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(sizeof(struct list_head) *
1811                         OWNER_HASH_SIZE, GFP_KERNEL);
1812         if (!clp->cl_ownerstr_hashtbl)
1813                 goto err_no_hashtbl;
1814         for (i = 0; i < OWNER_HASH_SIZE; i++)
1815                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1816         clp->cl_name.len = name.len;
1817         INIT_LIST_HEAD(&clp->cl_sessions);
1818         idr_init(&clp->cl_stateids);
1819         atomic_set(&clp->cl_refcount, 0);
1820         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1821         INIT_LIST_HEAD(&clp->cl_idhash);
1822         INIT_LIST_HEAD(&clp->cl_openowners);
1823         INIT_LIST_HEAD(&clp->cl_delegations);
1824         INIT_LIST_HEAD(&clp->cl_lru);
1825         INIT_LIST_HEAD(&clp->cl_revoked);
1826 #ifdef CONFIG_NFSD_PNFS
1827         INIT_LIST_HEAD(&clp->cl_lo_states);
1828 #endif
1829         spin_lock_init(&clp->cl_lock);
1830         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1831         return clp;
1832 err_no_hashtbl:
1833         kfree(clp->cl_name.data);
1834 err_no_name:
1835         kmem_cache_free(client_slab, clp);
1836         return NULL;
1837 }
1838
1839 static void
1840 free_client(struct nfs4_client *clp)
1841 {
1842         while (!list_empty(&clp->cl_sessions)) {
1843                 struct nfsd4_session *ses;
1844                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1845                                 se_perclnt);
1846                 list_del(&ses->se_perclnt);
1847                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1848                 free_session(ses);
1849         }
1850         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1851         free_svc_cred(&clp->cl_cred);
1852         kfree(clp->cl_ownerstr_hashtbl);
1853         kfree(clp->cl_name.data);
1854         idr_destroy(&clp->cl_stateids);
1855         kmem_cache_free(client_slab, clp);
1856 }
1857
1858 /* must be called under the client_lock */
1859 static void
1860 unhash_client_locked(struct nfs4_client *clp)
1861 {
1862         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1863         struct nfsd4_session *ses;
1864
1865         lockdep_assert_held(&nn->client_lock);
1866
1867         /* Mark the client as expired! */
1868         clp->cl_time = 0;
1869         /* Make it invisible */
1870         if (!list_empty(&clp->cl_idhash)) {
1871                 list_del_init(&clp->cl_idhash);
1872                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1873                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1874                 else
1875                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1876         }
1877         list_del_init(&clp->cl_lru);
1878         spin_lock(&clp->cl_lock);
1879         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1880                 list_del_init(&ses->se_hash);
1881         spin_unlock(&clp->cl_lock);
1882 }
1883
1884 static void
1885 unhash_client(struct nfs4_client *clp)
1886 {
1887         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1888
1889         spin_lock(&nn->client_lock);
1890         unhash_client_locked(clp);
1891         spin_unlock(&nn->client_lock);
1892 }
1893
1894 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1895 {
1896         if (atomic_read(&clp->cl_refcount))
1897                 return nfserr_jukebox;
1898         unhash_client_locked(clp);
1899         return nfs_ok;
1900 }
1901
1902 static void
1903 __destroy_client(struct nfs4_client *clp)
1904 {
1905         int i;
1906         struct nfs4_openowner *oo;
1907         struct nfs4_delegation *dp;
1908         struct list_head reaplist;
1909
1910         INIT_LIST_HEAD(&reaplist);
1911         spin_lock(&state_lock);
1912         while (!list_empty(&clp->cl_delegations)) {
1913                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1914                 WARN_ON(!unhash_delegation_locked(dp));
1915                 list_add(&dp->dl_recall_lru, &reaplist);
1916         }
1917         spin_unlock(&state_lock);
1918         while (!list_empty(&reaplist)) {
1919                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1920                 list_del_init(&dp->dl_recall_lru);
1921                 destroy_unhashed_deleg(dp);
1922         }
1923         while (!list_empty(&clp->cl_revoked)) {
1924                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1925                 list_del_init(&dp->dl_recall_lru);
1926                 nfs4_put_stid(&dp->dl_stid);
1927         }
1928         while (!list_empty(&clp->cl_openowners)) {
1929                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1930                 nfs4_get_stateowner(&oo->oo_owner);
1931                 release_openowner(oo);
1932         }
1933         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1934                 struct nfs4_stateowner *so, *tmp;
1935
1936                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1937                                          so_strhash) {
1938                         /* Should be no openowners at this point */
1939                         WARN_ON_ONCE(so->so_is_open_owner);
1940                         remove_blocked_locks(lockowner(so));
1941                 }
1942         }
1943         nfsd4_return_all_client_layouts(clp);
1944         nfsd4_shutdown_callback(clp);
1945         if (clp->cl_cb_conn.cb_xprt)
1946                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1947         free_client(clp);
1948 }
1949
1950 static void
1951 destroy_client(struct nfs4_client *clp)
1952 {
1953         unhash_client(clp);
1954         __destroy_client(clp);
1955 }
1956
1957 static void expire_client(struct nfs4_client *clp)
1958 {
1959         unhash_client(clp);
1960         nfsd4_client_record_remove(clp);
1961         __destroy_client(clp);
1962 }
1963
1964 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1965 {
1966         memcpy(target->cl_verifier.data, source->data,
1967                         sizeof(target->cl_verifier.data));
1968 }
1969
1970 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1971 {
1972         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1973         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1974 }
1975
1976 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1977 {
1978         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1979         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1980                                                                 GFP_KERNEL);
1981         if ((source->cr_principal && ! target->cr_principal) ||
1982             (source->cr_raw_principal && ! target->cr_raw_principal))
1983                 return -ENOMEM;
1984
1985         target->cr_flavor = source->cr_flavor;
1986         target->cr_uid = source->cr_uid;
1987         target->cr_gid = source->cr_gid;
1988         target->cr_group_info = source->cr_group_info;
1989         get_group_info(target->cr_group_info);
1990         target->cr_gss_mech = source->cr_gss_mech;
1991         if (source->cr_gss_mech)
1992                 gss_mech_get(source->cr_gss_mech);
1993         return 0;
1994 }
1995
1996 static int
1997 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1998 {
1999         if (o1->len < o2->len)
2000                 return -1;
2001         if (o1->len > o2->len)
2002                 return 1;
2003         return memcmp(o1->data, o2->data, o1->len);
2004 }
2005
2006 static int same_name(const char *n1, const char *n2)
2007 {
2008         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2009 }
2010
2011 static int
2012 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2013 {
2014         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2015 }
2016
2017 static int
2018 same_clid(clientid_t *cl1, clientid_t *cl2)
2019 {
2020         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2021 }
2022
2023 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2024 {
2025         int i;
2026
2027         if (g1->ngroups != g2->ngroups)
2028                 return false;
2029         for (i=0; i<g1->ngroups; i++)
2030                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2031                         return false;
2032         return true;
2033 }
2034
2035 /*
2036  * RFC 3530 language requires clid_inuse be returned when the
2037  * "principal" associated with a requests differs from that previously
2038  * used.  We use uid, gid's, and gss principal string as our best
2039  * approximation.  We also don't want to allow non-gss use of a client
2040  * established using gss: in theory cr_principal should catch that
2041  * change, but in practice cr_principal can be null even in the gss case
2042  * since gssd doesn't always pass down a principal string.
2043  */
2044 static bool is_gss_cred(struct svc_cred *cr)
2045 {
2046         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2047         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2048 }
2049
2050
2051 static bool
2052 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2053 {
2054         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2055                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2056                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2057                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2058                 return false;
2059         if (cr1->cr_principal == cr2->cr_principal)
2060                 return true;
2061         if (!cr1->cr_principal || !cr2->cr_principal)
2062                 return false;
2063         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2064 }
2065
2066 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2067 {
2068         struct svc_cred *cr = &rqstp->rq_cred;
2069         u32 service;
2070
2071         if (!cr->cr_gss_mech)
2072                 return false;
2073         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2074         return service == RPC_GSS_SVC_INTEGRITY ||
2075                service == RPC_GSS_SVC_PRIVACY;
2076 }
2077
2078 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2079 {
2080         struct svc_cred *cr = &rqstp->rq_cred;
2081
2082         if (!cl->cl_mach_cred)
2083                 return true;
2084         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2085                 return false;
2086         if (!svc_rqst_integrity_protected(rqstp))
2087                 return false;
2088         if (cl->cl_cred.cr_raw_principal)
2089                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2090                                                 cr->cr_raw_principal);
2091         if (!cr->cr_principal)
2092                 return false;
2093         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2094 }
2095
2096 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2097 {
2098         __be32 verf[2];
2099
2100         /*
2101          * This is opaque to client, so no need to byte-swap. Use
2102          * __force to keep sparse happy
2103          */
2104         verf[0] = (__force __be32)get_seconds();
2105         verf[1] = (__force __be32)nn->clverifier_counter++;
2106         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2107 }
2108
2109 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2110 {
2111         clp->cl_clientid.cl_boot = nn->boot_time;
2112         clp->cl_clientid.cl_id = nn->clientid_counter++;
2113         gen_confirm(clp, nn);
2114 }
2115
2116 static struct nfs4_stid *
2117 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2118 {
2119         struct nfs4_stid *ret;
2120
2121         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2122         if (!ret || !ret->sc_type)
2123                 return NULL;
2124         return ret;
2125 }
2126
2127 static struct nfs4_stid *
2128 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2129 {
2130         struct nfs4_stid *s;
2131
2132         spin_lock(&cl->cl_lock);
2133         s = find_stateid_locked(cl, t);
2134         if (s != NULL) {
2135                 if (typemask & s->sc_type)
2136                         refcount_inc(&s->sc_count);
2137                 else
2138                         s = NULL;
2139         }
2140         spin_unlock(&cl->cl_lock);
2141         return s;
2142 }
2143
2144 static struct nfs4_client *create_client(struct xdr_netobj name,
2145                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2146 {
2147         struct nfs4_client *clp;
2148         struct sockaddr *sa = svc_addr(rqstp);
2149         int ret;
2150         struct net *net = SVC_NET(rqstp);
2151
2152         clp = alloc_client(name);
2153         if (clp == NULL)
2154                 return NULL;
2155
2156         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2157         if (ret) {
2158                 free_client(clp);
2159                 return NULL;
2160         }
2161         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2162         clp->cl_time = get_seconds();
2163         clear_bit(0, &clp->cl_cb_slot_busy);
2164         copy_verf(clp, verf);
2165         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2166         clp->cl_cb_session = NULL;
2167         clp->net = net;
2168         return clp;
2169 }
2170
2171 static void
2172 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2173 {
2174         struct rb_node **new = &(root->rb_node), *parent = NULL;
2175         struct nfs4_client *clp;
2176
2177         while (*new) {
2178                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2179                 parent = *new;
2180
2181                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2182                         new = &((*new)->rb_left);
2183                 else
2184                         new = &((*new)->rb_right);
2185         }
2186
2187         rb_link_node(&new_clp->cl_namenode, parent, new);
2188         rb_insert_color(&new_clp->cl_namenode, root);
2189 }
2190
2191 static struct nfs4_client *
2192 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2193 {
2194         int cmp;
2195         struct rb_node *node = root->rb_node;
2196         struct nfs4_client *clp;
2197
2198         while (node) {
2199                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2200                 cmp = compare_blob(&clp->cl_name, name);
2201                 if (cmp > 0)
2202                         node = node->rb_left;
2203                 else if (cmp < 0)
2204                         node = node->rb_right;
2205                 else
2206                         return clp;
2207         }
2208         return NULL;
2209 }
2210
2211 static void
2212 add_to_unconfirmed(struct nfs4_client *clp)
2213 {
2214         unsigned int idhashval;
2215         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2216
2217         lockdep_assert_held(&nn->client_lock);
2218
2219         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2220         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2221         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2222         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2223         renew_client_locked(clp);
2224 }
2225
2226 static void
2227 move_to_confirmed(struct nfs4_client *clp)
2228 {
2229         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2230         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2231
2232         lockdep_assert_held(&nn->client_lock);
2233
2234         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2235         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2236         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2237         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2238         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2239         renew_client_locked(clp);
2240 }
2241
2242 static struct nfs4_client *
2243 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2244 {
2245         struct nfs4_client *clp;
2246         unsigned int idhashval = clientid_hashval(clid->cl_id);
2247
2248         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2249                 if (same_clid(&clp->cl_clientid, clid)) {
2250                         if ((bool)clp->cl_minorversion != sessions)
2251                                 return NULL;
2252                         renew_client_locked(clp);
2253                         return clp;
2254                 }
2255         }
2256         return NULL;
2257 }
2258
2259 static struct nfs4_client *
2260 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2261 {
2262         struct list_head *tbl = nn->conf_id_hashtbl;
2263
2264         lockdep_assert_held(&nn->client_lock);
2265         return find_client_in_id_table(tbl, clid, sessions);
2266 }
2267
2268 static struct nfs4_client *
2269 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2270 {
2271         struct list_head *tbl = nn->unconf_id_hashtbl;
2272
2273         lockdep_assert_held(&nn->client_lock);
2274         return find_client_in_id_table(tbl, clid, sessions);
2275 }
2276
2277 static bool clp_used_exchangeid(struct nfs4_client *clp)
2278 {
2279         return clp->cl_exchange_flags != 0;
2280 }
2281
2282 static struct nfs4_client *
2283 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2284 {
2285         lockdep_assert_held(&nn->client_lock);
2286         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2287 }
2288
2289 static struct nfs4_client *
2290 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2291 {
2292         lockdep_assert_held(&nn->client_lock);
2293         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2294 }
2295
2296 static void
2297 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2298 {
2299         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2300         struct sockaddr *sa = svc_addr(rqstp);
2301         u32 scopeid = rpc_get_scope_id(sa);
2302         unsigned short expected_family;
2303
2304         /* Currently, we only support tcp and tcp6 for the callback channel */
2305         if (se->se_callback_netid_len == 3 &&
2306             !memcmp(se->se_callback_netid_val, "tcp", 3))
2307                 expected_family = AF_INET;
2308         else if (se->se_callback_netid_len == 4 &&
2309                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2310                 expected_family = AF_INET6;
2311         else
2312                 goto out_err;
2313
2314         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2315                                             se->se_callback_addr_len,
2316                                             (struct sockaddr *)&conn->cb_addr,
2317                                             sizeof(conn->cb_addr));
2318
2319         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2320                 goto out_err;
2321
2322         if (conn->cb_addr.ss_family == AF_INET6)
2323                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2324
2325         conn->cb_prog = se->se_callback_prog;
2326         conn->cb_ident = se->se_callback_ident;
2327         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2328         return;
2329 out_err:
2330         conn->cb_addr.ss_family = AF_UNSPEC;
2331         conn->cb_addrlen = 0;
2332         dprintk("NFSD: this client (clientid %08x/%08x) "
2333                 "will not receive delegations\n",
2334                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2335
2336         return;
2337 }
2338
2339 /*
2340  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2341  */
2342 static void
2343 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2344 {
2345         struct xdr_buf *buf = resp->xdr.buf;
2346         struct nfsd4_slot *slot = resp->cstate.slot;
2347         unsigned int base;
2348
2349         dprintk("--> %s slot %p\n", __func__, slot);
2350
2351         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2352         slot->sl_opcnt = resp->opcnt;
2353         slot->sl_status = resp->cstate.status;
2354         free_svc_cred(&slot->sl_cred);
2355         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2356
2357         if (!nfsd4_cache_this(resp)) {
2358                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2359                 return;
2360         }
2361         slot->sl_flags |= NFSD4_SLOT_CACHED;
2362
2363         base = resp->cstate.data_offset;
2364         slot->sl_datalen = buf->len - base;
2365         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2366                 WARN(1, "%s: sessions DRC could not cache compound\n",
2367                      __func__);
2368         return;
2369 }
2370
2371 /*
2372  * Encode the replay sequence operation from the slot values.
2373  * If cachethis is FALSE encode the uncached rep error on the next
2374  * operation which sets resp->p and increments resp->opcnt for
2375  * nfs4svc_encode_compoundres.
2376  *
2377  */
2378 static __be32
2379 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2380                           struct nfsd4_compoundres *resp)
2381 {
2382         struct nfsd4_op *op;
2383         struct nfsd4_slot *slot = resp->cstate.slot;
2384
2385         /* Encode the replayed sequence operation */
2386         op = &args->ops[resp->opcnt - 1];
2387         nfsd4_encode_operation(resp, op);
2388
2389         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2390                 return op->status;
2391         if (args->opcnt == 1) {
2392                 /*
2393                  * The original operation wasn't a solo sequence--we
2394                  * always cache those--so this retry must not match the
2395                  * original:
2396                  */
2397                 op->status = nfserr_seq_false_retry;
2398         } else {
2399                 op = &args->ops[resp->opcnt++];
2400                 op->status = nfserr_retry_uncached_rep;
2401                 nfsd4_encode_operation(resp, op);
2402         }
2403         return op->status;
2404 }
2405
2406 /*
2407  * The sequence operation is not cached because we can use the slot and
2408  * session values.
2409  */
2410 static __be32
2411 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2412                          struct nfsd4_sequence *seq)
2413 {
2414         struct nfsd4_slot *slot = resp->cstate.slot;
2415         struct xdr_stream *xdr = &resp->xdr;
2416         __be32 *p;
2417         __be32 status;
2418
2419         dprintk("--> %s slot %p\n", __func__, slot);
2420
2421         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2422         if (status)
2423                 return status;
2424
2425         p = xdr_reserve_space(xdr, slot->sl_datalen);
2426         if (!p) {
2427                 WARN_ON_ONCE(1);
2428                 return nfserr_serverfault;
2429         }
2430         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2431         xdr_commit_encode(xdr);
2432
2433         resp->opcnt = slot->sl_opcnt;
2434         return slot->sl_status;
2435 }
2436
2437 /*
2438  * Set the exchange_id flags returned by the server.
2439  */
2440 static void
2441 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2442 {
2443 #ifdef CONFIG_NFSD_PNFS
2444         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2445 #else
2446         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2447 #endif
2448
2449         /* Referrals are supported, Migration is not. */
2450         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2451
2452         /* set the wire flags to return to client. */
2453         clid->flags = new->cl_exchange_flags;
2454 }
2455
2456 static bool client_has_openowners(struct nfs4_client *clp)
2457 {
2458         struct nfs4_openowner *oo;
2459
2460         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2461                 if (!list_empty(&oo->oo_owner.so_stateids))
2462                         return true;
2463         }
2464         return false;
2465 }
2466
2467 static bool client_has_state(struct nfs4_client *clp)
2468 {
2469         return client_has_openowners(clp)
2470 #ifdef CONFIG_NFSD_PNFS
2471                 || !list_empty(&clp->cl_lo_states)
2472 #endif
2473                 || !list_empty(&clp->cl_delegations)
2474                 || !list_empty(&clp->cl_sessions);
2475 }
2476
2477 __be32
2478 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2479                 union nfsd4_op_u *u)
2480 {
2481         struct nfsd4_exchange_id *exid = &u->exchange_id;
2482         struct nfs4_client *conf, *new;
2483         struct nfs4_client *unconf = NULL;
2484         __be32 status;
2485         char                    addr_str[INET6_ADDRSTRLEN];
2486         nfs4_verifier           verf = exid->verifier;
2487         struct sockaddr         *sa = svc_addr(rqstp);
2488         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2489         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2490
2491         rpc_ntop(sa, addr_str, sizeof(addr_str));
2492         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2493                 "ip_addr=%s flags %x, spa_how %d\n",
2494                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2495                 addr_str, exid->flags, exid->spa_how);
2496
2497         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2498                 return nfserr_inval;
2499
2500         new = create_client(exid->clname, rqstp, &verf);
2501         if (new == NULL)
2502                 return nfserr_jukebox;
2503
2504         switch (exid->spa_how) {
2505         case SP4_MACH_CRED:
2506                 exid->spo_must_enforce[0] = 0;
2507                 exid->spo_must_enforce[1] = (
2508                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2509                         1 << (OP_EXCHANGE_ID - 32) |
2510                         1 << (OP_CREATE_SESSION - 32) |
2511                         1 << (OP_DESTROY_SESSION - 32) |
2512                         1 << (OP_DESTROY_CLIENTID - 32));
2513
2514                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2515                                         1 << (OP_OPEN_DOWNGRADE) |
2516                                         1 << (OP_LOCKU) |
2517                                         1 << (OP_DELEGRETURN));
2518
2519                 exid->spo_must_allow[1] &= (
2520                                         1 << (OP_TEST_STATEID - 32) |
2521                                         1 << (OP_FREE_STATEID - 32));
2522                 if (!svc_rqst_integrity_protected(rqstp)) {
2523                         status = nfserr_inval;
2524                         goto out_nolock;
2525                 }
2526                 /*
2527                  * Sometimes userspace doesn't give us a principal.
