0, &clp->cl_cb_slot_busy);
2168         copy_verf(clp, verf);
2169         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2170         clp->cl_cb_session = NULL;
2171         clp->net = net;
2172         return clp;
2173 }
2174
2175 static void
2176 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2177 {
2178         struct rb_node **new = &(root->rb_node), *parent = NULL;
2179         struct nfs4_client *clp;
2180
2181         while (*new) {
2182                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2183                 parent = *new;
2184
2185                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2186                         new = &((*new)->rb_left);
2187                 else
2188                         new = &((*new)->rb_right);
2189         }
2190
2191         rb_link_node(&new_clp->cl_namenode, parent, new);
2192         rb_insert_color(&new_clp->cl_namenode, root);
2193 }
2194
2195 static struct nfs4_client *
2196 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2197 {
2198         int cmp;
2199         struct rb_node *node = root->rb_node;
2200         struct nfs4_client *clp;
2201
2202         while (node) {
2203                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2204                 cmp = compare_blob(&clp->cl_name, name);
2205                 if (cmp > 0)
2206                         node = node->rb_left;
2207                 else if (cmp < 0)
2208                         node = node->rb_right;
2209                 else
2210                         return clp;
2211         }
2212         return NULL;
2213 }
2214
2215 static void
2216 add_to_unconfirmed(struct nfs4_client *clp)
2217 {
2218         unsigned int idhashval;
2219         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2220
2221         lockdep_assert_held(&nn->client_lock);
2222
2223         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2224         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2225         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2226         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2227         renew_client_locked(clp);
2228 }
2229
2230 static void
2231 move_to_confirmed(struct nfs4_client *clp)
2232 {
2233         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2234         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2235
2236         lockdep_assert_held(&nn->client_lock);
2237
2238         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2239         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2240         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2241         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2242         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2243         renew_client_locked(clp);
2244 }
2245
2246 static struct nfs4_client *
2247 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2248 {
2249         struct nfs4_client *clp;
2250         unsigned int idhashval = clientid_hashval(clid->cl_id);
2251
2252         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2253                 if (same_clid(&clp->cl_clientid, clid)) {
2254                         if ((bool)clp->cl_minorversion != sessions)
2255                                 return NULL;
2256                         renew_client_locked(clp);
2257                         return clp;
2258                 }
2259         }
2260         return NULL;
2261 }
2262
2263 static struct nfs4_client *
2264 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2265 {
2266         struct list_head *tbl = nn->conf_id_hashtbl;
2267
2268         lockdep_assert_held(&nn->client_lock);
2269         return find_client_in_id_table(tbl, clid, sessions);
2270 }
2271
2272 static struct nfs4_client *
2273 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2274 {
2275         struct list_head *tbl = nn->unconf_id_hashtbl;
2276
2277         lockdep_assert_held(&nn->client_lock);
2278         return find_client_in_id_table(tbl, clid, sessions);
2279 }
2280
2281 static bool clp_used_exchangeid(struct nfs4_client *clp)
2282 {
2283         return clp->cl_exchange_flags != 0;
2284 }
2285
2286 static struct nfs4_client *
2287 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2288 {
2289         lockdep_assert_held(&nn->client_lock);
2290         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2291 }
2292
2293 static struct nfs4_client *
2294 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2295 {
2296         lockdep_assert_held(&nn->client_lock);
2297         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2298 }
2299
2300 static void
2301 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2302 {
2303         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2304         struct sockaddr *sa = svc_addr(rqstp);
2305         u32 scopeid = rpc_get_scope_id(sa);
2306         unsigned short expected_family;
2307
2308         /* Currently, we only support tcp and tcp6 for the callback channel */
2309         if (se->se_callback_netid_len == 3 &&
2310             !memcmp(se->se_callback_netid_val, "tcp", 3))
2311                 expected_family = AF_INET;
2312         else if (se->se_callback_netid_len == 4 &&
2313                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2314                 expected_family = AF_INET6;
2315         else
2316                 goto out_err;
2317
2318         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2319                                             se->se_callback_addr_len,
2320                                             (struct sockaddr *)&conn->cb_addr,
2321                                             sizeof(conn->cb_addr));
2322
2323         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2324                 goto out_err;
2325
2326         if (conn->cb_addr.ss_family == AF_INET6)
2327                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2328
2329         conn->cb_prog = se->se_callback_prog;
2330         conn->cb_ident = se->se_callback_ident;
2331         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2332         return;
2333 out_err:
2334         conn->cb_addr.ss_family = AF_UNSPEC;
2335         conn->cb_addrlen = 0;
2336         dprintk("NFSD: this client (clientid %08x/%08x) "
2337                 "will not receive delegations\n",
2338                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2339
2340         return;
2341 }
2342
2343 /*
2344  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2345  */
2346 static void
2347 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2348 {
2349         struct xdr_buf *buf = resp->xdr.buf;
2350         struct nfsd4_slot *slot = resp->cstate.slot;
2351         unsigned int base;
2352
2353         dprintk("--> %s slot %p\n", __func__, slot);
2354
2355         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2356         slot->sl_opcnt = resp->opcnt;
2357         slot->sl_status = resp->cstate.status;
2358         free_svc_cred(&slot->sl_cred);
2359         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2360
2361         if (!nfsd4_cache_this(resp)) {
2362                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2363                 return;
2364         }
2365         slot->sl_flags |= NFSD4_SLOT_CACHED;
2366
2367         base = resp->cstate.data_offset;
2368         slot->sl_datalen = buf->len - base;
2369         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2370                 WARN(1, "%s: sessions DRC could not cache compound\n",
2371                      __func__);
2372         return;
2373 }
2374
2375 /*
2376  * Encode the replay sequence operation from the slot values.
2377  * If cachethis is FALSE encode the uncached rep error on the next
2378  * operation which sets resp->p and increments resp->opcnt for
2379  * nfs4svc_encode_compoundres.
2380  *
2381  */
2382 static __be32
2383 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2384                           struct nfsd4_compoundres *resp)
2385 {
2386         struct nfsd4_op *op;
2387         struct nfsd4_slot *slot = resp->cstate.slot;
2388
2389         /* Encode the replayed sequence operation */
2390         op = &args->ops[resp->opcnt - 1];
2391         nfsd4_encode_operation(resp, op);
2392
2393         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2394                 return op->status;
2395         if (args->opcnt == 1) {
2396                 /*
2397                  * The original operation wasn't a solo sequence--we
2398                  * always cache those--so this retry must not match the
2399                  * original:
2400                  */
2401                 op->status = nfserr_seq_false_retry;
2402         } else {
2403                 op = &args->ops[resp->opcnt++];
2404                 op->status = nfserr_retry_uncached_rep;
2405                 nfsd4_encode_operation(resp, op);
2406         }
2407         return op->status;
2408 }
2409
2410 /*
2411  * The sequence operation is not cached because we can use the slot and
2412  * session values.
2413  */
2414 static __be32
2415 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2416                          struct nfsd4_sequence *seq)
2417 {
2418         struct nfsd4_slot *slot = resp->cstate.slot;
2419         struct xdr_stream *xdr = &resp->xdr;
2420         __be32 *p;
2421         __be32 status;
2422
2423         dprintk("--> %s slot %p\n", __func__, slot);
2424
2425         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2426         if (status)
2427                 return status;
2428
2429         p = xdr_reserve_space(xdr, slot->sl_datalen);
2430         if (!p) {
2431                 WARN_ON_ONCE(1);
2432                 return nfserr_serverfault;
2433         }
2434         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2435         xdr_commit_encode(xdr);
2436
2437         resp->opcnt = slot->sl_opcnt;
2438         return slot->sl_status;
2439 }
2440
2441 /*
2442  * Set the exchange_id flags returned by the server.
2443  */
2444 static void
2445 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2446 {
2447 #ifdef CONFIG_NFSD_PNFS
2448         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2449 #else
2450         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2451 #endif
2452
2453         /* Referrals are supported, Migration is not. */
2454         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2455
2456         /* set the wire flags to return to client. */
2457         clid->flags = new->cl_exchange_flags;
2458 }
2459
2460 static bool client_has_openowners(struct nfs4_client *clp)
2461 {
2462         struct nfs4_openowner *oo;
2463
2464         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2465                 if (!list_empty(&oo->oo_owner.so_stateids))
2466                         return true;
2467         }
2468         return false;
2469 }
2470
2471 static bool client_has_state(struct nfs4_client *clp)
2472 {
2473         return client_has_openowners(clp)
2474 #ifdef CONFIG_NFSD_PNFS
2475                 || !list_empty(&clp->cl_lo_states)
2476 #endif
2477                 || !list_empty(&clp->cl_delegations)
2478                 || !list_empty(&clp->cl_sessions);
2479 }
2480
2481 __be32
2482 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2483                 union nfsd4_op_u *u)
2484 {
2485         struct nfsd4_exchange_id *exid = &u->exchange_id;
2486         struct nfs4_client *conf, *new;
2487         struct nfs4_client *unconf = NULL;
2488         __be32 status;
2489         char                    addr_str[INET6_ADDRSTRLEN];
2490         nfs4_verifier           verf = exid->verifier;
2491         struct sockaddr         *sa = svc_addr(rqstp);
2492         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2493         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2494
2495         rpc_ntop(sa, addr_str, sizeof(addr_str));
2496         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2497                 "ip_addr=%s flags %x, spa_how %d\n",
2498                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2499                 addr_str, exid->flags, exid->spa_how);
2500
2501         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2502                 return nfserr_inval;
2503
2504         new = create_client(exid->clname, rqstp, &verf);
2505         if (new == NULL)
2506                 return nfserr_jukebox;
2507
2508         switch (exid->spa_how) {
2509         case SP4_MACH_CRED:
2510                 exid->spo_must_enforce[0] = 0;
2511                 exid->spo_must_enforce[1] = (
2512                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2513                         1 << (OP_EXCHANGE_ID - 32) |
2514                         1 << (OP_CREATE_SESSION - 32) |
2515                         1 << (OP_DESTROY_SESSION - 32) |
2516                         1 << (OP_DESTROY_CLIENTID - 32));
2517
2518                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2519                                         1 << (OP_OPEN_DOWNGRADE) |
2520                                         1 << (OP_LOCKU) |
2521                                         1 << (OP_DELEGRETURN));
2522
2523                 exid->spo_must_allow[1] &= (
2524                                         1 << (OP_TEST_STATEID - 32) |
2525                                         1 << (OP_FREE_STATEID - 32));
2526                 if (!svc_rqst_integrity_protected(rqstp)) {
2527                         status = nfserr_inval;
2528                         goto out_nolock;
2529                 }
2530                 /*
2531                  * Sometimes userspace doesn't give us a principal.
2532                  * Which is a bug, really.  Anyway, we can't enforce
2533                  * MACH_CRED in that case, better to give up now:
2534                  */
2535                 if (!new->cl_cred.cr_principal &&
2536                                         !new->cl_cred.cr_raw_principal) {
2537                         status = nfserr_serverfault;
2538                         goto out_nolock;
2539                 }
2540                 new->cl_mach_cred = true;
2541         case SP4_NONE:
2542                 break;
2543         default:                                /* checked by xdr code */
2544                 WARN_ON_ONCE(1);
2545         case SP4_SSV:
2546                 status = nfserr_encr_alg_unsupp;
2547                 goto out_nolock;
2548         }
2549
2550         /* Cases below refer to rfc 5661 section 18.35.4: */
2551         spin_lock(&nn->client_lock);
2552         conf = find_confirmed_client_by_name(&exid->clname, nn);
2553         if (conf) {
2554                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2555                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2556
2557                 if (update) {
2558                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2559                                 status = nfserr_inval;
2560                                 goto out;
2561                         }
2562                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2563                                 status = nfserr_wrong_cred;
2564                                 goto out;
2565                         }
2566                         if (!creds_match) { /* case 9 */
2567                                 status = nfserr_perm;
2568                                 goto out;
2569                         }
2570                         if (!verfs_match) { /* case 8 */
2571                                 status = nfserr_not_same;
2572                                 goto out;
2573                         }
2574                         /* case 6 */
2575                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2576                         goto out_copy;
2577                 }
2578                 if (!creds_match) { /* case 3 */
2579                         if (client_has_state(conf)) {
2580                                 status = nfserr_clid_inuse;
2581                                 goto out;
2582                         }
2583                         goto out_new;
2584                 }
2585                 if (verfs_match) { /* case 2 */
2586                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2587                         goto out_copy;
2588                 }
2589                 /* case 5, client reboot */
2590                 conf = NULL;
2591                 goto out_new;
2592         }
2593
2594         if (update) { /* case 7 */
2595                 status = nfserr_noent;
2596                 goto out;
2597         }
2598
2599         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2600         if (unconf) /* case 4, possible retry or client restart */
2601                 unhash_client_locked(unconf);
2602
2603         /* case 1 (normal case) */
2604 out_new:
2605         if (conf) {
2606                 status = mark_client_expired_locked(conf);
2607                 if (status)
2608                         goto out;
2609         }
2610         new->cl_minorversion = cstate->minorversion;
2611         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2612         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2613
2614         gen_clid(new, nn);
2615         add_to_unconfirmed(new);
2616         swap(new, conf);
2617 out_copy:
2618         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2619         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2620
2621         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2622         nfsd4_set_ex_flags(conf, exid);
2623
2624         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2625                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2626         status = nfs_ok;
2627
2628 out:
2629         spin_unlock(&nn->client_lock);
2630 out_nolock:
2631         if (new)
2632                 expire_client(new);
2633         if (unconf)
2634                 expire_client(unconf);
2635         return status;
2636 }
2637
2638 static __be32
2639 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2640 {
2641         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2642                 slot_seqid);
2643
2644         /* The slot is in use, and no response has been sent. */
2645         if (slot_inuse) {
2646                 if (seqid == slot_seqid)
2647                         return nfserr_jukebox;
2648                 else
2649                         return nfserr_seq_misordered;
2650         }
2651         /* Note unsigned 32-bit arithmetic handles wraparound: */
2652         if (likely(seqid == slot_seqid + 1))
2653                 return nfs_ok;
2654         if (seqid == slot_seqid)
2655                 return nfserr_replay_cache;
2656         return nfserr_seq_misordered;
2657 }
2658
2659 /*
2660  * Cache the create session result into the create session single DRC
2661  * slot cache by saving the xdr structure. sl_seqid has been set.
2662  * Do this for solo or embedded create session operations.
2663  */
2664 static void
2665 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2666                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2667 {
2668         slot->sl_status = nfserr;
2669         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2670 }
2671
2672 static __be32
2673 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2674                             struct nfsd4_clid_slot *slot)
2675 {
2676         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2677         return slot->sl_status;
2678 }
2679
2680 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2681                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2682                         1 +     /* MIN tag is length with zero, only length */ \
2683                         3 +     /* version, opcount, opcode */ \
2684                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2685                                 /* seqid, slotID, slotID, cache */ \
2686                         4 ) * sizeof(__be32))
2687
2688 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2689                         2 +     /* verifier: AUTH_NULL, length 0 */\
2690                         1 +     /* status */ \
2691                         1 +     /* MIN tag is length with zero, only length */ \
2692                         3 +     /* opcount, opcode, opstatus*/ \
2693                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2694                                 /* seqid, slotID, slotID, slotID, status */ \
2695                         5 ) * sizeof(__be32))
2696
2697 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2698 {
2699         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2700
2701         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2702                 return nfserr_toosmall;
2703         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2704                 return nfserr_toosmall;
2705         ca->headerpadsz = 0;
2706         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2707         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2708         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2709         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2710                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2711         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2712         /*
2713          * Note decreasing slot size below client's request may make it
2714          * difficult for client to function correctly, whereas
2715          * decreasing the number of slots will (just?) affect
2716          * performance.  When short on memory we therefore prefer to
2717          * decrease number of slots instead of their size.  Clients that
2718          * request larger slots than they need will get poor results:
2719          */
2720         ca->maxreqs = nfsd4_get_drc_mem(ca);
2721         if (!ca->maxreqs)
2722                 return nfserr_jukebox;
2723
2724         return nfs_ok;
2725 }
2726
2727 /*
2728  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2729  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2730  */
2731 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2732         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2733
2734 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2735         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2736
2737 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2738                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2739 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2740                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2741                                  sizeof(__be32))
2742
2743 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2744 {
2745         ca->headerpadsz = 0;
2746
2747         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2748                 return nfserr_toosmall;
2749         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2750                 return nfserr_toosmall;
2751         ca->maxresp_cached = 0;
2752         if (ca->maxops < 2)
2753                 return nfserr_toosmall;
2754
2755         return nfs_ok;
2756 }
2757
2758 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2759 {
2760         switch (cbs->flavor) {
2761         case RPC_AUTH_NULL:
2762         case RPC_AUTH_UNIX:
2763                 return nfs_ok;
2764         default:
2765                 /*
2766                  * GSS case: the spec doesn't allow us to return this
2767                  * error.  But it also doesn't allow us not to support
2768                  * GSS.
