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
  67 static u64 current_sessionid = 1;
  68
  69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  70 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  72
  73 /* forward declarations */
  74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  76
  77 /* Locking: */
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(state_lock);
  85
  86 /*
  87  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  88  * the refcount on the open stateid to drop.
  89  */
  90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  91
  92 static struct kmem_cache *openowner_slab;
  93 static struct kmem_cache *lockowner_slab;
  94 static struct kmem_cache *file_slab;
  95 static struct kmem_cache *stateid_slab;
  96 static struct kmem_cache *deleg_slab;
  97 static struct kmem_cache *odstate_slab;
  98
  99 static void free_session(struct nfsd4_session *);
 100
 101 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 102 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 103
 104 static bool is_session_dead(struct nfsd4_session *ses)
 105 {
 106         return ses->se_flags & NFS4_SESSION_DEAD;
 107 }
 108
 109 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 110 {
 111         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 112                 return nfserr_jukebox;
 113         ses->se_flags |= NFS4_SESSION_DEAD;
 114         return nfs_ok;
 115 }
 116
 117 static bool is_client_expired(struct nfs4_client *clp)
 118 {
 119         return clp->cl_time == 0;
 120 }
 121
 122 static __be32 get_client_locked(struct nfs4_client *clp)
 123 {
 124         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 125
 126         lockdep_assert_held(&nn->client_lock);
 127
 128         if (is_client_expired(clp))
 129                 return nfserr_expired;
 130         atomic_inc(&clp->cl_refcount);
 131         return nfs_ok;
 132 }
 133
 134 /* must be called under the client_lock */
 135 static inline void
 136 renew_client_locked(struct nfs4_client *clp)
 137 {
 138         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 139
 140         if (is_client_expired(clp)) {
 141                 WARN_ON(1);
 142                 printk("%s: client (clientid %08x/%08x) already expired\n",
 143                         __func__,
 144                         clp->cl_clientid.cl_boot,
 145                         clp->cl_clientid.cl_id);
 146                 return;
 147         }
 148
 149         dprintk("renewing client (clientid %08x/%08x)\n",
 150                         clp->cl_clientid.cl_boot,
 151                         clp->cl_clientid.cl_id);
 152         list_move_tail(&clp->cl_lru, &nn->client_lru);
 153         clp->cl_time = get_seconds();
 154 }
 155
 156 static void put_client_renew_locked(struct nfs4_client *clp)
 157 {
 158         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 159
 160         lockdep_assert_held(&nn->client_lock);
 161
 162         if (!atomic_dec_and_test(&clp->cl_refcount))
 163                 return;
 164         if (!is_client_expired(clp))
 165                 renew_client_locked(clp);
 166 }
 167
 168 static void put_client_renew(struct nfs4_client *clp)
 169 {
 170         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 171
 172         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 173                 return;
 174         if (!is_client_expired(clp))
 175                 renew_client_locked(clp);
 176         spin_unlock(&nn->client_lock);
 177 }
 178
 179 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 180 {
 181         __be32 status;
 182
 183         if (is_session_dead(ses))
 184                 return nfserr_badsession;
 185         status = get_client_locked(ses->se_client);
 186         if (status)
 187                 return status;
 188         atomic_inc(&ses->se_ref);
 189         return nfs_ok;
 190 }
 191
 192 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 193 {
 194         struct nfs4_client *clp = ses->se_client;
 195         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 196
 197         lockdep_assert_held(&nn->client_lock);
 198
 199         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 200                 free_session(ses);
 201         put_client_renew_locked(clp);
 202 }
 203
 204 static void nfsd4_put_session(struct nfsd4_session *ses)
 205 {
 206         struct nfs4_client *clp = ses->se_client;
 207         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 208
 209         spin_lock(&nn->client_lock);
 210         nfsd4_put_session_locked(ses);
 211         spin_unlock(&nn->client_lock);
 212 }
 213
 214 static struct nfsd4_blocked_lock *
 215 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 216                         struct nfsd_net *nn)
 217 {
 218         struct nfsd4_blocked_lock *cur, *found = NULL;
 219
 220         spin_lock(&nn->blocked_locks_lock);
 221         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 222                 if (fh_match(fh, &cur->nbl_fh)) {
 223                         list_del_init(&cur->nbl_list);
 224                         list_del_init(&cur->nbl_lru);
 225                         found = cur;
 226                         break;
 227                 }
 228         }
 229         spin_unlock(&nn->blocked_locks_lock);
 230         if (found)
 231                 posix_unblock_lock(&found->nbl_lock);
 232         return found;
 233 }
 234
 235 static struct nfsd4_blocked_lock *
 236 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 237                         struct nfsd_net *nn)
 238 {
 239         struct nfsd4_blocked_lock *nbl;
 240
 241         nbl = find_blocked_lock(lo, fh, nn);
 242         if (!nbl) {
 243                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 244                 if (nbl) {
 245                         fh_copy_shallow(&nbl->nbl_fh, fh);
 246                         locks_init_lock(&nbl->nbl_lock);
 247                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 248                                         &nfsd4_cb_notify_lock_ops,
 249                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 250                 }
 251         }
 252         return nbl;
 253 }
 254
 255 static void
 256 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 257 {
 258         locks_release_private(&nbl->nbl_lock);
 259         kfree(nbl);
 260 }
 261
 262 static int
 263 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 264 {
 265         /*
 266          * Since this is just an optimization, we don't try very hard if it
 267          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 268          * just quit trying on anything else.
 269          */
 270         switch (task->tk_status) {
 271         case -NFS4ERR_DELAY:
 272                 rpc_delay(task, 1 * HZ);
 273                 return 0;
 274         default:
 275                 return 1;
 276         }
 277 }
 278
 279 static void
 280 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 281 {
 282         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 283                                                 struct nfsd4_blocked_lock, nbl_cb);
 284
 285         free_blocked_lock(nbl);
 286 }
 287
 288 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 289         .done           = nfsd4_cb_notify_lock_done,
 290         .release        = nfsd4_cb_notify_lock_release,
 291 };
 292
 293 static inline struct nfs4_stateowner *
 294 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 295 {
 296         atomic_inc(&sop->so_count);
 297         return sop;
 298 }
 299
 300 static int
 301 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 302 {
 303         return (sop->so_owner.len == owner->len) &&
 304                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 305 }
 306
 307 static struct nfs4_openowner *
 308 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 309                         struct nfs4_client *clp)
 310 {
 311         struct nfs4_stateowner *so;
 312
 313         lockdep_assert_held(&clp->cl_lock);
 314
 315         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 316                             so_strhash) {
 317                 if (!so->so_is_open_owner)
 318                         continue;
 319                 if (same_owner_str(so, &open->op_owner))
 320                         return openowner(nfs4_get_stateowner(so));
 321         }
 322         return NULL;
 323 }
 324
 325 static struct nfs4_openowner *
 326 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 327                         struct nfs4_client *clp)
 328 {
 329         struct nfs4_openowner *oo;
 330
 331         spin_lock(&clp->cl_lock);
 332         oo = find_openstateowner_str_locked(hashval, open, clp);
 333         spin_unlock(&clp->cl_lock);
 334         return oo;
 335 }
 336
 337 static inline u32
 338 opaque_hashval(const void *ptr, int nbytes)
 339 {
 340         unsigned char *cptr = (unsigned char *) ptr;
 341
 342         u32 x = 0;
 343         while (nbytes--) {
 344                 x *= 37;
 345                 x += *cptr++;
 346         }
 347         return x;
 348 }
 349
 350 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 351 {
 352         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 353
 354         kmem_cache_free(file_slab, fp);
 355 }
 356
 357 void
 358 put_nfs4_file(struct nfs4_file *fi)
 359 {
 360         might_lock(&state_lock);
 361
 362         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 363                 hlist_del_rcu(&fi->fi_hash);
 364                 spin_unlock(&state_lock);
 365                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 366                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 367                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 368         }
 369 }
 370
 371 static struct file *
 372 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 373 {
 374         if (f->fi_fds[oflag])
 375                 return get_file(f->fi_fds[oflag]);
 376         return NULL;
 377 }
 378
 379 static struct file *
 380 find_writeable_file_locked(struct nfs4_file *f)
 381 {
 382         struct file *ret;
 383
 384         lockdep_assert_held(&f->fi_lock);
 385
 386         ret = __nfs4_get_fd(f, O_WRONLY);
 387         if (!ret)
 388                 ret = __nfs4_get_fd(f, O_RDWR);
 389         return ret;