2528                  * Which is a bug, really.  Anyway, we can't enforce
2529                  * MACH_CRED in that case, better to give up now:
2530                  */
2531                 if (!new->cl_cred.cr_principal &&
2532                                         !new->cl_cred.cr_raw_principal) {
2533                         status = nfserr_serverfault;
2534                         goto out_nolock;
2535                 }
2536                 new->cl_mach_cred = true;
2537         case SP4_NONE:
2538                 break;
2539         default:                                /* checked by xdr code */
2540                 WARN_ON_ONCE(1);
2541         case SP4_SSV:
2542                 status = nfserr_encr_alg_unsupp;
2543                 goto out_nolock;
2544         }
2545
2546         /* Cases below refer to rfc 5661 section 18.35.4: */
2547         spin_lock(&nn->client_lock);
2548         conf = find_confirmed_client_by_name(&exid->clname, nn);
2549         if (conf) {
2550                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2551                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2552
2553                 if (update) {
2554                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2555                                 status = nfserr_inval;
2556                                 goto out;
2557                         }
2558                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2559                                 status = nfserr_wrong_cred;
2560                                 goto out;
2561                         }
2562                         if (!creds_match) { /* case 9 */
2563                                 status = nfserr_perm;
2564                                 goto out;
2565                         }
2566                         if (!verfs_match) { /* case 8 */
2567                                 status = nfserr_not_same;
2568                                 goto out;
2569                         }
2570                         /* case 6 */
2571                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2572                         goto out_copy;
2573                 }
2574                 if (!creds_match) { /* case 3 */
2575                         if (client_has_state(conf)) {
2576                                 status = nfserr_clid_inuse;
2577                                 goto out;
2578                         }
2579                         goto out_new;
2580                 }
2581                 if (verfs_match) { /* case 2 */
2582                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2583                         goto out_copy;
2584                 }
2585                 /* case 5, client reboot */
2586                 conf = NULL;
2587                 goto out_new;
2588         }
2589
2590         if (update) { /* case 7 */
2591                 status = nfserr_noent;
2592                 goto out;
2593         }
2594
2595         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2596         if (unconf) /* case 4, possible retry or client restart */
2597                 unhash_client_locked(unconf);
2598
2599         /* case 1 (normal case) */
2600 out_new:
2601         if (conf) {
2602                 status = mark_client_expired_locked(conf);
2603                 if (status)
2604                         goto out;
2605         }
2606         new->cl_minorversion = cstate->minorversion;
2607         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2608         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2609
2610         gen_clid(new, nn);
2611         add_to_unconfirmed(new);
2612         swap(new, conf);
2613 out_copy:
2614         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2615         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2616
2617         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2618         nfsd4_set_ex_flags(conf, exid);
2619
2620         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2621                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2622         status = nfs_ok;
2623
2624 out:
2625         spin_unlock(&nn->client_lock);
2626 out_nolock:
2627         if (new)
2628                 expire_client(new);
2629         if (unconf)
2630                 expire_client(unconf);
2631         return status;
2632 }
2633
2634 static __be32
2635 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2636 {
2637         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2638                 slot_seqid);
2639
2640         /* The slot is in use, and no response has been sent. */
2641         if (slot_inuse) {
2642                 if (seqid == slot_seqid)
2643                         return nfserr_jukebox;
2644                 else
2645                         return nfserr_seq_misordered;
2646         }
2647         /* Note unsigned 32-bit arithmetic handles wraparound: */
2648         if (likely(seqid == slot_seqid + 1))
2649                 return nfs_ok;
2650         if (seqid == slot_seqid)
2651                 return nfserr_replay_cache;
2652         return nfserr_seq_misordered;
2653 }
2654
2655 /*
2656  * Cache the create session result into the create session single DRC
2657  * slot cache by saving the xdr structure. sl_seqid has been set.
2658  * Do this for solo or embedded create session operations.
2659  */
2660 static void
2661 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2662                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2663 {
2664         slot->sl_status = nfserr;
2665         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2666 }
2667
2668 static __be32
2669 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2670                             struct nfsd4_clid_slot *slot)
2671 {
2672         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2673         return slot->sl_status;
2674 }
2675
2676 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2677                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2678                         1 +     /* MIN tag is length with zero, only length */ \
2679                         3 +     /* version, opcount, opcode */ \
2680                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2681                                 /* seqid, slotID, slotID, cache */ \
2682                         4 ) * sizeof(__be32))
2683
2684 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2685                         2 +     /* verifier: AUTH_NULL, length 0 */\
2686                         1 +     /* status */ \
2687                         1 +     /* MIN tag is length with zero, only length */ \
2688                         3 +     /* opcount, opcode, opstatus*/ \
2689                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2690                                 /* seqid, slotID, slotID, slotID, status */ \
2691                         5 ) * sizeof(__be32))
2692
2693 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2694 {
2695         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2696
2697         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2698                 return nfserr_toosmall;
2699         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2700                 return nfserr_toosmall;
2701         ca->headerpadsz = 0;
2702         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2703         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2704         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2705         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2706                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2707         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2708         /*
2709          * Note decreasing slot size below client's request may make it
2710          * difficult for client to function correctly, whereas
2711          * decreasing the number of slots will (just?) affect
2712          * performance.  When short on memory we therefore prefer to
2713          * decrease number of slots instead of their size.  Clients that
2714          * request larger slots than they need will get poor results:
2715          */
2716         ca->maxreqs = nfsd4_get_drc_mem(ca);
2717         if (!ca->maxreqs)
2718                 return nfserr_jukebox;
2719
2720         return nfs_ok;
2721 }
2722
2723 /*
2724  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2725  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2726  */
2727 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2728         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2729
2730 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2731         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2732
2733 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2734                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2735 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2736                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2737                                  sizeof(__be32))
2738
2739 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2740 {
2741         ca->headerpadsz = 0;
2742
2743         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2744                 return nfserr_toosmall;
2745         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2746                 return nfserr_toosmall;
2747         ca->maxresp_cached = 0;
2748         if (ca->maxops < 2)
2749                 return nfserr_toosmall;
2750
2751         return nfs_ok;
2752 }
2753
2754 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2755 {
2756         switch (cbs->flavor) {
2757         case RPC_AUTH_NULL:
2758         case RPC_AUTH_UNIX:
2759                 return nfs_ok;
2760         default:
2761                 /*
2762                  * GSS case: the spec doesn't allow us to return this
2763                  * error.  But it also doesn't allow us not to support
2764                  * GSS.
2765                  * I'd rather this fail hard than return some error the
2766                  * client might think it can already handle:
2767                  */
2768                 return nfserr_encr_alg_unsupp;
2769         }
2770 }
2771
2772 __be32
2773 nfsd4_create_session(struct svc_rqst *rqstp,
2774                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2775 {
2776         struct nfsd4_create_session *cr_ses = &u->create_session;
2777         struct sockaddr *sa = svc_addr(rqstp);
2778         struct nfs4_client *conf, *unconf;
2779         struct nfs4_client *old = NULL;
2780         struct nfsd4_session *new;
2781         struct nfsd4_conn *conn;
2782         struct nfsd4_clid_slot *cs_slot = NULL;
2783         __be32 status = 0;
2784         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2785
2786         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2787                 return nfserr_inval;
2788         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2789         if (status)
2790                 return status;
2791         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2792         if (status)
2793                 return status;
2794         status = check_backchannel_attrs(&cr_ses->back_channel);
2795         if (status)
2796                 goto out_release_drc_mem;
2797         status = nfserr_jukebox;
2798         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2799         if (!new)
2800                 goto out_release_drc_mem;
2801         conn = alloc_conn_from_crses(rqstp, cr_ses);
2802         if (!conn)
2803                 goto out_free_session;
2804
2805         spin_lock(&nn->client_lock);
2806         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2807         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2808         WARN_ON_ONCE(conf && unconf);
2809
2810         if (conf) {
2811                 status = nfserr_wrong_cred;
2812                 if (!nfsd4_mach_creds_match(conf, rqstp))
2813                         goto out_free_conn;
2814                 cs_slot = &conf->cl_cs_slot;
2815                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2816                 if (status) {
2817                         if (status == nfserr_replay_cache)
2818                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2819                         goto out_free_conn;
2820                 }
2821         } else if (unconf) {
2822                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2823                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2824                         status = nfserr_clid_inuse;
2825                         goto out_free_conn;
2826                 }
2827                 status = nfserr_wrong_cred;
2828                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2829                         goto out_free_conn;
2830                 cs_slot = &unconf->cl_cs_slot;
2831                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2832                 if (status) {
2833                         /* an unconfirmed replay returns misordered */
2834                         status = nfserr_seq_misordered;
2835                         goto out_free_conn;
2836                 }
2837                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2838                 if (old) {
2839                         status = mark_client_expired_locked(old);
2840                         if (status) {
2841                                 old = NULL;
2842                                 goto out_free_conn;
2843                         }
2844                 }
2845                 move_to_confirmed(unconf);
2846                 conf = unconf;
2847         } else {
2848                 status = nfserr_stale_clientid;
2849                 goto out_free_conn;
2850         }
2851         status = nfs_ok;
2852         /* Persistent sessions are not supported */
2853         cr_ses->flags &= ~SESSION4_PERSIST;
2854         /* Upshifting from TCP to RDMA is not supported */
2855         cr_ses->flags &= ~SESSION4_RDMA;
2856
2857         init_session(rqstp, new, conf, cr_ses);
2858         nfsd4_get_session_locked(new);
2859
2860         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2861                NFS4_MAX_SESSIONID_LEN);
2862         cs_slot->sl_seqid++;
2863         cr_ses->seqid = cs_slot->sl_seqid;
2864
2865         /* cache solo and embedded create sessions under the client_lock */
2866         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2867         spin_unlock(&nn->client_lock);
2868         /* init connection and backchannel */
2869         nfsd4_init_conn(rqstp, conn, new);
2870         nfsd4_put_session(new);
2871         if (old)
2872                 expire_client(old);
2873         return status;
2874 out_free_conn:
2875         spin_unlock(&nn->client_lock);
2876         free_conn(conn);
2877         if (old)
2878                 expire_client(old);
2879 out_free_session:
2880         __free_session(new);
2881 out_release_drc_mem:
2882         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2883         return status;
2884 }
2885
2886 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2887 {
2888         switch (*dir) {
2889         case NFS4_CDFC4_FORE:
2890         case NFS4_CDFC4_BACK:
2891                 return nfs_ok;
2892         case NFS4_CDFC4_FORE_OR_BOTH:
2893         case NFS4_CDFC4_BACK_OR_BOTH:
2894                 *dir = NFS4_CDFC4_BOTH;
2895                 return nfs_ok;
2896         };
2897         return nfserr_inval;
2898 }
2899
2900 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2901                 struct nfsd4_compound_state *cstate,
2902                 union nfsd4_op_u *u)
2903 {
2904         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2905         struct nfsd4_session *session = cstate->session;
2906         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2907         __be32 status;
2908
2909         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2910         if (status)
2911                 return status;
2912         spin_lock(&nn->client_lock);
2913         session->se_cb_prog = bc->bc_cb_program;
2914         session->se_cb_sec = bc->bc_cb_sec;
2915         spin_unlock(&nn->client_lock);
2916
2917         nfsd4_probe_callback(session->se_client);
2918
2919         return nfs_ok;
2920 }
2921
2922 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2923                      struct nfsd4_compound_state *cstate,
2924                      union nfsd4_op_u *u)
2925 {
2926         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2927         __be32 status;
2928         struct nfsd4_conn *conn;
2929         struct nfsd4_session *session;
2930         struct net *net = SVC_NET(rqstp);
2931         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2932
2933         if (!nfsd4_last_compound_op(rqstp))
2934                 return nfserr_not_only_op;
2935         spin_lock(&nn->client_lock);
2936         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2937         spin_unlock(&nn->client_lock);
2938         if (!session)
2939                 goto out_no_session;
2940         status = nfserr_wrong_cred;
2941         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2942                 goto out;
2943         status = nfsd4_map_bcts_dir(&bcts->dir);
2944         if (status)
2945                 goto out;
2946         conn = alloc_conn(rqstp, bcts->dir);
2947         status = nfserr_jukebox;
2948         if (!conn)
2949                 goto out;
2950         nfsd4_init_conn(rqstp, conn, session);
2951         status = nfs_ok;
2952 out:
2953         nfsd4_put_session(session);
2954 out_no_session:
2955         return status;
2956 }
2957
2958 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2959 {
2960         if (!session)
2961                 return false;
2962         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2963 }
2964
2965 __be32
2966 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2967                 union nfsd4_op_u *u)
2968 {
2969         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2970         struct nfsd4_session *ses;
2971         __be32 status;
2972         int ref_held_by_me = 0;
2973         struct net *net = SVC_NET(r);
2974         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2975
2976         status = nfserr_not_only_op;
2977         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2978                 if (!nfsd4_last_compound_op(r))
2979                         goto out;
2980                 ref_held_by_me++;
2981         }
2982         dump_sessionid(__func__, &sessionid->sessionid);
2983         spin_lock(&nn->client_lock);
2984         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2985         if (!ses)
2986                 goto out_client_lock;
2987         status = nfserr_wrong_cred;
2988         if (!nfsd4_mach_creds_match(ses->se_client, r))
2989                 goto out_put_session;
2990         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2991         if (status)
2992                 goto out_put_session;
2993         unhash_session(ses);
2994         spin_unlock(&nn->client_lock);
2995
2996         nfsd4_probe_callback_sync(ses->se_client);
2997
2998         spin_lock(&nn->client_lock);
2999         status = nfs_ok;
3000 out_put_session:
3001         nfsd4_put_session_locked(ses);
3002 out_client_lock:
3003         spin_unlock(&nn->client_lock);
3004 out:
3005         return status;
3006 }
3007
3008 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3009 {
3010         struct nfsd4_conn *c;
3011
3012         list_for_each_entry(c, &s->se_conns, cn_persession) {
3013                 if (c->cn_xprt == xpt) {
3014                         return c;
3015                 }
3016         }
3017         return NULL;
3018 }
3019
3020 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3021 {
3022         struct nfs4_client *clp = ses->se_client;
3023         struct nfsd4_conn *c;
3024         __be32 status = nfs_ok;
3025         int ret;
3026
3027         spin_lock(&clp->cl_lock);
3028         c = __nfsd4_find_conn(new->cn_xprt, ses);
3029         if (c)
3030                 goto out_free;
3031         status = nfserr_conn_not_bound_to_session;
3032         if (clp->cl_mach_cred)
3033                 goto out_free;
3034         __nfsd4_hash_conn(new, ses);
3035         spin_unlock(&clp->cl_lock);
3036         ret = nfsd4_register_conn(new);
3037         if (ret)
3038                 /* oops; xprt is already down: */
3039                 nfsd4_conn_lost(&new->cn_xpt_user);
3040         return nfs_ok;
3041 out_free:
3042         spin_unlock(&clp->cl_lock);
3043         free_conn(new);
3044         return status;
3045 }
3046
3047 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3048 {
3049         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3050
3051         return args->opcnt > session->se_fchannel.maxops;
3052 }
3053
3054 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3055                                   struct nfsd4_session *session)
3056 {
3057         struct xdr_buf *xb = &rqstp->rq_arg;
3058
3059         return xb->len > session->se_fchannel.maxreq_sz;
3060 }
3061
3062 static bool replay_matches_cache(struct svc_rqst *rqstp,
3063                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3064 {
3065         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3066
3067         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3068             (bool)seq->cachethis)
3069                 return false;
3070         /*
3071          * If there's an error than the reply can have fewer ops than
3072          * the call.  But if we cached a reply with *more* ops than the
3073          * call you're sending us now, then this new call is clearly not
3074          * really a replay of the old one:
3075          */
3076         if (slot->sl_opcnt < argp->opcnt)
3077                 return false;
3078         /* This is the only check explicitly called by spec: */
3079         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3080                 return false;
3081         /*
3082          * There may be more comparisons we could actually do, but the
3083          * spec doesn't require us to catch every case where the calls
3084          * don't match (that would require caching the call as well as
3085          * the reply), so we don't bother.
3086          */
3087         return true;
3088 }
3089
3090 __be32
3091 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3092                 union nfsd4_op_u *u)
3093 {
3094         struct nfsd4_sequence *seq = &u->sequence;
3095         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3096         struct xdr_stream *xdr = &resp->xdr;
3097         struct nfsd4_session *session;
3098         struct nfs4_client *clp;
3099         struct nfsd4_slot *slot;
3100         struct nfsd4_conn *conn;
3101         __be32 status;
3102         int buflen;
3103         struct net *net = SVC_NET(rqstp);
3104         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3105
3106         if (resp->opcnt != 1)
3107                 return nfserr_sequence_pos;
3108
3109         /*
3110          * Will be either used or freed by nfsd4_sequence_check_conn
3111          * below.
3112          */
3113         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3114         if (!conn)
3115                 return nfserr_jukebox;
3116
3117         spin_lock(&nn->client_lock);
3118         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3119         if (!session)
3120                 goto out_no_session;
3121         clp = session->se_client;
3122
3123         status = nfserr_too_many_ops;
3124         if (nfsd4_session_too_many_ops(rqstp, session))
3125                 goto out_put_session;
3126
3127         status = nfserr_req_too_big;
3128         if (nfsd4_request_too_big(rqstp, session))
3129                 goto out_put_session;
3130
3131         status = nfserr_badslot;
3132         if (seq->slotid >= session->se_fchannel.maxreqs)
3133                 goto out_put_session;
3134
3135         slot = session->se_slots[seq->slotid];
3136         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3137
3138         /* We do not negotiate the number of slots yet, so set the
3139          * maxslots to the session maxreqs which is used to encode
3140          * sr_highest_slotid and the sr_target_slot id to maxslots */
3141         seq->maxslots = session->se_fchannel.maxreqs;
3142
3143         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3144                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3145         if (status == nfserr_replay_cache) {
3146                 status = nfserr_seq_misordered;
3147                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3148                         goto out_put_session;
3149                 status = nfserr_seq_false_retry;
3150                 if (!replay_matches_cache(rqstp, seq, slot))
3151                         goto out_put_session;
3152                 cstate->slot = slot;
3153                 cstate->session = session;
3154                 cstate->clp = clp;
3155                 /* Return the cached reply status and set cstate->status
3156                  * for nfsd4_proc_compound processing */
3157                 status = nfsd4_replay_cache_entry(resp, seq);
3158                 cstate->status = nfserr_replay_cache;
3159                 goto out;
3160         }
3161         if (status)
3162                 goto out_put_session;
3163
3164         status = nfsd4_sequence_check_conn(conn, session);
3165         conn = NULL;
3166         if (status)
3167                 goto out_put_session;
3168
3169         buflen = (seq->cachethis) ?
