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 *openowner_slab;
 102 static struct kmem_cache *lockowner_slab;
 103 static struct kmem_cache *file_slab;
 104 static struct kmem_cache *stateid_slab;
 105 static struct kmem_cache *deleg_slab;
 106 static struct kmem_cache *odstate_slab;
 107
 108 static void free_session(struct nfsd4_session *);
 109
 110 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 111 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 112
 113 static bool is_session_dead(struct nfsd4_session *ses)
 114 {
 115         return ses->se_flags & NFS4_SESSION_DEAD;
 116 }
 117
 118 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 119 {
 120         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 121                 return nfserr_jukebox;
 122         ses->se_flags |= NFS4_SESSION_DEAD;
 123         return nfs_ok;
 124 }
 125
 126 static bool is_client_expired(struct nfs4_client *clp)
 127 {
 128         return clp->cl_time == 0;
 129 }
 130
 131 static __be32 get_client_locked(struct nfs4_client *clp)
 132 {
 133         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 134
 135         lockdep_assert_held(&nn->client_lock);
 136
 137         if (is_client_expired(clp))
 138                 return nfserr_expired;
 139         atomic_inc(&clp->cl_refcount);
 140         return nfs_ok;
 141 }
 142
 143 /* must be called under the client_lock */
 144 static inline void
 145 renew_client_locked(struct nfs4_client *clp)
 146 {
 147         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 148
 149         if (is_client_expired(clp)) {
 150                 WARN_ON(1);
 151                 printk("%s: client (clientid %08x/%08x) already expired\n",
 152                         __func__,
 153                         clp->cl_clientid.cl_boot,
 154                         clp->cl_clientid.cl_id);
 155                 return;
 156         }
 157
 158         dprintk("renewing client (clientid %08x/%08x)\n",
 159                         clp->cl_clientid.cl_boot,
 160                         clp->cl_clientid.cl_id);
 161         list_move_tail(&clp->cl_lru, &nn->client_lru);
 162         clp->cl_time = get_seconds();
 163 }
 164
 165 static void put_client_renew_locked(struct nfs4_client *clp)
 166 {
 167         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 168
 169         lockdep_assert_held(&nn->client_lock);
 170
 171         if (!atomic_dec_and_test(&clp->cl_refcount))
 172                 return;
 173         if (!is_client_expired(clp))
 174                 renew_client_locked(clp);
 175 }
 176
 177 static void put_client_renew(struct nfs4_client *clp)
 178 {
 179         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 180
 181         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 182                 return;
 183         if (!is_client_expired(clp))
 184                 renew_client_locked(clp);
 185         spin_unlock(&nn->client_lock);
 186 }
 187
 188 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 189 {
 190         __be32 status;
 191
 192         if (is_session_dead(ses))
 193                 return nfserr_badsession;
 194         status = get_client_locked(ses->se_client);
 195         if (status)
 196                 return status;
 197         atomic_inc(&ses->se_ref);
 198         return nfs_ok;
 199 }
 200
 201 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 202 {
 203         struct nfs4_client *clp = ses->se_client;
 204         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 205
 206         lockdep_assert_held(&nn->client_lock);
 207
 208         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 209                 free_session(ses);
 210         put_client_renew_locked(clp);
 211 }
 212
 213 static void nfsd4_put_session(struct nfsd4_session *ses)
 214 {
 215         struct nfs4_client *clp = ses->se_client;
 216         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 217
 218         spin_lock(&nn->client_lock);
 219         nfsd4_put_session_locked(ses);
 220         spin_unlock(&nn->client_lock);
 221 }
 222
 223 static struct nfsd4_blocked_lock *
 224 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 225                         struct nfsd_net *nn)
 226 {
 227         struct nfsd4_blocked_lock *cur, *found = NULL;
 228
 229         spin_lock(&nn->blocked_locks_lock);
 230         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 231                 if (fh_match(fh, &cur->nbl_fh)) {
 232                         list_del_init(&cur->nbl_list);
 233                         list_del_init(&cur->nbl_lru);
 234                         found = cur;
 235                         break;
 236                 }
 237         }
 238         spin_unlock(&nn->blocked_locks_lock);
 239         if (found)
 240                 posix_unblock_lock(&found->nbl_lock);
 241         return found;
 242 }
 243
 244 static struct nfsd4_blocked_lock *
 245 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 246                         struct nfsd_net *nn)
 247 {
 248         struct nfsd4_blocked_lock *nbl;
 249
 250         nbl = find_blocked_lock(lo, fh, nn);
 251         if (!nbl) {
 252                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 253                 if (nbl) {
 254                         fh_copy_shallow(&nbl->nbl_fh, fh);
 255                         locks_init_lock(&nbl->nbl_lock);
 256                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 257                                         &nfsd4_cb_notify_lock_ops,
 258                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 259                 }
 260         }
 261         return nbl;
 262 }
 263
 264 static void
 265 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 266 {
 267         locks_release_private(&nbl->nbl_lock);
 268         kfree(nbl);
 269 }
 270
 271 static void
 272 remove_blocked_locks(struct nfs4_lockowner *lo)
 273 {
 274         struct nfs4_client *clp = lo->lo_owner.so_client;
 275         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 276         struct nfsd4_blocked_lock *nbl;
 277         LIST_HEAD(reaplist);
 278
 279         /* Dequeue all blocked locks */
 280         spin_lock(&nn->blocked_locks_lock);
 281         while (!list_empty(&lo->lo_blocked)) {
 282                 nbl = list_first_entry(&lo->lo_blocked,
 283                                         struct nfsd4_blocked_lock,
 284                                         nbl_list);
 285                 list_del_init(&nbl->nbl_list);
 286                 list_move(&nbl->nbl_lru, &reaplist);
 287         }
 288         spin_unlock(&nn->blocked_locks_lock);
 289
 290         /* Now free them */
 291         while (!list_empty(&reaplist)) {
 292                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 293                                         nbl_lru);
 294                 list_del_init(&nbl->nbl_lru);
 295                 posix_unblock_lock(&nbl->nbl_lock);
 296                 free_blocked_lock(nbl);
 297         }
 298 }
 299
 300 static int
 301 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 302 {
 303         /*
 304          * Since this is just an optimization, we don't try very hard if it
 305          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 306          * just quit trying on anything else.
 307          */
 308         switch (task->tk_status) {
 309         case -NFS4ERR_DELAY:
 310                 rpc_delay(task, 1 * HZ);
 311                 return 0;
 312         default:
 313                 return 1;
 314         }
 315 }
 316
 317 static void
 318 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 319 {
 320         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 321                                                 struct nfsd4_blocked_lock, nbl_cb);
 322
 323         free_blocked_lock(nbl);
 324 }
 325
 326 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 327         .done           = nfsd4_cb_notify_lock_done,
 328         .release        = nfsd4_cb_notify_lock_release,
 329 };
 330
 331 static inline struct nfs4_stateowner *
 332 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 333 {
 334         atomic_inc(&sop->so_count);
 335         return sop;
 336 }
 337
 338 static int
 339 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 340 {
 341         return (sop->so_owner.len == owner->len) &&
 342                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 343 }
 344
 345 static struct nfs4_openowner *
 346 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 347                         struct nfs4_client *clp)
 348 {
 349         struct nfs4_stateowner *so;
 350
 351         lockdep_assert_held(&clp->cl_lock);
 352
 353         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 354                             so_strhash) {
 355                 if (!so->so_is_open_owner)
 356                         continue;
 357                 if (same_owner_str(so, &open->op_owner))
 358                         return openowner(nfs4_get_stateowner(so));
 359         }
 360         return NULL;
 361 }
 362
 363 static struct nfs4_openowner *
 364 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 365                         struct nfs4_client *clp)
 366 {
 367         struct nfs4_openowner *oo;
 368
 369         spin_lock(&clp->cl_lock);
 370         oo = find_openstateowner_str_locked(hashval, open, clp);
 371         spin_unlock(&clp->cl_lock);
 372         return oo;
 373 }
 374
 375 static inline u32
 376 opaque_hashval(const void *ptr, int nbytes)
 377 {
 378         unsigned char *cptr = (unsigned char *) ptr;
 379
 380         u32 x = 0;
 381         while (nbytes--) {
 382                 x *= 37;
 383                 x += *cptr++;
 384         }
 385         return x;
 386 }
 387
 388 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 389 {
 390         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 391
 392         kmem_cache_free(file_slab, fp);
 393 }
 394
 395 void
 396 put_nfs4_file(struct nfs4_file *fi)
 397 {
 398         might_lock(&state_lock);
 399
 400         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 401                 hlist_del_rcu(&fi->fi_hash);
 402                 spin_unlock(&state_lock);
 403                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 404                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 405                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 406         }
 407 }
 408
 409 static struct file *
 410 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 411 {
 412         if (f->fi_fds[oflag])
 413                 return get_file(f->fi_fds[oflag]);
 414         return NULL;
 415 }
 416
 417 static struct file *
 418 find_writeable_file_locked(struct nfs4_file *f)
 419 {
 420         struct file *ret;
 421
 422         lockdep_assert_held(&f->fi_lock);
 423
 424         ret = __nfs4_get_fd(f, O_WRONLY);
 425         if (!ret)
 426                 ret = __nfs4_get_fd(f, O_RDWR);
 427         return ret;
 428 }
 429
 430 static struct file *
 431 find_writeable_file(struct nfs4_file *f)
 432 {
 433         struct file *ret;
 434
 435         spin_lock(&f->fi_lock);
 436         ret = find_writeable_file_locked(f);
 437         spin_unlock(&f->fi_lock);
 438
 439         return ret;
 440 }
 441
 442 static struct file *find_readable_file_locked(struct nfs4_file *f)
 443 {
 444         struct file *ret;
 445
 446         lockdep_assert_held(&f->fi_lock);
 447
 448         ret = __nfs4_get_fd(f, O_RDONLY);
 449         if (!ret)
 450                 ret = __nfs4_get_fd(f, O_RDWR);
 451         return ret;
 452 }
 453
 454 static struct file *
 455 find_readable_file(struct nfs4_file *f)
 456 {
 457         struct file *ret;
 458
 459         spin_lock(&f->fi_lock);
 460         ret = find_readable_file_locked(f);
 461         spin_unlock(&f->fi_lock);
 462
 463         return ret;
 464 }
 465
 466 struct file *
 467 find_any_file(struct nfs4_file *f)
 468 {
 469         struct file *ret;
 470
 471         spin_lock(&f->fi_lock);
 472         ret = __nfs4_get_fd(f, O_RDWR);
 473         if (!ret) {
 474                 ret = __nfs4_get_fd(f, O_WRONLY);
 475                 if (!ret)
 476                         ret = __nfs4_get_fd(f, O_RDONLY);
 477         }
 478         spin_unlock(&f->fi_lock);
 479         return ret;
 480 }
 481
 482 static atomic_long_t num_delegations;
 483 unsigned long max_delegations;
 484
 485 /*
 486  * Open owner state (share locks)
 487  */
 488
 489 /* hash tables for lock and open owners */
 490 #define OWNER_HASH_BITS              8
 491 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 492 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 493
 494 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 495 {
 496         unsigned int ret;
 497
 498         ret = opaque_hashval(ownername->data, ownername->len);
 499         return ret & OWNER_HASH_MASK;
 500 }
 501
 502 /* hash table for nfs4_file */
 503 #define FILE_HASH_BITS                   8
 504 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 505
 506 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 507 {
 508         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 509 }
 510
 511 static unsigned int file_hashval(struct knfsd_fh *fh)
 512 {
 513         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 514 }
 515
 516 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 517
 518 static void
 519 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 520 {
 521         lockdep_assert_held(&fp->fi_lock);
 522
 523         if (access & NFS4_SHARE_ACCESS_WRITE)
 524                 atomic_inc(&fp->fi_access[O_WRONLY]);
 525         if (access & NFS4_SHARE_ACCESS_READ)
 526                 atomic_inc(&fp->fi_access[O_RDONLY]);
 527 }
 528
 529 static __be32
 530 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 531 {
 532         lockdep_assert_held(&fp->fi_lock);
 533
 534         /* Does this access mode make sense? */
 535         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 536                 return nfserr_inval;
 537
 538         /* Does it conflict with a deny mode already set? */
 539         if ((access & fp->fi_share_deny) != 0)
 540                 return nfserr_share_denied;
 541
 542         __nfs4_file_get_access(fp, access);
 543         return nfs_ok;
 544 }
 545
 546 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 547 {
 548         /* Common case is that there is no deny mode. */
 549         if (deny) {
 550                 /* Does this deny mode make sense? */
 551                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 552                         return nfserr_inval;
 553
 554                 if ((deny & NFS4_SHARE_DENY_READ) &&
 555                     atomic_read(&fp->fi_access[O_RDONLY]))
 556                         return nfserr_share_denied;
 557
 558                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 559                     atomic_read(&fp->fi_access[O_WRONLY]))
 560                         return nfserr_share_denied;
 561         }
 562         return nfs_ok;
 563 }
 564
 565 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 566 {
 567         might_lock(&fp->fi_lock);
 568
 569         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 570                 struct file *f1 = NULL;
 571                 struct file *f2 = NULL;
 572
 573                 swap(f1, fp->fi_fds[oflag]);
 574                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 575                         swap(f2, fp->fi_fds[O_RDWR]);
 576                 spin_unlock(&fp->fi_lock);
 577                 if (f1)
 578                         fput(f1);
 579                 if (f2)
 580                         fput(f2);
 581         }
 582 }
 583
 584 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 585 {
 586         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 587
 588         if (access & NFS4_SHARE_ACCESS_WRITE)
 589                 __nfs4_file_put_access(fp, O_WRONLY);
 590         if (access & NFS4_SHARE_ACCESS_READ)
 591                 __nfs4_file_put_access(fp, O_RDONLY);
 592 }
 593
 594 /*
 595  * Allocate a new open/delegation state counter. This is needed for
 596  * pNFS for proper return on close semantics.
 597  *
 598  * Note that we only allocate it for pNFS-enabled exports, otherwise
 599  * all pointers to struct nfs4_clnt_odstate are always NULL.
 600  */
 601 static struct nfs4_clnt_odstate *
 602 alloc_clnt_odstate(struct nfs4_client *clp)
 603 {
 604         struct nfs4_clnt_odstate *co;
 605
 606         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 607         if (co) {
 608                 co->co_client = clp;
 609                 refcount_set(&co->co_odcount, 1);
 610         }
 611         return co;
 612 }
 613
 614 static void
 615 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 616 {
 617         struct nfs4_file *fp = co->co_file;
 618
 619         lockdep_assert_held(&fp->fi_lock);
 620         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 621 }
 622
 623 static inline void
 624 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 625 {
 626         if (co)
 627                 refcount_inc(&co->co_odcount);
 628 }
 629
 630 static void
 631 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 632 {
 633         struct nfs4_file *fp;
 634
 635         if (!co)
 636                 return;
 637
 638         fp = co->co_file;
 639         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 640                 list_del(&co->co_perfile);
 641                 spin_unlock(&fp->fi_lock);
 642
 643                 nfsd4_return_all_file_layouts(co->co_client, fp);
 644                 kmem_cache_free(odstate_slab, co);
 645         }
 646 }
 647
 648 static struct nfs4_clnt_odstate *
 649 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 650 {
 651         struct nfs4_clnt_odstate *co;
 652         struct nfs4_client *cl;
 653
 654         if (!new)
 655                 return NULL;
 656
 657         cl = new->co_client;
 658
 659         spin_lock(&fp->fi_lock);
 660         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 661                 if (co->co_client == cl) {
 662                         get_clnt_odstate(co);
 663                         goto out;
 664                 }
 665         }
 666         co = new;
 667         co->co_file = fp;
 668         hash_clnt_odstate_locked(new);
 669 out:
 670         spin_unlock(&fp->fi_lock);
 671         return co;
 672 }
 673
 674 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 675                                   void (*sc_free)(struct nfs4_stid *))
 676 {
 677         struct nfs4_stid *stid;
 678         int new_id;
 679
 680         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 681         if (!stid)
 682                 return NULL;
 683
 684         idr_preload(GFP_KERNEL);
 685         spin_lock(&cl->cl_lock);
 686         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 687         spin_unlock(&cl->cl_lock);
 688         idr_preload_end();
 689         if (new_id < 0)
 690                 goto out_free;
 691
 692         stid->sc_free = sc_free;
 693         stid->sc_client = cl;
 694         stid->sc_stateid.si_opaque.so_id = new_id;
 695         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 696         /* Will be incremented before return to client: */
 697         refcount_set(&stid->sc_count, 1);
 698         spin_lock_init(&stid->sc_lock);
 699
 700         /*
 701          * It shouldn't be a problem to reuse an opaque stateid value.
 702          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 703          * example, a stray write retransmission could be accepted by
 704          * the server when it should have been rejected.  Therefore,
 705          * adopt a trick from the sctp code to attempt to maximize the
 706          * amount of time until an id is reused, by ensuring they always
 707          * "increase" (mod INT_MAX):
 708          */
 709         return stid;
 710 out_free:
 711         kmem_cache_free(slab, stid);
 712         return NULL;
 713 }
 714
 715 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 716 {
 717         struct nfs4_stid *stid;
 718
 719         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 720         if (!stid)
 721                 return NULL;
 722
 723         return openlockstateid(stid);
 724 }
 725
 726 static void nfs4_free_deleg(struct nfs4_stid *stid)
 727 {
 728         kmem_cache_free(deleg_slab, stid);
 729         atomic_long_dec(&num_delegations);
 730 }
 731
 732 /*
 733  * When we recall a delegation, we should be careful not to hand it
 734  * out again straight away.
 735  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 736  * in which the filehandles of recalled delegations are "stored".
 737  * If a filehandle appear in either filter, a delegation is blocked.
 738  * When a delegation is recalled, the filehandle is stored in the "new"
 739  * filter.
 740  * Every 30 seconds we swap the filters and clear the "new" one,
 741  * unless both are empty of course.
 742  *
 743  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 744  * low 3 bytes as hash-table indices.
 745  *
 746  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 747  * is used to manage concurrent access.  Testing does not need the lock
 748  * except when swapping the two filters.
 749  */
 750 static DEFINE_SPINLOCK(blocked_delegations_lock);
 751 static struct bloom_pair {
 752         int     entries, old_entries;
 753         time_t  swap_time;
 754         int     new; /* index into 'set' */
 755         DECLARE_BITMAP(set[2], 256);
 756 } blocked_delegations;
 757
 758 static int delegation_blocked(struct knfsd_fh *fh)
 759 {
 760         u32 hash;
 761         struct bloom_pair *bd = &blocked_delegations;
 762
 763         if (bd->entries == 0)
 764                 return 0;
 765         if (seconds_since_boot() - bd->swap_time > 30) {
 766                 spin_lock(&blocked_delegations_lock);
 767                 if (seconds_since_boot() - bd->swap_time > 30) {
 768                         bd->entries -= bd->old_entries;
 769                         bd->old_entries = bd->entries;
 770                         memset(bd->set[bd->new], 0,
 771                                sizeof(bd->set[0]));
 772                         bd->new = 1-bd->new;
 773                         bd->swap_time = seconds_since_boot();
 774                 }
 775                 spin_unlock(&blocked_delegations_lock);
 776         }
 777         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 778         if (test_bit(hash&255, bd->set[0]) &&
 779             test_bit((hash>>8)&255, bd->set[0]) &&
 780             test_bit((hash>>16)&255, bd->set[0]))
 781                 return 1;
 782
 783         if (test_bit(hash&255, bd->set[1]) &&
 784             test_bit((hash>>8)&255, bd->set[1]) &&
 785             test_bit((hash>>16)&255, bd->set[1]))
 786                 return 1;
 787
 788         return 0;
 789 }
 790
 791 static void block_delegations(struct knfsd_fh *fh)
 792 {
 793         u32 hash;
 794         struct bloom_pair *bd = &blocked_delegations;
 795
 796         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 797
 798         spin_lock(&blocked_delegations_lock);
 799         __set_bit(hash&255, bd->set[bd->new]);
 800         __set_bit((hash>>8)&255, bd->set[bd->new]);
 801         __set_bit((hash>>16)&255, bd->set[bd->new]);
 802         if (bd->entries == 0)
 803                 bd->swap_time = seconds_since_boot();
 804         bd->entries += 1;
 805         spin_unlock(&blocked_delegations_lock);
 806 }
 807
 808 static struct nfs4_delegation *
 809 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
 810                  struct nfs4_clnt_odstate *odstate)
 811 {
 812         struct nfs4_delegation *dp;
 813         long n;
 814
 815         dprintk("NFSD alloc_init_deleg\n");
 816         n = atomic_long_inc_return(&num_delegations);
 817         if (n < 0 || n > max_delegations)
 818                 goto out_dec;
 819         if (delegation_blocked(&current_fh->fh_handle))
 820                 goto out_dec;
 821         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 822         if (dp == NULL)
 823                 goto out_dec;
 824
 825         /*
 826          * delegation seqid's are never incremented.  The 4.1 special
 827          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 828          * 0 anyway just for consistency and use 1:
 829          */
 830         dp->dl_stid.sc_stateid.si_generation = 1;
 831         INIT_LIST_HEAD(&dp->dl_perfile);
 832         INIT_LIST_HEAD(&dp->dl_perclnt);
 833         INIT_LIST_HEAD(&dp->dl_recall_lru);
 834         dp->dl_clnt_odstate = odstate;
 835         get_clnt_odstate(odstate);
 836         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 837         dp->dl_retries = 1;
 838         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 839                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 840         return dp;
 841 out_dec:
 842         atomic_long_dec(&num_delegations);
 843         return NULL;
 844 }
 845
 846 void
 847 nfs4_put_stid(struct nfs4_stid *s)
 848 {
 849         struct nfs4_file *fp = s->sc_file;
 850         struct nfs4_client *clp = s->sc_client;
 851
 852         might_lock(&clp->cl_lock);
 853
 854         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 855                 wake_up_all(&close_wq);
 856                 return;
 857         }
 858         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 859         spin_unlock(&clp->cl_lock);
 860         s->sc_free(s);
 861         if (fp)
 862                 put_nfs4_file(fp);
 863 }
 864
 865 void
 866 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 867 {
 868         stateid_t *src = &stid->sc_stateid;
 869
 870         spin_lock(&stid->sc_lock);
 871         if (unlikely(++src->si_generation == 0))
 872                 src->si_generation = 1;
 873         memcpy(dst, src, sizeof(*dst));
 874         spin_unlock(&stid->sc_lock);
 875 }
 876
 877 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 878 {
 879         struct file *filp = NULL;
 880
 881         spin_lock(&fp->fi_lock);
 882         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 883                 swap(filp, fp->fi_deleg_file);
 884         spin_unlock(&fp->fi_lock);
 885
 886         if (filp) {
 887                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 888                 fput(filp);
 889         }
 890 }
 891
 892 void nfs4_unhash_stid(struct nfs4_stid *s)
 893 {
 894         s->sc_type = 0;
 895 }
 896
 897 /**
 898  * nfs4_get_existing_delegation - Discover if this delegation already exists
 899  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 900  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 901  *
 902  * Return:
 903  *      On success: NULL if an existing delegation was not found.
 904  *
 905  *      On error: -EAGAIN if one was previously granted to this nfs4_client
 906  *                 for this nfs4_file.
 907  *
 908  */
 909
 910 static int
 911 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
 912 {
 913         struct nfs4_delegation *searchdp = NULL;
 914         struct nfs4_client *searchclp = NULL;
 915
 916         lockdep_assert_held(&state_lock);
 917         lockdep_assert_held(&fp->fi_lock);
 918
 919         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 920                 searchclp = searchdp->dl_stid.sc_client;
 921                 if (clp == searchclp) {
 922                         return -EAGAIN;
 923                 }
 924         }
 925         return 0;
 926 }
 927
 928 /**
 929  * hash_delegation_locked - Add a delegation to the appropriate lists
 930  * @dp:     a pointer to the nfs4_delegation we are adding.
 931  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 932  *
 933  * Return:
 934  *      On success: NULL if the delegation was successfully hashed.
 935  *
 936  *      On error: -EAGAIN if one was previously granted to this
 937  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 938  *
 939  */
 940
 941 static int
 942 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 943 {
 944         int status;
 945         struct nfs4_client *clp = dp->dl_stid.sc_client;
 946
 947         lockdep_assert_held(&state_lock);
 948         lockdep_assert_held(&fp->fi_lock);
 949
 950         status = nfs4_get_existing_delegation(clp, fp);
 951         if (status)
 952                 return status;
 953         ++fp->fi_delegees;
 954         refcount_inc(&dp->dl_stid.sc_count);
 955         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 956         list_add(&dp->dl_perfile, &fp->fi_delegations);
 957         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 958         return 0;
 959 }
 960
 961 static bool
 962 unhash_delegation_locked(struct nfs4_delegation *dp)
 963 {
 964         struct nfs4_file *fp = dp->dl_stid.sc_file;
 965
 966         lockdep_assert_held(&state_lock);
 967
 968         if (list_empty(&dp->dl_perfile))
 969                 return false;
 970
 971         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 972         /* Ensure that deleg break won't try to requeue it */
 973         ++dp->dl_time;
 974         spin_lock(&fp->fi_lock);
 975         list_del_init(&dp->dl_perclnt);
 976         list_del_init(&dp->dl_recall_lru);
 977         list_del_init(&dp->dl_perfile);
 978         spin_unlock(&fp->fi_lock);
 979         return true;
 980 }
 981
 982 static void destroy_delegation(struct nfs4_delegation *dp)
 983 {
 984         bool unhashed;
 985
 986         spin_lock(&state_lock);
 987         unhashed = unhash_delegation_locked(dp);
 988         spin_unlock(&state_lock);
 989         if (unhashed) {
 990                 put_clnt_odstate(dp->dl_clnt_odstate);
 991                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 992                 nfs4_put_stid(&dp->dl_stid);
 993         }
 994 }
 995
 996 static void revoke_delegation(struct nfs4_delegation *dp)
 997 {
 998         struct nfs4_client *clp = dp->dl_stid.sc_client;
 999
1000         WARN_ON(!list_empty(&dp->dl_recall_lru));
1001
1002         put_clnt_odstate(dp->dl_clnt_odstate);
1003         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1004
1005         if (clp->cl_minorversion == 0)
1006                 nfs4_put_stid(&dp->dl_stid);
1007         else {
1008                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1009                 spin_lock(&clp->cl_lock);
1010                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1011                 spin_unlock(&clp->cl_lock);
1012         }
1013 }
1014
1015 /*
1016  * SETCLIENTID state
1017  */
1018
1019 static unsigned int clientid_hashval(u32 id)
1020 {
1021         return id & CLIENT_HASH_MASK;
1022 }
1023
1024 static unsigned int clientstr_hashval(const char *name)
1025 {
1026         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1027 }
1028
1029 /*
1030  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1031  * st_{access,deny}_bmap field of the stateid, in order to track not
1032  * only what share bits are currently in force, but also what
1033  * combinations of share bits previous opens have used.  This allows us
1034  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1035  * return an error if the client attempt to downgrade to a combination
1036  * of share bits not explicable by closing some of its previous opens.