 390 }
 391
 392 static struct file *
 393 find_writeable_file(struct nfs4_file *f)
 394 {
 395         struct file *ret;
 396
 397         spin_lock(&f->fi_lock);
 398         ret = find_writeable_file_locked(f);
 399         spin_unlock(&f->fi_lock);
 400
 401         return ret;
 402 }
 403
 404 static struct file *find_readable_file_locked(struct nfs4_file *f)
 405 {
 406         struct file *ret;
 407
 408         lockdep_assert_held(&f->fi_lock);
 409
 410         ret = __nfs4_get_fd(f, O_RDONLY);
 411         if (!ret)
 412                 ret = __nfs4_get_fd(f, O_RDWR);
 413         return ret;
 414 }
 415
 416 static struct file *
 417 find_readable_file(struct nfs4_file *f)
 418 {
 419         struct file *ret;
 420
 421         spin_lock(&f->fi_lock);
 422         ret = find_readable_file_locked(f);
 423         spin_unlock(&f->fi_lock);
 424
 425         return ret;
 426 }
 427
 428 struct file *
 429 find_any_file(struct nfs4_file *f)
 430 {
 431         struct file *ret;
 432
 433         spin_lock(&f->fi_lock);
 434         ret = __nfs4_get_fd(f, O_RDWR);
 435         if (!ret) {
 436                 ret = __nfs4_get_fd(f, O_WRONLY);
 437                 if (!ret)
 438                         ret = __nfs4_get_fd(f, O_RDONLY);
 439         }
 440         spin_unlock(&f->fi_lock);
 441         return ret;
 442 }
 443
 444 static atomic_long_t num_delegations;
 445 unsigned long max_delegations;
 446
 447 /*
 448  * Open owner state (share locks)
 449  */
 450
 451 /* hash tables for lock and open owners */
 452 #define OWNER_HASH_BITS              8
 453 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 454 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 455
 456 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 457 {
 458         unsigned int ret;
 459
 460         ret = opaque_hashval(ownername->data, ownername->len);
 461         return ret & OWNER_HASH_MASK;
 462 }
 463
 464 /* hash table for nfs4_file */
 465 #define FILE_HASH_BITS                   8
 466 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 467
 468 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 469 {
 470         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 471 }
 472
 473 static unsigned int file_hashval(struct knfsd_fh *fh)
 474 {
 475         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 476 }
 477
 478 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 479
 480 static void
 481 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 482 {
 483         lockdep_assert_held(&fp->fi_lock);
 484
 485         if (access & NFS4_SHARE_ACCESS_WRITE)
 486                 atomic_inc(&fp->fi_access[O_WRONLY]);
 487         if (access & NFS4_SHARE_ACCESS_READ)
 488                 atomic_inc(&fp->fi_access[O_RDONLY]);
 489 }
 490
 491 static __be32
 492 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 493 {
 494         lockdep_assert_held(&fp->fi_lock);
 495
 496         /* Does this access mode make sense? */
 497         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 498                 return nfserr_inval;
 499
 500         /* Does it conflict with a deny mode already set? */
 501         if ((access & fp->fi_share_deny) != 0)
 502                 return nfserr_share_denied;
 503
 504         __nfs4_file_get_access(fp, access);
 505         return nfs_ok;
 506 }
 507
 508 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 509 {
 510         /* Common case is that there is no deny mode. */
 511         if (deny) {
 512                 /* Does this deny mode make sense? */
 513                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 514                         return nfserr_inval;
 515
 516                 if ((deny & NFS4_SHARE_DENY_READ) &&
 517                     atomic_read(&fp->fi_access[O_RDONLY]))
 518                         return nfserr_share_denied;
 519
 520                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 521                     atomic_read(&fp->fi_access[O_WRONLY]))
 522                         return nfserr_share_denied;
 523         }
 524         return nfs_ok;
 525 }
 526
 527 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 528 {
 529         might_lock(&fp->fi_lock);
 530
 531         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 532                 struct file *f1 = NULL;
 533                 struct file *f2 = NULL;
 534
 535                 swap(f1, fp->fi_fds[oflag]);
 536                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 537                         swap(f2, fp->fi_fds[O_RDWR]);
 538                 spin_unlock(&fp->fi_lock);
 539                 if (f1)
 540                         fput(f1);
 541                 if (f2)
 542                         fput(f2);
 543         }
 544 }
 545
 546 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 547 {
 548         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 549
 550         if (access & NFS4_SHARE_ACCESS_WRITE)
 551                 __nfs4_file_put_access(fp, O_WRONLY);
 552         if (access & NFS4_SHARE_ACCESS_READ)
 553                 __nfs4_file_put_access(fp, O_RDONLY);
 554 }
 555
 556 /*
 557  * Allocate a new open/delegation state counter. This is needed for
 558  * pNFS for proper return on close semantics.
 559  *
 560  * Note that we only allocate it for pNFS-enabled exports, otherwise
 561  * all pointers to struct nfs4_clnt_odstate are always NULL.
 562  */
 563 static struct nfs4_clnt_odstate *
 564 alloc_clnt_odstate(struct nfs4_client *clp)
 565 {
 566         struct nfs4_clnt_odstate *co;
 567
 568         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 569         if (co) {
 570                 co->co_client = clp;
 571                 atomic_set(&co->co_odcount, 1);
 572         }
 573         return co;
 574 }
 575
 576 static void
 577 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 578 {
 579         struct nfs4_file *fp = co->co_file;
 580
 581         lockdep_assert_held(&fp->fi_lock);
 582         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 583 }
 584
 585 static inline void
 586 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 587 {
 588         if (co)
 589                 atomic_inc(&co->co_odcount);
 590 }
 591
 592 static void
 593 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 594 {
 595         struct nfs4_file *fp;
 596
 597         if (!co)
 598                 return;
 599
 600         fp = co->co_file;
 601         if (atomic_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 602                 list_del(&co->co_perfile);
 603                 spin_unlock(&fp->fi_lock);
 604
 605                 nfsd4_return_all_file_layouts(co->co_client, fp);
 606                 kmem_cache_free(odstate_slab, co);
 607         }
 608 }
 609
 610 static struct nfs4_clnt_odstate *
 611 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 612 {
 613         struct nfs4_clnt_odstate *co;
 614         struct nfs4_client *cl;
 615
 616         if (!new)
 617                 return NULL;
 618
 619         cl = new->co_client;
 620
 621         spin_lock(&fp->fi_lock);
 622         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 623                 if (co->co_client == cl) {
 624                         get_clnt_odstate(co);
 625                         goto out;
 626                 }
 627         }
 628         co = new;
 629         co->co_file = fp;
 630         hash_clnt_odstate_locked(new);
 631 out:
 632         spin_unlock(&fp->fi_lock);
 633         return co;
 634 }
 635
 636 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 637                                   void (*sc_free)(struct nfs4_stid *))
 638 {
 639         struct nfs4_stid *stid;
 640         int new_id;
 641
 642         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 643         if (!stid)
 644                 return NULL;
 645
 646         idr_preload(GFP_KERNEL);
 647         spin_lock(&cl->cl_lock);
 648         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 649         spin_unlock(&cl->cl_lock);
 650         idr_preload_end();
 651         if (new_id < 0)
 652                 goto out_free;
 653
 654         stid->sc_free = sc_free;
 655         stid->sc_client = cl;
 656         stid->sc_stateid.si_opaque.so_id = new_id;
 657         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 658         /* Will be incremented before return to client: */
 659         atomic_set(&stid->sc_count, 1);
 660         spin_lock_init(&stid->sc_lock);
 661
 662         /*
 663          * It shouldn't be a problem to reuse an opaque stateid value.
 664          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 665          * example, a stray write retransmission could be accepted by
 666          * the server when it should have been rejected.  Therefore,
 667          * adopt a trick from the sctp code to attempt to maximize the
 668          * amount of time until an id is reused, by ensuring they always
 669          * "increase" (mod INT_MAX):
 670          */
 671         return stid;
 672 out_free:
 673         kmem_cache_free(slab, stid);
 674         return NULL;
 675 }
 676
 677 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 678 {
 679         struct nfs4_stid *stid;
 680
 681         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 682         if (!stid)
 683                 return NULL;
 684
 685         return openlockstateid(stid);
 686 }
 687
 688 static void nfs4_free_deleg(struct nfs4_stid *stid)
 689 {
 690         kmem_cache_free(deleg_slab, stid);
 691         atomic_long_dec(&num_delegations);
 692 }
 693
 694 /*
 695  * When we recall a delegation, we should be careful not to hand it
 696  * out again straight away.
 697  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 698  * in which the filehandles of recalled delegations are "stored".
 699  * If a filehandle appear in either filter, a delegation is blocked.
 700  * When a delegation is recalled, the filehandle is stored in the "new"
 701  * filter.
 702  * Every 30 seconds we swap the filters and clear the "new" one,
 703  * unless both are empty of course.
 704  *
 705  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 706  * low 3 bytes as hash-table indices.
 707  *
 708  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 709  * is used to manage concurrent access.  Testing does not need the lock
 710  * except when swapping the two filters.
