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                         INIT_LIST_HEAD(&nbl->nbl_list);
 256                         INIT_LIST_HEAD(&nbl->nbl_lru);
 257                         fh_copy_shallow(&nbl->nbl_fh, fh);
 258                         locks_init_lock(&nbl->nbl_lock);
 259                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 260                                         &nfsd4_cb_notify_lock_ops,
 261                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 262                 }
 263         }
 264         return nbl;
 265 }
 266
 267 static void
 268 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 269 {
 270         locks_release_private(&nbl->nbl_lock);
 271         kfree(nbl);
 272 }
 273
 274 static void
 275 remove_blocked_locks(struct nfs4_lockowner *lo)
 276 {
 277         struct nfs4_client *clp = lo->lo_owner.so_client;
 278         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 279         struct nfsd4_blocked_lock *nbl;
 280         LIST_HEAD(reaplist);
 281
 282         /* Dequeue all blocked locks */
 283         spin_lock(&nn->blocked_locks_lock);
 284         while (!list_empty(&lo->lo_blocked)) {
 285                 nbl = list_first_entry(&lo->lo_blocked,
 286                                         struct nfsd4_blocked_lock,
 287                                         nbl_list);
 288                 list_del_init(&nbl->nbl_list);
 289                 list_move(&nbl->nbl_lru, &reaplist);
 290         }
 291         spin_unlock(&nn->blocked_locks_lock);
 292
 293         /* Now free them */
 294         while (!list_empty(&reaplist)) {
 295                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 296                                         nbl_lru);
 297                 list_del_init(&nbl->nbl_lru);
 298                 posix_unblock_lock(&nbl->nbl_lock);
 299                 free_blocked_lock(nbl);
 300         }
 301 }
 302
 303 static int
 304 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 305 {
 306         /*
 307          * Since this is just an optimization, we don't try very hard if it
 308          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 309          * just quit trying on anything else.
 310          */
 311         switch (task->tk_status) {
 312         case -NFS4ERR_DELAY:
 313                 rpc_delay(task, 1 * HZ);
 314                 return 0;
 315         default:
 316                 return 1;
 317         }
 318 }
 319
 320 static void
 321 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 322 {
 323         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 324                                                 struct nfsd4_blocked_lock, nbl_cb);
 325
 326         free_blocked_lock(nbl);
 327 }
 328
 329 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 330         .done           = nfsd4_cb_notify_lock_done,
 331         .release        = nfsd4_cb_notify_lock_release,
 332 };
 333
 334 static inline struct nfs4_stateowner *
 335 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 336 {
 337         atomic_inc(&sop->so_count);
 338         return sop;
 339 }
 340
 341 static int
 342 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 343 {
 344         return (sop->so_owner.len == owner->len) &&
 345                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 346 }
 347
 348 static struct nfs4_openowner *
 349 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 350                         struct nfs4_client *clp)
 351 {
 352         struct nfs4_stateowner *so;
 353
 354         lockdep_assert_held(&clp->cl_lock);
 355
 356         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 357                             so_strhash) {
 358                 if (!so->so_is_open_owner)
 359                         continue;
 360                 if (same_owner_str(so, &open->op_owner))
 361                         return openowner(nfs4_get_stateowner(so));
 362         }
 363         return NULL;
 364 }
 365
 366 static struct nfs4_openowner *
 367 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 368                         struct nfs4_client *clp)
 369 {
 370         struct nfs4_openowner *oo;
 371
 372         spin_lock(&clp->cl_lock);
 373         oo = find_openstateowner_str_locked(hashval, open, clp);
 374         spin_unlock(&clp->cl_lock);
 375         return oo;
 376 }
 377
 378 static inline u32
 379 opaque_hashval(const void *ptr, int nbytes)
 380 {
 381         unsigned char *cptr = (unsigned char *) ptr;
 382
 383         u32 x = 0;
 384         while (nbytes--) {
 385                 x *= 37;
 386                 x += *cptr++;
 387         }
 388         return x;
 389 }
 390
 391 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 392 {
 393         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 394
 395         kmem_cache_free(file_slab, fp);
 396 }
 397
 398 void
 399 put_nfs4_file(struct nfs4_file *fi)
 400 {
 401         might_lock(&state_lock);
 402
 403         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 404                 hlist_del_rcu(&fi->fi_hash);
 405                 spin_unlock(&state_lock);
 406                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 407                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 408                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 409         }
 410 }
 411
 412 static struct file *
 413 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 414 {
 415         if (f->fi_fds[oflag])
 416                 return get_file(f->fi_fds[oflag]);
 417         return NULL;
 418 }
 419
 420 static struct file *
 421 find_writeable_file_locked(struct nfs4_file *f)
 422 {
 423         struct file *ret;
 424
 425         lockdep_assert_held(&f->fi_lock);
 426
 427         ret = __nfs4_get_fd(f, O_WRONLY);
 428         if (!ret)
 429                 ret = __nfs4_get_fd(f, O_RDWR);
 430         return ret;
 431 }
 432
 433 static struct file *
 434 find_writeable_file(struct nfs4_file *f)
 435 {
 436         struct file *ret;
 437
 438         spin_lock(&f->fi_lock);
 439         ret = find_writeable_file_locked(f);
 440         spin_unlock(&f->fi_lock);
 441
 442         return ret;
 443 }
 444
 445 static struct file *find_readable_file_locked(struct nfs4_file *f)
 446 {
 447         struct file *ret;
 448
 449         lockdep_assert_held(&f->fi_lock);
 450
 451         ret = __nfs4_get_fd(f, O_RDONLY);
 452         if (!ret)
 453                 ret = __nfs4_get_fd(f, O_RDWR);
 454         return ret;
 455 }
 456
 457 static struct file *
 458 find_readable_file(struct nfs4_file *f)
 459 {
 460         struct file *ret;
 461
 462         spin_lock(&f->fi_lock);
 463         ret = find_readable_file_locked(f);
 464         spin_unlock(&f->fi_lock);
 465
 466         return ret;
 467 }
 468
 469 struct file *
 470 find_any_file(struct nfs4_file *f)
 471 {
 472         struct file *ret;
 473
 474         if (!f)
 475                 return NULL;
 476         spin_lock(&f->fi_lock);
 477         ret = __nfs4_get_fd(f, O_RDWR);
 478         if (!ret) {
 479                 ret = __nfs4_get_fd(f, O_WRONLY);
 480                 if (!ret)
 481                         ret = __nfs4_get_fd(f, O_RDONLY);
 482         }
 483         spin_unlock(&f->fi_lock);
 484         return ret;
 485 }
 486
 487 static atomic_long_t num_delegations;
 488 unsigned long max_delegations;
 489
 490 /*
 491  * Open owner state (share locks)
 492  */
 493
 494 /* hash tables for lock and open owners */
 495 #define OWNER_HASH_BITS              8
 496 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 497 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 498
 499 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 500 {
 501         unsigned int ret;
 502
 503         ret = opaque_hashval(ownername->data, ownername->len);
 504         return ret & OWNER_HASH_MASK;
 505 }
 506
 507 /* hash table for nfs4_file */
 508 #define FILE_HASH_BITS                   8
 509 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 510
 511 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 512 {
 513         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 514 }
 515
 516 static unsigned int file_hashval(struct knfsd_fh *fh)
 517 {
 518         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 519 }
 520
 521 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 522
 523 static void
 524 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 525 {
 526         lockdep_assert_held(&fp->fi_lock);
 527
 528         if (access & NFS4_SHARE_ACCESS_WRITE)
 529                 atomic_inc(&fp->fi_access[O_WRONLY]);
 530         if (access & NFS4_SHARE_ACCESS_READ)
 531                 atomic_inc(&fp->fi_access[O_RDONLY]);
 532 }
 533
 534 static __be32
 535 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 536 {
 537         lockdep_assert_held(&fp->fi_lock);
 538
 539         /* Does this access mode make sense? */
 540         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 541                 return nfserr_inval;
 542
 543         /* Does it conflict with a deny mode already set? */
 544         if ((access & fp->fi_share_deny) != 0)
 545                 return nfserr_share_denied;
 546
 547         __nfs4_file_get_access(fp, access);
 548         return nfs_ok;
 549 }
 550
 551 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 552 {
 553         /* Common case is that there is no deny mode. */
 554         if (deny) {
 555                 /* Does this deny mode make sense? */
 556                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 557                         return nfserr_inval;
 558
 559                 if ((deny & NFS4_SHARE_DENY_READ) &&
 560                     atomic_read(&fp->fi_access[O_RDONLY]))
 561                         return nfserr_share_denied;
 562
 563                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 564                     atomic_read(&fp->fi_access[O_WRONLY]))
 565                         return nfserr_share_denied;
 566         }
 567         return nfs_ok;
 568 }
 569
 570 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 571 {
 572         might_lock(&fp->fi_lock);
 573
 574         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 575                 struct file *f1 = NULL;
 576                 struct file *f2 = NULL;
 577
 578                 swap(f1, fp->fi_fds[oflag]);
 579                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 580                         swap(f2, fp->fi_fds[O_RDWR]);
 581                 spin_unlock(&fp->fi_lock);
 582                 if (f1)
 583                         fput(f1);
 584                 if (f2)
 585                         fput(f2);
 586         }
 587 }
 588
 589 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 590 {
 591         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 592
 593         if (access & NFS4_SHARE_ACCESS_WRITE)
 594                 __nfs4_file_put_access(fp, O_WRONLY);
 595         if (access & NFS4_SHARE_ACCESS_READ)
 596                 __nfs4_file_put_access(fp, O_RDONLY);
 597 }
 598
 599 /*
 600  * Allocate a new open/delegation state counter. This is needed for
 601  * pNFS for proper return on close semantics.
 602  *
 603  * Note that we only allocate it for pNFS-enabled exports, otherwise
 604  * all pointers to struct nfs4_clnt_odstate are always NULL.
 605  */
 606 static struct nfs4_clnt_odstate *
 607 alloc_clnt_odstate(struct nfs4_client *clp)
 608 {
 609         struct nfs4_clnt_odstate *co;
 610
 611         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 612         if (co) {
 613                 co->co_client = clp;
 614                 refcount_set(&co->co_odcount, 1);
 615         }
 616         return co;
 617 }
 618
 619 static void
 620 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 621 {
 622         struct nfs4_file *fp = co->co_file;
 623
 624         lockdep_assert_held(&fp->fi_lock);
 625         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 626 }
 627
 628 static inline void
 629 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 630 {
 631         if (co)
 632                 refcount_inc(&co->co_odcount);
 633 }
 634
 635 static void
 636 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 637 {
 638         struct nfs4_file *fp;
 639
 640         if (!co)
 641                 return;
 642
 643         fp = co->co_file;
 644         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 645                 list_del(&co->co_perfile);
 646                 spin_unlock(&fp->fi_lock);
 647
 648                 nfsd4_return_all_file_layouts(co->co_client, fp);
 649                 kmem_cache_free(odstate_slab, co);
 650         }
 651 }
 652
 653 static struct nfs4_clnt_odstate *
 654 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 655 {
 656         struct nfs4_clnt_odstate *co;
 657         struct nfs4_client *cl;
 658
 659         if (!new)
 660                 return NULL;
 661
 662         cl = new->co_client;
 663
 664         spin_lock(&fp->fi_lock);
 665         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 666                 if (co->co_client == cl) {
 667                         get_clnt_odstate(co);
 668                         goto out;
 669                 }
 670         }
 671         co = new;
 672         co->co_file = fp;
 673         hash_clnt_odstate_locked(new);
 674 out:
 675         spin_unlock(&fp->fi_lock);
 676         return co;
 677 }
 678
 679 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 680                                   void (*sc_free)(struct nfs4_stid *))
 681 {
 682         struct nfs4_stid *stid;
 683         int new_id;
 684
 685         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 686         if (!stid)
 687                 return NULL;
 688
 689         idr_preload(GFP_KERNEL);
 690         spin_lock(&cl->cl_lock);
 691         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 692         spin_unlock(&cl->cl_lock);
 693         idr_preload_end();
 694         if (new_id < 0)
 695                 goto out_free;
 696
 697         stid->sc_free = sc_free;
 698         stid->sc_client = cl;
 699         stid->sc_stateid.si_opaque.so_id = new_id;
 700         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 701         /* Will be incremented before return to client: */
 702         refcount_set(&stid->sc_count, 1);
 703         spin_lock_init(&stid->sc_lock);
 704
 705         /*
 706          * It shouldn't be a problem to reuse an opaque stateid value.
 707          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 708          * example, a stray write retransmission could be accepted by
 709          * the server when it should have been rejected.  Therefore,
 710          * adopt a trick from the sctp code to attempt to maximize the
 711          * amount of time until an id is reused, by ensuring they always
 712          * "increase" (mod INT_MAX):
 713          */
 714         return stid;
 715 out_free:
 716         kmem_cache_free(slab, stid);
 717         return NULL;
 718 }
 719
 720 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 721 {
 722         struct nfs4_stid *stid;
 723
 724         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 725         if (!stid)
 726                 return NULL;
 727
 728         return openlockstateid(stid);
 729 }
 730
 731 static void nfs4_free_deleg(struct nfs4_stid *stid)
 732 {
 733         kmem_cache_free(deleg_slab, stid);
 734         atomic_long_dec(&num_delegations);
 735 }
 736
 737 /*
 738  * When we recall a delegation, we should be careful not to hand it
 739  * out again straight away.
 740  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 741  * in which the filehandles of recalled delegations are "stored".
 742  * If a filehandle appear in either filter, a delegation is blocked.
 743  * When a delegation is recalled, the filehandle is stored in the "new"
 744  * filter.
 745  * Every 30 seconds we swap the filters and clear the "new" one,
 746  * unless both are empty of course.
 747  *
 748  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 749  * low 3 bytes as hash-table indices.
 750  *
 751  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 752  * is used to manage concurrent access.  Testing does not need the lock
 753  * except when swapping the two filters.
 754  */
 755 static DEFINE_SPINLOCK(blocked_delegations_lock);
 756 static struct bloom_pair {
 757         int     entries, old_entries;
 758         time_t  swap_time;
 759         int     new; /* index into 'set' */
 760         DECLARE_BITMAP(set[2], 256);
 761 } blocked_delegations;
 762
 763 static int delegation_blocked(struct knfsd_fh *fh)
 764 {
 765         u32 hash;
 766         struct bloom_pair *bd = &blocked_delegations;
 767
 768         if (bd->entries == 0)
 769                 return 0;
 770         if (seconds_since_boot() - bd->swap_time > 30) {
 771                 spin_lock(&blocked_delegations_lock);
 772                 if (seconds_since_boot() - bd->swap_time > 30) {
 773                         bd->entries -= bd->old_entries;
 774                         bd->old_entries = bd->entries;
 775                         memset(bd->set[bd->new], 0,
 776                                sizeof(bd->set[0]));
 777                         bd->new = 1-bd->new;
 778                         bd->swap_time = seconds_since_boot();
 779                 }
 780                 spin_unlock(&blocked_delegations_lock);
 781         }
 782         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 783         if (test_bit(hash&255, bd->set[0]) &&
 784             test_bit((hash>>8)&255, bd->set[0]) &&
 785             test_bit((hash>>16)&255, bd->set[0]))
 786                 return 1;
 787
 788         if (test_bit(hash&255, bd->set[1]) &&
 789             test_bit((hash>>8)&255, bd->set[1]) &&
 790             test_bit((hash>>16)&255, bd->set[1]))
 791                 return 1;
 792
 793         return 0;
 794 }
 795
 796 static void block_delegations(struct knfsd_fh *fh)
 797 {
 798         u32 hash;
 799         struct bloom_pair *bd = &blocked_delegations;
 800
 801         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 802
 803         spin_lock(&blocked_delegations_lock);
 804         __set_bit(hash&255, bd->set[bd->new]);
 805         __set_bit((hash>>8)&255, bd->set[bd->new]);
 806         __set_bit((hash>>16)&255, bd->set[bd->new]);
 807         if (bd->entries == 0)
 808                 bd->swap_time = seconds_since_boot();
 809         bd->entries += 1;
 810         spin_unlock(&blocked_delegations_lock);
 811 }
 812
 813 static struct nfs4_delegation *
 814 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 815                  struct svc_fh *current_fh,
 816                  struct nfs4_clnt_odstate *odstate)
 817 {
 818         struct nfs4_delegation *dp;
 819         long n;
 820
 821         dprintk("NFSD alloc_init_deleg\n");
 822         n = atomic_long_inc_return(&num_delegations);
 823         if (n < 0 || n > max_delegations)
 824                 goto out_dec;
 825         if (delegation_blocked(&current_fh->fh_handle))
 826                 goto out_dec;
 827         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 828         if (dp == NULL)
 829                 goto out_dec;
 830
 831         /*
 832          * delegation seqid's are never incremented.  The 4.1 special
 833          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 834          * 0 anyway just for consistency and use 1:
 835          */
 836         dp->dl_stid.sc_stateid.si_generation = 1;
 837         INIT_LIST_HEAD(&dp->dl_perfile);
 838         INIT_LIST_HEAD(&dp->dl_perclnt);
 839         INIT_LIST_HEAD(&dp->dl_recall_lru);
 840         dp->dl_clnt_odstate = odstate;
 841         get_clnt_odstate(odstate);
 842         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 843         dp->dl_retries = 1;
 844         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 845                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 846         get_nfs4_file(fp);
 847         dp->dl_stid.sc_file = fp;
 848         return dp;
 849 out_dec:
 850         atomic_long_dec(&num_delegations);
 851         return NULL;
 852 }
 853
 854 void
 855 nfs4_put_stid(struct nfs4_stid *s)
 856 {
 857         struct nfs4_file *fp = s->sc_file;
 858         struct nfs4_client *clp = s->sc_client;
 859
 860         might_lock(&clp->cl_lock);
 861
 862         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 863                 wake_up_all(&close_wq);
 864                 return;
 865         }
 866         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 867         spin_unlock(&clp->cl_lock);
 868         s->sc_free(s);
 869         if (fp)
 870                 put_nfs4_file(fp);
 871 }
 872
 873 void
 874 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 875 {
 876         stateid_t *src = &stid->sc_stateid;
 877
 878         spin_lock(&stid->sc_lock);
 879         if (unlikely(++src->si_generation == 0))
 880                 src->si_generation = 1;
 881         memcpy(dst, src, sizeof(*dst));
 882         spin_unlock(&stid->sc_lock);
 883 }
 884
 885 static void put_deleg_file(struct nfs4_file *fp)
 886 {
 887         struct file *filp = NULL;
 888
 889         spin_lock(&fp->fi_lock);
 890         if (--fp->fi_delegees == 0)
 891                 swap(filp, fp->fi_deleg_file);
 892         spin_unlock(&fp->fi_lock);
 893
 894         if (filp)
 895                 fput(filp);
 896 }
 897
 898 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 899 {
 900         struct nfs4_file *fp = dp->dl_stid.sc_file;
 901         struct file *filp = fp->fi_deleg_file;
 902
 903         WARN_ON_ONCE(!fp->fi_delegees);
 904
 905         vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
 906         put_deleg_file(fp);
 907 }
 908
 909 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 910 {
 911         put_clnt_odstate(dp->dl_clnt_odstate);
 912         nfs4_unlock_deleg_lease(dp);
 913         nfs4_put_stid(&dp->dl_stid);
 914 }
 915
 916 void nfs4_unhash_stid(struct nfs4_stid *s)
 917 {
 918         s->sc_type = 0;
 919 }
 920
 921 /**
 922  * nfs4_delegation_exists - Discover if this delegation already exists
 923  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 924  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 925  *
 926  * Return:
 927  *      On success: true iff an existing delegation is found
 928  */
 929
 930 static bool
 931 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 932 {
 933         struct nfs4_delegation *searchdp = NULL;
 934         struct nfs4_client *searchclp = NULL;
 935
 936         lockdep_assert_held(&state_lock);
 937         lockdep_assert_held(&fp->fi_lock);
 938
 939         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 940                 searchclp = searchdp->dl_stid.sc_client;
 941                 if (clp == searchclp) {
 942                         return true;
 943                 }
 944         }
 945         return false;
 946 }
 947
 948 /**
 949  * hash_delegation_locked - Add a delegation to the appropriate lists
 950  * @dp:     a pointer to the nfs4_delegation we are adding.
 951  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 952  *
 953  * Return:
 954  *      On success: NULL if the delegation was successfully hashed.
 955  *
 956  *      On error: -EAGAIN if one was previously granted to this
 957  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 958  *
 959  */
 960
 961 static int
 962 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 963 {
 964         struct nfs4_client *clp = dp->dl_stid.sc_client;
 965
 966         lockdep_assert_held(&state_lock);
 967         lockdep_assert_held(&fp->fi_lock);
 968
 969         if (nfs4_delegation_exists(clp, fp))
 970                 return -EAGAIN;
 971         refcount_inc(&dp->dl_stid.sc_count);
 972         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 973         list_add(&dp->dl_perfile, &fp->fi_delegations);
 974         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 975         return 0;
 976 }
 977
 978 static bool delegation_hashed(struct nfs4_delegation *dp)
 979 {
 980         return !(list_empty(&dp->dl_perfile));
 981 }
 982
 983 static bool
 984 unhash_delegation_locked(struct nfs4_delegation *dp)
 985 {
 986         struct nfs4_file *fp = dp->dl_stid.sc_file;
 987
 988         lockdep_assert_held(&state_lock);
 989
 990         if (!delegation_hashed(dp))
 991                 return false;
 992
 993         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 994         /* Ensure that deleg break won't try to requeue it */
 995         ++dp->dl_time;
 996         spin_lock(&fp->fi_lock);
 997         list_del_init(&dp->dl_perclnt);
 998         list_del_init(&dp->dl_recall_lru);
 999         list_del_init(&dp->dl_perfile);
1000         spin_unlock(&fp->fi_lock);
1001         return true;
1002 }
1003
1004 static void destroy_delegation(struct nfs4_delegation *dp)
1005 {
1006         bool unhashed;
1007
1008         spin_lock(&state_lock);
1009         unhashed = unhash_delegation_locked(dp);
1010         spin_unlock(&state_lock);
1011         if (unhashed)
1012                 destroy_unhashed_deleg(dp);
1013 }
1014
1015 static void revoke_delegation(struct nfs4_delegation *dp)
1016 {
1017         struct nfs4_client *clp = dp->dl_stid.sc_client;
1018
1019         WARN_ON(!list_empty(&dp->dl_recall_lru));
1020
1021         if (clp->cl_minorversion) {
1022                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1023                 refcount_inc(&dp->dl_stid.sc_count);
1024                 spin_lock(&clp->cl_lock);
1025                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1026                 spin_unlock(&clp->cl_lock);
1027         }
1028         destroy_unhashed_deleg(dp);
1029 }
1030
1031 /*
1032  * SETCLIENTID state
1033  */
1034
1035 static unsigned int clientid_hashval(u32 id)
1036 {
1037         return id & CLIENT_HASH_MASK;
1038 }
1039
1040 static unsigned int clientstr_hashval(const char *name)
1041 {
1042         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1043 }
1044
1045 /*
1046  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1047  * st_{access,deny}_bmap field of the stateid, in order to track not
1048  * only what share bits are currently in force, but also what
1049  * combinations of share bits previous opens have used.  This allows us
1050  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1051  * return an error if the client attempt to downgrade to a combination
1052  * of share bits not explicable by closing some of its previous opens.
1053  *
1054  * XXX: This enforcement is actually incomplete, since we don't keep
1055  * track of access/deny bit combinations; so, e.g., we allow:
1056  *
1057  *      OPEN allow read, deny write
1058  *      OPEN allow both, deny none
1059  *      DOWNGRADE allow read, deny none
1060  *
1061  * which we should reject.
1062  */
1063 static unsigned int
1064 bmap_to_share_mode(unsigned long bmap) {
1065         int i;
1066         unsigned int access = 0;
1067
1068         for (i = 1; i < 4; i++) {
1069                 if (test_bit(i, &bmap))
1070                         access |= i;
1071         }
1072         return access;
1073 }
1074
1075 /* set share access for a given stateid */
1076 static inline void
1077 set_access(u32 access, struct nfs4_ol_stateid *stp)
1078 {
1079         unsigned char mask = 1 << access;
1080
1081         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1082         stp->st_access_bmap |= mask;
1083 }
1084
1085 /* clear share access for a given stateid */
1086 static inline void
1087 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1088 {
1089         unsigned char mask = 1 << access;
1090
1091         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1092         stp->st_access_bmap &= ~mask;
1093 }
1094
1095 /* test whether a given stateid has access */
1096 static inline bool
1097 test_access(u32 access, struct nfs4_ol_stateid *stp)
1098 {
1099         unsigned char mask = 1 << access;
1100
1101         return (bool)(stp->st_access_bmap & mask);
1102 }
1103
1104 /* set share deny for a given stateid */
1105 static inline void
1106 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1107 {
1108         unsigned char mask = 1 << deny;
1109
1110         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1111         stp->st_deny_bmap |= mask;
1112 }
1113
1114 /* clear share deny for a given stateid */
1115 static inline void
1116 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1117 {
1118         unsigned char mask = 1 << deny;
1119
1120         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1121         stp->st_deny_bmap &= ~mask;
1122 }
1123
1124 /* test whether a given stateid is denying specific access */
1125 static inline bool
1126 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1127 {
1128         unsigned char mask = 1 << deny;
1129
1130         return (bool)(stp->st_deny_bmap & mask);
1131 }
1132
1133 static int nfs4_access_to_omode(u32 access)
1134 {
1135         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1136         case NFS4_SHARE_ACCESS_READ:
1137                 return O_RDONLY;
1138         case NFS4_SHARE_ACCESS_WRITE:
1139                 return O_WRONLY;
1140         case NFS4_SHARE_ACCESS_BOTH:
1141                 return O_RDWR;
1142         }
1143         WARN_ON_ONCE(1);
1144         return O_RDONLY;
1145 }
1146
1147 /*
1148  * A stateid that had a deny mode associated with it is being released
1149  * or downgraded. Recalculate the deny mode on the file.
