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         if ((source->cr_principal && ! target->cr_principal) ||
1983             (source->cr_raw_principal && ! target->cr_raw_principal))
1984                 return -ENOMEM;
1985
1986         target->cr_flavor = source->cr_flavor;
1987         target->cr_uid = source->cr_uid;
1988         target->cr_gid = source->cr_gid;
1989         target->cr_group_info = source->cr_group_info;
1990         get_group_info(target->cr_group_info);
1991         target->cr_gss_mech = source->cr_gss_mech;
1992         if (source->cr_gss_mech)
1993                 gss_mech_get(source->cr_gss_mech);
1994         return 0;
1995 }
1996
1997 static int
1998 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1999 {
2000         if (o1->len < o2->len)
2001                 return -1;
2002         if (o1->len > o2->len)
2003                 return 1;
2004         return memcmp(o1->data, o2->data, o1->len);
2005 }
2006
2007 static int same_name(const char *n1, const char *n2)
2008 {
2009         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2010 }
2011
2012 static int
2013 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2014 {
2015         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2016 }
2017
2018 static int
2019 same_clid(clientid_t *cl1, clientid_t *cl2)
2020 {
2021         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2022 }
2023
2024 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2025 {
2026         int i;
2027
2028         if (g1->ngroups != g2->ngroups)
2029                 return false;
2030         for (i=0; i<g1->ngroups; i++)
2031                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2032                         return false;
2033         return true;
2034 }
2035
2036 /*
2037  * RFC 3530 language requires clid_inuse be returned when the
2038  * "principal" associated with a requests differs from that previously
2039  * used.  We use uid, gid's, and gss principal string as our best
2040  * approximation.  We also don't want to allow non-gss use of a client
2041  * established using gss: in theory cr_principal should catch that
2042  * change, but in practice cr_principal can be null even in the gss case
2043  * since gssd doesn't always pass down a principal string.
2044  */
2045 static bool is_gss_cred(struct svc_cred *cr)
2046 {
2047         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2048         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2049 }
2050
2051
2052 static bool
2053 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2054 {
2055         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2056                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2057                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2058                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2059                 return false;
2060         if (cr1->cr_principal == cr2->cr_principal)
2061                 return true;
2062         if (!cr1->cr_principal || !cr2->cr_principal)
2063                 return false;
2064         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2065 }
2066
2067 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2068 {
2069         struct svc_cred *cr = &rqstp->rq_cred;
2070         u32 service;
2071
2072         if (!cr->cr_gss_mech)
2073                 return false;
2074         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2075         return service == RPC_GSS_SVC_INTEGRITY ||
2076                service == RPC_GSS_SVC_PRIVACY;
2077 }
2078
2079 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2080 {
2081         struct svc_cred *cr = &rqstp->rq_cred;
2082
2083         if (!cl->cl_mach_cred)
2084                 return true;
2085         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2086                 return false;
2087         if (!svc_rqst_integrity_protected(rqstp))
2088                 return false;
2089         if (cl->cl_cred.cr_raw_principal)
2090                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2091                                                 cr->cr_raw_principal);
2092         if (!cr->cr_principal)
2093                 return false;
2094         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2095 }
2096
2097 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2098 {
2099         __be32 verf[2];
2100
2101         /*
2102          * This is opaque to client, so no need to byte-swap. Use
2103          * __force to keep sparse happy
2104          */
2105         verf[0] = (__force __be32)get_seconds();
2106         verf[1] = (__force __be32)nn->clverifier_counter++;
2107         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2108 }
2109
2110 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2111 {
2112         clp->cl_clientid.cl_boot = nn->boot_time;
2113         clp->cl_clientid.cl_id = nn->clientid_counter++;
2114         gen_confirm(clp, nn);
2115 }
2116
2117 static struct nfs4_stid *
2118 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2119 {
2120         struct nfs4_stid *ret;
2121
2122         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2123         if (!ret || !ret->sc_type)
2124                 return NULL;
2125         return ret;
2126 }
2127
2128 static struct nfs4_stid *
2129 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2130 {
2131         struct nfs4_stid *s;
2132
2133         spin_lock(&cl->cl_lock);
2134         s = find_stateid_locked(cl, t);
2135         if (s != NULL) {
2136                 if (typemask & s->sc_type)
2137                         refcount_inc(&s->sc_count);
2138                 else
2139                         s = NULL;
2140         }
2141         spin_unlock(&cl->cl_lock);
2142         return s;
2143 }
2144
2145 static struct nfs4_client *create_client(struct xdr_netobj name,
2146                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2147 {
2148         struct nfs4_client *clp;
2149         struct sockaddr *sa = svc_addr(rqstp);
2150         int ret;
2151         struct net *net = SVC_NET(rqstp);
2152
2153         clp = alloc_client(name);
2154         if (clp == NULL)
2155                 return NULL;
2156
2157         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2158         if (ret) {
2159                 free_client(clp);
2160                 return NULL;
2161         }
2162         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2163         clp->cl_time = get_seconds();
2164         clear_bit(0, &clp->cl_cb_slot_busy);
2165         copy_verf(clp, verf);
2166         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2167         clp->cl_cb_session = NULL;
2168         clp->net = net;
2169         return clp;
2170 }
2171
2172 static void
2173 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2174 {
2175         struct rb_node **new = &(root->rb_node), *parent = NULL;
2176         struct nfs4_client *clp;
2177
2178         while (*new) {
2179                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2180                 parent = *new;
2181
2182                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2183                         new = &((*new)->rb_left);
2184                 else
2185                         new = &((*new)->rb_right);
2186         }
2187
2188         rb_link_node(&new_clp->cl_namenode, parent, new);
2189         rb_insert_color(&new_clp->cl_namenode, root);
2190 }
2191
2192 static struct nfs4_client *
2193 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2194 {
2195         int cmp;
2196         struct rb_node *node = root->rb_node;
2197         struct nfs4_client *clp;
2198
2199         while (node) {
2200                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2201                 cmp = compare_blob(&clp->cl_name, name);
2202                 if (cmp > 0)
2203                         node = node->rb_left;
2204                 else if (cmp < 0)
2205                         node = node->rb_right;
2206                 else
2207                         return clp;
2208         }
2209         return NULL;
2210 }
2211
2212 static void
2213 add_to_unconfirmed(struct nfs4_client *clp)
2214 {
2215         unsigned int idhashval;
2216         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2217
2218         lockdep_assert_held(&nn->client_lock);
2219
2220         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2221         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2222         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2223         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2224         renew_client_locked(clp);
2225 }
2226
2227 static void
2228 move_to_confirmed(struct nfs4_client *clp)
2229 {
2230         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2231         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2232
2233         lockdep_assert_held(&nn->client_lock);
2234
2235         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2236         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2237         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2238         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2239         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2240         renew_client_locked(clp);
2241 }
2242
2243 static struct nfs4_client *
2244 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2245 {
2246         struct nfs4_client *clp;
2247         unsigned int idhashval = clientid_hashval(clid->cl_id);
2248
2249         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2250                 if (same_clid(&clp->cl_clientid, clid)) {
2251                         if ((bool)clp->cl_minorversion != sessions)
2252                                 return NULL;
2253                         renew_client_locked(clp);
2254                         return clp;
2255                 }
2256         }
2257         return NULL;
2258 }
2259
2260 static struct nfs4_client *
2261 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2262 {
2263         struct list_head *tbl = nn->conf_id_hashtbl;
2264
2265         lockdep_assert_held(&nn->client_lock);
2266         return find_client_in_id_table(tbl, clid, sessions);
2267 }
2268
2269 static struct nfs4_client *
2270 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2271 {
2272         struct list_head *tbl = nn->unconf_id_hashtbl;
2273
2274         lockdep_assert_held(&nn->client_lock);
2275         return find_client_in_id_table(tbl, clid, sessions);
2276 }
2277
2278 static bool clp_used_exchangeid(struct nfs4_client *clp)
2279 {
2280         return clp->cl_exchange_flags != 0;
2281 }
2282
2283 static struct nfs4_client *
2284 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2285 {
2286         lockdep_assert_held(&nn->client_lock);
2287         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2288 }
2289
2290 static struct nfs4_client *
2291 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2292 {
2293         lockdep_assert_held(&nn->client_lock);
2294         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2295 }
2296
2297 static void
2298 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2299 {
2300         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2301         struct sockaddr *sa = svc_addr(rqstp);
2302         u32 scopeid = rpc_get_scope_id(sa);
2303         unsigned short expected_family;
2304
2305         /* Currently, we only support tcp and tcp6 for the callback channel */
2306         if (se->se_callback_netid_len == 3 &&
2307             !memcmp(se->se_callback_netid_val, "tcp", 3))
2308                 expected_family = AF_INET;
2309         else if (se->se_callback_netid_len == 4 &&
2310                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2311                 expected_family = AF_INET6;
2312         else
2313                 goto out_err;
2314
2315         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2316                                             se->se_callback_addr_len,
2317                                             (struct sockaddr *)&conn->cb_addr,
2318                                             sizeof(conn->cb_addr));
2319
2320         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2321                 goto out_err;
2322
2323         if (conn->cb_addr.ss_family == AF_INET6)
2324                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2325
2326         conn->cb_prog = se->se_callback_prog;
2327         conn->cb_ident = se->se_callback_ident;
2328         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2329         return;
2330 out_err:
2331         conn->cb_addr.ss_family = AF_UNSPEC;
2332         conn->cb_addrlen = 0;
2333         dprintk("NFSD: this client (clientid %08x/%08x) "
2334                 "will not receive delegations\n",
2335                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2336
2337         return;
2338 }
2339
2340 /*
2341  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2342  */
2343 static void
2344 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2345 {
2346         struct xdr_buf *buf = resp->xdr.buf;
2347         struct nfsd4_slot *slot = resp->cstate.slot;
2348         unsigned int base;
2349
2350         dprintk("--> %s slot %p\n", __func__, slot);
2351
2352         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2353         slot->sl_opcnt = resp->opcnt;
2354         slot->sl_status = resp->cstate.status;
2355         free_svc_cred(&slot->sl_cred);
2356         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2357
2358         if (!nfsd4_cache_this(resp)) {
2359                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2360                 return;
2361         }
2362         slot->sl_flags |= NFSD4_SLOT_CACHED;
2363
2364         base = resp->cstate.data_offset;
2365         slot->sl_datalen = buf->len - base;
2366         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2367                 WARN(1, "%s: sessions DRC could not cache compound\n",
2368                      __func__);
2369         return;
2370 }
2371
2372 /*
2373  * Encode the replay sequence operation from the slot values.
2374  * If cachethis is FALSE encode the uncached rep error on the next
2375  * operation which sets resp->p and increments resp->opcnt for
2376  * nfs4svc_encode_compoundres.
2377  *
2378  */
2379 static __be32
2380 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2381                           struct nfsd4_compoundres *resp)
2382 {
2383         struct nfsd4_op *op;
2384         struct nfsd4_slot *slot = resp->cstate.slot;
2385
2386         /* Encode the replayed sequence operation */
2387         op = &args->ops[resp->opcnt - 1];
2388         nfsd4_encode_operation(resp, op);
2389
2390         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2391                 return op->status;
2392         if (args->opcnt == 1) {
2393                 /*
2394                  * The original operation wasn't a solo sequence--we
2395                  * always cache those--so this retry must not match the
2396                  * original:
2397                  */
2398                 op->status = nfserr_seq_false_retry;
2399         } else {
2400                 op = &args->ops[resp->opcnt++];
2401                 op->status = nfserr_retry_uncached_rep;
2402                 nfsd4_encode_operation(resp, op);
2403         }
2404         return op->status;
2405 }
2406
2407 /*
2408  * The sequence operation is not cached because we can use the slot and
2409  * session values.
2410  */
2411 static __be32
2412 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2413                          struct nfsd4_sequence *seq)
2414 {
2415         struct nfsd4_slot *slot = resp->cstate.slot;
2416         struct xdr_stream *xdr = &resp->xdr;
2417         __be32 *p;
2418         __be32 status;
2419
2420         dprintk("--> %s slot %p\n", __func__, slot);
2421
2422         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2423         if (status)
2424                 return status;
2425
2426         p = xdr_reserve_space(xdr, slot->sl_datalen);
2427         if (!p) {
2428                 WARN_ON_ONCE(1);
2429                 return nfserr_serverfault;
2430         }
2431         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2432         xdr_commit_encode(xdr);
2433
2434         resp->opcnt = slot->sl_opcnt;
2435         return slot->sl_status;
2436 }
2437
2438 /*
2439  * Set the exchange_id flags returned by the server.
2440  */
2441 static void
2442 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2443 {
2444 #ifdef CONFIG_NFSD_PNFS
2445         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2446 #else
2447         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2448 #endif
2449
2450         /* Referrals are supported, Migration is not. */
2451         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2452
2453         /* set the wire flags to return to client. */
2454         clid->flags = new->cl_exchange_flags;
2455 }
2456
2457 static bool client_has_openowners(struct nfs4_client *clp)
2458 {
2459         struct nfs4_openowner *oo;
2460
2461         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2462                 if (!list_empty(&oo->oo_owner.so_stateids))
2463                         return true;
2464         }
2465         return false;
2466 }
2467
2468 static bool client_has_state(struct nfs4_client *clp)
2469 {
2470         return client_has_openowners(clp)
2471 #ifdef CONFIG_NFSD_PNFS
2472                 || !list_empty(&clp->cl_lo_states)
2473 #endif
2474                 || !list_empty(&clp->cl_delegations)
2475                 || !list_empty(&clp->cl_sessions);
2476 }
2477
2478 __be32
2479 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2480                 union nfsd4_op_u *u)
2481 {
2482         struct nfsd4_exchange_id *exid = &u->exchange_id;
2483         struct nfs4_client *conf, *new;
2484         struct nfs4_client *unconf = NULL;
2485         __be32 status;
2486         char                    addr_str[INET6_ADDRSTRLEN];
2487         nfs4_verifier           verf = exid->verifier;
2488         struct sockaddr         *sa = svc_addr(rqstp);
2489         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2490         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2491
2492         rpc_ntop(sa, addr_str, sizeof(addr_str));
2493         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2494                 "ip_addr=%s flags %x, spa_how %d\n",
2495                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2496                 addr_str, exid->flags, exid->spa_how);
2497
2498         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2499                 return nfserr_inval;
2500
2501         new = create_client(exid->clname, rqstp, &verf);
2502         if (new == NULL)
2503                 return nfserr_jukebox;
2504
2505         switch (exid->spa_how) {
2506         case SP4_MACH_CRED:
2507                 exid->spo_must_enforce[0] = 0;
2508                 exid->spo_must_enforce[1] = (
2509                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2510                         1 << (OP_EXCHANGE_ID - 32) |
2511                         1 << (OP_CREATE_SESSION - 32) |
2512                         1 << (OP_DESTROY_SESSION - 32) |
2513                         1 << (OP_DESTROY_CLIENTID - 32));
2514
2515                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2516                                         1 << (OP_OPEN_DOWNGRADE) |
2517                                         1 << (OP_LOCKU) |
2518                                         1 << (OP_DELEGRETURN));
2519
2520                 exid->spo_must_allow[1] &= (
2521                                         1 << (OP_TEST_STATEID - 32) |
2522                                         1 << (OP_FREE_STATEID - 32));
2523                 if (!svc_rqst_integrity_protected(rqstp)) {
2524                         status = nfserr_inval;
2525                         goto out_nolock;
2526                 }
2527                 /*
2528                  * Sometimes userspace doesn't give us a principal.
2529                  * Which is a bug, really.  Anyway, we can't enforce
2530                  * MACH_CRED in that case, better to give up now:
2531                  */
2532                 if (!new->cl_cred.cr_principal &&
2533                                         !new->cl_cred.cr_raw_principal) {
2534                         status = nfserr_serverfault;
2535                         goto out_nolock;
2536                 }
2537                 new->cl_mach_cred = true;
2538         case SP4_NONE:
2539                 break;
2540         default:                                /* checked by xdr code */
2541                 WARN_ON_ONCE(1);
2542         case SP4_SSV:
2543                 status = nfserr_encr_alg_unsupp;
2544                 goto out_nolock;
2545         }
2546
2547         /* Cases below refer to rfc 5661 section 18.35.4: */
2548         spin_lock(&nn->client_lock);
2549         conf = find_confirmed_client_by_name(&exid->clname, nn);
2550         if (conf) {
2551                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2552                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2553
2554                 if (update) {
2555                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2556                                 status = nfserr_inval;
2557                                 goto out;
2558                         }
2559                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2560                                 status = nfserr_wrong_cred;
2561                                 goto out;
2562                         }
2563                         if (!creds_match) { /* case 9 */
2564                                 status = nfserr_perm;
2565                                 goto out;
2566                         }
2567                         if (!verfs_match) { /* case 8 */
2568                                 status = nfserr_not_same;
2569                                 goto out;
2570                         }
2571                         /* case 6 */
2572                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2573                         goto out_copy;
2574                 }
2575                 if (!creds_match) { /* case 3 */
2576                         if (client_has_state(conf)) {
2577                                 status = nfserr_clid_inuse;
2578                                 goto out;
2579                         }
2580                         goto out_new;
2581                 }
2582                 if (verfs_match) { /* case 2 */
2583                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2584                         goto out_copy;
2585                 }
2586                 /* case 5, client reboot */
2587                 conf = NULL;
2588                 goto out_new;
2589         }
2590
2591         if (update) { /* case 7 */
2592                 status = nfserr_noent;
2593                 goto out;
2594         }
2595
2596         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2597         if (unconf) /* case 4, possible retry or client restart */
2598                 unhash_client_locked(unconf);
2599
2600         /* case 1 (normal case) */
2601 out_new:
2602         if (conf) {
2603                 status = mark_client_expired_locked(conf);
2604                 if (status)
2605                         goto out;
2606         }
2607         new->cl_minorversion = cstate->minorversion;
2608         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2609         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2610
2611         gen_clid(new, nn);
2612         add_to_unconfirmed(new);
2613         swap(new, conf);
2614 out_copy:
2615         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2616         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2617
2618         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2619         nfsd4_set_ex_flags(conf, exid);
2620
2621         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2622                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2623         status = nfs_ok;
2624
2625 out:
2626         spin_unlock(&nn->client_lock);
2627 out_nolock:
2628         if (new)
2629                 expire_client(new);
2630         if (unconf)
2631                 expire_client(unconf);
2632         return status;
2633 }
2634
2635 static __be32
2636 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2637 {
2638         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2639                 slot_seqid);
2640
2641         /* The slot is in use, and no response has been sent. */
2642         if (slot_inuse) {
2643                 if (seqid == slot_seqid)
2644                         return nfserr_jukebox;
2645                 else
2646                         return nfserr_seq_misordered;
2647         }
2648         /* Note unsigned 32-bit arithmetic handles wraparound: */
2649         if (likely(seqid == slot_seqid + 1))
2650                 return nfs_ok;
2651         if (seqid == slot_seqid)
2652                 return nfserr_replay_cache;
2653         return nfserr_seq_misordered;
2654 }
2655
2656 /*
2657  * Cache the create session result into the create session single DRC
2658  * slot cache by saving the xdr structure. sl_seqid has been set.
2659  * Do this for solo or embedded create session operations.