 711  */
 712 static DEFINE_SPINLOCK(blocked_delegations_lock);
 713 static struct bloom_pair {
 714         int     entries, old_entries;
 715         time_t  swap_time;
 716         int     new; /* index into 'set' */
 717         DECLARE_BITMAP(set[2], 256);
 718 } blocked_delegations;
 719
 720 static int delegation_blocked(struct knfsd_fh *fh)
 721 {
 722         u32 hash;
 723         struct bloom_pair *bd = &blocked_delegations;
 724
 725         if (bd->entries == 0)
 726                 return 0;
 727         if (seconds_since_boot() - bd->swap_time > 30) {
 728                 spin_lock(&blocked_delegations_lock);
 729                 if (seconds_since_boot() - bd->swap_time > 30) {
 730                         bd->entries -= bd->old_entries;
 731                         bd->old_entries = bd->entries;
 732                         memset(bd->set[bd->new], 0,
 733                                sizeof(bd->set[0]));
 734                         bd->new = 1-bd->new;
 735                         bd->swap_time = seconds_since_boot();
 736                 }
 737                 spin_unlock(&blocked_delegations_lock);
 738         }
 739         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 740         if (test_bit(hash&255, bd->set[0]) &&
 741             test_bit((hash>>8)&255, bd->set[0]) &&
 742             test_bit((hash>>16)&255, bd->set[0]))
 743                 return 1;
 744
 745         if (test_bit(hash&255, bd->set[1]) &&
 746             test_bit((hash>>8)&255, bd->set[1]) &&
 747             test_bit((hash>>16)&255, bd->set[1]))
 748                 return 1;
 749
 750         return 0;
 751 }
 752
 753 static void block_delegations(struct knfsd_fh *fh)
 754 {
 755         u32 hash;
 756         struct bloom_pair *bd = &blocked_delegations;
 757
 758         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 759
 760         spin_lock(&blocked_delegations_lock);
 761         __set_bit(hash&255, bd->set[bd->new]);
 762         __set_bit((hash>>8)&255, bd->set[bd->new]);
 763         __set_bit((hash>>16)&255, bd->set[bd->new]);
 764         if (bd->entries == 0)
 765                 bd->swap_time = seconds_since_boot();
 766         bd->entries += 1;
 767         spin_unlock(&blocked_delegations_lock);
 768 }
 769
 770 static struct nfs4_delegation *
 771 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
 772                  struct nfs4_clnt_odstate *odstate)
 773 {
 774         struct nfs4_delegation *dp;
 775         long n;
 776
 777         dprintk("NFSD alloc_init_deleg\n");
 778         n = atomic_long_inc_return(&num_delegations);
 779         if (n < 0 || n > max_delegations)
 780                 goto out_dec;
 781         if (delegation_blocked(&current_fh->fh_handle))
 782                 goto out_dec;
 783         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 784         if (dp == NULL)
 785                 goto out_dec;
 786
 787         /*
 788          * delegation seqid's are never incremented.  The 4.1 special
 789          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 790          * 0 anyway just for consistency and use 1:
 791          */
 792         dp->dl_stid.sc_stateid.si_generation = 1;
 793         INIT_LIST_HEAD(&dp->dl_perfile);
 794         INIT_LIST_HEAD(&dp->dl_perclnt);
 795         INIT_LIST_HEAD(&dp->dl_recall_lru);
 796         dp->dl_clnt_odstate = odstate;
 797         get_clnt_odstate(odstate);
 798         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 799         dp->dl_retries = 1;
 800         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 801                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 802         return dp;
 803 out_dec:
 804         atomic_long_dec(&num_delegations);
 805         return NULL;
 806 }
 807
 808 void
 809 nfs4_put_stid(struct nfs4_stid *s)
 810 {
 811         struct nfs4_file *fp = s->sc_file;
 812         struct nfs4_client *clp = s->sc_client;
 813
 814         might_lock(&clp->cl_lock);
 815
 816         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 817                 wake_up_all(&close_wq);
 818                 return;
 819         }
 820         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 821         spin_unlock(&clp->cl_lock);
 822         s->sc_free(s);
 823         if (fp)
 824                 put_nfs4_file(fp);
 825 }
 826
 827 void
 828 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 829 {
 830         stateid_t *src = &stid->sc_stateid;
 831
 832         spin_lock(&stid->sc_lock);
 833         if (unlikely(++src->si_generation == 0))
 834                 src->si_generation = 1;
 835         memcpy(dst, src, sizeof(*dst));
 836         spin_unlock(&stid->sc_lock);
 837 }
 838
 839 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 840 {
 841         struct file *filp = NULL;
 842
 843         spin_lock(&fp->fi_lock);
 844         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 845                 swap(filp, fp->fi_deleg_file);
 846         spin_unlock(&fp->fi_lock);
 847
 848         if (filp) {
 849                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 850                 fput(filp);
 851         }
 852 }
 853
 854 void nfs4_unhash_stid(struct nfs4_stid *s)
 855 {
 856         s->sc_type = 0;
 857 }
 858
 859 /**
 860  * nfs4_get_existing_delegation - Discover if this delegation already exists
 861  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 862  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 863  *
 864  * Return:
 865  *      On success: NULL if an existing delegation was not found.
 866  *
 867  *      On error: -EAGAIN if one was previously granted to this nfs4_client
 868  *                 for this nfs4_file.
 869  *
 870  */
 871
 872 static int
 873 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
 874 {
 875         struct nfs4_delegation *searchdp = NULL;
 876         struct nfs4_client *searchclp = NULL;
 877
 878         lockdep_assert_held(&state_lock);
 879         lockdep_assert_held(&fp->fi_lock);
 880
 881         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 882                 searchclp = searchdp->dl_stid.sc_client;
 883                 if (clp == searchclp) {
 884                         return -EAGAIN;
 885                 }
 886         }
 887         return 0;
 888 }
 889
 890 /**
 891  * hash_delegation_locked - Add a delegation to the appropriate lists
 892  * @dp:     a pointer to the nfs4_delegation we are adding.
 893  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 894  *
 895  * Return:
 896  *      On success: NULL if the delegation was successfully hashed.
 897  *
 898  *      On error: -EAGAIN if one was previously granted to this
 899  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 900  *
 901  */
 902
 903 static int
 904 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 905 {
 906         int status;
 907         struct nfs4_client *clp = dp->dl_stid.sc_client;
 908
 909         lockdep_assert_held(&state_lock);
 910         lockdep_assert_held(&fp->fi_lock);
 911
 912         status = nfs4_get_existing_delegation(clp, fp);
 913         if (status)
 914                 return status;
 915         ++fp->fi_delegees;
 916         atomic_inc(&dp->dl_stid.sc_count);
 917         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 918         list_add(&dp->dl_perfile, &fp->fi_delegations);
 919         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 920         return 0;
 921 }
 922
 923 static bool
 924 unhash_delegation_locked(struct nfs4_delegation *dp)
 925 {
 926         struct nfs4_file *fp = dp->dl_stid.sc_file;
 927
 928         lockdep_assert_held(&state_lock);
 929
 930         if (list_empty(&dp->dl_perfile))
 931                 return false;
 932
 933         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 934         /* Ensure that deleg break won't try to requeue it */
 935         ++dp->dl_time;
 936         spin_lock(&fp->fi_lock);
 937         list_del_init(&dp->dl_perclnt);
 938         list_del_init(&dp->dl_recall_lru);
 939         list_del_init(&dp->dl_perfile);
 940         spin_unlock(&fp->fi_lock);
 941         return true;
 942 }
 943
 944 static void destroy_delegation(struct nfs4_delegation *dp)
 945 {
 946         bool unhashed;
 947
 948         spin_lock(&state_lock);
 949         unhashed = unhash_delegation_locked(dp);
 950         spin_unlock(&state_lock);
 951         if (unhashed) {
 952                 put_clnt_odstate(dp->dl_clnt_odstate);
 953                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 954                 nfs4_put_stid(&dp->dl_stid);
 955         }
 956 }
 957
 958 static void revoke_delegation(struct nfs4_delegation *dp)
 959 {
 960         struct nfs4_client *clp = dp->dl_stid.sc_client;
 961
 962         WARN_ON(!list_empty(&dp->dl_recall_lru));
 963
 964         put_clnt_odstate(dp->dl_clnt_odstate);
 965         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 966
 967         if (clp->cl_minorversion == 0)
 968                 nfs4_put_stid(&dp->dl_stid);
 969         else {
 970                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 971                 spin_lock(&clp->cl_lock);
 972                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 973                 spin_unlock(&clp->cl_lock);
 974         }
 975 }
 976
 977 /*
 978  * SETCLIENTID state
 979  */
 980
 981 static unsigned int clientid_hashval(u32 id)
 982 {
 983         return id & CLIENT_HASH_MASK;
 984 }
 985
 986 static unsigned int clientstr_hashval(const char *name)
 987 {
 988         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 989 }
 990
 991 /*
 992  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 993  * st_{access,deny}_bmap field of the stateid, in order to track not
 994  * only what share bits are currently in force, but also what
 995  * combinations of share bits previous opens have used.  This allows us
 996  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 997  * return an error if the client attempt to downgrade to a combination
 998  * of share bits not explicable by closing some of its previous opens.