1150  */
1151 static void
1152 recalculate_deny_mode(struct nfs4_file *fp)
1153 {
1154         struct nfs4_ol_stateid *stp;
1155
1156         spin_lock(&fp->fi_lock);
1157         fp->fi_share_deny = 0;
1158         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1159                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1160         spin_unlock(&fp->fi_lock);
1161 }
1162
1163 static void
1164 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1165 {
1166         int i;
1167         bool change = false;
1168
1169         for (i = 1; i < 4; i++) {
1170                 if ((i & deny) != i) {
1171                         change = true;
1172                         clear_deny(i, stp);
1173                 }
1174         }
1175
1176         /* Recalculate per-file deny mode if there was a change */
1177         if (change)
1178                 recalculate_deny_mode(stp->st_stid.sc_file);
1179 }
1180
1181 /* release all access and file references for a given stateid */
1182 static void
1183 release_all_access(struct nfs4_ol_stateid *stp)
1184 {
1185         int i;
1186         struct nfs4_file *fp = stp->st_stid.sc_file;
1187
1188         if (fp && stp->st_deny_bmap != 0)
1189                 recalculate_deny_mode(fp);
1190
1191         for (i = 1; i < 4; i++) {
1192                 if (test_access(i, stp))
1193                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1194                 clear_access(i, stp);
1195         }
1196 }
1197
1198 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1199 {
1200         kfree(sop->so_owner.data);
1201         sop->so_ops->so_free(sop);
1202 }
1203
1204 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1205 {
1206         struct nfs4_client *clp = sop->so_client;
1207
1208         might_lock(&clp->cl_lock);
1209
1210         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1211                 return;
1212         sop->so_ops->so_unhash(sop);
1213         spin_unlock(&clp->cl_lock);
1214         nfs4_free_stateowner(sop);
1215 }
1216
1217 static bool
1218 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1219 {
1220         return list_empty(&stp->st_perfile);
1221 }
1222
1223 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1224 {
1225         struct nfs4_file *fp = stp->st_stid.sc_file;
1226
1227         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1228
1229         if (list_empty(&stp->st_perfile))
1230                 return false;
1231
1232         spin_lock(&fp->fi_lock);
1233         list_del_init(&stp->st_perfile);
1234         spin_unlock(&fp->fi_lock);
1235         list_del(&stp->st_perstateowner);
1236         return true;
1237 }
1238
1239 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1240 {
1241         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1242
1243         put_clnt_odstate(stp->st_clnt_odstate);
1244         release_all_access(stp);
1245         if (stp->st_stateowner)
1246                 nfs4_put_stateowner(stp->st_stateowner);
1247         kmem_cache_free(stateid_slab, stid);
1248 }
1249
1250 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1251 {
1252         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1253         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1254         struct file *file;
1255
1256         file = find_any_file(stp->st_stid.sc_file);
1257         if (file)
1258                 filp_close(file, (fl_owner_t)lo);
1259         nfs4_free_ol_stateid(stid);
1260 }
1261
1262 /*
1263  * Put the persistent reference to an already unhashed generic stateid, while
1264  * holding the cl_lock. If it's the last reference, then put it onto the
1265  * reaplist for later destruction.
1266  */
1267 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1268                                        struct list_head *reaplist)
1269 {
1270         struct nfs4_stid *s = &stp->st_stid;
1271         struct nfs4_client *clp = s->sc_client;
1272
1273         lockdep_assert_held(&clp->cl_lock);
1274
1275         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1276
1277         if (!refcount_dec_and_test(&s->sc_count)) {
1278                 wake_up_all(&close_wq);
1279                 return;
1280         }
1281
1282         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1283         list_add(&stp->st_locks, reaplist);
1284 }
1285
1286 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1287 {
1288         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1289
1290         if (!unhash_ol_stateid(stp))
1291                 return false;
1292         list_del_init(&stp->st_locks);
1293         nfs4_unhash_stid(&stp->st_stid);
1294         return true;
1295 }
1296
1297 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1298 {
1299         struct nfs4_client *clp = stp->st_stid.sc_client;
1300         bool unhashed;
1301
1302         spin_lock(&clp->cl_lock);
1303         unhashed = unhash_lock_stateid(stp);
1304         spin_unlock(&clp->cl_lock);
1305         if (unhashed)
1306                 nfs4_put_stid(&stp->st_stid);
1307 }
1308
1309 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1310 {
1311         struct nfs4_client *clp = lo->lo_owner.so_client;
1312
1313         lockdep_assert_held(&clp->cl_lock);
1314
1315         list_del_init(&lo->lo_owner.so_strhash);
1316 }
1317
1318 /*
1319  * Free a list of generic stateids that were collected earlier after being
1320  * fully unhashed.
1321  */
1322 static void
1323 free_ol_stateid_reaplist(struct list_head *reaplist)
1324 {
1325         struct nfs4_ol_stateid *stp;
1326         struct nfs4_file *fp;
1327
1328         might_sleep();
1329
1330         while (!list_empty(reaplist)) {
1331                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1332                                        st_locks);
1333                 list_del(&stp->st_locks);
1334                 fp = stp->st_stid.sc_file;
1335                 stp->st_stid.sc_free(&stp->st_stid);
1336                 if (fp)
1337                         put_nfs4_file(fp);
1338         }
1339 }
1340
1341 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1342                                        struct list_head *reaplist)
1343 {
1344         struct nfs4_ol_stateid *stp;
1345
1346         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1347
1348         while (!list_empty(&open_stp->st_locks)) {
1349                 stp = list_entry(open_stp->st_locks.next,
1350                                 struct nfs4_ol_stateid, st_locks);
1351                 WARN_ON(!unhash_lock_stateid(stp));
1352                 put_ol_stateid_locked(stp, reaplist);
1353         }
1354 }
1355
1356 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1357                                 struct list_head *reaplist)
1358 {
1359         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1360
1361         if (!unhash_ol_stateid(stp))
1362                 return false;
1363         release_open_stateid_locks(stp, reaplist);
1364         return true;
1365 }
1366
1367 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1368 {
1369         LIST_HEAD(reaplist);
1370
1371         spin_lock(&stp->st_stid.sc_client->cl_lock);
1372         if (unhash_open_stateid(stp, &reaplist))
1373                 put_ol_stateid_locked(stp, &reaplist);
1374         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1375         free_ol_stateid_reaplist(&reaplist);
1376 }
1377
1378 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1379 {
1380         struct nfs4_client *clp = oo->oo_owner.so_client;
1381
1382         lockdep_assert_held(&clp->cl_lock);
1383
1384         list_del_init(&oo->oo_owner.so_strhash);
1385         list_del_init(&oo->oo_perclient);
1386 }
1387
1388 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1389 {
1390         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1391                                           nfsd_net_id);
1392         struct nfs4_ol_stateid *s;
1393
1394         spin_lock(&nn->client_lock);
1395         s = oo->oo_last_closed_stid;
1396         if (s) {
1397                 list_del_init(&oo->oo_close_lru);
1398                 oo->oo_last_closed_stid = NULL;
1399         }
1400         spin_unlock(&nn->client_lock);
1401         if (s)
1402                 nfs4_put_stid(&s->st_stid);
1403 }
1404
1405 static void release_openowner(struct nfs4_openowner *oo)
1406 {
1407         struct nfs4_ol_stateid *stp;
1408         struct nfs4_client *clp = oo->oo_owner.so_client;
1409         struct list_head reaplist;
1410
1411         INIT_LIST_HEAD(&reaplist);
1412
1413         spin_lock(&clp->cl_lock);
1414         unhash_openowner_locked(oo);
1415         while (!list_empty(&oo->oo_owner.so_stateids)) {
1416                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1417                                 struct nfs4_ol_stateid, st_perstateowner);
1418                 if (unhash_open_stateid(stp, &reaplist))
1419                         put_ol_stateid_locked(stp, &reaplist);
1420         }
1421         spin_unlock(&clp->cl_lock);
1422         free_ol_stateid_reaplist(&reaplist);
1423         release_last_closed_stateid(oo);
1424         nfs4_put_stateowner(&oo->oo_owner);
1425 }
1426
1427 static inline int
1428 hash_sessionid(struct nfs4_sessionid *sessionid)
1429 {
1430         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1431
1432         return sid->sequence % SESSION_HASH_SIZE;
1433 }
1434
1435 #ifdef CONFIG_SUNRPC_DEBUG
1436 static inline void
1437 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1438 {
1439         u32 *ptr = (u32 *)(&sessionid->data[0]);
1440         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1441 }
1442 #else
1443 static inline void
1444 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1445 {
1446 }
1447 #endif
1448
1449 /*
1450  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1451  * won't be used for replay.
1452  */
1453 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1454 {
1455         struct nfs4_stateowner *so = cstate->replay_owner;
1456
1457         if (nfserr == nfserr_replay_me)
1458                 return;
1459
1460         if (!seqid_mutating_err(ntohl(nfserr))) {
1461                 nfsd4_cstate_clear_replay(cstate);
1462                 return;
1463         }
1464         if (!so)
1465                 return;
1466         if (so->so_is_open_owner)
1467                 release_last_closed_stateid(openowner(so));
1468         so->so_seqid++;
1469         return;
1470 }
1471
1472 static void
1473 gen_sessionid(struct nfsd4_session *ses)
1474 {
1475         struct nfs4_client *clp = ses->se_client;
1476         struct nfsd4_sessionid *sid;
1477
1478         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1479         sid->clientid = clp->cl_clientid;
1480         sid->sequence = current_sessionid++;
1481         sid->reserved = 0;
1482 }
1483
1484 /*
1485  * The protocol defines ca_maxresponssize_cached to include the size of
1486  * the rpc header, but all we need to cache is the data starting after
1487  * the end of the initial SEQUENCE operation--the rest we regenerate
1488  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1489  * value that is the number of bytes in our cache plus a few additional
1490  * bytes.  In order to stay on the safe side, and not promise more than
1491  * we can cache, those additional bytes must be the minimum possible: 24
1492  * bytes of rpc header (xid through accept state, with AUTH_NULL
1493  * verifier), 12 for the compound header (with zero-length tag), and 44
1494  * for the SEQUENCE op response:
1495  */
1496 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1497
1498 static void
1499 free_session_slots(struct nfsd4_session *ses)
1500 {
1501         int i;
1502
1503         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1504                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1505                 kfree(ses->se_slots[i]);
1506         }
1507 }
1508
1509 /*
1510  * We don't actually need to cache the rpc and session headers, so we
1511  * can allocate a little less for each slot:
1512  */
1513 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1514 {
1515         u32 size;
1516
1517         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1518                 size = 0;
1519         else
1520                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1521         return size + sizeof(struct nfsd4_slot);
1522 }
1523
1524 /*
1525  * XXX: If we run out of reserved DRC memory we could (up to a point)
1526  * re-negotiate active sessions and reduce their slot usage to make
1527  * room for new connections. For now we just fail the create session.
1528  */
1529 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1530 {
1531         u32 slotsize = slot_bytes(ca);
1532         u32 num = ca->maxreqs;
1533         unsigned long avail, total_avail;
1534
1535         spin_lock(&nfsd_drc_lock);
1536         total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1537         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1538         /*
1539          * Never use more than a third of the remaining memory,
1540          * unless it's the only way to give this client a slot:
1541          */
1542         avail = clamp_t(unsigned long, avail, slotsize, total_avail/3);
1543         num = min_t(int, num, avail / slotsize);
1544         nfsd_drc_mem_used += num * slotsize;
1545         spin_unlock(&nfsd_drc_lock);
1546
1547         return num;
1548 }
1549
1550 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1551 {
1552         int slotsize = slot_bytes(ca);
1553
1554         spin_lock(&nfsd_drc_lock);
1555         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1556         spin_unlock(&nfsd_drc_lock);
1557 }
1558
1559 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1560                                            struct nfsd4_channel_attrs *battrs)
1561 {
1562         int numslots = fattrs->maxreqs;
1563         int slotsize = slot_bytes(fattrs);
1564         struct nfsd4_session *new;
1565         int mem, i;
1566
1567         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1568                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1569         mem = numslots * sizeof(struct nfsd4_slot *);
1570
1571         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1572         if (!new)
1573                 return NULL;
1574         /* allocate each struct nfsd4_slot and data cache in one piece */
1575         for (i = 0; i < numslots; i++) {
1576                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1577                 if (!new->se_slots[i])
1578                         goto out_free;
1579         }
1580
1581         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1582         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1583
1584         return new;
1585 out_free:
1586         while (i--)
1587                 kfree(new->se_slots[i]);
1588         kfree(new);
1589         return NULL;
1590 }
1591
1592 static void free_conn(struct nfsd4_conn *c)
1593 {
1594         svc_xprt_put(c->cn_xprt);
1595         kfree(c);
1596 }
1597
1598 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1599 {
1600         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1601         struct nfs4_client *clp = c->cn_session->se_client;
1602
1603         spin_lock(&clp->cl_lock);
1604         if (!list_empty(&c->cn_persession)) {
1605                 list_del(&c->cn_persession);
1606                 free_conn(c);
1607         }
1608         nfsd4_probe_callback(clp);
1609         spin_unlock(&clp->cl_lock);
1610 }
1611
1612 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1613 {
1614         struct nfsd4_conn *conn;
1615
1616         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1617         if (!conn)
1618                 return NULL;
1619         svc_xprt_get(rqstp->rq_xprt);
1620         conn->cn_xprt = rqstp->rq_xprt;
1621         conn->cn_flags = flags;
1622         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1623         return conn;
1624 }
1625
1626 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1627 {
1628         conn->cn_session = ses;
1629         list_add(&conn->cn_persession, &ses->se_conns);
1630 }
1631
1632 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1633 {
1634         struct nfs4_client *clp = ses->se_client;
1635
1636         spin_lock(&clp->cl_lock);
1637         __nfsd4_hash_conn(conn, ses);
1638         spin_unlock(&clp->cl_lock);
1639 }
1640
1641 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1642 {
1643         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1644         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1645 }
1646
1647 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1648 {
1649         int ret;
1650
1651         nfsd4_hash_conn(conn, ses);
1652         ret = nfsd4_register_conn(conn);
1653         if (ret)
1654                 /* oops; xprt is already down: */
1655                 nfsd4_conn_lost(&conn->cn_xpt_user);
1656         /* We may have gained or lost a callback channel: */
1657         nfsd4_probe_callback_sync(ses->se_client);
1658 }
1659
1660 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1661 {
1662         u32 dir = NFS4_CDFC4_FORE;
1663
1664         if (cses->flags & SESSION4_BACK_CHAN)
1665                 dir |= NFS4_CDFC4_BACK;
1666         return alloc_conn(rqstp, dir);
1667 }
1668
1669 /* must be called under client_lock */
1670 static void nfsd4_del_conns(struct nfsd4_session *s)
1671 {
1672         struct nfs4_client *clp = s->se_client;
1673         struct nfsd4_conn *c;
1674
1675         spin_lock(&clp->cl_lock);
1676         while (!list_empty(&s->se_conns)) {
1677                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1678                 list_del_init(&c->cn_persession);
1679                 spin_unlock(&clp->cl_lock);
1680
1681                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1682                 free_conn(c);
1683
1684                 spin_lock(&clp->cl_lock);
1685         }
1686         spin_unlock(&clp->cl_lock);
1687 }
1688
1689 static void __free_session(struct nfsd4_session *ses)
1690 {
1691         free_session_slots(ses);
1692         kfree(ses);
1693 }
1694
1695 static void free_session(struct nfsd4_session *ses)
1696 {
1697         nfsd4_del_conns(ses);
1698         nfsd4_put_drc_mem(&ses->se_fchannel);
1699         __free_session(ses);
1700 }
1701
1702 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1703 {
1704         int idx;
1705         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1706
1707         new->se_client = clp;
1708         gen_sessionid(new);
1709
1710         INIT_LIST_HEAD(&new->se_conns);
1711
1712         new->se_cb_seq_nr = 1;
1713         new->se_flags = cses->flags;
1714         new->se_cb_prog = cses->callback_prog;
1715         new->se_cb_sec = cses->cb_sec;
1716         atomic_set(&new->se_ref, 0);
1717         idx = hash_sessionid(&new->se_sessionid);
1718         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1719         spin_lock(&clp->cl_lock);
1720         list_add(&new->se_perclnt, &clp->cl_sessions);
1721         spin_unlock(&clp->cl_lock);
1722
1723         {
1724                 struct sockaddr *sa = svc_addr(rqstp);
1725                 /*
1726                  * This is a little silly; with sessions there's no real
1727                  * use for the callback address.  Use the peer address
1728                  * as a reasonable default for now, but consider fixing
1729                  * the rpc client not to require an address in the
1730                  * future:
1731                  */
1732                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1733                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1734         }
1735 }
1736
1737 /* caller must hold client_lock */
1738 static struct nfsd4_session *
1739 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1740 {
1741         struct nfsd4_session *elem;
1742         int idx;
1743         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1744
1745         lockdep_assert_held(&nn->client_lock);
1746
1747         dump_sessionid(__func__, sessionid);
1748         idx = hash_sessionid(sessionid);
1749         /* Search in the appropriate list */
1750         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1751                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1752                             NFS4_MAX_SESSIONID_LEN)) {
1753                         return elem;
1754                 }
1755         }
1756
1757         dprintk("%s: session not found\n", __func__);
1758         return NULL;
1759 }
1760
1761 static struct nfsd4_session *
1762 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1763                 __be32 *ret)
1764 {
1765         struct nfsd4_session *session;
1766         __be32 status = nfserr_badsession;
1767
1768         session = __find_in_sessionid_hashtbl(sessionid, net);
1769         if (!session)
1770                 goto out;
1771         status = nfsd4_get_session_locked(session);
1772         if (status)
1773                 session = NULL;
1774 out:
1775         *ret = status;
1776         return session;
1777 }
1778
1779 /* caller must hold client_lock */
1780 static void
1781 unhash_session(struct nfsd4_session *ses)
1782 {
1783         struct nfs4_client *clp = ses->se_client;
1784         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1785
1786         lockdep_assert_held(&nn->client_lock);
1787
1788         list_del(&ses->se_hash);
1789         spin_lock(&ses->se_client->cl_lock);
1790         list_del(&ses->se_perclnt);
1791         spin_unlock(&ses->se_client->cl_lock);
1792 }
1793
1794 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1795 static int
1796 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1797 {
1798         /*
1799          * We're assuming the clid was not given out from a boot
1800          * precisely 2^32 (about 136 years) before this one.  That seems
1801          * a safe assumption:
1802          */
1803         if (clid->cl_boot == (u32)nn->boot_time)
1804                 return 0;
1805         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1806                 clid->cl_boot, clid->cl_id, nn->boot_time);
1807         return 1;
1808 }
1809
1810 /*
1811  * XXX Should we use a slab cache ?
1812  * This type of memory management is somewhat inefficient, but we use it
1813  * anyway since SETCLIENTID is not a common operation.
1814  */
1815 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1816 {
1817         struct nfs4_client *clp;
1818         int i;
1819
1820         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1821         if (clp == NULL)
1822                 return NULL;
1823         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1824         if (clp->cl_name.data == NULL)
1825                 goto err_no_name;
1826         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1827                                                  sizeof(struct list_head),
1828                                                  GFP_KERNEL);
1829         if (!clp->cl_ownerstr_hashtbl)
1830                 goto err_no_hashtbl;
1831         for (i = 0; i < OWNER_HASH_SIZE; i++)
1832                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1833         clp->cl_name.len = name.len;
1834         INIT_LIST_HEAD(&clp->cl_sessions);
1835         idr_init(&clp->cl_stateids);
1836         atomic_set(&clp->cl_refcount, 0);
1837         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1838         INIT_LIST_HEAD(&clp->cl_idhash);
1839         INIT_LIST_HEAD(&clp->cl_openowners);
1840         INIT_LIST_HEAD(&clp->cl_delegations);
1841         INIT_LIST_HEAD(&clp->cl_lru);
1842         INIT_LIST_HEAD(&clp->cl_revoked);
1843 #ifdef CONFIG_NFSD_PNFS
1844         INIT_LIST_HEAD(&clp->cl_lo_states);
1845 #endif
1846         spin_lock_init(&clp->cl_lock);
1847         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1848         return clp;
1849 err_no_hashtbl:
1850         kfree(clp->cl_name.data);
1851 err_no_name:
1852         kmem_cache_free(client_slab, clp);
1853         return NULL;
1854 }
1855
1856 static void
1857 free_client(struct nfs4_client *clp)
1858 {
1859         while (!list_empty(&clp->cl_sessions)) {
1860                 struct nfsd4_session *ses;
1861                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1862                                 se_perclnt);
1863                 list_del(&ses->se_perclnt);
1864                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1865                 free_session(ses);
1866         }
1867         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1868         free_svc_cred(&clp->cl_cred);
1869         kfree(clp->cl_ownerstr_hashtbl);
1870         kfree(clp->cl_name.data);
1871         idr_destroy(&clp->cl_stateids);
1872         kmem_cache_free(client_slab, clp);
1873 }
1874
1875 /* must be called under the client_lock */
1876 static void
1877 unhash_client_locked(struct nfs4_client *clp)
1878 {
1879         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1880         struct nfsd4_session *ses;
1881
1882         lockdep_assert_held(&nn->client_lock);
1883
1884         /* Mark the client as expired! */
1885         clp->cl_time = 0;
1886         /* Make it invisible */
1887         if (!list_empty(&clp->cl_idhash)) {
1888                 list_del_init(&clp->cl_idhash);
1889                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1890                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1891                 else
1892                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1893         }
1894         list_del_init(&clp->cl_lru);
1895         spin_lock(&clp->cl_lock);
1896         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1897                 list_del_init(&ses->se_hash);
1898         spin_unlock(&clp->cl_lock);
1899 }
1900
1901 static void
1902 unhash_client(struct nfs4_client *clp)
1903 {
1904         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1905
1906         spin_lock(&nn->client_lock);
1907         unhash_client_locked(clp);
1908         spin_unlock(&nn->client_lock);
1909 }
1910
1911 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1912 {
1913         if (atomic_read(&clp->cl_refcount))
1914                 return nfserr_jukebox;
1915         unhash_client_locked(clp);
1916         return nfs_ok;
1917 }
1918
1919 static void
1920 __destroy_client(struct nfs4_client *clp)
1921 {
1922         int i;
1923         struct nfs4_openowner *oo;
1924         struct nfs4_delegation *dp;
1925         struct list_head reaplist;
1926
1927         INIT_LIST_HEAD(&reaplist);
1928         spin_lock(&state_lock);
1929         while (!list_empty(&clp->cl_delegations)) {
1930                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1931                 WARN_ON(!unhash_delegation_locked(dp));
1932                 list_add(&dp->dl_recall_lru, &reaplist);
1933         }
1934         spin_unlock(&state_lock);
1935         while (!list_empty(&reaplist)) {
1936                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1937                 list_del_init(&dp->dl_recall_lru);
1938                 destroy_unhashed_deleg(dp);
1939         }
1940         while (!list_empty(&clp->cl_revoked)) {
1941                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1942                 list_del_init(&dp->dl_recall_lru);
1943                 nfs4_put_stid(&dp->dl_stid);
1944         }
1945         while (!list_empty(&clp->cl_openowners)) {
1946                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1947                 nfs4_get_stateowner(&oo->oo_owner);
1948                 release_openowner(oo);
1949         }
1950         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1951                 struct nfs4_stateowner *so, *tmp;
1952
1953                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1954                                          so_strhash) {
1955                         /* Should be no openowners at this point */
1956                         WARN_ON_ONCE(so->so_is_open_owner);
1957                         remove_blocked_locks(lockowner(so));
1958                 }
1959         }
1960         nfsd4_return_all_client_layouts(clp);
1961         nfsd4_shutdown_callback(clp);
1962         if (clp->cl_cb_conn.cb_xprt)
1963                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1964         free_client(clp);
1965 }
1966
1967 static void
1968 destroy_client(struct nfs4_client *clp)
1969 {
1970         unhash_client(clp);
1971         __destroy_client(clp);
1972 }
1973
1974 static void expire_client(struct nfs4_client *clp)
1975 {
1976         unhash_client(clp);
1977         nfsd4_client_record_remove(clp);
1978         __destroy_client(clp);
1979 }
1980
1981 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1982 {
1983         memcpy(target->cl_verifier.data, source->data,
1984                         sizeof(target->cl_verifier.data));
1985 }
1986
1987 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1988 {
1989         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1990         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1991 }
1992
1993 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1994 {
1995         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1996         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1997                                                                 GFP_KERNEL);
1998         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
1999         if ((source->cr_principal && !target->cr_principal) ||
2000             (source->cr_raw_principal && !target->cr_raw_principal) ||
2001             (source->cr_targ_princ && !target->cr_targ_princ))
2002                 return -ENOMEM;
2003
2004         target->cr_flavor = source->cr_flavor;
2005         target->cr_uid = source->cr_uid;
2006         target->cr_gid = source->cr_gid;
2007         target->cr_group_info = source->cr_group_info;
2008         get_group_info(target->cr_group_info);
2009         target->cr_gss_mech = source->cr_gss_mech;
2010         if (source->cr_gss_mech)
2011                 gss_mech_get(source->cr_gss_mech);
2012         return 0;
2013 }
2014
2015 static int
2016 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2017 {
2018         if (o1->len < o2->len)
2019                 return -1;
2020         if (o1->len > o2->len)
2021                 return 1;
2022         return memcmp(o1->data, o2->data, o1->len);
2023 }
2024
2025 static int same_name(const char *n1, const char *n2)
2026 {
2027         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2028 }
2029
2030 static int
2031 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2032 {
2033         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2034 }
2035
2036 static int
2037 same_clid(clientid_t *cl1, clientid_t *cl2)
2038 {
2039         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2040 }
2041
2042 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2043 {
2044         int i;
2045
2046         if (g1->ngroups != g2->ngroups)
2047                 return false;
2048         for (i=0; i<g1->ngroups; i++)
2049                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2050                         return false;
2051         return true;
2052 }
2053
2054 /*
2055  * RFC 3530 language requires clid_inuse be returned when the
2056  * "principal" associated with a requests differs from that previously
2057  * used.  We use uid, gid's, and gss principal string as our best
2058  * approximation.  We also don't want to allow non-gss use of a client
2059  * established using gss: in theory cr_principal should catch that
2060  * change, but in practice cr_principal can be null even in the gss case
2061  * since gssd doesn't always pass down a principal string.