2769                  * I'd rather this fail hard than return some error the
2770                  * client might think it can already handle:
2771                  */
2772                 return nfserr_encr_alg_unsupp;
2773         }
2774 }
2775
2776 __be32
2777 nfsd4_create_session(struct svc_rqst *rqstp,
2778                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2779 {
2780         struct nfsd4_create_session *cr_ses = &u->create_session;
2781         struct sockaddr *sa = svc_addr(rqstp);
2782         struct nfs4_client *conf, *unconf;
2783         struct nfs4_client *old = NULL;
2784         struct nfsd4_session *new;
2785         struct nfsd4_conn *conn;
2786         struct nfsd4_clid_slot *cs_slot = NULL;
2787         __be32 status = 0;
2788         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2789
2790         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2791                 return nfserr_inval;
2792         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2793         if (status)
2794                 return status;
2795         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2796         if (status)
2797                 return status;
2798         status = check_backchannel_attrs(&cr_ses->back_channel);
2799         if (status)
2800                 goto out_release_drc_mem;
2801         status = nfserr_jukebox;
2802         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2803         if (!new)
2804                 goto out_release_drc_mem;
2805         conn = alloc_conn_from_crses(rqstp, cr_ses);
2806         if (!conn)
2807                 goto out_free_session;
2808
2809         spin_lock(&nn->client_lock);
2810         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2811         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2812         WARN_ON_ONCE(conf && unconf);
2813
2814         if (conf) {
2815                 status = nfserr_wrong_cred;
2816                 if (!nfsd4_mach_creds_match(conf, rqstp))
2817                         goto out_free_conn;
2818                 cs_slot = &conf->cl_cs_slot;
2819                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2820                 if (status) {
2821                         if (status == nfserr_replay_cache)
2822                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2823                         goto out_free_conn;
2824                 }
2825         } else if (unconf) {
2826                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2827                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2828                         status = nfserr_clid_inuse;
2829                         goto out_free_conn;
2830                 }
2831                 status = nfserr_wrong_cred;
2832                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2833                         goto out_free_conn;
2834                 cs_slot = &unconf->cl_cs_slot;
2835                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2836                 if (status) {
2837                         /* an unconfirmed replay returns misordered */
2838                         status = nfserr_seq_misordered;
2839                         goto out_free_conn;
2840                 }
2841                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2842                 if (old) {
2843                         status = mark_client_expired_locked(old);
2844                         if (status) {
2845                                 old = NULL;
2846                                 goto out_free_conn;
2847                         }
2848                 }
2849                 move_to_confirmed(unconf);
2850                 conf = unconf;
2851         } else {
2852                 status = nfserr_stale_clientid;
2853                 goto out_free_conn;
2854         }
2855         status = nfs_ok;
2856         /* Persistent sessions are not supported */
2857         cr_ses->flags &= ~SESSION4_PERSIST;
2858         /* Upshifting from TCP to RDMA is not supported */
2859         cr_ses->flags &= ~SESSION4_RDMA;
2860
2861         init_session(rqstp, new, conf, cr_ses);
2862         nfsd4_get_session_locked(new);
2863
2864         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2865                NFS4_MAX_SESSIONID_LEN);
2866         cs_slot->sl_seqid++;
2867         cr_ses->seqid = cs_slot->sl_seqid;
2868
2869         /* cache solo and embedded create sessions under the client_lock */
2870         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2871         spin_unlock(&nn->client_lock);
2872         /* init connection and backchannel */
2873         nfsd4_init_conn(rqstp, conn, new);
2874         nfsd4_put_session(new);
2875         if (old)
2876                 expire_client(old);
2877         return status;
2878 out_free_conn:
2879         spin_unlock(&nn->client_lock);
2880         free_conn(conn);
2881         if (old)
2882                 expire_client(old);
2883 out_free_session:
2884         __free_session(new);
2885 out_release_drc_mem:
2886         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2887         return status;
2888 }
2889
2890 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2891 {
2892         switch (*dir) {
2893         case NFS4_CDFC4_FORE:
2894         case NFS4_CDFC4_BACK:
2895                 return nfs_ok;
2896         case NFS4_CDFC4_FORE_OR_BOTH:
2897         case NFS4_CDFC4_BACK_OR_BOTH:
2898                 *dir = NFS4_CDFC4_BOTH;
2899                 return nfs_ok;
2900         };
2901         return nfserr_inval;
2902 }
2903
2904 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2905                 struct nfsd4_compound_state *cstate,
2906                 union nfsd4_op_u *u)
2907 {
2908         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2909         struct nfsd4_session *session = cstate->session;
2910         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2911         __be32 status;
2912
2913         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2914         if (status)
2915                 return status;
2916         spin_lock(&nn->client_lock);
2917         session->se_cb_prog = bc->bc_cb_program;
2918         session->se_cb_sec = bc->bc_cb_sec;
2919         spin_unlock(&nn->client_lock);
2920
2921         nfsd4_probe_callback(session->se_client);
2922
2923         return nfs_ok;
2924 }
2925
2926 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2927                      struct nfsd4_compound_state *cstate,
2928                      union nfsd4_op_u *u)
2929 {
2930         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2931         __be32 status;
2932         struct nfsd4_conn *conn;
2933         struct nfsd4_session *session;
2934         struct net *net = SVC_NET(rqstp);
2935         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2936
2937         if (!nfsd4_last_compound_op(rqstp))
2938                 return nfserr_not_only_op;
2939         spin_lock(&nn->client_lock);
2940         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2941         spin_unlock(&nn->client_lock);
2942         if (!session)
2943                 goto out_no_session;
2944         status = nfserr_wrong_cred;
2945         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2946                 goto out;
2947         status = nfsd4_map_bcts_dir(&bcts->dir);
2948         if (status)
2949                 goto out;
2950         conn = alloc_conn(rqstp, bcts->dir);
2951         status = nfserr_jukebox;
2952         if (!conn)
2953                 goto out;
2954         nfsd4_init_conn(rqstp, conn, session);
2955         status = nfs_ok;
2956 out:
2957         nfsd4_put_session(session);
2958 out_no_session:
2959         return status;
2960 }
2961
2962 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
2963 {
2964         if (!cstate->session)
2965                 return false;
2966         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
2967 }
2968
2969 __be32
2970 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2971                 union nfsd4_op_u *u)
2972 {
2973         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
2974         struct nfsd4_session *ses;
2975         __be32 status;
2976         int ref_held_by_me = 0;
2977         struct net *net = SVC_NET(r);
2978         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2979
2980         status = nfserr_not_only_op;
2981         if (nfsd4_compound_in_session(cstate, sessionid)) {
2982                 if (!nfsd4_last_compound_op(r))
2983                         goto out;
2984                 ref_held_by_me++;
2985         }
2986         dump_sessionid(__func__, sessionid);
2987         spin_lock(&nn->client_lock);
2988         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
2989         if (!ses)
2990                 goto out_client_lock;
2991         status = nfserr_wrong_cred;
2992         if (!nfsd4_mach_creds_match(ses->se_client, r))
2993                 goto out_put_session;
2994         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2995         if (status)
2996                 goto out_put_session;
2997         unhash_session(ses);
2998         spin_unlock(&nn->client_lock);
2999
3000         nfsd4_probe_callback_sync(ses->se_client);
3001
3002         spin_lock(&nn->client_lock);
3003         status = nfs_ok;
3004 out_put_session:
3005         nfsd4_put_session_locked(ses);
3006 out_client_lock:
3007         spin_unlock(&nn->client_lock);
3008 out:
3009         return status;
3010 }
3011
3012 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3013 {
3014         struct nfsd4_conn *c;
3015
3016         list_for_each_entry(c, &s->se_conns, cn_persession) {
3017                 if (c->cn_xprt == xpt) {
3018                         return c;
3019                 }
3020         }
3021         return NULL;
3022 }
3023
3024 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3025 {
3026         struct nfs4_client *clp = ses->se_client;
3027         struct nfsd4_conn *c;
3028         __be32 status = nfs_ok;
3029         int ret;
3030
3031         spin_lock(&clp->cl_lock);
3032         c = __nfsd4_find_conn(new->cn_xprt, ses);
3033         if (c)
3034                 goto out_free;
3035         status = nfserr_conn_not_bound_to_session;
3036         if (clp->cl_mach_cred)
3037                 goto out_free;
3038         __nfsd4_hash_conn(new, ses);
3039         spin_unlock(&clp->cl_lock);
3040         ret = nfsd4_register_conn(new);
3041         if (ret)
3042                 /* oops; xprt is already down: */
3043                 nfsd4_conn_lost(&new->cn_xpt_user);
3044         return nfs_ok;
3045 out_free:
3046         spin_unlock(&clp->cl_lock);
3047         free_conn(new);
3048         return status;
3049 }
3050
3051 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3052 {
3053         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3054
3055         return args->opcnt > session->se_fchannel.maxops;
3056 }
3057
3058 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3059                                   struct nfsd4_session *session)
3060 {
3061         struct xdr_buf *xb = &rqstp->rq_arg;
3062
3063         return xb->len > session->se_fchannel.maxreq_sz;
3064 }
3065
3066 static bool replay_matches_cache(struct svc_rqst *rqstp,
3067                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3068 {
3069         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3070
3071         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3072             (bool)seq->cachethis)
3073                 return false;
3074         /*
3075          * If there's an error than the reply can have fewer ops than
3076          * the call.  But if we cached a reply with *more* ops than the
3077          * call you're sending us now, then this new call is clearly not
3078          * really a replay of the old one:
3079          */
3080         if (slot->sl_opcnt < argp->opcnt)
3081                 return false;
3082         /* This is the only check explicitly called by spec: */
3083         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3084                 return false;
3085         /*
3086          * There may be more comparisons we could actually do, but the
3087          * spec doesn't require us to catch every case where the calls
3088          * don't match (that would require caching the call as well as
3089          * the reply), so we don't bother.
3090          */
3091         return true;
3092 }
3093
3094 __be32
3095 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3096                 union nfsd4_op_u *u)
3097 {
3098         struct nfsd4_sequence *seq = &u->sequence;
3099         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3100         struct xdr_stream *xdr = &resp->xdr;
3101         struct nfsd4_session *session;
3102         struct nfs4_client *clp;
3103         struct nfsd4_slot *slot;
3104         struct nfsd4_conn *conn;
3105         __be32 status;
3106         int buflen;
3107         struct net *net = SVC_NET(rqstp);
3108         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3109
3110         if (resp->opcnt != 1)
3111                 return nfserr_sequence_pos;
3112
3113         /*
3114          * Will be either used or freed by nfsd4_sequence_check_conn
3115          * below.
3116          */
3117         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3118         if (!conn)
3119                 return nfserr_jukebox;
3120
3121         spin_lock(&nn->client_lock);
3122         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3123         if (!session)
3124                 goto out_no_session;
3125         clp = session->se_client;
3126
3127         status = nfserr_too_many_ops;
3128         if (nfsd4_session_too_many_ops(rqstp, session))
3129                 goto out_put_session;
3130
3131         status = nfserr_req_too_big;
3132         if (nfsd4_request_too_big(rqstp, session))
3133                 goto out_put_session;
3134
3135         status = nfserr_badslot;
3136         if (seq->slotid >= session->se_fchannel.maxreqs)
3137                 goto out_put_session;
3138
3139         slot = session->se_slots[seq->slotid];
3140         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3141
3142         /* We do not negotiate the number of slots yet, so set the
3143          * maxslots to the session maxreqs which is used to encode
3144          * sr_highest_slotid and the sr_target_slot id to maxslots */
3145         seq->maxslots = session->se_fchannel.maxreqs;
3146
3147         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3148                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3149         if (status == nfserr_replay_cache) {
3150                 status = nfserr_seq_misordered;
3151                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3152                         goto out_put_session;
3153                 status = nfserr_seq_false_retry;
3154                 if (!replay_matches_cache(rqstp, seq, slot))
3155                         goto out_put_session;
3156                 cstate->slot = slot;
3157                 cstate->session = session;
3158                 cstate->clp = clp;
3159                 /* Return the cached reply status and set cstate->status
3160                  * for nfsd4_proc_compound processing */
3161                 status = nfsd4_replay_cache_entry(resp, seq);
3162                 cstate->status = nfserr_replay_cache;
3163                 goto out;
3164         }
3165         if (status)
3166                 goto out_put_session;
3167
3168         status = nfsd4_sequence_check_conn(conn, session);
3169         conn = NULL;
3170         if (status)
3171                 goto out_put_session;
3172
3173         buflen = (seq->cachethis) ?
3174                         session->se_fchannel.maxresp_cached :
3175                         session->se_fchannel.maxresp_sz;
3176         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3177                                     nfserr_rep_too_big;
3178         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3179                 goto out_put_session;
3180         svc_reserve(rqstp, buflen);
3181
3182         status = nfs_ok;
3183         /* Success! bump slot seqid */
3184         slot->sl_seqid = seq->seqid;
3185         slot->sl_flags |= NFSD4_SLOT_INUSE;
3186         if (seq->cachethis)
3187                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3188         else
3189                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3190
3191         cstate->slot = slot;
3192         cstate->session = session;
3193         cstate->clp = clp;
3194
3195 out:
3196         switch (clp->cl_cb_state) {
3197         case NFSD4_CB_DOWN:
3198                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3199                 break;
3200         case NFSD4_CB_FAULT:
3201                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3202                 break;
3203         default:
3204                 seq->status_flags = 0;
3205         }
3206         if (!list_empty(&clp->cl_revoked))
3207                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3208 out_no_session:
3209         if (conn)
3210                 free_conn(conn);
3211         spin_unlock(&nn->client_lock);
3212         return status;
3213 out_put_session:
3214         nfsd4_put_session_locked(session);
3215         goto out_no_session;
3216 }
3217
3218 void
3219 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3220 {
3221         struct nfsd4_compound_state *cs = &resp->cstate;
3222
3223         if (nfsd4_has_session(cs)) {
3224                 if (cs->status != nfserr_replay_cache) {
3225                         nfsd4_store_cache_entry(resp);
3226                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3227                 }
3228                 /* Drop session reference that was taken in nfsd4_sequence() */
3229                 nfsd4_put_session(cs->session);
3230         } else if (cs->clp)
3231                 put_client_renew(cs->clp);
3232 }
3233
3234 __be32
3235 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3236                 struct nfsd4_compound_state *cstate,
3237                 union nfsd4_op_u *u)
3238 {
3239         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3240         struct nfs4_client *conf, *unconf;
3241         struct nfs4_client *clp = NULL;
3242         __be32 status = 0;
3243         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3244
3245         spin_lock(&nn->client_lock);
3246         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3247         conf = find_confirmed_client(&dc->clientid, true, nn);
3248         WARN_ON_ONCE(conf && unconf);
3249
3250         if (conf) {
3251                 if (client_has_state(conf)) {
3252                         status = nfserr_clientid_busy;
3253                         goto out;
3254                 }
3255                 status = mark_client_expired_locked(conf);
3256                 if (status)
3257                         goto out;
3258                 clp = conf;
3259         } else if (unconf)
3260                 clp = unconf;
3261         else {
3262                 status = nfserr_stale_clientid;
3263                 goto out;
3264         }
3265         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3266                 clp = NULL;
3267                 status = nfserr_wrong_cred;
3268                 goto out;
3269         }
3270         unhash_client_locked(clp);
3271 out:
3272         spin_unlock(&nn->client_lock);
3273         if (clp)
3274                 expire_client(clp);
3275         return status;
3276 }
3277
3278 __be32
3279 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3280                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3281 {
3282         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3283         __be32 status = 0;
3284
3285         if (rc->rca_one_fs) {
3286                 if (!cstate->current_fh.fh_dentry)
3287                         return nfserr_nofilehandle;
3288                 /*
3289                  * We don't take advantage of the rca_one_fs case.
3290                  * That's OK, it's optional, we can safely ignore it.
3291                  */
3292                 return nfs_ok;
3293         }
3294
3295         status = nfserr_complete_already;
3296         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3297                              &cstate->session->se_client->cl_flags))
3298                 goto out;
3299
3300         status = nfserr_stale_clientid;
3301         if (is_client_expired(cstate->session->se_client))
3302                 /*
3303                  * The following error isn't really legal.
3304                  * But we only get here if the client just explicitly
3305                  * destroyed the client.  Surely it no longer cares what
3306                  * error it gets back on an operation for the dead
3307                  * client.
3308                  */
3309                 goto out;
3310
3311         status = nfs_ok;
3312         nfsd4_client_record_create(cstate->session->se_client);
3313 out:
3314         return status;
3315 }
3316
3317 __be32
3318 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3319                   union nfsd4_op_u *u)
3320 {
3321         struct nfsd4_setclientid *setclid = &u->setclientid;
3322         struct xdr_netobj       clname = setclid->se_name;
3323         nfs4_verifier           clverifier = setclid->se_verf;
3324         struct nfs4_client      *conf, *new;
3325         struct nfs4_client      *unconf = NULL;
3326         __be32                  status;
3327         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3328
3329         new = create_client(clname, rqstp, &clverifier);
3330         if (new == NULL)
3331                 return nfserr_jukebox;
3332         /* Cases below refer to rfc 3530 section 14.2.33: */
3333         spin_lock(&nn->client_lock);
3334         conf = find_confirmed_client_by_name(&clname, nn);
3335         if (conf && client_has_state(conf)) {
3336                 /* case 0: */
3337                 status = nfserr_clid_inuse;
3338                 if (clp_used_exchangeid(conf))
3339                         goto out;
3340                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3341                         char addr_str[INET6_ADDRSTRLEN];
3342                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3343                                  sizeof(addr_str));
3344                         dprintk("NFSD: setclientid: string in use by client "
3345                                 "at %s\n", addr_str);
3346                         goto out;
3347                 }
3348         }
3349         unconf = find_unconfirmed_client_by_name(&clname, nn);
3350         if (unconf)
3351                 unhash_client_locked(unconf);
3352         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3353                 /* case 1: probable callback update */
3354                 copy_clid(new, conf);
3355                 gen_confirm(new, nn);
3356         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3357                 gen_clid(new, nn);
3358         new->cl_minorversion = 0;
3359         gen_callback(new, setclid, rqstp);
3360         add_to_unconfirmed(new);
3361         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3362         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3363         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3364         new = NULL;
3365         status = nfs_ok;
3366 out:
3367         spin_unlock(&nn->client_lock);
3368         if (new)
3369                 free_client(new);
3370         if (unconf)
3371                 expire_client(unconf);
3372         return status;
3373 }
3374
3375
3376 __be32
3377 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3378                         struct nfsd4_compound_state *cstate,
3379                         union nfsd4_op_u *u)
3380 {
3381         struct nfsd4_setclientid_confirm *setclientid_confirm =
3382                         &u->setclientid_confirm;
3383         struct nfs4_client *conf, *unconf;
3384         struct nfs4_client *old = NULL;
3385         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3386         clientid_t * clid = &setclientid_confirm->sc_clientid;
3387         __be32 status;
3388         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3389
3390         if (STALE_CLIENTID(clid, nn))
3391                 return nfserr_stale_clientid;
3392
3393         spin_lock(&nn->client_lock);
3394         conf = find_confirmed_client(clid, false, nn);
3395         unconf = find_unconfirmed_client(clid, false, nn);
3396         /*
3397          * We try hard to give out unique clientid's, so if we get an
3398          * attempt to confirm the same clientid with a different cred,
3399          * the client may be buggy; this should never happen.