2660  */
2661 static void
2662 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2663                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2664 {
2665         slot->sl_status = nfserr;
2666         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2667 }
2668
2669 static __be32
2670 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2671                             struct nfsd4_clid_slot *slot)
2672 {
2673         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2674         return slot->sl_status;
2675 }
2676
2677 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2678                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2679                         1 +     /* MIN tag is length with zero, only length */ \
2680                         3 +     /* version, opcount, opcode */ \
2681                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2682                                 /* seqid, slotID, slotID, cache */ \
2683                         4 ) * sizeof(__be32))
2684
2685 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2686                         2 +     /* verifier: AUTH_NULL, length 0 */\
2687                         1 +     /* status */ \
2688                         1 +     /* MIN tag is length with zero, only length */ \
2689                         3 +     /* opcount, opcode, opstatus*/ \
2690                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2691                                 /* seqid, slotID, slotID, slotID, status */ \
2692                         5 ) * sizeof(__be32))
2693
2694 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2695 {
2696         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2697
2698         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2699                 return nfserr_toosmall;
2700         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2701                 return nfserr_toosmall;
2702         ca->headerpadsz = 0;
2703         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2704         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2705         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2706         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2707                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2708         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2709         /*
2710          * Note decreasing slot size below client's request may make it
2711          * difficult for client to function correctly, whereas
2712          * decreasing the number of slots will (just?) affect
2713          * performance.  When short on memory we therefore prefer to
2714          * decrease number of slots instead of their size.  Clients that
2715          * request larger slots than they need will get poor results:
2716          */
2717         ca->maxreqs = nfsd4_get_drc_mem(ca);
2718         if (!ca->maxreqs)
2719                 return nfserr_jukebox;
2720
2721         return nfs_ok;
2722 }
2723
2724 /*
2725  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2726  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2727  */
2728 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2729         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2730
2731 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2732         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2733
2734 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2735                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2736 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2737                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2738                                  sizeof(__be32))
2739
2740 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2741 {
2742         ca->headerpadsz = 0;
2743
2744         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2745                 return nfserr_toosmall;
2746         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2747                 return nfserr_toosmall;
2748         ca->maxresp_cached = 0;
2749         if (ca->maxops < 2)
2750                 return nfserr_toosmall;
2751
2752         return nfs_ok;
2753 }
2754
2755 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2756 {
2757         switch (cbs->flavor) {
2758         case RPC_AUTH_NULL:
2759         case RPC_AUTH_UNIX:
2760                 return nfs_ok;
2761         default:
2762                 /*
2763                  * GSS case: the spec doesn't allow us to return this
2764                  * error.  But it also doesn't allow us not to support
2765                  * GSS.
2766                  * I'd rather this fail hard than return some error the
2767                  * client might think it can already handle:
2768                  */
2769                 return nfserr_encr_alg_unsupp;
2770         }
2771 }
2772
2773 __be32
2774 nfsd4_create_session(struct svc_rqst *rqstp,
2775                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2776 {
2777         struct nfsd4_create_session *cr_ses = &u->create_session;
2778         struct sockaddr *sa = svc_addr(rqstp);
2779         struct nfs4_client *conf, *unconf;
2780         struct nfs4_client *old = NULL;
2781         struct nfsd4_session *new;
2782         struct nfsd4_conn *conn;
2783         struct nfsd4_clid_slot *cs_slot = NULL;
2784         __be32 status = 0;
2785         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2786
2787         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2788                 return nfserr_inval;
2789         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2790         if (status)
2791                 return status;
2792         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2793         if (status)
2794                 return status;
2795         status = check_backchannel_attrs(&cr_ses->back_channel);
2796         if (status)
2797                 goto out_release_drc_mem;
2798         status = nfserr_jukebox;
2799         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2800         if (!new)
2801                 goto out_release_drc_mem;
2802         conn = alloc_conn_from_crses(rqstp, cr_ses);
2803         if (!conn)
2804                 goto out_free_session;
2805
2806         spin_lock(&nn->client_lock);
2807         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2808         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2809         WARN_ON_ONCE(conf && unconf);
2810
2811         if (conf) {
2812                 status = nfserr_wrong_cred;
2813                 if (!nfsd4_mach_creds_match(conf, rqstp))
2814                         goto out_free_conn;
2815                 cs_slot = &conf->cl_cs_slot;
2816                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2817                 if (status) {
2818                         if (status == nfserr_replay_cache)
2819                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2820                         goto out_free_conn;
2821                 }
2822         } else if (unconf) {
2823                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2824                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2825                         status = nfserr_clid_inuse;
2826                         goto out_free_conn;
2827                 }
2828                 status = nfserr_wrong_cred;
2829                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2830                         goto out_free_conn;
2831                 cs_slot = &unconf->cl_cs_slot;
2832                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2833                 if (status) {
2834                         /* an unconfirmed replay returns misordered */
2835                         status = nfserr_seq_misordered;
2836                         goto out_free_conn;
2837                 }
2838                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2839                 if (old) {
2840                         status = mark_client_expired_locked(old);
2841                         if (status) {
2842                                 old = NULL;
2843                                 goto out_free_conn;
2844                         }
2845                 }
2846                 move_to_confirmed(unconf);
2847                 conf = unconf;
2848         } else {
2849                 status = nfserr_stale_clientid;
2850                 goto out_free_conn;
2851         }
2852         status = nfs_ok;
2853         /* Persistent sessions are not supported */
2854         cr_ses->flags &= ~SESSION4_PERSIST;
2855         /* Upshifting from TCP to RDMA is not supported */
2856         cr_ses->flags &= ~SESSION4_RDMA;
2857
2858         init_session(rqstp, new, conf, cr_ses);
2859         nfsd4_get_session_locked(new);
2860
2861         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2862                NFS4_MAX_SESSIONID_LEN);
2863         cs_slot->sl_seqid++;
2864         cr_ses->seqid = cs_slot->sl_seqid;
2865
2866         /* cache solo and embedded create sessions under the client_lock */
2867         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2868         spin_unlock(&nn->client_lock);
2869         /* init connection and backchannel */
2870         nfsd4_init_conn(rqstp, conn, new);
2871         nfsd4_put_session(new);
2872         if (old)
2873                 expire_client(old);
2874         return status;
2875 out_free_conn:
2876         spin_unlock(&nn->client_lock);
2877         free_conn(conn);
2878         if (old)
2879                 expire_client(old);
2880 out_free_session:
2881         __free_session(new);
2882 out_release_drc_mem:
2883         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2884         return status;
2885 }
2886
2887 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2888 {
2889         switch (*dir) {
2890         case NFS4_CDFC4_FORE:
2891         case NFS4_CDFC4_BACK:
2892                 return nfs_ok;
2893         case NFS4_CDFC4_FORE_OR_BOTH:
2894         case NFS4_CDFC4_BACK_OR_BOTH:
2895                 *dir = NFS4_CDFC4_BOTH;
2896                 return nfs_ok;
2897         };
2898         return nfserr_inval;
2899 }
2900
2901 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2902                 struct nfsd4_compound_state *cstate,
2903                 union nfsd4_op_u *u)
2904 {
2905         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2906         struct nfsd4_session *session = cstate->session;
2907         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2908         __be32 status;
2909
2910         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2911         if (status)
2912                 return status;
2913         spin_lock(&nn->client_lock);
2914         session->se_cb_prog = bc->bc_cb_program;
2915         session->se_cb_sec = bc->bc_cb_sec;
2916         spin_unlock(&nn->client_lock);
2917
2918         nfsd4_probe_callback(session->se_client);
2919
2920         return nfs_ok;
2921 }
2922
2923 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2924                      struct nfsd4_compound_state *cstate,
2925                      union nfsd4_op_u *u)
2926 {
2927         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2928         __be32 status;
2929         struct nfsd4_conn *conn;
2930         struct nfsd4_session *session;
2931         struct net *net = SVC_NET(rqstp);
2932         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2933
2934         if (!nfsd4_last_compound_op(rqstp))
2935                 return nfserr_not_only_op;
2936         spin_lock(&nn->client_lock);
2937         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2938         spin_unlock(&nn->client_lock);
2939         if (!session)
2940                 goto out_no_session;
2941         status = nfserr_wrong_cred;
2942         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2943                 goto out;
2944         status = nfsd4_map_bcts_dir(&bcts->dir);
2945         if (status)
2946                 goto out;
2947         conn = alloc_conn(rqstp, bcts->dir);
2948         status = nfserr_jukebox;
2949         if (!conn)
2950                 goto out;
2951         nfsd4_init_conn(rqstp, conn, session);
2952         status = nfs_ok;
2953 out:
2954         nfsd4_put_session(session);
2955 out_no_session:
2956         return status;
2957 }
2958
2959 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2960 {
2961         if (!session)
2962                 return false;
2963         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2964 }
2965
2966 __be32
2967 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2968                 union nfsd4_op_u *u)
2969 {
2970         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2971         struct nfsd4_session *ses;
2972         __be32 status;
2973         int ref_held_by_me = 0;
2974         struct net *net = SVC_NET(r);
2975         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2976
2977         status = nfserr_not_only_op;
2978         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2979                 if (!nfsd4_last_compound_op(r))
2980                         goto out;
2981                 ref_held_by_me++;
2982         }
2983         dump_sessionid(__func__, &sessionid->sessionid);
2984         spin_lock(&nn->client_lock);
2985         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2986         if (!ses)
2987                 goto out_client_lock;
2988         status = nfserr_wrong_cred;
2989         if (!nfsd4_mach_creds_match(ses->se_client, r))
2990                 goto out_put_session;
2991         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2992         if (status)
2993                 goto out_put_session;
2994         unhash_session(ses);
2995         spin_unlock(&nn->client_lock);
2996
2997         nfsd4_probe_callback_sync(ses->se_client);
2998
2999         spin_lock(&nn->client_lock);
3000         status = nfs_ok;
3001 out_put_session:
3002         nfsd4_put_session_locked(ses);
3003 out_client_lock:
3004         spin_unlock(&nn->client_lock);
3005 out:
3006         return status;
3007 }
3008
3009 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3010 {
3011         struct nfsd4_conn *c;
3012
3013         list_for_each_entry(c, &s->se_conns, cn_persession) {
3014                 if (c->cn_xprt == xpt) {
3015                         return c;
3016                 }
3017         }
3018         return NULL;
3019 }
3020
3021 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3022 {
3023         struct nfs4_client *clp = ses->se_client;
3024         struct nfsd4_conn *c;
3025         __be32 status = nfs_ok;
3026         int ret;
3027
3028         spin_lock(&clp->cl_lock);
3029         c = __nfsd4_find_conn(new->cn_xprt, ses);
3030         if (c)
3031                 goto out_free;
3032         status = nfserr_conn_not_bound_to_session;
3033         if (clp->cl_mach_cred)
3034                 goto out_free;
3035         __nfsd4_hash_conn(new, ses);
3036         spin_unlock(&clp->cl_lock);
3037         ret = nfsd4_register_conn(new);
3038         if (ret)
3039                 /* oops; xprt is already down: */
3040                 nfsd4_conn_lost(&new->cn_xpt_user);
3041         return nfs_ok;
3042 out_free:
3043         spin_unlock(&clp->cl_lock);
3044         free_conn(new);
3045         return status;
3046 }
3047
3048 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3049 {
3050         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3051
3052         return args->opcnt > session->se_fchannel.maxops;
3053 }
3054
3055 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3056                                   struct nfsd4_session *session)
3057 {
3058         struct xdr_buf *xb = &rqstp->rq_arg;
3059
3060         return xb->len > session->se_fchannel.maxreq_sz;
3061 }
3062
3063 static bool replay_matches_cache(struct svc_rqst *rqstp,
3064                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3065 {
3066         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3067
3068         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3069             (bool)seq->cachethis)
3070                 return false;
3071         /*
3072          * If there's an error than the reply can have fewer ops than
3073          * the call.  But if we cached a reply with *more* ops than the
3074          * call you're sending us now, then this new call is clearly not
3075          * really a replay of the old one:
3076          */
3077         if (slot->sl_opcnt < argp->opcnt)
3078                 return false;
3079         /* This is the only check explicitly called by spec: */
3080         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3081                 return false;
3082         /*
3083          * There may be more comparisons we could actually do, but the
3084          * spec doesn't require us to catch every case where the calls
3085          * don't match (that would require caching the call as well as
3086          * the reply), so we don't bother.
3087          */
3088         return true;
3089 }
3090
3091 __be32
3092 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3093                 union nfsd4_op_u *u)
3094 {
3095         struct nfsd4_sequence *seq = &u->sequence;
3096         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3097         struct xdr_stream *xdr = &resp->xdr;
3098         struct nfsd4_session *session;
3099         struct nfs4_client *clp;
3100         struct nfsd4_slot *slot;
3101         struct nfsd4_conn *conn;
3102         __be32 status;
3103         int buflen;
3104         struct net *net = SVC_NET(rqstp);
3105         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3106
3107         if (resp->opcnt != 1)
3108                 return nfserr_sequence_pos;
3109
3110         /*
3111          * Will be either used or freed by nfsd4_sequence_check_conn
3112          * below.
3113          */
3114         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3115         if (!conn)
3116                 return nfserr_jukebox;
3117
3118         spin_lock(&nn->client_lock);
3119         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3120         if (!session)
3121                 goto out_no_session;
3122         clp = session->se_client;
3123
3124         status = nfserr_too_many_ops;
3125         if (nfsd4_session_too_many_ops(rqstp, session))
3126                 goto out_put_session;
3127
3128         status = nfserr_req_too_big;
3129         if (nfsd4_request_too_big(rqstp, session))
3130                 goto out_put_session;
3131
3132         status = nfserr_badslot;
3133         if (seq->slotid >= session->se_fchannel.maxreqs)
3134                 goto out_put_session;
3135
3136         slot = session->se_slots[seq->slotid];
3137         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3138
3139         /* We do not negotiate the number of slots yet, so set the
3140          * maxslots to the session maxreqs which is used to encode
3141          * sr_highest_slotid and the sr_target_slot id to maxslots */
3142         seq->maxslots = session->se_fchannel.maxreqs;
3143
3144         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3145                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3146         if (status == nfserr_replay_cache) {
3147                 status = nfserr_seq_misordered;
3148                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3149                         goto out_put_session;
3150                 status = nfserr_seq_false_retry;
3151                 if (!replay_matches_cache(rqstp, seq, slot))
3152                         goto out_put_session;
3153                 cstate->slot = slot;
3154                 cstate->session = session;
3155                 cstate->clp = clp;
3156                 /* Return the cached reply status and set cstate->status
3157                  * for nfsd4_proc_compound processing */
3158                 status = nfsd4_replay_cache_entry(resp, seq);
3159                 cstate->status = nfserr_replay_cache;
3160                 goto out;
3161         }
3162         if (status)
3163                 goto out_put_session;
3164
3165         status = nfsd4_sequence_check_conn(conn, session);
3166         conn = NULL;
3167         if (status)
3168                 goto out_put_session;
3169
3170         buflen = (seq->cachethis) ?
3171                         session->se_fchannel.maxresp_cached :
3172                         session->se_fchannel.maxresp_sz;
3173         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3174                                     nfserr_rep_too_big;
3175         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3176                 goto out_put_session;
3177         svc_reserve(rqstp, buflen);
3178
3179         status = nfs_ok;
3180         /* Success! bump slot seqid */
3181         slot->sl_seqid = seq->seqid;
3182         slot->sl_flags |= NFSD4_SLOT_INUSE;
3183         if (seq->cachethis)
3184                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3185         else
3186                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3187
3188         cstate->slot = slot;
3189         cstate->session = session;
3190         cstate->clp = clp;
3191
3192 out:
3193         switch (clp->cl_cb_state) {
3194         case NFSD4_CB_DOWN:
3195                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3196                 break;
3197         case NFSD4_CB_FAULT:
3198                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3199                 break;
3200         default:
3201                 seq->status_flags = 0;
3202         }
3203         if (!list_empty(&clp->cl_revoked))
3204                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3205 out_no_session:
3206         if (conn)
3207                 free_conn(conn);
3208         spin_unlock(&nn->client_lock);
3209         return status;
3210 out_put_session:
3211         nfsd4_put_session_locked(session);
3212         goto out_no_session;
3213 }
3214
3215 void
3216 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3217 {
3218         struct nfsd4_compound_state *cs = &resp->cstate;
3219
3220         if (nfsd4_has_session(cs)) {
3221                 if (cs->status != nfserr_replay_cache) {
3222                         nfsd4_store_cache_entry(resp);
3223                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3224                 }
3225                 /* Drop session reference that was taken in nfsd4_sequence() */
3226                 nfsd4_put_session(cs->session);
3227         } else if (cs->clp)
3228                 put_client_renew(cs->clp);
3229 }
3230
3231 __be32
3232 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3233                 struct nfsd4_compound_state *cstate,
3234                 union nfsd4_op_u *u)
3235 {
3236         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3237         struct nfs4_client *conf, *unconf;
3238         struct nfs4_client *clp = NULL;
3239         __be32 status = 0;
3240         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3241
3242         spin_lock(&nn->client_lock);
3243         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3244         conf = find_confirmed_client(&dc->clientid, true, nn);
3245         WARN_ON_ONCE(conf && unconf);
3246
3247         if (conf) {
3248                 if (client_has_state(conf)) {
3249                         status = nfserr_clientid_busy;
3250                         goto out;
3251                 }
3252                 status = mark_client_expired_locked(conf);
3253                 if (status)
3254                         goto out;
3255                 clp = conf;
3256         } else if (unconf)
3257                 clp = unconf;
3258         else {
3259                 status = nfserr_stale_clientid;
3260                 goto out;
3261         }
3262         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3263                 clp = NULL;
3264                 status = nfserr_wrong_cred;
3265                 goto out;
3266         }
3267         unhash_client_locked(clp);
3268 out:
3269         spin_unlock(&nn->client_lock);
3270         if (clp)
3271                 expire_client(clp);
3272         return status;
3273 }
3274
3275 __be32
3276 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3277                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3278 {
3279         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3280         __be32 status = 0;
3281
3282         if (rc->rca_one_fs) {
3283                 if (!cstate->current_fh.fh_dentry)
3284                         return nfserr_nofilehandle;
3285                 /*
3286                  * We don't take advantage of the rca_one_fs case.
3287                  * That's OK, it's optional, we can safely ignore it.
3288                  */
3289                 return nfs_ok;
3290         }
3291
3292         status = nfserr_complete_already;
3293         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3294                              &cstate->session->se_client->cl_flags))
3295                 goto out;
3296
3297         status = nfserr_stale_clientid;
3298         if (is_client_expired(cstate->session->se_client))
3299                 /*
3300                  * The following error isn't really legal.
3301                  * But we only get here if the client just explicitly
3302                  * destroyed the client.  Surely it no longer cares what
3303                  * error it gets back on an operation for the dead
3304                  * client.