 999  *
1000  * XXX: This enforcement is actually incomplete, since we don't keep
1001  * track of access/deny bit combinations; so, e.g., we allow:
1002  *
1003  *      OPEN allow read, deny write
1004  *      OPEN allow both, deny none
1005  *      DOWNGRADE allow read, deny none
1006  *
1007  * which we should reject.
1008  */
1009 static unsigned int
1010 bmap_to_share_mode(unsigned long bmap) {
1011         int i;
1012         unsigned int access = 0;
1013
1014         for (i = 1; i < 4; i++) {
1015                 if (test_bit(i, &bmap))
1016                         access |= i;
1017         }
1018         return access;
1019 }
1020
1021 /* set share access for a given stateid */
1022 static inline void
1023 set_access(u32 access, struct nfs4_ol_stateid *stp)
1024 {
1025         unsigned char mask = 1 << access;
1026
1027         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1028         stp->st_access_bmap |= mask;
1029 }
1030
1031 /* clear share access for a given stateid */
1032 static inline void
1033 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1034 {
1035         unsigned char mask = 1 << access;
1036
1037         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1038         stp->st_access_bmap &= ~mask;
1039 }
1040
1041 /* test whether a given stateid has access */
1042 static inline bool
1043 test_access(u32 access, struct nfs4_ol_stateid *stp)
1044 {
1045         unsigned char mask = 1 << access;
1046
1047         return (bool)(stp->st_access_bmap & mask);
1048 }
1049
1050 /* set share deny for a given stateid */
1051 static inline void
1052 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1053 {
1054         unsigned char mask = 1 << deny;
1055
1056         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1057         stp->st_deny_bmap |= mask;
1058 }
1059
1060 /* clear share deny for a given stateid */
1061 static inline void
1062 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1063 {
1064         unsigned char mask = 1 << deny;
1065
1066         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1067         stp->st_deny_bmap &= ~mask;
1068 }
1069
1070 /* test whether a given stateid is denying specific access */
1071 static inline bool
1072 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1073 {
1074         unsigned char mask = 1 << deny;
1075
1076         return (bool)(stp->st_deny_bmap & mask);
1077 }
1078
1079 static int nfs4_access_to_omode(u32 access)
1080 {
1081         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1082         case NFS4_SHARE_ACCESS_READ:
1083                 return O_RDONLY;
1084         case NFS4_SHARE_ACCESS_WRITE:
1085                 return O_WRONLY;
1086         case NFS4_SHARE_ACCESS_BOTH:
1087                 return O_RDWR;
1088         }
1089         WARN_ON_ONCE(1);
1090         return O_RDONLY;
1091 }
1092
1093 /*
1094  * A stateid that had a deny mode associated with it is being released
1095  * or downgraded. Recalculate the deny mode on the file.
1096  */
1097 static void
1098 recalculate_deny_mode(struct nfs4_file *fp)
1099 {
1100         struct nfs4_ol_stateid *stp;
1101
1102         spin_lock(&fp->fi_lock);
1103         fp->fi_share_deny = 0;
1104         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1105                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1106         spin_unlock(&fp->fi_lock);
1107 }
1108
1109 static void
1110 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1111 {
1112         int i;
1113         bool change = false;
1114
1115         for (i = 1; i < 4; i++) {
1116                 if ((i & deny) != i) {
1117                         change = true;
1118                         clear_deny(i, stp);
1119                 }
1120         }
1121
1122         /* Recalculate per-file deny mode if there was a change */
1123         if (change)
1124                 recalculate_deny_mode(stp->st_stid.sc_file);
1125 }
1126
1127 /* release all access and file references for a given stateid */
1128 static void
1129 release_all_access(struct nfs4_ol_stateid *stp)
1130 {
1131         int i;
1132         struct nfs4_file *fp = stp->st_stid.sc_file;
1133
1134         if (fp && stp->st_deny_bmap != 0)
1135                 recalculate_deny_mode(fp);
1136
1137         for (i = 1; i < 4; i++) {
1138                 if (test_access(i, stp))
1139                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1140                 clear_access(i, stp);
1141         }
1142 }
1143
1144 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1145 {
1146         kfree(sop->so_owner.data);
1147         sop->so_ops->so_free(sop);
1148 }
1149
1150 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1151 {
1152         struct nfs4_client *clp = sop->so_client;
1153
1154         might_lock(&clp->cl_lock);
1155
1156         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1157                 return;
1158         sop->so_ops->so_unhash(sop);
1159         spin_unlock(&clp->cl_lock);
1160         nfs4_free_stateowner(sop);
1161 }
1162
1163 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1164 {
1165         struct nfs4_file *fp = stp->st_stid.sc_file;
1166
1167         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1168
1169         if (list_empty(&stp->st_perfile))
1170                 return false;
1171
1172         spin_lock(&fp->fi_lock);
1173         list_del_init(&stp->st_perfile);
1174         spin_unlock(&fp->fi_lock);
1175         list_del(&stp->st_perstateowner);
1176         return true;
1177 }
1178
1179 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1180 {
1181         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1182
1183         put_clnt_odstate(stp->st_clnt_odstate);
1184         release_all_access(stp);
1185         if (stp->st_stateowner)
1186                 nfs4_put_stateowner(stp->st_stateowner);
1187         kmem_cache_free(stateid_slab, stid);
1188 }
1189
1190 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1191 {
1192         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1193         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1194         struct file *file;
1195
1196         file = find_any_file(stp->st_stid.sc_file);
1197         if (file)
1198                 filp_close(file, (fl_owner_t)lo);
1199         nfs4_free_ol_stateid(stid);
1200 }
1201
1202 /*
1203  * Put the persistent reference to an already unhashed generic stateid, while
1204  * holding the cl_lock. If it's the last reference, then put it onto the
1205  * reaplist for later destruction.
1206  */
1207 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1208                                        struct list_head *reaplist)
1209 {
1210         struct nfs4_stid *s = &stp->st_stid;
1211         struct nfs4_client *clp = s->sc_client;
1212
1213         lockdep_assert_held(&clp->cl_lock);
1214
1215         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1216
1217         if (!atomic_dec_and_test(&s->sc_count)) {
1218                 wake_up_all(&close_wq);
1219                 return;
1220         }
1221
1222         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1223         list_add(&stp->st_locks, reaplist);
1224 }
1225
1226 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1227 {
1228         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1229
1230         list_del_init(&stp->st_locks);
1231         nfs4_unhash_stid(&stp->st_stid);
1232         return unhash_ol_stateid(stp);
1233 }
1234
1235 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1236 {
1237         struct nfs4_client *clp = stp->st_stid.sc_client;
1238         bool unhashed;
1239
1240         spin_lock(&clp->cl_lock);
1241         unhashed = unhash_lock_stateid(stp);
1242         spin_unlock(&clp->cl_lock);
1243         if (unhashed)
1244                 nfs4_put_stid(&stp->st_stid);
1245 }
1246
1247 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1248 {
1249         struct nfs4_client *clp = lo->lo_owner.so_client;
1250
1251         lockdep_assert_held(&clp->cl_lock);
1252
1253         list_del_init(&lo->lo_owner.so_strhash);
1254 }
1255
1256 /*
1257  * Free a list of generic stateids that were collected earlier after being
1258  * fully unhashed.