3400          *
3401          * Nevertheless, RFC 7530 recommends INUSE for this case:
3402          */
3403         status = nfserr_clid_inuse;
3404         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3405                 goto out;
3406         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3407                 goto out;
3408         /* cases below refer to rfc 3530 section 14.2.34: */
3409         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3410                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3411                         /* case 2: probable retransmit */
3412                         status = nfs_ok;
3413                 } else /* case 4: client hasn't noticed we rebooted yet? */
3414                         status = nfserr_stale_clientid;
3415                 goto out;
3416         }
3417         status = nfs_ok;
3418         if (conf) { /* case 1: callback update */
3419                 old = unconf;
3420                 unhash_client_locked(old);
3421                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3422         } else { /* case 3: normal case; new or rebooted client */
3423                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3424                 if (old) {
3425                         status = nfserr_clid_inuse;
3426                         if (client_has_state(old)
3427                                         && !same_creds(&unconf->cl_cred,
3428                                                         &old->cl_cred))
3429                                 goto out;
3430                         status = mark_client_expired_locked(old);
3431                         if (status) {
3432                                 old = NULL;
3433                                 goto out;
3434                         }
3435                 }
3436                 move_to_confirmed(unconf);
3437                 conf = unconf;
3438         }
3439         get_client_locked(conf);
3440         spin_unlock(&nn->client_lock);
3441         nfsd4_probe_callback(conf);
3442         spin_lock(&nn->client_lock);
3443         put_client_renew_locked(conf);
3444 out:
3445         spin_unlock(&nn->client_lock);
3446         if (old)
3447                 expire_client(old);
3448         return status;
3449 }
3450
3451 static struct nfs4_file *nfsd4_alloc_file(void)
3452 {
3453         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3454 }
3455
3456 /* OPEN Share state helper functions */
3457 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3458                                 struct nfs4_file *fp)
3459 {
3460         lockdep_assert_held(&state_lock);
3461
3462         refcount_set(&fp->fi_ref, 1);
3463         spin_lock_init(&fp->fi_lock);
3464         INIT_LIST_HEAD(&fp->fi_stateids);
3465         INIT_LIST_HEAD(&fp->fi_delegations);
3466         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3467         fh_copy_shallow(&fp->fi_fhandle, fh);
3468         fp->fi_deleg_file = NULL;
3469         fp->fi_had_conflict = false;
3470         fp->fi_share_deny = 0;
3471         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3472         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3473 #ifdef CONFIG_NFSD_PNFS
3474         INIT_LIST_HEAD(&fp->fi_lo_states);
3475         atomic_set(&fp->fi_lo_recalls, 0);
3476 #endif
3477         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3478 }
3479
3480 void
3481 nfsd4_free_slabs(void)
3482 {
3483         kmem_cache_destroy(client_slab);
3484         kmem_cache_destroy(openowner_slab);
3485         kmem_cache_destroy(lockowner_slab);
3486         kmem_cache_destroy(file_slab);
3487         kmem_cache_destroy(stateid_slab);
3488         kmem_cache_destroy(deleg_slab);
3489         kmem_cache_destroy(odstate_slab);
3490 }
3491
3492 int
3493 nfsd4_init_slabs(void)
3494 {
3495         client_slab = kmem_cache_create("nfsd4_clients",
3496                         sizeof(struct nfs4_client), 0, 0, NULL);
3497         if (client_slab == NULL)
3498                 goto out;
3499         openowner_slab = kmem_cache_create("nfsd4_openowners",
3500                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3501         if (openowner_slab == NULL)
3502                 goto out_free_client_slab;
3503         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3504                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3505         if (lockowner_slab == NULL)
3506                 goto out_free_openowner_slab;
3507         file_slab = kmem_cache_create("nfsd4_files",
3508                         sizeof(struct nfs4_file), 0, 0, NULL);
3509         if (file_slab == NULL)
3510                 goto out_free_lockowner_slab;
3511         stateid_slab = kmem_cache_create("nfsd4_stateids",
3512                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3513         if (stateid_slab == NULL)
3514                 goto out_free_file_slab;
3515         deleg_slab = kmem_cache_create("nfsd4_delegations",
3516                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3517         if (deleg_slab == NULL)
3518                 goto out_free_stateid_slab;
3519         odstate_slab = kmem_cache_create("nfsd4_odstate",
3520                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3521         if (odstate_slab == NULL)
3522                 goto out_free_deleg_slab;
3523         return 0;
3524
3525 out_free_deleg_slab:
3526         kmem_cache_destroy(deleg_slab);
3527 out_free_stateid_slab:
3528         kmem_cache_destroy(stateid_slab);
3529 out_free_file_slab:
3530         kmem_cache_destroy(file_slab);
3531 out_free_lockowner_slab:
3532         kmem_cache_destroy(lockowner_slab);
3533 out_free_openowner_slab:
3534         kmem_cache_destroy(openowner_slab);
3535 out_free_client_slab:
3536         kmem_cache_destroy(client_slab);
3537 out:
3538         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3539         return -ENOMEM;
3540 }
3541
3542 static void init_nfs4_replay(struct nfs4_replay *rp)
3543 {
3544         rp->rp_status = nfserr_serverfault;
3545         rp->rp_buflen = 0;
3546         rp->rp_buf = rp->rp_ibuf;
3547         mutex_init(&rp->rp_mutex);
3548 }
3549
3550 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3551                 struct nfs4_stateowner *so)
3552 {
3553         if (!nfsd4_has_session(cstate)) {
3554                 mutex_lock(&so->so_replay.rp_mutex);
3555                 cstate->replay_owner = nfs4_get_stateowner(so);
3556         }
3557 }
3558
3559 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3560 {
3561         struct nfs4_stateowner *so = cstate->replay_owner;
3562
3563         if (so != NULL) {
3564                 cstate->replay_owner = NULL;
3565                 mutex_unlock(&so->so_replay.rp_mutex);
3566                 nfs4_put_stateowner(so);
3567         }
3568 }
3569
3570 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3571 {
3572         struct nfs4_stateowner *sop;
3573
3574         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3575         if (!sop)
3576                 return NULL;
3577
3578         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3579         if (!sop->so_owner.data) {
3580                 kmem_cache_free(slab, sop);
3581                 return NULL;
3582         }
3583         sop->so_owner.len = owner->len;
3584
3585         INIT_LIST_HEAD(&sop->so_stateids);
3586         sop->so_client = clp;
3587         init_nfs4_replay(&sop->so_replay);
3588         atomic_set(&sop->so_count, 1);
3589         return sop;
3590 }
3591
3592 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3593 {
3594         lockdep_assert_held(&clp->cl_lock);
3595
3596         list_add(&oo->oo_owner.so_strhash,
3597                  &clp->cl_ownerstr_hashtbl[strhashval]);
3598         list_add(&oo->oo_perclient, &clp->cl_openowners);
3599 }
3600
3601 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3602 {
3603         unhash_openowner_locked(openowner(so));
3604 }
3605
3606 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3607 {
3608         struct nfs4_openowner *oo = openowner(so);
3609
3610         kmem_cache_free(openowner_slab, oo);
3611 }
3612
3613 static const struct nfs4_stateowner_operations openowner_ops = {
3614         .so_unhash =    nfs4_unhash_openowner,
3615         .so_free =      nfs4_free_openowner,
3616 };
3617
3618 static struct nfs4_ol_stateid *
3619 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3620 {
3621         struct nfs4_ol_stateid *local, *ret = NULL;
3622         struct nfs4_openowner *oo = open->op_openowner;
3623
3624         lockdep_assert_held(&fp->fi_lock);
3625
3626         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3627                 /* ignore lock owners */
3628                 if (local->st_stateowner->so_is_open_owner == 0)
3629                         continue;
3630                 if (local->st_stateowner != &oo->oo_owner)
3631                         continue;
3632                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3633                         ret = local;
3634                         refcount_inc(&ret->st_stid.sc_count);
3635                         break;
3636                 }
3637         }
3638         return ret;
3639 }
3640
3641 static __be32
3642 nfsd4_verify_open_stid(struct nfs4_stid *s)
3643 {
3644         __be32 ret = nfs_ok;
3645
3646         switch (s->sc_type) {
3647         default:
3648                 break;
3649         case 0:
3650         case NFS4_CLOSED_STID:
3651         case NFS4_CLOSED_DELEG_STID:
3652                 ret = nfserr_bad_stateid;
3653                 break;
3654         case NFS4_REVOKED_DELEG_STID:
3655                 ret = nfserr_deleg_revoked;
3656         }
3657         return ret;
3658 }
3659
3660 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3661 static __be32
3662 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3663 {
3664         __be32 ret;
3665
3666         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3667         ret = nfsd4_verify_open_stid(&stp->st_stid);
3668         if (ret != nfs_ok)
3669                 mutex_unlock(&stp->st_mutex);
3670         return ret;
3671 }
3672
3673 static struct nfs4_ol_stateid *
3674 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3675 {
3676         struct nfs4_ol_stateid *stp;
3677         for (;;) {
3678                 spin_lock(&fp->fi_lock);
3679                 stp = nfsd4_find_existing_open(fp, open);
3680                 spin_unlock(&fp->fi_lock);
3681                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3682                         break;
3683                 nfs4_put_stid(&stp->st_stid);
3684         }
3685         return stp;
3686 }
3687
3688 static struct nfs4_openowner *
3689 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3690                            struct nfsd4_compound_state *cstate)
3691 {
3692         struct nfs4_client *clp = cstate->clp;
3693         struct nfs4_openowner *oo, *ret;
3694
3695         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3696         if (!oo)
3697                 return NULL;
3698         oo->oo_owner.so_ops = &openowner_ops;
3699         oo->oo_owner.so_is_open_owner = 1;
3700         oo->oo_owner.so_seqid = open->op_seqid;
3701         oo->oo_flags = 0;
3702         if (nfsd4_has_session(cstate))
3703                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3704         oo->oo_time = 0;
3705         oo->oo_last_closed_stid = NULL;
3706         INIT_LIST_HEAD(&oo->oo_close_lru);
3707         spin_lock(&clp->cl_lock);
3708         ret = find_openstateowner_str_locked(strhashval, open, clp);
3709         if (ret == NULL) {
3710                 hash_openowner(oo, clp, strhashval);
3711                 ret = oo;
3712         } else
3713                 nfs4_free_stateowner(&oo->oo_owner);
3714
3715         spin_unlock(&clp->cl_lock);
3716         return ret;
3717 }
3718
3719 static struct nfs4_ol_stateid *
3720 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3721 {
3722
3723         struct nfs4_openowner *oo = open->op_openowner;
3724         struct nfs4_ol_stateid *retstp = NULL;
3725         struct nfs4_ol_stateid *stp;
3726
3727         stp = open->op_stp;
3728         /* We are moving these outside of the spinlocks to avoid the warnings */
3729         mutex_init(&stp->st_mutex);
3730         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3731
3732 retry:
3733         spin_lock(&oo->oo_owner.so_client->cl_lock);
3734         spin_lock(&fp->fi_lock);
3735
3736         retstp = nfsd4_find_existing_open(fp, open);
3737         if (retstp)
3738                 goto out_unlock;
3739
3740         open->op_stp = NULL;
3741         refcount_inc(&stp->st_stid.sc_count);
3742         stp->st_stid.sc_type = NFS4_OPEN_STID;
3743         INIT_LIST_HEAD(&stp->st_locks);
3744         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3745         get_nfs4_file(fp);
3746         stp->st_stid.sc_file = fp;
3747         stp->st_access_bmap = 0;
3748         stp->st_deny_bmap = 0;
3749         stp->st_openstp = NULL;
3750         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3751         list_add(&stp->st_perfile, &fp->fi_stateids);
3752
3753 out_unlock:
3754         spin_unlock(&fp->fi_lock);
3755         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3756         if (retstp) {
3757                 /* Handle races with CLOSE */
3758                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3759                         nfs4_put_stid(&retstp->st_stid);
3760                         goto retry;
3761                 }
3762                 /* To keep mutex tracking happy */
3763                 mutex_unlock(&stp->st_mutex);
3764                 stp = retstp;
3765         }
3766         return stp;
3767 }
3768
3769 /*
3770  * In the 4.0 case we need to keep the owners around a little while to handle
3771  * CLOSE replay. We still do need to release any file access that is held by
3772  * them before returning however.
3773  */
3774 static void
3775 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3776 {
3777         struct nfs4_ol_stateid *last;
3778         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3779         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3780                                                 nfsd_net_id);
3781
3782         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3783
3784         /*
3785          * We know that we hold one reference via nfsd4_close, and another
3786          * "persistent" reference for the client. If the refcount is higher
3787          * than 2, then there are still calls in progress that are using this
3788          * stateid. We can't put the sc_file reference until they are finished.
3789          * Wait for the refcount to drop to 2. Since it has been unhashed,
3790          * there should be no danger of the refcount going back up again at
3791          * this point.
3792          */
3793         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3794
3795         release_all_access(s);
3796         if (s->st_stid.sc_file) {
3797                 put_nfs4_file(s->st_stid.sc_file);
3798                 s->st_stid.sc_file = NULL;
3799         }
3800
3801         spin_lock(&nn->client_lock);
3802         last = oo->oo_last_closed_stid;
3803         oo->oo_last_closed_stid = s;
3804         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3805         oo->oo_time = get_seconds();
3806         spin_unlock(&nn->client_lock);
3807         if (last)
3808                 nfs4_put_stid(&last->st_stid);
3809 }
3810
3811 /* search file_hashtbl[] for file */
3812 static struct nfs4_file *
3813 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3814 {
3815         struct nfs4_file *fp;
3816
3817         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3818                 if (fh_match(&fp->fi_fhandle, fh)) {
3819                         if (refcount_inc_not_zero(&fp->fi_ref))
3820                                 return fp;
3821                 }
3822         }
3823         return NULL;
3824 }
3825
3826 struct nfs4_file *
3827 find_file(struct knfsd_fh *fh)
3828 {
3829         struct nfs4_file *fp;
3830         unsigned int hashval = file_hashval(fh);
3831
3832         rcu_read_lock();
3833         fp = find_file_locked(fh, hashval);
3834         rcu_read_unlock();
3835         return fp;
3836 }
3837
3838 static struct nfs4_file *
3839 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3840 {
3841         struct nfs4_file *fp;
3842         unsigned int hashval = file_hashval(fh);
3843
3844         rcu_read_lock();
3845         fp = find_file_locked(fh, hashval);
3846         rcu_read_unlock();
3847         if (fp)
3848                 return fp;
3849
3850         spin_lock(&state_lock);
3851         fp = find_file_locked(fh, hashval);
3852         if (likely(fp == NULL)) {
3853                 nfsd4_init_file(fh, hashval, new);
3854                 fp = new;
3855         }
3856         spin_unlock(&state_lock);
3857
3858         return fp;
3859 }
3860
3861 /*
3862  * Called to check deny when READ with all zero stateid or
3863  * WRITE with all zero or all one stateid
3864  */
3865 static __be32
3866 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3867 {
3868         struct nfs4_file *fp;
3869         __be32 ret = nfs_ok;
3870
3871         fp = find_file(&current_fh->fh_handle);
3872         if (!fp)
3873                 return ret;
3874         /* Check for conflicting share reservations */
3875         spin_lock(&fp->fi_lock);
3876         if (fp->fi_share_deny & deny_type)
3877                 ret = nfserr_locked;
3878         spin_unlock(&fp->fi_lock);
3879         put_nfs4_file(fp);
3880         return ret;
3881 }
3882
3883 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3884 {
3885         struct nfs4_delegation *dp = cb_to_delegation(cb);
3886         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3887                                           nfsd_net_id);
3888
3889         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3890
3891         /*
3892          * We can't do this in nfsd_break_deleg_cb because it is
3893          * already holding inode->i_lock.
3894          *
3895          * If the dl_time != 0, then we know that it has already been
3896          * queued for a lease break. Don't queue it again.
3897          */
3898         spin_lock(&state_lock);
3899         if (dp->dl_time == 0) {
3900                 dp->dl_time = get_seconds();
3901                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3902         }
3903         spin_unlock(&state_lock);
3904 }
3905
3906 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3907                 struct rpc_task *task)
3908 {
3909         struct nfs4_delegation *dp = cb_to_delegation(cb);
3910
3911         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3912                 return 1;
3913
3914         switch (task->tk_status) {
3915         case 0:
3916                 return 1;
3917         case -EBADHANDLE:
3918         case -NFS4ERR_BAD_STATEID:
3919                 /*
3920                  * Race: client probably got cb_recall before open reply
3921                  * granting delegation.
3922                  */
3923                 if (dp->dl_retries--) {
3924                         rpc_delay(task, 2 * HZ);
3925                         return 0;
3926                 }
3927                 /*FALLTHRU*/
3928         default:
3929                 return -1;
3930         }
3931 }
3932
3933 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3934 {
3935         struct nfs4_delegation *dp = cb_to_delegation(cb);
3936
3937         nfs4_put_stid(&dp->dl_stid);
3938 }
3939
3940 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3941         .prepare        = nfsd4_cb_recall_prepare,
3942         .done           = nfsd4_cb_recall_done,
3943         .release        = nfsd4_cb_recall_release,
3944 };
3945
3946 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3947 {
3948         /*
3949          * We're assuming the state code never drops its reference
3950          * without first removing the lease.  Since we're in this lease
3951          * callback (and since the lease code is serialized by the
3952          * i_lock) we know the server hasn't removed the lease yet, and
3953          * we know it's safe to take a reference.
3954          */
3955         refcount_inc(&dp->dl_stid.sc_count);
3956         nfsd4_run_cb(&dp->dl_recall);
3957 }
3958
3959 /* Called from break_lease() with i_lock held. */
3960 static bool
3961 nfsd_break_deleg_cb(struct file_lock *fl)
3962 {
3963         bool ret = false;
3964         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
3965         struct nfs4_file *fp = dp->dl_stid.sc_file;
3966
3967         /*
3968          * We don't want the locks code to timeout the lease for us;
3969          * we'll remove it ourself if a delegation isn't returned
3970          * in time:
3971          */
3972         fl->fl_break_time = 0;
3973
3974         spin_lock(&fp->fi_lock);
3975         fp->fi_had_conflict = true;
3976         nfsd_break_one_deleg(dp);
3977         spin_unlock(&fp->fi_lock);
3978         return ret;
3979 }
3980
3981 static int
3982 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3983                      struct list_head *dispose)
3984 {
3985         if (arg & F_UNLCK)
3986                 return lease_modify(onlist, arg, dispose);
3987         else
3988                 return -EAGAIN;
3989 }
3990
3991 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3992         .lm_break = nfsd_break_deleg_cb,
3993         .lm_change = nfsd_change_deleg_cb,
3994 };
3995
3996 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3997 {
3998         if (nfsd4_has_session(cstate))
3999                 return nfs_ok;
4000         if (seqid == so->so_seqid - 1)
4001                 return nfserr_replay_me;
4002         if (seqid == so->so_seqid)
4003                 return nfs_ok;
4004         return nfserr_bad_seqid;
4005 }
4006
4007 static __be32 lookup_clientid(clientid_t *clid,
4008                 struct nfsd4_compound_state *cstate,
4009                 struct nfsd_net *nn)
4010 {
4011         struct nfs4_client *found;
4012
4013         if (cstate->clp) {
4014                 found = cstate->clp;
4015                 if (!same_clid(&found->cl_clientid, clid))
4016                         return nfserr_stale_clientid;
4017                 return nfs_ok;
4018         }
4019
4020         if (STALE_CLIENTID(clid, nn))
4021                 return nfserr_stale_clientid;
4022
4023         /*
4024          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4025          * cached already then we know this is for is for v4.0 and "sessions"
4026          * will be false.