1037  *
1038  * XXX: This enforcement is actually incomplete, since we don't keep
1039  * track of access/deny bit combinations; so, e.g., we allow:
1040  *
1041  *      OPEN allow read, deny write
1042  *      OPEN allow both, deny none
1043  *      DOWNGRADE allow read, deny none
1044  *
1045  * which we should reject.
1046  */
1047 static unsigned int
1048 bmap_to_share_mode(unsigned long bmap) {
1049         int i;
1050         unsigned int access = 0;
1051
1052         for (i = 1; i < 4; i++) {
1053                 if (test_bit(i, &bmap))
1054                         access |= i;
1055         }
1056         return access;
1057 }
1058
1059 /* set share access for a given stateid */
1060 static inline void
1061 set_access(u32 access, struct nfs4_ol_stateid *stp)
1062 {
1063         unsigned char mask = 1 << access;
1064
1065         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1066         stp->st_access_bmap |= mask;
1067 }
1068
1069 /* clear share access for a given stateid */
1070 static inline void
1071 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1072 {
1073         unsigned char mask = 1 << access;
1074
1075         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1076         stp->st_access_bmap &= ~mask;
1077 }
1078
1079 /* test whether a given stateid has access */
1080 static inline bool
1081 test_access(u32 access, struct nfs4_ol_stateid *stp)
1082 {
1083         unsigned char mask = 1 << access;
1084
1085         return (bool)(stp->st_access_bmap & mask);
1086 }
1087
1088 /* set share deny for a given stateid */
1089 static inline void
1090 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1091 {
1092         unsigned char mask = 1 << deny;
1093
1094         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1095         stp->st_deny_bmap |= mask;
1096 }
1097
1098 /* clear share deny for a given stateid */
1099 static inline void
1100 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1101 {
1102         unsigned char mask = 1 << deny;
1103
1104         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1105         stp->st_deny_bmap &= ~mask;
1106 }
1107
1108 /* test whether a given stateid is denying specific access */
1109 static inline bool
1110 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1111 {
1112         unsigned char mask = 1 << deny;
1113
1114         return (bool)(stp->st_deny_bmap & mask);
1115 }
1116
1117 static int nfs4_access_to_omode(u32 access)
1118 {
1119         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1120         case NFS4_SHARE_ACCESS_READ:
1121                 return O_RDONLY;
1122         case NFS4_SHARE_ACCESS_WRITE:
1123                 return O_WRONLY;
1124         case NFS4_SHARE_ACCESS_BOTH:
1125                 return O_RDWR;
1126         }
1127         WARN_ON_ONCE(1);
1128         return O_RDONLY;
1129 }
1130
1131 /*
1132  * A stateid that had a deny mode associated with it is being released
1133  * or downgraded. Recalculate the deny mode on the file.
1134  */
1135 static void
1136 recalculate_deny_mode(struct nfs4_file *fp)
1137 {
1138         struct nfs4_ol_stateid *stp;
1139
1140         spin_lock(&fp->fi_lock);
1141         fp->fi_share_deny = 0;
1142         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1143                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1144         spin_unlock(&fp->fi_lock);
1145 }
1146
1147 static void
1148 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1149 {
1150         int i;
1151         bool change = false;
1152
1153         for (i = 1; i < 4; i++) {
1154                 if ((i & deny) != i) {
1155                         change = true;
1156                         clear_deny(i, stp);
1157                 }
1158         }
1159
1160         /* Recalculate per-file deny mode if there was a change */
1161         if (change)
1162                 recalculate_deny_mode(stp->st_stid.sc_file);
1163 }
1164
1165 /* release all access and file references for a given stateid */
1166 static void
1167 release_all_access(struct nfs4_ol_stateid *stp)
1168 {
1169         int i;
1170         struct nfs4_file *fp = stp->st_stid.sc_file;
1171
1172         if (fp && stp->st_deny_bmap != 0)
1173                 recalculate_deny_mode(fp);
1174
1175         for (i = 1; i < 4; i++) {
1176                 if (test_access(i, stp))
1177                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1178                 clear_access(i, stp);
1179         }
1180 }
1181
1182 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1183 {
1184         kfree(sop->so_owner.data);
1185         sop->so_ops->so_free(sop);
1186 }
1187
1188 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1189 {
1190         struct nfs4_client *clp = sop->so_client;
1191
1192         might_lock(&clp->cl_lock);
1193
1194         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1195                 return;
1196         sop->so_ops->so_unhash(sop);
1197         spin_unlock(&clp->cl_lock);
1198         nfs4_free_stateowner(sop);
1199 }
1200
1201 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1202 {
1203         struct nfs4_file *fp = stp->st_stid.sc_file;
1204
1205         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1206
1207         if (list_empty(&stp->st_perfile))
1208                 return false;
1209
1210         spin_lock(&fp->fi_lock);
1211         list_del_init(&stp->st_perfile);
1212         spin_unlock(&fp->fi_lock);
1213         list_del(&stp->st_perstateowner);
1214         return true;
1215 }
1216
1217 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1218 {
1219         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1220
1221         put_clnt_odstate(stp->st_clnt_odstate);
1222         release_all_access(stp);
1223         if (stp->st_stateowner)
1224                 nfs4_put_stateowner(stp->st_stateowner);
1225         kmem_cache_free(stateid_slab, stid);
1226 }
1227
1228 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1229 {
1230         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1231         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1232         struct file *file;
1233
1234         file = find_any_file(stp->st_stid.sc_file);
1235         if (file)
1236                 filp_close(file, (fl_owner_t)lo);
1237         nfs4_free_ol_stateid(stid);
1238 }
1239
1240 /*
1241  * Put the persistent reference to an already unhashed generic stateid, while
1242  * holding the cl_lock. If it's the last reference, then put it onto the
1243  * reaplist for later destruction.
1244  */
1245 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1246                                        struct list_head *reaplist)
1247 {
1248         struct nfs4_stid *s = &stp->st_stid;
1249         struct nfs4_client *clp = s->sc_client;
1250
1251         lockdep_assert_held(&clp->cl_lock);
1252
1253         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1254
1255         if (!refcount_dec_and_test(&s->sc_count)) {
1256                 wake_up_all(&close_wq);
1257                 return;
1258         }
1259
1260         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1261         list_add(&stp->st_locks, reaplist);
1262 }
1263
1264 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1265 {
1266         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1267
1268         list_del_init(&stp->st_locks);
1269         nfs4_unhash_stid(&stp->st_stid);
1270         return unhash_ol_stateid(stp);
1271 }
1272
1273 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1274 {
1275         struct nfs4_client *clp = stp->st_stid.sc_client;
1276         bool unhashed;
1277
1278         spin_lock(&clp->cl_lock);
1279         unhashed = unhash_lock_stateid(stp);
1280         spin_unlock(&clp->cl_lock);
1281         if (unhashed)
1282                 nfs4_put_stid(&stp->st_stid);
1283 }
1284
1285 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1286 {
1287         struct nfs4_client *clp = lo->lo_owner.so_client;
1288
1289         lockdep_assert_held(&clp->cl_lock);
1290
1291         list_del_init(&lo->lo_owner.so_strhash);
1292 }
1293
1294 /*
1295  * Free a list of generic stateids that were collected earlier after being
1296  * fully unhashed.
1297  */
1298 static void
1299 free_ol_stateid_reaplist(struct list_head *reaplist)
1300 {
1301         struct nfs4_ol_stateid *stp;
1302         struct nfs4_file *fp;
1303
1304         might_sleep();
1305
1306         while (!list_empty(reaplist)) {
1307                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1308                                        st_locks);
1309                 list_del(&stp->st_locks);
1310                 fp = stp->st_stid.sc_file;
1311                 stp->st_stid.sc_free(&stp->st_stid);
1312                 if (fp)
1313                         put_nfs4_file(fp);
1314         }
1315 }
1316
1317 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1318                                        struct list_head *reaplist)
1319 {
1320         struct nfs4_ol_stateid *stp;
1321
1322         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1323
1324         while (!list_empty(&open_stp->st_locks)) {
1325                 stp = list_entry(open_stp->st_locks.next,
1326                                 struct nfs4_ol_stateid, st_locks);
1327                 WARN_ON(!unhash_lock_stateid(stp));
1328                 put_ol_stateid_locked(stp, reaplist);
1329         }
1330 }
1331
1332 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1333                                 struct list_head *reaplist)
1334 {
1335         bool unhashed;
1336
1337         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1338
1339         unhashed = unhash_ol_stateid(stp);
1340         release_open_stateid_locks(stp, reaplist);
1341         return unhashed;
1342 }
1343
1344 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1345 {
1346         LIST_HEAD(reaplist);
1347
1348         spin_lock(&stp->st_stid.sc_client->cl_lock);
1349         if (unhash_open_stateid(stp, &reaplist))
1350                 put_ol_stateid_locked(stp, &reaplist);
1351         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1352         free_ol_stateid_reaplist(&reaplist);
1353 }
1354
1355 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1356 {
1357         struct nfs4_client *clp = oo->oo_owner.so_client;
1358
1359         lockdep_assert_held(&clp->cl_lock);
1360
1361         list_del_init(&oo->oo_owner.so_strhash);
1362         list_del_init(&oo->oo_perclient);
1363 }
1364
1365 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1366 {
1367         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1368                                           nfsd_net_id);
1369         struct nfs4_ol_stateid *s;
1370
1371         spin_lock(&nn->client_lock);
1372         s = oo->oo_last_closed_stid;
1373         if (s) {
1374                 list_del_init(&oo->oo_close_lru);
1375                 oo->oo_last_closed_stid = NULL;
1376         }
1377         spin_unlock(&nn->client_lock);
1378         if (s)
1379                 nfs4_put_stid(&s->st_stid);
1380 }
1381
1382 static void release_openowner(struct nfs4_openowner *oo)
1383 {
1384         struct nfs4_ol_stateid *stp;
1385         struct nfs4_client *clp = oo->oo_owner.so_client;
1386         struct list_head reaplist;
1387
1388         INIT_LIST_HEAD(&reaplist);
1389
1390         spin_lock(&clp->cl_lock);
1391         unhash_openowner_locked(oo);
1392         while (!list_empty(&oo->oo_owner.so_stateids)) {
1393                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1394                                 struct nfs4_ol_stateid, st_perstateowner);
1395                 if (unhash_open_stateid(stp, &reaplist))
1396                         put_ol_stateid_locked(stp, &reaplist);
1397         }
1398         spin_unlock(&clp->cl_lock);
1399         free_ol_stateid_reaplist(&reaplist);
1400         release_last_closed_stateid(oo);
1401         nfs4_put_stateowner(&oo->oo_owner);
1402 }
1403
1404 static inline int
1405 hash_sessionid(struct nfs4_sessionid *sessionid)
1406 {
1407         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1408
1409         return sid->sequence % SESSION_HASH_SIZE;
1410 }
1411
1412 #ifdef CONFIG_SUNRPC_DEBUG
1413 static inline void
1414 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1415 {
1416         u32 *ptr = (u32 *)(&sessionid->data[0]);
1417         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1418 }
1419 #else
1420 static inline void
1421 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1422 {
1423 }
1424 #endif
1425
1426 /*
1427  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1428  * won't be used for replay.
1429  */
1430 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1431 {
1432         struct nfs4_stateowner *so = cstate->replay_owner;
1433
1434         if (nfserr == nfserr_replay_me)
1435                 return;
1436
1437         if (!seqid_mutating_err(ntohl(nfserr))) {
1438                 nfsd4_cstate_clear_replay(cstate);
1439                 return;
1440         }
1441         if (!so)
1442                 return;
1443         if (so->so_is_open_owner)
1444                 release_last_closed_stateid(openowner(so));
1445         so->so_seqid++;
1446         return;
1447 }
1448
1449 static void
1450 gen_sessionid(struct nfsd4_session *ses)
1451 {
1452         struct nfs4_client *clp = ses->se_client;
1453         struct nfsd4_sessionid *sid;
1454
1455         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1456         sid->clientid = clp->cl_clientid;
1457         sid->sequence = current_sessionid++;
1458         sid->reserved = 0;
1459 }
1460
1461 /*
1462  * The protocol defines ca_maxresponssize_cached to include the size of
1463  * the rpc header, but all we need to cache is the data starting after
1464  * the end of the initial SEQUENCE operation--the rest we regenerate
1465  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1466  * value that is the number of bytes in our cache plus a few additional
1467  * bytes.  In order to stay on the safe side, and not promise more than
1468  * we can cache, those additional bytes must be the minimum possible: 24
1469  * bytes of rpc header (xid through accept state, with AUTH_NULL
1470  * verifier), 12 for the compound header (with zero-length tag), and 44
1471  * for the SEQUENCE op response:
1472  */
1473 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1474
1475 static void
1476 free_session_slots(struct nfsd4_session *ses)
1477 {
1478         int i;
1479
1480         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1481                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1482                 kfree(ses->se_slots[i]);
1483         }
1484 }
1485
1486 /*
1487  * We don't actually need to cache the rpc and session headers, so we
1488  * can allocate a little less for each slot:
1489  */
1490 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1491 {
1492         u32 size;
1493
1494         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1495                 size = 0;
1496         else
1497                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1498         return size + sizeof(struct nfsd4_slot);
1499 }
1500
1501 /*
1502  * XXX: If we run out of reserved DRC memory we could (up to a point)
1503  * re-negotiate active sessions and reduce their slot usage to make
1504  * room for new connections. For now we just fail the create session.
1505  */
1506 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1507 {
1508         u32 slotsize = slot_bytes(ca);
1509         u32 num = ca->maxreqs;
1510         int avail;
1511
1512         spin_lock(&nfsd_drc_lock);
1513         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1514                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1515         /*
1516          * Never use more than a third of the remaining memory,
1517          * unless it's the only way to give this client a slot:
1518          */
1519         avail = clamp_t(int, avail, slotsize, avail/3);
1520         num = min_t(int, num, avail / slotsize);
1521         nfsd_drc_mem_used += num * slotsize;
1522         spin_unlock(&nfsd_drc_lock);
1523
1524         return num;
1525 }
1526
1527 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1528 {
1529         int slotsize = slot_bytes(ca);
1530
1531         spin_lock(&nfsd_drc_lock);
1532         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1533         spin_unlock(&nfsd_drc_lock);
1534 }
1535
1536 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1537                                            struct nfsd4_channel_attrs *battrs)
1538 {
1539         int numslots = fattrs->maxreqs;
1540         int slotsize = slot_bytes(fattrs);
1541         struct nfsd4_session *new;
1542         int mem, i;
1543
1544         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1545                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1546         mem = numslots * sizeof(struct nfsd4_slot *);
1547
1548         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1549         if (!new)
1550                 return NULL;
1551         /* allocate each struct nfsd4_slot and data cache in one piece */
1552         for (i = 0; i < numslots; i++) {
1553                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1554                 if (!new->se_slots[i])
1555                         goto out_free;
1556         }
1557
1558         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1559         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1560
1561         return new;
1562 out_free:
1563         while (i--)
1564                 kfree(new->se_slots[i]);
1565         kfree(new);
1566         return NULL;
1567 }
1568
1569 static void free_conn(struct nfsd4_conn *c)
1570 {
1571         svc_xprt_put(c->cn_xprt);
1572         kfree(c);
1573 }
1574
1575 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1576 {
1577         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1578         struct nfs4_client *clp = c->cn_session->se_client;
1579
1580         spin_lock(&clp->cl_lock);
1581         if (!list_empty(&c->cn_persession)) {
1582                 list_del(&c->cn_persession);
1583                 free_conn(c);
1584         }
1585         nfsd4_probe_callback(clp);
1586         spin_unlock(&clp->cl_lock);
1587 }
1588
1589 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1590 {
1591         struct nfsd4_conn *conn;
1592
1593         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1594         if (!conn)
1595                 return NULL;
1596         svc_xprt_get(rqstp->rq_xprt);
1597         conn->cn_xprt = rqstp->rq_xprt;
1598         conn->cn_flags = flags;
1599         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1600         return conn;
1601 }
1602
1603 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1604 {
1605         conn->cn_session = ses;
1606         list_add(&conn->cn_persession, &ses->se_conns);
1607 }
1608
1609 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1610 {
1611         struct nfs4_client *clp = ses->se_client;
1612
1613         spin_lock(&clp->cl_lock);
1614         __nfsd4_hash_conn(conn, ses);
1615         spin_unlock(&clp->cl_lock);
1616 }
1617
1618 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1619 {
1620         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1621         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1622 }
1623
1624 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1625 {
1626         int ret;
1627
1628         nfsd4_hash_conn(conn, ses);
1629         ret = nfsd4_register_conn(conn);
1630         if (ret)
1631                 /* oops; xprt is already down: */
1632                 nfsd4_conn_lost(&conn->cn_xpt_user);
1633         /* We may have gained or lost a callback channel: */
1634         nfsd4_probe_callback_sync(ses->se_client);
1635 }
1636
1637 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1638 {
1639         u32 dir = NFS4_CDFC4_FORE;
1640
1641         if (cses->flags & SESSION4_BACK_CHAN)
1642                 dir |= NFS4_CDFC4_BACK;
1643         return alloc_conn(rqstp, dir);
1644 }
1645
1646 /* must be called under client_lock */
1647 static void nfsd4_del_conns(struct nfsd4_session *s)
1648 {
1649         struct nfs4_client *clp = s->se_client;
1650         struct nfsd4_conn *c;
1651
1652         spin_lock(&clp->cl_lock);
1653         while (!list_empty(&s->se_conns)) {
1654                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1655                 list_del_init(&c->cn_persession);
1656                 spin_unlock(&clp->cl_lock);
1657
1658                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1659                 free_conn(c);
1660
1661                 spin_lock(&clp->cl_lock);
1662         }
1663         spin_unlock(&clp->cl_lock);
1664 }
1665
1666 static void __free_session(struct nfsd4_session *ses)
1667 {
1668         free_session_slots(ses);
1669         kfree(ses);
1670 }
1671
1672 static void free_session(struct nfsd4_session *ses)
1673 {
1674         nfsd4_del_conns(ses);
1675         nfsd4_put_drc_mem(&ses->se_fchannel);
1676         __free_session(ses);
1677 }
1678
1679 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1680 {
1681         int idx;
1682         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1683
1684         new->se_client = clp;
1685         gen_sessionid(new);
1686
1687         INIT_LIST_HEAD(&new->se_conns);
1688
1689         new->se_cb_seq_nr = 1;
1690         new->se_flags = cses->flags;
1691         new->se_cb_prog = cses->callback_prog;
1692         new->se_cb_sec = cses->cb_sec;
1693         atomic_set(&new->se_ref, 0);
1694         idx = hash_sessionid(&new->se_sessionid);
1695         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1696         spin_lock(&clp->cl_lock);
1697         list_add(&new->se_perclnt, &clp->cl_sessions);
1698         spin_unlock(&clp->cl_lock);
1699
1700         {
1701                 struct sockaddr *sa = svc_addr(rqstp);
1702                 /*
1703                  * This is a little silly; with sessions there's no real
1704                  * use for the callback address.  Use the peer address
1705                  * as a reasonable default for now, but consider fixing
1706                  * the rpc client not to require an address in the
1707                  * future:
1708                  */
1709                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1710                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1711         }
1712 }
1713
1714 /* caller must hold client_lock */
1715 static struct nfsd4_session *
1716 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1717 {
1718         struct nfsd4_session *elem;
1719         int idx;
1720         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1721
1722         lockdep_assert_held(&nn->client_lock);
1723
1724         dump_sessionid(__func__, sessionid);
1725         idx = hash_sessionid(sessionid);
1726         /* Search in the appropriate list */
1727         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1728                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1729                             NFS4_MAX_SESSIONID_LEN)) {
1730                         return elem;
1731                 }
1732         }
1733
1734         dprintk("%s: session not found\n", __func__);
1735         return NULL;
1736 }
1737
1738 static struct nfsd4_session *
1739 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1740                 __be32 *ret)
1741 {
1742         struct nfsd4_session *session;
1743         __be32 status = nfserr_badsession;
1744
1745         session = __find_in_sessionid_hashtbl(sessionid, net);
1746         if (!session)
1747                 goto out;
1748         status = nfsd4_get_session_locked(session);
1749         if (status)
1750                 session = NULL;
1751 out:
1752         *ret = status;
1753         return session;
1754 }
1755
1756 /* caller must hold client_lock */
1757 static void
1758 unhash_session(struct nfsd4_session *ses)
1759 {
1760         struct nfs4_client *clp = ses->se_client;
1761         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1762
1763         lockdep_assert_held(&nn->client_lock);
1764
1765         list_del(&ses->se_hash);
1766         spin_lock(&ses->se_client->cl_lock);
1767         list_del(&ses->se_perclnt);
1768         spin_unlock(&ses->se_client->cl_lock);
1769 }
1770
1771 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1772 static int
1773 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1774 {
1775         /*
1776          * We're assuming the clid was not given out from a boot
1777          * precisely 2^32 (about 136 years) before this one.  That seems
1778          * a safe assumption:
1779          */
1780         if (clid->cl_boot == (u32)nn->boot_time)
1781                 return 0;
1782         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1783                 clid->cl_boot, clid->cl_id, nn->boot_time);
1784         return 1;
1785 }
1786
1787 /*
1788  * XXX Should we use a slab cache ?
1789  * This type of memory management is somewhat inefficient, but we use it
1790  * anyway since SETCLIENTID is not a common operation.
1791  */
1792 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1793 {
1794         struct nfs4_client *clp;
1795         int i;
1796
1797         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1798         if (clp == NULL)
1799                 return NULL;
1800         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1801         if (clp->cl_name.data == NULL)
1802                 goto err_no_name;
1803         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1804                         OWNER_HASH_SIZE, GFP_KERNEL);
1805         if (!clp->cl_ownerstr_hashtbl)
1806                 goto err_no_hashtbl;
1807         for (i = 0; i < OWNER_HASH_SIZE; i++)
1808                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1809         clp->cl_name.len = name.len;
1810         INIT_LIST_HEAD(&clp->cl_sessions);
1811         idr_init(&clp->cl_stateids);
1812         atomic_set(&clp->cl_refcount, 0);
1813         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1814         INIT_LIST_HEAD(&clp->cl_idhash);
1815         INIT_LIST_HEAD(&clp->cl_openowners);
1816         INIT_LIST_HEAD(&clp->cl_delegations);
1817         INIT_LIST_HEAD(&clp->cl_lru);
1818         INIT_LIST_HEAD(&clp->cl_revoked);
1819 #ifdef CONFIG_NFSD_PNFS
1820         INIT_LIST_HEAD(&clp->cl_lo_states);
1821 #endif
1822         spin_lock_init(&clp->cl_lock);
1823         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1824         return clp;
1825 err_no_hashtbl:
1826         kfree(clp->cl_name.data);
1827 err_no_name:
1828         kfree(clp);
1829         return NULL;
1830 }
1831
1832 static void
1833 free_client(struct nfs4_client *clp)
1834 {
1835         while (!list_empty(&clp->cl_sessions)) {
1836                 struct nfsd4_session *ses;
1837                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1838                                 se_perclnt);
1839                 list_del(&ses->se_perclnt);
1840                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1841                 free_session(ses);
1842         }
1843         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1844         free_svc_cred(&clp->cl_cred);
1845         kfree(clp->cl_ownerstr_hashtbl);
1846         kfree(clp->cl_name.data);
1847         idr_destroy(&clp->cl_stateids);
1848         kfree(clp);
1849 }
1850
1851 /* must be called under the client_lock */
1852 static void
1853 unhash_client_locked(struct nfs4_client *clp)
1854 {
1855         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1856         struct nfsd4_session *ses;
1857
1858         lockdep_assert_held(&nn->client_lock);
1859
1860         /* Mark the client as expired! */
1861         clp->cl_time = 0;
1862         /* Make it invisible */
1863         if (!list_empty(&clp->cl_idhash)) {
1864                 list_del_init(&clp->cl_idhash);
1865                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1866                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1867                 else
1868                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1869         }
1870         list_del_init(&clp->cl_lru);
1871         spin_lock(&clp->cl_lock);
1872         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1873                 list_del_init(&ses->se_hash);
1874         spin_unlock(&clp->cl_lock);
1875 }
1876
1877 static void
1878 unhash_client(struct nfs4_client *clp)
1879 {
1880         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1881
1882         spin_lock(&nn->client_lock);
1883         unhash_client_locked(clp);
1884         spin_unlock(&nn->client_lock);
1885 }
1886
1887 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1888 {
1889         if (atomic_read(&clp->cl_refcount))
1890                 return nfserr_jukebox;
1891         unhash_client_locked(clp);
1892         return nfs_ok;
1893 }
1894
1895 static void
1896 __destroy_client(struct nfs4_client *clp)
1897 {
1898         int i;
1899         struct nfs4_openowner *oo;
1900         struct nfs4_delegation *dp;
1901         struct list_head reaplist;
1902
1903         INIT_LIST_HEAD(&reaplist);
1904         spin_lock(&state_lock);
1905         while (!list_empty(&clp->cl_delegations)) {
1906                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1907                 WARN_ON(!unhash_delegation_locked(dp));
1908                 list_add(&dp->dl_recall_lru, &reaplist);
1909         }
1910         spin_unlock(&state_lock);
1911         while (!list_empty(&reaplist)) {
1912                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1913                 list_del_init(&dp->dl_recall_lru);
1914                 put_clnt_odstate(dp->dl_clnt_odstate);
1915                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1916                 nfs4_put_stid(&dp->dl_stid);
1917         }
1918         while (!list_empty(&clp->cl_revoked)) {
1919                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1920                 list_del_init(&dp->dl_recall_lru);
1921                 nfs4_put_stid(&dp->dl_stid);
1922         }
1923         while (!list_empty(&clp->cl_openowners)) {
1924                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1925                 nfs4_get_stateowner(&oo->oo_owner);
1926                 release_openowner(oo);
1927         }
1928         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1929                 struct nfs4_stateowner *so, *tmp;
1930
1931                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1932                                          so_strhash) {
1933                         /* Should be no openowners at this point */
1934                         WARN_ON_ONCE(so->so_is_open_owner);
1935                         remove_blocked_locks(lockowner(so));
1936                 }
1937         }
1938         nfsd4_return_all_client_layouts(clp);
1939         nfsd4_shutdown_callback(clp);
1940         if (clp->cl_cb_conn.cb_xprt)
1941                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1942         free_client(clp);
1943 }
1944
1945 static void
1946 destroy_client(struct nfs4_client *clp)
1947 {
1948         unhash_client(clp);
1949         __destroy_client(clp);
1950 }
1951
1952 static void expire_client(struct nfs4_client *clp)
1953 {
1954         unhash_client(clp);
1955         nfsd4_client_record_remove(clp);
1956         __destroy_client(clp);
1957 }
1958
1959 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1960 {
1961         memcpy(target->cl_verifier.data, source->data,
1962                         sizeof(target->cl_verifier.data));
1963 }
1964
1965 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1966 {
1967         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1968         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1969 }
1970
1971 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1972 {
1973         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1974         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1975                                                                 GFP_KERNEL);
1976         if ((source->cr_principal && ! target->cr_principal) ||
1977             (source->cr_raw_principal && ! target->cr_raw_principal))
1978                 return -ENOMEM;
1979
1980         target->cr_flavor = source->cr_flavor;
1981         target->cr_uid = source->cr_uid;
1982         target->cr_gid = source->cr_gid;
1983         target->cr_group_info = source->cr_group_info;
1984         get_group_info(target->cr_group_info);
1985         target->cr_gss_mech = source->cr_gss_mech;
1986         if (source->cr_gss_mech)
1987                 gss_mech_get(source->cr_gss_mech);
1988         return 0;
1989 }
1990
1991 static int
1992 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1993 {
1994         if (o1->len < o2->len)
1995                 return -1;
1996         if (o1->len > o2->len)
1997                 return 1;
1998         return memcmp(o1->data, o2->data, o1->len);
1999 }
2000
2001 static int same_name(const char *n1, const char *n2)
2002 {
2003         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2004 }
2005
2006 static int
2007 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2008 {
2009         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2010 }
2011
2012 static int
2013 same_clid(clientid_t *cl1, clientid_t *cl2)
2014 {
2015         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2016 }
2017
2018 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2019 {
2020         int i;
2021
2022         if (g1->ngroups != g2->ngroups)
2023                 return false;
2024         for (i=0; i<g1->ngroups; i++)
2025                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2026                         return false;
2027         return true;
2028 }
2029
2030 /*
2031  * RFC 3530 language requires clid_inuse be returned when the
2032  * "principal" associated with a requests differs from that previously
2033  * used.  We use uid, gid's, and gss principal string as our best
2034  * approximation.  We also don't want to allow non-gss use of a client
2035  * established using gss: in theory cr_principal should catch that
2036  * change, but in practice cr_principal can be null even in the gss case
2037  * since gssd doesn't always pass down a principal string.