2062  */
2063 static bool is_gss_cred(struct svc_cred *cr)
2064 {
2065         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2066         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2067 }
2068
2069
2070 static bool
2071 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2072 {
2073         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2074                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2075                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2076                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2077                 return false;
2078         /* XXX: check that cr_targ_princ fields match ? */
2079         if (cr1->cr_principal == cr2->cr_principal)
2080                 return true;
2081         if (!cr1->cr_principal || !cr2->cr_principal)
2082                 return false;
2083         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2084 }
2085
2086 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2087 {
2088         struct svc_cred *cr = &rqstp->rq_cred;
2089         u32 service;
2090
2091         if (!cr->cr_gss_mech)
2092                 return false;
2093         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2094         return service == RPC_GSS_SVC_INTEGRITY ||
2095                service == RPC_GSS_SVC_PRIVACY;
2096 }
2097
2098 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2099 {
2100         struct svc_cred *cr = &rqstp->rq_cred;
2101
2102         if (!cl->cl_mach_cred)
2103                 return true;
2104         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2105                 return false;
2106         if (!svc_rqst_integrity_protected(rqstp))
2107                 return false;
2108         if (cl->cl_cred.cr_raw_principal)
2109                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2110                                                 cr->cr_raw_principal);
2111         if (!cr->cr_principal)
2112                 return false;
2113         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2114 }
2115
2116 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2117 {
2118         __be32 verf[2];
2119
2120         /*
2121          * This is opaque to client, so no need to byte-swap. Use
2122          * __force to keep sparse happy
2123          */
2124         verf[0] = (__force __be32)get_seconds();
2125         verf[1] = (__force __be32)nn->clverifier_counter++;
2126         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2127 }
2128
2129 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2130 {
2131         clp->cl_clientid.cl_boot = nn->boot_time;
2132         clp->cl_clientid.cl_id = nn->clientid_counter++;
2133         gen_confirm(clp, nn);
2134 }
2135
2136 static struct nfs4_stid *
2137 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2138 {
2139         struct nfs4_stid *ret;
2140
2141         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2142         if (!ret || !ret->sc_type)
2143                 return NULL;
2144         return ret;
2145 }
2146
2147 static struct nfs4_stid *
2148 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2149 {
2150         struct nfs4_stid *s;
2151
2152         spin_lock(&cl->cl_lock);
2153         s = find_stateid_locked(cl, t);
2154         if (s != NULL) {
2155                 if (typemask & s->sc_type)
2156                         refcount_inc(&s->sc_count);
2157                 else
2158                         s = NULL;
2159         }
2160         spin_unlock(&cl->cl_lock);
2161         return s;
2162 }
2163
2164 static struct nfs4_client *create_client(struct xdr_netobj name,
2165                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2166 {
2167         struct nfs4_client *clp;
2168         struct sockaddr *sa = svc_addr(rqstp);
2169         int ret;
2170         struct net *net = SVC_NET(rqstp);
2171
2172         clp = alloc_client(name);
2173         if (clp == NULL)
2174                 return NULL;
2175
2176         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2177         if (ret) {
2178                 free_client(clp);
2179                 return NULL;
2180         }
2181         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2182         clp->cl_time = get_seconds();
2183         clear_bit(0, &clp->cl_cb_slot_busy);
2184         copy_verf(clp, verf);
2185         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2186         clp->cl_cb_session = NULL;
2187         clp->net = net;
2188         return clp;
2189 }
2190
2191 static void
2192 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2193 {
2194         struct rb_node **new = &(root->rb_node), *parent = NULL;
2195         struct nfs4_client *clp;
2196
2197         while (*new) {
2198                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2199                 parent = *new;
2200
2201                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2202                         new = &((*new)->rb_left);
2203                 else
2204                         new = &((*new)->rb_right);
2205         }
2206
2207         rb_link_node(&new_clp->cl_namenode, parent, new);
2208         rb_insert_color(&new_clp->cl_namenode, root);
2209 }
2210
2211 static struct nfs4_client *
2212 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2213 {
2214         int cmp;
2215         struct rb_node *node = root->rb_node;
2216         struct nfs4_client *clp;
2217
2218         while (node) {
2219                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2220                 cmp = compare_blob(&clp->cl_name, name);
2221                 if (cmp > 0)
2222                         node = node->rb_left;
2223                 else if (cmp < 0)
2224                         node = node->rb_right;
2225                 else
2226                         return clp;
2227         }
2228         return NULL;
2229 }
2230
2231 static void
2232 add_to_unconfirmed(struct nfs4_client *clp)
2233 {
2234         unsigned int idhashval;
2235         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2236
2237         lockdep_assert_held(&nn->client_lock);
2238
2239         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2240         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2241         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2242         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2243         renew_client_locked(clp);
2244 }
2245
2246 static void
2247 move_to_confirmed(struct nfs4_client *clp)
2248 {
2249         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2250         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2251
2252         lockdep_assert_held(&nn->client_lock);
2253
2254         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2255         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2256         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2257         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2258         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2259         renew_client_locked(clp);
2260 }
2261
2262 static struct nfs4_client *
2263 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2264 {
2265         struct nfs4_client *clp;
2266         unsigned int idhashval = clientid_hashval(clid->cl_id);
2267
2268         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2269                 if (same_clid(&clp->cl_clientid, clid)) {
2270                         if ((bool)clp->cl_minorversion != sessions)
2271                                 return NULL;
2272                         renew_client_locked(clp);
2273                         return clp;
2274                 }
2275         }
2276         return NULL;
2277 }
2278
2279 static struct nfs4_client *
2280 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2281 {
2282         struct list_head *tbl = nn->conf_id_hashtbl;
2283
2284         lockdep_assert_held(&nn->client_lock);
2285         return find_client_in_id_table(tbl, clid, sessions);
2286 }
2287
2288 static struct nfs4_client *
2289 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2290 {
2291         struct list_head *tbl = nn->unconf_id_hashtbl;
2292
2293         lockdep_assert_held(&nn->client_lock);
2294         return find_client_in_id_table(tbl, clid, sessions);
2295 }
2296
2297 static bool clp_used_exchangeid(struct nfs4_client *clp)
2298 {
2299         return clp->cl_exchange_flags != 0;
2300 }
2301
2302 static struct nfs4_client *
2303 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2304 {
2305         lockdep_assert_held(&nn->client_lock);
2306         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2307 }
2308
2309 static struct nfs4_client *
2310 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2311 {
2312         lockdep_assert_held(&nn->client_lock);
2313         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2314 }
2315
2316 static void
2317 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2318 {
2319         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2320         struct sockaddr *sa = svc_addr(rqstp);
2321         u32 scopeid = rpc_get_scope_id(sa);
2322         unsigned short expected_family;
2323
2324         /* Currently, we only support tcp and tcp6 for the callback channel */
2325         if (se->se_callback_netid_len == 3 &&
2326             !memcmp(se->se_callback_netid_val, "tcp", 3))
2327                 expected_family = AF_INET;
2328         else if (se->se_callback_netid_len == 4 &&
2329                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2330                 expected_family = AF_INET6;
2331         else
2332                 goto out_err;
2333
2334         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2335                                             se->se_callback_addr_len,
2336                                             (struct sockaddr *)&conn->cb_addr,
2337                                             sizeof(conn->cb_addr));
2338
2339         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2340                 goto out_err;
2341
2342         if (conn->cb_addr.ss_family == AF_INET6)
2343                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2344
2345         conn->cb_prog = se->se_callback_prog;
2346         conn->cb_ident = se->se_callback_ident;
2347         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2348         return;
2349 out_err:
2350         conn->cb_addr.ss_family = AF_UNSPEC;
2351         conn->cb_addrlen = 0;
2352         dprintk("NFSD: this client (clientid %08x/%08x) "
2353                 "will not receive delegations\n",
2354                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2355
2356         return;
2357 }
2358
2359 /*
2360  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2361  */
2362 static void
2363 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2364 {
2365         struct xdr_buf *buf = resp->xdr.buf;
2366         struct nfsd4_slot *slot = resp->cstate.slot;
2367         unsigned int base;
2368
2369         dprintk("--> %s slot %p\n", __func__, slot);
2370
2371         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2372         slot->sl_opcnt = resp->opcnt;
2373         slot->sl_status = resp->cstate.status;
2374         free_svc_cred(&slot->sl_cred);
2375         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2376
2377         if (!nfsd4_cache_this(resp)) {
2378                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2379                 return;
2380         }
2381         slot->sl_flags |= NFSD4_SLOT_CACHED;
2382
2383         base = resp->cstate.data_offset;
2384         slot->sl_datalen = buf->len - base;
2385         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2386                 WARN(1, "%s: sessions DRC could not cache compound\n",
2387                      __func__);
2388         return;
2389 }
2390
2391 /*
2392  * Encode the replay sequence operation from the slot values.
2393  * If cachethis is FALSE encode the uncached rep error on the next
2394  * operation which sets resp->p and increments resp->opcnt for
2395  * nfs4svc_encode_compoundres.
2396  *
2397  */
2398 static __be32
2399 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2400                           struct nfsd4_compoundres *resp)
2401 {
2402         struct nfsd4_op *op;
2403         struct nfsd4_slot *slot = resp->cstate.slot;
2404
2405         /* Encode the replayed sequence operation */
2406         op = &args->ops[resp->opcnt - 1];
2407         nfsd4_encode_operation(resp, op);
2408
2409         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2410                 return op->status;
2411         if (args->opcnt == 1) {
2412                 /*
2413                  * The original operation wasn't a solo sequence--we
2414                  * always cache those--so this retry must not match the
2415                  * original:
2416                  */
2417                 op->status = nfserr_seq_false_retry;
2418         } else {
2419                 op = &args->ops[resp->opcnt++];
2420                 op->status = nfserr_retry_uncached_rep;
2421                 nfsd4_encode_operation(resp, op);
2422         }
2423         return op->status;
2424 }
2425
2426 /*
2427  * The sequence operation is not cached because we can use the slot and
2428  * session values.
2429  */
2430 static __be32
2431 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2432                          struct nfsd4_sequence *seq)
2433 {
2434         struct nfsd4_slot *slot = resp->cstate.slot;
2435         struct xdr_stream *xdr = &resp->xdr;
2436         __be32 *p;
2437         __be32 status;
2438
2439         dprintk("--> %s slot %p\n", __func__, slot);
2440
2441         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2442         if (status)
2443                 return status;
2444
2445         p = xdr_reserve_space(xdr, slot->sl_datalen);
2446         if (!p) {
2447                 WARN_ON_ONCE(1);
2448                 return nfserr_serverfault;
2449         }
2450         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2451         xdr_commit_encode(xdr);
2452
2453         resp->opcnt = slot->sl_opcnt;
2454         return slot->sl_status;
2455 }
2456
2457 /*
2458  * Set the exchange_id flags returned by the server.
2459  */
2460 static void
2461 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2462 {
2463 #ifdef CONFIG_NFSD_PNFS
2464         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2465 #else
2466         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2467 #endif
2468
2469         /* Referrals are supported, Migration is not. */
2470         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2471
2472         /* set the wire flags to return to client. */
2473         clid->flags = new->cl_exchange_flags;
2474 }
2475
2476 static bool client_has_openowners(struct nfs4_client *clp)
2477 {
2478         struct nfs4_openowner *oo;
2479
2480         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2481                 if (!list_empty(&oo->oo_owner.so_stateids))
2482                         return true;
2483         }
2484         return false;
2485 }
2486
2487 static bool client_has_state(struct nfs4_client *clp)
2488 {
2489         return client_has_openowners(clp)
2490 #ifdef CONFIG_NFSD_PNFS
2491                 || !list_empty(&clp->cl_lo_states)
2492 #endif
2493                 || !list_empty(&clp->cl_delegations)
2494                 || !list_empty(&clp->cl_sessions);
2495 }
2496
2497 __be32
2498 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2499                 union nfsd4_op_u *u)
2500 {
2501         struct nfsd4_exchange_id *exid = &u->exchange_id;
2502         struct nfs4_client *conf, *new;
2503         struct nfs4_client *unconf = NULL;
2504         __be32 status;
2505         char                    addr_str[INET6_ADDRSTRLEN];
2506         nfs4_verifier           verf = exid->verifier;
2507         struct sockaddr         *sa = svc_addr(rqstp);
2508         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2509         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2510
2511         rpc_ntop(sa, addr_str, sizeof(addr_str));
2512         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2513                 "ip_addr=%s flags %x, spa_how %d\n",
2514                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2515                 addr_str, exid->flags, exid->spa_how);
2516
2517         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2518                 return nfserr_inval;
2519
2520         new = create_client(exid->clname, rqstp, &verf);
2521         if (new == NULL)
2522                 return nfserr_jukebox;
2523
2524         switch (exid->spa_how) {
2525         case SP4_MACH_CRED:
2526                 exid->spo_must_enforce[0] = 0;
2527                 exid->spo_must_enforce[1] = (
2528                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2529                         1 << (OP_EXCHANGE_ID - 32) |
2530                         1 << (OP_CREATE_SESSION - 32) |
2531                         1 << (OP_DESTROY_SESSION - 32) |
2532                         1 << (OP_DESTROY_CLIENTID - 32));
2533
2534                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2535                                         1 << (OP_OPEN_DOWNGRADE) |
2536                                         1 << (OP_LOCKU) |
2537                                         1 << (OP_DELEGRETURN));
2538
2539                 exid->spo_must_allow[1] &= (
2540                                         1 << (OP_TEST_STATEID - 32) |
2541                                         1 << (OP_FREE_STATEID - 32));
2542                 if (!svc_rqst_integrity_protected(rqstp)) {
2543                         status = nfserr_inval;
2544                         goto out_nolock;
2545                 }
2546                 /*
2547                  * Sometimes userspace doesn't give us a principal.
2548                  * Which is a bug, really.  Anyway, we can't enforce
2549                  * MACH_CRED in that case, better to give up now:
2550                  */
2551                 if (!new->cl_cred.cr_principal &&
2552                                         !new->cl_cred.cr_raw_principal) {
2553                         status = nfserr_serverfault;
2554                         goto out_nolock;
2555                 }
2556                 new->cl_mach_cred = true;
2557         case SP4_NONE:
2558                 break;
2559         default:                                /* checked by xdr code */
2560                 WARN_ON_ONCE(1);
2561         case SP4_SSV:
2562                 status = nfserr_encr_alg_unsupp;
2563                 goto out_nolock;
2564         }
2565
2566         /* Cases below refer to rfc 5661 section 18.35.4: */
2567         spin_lock(&nn->client_lock);
2568         conf = find_confirmed_client_by_name(&exid->clname, nn);
2569         if (conf) {
2570                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2571                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2572
2573                 if (update) {
2574                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2575                                 status = nfserr_inval;
2576                                 goto out;
2577                         }
2578                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2579                                 status = nfserr_wrong_cred;
2580                                 goto out;
2581                         }
2582                         if (!creds_match) { /* case 9 */
2583                                 status = nfserr_perm;
2584                                 goto out;
2585                         }
2586                         if (!verfs_match) { /* case 8 */
2587                                 status = nfserr_not_same;
2588                                 goto out;
2589                         }
2590                         /* case 6 */
2591                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2592                         goto out_copy;
2593                 }
2594                 if (!creds_match) { /* case 3 */
2595                         if (client_has_state(conf)) {
2596                                 status = nfserr_clid_inuse;
2597                                 goto out;
2598                         }
2599                         goto out_new;
2600                 }
2601                 if (verfs_match) { /* case 2 */
2602                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2603                         goto out_copy;
2604                 }
2605                 /* case 5, client reboot */
2606                 conf = NULL;
2607                 goto out_new;
2608         }
2609
2610         if (update) { /* case 7 */
2611                 status = nfserr_noent;
2612                 goto out;
2613         }
2614
2615         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2616         if (unconf) /* case 4, possible retry or client restart */
2617                 unhash_client_locked(unconf);
2618
2619         /* case 1 (normal case) */
2620 out_new:
2621         if (conf) {
2622                 status = mark_client_expired_locked(conf);
2623                 if (status)
2624                         goto out;
2625         }
2626         new->cl_minorversion = cstate->minorversion;
2627         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2628         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2629
2630         gen_clid(new, nn);
2631         add_to_unconfirmed(new);
2632         swap(new, conf);
2633 out_copy:
2634         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2635         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2636
2637         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2638         nfsd4_set_ex_flags(conf, exid);
2639
2640         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2641                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2642         status = nfs_ok;
2643
2644 out:
2645         spin_unlock(&nn->client_lock);
2646 out_nolock:
2647         if (new)
2648                 expire_client(new);
2649         if (unconf)
2650                 expire_client(unconf);
2651         return status;
2652 }
2653
2654 static __be32
2655 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2656 {
2657         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2658                 slot_seqid);
2659
2660         /* The slot is in use, and no response has been sent. */
2661         if (slot_inuse) {
2662                 if (seqid == slot_seqid)
2663                         return nfserr_jukebox;
2664                 else
2665                         return nfserr_seq_misordered;
2666         }
2667         /* Note unsigned 32-bit arithmetic handles wraparound: */
2668         if (likely(seqid == slot_seqid + 1))
2669                 return nfs_ok;
2670         if (seqid == slot_seqid)
2671                 return nfserr_replay_cache;
2672         return nfserr_seq_misordered;
2673 }
2674
2675 /*
2676  * Cache the create session result into the create session single DRC
2677  * slot cache by saving the xdr structure. sl_seqid has been set.
2678  * Do this for solo or embedded create session operations.
2679  */
2680 static void
2681 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2682                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2683 {
2684         slot->sl_status = nfserr;
2685         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2686 }
2687
2688 static __be32
2689 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2690                             struct nfsd4_clid_slot *slot)
2691 {
2692         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2693         return slot->sl_status;
2694 }
2695
2696 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2697                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2698                         1 +     /* MIN tag is length with zero, only length */ \
2699                         3 +     /* version, opcount, opcode */ \
2700                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2701                                 /* seqid, slotID, slotID, cache */ \
2702                         4 ) * sizeof(__be32))
2703
2704 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2705                         2 +     /* verifier: AUTH_NULL, length 0 */\
2706                         1 +     /* status */ \
2707                         1 +     /* MIN tag is length with zero, only length */ \
2708                         3 +     /* opcount, opcode, opstatus*/ \
2709                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2710                                 /* seqid, slotID, slotID, slotID, status */ \
2711                         5 ) * sizeof(__be32))
2712
2713 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2714 {
2715         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2716
2717         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2718                 return nfserr_toosmall;
2719         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2720                 return nfserr_toosmall;
2721         ca->headerpadsz = 0;
2722         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2723         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2724         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2725         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2726                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2727         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2728         /*
2729          * Note decreasing slot size below client's request may make it
2730          * difficult for client to function correctly, whereas
2731          * decreasing the number of slots will (just?) affect
2732          * performance.  When short on memory we therefore prefer to
2733          * decrease number of slots instead of their size.  Clients that
2734          * request larger slots than they need will get poor results:
2735          */
2736         ca->maxreqs = nfsd4_get_drc_mem(ca);
2737         if (!ca->maxreqs)
2738                 return nfserr_jukebox;
2739
2740         return nfs_ok;
2741 }
2742
2743 /*
2744  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2745  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2746  */
2747 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2748         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2749
2750 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2751         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2752
2753 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2754                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2755 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2756                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2757                                  sizeof(__be32))
2758
2759 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2760 {
2761         ca->headerpadsz = 0;
2762
2763         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2764                 return nfserr_toosmall;
2765         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2766                 return nfserr_toosmall;
2767         ca->maxresp_cached = 0;
2768         if (ca->maxops < 2)
2769                 return nfserr_toosmall;
2770
2771         return nfs_ok;
2772 }
2773
2774 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2775 {
2776         switch (cbs->flavor) {
2777         case RPC_AUTH_NULL:
2778         case RPC_AUTH_UNIX:
2779                 return nfs_ok;
2780         default:
2781                 /*
2782                  * GSS case: the spec doesn't allow us to return this
2783                  * error.  But it also doesn't allow us not to support
2784                  * GSS.
2785                  * I'd rather this fail hard than return some error the
2786                  * client might think it can already handle:
2787                  */
2788                 return nfserr_encr_alg_unsupp;
2789         }
2790 }
2791
2792 __be32
2793 nfsd4_create_session(struct svc_rqst *rqstp,
2794                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2795 {
2796         struct nfsd4_create_session *cr_ses = &u->create_session;
2797         struct sockaddr *sa = svc_addr(rqstp);
2798         struct nfs4_client *conf, *unconf;
2799         struct nfs4_client *old = NULL;
2800         struct nfsd4_session *new;
2801         struct nfsd4_conn *conn;
2802         struct nfsd4_clid_slot *cs_slot = NULL;
2803         __be32 status = 0;
2804         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2805
2806         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2807                 return nfserr_inval;
2808         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2809         if (status)
2810                 return status;
2811         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2812         if (status)
2813                 return status;
2814         status = check_backchannel_attrs(&cr_ses->back_channel);
2815         if (status)
2816                 goto out_release_drc_mem;
2817         status = nfserr_jukebox;
2818         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2819         if (!new)
2820                 goto out_release_drc_mem;
2821         conn = alloc_conn_from_crses(rqstp, cr_ses);
2822         if (!conn)
2823                 goto out_free_session;
2824
2825         spin_lock(&nn->client_lock);
2826         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2827         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2828         WARN_ON_ONCE(conf && unconf);
2829
2830         if (conf) {
2831                 status = nfserr_wrong_cred;
2832                 if (!nfsd4_mach_creds_match(conf, rqstp))
2833                         goto out_free_conn;
2834                 cs_slot = &conf->cl_cs_slot;
2835                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2836                 if (status) {
2837                         if (status == nfserr_replay_cache)
2838                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2839                         goto out_free_conn;
2840                 }
2841         } else if (unconf) {
2842                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2843                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2844                         status = nfserr_clid_inuse;
2845                         goto out_free_conn;
2846                 }
2847                 status = nfserr_wrong_cred;
2848                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2849                         goto out_free_conn;
2850                 cs_slot = &unconf->cl_cs_slot;
2851                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2852                 if (status) {
2853                         /* an unconfirmed replay returns misordered */
2854                         status = nfserr_seq_misordered;
2855                         goto out_free_conn;
2856                 }
2857                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2858                 if (old) {
2859                         status = mark_client_expired_locked(old);
2860                         if (status) {
2861                                 old = NULL;
2862                                 goto out_free_conn;
2863                         }
2864                 }
2865                 move_to_confirmed(unconf);
2866                 conf = unconf;
2867         } else {
2868                 status = nfserr_stale_clientid;
2869                 goto out_free_conn;
2870         }
2871         status = nfs_ok;
2872         /* Persistent sessions are not supported */
2873         cr_ses->flags &= ~SESSION4_PERSIST;
2874         /* Upshifting from TCP to RDMA is not supported */
2875         cr_ses->flags &= ~SESSION4_RDMA;
2876
2877         init_session(rqstp, new, conf, cr_ses);
2878         nfsd4_get_session_locked(new);
2879
2880         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2881                NFS4_MAX_SESSIONID_LEN);
2882         cs_slot->sl_seqid++;
2883         cr_ses->seqid = cs_slot->sl_seqid;
2884
2885         /* cache solo and embedded create sessions under the client_lock */
2886         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2887         spin_unlock(&nn->client_lock);
2888         /* init connection and backchannel */
2889         nfsd4_init_conn(rqstp, conn, new);
2890         nfsd4_put_session(new);
2891         if (old)
2892                 expire_client(old);
2893         return status;
2894 out_free_conn:
2895         spin_unlock(&nn->client_lock);
2896         free_conn(conn);
2897         if (old)
2898                 expire_client(old);
2899 out_free_session:
2900         __free_session(new);
2901 out_release_drc_mem:
2902         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2903         return status;
2904 }
2905
2906 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2907 {
2908         switch (*dir) {
2909         case NFS4_CDFC4_FORE:
2910         case NFS4_CDFC4_BACK:
2911                 return nfs_ok;
2912         case NFS4_CDFC4_FORE_OR_BOTH:
2913         case NFS4_CDFC4_BACK_OR_BOTH:
2914                 *dir = NFS4_CDFC4_BOTH;
2915                 return nfs_ok;
2916         };
2917         return nfserr_inval;
2918 }
2919
2920 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2921                 struct nfsd4_compound_state *cstate,
2922                 union nfsd4_op_u *u)
2923 {
2924         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2925         struct nfsd4_session *session = cstate->session;
2926         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2927         __be32 status;
2928
2929         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2930         if (status)
2931                 return status;
2932         spin_lock(&nn->client_lock);
2933         session->se_cb_prog = bc->bc_cb_program;
2934         session->se_cb_sec = bc->bc_cb_sec;
2935         spin_unlock(&nn->client_lock);
2936
2937         nfsd4_probe_callback(session->se_client);
2938
2939         return nfs_ok;
2940 }
2941
2942 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2943                      struct nfsd4_compound_state *cstate,
2944                      union nfsd4_op_u *u)
2945 {
2946         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2947         __be32 status;
2948         struct nfsd4_conn *conn;
2949         struct nfsd4_session *session;
2950         struct net *net = SVC_NET(rqstp);
2951         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2952
2953         if (!nfsd4_last_compound_op(rqstp))
2954                 return nfserr_not_only_op;
2955         spin_lock(&nn->client_lock);
2956         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2957         spin_unlock(&nn->client_lock);
2958         if (!session)
2959                 goto out_no_session;
2960         status = nfserr_wrong_cred;
2961         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2962                 goto out;
2963         status = nfsd4_map_bcts_dir(&bcts->dir);
2964         if (status)
2965                 goto out;
2966         conn = alloc_conn(rqstp, bcts->dir);
2967         status = nfserr_jukebox;
2968         if (!conn)
2969                 goto out;
2970         nfsd4_init_conn(rqstp, conn, session);
2971         status = nfs_ok;
2972 out:
2973         nfsd4_put_session(session);
2974 out_no_session:
2975         return status;
2976 }
2977
2978 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
2979 {
2980         if (!cstate->session)
2981                 return false;
2982         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
2983 }
2984
2985 __be32
2986 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2987                 union nfsd4_op_u *u)
2988 {
2989         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
2990         struct nfsd4_session *ses;
2991         __be32 status;
2992         int ref_held_by_me = 0;
2993         struct net *net = SVC_NET(r);
2994         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2995
2996         status = nfserr_not_only_op;
2997         if (nfsd4_compound_in_session(cstate, sessionid)) {
2998                 if (!nfsd4_last_compound_op(r))
2999                         goto out;
3000                 ref_held_by_me++;
3001         }
3002         dump_sessionid(__func__, sessionid);
3003         spin_lock(&nn->client_lock);
3004         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3005         if (!ses)
3006                 goto out_client_lock;
3007         status = nfserr_wrong_cred;
3008         if (!nfsd4_mach_creds_match(ses->se_client, r))
3009                 goto out_put_session;
3010         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3011         if (status)
3012                 goto out_put_session;
3013         unhash_session(ses);
3014         spin_unlock(&nn->client_lock);
3015
3016         nfsd4_probe_callback_sync(ses->se_client);
3017
3018         spin_lock(&nn->client_lock);
3019         status = nfs_ok;
3020 out_put_session:
3021         nfsd4_put_session_locked(ses);
3022 out_client_lock:
3023         spin_unlock(&nn->client_lock);
3024 out:
3025         return status;
3026 }
3027
3028 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3029 {
3030         struct nfsd4_conn *c;
3031
3032         list_for_each_entry(c, &s->se_conns, cn_persession) {
3033                 if (c->cn_xprt == xpt) {
3034                         return c;
3035                 }
3036         }
3037         return NULL;
3038 }
3039
3040 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3041 {
3042         struct nfs4_client *clp = ses->se_client;
3043         struct nfsd4_conn *c;
3044         __be32 status = nfs_ok;
3045         int ret;
3046
3047         spin_lock(&clp->cl_lock);
3048         c = __nfsd4_find_conn(new->cn_xprt, ses);
3049         if (c)
3050                 goto out_free;
3051         status = nfserr_conn_not_bound_to_session;
3052         if (clp->cl_mach_cred)
3053                 goto out_free;
3054         __nfsd4_hash_conn(new, ses);
3055         spin_unlock(&clp->cl_lock);
3056         ret = nfsd4_register_conn(new);
3057         if (ret)
3058                 /* oops; xprt is already down: */
3059                 nfsd4_conn_lost(&new->cn_xpt_user);
3060         return nfs_ok;
3061 out_free:
3062         spin_unlock(&clp->cl_lock);
3063         free_conn(new);
3064         return status;
3065 }
3066
3067 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3068 {
3069         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3070
3071         return args->opcnt > session->se_fchannel.maxops;
3072 }
3073
3074 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3075                                   struct nfsd4_session *session)
3076 {
3077         struct xdr_buf *xb = &rqstp->rq_arg;
3078
3079         return xb->len > session->se_fchannel.maxreq_sz;
3080 }
3081
3082 static bool replay_matches_cache(struct svc_rqst *rqstp,
3083                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3084 {
3085         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3086
3087         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3088             (bool)seq->cachethis)
3089                 return false;
3090         /*
3091          * If there's an error then the reply can have fewer ops than
3092          * the call.