4027          */
4028         WARN_ON_ONCE(cstate->session);
4029         spin_lock(&nn->client_lock);
4030         found = find_confirmed_client(clid, false, nn);
4031         if (!found) {
4032                 spin_unlock(&nn->client_lock);
4033                 return nfserr_expired;
4034         }
4035         atomic_inc(&found->cl_refcount);
4036         spin_unlock(&nn->client_lock);
4037
4038         /* Cache the nfs4_client in cstate! */
4039         cstate->clp = found;
4040         return nfs_ok;
4041 }
4042
4043 __be32
4044 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4045                     struct nfsd4_open *open, struct nfsd_net *nn)
4046 {
4047         clientid_t *clientid = &open->op_clientid;
4048         struct nfs4_client *clp = NULL;
4049         unsigned int strhashval;
4050         struct nfs4_openowner *oo = NULL;
4051         __be32 status;
4052
4053         if (STALE_CLIENTID(&open->op_clientid, nn))
4054                 return nfserr_stale_clientid;
4055         /*
4056          * In case we need it later, after we've already created the
4057          * file and don't want to risk a further failure:
4058          */
4059         open->op_file = nfsd4_alloc_file();
4060         if (open->op_file == NULL)
4061                 return nfserr_jukebox;
4062
4063         status = lookup_clientid(clientid, cstate, nn);
4064         if (status)
4065                 return status;
4066         clp = cstate->clp;
4067
4068         strhashval = ownerstr_hashval(&open->op_owner);
4069         oo = find_openstateowner_str(strhashval, open, clp);
4070         open->op_openowner = oo;
4071         if (!oo) {
4072                 goto new_owner;
4073         }
4074         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4075                 /* Replace unconfirmed owners without checking for replay. */
4076                 release_openowner(oo);
4077                 open->op_openowner = NULL;
4078                 goto new_owner;
4079         }
4080         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4081         if (status)
4082                 return status;
4083         goto alloc_stateid;
4084 new_owner:
4085         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4086         if (oo == NULL)
4087                 return nfserr_jukebox;
4088         open->op_openowner = oo;
4089 alloc_stateid:
4090         open->op_stp = nfs4_alloc_open_stateid(clp);
4091         if (!open->op_stp)
4092                 return nfserr_jukebox;
4093
4094         if (nfsd4_has_session(cstate) &&
4095             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4096                 open->op_odstate = alloc_clnt_odstate(clp);
4097                 if (!open->op_odstate)
4098                         return nfserr_jukebox;
4099         }
4100
4101         return nfs_ok;
4102 }
4103
4104 static inline __be32
4105 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4106 {
4107         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4108                 return nfserr_openmode;
4109         else
4110                 return nfs_ok;
4111 }
4112
4113 static int share_access_to_flags(u32 share_access)
4114 {
4115         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4116 }
4117
4118 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4119 {
4120         struct nfs4_stid *ret;
4121
4122         ret = find_stateid_by_type(cl, s,
4123                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4124         if (!ret)
4125                 return NULL;
4126         return delegstateid(ret);
4127 }
4128
4129 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4130 {
4131         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4132                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4133 }
4134
4135 static __be32
4136 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4137                 struct nfs4_delegation **dp)
4138 {
4139         int flags;
4140         __be32 status = nfserr_bad_stateid;
4141         struct nfs4_delegation *deleg;
4142
4143         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4144         if (deleg == NULL)
4145                 goto out;
4146         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4147                 nfs4_put_stid(&deleg->dl_stid);
4148                 if (cl->cl_minorversion)
4149                         status = nfserr_deleg_revoked;
4150                 goto out;
4151         }
4152         flags = share_access_to_flags(open->op_share_access);
4153         status = nfs4_check_delegmode(deleg, flags);
4154         if (status) {
4155                 nfs4_put_stid(&deleg->dl_stid);
4156                 goto out;
4157         }
4158         *dp = deleg;
4159 out:
4160         if (!nfsd4_is_deleg_cur(open))
4161                 return nfs_ok;
4162         if (status)
4163                 return status;
4164         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4165         return nfs_ok;
4166 }
4167
4168 static inline int nfs4_access_to_access(u32 nfs4_access)
4169 {
4170         int flags = 0;
4171
4172         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4173                 flags |= NFSD_MAY_READ;
4174         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4175                 flags |= NFSD_MAY_WRITE;
4176         return flags;
4177 }
4178
4179 static inline __be32
4180 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4181                 struct nfsd4_open *open)
4182 {
4183         struct iattr iattr = {
4184                 .ia_valid = ATTR_SIZE,
4185                 .ia_size = 0,
4186         };
4187         if (!open->op_truncate)
4188                 return 0;
4189         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4190                 return nfserr_inval;
4191         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4192 }
4193
4194 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4195                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4196                 struct nfsd4_open *open)
4197 {
4198         struct file *filp = NULL;
4199         __be32 status;
4200         int oflag = nfs4_access_to_omode(open->op_share_access);
4201         int access = nfs4_access_to_access(open->op_share_access);
4202         unsigned char old_access_bmap, old_deny_bmap;
4203
4204         spin_lock(&fp->fi_lock);
4205
4206         /*
4207          * Are we trying to set a deny mode that would conflict with
4208          * current access?
4209          */
4210         status = nfs4_file_check_deny(fp, open->op_share_deny);
4211         if (status != nfs_ok) {
4212                 spin_unlock(&fp->fi_lock);
4213                 goto out;
4214         }
4215
4216         /* set access to the file */
4217         status = nfs4_file_get_access(fp, open->op_share_access);
4218         if (status != nfs_ok) {
4219                 spin_unlock(&fp->fi_lock);
4220                 goto out;
4221         }
4222
4223         /* Set access bits in stateid */
4224         old_access_bmap = stp->st_access_bmap;
4225         set_access(open->op_share_access, stp);
4226
4227         /* Set new deny mask */
4228         old_deny_bmap = stp->st_deny_bmap;
4229         set_deny(open->op_share_deny, stp);
4230         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4231
4232         if (!fp->fi_fds[oflag]) {
4233                 spin_unlock(&fp->fi_lock);
4234                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4235                 if (status)
4236                         goto out_put_access;
4237                 spin_lock(&fp->fi_lock);
4238                 if (!fp->fi_fds[oflag]) {
4239                         fp->fi_fds[oflag] = filp;
4240                         filp = NULL;
4241                 }
4242         }
4243         spin_unlock(&fp->fi_lock);
4244         if (filp)
4245                 fput(filp);
4246
4247         status = nfsd4_truncate(rqstp, cur_fh, open);
4248         if (status)
4249                 goto out_put_access;
4250 out:
4251         return status;
4252 out_put_access:
4253         stp->st_access_bmap = old_access_bmap;
4254         nfs4_file_put_access(fp, open->op_share_access);
4255         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4256         goto out;
4257 }
4258
4259 static __be32
4260 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)
4261 {
4262         __be32 status;
4263         unsigned char old_deny_bmap = stp->st_deny_bmap;
4264
4265         if (!test_access(open->op_share_access, stp))
4266                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4267
4268         /* test and set deny mode */
4269         spin_lock(&fp->fi_lock);
4270         status = nfs4_file_check_deny(fp, open->op_share_deny);
4271         if (status == nfs_ok) {
4272                 set_deny(open->op_share_deny, stp);
4273                 fp->fi_share_deny |=
4274                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4275         }
4276         spin_unlock(&fp->fi_lock);
4277
4278         if (status != nfs_ok)
4279                 return status;
4280
4281         status = nfsd4_truncate(rqstp, cur_fh, open);
4282         if (status != nfs_ok)
4283                 reset_union_bmap_deny(old_deny_bmap, stp);
4284         return status;
4285 }
4286
4287 /* Should we give out recallable state?: */
4288 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4289 {
4290         if (clp->cl_cb_state == NFSD4_CB_UP)
4291                 return true;
4292         /*
4293          * In the sessions case, since we don't have to establish a
4294          * separate connection for callbacks, we assume it's OK
4295          * until we hear otherwise:
4296          */
4297         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4298 }
4299
4300 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4301                                                 int flag)
4302 {
4303         struct file_lock *fl;
4304
4305         fl = locks_alloc_lock();
4306         if (!fl)
4307                 return NULL;
4308         fl->fl_lmops = &nfsd_lease_mng_ops;
4309         fl->fl_flags = FL_DELEG;
4310         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4311         fl->fl_end = OFFSET_MAX;
4312         fl->fl_owner = (fl_owner_t)dp;
4313         fl->fl_pid = current->tgid;
4314         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4315         return fl;
4316 }
4317
4318 static struct nfs4_delegation *
4319 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4320                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4321 {
4322         int status = 0;
4323         struct nfs4_delegation *dp;
4324         struct file *filp;
4325         struct file_lock *fl;
4326
4327         /*
4328          * The fi_had_conflict and nfs_get_existing_delegation checks
4329          * here are just optimizations; we'll need to recheck them at
4330          * the end:
4331          */
4332         if (fp->fi_had_conflict)
4333                 return ERR_PTR(-EAGAIN);
4334
4335         filp = find_readable_file(fp);
4336         if (!filp) {
4337                 /* We should always have a readable file here */
4338                 WARN_ON_ONCE(1);
4339                 return ERR_PTR(-EBADF);
4340         }
4341         spin_lock(&state_lock);
4342         spin_lock(&fp->fi_lock);
4343         if (nfs4_delegation_exists(clp, fp))
4344                 status = -EAGAIN;
4345         else if (!fp->fi_deleg_file) {
4346                 fp->fi_deleg_file = filp;
4347                 /* increment early to prevent fi_deleg_file from being
4348                  * cleared */
4349                 fp->fi_delegees = 1;
4350                 filp = NULL;
4351         } else
4352                 fp->fi_delegees++;
4353         spin_unlock(&fp->fi_lock);
4354         spin_unlock(&state_lock);
4355         if (filp)
4356                 fput(filp);
4357         if (status)
4358                 return ERR_PTR(status);
4359
4360         status = -ENOMEM;
4361         dp = alloc_init_deleg(clp, fp, fh, odstate);
4362         if (!dp)
4363                 goto out_delegees;
4364
4365         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4366         if (!fl)
4367                 goto out_stid;
4368
4369         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4370         if (fl)
4371                 locks_free_lock(fl);
4372         if (status)
4373                 goto out_clnt_odstate;
4374
4375         spin_lock(&state_lock);
4376         spin_lock(&fp->fi_lock);
4377         if (fp->fi_had_conflict)
4378                 status = -EAGAIN;
4379         else
4380                 status = hash_delegation_locked(dp, fp);
4381         spin_unlock(&fp->fi_lock);
4382         spin_unlock(&state_lock);
4383
4384         if (status)
4385                 goto out_unlock;
4386
4387         return dp;
4388 out_unlock:
4389         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4390 out_clnt_odstate:
4391         put_clnt_odstate(dp->dl_clnt_odstate);
4392 out_stid:
4393         nfs4_put_stid(&dp->dl_stid);
4394 out_delegees:
4395         put_deleg_file(fp);
4396         return ERR_PTR(status);
4397 }
4398
4399 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4400 {
4401         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4402         if (status == -EAGAIN)
4403                 open->op_why_no_deleg = WND4_CONTENTION;
4404         else {
4405                 open->op_why_no_deleg = WND4_RESOURCE;
4406                 switch (open->op_deleg_want) {
4407                 case NFS4_SHARE_WANT_READ_DELEG:
4408                 case NFS4_SHARE_WANT_WRITE_DELEG:
4409                 case NFS4_SHARE_WANT_ANY_DELEG:
4410                         break;
4411                 case NFS4_SHARE_WANT_CANCEL:
4412                         open->op_why_no_deleg = WND4_CANCELLED;
4413                         break;
4414                 case NFS4_SHARE_WANT_NO_DELEG:
4415                         WARN_ON_ONCE(1);
4416                 }
4417         }
4418 }
4419
4420 /*
4421  * Attempt to hand out a delegation.
4422  *
4423  * Note we don't support write delegations, and won't until the vfs has
4424  * proper support for them.
4425  */
4426 static void
4427 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4428                         struct nfs4_ol_stateid *stp)
4429 {
4430         struct nfs4_delegation *dp;
4431         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4432         struct nfs4_client *clp = stp->st_stid.sc_client;
4433         int cb_up;
4434         int status = 0;
4435
4436         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4437         open->op_recall = 0;
4438         switch (open->op_claim_type) {
4439                 case NFS4_OPEN_CLAIM_PREVIOUS:
4440                         if (!cb_up)
4441                                 open->op_recall = 1;
4442                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4443                                 goto out_no_deleg;
4444                         break;
4445                 case NFS4_OPEN_CLAIM_NULL:
4446                 case NFS4_OPEN_CLAIM_FH:
4447                         /*
4448                          * Let's not give out any delegations till everyone's
4449                          * had the chance to reclaim theirs, *and* until
4450                          * NLM locks have all been reclaimed:
4451                          */
4452                         if (locks_in_grace(clp->net))
4453                                 goto out_no_deleg;
4454                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4455                                 goto out_no_deleg;
4456                         /*
4457                          * Also, if the file was opened for write or
4458                          * create, there's a good chance the client's
4459                          * about to write to it, resulting in an
4460                          * immediate recall (since we don't support
4461                          * write delegations):
4462                          */
4463                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4464                                 goto out_no_deleg;
4465                         if (open->op_create == NFS4_OPEN_CREATE)
4466                                 goto out_no_deleg;
4467                         break;
4468                 default:
4469                         goto out_no_deleg;
4470         }
4471         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4472         if (IS_ERR(dp))
4473                 goto out_no_deleg;
4474
4475         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4476
4477         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4478                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4479         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4480         nfs4_put_stid(&dp->dl_stid);
4481         return;
4482 out_no_deleg:
4483         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4484         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4485             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4486                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4487                 open->op_recall = 1;
4488         }
4489
4490         /* 4.1 client asking for a delegation? */
4491         if (open->op_deleg_want)
4492                 nfsd4_open_deleg_none_ext(open, status);
4493         return;
4494 }
4495
4496 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4497                                         struct nfs4_delegation *dp)
4498 {
4499         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4500             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4501                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4502                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4503         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4504                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4505                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4506                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4507         }
4508         /* Otherwise the client must be confused wanting a delegation
4509          * it already has, therefore we don't return
4510          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4511          */
4512 }
4513
4514 __be32
4515 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4516 {
4517         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4518         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4519         struct nfs4_file *fp = NULL;
4520         struct nfs4_ol_stateid *stp = NULL;
4521         struct nfs4_delegation *dp = NULL;
4522         __be32 status;
4523         bool new_stp = false;
4524
4525         /*
4526          * Lookup file; if found, lookup stateid and check open request,
4527          * and check for delegations in the process of being recalled.
4528          * If not found, create the nfs4_file struct
4529          */
4530         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4531         if (fp != open->op_file) {
4532                 status = nfs4_check_deleg(cl, open, &dp);
4533                 if (status)
4534                         goto out;
4535                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4536         } else {
4537                 open->op_file = NULL;
4538                 status = nfserr_bad_stateid;
4539                 if (nfsd4_is_deleg_cur(open))
4540                         goto out;
4541         }
4542
4543         if (!stp) {
4544                 stp = init_open_stateid(fp, open);
4545                 if (!open->op_stp)
4546                         new_stp = true;
4547         }
4548
4549         /*
4550          * OPEN the file, or upgrade an existing OPEN.
4551          * If truncate fails, the OPEN fails.
4552          *
4553          * stp is already locked.
4554          */
4555         if (!new_stp) {
4556                 /* Stateid was found, this is an OPEN upgrade */
4557                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4558                 if (status) {
4559                         mutex_unlock(&stp->st_mutex);
4560                         goto out;
4561                 }
4562         } else {
4563                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4564                 if (status) {
4565                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4566                         release_open_stateid(stp);
4567                         mutex_unlock(&stp->st_mutex);
4568                         goto out;
4569                 }
4570
4571                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4572                                                         open->op_odstate);
4573                 if (stp->st_clnt_odstate == open->op_odstate)
4574                         open->op_odstate = NULL;
4575         }
4576
4577         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4578         mutex_unlock(&stp->st_mutex);
4579
4580         if (nfsd4_has_session(&resp->cstate)) {
4581                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4582                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4583                         open->op_why_no_deleg = WND4_NOT_WANTED;
4584                         goto nodeleg;
4585                 }
4586         }
4587
4588         /*
4589         * Attempt to hand out a delegation. No error return, because the
4590         * OPEN succeeds even if we fail.
4591         */
4592         nfs4_open_delegation(current_fh, open, stp);
4593 nodeleg:
4594         status = nfs_ok;
4595
4596         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4597                 STATEID_VAL(&stp->st_stid.sc_stateid));
4598 out:
4599         /* 4.1 client trying to upgrade/downgrade delegation? */
4600         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4601             open->op_deleg_want)
4602                 nfsd4_deleg_xgrade_none_ext(open, dp);
4603
4604         if (fp)
4605                 put_nfs4_file(fp);
4606         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4607                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4608         /*
4609         * To finish the open response, we just need to set the rflags.
4610         */
4611         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4612         if (nfsd4_has_session(&resp->cstate))
4613                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4614         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4615                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4616
4617         if (dp)
4618                 nfs4_put_stid(&dp->dl_stid);
4619         if (stp)
4620                 nfs4_put_stid(&stp->st_stid);
4621
4622         return status;
4623 }
4624
4625 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4626                               struct nfsd4_open *open)
4627 {
4628         if (open->op_openowner) {
4629                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4630
4631                 nfsd4_cstate_assign_replay(cstate, so);
4632                 nfs4_put_stateowner(so);
4633         }
4634         if (open->op_file)
4635                 kmem_cache_free(file_slab, open->op_file);
4636         if (open->op_stp)
4637                 nfs4_put_stid(&open->op_stp->st_stid);
4638         if (open->op_odstate)
4639                 kmem_cache_free(odstate_slab, open->op_odstate);
4640 }
4641
4642 __be32
4643 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4644             union nfsd4_op_u *u)
4645 {
4646         clientid_t *clid = &u->renew;
4647         struct nfs4_client *clp;
4648         __be32 status;
4649         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4650
4651         dprintk("process_renew(%08x/%08x): starting\n",
4652                         clid->cl_boot, clid->cl_id);
4653         status = lookup_clientid(clid, cstate, nn);
4654         if (status)
4655                 goto out;
4656         clp = cstate->clp;
4657         status = nfserr_cb_path_down;
4658         if (!list_empty(&clp->cl_delegations)
4659                         && clp->cl_cb_state != NFSD4_CB_UP)
4660                 goto out;
4661         status = nfs_ok;
4662 out:
4663         return status;
4664 }
4665
4666 void
4667 nfsd4_end_grace(struct nfsd_net *nn)
4668 {
4669         /* do nothing if grace period already ended */
4670         if (nn->grace_ended)
4671                 return;
4672
4673         dprintk("NFSD: end of grace period\n");
4674         nn->grace_ended = true;
4675         /*
4676          * If the server goes down again right now, an NFSv4
4677          * client will still be allowed to reclaim after it comes back up,
4678          * even if it hasn't yet had a chance to reclaim state this time.
4679          *
4680          */
4681         nfsd4_record_grace_done(nn);
4682         /*
4683          * At this point, NFSv4 clients can still reclaim.  But if the
4684          * server crashes, any that have not yet reclaimed will be out
4685          * of luck on the next boot.
4686          *
4687          * (NFSv4.1+ clients are considered to have reclaimed once they
4688          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4689          * have reclaimed after their first OPEN.)
4690          */
4691         locks_end_grace(&nn->nfsd4_manager);
4692         /*
4693          * At this point, and once lockd and/or any other containers
4694          * exit their grace period, further reclaims will fail and
4695          * regular locking can resume.
4696          */
4697 }
4698
4699 /*
4700  * If we've waited a lease period but there are still clients trying to
4701  * reclaim, wait a little longer to give them a chance to finish.