3170                         session->se_fchannel.maxresp_cached :
3171                         session->se_fchannel.maxresp_sz;
3172         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3173                                     nfserr_rep_too_big;
3174         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3175                 goto out_put_session;
3176         svc_reserve(rqstp, buflen);
3177
3178         status = nfs_ok;
3179         /* Success! bump slot seqid */
3180         slot->sl_seqid = seq->seqid;
3181         slot->sl_flags |= NFSD4_SLOT_INUSE;
3182         if (seq->cachethis)
3183                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3184         else
3185                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3186
3187         cstate->slot = slot;
3188         cstate->session = session;
3189         cstate->clp = clp;
3190
3191 out:
3192         switch (clp->cl_cb_state) {
3193         case NFSD4_CB_DOWN:
3194                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3195                 break;
3196         case NFSD4_CB_FAULT:
3197                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3198                 break;
3199         default:
3200                 seq->status_flags = 0;
3201         }
3202         if (!list_empty(&clp->cl_revoked))
3203                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3204 out_no_session:
3205         if (conn)
3206                 free_conn(conn);
3207         spin_unlock(&nn->client_lock);
3208         return status;
3209 out_put_session:
3210         nfsd4_put_session_locked(session);
3211         goto out_no_session;
3212 }
3213
3214 void
3215 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3216 {
3217         struct nfsd4_compound_state *cs = &resp->cstate;
3218
3219         if (nfsd4_has_session(cs)) {
3220                 if (cs->status != nfserr_replay_cache) {
3221                         nfsd4_store_cache_entry(resp);
3222                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3223                 }
3224                 /* Drop session reference that was taken in nfsd4_sequence() */
3225                 nfsd4_put_session(cs->session);
3226         } else if (cs->clp)
3227                 put_client_renew(cs->clp);
3228 }
3229
3230 __be32
3231 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3232                 struct nfsd4_compound_state *cstate,
3233                 union nfsd4_op_u *u)
3234 {
3235         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3236         struct nfs4_client *conf, *unconf;
3237         struct nfs4_client *clp = NULL;
3238         __be32 status = 0;
3239         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3240
3241         spin_lock(&nn->client_lock);
3242         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3243         conf = find_confirmed_client(&dc->clientid, true, nn);
3244         WARN_ON_ONCE(conf && unconf);
3245
3246         if (conf) {
3247                 if (client_has_state(conf)) {
3248                         status = nfserr_clientid_busy;
3249                         goto out;
3250                 }
3251                 status = mark_client_expired_locked(conf);
3252                 if (status)
3253                         goto out;
3254                 clp = conf;
3255         } else if (unconf)
3256                 clp = unconf;
3257         else {
3258                 status = nfserr_stale_clientid;
3259                 goto out;
3260         }
3261         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3262                 clp = NULL;
3263                 status = nfserr_wrong_cred;
3264                 goto out;
3265         }
3266         unhash_client_locked(clp);
3267 out:
3268         spin_unlock(&nn->client_lock);
3269         if (clp)
3270                 expire_client(clp);
3271         return status;
3272 }
3273
3274 __be32
3275 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3276                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3277 {
3278         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3279         __be32 status = 0;
3280
3281         if (rc->rca_one_fs) {
3282                 if (!cstate->current_fh.fh_dentry)
3283                         return nfserr_nofilehandle;
3284                 /*
3285                  * We don't take advantage of the rca_one_fs case.
3286                  * That's OK, it's optional, we can safely ignore it.
3287                  */
3288                 return nfs_ok;
3289         }
3290
3291         status = nfserr_complete_already;
3292         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3293                              &cstate->session->se_client->cl_flags))
3294                 goto out;
3295
3296         status = nfserr_stale_clientid;
3297         if (is_client_expired(cstate->session->se_client))
3298                 /*
3299                  * The following error isn't really legal.
3300                  * But we only get here if the client just explicitly
3301                  * destroyed the client.  Surely it no longer cares what
3302                  * error it gets back on an operation for the dead
3303                  * client.
3304                  */
3305                 goto out;
3306
3307         status = nfs_ok;
3308         nfsd4_client_record_create(cstate->session->se_client);
3309 out:
3310         return status;
3311 }
3312
3313 __be32
3314 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3315                   union nfsd4_op_u *u)
3316 {
3317         struct nfsd4_setclientid *setclid = &u->setclientid;
3318         struct xdr_netobj       clname = setclid->se_name;
3319         nfs4_verifier           clverifier = setclid->se_verf;
3320         struct nfs4_client      *conf, *new;
3321         struct nfs4_client      *unconf = NULL;
3322         __be32                  status;
3323         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3324
3325         new = create_client(clname, rqstp, &clverifier);
3326         if (new == NULL)
3327                 return nfserr_jukebox;
3328         /* Cases below refer to rfc 3530 section 14.2.33: */
3329         spin_lock(&nn->client_lock);
3330         conf = find_confirmed_client_by_name(&clname, nn);
3331         if (conf && client_has_state(conf)) {
3332                 /* case 0: */
3333                 status = nfserr_clid_inuse;
3334                 if (clp_used_exchangeid(conf))
3335                         goto out;
3336                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3337                         char addr_str[INET6_ADDRSTRLEN];
3338                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3339                                  sizeof(addr_str));
3340                         dprintk("NFSD: setclientid: string in use by client "
3341                                 "at %s\n", addr_str);
3342                         goto out;
3343                 }
3344         }
3345         unconf = find_unconfirmed_client_by_name(&clname, nn);
3346         if (unconf)
3347                 unhash_client_locked(unconf);
3348         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3349                 /* case 1: probable callback update */
3350                 copy_clid(new, conf);
3351                 gen_confirm(new, nn);
3352         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3353                 gen_clid(new, nn);
3354         new->cl_minorversion = 0;
3355         gen_callback(new, setclid, rqstp);
3356         add_to_unconfirmed(new);
3357         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3358         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3359         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3360         new = NULL;
3361         status = nfs_ok;
3362 out:
3363         spin_unlock(&nn->client_lock);
3364         if (new)
3365                 free_client(new);
3366         if (unconf)
3367                 expire_client(unconf);
3368         return status;
3369 }
3370
3371
3372 __be32
3373 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3374                         struct nfsd4_compound_state *cstate,
3375                         union nfsd4_op_u *u)
3376 {
3377         struct nfsd4_setclientid_confirm *setclientid_confirm =
3378                         &u->setclientid_confirm;
3379         struct nfs4_client *conf, *unconf;
3380         struct nfs4_client *old = NULL;
3381         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3382         clientid_t * clid = &setclientid_confirm->sc_clientid;
3383         __be32 status;
3384         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3385
3386         if (STALE_CLIENTID(clid, nn))
3387                 return nfserr_stale_clientid;
3388
3389         spin_lock(&nn->client_lock);
3390         conf = find_confirmed_client(clid, false, nn);
3391         unconf = find_unconfirmed_client(clid, false, nn);
3392         /*
3393          * We try hard to give out unique clientid's, so if we get an
3394          * attempt to confirm the same clientid with a different cred,
3395          * the client may be buggy; this should never happen.
3396          *
3397          * Nevertheless, RFC 7530 recommends INUSE for this case:
3398          */
3399         status = nfserr_clid_inuse;
3400         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3401                 goto out;
3402         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3403                 goto out;
3404         /* cases below refer to rfc 3530 section 14.2.34: */
3405         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3406                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3407                         /* case 2: probable retransmit */
3408                         status = nfs_ok;
3409                 } else /* case 4: client hasn't noticed we rebooted yet? */
3410                         status = nfserr_stale_clientid;
3411                 goto out;
3412         }
3413         status = nfs_ok;
3414         if (conf) { /* case 1: callback update */
3415                 old = unconf;
3416                 unhash_client_locked(old);
3417                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3418         } else { /* case 3: normal case; new or rebooted client */
3419                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3420                 if (old) {
3421                         status = nfserr_clid_inuse;
3422                         if (client_has_state(old)
3423                                         && !same_creds(&unconf->cl_cred,
3424                                                         &old->cl_cred))
3425                                 goto out;
3426                         status = mark_client_expired_locked(old);
3427                         if (status) {
3428                                 old = NULL;
3429                                 goto out;
3430                         }
3431                 }
3432                 move_to_confirmed(unconf);
3433                 conf = unconf;
3434         }
3435         get_client_locked(conf);
3436         spin_unlock(&nn->client_lock);
3437         nfsd4_probe_callback(conf);
3438         spin_lock(&nn->client_lock);
3439         put_client_renew_locked(conf);
3440 out:
3441         spin_unlock(&nn->client_lock);
3442         if (old)
3443                 expire_client(old);
3444         return status;
3445 }
3446
3447 static struct nfs4_file *nfsd4_alloc_file(void)
3448 {
3449         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3450 }
3451
3452 /* OPEN Share state helper functions */
3453 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3454                                 struct nfs4_file *fp)
3455 {
3456         lockdep_assert_held(&state_lock);
3457
3458         refcount_set(&fp->fi_ref, 1);
3459         spin_lock_init(&fp->fi_lock);
3460         INIT_LIST_HEAD(&fp->fi_stateids);
3461         INIT_LIST_HEAD(&fp->fi_delegations);
3462         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3463         fh_copy_shallow(&fp->fi_fhandle, fh);
3464         fp->fi_deleg_file = NULL;
3465         fp->fi_had_conflict = false;
3466         fp->fi_share_deny = 0;
3467         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3468         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3469 #ifdef CONFIG_NFSD_PNFS
3470         INIT_LIST_HEAD(&fp->fi_lo_states);
3471         atomic_set(&fp->fi_lo_recalls, 0);
3472 #endif
3473         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3474 }
3475
3476 void
3477 nfsd4_free_slabs(void)
3478 {
3479         kmem_cache_destroy(client_slab);
3480         kmem_cache_destroy(openowner_slab);
3481         kmem_cache_destroy(lockowner_slab);
3482         kmem_cache_destroy(file_slab);
3483         kmem_cache_destroy(stateid_slab);
3484         kmem_cache_destroy(deleg_slab);
3485         kmem_cache_destroy(odstate_slab);
3486 }
3487
3488 int
3489 nfsd4_init_slabs(void)
3490 {
3491         client_slab = kmem_cache_create("nfsd4_clients",
3492                         sizeof(struct nfs4_client), 0, 0, NULL);
3493         if (client_slab == NULL)
3494                 goto out;
3495         openowner_slab = kmem_cache_create("nfsd4_openowners",
3496                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3497         if (openowner_slab == NULL)
3498                 goto out_free_client_slab;
3499         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3500                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3501         if (lockowner_slab == NULL)
3502                 goto out_free_openowner_slab;
3503         file_slab = kmem_cache_create("nfsd4_files",
3504                         sizeof(struct nfs4_file), 0, 0, NULL);
3505         if (file_slab == NULL)
3506                 goto out_free_lockowner_slab;
3507         stateid_slab = kmem_cache_create("nfsd4_stateids",
3508                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3509         if (stateid_slab == NULL)
3510                 goto out_free_file_slab;
3511         deleg_slab = kmem_cache_create("nfsd4_delegations",
3512                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3513         if (deleg_slab == NULL)
3514                 goto out_free_stateid_slab;
3515         odstate_slab = kmem_cache_create("nfsd4_odstate",
3516                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3517         if (odstate_slab == NULL)
3518                 goto out_free_deleg_slab;
3519         return 0;
3520
3521 out_free_deleg_slab:
3522         kmem_cache_destroy(deleg_slab);
3523 out_free_stateid_slab:
3524         kmem_cache_destroy(stateid_slab);
3525 out_free_file_slab:
3526         kmem_cache_destroy(file_slab);
3527 out_free_lockowner_slab:
3528         kmem_cache_destroy(lockowner_slab);
3529 out_free_openowner_slab:
3530         kmem_cache_destroy(openowner_slab);
3531 out_free_client_slab:
3532         kmem_cache_destroy(client_slab);
3533 out:
3534         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3535         return -ENOMEM;
3536 }
3537
3538 static void init_nfs4_replay(struct nfs4_replay *rp)
3539 {
3540         rp->rp_status = nfserr_serverfault;
3541         rp->rp_buflen = 0;
3542         rp->rp_buf = rp->rp_ibuf;
3543         mutex_init(&rp->rp_mutex);
3544 }
3545
3546 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3547                 struct nfs4_stateowner *so)
3548 {
3549         if (!nfsd4_has_session(cstate)) {
3550                 mutex_lock(&so->so_replay.rp_mutex);
3551                 cstate->replay_owner = nfs4_get_stateowner(so);
3552         }
3553 }
3554
3555 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3556 {
3557         struct nfs4_stateowner *so = cstate->replay_owner;
3558
3559         if (so != NULL) {
3560                 cstate->replay_owner = NULL;
3561                 mutex_unlock(&so->so_replay.rp_mutex);
3562                 nfs4_put_stateowner(so);
3563         }
3564 }
3565
3566 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3567 {
3568         struct nfs4_stateowner *sop;
3569
3570         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3571         if (!sop)
3572                 return NULL;
3573
3574         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3575         if (!sop->so_owner.data) {
3576                 kmem_cache_free(slab, sop);
3577                 return NULL;
3578         }
3579         sop->so_owner.len = owner->len;
3580
3581         INIT_LIST_HEAD(&sop->so_stateids);
3582         sop->so_client = clp;
3583         init_nfs4_replay(&sop->so_replay);
3584         atomic_set(&sop->so_count, 1);
3585         return sop;
3586 }
3587
3588 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3589 {
3590         lockdep_assert_held(&clp->cl_lock);
3591
3592         list_add(&oo->oo_owner.so_strhash,
3593                  &clp->cl_ownerstr_hashtbl[strhashval]);
3594         list_add(&oo->oo_perclient, &clp->cl_openowners);
3595 }
3596
3597 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3598 {
3599         unhash_openowner_locked(openowner(so));
3600 }
3601
3602 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3603 {
3604         struct nfs4_openowner *oo = openowner(so);
3605
3606         kmem_cache_free(openowner_slab, oo);
3607 }
3608
3609 static const struct nfs4_stateowner_operations openowner_ops = {
3610         .so_unhash =    nfs4_unhash_openowner,
3611         .so_free =      nfs4_free_openowner,
3612 };
3613
3614 static struct nfs4_ol_stateid *
3615 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3616 {
3617         struct nfs4_ol_stateid *local, *ret = NULL;
3618         struct nfs4_openowner *oo = open->op_openowner;
3619
3620         lockdep_assert_held(&fp->fi_lock);
3621
3622         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3623                 /* ignore lock owners */
3624                 if (local->st_stateowner->so_is_open_owner == 0)
3625                         continue;
3626                 if (local->st_stateowner != &oo->oo_owner)
3627                         continue;
3628                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3629                         ret = local;
3630                         refcount_inc(&ret->st_stid.sc_count);
3631                         break;
3632                 }
3633         }
3634         return ret;
3635 }
3636
3637 static __be32
3638 nfsd4_verify_open_stid(struct nfs4_stid *s)
3639 {
3640         __be32 ret = nfs_ok;
3641
3642         switch (s->sc_type) {
3643         default:
3644                 break;
3645         case 0:
3646         case NFS4_CLOSED_STID:
3647         case NFS4_CLOSED_DELEG_STID:
3648                 ret = nfserr_bad_stateid;
3649                 break;
3650         case NFS4_REVOKED_DELEG_STID:
3651                 ret = nfserr_deleg_revoked;
3652         }
3653         return ret;
3654 }
3655
3656 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3657 static __be32
3658 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3659 {
3660         __be32 ret;
3661
3662         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3663         ret = nfsd4_verify_open_stid(&stp->st_stid);
3664         if (ret != nfs_ok)
3665                 mutex_unlock(&stp->st_mutex);
3666         return ret;
3667 }
3668
3669 static struct nfs4_ol_stateid *
3670 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3671 {
3672         struct nfs4_ol_stateid *stp;
3673         for (;;) {
3674                 spin_lock(&fp->fi_lock);
3675                 stp = nfsd4_find_existing_open(fp, open);
3676                 spin_unlock(&fp->fi_lock);
3677                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3678                         break;
3679                 nfs4_put_stid(&stp->st_stid);
3680         }
3681         return stp;
3682 }
3683
3684 static struct nfs4_openowner *
3685 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3686                            struct nfsd4_compound_state *cstate)
3687 {
3688         struct nfs4_client *clp = cstate->clp;
3689         struct nfs4_openowner *oo, *ret;
3690
3691         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3692         if (!oo)
3693                 return NULL;
3694         oo->oo_owner.so_ops = &openowner_ops;
3695         oo->oo_owner.so_is_open_owner = 1;
3696         oo->oo_owner.so_seqid = open->op_seqid;
3697         oo->oo_flags = 0;
3698         if (nfsd4_has_session(cstate))
3699                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3700         oo->oo_time = 0;
3701         oo->oo_last_closed_stid = NULL;
3702         INIT_LIST_HEAD(&oo->oo_close_lru);
3703         spin_lock(&clp->cl_lock);
3704         ret = find_openstateowner_str_locked(strhashval, open, clp);
3705         if (ret == NULL) {
3706                 hash_openowner(oo, clp, strhashval);
3707                 ret = oo;
3708         } else
3709                 nfs4_free_stateowner(&oo->oo_owner);
3710
3711         spin_unlock(&clp->cl_lock);
3712         return ret;
3713 }
3714
3715 static struct nfs4_ol_stateid *
3716 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3717 {
3718
3719         struct nfs4_openowner *oo = open->op_openowner;
3720         struct nfs4_ol_stateid *retstp = NULL;
3721         struct nfs4_ol_stateid *stp;
3722
3723         stp = open->op_stp;
3724         /* We are moving these outside of the spinlocks to avoid the warnings */
3725         mutex_init(&stp->st_mutex);
3726         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3727
3728 retry:
3729         spin_lock(&oo->oo_owner.so_client->cl_lock);
3730         spin_lock(&fp->fi_lock);
3731
3732         retstp = nfsd4_find_existing_open(fp, open);
3733         if (retstp)
3734                 goto out_unlock;
3735
3736         open->op_stp = NULL;
3737         refcount_inc(&stp->st_stid.sc_count);
3738         stp->st_stid.sc_type = NFS4_OPEN_STID;
3739         INIT_LIST_HEAD(&stp->st_locks);
3740         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3741         get_nfs4_file(fp);
3742         stp->st_stid.sc_file = fp;
3743         stp->st_access_bmap = 0;
3744         stp->st_deny_bmap = 0;
3745         stp->st_openstp = NULL;
3746         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3747         list_add(&stp->st_perfile, &fp->fi_stateids);
3748
3749 out_unlock:
3750         spin_unlock(&fp->fi_lock);
3751         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3752         if (retstp) {
3753                 /* Handle races with CLOSE */
3754                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3755                         nfs4_put_stid(&retstp->st_stid);
3756                         goto retry;
3757                 }
3758                 /* To keep mutex tracking happy */
3759                 mutex_unlock(&stp->st_mutex);
3760                 stp = retstp;
3761         }
3762         return stp;
3763 }
3764
3765 /*
3766  * In the 4.0 case we need to keep the owners around a little while to handle
3767  * CLOSE replay. We still do need to release any file access that is held by
3768  * them before returning however.
3769  */
3770 static void
3771 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3772 {
3773         struct nfs4_ol_stateid *last;
3774         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3775         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3776                                                 nfsd_net_id);
3777
3778         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3779
3780         /*
3781          * We know that we hold one reference via nfsd4_close, and another
3782          * "persistent" reference for the client. If the refcount is higher
3783          * than 2, then there are still calls in progress that are using this
3784          * stateid. We can't put the sc_file reference until they are finished.
3785          * Wait for the refcount to drop to 2. Since it has been unhashed,
3786          * there should be no danger of the refcount going back up again at
3787          * this point.
3788          */
3789         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3790
3791         release_all_access(s);
3792         if (s->st_stid.sc_file) {
3793                 put_nfs4_file(s->st_stid.sc_file);
3794                 s->st_stid.sc_file = NULL;
3795         }
3796
3797         spin_lock(&nn->client_lock);
3798         last = oo->oo_last_closed_stid;
3799         oo->oo_last_closed_stid = s;
3800         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3801         oo->oo_time = get_seconds();
3802         spin_unlock(&nn->client_lock);
3803         if (last)
3804                 nfs4_put_stid(&last->st_stid);
3805 }
3806
3807 /* search file_hashtbl[] for file */
3808 static struct nfs4_file *
3809 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3810 {
3811         struct nfs4_file *fp;
3812
3813         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3814                 if (fh_match(&fp->fi_fhandle, fh)) {
3815                         if (refcount_inc_not_zero(&fp->fi_ref))
3816                                 return fp;
3817                 }
3818         }
3819         return NULL;
3820 }
3821
3822 struct nfs4_file *
3823 find_file(struct knfsd_fh *fh)
3824 {
3825         struct nfs4_file *fp;
3826         unsigned int hashval = file_hashval(fh);
3827
3828         rcu_read_lock();
3829         fp = find_file_locked(fh, hashval);
3830         rcu_read_unlock();
3831         return fp;
3832 }
3833
3834 static struct nfs4_file *
3835 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3836 {
3837         struct nfs4_file *fp;
3838         unsigned int hashval = file_hashval(fh);
3839
3840         rcu_read_lock();
3841         fp = find_file_locked(fh, hashval);
3842         rcu_read_unlock();
3843         if (fp)
3844                 return fp;
3845
3846         spin_lock(&state_lock);
3847         fp = find_file_locked(fh, hashval);
3848         if (likely(fp == NULL)) {
3849                 nfsd4_init_file(fh, hashval, new);
3850                 fp = new;
3851         }
3852         spin_unlock(&state_lock);
3853
3854         return fp;
3855 }
3856
3857 /*
3858  * Called to check deny when READ with all zero stateid or
3859  * WRITE with all zero or all one stateid
3860  */
3861 static __be32
3862 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3863 {
3864         struct nfs4_file *fp;
3865         __be32 ret = nfs_ok;
3866
3867         fp = find_file(&current_fh->fh_handle);
3868         if (!fp)
3869                 return ret;
3870         /* Check for conflicting share reservations */
3871         spin_lock(&fp->fi_lock);
3872         if (fp->fi_share_deny & deny_type)
3873                 ret = nfserr_locked;
3874         spin_unlock(&fp->fi_lock);
3875         put_nfs4_file(fp);
3876         return ret;
3877 }
3878
3879 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3880 {
3881         struct nfs4_delegation *dp = cb_to_delegation(cb);
3882         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3883                                           nfsd_net_id);
3884
3885         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3886
3887         /*
3888          * We can't do this in nfsd_break_deleg_cb because it is
3889          * already holding inode->i_lock.