3305                  */
3306                 goto out;
3307
3308         status = nfs_ok;
3309         nfsd4_client_record_create(cstate->session->se_client);
3310 out:
3311         return status;
3312 }
3313
3314 __be32
3315 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3316                   union nfsd4_op_u *u)
3317 {
3318         struct nfsd4_setclientid *setclid = &u->setclientid;
3319         struct xdr_netobj       clname = setclid->se_name;
3320         nfs4_verifier           clverifier = setclid->se_verf;
3321         struct nfs4_client      *conf, *new;
3322         struct nfs4_client      *unconf = NULL;
3323         __be32                  status;
3324         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3325
3326         new = create_client(clname, rqstp, &clverifier);
3327         if (new == NULL)
3328                 return nfserr_jukebox;
3329         /* Cases below refer to rfc 3530 section 14.2.33: */
3330         spin_lock(&nn->client_lock);
3331         conf = find_confirmed_client_by_name(&clname, nn);
3332         if (conf && client_has_state(conf)) {
3333                 /* case 0: */
3334                 status = nfserr_clid_inuse;
3335                 if (clp_used_exchangeid(conf))
3336                         goto out;
3337                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3338                         char addr_str[INET6_ADDRSTRLEN];
3339                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3340                                  sizeof(addr_str));
3341                         dprintk("NFSD: setclientid: string in use by client "
3342                                 "at %s\n", addr_str);
3343                         goto out;
3344                 }
3345         }
3346         unconf = find_unconfirmed_client_by_name(&clname, nn);
3347         if (unconf)
3348                 unhash_client_locked(unconf);
3349         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3350                 /* case 1: probable callback update */
3351                 copy_clid(new, conf);
3352                 gen_confirm(new, nn);
3353         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3354                 gen_clid(new, nn);
3355         new->cl_minorversion = 0;
3356         gen_callback(new, setclid, rqstp);
3357         add_to_unconfirmed(new);
3358         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3359         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3360         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3361         new = NULL;
3362         status = nfs_ok;
3363 out:
3364         spin_unlock(&nn->client_lock);
3365         if (new)
3366                 free_client(new);
3367         if (unconf)
3368                 expire_client(unconf);
3369         return status;
3370 }
3371
3372
3373 __be32
3374 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3375                         struct nfsd4_compound_state *cstate,
3376                         union nfsd4_op_u *u)
3377 {
3378         struct nfsd4_setclientid_confirm *setclientid_confirm =
3379                         &u->setclientid_confirm;
3380         struct nfs4_client *conf, *unconf;
3381         struct nfs4_client *old = NULL;
3382         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3383         clientid_t * clid = &setclientid_confirm->sc_clientid;
3384         __be32 status;
3385         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3386
3387         if (STALE_CLIENTID(clid, nn))
3388                 return nfserr_stale_clientid;
3389
3390         spin_lock(&nn->client_lock);
3391         conf = find_confirmed_client(clid, false, nn);
3392         unconf = find_unconfirmed_client(clid, false, nn);
3393         /*
3394          * We try hard to give out unique clientid's, so if we get an
3395          * attempt to confirm the same clientid with a different cred,
3396          * the client may be buggy; this should never happen.
3397          *
3398          * Nevertheless, RFC 7530 recommends INUSE for this case:
3399          */
3400         status = nfserr_clid_inuse;
3401         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3402                 goto out;
3403         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3404                 goto out;
3405         /* cases below refer to rfc 3530 section 14.2.34: */
3406         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3407                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3408                         /* case 2: probable retransmit */
3409                         status = nfs_ok;
3410                 } else /* case 4: client hasn't noticed we rebooted yet? */
3411                         status = nfserr_stale_clientid;
3412                 goto out;
3413         }
3414         status = nfs_ok;
3415         if (conf) { /* case 1: callback update */
3416                 old = unconf;
3417                 unhash_client_locked(old);
3418                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3419         } else { /* case 3: normal case; new or rebooted client */
3420                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3421                 if (old) {
3422                         status = nfserr_clid_inuse;
3423                         if (client_has_state(old)
3424                                         && !same_creds(&unconf->cl_cred,
3425                                                         &old->cl_cred))
3426                                 goto out;
3427                         status = mark_client_expired_locked(old);
3428                         if (status) {
3429                                 old = NULL;
3430                                 goto out;
3431                         }
3432                 }
3433                 move_to_confirmed(unconf);
3434                 conf = unconf;
3435         }
3436         get_client_locked(conf);
3437         spin_unlock(&nn->client_lock);
3438         nfsd4_probe_callback(conf);
3439         spin_lock(&nn->client_lock);
3440         put_client_renew_locked(conf);
3441 out:
3442         spin_unlock(&nn->client_lock);
3443         if (old)
3444                 expire_client(old);
3445         return status;
3446 }
3447
3448 static struct nfs4_file *nfsd4_alloc_file(void)
3449 {
3450         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3451 }
3452
3453 /* OPEN Share state helper functions */
3454 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3455                                 struct nfs4_file *fp)
3456 {
3457         lockdep_assert_held(&state_lock);
3458
3459         refcount_set(&fp->fi_ref, 1);
3460         spin_lock_init(&fp->fi_lock);
3461         INIT_LIST_HEAD(&fp->fi_stateids);
3462         INIT_LIST_HEAD(&fp->fi_delegations);
3463         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3464         fh_copy_shallow(&fp->fi_fhandle, fh);
3465         fp->fi_deleg_file = NULL;
3466         fp->fi_had_conflict = false;
3467         fp->fi_share_deny = 0;
3468         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3469         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3470 #ifdef CONFIG_NFSD_PNFS
3471         INIT_LIST_HEAD(&fp->fi_lo_states);
3472         atomic_set(&fp->fi_lo_recalls, 0);
3473 #endif
3474         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3475 }
3476
3477 void
3478 nfsd4_free_slabs(void)
3479 {
3480         kmem_cache_destroy(client_slab);
3481         kmem_cache_destroy(openowner_slab);
3482         kmem_cache_destroy(lockowner_slab);
3483         kmem_cache_destroy(file_slab);
3484         kmem_cache_destroy(stateid_slab);
3485         kmem_cache_destroy(deleg_slab);
3486         kmem_cache_destroy(odstate_slab);
3487 }
3488
3489 int
3490 nfsd4_init_slabs(void)
3491 {
3492         client_slab = kmem_cache_create("nfsd4_clients",
3493                         sizeof(struct nfs4_client), 0, 0, NULL);
3494         if (client_slab == NULL)
3495                 goto out;
3496         openowner_slab = kmem_cache_create("nfsd4_openowners",
3497                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3498         if (openowner_slab == NULL)
3499                 goto out_free_client_slab;
3500         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3501                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3502         if (lockowner_slab == NULL)
3503                 goto out_free_openowner_slab;
3504         file_slab = kmem_cache_create("nfsd4_files",
3505                         sizeof(struct nfs4_file), 0, 0, NULL);
3506         if (file_slab == NULL)
3507                 goto out_free_lockowner_slab;
3508         stateid_slab = kmem_cache_create("nfsd4_stateids",
3509                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3510         if (stateid_slab == NULL)
3511                 goto out_free_file_slab;
3512         deleg_slab = kmem_cache_create("nfsd4_delegations",
3513                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3514         if (deleg_slab == NULL)
3515                 goto out_free_stateid_slab;
3516         odstate_slab = kmem_cache_create("nfsd4_odstate",
3517                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3518         if (odstate_slab == NULL)
3519                 goto out_free_deleg_slab;
3520         return 0;
3521
3522 out_free_deleg_slab:
3523         kmem_cache_destroy(deleg_slab);
3524 out_free_stateid_slab:
3525         kmem_cache_destroy(stateid_slab);
3526 out_free_file_slab:
3527         kmem_cache_destroy(file_slab);
3528 out_free_lockowner_slab:
3529         kmem_cache_destroy(lockowner_slab);
3530 out_free_openowner_slab:
3531         kmem_cache_destroy(openowner_slab);
3532 out_free_client_slab:
3533         kmem_cache_destroy(client_slab);
3534 out:
3535         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3536         return -ENOMEM;
3537 }
3538
3539 static void init_nfs4_replay(struct nfs4_replay *rp)
3540 {
3541         rp->rp_status = nfserr_serverfault;
3542         rp->rp_buflen = 0;
3543         rp->rp_buf = rp->rp_ibuf;
3544         mutex_init(&rp->rp_mutex);
3545 }
3546
3547 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3548                 struct nfs4_stateowner *so)
3549 {
3550         if (!nfsd4_has_session(cstate)) {
3551                 mutex_lock(&so->so_replay.rp_mutex);
3552                 cstate->replay_owner = nfs4_get_stateowner(so);
3553         }
3554 }
3555
3556 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3557 {
3558         struct nfs4_stateowner *so = cstate->replay_owner;
3559
3560         if (so != NULL) {
3561                 cstate->replay_owner = NULL;
3562                 mutex_unlock(&so->so_replay.rp_mutex);
3563                 nfs4_put_stateowner(so);
3564         }
3565 }
3566
3567 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3568 {
3569         struct nfs4_stateowner *sop;
3570
3571         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3572         if (!sop)
3573                 return NULL;
3574
3575         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3576         if (!sop->so_owner.data) {
3577                 kmem_cache_free(slab, sop);
3578                 return NULL;
3579         }
3580         sop->so_owner.len = owner->len;
3581
3582         INIT_LIST_HEAD(&sop->so_stateids);
3583         sop->so_client = clp;
3584         init_nfs4_replay(&sop->so_replay);
3585         atomic_set(&sop->so_count, 1);
3586         return sop;
3587 }
3588
3589 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3590 {
3591         lockdep_assert_held(&clp->cl_lock);
3592
3593         list_add(&oo->oo_owner.so_strhash,
3594                  &clp->cl_ownerstr_hashtbl[strhashval]);
3595         list_add(&oo->oo_perclient, &clp->cl_openowners);
3596 }
3597
3598 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3599 {
3600         unhash_openowner_locked(openowner(so));
3601 }
3602
3603 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3604 {
3605         struct nfs4_openowner *oo = openowner(so);
3606
3607         kmem_cache_free(openowner_slab, oo);
3608 }
3609
3610 static const struct nfs4_stateowner_operations openowner_ops = {
3611         .so_unhash =    nfs4_unhash_openowner,
3612         .so_free =      nfs4_free_openowner,
3613 };
3614
3615 static struct nfs4_ol_stateid *
3616 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3617 {
3618         struct nfs4_ol_stateid *local, *ret = NULL;
3619         struct nfs4_openowner *oo = open->op_openowner;
3620
3621         lockdep_assert_held(&fp->fi_lock);
3622
3623         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3624                 /* ignore lock owners */
3625                 if (local->st_stateowner->so_is_open_owner == 0)
3626                         continue;
3627                 if (local->st_stateowner != &oo->oo_owner)
3628                         continue;
3629                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3630                         ret = local;
3631                         refcount_inc(&ret->st_stid.sc_count);
3632                         break;
3633                 }
3634         }
3635         return ret;
3636 }
3637
3638 static __be32
3639 nfsd4_verify_open_stid(struct nfs4_stid *s)
3640 {
3641         __be32 ret = nfs_ok;
3642
3643         switch (s->sc_type) {
3644         default:
3645                 break;
3646         case 0:
3647         case NFS4_CLOSED_STID:
3648         case NFS4_CLOSED_DELEG_STID:
3649                 ret = nfserr_bad_stateid;
3650                 break;
3651         case NFS4_REVOKED_DELEG_STID:
3652                 ret = nfserr_deleg_revoked;
3653         }
3654         return ret;
3655 }
3656
3657 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3658 static __be32
3659 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3660 {
3661         __be32 ret;
3662
3663         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3664         ret = nfsd4_verify_open_stid(&stp->st_stid);
3665         if (ret != nfs_ok)
3666                 mutex_unlock(&stp->st_mutex);
3667         return ret;
3668 }
3669
3670 static struct nfs4_ol_stateid *
3671 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3672 {
3673         struct nfs4_ol_stateid *stp;
3674         for (;;) {
3675                 spin_lock(&fp->fi_lock);
3676                 stp = nfsd4_find_existing_open(fp, open);
3677                 spin_unlock(&fp->fi_lock);
3678                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3679                         break;
3680                 nfs4_put_stid(&stp->st_stid);
3681         }
3682         return stp;
3683 }
3684
3685 static struct nfs4_openowner *
3686 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3687                            struct nfsd4_compound_state *cstate)
3688 {
3689         struct nfs4_client *clp = cstate->clp;
3690         struct nfs4_openowner *oo, *ret;
3691
3692         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3693         if (!oo)
3694                 return NULL;
3695         oo->oo_owner.so_ops = &openowner_ops;
3696         oo->oo_owner.so_is_open_owner = 1;
3697         oo->oo_owner.so_seqid = open->op_seqid;
3698         oo->oo_flags = 0;
3699         if (nfsd4_has_session(cstate))
3700                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3701         oo->oo_time = 0;
3702         oo->oo_last_closed_stid = NULL;
3703         INIT_LIST_HEAD(&oo->oo_close_lru);
3704         spin_lock(&clp->cl_lock);
3705         ret = find_openstateowner_str_locked(strhashval, open, clp);
3706         if (ret == NULL) {
3707                 hash_openowner(oo, clp, strhashval);
3708                 ret = oo;
3709         } else
3710                 nfs4_free_stateowner(&oo->oo_owner);
3711
3712         spin_unlock(&clp->cl_lock);
3713         return ret;
3714 }
3715
3716 static struct nfs4_ol_stateid *
3717 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3718 {
3719
3720         struct nfs4_openowner *oo = open->op_openowner;
3721         struct nfs4_ol_stateid *retstp = NULL;
3722         struct nfs4_ol_stateid *stp;
3723
3724         stp = open->op_stp;
3725         /* We are moving these outside of the spinlocks to avoid the warnings */
3726         mutex_init(&stp->st_mutex);
3727         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3728
3729 retry:
3730         spin_lock(&oo->oo_owner.so_client->cl_lock);
3731         spin_lock(&fp->fi_lock);
3732
3733         retstp = nfsd4_find_existing_open(fp, open);
3734         if (retstp)
3735                 goto out_unlock;
3736
3737         open->op_stp = NULL;
3738         refcount_inc(&stp->st_stid.sc_count);
3739         stp->st_stid.sc_type = NFS4_OPEN_STID;
3740         INIT_LIST_HEAD(&stp->st_locks);
3741         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3742         get_nfs4_file(fp);
3743         stp->st_stid.sc_file = fp;
3744         stp->st_access_bmap = 0;
3745         stp->st_deny_bmap = 0;
3746         stp->st_openstp = NULL;
3747         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3748         list_add(&stp->st_perfile, &fp->fi_stateids);
3749
3750 out_unlock:
3751         spin_unlock(&fp->fi_lock);
3752         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3753         if (retstp) {
3754                 /* Handle races with CLOSE */
3755                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3756                         nfs4_put_stid(&retstp->st_stid);
3757                         goto retry;
3758                 }
3759                 /* To keep mutex tracking happy */
3760                 mutex_unlock(&stp->st_mutex);
3761                 stp = retstp;
3762         }
3763         return stp;
3764 }
3765
3766 /*
3767  * In the 4.0 case we need to keep the owners around a little while to handle
3768  * CLOSE replay. We still do need to release any file access that is held by
3769  * them before returning however.
3770  */
3771 static void
3772 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3773 {
3774         struct nfs4_ol_stateid *last;
3775         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3776         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3777                                                 nfsd_net_id);
3778
3779         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3780
3781         /*
3782          * We know that we hold one reference via nfsd4_close, and another
3783          * "persistent" reference for the client. If the refcount is higher
3784          * than 2, then there are still calls in progress that are using this
3785          * stateid. We can't put the sc_file reference until they are finished.
3786          * Wait for the refcount to drop to 2. Since it has been unhashed,
3787          * there should be no danger of the refcount going back up again at
3788          * this point.
3789          */
3790         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3791
3792         release_all_access(s);
3793         if (s->st_stid.sc_file) {
3794                 put_nfs4_file(s->st_stid.sc_file);
3795                 s->st_stid.sc_file = NULL;
3796         }
3797
3798         spin_lock(&nn->client_lock);
3799         last = oo->oo_last_closed_stid;
3800         oo->oo_last_closed_stid = s;
3801         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3802         oo->oo_time = get_seconds();
3803         spin_unlock(&nn->client_lock);
3804         if (last)
3805                 nfs4_put_stid(&last->st_stid);
3806 }
3807
3808 /* search file_hashtbl[] for file */
3809 static struct nfs4_file *
3810 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3811 {
3812         struct nfs4_file *fp;
3813
3814         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3815                 if (fh_match(&fp->fi_fhandle, fh)) {
3816                         if (refcount_inc_not_zero(&fp->fi_ref))
3817                                 return fp;
3818                 }
3819         }
3820         return NULL;
3821 }
3822
3823 struct nfs4_file *
3824 find_file(struct knfsd_fh *fh)
3825 {
3826         struct nfs4_file *fp;
3827         unsigned int hashval = file_hashval(fh);
3828
3829         rcu_read_lock();
3830         fp = find_file_locked(fh, hashval);
3831         rcu_read_unlock();
3832         return fp;
3833 }
3834
3835 static struct nfs4_file *
3836 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3837 {
3838         struct nfs4_file *fp;
3839         unsigned int hashval = file_hashval(fh);
3840
3841         rcu_read_lock();
3842         fp = find_file_locked(fh, hashval);
3843         rcu_read_unlock();
3844         if (fp)
3845                 return fp;
3846
3847         spin_lock(&state_lock);
3848         fp = find_file_locked(fh, hashval);
3849         if (likely(fp == NULL)) {
3850                 nfsd4_init_file(fh, hashval, new);
3851                 fp = new;
3852         }
3853         spin_unlock(&state_lock);
3854
3855         return fp;
3856 }
3857
3858 /*
3859  * Called to check deny when READ with all zero stateid or
3860  * WRITE with all zero or all one stateid
3861  */
3862 static __be32
3863 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3864 {
3865         struct nfs4_file *fp;
3866         __be32 ret = nfs_ok;
3867
3868         fp = find_file(&current_fh->fh_handle);
3869         if (!fp)
3870                 return ret;
3871         /* Check for conflicting share reservations */
3872         spin_lock(&fp->fi_lock);
3873         if (fp->fi_share_deny & deny_type)
3874                 ret = nfserr_locked;
3875         spin_unlock(&fp->fi_lock);
3876         put_nfs4_file(fp);
3877         return ret;
3878 }
3879
3880 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3881 {
3882         struct nfs4_delegation *dp = cb_to_delegation(cb);
3883         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3884                                           nfsd_net_id);
3885
3886         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3887
3888         /*
3889          * We can't do this in nfsd_break_deleg_cb because it is
3890          * already holding inode->i_lock.
3891          *
3892          * If the dl_time != 0, then we know that it has already been
3893          * queued for a lease break. Don't queue it again.
3894          */
3895         spin_lock(&state_lock);
3896         if (dp->dl_time == 0) {
3897                 dp->dl_time = get_seconds();
3898                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3899         }
3900         spin_unlock(&state_lock);
3901 }
3902
3903 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3904                 struct rpc_task *task)
3905 {
3906         struct nfs4_delegation *dp = cb_to_delegation(cb);
3907
3908         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3909                 return 1;
3910
3911         switch (task->tk_status) {
3912         case 0:
3913                 return 1;
3914         case -EBADHANDLE:
3915         case -NFS4ERR_BAD_STATEID:
3916                 /*
3917                  * Race: client probably got cb_recall before open reply
3918                  * granting delegation.
3919                  */
3920                 if (dp->dl_retries--) {
3921                         rpc_delay(task, 2 * HZ);
3922                         return 0;
3923                 }
3924                 /*FALLTHRU*/
3925         default:
3926                 return -1;
3927         }
3928 }
3929
3930 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3931 {
3932         struct nfs4_delegation *dp = cb_to_delegation(cb);
3933
3934         nfs4_put_stid(&dp->dl_stid);
3935 }
3936
3937 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3938         .prepare        = nfsd4_cb_recall_prepare,
3939         .done           = nfsd4_cb_recall_done,
3940         .release        = nfsd4_cb_recall_release,
3941 };
3942
3943 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3944 {
3945         /*
3946          * We're assuming the state code never drops its reference
3947          * without first removing the lease.  Since we're in this lease
3948          * callback (and since the lease code is serialized by the kernel
3949          * lock) we know the server hasn't removed the lease yet, we know
3950          * it's safe to take a reference.