4702  */
4703 static bool clients_still_reclaiming(struct nfsd_net *nn)
4704 {
4705         unsigned long now = get_seconds();
4706         unsigned long double_grace_period_end = nn->boot_time +
4707                                                 2 * nn->nfsd4_lease;
4708
4709         if (!nn->somebody_reclaimed)
4710                 return false;
4711         nn->somebody_reclaimed = false;
4712         /*
4713          * If we've given them *two* lease times to reclaim, and they're
4714          * still not done, give up:
4715          */
4716         if (time_after(now, double_grace_period_end))
4717                 return false;
4718         return true;
4719 }
4720
4721 static time_t
4722 nfs4_laundromat(struct nfsd_net *nn)
4723 {
4724         struct nfs4_client *clp;
4725         struct nfs4_openowner *oo;
4726         struct nfs4_delegation *dp;
4727         struct nfs4_ol_stateid *stp;
4728         struct nfsd4_blocked_lock *nbl;
4729         struct list_head *pos, *next, reaplist;
4730         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4731         time_t t, new_timeo = nn->nfsd4_lease;
4732
4733         dprintk("NFSD: laundromat service - starting\n");
4734
4735         if (clients_still_reclaiming(nn)) {
4736                 new_timeo = 0;
4737                 goto out;
4738         }
4739         nfsd4_end_grace(nn);
4740         INIT_LIST_HEAD(&reaplist);
4741         spin_lock(&nn->client_lock);
4742         list_for_each_safe(pos, next, &nn->client_lru) {
4743                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4744                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4745                         t = clp->cl_time - cutoff;
4746                         new_timeo = min(new_timeo, t);
4747                         break;
4748                 }
4749                 if (mark_client_expired_locked(clp)) {
4750                         dprintk("NFSD: client in use (clientid %08x)\n",
4751                                 clp->cl_clientid.cl_id);
4752                         continue;
4753                 }
4754                 list_add(&clp->cl_lru, &reaplist);
4755         }
4756         spin_unlock(&nn->client_lock);
4757         list_for_each_safe(pos, next, &reaplist) {
4758                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4759                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4760                         clp->cl_clientid.cl_id);
4761                 list_del_init(&clp->cl_lru);
4762                 expire_client(clp);
4763         }
4764         spin_lock(&state_lock);
4765         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4766                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4767                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4768                         t = dp->dl_time - cutoff;
4769                         new_timeo = min(new_timeo, t);
4770                         break;
4771                 }
4772                 WARN_ON(!unhash_delegation_locked(dp));
4773                 list_add(&dp->dl_recall_lru, &reaplist);
4774         }
4775         spin_unlock(&state_lock);
4776         while (!list_empty(&reaplist)) {
4777                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4778                                         dl_recall_lru);
4779                 list_del_init(&dp->dl_recall_lru);
4780                 revoke_delegation(dp);
4781         }
4782
4783         spin_lock(&nn->client_lock);
4784         while (!list_empty(&nn->close_lru)) {
4785                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4786                                         oo_close_lru);
4787                 if (time_after((unsigned long)oo->oo_time,
4788                                (unsigned long)cutoff)) {
4789                         t = oo->oo_time - cutoff;
4790                         new_timeo = min(new_timeo, t);
4791                         break;
4792                 }
4793                 list_del_init(&oo->oo_close_lru);
4794                 stp = oo->oo_last_closed_stid;
4795                 oo->oo_last_closed_stid = NULL;
4796                 spin_unlock(&nn->client_lock);
4797                 nfs4_put_stid(&stp->st_stid);
4798                 spin_lock(&nn->client_lock);
4799         }
4800         spin_unlock(&nn->client_lock);
4801
4802         /*
4803          * It's possible for a client to try and acquire an already held lock
4804          * that is being held for a long time, and then lose interest in it.
4805          * So, we clean out any un-revisited request after a lease period
4806          * under the assumption that the client is no longer interested.
4807          *
4808          * RFC5661, sec. 9.6 states that the client must not rely on getting
4809          * notifications and must continue to poll for locks, even when the
4810          * server supports them. Thus this shouldn't lead to clients blocking
4811          * indefinitely once the lock does become free.
4812          */
4813         BUG_ON(!list_empty(&reaplist));
4814         spin_lock(&nn->blocked_locks_lock);
4815         while (!list_empty(&nn->blocked_locks_lru)) {
4816                 nbl = list_first_entry(&nn->blocked_locks_lru,
4817                                         struct nfsd4_blocked_lock, nbl_lru);
4818                 if (time_after((unsigned long)nbl->nbl_time,
4819                                (unsigned long)cutoff)) {
4820                         t = nbl->nbl_time - cutoff;
4821                         new_timeo = min(new_timeo, t);
4822                         break;
4823                 }
4824                 list_move(&nbl->nbl_lru, &reaplist);
4825                 list_del_init(&nbl->nbl_list);
4826         }
4827         spin_unlock(&nn->blocked_locks_lock);
4828
4829         while (!list_empty(&reaplist)) {
4830                 nbl = list_first_entry(&reaplist,
4831                                         struct nfsd4_blocked_lock, nbl_lru);
4832                 list_del_init(&nbl->nbl_lru);
4833                 posix_unblock_lock(&nbl->nbl_lock);
4834                 free_blocked_lock(nbl);
4835         }
4836 out:
4837         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4838         return new_timeo;
4839 }
4840
4841 static struct workqueue_struct *laundry_wq;
4842 static void laundromat_main(struct work_struct *);
4843
4844 static void
4845 laundromat_main(struct work_struct *laundry)
4846 {
4847         time_t t;
4848         struct delayed_work *dwork = to_delayed_work(laundry);
4849         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4850                                            laundromat_work);
4851
4852         t = nfs4_laundromat(nn);
4853         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4854         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4855 }
4856
4857 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4858 {
4859         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4860                 return nfserr_bad_stateid;
4861         return nfs_ok;
4862 }
4863
4864 static inline int
4865 access_permit_read(struct nfs4_ol_stateid *stp)
4866 {
4867         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4868                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4869                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4870 }
4871
4872 static inline int
4873 access_permit_write(struct nfs4_ol_stateid *stp)
4874 {
4875         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4876                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4877 }
4878
4879 static
4880 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4881 {
4882         __be32 status = nfserr_openmode;
4883
4884         /* For lock stateid's, we test the parent open, not the lock: */
4885         if (stp->st_openstp)
4886                 stp = stp->st_openstp;
4887         if ((flags & WR_STATE) && !access_permit_write(stp))
4888                 goto out;
4889         if ((flags & RD_STATE) && !access_permit_read(stp))
4890                 goto out;
4891         status = nfs_ok;
4892 out:
4893         return status;
4894 }
4895
4896 static inline __be32
4897 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4898 {
4899         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4900                 return nfs_ok;
4901         else if (opens_in_grace(net)) {
4902                 /* Answer in remaining cases depends on existence of
4903                  * conflicting state; so we must wait out the grace period. */
4904                 return nfserr_grace;
4905         } else if (flags & WR_STATE)
4906                 return nfs4_share_conflict(current_fh,
4907                                 NFS4_SHARE_DENY_WRITE);
4908         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4909                 return nfs4_share_conflict(current_fh,
4910                                 NFS4_SHARE_DENY_READ);
4911 }
4912
4913 /*
4914  * Allow READ/WRITE during grace period on recovered state only for files
4915  * that are not able to provide mandatory locking.
4916  */
4917 static inline int
4918 grace_disallows_io(struct net *net, struct inode *inode)
4919 {
4920         return opens_in_grace(net) && mandatory_lock(inode);
4921 }
4922
4923 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4924 {
4925         /*
4926          * When sessions are used the stateid generation number is ignored
4927          * when it is zero.
4928          */
4929         if (has_session && in->si_generation == 0)
4930                 return nfs_ok;
4931
4932         if (in->si_generation == ref->si_generation)
4933                 return nfs_ok;
4934
4935         /* If the client sends us a stateid from the future, it's buggy: */
4936         if (nfsd4_stateid_generation_after(in, ref))
4937                 return nfserr_bad_stateid;
4938         /*
4939          * However, we could see a stateid from the past, even from a
4940          * non-buggy client.  For example, if the client sends a lock
4941          * while some IO is outstanding, the lock may bump si_generation
4942          * while the IO is still in flight.  The client could avoid that
4943          * situation by waiting for responses on all the IO requests,
4944          * but better performance may result in retrying IO that
4945          * receives an old_stateid error if requests are rarely
4946          * reordered in flight:
4947          */
4948         return nfserr_old_stateid;
4949 }
4950
4951 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4952 {
4953         __be32 ret;
4954
4955         spin_lock(&s->sc_lock);
4956         ret = nfsd4_verify_open_stid(s);
4957         if (ret == nfs_ok)
4958                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4959         spin_unlock(&s->sc_lock);
4960         return ret;
4961 }
4962
4963 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4964 {
4965         if (ols->st_stateowner->so_is_open_owner &&
4966             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4967                 return nfserr_bad_stateid;
4968         return nfs_ok;
4969 }
4970
4971 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4972 {
4973         struct nfs4_stid *s;
4974         __be32 status = nfserr_bad_stateid;
4975
4976         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4977                 CLOSE_STATEID(stateid))
4978                 return status;
4979         /* Client debugging aid. */
4980         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4981                 char addr_str[INET6_ADDRSTRLEN];
4982                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4983                                  sizeof(addr_str));
4984                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4985                                         "with incorrect client ID\n", addr_str);
4986                 return status;
4987         }
4988         spin_lock(&cl->cl_lock);
4989         s = find_stateid_locked(cl, stateid);
4990         if (!s)
4991                 goto out_unlock;
4992         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
4993         if (status)
4994                 goto out_unlock;
4995         switch (s->sc_type) {
4996         case NFS4_DELEG_STID:
4997                 status = nfs_ok;
4998                 break;
4999         case NFS4_REVOKED_DELEG_STID:
5000                 status = nfserr_deleg_revoked;
5001                 break;
5002         case NFS4_OPEN_STID:
5003         case NFS4_LOCK_STID:
5004                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5005                 break;
5006         default:
5007                 printk("unknown stateid type %x\n", s->sc_type);
5008                 /* Fallthrough */
5009         case NFS4_CLOSED_STID:
5010         case NFS4_CLOSED_DELEG_STID:
5011                 status = nfserr_bad_stateid;
5012         }
5013 out_unlock:
5014         spin_unlock(&cl->cl_lock);
5015         return status;
5016 }
5017
5018 __be32
5019 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5020                      stateid_t *stateid, unsigned char typemask,
5021                      struct nfs4_stid **s, struct nfsd_net *nn)
5022 {
5023         __be32 status;
5024         bool return_revoked = false;
5025
5026         /*
5027          *  only return revoked delegations if explicitly asked.
5028          *  otherwise we report revoked or bad_stateid status.
5029          */
5030         if (typemask & NFS4_REVOKED_DELEG_STID)
5031                 return_revoked = true;
5032         else if (typemask & NFS4_DELEG_STID)
5033                 typemask |= NFS4_REVOKED_DELEG_STID;
5034
5035         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5036                 CLOSE_STATEID(stateid))
5037                 return nfserr_bad_stateid;
5038         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5039         if (status == nfserr_stale_clientid) {
5040                 if (cstate->session)
5041                         return nfserr_bad_stateid;
5042                 return nfserr_stale_stateid;
5043         }
5044         if (status)
5045                 return status;
5046         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5047         if (!*s)
5048                 return nfserr_bad_stateid;
5049         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5050                 nfs4_put_stid(*s);
5051                 if (cstate->minorversion)
5052                         return nfserr_deleg_revoked;
5053                 return nfserr_bad_stateid;
5054         }
5055         return nfs_ok;
5056 }
5057
5058 static struct file *
5059 nfs4_find_file(struct nfs4_stid *s, int flags)
5060 {
5061         if (!s)
5062                 return NULL;
5063
5064         switch (s->sc_type) {
5065         case NFS4_DELEG_STID:
5066                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5067                         return NULL;
5068                 return get_file(s->sc_file->fi_deleg_file);
5069         case NFS4_OPEN_STID:
5070         case NFS4_LOCK_STID:
5071                 if (flags & RD_STATE)
5072                         return find_readable_file(s->sc_file);
5073                 else
5074                         return find_writeable_file(s->sc_file);
5075                 break;
5076         }
5077
5078         return NULL;
5079 }
5080
5081 static __be32
5082 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5083 {
5084         __be32 status;
5085
5086         status = nfsd4_check_openowner_confirmed(ols);
5087         if (status)
5088                 return status;
5089         return nfs4_check_openmode(ols, flags);
5090 }
5091
5092 static __be32
5093 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5094                 struct file **filpp, bool *tmp_file, int flags)
5095 {
5096         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5097         struct file *file;
5098         __be32 status;
5099
5100         file = nfs4_find_file(s, flags);
5101         if (file) {
5102                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5103                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5104                 if (status) {
5105                         fput(file);
5106                         return status;
5107                 }
5108
5109                 *filpp = file;
5110         } else {
5111                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5112                 if (status)
5113                         return status;
5114
5115                 if (tmp_file)
5116                         *tmp_file = true;
5117         }
5118
5119         return 0;
5120 }
5121
5122 /*
5123  * Checks for stateid operations
5124  */
5125 __be32
5126 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5127                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5128                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5129 {
5130         struct inode *ino = d_inode(fhp->fh_dentry);
5131         struct net *net = SVC_NET(rqstp);
5132         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5133         struct nfs4_stid *s = NULL;
5134         __be32 status;
5135
5136         if (filpp)
5137                 *filpp = NULL;
5138         if (tmp_file)
5139                 *tmp_file = false;
5140
5141         if (grace_disallows_io(net, ino))
5142                 return nfserr_grace;
5143
5144         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5145                 status = check_special_stateids(net, fhp, stateid, flags);
5146                 goto done;
5147         }
5148
5149         status = nfsd4_lookup_stateid(cstate, stateid,
5150                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5151                                 &s, nn);
5152         if (status)
5153                 return status;
5154         status = nfsd4_stid_check_stateid_generation(stateid, s,
5155                         nfsd4_has_session(cstate));
5156         if (status)
5157                 goto out;
5158
5159         switch (s->sc_type) {
5160         case NFS4_DELEG_STID:
5161                 status = nfs4_check_delegmode(delegstateid(s), flags);
5162                 break;
5163         case NFS4_OPEN_STID:
5164         case NFS4_LOCK_STID:
5165                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5166                 break;
5167         default:
5168                 status = nfserr_bad_stateid;
5169                 break;
5170         }
5171         if (status)
5172                 goto out;
5173         status = nfs4_check_fh(fhp, s);
5174
5175 done:
5176         if (!status && filpp)
5177                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5178 out:
5179         if (s)
5180                 nfs4_put_stid(s);
5181         return status;
5182 }
5183
5184 /*
5185  * Test if the stateid is valid
5186  */
5187 __be32
5188 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5189                    union nfsd4_op_u *u)
5190 {
5191         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5192         struct nfsd4_test_stateid_id *stateid;
5193         struct nfs4_client *cl = cstate->session->se_client;
5194
5195         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5196                 stateid->ts_id_status =
5197                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5198
5199         return nfs_ok;
5200 }
5201
5202 static __be32
5203 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5204 {
5205         struct nfs4_ol_stateid *stp = openlockstateid(s);
5206         __be32 ret;
5207
5208         ret = nfsd4_lock_ol_stateid(stp);
5209         if (ret)
5210                 goto out_put_stid;
5211
5212         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5213         if (ret)
5214                 goto out;
5215
5216         ret = nfserr_locks_held;
5217         if (check_for_locks(stp->st_stid.sc_file,
5218                             lockowner(stp->st_stateowner)))
5219                 goto out;
5220
5221         release_lock_stateid(stp);
5222         ret = nfs_ok;
5223
5224 out:
5225         mutex_unlock(&stp->st_mutex);
5226 out_put_stid:
5227         nfs4_put_stid(s);
5228         return ret;
5229 }
5230
5231 __be32
5232 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5233                    union nfsd4_op_u *u)
5234 {
5235         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5236         stateid_t *stateid = &free_stateid->fr_stateid;
5237         struct nfs4_stid *s;
5238         struct nfs4_delegation *dp;
5239         struct nfs4_client *cl = cstate->session->se_client;
5240         __be32 ret = nfserr_bad_stateid;
5241
5242         spin_lock(&cl->cl_lock);
5243         s = find_stateid_locked(cl, stateid);
5244         if (!s)
5245                 goto out_unlock;
5246         spin_lock(&s->sc_lock);
5247         switch (s->sc_type) {
5248         case NFS4_DELEG_STID:
5249                 ret = nfserr_locks_held;
5250                 break;
5251         case NFS4_OPEN_STID:
5252                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5253                 if (ret)
5254                         break;
5255                 ret = nfserr_locks_held;
5256                 break;
5257         case NFS4_LOCK_STID:
5258                 spin_unlock(&s->sc_lock);
5259                 refcount_inc(&s->sc_count);
5260                 spin_unlock(&cl->cl_lock);
5261                 ret = nfsd4_free_lock_stateid(stateid, s);
5262                 goto out;
5263         case NFS4_REVOKED_DELEG_STID:
5264                 spin_unlock(&s->sc_lock);
5265                 dp = delegstateid(s);
5266                 list_del_init(&dp->dl_recall_lru);
5267                 spin_unlock(&cl->cl_lock);
5268                 nfs4_put_stid(s);
5269                 ret = nfs_ok;
5270                 goto out;
5271         /* Default falls through and returns nfserr_bad_stateid */
5272         }
5273         spin_unlock(&s->sc_lock);
5274 out_unlock:
5275         spin_unlock(&cl->cl_lock);
5276 out:
5277         return ret;
5278 }
5279
5280 static inline int
5281 setlkflg (int type)
5282 {
5283         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5284                 RD_STATE : WR_STATE;
5285 }
5286
5287 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5288 {
5289         struct svc_fh *current_fh = &cstate->current_fh;
5290         struct nfs4_stateowner *sop = stp->st_stateowner;
5291         __be32 status;
5292
5293         status = nfsd4_check_seqid(cstate, sop, seqid);
5294         if (status)
5295                 return status;
5296         status = nfsd4_lock_ol_stateid(stp);
5297         if (status != nfs_ok)
5298                 return status;
5299         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5300         if (status == nfs_ok)
5301                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5302         if (status != nfs_ok)
5303                 mutex_unlock(&stp->st_mutex);
5304         return status;
5305 }
5306
5307 /*
5308  * Checks for sequence id mutating operations.