3890          *
3891          * If the dl_time != 0, then we know that it has already been
3892          * queued for a lease break. Don't queue it again.
3893          */
3894         spin_lock(&state_lock);
3895         if (dp->dl_time == 0) {
3896                 dp->dl_time = get_seconds();
3897                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3898         }
3899         spin_unlock(&state_lock);
3900 }
3901
3902 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3903                 struct rpc_task *task)
3904 {
3905         struct nfs4_delegation *dp = cb_to_delegation(cb);
3906
3907         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3908                 return 1;
3909
3910         switch (task->tk_status) {
3911         case 0:
3912                 return 1;
3913         case -EBADHANDLE:
3914         case -NFS4ERR_BAD_STATEID:
3915                 /*
3916                  * Race: client probably got cb_recall before open reply
3917                  * granting delegation.
3918                  */
3919                 if (dp->dl_retries--) {
3920                         rpc_delay(task, 2 * HZ);
3921                         return 0;
3922                 }
3923                 /*FALLTHRU*/
3924         default:
3925                 return -1;
3926         }
3927 }
3928
3929 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3930 {
3931         struct nfs4_delegation *dp = cb_to_delegation(cb);
3932
3933         nfs4_put_stid(&dp->dl_stid);
3934 }
3935
3936 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3937         .prepare        = nfsd4_cb_recall_prepare,
3938         .done           = nfsd4_cb_recall_done,
3939         .release        = nfsd4_cb_recall_release,
3940 };
3941
3942 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3943 {
3944         /*
3945          * We're assuming the state code never drops its reference
3946          * without first removing the lease.  Since we're in this lease
3947          * callback (and since the lease code is serialized by the kernel
3948          * lock) we know the server hasn't removed the lease yet, we know
3949          * it's safe to take a reference.
3950          */
3951         refcount_inc(&dp->dl_stid.sc_count);
3952         nfsd4_run_cb(&dp->dl_recall);
3953 }
3954
3955 /* Called from break_lease() with i_lock held. */
3956 static bool
3957 nfsd_break_deleg_cb(struct file_lock *fl)
3958 {
3959         bool ret = false;
3960         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
3961         struct nfs4_file *fp = dp->dl_stid.sc_file;
3962
3963         /*
3964          * We don't want the locks code to timeout the lease for us;
3965          * we'll remove it ourself if a delegation isn't returned
3966          * in time:
3967          */
3968         fl->fl_break_time = 0;
3969
3970         spin_lock(&fp->fi_lock);
3971         fp->fi_had_conflict = true;
3972         nfsd_break_one_deleg(dp);
3973         spin_unlock(&fp->fi_lock);
3974         return ret;
3975 }
3976
3977 static int
3978 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3979                      struct list_head *dispose)
3980 {
3981         if (arg & F_UNLCK)
3982                 return lease_modify(onlist, arg, dispose);
3983         else
3984                 return -EAGAIN;
3985 }
3986
3987 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3988         .lm_break = nfsd_break_deleg_cb,
3989         .lm_change = nfsd_change_deleg_cb,
3990 };
3991
3992 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3993 {
3994         if (nfsd4_has_session(cstate))
3995                 return nfs_ok;
3996         if (seqid == so->so_seqid - 1)
3997                 return nfserr_replay_me;
3998         if (seqid == so->so_seqid)
3999                 return nfs_ok;
4000         return nfserr_bad_seqid;
4001 }
4002
4003 static __be32 lookup_clientid(clientid_t *clid,
4004                 struct nfsd4_compound_state *cstate,
4005                 struct nfsd_net *nn)
4006 {
4007         struct nfs4_client *found;
4008
4009         if (cstate->clp) {
4010                 found = cstate->clp;
4011                 if (!same_clid(&found->cl_clientid, clid))
4012                         return nfserr_stale_clientid;
4013                 return nfs_ok;
4014         }
4015
4016         if (STALE_CLIENTID(clid, nn))
4017                 return nfserr_stale_clientid;
4018
4019         /*
4020          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4021          * cached already then we know this is for is for v4.0 and "sessions"
4022          * will be false.
4023          */
4024         WARN_ON_ONCE(cstate->session);
4025         spin_lock(&nn->client_lock);
4026         found = find_confirmed_client(clid, false, nn);
4027         if (!found) {
4028                 spin_unlock(&nn->client_lock);
4029                 return nfserr_expired;
4030         }
4031         atomic_inc(&found->cl_refcount);
4032         spin_unlock(&nn->client_lock);
4033
4034         /* Cache the nfs4_client in cstate! */
4035         cstate->clp = found;
4036         return nfs_ok;
4037 }
4038
4039 __be32
4040 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4041                     struct nfsd4_open *open, struct nfsd_net *nn)
4042 {
4043         clientid_t *clientid = &open->op_clientid;
4044         struct nfs4_client *clp = NULL;
4045         unsigned int strhashval;
4046         struct nfs4_openowner *oo = NULL;
4047         __be32 status;
4048
4049         if (STALE_CLIENTID(&open->op_clientid, nn))
4050                 return nfserr_stale_clientid;
4051         /*
4052          * In case we need it later, after we've already created the
4053          * file and don't want to risk a further failure:
4054          */
4055         open->op_file = nfsd4_alloc_file();
4056         if (open->op_file == NULL)
4057                 return nfserr_jukebox;
4058
4059         status = lookup_clientid(clientid, cstate, nn);
4060         if (status)
4061                 return status;
4062         clp = cstate->clp;
4063
4064         strhashval = ownerstr_hashval(&open->op_owner);
4065         oo = find_openstateowner_str(strhashval, open, clp);
4066         open->op_openowner = oo;
4067         if (!oo) {
4068                 goto new_owner;
4069         }
4070         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4071                 /* Replace unconfirmed owners without checking for replay. */
4072                 release_openowner(oo);
4073                 open->op_openowner = NULL;
4074                 goto new_owner;
4075         }
4076         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4077         if (status)
4078                 return status;
4079         goto alloc_stateid;
4080 new_owner:
4081         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4082         if (oo == NULL)
4083                 return nfserr_jukebox;
4084         open->op_openowner = oo;
4085 alloc_stateid:
4086         open->op_stp = nfs4_alloc_open_stateid(clp);
4087         if (!open->op_stp)
4088                 return nfserr_jukebox;
4089
4090         if (nfsd4_has_session(cstate) &&
4091             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4092                 open->op_odstate = alloc_clnt_odstate(clp);
4093                 if (!open->op_odstate)
4094                         return nfserr_jukebox;
4095         }
4096
4097         return nfs_ok;
4098 }
4099
4100 static inline __be32
4101 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4102 {
4103         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4104                 return nfserr_openmode;
4105         else
4106                 return nfs_ok;
4107 }
4108
4109 static int share_access_to_flags(u32 share_access)
4110 {
4111         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4112 }
4113
4114 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4115 {
4116         struct nfs4_stid *ret;
4117
4118         ret = find_stateid_by_type(cl, s,
4119                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4120         if (!ret)
4121                 return NULL;
4122         return delegstateid(ret);
4123 }
4124
4125 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4126 {
4127         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4128                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4129 }
4130
4131 static __be32
4132 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4133                 struct nfs4_delegation **dp)
4134 {
4135         int flags;
4136         __be32 status = nfserr_bad_stateid;
4137         struct nfs4_delegation *deleg;
4138
4139         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4140         if (deleg == NULL)
4141                 goto out;
4142         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4143                 nfs4_put_stid(&deleg->dl_stid);
4144                 if (cl->cl_minorversion)
4145                         status = nfserr_deleg_revoked;
4146                 goto out;
4147         }
4148         flags = share_access_to_flags(open->op_share_access);
4149         status = nfs4_check_delegmode(deleg, flags);
4150         if (status) {
4151                 nfs4_put_stid(&deleg->dl_stid);
4152                 goto out;
4153         }
4154         *dp = deleg;
4155 out:
4156         if (!nfsd4_is_deleg_cur(open))
4157                 return nfs_ok;
4158         if (status)
4159                 return status;
4160         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4161         return nfs_ok;
4162 }
4163
4164 static inline int nfs4_access_to_access(u32 nfs4_access)
4165 {
4166         int flags = 0;
4167
4168         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4169                 flags |= NFSD_MAY_READ;
4170         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4171                 flags |= NFSD_MAY_WRITE;
4172         return flags;
4173 }
4174
4175 static inline __be32
4176 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4177                 struct nfsd4_open *open)
4178 {
4179         struct iattr iattr = {
4180                 .ia_valid = ATTR_SIZE,
4181                 .ia_size = 0,
4182         };
4183         if (!open->op_truncate)
4184                 return 0;
4185         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4186                 return nfserr_inval;
4187         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4188 }
4189
4190 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4191                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4192                 struct nfsd4_open *open)
4193 {
4194         struct file *filp = NULL;
4195         __be32 status;
4196         int oflag = nfs4_access_to_omode(open->op_share_access);
4197         int access = nfs4_access_to_access(open->op_share_access);
4198         unsigned char old_access_bmap, old_deny_bmap;
4199
4200         spin_lock(&fp->fi_lock);
4201
4202         /*
4203          * Are we trying to set a deny mode that would conflict with
4204          * current access?
4205          */
4206         status = nfs4_file_check_deny(fp, open->op_share_deny);
4207         if (status != nfs_ok) {
4208                 spin_unlock(&fp->fi_lock);
4209                 goto out;
4210         }
4211
4212         /* set access to the file */
4213         status = nfs4_file_get_access(fp, open->op_share_access);
4214         if (status != nfs_ok) {
4215                 spin_unlock(&fp->fi_lock);
4216                 goto out;
4217         }
4218
4219         /* Set access bits in stateid */
4220         old_access_bmap = stp->st_access_bmap;
4221         set_access(open->op_share_access, stp);
4222
4223         /* Set new deny mask */
4224         old_deny_bmap = stp->st_deny_bmap;
4225         set_deny(open->op_share_deny, stp);
4226         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4227
4228         if (!fp->fi_fds[oflag]) {
4229                 spin_unlock(&fp->fi_lock);
4230                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4231                 if (status)
4232                         goto out_put_access;
4233                 spin_lock(&fp->fi_lock);
4234                 if (!fp->fi_fds[oflag]) {
4235                         fp->fi_fds[oflag] = filp;
4236                         filp = NULL;
4237                 }
4238         }
4239         spin_unlock(&fp->fi_lock);
4240         if (filp)
4241                 fput(filp);
4242
4243         status = nfsd4_truncate(rqstp, cur_fh, open);
4244         if (status)
4245                 goto out_put_access;
4246 out:
4247         return status;
4248 out_put_access:
4249         stp->st_access_bmap = old_access_bmap;
4250         nfs4_file_put_access(fp, open->op_share_access);
4251         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4252         goto out;
4253 }
4254
4255 static __be32
4256 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)
4257 {
4258         __be32 status;
4259         unsigned char old_deny_bmap = stp->st_deny_bmap;
4260
4261         if (!test_access(open->op_share_access, stp))
4262                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4263
4264         /* test and set deny mode */
4265         spin_lock(&fp->fi_lock);
4266         status = nfs4_file_check_deny(fp, open->op_share_deny);
4267         if (status == nfs_ok) {
4268                 set_deny(open->op_share_deny, stp);
4269                 fp->fi_share_deny |=
4270                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4271         }
4272         spin_unlock(&fp->fi_lock);
4273
4274         if (status != nfs_ok)
4275                 return status;
4276
4277         status = nfsd4_truncate(rqstp, cur_fh, open);
4278         if (status != nfs_ok)
4279                 reset_union_bmap_deny(old_deny_bmap, stp);
4280         return status;
4281 }
4282
4283 /* Should we give out recallable state?: */
4284 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4285 {
4286         if (clp->cl_cb_state == NFSD4_CB_UP)
4287                 return true;
4288         /*
4289          * In the sessions case, since we don't have to establish a
4290          * separate connection for callbacks, we assume it's OK
4291          * until we hear otherwise:
4292          */
4293         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4294 }
4295
4296 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4297                                                 int flag)
4298 {
4299         struct file_lock *fl;
4300
4301         fl = locks_alloc_lock();
4302         if (!fl)
4303                 return NULL;
4304         fl->fl_lmops = &nfsd_lease_mng_ops;
4305         fl->fl_flags = FL_DELEG;
4306         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4307         fl->fl_end = OFFSET_MAX;
4308         fl->fl_owner = (fl_owner_t)dp;
4309         fl->fl_pid = current->tgid;
4310         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4311         return fl;
4312 }
4313
4314 static struct nfs4_delegation *
4315 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4316                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4317 {
4318         int status = 0;
4319         struct nfs4_delegation *dp;
4320         struct file *filp;
4321         struct file_lock *fl;
4322
4323         /*
4324          * The fi_had_conflict and nfs_get_existing_delegation checks
4325          * here are just optimizations; we'll need to recheck them at
4326          * the end:
4327          */
4328         if (fp->fi_had_conflict)
4329                 return ERR_PTR(-EAGAIN);
4330
4331         filp = find_readable_file(fp);
4332         if (!filp) {
4333                 /* We should always have a readable file here */
4334                 WARN_ON_ONCE(1);
4335                 return ERR_PTR(-EBADF);
4336         }
4337         spin_lock(&state_lock);
4338         spin_lock(&fp->fi_lock);
4339         if (nfs4_delegation_exists(clp, fp))
4340                 status = -EAGAIN;
4341         else if (!fp->fi_deleg_file) {
4342                 fp->fi_deleg_file = filp;
4343                 /* increment early to prevent fi_deleg_file from being
4344                  * cleared */
4345                 fp->fi_delegees = 1;
4346                 filp = NULL;
4347         } else
4348                 fp->fi_delegees++;
4349         spin_unlock(&fp->fi_lock);
4350         spin_unlock(&state_lock);
4351         if (filp)
4352                 fput(filp);
4353         if (status)
4354                 return ERR_PTR(status);
4355
4356         status = -ENOMEM;
4357         dp = alloc_init_deleg(clp, fp, fh, odstate);
4358         if (!dp)
4359                 goto out_delegees;
4360
4361         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4362         if (!fl)
4363                 goto out_stid;
4364
4365         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4366         if (fl)
4367                 locks_free_lock(fl);
4368         if (status)
4369                 goto out_clnt_odstate;
4370
4371         spin_lock(&state_lock);
4372         spin_lock(&fp->fi_lock);
4373         if (fp->fi_had_conflict)
4374                 status = -EAGAIN;
4375         else
4376                 status = hash_delegation_locked(dp, fp);
4377         spin_unlock(&fp->fi_lock);
4378         spin_unlock(&state_lock);
4379
4380         if (status)
4381                 destroy_unhashed_deleg(dp);
4382         return dp;
4383 out_clnt_odstate:
4384         put_clnt_odstate(dp->dl_clnt_odstate);
4385 out_stid:
4386         nfs4_put_stid(&dp->dl_stid);
4387 out_delegees:
4388         put_deleg_file(fp);
4389         return ERR_PTR(status);
4390 }
4391
4392 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4393 {
4394         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4395         if (status == -EAGAIN)
4396                 open->op_why_no_deleg = WND4_CONTENTION;
4397         else {
4398                 open->op_why_no_deleg = WND4_RESOURCE;
4399                 switch (open->op_deleg_want) {
4400                 case NFS4_SHARE_WANT_READ_DELEG:
4401                 case NFS4_SHARE_WANT_WRITE_DELEG:
4402                 case NFS4_SHARE_WANT_ANY_DELEG:
4403                         break;
4404                 case NFS4_SHARE_WANT_CANCEL:
4405                         open->op_why_no_deleg = WND4_CANCELLED;
4406                         break;
4407                 case NFS4_SHARE_WANT_NO_DELEG:
4408                         WARN_ON_ONCE(1);
4409                 }
4410         }
4411 }
4412
4413 /*
4414  * Attempt to hand out a delegation.
4415  *
4416  * Note we don't support write delegations, and won't until the vfs has
4417  * proper support for them.
4418  */
4419 static void
4420 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4421                         struct nfs4_ol_stateid *stp)
4422 {
4423         struct nfs4_delegation *dp;
4424         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4425         struct nfs4_client *clp = stp->st_stid.sc_client;
4426         int cb_up;
4427         int status = 0;
4428
4429         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4430         open->op_recall = 0;
4431         switch (open->op_claim_type) {
4432                 case NFS4_OPEN_CLAIM_PREVIOUS:
4433                         if (!cb_up)
4434                                 open->op_recall = 1;
4435                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4436                                 goto out_no_deleg;
4437                         break;
4438                 case NFS4_OPEN_CLAIM_NULL:
4439                 case NFS4_OPEN_CLAIM_FH:
4440                         /*
4441                          * Let's not give out any delegations till everyone's
4442                          * had the chance to reclaim theirs, *and* until
4443                          * NLM locks have all been reclaimed:
4444                          */
4445                         if (locks_in_grace(clp->net))
4446                                 goto out_no_deleg;
4447                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4448                                 goto out_no_deleg;
4449                         /*
4450                          * Also, if the file was opened for write or
4451                          * create, there's a good chance the client's
4452                          * about to write to it, resulting in an
4453                          * immediate recall (since we don't support
4454                          * write delegations):
4455                          */
4456                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4457                                 goto out_no_deleg;
4458                         if (open->op_create == NFS4_OPEN_CREATE)
4459                                 goto out_no_deleg;
4460                         break;
4461                 default:
4462                         goto out_no_deleg;
4463         }
4464         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4465         if (IS_ERR(dp))
4466                 goto out_no_deleg;
4467
4468         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4469
4470         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4471                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4472         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4473         nfs4_put_stid(&dp->dl_stid);
4474         return;
4475 out_no_deleg:
4476         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4477         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4478             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4479                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4480                 open->op_recall = 1;
4481         }
4482
4483         /* 4.1 client asking for a delegation? */
4484         if (open->op_deleg_want)
4485                 nfsd4_open_deleg_none_ext(open, status);
4486         return;
4487 }
4488
4489 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4490                                         struct nfs4_delegation *dp)
4491 {
4492         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4493             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4494                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4495                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4496         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_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_UPGRADE;
4500         }
4501         /* Otherwise the client must be confused wanting a delegation
4502          * it already has, therefore we don't return
4503          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4504          */
4505 }
4506
4507 __be32
4508 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4509 {
4510         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4511         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4512         struct nfs4_file *fp = NULL;
4513         struct nfs4_ol_stateid *stp = NULL;
4514         struct nfs4_delegation *dp = NULL;
4515         __be32 status;
4516         bool new_stp = false;
4517
4518         /*
4519          * Lookup file; if found, lookup stateid and check open request,
4520          * and check for delegations in the process of being recalled.
4521          * If not found, create the nfs4_file struct
4522          */
4523         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4524         if (fp != open->op_file) {
4525                 status = nfs4_check_deleg(cl, open, &dp);
4526                 if (status)
4527                         goto out;
4528                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4529         } else {
4530                 open->op_file = NULL;
4531                 status = nfserr_bad_stateid;
4532                 if (nfsd4_is_deleg_cur(open))
4533                         goto out;
4534         }
4535
4536         if (!stp) {
4537                 stp = init_open_stateid(fp, open);
4538                 if (!open->op_stp)
4539                         new_stp = true;
4540         }
4541
4542         /*
4543          * OPEN the file, or upgrade an existing OPEN.