2038  */
2039 static bool is_gss_cred(struct svc_cred *cr)
2040 {
2041         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2042         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2043 }
2044
2045
2046 static bool
2047 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2048 {
2049         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2050                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2051                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2052                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2053                 return false;
2054         if (cr1->cr_principal == cr2->cr_principal)
2055                 return true;
2056         if (!cr1->cr_principal || !cr2->cr_principal)
2057                 return false;
2058         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2059 }
2060
2061 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2062 {
2063         struct svc_cred *cr = &rqstp->rq_cred;
2064         u32 service;
2065
2066         if (!cr->cr_gss_mech)
2067                 return false;
2068         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2069         return service == RPC_GSS_SVC_INTEGRITY ||
2070                service == RPC_GSS_SVC_PRIVACY;
2071 }
2072
2073 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2074 {
2075         struct svc_cred *cr = &rqstp->rq_cred;
2076
2077         if (!cl->cl_mach_cred)
2078                 return true;
2079         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2080                 return false;
2081         if (!svc_rqst_integrity_protected(rqstp))
2082                 return false;
2083         if (cl->cl_cred.cr_raw_principal)
2084                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2085                                                 cr->cr_raw_principal);
2086         if (!cr->cr_principal)
2087                 return false;
2088         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2089 }
2090
2091 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2092 {
2093         __be32 verf[2];
2094
2095         /*
2096          * This is opaque to client, so no need to byte-swap. Use
2097          * __force to keep sparse happy
2098          */
2099         verf[0] = (__force __be32)get_seconds();
2100         verf[1] = (__force __be32)nn->clverifier_counter++;
2101         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2102 }
2103
2104 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2105 {
2106         clp->cl_clientid.cl_boot = nn->boot_time;
2107         clp->cl_clientid.cl_id = nn->clientid_counter++;
2108         gen_confirm(clp, nn);
2109 }
2110
2111 static struct nfs4_stid *
2112 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2113 {
2114         struct nfs4_stid *ret;
2115
2116         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2117         if (!ret || !ret->sc_type)
2118                 return NULL;
2119         return ret;
2120 }
2121
2122 static struct nfs4_stid *
2123 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2124 {
2125         struct nfs4_stid *s;
2126
2127         spin_lock(&cl->cl_lock);
2128         s = find_stateid_locked(cl, t);
2129         if (s != NULL) {
2130                 if (typemask & s->sc_type)
2131                         refcount_inc(&s->sc_count);
2132                 else
2133                         s = NULL;
2134         }
2135         spin_unlock(&cl->cl_lock);
2136         return s;
2137 }
2138
2139 static struct nfs4_client *create_client(struct xdr_netobj name,
2140                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2141 {
2142         struct nfs4_client *clp;
2143         struct sockaddr *sa = svc_addr(rqstp);
2144         int ret;
2145         struct net *net = SVC_NET(rqstp);
2146
2147         clp = alloc_client(name);
2148         if (clp == NULL)
2149                 return NULL;
2150
2151         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2152         if (ret) {
2153                 free_client(clp);
2154                 return NULL;
2155         }
2156         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2157         clp->cl_time = get_seconds();
2158         clear_bit(0, &clp->cl_cb_slot_busy);
2159         copy_verf(clp, verf);
2160         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2161         clp->cl_cb_session = NULL;
2162         clp->net = net;
2163         return clp;
2164 }
2165
2166 static void
2167 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2168 {
2169         struct rb_node **new = &(root->rb_node), *parent = NULL;
2170         struct nfs4_client *clp;
2171
2172         while (*new) {
2173                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2174                 parent = *new;
2175
2176                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2177                         new = &((*new)->rb_left);
2178                 else
2179                         new = &((*new)->rb_right);
2180         }
2181
2182         rb_link_node(&new_clp->cl_namenode, parent, new);
2183         rb_insert_color(&new_clp->cl_namenode, root);
2184 }
2185
2186 static struct nfs4_client *
2187 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2188 {
2189         int cmp;
2190         struct rb_node *node = root->rb_node;
2191         struct nfs4_client *clp;
2192
2193         while (node) {
2194                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2195                 cmp = compare_blob(&clp->cl_name, name);
2196                 if (cmp > 0)
2197                         node = node->rb_left;
2198                 else if (cmp < 0)
2199                         node = node->rb_right;
2200                 else
2201                         return clp;
2202         }
2203         return NULL;
2204 }
2205
2206 static void
2207 add_to_unconfirmed(struct nfs4_client *clp)
2208 {
2209         unsigned int idhashval;
2210         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2211
2212         lockdep_assert_held(&nn->client_lock);
2213
2214         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2215         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2216         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2217         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2218         renew_client_locked(clp);
2219 }
2220
2221 static void
2222 move_to_confirmed(struct nfs4_client *clp)
2223 {
2224         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2225         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2226
2227         lockdep_assert_held(&nn->client_lock);
2228
2229         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2230         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2231         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2232         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2233         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2234         renew_client_locked(clp);
2235 }
2236
2237 static struct nfs4_client *
2238 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2239 {
2240         struct nfs4_client *clp;
2241         unsigned int idhashval = clientid_hashval(clid->cl_id);
2242
2243         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2244                 if (same_clid(&clp->cl_clientid, clid)) {
2245                         if ((bool)clp->cl_minorversion != sessions)
2246                                 return NULL;
2247                         renew_client_locked(clp);
2248                         return clp;
2249                 }
2250         }
2251         return NULL;
2252 }
2253
2254 static struct nfs4_client *
2255 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2256 {
2257         struct list_head *tbl = nn->conf_id_hashtbl;
2258
2259         lockdep_assert_held(&nn->client_lock);
2260         return find_client_in_id_table(tbl, clid, sessions);
2261 }
2262
2263 static struct nfs4_client *
2264 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2265 {
2266         struct list_head *tbl = nn->unconf_id_hashtbl;
2267
2268         lockdep_assert_held(&nn->client_lock);
2269         return find_client_in_id_table(tbl, clid, sessions);
2270 }
2271
2272 static bool clp_used_exchangeid(struct nfs4_client *clp)
2273 {
2274         return clp->cl_exchange_flags != 0;
2275 }
2276
2277 static struct nfs4_client *
2278 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2279 {
2280         lockdep_assert_held(&nn->client_lock);
2281         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2282 }
2283
2284 static struct nfs4_client *
2285 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2286 {
2287         lockdep_assert_held(&nn->client_lock);
2288         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2289 }
2290
2291 static void
2292 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2293 {
2294         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2295         struct sockaddr *sa = svc_addr(rqstp);
2296         u32 scopeid = rpc_get_scope_id(sa);
2297         unsigned short expected_family;
2298
2299         /* Currently, we only support tcp and tcp6 for the callback channel */
2300         if (se->se_callback_netid_len == 3 &&
2301             !memcmp(se->se_callback_netid_val, "tcp", 3))
2302                 expected_family = AF_INET;
2303         else if (se->se_callback_netid_len == 4 &&
2304                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2305                 expected_family = AF_INET6;
2306         else
2307                 goto out_err;
2308
2309         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2310                                             se->se_callback_addr_len,
2311                                             (struct sockaddr *)&conn->cb_addr,
2312                                             sizeof(conn->cb_addr));
2313
2314         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2315                 goto out_err;
2316
2317         if (conn->cb_addr.ss_family == AF_INET6)
2318                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2319
2320         conn->cb_prog = se->se_callback_prog;
2321         conn->cb_ident = se->se_callback_ident;
2322         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2323         return;
2324 out_err:
2325         conn->cb_addr.ss_family = AF_UNSPEC;
2326         conn->cb_addrlen = 0;
2327         dprintk("NFSD: this client (clientid %08x/%08x) "
2328                 "will not receive delegations\n",
2329                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2330
2331         return;
2332 }
2333
2334 /*
2335  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2336  */
2337 static void
2338 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2339 {
2340         struct xdr_buf *buf = resp->xdr.buf;
2341         struct nfsd4_slot *slot = resp->cstate.slot;
2342         unsigned int base;
2343
2344         dprintk("--> %s slot %p\n", __func__, slot);
2345
2346         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2347         slot->sl_opcnt = resp->opcnt;
2348         slot->sl_status = resp->cstate.status;
2349         free_svc_cred(&slot->sl_cred);
2350         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2351
2352         if (!nfsd4_cache_this(resp)) {
2353                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2354                 return;
2355         }
2356         slot->sl_flags |= NFSD4_SLOT_CACHED;
2357
2358         base = resp->cstate.data_offset;
2359         slot->sl_datalen = buf->len - base;
2360         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2361                 WARN(1, "%s: sessions DRC could not cache compound\n",
2362                      __func__);
2363         return;
2364 }
2365
2366 /*
2367  * Encode the replay sequence operation from the slot values.
2368  * If cachethis is FALSE encode the uncached rep error on the next
2369  * operation which sets resp->p and increments resp->opcnt for
2370  * nfs4svc_encode_compoundres.
2371  *
2372  */
2373 static __be32
2374 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2375                           struct nfsd4_compoundres *resp)
2376 {
2377         struct nfsd4_op *op;
2378         struct nfsd4_slot *slot = resp->cstate.slot;
2379
2380         /* Encode the replayed sequence operation */
2381         op = &args->ops[resp->opcnt - 1];
2382         nfsd4_encode_operation(resp, op);
2383
2384         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2385                 return op->status;
2386         if (args->opcnt == 1) {
2387                 /*
2388                  * The original operation wasn't a solo sequence--we
2389                  * always cache those--so this retry must not match the
2390                  * original:
2391                  */
2392                 op->status = nfserr_seq_false_retry;
2393         } else {
2394                 op = &args->ops[resp->opcnt++];
2395                 op->status = nfserr_retry_uncached_rep;
2396                 nfsd4_encode_operation(resp, op);
2397         }
2398         return op->status;
2399 }
2400
2401 /*
2402  * The sequence operation is not cached because we can use the slot and
2403  * session values.
2404  */
2405 static __be32
2406 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2407                          struct nfsd4_sequence *seq)
2408 {
2409         struct nfsd4_slot *slot = resp->cstate.slot;
2410         struct xdr_stream *xdr = &resp->xdr;
2411         __be32 *p;
2412         __be32 status;
2413
2414         dprintk("--> %s slot %p\n", __func__, slot);
2415
2416         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2417         if (status)
2418                 return status;
2419
2420         p = xdr_reserve_space(xdr, slot->sl_datalen);
2421         if (!p) {
2422                 WARN_ON_ONCE(1);
2423                 return nfserr_serverfault;
2424         }
2425         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2426         xdr_commit_encode(xdr);
2427
2428         resp->opcnt = slot->sl_opcnt;
2429         return slot->sl_status;
2430 }
2431
2432 /*
2433  * Set the exchange_id flags returned by the server.
2434  */
2435 static void
2436 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2437 {
2438 #ifdef CONFIG_NFSD_PNFS
2439         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2440 #else
2441         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2442 #endif
2443
2444         /* Referrals are supported, Migration is not. */
2445         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2446
2447         /* set the wire flags to return to client. */
2448         clid->flags = new->cl_exchange_flags;
2449 }
2450
2451 static bool client_has_openowners(struct nfs4_client *clp)
2452 {
2453         struct nfs4_openowner *oo;
2454
2455         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2456                 if (!list_empty(&oo->oo_owner.so_stateids))
2457                         return true;
2458         }
2459         return false;
2460 }
2461
2462 static bool client_has_state(struct nfs4_client *clp)
2463 {
2464         return client_has_openowners(clp)
2465 #ifdef CONFIG_NFSD_PNFS
2466                 || !list_empty(&clp->cl_lo_states)
2467 #endif
2468                 || !list_empty(&clp->cl_delegations)
2469                 || !list_empty(&clp->cl_sessions);
2470 }
2471
2472 __be32
2473 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2474                 union nfsd4_op_u *u)
2475 {
2476         struct nfsd4_exchange_id *exid = &u->exchange_id;
2477         struct nfs4_client *conf, *new;
2478         struct nfs4_client *unconf = NULL;
2479         __be32 status;
2480         char                    addr_str[INET6_ADDRSTRLEN];
2481         nfs4_verifier           verf = exid->verifier;
2482         struct sockaddr         *sa = svc_addr(rqstp);
2483         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2484         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2485
2486         rpc_ntop(sa, addr_str, sizeof(addr_str));
2487         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2488                 "ip_addr=%s flags %x, spa_how %d\n",
2489                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2490                 addr_str, exid->flags, exid->spa_how);
2491
2492         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2493                 return nfserr_inval;
2494
2495         new = create_client(exid->clname, rqstp, &verf);
2496         if (new == NULL)
2497                 return nfserr_jukebox;
2498
2499         switch (exid->spa_how) {
2500         case SP4_MACH_CRED:
2501                 exid->spo_must_enforce[0] = 0;
2502                 exid->spo_must_enforce[1] = (
2503                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2504                         1 << (OP_EXCHANGE_ID - 32) |
2505                         1 << (OP_CREATE_SESSION - 32) |
2506                         1 << (OP_DESTROY_SESSION - 32) |
2507                         1 << (OP_DESTROY_CLIENTID - 32));
2508
2509                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2510                                         1 << (OP_OPEN_DOWNGRADE) |
2511                                         1 << (OP_LOCKU) |
2512                                         1 << (OP_DELEGRETURN));
2513
2514                 exid->spo_must_allow[1] &= (
2515                                         1 << (OP_TEST_STATEID - 32) |
2516                                         1 << (OP_FREE_STATEID - 32));
2517                 if (!svc_rqst_integrity_protected(rqstp)) {
2518                         status = nfserr_inval;
2519                         goto out_nolock;
2520                 }
2521                 /*
2522                  * Sometimes userspace doesn't give us a principal.
2523                  * Which is a bug, really.  Anyway, we can't enforce
2524                  * MACH_CRED in that case, better to give up now:
2525                  */
2526                 if (!new->cl_cred.cr_principal &&
2527                                         !new->cl_cred.cr_raw_principal) {
2528                         status = nfserr_serverfault;
2529                         goto out_nolock;
2530                 }
2531                 new->cl_mach_cred = true;
2532         case SP4_NONE:
2533                 break;
2534         default:                                /* checked by xdr code */
2535                 WARN_ON_ONCE(1);
2536         case SP4_SSV:
2537                 status = nfserr_encr_alg_unsupp;
2538                 goto out_nolock;
2539         }
2540
2541         /* Cases below refer to rfc 5661 section 18.35.4: */
2542         spin_lock(&nn->client_lock);
2543         conf = find_confirmed_client_by_name(&exid->clname, nn);
2544         if (conf) {
2545                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2546                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2547
2548                 if (update) {
2549                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2550                                 status = nfserr_inval;
2551                                 goto out;
2552                         }
2553                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2554                                 status = nfserr_wrong_cred;
2555                                 goto out;
2556                         }
2557                         if (!creds_match) { /* case 9 */
2558                                 status = nfserr_perm;
2559                                 goto out;
2560                         }
2561                         if (!verfs_match) { /* case 8 */
2562                                 status = nfserr_not_same;
2563                                 goto out;
2564                         }
2565                         /* case 6 */
2566                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2567                         goto out_copy;
2568                 }
2569                 if (!creds_match) { /* case 3 */
2570                         if (client_has_state(conf)) {
2571                                 status = nfserr_clid_inuse;
2572                                 goto out;
2573                         }
2574                         goto out_new;
2575                 }
2576                 if (verfs_match) { /* case 2 */
2577                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2578                         goto out_copy;
2579                 }
2580                 /* case 5, client reboot */
2581                 conf = NULL;
2582                 goto out_new;
2583         }
2584
2585         if (update) { /* case 7 */
2586                 status = nfserr_noent;
2587                 goto out;
2588         }
2589
2590         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2591         if (unconf) /* case 4, possible retry or client restart */
2592                 unhash_client_locked(unconf);
2593
2594         /* case 1 (normal case) */
2595 out_new:
2596         if (conf) {
2597                 status = mark_client_expired_locked(conf);
2598                 if (status)
2599                         goto out;
2600         }
2601         new->cl_minorversion = cstate->minorversion;
2602         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2603         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2604
2605         gen_clid(new, nn);
2606         add_to_unconfirmed(new);
2607         swap(new, conf);
2608 out_copy:
2609         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2610         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2611
2612         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2613         nfsd4_set_ex_flags(conf, exid);
2614
2615         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2616                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2617         status = nfs_ok;
2618
2619 out:
2620         spin_unlock(&nn->client_lock);
2621 out_nolock:
2622         if (new)
2623                 expire_client(new);
2624         if (unconf)
2625                 expire_client(unconf);
2626         return status;
2627 }
2628
2629 static __be32
2630 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2631 {
2632         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2633                 slot_seqid);
2634
2635         /* The slot is in use, and no response has been sent. */
2636         if (slot_inuse) {
2637                 if (seqid == slot_seqid)
2638                         return nfserr_jukebox;
2639                 else
2640                         return nfserr_seq_misordered;
2641         }
2642         /* Note unsigned 32-bit arithmetic handles wraparound: */
2643         if (likely(seqid == slot_seqid + 1))
2644                 return nfs_ok;
2645         if (seqid == slot_seqid)
2646                 return nfserr_replay_cache;
2647         return nfserr_seq_misordered;
2648 }
2649
2650 /*
2651  * Cache the create session result into the create session single DRC
2652  * slot cache by saving the xdr structure. sl_seqid has been set.
2653  * Do this for solo or embedded create session operations.
2654  */
2655 static void
2656 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2657                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2658 {
2659         slot->sl_status = nfserr;
2660         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2661 }
2662
2663 static __be32
2664 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2665                             struct nfsd4_clid_slot *slot)
2666 {
2667         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2668         return slot->sl_status;
2669 }
2670
2671 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2672                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2673                         1 +     /* MIN tag is length with zero, only length */ \
2674                         3 +     /* version, opcount, opcode */ \
2675                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2676                                 /* seqid, slotID, slotID, cache */ \
2677                         4 ) * sizeof(__be32))
2678
2679 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2680                         2 +     /* verifier: AUTH_NULL, length 0 */\
2681                         1 +     /* status */ \
2682                         1 +     /* MIN tag is length with zero, only length */ \
2683                         3 +     /* opcount, opcode, opstatus*/ \
2684                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2685                                 /* seqid, slotID, slotID, slotID, status */ \
2686                         5 ) * sizeof(__be32))
2687
2688 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2689 {
2690         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2691
2692         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2693                 return nfserr_toosmall;
2694         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2695                 return nfserr_toosmall;
2696         ca->headerpadsz = 0;
2697         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2698         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2699         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2700         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2701                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2702         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2703         /*
2704          * Note decreasing slot size below client's request may make it
2705          * difficult for client to function correctly, whereas
2706          * decreasing the number of slots will (just?) affect
2707          * performance.  When short on memory we therefore prefer to
2708          * decrease number of slots instead of their size.  Clients that
2709          * request larger slots than they need will get poor results:
2710          */
2711         ca->maxreqs = nfsd4_get_drc_mem(ca);
2712         if (!ca->maxreqs)
2713                 return nfserr_jukebox;
2714
2715         return nfs_ok;
2716 }
2717
2718 /*
2719  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2720  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2721  */
2722 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2723         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2724
2725 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2726         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2727
2728 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2729                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2730 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2731                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2732                                  sizeof(__be32))
2733
2734 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2735 {
2736         ca->headerpadsz = 0;
2737
2738         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2739                 return nfserr_toosmall;
2740         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2741                 return nfserr_toosmall;
2742         ca->maxresp_cached = 0;
2743         if (ca->maxops < 2)
2744                 return nfserr_toosmall;
2745
2746         return nfs_ok;
2747 }
2748
2749 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2750 {
2751         switch (cbs->flavor) {
2752         case RPC_AUTH_NULL:
2753         case RPC_AUTH_UNIX:
2754                 return nfs_ok;
2755         default:
2756                 /*
2757                  * GSS case: the spec doesn't allow us to return this
2758                  * error.  But it also doesn't allow us not to support
2759                  * GSS.