3951          */
3952         refcount_inc(&dp->dl_stid.sc_count);
3953         nfsd4_run_cb(&dp->dl_recall);
3954 }
3955
3956 /* Called from break_lease() with i_lock held. */
3957 static bool
3958 nfsd_break_deleg_cb(struct file_lock *fl)
3959 {
3960         bool ret = false;
3961         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
3962         struct nfs4_file *fp = dp->dl_stid.sc_file;
3963
3964         /*
3965          * We don't want the locks code to timeout the lease for us;
3966          * we'll remove it ourself if a delegation isn't returned
3967          * in time:
3968          */
3969         fl->fl_break_time = 0;
3970
3971         spin_lock(&fp->fi_lock);
3972         fp->fi_had_conflict = true;
3973         nfsd_break_one_deleg(dp);
3974         spin_unlock(&fp->fi_lock);
3975         return ret;
3976 }
3977
3978 static int
3979 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3980                      struct list_head *dispose)
3981 {
3982         if (arg & F_UNLCK)
3983                 return lease_modify(onlist, arg, dispose);
3984         else
3985                 return -EAGAIN;
3986 }
3987
3988 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3989         .lm_break = nfsd_break_deleg_cb,
3990         .lm_change = nfsd_change_deleg_cb,
3991 };
3992
3993 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3994 {
3995         if (nfsd4_has_session(cstate))
3996                 return nfs_ok;
3997         if (seqid == so->so_seqid - 1)
3998                 return nfserr_replay_me;
3999         if (seqid == so->so_seqid)
4000                 return nfs_ok;
4001         return nfserr_bad_seqid;
4002 }
4003
4004 static __be32 lookup_clientid(clientid_t *clid,
4005                 struct nfsd4_compound_state *cstate,
4006                 struct nfsd_net *nn)
4007 {
4008         struct nfs4_client *found;
4009
4010         if (cstate->clp) {
4011                 found = cstate->clp;
4012                 if (!same_clid(&found->cl_clientid, clid))
4013                         return nfserr_stale_clientid;
4014                 return nfs_ok;
4015         }
4016
4017         if (STALE_CLIENTID(clid, nn))
4018                 return nfserr_stale_clientid;
4019
4020         /*
4021          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4022          * cached already then we know this is for is for v4.0 and "sessions"
4023          * will be false.
4024          */
4025         WARN_ON_ONCE(cstate->session);
4026         spin_lock(&nn->client_lock);
4027         found = find_confirmed_client(clid, false, nn);
4028         if (!found) {
4029                 spin_unlock(&nn->client_lock);
4030                 return nfserr_expired;
4031         }
4032         atomic_inc(&found->cl_refcount);
4033         spin_unlock(&nn->client_lock);
4034
4035         /* Cache the nfs4_client in cstate! */
4036         cstate->clp = found;
4037         return nfs_ok;
4038 }
4039
4040 __be32
4041 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4042                     struct nfsd4_open *open, struct nfsd_net *nn)
4043 {
4044         clientid_t *clientid = &open->op_clientid;
4045         struct nfs4_client *clp = NULL;
4046         unsigned int strhashval;
4047         struct nfs4_openowner *oo = NULL;
4048         __be32 status;
4049
4050         if (STALE_CLIENTID(&open->op_clientid, nn))
4051                 return nfserr_stale_clientid;
4052         /*
4053          * In case we need it later, after we've already created the
4054          * file and don't want to risk a further failure:
4055          */
4056         open->op_file = nfsd4_alloc_file();
4057         if (open->op_file == NULL)
4058                 return nfserr_jukebox;
4059
4060         status = lookup_clientid(clientid, cstate, nn);
4061         if (status)
4062                 return status;
4063         clp = cstate->clp;
4064
4065         strhashval = ownerstr_hashval(&open->op_owner);
4066         oo = find_openstateowner_str(strhashval, open, clp);
4067         open->op_openowner = oo;
4068         if (!oo) {
4069                 goto new_owner;
4070         }
4071         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4072                 /* Replace unconfirmed owners without checking for replay. */
4073                 release_openowner(oo);
4074                 open->op_openowner = NULL;
4075                 goto new_owner;
4076         }
4077         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4078         if (status)
4079                 return status;
4080         goto alloc_stateid;
4081 new_owner:
4082         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4083         if (oo == NULL)
4084                 return nfserr_jukebox;
4085         open->op_openowner = oo;
4086 alloc_stateid:
4087         open->op_stp = nfs4_alloc_open_stateid(clp);
4088         if (!open->op_stp)
4089                 return nfserr_jukebox;
4090
4091         if (nfsd4_has_session(cstate) &&
4092             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4093                 open->op_odstate = alloc_clnt_odstate(clp);
4094                 if (!open->op_odstate)
4095                         return nfserr_jukebox;
4096         }
4097
4098         return nfs_ok;
4099 }
4100
4101 static inline __be32
4102 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4103 {
4104         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4105                 return nfserr_openmode;
4106         else
4107                 return nfs_ok;
4108 }
4109
4110 static int share_access_to_flags(u32 share_access)
4111 {
4112         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4113 }
4114
4115 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4116 {
4117         struct nfs4_stid *ret;
4118
4119         ret = find_stateid_by_type(cl, s,
4120                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4121         if (!ret)
4122                 return NULL;
4123         return delegstateid(ret);
4124 }
4125
4126 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4127 {
4128         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4129                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4130 }
4131
4132 static __be32
4133 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4134                 struct nfs4_delegation **dp)
4135 {
4136         int flags;
4137         __be32 status = nfserr_bad_stateid;
4138         struct nfs4_delegation *deleg;
4139
4140         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4141         if (deleg == NULL)
4142                 goto out;
4143         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4144                 nfs4_put_stid(&deleg->dl_stid);
4145                 if (cl->cl_minorversion)
4146                         status = nfserr_deleg_revoked;
4147                 goto out;
4148         }
4149         flags = share_access_to_flags(open->op_share_access);
4150         status = nfs4_check_delegmode(deleg, flags);
4151         if (status) {
4152                 nfs4_put_stid(&deleg->dl_stid);
4153                 goto out;
4154         }
4155         *dp = deleg;
4156 out:
4157         if (!nfsd4_is_deleg_cur(open))
4158                 return nfs_ok;
4159         if (status)
4160                 return status;
4161         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4162         return nfs_ok;
4163 }
4164
4165 static inline int nfs4_access_to_access(u32 nfs4_access)
4166 {
4167         int flags = 0;
4168
4169         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4170                 flags |= NFSD_MAY_READ;
4171         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4172                 flags |= NFSD_MAY_WRITE;
4173         return flags;
4174 }
4175
4176 static inline __be32
4177 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4178                 struct nfsd4_open *open)
4179 {
4180         struct iattr iattr = {
4181                 .ia_valid = ATTR_SIZE,
4182                 .ia_size = 0,
4183         };
4184         if (!open->op_truncate)
4185                 return 0;
4186         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4187                 return nfserr_inval;
4188         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4189 }
4190
4191 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4192                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4193                 struct nfsd4_open *open)
4194 {
4195         struct file *filp = NULL;
4196         __be32 status;
4197         int oflag = nfs4_access_to_omode(open->op_share_access);
4198         int access = nfs4_access_to_access(open->op_share_access);
4199         unsigned char old_access_bmap, old_deny_bmap;
4200
4201         spin_lock(&fp->fi_lock);
4202
4203         /*
4204          * Are we trying to set a deny mode that would conflict with
4205          * current access?
4206          */
4207         status = nfs4_file_check_deny(fp, open->op_share_deny);
4208         if (status != nfs_ok) {
4209                 spin_unlock(&fp->fi_lock);
4210                 goto out;
4211         }
4212
4213         /* set access to the file */
4214         status = nfs4_file_get_access(fp, open->op_share_access);
4215         if (status != nfs_ok) {
4216                 spin_unlock(&fp->fi_lock);
4217                 goto out;
4218         }
4219
4220         /* Set access bits in stateid */
4221         old_access_bmap = stp->st_access_bmap;
4222         set_access(open->op_share_access, stp);
4223
4224         /* Set new deny mask */
4225         old_deny_bmap = stp->st_deny_bmap;
4226         set_deny(open->op_share_deny, stp);
4227         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4228
4229         if (!fp->fi_fds[oflag]) {
4230                 spin_unlock(&fp->fi_lock);
4231                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4232                 if (status)
4233                         goto out_put_access;
4234                 spin_lock(&fp->fi_lock);
4235                 if (!fp->fi_fds[oflag]) {
4236                         fp->fi_fds[oflag] = filp;
4237                         filp = NULL;
4238                 }
4239         }
4240         spin_unlock(&fp->fi_lock);
4241         if (filp)
4242                 fput(filp);
4243
4244         status = nfsd4_truncate(rqstp, cur_fh, open);
4245         if (status)
4246                 goto out_put_access;
4247 out:
4248         return status;
4249 out_put_access:
4250         stp->st_access_bmap = old_access_bmap;
4251         nfs4_file_put_access(fp, open->op_share_access);
4252         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4253         goto out;
4254 }
4255
4256 static __be32
4257 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)
4258 {
4259         __be32 status;
4260         unsigned char old_deny_bmap = stp->st_deny_bmap;
4261
4262         if (!test_access(open->op_share_access, stp))
4263                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4264
4265         /* test and set deny mode */
4266         spin_lock(&fp->fi_lock);
4267         status = nfs4_file_check_deny(fp, open->op_share_deny);
4268         if (status == nfs_ok) {
4269                 set_deny(open->op_share_deny, stp);
4270                 fp->fi_share_deny |=
4271                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4272         }
4273         spin_unlock(&fp->fi_lock);
4274
4275         if (status != nfs_ok)
4276                 return status;
4277
4278         status = nfsd4_truncate(rqstp, cur_fh, open);
4279         if (status != nfs_ok)
4280                 reset_union_bmap_deny(old_deny_bmap, stp);
4281         return status;
4282 }
4283
4284 /* Should we give out recallable state?: */
4285 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4286 {
4287         if (clp->cl_cb_state == NFSD4_CB_UP)
4288                 return true;
4289         /*
4290          * In the sessions case, since we don't have to establish a
4291          * separate connection for callbacks, we assume it's OK
4292          * until we hear otherwise:
4293          */
4294         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4295 }
4296
4297 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4298                                                 int flag)
4299 {
4300         struct file_lock *fl;
4301
4302         fl = locks_alloc_lock();
4303         if (!fl)
4304                 return NULL;
4305         fl->fl_lmops = &nfsd_lease_mng_ops;
4306         fl->fl_flags = FL_DELEG;
4307         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4308         fl->fl_end = OFFSET_MAX;
4309         fl->fl_owner = (fl_owner_t)dp;
4310         fl->fl_pid = current->tgid;
4311         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4312         return fl;
4313 }
4314
4315 static struct nfs4_delegation *
4316 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4317                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4318 {
4319         int status = 0;
4320         struct nfs4_delegation *dp;
4321         struct file *filp;
4322         struct file_lock *fl;
4323
4324         /*
4325          * The fi_had_conflict and nfs_get_existing_delegation checks
4326          * here are just optimizations; we'll need to recheck them at
4327          * the end:
4328          */
4329         if (fp->fi_had_conflict)
4330                 return ERR_PTR(-EAGAIN);
4331
4332         filp = find_readable_file(fp);
4333         if (!filp) {
4334                 /* We should always have a readable file here */
4335                 WARN_ON_ONCE(1);
4336                 return ERR_PTR(-EBADF);
4337         }
4338         spin_lock(&state_lock);
4339         spin_lock(&fp->fi_lock);
4340         if (nfs4_delegation_exists(clp, fp))
4341                 status = -EAGAIN;
4342         else if (!fp->fi_deleg_file) {
4343                 fp->fi_deleg_file = filp;
4344                 /* increment early to prevent fi_deleg_file from being
4345                  * cleared */
4346                 fp->fi_delegees = 1;
4347                 filp = NULL;
4348         } else
4349                 fp->fi_delegees++;
4350         spin_unlock(&fp->fi_lock);
4351         spin_unlock(&state_lock);
4352         if (filp)
4353                 fput(filp);
4354         if (status)
4355                 return ERR_PTR(status);
4356
4357         status = -ENOMEM;
4358         dp = alloc_init_deleg(clp, fp, fh, odstate);
4359         if (!dp)
4360                 goto out_delegees;
4361
4362         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4363         if (!fl)
4364                 goto out_stid;
4365
4366         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4367         if (fl)
4368                 locks_free_lock(fl);
4369         if (status)
4370                 goto out_clnt_odstate;
4371
4372         spin_lock(&state_lock);
4373         spin_lock(&fp->fi_lock);
4374         if (fp->fi_had_conflict)
4375                 status = -EAGAIN;
4376         else
4377                 status = hash_delegation_locked(dp, fp);
4378         spin_unlock(&fp->fi_lock);
4379         spin_unlock(&state_lock);
4380
4381         if (status)
4382                 goto out_unlock;
4383
4384         return dp;
4385 out_unlock:
4386         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4387 out_clnt_odstate:
4388         put_clnt_odstate(dp->dl_clnt_odstate);
4389 out_stid:
4390         nfs4_put_stid(&dp->dl_stid);
4391 out_delegees:
4392         put_deleg_file(fp);
4393         return ERR_PTR(status);
4394 }
4395
4396 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4397 {
4398         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4399         if (status == -EAGAIN)
4400                 open->op_why_no_deleg = WND4_CONTENTION;
4401         else {
4402                 open->op_why_no_deleg = WND4_RESOURCE;
4403                 switch (open->op_deleg_want) {
4404                 case NFS4_SHARE_WANT_READ_DELEG:
4405                 case NFS4_SHARE_WANT_WRITE_DELEG:
4406                 case NFS4_SHARE_WANT_ANY_DELEG:
4407                         break;
4408                 case NFS4_SHARE_WANT_CANCEL:
4409                         open->op_why_no_deleg = WND4_CANCELLED;
4410                         break;
4411                 case NFS4_SHARE_WANT_NO_DELEG:
4412                         WARN_ON_ONCE(1);
4413                 }
4414         }
4415 }
4416
4417 /*
4418  * Attempt to hand out a delegation.
4419  *
4420  * Note we don't support write delegations, and won't until the vfs has
4421  * proper support for them.
4422  */
4423 static void
4424 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4425                         struct nfs4_ol_stateid *stp)
4426 {
4427         struct nfs4_delegation *dp;
4428         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4429         struct nfs4_client *clp = stp->st_stid.sc_client;
4430         int cb_up;
4431         int status = 0;
4432
4433         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4434         open->op_recall = 0;
4435         switch (open->op_claim_type) {
4436                 case NFS4_OPEN_CLAIM_PREVIOUS:
4437                         if (!cb_up)
4438                                 open->op_recall = 1;
4439                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4440                                 goto out_no_deleg;
4441                         break;
4442                 case NFS4_OPEN_CLAIM_NULL:
4443                 case NFS4_OPEN_CLAIM_FH:
4444                         /*
4445                          * Let's not give out any delegations till everyone's
4446                          * had the chance to reclaim theirs, *and* until
4447                          * NLM locks have all been reclaimed:
4448                          */
4449                         if (locks_in_grace(clp->net))
4450                                 goto out_no_deleg;
4451                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4452                                 goto out_no_deleg;
4453                         /*
4454                          * Also, if the file was opened for write or
4455                          * create, there's a good chance the client's
4456                          * about to write to it, resulting in an
4457                          * immediate recall (since we don't support
4458                          * write delegations):
4459                          */
4460                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4461                                 goto out_no_deleg;
4462                         if (open->op_create == NFS4_OPEN_CREATE)
4463                                 goto out_no_deleg;
4464                         break;
4465                 default:
4466                         goto out_no_deleg;
4467         }
4468         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4469         if (IS_ERR(dp))
4470                 goto out_no_deleg;
4471
4472         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4473
4474         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4475                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4476         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4477         nfs4_put_stid(&dp->dl_stid);
4478         return;
4479 out_no_deleg:
4480         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4481         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4482             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4483                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4484                 open->op_recall = 1;
4485         }
4486
4487         /* 4.1 client asking for a delegation? */
4488         if (open->op_deleg_want)
4489                 nfsd4_open_deleg_none_ext(open, status);
4490         return;
4491 }
4492
4493 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4494                                         struct nfs4_delegation *dp)
4495 {
4496         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4497             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4498                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4499                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4500         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4501                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4502                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4503                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4504         }
4505         /* Otherwise the client must be confused wanting a delegation
4506          * it already has, therefore we don't return
4507          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4508          */
4509 }
4510
4511 __be32
4512 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4513 {
4514         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4515         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4516         struct nfs4_file *fp = NULL;
4517         struct nfs4_ol_stateid *stp = NULL;
4518         struct nfs4_delegation *dp = NULL;
4519         __be32 status;
4520         bool new_stp = false;
4521
4522         /*
4523          * Lookup file; if found, lookup stateid and check open request,
4524          * and check for delegations in the process of being recalled.
4525          * If not found, create the nfs4_file struct
4526          */
4527         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4528         if (fp != open->op_file) {
4529                 status = nfs4_check_deleg(cl, open, &dp);
4530                 if (status)
4531                         goto out;
4532                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4533         } else {
4534                 open->op_file = NULL;
4535                 status = nfserr_bad_stateid;
4536                 if (nfsd4_is_deleg_cur(open))
4537                         goto out;
4538         }
4539
4540         if (!stp) {
4541                 stp = init_open_stateid(fp, open);
4542                 if (!open->op_stp)
4543                         new_stp = true;
4544         }
4545
4546         /*
4547          * OPEN the file, or upgrade an existing OPEN.
4548          * If truncate fails, the OPEN fails.
4549          *
4550          * stp is already locked.
4551          */
4552         if (!new_stp) {
4553                 /* Stateid was found, this is an OPEN upgrade */
4554                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4555                 if (status) {
4556                         mutex_unlock(&stp->st_mutex);
4557                         goto out;
4558                 }
4559         } else {
4560                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4561                 if (status) {
4562                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4563                         release_open_stateid(stp);
4564                         mutex_unlock(&stp->st_mutex);
4565                         goto out;
4566                 }
4567
4568                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4569                                                         open->op_odstate);
4570                 if (stp->st_clnt_odstate == open->op_odstate)
4571                         open->op_odstate = NULL;
4572         }
4573
4574         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4575         mutex_unlock(&stp->st_mutex);
4576
4577         if (nfsd4_has_session(&resp->cstate)) {
4578                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4579                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4580                         open->op_why_no_deleg = WND4_NOT_WANTED;
4581                         goto nodeleg;
4582                 }
4583         }
4584
4585         /*
4586         * Attempt to hand out a delegation. No error return, because the
4587         * OPEN succeeds even if we fail.
4588         */
4589         nfs4_open_delegation(current_fh, open, stp);
4590 nodeleg:
4591         status = nfs_ok;
4592
4593         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4594                 STATEID_VAL(&stp->st_stid.sc_stateid));
4595 out:
4596         /* 4.1 client trying to upgrade/downgrade delegation? */
4597         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4598             open->op_deleg_want)
4599                 nfsd4_deleg_xgrade_none_ext(open, dp);
4600
4601         if (fp)
4602                 put_nfs4_file(fp);
4603         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4604                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4605         /*
4606         * To finish the open response, we just need to set the rflags.