4544          * If truncate fails, the OPEN fails.
4545          *
4546          * stp is already locked.
4547          */
4548         if (!new_stp) {
4549                 /* Stateid was found, this is an OPEN upgrade */
4550                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4551                 if (status) {
4552                         mutex_unlock(&stp->st_mutex);
4553                         goto out;
4554                 }
4555         } else {
4556                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4557                 if (status) {
4558                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4559                         release_open_stateid(stp);
4560                         mutex_unlock(&stp->st_mutex);
4561                         goto out;
4562                 }
4563
4564                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4565                                                         open->op_odstate);
4566                 if (stp->st_clnt_odstate == open->op_odstate)
4567                         open->op_odstate = NULL;
4568         }
4569
4570         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4571         mutex_unlock(&stp->st_mutex);
4572
4573         if (nfsd4_has_session(&resp->cstate)) {
4574                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4575                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4576                         open->op_why_no_deleg = WND4_NOT_WANTED;
4577                         goto nodeleg;
4578                 }
4579         }
4580
4581         /*
4582         * Attempt to hand out a delegation. No error return, because the
4583         * OPEN succeeds even if we fail.
4584         */
4585         nfs4_open_delegation(current_fh, open, stp);
4586 nodeleg:
4587         status = nfs_ok;
4588
4589         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4590                 STATEID_VAL(&stp->st_stid.sc_stateid));
4591 out:
4592         /* 4.1 client trying to upgrade/downgrade delegation? */
4593         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4594             open->op_deleg_want)
4595                 nfsd4_deleg_xgrade_none_ext(open, dp);
4596
4597         if (fp)
4598                 put_nfs4_file(fp);
4599         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4600                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4601         /*
4602         * To finish the open response, we just need to set the rflags.
4603         */
4604         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4605         if (nfsd4_has_session(&resp->cstate))
4606                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4607         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4608                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4609
4610         if (dp)
4611                 nfs4_put_stid(&dp->dl_stid);
4612         if (stp)
4613                 nfs4_put_stid(&stp->st_stid);
4614
4615         return status;
4616 }
4617
4618 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4619                               struct nfsd4_open *open)
4620 {
4621         if (open->op_openowner) {
4622                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4623
4624                 nfsd4_cstate_assign_replay(cstate, so);
4625                 nfs4_put_stateowner(so);
4626         }
4627         if (open->op_file)
4628                 kmem_cache_free(file_slab, open->op_file);
4629         if (open->op_stp)
4630                 nfs4_put_stid(&open->op_stp->st_stid);
4631         if (open->op_odstate)
4632                 kmem_cache_free(odstate_slab, open->op_odstate);
4633 }
4634
4635 __be32
4636 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4637             union nfsd4_op_u *u)
4638 {
4639         clientid_t *clid = &u->renew;
4640         struct nfs4_client *clp;
4641         __be32 status;
4642         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4643
4644         dprintk("process_renew(%08x/%08x): starting\n",
4645                         clid->cl_boot, clid->cl_id);
4646         status = lookup_clientid(clid, cstate, nn);
4647         if (status)
4648                 goto out;
4649         clp = cstate->clp;
4650         status = nfserr_cb_path_down;
4651         if (!list_empty(&clp->cl_delegations)
4652                         && clp->cl_cb_state != NFSD4_CB_UP)
4653                 goto out;
4654         status = nfs_ok;
4655 out:
4656         return status;
4657 }
4658
4659 void
4660 nfsd4_end_grace(struct nfsd_net *nn)
4661 {
4662         /* do nothing if grace period already ended */
4663         if (nn->grace_ended)
4664                 return;
4665
4666         dprintk("NFSD: end of grace period\n");
4667         nn->grace_ended = true;
4668         /*
4669          * If the server goes down again right now, an NFSv4
4670          * client will still be allowed to reclaim after it comes back up,
4671          * even if it hasn't yet had a chance to reclaim state this time.
4672          *
4673          */
4674         nfsd4_record_grace_done(nn);
4675         /*
4676          * At this point, NFSv4 clients can still reclaim.  But if the
4677          * server crashes, any that have not yet reclaimed will be out
4678          * of luck on the next boot.
4679          *
4680          * (NFSv4.1+ clients are considered to have reclaimed once they
4681          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4682          * have reclaimed after their first OPEN.)
4683          */
4684         locks_end_grace(&nn->nfsd4_manager);
4685         /*
4686          * At this point, and once lockd and/or any other containers
4687          * exit their grace period, further reclaims will fail and
4688          * regular locking can resume.
4689          */
4690 }
4691
4692 static time_t
4693 nfs4_laundromat(struct nfsd_net *nn)
4694 {
4695         struct nfs4_client *clp;
4696         struct nfs4_openowner *oo;
4697         struct nfs4_delegation *dp;
4698         struct nfs4_ol_stateid *stp;
4699         struct nfsd4_blocked_lock *nbl;
4700         struct list_head *pos, *next, reaplist;
4701         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4702         time_t t, new_timeo = nn->nfsd4_lease;
4703
4704         dprintk("NFSD: laundromat service - starting\n");
4705         nfsd4_end_grace(nn);
4706         INIT_LIST_HEAD(&reaplist);
4707         spin_lock(&nn->client_lock);
4708         list_for_each_safe(pos, next, &nn->client_lru) {
4709                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4710                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4711                         t = clp->cl_time - cutoff;
4712                         new_timeo = min(new_timeo, t);
4713                         break;
4714                 }
4715                 if (mark_client_expired_locked(clp)) {
4716                         dprintk("NFSD: client in use (clientid %08x)\n",
4717                                 clp->cl_clientid.cl_id);
4718                         continue;
4719                 }
4720                 list_add(&clp->cl_lru, &reaplist);
4721         }
4722         spin_unlock(&nn->client_lock);
4723         list_for_each_safe(pos, next, &reaplist) {
4724                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4725                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4726                         clp->cl_clientid.cl_id);
4727                 list_del_init(&clp->cl_lru);
4728                 expire_client(clp);
4729         }
4730         spin_lock(&state_lock);
4731         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4732                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4733                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4734                         t = dp->dl_time - cutoff;
4735                         new_timeo = min(new_timeo, t);
4736                         break;
4737                 }
4738                 WARN_ON(!unhash_delegation_locked(dp));
4739                 list_add(&dp->dl_recall_lru, &reaplist);
4740         }
4741         spin_unlock(&state_lock);
4742         while (!list_empty(&reaplist)) {
4743                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4744                                         dl_recall_lru);
4745                 list_del_init(&dp->dl_recall_lru);
4746                 revoke_delegation(dp);
4747         }
4748
4749         spin_lock(&nn->client_lock);
4750         while (!list_empty(&nn->close_lru)) {
4751                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4752                                         oo_close_lru);
4753                 if (time_after((unsigned long)oo->oo_time,
4754                                (unsigned long)cutoff)) {
4755                         t = oo->oo_time - cutoff;
4756                         new_timeo = min(new_timeo, t);
4757                         break;
4758                 }
4759                 list_del_init(&oo->oo_close_lru);
4760                 stp = oo->oo_last_closed_stid;
4761                 oo->oo_last_closed_stid = NULL;
4762                 spin_unlock(&nn->client_lock);
4763                 nfs4_put_stid(&stp->st_stid);
4764                 spin_lock(&nn->client_lock);
4765         }
4766         spin_unlock(&nn->client_lock);
4767
4768         /*
4769          * It's possible for a client to try and acquire an already held lock
4770          * that is being held for a long time, and then lose interest in it.
4771          * So, we clean out any un-revisited request after a lease period
4772          * under the assumption that the client is no longer interested.
4773          *
4774          * RFC5661, sec. 9.6 states that the client must not rely on getting
4775          * notifications and must continue to poll for locks, even when the
4776          * server supports them. Thus this shouldn't lead to clients blocking
4777          * indefinitely once the lock does become free.
4778          */
4779         BUG_ON(!list_empty(&reaplist));
4780         spin_lock(&nn->blocked_locks_lock);
4781         while (!list_empty(&nn->blocked_locks_lru)) {
4782                 nbl = list_first_entry(&nn->blocked_locks_lru,
4783                                         struct nfsd4_blocked_lock, nbl_lru);
4784                 if (time_after((unsigned long)nbl->nbl_time,
4785                                (unsigned long)cutoff)) {
4786                         t = nbl->nbl_time - cutoff;
4787                         new_timeo = min(new_timeo, t);
4788                         break;
4789                 }
4790                 list_move(&nbl->nbl_lru, &reaplist);
4791                 list_del_init(&nbl->nbl_list);
4792         }
4793         spin_unlock(&nn->blocked_locks_lock);
4794
4795         while (!list_empty(&reaplist)) {
4796                 nbl = list_first_entry(&reaplist,
4797                                         struct nfsd4_blocked_lock, nbl_lru);
4798                 list_del_init(&nbl->nbl_lru);
4799                 posix_unblock_lock(&nbl->nbl_lock);
4800                 free_blocked_lock(nbl);
4801         }
4802
4803         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4804         return new_timeo;
4805 }
4806
4807 static struct workqueue_struct *laundry_wq;
4808 static void laundromat_main(struct work_struct *);
4809
4810 static void
4811 laundromat_main(struct work_struct *laundry)
4812 {
4813         time_t t;
4814         struct delayed_work *dwork = to_delayed_work(laundry);
4815         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4816                                            laundromat_work);
4817
4818         t = nfs4_laundromat(nn);
4819         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4820         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4821 }
4822
4823 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4824 {
4825         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4826                 return nfserr_bad_stateid;
4827         return nfs_ok;
4828 }
4829
4830 static inline int
4831 access_permit_read(struct nfs4_ol_stateid *stp)
4832 {
4833         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4834                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4835                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4836 }
4837
4838 static inline int
4839 access_permit_write(struct nfs4_ol_stateid *stp)
4840 {
4841         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4842                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4843 }
4844
4845 static
4846 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4847 {
4848         __be32 status = nfserr_openmode;
4849
4850         /* For lock stateid's, we test the parent open, not the lock: */
4851         if (stp->st_openstp)
4852                 stp = stp->st_openstp;
4853         if ((flags & WR_STATE) && !access_permit_write(stp))
4854                 goto out;
4855         if ((flags & RD_STATE) && !access_permit_read(stp))
4856                 goto out;
4857         status = nfs_ok;
4858 out:
4859         return status;
4860 }
4861
4862 static inline __be32
4863 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4864 {
4865         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4866                 return nfs_ok;
4867         else if (opens_in_grace(net)) {
4868                 /* Answer in remaining cases depends on existence of
4869                  * conflicting state; so we must wait out the grace period. */
4870                 return nfserr_grace;
4871         } else if (flags & WR_STATE)
4872                 return nfs4_share_conflict(current_fh,
4873                                 NFS4_SHARE_DENY_WRITE);
4874         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4875                 return nfs4_share_conflict(current_fh,
4876                                 NFS4_SHARE_DENY_READ);
4877 }
4878
4879 /*
4880  * Allow READ/WRITE during grace period on recovered state only for files
4881  * that are not able to provide mandatory locking.
4882  */
4883 static inline int
4884 grace_disallows_io(struct net *net, struct inode *inode)
4885 {
4886         return opens_in_grace(net) && mandatory_lock(inode);
4887 }
4888
4889 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4890 {
4891         /*
4892          * When sessions are used the stateid generation number is ignored
4893          * when it is zero.
4894          */
4895         if (has_session && in->si_generation == 0)
4896                 return nfs_ok;
4897
4898         if (in->si_generation == ref->si_generation)
4899                 return nfs_ok;
4900
4901         /* If the client sends us a stateid from the future, it's buggy: */
4902         if (nfsd4_stateid_generation_after(in, ref))
4903                 return nfserr_bad_stateid;
4904         /*
4905          * However, we could see a stateid from the past, even from a
4906          * non-buggy client.  For example, if the client sends a lock
4907          * while some IO is outstanding, the lock may bump si_generation
4908          * while the IO is still in flight.  The client could avoid that
4909          * situation by waiting for responses on all the IO requests,
4910          * but better performance may result in retrying IO that
4911          * receives an old_stateid error if requests are rarely
4912          * reordered in flight:
4913          */
4914         return nfserr_old_stateid;
4915 }
4916
4917 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4918 {
4919         __be32 ret;
4920
4921         spin_lock(&s->sc_lock);
4922         ret = nfsd4_verify_open_stid(s);
4923         if (ret == nfs_ok)
4924                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4925         spin_unlock(&s->sc_lock);
4926         return ret;
4927 }
4928
4929 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4930 {
4931         if (ols->st_stateowner->so_is_open_owner &&
4932             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4933                 return nfserr_bad_stateid;
4934         return nfs_ok;
4935 }
4936
4937 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4938 {
4939         struct nfs4_stid *s;
4940         __be32 status = nfserr_bad_stateid;
4941
4942         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4943                 CLOSE_STATEID(stateid))
4944                 return status;
4945         /* Client debugging aid. */
4946         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4947                 char addr_str[INET6_ADDRSTRLEN];
4948                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4949                                  sizeof(addr_str));
4950                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4951                                         "with incorrect client ID\n", addr_str);
4952                 return status;
4953         }
4954         spin_lock(&cl->cl_lock);
4955         s = find_stateid_locked(cl, stateid);
4956         if (!s)
4957                 goto out_unlock;
4958         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
4959         if (status)
4960                 goto out_unlock;
4961         switch (s->sc_type) {
4962         case NFS4_DELEG_STID:
4963                 status = nfs_ok;
4964                 break;
4965         case NFS4_REVOKED_DELEG_STID:
4966                 status = nfserr_deleg_revoked;
4967                 break;
4968         case NFS4_OPEN_STID:
4969         case NFS4_LOCK_STID:
4970                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4971                 break;
4972         default:
4973                 printk("unknown stateid type %x\n", s->sc_type);
4974                 /* Fallthrough */
4975         case NFS4_CLOSED_STID:
4976         case NFS4_CLOSED_DELEG_STID:
4977                 status = nfserr_bad_stateid;
4978         }
4979 out_unlock:
4980         spin_unlock(&cl->cl_lock);
4981         return status;
4982 }
4983
4984 __be32
4985 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4986                      stateid_t *stateid, unsigned char typemask,
4987                      struct nfs4_stid **s, struct nfsd_net *nn)
4988 {
4989         __be32 status;
4990         bool return_revoked = false;
4991
4992         /*
4993          *  only return revoked delegations if explicitly asked.
4994          *  otherwise we report revoked or bad_stateid status.
4995          */
4996         if (typemask & NFS4_REVOKED_DELEG_STID)
4997                 return_revoked = true;
4998         else if (typemask & NFS4_DELEG_STID)
4999                 typemask |= NFS4_REVOKED_DELEG_STID;
5000
5001         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5002                 CLOSE_STATEID(stateid))
5003                 return nfserr_bad_stateid;
5004         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5005         if (status == nfserr_stale_clientid) {
5006                 if (cstate->session)
5007                         return nfserr_bad_stateid;
5008                 return nfserr_stale_stateid;
5009         }
5010         if (status)
5011                 return status;
5012         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5013         if (!*s)
5014                 return nfserr_bad_stateid;
5015         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5016                 nfs4_put_stid(*s);
5017                 if (cstate->minorversion)
5018                         return nfserr_deleg_revoked;
5019                 return nfserr_bad_stateid;
5020         }
5021         return nfs_ok;
5022 }
5023
5024 static struct file *
5025 nfs4_find_file(struct nfs4_stid *s, int flags)
5026 {
5027         if (!s)
5028                 return NULL;
5029
5030         switch (s->sc_type) {
5031         case NFS4_DELEG_STID:
5032                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5033                         return NULL;
5034                 return get_file(s->sc_file->fi_deleg_file);
5035         case NFS4_OPEN_STID:
5036         case NFS4_LOCK_STID:
5037                 if (flags & RD_STATE)
5038                         return find_readable_file(s->sc_file);
5039                 else
5040                         return find_writeable_file(s->sc_file);
5041                 break;
5042         }
5043
5044         return NULL;
5045 }
5046
5047 static __be32
5048 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5049 {
5050         __be32 status;
5051
5052         status = nfsd4_check_openowner_confirmed(ols);
5053         if (status)
5054                 return status;
5055         return nfs4_check_openmode(ols, flags);
5056 }
5057
5058 static __be32
5059 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5060                 struct file **filpp, bool *tmp_file, int flags)
5061 {
5062         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5063         struct file *file;
5064         __be32 status;
5065
5066         file = nfs4_find_file(s, flags);
5067         if (file) {
5068                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5069                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5070                 if (status) {
5071                         fput(file);
5072                         return status;
5073                 }
5074
5075                 *filpp = file;
5076         } else {
5077                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5078                 if (status)
5079                         return status;
5080
5081                 if (tmp_file)
5082                         *tmp_file = true;
5083         }
5084
5085         return 0;
5086 }
5087
5088 /*
5089  * Checks for stateid operations
5090  */
5091 __be32
5092 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5093                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5094                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5095 {
5096         struct inode *ino = d_inode(fhp->fh_dentry);
5097         struct net *net = SVC_NET(rqstp);
5098         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5099         struct nfs4_stid *s = NULL;
5100         __be32 status;
5101
5102         if (filpp)
5103                 *filpp = NULL;
5104         if (tmp_file)
5105                 *tmp_file = false;
5106
5107         if (grace_disallows_io(net, ino))
5108                 return nfserr_grace;
5109
5110         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5111                 status = check_special_stateids(net, fhp, stateid, flags);
5112                 goto done;
5113         }
5114
5115         status = nfsd4_lookup_stateid(cstate, stateid,
5116                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5117                                 &s, nn);
5118         if (status)
5119                 return status;
5120         status = nfsd4_stid_check_stateid_generation(stateid, s,
5121                         nfsd4_has_session(cstate));
5122         if (status)
5123                 goto out;
5124
5125         switch (s->sc_type) {
5126         case NFS4_DELEG_STID:
5127                 status = nfs4_check_delegmode(delegstateid(s), flags);
5128                 break;
5129         case NFS4_OPEN_STID:
5130         case NFS4_LOCK_STID:
5131                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5132                 break;
5133         default:
5134                 status = nfserr_bad_stateid;
5135                 break;
5136         }
5137         if (status)
5138                 goto out;
5139         status = nfs4_check_fh(fhp, s);
5140
5141 done:
5142         if (!status && filpp)
5143                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5144 out:
5145         if (s)
5146                 nfs4_put_stid(s);
5147         return status;
5148 }
5149
5150 /*
5151  * Test if the stateid is valid
5152  */
5153 __be32
5154 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5155                    union nfsd4_op_u *u)
5156 {
5157         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5158         struct nfsd4_test_stateid_id *stateid;
5159         struct nfs4_client *cl = cstate->session->se_client;
5160
5161         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5162                 stateid->ts_id_status =
5163                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5164
5165         return nfs_ok;
5166 }
5167
5168 static __be32
5169 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5170 {
5171         struct nfs4_ol_stateid *stp = openlockstateid(s);
5172         __be32 ret;
5173
5174         ret = nfsd4_lock_ol_stateid(stp);
5175         if (ret)
5176                 goto out_put_stid;
5177
5178         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5179         if (ret)
5180                 goto out;
5181
5182         ret = nfserr_locks_held;
5183         if (check_for_locks(stp->st_stid.sc_file,
5184                             lockowner(stp->st_stateowner)))
5185                 goto out;
5186
5187         release_lock_stateid(stp);
5188         ret = nfs_ok;
5189
5190 out:
5191         mutex_unlock(&stp->st_mutex);
5192 out_put_stid:
5193         nfs4_put_stid(s);
5194         return ret;
5195 }
5196
5197 __be32
5198 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5199                    union nfsd4_op_u *u)
5200 {
5201         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5202         stateid_t *stateid = &free_stateid->fr_stateid;
5203         struct nfs4_stid *s;
5204         struct nfs4_delegation *dp;
5205         struct nfs4_client *cl = cstate->session->se_client;
5206         __be32 ret = nfserr_bad_stateid;
5207
5208         spin_lock(&cl->cl_lock);
5209         s = find_stateid_locked(cl, stateid);
5210         if (!s)
5211                 goto out_unlock;
5212         spin_lock(&s->sc_lock);
5213         switch (s->sc_type) {
5214         case NFS4_DELEG_STID:
5215                 ret = nfserr_locks_held;
5216                 break;
5217         case NFS4_OPEN_STID:
5218                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5219                 if (ret)
5220                         break;
5221                 ret = nfserr_locks_held;
5222                 break;
5223         case NFS4_LOCK_STID:
5224                 spin_unlock(&s->sc_lock);
5225                 refcount_inc(&s->sc_count);
5226                 spin_unlock(&cl->cl_lock);
5227                 ret = nfsd4_free_lock_stateid(stateid, s);
5228                 goto out;
5229         case NFS4_REVOKED_DELEG_STID:
5230                 spin_unlock(&s->sc_lock);
5231                 dp = delegstateid(s);
5232                 list_del_init(&dp->dl_recall_lru);
5233                 spin_unlock(&cl->cl_lock);
5234                 nfs4_put_stid(s);
5235                 ret = nfs_ok;
5236                 goto out;
5237         /* Default falls through and returns nfserr_bad_stateid */
5238         }
5239         spin_unlock(&s->sc_lock);
5240 out_unlock:
5241         spin_unlock(&cl->cl_lock);
5242 out:
5243         return ret;
5244 }
5245
5246 static inline int
5247 setlkflg (int type)
5248 {
5249         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5250                 RD_STATE : WR_STATE;
5251 }
5252
5253 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5254 {
5255         struct svc_fh *current_fh = &cstate->current_fh;
5256         struct nfs4_stateowner *sop = stp->st_stateowner;
5257         __be32 status;
5258
5259         status = nfsd4_check_seqid(cstate, sop, seqid);
5260         if (status)
5261                 return status;
5262         status = nfsd4_lock_ol_stateid(stp);
5263         if (status != nfs_ok)
5264                 return status;
5265         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5266         if (status == nfs_ok)
5267                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5268         if (status != nfs_ok)
5269                 mutex_unlock(&stp->st_mutex);
5270         return status;
5271 }
5272
5273 /*
5274  * Checks for sequence id mutating operations.