2760                  * I'd rather this fail hard than return some error the
2761                  * client might think it can already handle:
2762                  */
2763                 return nfserr_encr_alg_unsupp;
2764         }
2765 }
2766
2767 __be32
2768 nfsd4_create_session(struct svc_rqst *rqstp,
2769                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2770 {
2771         struct nfsd4_create_session *cr_ses = &u->create_session;
2772         struct sockaddr *sa = svc_addr(rqstp);
2773         struct nfs4_client *conf, *unconf;
2774         struct nfs4_client *old = NULL;
2775         struct nfsd4_session *new;
2776         struct nfsd4_conn *conn;
2777         struct nfsd4_clid_slot *cs_slot = NULL;
2778         __be32 status = 0;
2779         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2780
2781         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2782                 return nfserr_inval;
2783         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2784         if (status)
2785                 return status;
2786         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2787         if (status)
2788                 return status;
2789         status = check_backchannel_attrs(&cr_ses->back_channel);
2790         if (status)
2791                 goto out_release_drc_mem;
2792         status = nfserr_jukebox;
2793         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2794         if (!new)
2795                 goto out_release_drc_mem;
2796         conn = alloc_conn_from_crses(rqstp, cr_ses);
2797         if (!conn)
2798                 goto out_free_session;
2799
2800         spin_lock(&nn->client_lock);
2801         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2802         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2803         WARN_ON_ONCE(conf && unconf);
2804
2805         if (conf) {
2806                 status = nfserr_wrong_cred;
2807                 if (!nfsd4_mach_creds_match(conf, rqstp))
2808                         goto out_free_conn;
2809                 cs_slot = &conf->cl_cs_slot;
2810                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2811                 if (status) {
2812                         if (status == nfserr_replay_cache)
2813                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2814                         goto out_free_conn;
2815                 }
2816         } else if (unconf) {
2817                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2818                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2819                         status = nfserr_clid_inuse;
2820                         goto out_free_conn;
2821                 }
2822                 status = nfserr_wrong_cred;
2823                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2824                         goto out_free_conn;
2825                 cs_slot = &unconf->cl_cs_slot;
2826                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2827                 if (status) {
2828                         /* an unconfirmed replay returns misordered */
2829                         status = nfserr_seq_misordered;
2830                         goto out_free_conn;
2831                 }
2832                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2833                 if (old) {
2834                         status = mark_client_expired_locked(old);
2835                         if (status) {
2836                                 old = NULL;
2837                                 goto out_free_conn;
2838                         }
2839                 }
2840                 move_to_confirmed(unconf);
2841                 conf = unconf;
2842         } else {
2843                 status = nfserr_stale_clientid;
2844                 goto out_free_conn;
2845         }
2846         status = nfs_ok;
2847         /* Persistent sessions are not supported */
2848         cr_ses->flags &= ~SESSION4_PERSIST;
2849         /* Upshifting from TCP to RDMA is not supported */
2850         cr_ses->flags &= ~SESSION4_RDMA;
2851
2852         init_session(rqstp, new, conf, cr_ses);
2853         nfsd4_get_session_locked(new);
2854
2855         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2856                NFS4_MAX_SESSIONID_LEN);
2857         cs_slot->sl_seqid++;
2858         cr_ses->seqid = cs_slot->sl_seqid;
2859
2860         /* cache solo and embedded create sessions under the client_lock */
2861         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2862         spin_unlock(&nn->client_lock);
2863         /* init connection and backchannel */
2864         nfsd4_init_conn(rqstp, conn, new);
2865         nfsd4_put_session(new);
2866         if (old)
2867                 expire_client(old);
2868         return status;
2869 out_free_conn:
2870         spin_unlock(&nn->client_lock);
2871         free_conn(conn);
2872         if (old)
2873                 expire_client(old);
2874 out_free_session:
2875         __free_session(new);
2876 out_release_drc_mem:
2877         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2878         return status;
2879 }
2880
2881 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2882 {
2883         switch (*dir) {
2884         case NFS4_CDFC4_FORE:
2885         case NFS4_CDFC4_BACK:
2886                 return nfs_ok;
2887         case NFS4_CDFC4_FORE_OR_BOTH:
2888         case NFS4_CDFC4_BACK_OR_BOTH:
2889                 *dir = NFS4_CDFC4_BOTH;
2890                 return nfs_ok;
2891         };
2892         return nfserr_inval;
2893 }
2894
2895 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2896                 struct nfsd4_compound_state *cstate,
2897                 union nfsd4_op_u *u)
2898 {
2899         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2900         struct nfsd4_session *session = cstate->session;
2901         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2902         __be32 status;
2903
2904         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2905         if (status)
2906                 return status;
2907         spin_lock(&nn->client_lock);
2908         session->se_cb_prog = bc->bc_cb_program;
2909         session->se_cb_sec = bc->bc_cb_sec;
2910         spin_unlock(&nn->client_lock);
2911
2912         nfsd4_probe_callback(session->se_client);
2913
2914         return nfs_ok;
2915 }
2916
2917 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2918                      struct nfsd4_compound_state *cstate,
2919                      union nfsd4_op_u *u)
2920 {
2921         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2922         __be32 status;
2923         struct nfsd4_conn *conn;
2924         struct nfsd4_session *session;
2925         struct net *net = SVC_NET(rqstp);
2926         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2927
2928         if (!nfsd4_last_compound_op(rqstp))
2929                 return nfserr_not_only_op;
2930         spin_lock(&nn->client_lock);
2931         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2932         spin_unlock(&nn->client_lock);
2933         if (!session)
2934                 goto out_no_session;
2935         status = nfserr_wrong_cred;
2936         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2937                 goto out;
2938         status = nfsd4_map_bcts_dir(&bcts->dir);
2939         if (status)
2940                 goto out;
2941         conn = alloc_conn(rqstp, bcts->dir);
2942         status = nfserr_jukebox;
2943         if (!conn)
2944                 goto out;
2945         nfsd4_init_conn(rqstp, conn, session);
2946         status = nfs_ok;
2947 out:
2948         nfsd4_put_session(session);
2949 out_no_session:
2950         return status;
2951 }
2952
2953 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2954 {
2955         if (!session)
2956                 return 0;
2957         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2958 }
2959
2960 __be32
2961 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2962                 union nfsd4_op_u *u)
2963 {
2964         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2965         struct nfsd4_session *ses;
2966         __be32 status;
2967         int ref_held_by_me = 0;
2968         struct net *net = SVC_NET(r);
2969         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2970
2971         status = nfserr_not_only_op;
2972         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2973                 if (!nfsd4_last_compound_op(r))
2974                         goto out;
2975                 ref_held_by_me++;
2976         }
2977         dump_sessionid(__func__, &sessionid->sessionid);
2978         spin_lock(&nn->client_lock);
2979         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2980         if (!ses)
2981                 goto out_client_lock;
2982         status = nfserr_wrong_cred;
2983         if (!nfsd4_mach_creds_match(ses->se_client, r))
2984                 goto out_put_session;
2985         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2986         if (status)
2987                 goto out_put_session;
2988         unhash_session(ses);
2989         spin_unlock(&nn->client_lock);
2990
2991         nfsd4_probe_callback_sync(ses->se_client);
2992
2993         spin_lock(&nn->client_lock);
2994         status = nfs_ok;
2995 out_put_session:
2996         nfsd4_put_session_locked(ses);
2997 out_client_lock:
2998         spin_unlock(&nn->client_lock);
2999 out:
3000         return status;
3001 }
3002
3003 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3004 {
3005         struct nfsd4_conn *c;
3006
3007         list_for_each_entry(c, &s->se_conns, cn_persession) {
3008                 if (c->cn_xprt == xpt) {
3009                         return c;
3010                 }
3011         }
3012         return NULL;
3013 }
3014
3015 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3016 {
3017         struct nfs4_client *clp = ses->se_client;
3018         struct nfsd4_conn *c;
3019         __be32 status = nfs_ok;
3020         int ret;
3021
3022         spin_lock(&clp->cl_lock);
3023         c = __nfsd4_find_conn(new->cn_xprt, ses);
3024         if (c)
3025                 goto out_free;
3026         status = nfserr_conn_not_bound_to_session;
3027         if (clp->cl_mach_cred)
3028                 goto out_free;
3029         __nfsd4_hash_conn(new, ses);
3030         spin_unlock(&clp->cl_lock);
3031         ret = nfsd4_register_conn(new);
3032         if (ret)
3033                 /* oops; xprt is already down: */
3034                 nfsd4_conn_lost(&new->cn_xpt_user);
3035         return nfs_ok;
3036 out_free:
3037         spin_unlock(&clp->cl_lock);
3038         free_conn(new);
3039         return status;
3040 }
3041
3042 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3043 {
3044         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3045
3046         return args->opcnt > session->se_fchannel.maxops;
3047 }
3048
3049 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3050                                   struct nfsd4_session *session)
3051 {
3052         struct xdr_buf *xb = &rqstp->rq_arg;
3053
3054         return xb->len > session->se_fchannel.maxreq_sz;
3055 }
3056
3057 static bool replay_matches_cache(struct svc_rqst *rqstp,
3058                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3059 {
3060         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3061
3062         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3063             (bool)seq->cachethis)
3064                 return false;
3065         /*
3066          * If there's an error than the reply can have fewer ops than
3067          * the call.  But if we cached a reply with *more* ops than the
3068          * call you're sending us now, then this new call is clearly not
3069          * really a replay of the old one:
3070          */
3071         if (slot->sl_opcnt < argp->opcnt)
3072                 return false;
3073         /* This is the only check explicitly called by spec: */
3074         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3075                 return false;
3076         /*
3077          * There may be more comparisons we could actually do, but the
3078          * spec doesn't require us to catch every case where the calls
3079          * don't match (that would require caching the call as well as
3080          * the reply), so we don't bother.
3081          */
3082         return true;
3083 }
3084
3085 __be32
3086 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3087                 union nfsd4_op_u *u)
3088 {
3089         struct nfsd4_sequence *seq = &u->sequence;
3090         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3091         struct xdr_stream *xdr = &resp->xdr;
3092         struct nfsd4_session *session;
3093         struct nfs4_client *clp;
3094         struct nfsd4_slot *slot;
3095         struct nfsd4_conn *conn;
3096         __be32 status;
3097         int buflen;
3098         struct net *net = SVC_NET(rqstp);
3099         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3100
3101         if (resp->opcnt != 1)
3102                 return nfserr_sequence_pos;
3103
3104         /*
3105          * Will be either used or freed by nfsd4_sequence_check_conn
3106          * below.
3107          */
3108         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3109         if (!conn)
3110                 return nfserr_jukebox;
3111
3112         spin_lock(&nn->client_lock);
3113         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3114         if (!session)
3115                 goto out_no_session;
3116         clp = session->se_client;
3117
3118         status = nfserr_too_many_ops;
3119         if (nfsd4_session_too_many_ops(rqstp, session))
3120                 goto out_put_session;
3121
3122         status = nfserr_req_too_big;
3123         if (nfsd4_request_too_big(rqstp, session))
3124                 goto out_put_session;
3125
3126         status = nfserr_badslot;
3127         if (seq->slotid >= session->se_fchannel.maxreqs)
3128                 goto out_put_session;
3129
3130         slot = session->se_slots[seq->slotid];
3131         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3132
3133         /* We do not negotiate the number of slots yet, so set the
3134          * maxslots to the session maxreqs which is used to encode
3135          * sr_highest_slotid and the sr_target_slot id to maxslots */
3136         seq->maxslots = session->se_fchannel.maxreqs;
3137
3138         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3139                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3140         if (status == nfserr_replay_cache) {
3141                 status = nfserr_seq_misordered;
3142                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3143                         goto out_put_session;
3144                 status = nfserr_seq_false_retry;
3145                 if (!replay_matches_cache(rqstp, seq, slot))
3146                         goto out_put_session;
3147                 cstate->slot = slot;
3148                 cstate->session = session;
3149                 cstate->clp = clp;
3150                 /* Return the cached reply status and set cstate->status
3151                  * for nfsd4_proc_compound processing */
3152                 status = nfsd4_replay_cache_entry(resp, seq);
3153                 cstate->status = nfserr_replay_cache;
3154                 goto out;
3155         }
3156         if (status)
3157                 goto out_put_session;
3158
3159         status = nfsd4_sequence_check_conn(conn, session);
3160         conn = NULL;
3161         if (status)
3162                 goto out_put_session;
3163
3164         buflen = (seq->cachethis) ?
3165                         session->se_fchannel.maxresp_cached :
3166                         session->se_fchannel.maxresp_sz;
3167         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3168                                     nfserr_rep_too_big;
3169         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3170                 goto out_put_session;
3171         svc_reserve(rqstp, buflen);
3172
3173         status = nfs_ok;
3174         /* Success! bump slot seqid */
3175         slot->sl_seqid = seq->seqid;
3176         slot->sl_flags |= NFSD4_SLOT_INUSE;
3177         if (seq->cachethis)
3178                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3179         else
3180                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3181
3182         cstate->slot = slot;
3183         cstate->session = session;
3184         cstate->clp = clp;
3185
3186 out:
3187         switch (clp->cl_cb_state) {
3188         case NFSD4_CB_DOWN:
3189                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3190                 break;
3191         case NFSD4_CB_FAULT:
3192                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3193                 break;
3194         default:
3195                 seq->status_flags = 0;
3196         }
3197         if (!list_empty(&clp->cl_revoked))
3198                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3199 out_no_session:
3200         if (conn)
3201                 free_conn(conn);
3202         spin_unlock(&nn->client_lock);
3203         return status;
3204 out_put_session:
3205         nfsd4_put_session_locked(session);
3206         goto out_no_session;
3207 }
3208
3209 void
3210 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3211 {
3212         struct nfsd4_compound_state *cs = &resp->cstate;
3213
3214         if (nfsd4_has_session(cs)) {
3215                 if (cs->status != nfserr_replay_cache) {
3216                         nfsd4_store_cache_entry(resp);
3217                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3218                 }
3219                 /* Drop session reference that was taken in nfsd4_sequence() */
3220                 nfsd4_put_session(cs->session);
3221         } else if (cs->clp)
3222                 put_client_renew(cs->clp);
3223 }
3224
3225 __be32
3226 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3227                 struct nfsd4_compound_state *cstate,
3228                 union nfsd4_op_u *u)
3229 {
3230         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3231         struct nfs4_client *conf, *unconf;
3232         struct nfs4_client *clp = NULL;
3233         __be32 status = 0;
3234         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3235
3236         spin_lock(&nn->client_lock);
3237         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3238         conf = find_confirmed_client(&dc->clientid, true, nn);
3239         WARN_ON_ONCE(conf && unconf);
3240
3241         if (conf) {
3242                 if (client_has_state(conf)) {
3243                         status = nfserr_clientid_busy;
3244                         goto out;
3245                 }
3246                 status = mark_client_expired_locked(conf);
3247                 if (status)
3248                         goto out;
3249                 clp = conf;
3250         } else if (unconf)
3251                 clp = unconf;
3252         else {
3253                 status = nfserr_stale_clientid;
3254                 goto out;
3255         }
3256         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3257                 clp = NULL;
3258                 status = nfserr_wrong_cred;
3259                 goto out;
3260         }
3261         unhash_client_locked(clp);
3262 out:
3263         spin_unlock(&nn->client_lock);
3264         if (clp)
3265                 expire_client(clp);
3266         return status;
3267 }
3268
3269 __be32
3270 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3271                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3272 {
3273         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3274         __be32 status = 0;
3275
3276         if (rc->rca_one_fs) {
3277                 if (!cstate->current_fh.fh_dentry)
3278                         return nfserr_nofilehandle;
3279                 /*
3280                  * We don't take advantage of the rca_one_fs case.
3281                  * That's OK, it's optional, we can safely ignore it.
3282                  */
3283                 return nfs_ok;
3284         }
3285
3286         status = nfserr_complete_already;
3287         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3288                              &cstate->session->se_client->cl_flags))
3289                 goto out;
3290
3291         status = nfserr_stale_clientid;
3292         if (is_client_expired(cstate->session->se_client))
3293                 /*
3294                  * The following error isn't really legal.
3295                  * But we only get here if the client just explicitly
3296                  * destroyed the client.  Surely it no longer cares what
3297                  * error it gets back on an operation for the dead
3298                  * client.
3299                  */
3300                 goto out;
3301
3302         status = nfs_ok;
3303         nfsd4_client_record_create(cstate->session->se_client);
3304 out:
3305         return status;
3306 }
3307
3308 __be32
3309 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3310                   union nfsd4_op_u *u)
3311 {
3312         struct nfsd4_setclientid *setclid = &u->setclientid;
3313         struct xdr_netobj       clname = setclid->se_name;
3314         nfs4_verifier           clverifier = setclid->se_verf;
3315         struct nfs4_client      *conf, *new;
3316         struct nfs4_client      *unconf = NULL;
3317         __be32                  status;
3318         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3319
3320         new = create_client(clname, rqstp, &clverifier);
3321         if (new == NULL)
3322                 return nfserr_jukebox;
3323         /* Cases below refer to rfc 3530 section 14.2.33: */
3324         spin_lock(&nn->client_lock);
3325         conf = find_confirmed_client_by_name(&clname, nn);
3326         if (conf && client_has_state(conf)) {
3327                 /* case 0: */
3328                 status = nfserr_clid_inuse;
3329                 if (clp_used_exchangeid(conf))
3330                         goto out;
3331                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3332                         char addr_str[INET6_ADDRSTRLEN];
3333                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3334                                  sizeof(addr_str));
3335                         dprintk("NFSD: setclientid: string in use by client "
3336                                 "at %s\n", addr_str);
3337                         goto out;
3338                 }
3339         }
3340         unconf = find_unconfirmed_client_by_name(&clname, nn);
3341         if (unconf)
3342                 unhash_client_locked(unconf);
3343         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3344                 /* case 1: probable callback update */
3345                 copy_clid(new, conf);
3346                 gen_confirm(new, nn);
3347         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3348                 gen_clid(new, nn);
3349         new->cl_minorversion = 0;
3350         gen_callback(new, setclid, rqstp);
3351         add_to_unconfirmed(new);
3352         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3353         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3354         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3355         new = NULL;
3356         status = nfs_ok;
3357 out:
3358         spin_unlock(&nn->client_lock);
3359         if (new)
3360                 free_client(new);
3361         if (unconf)
3362                 expire_client(unconf);
3363         return status;
3364 }
3365
3366
3367 __be32
3368 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3369                         struct nfsd4_compound_state *cstate,
3370                         union nfsd4_op_u *u)
3371 {
3372         struct nfsd4_setclientid_confirm *setclientid_confirm =
3373                         &u->setclientid_confirm;
3374         struct nfs4_client *conf, *unconf;
3375         struct nfs4_client *old = NULL;
3376         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3377         clientid_t * clid = &setclientid_confirm->sc_clientid;
3378         __be32 status;
3379         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3380
3381         if (STALE_CLIENTID(clid, nn))
3382                 return nfserr_stale_clientid;
3383
3384         spin_lock(&nn->client_lock);
3385         conf = find_confirmed_client(clid, false, nn);
3386         unconf = find_unconfirmed_client(clid, false, nn);
3387         /*
3388          * We try hard to give out unique clientid's, so if we get an
3389          * attempt to confirm the same clientid with a different cred,
3390          * the client may be buggy; this should never happen.
3391          *
3392          * Nevertheless, RFC 7530 recommends INUSE for this case:
3393          */
3394         status = nfserr_clid_inuse;
3395         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3396                 goto out;
3397         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3398                 goto out;
3399         /* cases below refer to rfc 3530 section 14.2.34: */
3400         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3401                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3402                         /* case 2: probable retransmit */
3403                         status = nfs_ok;
3404                 } else /* case 4: client hasn't noticed we rebooted yet? */
3405                         status = nfserr_stale_clientid;
3406                 goto out;
3407         }
3408         status = nfs_ok;
3409         if (conf) { /* case 1: callback update */
3410                 old = unconf;
3411                 unhash_client_locked(old);
3412                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3413         } else { /* case 3: normal case; new or rebooted client */
3414                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3415                 if (old) {
3416                         status = nfserr_clid_inuse;
3417                         if (client_has_state(old)
3418                                         && !same_creds(&unconf->cl_cred,
3419                                                         &old->cl_cred))
3420                                 goto out;
3421                         status = mark_client_expired_locked(old);
3422                         if (status) {
3423                                 old = NULL;
3424                                 goto out;
3425                         }
3426                 }
3427                 move_to_confirmed(unconf);
3428                 conf = unconf;
3429         }
3430         get_client_locked(conf);
3431         spin_unlock(&nn->client_lock);
3432         nfsd4_probe_callback(conf);
3433         spin_lock(&nn->client_lock);
3434         put_client_renew_locked(conf);
3435 out:
3436         spin_unlock(&nn->client_lock);
3437         if (old)
3438                 expire_client(old);
3439         return status;
3440 }
3441
3442 static struct nfs4_file *nfsd4_alloc_file(void)
3443 {
3444         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3445 }
3446
3447 /* OPEN Share state helper functions */
3448 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3449                                 struct nfs4_file *fp)
3450 {
3451         lockdep_assert_held(&state_lock);
3452
3453         refcount_set(&fp->fi_ref, 1);
3454         spin_lock_init(&fp->fi_lock);
3455         INIT_LIST_HEAD(&fp->fi_stateids);
3456         INIT_LIST_HEAD(&fp->fi_delegations);
3457         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3458         fh_copy_shallow(&fp->fi_fhandle, fh);
3459         fp->fi_deleg_file = NULL;
3460         fp->fi_had_conflict = false;
3461         fp->fi_share_deny = 0;
3462         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3463         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3464 #ifdef CONFIG_NFSD_PNFS
3465         INIT_LIST_HEAD(&fp->fi_lo_states);
3466         atomic_set(&fp->fi_lo_recalls, 0);
3467 #endif
3468         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3469 }
3470
3471 void
3472 nfsd4_free_slabs(void)
3473 {
3474         kmem_cache_destroy(odstate_slab);
3475         kmem_cache_destroy(openowner_slab);
3476         kmem_cache_destroy(lockowner_slab);
3477         kmem_cache_destroy(file_slab);
3478         kmem_cache_destroy(stateid_slab);
3479         kmem_cache_destroy(deleg_slab);
3480 }
3481
3482 int
3483 nfsd4_init_slabs(void)
3484 {
3485         openowner_slab = kmem_cache_create("nfsd4_openowners",
3486                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3487         if (openowner_slab == NULL)
3488                 goto out;
3489         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3490                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3491         if (lockowner_slab == NULL)
3492                 goto out_free_openowner_slab;
3493         file_slab = kmem_cache_create("nfsd4_files",
3494                         sizeof(struct nfs4_file), 0, 0, NULL);
3495         if (file_slab == NULL)
3496                 goto out_free_lockowner_slab;
3497         stateid_slab = kmem_cache_create("nfsd4_stateids",
3498                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3499         if (stateid_slab == NULL)
3500                 goto out_free_file_slab;
3501         deleg_slab = kmem_cache_create("nfsd4_delegations",
3502                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3503         if (deleg_slab == NULL)
3504                 goto out_free_stateid_slab;
3505         odstate_slab = kmem_cache_create("nfsd4_odstate",
3506                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3507         if (odstate_slab == NULL)
3508                 goto out_free_deleg_slab;
3509         return 0;
3510
3511 out_free_deleg_slab:
3512         kmem_cache_destroy(deleg_slab);
3513 out_free_stateid_slab:
3514         kmem_cache_destroy(stateid_slab);
3515 out_free_file_slab:
3516         kmem_cache_destroy(file_slab);
3517 out_free_lockowner_slab:
3518         kmem_cache_destroy(lockowner_slab);
3519 out_free_openowner_slab:
3520         kmem_cache_destroy(openowner_slab);
3521 out:
3522         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3523         return -ENOMEM;
3524 }
3525
3526 static void init_nfs4_replay(struct nfs4_replay *rp)
3527 {
3528         rp->rp_status = nfserr_serverfault;
3529         rp->rp_buflen = 0;
3530         rp->rp_buf = rp->rp_ibuf;
3531         mutex_init(&rp->rp_mutex);
3532 }
3533
3534 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3535                 struct nfs4_stateowner *so)
3536 {
3537         if (!nfsd4_has_session(cstate)) {
3538                 mutex_lock(&so->so_replay.rp_mutex);
3539                 cstate->replay_owner = nfs4_get_stateowner(so);
3540         }
3541 }
3542
3543 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3544 {
3545         struct nfs4_stateowner *so = cstate->replay_owner;
3546
3547         if (so != NULL) {
3548                 cstate->replay_owner = NULL;
3549                 mutex_unlock(&so->so_replay.rp_mutex);
3550                 nfs4_put_stateowner(so);
3551         }
3552 }
3553
3554 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3555 {
3556         struct nfs4_stateowner *sop;
3557
3558         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3559         if (!sop)
3560                 return NULL;
3561
3562         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3563         if (!sop->so_owner.data) {
3564                 kmem_cache_free(slab, sop);
3565                 return NULL;
3566         }
3567         sop->so_owner.len = owner->len;
3568
3569         INIT_LIST_HEAD(&sop->so_stateids);
3570         sop->so_client = clp;
3571         init_nfs4_replay(&sop->so_replay);
3572         atomic_set(&sop->so_count, 1);
3573         return sop;
3574 }
3575
3576 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3577 {
3578         lockdep_assert_held(&clp->cl_lock);
3579
3580         list_add(&oo->oo_owner.so_strhash,
3581                  &clp->cl_ownerstr_hashtbl[strhashval]);
3582         list_add(&oo->oo_perclient, &clp->cl_openowners);
3583 }
3584
3585 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3586 {
3587         unhash_openowner_locked(openowner(so));
3588 }
3589
3590 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3591 {
3592         struct nfs4_openowner *oo = openowner(so);
3593
3594         kmem_cache_free(openowner_slab, oo);
3595 }
3596
3597 static const struct nfs4_stateowner_operations openowner_ops = {
3598         .so_unhash =    nfs4_unhash_openowner,
3599         .so_free =      nfs4_free_openowner,
3600 };
3601
3602 static struct nfs4_ol_stateid *
3603 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3604 {
3605         struct nfs4_ol_stateid *local, *ret = NULL;
3606         struct nfs4_openowner *oo = open->op_openowner;
3607
3608         lockdep_assert_held(&fp->fi_lock);
3609
3610         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3611                 /* ignore lock owners */
3612                 if (local->st_stateowner->so_is_open_owner == 0)
3613                         continue;
3614                 if (local->st_stateowner != &oo->oo_owner)
3615                         continue;
3616                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3617                         ret = local;
3618                         refcount_inc(&ret->st_stid.sc_count);
3619                         break;
3620                 }
3621         }
3622         return ret;
3623 }
3624
3625 static __be32
3626 nfsd4_verify_open_stid(struct nfs4_stid *s)
3627 {
3628         __be32 ret = nfs_ok;
3629
3630         switch (s->sc_type) {
3631         default:
3632                 break;
3633         case 0:
3634         case NFS4_CLOSED_STID:
3635         case NFS4_CLOSED_DELEG_STID:
3636                 ret = nfserr_bad_stateid;
3637                 break;
3638         case NFS4_REVOKED_DELEG_STID:
3639                 ret = nfserr_deleg_revoked;
3640         }
3641         return ret;
3642 }
3643
3644 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3645 static __be32
3646 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3647 {
3648         __be32 ret;
3649
3650         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3651         ret = nfsd4_verify_open_stid(&stp->st_stid);
3652         if (ret != nfs_ok)
3653                 mutex_unlock(&stp->st_mutex);
3654         return ret;
3655 }
3656
3657 static struct nfs4_ol_stateid *
3658 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3659 {
3660         struct nfs4_ol_stateid *stp;
3661         for (;;) {
3662                 spin_lock(&fp->fi_lock);
3663                 stp = nfsd4_find_existing_open(fp, open);
3664                 spin_unlock(&fp->fi_lock);
3665                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3666                         break;
3667                 nfs4_put_stid(&stp->st_stid);
3668         }
3669         return stp;
3670 }
3671
3672 static struct nfs4_openowner *
3673 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3674                            struct nfsd4_compound_state *cstate)
3675 {
3676         struct nfs4_client *clp = cstate->clp;
3677         struct nfs4_openowner *oo, *ret;
3678
3679         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3680         if (!oo)
3681                 return NULL;
3682         oo->oo_owner.so_ops = &openowner_ops;
3683         oo->oo_owner.so_is_open_owner = 1;
3684         oo->oo_owner.so_seqid = open->op_seqid;
3685         oo->oo_flags = 0;
3686         if (nfsd4_has_session(cstate))
3687                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3688         oo->oo_time = 0;
3689         oo->oo_last_closed_stid = NULL;
3690         INIT_LIST_HEAD(&oo->oo_close_lru);
3691         spin_lock(&clp->cl_lock);
3692         ret = find_openstateowner_str_locked(strhashval, open, clp);
3693         if (ret == NULL) {
3694                 hash_openowner(oo, clp, strhashval);
3695                 ret = oo;
3696         } else
3697                 nfs4_free_stateowner(&oo->oo_owner);
3698
3699         spin_unlock(&clp->cl_lock);
3700         return ret;
3701 }
3702
3703 static struct nfs4_ol_stateid *
3704 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3705 {
3706
3707         struct nfs4_openowner *oo = open->op_openowner;
3708         struct nfs4_ol_stateid *retstp = NULL;
3709         struct nfs4_ol_stateid *stp;
3710
3711         stp = open->op_stp;
3712         /* We are moving these outside of the spinlocks to avoid the warnings */
3713         mutex_init(&stp->st_mutex);
3714         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3715
3716 retry:
3717         spin_lock(&oo->oo_owner.so_client->cl_lock);
3718         spin_lock(&fp->fi_lock);
3719
3720         retstp = nfsd4_find_existing_open(fp, open);
3721         if (retstp)
3722                 goto out_unlock;
3723
3724         open->op_stp = NULL;
3725         refcount_inc(&stp->st_stid.sc_count);
3726         stp->st_stid.sc_type = NFS4_OPEN_STID;
3727         INIT_LIST_HEAD(&stp->st_locks);
3728         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3729         get_nfs4_file(fp);
3730         stp->st_stid.sc_file = fp;
3731         stp->st_access_bmap = 0;
3732         stp->st_deny_bmap = 0;
3733         stp->st_openstp = NULL;
3734         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3735         list_add(&stp->st_perfile, &fp->fi_stateids);
3736
3737 out_unlock:
3738         spin_unlock(&fp->fi_lock);
3739         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3740         if (retstp) {
3741                 /* Handle races with CLOSE */
3742                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3743                         nfs4_put_stid(&retstp->st_stid);
3744                         goto retry;
3745                 }
3746                 /* To keep mutex tracking happy */
3747                 mutex_unlock(&stp->st_mutex);
3748                 stp = retstp;
3749         }
3750         return stp;
3751 }
3752
3753 /*
3754  * In the 4.0 case we need to keep the owners around a little while to handle
3755  * CLOSE replay. We still do need to release any file access that is held by
3756  * them before returning however.