3093          */
3094         if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3095                 return false;
3096         /*
3097          * But if we cached a reply with *more* ops than the call you're
3098          * sending us now, then this new call is clearly not really a
3099          * replay of the old one:
3100          */
3101         if (slot->sl_opcnt > argp->opcnt)
3102                 return false;
3103         /* This is the only check explicitly called by spec: */
3104         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3105                 return false;
3106         /*
3107          * There may be more comparisons we could actually do, but the
3108          * spec doesn't require us to catch every case where the calls
3109          * don't match (that would require caching the call as well as
3110          * the reply), so we don't bother.
3111          */
3112         return true;
3113 }
3114
3115 __be32
3116 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3117                 union nfsd4_op_u *u)
3118 {
3119         struct nfsd4_sequence *seq = &u->sequence;
3120         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3121         struct xdr_stream *xdr = &resp->xdr;
3122         struct nfsd4_session *session;
3123         struct nfs4_client *clp;
3124         struct nfsd4_slot *slot;
3125         struct nfsd4_conn *conn;
3126         __be32 status;
3127         int buflen;
3128         struct net *net = SVC_NET(rqstp);
3129         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3130
3131         if (resp->opcnt != 1)
3132                 return nfserr_sequence_pos;
3133
3134         /*
3135          * Will be either used or freed by nfsd4_sequence_check_conn
3136          * below.
3137          */
3138         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3139         if (!conn)
3140                 return nfserr_jukebox;
3141
3142         spin_lock(&nn->client_lock);
3143         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3144         if (!session)
3145                 goto out_no_session;
3146         clp = session->se_client;
3147
3148         status = nfserr_too_many_ops;
3149         if (nfsd4_session_too_many_ops(rqstp, session))
3150                 goto out_put_session;
3151
3152         status = nfserr_req_too_big;
3153         if (nfsd4_request_too_big(rqstp, session))
3154                 goto out_put_session;
3155
3156         status = nfserr_badslot;
3157         if (seq->slotid >= session->se_fchannel.maxreqs)
3158                 goto out_put_session;
3159
3160         slot = session->se_slots[seq->slotid];
3161         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3162
3163         /* We do not negotiate the number of slots yet, so set the
3164          * maxslots to the session maxreqs which is used to encode
3165          * sr_highest_slotid and the sr_target_slot id to maxslots */
3166         seq->maxslots = session->se_fchannel.maxreqs;
3167
3168         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3169                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3170         if (status == nfserr_replay_cache) {
3171                 status = nfserr_seq_misordered;
3172                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3173                         goto out_put_session;
3174                 status = nfserr_seq_false_retry;
3175                 if (!replay_matches_cache(rqstp, seq, slot))
3176                         goto out_put_session;
3177                 cstate->slot = slot;
3178                 cstate->session = session;
3179                 cstate->clp = clp;
3180                 /* Return the cached reply status and set cstate->status
3181                  * for nfsd4_proc_compound processing */
3182                 status = nfsd4_replay_cache_entry(resp, seq);
3183                 cstate->status = nfserr_replay_cache;
3184                 goto out;
3185         }
3186         if (status)
3187                 goto out_put_session;
3188
3189         status = nfsd4_sequence_check_conn(conn, session);
3190         conn = NULL;
3191         if (status)
3192                 goto out_put_session;
3193
3194         buflen = (seq->cachethis) ?
3195                         session->se_fchannel.maxresp_cached :
3196                         session->se_fchannel.maxresp_sz;
3197         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3198                                     nfserr_rep_too_big;
3199         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3200                 goto out_put_session;
3201         svc_reserve(rqstp, buflen);
3202
3203         status = nfs_ok;
3204         /* Success! bump slot seqid */
3205         slot->sl_seqid = seq->seqid;
3206         slot->sl_flags |= NFSD4_SLOT_INUSE;
3207         if (seq->cachethis)
3208                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3209         else
3210                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3211
3212         cstate->slot = slot;
3213         cstate->session = session;
3214         cstate->clp = clp;
3215
3216 out:
3217         switch (clp->cl_cb_state) {
3218         case NFSD4_CB_DOWN:
3219                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3220                 break;
3221         case NFSD4_CB_FAULT:
3222                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3223                 break;
3224         default:
3225                 seq->status_flags = 0;
3226         }
3227         if (!list_empty(&clp->cl_revoked))
3228                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3229 out_no_session:
3230         if (conn)
3231                 free_conn(conn);
3232         spin_unlock(&nn->client_lock);
3233         return status;
3234 out_put_session:
3235         nfsd4_put_session_locked(session);
3236         goto out_no_session;
3237 }
3238
3239 void
3240 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3241 {
3242         struct nfsd4_compound_state *cs = &resp->cstate;
3243
3244         if (nfsd4_has_session(cs)) {
3245                 if (cs->status != nfserr_replay_cache) {
3246                         nfsd4_store_cache_entry(resp);
3247                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3248                 }
3249                 /* Drop session reference that was taken in nfsd4_sequence() */
3250                 nfsd4_put_session(cs->session);
3251         } else if (cs->clp)
3252                 put_client_renew(cs->clp);
3253 }
3254
3255 __be32
3256 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3257                 struct nfsd4_compound_state *cstate,
3258                 union nfsd4_op_u *u)
3259 {
3260         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3261         struct nfs4_client *conf, *unconf;
3262         struct nfs4_client *clp = NULL;
3263         __be32 status = 0;
3264         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3265
3266         spin_lock(&nn->client_lock);
3267         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3268         conf = find_confirmed_client(&dc->clientid, true, nn);
3269         WARN_ON_ONCE(conf && unconf);
3270
3271         if (conf) {
3272                 if (client_has_state(conf)) {
3273                         status = nfserr_clientid_busy;
3274                         goto out;
3275                 }
3276                 status = mark_client_expired_locked(conf);
3277                 if (status)
3278                         goto out;
3279                 clp = conf;
3280         } else if (unconf)
3281                 clp = unconf;
3282         else {
3283                 status = nfserr_stale_clientid;
3284                 goto out;
3285         }
3286         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3287                 clp = NULL;
3288                 status = nfserr_wrong_cred;
3289                 goto out;
3290         }
3291         unhash_client_locked(clp);
3292 out:
3293         spin_unlock(&nn->client_lock);
3294         if (clp)
3295                 expire_client(clp);
3296         return status;
3297 }
3298
3299 __be32
3300 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3301                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3302 {
3303         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3304         __be32 status = 0;
3305
3306         if (rc->rca_one_fs) {
3307                 if (!cstate->current_fh.fh_dentry)
3308                         return nfserr_nofilehandle;
3309                 /*
3310                  * We don't take advantage of the rca_one_fs case.
3311                  * That's OK, it's optional, we can safely ignore it.
3312                  */
3313                 return nfs_ok;
3314         }
3315
3316         status = nfserr_complete_already;
3317         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3318                              &cstate->session->se_client->cl_flags))
3319                 goto out;
3320
3321         status = nfserr_stale_clientid;
3322         if (is_client_expired(cstate->session->se_client))
3323                 /*
3324                  * The following error isn't really legal.
3325                  * But we only get here if the client just explicitly
3326                  * destroyed the client.  Surely it no longer cares what
3327                  * error it gets back on an operation for the dead
3328                  * client.
3329                  */
3330                 goto out;
3331
3332         status = nfs_ok;
3333         nfsd4_client_record_create(cstate->session->se_client);
3334 out:
3335         return status;
3336 }
3337
3338 __be32
3339 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3340                   union nfsd4_op_u *u)
3341 {
3342         struct nfsd4_setclientid *setclid = &u->setclientid;
3343         struct xdr_netobj       clname = setclid->se_name;
3344         nfs4_verifier           clverifier = setclid->se_verf;
3345         struct nfs4_client      *conf, *new;
3346         struct nfs4_client      *unconf = NULL;
3347         __be32                  status;
3348         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3349
3350         new = create_client(clname, rqstp, &clverifier);
3351         if (new == NULL)
3352                 return nfserr_jukebox;
3353         /* Cases below refer to rfc 3530 section 14.2.33: */
3354         spin_lock(&nn->client_lock);
3355         conf = find_confirmed_client_by_name(&clname, nn);
3356         if (conf && client_has_state(conf)) {
3357                 /* case 0: */
3358                 status = nfserr_clid_inuse;
3359                 if (clp_used_exchangeid(conf))
3360                         goto out;
3361                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3362                         char addr_str[INET6_ADDRSTRLEN];
3363                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3364                                  sizeof(addr_str));
3365                         dprintk("NFSD: setclientid: string in use by client "
3366                                 "at %s\n", addr_str);
3367                         goto out;
3368                 }
3369         }
3370         unconf = find_unconfirmed_client_by_name(&clname, nn);
3371         if (unconf)
3372                 unhash_client_locked(unconf);
3373         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3374                 /* case 1: probable callback update */
3375                 copy_clid(new, conf);
3376                 gen_confirm(new, nn);
3377         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3378                 gen_clid(new, nn);
3379         new->cl_minorversion = 0;
3380         gen_callback(new, setclid, rqstp);
3381         add_to_unconfirmed(new);
3382         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3383         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3384         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3385         new = NULL;
3386         status = nfs_ok;
3387 out:
3388         spin_unlock(&nn->client_lock);
3389         if (new)
3390                 free_client(new);
3391         if (unconf)
3392                 expire_client(unconf);
3393         return status;
3394 }
3395
3396
3397 __be32
3398 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3399                         struct nfsd4_compound_state *cstate,
3400                         union nfsd4_op_u *u)
3401 {
3402         struct nfsd4_setclientid_confirm *setclientid_confirm =
3403                         &u->setclientid_confirm;
3404         struct nfs4_client *conf, *unconf;
3405         struct nfs4_client *old = NULL;
3406         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3407         clientid_t * clid = &setclientid_confirm->sc_clientid;
3408         __be32 status;
3409         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3410
3411         if (STALE_CLIENTID(clid, nn))
3412                 return nfserr_stale_clientid;
3413
3414         spin_lock(&nn->client_lock);
3415         conf = find_confirmed_client(clid, false, nn);
3416         unconf = find_unconfirmed_client(clid, false, nn);
3417         /*
3418          * We try hard to give out unique clientid's, so if we get an
3419          * attempt to confirm the same clientid with a different cred,
3420          * the client may be buggy; this should never happen.
3421          *
3422          * Nevertheless, RFC 7530 recommends INUSE for this case:
3423          */
3424         status = nfserr_clid_inuse;
3425         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3426                 goto out;
3427         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3428                 goto out;
3429         /* cases below refer to rfc 3530 section 14.2.34: */
3430         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3431                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3432                         /* case 2: probable retransmit */
3433                         status = nfs_ok;
3434                 } else /* case 4: client hasn't noticed we rebooted yet? */
3435                         status = nfserr_stale_clientid;
3436                 goto out;
3437         }
3438         status = nfs_ok;
3439         if (conf) { /* case 1: callback update */
3440                 old = unconf;
3441                 unhash_client_locked(old);
3442                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3443         } else { /* case 3: normal case; new or rebooted client */
3444                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3445                 if (old) {
3446                         status = nfserr_clid_inuse;
3447                         if (client_has_state(old)
3448                                         && !same_creds(&unconf->cl_cred,
3449                                                         &old->cl_cred)) {
3450                                 old = NULL;
3451                                 goto out;
3452                         }
3453                         status = mark_client_expired_locked(old);
3454                         if (status) {
3455                                 old = NULL;
3456                                 goto out;
3457                         }
3458                 }
3459                 move_to_confirmed(unconf);
3460                 conf = unconf;
3461         }
3462         get_client_locked(conf);
3463         spin_unlock(&nn->client_lock);
3464         nfsd4_probe_callback(conf);
3465         spin_lock(&nn->client_lock);
3466         put_client_renew_locked(conf);
3467 out:
3468         spin_unlock(&nn->client_lock);
3469         if (old)
3470                 expire_client(old);
3471         return status;
3472 }
3473
3474 static struct nfs4_file *nfsd4_alloc_file(void)
3475 {
3476         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3477 }
3478
3479 /* OPEN Share state helper functions */
3480 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3481                                 struct nfs4_file *fp)
3482 {
3483         lockdep_assert_held(&state_lock);
3484
3485         refcount_set(&fp->fi_ref, 1);
3486         spin_lock_init(&fp->fi_lock);
3487         INIT_LIST_HEAD(&fp->fi_stateids);
3488         INIT_LIST_HEAD(&fp->fi_delegations);
3489         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3490         fh_copy_shallow(&fp->fi_fhandle, fh);
3491         fp->fi_deleg_file = NULL;
3492         fp->fi_had_conflict = false;
3493         fp->fi_share_deny = 0;
3494         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3495         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3496 #ifdef CONFIG_NFSD_PNFS
3497         INIT_LIST_HEAD(&fp->fi_lo_states);
3498         atomic_set(&fp->fi_lo_recalls, 0);
3499 #endif
3500         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3501 }
3502
3503 void
3504 nfsd4_free_slabs(void)
3505 {
3506         kmem_cache_destroy(client_slab);
3507         kmem_cache_destroy(openowner_slab);
3508         kmem_cache_destroy(lockowner_slab);
3509         kmem_cache_destroy(file_slab);
3510         kmem_cache_destroy(stateid_slab);
3511         kmem_cache_destroy(deleg_slab);
3512         kmem_cache_destroy(odstate_slab);
3513 }
3514
3515 int
3516 nfsd4_init_slabs(void)
3517 {
3518         client_slab = kmem_cache_create("nfsd4_clients",
3519                         sizeof(struct nfs4_client), 0, 0, NULL);
3520         if (client_slab == NULL)
3521                 goto out;
3522         openowner_slab = kmem_cache_create("nfsd4_openowners",
3523                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3524         if (openowner_slab == NULL)
3525                 goto out_free_client_slab;
3526         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3527                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3528         if (lockowner_slab == NULL)
3529                 goto out_free_openowner_slab;
3530         file_slab = kmem_cache_create("nfsd4_files",
3531                         sizeof(struct nfs4_file), 0, 0, NULL);
3532         if (file_slab == NULL)
3533                 goto out_free_lockowner_slab;
3534         stateid_slab = kmem_cache_create("nfsd4_stateids",
3535                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3536         if (stateid_slab == NULL)
3537                 goto out_free_file_slab;
3538         deleg_slab = kmem_cache_create("nfsd4_delegations",
3539                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3540         if (deleg_slab == NULL)
3541                 goto out_free_stateid_slab;
3542         odstate_slab = kmem_cache_create("nfsd4_odstate",
3543                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3544         if (odstate_slab == NULL)
3545                 goto out_free_deleg_slab;
3546         return 0;
3547
3548 out_free_deleg_slab:
3549         kmem_cache_destroy(deleg_slab);
3550 out_free_stateid_slab:
3551         kmem_cache_destroy(stateid_slab);
3552 out_free_file_slab:
3553         kmem_cache_destroy(file_slab);
3554 out_free_lockowner_slab:
3555         kmem_cache_destroy(lockowner_slab);
3556 out_free_openowner_slab:
3557         kmem_cache_destroy(openowner_slab);
3558 out_free_client_slab:
3559         kmem_cache_destroy(client_slab);
3560 out:
3561         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3562         return -ENOMEM;
3563 }
3564
3565 static void init_nfs4_replay(struct nfs4_replay *rp)
3566 {
3567         rp->rp_status = nfserr_serverfault;
3568         rp->rp_buflen = 0;
3569         rp->rp_buf = rp->rp_ibuf;
3570         mutex_init(&rp->rp_mutex);
3571 }
3572
3573 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3574                 struct nfs4_stateowner *so)
3575 {
3576         if (!nfsd4_has_session(cstate)) {
3577                 mutex_lock(&so->so_replay.rp_mutex);
3578                 cstate->replay_owner = nfs4_get_stateowner(so);
3579         }
3580 }
3581
3582 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3583 {
3584         struct nfs4_stateowner *so = cstate->replay_owner;
3585
3586         if (so != NULL) {
3587                 cstate->replay_owner = NULL;
3588                 mutex_unlock(&so->so_replay.rp_mutex);
3589                 nfs4_put_stateowner(so);
3590         }
3591 }
3592
3593 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3594 {
3595         struct nfs4_stateowner *sop;
3596
3597         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3598         if (!sop)
3599                 return NULL;
3600
3601         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3602         if (!sop->so_owner.data) {
3603                 kmem_cache_free(slab, sop);
3604                 return NULL;
3605         }
3606         sop->so_owner.len = owner->len;
3607
3608         INIT_LIST_HEAD(&sop->so_stateids);
3609         sop->so_client = clp;
3610         init_nfs4_replay(&sop->so_replay);
3611         atomic_set(&sop->so_count, 1);
3612         return sop;
3613 }
3614
3615 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3616 {
3617         lockdep_assert_held(&clp->cl_lock);
3618
3619         list_add(&oo->oo_owner.so_strhash,
3620                  &clp->cl_ownerstr_hashtbl[strhashval]);
3621         list_add(&oo->oo_perclient, &clp->cl_openowners);
3622 }
3623
3624 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3625 {
3626         unhash_openowner_locked(openowner(so));
3627 }
3628
3629 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3630 {
3631         struct nfs4_openowner *oo = openowner(so);
3632
3633         kmem_cache_free(openowner_slab, oo);
3634 }
3635
3636 static const struct nfs4_stateowner_operations openowner_ops = {
3637         .so_unhash =    nfs4_unhash_openowner,
3638         .so_free =      nfs4_free_openowner,
3639 };
3640
3641 static struct nfs4_ol_stateid *
3642 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3643 {
3644         struct nfs4_ol_stateid *local, *ret = NULL;
3645         struct nfs4_openowner *oo = open->op_openowner;
3646
3647         lockdep_assert_held(&fp->fi_lock);
3648
3649         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3650                 /* ignore lock owners */
3651                 if (local->st_stateowner->so_is_open_owner == 0)
3652                         continue;
3653                 if (local->st_stateowner != &oo->oo_owner)
3654                         continue;
3655                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3656                         ret = local;
3657                         refcount_inc(&ret->st_stid.sc_count);
3658                         break;
3659                 }
3660         }
3661         return ret;
3662 }
3663
3664 static __be32
3665 nfsd4_verify_open_stid(struct nfs4_stid *s)
3666 {
3667         __be32 ret = nfs_ok;
3668
3669         switch (s->sc_type) {
3670         default:
3671                 break;
3672         case 0:
3673         case NFS4_CLOSED_STID:
3674         case NFS4_CLOSED_DELEG_STID:
3675                 ret = nfserr_bad_stateid;
3676                 break;
3677         case NFS4_REVOKED_DELEG_STID:
3678                 ret = nfserr_deleg_revoked;
3679         }
3680         return ret;
3681 }
3682
3683 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3684 static __be32
3685 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3686 {
3687         __be32 ret;
3688
3689         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3690         ret = nfsd4_verify_open_stid(&stp->st_stid);
3691         if (ret != nfs_ok)
3692                 mutex_unlock(&stp->st_mutex);
3693         return ret;
3694 }
3695
3696 static struct nfs4_ol_stateid *
3697 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3698 {
3699         struct nfs4_ol_stateid *stp;
3700         for (;;) {
3701                 spin_lock(&fp->fi_lock);
3702                 stp = nfsd4_find_existing_open(fp, open);
3703                 spin_unlock(&fp->fi_lock);
3704                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3705                         break;
3706                 nfs4_put_stid(&stp->st_stid);
3707         }
3708         return stp;
3709 }
3710
3711 static struct nfs4_openowner *
3712 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3713                            struct nfsd4_compound_state *cstate)
3714 {
3715         struct nfs4_client *clp = cstate->clp;
3716         struct nfs4_openowner *oo, *ret;
3717
3718         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3719         if (!oo)
3720                 return NULL;
3721         oo->oo_owner.so_ops = &openowner_ops;
3722         oo->oo_owner.so_is_open_owner = 1;
3723         oo->oo_owner.so_seqid = open->op_seqid;
3724         oo->oo_flags = 0;
3725         if (nfsd4_has_session(cstate))
3726                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3727         oo->oo_time = 0;
3728         oo->oo_last_closed_stid = NULL;
3729         INIT_LIST_HEAD(&oo->oo_close_lru);
3730         spin_lock(&clp->cl_lock);
3731         ret = find_openstateowner_str_locked(strhashval, open, clp);
3732         if (ret == NULL) {
3733                 hash_openowner(oo, clp, strhashval);
3734                 ret = oo;
3735         } else
3736                 nfs4_free_stateowner(&oo->oo_owner);
3737
3738         spin_unlock(&clp->cl_lock);
3739         return ret;
3740 }
3741
3742 static struct nfs4_ol_stateid *
3743 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3744 {
3745
3746         struct nfs4_openowner *oo = open->op_openowner;
3747         struct nfs4_ol_stateid *retstp = NULL;
3748         struct nfs4_ol_stateid *stp;
3749
3750         stp = open->op_stp;
3751         /* We are moving these outside of the spinlocks to avoid the warnings */
3752         mutex_init(&stp->st_mutex);
3753         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3754
3755 retry:
3756         spin_lock(&oo->oo_owner.so_client->cl_lock);
3757         spin_lock(&fp->fi_lock);
3758
3759         retstp = nfsd4_find_existing_open(fp, open);
3760         if (retstp)
3761                 goto out_unlock;
3762
3763         open->op_stp = NULL;
3764         refcount_inc(&stp->st_stid.sc_count);
3765         stp->st_stid.sc_type = NFS4_OPEN_STID;
3766         INIT_LIST_HEAD(&stp->st_locks);
3767         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3768         get_nfs4_file(fp);
3769         stp->st_stid.sc_file = fp;
3770         stp->st_access_bmap = 0;
3771         stp->st_deny_bmap = 0;
3772         stp->st_openstp = NULL;
3773         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3774         list_add(&stp->st_perfile, &fp->fi_stateids);
3775
3776 out_unlock:
3777         spin_unlock(&fp->fi_lock);
3778         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3779         if (retstp) {
3780                 /* Handle races with CLOSE */
3781                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3782                         nfs4_put_stid(&retstp->st_stid);
3783                         goto retry;
3784                 }
3785                 /* To keep mutex tracking happy */
3786                 mutex_unlock(&stp->st_mutex);
3787                 stp = retstp;
3788         }
3789         return stp;
3790 }
3791
3792 /*
3793  * In the 4.0 case we need to keep the owners around a little while to handle
3794  * CLOSE replay. We still do need to release any file access that is held by
3795  * them before returning however.
3796  */
3797 static void
3798 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3799 {
3800         struct nfs4_ol_stateid *last;
3801         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3802         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3803                                                 nfsd_net_id);
3804
3805         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3806
3807         /*
3808          * We know that we hold one reference via nfsd4_close, and another
3809          * "persistent" reference for the client. If the refcount is higher
3810          * than 2, then there are still calls in progress that are using this
3811          * stateid. We can't put the sc_file reference until they are finished.
3812          * Wait for the refcount to drop to 2. Since it has been unhashed,
3813          * there should be no danger of the refcount going back up again at
3814          * this point.