5275  */
5276 static __be32
5277 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5278                          stateid_t *stateid, char typemask,
5279                          struct nfs4_ol_stateid **stpp,
5280                          struct nfsd_net *nn)
5281 {
5282         __be32 status;
5283         struct nfs4_stid *s;
5284         struct nfs4_ol_stateid *stp = NULL;
5285
5286         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5287                 seqid, STATEID_VAL(stateid));
5288
5289         *stpp = NULL;
5290         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5291         if (status)
5292                 return status;
5293         stp = openlockstateid(s);
5294         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5295
5296         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5297         if (!status)
5298                 *stpp = stp;
5299         else
5300                 nfs4_put_stid(&stp->st_stid);
5301         return status;
5302 }
5303
5304 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5305                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5306 {
5307         __be32 status;
5308         struct nfs4_openowner *oo;
5309         struct nfs4_ol_stateid *stp;
5310
5311         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5312                                                 NFS4_OPEN_STID, &stp, nn);
5313         if (status)
5314                 return status;
5315         oo = openowner(stp->st_stateowner);
5316         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5317                 mutex_unlock(&stp->st_mutex);
5318                 nfs4_put_stid(&stp->st_stid);
5319                 return nfserr_bad_stateid;
5320         }
5321         *stpp = stp;
5322         return nfs_ok;
5323 }
5324
5325 __be32
5326 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5327                    union nfsd4_op_u *u)
5328 {
5329         struct nfsd4_open_confirm *oc = &u->open_confirm;
5330         __be32 status;
5331         struct nfs4_openowner *oo;
5332         struct nfs4_ol_stateid *stp;
5333         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5334
5335         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5336                         cstate->current_fh.fh_dentry);
5337
5338         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5339         if (status)
5340                 return status;
5341
5342         status = nfs4_preprocess_seqid_op(cstate,
5343                                         oc->oc_seqid, &oc->oc_req_stateid,
5344                                         NFS4_OPEN_STID, &stp, nn);
5345         if (status)
5346                 goto out;
5347         oo = openowner(stp->st_stateowner);
5348         status = nfserr_bad_stateid;
5349         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5350                 mutex_unlock(&stp->st_mutex);
5351                 goto put_stateid;
5352         }
5353         oo->oo_flags |= NFS4_OO_CONFIRMED;
5354         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5355         mutex_unlock(&stp->st_mutex);
5356         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5357                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5358
5359         nfsd4_client_record_create(oo->oo_owner.so_client);
5360         status = nfs_ok;
5361 put_stateid:
5362         nfs4_put_stid(&stp->st_stid);
5363 out:
5364         nfsd4_bump_seqid(cstate, status);
5365         return status;
5366 }
5367
5368 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5369 {
5370         if (!test_access(access, stp))
5371                 return;
5372         nfs4_file_put_access(stp->st_stid.sc_file, access);
5373         clear_access(access, stp);
5374 }
5375
5376 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5377 {
5378         switch (to_access) {
5379         case NFS4_SHARE_ACCESS_READ:
5380                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5381                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5382                 break;
5383         case NFS4_SHARE_ACCESS_WRITE:
5384                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5385                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5386                 break;
5387         case NFS4_SHARE_ACCESS_BOTH:
5388                 break;
5389         default:
5390                 WARN_ON_ONCE(1);
5391         }
5392 }
5393
5394 __be32
5395 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5396                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5397 {
5398         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5399         __be32 status;
5400         struct nfs4_ol_stateid *stp;
5401         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5402
5403         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5404                         cstate->current_fh.fh_dentry);
5405
5406         /* We don't yet support WANT bits: */
5407         if (od->od_deleg_want)
5408                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5409                         od->od_deleg_want);
5410
5411         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5412                                         &od->od_stateid, &stp, nn);
5413         if (status)
5414                 goto out;
5415         status = nfserr_inval;
5416         if (!test_access(od->od_share_access, stp)) {
5417                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5418                         stp->st_access_bmap, od->od_share_access);
5419                 goto put_stateid;
5420         }
5421         if (!test_deny(od->od_share_deny, stp)) {
5422                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5423                         stp->st_deny_bmap, od->od_share_deny);
5424                 goto put_stateid;
5425         }
5426         nfs4_stateid_downgrade(stp, od->od_share_access);
5427         reset_union_bmap_deny(od->od_share_deny, stp);
5428         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5429         status = nfs_ok;
5430 put_stateid:
5431         mutex_unlock(&stp->st_mutex);
5432         nfs4_put_stid(&stp->st_stid);
5433 out:
5434         nfsd4_bump_seqid(cstate, status);
5435         return status;
5436 }
5437
5438 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5439 {
5440         struct nfs4_client *clp = s->st_stid.sc_client;
5441         bool unhashed;
5442         LIST_HEAD(reaplist);
5443
5444         spin_lock(&clp->cl_lock);
5445         unhashed = unhash_open_stateid(s, &reaplist);
5446
5447         if (clp->cl_minorversion) {
5448                 if (unhashed)
5449                         put_ol_stateid_locked(s, &reaplist);
5450                 spin_unlock(&clp->cl_lock);
5451                 free_ol_stateid_reaplist(&reaplist);
5452         } else {
5453                 spin_unlock(&clp->cl_lock);
5454                 free_ol_stateid_reaplist(&reaplist);
5455                 if (unhashed)
5456                         move_to_close_lru(s, clp->net);
5457         }
5458 }
5459
5460 /*
5461  * nfs4_unlock_state() called after encode
5462  */
5463 __be32
5464 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5465                 union nfsd4_op_u *u)
5466 {
5467         struct nfsd4_close *close = &u->close;
5468         __be32 status;
5469         struct nfs4_ol_stateid *stp;
5470         struct net *net = SVC_NET(rqstp);
5471         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5472
5473         dprintk("NFSD: nfsd4_close on file %pd\n",
5474                         cstate->current_fh.fh_dentry);
5475
5476         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5477                                         &close->cl_stateid,
5478                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5479                                         &stp, nn);
5480         nfsd4_bump_seqid(cstate, status);
5481         if (status)
5482                 goto out;
5483
5484         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5485
5486         /*
5487          * Technically we don't _really_ have to increment or copy it, since
5488          * it should just be gone after this operation and we clobber the
5489          * copied value below, but we continue to do so here just to ensure
5490          * that racing ops see that there was a state change.
5491          */
5492         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5493
5494         nfsd4_close_open_stateid(stp);
5495         mutex_unlock(&stp->st_mutex);
5496
5497         /* v4.1+ suggests that we send a special stateid in here, since the
5498          * clients should just ignore this anyway. Since this is not useful
5499          * for v4.0 clients either, we set it to the special close_stateid
5500          * universally.
5501          *
5502          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5503          */
5504         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5505
5506         /* put reference from nfs4_preprocess_seqid_op */
5507         nfs4_put_stid(&stp->st_stid);
5508 out:
5509         return status;
5510 }
5511
5512 __be32
5513 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5514                   union nfsd4_op_u *u)
5515 {
5516         struct nfsd4_delegreturn *dr = &u->delegreturn;
5517         struct nfs4_delegation *dp;
5518         stateid_t *stateid = &dr->dr_stateid;
5519         struct nfs4_stid *s;
5520         __be32 status;
5521         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5522
5523         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5524                 return status;
5525
5526         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5527         if (status)
5528                 goto out;
5529         dp = delegstateid(s);
5530         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5531         if (status)
5532                 goto put_stateid;
5533
5534         destroy_delegation(dp);
5535 put_stateid:
5536         nfs4_put_stid(&dp->dl_stid);
5537 out:
5538         return status;
5539 }
5540
5541 static inline u64
5542 end_offset(u64 start, u64 len)
5543 {
5544         u64 end;
5545
5546         end = start + len;
5547         return end >= start ? end: NFS4_MAX_UINT64;
5548 }
5549
5550 /* last octet in a range */
5551 static inline u64
5552 last_byte_offset(u64 start, u64 len)
5553 {
5554         u64 end;
5555
5556         WARN_ON_ONCE(!len);
5557         end = start + len;
5558         return end > start ? end - 1: NFS4_MAX_UINT64;
5559 }
5560
5561 /*
5562  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5563  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5564  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5565  * locking, this prevents us from being completely protocol-compliant.  The
5566  * real solution to this problem is to start using unsigned file offsets in
5567  * the VFS, but this is a very deep change!
5568  */
5569 static inline void
5570 nfs4_transform_lock_offset(struct file_lock *lock)
5571 {
5572         if (lock->fl_start < 0)
5573                 lock->fl_start = OFFSET_MAX;
5574         if (lock->fl_end < 0)
5575                 lock->fl_end = OFFSET_MAX;
5576 }
5577
5578 static fl_owner_t
5579 nfsd4_fl_get_owner(fl_owner_t owner)
5580 {
5581         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5582
5583         nfs4_get_stateowner(&lo->lo_owner);
5584         return owner;
5585 }
5586
5587 static void
5588 nfsd4_fl_put_owner(fl_owner_t owner)
5589 {
5590         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5591
5592         if (lo)
5593                 nfs4_put_stateowner(&lo->lo_owner);
5594 }
5595
5596 static void
5597 nfsd4_lm_notify(struct file_lock *fl)
5598 {
5599         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5600         struct net                      *net = lo->lo_owner.so_client->net;
5601         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5602         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5603                                                 struct nfsd4_blocked_lock, nbl_lock);
5604         bool queue = false;
5605
5606         /* An empty list means that something else is going to be using it */
5607         spin_lock(&nn->blocked_locks_lock);
5608         if (!list_empty(&nbl->nbl_list)) {
5609                 list_del_init(&nbl->nbl_list);
5610                 list_del_init(&nbl->nbl_lru);
5611                 queue = true;
5612         }
5613         spin_unlock(&nn->blocked_locks_lock);
5614
5615         if (queue)
5616                 nfsd4_run_cb(&nbl->nbl_cb);
5617 }
5618
5619 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5620         .lm_notify = nfsd4_lm_notify,
5621         .lm_get_owner = nfsd4_fl_get_owner,
5622         .lm_put_owner = nfsd4_fl_put_owner,
5623 };
5624
5625 static inline void
5626 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5627 {
5628         struct nfs4_lockowner *lo;
5629
5630         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5631                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5632                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5633                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5634                 if (!deny->ld_owner.data)
5635                         /* We just don't care that much */
5636                         goto nevermind;
5637                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5638                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5639         } else {
5640 nevermind:
5641                 deny->ld_owner.len = 0;
5642                 deny->ld_owner.data = NULL;
5643                 deny->ld_clientid.cl_boot = 0;
5644                 deny->ld_clientid.cl_id = 0;
5645         }
5646         deny->ld_start = fl->fl_start;
5647         deny->ld_length = NFS4_MAX_UINT64;
5648         if (fl->fl_end != NFS4_MAX_UINT64)
5649                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5650         deny->ld_type = NFS4_READ_LT;
5651         if (fl->fl_type != F_RDLCK)
5652                 deny->ld_type = NFS4_WRITE_LT;
5653 }
5654
5655 static struct nfs4_lockowner *
5656 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5657 {
5658         unsigned int strhashval = ownerstr_hashval(owner);
5659         struct nfs4_stateowner *so;
5660
5661         lockdep_assert_held(&clp->cl_lock);
5662
5663         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5664                             so_strhash) {
5665                 if (so->so_is_open_owner)
5666                         continue;
5667                 if (same_owner_str(so, owner))
5668                         return lockowner(nfs4_get_stateowner(so));
5669         }
5670         return NULL;
5671 }
5672
5673 static struct nfs4_lockowner *
5674 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5675 {
5676         struct nfs4_lockowner *lo;
5677
5678         spin_lock(&clp->cl_lock);
5679         lo = find_lockowner_str_locked(clp, owner);
5680         spin_unlock(&clp->cl_lock);
5681         return lo;
5682 }
5683
5684 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5685 {
5686         unhash_lockowner_locked(lockowner(sop));
5687 }
5688
5689 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5690 {
5691         struct nfs4_lockowner *lo = lockowner(sop);
5692
5693         kmem_cache_free(lockowner_slab, lo);
5694 }
5695
5696 static const struct nfs4_stateowner_operations lockowner_ops = {
5697         .so_unhash =    nfs4_unhash_lockowner,
5698         .so_free =      nfs4_free_lockowner,
5699 };
5700
5701 /*
5702  * Alloc a lock owner structure.
5703  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5704  * occurred.
5705  *
5706  * strhashval = ownerstr_hashval
5707  */
5708 static struct nfs4_lockowner *
5709 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5710                            struct nfs4_ol_stateid *open_stp,
5711                            struct nfsd4_lock *lock)
5712 {
5713         struct nfs4_lockowner *lo, *ret;
5714
5715         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5716         if (!lo)
5717                 return NULL;
5718         INIT_LIST_HEAD(&lo->lo_blocked);
5719         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5720         lo->lo_owner.so_is_open_owner = 0;
5721         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5722         lo->lo_owner.so_ops = &lockowner_ops;
5723         spin_lock(&clp->cl_lock);
5724         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5725         if (ret == NULL) {
5726                 list_add(&lo->lo_owner.so_strhash,
5727                          &clp->cl_ownerstr_hashtbl[strhashval]);
5728                 ret = lo;
5729         } else
5730                 nfs4_free_stateowner(&lo->lo_owner);
5731
5732         spin_unlock(&clp->cl_lock);
5733         return ret;
5734 }
5735
5736 static struct nfs4_ol_stateid *
5737 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5738 {
5739         struct nfs4_ol_stateid *lst;
5740         struct nfs4_client *clp = lo->lo_owner.so_client;
5741
5742         lockdep_assert_held(&clp->cl_lock);
5743
5744         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5745                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5746                         continue;
5747                 if (lst->st_stid.sc_file == fp) {
5748                         refcount_inc(&lst->st_stid.sc_count);
5749                         return lst;
5750                 }
5751         }
5752         return NULL;
5753 }
5754
5755 static struct nfs4_ol_stateid *
5756 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5757                   struct nfs4_file *fp, struct inode *inode,
5758                   struct nfs4_ol_stateid *open_stp)
5759 {
5760         struct nfs4_client *clp = lo->lo_owner.so_client;
5761         struct nfs4_ol_stateid *retstp;
5762
5763         mutex_init(&stp->st_mutex);
5764         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5765 retry:
5766         spin_lock(&clp->cl_lock);
5767         spin_lock(&fp->fi_lock);
5768         retstp = find_lock_stateid(lo, fp);
5769         if (retstp)
5770                 goto out_unlock;
5771
5772         refcount_inc(&stp->st_stid.sc_count);
5773         stp->st_stid.sc_type = NFS4_LOCK_STID;
5774         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5775         get_nfs4_file(fp);
5776         stp->st_stid.sc_file = fp;
5777         stp->st_access_bmap = 0;
5778         stp->st_deny_bmap = open_stp->st_deny_bmap;
5779         stp->st_openstp = open_stp;
5780         list_add(&stp->st_locks, &open_stp->st_locks);
5781         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5782         list_add(&stp->st_perfile, &fp->fi_stateids);
5783 out_unlock:
5784         spin_unlock(&fp->fi_lock);
5785         spin_unlock(&clp->cl_lock);
5786         if (retstp) {
5787                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5788                         nfs4_put_stid(&retstp->st_stid);
5789                         goto retry;
5790                 }
5791                 /* To keep mutex tracking happy */
5792                 mutex_unlock(&stp->st_mutex);
5793                 stp = retstp;
5794         }
5795         return stp;
5796 }
5797
5798 static struct nfs4_ol_stateid *
5799 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5800                             struct inode *inode, struct nfs4_ol_stateid *ost,
5801                             bool *new)
5802 {
5803         struct nfs4_stid *ns = NULL;
5804         struct nfs4_ol_stateid *lst;
5805         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5806         struct nfs4_client *clp = oo->oo_owner.so_client;
5807
5808         *new = false;
5809         spin_lock(&clp->cl_lock);
5810         lst = find_lock_stateid(lo, fi);
5811         spin_unlock(&clp->cl_lock);
5812         if (lst != NULL) {
5813                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5814                         goto out;
5815                 nfs4_put_stid(&lst->st_stid);
5816         }
5817         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5818         if (ns == NULL)
5819                 return NULL;
5820
5821         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5822         if (lst == openlockstateid(ns))
5823                 *new = true;
5824         else
5825                 nfs4_put_stid(ns);
5826 out:
5827         return lst;
5828 }
5829
5830 static int
5831 check_lock_length(u64 offset, u64 length)
5832 {
5833         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5834                 (length > ~offset)));
5835 }
5836
5837 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5838 {
5839         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5840
5841         lockdep_assert_held(&fp->fi_lock);
5842
5843         if (test_access(access, lock_stp))
5844                 return;
5845         __nfs4_file_get_access(fp, access);
5846         set_access(access, lock_stp);
5847 }
5848
5849 static __be32
5850 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5851                             struct nfs4_ol_stateid *ost,
5852                             struct nfsd4_lock *lock,
5853                             struct nfs4_ol_stateid **plst, bool *new)
5854 {
5855         __be32 status;
5856         struct nfs4_file *fi = ost->st_stid.sc_file;
5857         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5858         struct nfs4_client *cl = oo->oo_owner.so_client;
5859         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5860         struct nfs4_lockowner *lo;
5861         struct nfs4_ol_stateid *lst;
5862         unsigned int strhashval;
5863
5864         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5865         if (!lo) {
5866                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5867                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5868                 if (lo == NULL)
5869                         return nfserr_jukebox;
5870         } else {
5871                 /* with an existing lockowner, seqids must be the same */
5872                 status = nfserr_bad_seqid;
5873                 if (!cstate->minorversion &&
5874                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5875                         goto out;
5876         }
5877
5878         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5879         if (lst == NULL) {
5880                 status = nfserr_jukebox;
5881                 goto out;
5882         }
5883
5884         status = nfs_ok;
5885         *plst = lst;
5886 out:
5887         nfs4_put_stateowner(&lo->lo_owner);
5888         return status;
5889 }
5890
5891 /*
5892  *  LOCK operation
5893  */
5894 __be32
5895 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5896            union nfsd4_op_u *u)
5897 {
5898         struct nfsd4_lock *lock = &u->lock;
5899         struct nfs4_openowner *open_sop = NULL;
5900         struct nfs4_lockowner *lock_sop = NULL;
5901         struct nfs4_ol_stateid *lock_stp = NULL;
5902         struct nfs4_ol_stateid *open_stp = NULL;
5903         struct nfs4_file *fp;
5904         struct file *filp = NULL;
5905         struct nfsd4_blocked_lock *nbl = NULL;
5906         struct file_lock *file_lock = NULL;
5907         struct file_lock *conflock = NULL;
5908         __be32 status = 0;
5909         int lkflg;
5910         int err;
5911         bool new = false;
5912         unsigned char fl_type;
5913         unsigned int fl_flags = FL_POSIX;
5914         struct net *net = SVC_NET(rqstp);
5915         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5916
5917         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5918                 (long long) lock->lk_offset,
5919                 (long long) lock->lk_length);
5920
5921         if (check_lock_length(lock->lk_offset, lock->lk_length))
5922                  return nfserr_inval;
5923
5924         if ((status = fh_verify(rqstp, &cstate->current_fh,
5925                                 S_IFREG, NFSD_MAY_LOCK))) {
5926                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5927                 return status;
5928         }
5929
5930         if (lock->lk_is_new) {
5931                 if (nfsd4_has_session(cstate))
5932                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5933                         memcpy(&lock->lk_new_clientid,
5934                                 &cstate->session->se_client->cl_clientid,
5935                                 sizeof(clientid_t));
5936
5937                 status = nfserr_stale_clientid;
5938                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5939                         goto out;
5940
5941                 /* validate and update open stateid and open seqid */
5942                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5943                                         lock->lk_new_open_seqid,
5944                                         &lock->lk_new_open_stateid,
5945                                         &open_stp, nn);
5946                 if (status)
5947                         goto out;
5948                 mutex_unlock(&open_stp->st_mutex);
5949                 open_sop = openowner(open_stp->st_stateowner);
5950                 status = nfserr_bad_stateid;
5951                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5952                                                 &lock->lk_new_clientid))