1259  */
1260 static void
1261 free_ol_stateid_reaplist(struct list_head *reaplist)
1262 {
1263         struct nfs4_ol_stateid *stp;
1264         struct nfs4_file *fp;
1265
1266         might_sleep();
1267
1268         while (!list_empty(reaplist)) {
1269                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1270                                        st_locks);
1271                 list_del(&stp->st_locks);
1272                 fp = stp->st_stid.sc_file;
1273                 stp->st_stid.sc_free(&stp->st_stid);
1274                 if (fp)
1275                         put_nfs4_file(fp);
1276         }
1277 }
1278
1279 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1280                                        struct list_head *reaplist)
1281 {
1282         struct nfs4_ol_stateid *stp;
1283
1284         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1285
1286         while (!list_empty(&open_stp->st_locks)) {
1287                 stp = list_entry(open_stp->st_locks.next,
1288                                 struct nfs4_ol_stateid, st_locks);
1289                 WARN_ON(!unhash_lock_stateid(stp));
1290                 put_ol_stateid_locked(stp, reaplist);
1291         }
1292 }
1293
1294 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1295                                 struct list_head *reaplist)
1296 {
1297         bool unhashed;
1298
1299         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1300
1301         unhashed = unhash_ol_stateid(stp);
1302         release_open_stateid_locks(stp, reaplist);
1303         return unhashed;
1304 }
1305
1306 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1307 {
1308         LIST_HEAD(reaplist);
1309
1310         spin_lock(&stp->st_stid.sc_client->cl_lock);
1311         if (unhash_open_stateid(stp, &reaplist))
1312                 put_ol_stateid_locked(stp, &reaplist);
1313         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1314         free_ol_stateid_reaplist(&reaplist);
1315 }
1316
1317 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1318 {
1319         struct nfs4_client *clp = oo->oo_owner.so_client;
1320
1321         lockdep_assert_held(&clp->cl_lock);
1322
1323         list_del_init(&oo->oo_owner.so_strhash);
1324         list_del_init(&oo->oo_perclient);
1325 }
1326
1327 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1328 {
1329         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1330                                           nfsd_net_id);
1331         struct nfs4_ol_stateid *s;
1332
1333         spin_lock(&nn->client_lock);
1334         s = oo->oo_last_closed_stid;
1335         if (s) {
1336                 list_del_init(&oo->oo_close_lru);
1337                 oo->oo_last_closed_stid = NULL;
1338         }
1339         spin_unlock(&nn->client_lock);
1340         if (s)
1341                 nfs4_put_stid(&s->st_stid);
1342 }
1343
1344 static void release_openowner(struct nfs4_openowner *oo)
1345 {
1346         struct nfs4_ol_stateid *stp;
1347         struct nfs4_client *clp = oo->oo_owner.so_client;
1348         struct list_head reaplist;
1349
1350         INIT_LIST_HEAD(&reaplist);
1351
1352         spin_lock(&clp->cl_lock);
1353         unhash_openowner_locked(oo);
1354         while (!list_empty(&oo->oo_owner.so_stateids)) {
1355                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1356                                 struct nfs4_ol_stateid, st_perstateowner);
1357                 if (unhash_open_stateid(stp, &reaplist))
1358                         put_ol_stateid_locked(stp, &reaplist);
1359         }
1360         spin_unlock(&clp->cl_lock);
1361         free_ol_stateid_reaplist(&reaplist);
1362         release_last_closed_stateid(oo);
1363         nfs4_put_stateowner(&oo->oo_owner);
1364 }
1365
1366 static inline int
1367 hash_sessionid(struct nfs4_sessionid *sessionid)
1368 {
1369         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1370
1371         return sid->sequence % SESSION_HASH_SIZE;
1372 }
1373
1374 #ifdef CONFIG_SUNRPC_DEBUG
1375 static inline void
1376 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1377 {
1378         u32 *ptr = (u32 *)(&sessionid->data[0]);
1379         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1380 }
1381 #else
1382 static inline void
1383 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1384 {
1385 }
1386 #endif
1387
1388 /*
1389  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1390  * won't be used for replay.
1391  */
1392 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1393 {
1394         struct nfs4_stateowner *so = cstate->replay_owner;
1395
1396         if (nfserr == nfserr_replay_me)
1397                 return;
1398
1399         if (!seqid_mutating_err(ntohl(nfserr))) {
1400                 nfsd4_cstate_clear_replay(cstate);
1401                 return;
1402         }
1403         if (!so)
1404                 return;
1405         if (so->so_is_open_owner)
1406                 release_last_closed_stateid(openowner(so));
1407         so->so_seqid++;
1408         return;
1409 }
1410
1411 static void
1412 gen_sessionid(struct nfsd4_session *ses)
1413 {
1414         struct nfs4_client *clp = ses->se_client;
1415         struct nfsd4_sessionid *sid;
1416
1417         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1418         sid->clientid = clp->cl_clientid;
1419         sid->sequence = current_sessionid++;
1420         sid->reserved = 0;
1421 }
1422
1423 /*
1424  * The protocol defines ca_maxresponssize_cached to include the size of
1425  * the rpc header, but all we need to cache is the data starting after
1426  * the end of the initial SEQUENCE operation--the rest we regenerate
1427  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1428  * value that is the number of bytes in our cache plus a few additional
1429  * bytes.  In order to stay on the safe side, and not promise more than
1430  * we can cache, those additional bytes must be the minimum possible: 24
1431  * bytes of rpc header (xid through accept state, with AUTH_NULL
1432  * verifier), 12 for the compound header (with zero-length tag), and 44
1433  * for the SEQUENCE op response:
1434  */
1435 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1436
1437 static void
1438 free_session_slots(struct nfsd4_session *ses)
1439 {
1440         int i;
1441
1442         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1443                 kfree(ses->se_slots[i]);
1444 }
1445
1446 /*
1447  * We don't actually need to cache the rpc and session headers, so we
1448  * can allocate a little less for each slot:
1449  */
1450 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1451 {
1452         u32 size;
1453
1454         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1455                 size = 0;
1456         else
1457                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1458         return size + sizeof(struct nfsd4_slot);
1459 }
1460
1461 /*
1462  * XXX: If we run out of reserved DRC memory we could (up to a point)
1463  * re-negotiate active sessions and reduce their slot usage to make
1464  * room for new connections. For now we just fail the create session.
1465  */
1466 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1467 {
1468         u32 slotsize = slot_bytes(ca);
1469         u32 num = ca->maxreqs;
1470         int avail;
1471
1472         spin_lock(&nfsd_drc_lock);
1473         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1474                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1475         num = min_t(int, num, avail / slotsize);
1476         nfsd_drc_mem_used += num * slotsize;
1477         spin_unlock(&nfsd_drc_lock);
1478
1479         return num;
1480 }
1481
1482 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1483 {
1484         int slotsize = slot_bytes(ca);
1485
1486         spin_lock(&nfsd_drc_lock);
1487         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1488         spin_unlock(&nfsd_drc_lock);
1489 }
1490
1491 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1492                                            struct nfsd4_channel_attrs *battrs)
1493 {
1494         int numslots = fattrs->maxreqs;
1495         int slotsize = slot_bytes(fattrs);
1496         struct nfsd4_session *new;
1497         int mem, i;
1498
1499         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1500                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1501         mem = numslots * sizeof(struct nfsd4_slot *);
1502
1503         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1504         if (!new)
1505                 return NULL;
1506         /* allocate each struct nfsd4_slot and data cache in one piece */
1507         for (i = 0; i < numslots; i++) {
1508                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1509                 if (!new->se_slots[i])
1510                         goto out_free;
1511         }
1512
1513         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1514         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1515
1516         return new;
1517 out_free:
1518         while (i--)
1519                 kfree(new->se_slots[i]);
1520         kfree(new);
1521         return NULL;
1522 }
1523
1524 static void free_conn(struct nfsd4_conn *c)
1525 {
1526         svc_xprt_put(c->cn_xprt);
1527         kfree(c);
1528 }
1529
1530 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1531 {
1532         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1533         struct nfs4_client *clp = c->cn_session->se_client;
1534
1535         spin_lock(&clp->cl_lock);
1536         if (!list_empty(&c->cn_persession)) {
1537                 list_del(&c->cn_persession);
1538                 free_conn(c);
1539         }
1540         nfsd4_probe_callback(clp);
1541         spin_unlock(&clp->cl_lock);
1542 }
1543
1544 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1545 {
1546         struct nfsd4_conn *conn;
1547
1548         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1549         if (!conn)
1550                 return NULL;
1551         svc_xprt_get(rqstp->rq_xprt);
1552         conn->cn_xprt = rqstp->rq_xprt;
1553         conn->cn_flags = flags;
1554         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1555         return conn;
1556 }
1557
1558 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1559 {
1560         conn->cn_session = ses;
1561         list_add(&conn->cn_persession, &ses->se_conns);
1562 }
1563
1564 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1565 {
1566         struct nfs4_client *clp = ses->se_client;
1567
1568         spin_lock(&clp->cl_lock);
1569         __nfsd4_hash_conn(conn, ses);
1570         spin_unlock(&clp->cl_lock);
1571 }
1572
1573 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1574 {
1575         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1576         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1577 }
1578
1579 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1580 {
1581         int ret;
1582
1583         nfsd4_hash_conn(conn, ses);
1584         ret = nfsd4_register_conn(conn);
1585         if (ret)