3815          */
3816         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3817
3818         release_all_access(s);
3819         if (s->st_stid.sc_file) {
3820                 put_nfs4_file(s->st_stid.sc_file);
3821                 s->st_stid.sc_file = NULL;
3822         }
3823
3824         spin_lock(&nn->client_lock);
3825         last = oo->oo_last_closed_stid;
3826         oo->oo_last_closed_stid = s;
3827         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3828         oo->oo_time = get_seconds();
3829         spin_unlock(&nn->client_lock);
3830         if (last)
3831                 nfs4_put_stid(&last->st_stid);
3832 }
3833
3834 /* search file_hashtbl[] for file */
3835 static struct nfs4_file *
3836 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3837 {
3838         struct nfs4_file *fp;
3839
3840         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3841                 if (fh_match(&fp->fi_fhandle, fh)) {
3842                         if (refcount_inc_not_zero(&fp->fi_ref))
3843                                 return fp;
3844                 }
3845         }
3846         return NULL;
3847 }
3848
3849 struct nfs4_file *
3850 find_file(struct knfsd_fh *fh)
3851 {
3852         struct nfs4_file *fp;
3853         unsigned int hashval = file_hashval(fh);
3854
3855         rcu_read_lock();
3856         fp = find_file_locked(fh, hashval);
3857         rcu_read_unlock();
3858         return fp;
3859 }
3860
3861 static struct nfs4_file *
3862 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3863 {
3864         struct nfs4_file *fp;
3865         unsigned int hashval = file_hashval(fh);
3866
3867         rcu_read_lock();
3868         fp = find_file_locked(fh, hashval);
3869         rcu_read_unlock();
3870         if (fp)
3871                 return fp;
3872
3873         spin_lock(&state_lock);
3874         fp = find_file_locked(fh, hashval);
3875         if (likely(fp == NULL)) {
3876                 nfsd4_init_file(fh, hashval, new);
3877                 fp = new;
3878         }
3879         spin_unlock(&state_lock);
3880
3881         return fp;
3882 }
3883
3884 /*
3885  * Called to check deny when READ with all zero stateid or
3886  * WRITE with all zero or all one stateid
3887  */
3888 static __be32
3889 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3890 {
3891         struct nfs4_file *fp;
3892         __be32 ret = nfs_ok;
3893
3894         fp = find_file(&current_fh->fh_handle);
3895         if (!fp)
3896                 return ret;
3897         /* Check for conflicting share reservations */
3898         spin_lock(&fp->fi_lock);
3899         if (fp->fi_share_deny & deny_type)
3900                 ret = nfserr_locked;
3901         spin_unlock(&fp->fi_lock);
3902         put_nfs4_file(fp);
3903         return ret;
3904 }
3905
3906 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3907 {
3908         struct nfs4_delegation *dp = cb_to_delegation(cb);
3909         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3910                                           nfsd_net_id);
3911
3912         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3913
3914         /*
3915          * We can't do this in nfsd_break_deleg_cb because it is
3916          * already holding inode->i_lock.
3917          *
3918          * If the dl_time != 0, then we know that it has already been
3919          * queued for a lease break. Don't queue it again.
3920          */
3921         spin_lock(&state_lock);
3922         if (delegation_hashed(dp) && dp->dl_time == 0) {
3923                 dp->dl_time = get_seconds();
3924                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3925         }
3926         spin_unlock(&state_lock);
3927 }
3928
3929 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3930                 struct rpc_task *task)
3931 {
3932         struct nfs4_delegation *dp = cb_to_delegation(cb);
3933
3934         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3935                 return 1;
3936
3937         switch (task->tk_status) {
3938         case 0:
3939                 return 1;
3940         case -EBADHANDLE:
3941         case -NFS4ERR_BAD_STATEID:
3942                 /*
3943                  * Race: client probably got cb_recall before open reply
3944                  * granting delegation.
3945                  */
3946                 if (dp->dl_retries--) {
3947                         rpc_delay(task, 2 * HZ);
3948                         return 0;
3949                 }
3950                 /*FALLTHRU*/
3951         default:
3952                 return -1;
3953         }
3954 }
3955
3956 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3957 {
3958         struct nfs4_delegation *dp = cb_to_delegation(cb);
3959
3960         nfs4_put_stid(&dp->dl_stid);
3961 }
3962
3963 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3964         .prepare        = nfsd4_cb_recall_prepare,
3965         .done           = nfsd4_cb_recall_done,
3966         .release        = nfsd4_cb_recall_release,
3967 };
3968
3969 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3970 {
3971         /*
3972          * We're assuming the state code never drops its reference
3973          * without first removing the lease.  Since we're in this lease
3974          * callback (and since the lease code is serialized by the
3975          * i_lock) we know the server hasn't removed the lease yet, and
3976          * we know it's safe to take a reference.
3977          */
3978         refcount_inc(&dp->dl_stid.sc_count);
3979         nfsd4_run_cb(&dp->dl_recall);
3980 }
3981
3982 /* Called from break_lease() with i_lock held. */
3983 static bool
3984 nfsd_break_deleg_cb(struct file_lock *fl)
3985 {
3986         bool ret = false;
3987         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
3988         struct nfs4_file *fp = dp->dl_stid.sc_file;
3989
3990         /*
3991          * We don't want the locks code to timeout the lease for us;
3992          * we'll remove it ourself if a delegation isn't returned
3993          * in time:
3994          */
3995         fl->fl_break_time = 0;
3996
3997         spin_lock(&fp->fi_lock);
3998         fp->fi_had_conflict = true;
3999         nfsd_break_one_deleg(dp);
4000         spin_unlock(&fp->fi_lock);
4001         return ret;
4002 }
4003
4004 static int
4005 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4006                      struct list_head *dispose)
4007 {
4008         if (arg & F_UNLCK)
4009                 return lease_modify(onlist, arg, dispose);
4010         else
4011                 return -EAGAIN;
4012 }
4013
4014 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4015         .lm_break = nfsd_break_deleg_cb,
4016         .lm_change = nfsd_change_deleg_cb,
4017 };
4018
4019 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4020 {
4021         if (nfsd4_has_session(cstate))
4022                 return nfs_ok;
4023         if (seqid == so->so_seqid - 1)
4024                 return nfserr_replay_me;
4025         if (seqid == so->so_seqid)
4026                 return nfs_ok;
4027         return nfserr_bad_seqid;
4028 }
4029
4030 static __be32 lookup_clientid(clientid_t *clid,
4031                 struct nfsd4_compound_state *cstate,
4032                 struct nfsd_net *nn)
4033 {
4034         struct nfs4_client *found;
4035
4036         if (cstate->clp) {
4037                 found = cstate->clp;
4038                 if (!same_clid(&found->cl_clientid, clid))
4039                         return nfserr_stale_clientid;
4040                 return nfs_ok;
4041         }
4042
4043         if (STALE_CLIENTID(clid, nn))
4044                 return nfserr_stale_clientid;
4045
4046         /*
4047          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4048          * cached already then we know this is for is for v4.0 and "sessions"
4049          * will be false.
4050          */
4051         WARN_ON_ONCE(cstate->session);
4052         spin_lock(&nn->client_lock);
4053         found = find_confirmed_client(clid, false, nn);
4054         if (!found) {
4055                 spin_unlock(&nn->client_lock);
4056                 return nfserr_expired;
4057         }
4058         atomic_inc(&found->cl_refcount);
4059         spin_unlock(&nn->client_lock);
4060
4061         /* Cache the nfs4_client in cstate! */
4062         cstate->clp = found;
4063         return nfs_ok;
4064 }
4065
4066 __be32
4067 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4068                     struct nfsd4_open *open, struct nfsd_net *nn)
4069 {
4070         clientid_t *clientid = &open->op_clientid;
4071         struct nfs4_client *clp = NULL;
4072         unsigned int strhashval;
4073         struct nfs4_openowner *oo = NULL;
4074         __be32 status;
4075
4076         if (STALE_CLIENTID(&open->op_clientid, nn))
4077                 return nfserr_stale_clientid;
4078         /*
4079          * In case we need it later, after we've already created the
4080          * file and don't want to risk a further failure:
4081          */
4082         open->op_file = nfsd4_alloc_file();
4083         if (open->op_file == NULL)
4084                 return nfserr_jukebox;
4085
4086         status = lookup_clientid(clientid, cstate, nn);
4087         if (status)
4088                 return status;
4089         clp = cstate->clp;
4090
4091         strhashval = ownerstr_hashval(&open->op_owner);
4092         oo = find_openstateowner_str(strhashval, open, clp);
4093         open->op_openowner = oo;
4094         if (!oo) {
4095                 goto new_owner;
4096         }
4097         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4098                 /* Replace unconfirmed owners without checking for replay. */
4099                 release_openowner(oo);
4100                 open->op_openowner = NULL;
4101                 goto new_owner;
4102         }
4103         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4104         if (status)
4105                 return status;
4106         goto alloc_stateid;
4107 new_owner:
4108         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4109         if (oo == NULL)
4110                 return nfserr_jukebox;
4111         open->op_openowner = oo;
4112 alloc_stateid:
4113         open->op_stp = nfs4_alloc_open_stateid(clp);
4114         if (!open->op_stp)
4115                 return nfserr_jukebox;
4116
4117         if (nfsd4_has_session(cstate) &&
4118             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4119                 open->op_odstate = alloc_clnt_odstate(clp);
4120                 if (!open->op_odstate)
4121                         return nfserr_jukebox;
4122         }
4123
4124         return nfs_ok;
4125 }
4126
4127 static inline __be32
4128 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4129 {
4130         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4131                 return nfserr_openmode;
4132         else
4133                 return nfs_ok;
4134 }
4135
4136 static int share_access_to_flags(u32 share_access)
4137 {
4138         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4139 }
4140
4141 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4142 {
4143         struct nfs4_stid *ret;
4144
4145         ret = find_stateid_by_type(cl, s,
4146                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4147         if (!ret)
4148                 return NULL;
4149         return delegstateid(ret);
4150 }
4151
4152 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4153 {
4154         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4155                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4156 }
4157
4158 static __be32
4159 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4160                 struct nfs4_delegation **dp)
4161 {
4162         int flags;
4163         __be32 status = nfserr_bad_stateid;
4164         struct nfs4_delegation *deleg;
4165
4166         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4167         if (deleg == NULL)
4168                 goto out;
4169         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4170                 nfs4_put_stid(&deleg->dl_stid);
4171                 if (cl->cl_minorversion)
4172                         status = nfserr_deleg_revoked;
4173                 goto out;
4174         }
4175         flags = share_access_to_flags(open->op_share_access);
4176         status = nfs4_check_delegmode(deleg, flags);
4177         if (status) {
4178                 nfs4_put_stid(&deleg->dl_stid);
4179                 goto out;
4180         }
4181         *dp = deleg;
4182 out:
4183         if (!nfsd4_is_deleg_cur(open))
4184                 return nfs_ok;
4185         if (status)
4186                 return status;
4187         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4188         return nfs_ok;
4189 }
4190
4191 static inline int nfs4_access_to_access(u32 nfs4_access)
4192 {
4193         int flags = 0;
4194
4195         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4196                 flags |= NFSD_MAY_READ;
4197         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4198                 flags |= NFSD_MAY_WRITE;
4199         return flags;
4200 }
4201
4202 static inline __be32
4203 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4204                 struct nfsd4_open *open)
4205 {
4206         struct iattr iattr = {
4207                 .ia_valid = ATTR_SIZE,
4208                 .ia_size = 0,
4209         };
4210         if (!open->op_truncate)
4211                 return 0;
4212         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4213                 return nfserr_inval;
4214         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4215 }
4216
4217 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4218                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4219                 struct nfsd4_open *open)
4220 {
4221         struct file *filp = NULL;
4222         __be32 status;
4223         int oflag = nfs4_access_to_omode(open->op_share_access);
4224         int access = nfs4_access_to_access(open->op_share_access);
4225         unsigned char old_access_bmap, old_deny_bmap;
4226
4227         spin_lock(&fp->fi_lock);
4228
4229         /*
4230          * Are we trying to set a deny mode that would conflict with
4231          * current access?
4232          */
4233         status = nfs4_file_check_deny(fp, open->op_share_deny);
4234         if (status != nfs_ok) {
4235                 spin_unlock(&fp->fi_lock);
4236                 goto out;
4237         }
4238
4239         /* set access to the file */
4240         status = nfs4_file_get_access(fp, open->op_share_access);
4241         if (status != nfs_ok) {
4242                 spin_unlock(&fp->fi_lock);
4243                 goto out;
4244         }
4245
4246         /* Set access bits in stateid */
4247         old_access_bmap = stp->st_access_bmap;
4248         set_access(open->op_share_access, stp);
4249
4250         /* Set new deny mask */
4251         old_deny_bmap = stp->st_deny_bmap;
4252         set_deny(open->op_share_deny, stp);
4253         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4254
4255         if (!fp->fi_fds[oflag]) {
4256                 spin_unlock(&fp->fi_lock);
4257                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4258                 if (status)
4259                         goto out_put_access;
4260                 spin_lock(&fp->fi_lock);
4261                 if (!fp->fi_fds[oflag]) {
4262                         fp->fi_fds[oflag] = filp;
4263                         filp = NULL;
4264                 }
4265         }
4266         spin_unlock(&fp->fi_lock);
4267         if (filp)
4268                 fput(filp);
4269
4270         status = nfsd4_truncate(rqstp, cur_fh, open);
4271         if (status)
4272                 goto out_put_access;
4273 out:
4274         return status;
4275 out_put_access:
4276         stp->st_access_bmap = old_access_bmap;
4277         nfs4_file_put_access(fp, open->op_share_access);
4278         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4279         goto out;
4280 }
4281
4282 static __be32
4283 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)
4284 {
4285         __be32 status;
4286         unsigned char old_deny_bmap = stp->st_deny_bmap;
4287
4288         if (!test_access(open->op_share_access, stp))
4289                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4290
4291         /* test and set deny mode */
4292         spin_lock(&fp->fi_lock);
4293         status = nfs4_file_check_deny(fp, open->op_share_deny);
4294         if (status == nfs_ok) {
4295                 set_deny(open->op_share_deny, stp);
4296                 fp->fi_share_deny |=
4297                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4298         }
4299         spin_unlock(&fp->fi_lock);
4300
4301         if (status != nfs_ok)
4302                 return status;
4303
4304         status = nfsd4_truncate(rqstp, cur_fh, open);
4305         if (status != nfs_ok)
4306                 reset_union_bmap_deny(old_deny_bmap, stp);
4307         return status;
4308 }
4309
4310 /* Should we give out recallable state?: */
4311 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4312 {
4313         if (clp->cl_cb_state == NFSD4_CB_UP)
4314                 return true;
4315         /*
4316          * In the sessions case, since we don't have to establish a
4317          * separate connection for callbacks, we assume it's OK
4318          * until we hear otherwise:
4319          */
4320         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4321 }
4322
4323 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4324                                                 int flag)
4325 {
4326         struct file_lock *fl;
4327
4328         fl = locks_alloc_lock();
4329         if (!fl)
4330                 return NULL;
4331         fl->fl_lmops = &nfsd_lease_mng_ops;
4332         fl->fl_flags = FL_DELEG;
4333         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4334         fl->fl_end = OFFSET_MAX;
4335         fl->fl_owner = (fl_owner_t)dp;
4336         fl->fl_pid = current->tgid;
4337         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4338         return fl;
4339 }
4340
4341 static struct nfs4_delegation *
4342 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4343                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4344 {
4345         int status = 0;
4346         struct nfs4_delegation *dp;
4347         struct file *filp;
4348         struct file_lock *fl;
4349
4350         /*
4351          * The fi_had_conflict and nfs_get_existing_delegation checks
4352          * here are just optimizations; we'll need to recheck them at
4353          * the end:
4354          */
4355         if (fp->fi_had_conflict)
4356                 return ERR_PTR(-EAGAIN);
4357
4358         filp = find_readable_file(fp);
4359         if (!filp) {
4360                 /* We should always have a readable file here */
4361                 WARN_ON_ONCE(1);
4362                 return ERR_PTR(-EBADF);
4363         }
4364         spin_lock(&state_lock);
4365         spin_lock(&fp->fi_lock);
4366         if (nfs4_delegation_exists(clp, fp))
4367                 status = -EAGAIN;
4368         else if (!fp->fi_deleg_file) {
4369                 fp->fi_deleg_file = filp;
4370                 /* increment early to prevent fi_deleg_file from being
4371                  * cleared */
4372                 fp->fi_delegees = 1;
4373                 filp = NULL;
4374         } else
4375                 fp->fi_delegees++;
4376         spin_unlock(&fp->fi_lock);
4377         spin_unlock(&state_lock);
4378         if (filp)
4379                 fput(filp);
4380         if (status)
4381                 return ERR_PTR(status);
4382
4383         status = -ENOMEM;
4384         dp = alloc_init_deleg(clp, fp, fh, odstate);
4385         if (!dp)
4386                 goto out_delegees;
4387
4388         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4389         if (!fl)
4390                 goto out_clnt_odstate;
4391
4392         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4393         if (fl)
4394                 locks_free_lock(fl);
4395         if (status)
4396                 goto out_clnt_odstate;
4397
4398         spin_lock(&state_lock);
4399         spin_lock(&fp->fi_lock);
4400         if (fp->fi_had_conflict)
4401                 status = -EAGAIN;
4402         else
4403                 status = hash_delegation_locked(dp, fp);
4404         spin_unlock(&fp->fi_lock);
4405         spin_unlock(&state_lock);
4406
4407         if (status)
4408                 goto out_unlock;
4409
4410         return dp;
4411 out_unlock:
4412         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4413 out_clnt_odstate:
4414         put_clnt_odstate(dp->dl_clnt_odstate);
4415         nfs4_put_stid(&dp->dl_stid);
4416 out_delegees:
4417         put_deleg_file(fp);
4418         return ERR_PTR(status);
4419 }
4420
4421 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4422 {
4423         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4424         if (status == -EAGAIN)
4425                 open->op_why_no_deleg = WND4_CONTENTION;
4426         else {
4427                 open->op_why_no_deleg = WND4_RESOURCE;
4428                 switch (open->op_deleg_want) {
4429                 case NFS4_SHARE_WANT_READ_DELEG:
4430                 case NFS4_SHARE_WANT_WRITE_DELEG:
4431                 case NFS4_SHARE_WANT_ANY_DELEG:
4432                         break;
4433                 case NFS4_SHARE_WANT_CANCEL:
4434                         open->op_why_no_deleg = WND4_CANCELLED;
4435                         break;
4436                 case NFS4_SHARE_WANT_NO_DELEG:
4437                         WARN_ON_ONCE(1);
4438                 }
4439         }
4440 }
4441
4442 /*
4443  * Attempt to hand out a delegation.
4444  *
4445  * Note we don't support write delegations, and won't until the vfs has
4446  * proper support for them.
4447  */
4448 static void
4449 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4450                         struct nfs4_ol_stateid *stp)
4451 {
4452         struct nfs4_delegation *dp;
4453         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4454         struct nfs4_client *clp = stp->st_stid.sc_client;
4455         int cb_up;
4456         int status = 0;
4457
4458         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4459         open->op_recall = 0;
4460         switch (open->op_claim_type) {
4461                 case NFS4_OPEN_CLAIM_PREVIOUS:
4462                         if (!cb_up)
4463                                 open->op_recall = 1;
4464                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4465                                 goto out_no_deleg;
4466                         break;
4467                 case NFS4_OPEN_CLAIM_NULL:
4468                 case NFS4_OPEN_CLAIM_FH:
4469                         /*
4470                          * Let's not give out any delegations till everyone's
4471                          * had the chance to reclaim theirs, *and* until
4472                          * NLM locks have all been reclaimed:
4473                          */
4474                         if (locks_in_grace(clp->net))
4475                                 goto out_no_deleg;
4476                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4477                                 goto out_no_deleg;
4478                         /*
4479                          * Also, if the file was opened for write or
4480                          * create, there's a good chance the client's
4481                          * about to write to it, resulting in an
4482                          * immediate recall (since we don't support
4483                          * write delegations):
4484                          */
4485                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4486                                 goto out_no_deleg;
4487                         if (open->op_create == NFS4_OPEN_CREATE)
4488                                 goto out_no_deleg;
4489                         break;
4490                 default:
4491                         goto out_no_deleg;
4492         }
4493         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4494         if (IS_ERR(dp))
4495                 goto out_no_deleg;
4496
4497         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4498
4499         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4500                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4501         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4502         nfs4_put_stid(&dp->dl_stid);
4503         return;
4504 out_no_deleg:
4505         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4506         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4507             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4508                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4509                 open->op_recall = 1;
4510         }
4511
4512         /* 4.1 client asking for a delegation? */
4513         if (open->op_deleg_want)
4514                 nfsd4_open_deleg_none_ext(open, status);
4515         return;
4516 }
4517
4518 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4519                                         struct nfs4_delegation *dp)
4520 {
4521         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4522             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4523                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4524                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4525         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4526                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4527                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4528                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4529         }
4530         /* Otherwise the client must be confused wanting a delegation
4531          * it already has, therefore we don't return
4532          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4533          */
4534 }
4535
4536 __be32
4537 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4538 {
4539         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4540         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4541         struct nfs4_file *fp = NULL;
4542         struct nfs4_ol_stateid *stp = NULL;
4543         struct nfs4_delegation *dp = NULL;
4544         __be32 status;
4545         bool new_stp = false;
4546
4547         /*
4548          * Lookup file; if found, lookup stateid and check open request,
4549          * and check for delegations in the process of being recalled.
4550          * If not found, create the nfs4_file struct
4551          */
4552         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4553         if (fp != open->op_file) {
4554                 status = nfs4_check_deleg(cl, open, &dp);
4555                 if (status)
4556                         goto out;
4557                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4558         } else {
4559                 open->op_file = NULL;
4560                 status = nfserr_bad_stateid;
4561                 if (nfsd4_is_deleg_cur(open))
4562                         goto out;
4563         }
4564
4565         if (!stp) {
4566                 stp = init_open_stateid(fp, open);
4567                 if (!open->op_stp)
4568                         new_stp = true;
4569         }
4570
4571         /*
4572          * OPEN the file, or upgrade an existing OPEN.
4573          * If truncate fails, the OPEN fails.
4574          *
4575          * stp is already locked.
4576          */
4577         if (!new_stp) {
4578                 /* Stateid was found, this is an OPEN upgrade */
4579                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4580                 if (status) {
4581                         mutex_unlock(&stp->st_mutex);
4582                         goto out;
4583                 }
4584         } else {
4585                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4586                 if (status) {
4587                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4588                         release_open_stateid(stp);
4589                         mutex_unlock(&stp->st_mutex);
4590                         goto out;
4591                 }
4592
4593                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4594                                                         open->op_odstate);
4595                 if (stp->st_clnt_odstate == open->op_odstate)
4596                         open->op_odstate = NULL;
4597         }
4598
4599         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4600         mutex_unlock(&stp->st_mutex);
4601
4602         if (nfsd4_has_session(&resp->cstate)) {
4603                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4604                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4605                         open->op_why_no_deleg = WND4_NOT_WANTED;
4606                         goto nodeleg;
4607                 }
4608         }
4609
4610         /*
4611         * Attempt to hand out a delegation. No error return, because the
4612         * OPEN succeeds even if we fail.
4613         */
4614         nfs4_open_delegation(current_fh, open, stp);
4615 nodeleg:
4616         status = nfs_ok;
4617
4618         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4619                 STATEID_VAL(&stp->st_stid.sc_stateid));
4620 out:
4621         /* 4.1 client trying to upgrade/downgrade delegation? */
4622         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4623             open->op_deleg_want)
4624                 nfsd4_deleg_xgrade_none_ext(open, dp);
4625
4626         if (fp)
4627                 put_nfs4_file(fp);
4628         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4629                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4630         /*
4631         * To finish the open response, we just need to set the rflags.
4632         */
4633         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4634         if (nfsd4_has_session(&resp->cstate))
4635                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4636         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4637                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4638
4639         if (dp)
4640                 nfs4_put_stid(&dp->dl_stid);
4641         if (stp)
4642                 nfs4_put_stid(&stp->st_stid);
4643
4644         return status;
4645 }
4646
4647 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4648                               struct nfsd4_open *open)
4649 {
4650         if (open->op_openowner) {
4651                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4652
4653                 nfsd4_cstate_assign_replay(cstate, so);
4654                 nfs4_put_stateowner(so);
4655         }
4656         if (open->op_file)
4657                 kmem_cache_free(file_slab, open->op_file);
4658         if (open->op_stp)
4659                 nfs4_put_stid(&open->op_stp->st_stid);
4660         if (open->op_odstate)
4661                 kmem_cache_free(odstate_slab, open->op_odstate);
4662 }
4663
4664 __be32
4665 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4666             union nfsd4_op_u *u)
4667 {
4668         clientid_t *clid = &u->renew;
4669         struct nfs4_client *clp;
4670         __be32 status;
4671         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4672
4673         dprintk("process_renew(%08x/%08x): starting\n",
4674                         clid->cl_boot, clid->cl_id);
4675         status = lookup_clientid(clid, cstate, nn);
4676         if (status)
4677                 goto out;
4678         clp = cstate->clp;
4679         status = nfserr_cb_path_down;
4680         if (!list_empty(&clp->cl_delegations)
4681                         && clp->cl_cb_state != NFSD4_CB_UP)
4682                 goto out;
4683         status = nfs_ok;
4684 out:
4685         return status;
4686 }
4687
4688 void
4689 nfsd4_end_grace(struct nfsd_net *nn)
4690 {
4691         /* do nothing if grace period already ended */
4692         if (nn->grace_ended)
4693                 return;
4694
4695         dprintk("NFSD: end of grace period\n");
4696         nn->grace_ended = true;
4697         /*
4698          * If the server goes down again right now, an NFSv4
4699          * client will still be allowed to reclaim after it comes back up,
4700          * even if it hasn't yet had a chance to reclaim state this time.
4701          *
4702          */
4703         nfsd4_record_grace_done(nn);
4704         /*
4705          * At this point, NFSv4 clients can still reclaim.  But if the
4706          * server crashes, any that have not yet reclaimed will be out
4707          * of luck on the next boot.