5953                         goto out;
5954                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5955                                                         &lock_stp, &new);
5956         } else {
5957                 status = nfs4_preprocess_seqid_op(cstate,
5958                                        lock->lk_old_lock_seqid,
5959                                        &lock->lk_old_lock_stateid,
5960                                        NFS4_LOCK_STID, &lock_stp, nn);
5961         }
5962         if (status)
5963                 goto out;
5964         lock_sop = lockowner(lock_stp->st_stateowner);
5965
5966         lkflg = setlkflg(lock->lk_type);
5967         status = nfs4_check_openmode(lock_stp, lkflg);
5968         if (status)
5969                 goto out;
5970
5971         status = nfserr_grace;
5972         if (locks_in_grace(net) && !lock->lk_reclaim)
5973                 goto out;
5974         status = nfserr_no_grace;
5975         if (!locks_in_grace(net) && lock->lk_reclaim)
5976                 goto out;
5977
5978         fp = lock_stp->st_stid.sc_file;
5979         switch (lock->lk_type) {
5980                 case NFS4_READW_LT:
5981                         if (nfsd4_has_session(cstate))
5982                                 fl_flags |= FL_SLEEP;
5983                         /* Fallthrough */
5984                 case NFS4_READ_LT:
5985                         spin_lock(&fp->fi_lock);
5986                         filp = find_readable_file_locked(fp);
5987                         if (filp)
5988                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5989                         spin_unlock(&fp->fi_lock);
5990                         fl_type = F_RDLCK;
5991                         break;
5992                 case NFS4_WRITEW_LT:
5993                         if (nfsd4_has_session(cstate))
5994                                 fl_flags |= FL_SLEEP;
5995                         /* Fallthrough */
5996                 case NFS4_WRITE_LT:
5997                         spin_lock(&fp->fi_lock);
5998                         filp = find_writeable_file_locked(fp);
5999                         if (filp)
6000                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6001                         spin_unlock(&fp->fi_lock);
6002                         fl_type = F_WRLCK;
6003                         break;
6004                 default:
6005                         status = nfserr_inval;
6006                 goto out;
6007         }
6008
6009         if (!filp) {
6010                 status = nfserr_openmode;
6011                 goto out;
6012         }
6013
6014         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6015         if (!nbl) {
6016                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6017                 status = nfserr_jukebox;
6018                 goto out;
6019         }
6020
6021         file_lock = &nbl->nbl_lock;
6022         file_lock->fl_type = fl_type;
6023         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6024         file_lock->fl_pid = current->tgid;
6025         file_lock->fl_file = filp;
6026         file_lock->fl_flags = fl_flags;
6027         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6028         file_lock->fl_start = lock->lk_offset;
6029         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6030         nfs4_transform_lock_offset(file_lock);
6031
6032         conflock = locks_alloc_lock();
6033         if (!conflock) {
6034                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6035                 status = nfserr_jukebox;
6036                 goto out;
6037         }
6038
6039         if (fl_flags & FL_SLEEP) {
6040                 nbl->nbl_time = jiffies;
6041                 spin_lock(&nn->blocked_locks_lock);
6042                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6043                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6044                 spin_unlock(&nn->blocked_locks_lock);
6045         }
6046
6047         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6048         switch (err) {
6049         case 0: /* success! */
6050                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6051                 status = 0;
6052                 break;
6053         case FILE_LOCK_DEFERRED:
6054                 nbl = NULL;
6055                 /* Fallthrough */
6056         case -EAGAIN:           /* conflock holds conflicting lock */
6057                 status = nfserr_denied;
6058                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6059                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6060                 break;
6061         case -EDEADLK:
6062                 status = nfserr_deadlock;
6063                 break;
6064         default:
6065                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6066                 status = nfserrno(err);
6067                 break;
6068         }
6069 out:
6070         if (nbl) {
6071                 /* dequeue it if we queued it before */
6072                 if (fl_flags & FL_SLEEP) {
6073                         spin_lock(&nn->blocked_locks_lock);
6074                         list_del_init(&nbl->nbl_list);
6075                         list_del_init(&nbl->nbl_lru);
6076                         spin_unlock(&nn->blocked_locks_lock);
6077                 }
6078                 free_blocked_lock(nbl);
6079         }
6080         if (filp)
6081                 fput(filp);
6082         if (lock_stp) {
6083                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6084                 if (cstate->replay_owner &&
6085                     cstate->replay_owner != &lock_sop->lo_owner &&
6086                     seqid_mutating_err(ntohl(status)))
6087                         lock_sop->lo_owner.so_seqid++;
6088
6089                 /*
6090                  * If this is a new, never-before-used stateid, and we are
6091                  * returning an error, then just go ahead and release it.
6092                  */
6093                 if (status && new)
6094                         release_lock_stateid(lock_stp);
6095
6096                 mutex_unlock(&lock_stp->st_mutex);
6097
6098                 nfs4_put_stid(&lock_stp->st_stid);
6099         }
6100         if (open_stp)
6101                 nfs4_put_stid(&open_stp->st_stid);
6102         nfsd4_bump_seqid(cstate, status);
6103         if (conflock)
6104                 locks_free_lock(conflock);
6105         return status;
6106 }
6107
6108 /*
6109  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6110  * so we do a temporary open here just to get an open file to pass to
6111  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6112  * inode operation.)
6113  */
6114 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6115 {
6116         struct file *file;
6117         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6118         if (!err) {
6119                 err = nfserrno(vfs_test_lock(file, lock));
6120                 fput(file);
6121         }
6122         return err;
6123 }
6124
6125 /*
6126  * LOCKT operation
6127  */
6128 __be32
6129 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6130             union nfsd4_op_u *u)
6131 {
6132         struct nfsd4_lockt *lockt = &u->lockt;
6133         struct file_lock *file_lock = NULL;
6134         struct nfs4_lockowner *lo = NULL;
6135         __be32 status;
6136         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6137
6138         if (locks_in_grace(SVC_NET(rqstp)))
6139                 return nfserr_grace;
6140
6141         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6142                  return nfserr_inval;
6143
6144         if (!nfsd4_has_session(cstate)) {
6145                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6146                 if (status)
6147                         goto out;
6148         }
6149
6150         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6151                 goto out;
6152
6153         file_lock = locks_alloc_lock();
6154         if (!file_lock) {
6155                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6156                 status = nfserr_jukebox;
6157                 goto out;
6158         }
6159
6160         switch (lockt->lt_type) {
6161                 case NFS4_READ_LT:
6162                 case NFS4_READW_LT:
6163                         file_lock->fl_type = F_RDLCK;
6164                 break;
6165                 case NFS4_WRITE_LT:
6166                 case NFS4_WRITEW_LT:
6167                         file_lock->fl_type = F_WRLCK;
6168                 break;
6169                 default:
6170                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6171                         status = nfserr_inval;
6172                 goto out;
6173         }
6174
6175         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6176         if (lo)
6177                 file_lock->fl_owner = (fl_owner_t)lo;
6178         file_lock->fl_pid = current->tgid;
6179         file_lock->fl_flags = FL_POSIX;
6180
6181         file_lock->fl_start = lockt->lt_offset;
6182         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6183
6184         nfs4_transform_lock_offset(file_lock);
6185
6186         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6187         if (status)
6188                 goto out;
6189
6190         if (file_lock->fl_type != F_UNLCK) {
6191                 status = nfserr_denied;
6192                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6193         }
6194 out:
6195         if (lo)
6196                 nfs4_put_stateowner(&lo->lo_owner);
6197         if (file_lock)
6198                 locks_free_lock(file_lock);
6199         return status;
6200 }
6201
6202 __be32
6203 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6204             union nfsd4_op_u *u)
6205 {
6206         struct nfsd4_locku *locku = &u->locku;
6207         struct nfs4_ol_stateid *stp;
6208         struct file *filp = NULL;
6209         struct file_lock *file_lock = NULL;
6210         __be32 status;
6211         int err;
6212         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6213
6214         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6215                 (long long) locku->lu_offset,
6216                 (long long) locku->lu_length);
6217
6218         if (check_lock_length(locku->lu_offset, locku->lu_length))
6219                  return nfserr_inval;
6220
6221         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6222                                         &locku->lu_stateid, NFS4_LOCK_STID,
6223                                         &stp, nn);
6224         if (status)
6225                 goto out;
6226         filp = find_any_file(stp->st_stid.sc_file);
6227         if (!filp) {
6228                 status = nfserr_lock_range;
6229                 goto put_stateid;
6230         }
6231         file_lock = locks_alloc_lock();
6232         if (!file_lock) {
6233                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6234                 status = nfserr_jukebox;
6235                 goto fput;
6236         }
6237
6238         file_lock->fl_type = F_UNLCK;
6239         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6240         file_lock->fl_pid = current->tgid;
6241         file_lock->fl_file = filp;
6242         file_lock->fl_flags = FL_POSIX;
6243         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6244         file_lock->fl_start = locku->lu_offset;
6245
6246         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6247                                                 locku->lu_length);
6248         nfs4_transform_lock_offset(file_lock);
6249
6250         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6251         if (err) {
6252                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6253                 goto out_nfserr;
6254         }
6255         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6256 fput:
6257         fput(filp);
6258 put_stateid:
6259         mutex_unlock(&stp->st_mutex);
6260         nfs4_put_stid(&stp->st_stid);
6261 out:
6262         nfsd4_bump_seqid(cstate, status);
6263         if (file_lock)
6264                 locks_free_lock(file_lock);
6265         return status;
6266
6267 out_nfserr:
6268         status = nfserrno(err);
6269         goto fput;
6270 }
6271
6272 /*
6273  * returns
6274  *      true:  locks held by lockowner
6275  *      false: no locks held by lockowner
6276  */
6277 static bool
6278 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6279 {
6280         struct file_lock *fl;
6281         int status = false;
6282         struct file *filp = find_any_file(fp);
6283         struct inode *inode;
6284         struct file_lock_context *flctx;
6285
6286         if (!filp) {
6287                 /* Any valid lock stateid should have some sort of access */
6288                 WARN_ON_ONCE(1);
6289                 return status;
6290         }
6291
6292         inode = file_inode(filp);
6293         flctx = inode->i_flctx;
6294
6295         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6296                 spin_lock(&flctx->flc_lock);
6297                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6298                         if (fl->fl_owner == (fl_owner_t)lowner) {
6299                                 status = true;
6300                                 break;
6301                         }
6302                 }
6303                 spin_unlock(&flctx->flc_lock);
6304         }
6305         fput(filp);
6306         return status;
6307 }
6308
6309 __be32
6310 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6311                         struct nfsd4_compound_state *cstate,
6312                         union nfsd4_op_u *u)
6313 {
6314         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6315         clientid_t *clid = &rlockowner->rl_clientid;
6316         struct nfs4_stateowner *sop;
6317         struct nfs4_lockowner *lo = NULL;
6318         struct nfs4_ol_stateid *stp;
6319         struct xdr_netobj *owner = &rlockowner->rl_owner;
6320         unsigned int hashval = ownerstr_hashval(owner);
6321         __be32 status;
6322         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6323         struct nfs4_client *clp;
6324         LIST_HEAD (reaplist);
6325
6326         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6327                 clid->cl_boot, clid->cl_id);
6328
6329         status = lookup_clientid(clid, cstate, nn);
6330         if (status)
6331                 return status;
6332
6333         clp = cstate->clp;
6334         /* Find the matching lock stateowner */
6335         spin_lock(&clp->cl_lock);
6336         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6337                             so_strhash) {
6338
6339                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6340                         continue;
6341
6342                 /* see if there are still any locks associated with it */
6343                 lo = lockowner(sop);
6344                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6345                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6346                                 status = nfserr_locks_held;
6347                                 spin_unlock(&clp->cl_lock);
6348                                 return status;
6349                         }
6350                 }
6351
6352                 nfs4_get_stateowner(sop);
6353                 break;
6354         }
6355         if (!lo) {
6356                 spin_unlock(&clp->cl_lock);
6357                 return status;
6358         }
6359
6360         unhash_lockowner_locked(lo);
6361         while (!list_empty(&lo->lo_owner.so_stateids)) {
6362                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6363                                        struct nfs4_ol_stateid,
6364                                        st_perstateowner);
6365                 WARN_ON(!unhash_lock_stateid(stp));
6366                 put_ol_stateid_locked(stp, &reaplist);
6367         }
6368         spin_unlock(&clp->cl_lock);
6369         free_ol_stateid_reaplist(&reaplist);
6370         remove_blocked_locks(lo);
6371         nfs4_put_stateowner(&lo->lo_owner);
6372
6373         return status;
6374 }
6375
6376 static inline struct nfs4_client_reclaim *
6377 alloc_reclaim(void)
6378 {
6379         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6380 }
6381
6382 bool
6383 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6384 {
6385         struct nfs4_client_reclaim *crp;
6386
6387         crp = nfsd4_find_reclaim_client(name, nn);
6388         return (crp && crp->cr_clp);
6389 }
6390
6391 /*
6392  * failure => all reset bets are off, nfserr_no_grace...
6393  */
6394 struct nfs4_client_reclaim *
6395 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6396 {
6397         unsigned int strhashval;
6398         struct nfs4_client_reclaim *crp;
6399
6400         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6401         crp = alloc_reclaim();
6402         if (crp) {
6403                 strhashval = clientstr_hashval(name);
6404                 INIT_LIST_HEAD(&crp->cr_strhash);
6405                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6406                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6407                 crp->cr_clp = NULL;
6408                 nn->reclaim_str_hashtbl_size++;
6409         }
6410         return crp;
6411 }
6412
6413 void
6414 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6415 {
6416         list_del(&crp->cr_strhash);
6417         kfree(crp);
6418         nn->reclaim_str_hashtbl_size--;
6419 }
6420
6421 void
6422 nfs4_release_reclaim(struct nfsd_net *nn)
6423 {
6424         struct nfs4_client_reclaim *crp = NULL;
6425         int i;
6426
6427         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6428                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6429                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6430                                         struct nfs4_client_reclaim, cr_strhash);
6431                         nfs4_remove_reclaim_record(crp, nn);
6432                 }
6433         }
6434         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6435 }
6436
6437 /*
6438  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6439 struct nfs4_client_reclaim *
6440 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6441 {
6442         unsigned int strhashval;
6443         struct nfs4_client_reclaim *crp = NULL;
6444
6445         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6446
6447         strhashval = clientstr_hashval(recdir);
6448         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6449                 if (same_name(crp->cr_recdir, recdir)) {
6450                         return crp;
6451                 }
6452         }
6453         return NULL;
6454 }
6455
6456 /*
6457 * Called from OPEN. Look for clientid in reclaim list.
6458 */
6459 __be32
6460 nfs4_check_open_reclaim(clientid_t *clid,
6461                 struct nfsd4_compound_state *cstate,
6462                 struct nfsd_net *nn)
6463 {
6464         __be32 status;
6465
6466         /* find clientid in conf_id_hashtbl */
6467         status = lookup_clientid(clid, cstate, nn);
6468         if (status)
6469                 return nfserr_reclaim_bad;
6470
6471         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6472                 return nfserr_no_grace;
6473
6474         if (nfsd4_client_record_check(cstate->clp))
6475                 return nfserr_reclaim_bad;
6476
6477         return nfs_ok;
6478 }
6479
6480 #ifdef CONFIG_NFSD_FAULT_INJECTION
6481 static inline void
6482 put_client(struct nfs4_client *clp)
6483 {
6484         atomic_dec(&clp->cl_refcount);
6485 }
6486
6487 static struct nfs4_client *
6488 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6489 {
6490         struct nfs4_client *clp;
6491         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6492                                           nfsd_net_id);
6493
6494         if (!nfsd_netns_ready(nn))
6495                 return NULL;
6496
6497         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6498                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6499                         return clp;
6500         }
6501         return NULL;
6502 }
6503
6504 u64
6505 nfsd_inject_print_clients(void)
6506 {
6507         struct nfs4_client *clp;
6508         u64 count = 0;
6509         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6510                                           nfsd_net_id);
6511         char buf[INET6_ADDRSTRLEN];
6512
6513         if (!nfsd_netns_ready(nn))
6514                 return 0;
6515
6516         spin_lock(&nn->client_lock);
6517         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6518                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6519                 pr_info("NFS Client: %s\n", buf);
6520                 ++count;
6521         }
6522         spin_unlock(&nn->client_lock);
6523
6524         return count;
6525 }
6526
6527 u64
6528 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6529 {
6530         u64 count = 0;
6531         struct nfs4_client *clp;
6532         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6533                                           nfsd_net_id);
6534
6535         if (!nfsd_netns_ready(nn))
6536                 return count;
6537
6538         spin_lock(&nn->client_lock);
6539         clp = nfsd_find_client(addr, addr_size);
6540         if (clp) {
6541                 if (mark_client_expired_locked(clp) == nfs_ok)
6542                         ++count;
6543                 else
6544                         clp = NULL;
6545         }
6546         spin_unlock(&nn->client_lock);
6547
6548         if (clp)
6549                 expire_client(clp);
6550
6551         return count;
6552 }
6553
6554 u64
6555 nfsd_inject_forget_clients(u64 max)
6556 {
6557         u64 count = 0;
6558         struct nfs4_client *clp, *next;
6559         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6560                                                 nfsd_net_id);
6561         LIST_HEAD(reaplist);
6562
6563         if (!nfsd_netns_ready(nn))
6564                 return count;
6565
6566         spin_lock(&nn->client_lock);
6567         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6568                 if (mark_client_expired_locked(clp) == nfs_ok) {
6569                         list_add(&clp->cl_lru, &reaplist);
6570                         if (max != 0 && ++count >= max)
6571                                 break;
6572                 }
6573         }
6574         spin_unlock(&nn->client_lock);
6575
6576         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6577                 expire_client(clp);
6578
6579         return count;
6580 }
6581
6582 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6583                              const char *type)
6584 {
6585         char buf[INET6_ADDRSTRLEN];
6586         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6587         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6588 }
6589
6590 static void
6591 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6592                              struct list_head *collect)
6593 {
6594         struct nfs4_client *clp = lst->st_stid.sc_client;
6595         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6596                                           nfsd_net_id);
6597
6598         if (!collect)
6599                 return;
6600
6601         lockdep_assert_held(&nn->client_lock);
6602         atomic_inc(&clp->cl_refcount);
6603         list_add(&lst->st_locks, collect);
6604 }
6605
6606 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6607                                     struct list_head *collect,
6608                                     bool (*func)(struct nfs4_ol_stateid *))
6609 {
6610         struct nfs4_openowner *oop;
6611         struct nfs4_ol_stateid *stp, *st_next;
6612         struct nfs4_ol_stateid *lst, *lst_next;
6613         u64 count = 0;
6614
6615         spin_lock(&clp->cl_lock);
6616         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6617                 list_for_each_entry_safe(stp, st_next,
6618                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6619                         list_for_each_entry_safe(lst, lst_next,
6620                                         &stp->st_locks, st_locks) {
6621                                 if (func) {
6622                                         if (func(lst))
6623                                                 nfsd_inject_add_lock_to_list(lst,
6624                                                                         collect);
6625                                 }
6626                                 ++count;
6627                                 /*
6628                                  * Despite the fact that these functions deal
6629                                  * with 64-bit integers for "count", we must
6630                                  * ensure that it doesn't blow up the
6631                                  * clp->cl_refcount. Throw a warning if we
6632                                  * start to approach INT_MAX here.