4607         */
4608         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4609         if (nfsd4_has_session(&resp->cstate))
4610                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4611         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4612                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4613
4614         if (dp)
4615                 nfs4_put_stid(&dp->dl_stid);
4616         if (stp)
4617                 nfs4_put_stid(&stp->st_stid);
4618
4619         return status;
4620 }
4621
4622 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4623                               struct nfsd4_open *open)
4624 {
4625         if (open->op_openowner) {
4626                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4627
4628                 nfsd4_cstate_assign_replay(cstate, so);
4629                 nfs4_put_stateowner(so);
4630         }
4631         if (open->op_file)
4632                 kmem_cache_free(file_slab, open->op_file);
4633         if (open->op_stp)
4634                 nfs4_put_stid(&open->op_stp->st_stid);
4635         if (open->op_odstate)
4636                 kmem_cache_free(odstate_slab, open->op_odstate);
4637 }
4638
4639 __be32
4640 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4641             union nfsd4_op_u *u)
4642 {
4643         clientid_t *clid = &u->renew;
4644         struct nfs4_client *clp;
4645         __be32 status;
4646         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4647
4648         dprintk("process_renew(%08x/%08x): starting\n",
4649                         clid->cl_boot, clid->cl_id);
4650         status = lookup_clientid(clid, cstate, nn);
4651         if (status)
4652                 goto out;
4653         clp = cstate->clp;
4654         status = nfserr_cb_path_down;
4655         if (!list_empty(&clp->cl_delegations)
4656                         && clp->cl_cb_state != NFSD4_CB_UP)
4657                 goto out;
4658         status = nfs_ok;
4659 out:
4660         return status;
4661 }
4662
4663 void
4664 nfsd4_end_grace(struct nfsd_net *nn)
4665 {
4666         /* do nothing if grace period already ended */
4667         if (nn->grace_ended)
4668                 return;
4669
4670         dprintk("NFSD: end of grace period\n");
4671         nn->grace_ended = true;
4672         /*
4673          * If the server goes down again right now, an NFSv4
4674          * client will still be allowed to reclaim after it comes back up,
4675          * even if it hasn't yet had a chance to reclaim state this time.
4676          *
4677          */
4678         nfsd4_record_grace_done(nn);
4679         /*
4680          * At this point, NFSv4 clients can still reclaim.  But if the
4681          * server crashes, any that have not yet reclaimed will be out
4682          * of luck on the next boot.
4683          *
4684          * (NFSv4.1+ clients are considered to have reclaimed once they
4685          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4686          * have reclaimed after their first OPEN.)
4687          */
4688         locks_end_grace(&nn->nfsd4_manager);
4689         /*
4690          * At this point, and once lockd and/or any other containers
4691          * exit their grace period, further reclaims will fail and
4692          * regular locking can resume.
4693          */
4694 }
4695
4696 static time_t
4697 nfs4_laundromat(struct nfsd_net *nn)
4698 {
4699         struct nfs4_client *clp;
4700         struct nfs4_openowner *oo;
4701         struct nfs4_delegation *dp;
4702         struct nfs4_ol_stateid *stp;
4703         struct nfsd4_blocked_lock *nbl;
4704         struct list_head *pos, *next, reaplist;
4705         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4706         time_t t, new_timeo = nn->nfsd4_lease;
4707
4708         dprintk("NFSD: laundromat service - starting\n");
4709         nfsd4_end_grace(nn);
4710         INIT_LIST_HEAD(&reaplist);
4711         spin_lock(&nn->client_lock);
4712         list_for_each_safe(pos, next, &nn->client_lru) {
4713                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4714                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4715                         t = clp->cl_time - cutoff;
4716                         new_timeo = min(new_timeo, t);
4717                         break;
4718                 }
4719                 if (mark_client_expired_locked(clp)) {
4720                         dprintk("NFSD: client in use (clientid %08x)\n",
4721                                 clp->cl_clientid.cl_id);
4722                         continue;
4723                 }
4724                 list_add(&clp->cl_lru, &reaplist);
4725         }
4726         spin_unlock(&nn->client_lock);
4727         list_for_each_safe(pos, next, &reaplist) {
4728                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4729                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4730                         clp->cl_clientid.cl_id);
4731                 list_del_init(&clp->cl_lru);
4732                 expire_client(clp);
4733         }
4734         spin_lock(&state_lock);
4735         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4736                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4737                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4738                         t = dp->dl_time - cutoff;
4739                         new_timeo = min(new_timeo, t);
4740                         break;
4741                 }
4742                 WARN_ON(!unhash_delegation_locked(dp));
4743                 list_add(&dp->dl_recall_lru, &reaplist);
4744         }
4745         spin_unlock(&state_lock);
4746         while (!list_empty(&reaplist)) {
4747                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4748                                         dl_recall_lru);
4749                 list_del_init(&dp->dl_recall_lru);
4750                 revoke_delegation(dp);
4751         }
4752
4753         spin_lock(&nn->client_lock);
4754         while (!list_empty(&nn->close_lru)) {
4755                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4756                                         oo_close_lru);
4757                 if (time_after((unsigned long)oo->oo_time,
4758                                (unsigned long)cutoff)) {
4759                         t = oo->oo_time - cutoff;
4760                         new_timeo = min(new_timeo, t);
4761                         break;
4762                 }
4763                 list_del_init(&oo->oo_close_lru);
4764                 stp = oo->oo_last_closed_stid;
4765                 oo->oo_last_closed_stid = NULL;
4766                 spin_unlock(&nn->client_lock);
4767                 nfs4_put_stid(&stp->st_stid);
4768                 spin_lock(&nn->client_lock);
4769         }
4770         spin_unlock(&nn->client_lock);
4771
4772         /*
4773          * It's possible for a client to try and acquire an already held lock
4774          * that is being held for a long time, and then lose interest in it.
4775          * So, we clean out any un-revisited request after a lease period
4776          * under the assumption that the client is no longer interested.
4777          *
4778          * RFC5661, sec. 9.6 states that the client must not rely on getting
4779          * notifications and must continue to poll for locks, even when the
4780          * server supports them. Thus this shouldn't lead to clients blocking
4781          * indefinitely once the lock does become free.
4782          */
4783         BUG_ON(!list_empty(&reaplist));
4784         spin_lock(&nn->blocked_locks_lock);
4785         while (!list_empty(&nn->blocked_locks_lru)) {
4786                 nbl = list_first_entry(&nn->blocked_locks_lru,
4787                                         struct nfsd4_blocked_lock, nbl_lru);
4788                 if (time_after((unsigned long)nbl->nbl_time,
4789                                (unsigned long)cutoff)) {
4790                         t = nbl->nbl_time - cutoff;
4791                         new_timeo = min(new_timeo, t);
4792                         break;
4793                 }
4794                 list_move(&nbl->nbl_lru, &reaplist);
4795                 list_del_init(&nbl->nbl_list);
4796         }
4797         spin_unlock(&nn->blocked_locks_lock);
4798
4799         while (!list_empty(&reaplist)) {
4800                 nbl = list_first_entry(&reaplist,
4801                                         struct nfsd4_blocked_lock, nbl_lru);
4802                 list_del_init(&nbl->nbl_lru);
4803                 posix_unblock_lock(&nbl->nbl_lock);
4804                 free_blocked_lock(nbl);
4805         }
4806
4807         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4808         return new_timeo;
4809 }
4810
4811 static struct workqueue_struct *laundry_wq;
4812 static void laundromat_main(struct work_struct *);
4813
4814 static void
4815 laundromat_main(struct work_struct *laundry)
4816 {
4817         time_t t;
4818         struct delayed_work *dwork = to_delayed_work(laundry);
4819         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4820                                            laundromat_work);
4821
4822         t = nfs4_laundromat(nn);
4823         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4824         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4825 }
4826
4827 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4828 {
4829         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4830                 return nfserr_bad_stateid;
4831         return nfs_ok;
4832 }
4833
4834 static inline int
4835 access_permit_read(struct nfs4_ol_stateid *stp)
4836 {
4837         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4838                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4839                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4840 }
4841
4842 static inline int
4843 access_permit_write(struct nfs4_ol_stateid *stp)
4844 {
4845         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4846                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4847 }
4848
4849 static
4850 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4851 {
4852         __be32 status = nfserr_openmode;
4853
4854         /* For lock stateid's, we test the parent open, not the lock: */
4855         if (stp->st_openstp)
4856                 stp = stp->st_openstp;
4857         if ((flags & WR_STATE) && !access_permit_write(stp))
4858                 goto out;
4859         if ((flags & RD_STATE) && !access_permit_read(stp))
4860                 goto out;
4861         status = nfs_ok;
4862 out:
4863         return status;
4864 }
4865
4866 static inline __be32
4867 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4868 {
4869         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4870                 return nfs_ok;
4871         else if (opens_in_grace(net)) {
4872                 /* Answer in remaining cases depends on existence of
4873                  * conflicting state; so we must wait out the grace period. */
4874                 return nfserr_grace;
4875         } else if (flags & WR_STATE)
4876                 return nfs4_share_conflict(current_fh,
4877                                 NFS4_SHARE_DENY_WRITE);
4878         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4879                 return nfs4_share_conflict(current_fh,
4880                                 NFS4_SHARE_DENY_READ);
4881 }
4882
4883 /*
4884  * Allow READ/WRITE during grace period on recovered state only for files
4885  * that are not able to provide mandatory locking.
4886  */
4887 static inline int
4888 grace_disallows_io(struct net *net, struct inode *inode)
4889 {
4890         return opens_in_grace(net) && mandatory_lock(inode);
4891 }
4892
4893 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4894 {
4895         /*
4896          * When sessions are used the stateid generation number is ignored
4897          * when it is zero.
4898          */
4899         if (has_session && in->si_generation == 0)
4900                 return nfs_ok;
4901
4902         if (in->si_generation == ref->si_generation)
4903                 return nfs_ok;
4904
4905         /* If the client sends us a stateid from the future, it's buggy: */
4906         if (nfsd4_stateid_generation_after(in, ref))
4907                 return nfserr_bad_stateid;
4908         /*
4909          * However, we could see a stateid from the past, even from a
4910          * non-buggy client.  For example, if the client sends a lock
4911          * while some IO is outstanding, the lock may bump si_generation
4912          * while the IO is still in flight.  The client could avoid that
4913          * situation by waiting for responses on all the IO requests,
4914          * but better performance may result in retrying IO that
4915          * receives an old_stateid error if requests are rarely
4916          * reordered in flight:
4917          */
4918         return nfserr_old_stateid;
4919 }
4920
4921 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4922 {
4923         __be32 ret;
4924
4925         spin_lock(&s->sc_lock);
4926         ret = nfsd4_verify_open_stid(s);
4927         if (ret == nfs_ok)
4928                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4929         spin_unlock(&s->sc_lock);
4930         return ret;
4931 }
4932
4933 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4934 {
4935         if (ols->st_stateowner->so_is_open_owner &&
4936             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4937                 return nfserr_bad_stateid;
4938         return nfs_ok;
4939 }
4940
4941 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4942 {
4943         struct nfs4_stid *s;
4944         __be32 status = nfserr_bad_stateid;
4945
4946         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4947                 CLOSE_STATEID(stateid))
4948                 return status;
4949         /* Client debugging aid. */
4950         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4951                 char addr_str[INET6_ADDRSTRLEN];
4952                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4953                                  sizeof(addr_str));
4954                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4955                                         "with incorrect client ID\n", addr_str);
4956                 return status;
4957         }
4958         spin_lock(&cl->cl_lock);
4959         s = find_stateid_locked(cl, stateid);
4960         if (!s)
4961                 goto out_unlock;
4962         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
4963         if (status)
4964                 goto out_unlock;
4965         switch (s->sc_type) {
4966         case NFS4_DELEG_STID:
4967                 status = nfs_ok;
4968                 break;
4969         case NFS4_REVOKED_DELEG_STID:
4970                 status = nfserr_deleg_revoked;
4971                 break;
4972         case NFS4_OPEN_STID:
4973         case NFS4_LOCK_STID:
4974                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4975                 break;
4976         default:
4977                 printk("unknown stateid type %x\n", s->sc_type);
4978                 /* Fallthrough */
4979         case NFS4_CLOSED_STID:
4980         case NFS4_CLOSED_DELEG_STID:
4981                 status = nfserr_bad_stateid;
4982         }
4983 out_unlock:
4984         spin_unlock(&cl->cl_lock);
4985         return status;
4986 }
4987
4988 __be32
4989 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4990                      stateid_t *stateid, unsigned char typemask,
4991                      struct nfs4_stid **s, struct nfsd_net *nn)
4992 {
4993         __be32 status;
4994         bool return_revoked = false;
4995
4996         /*
4997          *  only return revoked delegations if explicitly asked.
4998          *  otherwise we report revoked or bad_stateid status.
4999          */
5000         if (typemask & NFS4_REVOKED_DELEG_STID)
5001                 return_revoked = true;
5002         else if (typemask & NFS4_DELEG_STID)
5003                 typemask |= NFS4_REVOKED_DELEG_STID;
5004
5005         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5006                 CLOSE_STATEID(stateid))
5007                 return nfserr_bad_stateid;
5008         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5009         if (status == nfserr_stale_clientid) {
5010                 if (cstate->session)
5011                         return nfserr_bad_stateid;
5012                 return nfserr_stale_stateid;
5013         }
5014         if (status)
5015                 return status;
5016         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5017         if (!*s)
5018                 return nfserr_bad_stateid;
5019         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5020                 nfs4_put_stid(*s);
5021                 if (cstate->minorversion)
5022                         return nfserr_deleg_revoked;
5023                 return nfserr_bad_stateid;
5024         }
5025         return nfs_ok;
5026 }
5027
5028 static struct file *
5029 nfs4_find_file(struct nfs4_stid *s, int flags)
5030 {
5031         if (!s)
5032                 return NULL;
5033
5034         switch (s->sc_type) {
5035         case NFS4_DELEG_STID:
5036                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5037                         return NULL;
5038                 return get_file(s->sc_file->fi_deleg_file);
5039         case NFS4_OPEN_STID:
5040         case NFS4_LOCK_STID:
5041                 if (flags & RD_STATE)
5042                         return find_readable_file(s->sc_file);
5043                 else
5044                         return find_writeable_file(s->sc_file);
5045                 break;
5046         }
5047
5048         return NULL;
5049 }
5050
5051 static __be32
5052 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5053 {
5054         __be32 status;
5055
5056         status = nfsd4_check_openowner_confirmed(ols);
5057         if (status)
5058                 return status;
5059         return nfs4_check_openmode(ols, flags);
5060 }
5061
5062 static __be32
5063 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5064                 struct file **filpp, bool *tmp_file, int flags)
5065 {
5066         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5067         struct file *file;
5068         __be32 status;
5069
5070         file = nfs4_find_file(s, flags);
5071         if (file) {
5072                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5073                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5074                 if (status) {
5075                         fput(file);
5076                         return status;
5077                 }
5078
5079                 *filpp = file;
5080         } else {
5081                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5082                 if (status)
5083                         return status;
5084
5085                 if (tmp_file)
5086                         *tmp_file = true;
5087         }
5088
5089         return 0;
5090 }
5091
5092 /*
5093  * Checks for stateid operations
5094  */
5095 __be32
5096 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5097                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5098                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5099 {
5100         struct inode *ino = d_inode(fhp->fh_dentry);
5101         struct net *net = SVC_NET(rqstp);
5102         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5103         struct nfs4_stid *s = NULL;
5104         __be32 status;
5105
5106         if (filpp)
5107                 *filpp = NULL;
5108         if (tmp_file)
5109                 *tmp_file = false;
5110
5111         if (grace_disallows_io(net, ino))
5112                 return nfserr_grace;
5113
5114         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5115                 status = check_special_stateids(net, fhp, stateid, flags);
5116                 goto done;
5117         }
5118
5119         status = nfsd4_lookup_stateid(cstate, stateid,
5120                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5121                                 &s, nn);
5122         if (status)
5123                 return status;
5124         status = nfsd4_stid_check_stateid_generation(stateid, s,
5125                         nfsd4_has_session(cstate));
5126         if (status)
5127                 goto out;
5128
5129         switch (s->sc_type) {
5130         case NFS4_DELEG_STID:
5131                 status = nfs4_check_delegmode(delegstateid(s), flags);
5132                 break;
5133         case NFS4_OPEN_STID:
5134         case NFS4_LOCK_STID:
5135                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5136                 break;
5137         default:
5138                 status = nfserr_bad_stateid;
5139                 break;
5140         }
5141         if (status)
5142                 goto out;
5143         status = nfs4_check_fh(fhp, s);
5144
5145 done:
5146         if (!status && filpp)
5147                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5148 out:
5149         if (s)
5150                 nfs4_put_stid(s);
5151         return status;
5152 }
5153
5154 /*
5155  * Test if the stateid is valid
5156  */
5157 __be32
5158 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5159                    union nfsd4_op_u *u)
5160 {
5161         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5162         struct nfsd4_test_stateid_id *stateid;
5163         struct nfs4_client *cl = cstate->session->se_client;
5164
5165         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5166                 stateid->ts_id_status =
5167                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5168
5169         return nfs_ok;
5170 }
5171
5172 static __be32
5173 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5174 {
5175         struct nfs4_ol_stateid *stp = openlockstateid(s);
5176         __be32 ret;
5177
5178         ret = nfsd4_lock_ol_stateid(stp);
5179         if (ret)
5180                 goto out_put_stid;
5181
5182         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5183         if (ret)
5184                 goto out;
5185
5186         ret = nfserr_locks_held;
5187         if (check_for_locks(stp->st_stid.sc_file,
5188                             lockowner(stp->st_stateowner)))
5189                 goto out;
5190
5191         release_lock_stateid(stp);
5192         ret = nfs_ok;
5193
5194 out:
5195         mutex_unlock(&stp->st_mutex);
5196 out_put_stid:
5197         nfs4_put_stid(s);
5198         return ret;
5199 }
5200
5201 __be32
5202 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5203                    union nfsd4_op_u *u)
5204 {
5205         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5206         stateid_t *stateid = &free_stateid->fr_stateid;
5207         struct nfs4_stid *s;
5208         struct nfs4_delegation *dp;
5209         struct nfs4_client *cl = cstate->session->se_client;
5210         __be32 ret = nfserr_bad_stateid;
5211
5212         spin_lock(&cl->cl_lock);
5213         s = find_stateid_locked(cl, stateid);
5214         if (!s)
5215                 goto out_unlock;
5216         spin_lock(&s->sc_lock);
5217         switch (s->sc_type) {
5218         case NFS4_DELEG_STID:
5219                 ret = nfserr_locks_held;
5220                 break;
5221         case NFS4_OPEN_STID:
5222                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5223                 if (ret)
5224                         break;
5225                 ret = nfserr_locks_held;
5226                 break;
5227         case NFS4_LOCK_STID:
5228                 spin_unlock(&s->sc_lock);
5229                 refcount_inc(&s->sc_count);
5230                 spin_unlock(&cl->cl_lock);
5231                 ret = nfsd4_free_lock_stateid(stateid, s);
5232                 goto out;
5233         case NFS4_REVOKED_DELEG_STID:
5234                 spin_unlock(&s->sc_lock);
5235                 dp = delegstateid(s);
5236                 list_del_init(&dp->dl_recall_lru);
5237                 spin_unlock(&cl->cl_lock);
5238                 nfs4_put_stid(s);
5239                 ret = nfs_ok;
5240                 goto out;
5241         /* Default falls through and returns nfserr_bad_stateid */
5242         }
5243         spin_unlock(&s->sc_lock);
5244 out_unlock:
5245         spin_unlock(&cl->cl_lock);
5246 out:
5247         return ret;
5248 }
5249
5250 static inline int
5251 setlkflg (int type)
5252 {
5253         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5254                 RD_STATE : WR_STATE;
5255 }
5256
5257 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5258 {
5259         struct svc_fh *current_fh = &cstate->current_fh;
5260         struct nfs4_stateowner *sop = stp->st_stateowner;
5261         __be32 status;
5262
5263         status = nfsd4_check_seqid(cstate, sop, seqid);
5264         if (status)
5265                 return status;
5266         status = nfsd4_lock_ol_stateid(stp);
5267         if (status != nfs_ok)
5268                 return status;
5269         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5270         if (status == nfs_ok)
5271                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5272         if (status != nfs_ok)
5273                 mutex_unlock(&stp->st_mutex);
5274         return status;
5275 }
5276
5277 /*
5278  * Checks for sequence id mutating operations.