4708          *
4709          * (NFSv4.1+ clients are considered to have reclaimed once they
4710          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4711          * have reclaimed after their first OPEN.)
4712          */
4713         locks_end_grace(&nn->nfsd4_manager);
4714         /*
4715          * At this point, and once lockd and/or any other containers
4716          * exit their grace period, further reclaims will fail and
4717          * regular locking can resume.
4718          */
4719 }
4720
4721 /*
4722  * If we've waited a lease period but there are still clients trying to
4723  * reclaim, wait a little longer to give them a chance to finish.
4724  */
4725 static bool clients_still_reclaiming(struct nfsd_net *nn)
4726 {
4727         unsigned long now = get_seconds();
4728         unsigned long double_grace_period_end = nn->boot_time +
4729                                                 2 * nn->nfsd4_lease;
4730
4731         if (!nn->somebody_reclaimed)
4732                 return false;
4733         nn->somebody_reclaimed = false;
4734         /*
4735          * If we've given them *two* lease times to reclaim, and they're
4736          * still not done, give up:
4737          */
4738         if (time_after(now, double_grace_period_end))
4739                 return false;
4740         return true;
4741 }
4742
4743 static time_t
4744 nfs4_laundromat(struct nfsd_net *nn)
4745 {
4746         struct nfs4_client *clp;
4747         struct nfs4_openowner *oo;
4748         struct nfs4_delegation *dp;
4749         struct nfs4_ol_stateid *stp;
4750         struct nfsd4_blocked_lock *nbl;
4751         struct list_head *pos, *next, reaplist;
4752         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4753         time_t t, new_timeo = nn->nfsd4_lease;
4754
4755         dprintk("NFSD: laundromat service - starting\n");
4756
4757         if (clients_still_reclaiming(nn)) {
4758                 new_timeo = 0;
4759                 goto out;
4760         }
4761         nfsd4_end_grace(nn);
4762         INIT_LIST_HEAD(&reaplist);
4763         spin_lock(&nn->client_lock);
4764         list_for_each_safe(pos, next, &nn->client_lru) {
4765                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4766                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4767                         t = clp->cl_time - cutoff;
4768                         new_timeo = min(new_timeo, t);
4769                         break;
4770                 }
4771                 if (mark_client_expired_locked(clp)) {
4772                         dprintk("NFSD: client in use (clientid %08x)\n",
4773                                 clp->cl_clientid.cl_id);
4774                         continue;
4775                 }
4776                 list_add(&clp->cl_lru, &reaplist);
4777         }
4778         spin_unlock(&nn->client_lock);
4779         list_for_each_safe(pos, next, &reaplist) {
4780                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4781                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4782                         clp->cl_clientid.cl_id);
4783                 list_del_init(&clp->cl_lru);
4784                 expire_client(clp);
4785         }
4786         spin_lock(&state_lock);
4787         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4788                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4789                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4790                         t = dp->dl_time - cutoff;
4791                         new_timeo = min(new_timeo, t);
4792                         break;
4793                 }
4794                 WARN_ON(!unhash_delegation_locked(dp));
4795                 list_add(&dp->dl_recall_lru, &reaplist);
4796         }
4797         spin_unlock(&state_lock);
4798         while (!list_empty(&reaplist)) {
4799                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4800                                         dl_recall_lru);
4801                 list_del_init(&dp->dl_recall_lru);
4802                 revoke_delegation(dp);
4803         }
4804
4805         spin_lock(&nn->client_lock);
4806         while (!list_empty(&nn->close_lru)) {
4807                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4808                                         oo_close_lru);
4809                 if (time_after((unsigned long)oo->oo_time,
4810                                (unsigned long)cutoff)) {
4811                         t = oo->oo_time - cutoff;
4812                         new_timeo = min(new_timeo, t);
4813                         break;
4814                 }
4815                 list_del_init(&oo->oo_close_lru);
4816                 stp = oo->oo_last_closed_stid;
4817                 oo->oo_last_closed_stid = NULL;
4818                 spin_unlock(&nn->client_lock);
4819                 nfs4_put_stid(&stp->st_stid);
4820                 spin_lock(&nn->client_lock);
4821         }
4822         spin_unlock(&nn->client_lock);
4823
4824         /*
4825          * It's possible for a client to try and acquire an already held lock
4826          * that is being held for a long time, and then lose interest in it.
4827          * So, we clean out any un-revisited request after a lease period
4828          * under the assumption that the client is no longer interested.
4829          *
4830          * RFC5661, sec. 9.6 states that the client must not rely on getting
4831          * notifications and must continue to poll for locks, even when the
4832          * server supports them. Thus this shouldn't lead to clients blocking
4833          * indefinitely once the lock does become free.
4834          */
4835         BUG_ON(!list_empty(&reaplist));
4836         spin_lock(&nn->blocked_locks_lock);
4837         while (!list_empty(&nn->blocked_locks_lru)) {
4838                 nbl = list_first_entry(&nn->blocked_locks_lru,
4839                                         struct nfsd4_blocked_lock, nbl_lru);
4840                 if (time_after((unsigned long)nbl->nbl_time,
4841                                (unsigned long)cutoff)) {
4842                         t = nbl->nbl_time - cutoff;
4843                         new_timeo = min(new_timeo, t);
4844                         break;
4845                 }
4846                 list_move(&nbl->nbl_lru, &reaplist);
4847                 list_del_init(&nbl->nbl_list);
4848         }
4849         spin_unlock(&nn->blocked_locks_lock);
4850
4851         while (!list_empty(&reaplist)) {
4852                 nbl = list_first_entry(&reaplist,
4853                                         struct nfsd4_blocked_lock, nbl_lru);
4854                 list_del_init(&nbl->nbl_lru);
4855                 posix_unblock_lock(&nbl->nbl_lock);
4856                 free_blocked_lock(nbl);
4857         }
4858 out:
4859         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4860         return new_timeo;
4861 }
4862
4863 static struct workqueue_struct *laundry_wq;
4864 static void laundromat_main(struct work_struct *);
4865
4866 static void
4867 laundromat_main(struct work_struct *laundry)
4868 {
4869         time_t t;
4870         struct delayed_work *dwork = to_delayed_work(laundry);
4871         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4872                                            laundromat_work);
4873
4874         t = nfs4_laundromat(nn);
4875         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4876         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4877 }
4878
4879 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4880 {
4881         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4882                 return nfserr_bad_stateid;
4883         return nfs_ok;
4884 }
4885
4886 static inline int
4887 access_permit_read(struct nfs4_ol_stateid *stp)
4888 {
4889         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4890                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4891                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4892 }
4893
4894 static inline int
4895 access_permit_write(struct nfs4_ol_stateid *stp)
4896 {
4897         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4898                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4899 }
4900
4901 static
4902 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4903 {
4904         __be32 status = nfserr_openmode;
4905
4906         /* For lock stateid's, we test the parent open, not the lock: */
4907         if (stp->st_openstp)
4908                 stp = stp->st_openstp;
4909         if ((flags & WR_STATE) && !access_permit_write(stp))
4910                 goto out;
4911         if ((flags & RD_STATE) && !access_permit_read(stp))
4912                 goto out;
4913         status = nfs_ok;
4914 out:
4915         return status;
4916 }
4917
4918 static inline __be32
4919 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4920 {
4921         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4922                 return nfs_ok;
4923         else if (opens_in_grace(net)) {
4924                 /* Answer in remaining cases depends on existence of
4925                  * conflicting state; so we must wait out the grace period. */
4926                 return nfserr_grace;
4927         } else if (flags & WR_STATE)
4928                 return nfs4_share_conflict(current_fh,
4929                                 NFS4_SHARE_DENY_WRITE);
4930         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4931                 return nfs4_share_conflict(current_fh,
4932                                 NFS4_SHARE_DENY_READ);
4933 }
4934
4935 /*
4936  * Allow READ/WRITE during grace period on recovered state only for files
4937  * that are not able to provide mandatory locking.
4938  */
4939 static inline int
4940 grace_disallows_io(struct net *net, struct inode *inode)
4941 {
4942         return opens_in_grace(net) && mandatory_lock(inode);
4943 }
4944
4945 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4946 {
4947         /*
4948          * When sessions are used the stateid generation number is ignored
4949          * when it is zero.
4950          */
4951         if (has_session && in->si_generation == 0)
4952                 return nfs_ok;
4953
4954         if (in->si_generation == ref->si_generation)
4955                 return nfs_ok;
4956
4957         /* If the client sends us a stateid from the future, it's buggy: */
4958         if (nfsd4_stateid_generation_after(in, ref))
4959                 return nfserr_bad_stateid;
4960         /*
4961          * However, we could see a stateid from the past, even from a
4962          * non-buggy client.  For example, if the client sends a lock
4963          * while some IO is outstanding, the lock may bump si_generation
4964          * while the IO is still in flight.  The client could avoid that
4965          * situation by waiting for responses on all the IO requests,
4966          * but better performance may result in retrying IO that
4967          * receives an old_stateid error if requests are rarely
4968          * reordered in flight:
4969          */
4970         return nfserr_old_stateid;
4971 }
4972
4973 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4974 {
4975         __be32 ret;
4976
4977         spin_lock(&s->sc_lock);
4978         ret = nfsd4_verify_open_stid(s);
4979         if (ret == nfs_ok)
4980                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4981         spin_unlock(&s->sc_lock);
4982         return ret;
4983 }
4984
4985 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4986 {
4987         if (ols->st_stateowner->so_is_open_owner &&
4988             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4989                 return nfserr_bad_stateid;
4990         return nfs_ok;
4991 }
4992
4993 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4994 {
4995         struct nfs4_stid *s;
4996         __be32 status = nfserr_bad_stateid;
4997
4998         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4999                 CLOSE_STATEID(stateid))
5000                 return status;
5001         /* Client debugging aid. */
5002         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5003                 char addr_str[INET6_ADDRSTRLEN];
5004                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5005                                  sizeof(addr_str));
5006                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5007                                         "with incorrect client ID\n", addr_str);
5008                 return status;
5009         }
5010         spin_lock(&cl->cl_lock);
5011         s = find_stateid_locked(cl, stateid);
5012         if (!s)
5013                 goto out_unlock;
5014         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5015         if (status)
5016                 goto out_unlock;
5017         switch (s->sc_type) {
5018         case NFS4_DELEG_STID:
5019                 status = nfs_ok;
5020                 break;
5021         case NFS4_REVOKED_DELEG_STID:
5022                 status = nfserr_deleg_revoked;
5023                 break;
5024         case NFS4_OPEN_STID:
5025         case NFS4_LOCK_STID:
5026                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5027                 break;
5028         default:
5029                 printk("unknown stateid type %x\n", s->sc_type);
5030                 /* Fallthrough */
5031         case NFS4_CLOSED_STID:
5032         case NFS4_CLOSED_DELEG_STID:
5033                 status = nfserr_bad_stateid;
5034         }
5035 out_unlock:
5036         spin_unlock(&cl->cl_lock);
5037         return status;
5038 }
5039
5040 __be32
5041 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5042                      stateid_t *stateid, unsigned char typemask,
5043                      struct nfs4_stid **s, struct nfsd_net *nn)
5044 {
5045         __be32 status;
5046         bool return_revoked = false;
5047
5048         /*
5049          *  only return revoked delegations if explicitly asked.
5050          *  otherwise we report revoked or bad_stateid status.
5051          */
5052         if (typemask & NFS4_REVOKED_DELEG_STID)
5053                 return_revoked = true;
5054         else if (typemask & NFS4_DELEG_STID)
5055                 typemask |= NFS4_REVOKED_DELEG_STID;
5056
5057         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5058                 CLOSE_STATEID(stateid))
5059                 return nfserr_bad_stateid;
5060         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5061         if (status == nfserr_stale_clientid) {
5062                 if (cstate->session)
5063                         return nfserr_bad_stateid;
5064                 return nfserr_stale_stateid;
5065         }
5066         if (status)
5067                 return status;
5068         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5069         if (!*s)
5070                 return nfserr_bad_stateid;
5071         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5072                 nfs4_put_stid(*s);
5073                 if (cstate->minorversion)
5074                         return nfserr_deleg_revoked;
5075                 return nfserr_bad_stateid;
5076         }
5077         return nfs_ok;
5078 }
5079
5080 static struct file *
5081 nfs4_find_file(struct nfs4_stid *s, int flags)
5082 {
5083         if (!s)
5084                 return NULL;
5085
5086         switch (s->sc_type) {
5087         case NFS4_DELEG_STID:
5088                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5089                         return NULL;
5090                 return get_file(s->sc_file->fi_deleg_file);
5091         case NFS4_OPEN_STID:
5092         case NFS4_LOCK_STID:
5093                 if (flags & RD_STATE)
5094                         return find_readable_file(s->sc_file);
5095                 else
5096                         return find_writeable_file(s->sc_file);
5097                 break;
5098         }
5099
5100         return NULL;
5101 }
5102
5103 static __be32
5104 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5105 {
5106         __be32 status;
5107
5108         status = nfsd4_check_openowner_confirmed(ols);
5109         if (status)
5110                 return status;
5111         return nfs4_check_openmode(ols, flags);
5112 }
5113
5114 static __be32
5115 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5116                 struct file **filpp, bool *tmp_file, int flags)
5117 {
5118         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5119         struct file *file;
5120         __be32 status;
5121
5122         file = nfs4_find_file(s, flags);
5123         if (file) {
5124                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5125                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5126                 if (status) {
5127                         fput(file);
5128                         return status;
5129                 }
5130
5131                 *filpp = file;
5132         } else {
5133                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5134                 if (status)
5135                         return status;
5136
5137                 if (tmp_file)
5138                         *tmp_file = true;
5139         }
5140
5141         return 0;
5142 }
5143
5144 /*
5145  * Checks for stateid operations
5146  */
5147 __be32
5148 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5149                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5150                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5151 {
5152         struct inode *ino = d_inode(fhp->fh_dentry);
5153         struct net *net = SVC_NET(rqstp);
5154         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5155         struct nfs4_stid *s = NULL;
5156         __be32 status;
5157
5158         if (filpp)
5159                 *filpp = NULL;
5160         if (tmp_file)
5161                 *tmp_file = false;
5162
5163         if (grace_disallows_io(net, ino))
5164                 return nfserr_grace;
5165
5166         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5167                 status = check_special_stateids(net, fhp, stateid, flags);
5168                 goto done;
5169         }
5170
5171         status = nfsd4_lookup_stateid(cstate, stateid,
5172                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5173                                 &s, nn);
5174         if (status)
5175                 return status;
5176         status = nfsd4_stid_check_stateid_generation(stateid, s,
5177                         nfsd4_has_session(cstate));
5178         if (status)
5179                 goto out;
5180
5181         switch (s->sc_type) {
5182         case NFS4_DELEG_STID:
5183                 status = nfs4_check_delegmode(delegstateid(s), flags);
5184                 break;
5185         case NFS4_OPEN_STID:
5186         case NFS4_LOCK_STID:
5187                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5188                 break;
5189         default:
5190                 status = nfserr_bad_stateid;
5191                 break;
5192         }
5193         if (status)
5194                 goto out;
5195         status = nfs4_check_fh(fhp, s);
5196
5197 done:
5198         if (!status && filpp)
5199                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5200 out:
5201         if (s)
5202                 nfs4_put_stid(s);
5203         return status;
5204 }
5205
5206 /*
5207  * Test if the stateid is valid
5208  */
5209 __be32
5210 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5211                    union nfsd4_op_u *u)
5212 {
5213         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5214         struct nfsd4_test_stateid_id *stateid;
5215         struct nfs4_client *cl = cstate->session->se_client;
5216
5217         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5218                 stateid->ts_id_status =
5219                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5220
5221         return nfs_ok;
5222 }
5223
5224 static __be32
5225 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5226 {
5227         struct nfs4_ol_stateid *stp = openlockstateid(s);
5228         __be32 ret;
5229
5230         ret = nfsd4_lock_ol_stateid(stp);
5231         if (ret)
5232                 goto out_put_stid;
5233
5234         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5235         if (ret)
5236                 goto out;
5237
5238         ret = nfserr_locks_held;
5239         if (check_for_locks(stp->st_stid.sc_file,
5240                             lockowner(stp->st_stateowner)))
5241                 goto out;
5242
5243         release_lock_stateid(stp);
5244         ret = nfs_ok;
5245
5246 out:
5247         mutex_unlock(&stp->st_mutex);
5248 out_put_stid:
5249         nfs4_put_stid(s);
5250         return ret;
5251 }
5252
5253 __be32
5254 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5255                    union nfsd4_op_u *u)
5256 {
5257         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5258         stateid_t *stateid = &free_stateid->fr_stateid;
5259         struct nfs4_stid *s;
5260         struct nfs4_delegation *dp;
5261         struct nfs4_client *cl = cstate->session->se_client;
5262         __be32 ret = nfserr_bad_stateid;
5263
5264         spin_lock(&cl->cl_lock);
5265         s = find_stateid_locked(cl, stateid);
5266         if (!s)
5267                 goto out_unlock;
5268         spin_lock(&s->sc_lock);
5269         switch (s->sc_type) {
5270         case NFS4_DELEG_STID:
5271                 ret = nfserr_locks_held;
5272                 break;
5273         case NFS4_OPEN_STID:
5274                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5275                 if (ret)
5276                         break;
5277                 ret = nfserr_locks_held;
5278                 break;
5279         case NFS4_LOCK_STID:
5280                 spin_unlock(&s->sc_lock);
5281                 refcount_inc(&s->sc_count);
5282                 spin_unlock(&cl->cl_lock);
5283                 ret = nfsd4_free_lock_stateid(stateid, s);
5284                 goto out;
5285         case NFS4_REVOKED_DELEG_STID:
5286                 spin_unlock(&s->sc_lock);
5287                 dp = delegstateid(s);
5288                 list_del_init(&dp->dl_recall_lru);
5289                 spin_unlock(&cl->cl_lock);
5290                 nfs4_put_stid(s);
5291                 ret = nfs_ok;
5292                 goto out;
5293         /* Default falls through and returns nfserr_bad_stateid */
5294         }
5295         spin_unlock(&s->sc_lock);
5296 out_unlock:
5297         spin_unlock(&cl->cl_lock);
5298 out:
5299         return ret;
5300 }
5301
5302 static inline int
5303 setlkflg (int type)
5304 {
5305         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5306                 RD_STATE : WR_STATE;
5307 }
5308
5309 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5310 {
5311         struct svc_fh *current_fh = &cstate->current_fh;
5312         struct nfs4_stateowner *sop = stp->st_stateowner;
5313         __be32 status;
5314
5315         status = nfsd4_check_seqid(cstate, sop, seqid);
5316         if (status)
5317                 return status;
5318         status = nfsd4_lock_ol_stateid(stp);
5319         if (status != nfs_ok)
5320                 return status;
5321         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5322         if (status == nfs_ok)
5323                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5324         if (status != nfs_ok)
5325                 mutex_unlock(&stp->st_mutex);
5326         return status;
5327 }
5328
5329 /*
5330  * Checks for sequence id mutating operations.
5331  */
5332 static __be32
5333 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5334                          stateid_t *stateid, char typemask,
5335                          struct nfs4_ol_stateid **stpp,
5336                          struct nfsd_net *nn)
5337 {
5338         __be32 status;
5339         struct nfs4_stid *s;
5340         struct nfs4_ol_stateid *stp = NULL;
5341
5342         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5343                 seqid, STATEID_VAL(stateid));
5344
5345         *stpp = NULL;
5346         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5347         if (status)
5348                 return status;
5349         stp = openlockstateid(s);
5350         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5351
5352         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5353         if (!status)
5354                 *stpp = stp;
5355         else
5356                 nfs4_put_stid(&stp->st_stid);
5357         return status;
5358 }
5359
5360 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5361                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5362 {
5363         __be32 status;
5364         struct nfs4_openowner *oo;
5365         struct nfs4_ol_stateid *stp;
5366
5367         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5368                                                 NFS4_OPEN_STID, &stp, nn);
5369         if (status)
5370                 return status;
5371         oo = openowner(stp->st_stateowner);
5372         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5373                 mutex_unlock(&stp->st_mutex);
5374                 nfs4_put_stid(&stp->st_stid);
5375                 return nfserr_bad_stateid;
5376         }
5377         *stpp = stp;
5378         return nfs_ok;
5379 }
5380
5381 __be32
5382 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5383                    union nfsd4_op_u *u)
5384 {
5385         struct nfsd4_open_confirm *oc = &u->open_confirm;
5386         __be32 status;
5387         struct nfs4_openowner *oo;
5388         struct nfs4_ol_stateid *stp;
5389         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5390
5391         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5392                         cstate->current_fh.fh_dentry);
5393
5394         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5395         if (status)
5396                 return status;
5397
5398         status = nfs4_preprocess_seqid_op(cstate,
5399                                         oc->oc_seqid, &oc->oc_req_stateid,
5400                                         NFS4_OPEN_STID, &stp, nn);
5401         if (status)
5402                 goto out;
5403         oo = openowner(stp->st_stateowner);
5404         status = nfserr_bad_stateid;
5405         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5406                 mutex_unlock(&stp->st_mutex);
5407                 goto put_stateid;
5408         }
5409         oo->oo_flags |= NFS4_OO_CONFIRMED;
5410         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5411         mutex_unlock(&stp->st_mutex);
5412         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5413                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5414
5415         nfsd4_client_record_create(oo->oo_owner.so_client);
5416         status = nfs_ok;
5417 put_stateid:
5418         nfs4_put_stid(&stp->st_stid);
5419 out:
5420         nfsd4_bump_seqid(cstate, status);
5421         return status;
5422 }
5423
5424 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5425 {
5426         if (!test_access(access, stp))
5427                 return;
5428         nfs4_file_put_access(stp->st_stid.sc_file, access);
5429         clear_access(access, stp);
5430 }
5431
5432 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5433 {
5434         switch (to_access) {
5435         case NFS4_SHARE_ACCESS_READ:
5436                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5437                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5438                 break;
5439         case NFS4_SHARE_ACCESS_WRITE:
5440                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5441                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5442                 break;
5443         case NFS4_SHARE_ACCESS_BOTH:
5444                 break;
5445         default:
5446                 WARN_ON_ONCE(1);
5447         }
5448 }
5449
5450 __be32
5451 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5452                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5453 {
5454         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5455         __be32 status;
5456         struct nfs4_ol_stateid *stp;
5457         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5458
5459         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5460                         cstate->current_fh.fh_dentry);
5461
5462         /* We don't yet support WANT bits: */
5463         if (od->od_deleg_want)
5464                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5465                         od->od_deleg_want);
5466
5467         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5468                                         &od->od_stateid, &stp, nn);
5469         if (status)
5470                 goto out;
5471         status = nfserr_inval;
5472         if (!test_access(od->od_share_access, stp)) {
5473                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5474                         stp->st_access_bmap, od->od_share_access);
5475                 goto put_stateid;
5476         }
5477         if (!test_deny(od->od_share_deny, stp)) {
5478                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5479                         stp->st_deny_bmap, od->od_share_deny);
5480                 goto put_stateid;
5481         }
5482         nfs4_stateid_downgrade(stp, od->od_share_access);
5483         reset_union_bmap_deny(od->od_share_deny, stp);
5484         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5485         status = nfs_ok;
5486 put_stateid:
5487         mutex_unlock(&stp->st_mutex);
5488         nfs4_put_stid(&stp->st_stid);
5489 out:
5490         nfsd4_bump_seqid(cstate, status);
5491         return status;
5492 }
5493
5494 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5495 {
5496         struct nfs4_client *clp = s->st_stid.sc_client;
5497         bool unhashed;
5498         LIST_HEAD(reaplist);
5499
5500         spin_lock(&clp->cl_lock);
5501         unhashed = unhash_open_stateid(s, &reaplist);
5502
5503         if (clp->cl_minorversion) {
5504                 if (unhashed)
5505                         put_ol_stateid_locked(s, &reaplist);
5506                 spin_unlock(&clp->cl_lock);
5507                 free_ol_stateid_reaplist(&reaplist);
5508         } else {
5509                 spin_unlock(&clp->cl_lock);
5510                 free_ol_stateid_reaplist(&reaplist);
5511                 if (unhashed)
5512                         move_to_close_lru(s, clp->net);
5513         }
5514 }
5515
5516 /*
5517  * nfs4_unlock_state() called after encode
5518  */
5519 __be32
5520 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5521                 union nfsd4_op_u *u)
5522 {
5523         struct nfsd4_close *close = &u->close;
5524         __be32 status;
5525         struct nfs4_ol_stateid *stp;
5526         struct net *net = SVC_NET(rqstp);
5527         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5528
5529         dprintk("NFSD: nfsd4_close on file %pd\n",
5530                         cstate->current_fh.fh_dentry);
5531
5532         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5533                                         &close->cl_stateid,
5534                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5535                                         &stp, nn);
5536         nfsd4_bump_seqid(cstate, status);
5537         if (status)
5538                 goto out;
5539
5540         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5541
5542         /*
5543          * Technically we don't _really_ have to increment or copy it, since
5544          * it should just be gone after this operation and we clobber the
5545          * copied value below, but we continue to do so here just to ensure
5546          * that racing ops see that there was a state change.
5547          */
5548         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5549
5550         nfsd4_close_open_stateid(stp);
5551         mutex_unlock(&stp->st_mutex);
5552
5553         /* v4.1+ suggests that we send a special stateid in here, since the
5554          * clients should just ignore this anyway. Since this is not useful
5555          * for v4.0 clients either, we set it to the special close_stateid
5556          * universally.