5309  */
5310 static __be32
5311 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5312                          stateid_t *stateid, char typemask,
5313                          struct nfs4_ol_stateid **stpp,
5314                          struct nfsd_net *nn)
5315 {
5316         __be32 status;
5317         struct nfs4_stid *s;
5318         struct nfs4_ol_stateid *stp = NULL;
5319
5320         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5321                 seqid, STATEID_VAL(stateid));
5322
5323         *stpp = NULL;
5324         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5325         if (status)
5326                 return status;
5327         stp = openlockstateid(s);
5328         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5329
5330         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5331         if (!status)
5332                 *stpp = stp;
5333         else
5334                 nfs4_put_stid(&stp->st_stid);
5335         return status;
5336 }
5337
5338 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5339                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5340 {
5341         __be32 status;
5342         struct nfs4_openowner *oo;
5343         struct nfs4_ol_stateid *stp;
5344
5345         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5346                                                 NFS4_OPEN_STID, &stp, nn);
5347         if (status)
5348                 return status;
5349         oo = openowner(stp->st_stateowner);
5350         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5351                 mutex_unlock(&stp->st_mutex);
5352                 nfs4_put_stid(&stp->st_stid);
5353                 return nfserr_bad_stateid;
5354         }
5355         *stpp = stp;
5356         return nfs_ok;
5357 }
5358
5359 __be32
5360 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5361                    union nfsd4_op_u *u)
5362 {
5363         struct nfsd4_open_confirm *oc = &u->open_confirm;
5364         __be32 status;
5365         struct nfs4_openowner *oo;
5366         struct nfs4_ol_stateid *stp;
5367         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5368
5369         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5370                         cstate->current_fh.fh_dentry);
5371
5372         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5373         if (status)
5374                 return status;
5375
5376         status = nfs4_preprocess_seqid_op(cstate,
5377                                         oc->oc_seqid, &oc->oc_req_stateid,
5378                                         NFS4_OPEN_STID, &stp, nn);
5379         if (status)
5380                 goto out;
5381         oo = openowner(stp->st_stateowner);
5382         status = nfserr_bad_stateid;
5383         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5384                 mutex_unlock(&stp->st_mutex);
5385                 goto put_stateid;
5386         }
5387         oo->oo_flags |= NFS4_OO_CONFIRMED;
5388         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5389         mutex_unlock(&stp->st_mutex);
5390         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5391                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5392
5393         nfsd4_client_record_create(oo->oo_owner.so_client);
5394         status = nfs_ok;
5395 put_stateid:
5396         nfs4_put_stid(&stp->st_stid);
5397 out:
5398         nfsd4_bump_seqid(cstate, status);
5399         return status;
5400 }
5401
5402 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5403 {
5404         if (!test_access(access, stp))
5405                 return;
5406         nfs4_file_put_access(stp->st_stid.sc_file, access);
5407         clear_access(access, stp);
5408 }
5409
5410 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5411 {
5412         switch (to_access) {
5413         case NFS4_SHARE_ACCESS_READ:
5414                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5415                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5416                 break;
5417         case NFS4_SHARE_ACCESS_WRITE:
5418                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5419                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5420                 break;
5421         case NFS4_SHARE_ACCESS_BOTH:
5422                 break;
5423         default:
5424                 WARN_ON_ONCE(1);
5425         }
5426 }
5427
5428 __be32
5429 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5430                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5431 {
5432         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5433         __be32 status;
5434         struct nfs4_ol_stateid *stp;
5435         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5436
5437         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5438                         cstate->current_fh.fh_dentry);
5439
5440         /* We don't yet support WANT bits: */
5441         if (od->od_deleg_want)
5442                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5443                         od->od_deleg_want);
5444
5445         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5446                                         &od->od_stateid, &stp, nn);
5447         if (status)
5448                 goto out;
5449         status = nfserr_inval;
5450         if (!test_access(od->od_share_access, stp)) {
5451                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5452                         stp->st_access_bmap, od->od_share_access);
5453                 goto put_stateid;
5454         }
5455         if (!test_deny(od->od_share_deny, stp)) {
5456                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5457                         stp->st_deny_bmap, od->od_share_deny);
5458                 goto put_stateid;
5459         }
5460         nfs4_stateid_downgrade(stp, od->od_share_access);
5461         reset_union_bmap_deny(od->od_share_deny, stp);
5462         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5463         status = nfs_ok;
5464 put_stateid:
5465         mutex_unlock(&stp->st_mutex);
5466         nfs4_put_stid(&stp->st_stid);
5467 out:
5468         nfsd4_bump_seqid(cstate, status);
5469         return status;
5470 }
5471
5472 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5473 {
5474         struct nfs4_client *clp = s->st_stid.sc_client;
5475         bool unhashed;
5476         LIST_HEAD(reaplist);
5477
5478         spin_lock(&clp->cl_lock);
5479         unhashed = unhash_open_stateid(s, &reaplist);
5480
5481         if (clp->cl_minorversion) {
5482                 if (unhashed)
5483                         put_ol_stateid_locked(s, &reaplist);
5484                 spin_unlock(&clp->cl_lock);
5485                 free_ol_stateid_reaplist(&reaplist);
5486         } else {
5487                 spin_unlock(&clp->cl_lock);
5488                 free_ol_stateid_reaplist(&reaplist);
5489                 if (unhashed)
5490                         move_to_close_lru(s, clp->net);
5491         }
5492 }
5493
5494 /*
5495  * nfs4_unlock_state() called after encode
5496  */
5497 __be32
5498 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5499                 union nfsd4_op_u *u)
5500 {
5501         struct nfsd4_close *close = &u->close;
5502         __be32 status;
5503         struct nfs4_ol_stateid *stp;
5504         struct net *net = SVC_NET(rqstp);
5505         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5506
5507         dprintk("NFSD: nfsd4_close on file %pd\n",
5508                         cstate->current_fh.fh_dentry);
5509
5510         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5511                                         &close->cl_stateid,
5512                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5513                                         &stp, nn);
5514         nfsd4_bump_seqid(cstate, status);
5515         if (status)
5516                 goto out;
5517
5518         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5519
5520         /*
5521          * Technically we don't _really_ have to increment or copy it, since
5522          * it should just be gone after this operation and we clobber the
5523          * copied value below, but we continue to do so here just to ensure
5524          * that racing ops see that there was a state change.
5525          */
5526         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5527
5528         nfsd4_close_open_stateid(stp);
5529         mutex_unlock(&stp->st_mutex);
5530
5531         /* v4.1+ suggests that we send a special stateid in here, since the
5532          * clients should just ignore this anyway. Since this is not useful
5533          * for v4.0 clients either, we set it to the special close_stateid
5534          * universally.
5535          *
5536          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5537          */
5538         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5539
5540         /* put reference from nfs4_preprocess_seqid_op */
5541         nfs4_put_stid(&stp->st_stid);
5542 out:
5543         return status;
5544 }
5545
5546 __be32
5547 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5548                   union nfsd4_op_u *u)
5549 {
5550         struct nfsd4_delegreturn *dr = &u->delegreturn;
5551         struct nfs4_delegation *dp;
5552         stateid_t *stateid = &dr->dr_stateid;
5553         struct nfs4_stid *s;
5554         __be32 status;
5555         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5556
5557         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5558                 return status;
5559
5560         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5561         if (status)
5562                 goto out;
5563         dp = delegstateid(s);
5564         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5565         if (status)
5566                 goto put_stateid;
5567
5568         destroy_delegation(dp);
5569 put_stateid:
5570         nfs4_put_stid(&dp->dl_stid);
5571 out:
5572         return status;
5573 }
5574
5575 static inline u64
5576 end_offset(u64 start, u64 len)
5577 {
5578         u64 end;
5579
5580         end = start + len;
5581         return end >= start ? end: NFS4_MAX_UINT64;
5582 }
5583
5584 /* last octet in a range */
5585 static inline u64
5586 last_byte_offset(u64 start, u64 len)
5587 {
5588         u64 end;
5589
5590         WARN_ON_ONCE(!len);
5591         end = start + len;
5592         return end > start ? end - 1: NFS4_MAX_UINT64;
5593 }
5594
5595 /*
5596  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5597  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5598  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5599  * locking, this prevents us from being completely protocol-compliant.  The
5600  * real solution to this problem is to start using unsigned file offsets in
5601  * the VFS, but this is a very deep change!
5602  */
5603 static inline void
5604 nfs4_transform_lock_offset(struct file_lock *lock)
5605 {
5606         if (lock->fl_start < 0)
5607                 lock->fl_start = OFFSET_MAX;
5608         if (lock->fl_end < 0)
5609                 lock->fl_end = OFFSET_MAX;
5610 }
5611
5612 static fl_owner_t
5613 nfsd4_fl_get_owner(fl_owner_t owner)
5614 {
5615         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5616
5617         nfs4_get_stateowner(&lo->lo_owner);
5618         return owner;
5619 }
5620
5621 static void
5622 nfsd4_fl_put_owner(fl_owner_t owner)
5623 {
5624         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5625
5626         if (lo)
5627                 nfs4_put_stateowner(&lo->lo_owner);
5628 }
5629
5630 static void
5631 nfsd4_lm_notify(struct file_lock *fl)
5632 {
5633         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5634         struct net                      *net = lo->lo_owner.so_client->net;
5635         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5636         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5637                                                 struct nfsd4_blocked_lock, nbl_lock);
5638         bool queue = false;
5639
5640         /* An empty list means that something else is going to be using it */
5641         spin_lock(&nn->blocked_locks_lock);
5642         if (!list_empty(&nbl->nbl_list)) {
5643                 list_del_init(&nbl->nbl_list);
5644                 list_del_init(&nbl->nbl_lru);
5645                 queue = true;
5646         }
5647         spin_unlock(&nn->blocked_locks_lock);
5648
5649         if (queue)
5650                 nfsd4_run_cb(&nbl->nbl_cb);
5651 }
5652
5653 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5654         .lm_notify = nfsd4_lm_notify,
5655         .lm_get_owner = nfsd4_fl_get_owner,
5656         .lm_put_owner = nfsd4_fl_put_owner,
5657 };
5658
5659 static inline void
5660 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5661 {
5662         struct nfs4_lockowner *lo;
5663
5664         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5665                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5666                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5667                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5668                 if (!deny->ld_owner.data)
5669                         /* We just don't care that much */
5670                         goto nevermind;
5671                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5672                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5673         } else {
5674 nevermind:
5675                 deny->ld_owner.len = 0;
5676                 deny->ld_owner.data = NULL;
5677                 deny->ld_clientid.cl_boot = 0;
5678                 deny->ld_clientid.cl_id = 0;
5679         }
5680         deny->ld_start = fl->fl_start;
5681         deny->ld_length = NFS4_MAX_UINT64;
5682         if (fl->fl_end != NFS4_MAX_UINT64)
5683                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5684         deny->ld_type = NFS4_READ_LT;
5685         if (fl->fl_type != F_RDLCK)
5686                 deny->ld_type = NFS4_WRITE_LT;
5687 }
5688
5689 static struct nfs4_lockowner *
5690 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5691 {
5692         unsigned int strhashval = ownerstr_hashval(owner);
5693         struct nfs4_stateowner *so;
5694
5695         lockdep_assert_held(&clp->cl_lock);
5696
5697         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5698                             so_strhash) {
5699                 if (so->so_is_open_owner)
5700                         continue;
5701                 if (same_owner_str(so, owner))
5702                         return lockowner(nfs4_get_stateowner(so));
5703         }
5704         return NULL;
5705 }
5706
5707 static struct nfs4_lockowner *
5708 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5709 {
5710         struct nfs4_lockowner *lo;
5711
5712         spin_lock(&clp->cl_lock);
5713         lo = find_lockowner_str_locked(clp, owner);
5714         spin_unlock(&clp->cl_lock);
5715         return lo;
5716 }
5717
5718 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5719 {
5720         unhash_lockowner_locked(lockowner(sop));
5721 }
5722
5723 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5724 {
5725         struct nfs4_lockowner *lo = lockowner(sop);
5726
5727         kmem_cache_free(lockowner_slab, lo);
5728 }
5729
5730 static const struct nfs4_stateowner_operations lockowner_ops = {
5731         .so_unhash =    nfs4_unhash_lockowner,
5732         .so_free =      nfs4_free_lockowner,
5733 };
5734
5735 /*
5736  * Alloc a lock owner structure.
5737  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5738  * occurred.
5739  *
5740  * strhashval = ownerstr_hashval
5741  */
5742 static struct nfs4_lockowner *
5743 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5744                            struct nfs4_ol_stateid *open_stp,
5745                            struct nfsd4_lock *lock)
5746 {
5747         struct nfs4_lockowner *lo, *ret;
5748
5749         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5750         if (!lo)
5751                 return NULL;
5752         INIT_LIST_HEAD(&lo->lo_blocked);
5753         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5754         lo->lo_owner.so_is_open_owner = 0;
5755         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5756         lo->lo_owner.so_ops = &lockowner_ops;
5757         spin_lock(&clp->cl_lock);
5758         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5759         if (ret == NULL) {
5760                 list_add(&lo->lo_owner.so_strhash,
5761                          &clp->cl_ownerstr_hashtbl[strhashval]);
5762                 ret = lo;
5763         } else
5764                 nfs4_free_stateowner(&lo->lo_owner);
5765
5766         spin_unlock(&clp->cl_lock);
5767         return ret;
5768 }
5769
5770 static struct nfs4_ol_stateid *
5771 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5772 {
5773         struct nfs4_ol_stateid *lst;
5774         struct nfs4_client *clp = lo->lo_owner.so_client;
5775
5776         lockdep_assert_held(&clp->cl_lock);
5777
5778         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5779                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5780                         continue;
5781                 if (lst->st_stid.sc_file == fp) {
5782                         refcount_inc(&lst->st_stid.sc_count);
5783                         return lst;
5784                 }
5785         }
5786         return NULL;
5787 }
5788
5789 static struct nfs4_ol_stateid *
5790 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5791                   struct nfs4_file *fp, struct inode *inode,
5792                   struct nfs4_ol_stateid *open_stp)
5793 {
5794         struct nfs4_client *clp = lo->lo_owner.so_client;
5795         struct nfs4_ol_stateid *retstp;
5796
5797         mutex_init(&stp->st_mutex);
5798         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5799 retry:
5800         spin_lock(&clp->cl_lock);
5801         spin_lock(&fp->fi_lock);
5802         retstp = find_lock_stateid(lo, fp);
5803         if (retstp)
5804                 goto out_unlock;
5805
5806         refcount_inc(&stp->st_stid.sc_count);
5807         stp->st_stid.sc_type = NFS4_LOCK_STID;
5808         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5809         get_nfs4_file(fp);
5810         stp->st_stid.sc_file = fp;
5811         stp->st_access_bmap = 0;
5812         stp->st_deny_bmap = open_stp->st_deny_bmap;
5813         stp->st_openstp = open_stp;
5814         list_add(&stp->st_locks, &open_stp->st_locks);
5815         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5816         list_add(&stp->st_perfile, &fp->fi_stateids);
5817 out_unlock:
5818         spin_unlock(&fp->fi_lock);
5819         spin_unlock(&clp->cl_lock);
5820         if (retstp) {
5821                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5822                         nfs4_put_stid(&retstp->st_stid);
5823                         goto retry;
5824                 }
5825                 /* To keep mutex tracking happy */
5826                 mutex_unlock(&stp->st_mutex);
5827                 stp = retstp;
5828         }
5829         return stp;
5830 }
5831
5832 static struct nfs4_ol_stateid *
5833 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5834                             struct inode *inode, struct nfs4_ol_stateid *ost,
5835                             bool *new)
5836 {
5837         struct nfs4_stid *ns = NULL;
5838         struct nfs4_ol_stateid *lst;
5839         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5840         struct nfs4_client *clp = oo->oo_owner.so_client;
5841
5842         *new = false;
5843         spin_lock(&clp->cl_lock);
5844         lst = find_lock_stateid(lo, fi);
5845         spin_unlock(&clp->cl_lock);
5846         if (lst != NULL) {
5847                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5848                         goto out;
5849                 nfs4_put_stid(&lst->st_stid);
5850         }
5851         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5852         if (ns == NULL)
5853                 return NULL;
5854
5855         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5856         if (lst == openlockstateid(ns))
5857                 *new = true;
5858         else
5859                 nfs4_put_stid(ns);
5860 out:
5861         return lst;
5862 }
5863
5864 static int
5865 check_lock_length(u64 offset, u64 length)
5866 {
5867         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5868                 (length > ~offset)));
5869 }
5870
5871 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5872 {
5873         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5874
5875         lockdep_assert_held(&fp->fi_lock);
5876
5877         if (test_access(access, lock_stp))
5878                 return;
5879         __nfs4_file_get_access(fp, access);
5880         set_access(access, lock_stp);
5881 }
5882
5883 static __be32
5884 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5885                             struct nfs4_ol_stateid *ost,
5886                             struct nfsd4_lock *lock,
5887                             struct nfs4_ol_stateid **plst, bool *new)
5888 {
5889         __be32 status;
5890         struct nfs4_file *fi = ost->st_stid.sc_file;
5891         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5892         struct nfs4_client *cl = oo->oo_owner.so_client;
5893         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5894         struct nfs4_lockowner *lo;
5895         struct nfs4_ol_stateid *lst;
5896         unsigned int strhashval;
5897
5898         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5899         if (!lo) {
5900                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5901                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5902                 if (lo == NULL)
5903                         return nfserr_jukebox;
5904         } else {
5905                 /* with an existing lockowner, seqids must be the same */
5906                 status = nfserr_bad_seqid;
5907                 if (!cstate->minorversion &&
5908                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5909                         goto out;
5910         }
5911
5912         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5913         if (lst == NULL) {
5914                 status = nfserr_jukebox;
5915                 goto out;
5916         }
5917
5918         status = nfs_ok;
5919         *plst = lst;
5920 out:
5921         nfs4_put_stateowner(&lo->lo_owner);
5922         return status;
5923 }
5924
5925 /*
5926  *  LOCK operation
5927  */
5928 __be32
5929 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5930            union nfsd4_op_u *u)
5931 {
5932         struct nfsd4_lock *lock = &u->lock;
5933         struct nfs4_openowner *open_sop = NULL;
5934         struct nfs4_lockowner *lock_sop = NULL;
5935         struct nfs4_ol_stateid *lock_stp = NULL;
5936         struct nfs4_ol_stateid *open_stp = NULL;
5937         struct nfs4_file *fp;
5938         struct file *filp = NULL;
5939         struct nfsd4_blocked_lock *nbl = NULL;
5940         struct file_lock *file_lock = NULL;
5941         struct file_lock *conflock = NULL;
5942         __be32 status = 0;
5943         int lkflg;
5944         int err;
5945         bool new = false;
5946         unsigned char fl_type;
5947         unsigned int fl_flags = FL_POSIX;
5948         struct net *net = SVC_NET(rqstp);
5949         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5950
5951         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5952                 (long long) lock->lk_offset,
5953                 (long long) lock->lk_length);
5954
5955         if (check_lock_length(lock->lk_offset, lock->lk_length))
5956                  return nfserr_inval;
5957
5958         if ((status = fh_verify(rqstp, &cstate->current_fh,
5959                                 S_IFREG, NFSD_MAY_LOCK))) {
5960                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5961                 return status;
5962         }
5963
5964         if (lock->lk_is_new) {
5965                 if (nfsd4_has_session(cstate))
5966                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5967                         memcpy(&lock->lk_new_clientid,
5968                                 &cstate->session->se_client->cl_clientid,
5969                                 sizeof(clientid_t));
5970
5971                 status = nfserr_stale_clientid;
5972                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5973                         goto out;
5974
5975                 /* validate and update open stateid and open seqid */
5976                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5977                                         lock->lk_new_open_seqid,
5978                                         &lock->lk_new_open_stateid,
5979                                         &open_stp, nn);
5980                 if (status)
5981                         goto out;
5982                 mutex_unlock(&open_stp->st_mutex);
5983                 open_sop = openowner(open_stp->st_stateowner);
5984                 status = nfserr_bad_stateid;
5985                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5986                                                 &lock->lk_new_clientid))
5987                         goto out;
5988                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5989                                                         &lock_stp, &new);
5990         } else {
5991                 status = nfs4_preprocess_seqid_op(cstate,
5992                                        lock->lk_old_lock_seqid,
5993                                        &lock->lk_old_lock_stateid,
5994                                        NFS4_LOCK_STID, &lock_stp, nn);
5995         }
5996         if (status)
5997                 goto out;
5998         lock_sop = lockowner(lock_stp->st_stateowner);
5999
6000         lkflg = setlkflg(lock->lk_type);
6001         status = nfs4_check_openmode(lock_stp, lkflg);
6002         if (status)
6003                 goto out;
6004
6005         status = nfserr_grace;
6006         if (locks_in_grace(net) && !lock->lk_reclaim)
6007                 goto out;
6008         status = nfserr_no_grace;
6009         if (!locks_in_grace(net) && lock->lk_reclaim)
6010                 goto out;
6011
6012         fp = lock_stp->st_stid.sc_file;
6013         switch (lock->lk_type) {
6014                 case NFS4_READW_LT:
6015                         if (nfsd4_has_session(cstate))
6016                                 fl_flags |= FL_SLEEP;
6017                         /* Fallthrough */
6018                 case NFS4_READ_LT:
6019                         spin_lock(&fp->fi_lock);
6020                         filp = find_readable_file_locked(fp);
6021                         if (filp)
6022                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6023                         spin_unlock(&fp->fi_lock);
6024                         fl_type = F_RDLCK;
6025                         break;
6026                 case NFS4_WRITEW_LT:
6027                         if (nfsd4_has_session(cstate))
6028                                 fl_flags |= FL_SLEEP;
6029                         /* Fallthrough */
6030                 case NFS4_WRITE_LT:
6031                         spin_lock(&fp->fi_lock);
6032                         filp = find_writeable_file_locked(fp);
6033                         if (filp)
6034                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6035                         spin_unlock(&fp->fi_lock);
6036                         fl_type = F_WRLCK;
6037                         break;
6038                 default:
6039                         status = nfserr_inval;
6040                 goto out;
6041         }
6042
6043         if (!filp) {
6044                 status = nfserr_openmode;
6045                 goto out;
6046         }
6047
6048         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6049         if (!nbl) {
6050                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6051                 status = nfserr_jukebox;
6052                 goto out;
6053         }
6054
6055         file_lock = &nbl->nbl_lock;
6056         file_lock->fl_type = fl_type;
6057         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6058         file_lock->fl_pid = current->tgid;
6059         file_lock->fl_file = filp;
6060         file_lock->fl_flags = fl_flags;
6061         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6062         file_lock->fl_start = lock->lk_offset;
6063         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6064         nfs4_transform_lock_offset(file_lock);
6065
6066         conflock = locks_alloc_lock();
6067         if (!conflock) {
6068                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6069                 status = nfserr_jukebox;
6070                 goto out;
6071         }
6072
6073         if (fl_flags & FL_SLEEP) {
6074                 nbl->nbl_time = jiffies;
6075                 spin_lock(&nn->blocked_locks_lock);
6076                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6077                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6078                 spin_unlock(&nn->blocked_locks_lock);
6079         }
6080
6081         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6082         switch (err) {
6083         case 0: /* success! */
6084                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6085                 status = 0;
6086                 if (lock->lk_reclaim)
6087                         nn->somebody_reclaimed = true;
6088                 break;
6089         case FILE_LOCK_DEFERRED:
6090                 nbl = NULL;
6091                 /* Fallthrough */
6092         case -EAGAIN:           /* conflock holds conflicting lock */
6093                 status = nfserr_denied;
6094                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6095                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6096                 break;
6097         case -EDEADLK:
6098                 status = nfserr_deadlock;
6099                 break;
6100         default:
6101                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6102                 status = nfserrno(err);
6103                 break;
6104         }
6105 out:
6106         if (nbl) {
6107                 /* dequeue it if we queued it before */
6108                 if (fl_flags & FL_SLEEP) {
6109                         spin_lock(&nn->blocked_locks_lock);
6110                         list_del_init(&nbl->nbl_list);
6111                         list_del_init(&nbl->nbl_lru);
6112                         spin_unlock(&nn->blocked_locks_lock);
6113                 }
6114                 free_blocked_lock(nbl);
6115         }
6116         if (filp)
6117                 fput(filp);
6118         if (lock_stp) {
6119                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6120                 if (cstate->replay_owner &&
6121                     cstate->replay_owner != &lock_sop->lo_owner &&
6122                     seqid_mutating_err(ntohl(status)))
6123                         lock_sop->lo_owner.so_seqid++;
6124
6125                 /*
6126                  * If this is a new, never-before-used stateid, and we are
6127                  * returning an error, then just go ahead and release it.