5279  */
5280 static __be32
5281 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5282                          stateid_t *stateid, char typemask,
5283                          struct nfs4_ol_stateid **stpp,
5284                          struct nfsd_net *nn)
5285 {
5286         __be32 status;
5287         struct nfs4_stid *s;
5288         struct nfs4_ol_stateid *stp = NULL;
5289
5290         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5291                 seqid, STATEID_VAL(stateid));
5292
5293         *stpp = NULL;
5294         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5295         if (status)
5296                 return status;
5297         stp = openlockstateid(s);
5298         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5299
5300         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5301         if (!status)
5302                 *stpp = stp;
5303         else
5304                 nfs4_put_stid(&stp->st_stid);
5305         return status;
5306 }
5307
5308 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5309                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5310 {
5311         __be32 status;
5312         struct nfs4_openowner *oo;
5313         struct nfs4_ol_stateid *stp;
5314
5315         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5316                                                 NFS4_OPEN_STID, &stp, nn);
5317         if (status)
5318                 return status;
5319         oo = openowner(stp->st_stateowner);
5320         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5321                 mutex_unlock(&stp->st_mutex);
5322                 nfs4_put_stid(&stp->st_stid);
5323                 return nfserr_bad_stateid;
5324         }
5325         *stpp = stp;
5326         return nfs_ok;
5327 }
5328
5329 __be32
5330 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5331                    union nfsd4_op_u *u)
5332 {
5333         struct nfsd4_open_confirm *oc = &u->open_confirm;
5334         __be32 status;
5335         struct nfs4_openowner *oo;
5336         struct nfs4_ol_stateid *stp;
5337         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5338
5339         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5340                         cstate->current_fh.fh_dentry);
5341
5342         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5343         if (status)
5344                 return status;
5345
5346         status = nfs4_preprocess_seqid_op(cstate,
5347                                         oc->oc_seqid, &oc->oc_req_stateid,
5348                                         NFS4_OPEN_STID, &stp, nn);
5349         if (status)
5350                 goto out;
5351         oo = openowner(stp->st_stateowner);
5352         status = nfserr_bad_stateid;
5353         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5354                 mutex_unlock(&stp->st_mutex);
5355                 goto put_stateid;
5356         }
5357         oo->oo_flags |= NFS4_OO_CONFIRMED;
5358         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5359         mutex_unlock(&stp->st_mutex);
5360         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5361                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5362
5363         nfsd4_client_record_create(oo->oo_owner.so_client);
5364         status = nfs_ok;
5365 put_stateid:
5366         nfs4_put_stid(&stp->st_stid);
5367 out:
5368         nfsd4_bump_seqid(cstate, status);
5369         return status;
5370 }
5371
5372 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5373 {
5374         if (!test_access(access, stp))
5375                 return;
5376         nfs4_file_put_access(stp->st_stid.sc_file, access);
5377         clear_access(access, stp);
5378 }
5379
5380 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5381 {
5382         switch (to_access) {
5383         case NFS4_SHARE_ACCESS_READ:
5384                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5385                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5386                 break;
5387         case NFS4_SHARE_ACCESS_WRITE:
5388                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5389                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5390                 break;
5391         case NFS4_SHARE_ACCESS_BOTH:
5392                 break;
5393         default:
5394                 WARN_ON_ONCE(1);
5395         }
5396 }
5397
5398 __be32
5399 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5400                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5401 {
5402         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5403         __be32 status;
5404         struct nfs4_ol_stateid *stp;
5405         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5406
5407         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5408                         cstate->current_fh.fh_dentry);
5409
5410         /* We don't yet support WANT bits: */
5411         if (od->od_deleg_want)
5412                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5413                         od->od_deleg_want);
5414
5415         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5416                                         &od->od_stateid, &stp, nn);
5417         if (status)
5418                 goto out;
5419         status = nfserr_inval;
5420         if (!test_access(od->od_share_access, stp)) {
5421                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5422                         stp->st_access_bmap, od->od_share_access);
5423                 goto put_stateid;
5424         }
5425         if (!test_deny(od->od_share_deny, stp)) {
5426                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5427                         stp->st_deny_bmap, od->od_share_deny);
5428                 goto put_stateid;
5429         }
5430         nfs4_stateid_downgrade(stp, od->od_share_access);
5431         reset_union_bmap_deny(od->od_share_deny, stp);
5432         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5433         status = nfs_ok;
5434 put_stateid:
5435         mutex_unlock(&stp->st_mutex);
5436         nfs4_put_stid(&stp->st_stid);
5437 out:
5438         nfsd4_bump_seqid(cstate, status);
5439         return status;
5440 }
5441
5442 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5443 {
5444         struct nfs4_client *clp = s->st_stid.sc_client;
5445         bool unhashed;
5446         LIST_HEAD(reaplist);
5447
5448         spin_lock(&clp->cl_lock);
5449         unhashed = unhash_open_stateid(s, &reaplist);
5450
5451         if (clp->cl_minorversion) {
5452                 if (unhashed)
5453                         put_ol_stateid_locked(s, &reaplist);
5454                 spin_unlock(&clp->cl_lock);
5455                 free_ol_stateid_reaplist(&reaplist);
5456         } else {
5457                 spin_unlock(&clp->cl_lock);
5458                 free_ol_stateid_reaplist(&reaplist);
5459                 if (unhashed)
5460                         move_to_close_lru(s, clp->net);
5461         }
5462 }
5463
5464 /*
5465  * nfs4_unlock_state() called after encode
5466  */
5467 __be32
5468 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5469                 union nfsd4_op_u *u)
5470 {
5471         struct nfsd4_close *close = &u->close;
5472         __be32 status;
5473         struct nfs4_ol_stateid *stp;
5474         struct net *net = SVC_NET(rqstp);
5475         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5476
5477         dprintk("NFSD: nfsd4_close on file %pd\n",
5478                         cstate->current_fh.fh_dentry);
5479
5480         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5481                                         &close->cl_stateid,
5482                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5483                                         &stp, nn);
5484         nfsd4_bump_seqid(cstate, status);
5485         if (status)
5486                 goto out;
5487
5488         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5489
5490         /*
5491          * Technically we don't _really_ have to increment or copy it, since
5492          * it should just be gone after this operation and we clobber the
5493          * copied value below, but we continue to do so here just to ensure
5494          * that racing ops see that there was a state change.
5495          */
5496         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5497
5498         nfsd4_close_open_stateid(stp);
5499         mutex_unlock(&stp->st_mutex);
5500
5501         /* v4.1+ suggests that we send a special stateid in here, since the
5502          * clients should just ignore this anyway. Since this is not useful
5503          * for v4.0 clients either, we set it to the special close_stateid
5504          * universally.
5505          *
5506          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5507          */
5508         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5509
5510         /* put reference from nfs4_preprocess_seqid_op */
5511         nfs4_put_stid(&stp->st_stid);
5512 out:
5513         return status;
5514 }
5515
5516 __be32
5517 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5518                   union nfsd4_op_u *u)
5519 {
5520         struct nfsd4_delegreturn *dr = &u->delegreturn;
5521         struct nfs4_delegation *dp;
5522         stateid_t *stateid = &dr->dr_stateid;
5523         struct nfs4_stid *s;
5524         __be32 status;
5525         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5526
5527         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5528                 return status;
5529
5530         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5531         if (status)
5532                 goto out;
5533         dp = delegstateid(s);
5534         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5535         if (status)
5536                 goto put_stateid;
5537
5538         destroy_delegation(dp);
5539 put_stateid:
5540         nfs4_put_stid(&dp->dl_stid);
5541 out:
5542         return status;
5543 }
5544
5545 static inline u64
5546 end_offset(u64 start, u64 len)
5547 {
5548         u64 end;
5549
5550         end = start + len;
5551         return end >= start ? end: NFS4_MAX_UINT64;
5552 }
5553
5554 /* last octet in a range */
5555 static inline u64
5556 last_byte_offset(u64 start, u64 len)
5557 {
5558         u64 end;
5559
5560         WARN_ON_ONCE(!len);
5561         end = start + len;
5562         return end > start ? end - 1: NFS4_MAX_UINT64;
5563 }
5564
5565 /*
5566  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5567  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5568  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5569  * locking, this prevents us from being completely protocol-compliant.  The
5570  * real solution to this problem is to start using unsigned file offsets in
5571  * the VFS, but this is a very deep change!
5572  */
5573 static inline void
5574 nfs4_transform_lock_offset(struct file_lock *lock)
5575 {
5576         if (lock->fl_start < 0)
5577                 lock->fl_start = OFFSET_MAX;
5578         if (lock->fl_end < 0)
5579                 lock->fl_end = OFFSET_MAX;
5580 }
5581
5582 static fl_owner_t
5583 nfsd4_fl_get_owner(fl_owner_t owner)
5584 {
5585         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5586
5587         nfs4_get_stateowner(&lo->lo_owner);
5588         return owner;
5589 }
5590
5591 static void
5592 nfsd4_fl_put_owner(fl_owner_t owner)
5593 {
5594         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5595
5596         if (lo)
5597                 nfs4_put_stateowner(&lo->lo_owner);
5598 }
5599
5600 static void
5601 nfsd4_lm_notify(struct file_lock *fl)
5602 {
5603         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5604         struct net                      *net = lo->lo_owner.so_client->net;
5605         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5606         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5607                                                 struct nfsd4_blocked_lock, nbl_lock);
5608         bool queue = false;
5609
5610         /* An empty list means that something else is going to be using it */
5611         spin_lock(&nn->blocked_locks_lock);
5612         if (!list_empty(&nbl->nbl_list)) {
5613                 list_del_init(&nbl->nbl_list);
5614                 list_del_init(&nbl->nbl_lru);
5615                 queue = true;
5616         }
5617         spin_unlock(&nn->blocked_locks_lock);
5618
5619         if (queue)
5620                 nfsd4_run_cb(&nbl->nbl_cb);
5621 }
5622
5623 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5624         .lm_notify = nfsd4_lm_notify,
5625         .lm_get_owner = nfsd4_fl_get_owner,
5626         .lm_put_owner = nfsd4_fl_put_owner,
5627 };
5628
5629 static inline void
5630 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5631 {
5632         struct nfs4_lockowner *lo;
5633
5634         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5635                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5636                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5637                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5638                 if (!deny->ld_owner.data)
5639                         /* We just don't care that much */
5640                         goto nevermind;
5641                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5642                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5643         } else {
5644 nevermind:
5645                 deny->ld_owner.len = 0;
5646                 deny->ld_owner.data = NULL;
5647                 deny->ld_clientid.cl_boot = 0;
5648                 deny->ld_clientid.cl_id = 0;
5649         }
5650         deny->ld_start = fl->fl_start;
5651         deny->ld_length = NFS4_MAX_UINT64;
5652         if (fl->fl_end != NFS4_MAX_UINT64)
5653                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5654         deny->ld_type = NFS4_READ_LT;
5655         if (fl->fl_type != F_RDLCK)
5656                 deny->ld_type = NFS4_WRITE_LT;
5657 }
5658
5659 static struct nfs4_lockowner *
5660 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5661 {
5662         unsigned int strhashval = ownerstr_hashval(owner);
5663         struct nfs4_stateowner *so;
5664
5665         lockdep_assert_held(&clp->cl_lock);
5666
5667         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5668                             so_strhash) {
5669                 if (so->so_is_open_owner)
5670                         continue;
5671                 if (same_owner_str(so, owner))
5672                         return lockowner(nfs4_get_stateowner(so));
5673         }
5674         return NULL;
5675 }
5676
5677 static struct nfs4_lockowner *
5678 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5679 {
5680         struct nfs4_lockowner *lo;
5681
5682         spin_lock(&clp->cl_lock);
5683         lo = find_lockowner_str_locked(clp, owner);
5684         spin_unlock(&clp->cl_lock);
5685         return lo;
5686 }
5687
5688 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5689 {
5690         unhash_lockowner_locked(lockowner(sop));
5691 }
5692
5693 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5694 {
5695         struct nfs4_lockowner *lo = lockowner(sop);
5696
5697         kmem_cache_free(lockowner_slab, lo);
5698 }
5699
5700 static const struct nfs4_stateowner_operations lockowner_ops = {
5701         .so_unhash =    nfs4_unhash_lockowner,
5702         .so_free =      nfs4_free_lockowner,
5703 };
5704
5705 /*
5706  * Alloc a lock owner structure.
5707  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5708  * occurred.
5709  *
5710  * strhashval = ownerstr_hashval
5711  */
5712 static struct nfs4_lockowner *
5713 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5714                            struct nfs4_ol_stateid *open_stp,
5715                            struct nfsd4_lock *lock)
5716 {
5717         struct nfs4_lockowner *lo, *ret;
5718
5719         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5720         if (!lo)
5721                 return NULL;
5722         INIT_LIST_HEAD(&lo->lo_blocked);
5723         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5724         lo->lo_owner.so_is_open_owner = 0;
5725         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5726         lo->lo_owner.so_ops = &lockowner_ops;
5727         spin_lock(&clp->cl_lock);
5728         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5729         if (ret == NULL) {
5730                 list_add(&lo->lo_owner.so_strhash,
5731                          &clp->cl_ownerstr_hashtbl[strhashval]);
5732                 ret = lo;
5733         } else
5734                 nfs4_free_stateowner(&lo->lo_owner);
5735
5736         spin_unlock(&clp->cl_lock);
5737         return ret;
5738 }
5739
5740 static struct nfs4_ol_stateid *
5741 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5742 {
5743         struct nfs4_ol_stateid *lst;
5744         struct nfs4_client *clp = lo->lo_owner.so_client;
5745
5746         lockdep_assert_held(&clp->cl_lock);
5747
5748         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5749                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5750                         continue;
5751                 if (lst->st_stid.sc_file == fp) {
5752                         refcount_inc(&lst->st_stid.sc_count);
5753                         return lst;
5754                 }
5755         }
5756         return NULL;
5757 }
5758
5759 static struct nfs4_ol_stateid *
5760 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5761                   struct nfs4_file *fp, struct inode *inode,
5762                   struct nfs4_ol_stateid *open_stp)
5763 {
5764         struct nfs4_client *clp = lo->lo_owner.so_client;
5765         struct nfs4_ol_stateid *retstp;
5766
5767         mutex_init(&stp->st_mutex);
5768         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5769 retry:
5770         spin_lock(&clp->cl_lock);
5771         spin_lock(&fp->fi_lock);
5772         retstp = find_lock_stateid(lo, fp);
5773         if (retstp)
5774                 goto out_unlock;
5775
5776         refcount_inc(&stp->st_stid.sc_count);
5777         stp->st_stid.sc_type = NFS4_LOCK_STID;
5778         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5779         get_nfs4_file(fp);
5780         stp->st_stid.sc_file = fp;
5781         stp->st_access_bmap = 0;
5782         stp->st_deny_bmap = open_stp->st_deny_bmap;
5783         stp->st_openstp = open_stp;
5784         list_add(&stp->st_locks, &open_stp->st_locks);
5785         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5786         list_add(&stp->st_perfile, &fp->fi_stateids);
5787 out_unlock:
5788         spin_unlock(&fp->fi_lock);
5789         spin_unlock(&clp->cl_lock);
5790         if (retstp) {
5791                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5792                         nfs4_put_stid(&retstp->st_stid);
5793                         goto retry;
5794                 }
5795                 /* To keep mutex tracking happy */
5796                 mutex_unlock(&stp->st_mutex);
5797                 stp = retstp;
5798         }
5799         return stp;
5800 }
5801
5802 static struct nfs4_ol_stateid *
5803 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5804                             struct inode *inode, struct nfs4_ol_stateid *ost,
5805                             bool *new)
5806 {
5807         struct nfs4_stid *ns = NULL;
5808         struct nfs4_ol_stateid *lst;
5809         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5810         struct nfs4_client *clp = oo->oo_owner.so_client;
5811
5812         *new = false;
5813         spin_lock(&clp->cl_lock);
5814         lst = find_lock_stateid(lo, fi);
5815         spin_unlock(&clp->cl_lock);
5816         if (lst != NULL) {
5817                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5818                         goto out;
5819                 nfs4_put_stid(&lst->st_stid);
5820         }
5821         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5822         if (ns == NULL)
5823                 return NULL;
5824
5825         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5826         if (lst == openlockstateid(ns))
5827                 *new = true;
5828         else
5829                 nfs4_put_stid(ns);
5830 out:
5831         return lst;
5832 }
5833
5834 static int
5835 check_lock_length(u64 offset, u64 length)
5836 {
5837         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5838                 (length > ~offset)));
5839 }
5840
5841 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5842 {
5843         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5844
5845         lockdep_assert_held(&fp->fi_lock);
5846
5847         if (test_access(access, lock_stp))
5848                 return;
5849         __nfs4_file_get_access(fp, access);
5850         set_access(access, lock_stp);
5851 }
5852
5853 static __be32
5854 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5855                             struct nfs4_ol_stateid *ost,
5856                             struct nfsd4_lock *lock,
5857                             struct nfs4_ol_stateid **plst, bool *new)
5858 {
5859         __be32 status;
5860         struct nfs4_file *fi = ost->st_stid.sc_file;
5861         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5862         struct nfs4_client *cl = oo->oo_owner.so_client;
5863         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5864         struct nfs4_lockowner *lo;
5865         struct nfs4_ol_stateid *lst;
5866         unsigned int strhashval;
5867
5868         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5869         if (!lo) {
5870                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5871                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5872                 if (lo == NULL)
5873                         return nfserr_jukebox;
5874         } else {
5875                 /* with an existing lockowner, seqids must be the same */
5876                 status = nfserr_bad_seqid;
5877                 if (!cstate->minorversion &&
5878                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5879                         goto out;
5880         }
5881
5882         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5883         if (lst == NULL) {
5884                 status = nfserr_jukebox;
5885                 goto out;
5886         }
5887
5888         status = nfs_ok;
5889         *plst = lst;
5890 out:
5891         nfs4_put_stateowner(&lo->lo_owner);
5892         return status;
5893 }
5894
5895 /*
5896  *  LOCK operation
5897  */
5898 __be32
5899 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5900            union nfsd4_op_u *u)
5901 {
5902         struct nfsd4_lock *lock = &u->lock;
5903         struct nfs4_openowner *open_sop = NULL;
5904         struct nfs4_lockowner *lock_sop = NULL;
5905         struct nfs4_ol_stateid *lock_stp = NULL;
5906         struct nfs4_ol_stateid *open_stp = NULL;
5907         struct nfs4_file *fp;
5908         struct file *filp = NULL;
5909         struct nfsd4_blocked_lock *nbl = NULL;
5910         struct file_lock *file_lock = NULL;
5911         struct file_lock *conflock = NULL;
5912         __be32 status = 0;
5913         int lkflg;
5914         int err;
5915         bool new = false;
5916         unsigned char fl_type;
5917         unsigned int fl_flags = FL_POSIX;
5918         struct net *net = SVC_NET(rqstp);
5919         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5920
5921         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5922                 (long long) lock->lk_offset,
5923                 (long long) lock->lk_length);
5924
5925         if (check_lock_length(lock->lk_offset, lock->lk_length))
5926                  return nfserr_inval;
5927
5928         if ((status = fh_verify(rqstp, &cstate->current_fh,
5929                                 S_IFREG, NFSD_MAY_LOCK))) {
5930                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5931                 return status;
5932         }
5933
5934         if (lock->lk_is_new) {
5935                 if (nfsd4_has_session(cstate))
5936                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5937                         memcpy(&lock->lk_new_clientid,
5938                                 &cstate->session->se_client->cl_clientid,
5939                                 sizeof(clientid_t));
5940
5941                 status = nfserr_stale_clientid;
5942                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5943                         goto out;
5944
5945                 /* validate and update open stateid and open seqid */
5946                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5947                                         lock->lk_new_open_seqid,
5948                                         &lock->lk_new_open_stateid,
5949                                         &open_stp, nn);
5950                 if (status)
5951                         goto out;
5952                 mutex_unlock(&open_stp->st_mutex);
5953                 open_sop = openowner(open_stp->st_stateowner);
5954                 status = nfserr_bad_stateid;
5955                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5956                                                 &lock->lk_new_clientid))
5957                         goto out;
5958                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5959                                                         &lock_stp, &new);
5960         } else {
5961                 status = nfs4_preprocess_seqid_op(cstate,
5962                                        lock->lk_old_lock_seqid,
5963                                        &lock->lk_old_lock_stateid,
5964                                        NFS4_LOCK_STID, &lock_stp, nn);
5965         }
5966         if (status)
5967                 goto out;
5968         lock_sop = lockowner(lock_stp->st_stateowner);
5969
5970         lkflg = setlkflg(lock->lk_type);
5971         status = nfs4_check_openmode(lock_stp, lkflg);
5972         if (status)
5973                 goto out;
5974
5975         status = nfserr_grace;
5976         if (locks_in_grace(net) && !lock->lk_reclaim)
5977                 goto out;
5978         status = nfserr_no_grace;
5979         if (!locks_in_grace(net) && lock->lk_reclaim)
5980                 goto out;
5981
5982         fp = lock_stp->st_stid.sc_file;
5983         switch (lock->lk_type) {
5984                 case NFS4_READW_LT:
5985                         if (nfsd4_has_session(cstate))
5986                                 fl_flags |= FL_SLEEP;
5987                         /* Fallthrough */
5988                 case NFS4_READ_LT:
5989                         spin_lock(&fp->fi_lock);
5990                         filp = find_readable_file_locked(fp);
5991                         if (filp)
5992                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5993                         spin_unlock(&fp->fi_lock);
5994                         fl_type = F_RDLCK;
5995                         break;
5996                 case NFS4_WRITEW_LT:
5997                         if (nfsd4_has_session(cstate))
5998                                 fl_flags |= FL_SLEEP;
5999                         /* Fallthrough */
6000                 case NFS4_WRITE_LT:
6001                         spin_lock(&fp->fi_lock);
6002                         filp = find_writeable_file_locked(fp);
6003                         if (filp)
6004                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6005                         spin_unlock(&fp->fi_lock);
6006                         fl_type = F_WRLCK;
6007                         break;
6008                 default:
6009                         status = nfserr_inval;
6010                 goto out;
6011         }
6012
6013         if (!filp) {
6014                 status = nfserr_openmode;
6015                 goto out;
6016         }
6017
6018         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6019         if (!nbl) {
6020                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6021                 status = nfserr_jukebox;
6022                 goto out;
6023         }
6024
6025         file_lock = &nbl->nbl_lock;
6026         file_lock->fl_type = fl_type;
6027         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6028         file_lock->fl_pid = current->tgid;
6029         file_lock->fl_file = filp;
6030         file_lock->fl_flags = fl_flags;
6031         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6032         file_lock->fl_start = lock->lk_offset;
6033         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6034         nfs4_transform_lock_offset(file_lock);
6035
6036         conflock = locks_alloc_lock();
6037         if (!conflock) {
6038                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6039                 status = nfserr_jukebox;
6040                 goto out;
6041         }
6042
6043         if (fl_flags & FL_SLEEP) {
6044                 nbl->nbl_time = jiffies;
6045                 spin_lock(&nn->blocked_locks_lock);
6046                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6047                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6048                 spin_unlock(&nn->blocked_locks_lock);
6049         }
6050
6051         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6052         switch (err) {
6053         case 0: /* success! */
6054                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6055                 status = 0;
6056                 break;
6057         case FILE_LOCK_DEFERRED:
6058                 nbl = NULL;
6059                 /* Fallthrough */
6060         case -EAGAIN:           /* conflock holds conflicting lock */
6061                 status = nfserr_denied;
6062                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6063                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6064                 break;
6065         case -EDEADLK:
6066                 status = nfserr_deadlock;
6067                 break;
6068         default:
6069                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6070                 status = nfserrno(err);
6071                 break;
6072         }
6073 out:
6074         if (nbl) {
6075                 /* dequeue it if we queued it before */
6076                 if (fl_flags & FL_SLEEP) {
6077                         spin_lock(&nn->blocked_locks_lock);
6078                         list_del_init(&nbl->nbl_list);
6079                         list_del_init(&nbl->nbl_lru);
6080                         spin_unlock(&nn->blocked_locks_lock);
6081                 }
6082                 free_blocked_lock(nbl);
6083         }
6084         if (filp)
6085                 fput(filp);
6086         if (lock_stp) {
6087                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6088                 if (cstate->replay_owner &&
6089                     cstate->replay_owner != &lock_sop->lo_owner &&
6090                     seqid_mutating_err(ntohl(status)))
6091                         lock_sop->lo_owner.so_seqid++;
6092
6093                 /*
6094                  * If this is a new, never-before-used stateid, and we are
6095                  * returning an error, then just go ahead and release it.