3757  */
3758 static void
3759 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3760 {
3761         struct nfs4_ol_stateid *last;
3762         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3763         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3764                                                 nfsd_net_id);
3765
3766         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3767
3768         /*
3769          * We know that we hold one reference via nfsd4_close, and another
3770          * "persistent" reference for the client. If the refcount is higher
3771          * than 2, then there are still calls in progress that are using this
3772          * stateid. We can't put the sc_file reference until they are finished.
3773          * Wait for the refcount to drop to 2. Since it has been unhashed,
3774          * there should be no danger of the refcount going back up again at
3775          * this point.
3776          */
3777         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3778
3779         release_all_access(s);
3780         if (s->st_stid.sc_file) {
3781                 put_nfs4_file(s->st_stid.sc_file);
3782                 s->st_stid.sc_file = NULL;
3783         }
3784
3785         spin_lock(&nn->client_lock);
3786         last = oo->oo_last_closed_stid;
3787         oo->oo_last_closed_stid = s;
3788         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3789         oo->oo_time = get_seconds();
3790         spin_unlock(&nn->client_lock);
3791         if (last)
3792                 nfs4_put_stid(&last->st_stid);
3793 }
3794
3795 /* search file_hashtbl[] for file */
3796 static struct nfs4_file *
3797 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3798 {
3799         struct nfs4_file *fp;
3800
3801         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3802                 if (fh_match(&fp->fi_fhandle, fh)) {
3803                         if (refcount_inc_not_zero(&fp->fi_ref))
3804                                 return fp;
3805                 }
3806         }
3807         return NULL;
3808 }
3809
3810 struct nfs4_file *
3811 find_file(struct knfsd_fh *fh)
3812 {
3813         struct nfs4_file *fp;
3814         unsigned int hashval = file_hashval(fh);
3815
3816         rcu_read_lock();
3817         fp = find_file_locked(fh, hashval);
3818         rcu_read_unlock();
3819         return fp;
3820 }
3821
3822 static struct nfs4_file *
3823 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3824 {
3825         struct nfs4_file *fp;
3826         unsigned int hashval = file_hashval(fh);
3827
3828         rcu_read_lock();
3829         fp = find_file_locked(fh, hashval);
3830         rcu_read_unlock();
3831         if (fp)
3832                 return fp;
3833
3834         spin_lock(&state_lock);
3835         fp = find_file_locked(fh, hashval);
3836         if (likely(fp == NULL)) {
3837                 nfsd4_init_file(fh, hashval, new);
3838                 fp = new;
3839         }
3840         spin_unlock(&state_lock);
3841
3842         return fp;
3843 }
3844
3845 /*
3846  * Called to check deny when READ with all zero stateid or
3847  * WRITE with all zero or all one stateid
3848  */
3849 static __be32
3850 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3851 {
3852         struct nfs4_file *fp;
3853         __be32 ret = nfs_ok;
3854
3855         fp = find_file(&current_fh->fh_handle);
3856         if (!fp)
3857                 return ret;
3858         /* Check for conflicting share reservations */
3859         spin_lock(&fp->fi_lock);
3860         if (fp->fi_share_deny & deny_type)
3861                 ret = nfserr_locked;
3862         spin_unlock(&fp->fi_lock);
3863         put_nfs4_file(fp);
3864         return ret;
3865 }
3866
3867 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3868 {
3869         struct nfs4_delegation *dp = cb_to_delegation(cb);
3870         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3871                                           nfsd_net_id);
3872
3873         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3874
3875         /*
3876          * We can't do this in nfsd_break_deleg_cb because it is
3877          * already holding inode->i_lock.
3878          *
3879          * If the dl_time != 0, then we know that it has already been
3880          * queued for a lease break. Don't queue it again.
3881          */
3882         spin_lock(&state_lock);
3883         if (dp->dl_time == 0) {
3884                 dp->dl_time = get_seconds();
3885                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3886         }
3887         spin_unlock(&state_lock);
3888 }
3889
3890 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3891                 struct rpc_task *task)
3892 {
3893         struct nfs4_delegation *dp = cb_to_delegation(cb);
3894
3895         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3896                 return 1;
3897
3898         switch (task->tk_status) {
3899         case 0:
3900                 return 1;
3901         case -EBADHANDLE:
3902         case -NFS4ERR_BAD_STATEID:
3903                 /*
3904                  * Race: client probably got cb_recall before open reply
3905                  * granting delegation.
3906                  */
3907                 if (dp->dl_retries--) {
3908                         rpc_delay(task, 2 * HZ);
3909                         return 0;
3910                 }
3911                 /*FALLTHRU*/
3912         default:
3913                 return -1;
3914         }
3915 }
3916
3917 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3918 {
3919         struct nfs4_delegation *dp = cb_to_delegation(cb);
3920
3921         nfs4_put_stid(&dp->dl_stid);
3922 }
3923
3924 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3925         .prepare        = nfsd4_cb_recall_prepare,
3926         .done           = nfsd4_cb_recall_done,
3927         .release        = nfsd4_cb_recall_release,
3928 };
3929
3930 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3931 {
3932         /*
3933          * We're assuming the state code never drops its reference
3934          * without first removing the lease.  Since we're in this lease
3935          * callback (and since the lease code is serialized by the kernel
3936          * lock) we know the server hasn't removed the lease yet, we know
3937          * it's safe to take a reference.
3938          */
3939         refcount_inc(&dp->dl_stid.sc_count);
3940         nfsd4_run_cb(&dp->dl_recall);
3941 }
3942
3943 /* Called from break_lease() with i_lock held. */
3944 static bool
3945 nfsd_break_deleg_cb(struct file_lock *fl)
3946 {
3947         bool ret = false;
3948         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3949         struct nfs4_delegation *dp;
3950
3951         if (!fp) {
3952                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3953                 return ret;
3954         }
3955         if (fp->fi_had_conflict) {
3956                 WARN(1, "duplicate break on %p\n", fp);
3957                 return ret;
3958         }
3959         /*
3960          * We don't want the locks code to timeout the lease for us;
3961          * we'll remove it ourself if a delegation isn't returned
3962          * in time:
3963          */
3964         fl->fl_break_time = 0;
3965
3966         spin_lock(&fp->fi_lock);
3967         fp->fi_had_conflict = true;
3968         /*
3969          * If there are no delegations on the list, then return true
3970          * so that the lease code will go ahead and delete it.
3971          */
3972         if (list_empty(&fp->fi_delegations))
3973                 ret = true;
3974         else
3975                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3976                         nfsd_break_one_deleg(dp);
3977         spin_unlock(&fp->fi_lock);
3978         return ret;
3979 }
3980
3981 static int
3982 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3983                      struct list_head *dispose)
3984 {
3985         if (arg & F_UNLCK)
3986                 return lease_modify(onlist, arg, dispose);
3987         else
3988                 return -EAGAIN;
3989 }
3990
3991 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3992         .lm_break = nfsd_break_deleg_cb,
3993         .lm_change = nfsd_change_deleg_cb,
3994 };
3995
3996 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3997 {
3998         if (nfsd4_has_session(cstate))
3999                 return nfs_ok;
4000         if (seqid == so->so_seqid - 1)
4001                 return nfserr_replay_me;
4002         if (seqid == so->so_seqid)
4003                 return nfs_ok;
4004         return nfserr_bad_seqid;
4005 }
4006
4007 static __be32 lookup_clientid(clientid_t *clid,
4008                 struct nfsd4_compound_state *cstate,
4009                 struct nfsd_net *nn)
4010 {
4011         struct nfs4_client *found;
4012
4013         if (cstate->clp) {
4014                 found = cstate->clp;
4015                 if (!same_clid(&found->cl_clientid, clid))
4016                         return nfserr_stale_clientid;
4017                 return nfs_ok;
4018         }
4019
4020         if (STALE_CLIENTID(clid, nn))
4021                 return nfserr_stale_clientid;
4022
4023         /*
4024          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4025          * cached already then we know this is for is for v4.0 and "sessions"
4026          * will be false.
4027          */
4028         WARN_ON_ONCE(cstate->session);
4029         spin_lock(&nn->client_lock);
4030         found = find_confirmed_client(clid, false, nn);
4031         if (!found) {
4032                 spin_unlock(&nn->client_lock);
4033                 return nfserr_expired;
4034         }
4035         atomic_inc(&found->cl_refcount);
4036         spin_unlock(&nn->client_lock);
4037
4038         /* Cache the nfs4_client in cstate! */
4039         cstate->clp = found;
4040         return nfs_ok;
4041 }
4042
4043 __be32
4044 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4045                     struct nfsd4_open *open, struct nfsd_net *nn)
4046 {
4047         clientid_t *clientid = &open->op_clientid;
4048         struct nfs4_client *clp = NULL;
4049         unsigned int strhashval;
4050         struct nfs4_openowner *oo = NULL;
4051         __be32 status;
4052
4053         if (STALE_CLIENTID(&open->op_clientid, nn))
4054                 return nfserr_stale_clientid;
4055         /*
4056          * In case we need it later, after we've already created the
4057          * file and don't want to risk a further failure:
4058          */
4059         open->op_file = nfsd4_alloc_file();
4060         if (open->op_file == NULL)
4061                 return nfserr_jukebox;
4062
4063         status = lookup_clientid(clientid, cstate, nn);
4064         if (status)
4065                 return status;
4066         clp = cstate->clp;
4067
4068         strhashval = ownerstr_hashval(&open->op_owner);
4069         oo = find_openstateowner_str(strhashval, open, clp);
4070         open->op_openowner = oo;
4071         if (!oo) {
4072                 goto new_owner;
4073         }
4074         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4075                 /* Replace unconfirmed owners without checking for replay. */
4076                 release_openowner(oo);
4077                 open->op_openowner = NULL;
4078                 goto new_owner;
4079         }
4080         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4081         if (status)
4082                 return status;
4083         goto alloc_stateid;
4084 new_owner:
4085         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4086         if (oo == NULL)
4087                 return nfserr_jukebox;
4088         open->op_openowner = oo;
4089 alloc_stateid:
4090         open->op_stp = nfs4_alloc_open_stateid(clp);
4091         if (!open->op_stp)
4092                 return nfserr_jukebox;
4093
4094         if (nfsd4_has_session(cstate) &&
4095             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4096                 open->op_odstate = alloc_clnt_odstate(clp);
4097                 if (!open->op_odstate)
4098                         return nfserr_jukebox;
4099         }
4100
4101         return nfs_ok;
4102 }
4103
4104 static inline __be32
4105 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4106 {
4107         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4108                 return nfserr_openmode;
4109         else
4110                 return nfs_ok;
4111 }
4112
4113 static int share_access_to_flags(u32 share_access)
4114 {
4115         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4116 }
4117
4118 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4119 {
4120         struct nfs4_stid *ret;
4121
4122         ret = find_stateid_by_type(cl, s,
4123                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4124         if (!ret)
4125                 return NULL;
4126         return delegstateid(ret);
4127 }
4128
4129 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4130 {
4131         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4132                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4133 }
4134
4135 static __be32
4136 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4137                 struct nfs4_delegation **dp)
4138 {
4139         int flags;
4140         __be32 status = nfserr_bad_stateid;
4141         struct nfs4_delegation *deleg;
4142
4143         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4144         if (deleg == NULL)
4145                 goto out;
4146         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4147                 nfs4_put_stid(&deleg->dl_stid);
4148                 if (cl->cl_minorversion)
4149                         status = nfserr_deleg_revoked;
4150                 goto out;
4151         }
4152         flags = share_access_to_flags(open->op_share_access);
4153         status = nfs4_check_delegmode(deleg, flags);
4154         if (status) {
4155                 nfs4_put_stid(&deleg->dl_stid);
4156                 goto out;
4157         }
4158         *dp = deleg;
4159 out:
4160         if (!nfsd4_is_deleg_cur(open))
4161                 return nfs_ok;
4162         if (status)
4163                 return status;
4164         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4165         return nfs_ok;
4166 }
4167
4168 static inline int nfs4_access_to_access(u32 nfs4_access)
4169 {
4170         int flags = 0;
4171
4172         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4173                 flags |= NFSD_MAY_READ;
4174         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4175                 flags |= NFSD_MAY_WRITE;
4176         return flags;
4177 }
4178
4179 static inline __be32
4180 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4181                 struct nfsd4_open *open)
4182 {
4183         struct iattr iattr = {
4184                 .ia_valid = ATTR_SIZE,
4185                 .ia_size = 0,
4186         };
4187         if (!open->op_truncate)
4188                 return 0;
4189         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4190                 return nfserr_inval;
4191         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4192 }
4193
4194 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4195                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4196                 struct nfsd4_open *open)
4197 {
4198         struct file *filp = NULL;
4199         __be32 status;
4200         int oflag = nfs4_access_to_omode(open->op_share_access);
4201         int access = nfs4_access_to_access(open->op_share_access);
4202         unsigned char old_access_bmap, old_deny_bmap;
4203
4204         spin_lock(&fp->fi_lock);
4205
4206         /*
4207          * Are we trying to set a deny mode that would conflict with
4208          * current access?
4209          */
4210         status = nfs4_file_check_deny(fp, open->op_share_deny);
4211         if (status != nfs_ok) {
4212                 spin_unlock(&fp->fi_lock);
4213                 goto out;
4214         }
4215
4216         /* set access to the file */
4217         status = nfs4_file_get_access(fp, open->op_share_access);
4218         if (status != nfs_ok) {
4219                 spin_unlock(&fp->fi_lock);
4220                 goto out;
4221         }
4222
4223         /* Set access bits in stateid */
4224         old_access_bmap = stp->st_access_bmap;
4225         set_access(open->op_share_access, stp);
4226
4227         /* Set new deny mask */
4228         old_deny_bmap = stp->st_deny_bmap;
4229         set_deny(open->op_share_deny, stp);
4230         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4231
4232         if (!fp->fi_fds[oflag]) {
4233                 spin_unlock(&fp->fi_lock);
4234                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4235                 if (status)
4236                         goto out_put_access;
4237                 spin_lock(&fp->fi_lock);
4238                 if (!fp->fi_fds[oflag]) {
4239                         fp->fi_fds[oflag] = filp;
4240                         filp = NULL;
4241                 }
4242         }
4243         spin_unlock(&fp->fi_lock);
4244         if (filp)
4245                 fput(filp);
4246
4247         status = nfsd4_truncate(rqstp, cur_fh, open);
4248         if (status)
4249                 goto out_put_access;
4250 out:
4251         return status;
4252 out_put_access:
4253         stp->st_access_bmap = old_access_bmap;
4254         nfs4_file_put_access(fp, open->op_share_access);
4255         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4256         goto out;
4257 }
4258
4259 static __be32
4260 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4261 {
4262         __be32 status;
4263         unsigned char old_deny_bmap = stp->st_deny_bmap;
4264
4265         if (!test_access(open->op_share_access, stp))
4266                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4267
4268         /* test and set deny mode */
4269         spin_lock(&fp->fi_lock);
4270         status = nfs4_file_check_deny(fp, open->op_share_deny);
4271         if (status == nfs_ok) {
4272                 set_deny(open->op_share_deny, stp);
4273                 fp->fi_share_deny |=
4274                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4275         }
4276         spin_unlock(&fp->fi_lock);
4277
4278         if (status != nfs_ok)
4279                 return status;
4280
4281         status = nfsd4_truncate(rqstp, cur_fh, open);
4282         if (status != nfs_ok)
4283                 reset_union_bmap_deny(old_deny_bmap, stp);
4284         return status;
4285 }
4286
4287 /* Should we give out recallable state?: */
4288 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4289 {
4290         if (clp->cl_cb_state == NFSD4_CB_UP)
4291                 return true;
4292         /*
4293          * In the sessions case, since we don't have to establish a
4294          * separate connection for callbacks, we assume it's OK
4295          * until we hear otherwise:
4296          */
4297         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4298 }
4299
4300 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4301 {
4302         struct file_lock *fl;
4303
4304         fl = locks_alloc_lock();
4305         if (!fl)
4306                 return NULL;
4307         fl->fl_lmops = &nfsd_lease_mng_ops;
4308         fl->fl_flags = FL_DELEG;
4309         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4310         fl->fl_end = OFFSET_MAX;
4311         fl->fl_owner = (fl_owner_t)fp;
4312         fl->fl_pid = current->tgid;
4313         return fl;
4314 }
4315
4316 /**
4317  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4318  * @dp:   a pointer to the nfs4_delegation we're adding.
4319  *
4320  * Return:
4321  *      On success: Return code will be 0 on success.
4322  *
4323  *      On error: -EAGAIN if there was an existing delegation.
4324  *                 nonzero if there is an error in other cases.
4325  *
4326  */
4327
4328 static int nfs4_setlease(struct nfs4_delegation *dp)
4329 {
4330         struct nfs4_file *fp = dp->dl_stid.sc_file;
4331         struct file_lock *fl;
4332         struct file *filp;
4333         int status = 0;
4334
4335         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4336         if (!fl)
4337                 return -ENOMEM;
4338         filp = find_readable_file(fp);
4339         if (!filp) {
4340                 /* We should always have a readable file here */
4341                 WARN_ON_ONCE(1);
4342                 locks_free_lock(fl);
4343                 return -EBADF;
4344         }
4345         fl->fl_file = filp;
4346         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4347         if (fl)
4348                 locks_free_lock(fl);
4349         if (status)
4350                 goto out_fput;
4351         spin_lock(&state_lock);
4352         spin_lock(&fp->fi_lock);
4353         /* Did the lease get broken before we took the lock? */
4354         status = -EAGAIN;
4355         if (fp->fi_had_conflict)
4356                 goto out_unlock;
4357         /* Race breaker */
4358         if (fp->fi_deleg_file) {
4359                 status = hash_delegation_locked(dp, fp);
4360                 goto out_unlock;
4361         }
4362         fp->fi_deleg_file = filp;
4363         fp->fi_delegees = 0;
4364         status = hash_delegation_locked(dp, fp);
4365         spin_unlock(&fp->fi_lock);
4366         spin_unlock(&state_lock);
4367         if (status) {
4368                 /* Should never happen, this is a new fi_deleg_file  */
4369                 WARN_ON_ONCE(1);
4370                 goto out_fput;
4371         }
4372         return 0;
4373 out_unlock:
4374         spin_unlock(&fp->fi_lock);
4375         spin_unlock(&state_lock);
4376 out_fput:
4377         fput(filp);
4378         return status;
4379 }
4380
4381 static struct nfs4_delegation *
4382 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4383                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4384 {
4385         int status;
4386         struct nfs4_delegation *dp;
4387
4388         if (fp->fi_had_conflict)
4389                 return ERR_PTR(-EAGAIN);
4390
4391         spin_lock(&state_lock);
4392         spin_lock(&fp->fi_lock);
4393         status = nfs4_get_existing_delegation(clp, fp);
4394         spin_unlock(&fp->fi_lock);
4395         spin_unlock(&state_lock);
4396
4397         if (status)
4398                 return ERR_PTR(status);
4399
4400         dp = alloc_init_deleg(clp, fh, odstate);
4401         if (!dp)
4402                 return ERR_PTR(-ENOMEM);
4403
4404         get_nfs4_file(fp);
4405         spin_lock(&state_lock);
4406         spin_lock(&fp->fi_lock);
4407         dp->dl_stid.sc_file = fp;
4408         if (!fp->fi_deleg_file) {
4409                 spin_unlock(&fp->fi_lock);
4410                 spin_unlock(&state_lock);
4411                 status = nfs4_setlease(dp);
4412                 goto out;
4413         }
4414         if (fp->fi_had_conflict) {
4415                 status = -EAGAIN;
4416                 goto out_unlock;
4417         }
4418         status = hash_delegation_locked(dp, fp);
4419 out_unlock:
4420         spin_unlock(&fp->fi_lock);
4421         spin_unlock(&state_lock);
4422 out:
4423         if (status) {
4424                 put_clnt_odstate(dp->dl_clnt_odstate);
4425                 nfs4_put_stid(&dp->dl_stid);
4426                 return ERR_PTR(status);
4427         }
4428         return dp;
4429 }
4430
4431 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4432 {
4433         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4434         if (status == -EAGAIN)
4435                 open->op_why_no_deleg = WND4_CONTENTION;
4436         else {
4437                 open->op_why_no_deleg = WND4_RESOURCE;
4438                 switch (open->op_deleg_want) {
4439                 case NFS4_SHARE_WANT_READ_DELEG:
4440                 case NFS4_SHARE_WANT_WRITE_DELEG:
4441                 case NFS4_SHARE_WANT_ANY_DELEG:
4442                         break;
4443                 case NFS4_SHARE_WANT_CANCEL:
4444                         open->op_why_no_deleg = WND4_CANCELLED;
4445                         break;
4446                 case NFS4_SHARE_WANT_NO_DELEG:
4447                         WARN_ON_ONCE(1);
4448                 }
4449         }
4450 }
4451
4452 /*
4453  * Attempt to hand out a delegation.
4454  *
4455  * Note we don't support write delegations, and won't until the vfs has
4456  * proper support for them.
4457  */
4458 static void
4459 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4460                         struct nfs4_ol_stateid *stp)
4461 {
4462         struct nfs4_delegation *dp;
4463         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4464         struct nfs4_client *clp = stp->st_stid.sc_client;
4465         int cb_up;
4466         int status = 0;
4467
4468         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4469         open->op_recall = 0;
4470         switch (open->op_claim_type) {
4471                 case NFS4_OPEN_CLAIM_PREVIOUS:
4472                         if (!cb_up)
4473                                 open->op_recall = 1;
4474                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4475                                 goto out_no_deleg;
4476                         break;
4477                 case NFS4_OPEN_CLAIM_NULL:
4478                 case NFS4_OPEN_CLAIM_FH:
4479                         /*
4480                          * Let's not give out any delegations till everyone's
4481                          * had the chance to reclaim theirs, *and* until
4482                          * NLM locks have all been reclaimed:
4483                          */
4484                         if (locks_in_grace(clp->net))
4485                                 goto out_no_deleg;
4486                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4487                                 goto out_no_deleg;
4488                         /*
4489                          * Also, if the file was opened for write or
4490                          * create, there's a good chance the client's
4491                          * about to write to it, resulting in an
4492                          * immediate recall (since we don't support
4493                          * write delegations):
4494                          */
4495                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4496                                 goto out_no_deleg;
4497                         if (open->op_create == NFS4_OPEN_CREATE)
4498                                 goto out_no_deleg;
4499                         break;
4500                 default:
4501                         goto out_no_deleg;
4502         }
4503         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4504         if (IS_ERR(dp))
4505                 goto out_no_deleg;
4506
4507         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4508
4509         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4510                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4511         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4512         nfs4_put_stid(&dp->dl_stid);
4513         return;
4514 out_no_deleg:
4515         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4516         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4517             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4518                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4519                 open->op_recall = 1;
4520         }
4521
4522         /* 4.1 client asking for a delegation? */
4523         if (open->op_deleg_want)
4524                 nfsd4_open_deleg_none_ext(open, status);
4525         return;
4526 }
4527
4528 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4529                                         struct nfs4_delegation *dp)
4530 {
4531         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4532             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4533                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4534                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4535         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4536                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4537                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4538                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4539         }
4540         /* Otherwise the client must be confused wanting a delegation
4541          * it already has, therefore we don't return
4542          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4543          */
4544 }
4545
4546 __be32
4547 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4548 {
4549         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4550         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4551         struct nfs4_file *fp = NULL;
4552         struct nfs4_ol_stateid *stp = NULL;
4553         struct nfs4_delegation *dp = NULL;
4554         __be32 status;
4555         bool new_stp = false;
4556
4557         /*
4558          * Lookup file; if found, lookup stateid and check open request,
4559          * and check for delegations in the process of being recalled.
4560          * If not found, create the nfs4_file struct
4561          */
4562         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4563         if (fp != open->op_file) {
4564                 status = nfs4_check_deleg(cl, open, &dp);
4565                 if (status)
4566                         goto out;
4567                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4568         } else {
4569                 open->op_file = NULL;
4570                 status = nfserr_bad_stateid;
4571                 if (nfsd4_is_deleg_cur(open))
4572                         goto out;
4573         }
4574
4575         if (!stp) {
4576                 stp = init_open_stateid(fp, open);
4577                 if (!open->op_stp)
4578                         new_stp = true;
4579         }
4580
4581         /*
4582          * OPEN the file, or upgrade an existing OPEN.
4583          * If truncate fails, the OPEN fails.
4584          *
4585          * stp is already locked.
4586          */
4587         if (!new_stp) {
4588                 /* Stateid was found, this is an OPEN upgrade */
4589                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4590                 if (status) {
4591                         mutex_unlock(&stp->st_mutex);
4592                         goto out;
4593                 }
4594         } else {
4595                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4596                 if (status) {
4597                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4598                         release_open_stateid(stp);
4599                         mutex_unlock(&stp->st_mutex);
4600                         goto out;
4601                 }
4602
4603                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4604                                                         open->op_odstate);
4605                 if (stp->st_clnt_odstate == open->op_odstate)
4606                         open->op_odstate = NULL;
4607         }
4608
4609         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4610         mutex_unlock(&stp->st_mutex);
4611
4612         if (nfsd4_has_session(&resp->cstate)) {
4613                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4614                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4615                         open->op_why_no_deleg = WND4_NOT_WANTED;
4616                         goto nodeleg;
4617                 }
4618         }
4619
4620         /*
4621         * Attempt to hand out a delegation. No error return, because the
4622         * OPEN succeeds even if we fail.
4623         */
4624         nfs4_open_delegation(current_fh, open, stp);
4625 nodeleg:
4626         status = nfs_ok;
4627
4628         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4629                 STATEID_VAL(&stp->st_stid.sc_stateid));
4630 out:
4631         /* 4.1 client trying to upgrade/downgrade delegation? */
4632         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4633             open->op_deleg_want)
4634                 nfsd4_deleg_xgrade_none_ext(open, dp);
4635
4636         if (fp)
4637                 put_nfs4_file(fp);
4638         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4639                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4640         /*
4641         * To finish the open response, we just need to set the rflags.