1586                 /* oops; xprt is already down: */
1587                 nfsd4_conn_lost(&conn->cn_xpt_user);
1588         /* We may have gained or lost a callback channel: */
1589         nfsd4_probe_callback_sync(ses->se_client);
1590 }
1591
1592 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1593 {
1594         u32 dir = NFS4_CDFC4_FORE;
1595
1596         if (cses->flags & SESSION4_BACK_CHAN)
1597                 dir |= NFS4_CDFC4_BACK;
1598         return alloc_conn(rqstp, dir);
1599 }
1600
1601 /* must be called under client_lock */
1602 static void nfsd4_del_conns(struct nfsd4_session *s)
1603 {
1604         struct nfs4_client *clp = s->se_client;
1605         struct nfsd4_conn *c;
1606
1607         spin_lock(&clp->cl_lock);
1608         while (!list_empty(&s->se_conns)) {
1609                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1610                 list_del_init(&c->cn_persession);
1611                 spin_unlock(&clp->cl_lock);
1612
1613                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1614                 free_conn(c);
1615
1616                 spin_lock(&clp->cl_lock);
1617         }
1618         spin_unlock(&clp->cl_lock);
1619 }
1620
1621 static void __free_session(struct nfsd4_session *ses)
1622 {
1623         free_session_slots(ses);
1624         kfree(ses);
1625 }
1626
1627 static void free_session(struct nfsd4_session *ses)
1628 {
1629         nfsd4_del_conns(ses);
1630         nfsd4_put_drc_mem(&ses->se_fchannel);
1631         __free_session(ses);
1632 }
1633
1634 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1635 {
1636         int idx;
1637         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1638
1639         new->se_client = clp;
1640         gen_sessionid(new);
1641
1642         INIT_LIST_HEAD(&new->se_conns);
1643
1644         new->se_cb_seq_nr = 1;
1645         new->se_flags = cses->flags;
1646         new->se_cb_prog = cses->callback_prog;
1647         new->se_cb_sec = cses->cb_sec;
1648         atomic_set(&new->se_ref, 0);
1649         idx = hash_sessionid(&new->se_sessionid);
1650         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1651         spin_lock(&clp->cl_lock);
1652         list_add(&new->se_perclnt, &clp->cl_sessions);
1653         spin_unlock(&clp->cl_lock);
1654
1655         {
1656                 struct sockaddr *sa = svc_addr(rqstp);
1657                 /*
1658                  * This is a little silly; with sessions there's no real
1659                  * use for the callback address.  Use the peer address
1660                  * as a reasonable default for now, but consider fixing
1661                  * the rpc client not to require an address in the
1662                  * future:
1663                  */
1664                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1665                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1666         }
1667 }
1668
1669 /* caller must hold client_lock */
1670 static struct nfsd4_session *
1671 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1672 {
1673         struct nfsd4_session *elem;
1674         int idx;
1675         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1676
1677         lockdep_assert_held(&nn->client_lock);
1678
1679         dump_sessionid(__func__, sessionid);
1680         idx = hash_sessionid(sessionid);
1681         /* Search in the appropriate list */
1682         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1683                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1684                             NFS4_MAX_SESSIONID_LEN)) {
1685                         return elem;
1686                 }
1687         }
1688
1689         dprintk("%s: session not found\n", __func__);
1690         return NULL;
1691 }
1692
1693 static struct nfsd4_session *
1694 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1695                 __be32 *ret)
1696 {
1697         struct nfsd4_session *session;
1698         __be32 status = nfserr_badsession;
1699
1700         session = __find_in_sessionid_hashtbl(sessionid, net);
1701         if (!session)
1702                 goto out;
1703         status = nfsd4_get_session_locked(session);
1704         if (status)
1705                 session = NULL;
1706 out:
1707         *ret = status;
1708         return session;
1709 }
1710
1711 /* caller must hold client_lock */
1712 static void
1713 unhash_session(struct nfsd4_session *ses)
1714 {
1715         struct nfs4_client *clp = ses->se_client;
1716         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1717
1718         lockdep_assert_held(&nn->client_lock);
1719
1720         list_del(&ses->se_hash);
1721         spin_lock(&ses->se_client->cl_lock);
1722         list_del(&ses->se_perclnt);
1723         spin_unlock(&ses->se_client->cl_lock);
1724 }
1725
1726 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1727 static int
1728 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1729 {
1730         /*
1731          * We're assuming the clid was not given out from a boot
1732          * precisely 2^32 (about 136 years) before this one.  That seems
1733          * a safe assumption:
1734          */
1735         if (clid->cl_boot == (u32)nn->boot_time)
1736                 return 0;
1737         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1738                 clid->cl_boot, clid->cl_id, nn->boot_time);
1739         return 1;
1740 }
1741
1742 /*
1743  * XXX Should we use a slab cache ?
1744  * This type of memory management is somewhat inefficient, but we use it
1745  * anyway since SETCLIENTID is not a common operation.
1746  */
1747 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1748 {
1749         struct nfs4_client *clp;
1750         int i;
1751
1752         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1753         if (clp == NULL)
1754                 return NULL;
1755         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1756         if (clp->cl_name.data == NULL)
1757                 goto err_no_name;
1758         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1759                         OWNER_HASH_SIZE, GFP_KERNEL);
1760         if (!clp->cl_ownerstr_hashtbl)
1761                 goto err_no_hashtbl;
1762         for (i = 0; i < OWNER_HASH_SIZE; i++)
1763                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1764         clp->cl_name.len = name.len;
1765         INIT_LIST_HEAD(&clp->cl_sessions);
1766         idr_init(&clp->cl_stateids);
1767         atomic_set(&clp->cl_refcount, 0);
1768         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1769         INIT_LIST_HEAD(&clp->cl_idhash);
1770         INIT_LIST_HEAD(&clp->cl_openowners);
1771         INIT_LIST_HEAD(&clp->cl_delegations);
1772         INIT_LIST_HEAD(&clp->cl_lru);
1773         INIT_LIST_HEAD(&clp->cl_revoked);
1774 #ifdef CONFIG_NFSD_PNFS
1775         INIT_LIST_HEAD(&clp->cl_lo_states);
1776 #endif
1777         spin_lock_init(&clp->cl_lock);
1778         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1779         return clp;
1780 err_no_hashtbl:
1781         kfree(clp->cl_name.data);
1782 err_no_name:
1783         kfree(clp);
1784         return NULL;
1785 }
1786
1787 static void
1788 free_client(struct nfs4_client *clp)
1789 {
1790         while (!list_empty(&clp->cl_sessions)) {
1791                 struct nfsd4_session *ses;
1792                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1793                                 se_perclnt);
1794                 list_del(&ses->se_perclnt);
1795                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1796                 free_session(ses);
1797         }
1798         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1799         free_svc_cred(&clp->cl_cred);
1800         kfree(clp->cl_ownerstr_hashtbl);
1801         kfree(clp->cl_name.data);
1802         idr_destroy(&clp->cl_stateids);
1803         kfree(clp);
1804 }
1805
1806 /* must be called under the client_lock */
1807 static void
1808 unhash_client_locked(struct nfs4_client *clp)
1809 {
1810         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1811         struct nfsd4_session *ses;
1812
1813         lockdep_assert_held(&nn->client_lock);
1814
1815         /* Mark the client as expired! */
1816         clp->cl_time = 0;
1817         /* Make it invisible */
1818         if (!list_empty(&clp->cl_idhash)) {
1819                 list_del_init(&clp->cl_idhash);
1820                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1821                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1822                 else
1823                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1824         }
1825         list_del_init(&clp->cl_lru);
1826         spin_lock(&clp->cl_lock);
1827         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1828                 list_del_init(&ses->se_hash);
1829         spin_unlock(&clp->cl_lock);
1830 }
1831
1832 static void
1833 unhash_client(struct nfs4_client *clp)
1834 {
1835         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1836
1837         spin_lock(&nn->client_lock);
1838         unhash_client_locked(clp);
1839         spin_unlock(&nn->client_lock);
1840 }
1841
1842 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1843 {
1844         if (atomic_read(&clp->cl_refcount))
1845                 return nfserr_jukebox;
1846         unhash_client_locked(clp);
1847         return nfs_ok;
1848 }
1849
1850 static void
1851 __destroy_client(struct nfs4_client *clp)
1852 {
1853         struct nfs4_openowner *oo;
1854         struct nfs4_delegation *dp;
1855         struct list_head reaplist;
1856
1857         INIT_LIST_HEAD(&reaplist);
1858         spin_lock(&state_lock);
1859         while (!list_empty(&clp->cl_delegations)) {
1860                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1861                 WARN_ON(!unhash_delegation_locked(dp));
1862                 list_add(&dp->dl_recall_lru, &reaplist);
1863         }
1864         spin_unlock(&state_lock);
1865         while (!list_empty(&reaplist)) {
1866                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1867                 list_del_init(&dp->dl_recall_lru);
1868                 put_clnt_odstate(dp->dl_clnt_odstate);
1869                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1870                 nfs4_put_stid(&dp->dl_stid);
1871         }
1872         while (!list_empty(&clp->cl_revoked)) {
1873                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1874                 list_del_init(&dp->dl_recall_lru);
1875                 nfs4_put_stid(&dp->dl_stid);
1876         }
1877         while (!list_empty(&clp->cl_openowners)) {
1878                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1879                 nfs4_get_stateowner(&oo->oo_owner);
1880                 release_openowner(oo);
1881         }
1882         nfsd4_return_all_client_layouts(clp);
1883         nfsd4_shutdown_callback(clp);
1884         if (clp->cl_cb_conn.cb_xprt)
1885                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1886         free_client(clp);