6633                                  */
6634                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6635                                 if (count == max)
6636                                         goto out;
6637                         }
6638                 }
6639         }
6640 out:
6641         spin_unlock(&clp->cl_lock);
6642
6643         return count;
6644 }
6645
6646 static u64
6647 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6648                           u64 max)
6649 {
6650         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6651 }
6652
6653 static u64
6654 nfsd_print_client_locks(struct nfs4_client *clp)
6655 {
6656         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6657         nfsd_print_count(clp, count, "locked files");
6658         return count;
6659 }
6660
6661 u64
6662 nfsd_inject_print_locks(void)
6663 {
6664         struct nfs4_client *clp;
6665         u64 count = 0;
6666         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6667                                                 nfsd_net_id);
6668
6669         if (!nfsd_netns_ready(nn))
6670                 return 0;
6671
6672         spin_lock(&nn->client_lock);
6673         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6674                 count += nfsd_print_client_locks(clp);
6675         spin_unlock(&nn->client_lock);
6676
6677         return count;
6678 }
6679
6680 static void
6681 nfsd_reap_locks(struct list_head *reaplist)
6682 {
6683         struct nfs4_client *clp;
6684         struct nfs4_ol_stateid *stp, *next;
6685
6686         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6687                 list_del_init(&stp->st_locks);
6688                 clp = stp->st_stid.sc_client;
6689                 nfs4_put_stid(&stp->st_stid);
6690                 put_client(clp);
6691         }
6692 }
6693
6694 u64
6695 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6696 {
6697         unsigned int count = 0;
6698         struct nfs4_client *clp;
6699         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6700                                                 nfsd_net_id);
6701         LIST_HEAD(reaplist);
6702
6703         if (!nfsd_netns_ready(nn))
6704                 return count;
6705
6706         spin_lock(&nn->client_lock);
6707         clp = nfsd_find_client(addr, addr_size);
6708         if (clp)
6709                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6710         spin_unlock(&nn->client_lock);
6711         nfsd_reap_locks(&reaplist);
6712         return count;
6713 }
6714
6715 u64
6716 nfsd_inject_forget_locks(u64 max)
6717 {
6718         u64 count = 0;
6719         struct nfs4_client *clp;
6720         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6721                                                 nfsd_net_id);
6722         LIST_HEAD(reaplist);
6723
6724         if (!nfsd_netns_ready(nn))
6725                 return count;
6726
6727         spin_lock(&nn->client_lock);
6728         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6729                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6730                 if (max != 0 && count >= max)
6731                         break;
6732         }
6733         spin_unlock(&nn->client_lock);
6734         nfsd_reap_locks(&reaplist);
6735         return count;
6736 }
6737
6738 static u64
6739 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6740                               struct list_head *collect,
6741                               void (*func)(struct nfs4_openowner *))
6742 {
6743         struct nfs4_openowner *oop, *next;
6744         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6745                                                 nfsd_net_id);
6746         u64 count = 0;
6747
6748         lockdep_assert_held(&nn->client_lock);
6749
6750         spin_lock(&clp->cl_lock);
6751         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6752                 if (func) {
6753                         func(oop);
6754                         if (collect) {
6755                                 atomic_inc(&clp->cl_refcount);
6756                                 list_add(&oop->oo_perclient, collect);
6757                         }
6758                 }
6759                 ++count;
6760                 /*
6761                  * Despite the fact that these functions deal with
6762                  * 64-bit integers for "count", we must ensure that
6763                  * it doesn't blow up the clp->cl_refcount. Throw a
6764                  * warning if we start to approach INT_MAX here.
6765                  */
6766                 WARN_ON_ONCE(count == (INT_MAX / 2));
6767                 if (count == max)
6768                         break;
6769         }
6770         spin_unlock(&clp->cl_lock);
6771
6772         return count;
6773 }
6774
6775 static u64
6776 nfsd_print_client_openowners(struct nfs4_client *clp)
6777 {
6778         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6779
6780         nfsd_print_count(clp, count, "openowners");
6781         return count;
6782 }
6783
6784 static u64
6785 nfsd_collect_client_openowners(struct nfs4_client *clp,
6786                                struct list_head *collect, u64 max)
6787 {
6788         return nfsd_foreach_client_openowner(clp, max, collect,
6789                                                 unhash_openowner_locked);
6790 }
6791
6792 u64
6793 nfsd_inject_print_openowners(void)
6794 {
6795         struct nfs4_client *clp;
6796         u64 count = 0;
6797         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6798                                                 nfsd_net_id);
6799
6800         if (!nfsd_netns_ready(nn))
6801                 return 0;
6802
6803         spin_lock(&nn->client_lock);
6804         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6805                 count += nfsd_print_client_openowners(clp);
6806         spin_unlock(&nn->client_lock);
6807
6808         return count;
6809 }
6810
6811 static void
6812 nfsd_reap_openowners(struct list_head *reaplist)
6813 {
6814         struct nfs4_client *clp;
6815         struct nfs4_openowner *oop, *next;
6816
6817         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6818                 list_del_init(&oop->oo_perclient);
6819                 clp = oop->oo_owner.so_client;
6820                 release_openowner(oop);
6821                 put_client(clp);
6822         }
6823 }
6824
6825 u64
6826 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6827                                      size_t addr_size)
6828 {
6829         unsigned int count = 0;
6830         struct nfs4_client *clp;
6831         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6832                                                 nfsd_net_id);
6833         LIST_HEAD(reaplist);
6834
6835         if (!nfsd_netns_ready(nn))
6836                 return count;
6837
6838         spin_lock(&nn->client_lock);
6839         clp = nfsd_find_client(addr, addr_size);
6840         if (clp)
6841                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6842         spin_unlock(&nn->client_lock);
6843         nfsd_reap_openowners(&reaplist);
6844         return count;
6845 }
6846
6847 u64
6848 nfsd_inject_forget_openowners(u64 max)
6849 {
6850         u64 count = 0;
6851         struct nfs4_client *clp;
6852         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6853                                                 nfsd_net_id);
6854         LIST_HEAD(reaplist);
6855
6856         if (!nfsd_netns_ready(nn))
6857                 return count;
6858
6859         spin_lock(&nn->client_lock);
6860         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6861                 count += nfsd_collect_client_openowners(clp, &reaplist,
6862                                                         max - count);
6863                 if (max != 0 && count >= max)
6864                         break;
6865         }
6866         spin_unlock(&nn->client_lock);
6867         nfsd_reap_openowners(&reaplist);
6868         return count;
6869 }
6870
6871 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6872                                      struct list_head *victims)
6873 {
6874         struct nfs4_delegation *dp, *next;
6875         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6876                                                 nfsd_net_id);
6877         u64 count = 0;
6878
6879         lockdep_assert_held(&nn->client_lock);
6880
6881         spin_lock(&state_lock);
6882         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6883                 if (victims) {
6884                         /*
6885                          * It's not safe to mess with delegations that have a
6886                          * non-zero dl_time. They might have already been broken
6887                          * and could be processed by the laundromat outside of
6888                          * the state_lock. Just leave them be.
6889                          */
6890                         if (dp->dl_time != 0)
6891                                 continue;
6892
6893                         atomic_inc(&clp->cl_refcount);
6894                         WARN_ON(!unhash_delegation_locked(dp));
6895                         list_add(&dp->dl_recall_lru, victims);
6896                 }
6897                 ++count;
6898                 /*
6899                  * Despite the fact that these functions deal with
6900                  * 64-bit integers for "count", we must ensure that
6901                  * it doesn't blow up the clp->cl_refcount. Throw a
6902                  * warning if we start to approach INT_MAX here.
6903                  */
6904                 WARN_ON_ONCE(count == (INT_MAX / 2));
6905                 if (count == max)
6906                         break;
6907         }
6908         spin_unlock(&state_lock);
6909         return count;
6910 }
6911
6912 static u64
6913 nfsd_print_client_delegations(struct nfs4_client *clp)
6914 {
6915         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6916
6917         nfsd_print_count(clp, count, "delegations");
6918         return count;
6919 }
6920
6921 u64
6922 nfsd_inject_print_delegations(void)
6923 {
6924         struct nfs4_client *clp;
6925         u64 count = 0;
6926         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6927                                                 nfsd_net_id);
6928
6929         if (!nfsd_netns_ready(nn))
6930                 return 0;
6931
6932         spin_lock(&nn->client_lock);
6933         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6934                 count += nfsd_print_client_delegations(clp);
6935         spin_unlock(&nn->client_lock);
6936
6937         return count;
6938 }
6939
6940 static void
6941 nfsd_forget_delegations(struct list_head *reaplist)
6942 {
6943         struct nfs4_client *clp;
6944         struct nfs4_delegation *dp, *next;
6945
6946         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6947                 list_del_init(&dp->dl_recall_lru);
6948                 clp = dp->dl_stid.sc_client;
6949                 revoke_delegation(dp);
6950                 put_client(clp);
6951         }
6952 }
6953
6954 u64
6955 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6956                                       size_t addr_size)
6957 {
6958         u64 count = 0;
6959         struct nfs4_client *clp;
6960         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6961                                                 nfsd_net_id);
6962         LIST_HEAD(reaplist);
6963
6964         if (!nfsd_netns_ready(nn))
6965                 return count;
6966
6967         spin_lock(&nn->client_lock);
6968         clp = nfsd_find_client(addr, addr_size);
6969         if (clp)
6970                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6971         spin_unlock(&nn->client_lock);
6972
6973         nfsd_forget_delegations(&reaplist);
6974         return count;
6975 }
6976
6977 u64
6978 nfsd_inject_forget_delegations(u64 max)
6979 {
6980         u64 count = 0;
6981         struct nfs4_client *clp;
6982         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6983                                                 nfsd_net_id);
6984         LIST_HEAD(reaplist);
6985
6986         if (!nfsd_netns_ready(nn))
6987                 return count;
6988
6989         spin_lock(&nn->client_lock);
6990         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6991                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6992                 if (max != 0 && count >= max)
6993                         break;
6994         }
6995         spin_unlock(&nn->client_lock);
6996         nfsd_forget_delegations(&reaplist);
6997         return count;
6998 }
6999
7000 static void
7001 nfsd_recall_delegations(struct list_head *reaplist)
7002 {
7003         struct nfs4_client *clp;
7004         struct nfs4_delegation *dp, *next;
7005
7006         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7007                 list_del_init(&dp->dl_recall_lru);
7008                 clp = dp->dl_stid.sc_client;
7009                 /*
7010                  * We skipped all entries that had a zero dl_time before,
7011                  * so we can now reset the dl_time back to 0. If a delegation
7012                  * break comes in now, then it won't make any difference since
7013                  * we're recalling it either way.
7014                  */
7015                 spin_lock(&state_lock);
7016                 dp->dl_time = 0;
7017                 spin_unlock(&state_lock);
7018                 nfsd_break_one_deleg(dp);
7019                 put_client(clp);
7020         }
7021 }
7022
7023 u64
7024 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7025                                       size_t addr_size)
7026 {
7027         u64 count = 0;
7028         struct nfs4_client *clp;
7029         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7030                                                 nfsd_net_id);
7031         LIST_HEAD(reaplist);
7032
7033         if (!nfsd_netns_ready(nn))
7034                 return count;
7035
7036         spin_lock(&nn->client_lock);
7037         clp = nfsd_find_client(addr, addr_size);
7038         if (clp)
7039                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7040         spin_unlock(&nn->client_lock);
7041
7042         nfsd_recall_delegations(&reaplist);
7043         return count;
7044 }
7045
7046 u64
7047 nfsd_inject_recall_delegations(u64 max)
7048 {
7049         u64 count = 0;
7050         struct nfs4_client *clp, *next;
7051         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7052                                                 nfsd_net_id);
7053         LIST_HEAD(reaplist);
7054
7055         if (!nfsd_netns_ready(nn))
7056                 return count;
7057
7058         spin_lock(&nn->client_lock);
7059         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7060                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7061                 if (max != 0 && ++count >= max)
7062                         break;
7063         }
7064         spin_unlock(&nn->client_lock);
7065         nfsd_recall_delegations(&reaplist);
7066         return count;
7067 }
7068 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7069
7070 /*
7071  * Since the lifetime of a delegation isn't limited to that of an open, a
7072  * client may quite reasonably hang on to a delegation as long as it has
7073  * the inode cached.  This becomes an obvious problem the first time a
7074  * client's inode cache approaches the size of the server's total memory.
7075  *
7076  * For now we avoid this problem by imposing a hard limit on the number
7077  * of delegations, which varies according to the server's memory size.
7078  */
7079 static void
7080 set_max_delegations(void)
7081 {
7082         /*
7083          * Allow at most 4 delegations per megabyte of RAM.  Quick
7084          * estimates suggest that in the worst case (where every delegation
7085          * is for a different inode), a delegation could take about 1.5K,
7086          * giving a worst case usage of about 6% of memory.
7087          */
7088         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7089 }
7090
7091 static int nfs4_state_create_net(struct net *net)
7092 {
7093         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7094         int i;
7095
7096         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7097                         CLIENT_HASH_SIZE, GFP_KERNEL);
7098         if (!nn->conf_id_hashtbl)
7099                 goto err;
7100         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7101                         CLIENT_HASH_SIZE, GFP_KERNEL);
7102         if (!nn->unconf_id_hashtbl)
7103                 goto err_unconf_id;
7104         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
7105                         SESSION_HASH_SIZE, GFP_KERNEL);
7106         if (!nn->sessionid_hashtbl)
7107                 goto err_sessionid;
7108
7109         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7110                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7111                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7112         }
7113         for (i = 0; i < SESSION_HASH_SIZE; i++)
7114                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7115         nn->conf_name_tree = RB_ROOT;
7116         nn->unconf_name_tree = RB_ROOT;
7117         nn->boot_time = get_seconds();
7118         nn->grace_ended = false;
7119         nn->nfsd4_manager.block_opens = true;
7120         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7121         INIT_LIST_HEAD(&nn->client_lru);
7122         INIT_LIST_HEAD(&nn->close_lru);
7123         INIT_LIST_HEAD(&nn->del_recall_lru);
7124         spin_lock_init(&nn->client_lock);
7125
7126         spin_lock_init(&nn->blocked_locks_lock);
7127         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7128
7129         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7130         get_net(net);
7131
7132         return 0;
7133
7134 err_sessionid:
7135         kfree(nn->unconf_id_hashtbl);
7136 err_unconf_id:
7137         kfree(nn->conf_id_hashtbl);
7138 err:
7139         return -ENOMEM;
7140 }
7141
7142 static void
7143 nfs4_state_destroy_net(struct net *net)
7144 {
7145         int i;
7146         struct nfs4_client *clp = NULL;
7147         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7148
7149         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7150                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7151                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7152                         destroy_client(clp);
7153                 }
7154         }
7155
7156         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7157
7158         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7159                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7160                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7161                         destroy_client(clp);
7162                 }
7163         }
7164
7165         kfree(nn->sessionid_hashtbl);
7166         kfree(nn->unconf_id_hashtbl);
7167         kfree(nn->conf_id_hashtbl);
7168         put_net(net);
7169 }
7170
7171 int
7172 nfs4_state_start_net(struct net *net)
7173 {
7174         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7175         int ret;
7176
7177         ret = nfs4_state_create_net(net);
7178         if (ret)
7179                 return ret;
7180         locks_start_grace(net, &nn->nfsd4_manager);
7181         nfsd4_client_tracking_init(net);
7182         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7183                nn->nfsd4_grace, net->ns.inum);
7184         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7185         return 0;
7186 }
7187
7188 /* initialization to perform when the nfsd service is started: */
7189
7190 int
7191 nfs4_state_start(void)
7192 {
7193         int ret;
7194
7195         ret = set_callback_cred();
7196         if (ret)
7197                 return ret;
7198
7199         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7200         if (laundry_wq == NULL) {
7201                 ret = -ENOMEM;
7202                 goto out_cleanup_cred;
7203         }
7204         ret = nfsd4_create_callback_queue();
7205         if (ret)
7206                 goto out_free_laundry;
7207
7208         set_max_delegations();
7209         return 0;
7210
7211 out_free_laundry:
7212         destroy_workqueue(laundry_wq);
7213 out_cleanup_cred:
7214         cleanup_callback_cred();
7215         return ret;
7216 }
7217
7218 void
7219 nfs4_state_shutdown_net(struct net *net)
7220 {
7221         struct nfs4_delegation *dp = NULL;
7222         struct list_head *pos, *next, reaplist;
7223         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7224
7225         cancel_delayed_work_sync(&nn->laundromat_work);
7226         locks_end_grace(&nn->nfsd4_manager);
7227
7228         INIT_LIST_HEAD(&reaplist);
7229         spin_lock(&state_lock);
7230         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7231                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7232                 WARN_ON(!unhash_delegation_locked(dp));
7233                 list_add(&dp->dl_recall_lru, &reaplist);
7234         }
7235         spin_unlock(&state_lock);
7236         list_for_each_safe(pos, next, &reaplist) {
7237                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7238                 list_del_init(&dp->dl_recall_lru);
7239                 destroy_unhashed_deleg(dp);
7240         }
7241
7242         nfsd4_client_tracking_exit(net);
7243         nfs4_state_destroy_net(net);
7244 }
7245
7246 void
7247 nfs4_state_shutdown(void)
7248 {
7249         destroy_workqueue(laundry_wq);
7250         nfsd4_destroy_callback_queue();
7251         cleanup_callback_cred();
7252 }
7253
7254 static void
7255 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7256 {
7257         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7258                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7259 }
7260
7261 static void
7262 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7263 {
7264         if (cstate->minorversion) {
7265                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7266                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7267         }
7268 }
7269
7270 void
7271 clear_current_stateid(struct nfsd4_compound_state *cstate)
7272 {
7273         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7274 }
7275
7276 /*
7277  * functions to set current state id
7278  */
7279 void
7280 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7281                 union nfsd4_op_u *u)
7282 {
7283         put_stateid(cstate, &u->open_downgrade.od_stateid);
7284 }
7285
7286 void
7287 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7288                 union nfsd4_op_u *u)
7289 {
7290         put_stateid(cstate, &u->open.op_stateid);
7291 }
7292
7293 void
7294 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7295                 union nfsd4_op_u *u)
7296 {
7297         put_stateid(cstate, &u->close.cl_stateid);
7298 }
7299
7300 void
7301 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7302                 union nfsd4_op_u *u)
7303 {
7304         put_stateid(cstate, &u->lock.lk_resp_stateid);
7305 }
7306
7307 /*
7308  * functions to consume current state id
7309  */
7310
7311 void
7312 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7313                 union nfsd4_op_u *u)
7314 {
7315         get_stateid(cstate, &u->open_downgrade.od_stateid);
7316 }
7317
7318 void
7319 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7320                 union nfsd4_op_u *u)
7321 {
7322         get_stateid(cstate, &u->delegreturn.dr_stateid);
7323 }
7324
7325 void
7326 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7327                 union nfsd4_op_u *u)
7328 {
7329         get_stateid(cstate, &u->free_stateid.fr_stateid);
7330 }
7331
7332 void
7333 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7334                 union nfsd4_op_u *u)
7335 {
7336         get_stateid(cstate, &u->setattr.sa_stateid);
7337 }
7338
7339 void
7340 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7341                 union nfsd4_op_u *u)
7342 {
7343         get_stateid(cstate, &u->close.cl_stateid);
7344 }
7345
7346 void
7347 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7348                 union nfsd4_op_u *u)
7349 {
7350         get_stateid(cstate, &u->locku.lu_stateid);
7351 }
7352
7353 void
7354 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7355                 union nfsd4_op_u *u)
7356 {
7357         get_stateid(cstate, &u->read.rd_stateid);
7358 }
7359
7360 void
7361 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7362                 union nfsd4_op_u *u)
7363 {
7364         get_stateid(cstate, &u->write.wr_stateid);
7365 }