4642         */
4643         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4644         if (nfsd4_has_session(&resp->cstate))
4645                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4646         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4647                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4648
4649         if (dp)
4650                 nfs4_put_stid(&dp->dl_stid);
4651         if (stp)
4652                 nfs4_put_stid(&stp->st_stid);
4653
4654         return status;
4655 }
4656
4657 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4658                               struct nfsd4_open *open)
4659 {
4660         if (open->op_openowner) {
4661                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4662
4663                 nfsd4_cstate_assign_replay(cstate, so);
4664                 nfs4_put_stateowner(so);
4665         }
4666         if (open->op_file)
4667                 kmem_cache_free(file_slab, open->op_file);
4668         if (open->op_stp)
4669                 nfs4_put_stid(&open->op_stp->st_stid);
4670         if (open->op_odstate)
4671                 kmem_cache_free(odstate_slab, open->op_odstate);
4672 }
4673
4674 __be32
4675 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4676             union nfsd4_op_u *u)
4677 {
4678         clientid_t *clid = &u->renew;
4679         struct nfs4_client *clp;
4680         __be32 status;
4681         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4682
4683         dprintk("process_renew(%08x/%08x): starting\n",
4684                         clid->cl_boot, clid->cl_id);
4685         status = lookup_clientid(clid, cstate, nn);
4686         if (status)
4687                 goto out;
4688         clp = cstate->clp;
4689         status = nfserr_cb_path_down;
4690         if (!list_empty(&clp->cl_delegations)
4691                         && clp->cl_cb_state != NFSD4_CB_UP)
4692                 goto out;
4693         status = nfs_ok;
4694 out:
4695         return status;
4696 }
4697
4698 void
4699 nfsd4_end_grace(struct nfsd_net *nn)
4700 {
4701         /* do nothing if grace period already ended */
4702         if (nn->grace_ended)
4703                 return;
4704
4705         dprintk("NFSD: end of grace period\n");
4706         nn->grace_ended = true;
4707         /*
4708          * If the server goes down again right now, an NFSv4
4709          * client will still be allowed to reclaim after it comes back up,
4710          * even if it hasn't yet had a chance to reclaim state this time.
4711          *
4712          */
4713         nfsd4_record_grace_done(nn);
4714         /*
4715          * At this point, NFSv4 clients can still reclaim.  But if the
4716          * server crashes, any that have not yet reclaimed will be out
4717          * of luck on the next boot.
4718          *
4719          * (NFSv4.1+ clients are considered to have reclaimed once they
4720          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4721          * have reclaimed after their first OPEN.)
4722          */
4723         locks_end_grace(&nn->nfsd4_manager);
4724         /*
4725          * At this point, and once lockd and/or any other containers
4726          * exit their grace period, further reclaims will fail and
4727          * regular locking can resume.
4728          */
4729 }
4730
4731 static time_t
4732 nfs4_laundromat(struct nfsd_net *nn)
4733 {
4734         struct nfs4_client *clp;
4735         struct nfs4_openowner *oo;
4736         struct nfs4_delegation *dp;
4737         struct nfs4_ol_stateid *stp;
4738         struct nfsd4_blocked_lock *nbl;
4739         struct list_head *pos, *next, reaplist;
4740         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4741         time_t t, new_timeo = nn->nfsd4_lease;
4742
4743         dprintk("NFSD: laundromat service - starting\n");
4744         nfsd4_end_grace(nn);
4745         INIT_LIST_HEAD(&reaplist);
4746         spin_lock(&nn->client_lock);
4747         list_for_each_safe(pos, next, &nn->client_lru) {
4748                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4749                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4750                         t = clp->cl_time - cutoff;
4751                         new_timeo = min(new_timeo, t);
4752                         break;
4753                 }
4754                 if (mark_client_expired_locked(clp)) {
4755                         dprintk("NFSD: client in use (clientid %08x)\n",
4756                                 clp->cl_clientid.cl_id);
4757                         continue;
4758                 }
4759                 list_add(&clp->cl_lru, &reaplist);
4760         }
4761         spin_unlock(&nn->client_lock);
4762         list_for_each_safe(pos, next, &reaplist) {
4763                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4764                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4765                         clp->cl_clientid.cl_id);
4766                 list_del_init(&clp->cl_lru);
4767                 expire_client(clp);
4768         }
4769         spin_lock(&state_lock);
4770         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4771                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4772                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4773                         t = dp->dl_time - cutoff;
4774                         new_timeo = min(new_timeo, t);
4775                         break;
4776                 }
4777                 WARN_ON(!unhash_delegation_locked(dp));
4778                 list_add(&dp->dl_recall_lru, &reaplist);
4779         }
4780         spin_unlock(&state_lock);
4781         while (!list_empty(&reaplist)) {
4782                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4783                                         dl_recall_lru);
4784                 list_del_init(&dp->dl_recall_lru);
4785                 revoke_delegation(dp);
4786         }
4787
4788         spin_lock(&nn->client_lock);
4789         while (!list_empty(&nn->close_lru)) {
4790                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4791                                         oo_close_lru);
4792                 if (time_after((unsigned long)oo->oo_time,
4793                                (unsigned long)cutoff)) {
4794                         t = oo->oo_time - cutoff;
4795                         new_timeo = min(new_timeo, t);
4796                         break;
4797                 }
4798                 list_del_init(&oo->oo_close_lru);
4799                 stp = oo->oo_last_closed_stid;
4800                 oo->oo_last_closed_stid = NULL;
4801                 spin_unlock(&nn->client_lock);
4802                 nfs4_put_stid(&stp->st_stid);
4803                 spin_lock(&nn->client_lock);
4804         }
4805         spin_unlock(&nn->client_lock);
4806
4807         /*
4808          * It's possible for a client to try and acquire an already held lock
4809          * that is being held for a long time, and then lose interest in it.
4810          * So, we clean out any un-revisited request after a lease period
4811          * under the assumption that the client is no longer interested.
4812          *
4813          * RFC5661, sec. 9.6 states that the client must not rely on getting
4814          * notifications and must continue to poll for locks, even when the
4815          * server supports them. Thus this shouldn't lead to clients blocking
4816          * indefinitely once the lock does become free.
4817          */
4818         BUG_ON(!list_empty(&reaplist));
4819         spin_lock(&nn->blocked_locks_lock);
4820         while (!list_empty(&nn->blocked_locks_lru)) {
4821                 nbl = list_first_entry(&nn->blocked_locks_lru,
4822                                         struct nfsd4_blocked_lock, nbl_lru);
4823                 if (time_after((unsigned long)nbl->nbl_time,
4824                                (unsigned long)cutoff)) {
4825                         t = nbl->nbl_time - cutoff;
4826                         new_timeo = min(new_timeo, t);
4827                         break;
4828                 }
4829                 list_move(&nbl->nbl_lru, &reaplist);
4830                 list_del_init(&nbl->nbl_list);
4831         }
4832         spin_unlock(&nn->blocked_locks_lock);
4833
4834         while (!list_empty(&reaplist)) {
4835                 nbl = list_first_entry(&reaplist,
4836                                         struct nfsd4_blocked_lock, nbl_lru);
4837                 list_del_init(&nbl->nbl_lru);
4838                 posix_unblock_lock(&nbl->nbl_lock);
4839                 free_blocked_lock(nbl);
4840         }
4841
4842         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4843         return new_timeo;
4844 }
4845
4846 static struct workqueue_struct *laundry_wq;
4847 static void laundromat_main(struct work_struct *);
4848
4849 static void
4850 laundromat_main(struct work_struct *laundry)
4851 {
4852         time_t t;
4853         struct delayed_work *dwork = to_delayed_work(laundry);
4854         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4855                                            laundromat_work);
4856
4857         t = nfs4_laundromat(nn);
4858         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4859         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4860 }
4861
4862 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4863 {
4864         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4865                 return nfserr_bad_stateid;
4866         return nfs_ok;
4867 }
4868
4869 static inline int
4870 access_permit_read(struct nfs4_ol_stateid *stp)
4871 {
4872         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4873                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4874                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4875 }
4876
4877 static inline int
4878 access_permit_write(struct nfs4_ol_stateid *stp)
4879 {
4880         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4881                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4882 }
4883
4884 static
4885 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4886 {
4887         __be32 status = nfserr_openmode;
4888
4889         /* For lock stateid's, we test the parent open, not the lock: */
4890         if (stp->st_openstp)
4891                 stp = stp->st_openstp;
4892         if ((flags & WR_STATE) && !access_permit_write(stp))
4893                 goto out;
4894         if ((flags & RD_STATE) && !access_permit_read(stp))
4895                 goto out;
4896         status = nfs_ok;
4897 out:
4898         return status;
4899 }
4900
4901 static inline __be32
4902 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4903 {
4904         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4905                 return nfs_ok;
4906         else if (opens_in_grace(net)) {
4907                 /* Answer in remaining cases depends on existence of
4908                  * conflicting state; so we must wait out the grace period. */
4909                 return nfserr_grace;
4910         } else if (flags & WR_STATE)
4911                 return nfs4_share_conflict(current_fh,
4912                                 NFS4_SHARE_DENY_WRITE);
4913         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4914                 return nfs4_share_conflict(current_fh,
4915                                 NFS4_SHARE_DENY_READ);
4916 }
4917
4918 /*
4919  * Allow READ/WRITE during grace period on recovered state only for files
4920  * that are not able to provide mandatory locking.
4921  */
4922 static inline int
4923 grace_disallows_io(struct net *net, struct inode *inode)
4924 {
4925         return opens_in_grace(net) && mandatory_lock(inode);
4926 }
4927
4928 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4929 {
4930         /*
4931          * When sessions are used the stateid generation number is ignored
4932          * when it is zero.
4933          */
4934         if (has_session && in->si_generation == 0)
4935                 return nfs_ok;
4936
4937         if (in->si_generation == ref->si_generation)
4938                 return nfs_ok;
4939
4940         /* If the client sends us a stateid from the future, it's buggy: */
4941         if (nfsd4_stateid_generation_after(in, ref))
4942                 return nfserr_bad_stateid;
4943         /*
4944          * However, we could see a stateid from the past, even from a
4945          * non-buggy client.  For example, if the client sends a lock
4946          * while some IO is outstanding, the lock may bump si_generation
4947          * while the IO is still in flight.  The client could avoid that
4948          * situation by waiting for responses on all the IO requests,
4949          * but better performance may result in retrying IO that
4950          * receives an old_stateid error if requests are rarely
4951          * reordered in flight:
4952          */
4953         return nfserr_old_stateid;
4954 }
4955
4956 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4957 {
4958         __be32 ret;
4959
4960         spin_lock(&s->sc_lock);
4961         ret = nfsd4_verify_open_stid(s);
4962         if (ret == nfs_ok)
4963                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4964         spin_unlock(&s->sc_lock);
4965         return ret;
4966 }
4967
4968 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4969 {
4970         if (ols->st_stateowner->so_is_open_owner &&
4971             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4972                 return nfserr_bad_stateid;
4973         return nfs_ok;
4974 }
4975
4976 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4977 {
4978         struct nfs4_stid *s;
4979         __be32 status = nfserr_bad_stateid;
4980
4981         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4982                 CLOSE_STATEID(stateid))
4983                 return status;
4984         /* Client debugging aid. */
4985         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4986                 char addr_str[INET6_ADDRSTRLEN];
4987                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4988                                  sizeof(addr_str));
4989                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4990                                         "with incorrect client ID\n", addr_str);
4991                 return status;
4992         }
4993         spin_lock(&cl->cl_lock);
4994         s = find_stateid_locked(cl, stateid);
4995         if (!s)
4996                 goto out_unlock;
4997         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
4998         if (status)
4999                 goto out_unlock;
5000         switch (s->sc_type) {
5001         case NFS4_DELEG_STID:
5002                 status = nfs_ok;
5003                 break;
5004         case NFS4_REVOKED_DELEG_STID:
5005                 status = nfserr_deleg_revoked;
5006                 break;
5007         case NFS4_OPEN_STID:
5008         case NFS4_LOCK_STID:
5009                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5010                 break;
5011         default:
5012                 printk("unknown stateid type %x\n", s->sc_type);
5013                 /* Fallthrough */
5014         case NFS4_CLOSED_STID:
5015         case NFS4_CLOSED_DELEG_STID:
5016                 status = nfserr_bad_stateid;
5017         }
5018 out_unlock:
5019         spin_unlock(&cl->cl_lock);
5020         return status;
5021 }
5022
5023 __be32
5024 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5025                      stateid_t *stateid, unsigned char typemask,
5026                      struct nfs4_stid **s, struct nfsd_net *nn)
5027 {
5028         __be32 status;
5029         bool return_revoked = false;
5030
5031         /*
5032          *  only return revoked delegations if explicitly asked.
5033          *  otherwise we report revoked or bad_stateid status.
5034          */
5035         if (typemask & NFS4_REVOKED_DELEG_STID)
5036                 return_revoked = true;
5037         else if (typemask & NFS4_DELEG_STID)
5038                 typemask |= NFS4_REVOKED_DELEG_STID;
5039
5040         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5041                 CLOSE_STATEID(stateid))
5042                 return nfserr_bad_stateid;
5043         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5044         if (status == nfserr_stale_clientid) {
5045                 if (cstate->session)
5046                         return nfserr_bad_stateid;
5047                 return nfserr_stale_stateid;
5048         }
5049         if (status)
5050                 return status;
5051         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5052         if (!*s)
5053                 return nfserr_bad_stateid;
5054         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5055                 nfs4_put_stid(*s);
5056                 if (cstate->minorversion)
5057                         return nfserr_deleg_revoked;
5058                 return nfserr_bad_stateid;
5059         }
5060         return nfs_ok;
5061 }
5062
5063 static struct file *
5064 nfs4_find_file(struct nfs4_stid *s, int flags)
5065 {
5066         if (!s)
5067                 return NULL;
5068
5069         switch (s->sc_type) {
5070         case NFS4_DELEG_STID:
5071                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5072                         return NULL;
5073                 return get_file(s->sc_file->fi_deleg_file);
5074         case NFS4_OPEN_STID:
5075         case NFS4_LOCK_STID:
5076                 if (flags & RD_STATE)
5077                         return find_readable_file(s->sc_file);
5078                 else
5079                         return find_writeable_file(s->sc_file);
5080                 break;
5081         }
5082
5083         return NULL;
5084 }
5085
5086 static __be32
5087 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5088 {
5089         __be32 status;
5090
5091         status = nfsd4_check_openowner_confirmed(ols);
5092         if (status)
5093                 return status;
5094         return nfs4_check_openmode(ols, flags);
5095 }
5096
5097 static __be32
5098 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5099                 struct file **filpp, bool *tmp_file, int flags)
5100 {
5101         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5102         struct file *file;
5103         __be32 status;
5104
5105         file = nfs4_find_file(s, flags);
5106         if (file) {
5107                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5108                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5109                 if (status) {
5110                         fput(file);
5111                         return status;
5112                 }
5113
5114                 *filpp = file;
5115         } else {
5116                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5117                 if (status)
5118                         return status;
5119
5120                 if (tmp_file)
5121                         *tmp_file = true;
5122         }
5123
5124         return 0;
5125 }
5126
5127 /*
5128  * Checks for stateid operations
5129  */
5130 __be32
5131 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5132                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5133                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5134 {
5135         struct inode *ino = d_inode(fhp->fh_dentry);
5136         struct net *net = SVC_NET(rqstp);
5137         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5138         struct nfs4_stid *s = NULL;
5139         __be32 status;
5140
5141         if (filpp)
5142                 *filpp = NULL;
5143         if (tmp_file)
5144                 *tmp_file = false;
5145
5146         if (grace_disallows_io(net, ino))
5147                 return nfserr_grace;
5148
5149         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5150                 status = check_special_stateids(net, fhp, stateid, flags);
5151                 goto done;
5152         }
5153
5154         status = nfsd4_lookup_stateid(cstate, stateid,
5155                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5156                                 &s, nn);
5157         if (status)
5158                 return status;
5159         status = nfsd4_stid_check_stateid_generation(stateid, s,
5160                         nfsd4_has_session(cstate));
5161         if (status)
5162                 goto out;
5163
5164         switch (s->sc_type) {
5165         case NFS4_DELEG_STID:
5166                 status = nfs4_check_delegmode(delegstateid(s), flags);
5167                 break;
5168         case NFS4_OPEN_STID:
5169         case NFS4_LOCK_STID:
5170                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5171                 break;
5172         default:
5173                 status = nfserr_bad_stateid;
5174                 break;
5175         }
5176         if (status)
5177                 goto out;
5178         status = nfs4_check_fh(fhp, s);
5179
5180 done:
5181         if (!status && filpp)
5182                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5183 out:
5184         if (s)
5185                 nfs4_put_stid(s);
5186         return status;
5187 }
5188
5189 /*
5190  * Test if the stateid is valid
5191  */
5192 __be32
5193 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5194                    union nfsd4_op_u *u)
5195 {
5196         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5197         struct nfsd4_test_stateid_id *stateid;
5198         struct nfs4_client *cl = cstate->session->se_client;
5199
5200         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5201                 stateid->ts_id_status =
5202                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5203
5204         return nfs_ok;
5205 }
5206
5207 static __be32
5208 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5209 {
5210         struct nfs4_ol_stateid *stp = openlockstateid(s);
5211         __be32 ret;
5212
5213         ret = nfsd4_lock_ol_stateid(stp);
5214         if (ret)
5215                 goto out_put_stid;
5216
5217         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5218         if (ret)
5219                 goto out;
5220
5221         ret = nfserr_locks_held;
5222         if (check_for_locks(stp->st_stid.sc_file,
5223                             lockowner(stp->st_stateowner)))
5224                 goto out;
5225
5226         release_lock_stateid(stp);
5227         ret = nfs_ok;
5228
5229 out:
5230         mutex_unlock(&stp->st_mutex);
5231 out_put_stid:
5232         nfs4_put_stid(s);
5233         return ret;
5234 }
5235
5236 __be32
5237 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5238                    union nfsd4_op_u *u)
5239 {
5240         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5241         stateid_t *stateid = &free_stateid->fr_stateid;
5242         struct nfs4_stid *s;
5243         struct nfs4_delegation *dp;
5244         struct nfs4_client *cl = cstate->session->se_client;
5245         __be32 ret = nfserr_bad_stateid;
5246
5247         spin_lock(&cl->cl_lock);
5248         s = find_stateid_locked(cl, stateid);
5249         if (!s)
5250                 goto out_unlock;
5251         spin_lock(&s->sc_lock);
5252         switch (s->sc_type) {
5253         case NFS4_DELEG_STID:
5254                 ret = nfserr_locks_held;
5255                 break;
5256         case NFS4_OPEN_STID:
5257                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5258                 if (ret)
5259                         break;
5260                 ret = nfserr_locks_held;
5261                 break;
5262         case NFS4_LOCK_STID:
5263                 spin_unlock(&s->sc_lock);
5264                 refcount_inc(&s->sc_count);
5265                 spin_unlock(&cl->cl_lock);
5266                 ret = nfsd4_free_lock_stateid(stateid, s);
5267                 goto out;
5268         case NFS4_REVOKED_DELEG_STID:
5269                 spin_unlock(&s->sc_lock);
5270                 dp = delegstateid(s);
5271                 list_del_init(&dp->dl_recall_lru);
5272                 spin_unlock(&cl->cl_lock);
5273                 nfs4_put_stid(s);
5274                 ret = nfs_ok;
5275                 goto out;
5276         /* Default falls through and returns nfserr_bad_stateid */
5277         }
5278         spin_unlock(&s->sc_lock);
5279 out_unlock:
5280         spin_unlock(&cl->cl_lock);
5281 out:
5282         return ret;
5283 }
5284
5285 static inline int
5286 setlkflg (int type)
5287 {
5288         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5289                 RD_STATE : WR_STATE;
5290 }
5291
5292 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5293 {
5294         struct svc_fh *current_fh = &cstate->current_fh;
5295         struct nfs4_stateowner *sop = stp->st_stateowner;
5296         __be32 status;
5297
5298         status = nfsd4_check_seqid(cstate, sop, seqid);
5299         if (status)
5300                 return status;
5301         status = nfsd4_lock_ol_stateid(stp);
5302         if (status != nfs_ok)
5303                 return status;
5304         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5305         if (status == nfs_ok)
5306                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5307         if (status != nfs_ok)
5308                 mutex_unlock(&stp->st_mutex);
5309         return status;
5310 }
5311
5312 /*
5313  * Checks for sequence id mutating operations.
5314  */
5315 static __be32
5316 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5317                          stateid_t *stateid, char typemask,
5318                          struct nfs4_ol_stateid **stpp,
5319                          struct nfsd_net *nn)
5320 {
5321         __be32 status;
5322         struct nfs4_stid *s;
5323         struct nfs4_ol_stateid *stp = NULL;
5324
5325         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5326                 seqid, STATEID_VAL(stateid));
5327
5328         *stpp = NULL;
5329         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5330         if (status)
5331                 return status;
5332         stp = openlockstateid(s);
5333         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5334
5335         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5336         if (!status)
5337                 *stpp = stp;
5338         else
5339                 nfs4_put_stid(&stp->st_stid);
5340         return status;
5341 }
5342
5343 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5344                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5345 {
5346         __be32 status;
5347         struct nfs4_openowner *oo;
5348         struct nfs4_ol_stateid *stp;
5349
5350         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5351                                                 NFS4_OPEN_STID, &stp, nn);
5352         if (status)
5353                 return status;
5354         oo = openowner(stp->st_stateowner);
5355         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5356                 mutex_unlock(&stp->st_mutex);
5357                 nfs4_put_stid(&stp->st_stid);
5358                 return nfserr_bad_stateid;
5359         }
5360         *stpp = stp;
5361         return nfs_ok;
5362 }
5363
5364 __be32
5365 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5366                    union nfsd4_op_u *u)
5367 {
5368         struct nfsd4_open_confirm *oc = &u->open_confirm;
5369         __be32 status;
5370         struct nfs4_openowner *oo;
5371         struct nfs4_ol_stateid *stp;
5372         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5373
5374         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5375                         cstate->current_fh.fh_dentry);
5376
5377         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5378         if (status)
5379                 return status;
5380
5381         status = nfs4_preprocess_seqid_op(cstate,
5382                                         oc->oc_seqid, &oc->oc_req_stateid,
5383                                         NFS4_OPEN_STID, &stp, nn);
5384         if (status)
5385                 goto out;
5386         oo = openowner(stp->st_stateowner);
5387         status = nfserr_bad_stateid;
5388         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5389                 mutex_unlock(&stp->st_mutex);
5390                 goto put_stateid;
5391         }
5392         oo->oo_flags |= NFS4_OO_CONFIRMED;
5393         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5394         mutex_unlock(&stp->st_mutex);
5395         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5396                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5397
5398         nfsd4_client_record_create(oo->oo_owner.so_client);
5399         status = nfs_ok;
5400 put_stateid:
5401         nfs4_put_stid(&stp->st_stid);
5402 out:
5403         nfsd4_bump_seqid(cstate, status);
5404         return status;
5405 }
5406
5407 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5408 {
5409         if (!test_access(access, stp))
5410                 return;
5411         nfs4_file_put_access(stp->st_stid.sc_file, access);
5412         clear_access(access, stp);
5413 }
5414
5415 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5416 {
5417         switch (to_access) {
5418         case NFS4_SHARE_ACCESS_READ:
5419                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5420                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5421                 break;
5422         case NFS4_SHARE_ACCESS_WRITE:
5423                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5424                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5425                 break;
5426         case NFS4_SHARE_ACCESS_BOTH:
5427                 break;
5428         default:
5429                 WARN_ON_ONCE(1);
5430         }
5431 }
5432
5433 __be32
5434 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5435                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5436 {
5437         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5438         __be32 status;
5439         struct nfs4_ol_stateid *stp;
5440         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5441
5442         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5443                         cstate->current_fh.fh_dentry);
5444
5445         /* We don't yet support WANT bits: */
5446         if (od->od_deleg_want)
5447                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5448                         od->od_deleg_want);
5449
5450         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5451                                         &od->od_stateid, &stp, nn);
5452         if (status)
5453                 goto out;
5454         status = nfserr_inval;
5455         if (!test_access(od->od_share_access, stp)) {
5456                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5457                         stp->st_access_bmap, od->od_share_access);
5458                 goto put_stateid;
5459         }
5460         if (!test_deny(od->od_share_deny, stp)) {
5461                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5462                         stp->st_deny_bmap, od->od_share_deny);
5463                 goto put_stateid;
5464         }
5465         nfs4_stateid_downgrade(stp, od->od_share_access);
5466         reset_union_bmap_deny(od->od_share_deny, stp);
5467         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5468         status = nfs_ok;
5469 put_stateid:
5470         mutex_unlock(&stp->st_mutex);
5471         nfs4_put_stid(&stp->st_stid);
5472 out:
5473         nfsd4_bump_seqid(cstate, status);
5474         return status;
5475 }
5476
5477 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5478 {
5479         struct nfs4_client *clp = s->st_stid.sc_client;
5480         bool unhashed;
5481         LIST_HEAD(reaplist);
5482
5483         spin_lock(&clp->cl_lock);
5484         unhashed = unhash_open_stateid(s, &reaplist);
5485
5486         if (clp->cl_minorversion) {
5487                 if (unhashed)
5488                         put_ol_stateid_locked(s, &reaplist);
5489                 spin_unlock(&clp->cl_lock);
5490                 free_ol_stateid_reaplist(&reaplist);
5491         } else {
5492                 spin_unlock(&clp->cl_lock);
5493                 free_ol_stateid_reaplist(&reaplist);
5494                 if (unhashed)
5495                         move_to_close_lru(s, clp->net);
5496         }
5497 }
5498
5499 /*
5500  * nfs4_unlock_state() called after encode
5501  */
5502 __be32
5503 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5504                 union nfsd4_op_u *u)
5505 {
5506         struct nfsd4_close *close = &u->close;
5507         __be32 status;
5508         struct nfs4_ol_stateid *stp;
5509         struct net *net = SVC_NET(rqstp);
5510         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5511
5512         dprintk("NFSD: nfsd4_close on file %pd\n",
5513                         cstate->current_fh.fh_dentry);
5514
5515         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5516                                         &close->cl_stateid,
5517                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5518                                         &stp, nn);
5519         nfsd4_bump_seqid(cstate, status);
5520         if (status)
5521                 goto out;
5522
5523         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5524         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5525
5526         nfsd4_close_open_stateid(stp);
5527         mutex_unlock(&stp->st_mutex);
5528
5529         /* See RFC5661 sectionm 18.2.4 */
5530         if (stp->st_stid.sc_client->cl_minorversion)
5531                 memcpy(&close->cl_stateid, &close_stateid,
5532                                 sizeof(close->cl_stateid));
5533
5534         /* put reference from nfs4_preprocess_seqid_op */
5535         nfs4_put_stid(&stp->st_stid);
5536 out:
5537         return status;
5538 }
5539
5540 __be32
5541 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5542                   union nfsd4_op_u *u)
5543 {
5544         struct nfsd4_delegreturn *dr = &u->delegreturn;
5545         struct nfs4_delegation *dp;
5546         stateid_t *stateid = &dr->dr_stateid;
5547         struct nfs4_stid *s;
5548         __be32 status;
5549         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5550
5551         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5552                 return status;
5553
5554         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5555         if (status)
5556                 goto out;
5557         dp = delegstateid(s);
5558         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5559         if (status)
5560                 goto put_stateid;
5561
5562         destroy_delegation(dp);
5563 put_stateid:
5564         nfs4_put_stid(&dp->dl_stid);
5565 out:
5566         return status;
5567 }
5568
5569 static inline u64
5570 end_offset(u64 start, u64 len)
5571 {
5572         u64 end;
5573
5574         end = start + len;
5575         return end >= start ? end: NFS4_MAX_UINT64;
5576 }
5577
5578 /* last octet in a range */
5579 static inline u64
5580 last_byte_offset(u64 start, u64 len)
5581 {
5582         u64 end;
5583
5584         WARN_ON_ONCE(!len);
5585         end = start + len;
5586         return end > start ? end - 1: NFS4_MAX_UINT64;
5587 }
5588
5589 /*
5590  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5591  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5592  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5593  * locking, this prevents us from being completely protocol-compliant.  The
5594  * real solution to this problem is to start using unsigned file offsets in
5595  * the VFS, but this is a very deep change!