1887 }
1888
1889 static void
1890 destroy_client(struct nfs4_client *clp)
1891 {
1892         unhash_client(clp);
1893         __destroy_client(clp);
1894 }
1895
1896 static void expire_client(struct nfs4_client *clp)
1897 {
1898         unhash_client(clp);
1899         nfsd4_client_record_remove(clp);
1900         __destroy_client(clp);
1901 }
1902
1903 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1904 {
1905         memcpy(target->cl_verifier.data, source->data,
1906                         sizeof(target->cl_verifier.data));
1907 }
1908
1909 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1910 {
1911         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1912         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1913 }
1914
1915 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1916 {
1917         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1918         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1919                                                                 GFP_KERNEL);
1920         if ((source->cr_principal && ! target->cr_principal) ||
1921             (source->cr_raw_principal && ! target->cr_raw_principal))
1922                 return -ENOMEM;
1923
1924         target->cr_flavor = source->cr_flavor;
1925         target->cr_uid = source->cr_uid;
1926         target->cr_gid = source->cr_gid;
1927         target->cr_group_info = source->cr_group_info;
1928         get_group_info(target->cr_group_info);
1929         target->cr_gss_mech = source->cr_gss_mech;
1930         if (source->cr_gss_mech)
1931                 gss_mech_get(source->cr_gss_mech);
1932         return 0;
1933 }
1934
1935 static int
1936 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1937 {
1938         if (o1->len < o2->len)
1939                 return -1;
1940         if (o1->len > o2->len)
1941                 return 1;
1942         return memcmp(o1->data, o2->data, o1->len);
1943 }
1944
1945 static int same_name(const char *n1, const char *n2)
1946 {
1947         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1948 }
1949
1950 static int
1951 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1952 {
1953         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1954 }
1955
1956 static int
1957 same_clid(clientid_t *cl1, clientid_t *cl2)
1958 {
1959         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1960 }
1961
1962 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1963 {
1964         int i;
1965
1966         if (g1->ngroups != g2->ngroups)
1967                 return false;
1968         for (i=0; i<g1->ngroups; i++)
1969                 if (!gid_eq(g1->gid[i], g2->gid[i]))
1970                         return false;
1971         return true;
1972 }
1973
1974 /*
1975  * RFC 3530 language requires clid_inuse be returned when the
1976  * "principal" associated with a requests differs from that previously
1977  * used.  We use uid, gid's, and gss principal string as our best
1978  * approximation.  We also don't want to allow non-gss use of a client
1979  * established using gss: in theory cr_principal should catch that
1980  * change, but in practice cr_principal can be null even in the gss case
1981  * since gssd doesn't always pass down a principal string.
1982  */
1983 static bool is_gss_cred(struct svc_cred *cr)
1984 {
1985         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1986         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1987 }
1988
1989
1990 static bool
1991 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1992 {
1993         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1994                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1995                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1996                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1997                 return false;
1998         if (cr1->cr_principal == cr2->cr_principal)
1999                 return true;
2000         if (!cr1->cr_principal || !cr2->cr_principal)
2001                 return false;
2002         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2003 }
2004
2005 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2006 {
2007         struct svc_cred *cr = &rqstp->rq_cred;
2008         u32 service;
2009
2010         if (!cr->cr_gss_mech)
2011                 return false;
2012         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2013         return service == RPC_GSS_SVC_INTEGRITY ||
2014                service == RPC_GSS_SVC_PRIVACY;
2015 }
2016
2017 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2018 {
2019         struct svc_cred *cr = &rqstp->rq_cred;
2020
2021         if (!cl->cl_mach_cred)
2022                 return true;
2023         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2024                 return false;
2025         if (!svc_rqst_integrity_protected(rqstp))
2026                 return false;
2027         if (cl->cl_cred.cr_raw_principal)
2028                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2029                                                 cr->cr_raw_principal);
2030         if (!cr->cr_principal)
2031                 return false;
2032         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2033 }
2034
2035 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2036 {
2037         __be32 verf[2];
2038
2039         /*
2040          * This is opaque to client, so no need to byte-swap. Use
2041          * __force to keep sparse happy
2042          */
2043         verf[0] = (__force __be32)get_seconds();
2044         verf[1] = (__force __be32)nn->clverifier_counter++;
2045         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2046 }
2047
2048 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2049 {
2050         clp->cl_clientid.cl_boot = nn->boot_time;
2051         clp->cl_clientid.cl_id = nn->clientid_counter++;
2052         gen_confirm(clp, nn);
2053 }
2054
2055 static struct nfs4_stid *
2056 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2057 {
2058         struct nfs4_stid *ret;
2059
2060         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2061         if (!ret || !ret->sc_type)
2062                 return NULL;
2063         return ret;
2064 }
2065
2066 static struct nfs4_stid *
2067 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2068 {
2069         struct nfs4_stid *s;
2070
2071         spin_lock(&cl->cl_lock);
2072         s = find_stateid_locked(cl, t);
2073         if (s != NULL) {
2074                 if (typemask & s->sc_type)
2075                         atomic_inc(&s->sc_count);
2076                 else
2077                         s = NULL;
2078         }
2079         spin_unlock(&cl->cl_lock);
2080         return s;
2081 }
2082
2083 static struct nfs4_client *create_client(struct xdr_netobj name,
2084                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2085 {
2086         struct nfs4_client *clp;
2087         struct sockaddr *sa = svc_addr(rqstp);
2088         int ret;
2089         struct net *net = SVC_NET(rqstp);
2090
2091         clp = alloc_client(name);
2092         if (clp == NULL)
2093                 return NULL;
2094
2095         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2096         if (ret) {
2097                 free_client(clp);
2098                 return NULL;
2099         }
2100         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2101         clp->cl_time = get_seconds();
2102         clear_bit(0, &clp->cl_cb_slot_busy);
2103         copy_verf(clp, verf);
2104         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2105         clp->cl_cb_session = NULL;
2106         clp->net = net;
2107         return clp;
2108 }
2109
2110 static void
2111 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2112 {
2113         struct rb_node **new = &(root->rb_node), *parent = NULL;
2114         struct nfs4_client *clp;
2115
2116         while (*new) {
2117                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2118                 parent = *new;
2119
2120                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2121                         new = &((*new)->rb_left);
2122                 else
2123                         new = &((*new)->rb_right);
2124         }
2125
2126         rb_link_node(&new_clp->cl_namenode, parent, new);
2127         rb_insert_color(&new_clp->cl_namenode, root);
2128 }
2129
2130 static struct nfs4_client *
2131 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2132 {
2133         int cmp;
2134         struct rb_node *node = root->rb_node;
2135         struct nfs4_client *clp;
2136
2137         while (node) {
2138                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2139                 cmp = compare_blob(&clp->cl_name, name);
2140                 if (cmp > 0)
2141                         node = node->rb_left;
2142                 else if (cmp < 0)
2143                         node = node->rb_right;
2144                 else
2145                         return clp;
2146         }
2147         return NULL;
2148 }
2149
2150 static void
2151 add_to_unconfirmed(struct nfs4_client *clp)
2152 {
2153         unsigned int idhashval;
2154         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2155
2156         lockdep_assert_held(&nn->client_lock);
2157
2158         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2159         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2160         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2161         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2162         renew_client_locked(clp);
2163 }
2164
2165 static void
2166 move_to_confirmed(struct nfs4_client *clp)
2167 {
2168         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2170
2171         lockdep_assert_held(&nn->client_lock);
2172
2173         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2174         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2175         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2176         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2177         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2178         renew_client_locked(clp);
2179 }
2180
2181 static struct nfs4_client *
2182 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2183 {
2184         struct nfs4_client *clp;
2185         unsigned int idhashval = clientid_hashval(clid->cl_id);
2186
2187         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2188                 if (same_clid(&clp->cl_clientid, clid)) {
2189                         if ((bool)clp->cl_minorversion != sessions)
2190                                 return NULL;
2191                         renew_client_locked(clp);
2192                         return clp;
2193                 }
2194         }
2195         return NULL;
2196 }
2197
2198 static struct nfs4_client *
2199 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2200 {
2201         struct list_head *tbl = nn->conf_id_hashtbl;
2202
2203         lockdep_assert_held(&nn->client_lock);
2204         return find_client_in_id_table(tbl, clid, sessions);
2205 }
2206