6128                  */
6129                 if (status && new)
6130                         release_lock_stateid(lock_stp);
6131
6132                 mutex_unlock(&lock_stp->st_mutex);
6133
6134                 nfs4_put_stid(&lock_stp->st_stid);
6135         }
6136         if (open_stp)
6137                 nfs4_put_stid(&open_stp->st_stid);
6138         nfsd4_bump_seqid(cstate, status);
6139         if (conflock)
6140                 locks_free_lock(conflock);
6141         return status;
6142 }
6143
6144 /*
6145  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6146  * so we do a temporary open here just to get an open file to pass to
6147  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6148  * inode operation.)
6149  */
6150 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6151 {
6152         struct file *file;
6153         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6154         if (!err) {
6155                 err = nfserrno(vfs_test_lock(file, lock));
6156                 fput(file);
6157         }
6158         return err;
6159 }
6160
6161 /*
6162  * LOCKT operation
6163  */
6164 __be32
6165 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6166             union nfsd4_op_u *u)
6167 {
6168         struct nfsd4_lockt *lockt = &u->lockt;
6169         struct file_lock *file_lock = NULL;
6170         struct nfs4_lockowner *lo = NULL;
6171         __be32 status;
6172         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6173
6174         if (locks_in_grace(SVC_NET(rqstp)))
6175                 return nfserr_grace;
6176
6177         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6178                  return nfserr_inval;
6179
6180         if (!nfsd4_has_session(cstate)) {
6181                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6182                 if (status)
6183                         goto out;
6184         }
6185
6186         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6187                 goto out;
6188
6189         file_lock = locks_alloc_lock();
6190         if (!file_lock) {
6191                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6192                 status = nfserr_jukebox;
6193                 goto out;
6194         }
6195
6196         switch (lockt->lt_type) {
6197                 case NFS4_READ_LT:
6198                 case NFS4_READW_LT:
6199                         file_lock->fl_type = F_RDLCK;
6200                 break;
6201                 case NFS4_WRITE_LT:
6202                 case NFS4_WRITEW_LT:
6203                         file_lock->fl_type = F_WRLCK;
6204                 break;
6205                 default:
6206                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6207                         status = nfserr_inval;
6208                 goto out;
6209         }
6210
6211         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6212         if (lo)
6213                 file_lock->fl_owner = (fl_owner_t)lo;
6214         file_lock->fl_pid = current->tgid;
6215         file_lock->fl_flags = FL_POSIX;
6216
6217         file_lock->fl_start = lockt->lt_offset;
6218         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6219
6220         nfs4_transform_lock_offset(file_lock);
6221
6222         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6223         if (status)
6224                 goto out;
6225
6226         if (file_lock->fl_type != F_UNLCK) {
6227                 status = nfserr_denied;
6228                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6229         }
6230 out:
6231         if (lo)
6232                 nfs4_put_stateowner(&lo->lo_owner);
6233         if (file_lock)
6234                 locks_free_lock(file_lock);
6235         return status;
6236 }
6237
6238 __be32
6239 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6240             union nfsd4_op_u *u)
6241 {
6242         struct nfsd4_locku *locku = &u->locku;
6243         struct nfs4_ol_stateid *stp;
6244         struct file *filp = NULL;
6245         struct file_lock *file_lock = NULL;
6246         __be32 status;
6247         int err;
6248         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6249
6250         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6251                 (long long) locku->lu_offset,
6252                 (long long) locku->lu_length);
6253
6254         if (check_lock_length(locku->lu_offset, locku->lu_length))
6255                  return nfserr_inval;
6256
6257         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6258                                         &locku->lu_stateid, NFS4_LOCK_STID,
6259                                         &stp, nn);
6260         if (status)
6261                 goto out;
6262         filp = find_any_file(stp->st_stid.sc_file);
6263         if (!filp) {
6264                 status = nfserr_lock_range;
6265                 goto put_stateid;
6266         }
6267         file_lock = locks_alloc_lock();
6268         if (!file_lock) {
6269                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6270                 status = nfserr_jukebox;
6271                 goto fput;
6272         }
6273
6274         file_lock->fl_type = F_UNLCK;
6275         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6276         file_lock->fl_pid = current->tgid;
6277         file_lock->fl_file = filp;
6278         file_lock->fl_flags = FL_POSIX;
6279         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6280         file_lock->fl_start = locku->lu_offset;
6281
6282         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6283                                                 locku->lu_length);
6284         nfs4_transform_lock_offset(file_lock);
6285
6286         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6287         if (err) {
6288                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6289                 goto out_nfserr;
6290         }
6291         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6292 fput:
6293         fput(filp);
6294 put_stateid:
6295         mutex_unlock(&stp->st_mutex);
6296         nfs4_put_stid(&stp->st_stid);
6297 out:
6298         nfsd4_bump_seqid(cstate, status);
6299         if (file_lock)
6300                 locks_free_lock(file_lock);
6301         return status;
6302
6303 out_nfserr:
6304         status = nfserrno(err);
6305         goto fput;
6306 }
6307
6308 /*
6309  * returns
6310  *      true:  locks held by lockowner
6311  *      false: no locks held by lockowner
6312  */
6313 static bool
6314 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6315 {
6316         struct file_lock *fl;
6317         int status = false;
6318         struct file *filp = find_any_file(fp);
6319         struct inode *inode;
6320         struct file_lock_context *flctx;
6321
6322         if (!filp) {
6323                 /* Any valid lock stateid should have some sort of access */
6324                 WARN_ON_ONCE(1);
6325                 return status;
6326         }
6327
6328         inode = locks_inode(filp);
6329         flctx = inode->i_flctx;
6330
6331         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6332                 spin_lock(&flctx->flc_lock);
6333                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6334                         if (fl->fl_owner == (fl_owner_t)lowner) {
6335                                 status = true;
6336                                 break;
6337                         }
6338                 }
6339                 spin_unlock(&flctx->flc_lock);
6340         }
6341         fput(filp);
6342         return status;
6343 }
6344
6345 __be32
6346 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6347                         struct nfsd4_compound_state *cstate,
6348                         union nfsd4_op_u *u)
6349 {
6350         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6351         clientid_t *clid = &rlockowner->rl_clientid;
6352         struct nfs4_stateowner *sop;
6353         struct nfs4_lockowner *lo = NULL;
6354         struct nfs4_ol_stateid *stp;
6355         struct xdr_netobj *owner = &rlockowner->rl_owner;
6356         unsigned int hashval = ownerstr_hashval(owner);
6357         __be32 status;
6358         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6359         struct nfs4_client *clp;
6360         LIST_HEAD (reaplist);
6361
6362         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6363                 clid->cl_boot, clid->cl_id);
6364
6365         status = lookup_clientid(clid, cstate, nn);
6366         if (status)
6367                 return status;
6368
6369         clp = cstate->clp;
6370         /* Find the matching lock stateowner */
6371         spin_lock(&clp->cl_lock);
6372         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6373                             so_strhash) {
6374
6375                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6376                         continue;
6377
6378                 /* see if there are still any locks associated with it */
6379                 lo = lockowner(sop);
6380                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6381                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6382                                 status = nfserr_locks_held;
6383                                 spin_unlock(&clp->cl_lock);
6384                                 return status;
6385                         }
6386                 }
6387
6388                 nfs4_get_stateowner(sop);
6389                 break;
6390         }
6391         if (!lo) {
6392                 spin_unlock(&clp->cl_lock);
6393                 return status;
6394         }
6395
6396         unhash_lockowner_locked(lo);
6397         while (!list_empty(&lo->lo_owner.so_stateids)) {
6398                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6399                                        struct nfs4_ol_stateid,
6400                                        st_perstateowner);
6401                 WARN_ON(!unhash_lock_stateid(stp));
6402                 put_ol_stateid_locked(stp, &reaplist);
6403         }
6404         spin_unlock(&clp->cl_lock);
6405         free_ol_stateid_reaplist(&reaplist);
6406         remove_blocked_locks(lo);
6407         nfs4_put_stateowner(&lo->lo_owner);
6408
6409         return status;
6410 }
6411
6412 static inline struct nfs4_client_reclaim *
6413 alloc_reclaim(void)
6414 {
6415         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6416 }
6417
6418 bool
6419 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6420 {
6421         struct nfs4_client_reclaim *crp;
6422
6423         crp = nfsd4_find_reclaim_client(name, nn);
6424         return (crp && crp->cr_clp);
6425 }
6426
6427 /*
6428  * failure => all reset bets are off, nfserr_no_grace...
6429  */
6430 struct nfs4_client_reclaim *
6431 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6432 {
6433         unsigned int strhashval;
6434         struct nfs4_client_reclaim *crp;
6435
6436         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6437         crp = alloc_reclaim();
6438         if (crp) {
6439                 strhashval = clientstr_hashval(name);
6440                 INIT_LIST_HEAD(&crp->cr_strhash);
6441                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6442                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6443                 crp->cr_clp = NULL;
6444                 nn->reclaim_str_hashtbl_size++;
6445         }
6446         return crp;
6447 }
6448
6449 void
6450 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6451 {
6452         list_del(&crp->cr_strhash);
6453         kfree(crp);
6454         nn->reclaim_str_hashtbl_size--;
6455 }
6456
6457 void
6458 nfs4_release_reclaim(struct nfsd_net *nn)
6459 {
6460         struct nfs4_client_reclaim *crp = NULL;
6461         int i;
6462
6463         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6464                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6465                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6466                                         struct nfs4_client_reclaim, cr_strhash);
6467                         nfs4_remove_reclaim_record(crp, nn);
6468                 }
6469         }
6470         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6471 }
6472
6473 /*
6474  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6475 struct nfs4_client_reclaim *
6476 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6477 {
6478         unsigned int strhashval;
6479         struct nfs4_client_reclaim *crp = NULL;
6480
6481         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6482
6483         strhashval = clientstr_hashval(recdir);
6484         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6485                 if (same_name(crp->cr_recdir, recdir)) {
6486                         return crp;
6487                 }
6488         }
6489         return NULL;
6490 }
6491
6492 /*
6493 * Called from OPEN. Look for clientid in reclaim list.
6494 */
6495 __be32
6496 nfs4_check_open_reclaim(clientid_t *clid,
6497                 struct nfsd4_compound_state *cstate,
6498                 struct nfsd_net *nn)
6499 {
6500         __be32 status;
6501
6502         /* find clientid in conf_id_hashtbl */
6503         status = lookup_clientid(clid, cstate, nn);
6504         if (status)
6505                 return nfserr_reclaim_bad;
6506
6507         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6508                 return nfserr_no_grace;
6509
6510         if (nfsd4_client_record_check(cstate->clp))
6511                 return nfserr_reclaim_bad;
6512
6513         return nfs_ok;
6514 }
6515
6516 #ifdef CONFIG_NFSD_FAULT_INJECTION
6517 static inline void
6518 put_client(struct nfs4_client *clp)
6519 {
6520         atomic_dec(&clp->cl_refcount);
6521 }
6522
6523 static struct nfs4_client *
6524 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6525 {
6526         struct nfs4_client *clp;
6527         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6528                                           nfsd_net_id);
6529
6530         if (!nfsd_netns_ready(nn))
6531                 return NULL;
6532
6533         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6534                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6535                         return clp;
6536         }
6537         return NULL;
6538 }
6539
6540 u64
6541 nfsd_inject_print_clients(void)
6542 {
6543         struct nfs4_client *clp;
6544         u64 count = 0;
6545         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6546                                           nfsd_net_id);
6547         char buf[INET6_ADDRSTRLEN];
6548
6549         if (!nfsd_netns_ready(nn))
6550                 return 0;
6551
6552         spin_lock(&nn->client_lock);
6553         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6554                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6555                 pr_info("NFS Client: %s\n", buf);
6556                 ++count;
6557         }
6558         spin_unlock(&nn->client_lock);
6559
6560         return count;
6561 }
6562
6563 u64
6564 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6565 {
6566         u64 count = 0;
6567         struct nfs4_client *clp;
6568         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6569                                           nfsd_net_id);
6570
6571         if (!nfsd_netns_ready(nn))
6572                 return count;
6573
6574         spin_lock(&nn->client_lock);
6575         clp = nfsd_find_client(addr, addr_size);
6576         if (clp) {
6577                 if (mark_client_expired_locked(clp) == nfs_ok)
6578                         ++count;
6579                 else
6580                         clp = NULL;
6581         }
6582         spin_unlock(&nn->client_lock);
6583
6584         if (clp)
6585                 expire_client(clp);
6586
6587         return count;
6588 }
6589
6590 u64
6591 nfsd_inject_forget_clients(u64 max)
6592 {
6593         u64 count = 0;
6594         struct nfs4_client *clp, *next;
6595         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6596                                                 nfsd_net_id);
6597         LIST_HEAD(reaplist);
6598
6599         if (!nfsd_netns_ready(nn))
6600                 return count;
6601
6602         spin_lock(&nn->client_lock);
6603         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6604                 if (mark_client_expired_locked(clp) == nfs_ok) {
6605                         list_add(&clp->cl_lru, &reaplist);
6606                         if (max != 0 && ++count >= max)
6607                                 break;
6608                 }
6609         }
6610         spin_unlock(&nn->client_lock);
6611
6612         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6613                 expire_client(clp);
6614
6615         return count;
6616 }
6617
6618 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6619                              const char *type)
6620 {
6621         char buf[INET6_ADDRSTRLEN];
6622         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6623         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6624 }
6625
6626 static void
6627 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6628                              struct list_head *collect)
6629 {
6630         struct nfs4_client *clp = lst->st_stid.sc_client;
6631         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6632                                           nfsd_net_id);
6633
6634         if (!collect)
6635                 return;
6636
6637         lockdep_assert_held(&nn->client_lock);
6638         atomic_inc(&clp->cl_refcount);
6639         list_add(&lst->st_locks, collect);
6640 }
6641
6642 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6643                                     struct list_head *collect,
6644                                     bool (*func)(struct nfs4_ol_stateid *))
6645 {
6646         struct nfs4_openowner *oop;
6647         struct nfs4_ol_stateid *stp, *st_next;
6648         struct nfs4_ol_stateid *lst, *lst_next;
6649         u64 count = 0;
6650
6651         spin_lock(&clp->cl_lock);
6652         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6653                 list_for_each_entry_safe(stp, st_next,
6654                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6655                         list_for_each_entry_safe(lst, lst_next,
6656                                         &stp->st_locks, st_locks) {
6657                                 if (func) {
6658                                         if (func(lst))
6659                                                 nfsd_inject_add_lock_to_list(lst,
6660                                                                         collect);
6661                                 }
6662                                 ++count;
6663                                 /*
6664                                  * Despite the fact that these functions deal
6665                                  * with 64-bit integers for "count", we must
6666                                  * ensure that it doesn't blow up the
6667                                  * clp->cl_refcount. Throw a warning if we
6668                                  * start to approach INT_MAX here.