6096                  */
6097                 if (status && new)
6098                         release_lock_stateid(lock_stp);
6099
6100                 mutex_unlock(&lock_stp->st_mutex);
6101
6102                 nfs4_put_stid(&lock_stp->st_stid);
6103         }
6104         if (open_stp)
6105                 nfs4_put_stid(&open_stp->st_stid);
6106         nfsd4_bump_seqid(cstate, status);
6107         if (conflock)
6108                 locks_free_lock(conflock);
6109         return status;
6110 }
6111
6112 /*
6113  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6114  * so we do a temporary open here just to get an open file to pass to
6115  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6116  * inode operation.)
6117  */
6118 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6119 {
6120         struct file *file;
6121         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6122         if (!err) {
6123                 err = nfserrno(vfs_test_lock(file, lock));
6124                 fput(file);
6125         }
6126         return err;
6127 }
6128
6129 /*
6130  * LOCKT operation
6131  */
6132 __be32
6133 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6134             union nfsd4_op_u *u)
6135 {
6136         struct nfsd4_lockt *lockt = &u->lockt;
6137         struct file_lock *file_lock = NULL;
6138         struct nfs4_lockowner *lo = NULL;
6139         __be32 status;
6140         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6141
6142         if (locks_in_grace(SVC_NET(rqstp)))
6143                 return nfserr_grace;
6144
6145         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6146                  return nfserr_inval;
6147
6148         if (!nfsd4_has_session(cstate)) {
6149                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6150                 if (status)
6151                         goto out;
6152         }
6153
6154         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6155                 goto out;
6156
6157         file_lock = locks_alloc_lock();
6158         if (!file_lock) {
6159                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6160                 status = nfserr_jukebox;
6161                 goto out;
6162         }
6163
6164         switch (lockt->lt_type) {
6165                 case NFS4_READ_LT:
6166                 case NFS4_READW_LT:
6167                         file_lock->fl_type = F_RDLCK;
6168                 break;
6169                 case NFS4_WRITE_LT:
6170                 case NFS4_WRITEW_LT:
6171                         file_lock->fl_type = F_WRLCK;
6172                 break;
6173                 default:
6174                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6175                         status = nfserr_inval;
6176                 goto out;
6177         }
6178
6179         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6180         if (lo)
6181                 file_lock->fl_owner = (fl_owner_t)lo;
6182         file_lock->fl_pid = current->tgid;
6183         file_lock->fl_flags = FL_POSIX;
6184
6185         file_lock->fl_start = lockt->lt_offset;
6186         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6187
6188         nfs4_transform_lock_offset(file_lock);
6189
6190         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6191         if (status)
6192                 goto out;
6193
6194         if (file_lock->fl_type != F_UNLCK) {
6195                 status = nfserr_denied;
6196                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6197         }
6198 out:
6199         if (lo)
6200                 nfs4_put_stateowner(&lo->lo_owner);
6201         if (file_lock)
6202                 locks_free_lock(file_lock);
6203         return status;
6204 }
6205
6206 __be32
6207 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6208             union nfsd4_op_u *u)
6209 {
6210         struct nfsd4_locku *locku = &u->locku;
6211         struct nfs4_ol_stateid *stp;
6212         struct file *filp = NULL;
6213         struct file_lock *file_lock = NULL;
6214         __be32 status;
6215         int err;
6216         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6217
6218         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6219                 (long long) locku->lu_offset,
6220                 (long long) locku->lu_length);
6221
6222         if (check_lock_length(locku->lu_offset, locku->lu_length))
6223                  return nfserr_inval;
6224
6225         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6226                                         &locku->lu_stateid, NFS4_LOCK_STID,
6227                                         &stp, nn);
6228         if (status)
6229                 goto out;
6230         filp = find_any_file(stp->st_stid.sc_file);
6231         if (!filp) {
6232                 status = nfserr_lock_range;
6233                 goto put_stateid;
6234         }
6235         file_lock = locks_alloc_lock();
6236         if (!file_lock) {
6237                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6238                 status = nfserr_jukebox;
6239                 goto fput;
6240         }
6241
6242         file_lock->fl_type = F_UNLCK;
6243         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6244         file_lock->fl_pid = current->tgid;
6245         file_lock->fl_file = filp;
6246         file_lock->fl_flags = FL_POSIX;
6247         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6248         file_lock->fl_start = locku->lu_offset;
6249
6250         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6251                                                 locku->lu_length);
6252         nfs4_transform_lock_offset(file_lock);
6253
6254         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6255         if (err) {
6256                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6257                 goto out_nfserr;
6258         }
6259         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6260 fput:
6261         fput(filp);
6262 put_stateid:
6263         mutex_unlock(&stp->st_mutex);
6264         nfs4_put_stid(&stp->st_stid);
6265 out:
6266         nfsd4_bump_seqid(cstate, status);
6267         if (file_lock)
6268                 locks_free_lock(file_lock);
6269         return status;
6270
6271 out_nfserr:
6272         status = nfserrno(err);
6273         goto fput;
6274 }
6275
6276 /*
6277  * returns
6278  *      true:  locks held by lockowner
6279  *      false: no locks held by lockowner
6280  */
6281 static bool
6282 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6283 {
6284         struct file_lock *fl;
6285         int status = false;
6286         struct file *filp = find_any_file(fp);
6287         struct inode *inode;
6288         struct file_lock_context *flctx;
6289
6290         if (!filp) {
6291                 /* Any valid lock stateid should have some sort of access */
6292                 WARN_ON_ONCE(1);
6293                 return status;
6294         }
6295
6296         inode = file_inode(filp);
6297         flctx = inode->i_flctx;
6298
6299         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6300                 spin_lock(&flctx->flc_lock);
6301                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6302                         if (fl->fl_owner == (fl_owner_t)lowner) {
6303                                 status = true;
6304                                 break;
6305                         }
6306                 }
6307                 spin_unlock(&flctx->flc_lock);
6308         }
6309         fput(filp);
6310         return status;
6311 }
6312
6313 __be32
6314 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6315                         struct nfsd4_compound_state *cstate,
6316                         union nfsd4_op_u *u)
6317 {
6318         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6319         clientid_t *clid = &rlockowner->rl_clientid;
6320         struct nfs4_stateowner *sop;
6321         struct nfs4_lockowner *lo = NULL;
6322         struct nfs4_ol_stateid *stp;
6323         struct xdr_netobj *owner = &rlockowner->rl_owner;
6324         unsigned int hashval = ownerstr_hashval(owner);
6325         __be32 status;
6326         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6327         struct nfs4_client *clp;
6328         LIST_HEAD (reaplist);
6329
6330         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6331                 clid->cl_boot, clid->cl_id);
6332
6333         status = lookup_clientid(clid, cstate, nn);
6334         if (status)
6335                 return status;
6336
6337         clp = cstate->clp;
6338         /* Find the matching lock stateowner */
6339         spin_lock(&clp->cl_lock);
6340         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6341                             so_strhash) {
6342
6343                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6344                         continue;
6345
6346                 /* see if there are still any locks associated with it */
6347                 lo = lockowner(sop);
6348                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6349                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6350                                 status = nfserr_locks_held;
6351                                 spin_unlock(&clp->cl_lock);
6352                                 return status;
6353                         }
6354                 }
6355
6356                 nfs4_get_stateowner(sop);
6357                 break;
6358         }
6359         if (!lo) {
6360                 spin_unlock(&clp->cl_lock);
6361                 return status;
6362         }
6363
6364         unhash_lockowner_locked(lo);
6365         while (!list_empty(&lo->lo_owner.so_stateids)) {
6366                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6367                                        struct nfs4_ol_stateid,
6368                                        st_perstateowner);
6369                 WARN_ON(!unhash_lock_stateid(stp));
6370                 put_ol_stateid_locked(stp, &reaplist);
6371         }
6372         spin_unlock(&clp->cl_lock);
6373         free_ol_stateid_reaplist(&reaplist);
6374         remove_blocked_locks(lo);
6375         nfs4_put_stateowner(&lo->lo_owner);
6376
6377         return status;
6378 }
6379
6380 static inline struct nfs4_client_reclaim *
6381 alloc_reclaim(void)
6382 {
6383         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6384 }
6385
6386 bool
6387 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6388 {
6389         struct nfs4_client_reclaim *crp;
6390
6391         crp = nfsd4_find_reclaim_client(name, nn);
6392         return (crp && crp->cr_clp);
6393 }
6394
6395 /*
6396  * failure => all reset bets are off, nfserr_no_grace...
6397  */
6398 struct nfs4_client_reclaim *
6399 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6400 {
6401         unsigned int strhashval;
6402         struct nfs4_client_reclaim *crp;
6403
6404         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6405         crp = alloc_reclaim();
6406         if (crp) {
6407                 strhashval = clientstr_hashval(name);
6408                 INIT_LIST_HEAD(&crp->cr_strhash);
6409                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6410                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6411                 crp->cr_clp = NULL;
6412                 nn->reclaim_str_hashtbl_size++;
6413         }
6414         return crp;
6415 }
6416
6417 void
6418 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6419 {
6420         list_del(&crp->cr_strhash);
6421         kfree(crp);
6422         nn->reclaim_str_hashtbl_size--;
6423 }
6424
6425 void
6426 nfs4_release_reclaim(struct nfsd_net *nn)
6427 {
6428         struct nfs4_client_reclaim *crp = NULL;
6429         int i;
6430
6431         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6432                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6433                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6434                                         struct nfs4_client_reclaim, cr_strhash);
6435                         nfs4_remove_reclaim_record(crp, nn);
6436                 }
6437         }
6438         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6439 }
6440
6441 /*
6442  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6443 struct nfs4_client_reclaim *
6444 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6445 {
6446         unsigned int strhashval;
6447         struct nfs4_client_reclaim *crp = NULL;
6448
6449         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6450
6451         strhashval = clientstr_hashval(recdir);
6452         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6453                 if (same_name(crp->cr_recdir, recdir)) {
6454                         return crp;
6455                 }
6456         }
6457         return NULL;
6458 }
6459
6460 /*
6461 * Called from OPEN. Look for clientid in reclaim list.
6462 */
6463 __be32
6464 nfs4_check_open_reclaim(clientid_t *clid,
6465                 struct nfsd4_compound_state *cstate,
6466                 struct nfsd_net *nn)
6467 {
6468         __be32 status;
6469
6470         /* find clientid in conf_id_hashtbl */
6471         status = lookup_clientid(clid, cstate, nn);
6472         if (status)
6473                 return nfserr_reclaim_bad;
6474
6475         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6476                 return nfserr_no_grace;
6477
6478         if (nfsd4_client_record_check(cstate->clp))
6479                 return nfserr_reclaim_bad;
6480
6481         return nfs_ok;
6482 }
6483
6484 #ifdef CONFIG_NFSD_FAULT_INJECTION
6485 static inline void
6486 put_client(struct nfs4_client *clp)
6487 {
6488         atomic_dec(&clp->cl_refcount);
6489 }
6490
6491 static struct nfs4_client *
6492 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6493 {
6494         struct nfs4_client *clp;
6495         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6496                                           nfsd_net_id);
6497
6498         if (!nfsd_netns_ready(nn))
6499                 return NULL;
6500
6501         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6502                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6503                         return clp;
6504         }
6505         return NULL;
6506 }
6507
6508 u64
6509 nfsd_inject_print_clients(void)
6510 {
6511         struct nfs4_client *clp;
6512         u64 count = 0;
6513         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6514                                           nfsd_net_id);
6515         char buf[INET6_ADDRSTRLEN];
6516
6517         if (!nfsd_netns_ready(nn))
6518                 return 0;
6519
6520         spin_lock(&nn->client_lock);
6521         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6522                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6523                 pr_info("NFS Client: %s\n", buf);
6524                 ++count;
6525         }
6526         spin_unlock(&nn->client_lock);
6527
6528         return count;
6529 }
6530
6531 u64
6532 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6533 {
6534         u64 count = 0;
6535         struct nfs4_client *clp;
6536         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6537                                           nfsd_net_id);
6538
6539         if (!nfsd_netns_ready(nn))
6540                 return count;
6541
6542         spin_lock(&nn->client_lock);
6543         clp = nfsd_find_client(addr, addr_size);
6544         if (clp) {
6545                 if (mark_client_expired_locked(clp) == nfs_ok)
6546                         ++count;
6547                 else
6548                         clp = NULL;
6549         }
6550         spin_unlock(&nn->client_lock);
6551
6552         if (clp)
6553                 expire_client(clp);
6554
6555         return count;
6556 }
6557
6558 u64
6559 nfsd_inject_forget_clients(u64 max)
6560 {
6561         u64 count = 0;
6562         struct nfs4_client *clp, *next;
6563         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6564                                                 nfsd_net_id);
6565         LIST_HEAD(reaplist);
6566
6567         if (!nfsd_netns_ready(nn))
6568                 return count;
6569
6570         spin_lock(&nn->client_lock);
6571         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6572                 if (mark_client_expired_locked(clp) == nfs_ok) {
6573                         list_add(&clp->cl_lru, &reaplist);
6574                         if (max != 0 && ++count >= max)
6575                                 break;
6576                 }
6577         }
6578         spin_unlock(&nn->client_lock);
6579
6580         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6581                 expire_client(clp);
6582
6583         return count;
6584 }
6585
6586 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6587                              const char *type)
6588 {
6589         char buf[INET6_ADDRSTRLEN];
6590         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6591         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6592 }
6593
6594 static void
6595 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6596                              struct list_head *collect)
6597 {
6598         struct nfs4_client *clp = lst->st_stid.sc_client;
6599         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6600                                           nfsd_net_id);
6601
6602         if (!collect)
6603                 return;
6604
6605         lockdep_assert_held(&nn->client_lock);
6606         atomic_inc(&clp->cl_refcount);
6607         list_add(&lst->st_locks, collect);
6608 }
6609
6610 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6611                                     struct list_head *collect,
6612                                     bool (*func)(struct nfs4_ol_stateid *))
6613 {
6614         struct nfs4_openowner *oop;
6615         struct nfs4_ol_stateid *stp, *st_next;
6616         struct nfs4_ol_stateid *lst, *lst_next;
6617         u64 count = 0;
6618
6619         spin_lock(&clp->cl_lock);
6620         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6621                 list_for_each_entry_safe(stp, st_next,
6622                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6623                         list_for_each_entry_safe(lst, lst_next,
6624                                         &stp->st_locks, st_locks) {
6625                                 if (func) {
6626                                         if (func(lst))
6627                                                 nfsd_inject_add_lock_to_list(lst,
6628                                                                         collect);
6629                                 }
6630                                 ++count;
6631                                 /*
6632                                  * Despite the fact that these functions deal
6633                                  * with 64-bit integers for "count", we must
6634                                  * ensure that it doesn't blow up the
6635                                  * clp->cl_refcount. Throw a warning if we
6636                                  * start to approach INT_MAX here.