5557          *
5558          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5559          */
5560         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5561
5562         /* put reference from nfs4_preprocess_seqid_op */
5563         nfs4_put_stid(&stp->st_stid);
5564 out:
5565         return status;
5566 }
5567
5568 __be32
5569 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5570                   union nfsd4_op_u *u)
5571 {
5572         struct nfsd4_delegreturn *dr = &u->delegreturn;
5573         struct nfs4_delegation *dp;
5574         stateid_t *stateid = &dr->dr_stateid;
5575         struct nfs4_stid *s;
5576         __be32 status;
5577         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5578
5579         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5580                 return status;
5581
5582         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5583         if (status)
5584                 goto out;
5585         dp = delegstateid(s);
5586         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5587         if (status)
5588                 goto put_stateid;
5589
5590         destroy_delegation(dp);
5591 put_stateid:
5592         nfs4_put_stid(&dp->dl_stid);
5593 out:
5594         return status;
5595 }
5596
5597 static inline u64
5598 end_offset(u64 start, u64 len)
5599 {
5600         u64 end;
5601
5602         end = start + len;
5603         return end >= start ? end: NFS4_MAX_UINT64;
5604 }
5605
5606 /* last octet in a range */
5607 static inline u64
5608 last_byte_offset(u64 start, u64 len)
5609 {
5610         u64 end;
5611
5612         WARN_ON_ONCE(!len);
5613         end = start + len;
5614         return end > start ? end - 1: NFS4_MAX_UINT64;
5615 }
5616
5617 /*
5618  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5619  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5620  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5621  * locking, this prevents us from being completely protocol-compliant.  The
5622  * real solution to this problem is to start using unsigned file offsets in
5623  * the VFS, but this is a very deep change!
5624  */
5625 static inline void
5626 nfs4_transform_lock_offset(struct file_lock *lock)
5627 {
5628         if (lock->fl_start < 0)
5629                 lock->fl_start = OFFSET_MAX;
5630         if (lock->fl_end < 0)
5631                 lock->fl_end = OFFSET_MAX;
5632 }
5633
5634 static fl_owner_t
5635 nfsd4_fl_get_owner(fl_owner_t owner)
5636 {
5637         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5638
5639         nfs4_get_stateowner(&lo->lo_owner);
5640         return owner;
5641 }
5642
5643 static void
5644 nfsd4_fl_put_owner(fl_owner_t owner)
5645 {
5646         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5647
5648         if (lo)
5649                 nfs4_put_stateowner(&lo->lo_owner);
5650 }
5651
5652 static void
5653 nfsd4_lm_notify(struct file_lock *fl)
5654 {
5655         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5656         struct net                      *net = lo->lo_owner.so_client->net;
5657         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5658         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5659                                                 struct nfsd4_blocked_lock, nbl_lock);
5660         bool queue = false;
5661
5662         /* An empty list means that something else is going to be using it */
5663         spin_lock(&nn->blocked_locks_lock);
5664         if (!list_empty(&nbl->nbl_list)) {
5665                 list_del_init(&nbl->nbl_list);
5666                 list_del_init(&nbl->nbl_lru);
5667                 queue = true;
5668         }
5669         spin_unlock(&nn->blocked_locks_lock);
5670
5671         if (queue)
5672                 nfsd4_run_cb(&nbl->nbl_cb);
5673 }
5674
5675 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5676         .lm_notify = nfsd4_lm_notify,
5677         .lm_get_owner = nfsd4_fl_get_owner,
5678         .lm_put_owner = nfsd4_fl_put_owner,
5679 };
5680
5681 static inline void
5682 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5683 {
5684         struct nfs4_lockowner *lo;
5685
5686         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5687                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5688                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5689                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5690                 if (!deny->ld_owner.data)
5691                         /* We just don't care that much */
5692                         goto nevermind;
5693                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5694                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5695         } else {
5696 nevermind:
5697                 deny->ld_owner.len = 0;
5698                 deny->ld_owner.data = NULL;
5699                 deny->ld_clientid.cl_boot = 0;
5700                 deny->ld_clientid.cl_id = 0;
5701         }
5702         deny->ld_start = fl->fl_start;
5703         deny->ld_length = NFS4_MAX_UINT64;
5704         if (fl->fl_end != NFS4_MAX_UINT64)
5705                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5706         deny->ld_type = NFS4_READ_LT;
5707         if (fl->fl_type != F_RDLCK)
5708                 deny->ld_type = NFS4_WRITE_LT;
5709 }
5710
5711 static struct nfs4_lockowner *
5712 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5713 {
5714         unsigned int strhashval = ownerstr_hashval(owner);
5715         struct nfs4_stateowner *so;
5716
5717         lockdep_assert_held(&clp->cl_lock);
5718
5719         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5720                             so_strhash) {
5721                 if (so->so_is_open_owner)
5722                         continue;
5723                 if (same_owner_str(so, owner))
5724                         return lockowner(nfs4_get_stateowner(so));
5725         }
5726         return NULL;
5727 }
5728
5729 static struct nfs4_lockowner *
5730 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5731 {
5732         struct nfs4_lockowner *lo;
5733
5734         spin_lock(&clp->cl_lock);
5735         lo = find_lockowner_str_locked(clp, owner);
5736         spin_unlock(&clp->cl_lock);
5737         return lo;
5738 }
5739
5740 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5741 {
5742         unhash_lockowner_locked(lockowner(sop));
5743 }
5744
5745 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5746 {
5747         struct nfs4_lockowner *lo = lockowner(sop);
5748
5749         kmem_cache_free(lockowner_slab, lo);
5750 }
5751
5752 static const struct nfs4_stateowner_operations lockowner_ops = {
5753         .so_unhash =    nfs4_unhash_lockowner,
5754         .so_free =      nfs4_free_lockowner,
5755 };
5756
5757 /*
5758  * Alloc a lock owner structure.
5759  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5760  * occurred.
5761  *
5762  * strhashval = ownerstr_hashval
5763  */
5764 static struct nfs4_lockowner *
5765 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5766                            struct nfs4_ol_stateid *open_stp,
5767                            struct nfsd4_lock *lock)
5768 {
5769         struct nfs4_lockowner *lo, *ret;
5770
5771         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5772         if (!lo)
5773                 return NULL;
5774         INIT_LIST_HEAD(&lo->lo_blocked);
5775         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5776         lo->lo_owner.so_is_open_owner = 0;
5777         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5778         lo->lo_owner.so_ops = &lockowner_ops;
5779         spin_lock(&clp->cl_lock);
5780         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5781         if (ret == NULL) {
5782                 list_add(&lo->lo_owner.so_strhash,
5783                          &clp->cl_ownerstr_hashtbl[strhashval]);
5784                 ret = lo;
5785         } else
5786                 nfs4_free_stateowner(&lo->lo_owner);
5787
5788         spin_unlock(&clp->cl_lock);
5789         return ret;
5790 }
5791
5792 static struct nfs4_ol_stateid *
5793 find_lock_stateid(const struct nfs4_lockowner *lo,
5794                   const struct nfs4_ol_stateid *ost)
5795 {
5796         struct nfs4_ol_stateid *lst;
5797
5798         lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
5799
5800         /* If ost is not hashed, ost->st_locks will not be valid */
5801         if (!nfs4_ol_stateid_unhashed(ost))
5802                 list_for_each_entry(lst, &ost->st_locks, st_locks) {
5803                         if (lst->st_stateowner == &lo->lo_owner) {
5804                                 refcount_inc(&lst->st_stid.sc_count);
5805                                 return lst;
5806                         }
5807                 }
5808         return NULL;
5809 }
5810
5811 static struct nfs4_ol_stateid *
5812 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5813                   struct nfs4_file *fp, struct inode *inode,
5814                   struct nfs4_ol_stateid *open_stp)
5815 {
5816         struct nfs4_client *clp = lo->lo_owner.so_client;
5817         struct nfs4_ol_stateid *retstp;
5818
5819         mutex_init(&stp->st_mutex);
5820         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5821 retry:
5822         spin_lock(&clp->cl_lock);
5823         if (nfs4_ol_stateid_unhashed(open_stp))
5824                 goto out_close;
5825         retstp = find_lock_stateid(lo, open_stp);
5826         if (retstp)
5827                 goto out_found;
5828         refcount_inc(&stp->st_stid.sc_count);
5829         stp->st_stid.sc_type = NFS4_LOCK_STID;
5830         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5831         get_nfs4_file(fp);
5832         stp->st_stid.sc_file = fp;
5833         stp->st_access_bmap = 0;
5834         stp->st_deny_bmap = open_stp->st_deny_bmap;
5835         stp->st_openstp = open_stp;
5836         spin_lock(&fp->fi_lock);
5837         list_add(&stp->st_locks, &open_stp->st_locks);
5838         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5839         list_add(&stp->st_perfile, &fp->fi_stateids);
5840         spin_unlock(&fp->fi_lock);
5841         spin_unlock(&clp->cl_lock);
5842         return stp;
5843 out_found:
5844         spin_unlock(&clp->cl_lock);
5845         if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5846                 nfs4_put_stid(&retstp->st_stid);
5847                 goto retry;
5848         }
5849         /* To keep mutex tracking happy */
5850         mutex_unlock(&stp->st_mutex);
5851         return retstp;
5852 out_close:
5853         spin_unlock(&clp->cl_lock);
5854         mutex_unlock(&stp->st_mutex);
5855         return NULL;
5856 }
5857
5858 static struct nfs4_ol_stateid *
5859 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5860                             struct inode *inode, struct nfs4_ol_stateid *ost,
5861                             bool *new)
5862 {
5863         struct nfs4_stid *ns = NULL;
5864         struct nfs4_ol_stateid *lst;
5865         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5866         struct nfs4_client *clp = oo->oo_owner.so_client;
5867
5868         *new = false;
5869         spin_lock(&clp->cl_lock);
5870         lst = find_lock_stateid(lo, ost);
5871         spin_unlock(&clp->cl_lock);
5872         if (lst != NULL) {
5873                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5874                         goto out;
5875                 nfs4_put_stid(&lst->st_stid);
5876         }
5877         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5878         if (ns == NULL)
5879                 return NULL;
5880
5881         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5882         if (lst == openlockstateid(ns))
5883                 *new = true;
5884         else
5885                 nfs4_put_stid(ns);
5886 out:
5887         return lst;
5888 }
5889
5890 static int
5891 check_lock_length(u64 offset, u64 length)
5892 {
5893         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5894                 (length > ~offset)));
5895 }
5896
5897 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5898 {
5899         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5900
5901         lockdep_assert_held(&fp->fi_lock);
5902
5903         if (test_access(access, lock_stp))
5904                 return;
5905         __nfs4_file_get_access(fp, access);
5906         set_access(access, lock_stp);
5907 }
5908
5909 static __be32
5910 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5911                             struct nfs4_ol_stateid *ost,
5912                             struct nfsd4_lock *lock,
5913                             struct nfs4_ol_stateid **plst, bool *new)
5914 {
5915         __be32 status;
5916         struct nfs4_file *fi = ost->st_stid.sc_file;
5917         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5918         struct nfs4_client *cl = oo->oo_owner.so_client;
5919         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5920         struct nfs4_lockowner *lo;
5921         struct nfs4_ol_stateid *lst;
5922         unsigned int strhashval;
5923
5924         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5925         if (!lo) {
5926                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5927                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5928                 if (lo == NULL)
5929                         return nfserr_jukebox;
5930         } else {
5931                 /* with an existing lockowner, seqids must be the same */
5932                 status = nfserr_bad_seqid;
5933                 if (!cstate->minorversion &&
5934                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5935                         goto out;
5936         }
5937
5938         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5939         if (lst == NULL) {
5940                 status = nfserr_jukebox;
5941                 goto out;
5942         }
5943
5944         status = nfs_ok;
5945         *plst = lst;
5946 out:
5947         nfs4_put_stateowner(&lo->lo_owner);
5948         return status;
5949 }
5950
5951 /*
5952  *  LOCK operation
5953  */
5954 __be32
5955 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5956            union nfsd4_op_u *u)
5957 {
5958         struct nfsd4_lock *lock = &u->lock;
5959         struct nfs4_openowner *open_sop = NULL;
5960         struct nfs4_lockowner *lock_sop = NULL;
5961         struct nfs4_ol_stateid *lock_stp = NULL;
5962         struct nfs4_ol_stateid *open_stp = NULL;
5963         struct nfs4_file *fp;
5964         struct file *filp = NULL;
5965         struct nfsd4_blocked_lock *nbl = NULL;
5966         struct file_lock *file_lock = NULL;
5967         struct file_lock *conflock = NULL;
5968         __be32 status = 0;
5969         int lkflg;
5970         int err;
5971         bool new = false;
5972         unsigned char fl_type;
5973         unsigned int fl_flags = FL_POSIX;
5974         struct net *net = SVC_NET(rqstp);
5975         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5976
5977         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5978                 (long long) lock->lk_offset,
5979                 (long long) lock->lk_length);
5980
5981         if (check_lock_length(lock->lk_offset, lock->lk_length))
5982                  return nfserr_inval;
5983
5984         if ((status = fh_verify(rqstp, &cstate->current_fh,
5985                                 S_IFREG, NFSD_MAY_LOCK))) {
5986                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5987                 return status;
5988         }
5989
5990         if (lock->lk_is_new) {
5991                 if (nfsd4_has_session(cstate))
5992                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5993                         memcpy(&lock->lk_new_clientid,
5994                                 &cstate->session->se_client->cl_clientid,
5995                                 sizeof(clientid_t));
5996
5997                 status = nfserr_stale_clientid;
5998                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5999                         goto out;
6000
6001                 /* validate and update open stateid and open seqid */
6002                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6003                                         lock->lk_new_open_seqid,
6004                                         &lock->lk_new_open_stateid,
6005                                         &open_stp, nn);
6006                 if (status)
6007                         goto out;
6008                 mutex_unlock(&open_stp->st_mutex);
6009                 open_sop = openowner(open_stp->st_stateowner);
6010                 status = nfserr_bad_stateid;
6011                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6012                                                 &lock->lk_new_clientid))
6013                         goto out;
6014                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6015                                                         &lock_stp, &new);
6016         } else {
6017                 status = nfs4_preprocess_seqid_op(cstate,
6018                                        lock->lk_old_lock_seqid,
6019                                        &lock->lk_old_lock_stateid,
6020                                        NFS4_LOCK_STID, &lock_stp, nn);
6021         }
6022         if (status)
6023                 goto out;
6024         lock_sop = lockowner(lock_stp->st_stateowner);
6025
6026         lkflg = setlkflg(lock->lk_type);
6027         status = nfs4_check_openmode(lock_stp, lkflg);
6028         if (status)
6029                 goto out;
6030
6031         status = nfserr_grace;
6032         if (locks_in_grace(net) && !lock->lk_reclaim)
6033                 goto out;
6034         status = nfserr_no_grace;
6035         if (!locks_in_grace(net) && lock->lk_reclaim)
6036                 goto out;
6037
6038         fp = lock_stp->st_stid.sc_file;
6039         switch (lock->lk_type) {
6040                 case NFS4_READW_LT:
6041                         if (nfsd4_has_session(cstate))
6042                                 fl_flags |= FL_SLEEP;
6043                         /* Fallthrough */
6044                 case NFS4_READ_LT:
6045                         spin_lock(&fp->fi_lock);
6046                         filp = find_readable_file_locked(fp);
6047                         if (filp)
6048                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6049                         spin_unlock(&fp->fi_lock);
6050                         fl_type = F_RDLCK;
6051                         break;
6052                 case NFS4_WRITEW_LT:
6053                         if (nfsd4_has_session(cstate))
6054                                 fl_flags |= FL_SLEEP;
6055                         /* Fallthrough */
6056                 case NFS4_WRITE_LT:
6057                         spin_lock(&fp->fi_lock);
6058                         filp = find_writeable_file_locked(fp);
6059                         if (filp)
6060                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6061                         spin_unlock(&fp->fi_lock);
6062                         fl_type = F_WRLCK;
6063                         break;
6064                 default:
6065                         status = nfserr_inval;
6066                 goto out;
6067         }
6068
6069         if (!filp) {
6070                 status = nfserr_openmode;
6071                 goto out;
6072         }
6073
6074         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6075         if (!nbl) {
6076                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6077                 status = nfserr_jukebox;
6078                 goto out;
6079         }
6080
6081         file_lock = &nbl->nbl_lock;
6082         file_lock->fl_type = fl_type;
6083         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6084         file_lock->fl_pid = current->tgid;
6085         file_lock->fl_file = filp;
6086         file_lock->fl_flags = fl_flags;
6087         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6088         file_lock->fl_start = lock->lk_offset;
6089         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6090         nfs4_transform_lock_offset(file_lock);
6091
6092         conflock = locks_alloc_lock();
6093         if (!conflock) {
6094                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6095                 status = nfserr_jukebox;
6096                 goto out;
6097         }
6098
6099         if (fl_flags & FL_SLEEP) {
6100                 nbl->nbl_time = get_seconds();
6101                 spin_lock(&nn->blocked_locks_lock);
6102                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6103                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6104                 spin_unlock(&nn->blocked_locks_lock);
6105         }
6106
6107         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6108         switch (err) {
6109         case 0: /* success! */
6110                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6111                 status = 0;
6112                 if (lock->lk_reclaim)
6113                         nn->somebody_reclaimed = true;
6114                 break;
6115         case FILE_LOCK_DEFERRED:
6116                 nbl = NULL;
6117                 /* Fallthrough */
6118         case -EAGAIN:           /* conflock holds conflicting lock */
6119                 status = nfserr_denied;
6120                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6121                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6122                 break;
6123         case -EDEADLK:
6124                 status = nfserr_deadlock;
6125                 break;
6126         default:
6127                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6128                 status = nfserrno(err);
6129                 break;
6130         }
6131 out:
6132         if (nbl) {
6133                 /* dequeue it if we queued it before */
6134                 if (fl_flags & FL_SLEEP) {
6135                         spin_lock(&nn->blocked_locks_lock);
6136                         list_del_init(&nbl->nbl_list);
6137                         list_del_init(&nbl->nbl_lru);
6138                         spin_unlock(&nn->blocked_locks_lock);
6139                 }
6140                 free_blocked_lock(nbl);
6141         }
6142         if (filp)
6143                 fput(filp);
6144         if (lock_stp) {
6145                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6146                 if (cstate->replay_owner &&
6147                     cstate->replay_owner != &lock_sop->lo_owner &&
6148                     seqid_mutating_err(ntohl(status)))
6149                         lock_sop->lo_owner.so_seqid++;
6150
6151                 /*
6152                  * If this is a new, never-before-used stateid, and we are
6153                  * returning an error, then just go ahead and release it.
6154                  */
6155                 if (status && new)
6156                         release_lock_stateid(lock_stp);
6157
6158                 mutex_unlock(&lock_stp->st_mutex);
6159
6160                 nfs4_put_stid(&lock_stp->st_stid);
6161         }
6162         if (open_stp)
6163                 nfs4_put_stid(&open_stp->st_stid);
6164         nfsd4_bump_seqid(cstate, status);
6165         if (conflock)
6166                 locks_free_lock(conflock);
6167         return status;
6168 }
6169
6170 /*
6171  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6172  * so we do a temporary open here just to get an open file to pass to
6173  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6174  * inode operation.)
6175  */
6176 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6177 {
6178         struct file *file;
6179         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6180         if (!err) {
6181                 err = nfserrno(vfs_test_lock(file, lock));
6182                 fput(file);
6183         }
6184         return err;
6185 }
6186
6187 /*
6188  * LOCKT operation
6189  */
6190 __be32
6191 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6192             union nfsd4_op_u *u)
6193 {
6194         struct nfsd4_lockt *lockt = &u->lockt;
6195         struct file_lock *file_lock = NULL;
6196         struct nfs4_lockowner *lo = NULL;
6197         __be32 status;
6198         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6199
6200         if (locks_in_grace(SVC_NET(rqstp)))
6201                 return nfserr_grace;
6202
6203         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6204                  return nfserr_inval;
6205
6206         if (!nfsd4_has_session(cstate)) {
6207                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6208                 if (status)
6209                         goto out;
6210         }
6211
6212         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6213                 goto out;
6214
6215         file_lock = locks_alloc_lock();
6216         if (!file_lock) {
6217                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6218                 status = nfserr_jukebox;
6219                 goto out;
6220         }
6221
6222         switch (lockt->lt_type) {
6223                 case NFS4_READ_LT:
6224                 case NFS4_READW_LT:
6225                         file_lock->fl_type = F_RDLCK;
6226                 break;
6227                 case NFS4_WRITE_LT:
6228                 case NFS4_WRITEW_LT:
6229                         file_lock->fl_type = F_WRLCK;
6230                 break;
6231                 default:
6232                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6233                         status = nfserr_inval;
6234                 goto out;
6235         }
6236
6237         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6238         if (lo)
6239                 file_lock->fl_owner = (fl_owner_t)lo;
6240         file_lock->fl_pid = current->tgid;
6241         file_lock->fl_flags = FL_POSIX;
6242
6243         file_lock->fl_start = lockt->lt_offset;
6244         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6245
6246         nfs4_transform_lock_offset(file_lock);
6247
6248         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6249         if (status)
6250                 goto out;
6251
6252         if (file_lock->fl_type != F_UNLCK) {
6253                 status = nfserr_denied;
6254                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6255         }
6256 out:
6257         if (lo)
6258                 nfs4_put_stateowner(&lo->lo_owner);
6259         if (file_lock)
6260                 locks_free_lock(file_lock);
6261         return status;
6262 }
6263
6264 __be32
6265 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6266             union nfsd4_op_u *u)
6267 {
6268         struct nfsd4_locku *locku = &u->locku;
6269         struct nfs4_ol_stateid *stp;
6270         struct file *filp = NULL;
6271         struct file_lock *file_lock = NULL;
6272         __be32 status;
6273         int err;
6274         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6275
6276         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6277                 (long long) locku->lu_offset,
6278                 (long long) locku->lu_length);
6279
6280         if (check_lock_length(locku->lu_offset, locku->lu_length))
6281                  return nfserr_inval;
6282
6283         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6284                                         &locku->lu_stateid, NFS4_LOCK_STID,
6285                                         &stp, nn);
6286         if (status)
6287                 goto out;
6288         filp = find_any_file(stp->st_stid.sc_file);
6289         if (!filp) {
6290                 status = nfserr_lock_range;
6291                 goto put_stateid;
6292         }
6293         file_lock = locks_alloc_lock();
6294         if (!file_lock) {
6295                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6296                 status = nfserr_jukebox;
6297                 goto fput;
6298         }
6299
6300         file_lock->fl_type = F_UNLCK;
6301         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6302         file_lock->fl_pid = current->tgid;
6303         file_lock->fl_file = filp;
6304         file_lock->fl_flags = FL_POSIX;
6305         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6306         file_lock->fl_start = locku->lu_offset;
6307
6308         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6309                                                 locku->lu_length);
6310         nfs4_transform_lock_offset(file_lock);
6311
6312         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6313         if (err) {
6314                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6315                 goto out_nfserr;
6316         }
6317         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6318 fput:
6319         fput(filp);
6320 put_stateid:
6321         mutex_unlock(&stp->st_mutex);
6322         nfs4_put_stid(&stp->st_stid);
6323 out:
6324         nfsd4_bump_seqid(cstate, status);
6325         if (file_lock)
6326                 locks_free_lock(file_lock);
6327         return status;
6328
6329 out_nfserr:
6330         status = nfserrno(err);
6331         goto fput;
6332 }
6333
6334 /*
6335  * returns
6336  *      true:  locks held by lockowner
6337  *      false: no locks held by lockowner
6338  */
6339 static bool
6340 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6341 {
6342         struct file_lock *fl;
6343         int status = false;
6344         struct file *filp = find_any_file(fp);
6345         struct inode *inode;
6346         struct file_lock_context *flctx;
6347
6348         if (!filp) {
6349                 /* Any valid lock stateid should have some sort of access */
6350                 WARN_ON_ONCE(1);
6351                 return status;
6352         }
6353
6354         inode = locks_inode(filp);
6355         flctx = inode->i_flctx;
6356
6357         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6358                 spin_lock(&flctx->flc_lock);
6359                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6360                         if (fl->fl_owner == (fl_owner_t)lowner) {
6361                                 status = true;
6362                                 break;
6363                         }
6364                 }
6365                 spin_unlock(&flctx->flc_lock);
6366         }
6367         fput(filp);
6368         return status;
6369 }
6370
6371 __be32
6372 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6373                         struct nfsd4_compound_state *cstate,
6374                         union nfsd4_op_u *u)
6375 {
6376         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6377         clientid_t *clid = &rlockowner->rl_clientid;
6378         struct nfs4_stateowner *sop;
6379         struct nfs4_lockowner *lo = NULL;
6380         struct nfs4_ol_stateid *stp;
6381         struct xdr_netobj *owner = &rlockowner->rl_owner;
6382         unsigned int hashval = ownerstr_hashval(owner);
6383         __be32 status;
6384         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6385         struct nfs4_client *clp;
6386         LIST_HEAD (reaplist);
6387
6388         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6389                 clid->cl_boot, clid->cl_id);
6390
6391         status = lookup_clientid(clid, cstate, nn);
6392         if (status)
6393                 return status;
6394
6395         clp = cstate->clp;
6396         /* Find the matching lock stateowner */
6397         spin_lock(&clp->cl_lock);
6398         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6399                             so_strhash) {
6400
6401                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6402                         continue;
6403
6404                 if (atomic_read(&sop->so_count) != 1) {
6405                         spin_unlock(&clp->cl_lock);
6406                         return nfserr_locks_held;
6407                 }
6408
6409                 lo = lockowner(sop);
6410                 nfs4_get_stateowner(sop);
6411                 break;
6412         }
6413         if (!lo) {
6414                 spin_unlock(&clp->cl_lock);
6415                 return status;
6416         }
6417
6418         unhash_lockowner_locked(lo);
6419         while (!list_empty(&lo->lo_owner.so_stateids)) {
6420                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6421                                        struct nfs4_ol_stateid,
6422                                        st_perstateowner);
6423                 WARN_ON(!unhash_lock_stateid(stp));
6424                 put_ol_stateid_locked(stp, &reaplist);
6425         }
6426         spin_unlock(&clp->cl_lock);
6427         free_ol_stateid_reaplist(&reaplist);
6428         remove_blocked_locks(lo);
6429         nfs4_put_stateowner(&lo->lo_owner);
6430
6431         return status;
6432 }
6433
6434 static inline struct nfs4_client_reclaim *
6435 alloc_reclaim(void)
6436 {
6437         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6438 }
6439
6440 bool
6441 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6442 {
6443         struct nfs4_client_reclaim *crp;
6444
6445         crp = nfsd4_find_reclaim_client(name, nn);
6446         return (crp && crp->cr_clp);
6447 }
6448
6449 /*
6450  * failure => all reset bets are off, nfserr_no_grace...