6669                                  */
6670                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6671                                 if (count == max)
6672                                         goto out;
6673                         }
6674                 }
6675         }
6676 out:
6677         spin_unlock(&clp->cl_lock);
6678
6679         return count;
6680 }
6681
6682 static u64
6683 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6684                           u64 max)
6685 {
6686         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6687 }
6688
6689 static u64
6690 nfsd_print_client_locks(struct nfs4_client *clp)
6691 {
6692         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6693         nfsd_print_count(clp, count, "locked files");
6694         return count;
6695 }
6696
6697 u64
6698 nfsd_inject_print_locks(void)
6699 {
6700         struct nfs4_client *clp;
6701         u64 count = 0;
6702         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6703                                                 nfsd_net_id);
6704
6705         if (!nfsd_netns_ready(nn))
6706                 return 0;
6707
6708         spin_lock(&nn->client_lock);
6709         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6710                 count += nfsd_print_client_locks(clp);
6711         spin_unlock(&nn->client_lock);
6712
6713         return count;
6714 }
6715
6716 static void
6717 nfsd_reap_locks(struct list_head *reaplist)
6718 {
6719         struct nfs4_client *clp;
6720         struct nfs4_ol_stateid *stp, *next;
6721
6722         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6723                 list_del_init(&stp->st_locks);
6724                 clp = stp->st_stid.sc_client;
6725                 nfs4_put_stid(&stp->st_stid);
6726                 put_client(clp);
6727         }
6728 }
6729
6730 u64
6731 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6732 {
6733         unsigned int count = 0;
6734         struct nfs4_client *clp;
6735         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6736                                                 nfsd_net_id);
6737         LIST_HEAD(reaplist);
6738
6739         if (!nfsd_netns_ready(nn))
6740                 return count;
6741
6742         spin_lock(&nn->client_lock);
6743         clp = nfsd_find_client(addr, addr_size);
6744         if (clp)
6745                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6746         spin_unlock(&nn->client_lock);
6747         nfsd_reap_locks(&reaplist);
6748         return count;
6749 }
6750
6751 u64
6752 nfsd_inject_forget_locks(u64 max)
6753 {
6754         u64 count = 0;
6755         struct nfs4_client *clp;
6756         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6757                                                 nfsd_net_id);
6758         LIST_HEAD(reaplist);
6759
6760         if (!nfsd_netns_ready(nn))
6761                 return count;
6762
6763         spin_lock(&nn->client_lock);
6764         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6765                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6766                 if (max != 0 && count >= max)
6767                         break;
6768         }
6769         spin_unlock(&nn->client_lock);
6770         nfsd_reap_locks(&reaplist);
6771         return count;
6772 }
6773
6774 static u64
6775 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6776                               struct list_head *collect,
6777                               void (*func)(struct nfs4_openowner *))
6778 {
6779         struct nfs4_openowner *oop, *next;
6780         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6781                                                 nfsd_net_id);
6782         u64 count = 0;
6783
6784         lockdep_assert_held(&nn->client_lock);
6785
6786         spin_lock(&clp->cl_lock);
6787         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6788                 if (func) {
6789                         func(oop);
6790                         if (collect) {
6791                                 atomic_inc(&clp->cl_refcount);
6792                                 list_add(&oop->oo_perclient, collect);
6793                         }
6794                 }
6795                 ++count;
6796                 /*
6797                  * Despite the fact that these functions deal with
6798                  * 64-bit integers for "count", we must ensure that
6799                  * it doesn't blow up the clp->cl_refcount. Throw a
6800                  * warning if we start to approach INT_MAX here.
6801                  */
6802                 WARN_ON_ONCE(count == (INT_MAX / 2));
6803                 if (count == max)
6804                         break;
6805         }
6806         spin_unlock(&clp->cl_lock);
6807
6808         return count;
6809 }
6810
6811 static u64
6812 nfsd_print_client_openowners(struct nfs4_client *clp)
6813 {
6814         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6815
6816         nfsd_print_count(clp, count, "openowners");
6817         return count;
6818 }
6819
6820 static u64
6821 nfsd_collect_client_openowners(struct nfs4_client *clp,
6822                                struct list_head *collect, u64 max)
6823 {
6824         return nfsd_foreach_client_openowner(clp, max, collect,
6825                                                 unhash_openowner_locked);
6826 }
6827
6828 u64
6829 nfsd_inject_print_openowners(void)
6830 {
6831         struct nfs4_client *clp;
6832         u64 count = 0;
6833         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6834                                                 nfsd_net_id);
6835
6836         if (!nfsd_netns_ready(nn))
6837                 return 0;
6838
6839         spin_lock(&nn->client_lock);
6840         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6841                 count += nfsd_print_client_openowners(clp);
6842         spin_unlock(&nn->client_lock);
6843
6844         return count;
6845 }
6846
6847 static void
6848 nfsd_reap_openowners(struct list_head *reaplist)
6849 {
6850         struct nfs4_client *clp;
6851         struct nfs4_openowner *oop, *next;
6852
6853         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6854                 list_del_init(&oop->oo_perclient);
6855                 clp = oop->oo_owner.so_client;
6856                 release_openowner(oop);
6857                 put_client(clp);
6858         }
6859 }
6860
6861 u64
6862 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6863                                      size_t addr_size)
6864 {
6865         unsigned int count = 0;
6866         struct nfs4_client *clp;
6867         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6868                                                 nfsd_net_id);
6869         LIST_HEAD(reaplist);
6870
6871         if (!nfsd_netns_ready(nn))
6872                 return count;
6873
6874         spin_lock(&nn->client_lock);
6875         clp = nfsd_find_client(addr, addr_size);
6876         if (clp)
6877                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6878         spin_unlock(&nn->client_lock);
6879         nfsd_reap_openowners(&reaplist);
6880         return count;
6881 }
6882
6883 u64
6884 nfsd_inject_forget_openowners(u64 max)
6885 {
6886         u64 count = 0;
6887         struct nfs4_client *clp;
6888         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6889                                                 nfsd_net_id);
6890         LIST_HEAD(reaplist);
6891
6892         if (!nfsd_netns_ready(nn))
6893                 return count;
6894
6895         spin_lock(&nn->client_lock);
6896         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6897                 count += nfsd_collect_client_openowners(clp, &reaplist,
6898                                                         max - count);
6899                 if (max != 0 && count >= max)
6900                         break;
6901         }
6902         spin_unlock(&nn->client_lock);
6903         nfsd_reap_openowners(&reaplist);
6904         return count;
6905 }
6906
6907 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6908                                      struct list_head *victims)
6909 {
6910         struct nfs4_delegation *dp, *next;
6911         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6912                                                 nfsd_net_id);
6913         u64 count = 0;
6914
6915         lockdep_assert_held(&nn->client_lock);
6916
6917         spin_lock(&state_lock);
6918         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6919                 if (victims) {
6920                         /*
6921                          * It's not safe to mess with delegations that have a
6922                          * non-zero dl_time. They might have already been broken
6923                          * and could be processed by the laundromat outside of
6924                          * the state_lock. Just leave them be.
6925                          */
6926                         if (dp->dl_time != 0)
6927                                 continue;
6928
6929                         atomic_inc(&clp->cl_refcount);
6930                         WARN_ON(!unhash_delegation_locked(dp));
6931                         list_add(&dp->dl_recall_lru, victims);
6932                 }
6933                 ++count;
6934                 /*
6935                  * Despite the fact that these functions deal with
6936                  * 64-bit integers for "count", we must ensure that
6937                  * it doesn't blow up the clp->cl_refcount. Throw a
6938                  * warning if we start to approach INT_MAX here.
6939                  */
6940                 WARN_ON_ONCE(count == (INT_MAX / 2));
6941                 if (count == max)
6942                         break;
6943         }
6944         spin_unlock(&state_lock);
6945         return count;
6946 }
6947
6948 static u64
6949 nfsd_print_client_delegations(struct nfs4_client *clp)
6950 {
6951         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6952
6953         nfsd_print_count(clp, count, "delegations");
6954         return count;
6955 }
6956
6957 u64
6958 nfsd_inject_print_delegations(void)
6959 {
6960         struct nfs4_client *clp;
6961         u64 count = 0;
6962         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6963                                                 nfsd_net_id);
6964
6965         if (!nfsd_netns_ready(nn))
6966                 return 0;
6967
6968         spin_lock(&nn->client_lock);
6969         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6970                 count += nfsd_print_client_delegations(clp);
6971         spin_unlock(&nn->client_lock);
6972
6973         return count;
6974 }
6975
6976 static void
6977 nfsd_forget_delegations(struct list_head *reaplist)
6978 {
6979         struct nfs4_client *clp;
6980         struct nfs4_delegation *dp, *next;
6981
6982         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6983                 list_del_init(&dp->dl_recall_lru);
6984                 clp = dp->dl_stid.sc_client;
6985                 revoke_delegation(dp);
6986                 put_client(clp);
6987         }
6988 }
6989
6990 u64
6991 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6992                                       size_t addr_size)
6993 {
6994         u64 count = 0;
6995         struct nfs4_client *clp;
6996         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6997                                                 nfsd_net_id);
6998         LIST_HEAD(reaplist);
6999
7000         if (!nfsd_netns_ready(nn))
7001                 return count;
7002
7003         spin_lock(&nn->client_lock);
7004         clp = nfsd_find_client(addr, addr_size);
7005         if (clp)
7006                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7007         spin_unlock(&nn->client_lock);
7008
7009         nfsd_forget_delegations(&reaplist);
7010         return count;
7011 }
7012
7013 u64
7014 nfsd_inject_forget_delegations(u64 max)
7015 {
7016         u64 count = 0;
7017         struct nfs4_client *clp;
7018         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7019                                                 nfsd_net_id);
7020         LIST_HEAD(reaplist);
7021
7022         if (!nfsd_netns_ready(nn))
7023                 return count;
7024
7025         spin_lock(&nn->client_lock);
7026         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7027                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7028                 if (max != 0 && count >= max)
7029                         break;
7030         }
7031         spin_unlock(&nn->client_lock);
7032         nfsd_forget_delegations(&reaplist);
7033         return count;
7034 }
7035
7036 static void
7037 nfsd_recall_delegations(struct list_head *reaplist)
7038 {
7039         struct nfs4_client *clp;
7040         struct nfs4_delegation *dp, *next;
7041
7042         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7043                 list_del_init(&dp->dl_recall_lru);
7044                 clp = dp->dl_stid.sc_client;
7045                 /*
7046                  * We skipped all entries that had a zero dl_time before,
7047                  * so we can now reset the dl_time back to 0. If a delegation
7048                  * break comes in now, then it won't make any difference since
7049                  * we're recalling it either way.
7050                  */
7051                 spin_lock(&state_lock);
7052                 dp->dl_time = 0;
7053                 spin_unlock(&state_lock);
7054                 nfsd_break_one_deleg(dp);
7055                 put_client(clp);
7056         }
7057 }
7058
7059 u64
7060 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7061                                       size_t addr_size)
7062 {
7063         u64 count = 0;
7064         struct nfs4_client *clp;
7065         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7066                                                 nfsd_net_id);
7067         LIST_HEAD(reaplist);
7068
7069         if (!nfsd_netns_ready(nn))
7070                 return count;
7071
7072         spin_lock(&nn->client_lock);
7073         clp = nfsd_find_client(addr, addr_size);
7074         if (clp)
7075                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7076         spin_unlock(&nn->client_lock);
7077
7078         nfsd_recall_delegations(&reaplist);
7079         return count;
7080 }
7081
7082 u64
7083 nfsd_inject_recall_delegations(u64 max)
7084 {
7085         u64 count = 0;
7086         struct nfs4_client *clp, *next;
7087         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7088                                                 nfsd_net_id);
7089         LIST_HEAD(reaplist);
7090
7091         if (!nfsd_netns_ready(nn))
7092                 return count;
7093
7094         spin_lock(&nn->client_lock);
7095         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7096                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7097                 if (max != 0 && ++count >= max)
7098                         break;
7099         }
7100         spin_unlock(&nn->client_lock);
7101         nfsd_recall_delegations(&reaplist);
7102         return count;
7103 }
7104 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7105
7106 /*
7107  * Since the lifetime of a delegation isn't limited to that of an open, a
7108  * client may quite reasonably hang on to a delegation as long as it has
7109  * the inode cached.  This becomes an obvious problem the first time a
7110  * client's inode cache approaches the size of the server's total memory.
7111  *
7112  * For now we avoid this problem by imposing a hard limit on the number
7113  * of delegations, which varies according to the server's memory size.
7114  */
7115 static void
7116 set_max_delegations(void)
7117 {
7118         /*
7119          * Allow at most 4 delegations per megabyte of RAM.  Quick
7120          * estimates suggest that in the worst case (where every delegation
7121          * is for a different inode), a delegation could take about 1.5K,
7122          * giving a worst case usage of about 6% of memory.
7123          */
7124         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7125 }
7126
7127 static int nfs4_state_create_net(struct net *net)
7128 {
7129         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7130         int i;
7131
7132         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7133                                             sizeof(struct list_head),
7134                                             GFP_KERNEL);
7135         if (!nn->conf_id_hashtbl)
7136                 goto err;
7137         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7138                                               sizeof(struct list_head),
7139                                               GFP_KERNEL);
7140         if (!nn->unconf_id_hashtbl)
7141                 goto err_unconf_id;
7142         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7143                                               sizeof(struct list_head),
7144                                               GFP_KERNEL);
7145         if (!nn->sessionid_hashtbl)
7146                 goto err_sessionid;
7147
7148         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7149                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7150                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7151         }
7152         for (i = 0; i < SESSION_HASH_SIZE; i++)
7153                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7154         nn->conf_name_tree = RB_ROOT;
7155         nn->unconf_name_tree = RB_ROOT;
7156         nn->boot_time = get_seconds();
7157         nn->grace_ended = false;
7158         nn->nfsd4_manager.block_opens = true;
7159         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7160         INIT_LIST_HEAD(&nn->client_lru);
7161         INIT_LIST_HEAD(&nn->close_lru);
7162         INIT_LIST_HEAD(&nn->del_recall_lru);
7163         spin_lock_init(&nn->client_lock);
7164
7165         spin_lock_init(&nn->blocked_locks_lock);
7166         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7167
7168         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7169         get_net(net);
7170
7171         return 0;
7172
7173 err_sessionid:
7174         kfree(nn->unconf_id_hashtbl);
7175 err_unconf_id:
7176         kfree(nn->conf_id_hashtbl);
7177 err:
7178         return -ENOMEM;
7179 }
7180
7181 static void
7182 nfs4_state_destroy_net(struct net *net)
7183 {
7184         int i;
7185         struct nfs4_client *clp = NULL;
7186         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7187
7188         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7189                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7190                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7191                         destroy_client(clp);
7192                 }
7193         }
7194
7195         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7196
7197         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7198                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7199                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7200                         destroy_client(clp);
7201                 }
7202         }
7203
7204         kfree(nn->sessionid_hashtbl);
7205         kfree(nn->unconf_id_hashtbl);
7206         kfree(nn->conf_id_hashtbl);
7207         put_net(net);
7208 }
7209
7210 int
7211 nfs4_state_start_net(struct net *net)
7212 {
7213         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7214         int ret;
7215
7216         ret = nfs4_state_create_net(net);
7217         if (ret)
7218                 return ret;
7219         locks_start_grace(net, &nn->nfsd4_manager);
7220         nfsd4_client_tracking_init(net);
7221         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7222                nn->nfsd4_grace, net->ns.inum);
7223         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7224         return 0;
7225 }
7226
7227 /* initialization to perform when the nfsd service is started: */
7228
7229 int
7230 nfs4_state_start(void)
7231 {
7232         int ret;
7233
7234         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7235         if (laundry_wq == NULL) {
7236                 ret = -ENOMEM;
7237                 goto out;
7238         }
7239         ret = nfsd4_create_callback_queue();
7240         if (ret)
7241                 goto out_free_laundry;
7242
7243         set_max_delegations();
7244         return 0;
7245
7246 out_free_laundry:
7247         destroy_workqueue(laundry_wq);
7248 out:
7249         return ret;
7250 }
7251
7252 void
7253 nfs4_state_shutdown_net(struct net *net)
7254 {
7255         struct nfs4_delegation *dp = NULL;
7256         struct list_head *pos, *next, reaplist;
7257         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7258
7259         cancel_delayed_work_sync(&nn->laundromat_work);
7260         locks_end_grace(&nn->nfsd4_manager);
7261
7262         INIT_LIST_HEAD(&reaplist);
7263         spin_lock(&state_lock);
7264         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7265                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7266                 WARN_ON(!unhash_delegation_locked(dp));
7267                 list_add(&dp->dl_recall_lru, &reaplist);
7268         }
7269         spin_unlock(&state_lock);
7270         list_for_each_safe(pos, next, &reaplist) {
7271                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7272                 list_del_init(&dp->dl_recall_lru);
7273                 destroy_unhashed_deleg(dp);
7274         }
7275
7276         nfsd4_client_tracking_exit(net);
7277         nfs4_state_destroy_net(net);
7278 }
7279
7280 void
7281 nfs4_state_shutdown(void)
7282 {
7283         destroy_workqueue(laundry_wq);
7284         nfsd4_destroy_callback_queue();
7285 }
7286
7287 static void
7288 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7289 {
7290         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7291                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7292 }
7293
7294 static void
7295 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7296 {
7297         if (cstate->minorversion) {
7298                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7299                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7300         }
7301 }
7302
7303 void
7304 clear_current_stateid(struct nfsd4_compound_state *cstate)
7305 {
7306         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7307 }
7308
7309 /*
7310  * functions to set current state id
7311  */
7312 void
7313 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7314                 union nfsd4_op_u *u)
7315 {
7316         put_stateid(cstate, &u->open_downgrade.od_stateid);
7317 }
7318
7319 void
7320 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7321                 union nfsd4_op_u *u)
7322 {
7323         put_stateid(cstate, &u->open.op_stateid);
7324 }
7325
7326 void
7327 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7328                 union nfsd4_op_u *u)
7329 {
7330         put_stateid(cstate, &u->close.cl_stateid);
7331 }
7332
7333 void
7334 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7335                 union nfsd4_op_u *u)
7336 {
7337         put_stateid(cstate, &u->lock.lk_resp_stateid);
7338 }
7339
7340 /*
7341  * functions to consume current state id
7342  */
7343
7344 void
7345 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7346                 union nfsd4_op_u *u)
7347 {
7348         get_stateid(cstate, &u->open_downgrade.od_stateid);
7349 }
7350
7351 void
7352 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7353                 union nfsd4_op_u *u)
7354 {
7355         get_stateid(cstate, &u->delegreturn.dr_stateid);
7356 }
7357
7358 void
7359 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7360                 union nfsd4_op_u *u)
7361 {
7362         get_stateid(cstate, &u->free_stateid.fr_stateid);
7363 }
7364
7365 void
7366 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7367                 union nfsd4_op_u *u)
7368 {
7369         get_stateid(cstate, &u->setattr.sa_stateid);
7370 }
7371
7372 void
7373 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7374                 union nfsd4_op_u *u)
7375 {
7376         get_stateid(cstate, &u->close.cl_stateid);
7377 }
7378
7379 void
7380 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7381                 union nfsd4_op_u *u)
7382 {
7383         get_stateid(cstate, &u->locku.lu_stateid);
7384 }
7385
7386 void
7387 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7388                 union nfsd4_op_u *u)
7389 {
7390         get_stateid(cstate, &u->read.rd_stateid);
7391 }
7392
7393 void
7394 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7395                 union nfsd4_op_u *u)
7396 {
7397         get_stateid(cstate, &u->write.wr_stateid);
7398 }