5596  */
5597 static inline void
5598 nfs4_transform_lock_offset(struct file_lock *lock)
5599 {
5600         if (lock->fl_start < 0)
5601                 lock->fl_start = OFFSET_MAX;
5602         if (lock->fl_end < 0)
5603                 lock->fl_end = OFFSET_MAX;
5604 }
5605
5606 static fl_owner_t
5607 nfsd4_fl_get_owner(fl_owner_t owner)
5608 {
5609         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5610
5611         nfs4_get_stateowner(&lo->lo_owner);
5612         return owner;
5613 }
5614
5615 static void
5616 nfsd4_fl_put_owner(fl_owner_t owner)
5617 {
5618         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5619
5620         if (lo)
5621                 nfs4_put_stateowner(&lo->lo_owner);
5622 }
5623
5624 static void
5625 nfsd4_lm_notify(struct file_lock *fl)
5626 {
5627         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5628         struct net                      *net = lo->lo_owner.so_client->net;
5629         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5630         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5631                                                 struct nfsd4_blocked_lock, nbl_lock);
5632         bool queue = false;
5633
5634         /* An empty list means that something else is going to be using it */
5635         spin_lock(&nn->blocked_locks_lock);
5636         if (!list_empty(&nbl->nbl_list)) {
5637                 list_del_init(&nbl->nbl_list);
5638                 list_del_init(&nbl->nbl_lru);
5639                 queue = true;
5640         }
5641         spin_unlock(&nn->blocked_locks_lock);
5642
5643         if (queue)
5644                 nfsd4_run_cb(&nbl->nbl_cb);
5645 }
5646
5647 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5648         .lm_notify = nfsd4_lm_notify,
5649         .lm_get_owner = nfsd4_fl_get_owner,
5650         .lm_put_owner = nfsd4_fl_put_owner,
5651 };
5652
5653 static inline void
5654 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5655 {
5656         struct nfs4_lockowner *lo;
5657
5658         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5659                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5660                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5661                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5662                 if (!deny->ld_owner.data)
5663                         /* We just don't care that much */
5664                         goto nevermind;
5665                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5666                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5667         } else {
5668 nevermind:
5669                 deny->ld_owner.len = 0;
5670                 deny->ld_owner.data = NULL;
5671                 deny->ld_clientid.cl_boot = 0;
5672                 deny->ld_clientid.cl_id = 0;
5673         }
5674         deny->ld_start = fl->fl_start;
5675         deny->ld_length = NFS4_MAX_UINT64;
5676         if (fl->fl_end != NFS4_MAX_UINT64)
5677                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5678         deny->ld_type = NFS4_READ_LT;
5679         if (fl->fl_type != F_RDLCK)
5680                 deny->ld_type = NFS4_WRITE_LT;
5681 }
5682
5683 static struct nfs4_lockowner *
5684 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5685 {
5686         unsigned int strhashval = ownerstr_hashval(owner);
5687         struct nfs4_stateowner *so;
5688
5689         lockdep_assert_held(&clp->cl_lock);
5690
5691         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5692                             so_strhash) {
5693                 if (so->so_is_open_owner)
5694                         continue;
5695                 if (same_owner_str(so, owner))
5696                         return lockowner(nfs4_get_stateowner(so));
5697         }
5698         return NULL;
5699 }
5700
5701 static struct nfs4_lockowner *
5702 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5703 {
5704         struct nfs4_lockowner *lo;
5705
5706         spin_lock(&clp->cl_lock);
5707         lo = find_lockowner_str_locked(clp, owner);
5708         spin_unlock(&clp->cl_lock);
5709         return lo;
5710 }
5711
5712 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5713 {
5714         unhash_lockowner_locked(lockowner(sop));
5715 }
5716
5717 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5718 {
5719         struct nfs4_lockowner *lo = lockowner(sop);
5720
5721         kmem_cache_free(lockowner_slab, lo);
5722 }
5723
5724 static const struct nfs4_stateowner_operations lockowner_ops = {
5725         .so_unhash =    nfs4_unhash_lockowner,
5726         .so_free =      nfs4_free_lockowner,
5727 };
5728
5729 /*
5730  * Alloc a lock owner structure.
5731  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5732  * occurred.
5733  *
5734  * strhashval = ownerstr_hashval
5735  */
5736 static struct nfs4_lockowner *
5737 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5738                            struct nfs4_ol_stateid *open_stp,
5739                            struct nfsd4_lock *lock)
5740 {
5741         struct nfs4_lockowner *lo, *ret;
5742
5743         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5744         if (!lo)
5745                 return NULL;
5746         INIT_LIST_HEAD(&lo->lo_blocked);
5747         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5748         lo->lo_owner.so_is_open_owner = 0;
5749         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5750         lo->lo_owner.so_ops = &lockowner_ops;
5751         spin_lock(&clp->cl_lock);
5752         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5753         if (ret == NULL) {
5754                 list_add(&lo->lo_owner.so_strhash,
5755                          &clp->cl_ownerstr_hashtbl[strhashval]);
5756                 ret = lo;
5757         } else
5758                 nfs4_free_stateowner(&lo->lo_owner);
5759
5760         spin_unlock(&clp->cl_lock);
5761         return ret;
5762 }
5763
5764 static struct nfs4_ol_stateid *
5765 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5766 {
5767         struct nfs4_ol_stateid *lst;
5768         struct nfs4_client *clp = lo->lo_owner.so_client;
5769
5770         lockdep_assert_held(&clp->cl_lock);
5771
5772         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5773                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5774                         continue;
5775                 if (lst->st_stid.sc_file == fp) {
5776                         refcount_inc(&lst->st_stid.sc_count);
5777                         return lst;
5778                 }
5779         }
5780         return NULL;
5781 }
5782
5783 static struct nfs4_ol_stateid *
5784 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5785                   struct nfs4_file *fp, struct inode *inode,
5786                   struct nfs4_ol_stateid *open_stp)
5787 {
5788         struct nfs4_client *clp = lo->lo_owner.so_client;
5789         struct nfs4_ol_stateid *retstp;
5790
5791         mutex_init(&stp->st_mutex);
5792         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5793 retry:
5794         spin_lock(&clp->cl_lock);
5795         spin_lock(&fp->fi_lock);
5796         retstp = find_lock_stateid(lo, fp);
5797         if (retstp)
5798                 goto out_unlock;
5799
5800         refcount_inc(&stp->st_stid.sc_count);
5801         stp->st_stid.sc_type = NFS4_LOCK_STID;
5802         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5803         get_nfs4_file(fp);
5804         stp->st_stid.sc_file = fp;
5805         stp->st_access_bmap = 0;
5806         stp->st_deny_bmap = open_stp->st_deny_bmap;
5807         stp->st_openstp = open_stp;
5808         list_add(&stp->st_locks, &open_stp->st_locks);
5809         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5810         list_add(&stp->st_perfile, &fp->fi_stateids);
5811 out_unlock:
5812         spin_unlock(&fp->fi_lock);
5813         spin_unlock(&clp->cl_lock);
5814         if (retstp) {
5815                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5816                         nfs4_put_stid(&retstp->st_stid);
5817                         goto retry;
5818                 }
5819                 /* To keep mutex tracking happy */
5820                 mutex_unlock(&stp->st_mutex);
5821                 stp = retstp;
5822         }
5823         return stp;
5824 }
5825
5826 static struct nfs4_ol_stateid *
5827 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5828                             struct inode *inode, struct nfs4_ol_stateid *ost,
5829                             bool *new)
5830 {
5831         struct nfs4_stid *ns = NULL;
5832         struct nfs4_ol_stateid *lst;
5833         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5834         struct nfs4_client *clp = oo->oo_owner.so_client;
5835
5836         *new = false;
5837         spin_lock(&clp->cl_lock);
5838         lst = find_lock_stateid(lo, fi);
5839         spin_unlock(&clp->cl_lock);
5840         if (lst != NULL) {
5841                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5842                         goto out;
5843                 nfs4_put_stid(&lst->st_stid);
5844         }
5845         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5846         if (ns == NULL)
5847                 return NULL;
5848
5849         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5850         if (lst == openlockstateid(ns))
5851                 *new = true;
5852         else
5853                 nfs4_put_stid(ns);
5854 out:
5855         return lst;
5856 }
5857
5858 static int
5859 check_lock_length(u64 offset, u64 length)
5860 {
5861         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5862                 (length > ~offset)));
5863 }
5864
5865 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5866 {
5867         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5868
5869         lockdep_assert_held(&fp->fi_lock);
5870
5871         if (test_access(access, lock_stp))
5872                 return;
5873         __nfs4_file_get_access(fp, access);
5874         set_access(access, lock_stp);
5875 }
5876
5877 static __be32
5878 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5879                             struct nfs4_ol_stateid *ost,
5880                             struct nfsd4_lock *lock,
5881                             struct nfs4_ol_stateid **plst, bool *new)
5882 {
5883         __be32 status;
5884         struct nfs4_file *fi = ost->st_stid.sc_file;
5885         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5886         struct nfs4_client *cl = oo->oo_owner.so_client;
5887         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5888         struct nfs4_lockowner *lo;
5889         struct nfs4_ol_stateid *lst;
5890         unsigned int strhashval;
5891
5892         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5893         if (!lo) {
5894                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5895                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5896                 if (lo == NULL)
5897                         return nfserr_jukebox;
5898         } else {
5899                 /* with an existing lockowner, seqids must be the same */
5900                 status = nfserr_bad_seqid;
5901                 if (!cstate->minorversion &&
5902                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5903                         goto out;
5904         }
5905
5906         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5907         if (lst == NULL) {
5908                 status = nfserr_jukebox;
5909                 goto out;
5910         }
5911
5912         status = nfs_ok;
5913         *plst = lst;
5914 out:
5915         nfs4_put_stateowner(&lo->lo_owner);
5916         return status;
5917 }
5918
5919 /*
5920  *  LOCK operation
5921  */
5922 __be32
5923 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5924            union nfsd4_op_u *u)
5925 {
5926         struct nfsd4_lock *lock = &u->lock;
5927         struct nfs4_openowner *open_sop = NULL;
5928         struct nfs4_lockowner *lock_sop = NULL;
5929         struct nfs4_ol_stateid *lock_stp = NULL;
5930         struct nfs4_ol_stateid *open_stp = NULL;
5931         struct nfs4_file *fp;
5932         struct file *filp = NULL;
5933         struct nfsd4_blocked_lock *nbl = NULL;
5934         struct file_lock *file_lock = NULL;
5935         struct file_lock *conflock = NULL;
5936         __be32 status = 0;
5937         int lkflg;
5938         int err;
5939         bool new = false;
5940         unsigned char fl_type;
5941         unsigned int fl_flags = FL_POSIX;
5942         struct net *net = SVC_NET(rqstp);
5943         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5944
5945         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5946                 (long long) lock->lk_offset,
5947                 (long long) lock->lk_length);
5948
5949         if (check_lock_length(lock->lk_offset, lock->lk_length))
5950                  return nfserr_inval;
5951
5952         if ((status = fh_verify(rqstp, &cstate->current_fh,
5953                                 S_IFREG, NFSD_MAY_LOCK))) {
5954                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5955                 return status;
5956         }
5957
5958         if (lock->lk_is_new) {
5959                 if (nfsd4_has_session(cstate))
5960                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5961                         memcpy(&lock->lk_new_clientid,
5962                                 &cstate->session->se_client->cl_clientid,
5963                                 sizeof(clientid_t));
5964
5965                 status = nfserr_stale_clientid;
5966                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5967                         goto out;
5968
5969                 /* validate and update open stateid and open seqid */
5970                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5971                                         lock->lk_new_open_seqid,
5972                                         &lock->lk_new_open_stateid,
5973                                         &open_stp, nn);
5974                 if (status)
5975                         goto out;
5976                 mutex_unlock(&open_stp->st_mutex);
5977                 open_sop = openowner(open_stp->st_stateowner);
5978                 status = nfserr_bad_stateid;
5979                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5980                                                 &lock->lk_new_clientid))
5981                         goto out;
5982                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5983                                                         &lock_stp, &new);
5984         } else {
5985                 status = nfs4_preprocess_seqid_op(cstate,
5986                                        lock->lk_old_lock_seqid,
5987                                        &lock->lk_old_lock_stateid,
5988                                        NFS4_LOCK_STID, &lock_stp, nn);
5989         }
5990         if (status)
5991                 goto out;
5992         lock_sop = lockowner(lock_stp->st_stateowner);
5993
5994         lkflg = setlkflg(lock->lk_type);
5995         status = nfs4_check_openmode(lock_stp, lkflg);
5996         if (status)
5997                 goto out;
5998
5999         status = nfserr_grace;
6000         if (locks_in_grace(net) && !lock->lk_reclaim)
6001                 goto out;
6002         status = nfserr_no_grace;
6003         if (!locks_in_grace(net) && lock->lk_reclaim)
6004                 goto out;
6005
6006         fp = lock_stp->st_stid.sc_file;
6007         switch (lock->lk_type) {
6008                 case NFS4_READW_LT:
6009                         if (nfsd4_has_session(cstate))
6010                                 fl_flags |= FL_SLEEP;
6011                         /* Fallthrough */
6012                 case NFS4_READ_LT:
6013                         spin_lock(&fp->fi_lock);
6014                         filp = find_readable_file_locked(fp);
6015                         if (filp)
6016                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6017                         spin_unlock(&fp->fi_lock);
6018                         fl_type = F_RDLCK;
6019                         break;
6020                 case NFS4_WRITEW_LT:
6021                         if (nfsd4_has_session(cstate))
6022                                 fl_flags |= FL_SLEEP;
6023                         /* Fallthrough */
6024                 case NFS4_WRITE_LT:
6025                         spin_lock(&fp->fi_lock);
6026                         filp = find_writeable_file_locked(fp);
6027                         if (filp)
6028                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6029                         spin_unlock(&fp->fi_lock);
6030                         fl_type = F_WRLCK;
6031                         break;
6032                 default:
6033                         status = nfserr_inval;
6034                 goto out;
6035         }
6036
6037         if (!filp) {
6038                 status = nfserr_openmode;
6039                 goto out;
6040         }
6041
6042         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6043         if (!nbl) {
6044                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6045                 status = nfserr_jukebox;
6046                 goto out;
6047         }
6048
6049         file_lock = &nbl->nbl_lock;
6050         file_lock->fl_type = fl_type;
6051         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6052         file_lock->fl_pid = current->tgid;
6053         file_lock->fl_file = filp;
6054         file_lock->fl_flags = fl_flags;
6055         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6056         file_lock->fl_start = lock->lk_offset;
6057         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6058         nfs4_transform_lock_offset(file_lock);
6059
6060         conflock = locks_alloc_lock();
6061         if (!conflock) {
6062                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6063                 status = nfserr_jukebox;
6064                 goto out;
6065         }
6066
6067         if (fl_flags & FL_SLEEP) {
6068                 nbl->nbl_time = jiffies;
6069                 spin_lock(&nn->blocked_locks_lock);
6070                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6071                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6072                 spin_unlock(&nn->blocked_locks_lock);
6073         }
6074
6075         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6076         switch (err) {
6077         case 0: /* success! */
6078                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6079                 status = 0;
6080                 break;
6081         case FILE_LOCK_DEFERRED:
6082                 nbl = NULL;
6083                 /* Fallthrough */
6084         case -EAGAIN:           /* conflock holds conflicting lock */
6085                 status = nfserr_denied;
6086                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6087                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6088                 break;
6089         case -EDEADLK:
6090                 status = nfserr_deadlock;
6091                 break;
6092         default:
6093                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6094                 status = nfserrno(err);
6095                 break;
6096         }
6097 out:
6098         if (nbl) {
6099                 /* dequeue it if we queued it before */
6100                 if (fl_flags & FL_SLEEP) {
6101                         spin_lock(&nn->blocked_locks_lock);
6102                         list_del_init(&nbl->nbl_list);
6103                         list_del_init(&nbl->nbl_lru);
6104                         spin_unlock(&nn->blocked_locks_lock);
6105                 }
6106                 free_blocked_lock(nbl);
6107         }
6108         if (filp)
6109                 fput(filp);
6110         if (lock_stp) {
6111                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6112                 if (cstate->replay_owner &&
6113                     cstate->replay_owner != &lock_sop->lo_owner &&
6114                     seqid_mutating_err(ntohl(status)))
6115                         lock_sop->lo_owner.so_seqid++;
6116
6117                 /*
6118                  * If this is a new, never-before-used stateid, and we are
6119                  * returning an error, then just go ahead and release it.
6120                  */
6121                 if (status && new)
6122                         release_lock_stateid(lock_stp);
6123
6124                 mutex_unlock(&lock_stp->st_mutex);
6125
6126                 nfs4_put_stid(&lock_stp->st_stid);
6127         }
6128         if (open_stp)
6129                 nfs4_put_stid(&open_stp->st_stid);
6130         nfsd4_bump_seqid(cstate, status);
6131         if (conflock)
6132                 locks_free_lock(conflock);
6133         return status;
6134 }
6135
6136 /*
6137  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6138  * so we do a temporary open here just to get an open file to pass to
6139  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6140  * inode operation.)
6141  */
6142 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6143 {
6144         struct file *file;
6145         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6146         if (!err) {
6147                 err = nfserrno(vfs_test_lock(file, lock));
6148                 fput(file);
6149         }
6150         return err;
6151 }
6152
6153 /*
6154  * LOCKT operation
6155  */
6156 __be32
6157 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6158             union nfsd4_op_u *u)
6159 {
6160         struct nfsd4_lockt *lockt = &u->lockt;
6161         struct file_lock *file_lock = NULL;
6162         struct nfs4_lockowner *lo = NULL;
6163         __be32 status;
6164         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6165
6166         if (locks_in_grace(SVC_NET(rqstp)))
6167                 return nfserr_grace;
6168
6169         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6170                  return nfserr_inval;
6171
6172         if (!nfsd4_has_session(cstate)) {
6173                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6174                 if (status)
6175                         goto out;
6176         }
6177
6178         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6179                 goto out;
6180
6181         file_lock = locks_alloc_lock();
6182         if (!file_lock) {
6183                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6184                 status = nfserr_jukebox;
6185                 goto out;
6186         }
6187
6188         switch (lockt->lt_type) {
6189                 case NFS4_READ_LT:
6190                 case NFS4_READW_LT:
6191                         file_lock->fl_type = F_RDLCK;
6192                 break;
6193                 case NFS4_WRITE_LT:
6194                 case NFS4_WRITEW_LT:
6195                         file_lock->fl_type = F_WRLCK;
6196                 break;
6197                 default:
6198                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6199                         status = nfserr_inval;
6200                 goto out;
6201         }
6202
6203         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6204         if (lo)
6205                 file_lock->fl_owner = (fl_owner_t)lo;
6206         file_lock->fl_pid = current->tgid;
6207         file_lock->fl_flags = FL_POSIX;
6208
6209         file_lock->fl_start = lockt->lt_offset;
6210         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6211
6212         nfs4_transform_lock_offset(file_lock);
6213
6214         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6215         if (status)
6216                 goto out;
6217
6218         if (file_lock->fl_type != F_UNLCK) {
6219                 status = nfserr_denied;
6220                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6221         }
6222 out:
6223         if (lo)
6224                 nfs4_put_stateowner(&lo->lo_owner);
6225         if (file_lock)
6226                 locks_free_lock(file_lock);
6227         return status;
6228 }
6229
6230 __be32
6231 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6232             union nfsd4_op_u *u)
6233 {
6234         struct nfsd4_locku *locku = &u->locku;
6235         struct nfs4_ol_stateid *stp;
6236         struct file *filp = NULL;
6237         struct file_lock *file_lock = NULL;
6238         __be32 status;
6239         int err;
6240         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6241
6242         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6243                 (long long) locku->lu_offset,
6244                 (long long) locku->lu_length);
6245
6246         if (check_lock_length(locku->lu_offset, locku->lu_length))
6247                  return nfserr_inval;
6248
6249         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6250                                         &locku->lu_stateid, NFS4_LOCK_STID,
6251                                         &stp, nn);
6252         if (status)
6253                 goto out;
6254         filp = find_any_file(stp->st_stid.sc_file);
6255         if (!filp) {
6256                 status = nfserr_lock_range;
6257                 goto put_stateid;
6258         }
6259         file_lock = locks_alloc_lock();
6260         if (!file_lock) {
6261                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6262                 status = nfserr_jukebox;
6263                 goto fput;
6264         }
6265
6266         file_lock->fl_type = F_UNLCK;
6267         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6268         file_lock->fl_pid = current->tgid;
6269         file_lock->fl_file = filp;
6270         file_lock->fl_flags = FL_POSIX;
6271         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6272         file_lock->fl_start = locku->lu_offset;
6273
6274         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6275                                                 locku->lu_length);
6276         nfs4_transform_lock_offset(file_lock);
6277
6278         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6279         if (err) {
6280                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6281                 goto out_nfserr;
6282         }
6283         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6284 fput:
6285         fput(filp);
6286 put_stateid:
6287         mutex_unlock(&stp->st_mutex);
6288         nfs4_put_stid(&stp->st_stid);
6289 out:
6290         nfsd4_bump_seqid(cstate, status);
6291         if (file_lock)
6292                 locks_free_lock(file_lock);
6293         return status;
6294
6295 out_nfserr:
6296         status = nfserrno(err);
6297         goto fput;
6298 }
6299
6300 /*
6301  * returns
6302  *      true:  locks held by lockowner
6303  *      false: no locks held by lockowner
6304  */
6305 static bool
6306 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6307 {
6308         struct file_lock *fl;
6309         int status = false;
6310         struct file *filp = find_any_file(fp);
6311         struct inode *inode;
6312         struct file_lock_context *flctx;
6313
6314         if (!filp) {
6315                 /* Any valid lock stateid should have some sort of access */
6316                 WARN_ON_ONCE(1);
6317                 return status;
6318         }
6319
6320         inode = file_inode(filp);
6321         flctx = inode->i_flctx;
6322
6323         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6324                 spin_lock(&flctx->flc_lock);
6325                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6326                         if (fl->fl_owner == (fl_owner_t)lowner) {
6327                                 status = true;
6328                                 break;
6329                         }
6330                 }
6331                 spin_unlock(&flctx->flc_lock);
6332         }
6333         fput(filp);
6334         return status;
6335 }
6336
6337 __be32
6338 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6339                         struct nfsd4_compound_state *cstate,
6340                         union nfsd4_op_u *u)
6341 {
6342         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6343         clientid_t *clid = &rlockowner->rl_clientid;
6344         struct nfs4_stateowner *sop;
6345         struct nfs4_lockowner *lo = NULL;
6346         struct nfs4_ol_stateid *stp;
6347         struct xdr_netobj *owner = &rlockowner->rl_owner;
6348         unsigned int hashval = ownerstr_hashval(owner);
6349         __be32 status;
6350         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6351         struct nfs4_client *clp;
6352         LIST_HEAD (reaplist);
6353
6354         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6355                 clid->cl_boot, clid->cl_id);
6356
6357         status = lookup_clientid(clid, cstate, nn);
6358         if (status)
6359                 return status;
6360
6361         clp = cstate->clp;
6362         /* Find the matching lock stateowner */
6363         spin_lock(&clp->cl_lock);
6364         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6365                             so_strhash) {
6366
6367                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6368                         continue;
6369
6370                 /* see if there are still any locks associated with it */
6371                 lo = lockowner(sop);
6372                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6373                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6374                                 status = nfserr_locks_held;
6375                                 spin_unlock(&clp->cl_lock);
6376                                 return status;
6377                         }
6378                 }
6379
6380                 nfs4_get_stateowner(sop);
6381                 break;
6382         }
6383         if (!lo) {
6384                 spin_unlock(&clp->cl_lock);
6385                 return status;
6386         }
6387
6388         unhash_lockowner_locked(lo);
6389         while (!list_empty(&lo->lo_owner.so_stateids)) {
6390                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6391                                        struct nfs4_ol_stateid,
6392                                        st_perstateowner);
6393                 WARN_ON(!unhash_lock_stateid(stp));
6394                 put_ol_stateid_locked(stp, &reaplist);
6395         }
6396         spin_unlock(&clp->cl_lock);
6397         free_ol_stateid_reaplist(&reaplist);
6398         remove_blocked_locks(lo);
6399         nfs4_put_stateowner(&lo->lo_owner);
6400
6401         return status;
6402 }
6403
6404 static inline struct nfs4_client_reclaim *
6405 alloc_reclaim(void)
6406 {
6407         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6408 }
6409
6410 bool
6411 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6412 {
6413         struct nfs4_client_reclaim *crp;
6414
6415         crp = nfsd4_find_reclaim_client(name, nn);
6416         return (crp && crp->cr_clp);
6417 }
6418
6419 /*
6420  * failure => all reset bets are off, nfserr_no_grace...