6451  */
6452 struct nfs4_client_reclaim *
6453 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6454 {
6455         unsigned int strhashval;
6456         struct nfs4_client_reclaim *crp;
6457
6458         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6459         crp = alloc_reclaim();
6460         if (crp) {
6461                 strhashval = clientstr_hashval(name);
6462                 INIT_LIST_HEAD(&crp->cr_strhash);
6463                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6464                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6465                 crp->cr_clp = NULL;
6466                 nn->reclaim_str_hashtbl_size++;
6467         }
6468         return crp;
6469 }
6470
6471 void
6472 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6473 {
6474         list_del(&crp->cr_strhash);
6475         kfree(crp);
6476         nn->reclaim_str_hashtbl_size--;
6477 }
6478
6479 void
6480 nfs4_release_reclaim(struct nfsd_net *nn)
6481 {
6482         struct nfs4_client_reclaim *crp = NULL;
6483         int i;
6484
6485         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6486                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6487                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6488                                         struct nfs4_client_reclaim, cr_strhash);
6489                         nfs4_remove_reclaim_record(crp, nn);
6490                 }
6491         }
6492         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6493 }
6494
6495 /*
6496  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6497 struct nfs4_client_reclaim *
6498 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6499 {
6500         unsigned int strhashval;
6501         struct nfs4_client_reclaim *crp = NULL;
6502
6503         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6504
6505         strhashval = clientstr_hashval(recdir);
6506         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6507                 if (same_name(crp->cr_recdir, recdir)) {
6508                         return crp;
6509                 }
6510         }
6511         return NULL;
6512 }
6513
6514 /*
6515 * Called from OPEN. Look for clientid in reclaim list.
6516 */
6517 __be32
6518 nfs4_check_open_reclaim(clientid_t *clid,
6519                 struct nfsd4_compound_state *cstate,
6520                 struct nfsd_net *nn)
6521 {
6522         __be32 status;
6523
6524         /* find clientid in conf_id_hashtbl */
6525         status = lookup_clientid(clid, cstate, nn);
6526         if (status)
6527                 return nfserr_reclaim_bad;
6528
6529         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6530                 return nfserr_no_grace;
6531
6532         if (nfsd4_client_record_check(cstate->clp))
6533                 return nfserr_reclaim_bad;
6534
6535         return nfs_ok;
6536 }
6537
6538 #ifdef CONFIG_NFSD_FAULT_INJECTION
6539 static inline void
6540 put_client(struct nfs4_client *clp)
6541 {
6542         atomic_dec(&clp->cl_refcount);
6543 }
6544
6545 static struct nfs4_client *
6546 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6547 {
6548         struct nfs4_client *clp;
6549         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6550                                           nfsd_net_id);
6551
6552         if (!nfsd_netns_ready(nn))
6553                 return NULL;
6554
6555         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6556                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6557                         return clp;
6558         }
6559         return NULL;
6560 }
6561
6562 u64
6563 nfsd_inject_print_clients(void)
6564 {
6565         struct nfs4_client *clp;
6566         u64 count = 0;
6567         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6568                                           nfsd_net_id);
6569         char buf[INET6_ADDRSTRLEN];
6570
6571         if (!nfsd_netns_ready(nn))
6572                 return 0;
6573
6574         spin_lock(&nn->client_lock);
6575         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6576                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6577                 pr_info("NFS Client: %s\n", buf);
6578                 ++count;
6579         }
6580         spin_unlock(&nn->client_lock);
6581
6582         return count;
6583 }
6584
6585 u64
6586 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6587 {
6588         u64 count = 0;
6589         struct nfs4_client *clp;
6590         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6591                                           nfsd_net_id);
6592
6593         if (!nfsd_netns_ready(nn))
6594                 return count;
6595
6596         spin_lock(&nn->client_lock);
6597         clp = nfsd_find_client(addr, addr_size);
6598         if (clp) {
6599                 if (mark_client_expired_locked(clp) == nfs_ok)
6600                         ++count;
6601                 else
6602                         clp = NULL;
6603         }
6604         spin_unlock(&nn->client_lock);
6605
6606         if (clp)
6607                 expire_client(clp);
6608
6609         return count;
6610 }
6611
6612 u64
6613 nfsd_inject_forget_clients(u64 max)
6614 {
6615         u64 count = 0;
6616         struct nfs4_client *clp, *next;
6617         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6618                                                 nfsd_net_id);
6619         LIST_HEAD(reaplist);
6620
6621         if (!nfsd_netns_ready(nn))
6622                 return count;
6623
6624         spin_lock(&nn->client_lock);
6625         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6626                 if (mark_client_expired_locked(clp) == nfs_ok) {
6627                         list_add(&clp->cl_lru, &reaplist);
6628                         if (max != 0 && ++count >= max)
6629                                 break;
6630                 }
6631         }
6632         spin_unlock(&nn->client_lock);
6633
6634         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6635                 expire_client(clp);
6636
6637         return count;
6638 }
6639
6640 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6641                              const char *type)
6642 {
6643         char buf[INET6_ADDRSTRLEN];
6644         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6645         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6646 }
6647
6648 static void
6649 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6650                              struct list_head *collect)
6651 {
6652         struct nfs4_client *clp = lst->st_stid.sc_client;
6653         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6654                                           nfsd_net_id);
6655
6656         if (!collect)
6657                 return;
6658
6659         lockdep_assert_held(&nn->client_lock);
6660         atomic_inc(&clp->cl_refcount);
6661         list_add(&lst->st_locks, collect);
6662 }
6663
6664 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6665                                     struct list_head *collect,
6666                                     bool (*func)(struct nfs4_ol_stateid *))
6667 {
6668         struct nfs4_openowner *oop;
6669         struct nfs4_ol_stateid *stp, *st_next;
6670         struct nfs4_ol_stateid *lst, *lst_next;
6671         u64 count = 0;
6672
6673         spin_lock(&clp->cl_lock);
6674         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6675                 list_for_each_entry_safe(stp, st_next,
6676                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6677                         list_for_each_entry_safe(lst, lst_next,
6678                                         &stp->st_locks, st_locks) {
6679                                 if (func) {
6680                                         if (func(lst))
6681                                                 nfsd_inject_add_lock_to_list(lst,
6682                                                                         collect);
6683                                 }
6684                                 ++count;
6685                                 /*
6686                                  * Despite the fact that these functions deal
6687                                  * with 64-bit integers for "count", we must
6688                                  * ensure that it doesn't blow up the
6689                                  * clp->cl_refcount. Throw a warning if we
6690                                  * start to approach INT_MAX here.
6691                                  */
6692                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6693                                 if (count == max)
6694                                         goto out;
6695                         }
6696                 }
6697         }
6698 out:
6699         spin_unlock(&clp->cl_lock);
6700
6701         return count;
6702 }
6703
6704 static u64
6705 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6706                           u64 max)
6707 {
6708         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6709 }
6710
6711 static u64
6712 nfsd_print_client_locks(struct nfs4_client *clp)
6713 {
6714         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6715         nfsd_print_count(clp, count, "locked files");
6716         return count;
6717 }
6718
6719 u64
6720 nfsd_inject_print_locks(void)
6721 {
6722         struct nfs4_client *clp;
6723         u64 count = 0;
6724         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6725                                                 nfsd_net_id);
6726
6727         if (!nfsd_netns_ready(nn))
6728                 return 0;
6729
6730         spin_lock(&nn->client_lock);
6731         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6732                 count += nfsd_print_client_locks(clp);
6733         spin_unlock(&nn->client_lock);
6734
6735         return count;
6736 }
6737
6738 static void
6739 nfsd_reap_locks(struct list_head *reaplist)
6740 {
6741         struct nfs4_client *clp;
6742         struct nfs4_ol_stateid *stp, *next;
6743
6744         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6745                 list_del_init(&stp->st_locks);
6746                 clp = stp->st_stid.sc_client;
6747                 nfs4_put_stid(&stp->st_stid);
6748                 put_client(clp);
6749         }
6750 }
6751
6752 u64
6753 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6754 {
6755         unsigned int count = 0;
6756         struct nfs4_client *clp;
6757         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6758                                                 nfsd_net_id);
6759         LIST_HEAD(reaplist);
6760
6761         if (!nfsd_netns_ready(nn))
6762                 return count;
6763
6764         spin_lock(&nn->client_lock);
6765         clp = nfsd_find_client(addr, addr_size);
6766         if (clp)
6767                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6768         spin_unlock(&nn->client_lock);
6769         nfsd_reap_locks(&reaplist);
6770         return count;
6771 }
6772
6773 u64
6774 nfsd_inject_forget_locks(u64 max)
6775 {
6776         u64 count = 0;
6777         struct nfs4_client *clp;
6778         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6779                                                 nfsd_net_id);
6780         LIST_HEAD(reaplist);
6781
6782         if (!nfsd_netns_ready(nn))
6783                 return count;
6784
6785         spin_lock(&nn->client_lock);
6786         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6787                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6788                 if (max != 0 && count >= max)
6789                         break;
6790         }
6791         spin_unlock(&nn->client_lock);
6792         nfsd_reap_locks(&reaplist);
6793         return count;
6794 }
6795
6796 static u64
6797 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6798                               struct list_head *collect,
6799                               void (*func)(struct nfs4_openowner *))
6800 {
6801         struct nfs4_openowner *oop, *next;
6802         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6803                                                 nfsd_net_id);
6804         u64 count = 0;
6805
6806         lockdep_assert_held(&nn->client_lock);
6807
6808         spin_lock(&clp->cl_lock);
6809         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6810                 if (func) {
6811                         func(oop);
6812                         if (collect) {
6813                                 atomic_inc(&clp->cl_refcount);
6814                                 list_add(&oop->oo_perclient, collect);
6815                         }
6816                 }
6817                 ++count;
6818                 /*
6819                  * Despite the fact that these functions deal with
6820                  * 64-bit integers for "count", we must ensure that
6821                  * it doesn't blow up the clp->cl_refcount. Throw a
6822                  * warning if we start to approach INT_MAX here.
6823                  */
6824                 WARN_ON_ONCE(count == (INT_MAX / 2));
6825                 if (count == max)
6826                         break;
6827         }
6828         spin_unlock(&clp->cl_lock);
6829
6830         return count;
6831 }
6832
6833 static u64
6834 nfsd_print_client_openowners(struct nfs4_client *clp)
6835 {
6836         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6837
6838         nfsd_print_count(clp, count, "openowners");
6839         return count;
6840 }
6841
6842 static u64
6843 nfsd_collect_client_openowners(struct nfs4_client *clp,
6844                                struct list_head *collect, u64 max)
6845 {
6846         return nfsd_foreach_client_openowner(clp, max, collect,
6847                                                 unhash_openowner_locked);
6848 }
6849
6850 u64
6851 nfsd_inject_print_openowners(void)
6852 {
6853         struct nfs4_client *clp;
6854         u64 count = 0;
6855         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6856                                                 nfsd_net_id);
6857
6858         if (!nfsd_netns_ready(nn))
6859                 return 0;
6860
6861         spin_lock(&nn->client_lock);
6862         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6863                 count += nfsd_print_client_openowners(clp);
6864         spin_unlock(&nn->client_lock);
6865
6866         return count;
6867 }
6868
6869 static void
6870 nfsd_reap_openowners(struct list_head *reaplist)
6871 {
6872         struct nfs4_client *clp;
6873         struct nfs4_openowner *oop, *next;
6874
6875         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6876                 list_del_init(&oop->oo_perclient);
6877                 clp = oop->oo_owner.so_client;
6878                 release_openowner(oop);
6879                 put_client(clp);
6880         }
6881 }
6882
6883 u64
6884 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6885                                      size_t addr_size)
6886 {
6887         unsigned int count = 0;
6888         struct nfs4_client *clp;
6889         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6890                                                 nfsd_net_id);
6891         LIST_HEAD(reaplist);
6892
6893         if (!nfsd_netns_ready(nn))
6894                 return count;
6895
6896         spin_lock(&nn->client_lock);
6897         clp = nfsd_find_client(addr, addr_size);
6898         if (clp)
6899                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6900         spin_unlock(&nn->client_lock);
6901         nfsd_reap_openowners(&reaplist);
6902         return count;
6903 }
6904
6905 u64
6906 nfsd_inject_forget_openowners(u64 max)
6907 {
6908         u64 count = 0;
6909         struct nfs4_client *clp;
6910         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6911                                                 nfsd_net_id);
6912         LIST_HEAD(reaplist);
6913
6914         if (!nfsd_netns_ready(nn))
6915                 return count;
6916
6917         spin_lock(&nn->client_lock);
6918         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6919                 count += nfsd_collect_client_openowners(clp, &reaplist,
6920                                                         max - count);
6921                 if (max != 0 && count >= max)
6922                         break;
6923         }
6924         spin_unlock(&nn->client_lock);
6925         nfsd_reap_openowners(&reaplist);
6926         return count;
6927 }
6928
6929 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6930                                      struct list_head *victims)
6931 {
6932         struct nfs4_delegation *dp, *next;
6933         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6934                                                 nfsd_net_id);
6935         u64 count = 0;
6936
6937         lockdep_assert_held(&nn->client_lock);
6938
6939         spin_lock(&state_lock);
6940         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6941                 if (victims) {
6942                         /*
6943                          * It's not safe to mess with delegations that have a
6944                          * non-zero dl_time. They might have already been broken
6945                          * and could be processed by the laundromat outside of
6946                          * the state_lock. Just leave them be.
6947                          */
6948                         if (dp->dl_time != 0)
6949                                 continue;
6950
6951                         atomic_inc(&clp->cl_refcount);
6952                         WARN_ON(!unhash_delegation_locked(dp));
6953                         list_add(&dp->dl_recall_lru, victims);
6954                 }
6955                 ++count;
6956                 /*
6957                  * Despite the fact that these functions deal with
6958                  * 64-bit integers for "count", we must ensure that
6959                  * it doesn't blow up the clp->cl_refcount. Throw a
6960                  * warning if we start to approach INT_MAX here.
6961                  */
6962                 WARN_ON_ONCE(count == (INT_MAX / 2));
6963                 if (count == max)
6964                         break;
6965         }
6966         spin_unlock(&state_lock);
6967         return count;
6968 }
6969
6970 static u64
6971 nfsd_print_client_delegations(struct nfs4_client *clp)
6972 {
6973         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6974
6975         nfsd_print_count(clp, count, "delegations");
6976         return count;
6977 }
6978
6979 u64
6980 nfsd_inject_print_delegations(void)
6981 {
6982         struct nfs4_client *clp;
6983         u64 count = 0;
6984         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6985                                                 nfsd_net_id);
6986
6987         if (!nfsd_netns_ready(nn))
6988                 return 0;
6989
6990         spin_lock(&nn->client_lock);
6991         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6992                 count += nfsd_print_client_delegations(clp);
6993         spin_unlock(&nn->client_lock);
6994
6995         return count;
6996 }
6997
6998 static void
6999 nfsd_forget_delegations(struct list_head *reaplist)
7000 {
7001         struct nfs4_client *clp;
7002         struct nfs4_delegation *dp, *next;
7003
7004         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7005                 list_del_init(&dp->dl_recall_lru);
7006                 clp = dp->dl_stid.sc_client;
7007                 revoke_delegation(dp);
7008                 put_client(clp);
7009         }
7010 }
7011
7012 u64
7013 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7014                                       size_t addr_size)
7015 {
7016         u64 count = 0;
7017         struct nfs4_client *clp;
7018         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7019                                                 nfsd_net_id);
7020         LIST_HEAD(reaplist);
7021
7022         if (!nfsd_netns_ready(nn))
7023                 return count;
7024
7025         spin_lock(&nn->client_lock);
7026         clp = nfsd_find_client(addr, addr_size);
7027         if (clp)
7028                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7029         spin_unlock(&nn->client_lock);
7030
7031         nfsd_forget_delegations(&reaplist);
7032         return count;
7033 }
7034
7035 u64
7036 nfsd_inject_forget_delegations(u64 max)
7037 {
7038         u64 count = 0;
7039         struct nfs4_client *clp;
7040         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7041                                                 nfsd_net_id);
7042         LIST_HEAD(reaplist);
7043
7044         if (!nfsd_netns_ready(nn))
7045                 return count;
7046
7047         spin_lock(&nn->client_lock);
7048         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7049                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7050                 if (max != 0 && count >= max)
7051                         break;
7052         }
7053         spin_unlock(&nn->client_lock);
7054         nfsd_forget_delegations(&reaplist);
7055         return count;
7056 }
7057
7058 static void
7059 nfsd_recall_delegations(struct list_head *reaplist)
7060 {
7061         struct nfs4_client *clp;
7062         struct nfs4_delegation *dp, *next;
7063
7064         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7065                 list_del_init(&dp->dl_recall_lru);
7066                 clp = dp->dl_stid.sc_client;
7067                 /*
7068                  * We skipped all entries that had a zero dl_time before,
7069                  * so we can now reset the dl_time back to 0. If a delegation
7070                  * break comes in now, then it won't make any difference since
7071                  * we're recalling it either way.
7072                  */
7073                 spin_lock(&state_lock);
7074                 dp->dl_time = 0;
7075                 spin_unlock(&state_lock);
7076                 nfsd_break_one_deleg(dp);
7077                 put_client(clp);
7078         }
7079 }
7080
7081 u64
7082 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7083                                       size_t addr_size)
7084 {
7085         u64 count = 0;
7086         struct nfs4_client *clp;
7087         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7088                                                 nfsd_net_id);
7089         LIST_HEAD(reaplist);
7090
7091         if (!nfsd_netns_ready(nn))
7092                 return count;
7093
7094         spin_lock(&nn->client_lock);
7095         clp = nfsd_find_client(addr, addr_size);
7096         if (clp)
7097                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7098         spin_unlock(&nn->client_lock);
7099
7100         nfsd_recall_delegations(&reaplist);
7101         return count;
7102 }
7103
7104 u64
7105 nfsd_inject_recall_delegations(u64 max)
7106 {
7107         u64 count = 0;
7108         struct nfs4_client *clp, *next;
7109         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7110                                                 nfsd_net_id);
7111         LIST_HEAD(reaplist);
7112
7113         if (!nfsd_netns_ready(nn))
7114                 return count;
7115
7116         spin_lock(&nn->client_lock);
7117         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7118                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7119                 if (max != 0 && ++count >= max)
7120                         break;
7121         }
7122         spin_unlock(&nn->client_lock);
7123         nfsd_recall_delegations(&reaplist);
7124         return count;
7125 }
7126 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7127
7128 /*
7129  * Since the lifetime of a delegation isn't limited to that of an open, a
7130  * client may quite reasonably hang on to a delegation as long as it has
7131  * the inode cached.  This becomes an obvious problem the first time a
7132  * client's inode cache approaches the size of the server's total memory.
7133  *
7134  * For now we avoid this problem by imposing a hard limit on the number
7135  * of delegations, which varies according to the server's memory size.
7136  */
7137 static void
7138 set_max_delegations(void)
7139 {
7140         /*
7141          * Allow at most 4 delegations per megabyte of RAM.  Quick
7142          * estimates suggest that in the worst case (where every delegation
7143          * is for a different inode), a delegation could take about 1.5K,
7144          * giving a worst case usage of about 6% of memory.
7145          */
7146         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7147 }
7148
7149 static int nfs4_state_create_net(struct net *net)
7150 {
7151         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7152         int i;
7153
7154         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7155                                             sizeof(struct list_head),
7156                                             GFP_KERNEL);
7157         if (!nn->conf_id_hashtbl)
7158                 goto err;
7159         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7160                                               sizeof(struct list_head),
7161                                               GFP_KERNEL);
7162         if (!nn->unconf_id_hashtbl)
7163                 goto err_unconf_id;
7164         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7165                                               sizeof(struct list_head),
7166                                               GFP_KERNEL);
7167         if (!nn->sessionid_hashtbl)
7168                 goto err_sessionid;
7169
7170         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7171                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7172                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7173         }
7174         for (i = 0; i < SESSION_HASH_SIZE; i++)
7175                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7176         nn->conf_name_tree = RB_ROOT;
7177         nn->unconf_name_tree = RB_ROOT;
7178         nn->boot_time = get_seconds();
7179         nn->grace_ended = false;
7180         nn->nfsd4_manager.block_opens = true;
7181         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7182         INIT_LIST_HEAD(&nn->client_lru);
7183         INIT_LIST_HEAD(&nn->close_lru);
7184         INIT_LIST_HEAD(&nn->del_recall_lru);
7185         spin_lock_init(&nn->client_lock);
7186
7187         spin_lock_init(&nn->blocked_locks_lock);
7188         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7189
7190         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7191         get_net(net);
7192
7193         return 0;
7194
7195 err_sessionid:
7196         kfree(nn->unconf_id_hashtbl);
7197 err_unconf_id:
7198         kfree(nn->conf_id_hashtbl);
7199 err:
7200         return -ENOMEM;
7201 }
7202
7203 static void
7204 nfs4_state_destroy_net(struct net *net)
7205 {
7206         int i;
7207         struct nfs4_client *clp = NULL;
7208         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7209
7210         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7211                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7212                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7213                         destroy_client(clp);
7214                 }
7215         }
7216
7217         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7218
7219         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7220                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7221                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7222                         destroy_client(clp);
7223                 }
7224         }
7225
7226         kfree(nn->sessionid_hashtbl);
7227         kfree(nn->unconf_id_hashtbl);
7228         kfree(nn->conf_id_hashtbl);
7229         put_net(net);
7230 }
7231
7232 int
7233 nfs4_state_start_net(struct net *net)
7234 {
7235         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7236         int ret;
7237
7238         ret = nfs4_state_create_net(net);
7239         if (ret)
7240                 return ret;
7241         locks_start_grace(net, &nn->nfsd4_manager);
7242         nfsd4_client_tracking_init(net);
7243         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7244                nn->nfsd4_grace, net->ns.inum);
7245         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7246         return 0;
7247 }
7248
7249 /* initialization to perform when the nfsd service is started: */
7250
7251 int
7252 nfs4_state_start(void)
7253 {
7254         int ret;
7255
7256         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7257         if (laundry_wq == NULL) {
7258                 ret = -ENOMEM;
7259                 goto out;
7260         }
7261         ret = nfsd4_create_callback_queue();
7262         if (ret)
7263                 goto out_free_laundry;
7264
7265         set_max_delegations();
7266         return 0;
7267
7268 out_free_laundry:
7269         destroy_workqueue(laundry_wq);
7270 out:
7271         return ret;
7272 }
7273
7274 void
7275 nfs4_state_shutdown_net(struct net *net)
7276 {
7277         struct nfs4_delegation *dp = NULL;
7278         struct list_head *pos, *next, reaplist;
7279         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7280
7281         cancel_delayed_work_sync(&nn->laundromat_work);
7282         locks_end_grace(&nn->nfsd4_manager);
7283
7284         INIT_LIST_HEAD(&reaplist);
7285         spin_lock(&state_lock);
7286         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7287                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7288                 WARN_ON(!unhash_delegation_locked(dp));
7289                 list_add(&dp->dl_recall_lru, &reaplist);
7290         }
7291         spin_unlock(&state_lock);
7292         list_for_each_safe(pos, next, &reaplist) {
7293                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7294                 list_del_init(&dp->dl_recall_lru);
7295                 destroy_unhashed_deleg(dp);
7296         }
7297
7298         nfsd4_client_tracking_exit(net);
7299         nfs4_state_destroy_net(net);
7300 }
7301
7302 void
7303 nfs4_state_shutdown(void)
7304 {
7305         destroy_workqueue(laundry_wq);
7306         nfsd4_destroy_callback_queue();
7307 }
7308
7309 static void
7310 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7311 {
7312         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7313                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7314 }
7315
7316 static void
7317 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7318 {
7319         if (cstate->minorversion) {
7320                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7321                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7322         }
7323 }
7324
7325 void
7326 clear_current_stateid(struct nfsd4_compound_state *cstate)
7327 {
7328         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7329 }
7330
7331 /*
7332  * functions to set current state id
7333  */
7334 void
7335 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7336                 union nfsd4_op_u *u)
7337 {
7338         put_stateid(cstate, &u->open_downgrade.od_stateid);
7339 }
7340
7341 void
7342 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7343                 union nfsd4_op_u *u)
7344 {
7345         put_stateid(cstate, &u->open.op_stateid);
7346 }
7347
7348 void
7349 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7350                 union nfsd4_op_u *u)
7351 {
7352         put_stateid(cstate, &u->close.cl_stateid);
7353 }
7354
7355 void
7356 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7357                 union nfsd4_op_u *u)
7358 {
7359         put_stateid(cstate, &u->lock.lk_resp_stateid);
7360 }
7361
7362 /*
7363  * functions to consume current state id
7364  */
7365
7366 void
7367 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7368                 union nfsd4_op_u *u)
7369 {
7370         get_stateid(cstate, &u->open_downgrade.od_stateid);
7371 }
7372
7373 void
7374 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7375                 union nfsd4_op_u *u)
7376 {
7377         get_stateid(cstate, &u->delegreturn.dr_stateid);
7378 }
7379
7380 void
7381 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7382                 union nfsd4_op_u *u)
7383 {
7384         get_stateid(cstate, &u->free_stateid.fr_stateid);
7385 }
7386
7387 void
7388 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7389                 union nfsd4_op_u *u)
7390 {
7391         get_stateid(cstate, &u->setattr.sa_stateid);
7392 }
7393
7394 void
7395 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7396                 union nfsd4_op_u *u)
7397 {
7398         get_stateid(cstate, &u->close.cl_stateid);
7399 }
7400
7401 void
7402 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7403                 union nfsd4_op_u *u)
7404 {
7405         get_stateid(cstate, &u->locku.lu_stateid);
7406 }
7407
7408 void
7409 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7410                 union nfsd4_op_u *u)
7411 {
7412         get_stateid(cstate, &u->read.rd_stateid);
7413 }
7414
7415 void
7416 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7417                 union nfsd4_op_u *u)
7418 {
7419         get_stateid(cstate, &u->write.wr_stateid);
7420 }