6637                                  */
6638                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6639                                 if (count == max)
6640                                         goto out;
6641                         }
6642                 }
6643         }
6644 out:
6645         spin_unlock(&clp->cl_lock);
6646
6647         return count;
6648 }
6649
6650 static u64
6651 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6652                           u64 max)
6653 {
6654         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6655 }
6656
6657 static u64
6658 nfsd_print_client_locks(struct nfs4_client *clp)
6659 {
6660         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6661         nfsd_print_count(clp, count, "locked files");
6662         return count;
6663 }
6664
6665 u64
6666 nfsd_inject_print_locks(void)
6667 {
6668         struct nfs4_client *clp;
6669         u64 count = 0;
6670         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6671                                                 nfsd_net_id);
6672
6673         if (!nfsd_netns_ready(nn))
6674                 return 0;
6675
6676         spin_lock(&nn->client_lock);
6677         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6678                 count += nfsd_print_client_locks(clp);
6679         spin_unlock(&nn->client_lock);
6680
6681         return count;
6682 }
6683
6684 static void
6685 nfsd_reap_locks(struct list_head *reaplist)
6686 {
6687         struct nfs4_client *clp;
6688         struct nfs4_ol_stateid *stp, *next;
6689
6690         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6691                 list_del_init(&stp->st_locks);
6692                 clp = stp->st_stid.sc_client;
6693                 nfs4_put_stid(&stp->st_stid);
6694                 put_client(clp);
6695         }
6696 }
6697
6698 u64
6699 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6700 {
6701         unsigned int count = 0;
6702         struct nfs4_client *clp;
6703         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6704                                                 nfsd_net_id);
6705         LIST_HEAD(reaplist);
6706
6707         if (!nfsd_netns_ready(nn))
6708                 return count;
6709
6710         spin_lock(&nn->client_lock);
6711         clp = nfsd_find_client(addr, addr_size);
6712         if (clp)
6713                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6714         spin_unlock(&nn->client_lock);
6715         nfsd_reap_locks(&reaplist);
6716         return count;
6717 }
6718
6719 u64
6720 nfsd_inject_forget_locks(u64 max)
6721 {
6722         u64 count = 0;
6723         struct nfs4_client *clp;
6724         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6725                                                 nfsd_net_id);
6726         LIST_HEAD(reaplist);
6727
6728         if (!nfsd_netns_ready(nn))
6729                 return count;
6730
6731         spin_lock(&nn->client_lock);
6732         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6733                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6734                 if (max != 0 && count >= max)
6735                         break;
6736         }
6737         spin_unlock(&nn->client_lock);
6738         nfsd_reap_locks(&reaplist);
6739         return count;
6740 }
6741
6742 static u64
6743 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6744                               struct list_head *collect,
6745                               void (*func)(struct nfs4_openowner *))
6746 {
6747         struct nfs4_openowner *oop, *next;
6748         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6749                                                 nfsd_net_id);
6750         u64 count = 0;
6751
6752         lockdep_assert_held(&nn->client_lock);
6753
6754         spin_lock(&clp->cl_lock);
6755         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6756                 if (func) {
6757                         func(oop);
6758                         if (collect) {
6759                                 atomic_inc(&clp->cl_refcount);
6760                                 list_add(&oop->oo_perclient, collect);
6761                         }
6762                 }
6763                 ++count;
6764                 /*
6765                  * Despite the fact that these functions deal with
6766                  * 64-bit integers for "count", we must ensure that
6767                  * it doesn't blow up the clp->cl_refcount. Throw a
6768                  * warning if we start to approach INT_MAX here.
6769                  */
6770                 WARN_ON_ONCE(count == (INT_MAX / 2));
6771                 if (count == max)
6772                         break;
6773         }
6774         spin_unlock(&clp->cl_lock);
6775
6776         return count;
6777 }
6778
6779 static u64
6780 nfsd_print_client_openowners(struct nfs4_client *clp)
6781 {
6782         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6783
6784         nfsd_print_count(clp, count, "openowners");
6785         return count;
6786 }
6787
6788 static u64
6789 nfsd_collect_client_openowners(struct nfs4_client *clp,
6790                                struct list_head *collect, u64 max)
6791 {
6792         return nfsd_foreach_client_openowner(clp, max, collect,
6793                                                 unhash_openowner_locked);
6794 }
6795
6796 u64
6797 nfsd_inject_print_openowners(void)
6798 {
6799         struct nfs4_client *clp;
6800         u64 count = 0;
6801         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6802                                                 nfsd_net_id);
6803
6804         if (!nfsd_netns_ready(nn))
6805                 return 0;
6806
6807         spin_lock(&nn->client_lock);
6808         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6809                 count += nfsd_print_client_openowners(clp);
6810         spin_unlock(&nn->client_lock);
6811
6812         return count;
6813 }
6814
6815 static void
6816 nfsd_reap_openowners(struct list_head *reaplist)
6817 {
6818         struct nfs4_client *clp;
6819         struct nfs4_openowner *oop, *next;
6820
6821         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6822                 list_del_init(&oop->oo_perclient);
6823                 clp = oop->oo_owner.so_client;
6824                 release_openowner(oop);
6825                 put_client(clp);
6826         }
6827 }
6828
6829 u64
6830 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6831                                      size_t addr_size)
6832 {
6833         unsigned int count = 0;
6834         struct nfs4_client *clp;
6835         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6836                                                 nfsd_net_id);
6837         LIST_HEAD(reaplist);
6838
6839         if (!nfsd_netns_ready(nn))
6840                 return count;
6841
6842         spin_lock(&nn->client_lock);
6843         clp = nfsd_find_client(addr, addr_size);
6844         if (clp)
6845                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6846         spin_unlock(&nn->client_lock);
6847         nfsd_reap_openowners(&reaplist);
6848         return count;
6849 }
6850
6851 u64
6852 nfsd_inject_forget_openowners(u64 max)
6853 {
6854         u64 count = 0;
6855         struct nfs4_client *clp;
6856         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6857                                                 nfsd_net_id);
6858         LIST_HEAD(reaplist);
6859
6860         if (!nfsd_netns_ready(nn))
6861                 return count;
6862
6863         spin_lock(&nn->client_lock);
6864         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6865                 count += nfsd_collect_client_openowners(clp, &reaplist,
6866                                                         max - count);
6867                 if (max != 0 && count >= max)
6868                         break;
6869         }
6870         spin_unlock(&nn->client_lock);
6871         nfsd_reap_openowners(&reaplist);
6872         return count;
6873 }
6874
6875 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6876                                      struct list_head *victims)
6877 {
6878         struct nfs4_delegation *dp, *next;
6879         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6880                                                 nfsd_net_id);
6881         u64 count = 0;
6882
6883         lockdep_assert_held(&nn->client_lock);
6884
6885         spin_lock(&state_lock);
6886         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6887                 if (victims) {
6888                         /*
6889                          * It's not safe to mess with delegations that have a
6890                          * non-zero dl_time. They might have already been broken
6891                          * and could be processed by the laundromat outside of
6892                          * the state_lock. Just leave them be.
6893                          */
6894                         if (dp->dl_time != 0)
6895                                 continue;
6896
6897                         atomic_inc(&clp->cl_refcount);
6898                         WARN_ON(!unhash_delegation_locked(dp));
6899                         list_add(&dp->dl_recall_lru, victims);
6900                 }
6901                 ++count;
6902                 /*
6903                  * Despite the fact that these functions deal with
6904                  * 64-bit integers for "count", we must ensure that
6905                  * it doesn't blow up the clp->cl_refcount. Throw a
6906                  * warning if we start to approach INT_MAX here.
6907                  */
6908                 WARN_ON_ONCE(count == (INT_MAX / 2));
6909                 if (count == max)
6910                         break;
6911         }
6912         spin_unlock(&state_lock);
6913         return count;
6914 }
6915
6916 static u64
6917 nfsd_print_client_delegations(struct nfs4_client *clp)
6918 {
6919         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6920
6921         nfsd_print_count(clp, count, "delegations");
6922         return count;
6923 }
6924
6925 u64
6926 nfsd_inject_print_delegations(void)
6927 {
6928         struct nfs4_client *clp;
6929         u64 count = 0;
6930         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6931                                                 nfsd_net_id);
6932
6933         if (!nfsd_netns_ready(nn))
6934                 return 0;
6935
6936         spin_lock(&nn->client_lock);
6937         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6938                 count += nfsd_print_client_delegations(clp);
6939         spin_unlock(&nn->client_lock);
6940
6941         return count;
6942 }
6943
6944 static void
6945 nfsd_forget_delegations(struct list_head *reaplist)
6946 {
6947         struct nfs4_client *clp;
6948         struct nfs4_delegation *dp, *next;
6949
6950         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6951                 list_del_init(&dp->dl_recall_lru);
6952                 clp = dp->dl_stid.sc_client;
6953                 revoke_delegation(dp);
6954                 put_client(clp);
6955         }
6956 }
6957
6958 u64
6959 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6960                                       size_t addr_size)
6961 {
6962         u64 count = 0;
6963         struct nfs4_client *clp;
6964         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6965                                                 nfsd_net_id);
6966         LIST_HEAD(reaplist);
6967
6968         if (!nfsd_netns_ready(nn))
6969                 return count;
6970
6971         spin_lock(&nn->client_lock);
6972         clp = nfsd_find_client(addr, addr_size);
6973         if (clp)
6974                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6975         spin_unlock(&nn->client_lock);
6976
6977         nfsd_forget_delegations(&reaplist);
6978         return count;
6979 }
6980
6981 u64
6982 nfsd_inject_forget_delegations(u64 max)
6983 {
6984         u64 count = 0;
6985         struct nfs4_client *clp;
6986         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6987                                                 nfsd_net_id);
6988         LIST_HEAD(reaplist);
6989
6990         if (!nfsd_netns_ready(nn))
6991                 return count;
6992
6993         spin_lock(&nn->client_lock);
6994         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6995                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6996                 if (max != 0 && count >= max)
6997                         break;
6998         }
6999         spin_unlock(&nn->client_lock);
7000         nfsd_forget_delegations(&reaplist);
7001         return count;
7002 }
7003
7004 static void
7005 nfsd_recall_delegations(struct list_head *reaplist)
7006 {
7007         struct nfs4_client *clp;
7008         struct nfs4_delegation *dp, *next;
7009
7010         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7011                 list_del_init(&dp->dl_recall_lru);
7012                 clp = dp->dl_stid.sc_client;
7013                 /*
7014                  * We skipped all entries that had a zero dl_time before,
7015                  * so we can now reset the dl_time back to 0. If a delegation
7016                  * break comes in now, then it won't make any difference since
7017                  * we're recalling it either way.
7018                  */
7019                 spin_lock(&state_lock);
7020                 dp->dl_time = 0;
7021                 spin_unlock(&state_lock);
7022                 nfsd_break_one_deleg(dp);
7023                 put_client(clp);
7024         }
7025 }
7026
7027 u64
7028 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7029                                       size_t addr_size)
7030 {
7031         u64 count = 0;
7032         struct nfs4_client *clp;
7033         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7034                                                 nfsd_net_id);
7035         LIST_HEAD(reaplist);
7036
7037         if (!nfsd_netns_ready(nn))
7038                 return count;
7039
7040         spin_lock(&nn->client_lock);
7041         clp = nfsd_find_client(addr, addr_size);
7042         if (clp)
7043                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7044         spin_unlock(&nn->client_lock);
7045
7046         nfsd_recall_delegations(&reaplist);
7047         return count;
7048 }
7049
7050 u64
7051 nfsd_inject_recall_delegations(u64 max)
7052 {
7053         u64 count = 0;
7054         struct nfs4_client *clp, *next;
7055         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7056                                                 nfsd_net_id);
7057         LIST_HEAD(reaplist);
7058
7059         if (!nfsd_netns_ready(nn))
7060                 return count;
7061
7062         spin_lock(&nn->client_lock);
7063         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7064                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7065                 if (max != 0 && ++count >= max)
7066                         break;
7067         }
7068         spin_unlock(&nn->client_lock);
7069         nfsd_recall_delegations(&reaplist);
7070         return count;
7071 }
7072 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7073
7074 /*
7075  * Since the lifetime of a delegation isn't limited to that of an open, a
7076  * client may quite reasonably hang on to a delegation as long as it has
7077  * the inode cached.  This becomes an obvious problem the first time a
7078  * client's inode cache approaches the size of the server's total memory.
7079  *
7080  * For now we avoid this problem by imposing a hard limit on the number
7081  * of delegations, which varies according to the server's memory size.
7082  */
7083 static void
7084 set_max_delegations(void)
7085 {
7086         /*
7087          * Allow at most 4 delegations per megabyte of RAM.  Quick
7088          * estimates suggest that in the worst case (where every delegation
7089          * is for a different inode), a delegation could take about 1.5K,
7090          * giving a worst case usage of about 6% of memory.
7091          */
7092         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7093 }
7094
7095 static int nfs4_state_create_net(struct net *net)
7096 {
7097         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7098         int i;
7099
7100         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7101                                             sizeof(struct list_head),
7102                                             GFP_KERNEL);
7103         if (!nn->conf_id_hashtbl)
7104                 goto err;
7105         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7106                                               sizeof(struct list_head),
7107                                               GFP_KERNEL);
7108         if (!nn->unconf_id_hashtbl)
7109                 goto err_unconf_id;
7110         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7111                                               sizeof(struct list_head),
7112                                               GFP_KERNEL);
7113         if (!nn->sessionid_hashtbl)
7114                 goto err_sessionid;
7115
7116         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7117                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7118                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7119         }
7120         for (i = 0; i < SESSION_HASH_SIZE; i++)
7121                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7122         nn->conf_name_tree = RB_ROOT;
7123         nn->unconf_name_tree = RB_ROOT;
7124         nn->boot_time = get_seconds();
7125         nn->grace_ended = false;
7126         nn->nfsd4_manager.block_opens = true;
7127         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7128         INIT_LIST_HEAD(&nn->client_lru);
7129         INIT_LIST_HEAD(&nn->close_lru);
7130         INIT_LIST_HEAD(&nn->del_recall_lru);
7131         spin_lock_init(&nn->client_lock);
7132
7133         spin_lock_init(&nn->blocked_locks_lock);
7134         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7135
7136         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7137         get_net(net);
7138
7139         return 0;
7140
7141 err_sessionid:
7142         kfree(nn->unconf_id_hashtbl);
7143 err_unconf_id:
7144         kfree(nn->conf_id_hashtbl);
7145 err:
7146         return -ENOMEM;
7147 }
7148
7149 static void
7150 nfs4_state_destroy_net(struct net *net)
7151 {
7152         int i;
7153         struct nfs4_client *clp = NULL;
7154         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7155
7156         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7157                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7158                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7159                         destroy_client(clp);
7160                 }
7161         }
7162
7163         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7164
7165         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7166                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7167                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7168                         destroy_client(clp);
7169                 }
7170         }
7171
7172         kfree(nn->sessionid_hashtbl);
7173         kfree(nn->unconf_id_hashtbl);
7174         kfree(nn->conf_id_hashtbl);
7175         put_net(net);
7176 }
7177
7178 int
7179 nfs4_state_start_net(struct net *net)
7180 {
7181         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7182         int ret;
7183
7184         ret = nfs4_state_create_net(net);
7185         if (ret)
7186                 return ret;
7187         locks_start_grace(net, &nn->nfsd4_manager);
7188         nfsd4_client_tracking_init(net);
7189         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7190                nn->nfsd4_grace, net->ns.inum);
7191         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7192         return 0;
7193 }
7194
7195 /* initialization to perform when the nfsd service is started: */
7196
7197 int
7198 nfs4_state_start(void)
7199 {
7200         int ret;
7201
7202         ret = set_callback_cred();
7203         if (ret)
7204                 return ret;
7205
7206         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7207         if (laundry_wq == NULL) {
7208                 ret = -ENOMEM;
7209                 goto out_cleanup_cred;
7210         }
7211         ret = nfsd4_create_callback_queue();
7212         if (ret)
7213                 goto out_free_laundry;
7214
7215         set_max_delegations();
7216         return 0;
7217
7218 out_free_laundry:
7219         destroy_workqueue(laundry_wq);
7220 out_cleanup_cred:
7221         cleanup_callback_cred();
7222         return ret;
7223 }
7224
7225 void
7226 nfs4_state_shutdown_net(struct net *net)
7227 {
7228         struct nfs4_delegation *dp = NULL;
7229         struct list_head *pos, *next, reaplist;
7230         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7231
7232         cancel_delayed_work_sync(&nn->laundromat_work);
7233         locks_end_grace(&nn->nfsd4_manager);
7234
7235         INIT_LIST_HEAD(&reaplist);
7236         spin_lock(&state_lock);
7237         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7238                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7239                 WARN_ON(!unhash_delegation_locked(dp));
7240                 list_add(&dp->dl_recall_lru, &reaplist);
7241         }
7242         spin_unlock(&state_lock);
7243         list_for_each_safe(pos, next, &reaplist) {
7244                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7245                 list_del_init(&dp->dl_recall_lru);
7246                 destroy_unhashed_deleg(dp);
7247         }
7248
7249         nfsd4_client_tracking_exit(net);
7250         nfs4_state_destroy_net(net);
7251 }
7252
7253 void
7254 nfs4_state_shutdown(void)
7255 {
7256         destroy_workqueue(laundry_wq);
7257         nfsd4_destroy_callback_queue();
7258         cleanup_callback_cred();
7259 }
7260
7261 static void
7262 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7263 {
7264         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7265                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7266 }
7267
7268 static void
7269 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7270 {
7271         if (cstate->minorversion) {
7272                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7273                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7274         }
7275 }
7276
7277 void
7278 clear_current_stateid(struct nfsd4_compound_state *cstate)
7279 {
7280         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7281 }
7282
7283 /*
7284  * functions to set current state id
7285  */
7286 void
7287 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7288                 union nfsd4_op_u *u)
7289 {
7290         put_stateid(cstate, &u->open_downgrade.od_stateid);
7291 }
7292
7293 void
7294 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7295                 union nfsd4_op_u *u)
7296 {
7297         put_stateid(cstate, &u->open.op_stateid);
7298 }
7299
7300 void
7301 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7302                 union nfsd4_op_u *u)
7303 {
7304         put_stateid(cstate, &u->close.cl_stateid);
7305 }
7306
7307 void
7308 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7309                 union nfsd4_op_u *u)
7310 {
7311         put_stateid(cstate, &u->lock.lk_resp_stateid);
7312 }
7313
7314 /*
7315  * functions to consume current state id
7316  */
7317
7318 void
7319 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7320                 union nfsd4_op_u *u)
7321 {
7322         get_stateid(cstate, &u->open_downgrade.od_stateid);
7323 }
7324
7325 void
7326 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7327                 union nfsd4_op_u *u)
7328 {
7329         get_stateid(cstate, &u->delegreturn.dr_stateid);
7330 }
7331
7332 void
7333 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7334                 union nfsd4_op_u *u)
7335 {
7336         get_stateid(cstate, &u->free_stateid.fr_stateid);
7337 }
7338
7339 void
7340 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7341                 union nfsd4_op_u *u)
7342 {
7343         get_stateid(cstate, &u->setattr.sa_stateid);
7344 }
7345
7346 void
7347 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7348                 union nfsd4_op_u *u)
7349 {
7350         get_stateid(cstate, &u->close.cl_stateid);
7351 }
7352
7353 void
7354 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7355                 union nfsd4_op_u *u)
7356 {
7357         get_stateid(cstate, &u->locku.lu_stateid);
7358 }
7359
7360 void
7361 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7362                 union nfsd4_op_u *u)
7363 {
7364         get_stateid(cstate, &u->read.rd_stateid);
7365 }
7366
7367 void
7368 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7369                 union nfsd4_op_u *u)
7370 {
7371         get_stateid(cstate, &u->write.wr_stateid);
7372 }