6421  */
6422 struct nfs4_client_reclaim *
6423 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6424 {
6425         unsigned int strhashval;
6426         struct nfs4_client_reclaim *crp;
6427
6428         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6429         crp = alloc_reclaim();
6430         if (crp) {
6431                 strhashval = clientstr_hashval(name);
6432                 INIT_LIST_HEAD(&crp->cr_strhash);
6433                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6434                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6435                 crp->cr_clp = NULL;
6436                 nn->reclaim_str_hashtbl_size++;
6437         }
6438         return crp;
6439 }
6440
6441 void
6442 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6443 {
6444         list_del(&crp->cr_strhash);
6445         kfree(crp);
6446         nn->reclaim_str_hashtbl_size--;
6447 }
6448
6449 void
6450 nfs4_release_reclaim(struct nfsd_net *nn)
6451 {
6452         struct nfs4_client_reclaim *crp = NULL;
6453         int i;
6454
6455         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6456                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6457                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6458                                         struct nfs4_client_reclaim, cr_strhash);
6459                         nfs4_remove_reclaim_record(crp, nn);
6460                 }
6461         }
6462         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6463 }
6464
6465 /*
6466  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6467 struct nfs4_client_reclaim *
6468 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6469 {
6470         unsigned int strhashval;
6471         struct nfs4_client_reclaim *crp = NULL;
6472
6473         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6474
6475         strhashval = clientstr_hashval(recdir);
6476         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6477                 if (same_name(crp->cr_recdir, recdir)) {
6478                         return crp;
6479                 }
6480         }
6481         return NULL;
6482 }
6483
6484 /*
6485 * Called from OPEN. Look for clientid in reclaim list.
6486 */
6487 __be32
6488 nfs4_check_open_reclaim(clientid_t *clid,
6489                 struct nfsd4_compound_state *cstate,
6490                 struct nfsd_net *nn)
6491 {
6492         __be32 status;
6493
6494         /* find clientid in conf_id_hashtbl */
6495         status = lookup_clientid(clid, cstate, nn);
6496         if (status)
6497                 return nfserr_reclaim_bad;
6498
6499         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6500                 return nfserr_no_grace;
6501
6502         if (nfsd4_client_record_check(cstate->clp))
6503                 return nfserr_reclaim_bad;
6504
6505         return nfs_ok;
6506 }
6507
6508 #ifdef CONFIG_NFSD_FAULT_INJECTION
6509 static inline void
6510 put_client(struct nfs4_client *clp)
6511 {
6512         atomic_dec(&clp->cl_refcount);
6513 }
6514
6515 static struct nfs4_client *
6516 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6517 {
6518         struct nfs4_client *clp;
6519         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6520                                           nfsd_net_id);
6521
6522         if (!nfsd_netns_ready(nn))
6523                 return NULL;
6524
6525         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6526                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6527                         return clp;
6528         }
6529         return NULL;
6530 }
6531
6532 u64
6533 nfsd_inject_print_clients(void)
6534 {
6535         struct nfs4_client *clp;
6536         u64 count = 0;
6537         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6538                                           nfsd_net_id);
6539         char buf[INET6_ADDRSTRLEN];
6540
6541         if (!nfsd_netns_ready(nn))
6542                 return 0;
6543
6544         spin_lock(&nn->client_lock);
6545         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6546                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6547                 pr_info("NFS Client: %s\n", buf);
6548                 ++count;
6549         }
6550         spin_unlock(&nn->client_lock);
6551
6552         return count;
6553 }
6554
6555 u64
6556 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6557 {
6558         u64 count = 0;
6559         struct nfs4_client *clp;
6560         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6561                                           nfsd_net_id);
6562
6563         if (!nfsd_netns_ready(nn))
6564                 return count;
6565
6566         spin_lock(&nn->client_lock);
6567         clp = nfsd_find_client(addr, addr_size);
6568         if (clp) {
6569                 if (mark_client_expired_locked(clp) == nfs_ok)
6570                         ++count;
6571                 else
6572                         clp = NULL;
6573         }
6574         spin_unlock(&nn->client_lock);
6575
6576         if (clp)
6577                 expire_client(clp);
6578
6579         return count;
6580 }
6581
6582 u64
6583 nfsd_inject_forget_clients(u64 max)
6584 {
6585         u64 count = 0;
6586         struct nfs4_client *clp, *next;
6587         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6588                                                 nfsd_net_id);
6589         LIST_HEAD(reaplist);
6590
6591         if (!nfsd_netns_ready(nn))
6592                 return count;
6593
6594         spin_lock(&nn->client_lock);
6595         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6596                 if (mark_client_expired_locked(clp) == nfs_ok) {
6597                         list_add(&clp->cl_lru, &reaplist);
6598                         if (max != 0 && ++count >= max)
6599                                 break;
6600                 }
6601         }
6602         spin_unlock(&nn->client_lock);
6603
6604         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6605                 expire_client(clp);
6606
6607         return count;
6608 }
6609
6610 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6611                              const char *type)
6612 {
6613         char buf[INET6_ADDRSTRLEN];
6614         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6615         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6616 }
6617
6618 static void
6619 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6620                              struct list_head *collect)
6621 {
6622         struct nfs4_client *clp = lst->st_stid.sc_client;
6623         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6624                                           nfsd_net_id);
6625
6626         if (!collect)
6627                 return;
6628
6629         lockdep_assert_held(&nn->client_lock);
6630         atomic_inc(&clp->cl_refcount);
6631         list_add(&lst->st_locks, collect);
6632 }
6633
6634 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6635                                     struct list_head *collect,
6636                                     bool (*func)(struct nfs4_ol_stateid *))
6637 {
6638         struct nfs4_openowner *oop;
6639         struct nfs4_ol_stateid *stp, *st_next;
6640         struct nfs4_ol_stateid *lst, *lst_next;
6641         u64 count = 0;
6642
6643         spin_lock(&clp->cl_lock);
6644         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6645                 list_for_each_entry_safe(stp, st_next,
6646                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6647                         list_for_each_entry_safe(lst, lst_next,
6648                                         &stp->st_locks, st_locks) {
6649                                 if (func) {
6650                                         if (func(lst))
6651                                                 nfsd_inject_add_lock_to_list(lst,
6652                                                                         collect);
6653                                 }
6654                                 ++count;
6655                                 /*
6656                                  * Despite the fact that these functions deal
6657                                  * with 64-bit integers for "count", we must
6658                                  * ensure that it doesn't blow up the
6659                                  * clp->cl_refcount. Throw a warning if we
6660                                  * start to approach INT_MAX here.
6661                                  */
6662                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6663                                 if (count == max)
6664                                         goto out;
6665                         }
6666                 }
6667         }
6668 out:
6669         spin_unlock(&clp->cl_lock);
6670
6671         return count;
6672 }
6673
6674 static u64
6675 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6676                           u64 max)
6677 {
6678         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6679 }
6680
6681 static u64
6682 nfsd_print_client_locks(struct nfs4_client *clp)
6683 {
6684         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6685         nfsd_print_count(clp, count, "locked files");
6686         return count;
6687 }
6688
6689 u64
6690 nfsd_inject_print_locks(void)
6691 {
6692         struct nfs4_client *clp;
6693         u64 count = 0;
6694         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6695                                                 nfsd_net_id);
6696
6697         if (!nfsd_netns_ready(nn))
6698                 return 0;
6699
6700         spin_lock(&nn->client_lock);
6701         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6702                 count += nfsd_print_client_locks(clp);
6703         spin_unlock(&nn->client_lock);
6704
6705         return count;
6706 }
6707
6708 static void
6709 nfsd_reap_locks(struct list_head *reaplist)
6710 {
6711         struct nfs4_client *clp;
6712         struct nfs4_ol_stateid *stp, *next;
6713
6714         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6715                 list_del_init(&stp->st_locks);
6716                 clp = stp->st_stid.sc_client;
6717                 nfs4_put_stid(&stp->st_stid);
6718                 put_client(clp);
6719         }
6720 }
6721
6722 u64
6723 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6724 {
6725         unsigned int count = 0;
6726         struct nfs4_client *clp;
6727         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6728                                                 nfsd_net_id);
6729         LIST_HEAD(reaplist);
6730
6731         if (!nfsd_netns_ready(nn))
6732                 return count;
6733
6734         spin_lock(&nn->client_lock);
6735         clp = nfsd_find_client(addr, addr_size);
6736         if (clp)
6737                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6738         spin_unlock(&nn->client_lock);
6739         nfsd_reap_locks(&reaplist);
6740         return count;
6741 }
6742
6743 u64
6744 nfsd_inject_forget_locks(u64 max)
6745 {
6746         u64 count = 0;
6747         struct nfs4_client *clp;
6748         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6749                                                 nfsd_net_id);
6750         LIST_HEAD(reaplist);
6751
6752         if (!nfsd_netns_ready(nn))
6753                 return count;
6754
6755         spin_lock(&nn->client_lock);
6756         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6757                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6758                 if (max != 0 && count >= max)
6759                         break;
6760         }
6761         spin_unlock(&nn->client_lock);
6762         nfsd_reap_locks(&reaplist);
6763         return count;
6764 }
6765
6766 static u64
6767 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6768                               struct list_head *collect,
6769                               void (*func)(struct nfs4_openowner *))
6770 {
6771         struct nfs4_openowner *oop, *next;
6772         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6773                                                 nfsd_net_id);
6774         u64 count = 0;
6775
6776         lockdep_assert_held(&nn->client_lock);
6777
6778         spin_lock(&clp->cl_lock);
6779         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6780                 if (func) {
6781                         func(oop);
6782                         if (collect) {
6783                                 atomic_inc(&clp->cl_refcount);
6784                                 list_add(&oop->oo_perclient, collect);
6785                         }
6786                 }
6787                 ++count;
6788                 /*
6789                  * Despite the fact that these functions deal with
6790                  * 64-bit integers for "count", we must ensure that
6791                  * it doesn't blow up the clp->cl_refcount. Throw a
6792                  * warning if we start to approach INT_MAX here.
6793                  */
6794                 WARN_ON_ONCE(count == (INT_MAX / 2));
6795                 if (count == max)
6796                         break;
6797         }
6798         spin_unlock(&clp->cl_lock);
6799
6800         return count;
6801 }
6802
6803 static u64
6804 nfsd_print_client_openowners(struct nfs4_client *clp)
6805 {
6806         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6807
6808         nfsd_print_count(clp, count, "openowners");
6809         return count;
6810 }
6811
6812 static u64
6813 nfsd_collect_client_openowners(struct nfs4_client *clp,
6814                                struct list_head *collect, u64 max)
6815 {
6816         return nfsd_foreach_client_openowner(clp, max, collect,
6817                                                 unhash_openowner_locked);
6818 }
6819
6820 u64
6821 nfsd_inject_print_openowners(void)
6822 {
6823         struct nfs4_client *clp;
6824         u64 count = 0;
6825         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6826                                                 nfsd_net_id);
6827
6828         if (!nfsd_netns_ready(nn))
6829                 return 0;
6830
6831         spin_lock(&nn->client_lock);
6832         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6833                 count += nfsd_print_client_openowners(clp);
6834         spin_unlock(&nn->client_lock);
6835
6836         return count;
6837 }
6838
6839 static void
6840 nfsd_reap_openowners(struct list_head *reaplist)
6841 {
6842         struct nfs4_client *clp;
6843         struct nfs4_openowner *oop, *next;
6844
6845         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6846                 list_del_init(&oop->oo_perclient);
6847                 clp = oop->oo_owner.so_client;
6848                 release_openowner(oop);
6849                 put_client(clp);
6850         }
6851 }
6852
6853 u64
6854 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6855                                      size_t addr_size)
6856 {
6857         unsigned int count = 0;
6858         struct nfs4_client *clp;
6859         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6860                                                 nfsd_net_id);
6861         LIST_HEAD(reaplist);
6862
6863         if (!nfsd_netns_ready(nn))
6864                 return count;
6865
6866         spin_lock(&nn->client_lock);
6867         clp = nfsd_find_client(addr, addr_size);
6868         if (clp)
6869                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6870         spin_unlock(&nn->client_lock);
6871         nfsd_reap_openowners(&reaplist);
6872         return count;
6873 }
6874
6875 u64
6876 nfsd_inject_forget_openowners(u64 max)
6877 {
6878         u64 count = 0;
6879         struct nfs4_client *clp;
6880         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6881                                                 nfsd_net_id);
6882         LIST_HEAD(reaplist);
6883
6884         if (!nfsd_netns_ready(nn))
6885                 return count;
6886
6887         spin_lock(&nn->client_lock);
6888         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6889                 count += nfsd_collect_client_openowners(clp, &reaplist,
6890                                                         max - count);
6891                 if (max != 0 && count >= max)
6892                         break;
6893         }
6894         spin_unlock(&nn->client_lock);
6895         nfsd_reap_openowners(&reaplist);
6896         return count;
6897 }
6898
6899 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6900                                      struct list_head *victims)
6901 {
6902         struct nfs4_delegation *dp, *next;
6903         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6904                                                 nfsd_net_id);
6905         u64 count = 0;
6906
6907         lockdep_assert_held(&nn->client_lock);
6908
6909         spin_lock(&state_lock);
6910         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6911                 if (victims) {
6912                         /*
6913                          * It's not safe to mess with delegations that have a
6914                          * non-zero dl_time. They might have already been broken
6915                          * and could be processed by the laundromat outside of
6916                          * the state_lock. Just leave them be.
6917                          */
6918                         if (dp->dl_time != 0)
6919                                 continue;
6920
6921                         atomic_inc(&clp->cl_refcount);
6922                         WARN_ON(!unhash_delegation_locked(dp));
6923                         list_add(&dp->dl_recall_lru, victims);
6924                 }
6925                 ++count;
6926                 /*
6927                  * Despite the fact that these functions deal with
6928                  * 64-bit integers for "count", we must ensure that
6929                  * it doesn't blow up the clp->cl_refcount. Throw a
6930                  * warning if we start to approach INT_MAX here.
6931                  */
6932                 WARN_ON_ONCE(count == (INT_MAX / 2));
6933                 if (count == max)
6934                         break;
6935         }
6936         spin_unlock(&state_lock);
6937         return count;
6938 }
6939
6940 static u64
6941 nfsd_print_client_delegations(struct nfs4_client *clp)
6942 {
6943         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6944
6945         nfsd_print_count(clp, count, "delegations");
6946         return count;
6947 }
6948
6949 u64
6950 nfsd_inject_print_delegations(void)
6951 {
6952         struct nfs4_client *clp;
6953         u64 count = 0;
6954         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6955                                                 nfsd_net_id);
6956
6957         if (!nfsd_netns_ready(nn))
6958                 return 0;
6959
6960         spin_lock(&nn->client_lock);
6961         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6962                 count += nfsd_print_client_delegations(clp);
6963         spin_unlock(&nn->client_lock);
6964
6965         return count;
6966 }
6967
6968 static void
6969 nfsd_forget_delegations(struct list_head *reaplist)
6970 {
6971         struct nfs4_client *clp;
6972         struct nfs4_delegation *dp, *next;
6973
6974         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6975                 list_del_init(&dp->dl_recall_lru);
6976                 clp = dp->dl_stid.sc_client;
6977                 revoke_delegation(dp);
6978                 put_client(clp);
6979         }
6980 }
6981
6982 u64
6983 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6984                                       size_t addr_size)
6985 {
6986         u64 count = 0;
6987         struct nfs4_client *clp;
6988         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6989                                                 nfsd_net_id);
6990         LIST_HEAD(reaplist);
6991
6992         if (!nfsd_netns_ready(nn))
6993                 return count;
6994
6995         spin_lock(&nn->client_lock);
6996         clp = nfsd_find_client(addr, addr_size);
6997         if (clp)
6998                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6999         spin_unlock(&nn->client_lock);
7000
7001         nfsd_forget_delegations(&reaplist);
7002         return count;
7003 }
7004
7005 u64
7006 nfsd_inject_forget_delegations(u64 max)
7007 {
7008         u64 count = 0;
7009         struct nfs4_client *clp;
7010         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7011                                                 nfsd_net_id);
7012         LIST_HEAD(reaplist);
7013
7014         if (!nfsd_netns_ready(nn))
7015                 return count;
7016
7017         spin_lock(&nn->client_lock);
7018         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7019                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7020                 if (max != 0 && count >= max)
7021                         break;
7022         }
7023         spin_unlock(&nn->client_lock);
7024         nfsd_forget_delegations(&reaplist);
7025         return count;
7026 }
7027
7028 static void
7029 nfsd_recall_delegations(struct list_head *reaplist)
7030 {
7031         struct nfs4_client *clp;
7032         struct nfs4_delegation *dp, *next;
7033
7034         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7035                 list_del_init(&dp->dl_recall_lru);
7036                 clp = dp->dl_stid.sc_client;
7037                 /*
7038                  * We skipped all entries that had a zero dl_time before,
7039                  * so we can now reset the dl_time back to 0. If a delegation
7040                  * break comes in now, then it won't make any difference since
7041                  * we're recalling it either way.
7042                  */
7043                 spin_lock(&state_lock);
7044                 dp->dl_time = 0;
7045                 spin_unlock(&state_lock);
7046                 nfsd_break_one_deleg(dp);
7047                 put_client(clp);
7048         }
7049 }
7050
7051 u64
7052 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7053                                       size_t addr_size)
7054 {
7055         u64 count = 0;
7056         struct nfs4_client *clp;
7057         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7058                                                 nfsd_net_id);
7059         LIST_HEAD(reaplist);
7060
7061         if (!nfsd_netns_ready(nn))
7062                 return count;
7063
7064         spin_lock(&nn->client_lock);
7065         clp = nfsd_find_client(addr, addr_size);
7066         if (clp)
7067                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7068         spin_unlock(&nn->client_lock);
7069
7070         nfsd_recall_delegations(&reaplist);
7071         return count;
7072 }
7073
7074 u64
7075 nfsd_inject_recall_delegations(u64 max)
7076 {
7077         u64 count = 0;
7078         struct nfs4_client *clp, *next;
7079         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7080                                                 nfsd_net_id);
7081         LIST_HEAD(reaplist);
7082
7083         if (!nfsd_netns_ready(nn))
7084                 return count;
7085
7086         spin_lock(&nn->client_lock);
7087         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7088                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7089                 if (max != 0 && ++count >= max)
7090                         break;
7091         }
7092         spin_unlock(&nn->client_lock);
7093         nfsd_recall_delegations(&reaplist);
7094         return count;
7095 }
7096 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7097
7098 /*
7099  * Since the lifetime of a delegation isn't limited to that of an open, a
7100  * client may quite reasonably hang on to a delegation as long as it has
7101  * the inode cached.  This becomes an obvious problem the first time a
7102  * client's inode cache approaches the size of the server's total memory.
7103  *
7104  * For now we avoid this problem by imposing a hard limit on the number
7105  * of delegations, which varies according to the server's memory size.
7106  */
7107 static void
7108 set_max_delegations(void)
7109 {
7110         /*
7111          * Allow at most 4 delegations per megabyte of RAM.  Quick
7112          * estimates suggest that in the worst case (where every delegation
7113          * is for a different inode), a delegation could take about 1.5K,
7114          * giving a worst case usage of about 6% of memory.
7115          */
7116         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7117 }
7118
7119 static int nfs4_state_create_net(struct net *net)
7120 {
7121         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7122         int i;
7123
7124         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7125                         CLIENT_HASH_SIZE, GFP_KERNEL);
7126         if (!nn->conf_id_hashtbl)
7127                 goto err;
7128         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7129                         CLIENT_HASH_SIZE, GFP_KERNEL);
7130         if (!nn->unconf_id_hashtbl)
7131                 goto err_unconf_id;
7132         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
7133                         SESSION_HASH_SIZE, GFP_KERNEL);
7134         if (!nn->sessionid_hashtbl)
7135                 goto err_sessionid;
7136
7137         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7138                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7139                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7140         }
7141         for (i = 0; i < SESSION_HASH_SIZE; i++)
7142                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7143         nn->conf_name_tree = RB_ROOT;
7144         nn->unconf_name_tree = RB_ROOT;
7145         nn->boot_time = get_seconds();
7146         nn->grace_ended = false;
7147         nn->nfsd4_manager.block_opens = true;
7148         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7149         INIT_LIST_HEAD(&nn->client_lru);
7150         INIT_LIST_HEAD(&nn->close_lru);
7151         INIT_LIST_HEAD(&nn->del_recall_lru);
7152         spin_lock_init(&nn->client_lock);
7153
7154         spin_lock_init(&nn->blocked_locks_lock);
7155         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7156
7157         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7158         get_net(net);
7159
7160         return 0;
7161
7162 err_sessionid:
7163         kfree(nn->unconf_id_hashtbl);
7164 err_unconf_id:
7165         kfree(nn->conf_id_hashtbl);
7166 err:
7167         return -ENOMEM;
7168 }
7169
7170 static void
7171 nfs4_state_destroy_net(struct net *net)
7172 {
7173         int i;
7174         struct nfs4_client *clp = NULL;
7175         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7176
7177         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7178                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7179                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7180                         destroy_client(clp);
7181                 }
7182         }
7183
7184         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7185
7186         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7187                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7188                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7189                         destroy_client(clp);
7190                 }
7191         }
7192
7193         kfree(nn->sessionid_hashtbl);
7194         kfree(nn->unconf_id_hashtbl);
7195         kfree(nn->conf_id_hashtbl);
7196         put_net(net);
7197 }
7198
7199 int
7200 nfs4_state_start_net(struct net *net)
7201 {
7202         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7203         int ret;
7204
7205         ret = nfs4_state_create_net(net);
7206         if (ret)
7207                 return ret;
7208         locks_start_grace(net, &nn->nfsd4_manager);
7209         nfsd4_client_tracking_init(net);
7210         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7211                nn->nfsd4_grace, net->ns.inum);
7212         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7213         return 0;
7214 }
7215
7216 /* initialization to perform when the nfsd service is started: */
7217
7218 int
7219 nfs4_state_start(void)
7220 {
7221         int ret;
7222
7223         ret = set_callback_cred();
7224         if (ret)
7225                 return ret;
7226
7227         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7228         if (laundry_wq == NULL) {
7229                 ret = -ENOMEM;
7230                 goto out_cleanup_cred;
7231         }
7232         ret = nfsd4_create_callback_queue();
7233         if (ret)
7234                 goto out_free_laundry;
7235
7236         set_max_delegations();
7237         return 0;
7238
7239 out_free_laundry:
7240         destroy_workqueue(laundry_wq);
7241 out_cleanup_cred:
7242         cleanup_callback_cred();
7243         return ret;
7244 }
7245
7246 void
7247 nfs4_state_shutdown_net(struct net *net)
7248 {
7249         struct nfs4_delegation *dp = NULL;
7250         struct list_head *pos, *next, reaplist;
7251         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7252
7253         cancel_delayed_work_sync(&nn->laundromat_work);
7254         locks_end_grace(&nn->nfsd4_manager);
7255
7256         INIT_LIST_HEAD(&reaplist);
7257         spin_lock(&state_lock);
7258         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7259                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7260                 WARN_ON(!unhash_delegation_locked(dp));
7261                 list_add(&dp->dl_recall_lru, &reaplist);
7262         }
7263         spin_unlock(&state_lock);
7264         list_for_each_safe(pos, next, &reaplist) {
7265                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7266                 list_del_init(&dp->dl_recall_lru);
7267                 put_clnt_odstate(dp->dl_clnt_odstate);
7268                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7269                 nfs4_put_stid(&dp->dl_stid);
7270         }
7271
7272         nfsd4_client_tracking_exit(net);
7273         nfs4_state_destroy_net(net);
7274 }
7275
7276 void
7277 nfs4_state_shutdown(void)
7278 {
7279         destroy_workqueue(laundry_wq);
7280         nfsd4_destroy_callback_queue();
7281         cleanup_callback_cred();
7282 }
7283
7284 static void
7285 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7286 {
7287         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7288                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7289 }
7290
7291 static void
7292 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7293 {
7294         if (cstate->minorversion) {
7295                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7296                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7297         }
7298 }
7299
7300 void
7301 clear_current_stateid(struct nfsd4_compound_state *cstate)
7302 {
7303         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7304 }
7305
7306 /*
7307  * functions to set current state id
7308  */
7309 void
7310 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7311                 union nfsd4_op_u *u)
7312 {
7313         put_stateid(cstate, &u->open_downgrade.od_stateid);
7314 }
7315
7316 void
7317 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7318                 union nfsd4_op_u *u)
7319 {
7320         put_stateid(cstate, &u->open.op_stateid);
7321 }
7322
7323 void
7324 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7325                 union nfsd4_op_u *u)
7326 {
7327         put_stateid(cstate, &u->close.cl_stateid);
7328 }
7329
7330 void
7331 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7332                 union nfsd4_op_u *u)
7333 {
7334         put_stateid(cstate, &u->lock.lk_resp_stateid);
7335 }
7336
7337 /*
7338  * functions to consume current state id
7339  */
7340
7341 void
7342 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7343                 union nfsd4_op_u *u)
7344 {
7345         get_stateid(cstate, &u->open_downgrade.od_stateid);
7346 }
7347
7348 void
7349 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7350                 union nfsd4_op_u *u)
7351 {
7352         get_stateid(cstate, &u->delegreturn.dr_stateid);
7353 }
7354
7355 void
7356 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7357                 union nfsd4_op_u *u)
7358 {
7359         get_stateid(cstate, &u->free_stateid.fr_stateid);
7360 }
7361
7362 void
7363 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7364                 union nfsd4_op_u *u)
7365 {
7366         get_stateid(cstate, &u->setattr.sa_stateid);
7367 }
7368
7369 void
7370 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7371                 union nfsd4_op_u *u)
7372 {
7373         get_stateid(cstate, &u->close.cl_stateid);
7374 }
7375
7376 void
7377 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7378                 union nfsd4_op_u *u)
7379 {
7380         get_stateid(cstate, &u->locku.lu_stateid);
7381 }
7382
7383 void
7384 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7385                 union nfsd4_op_u *u)
7386 {
7387         get_stateid(cstate, &u->read.rd_stateid);
7388 }
7389
7390 void
7391 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7392                 union nfsd4_op_u *u)
7393 {
7394         get_stateid(cstate, &u->write.wr_stateid);
7395 }