2207 static struct nfs4_client *
2208 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2209 {
2210         struct list_head *tbl = nn->unconf_id_hashtbl;
2211
2212         lockdep_assert_held(&nn->client_lock);
2213         return find_client_in_id_table(tbl, clid, sessions);
2214 }
2215
2216 static bool clp_used_exchangeid(struct nfs4_client *clp)
2217 {
2218         return clp->cl_exchange_flags != 0;
2219 }
2220
2221 static struct nfs4_client *
2222 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2223 {
2224         lockdep_assert_held(&nn->client_lock);
2225         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2226 }
2227
2228 static struct nfs4_client *
2229 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2230 {
2231         lockdep_assert_held(&nn->client_lock);
2232         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2233 }
2234
2235 static void
2236 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2237 {
2238         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2239         struct sockaddr *sa = svc_addr(rqstp);
2240         u32 scopeid = rpc_get_scope_id(sa);
2241         unsigned short expected_family;
2242
2243         /* Currently, we only support tcp and tcp6 for the callback channel */
2244         if (se->se_callback_netid_len == 3 &&
2245             !memcmp(se->se_callback_netid_val, "tcp", 3))
2246                 expected_family = AF_INET;
2247         else if (se->se_callback_netid_len == 4 &&
2248                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2249                 expected_family = AF_INET6;
2250         else
2251                 goto out_err;
2252
2253         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2254                                             se->se_callback_addr_len,
2255                                             (struct sockaddr *)&conn->cb_addr,
2256                                             sizeof(conn->cb_addr));
2257
2258         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2259                 goto out_err;
2260
2261         if (conn->cb_addr.ss_family == AF_INET6)
2262                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2263
2264         conn->cb_prog = se->se_callback_prog;
2265         conn->cb_ident = se->se_callback_ident;
2266         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2267         return;
2268 out_err:
2269         conn->cb_addr.ss_family = AF_UNSPEC;
2270         conn->cb_addrlen = 0;
2271         dprintk("NFSD: this client (clientid %08x/%08x) "
2272                 "will not receive delegations\n",
2273                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2274
2275         return;
2276 }
2277
2278 /*
2279  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2280  */
2281 static void
2282 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2283 {
2284         struct xdr_buf *buf = resp->xdr.buf;
2285         struct nfsd4_slot *slot = resp->cstate.slot;
2286         unsigned int base;
2287
2288         dprintk("--> %s slot %p\n", __func__, slot);
2289
2290         slot->sl_opcnt = resp->opcnt;
2291         slot->sl_status = resp->cstate.status;
2292
2293         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2294         if (nfsd4_not_cached(resp)) {
2295                 slot->sl_datalen = 0;
2296                 return;
2297         }
2298         base = resp->cstate.data_offset;
2299         slot->sl_datalen = buf->len - base;
2300         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2301                 WARN(1, "%s: sessions DRC could not cache compound\n",
2302                      __func__);
2303         return;
2304 }
2305
2306 /*
2307  * Encode the replay sequence operation from the slot values.
2308  * If cachethis is FALSE encode the uncached rep error on the next
2309  * operation which sets resp->p and increments resp->opcnt for
2310  * nfs4svc_encode_compoundres.
2311  *
2312  */
2313 static __be32
2314 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2315                           struct nfsd4_compoundres *resp)
2316 {
2317         struct nfsd4_op *op;
2318         struct nfsd4_slot *slot = resp->cstate.slot;
2319
2320         /* Encode the replayed sequence operation */
2321         op = &args->ops[resp->opcnt - 1];
2322         nfsd4_encode_operation(resp, op);
2323
2324         /* Return nfserr_retry_uncached_rep in next operation. */
2325         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2326                 op = &args->ops[resp->opcnt++];
2327                 op->status = nfserr_retry_uncached_rep;
2328                 nfsd4_encode_operation(resp, op);
2329         }
2330         return op->status;
2331 }
2332
2333 /*
2334  * The sequence operation is not cached because we can use the slot and
2335  * session values.
2336  */
2337 static __be32
2338 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2339                          struct nfsd4_sequence *seq)
2340 {
2341         struct nfsd4_slot *slot = resp->cstate.slot;
2342         struct xdr_stream *xdr = &resp->xdr;
2343         __be32 *p;
2344         __be32 status;
2345
2346         dprintk("--> %s slot %p\n", __func__, slot);
2347
2348         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2349         if (status)
2350                 return status;
2351
2352         p = xdr_reserve_space(xdr, slot->sl_datalen);
2353         if (!p) {
2354                 WARN_ON_ONCE(1);
2355                 return nfserr_serverfault;
2356         }
2357         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2358         xdr_commit_encode(xdr);
2359
2360         resp->opcnt = slot->sl_opcnt;
2361         return slot->sl_status;
2362 }
2363
2364 /*
2365  * Set the exchange_id flags returned by the server.
2366  */
2367 static void
2368 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2369 {
2370 #ifdef CONFIG_NFSD_PNFS
2371         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2372 #else
2373         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2374 #endif
2375
2376         /* Referrals are supported, Migration is not. */
2377         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2378
2379         /* set the wire flags to return to client. */
2380         clid->flags = new->cl_exchange_flags;
2381 }
2382
2383 static bool client_has_openowners(struct nfs4_client *clp)
2384 {
2385         struct nfs4_openowner *oo;
2386
2387         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2388                 if (!list_empty(&oo->oo_owner.so_stateids))
2389                         return true;
2390         }
2391         return false;
2392 }
2393
2394 static bool client_has_state(struct nfs4_client *clp)
2395 {
2396         return client_has_openowners(clp)
2397 #ifdef CONFIG_NFSD_PNFS
2398                 || !list_empty(&clp->cl_lo_states)
2399 #endif
2400                 || !list_empty(&clp->cl_delegations)
2401                 || !list_empty(&clp->cl_sessions);
2402 }
2403
2404 __be32
2405 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2406                 union nfsd4_op_u *u)
2407 {
2408         struct nfsd4_exchange_id *exid = &u->exchange_id;
2409         struct nfs4_client *conf, *new;
2410         struct nfs4_client *unconf = NULL;
2411         __be32 status;
2412         char                    addr_str[INET6_ADDRSTRLEN];
2413         nfs4_verifier           verf = exid->verifier;
2414         struct sockaddr         *sa = svc_addr(rqstp);
2415         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2416         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2417
2418         rpc_ntop(sa, addr_str, sizeof(addr_str));
2419         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2420                 "ip_addr=%s flags %x, spa_how %d\n",
2421                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2422                 addr_str, exid->flags, exid->spa_how);
2423
2424         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2425                 return nfserr_inval;
2426
2427         new = create_client(exid->clname, rqstp, &verf);
2428         if (new == NULL)
2429                 return nfserr_jukebox;
2430
2431         switch (exid->spa_how) {
2432         case SP4_MACH_CRED:
2433                 exid->spo_must_enforce[0] = 0;
2434                 exid->spo_must_enforce[1] = (
2435                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2436                         1 << (OP_EXCHANGE_ID - 32) |
2437                         1 << (OP_CREATE_SESSION - 32) |
2438                         1 << (OP_DESTROY_SESSION - 32) |
2439                         1 << (OP_DESTROY_CLIENTID - 32));
2440
2441                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2442                                         1 << (OP_OPEN_DOWNGRADE) |
2443                                         1 << (OP_LOCKU) |
2444                                         1 << (OP_DELEGRETURN));
2445
2446                 exid->spo_must_allow[1] &= (
2447                                         1 << (OP_TEST_STATEID - 32) |
2448                                         1 << (OP_FREE_STATEID - 32));
2449                 if (!svc_rqst_integrity_protected(rqstp)) {
2450                         status = nfserr_inval;
2451                         goto out_nolock;
2452                 }
2453                 /*
2454                  * Sometimes userspace doesn't give us a principal.
2455                  * Which is a bug, really.  Anyway, we can't enforce
2456                  * MACH_CRED in that case, better to give up now:
2457                  */
2458                 if (!new->cl_cred.cr_principal &&
2459                                         !new->cl_cred.cr_raw_principal) {
2460                         status = nfserr_serverfault;
2461                         goto out_nolock;
2462                 }
2463                 new->cl_mach_cred = true;
2464         case SP4_NONE:
2465                 break;
2466         default:                                /* checked by xdr code */
2467                 WARN_ON_ONCE(1);
2468         case SP4_SSV:
2469                 status = nfserr_encr_alg_unsupp;
2470                 goto out_nolock;
2471         }
2472
2473         /* Cases below refer to rfc 5661 section 18.35.4: */
2474         spin_lock(&nn->client_lock);
2475         conf = find_confirmed_client_by_name(&exid->clname, nn);
2476         if (conf) {
2477                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2478                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2479
2480                 if (update) {
2481                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2482                                 status = nfserr_inval;
2483                                 goto out;
2484                         }
2485                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2486                                 status = nfserr_wrong_cred;
2487                                 goto out;
2488                         }
2489                         if (!creds_match) { /* case 9 */
2490                                 status = nfserr_perm;
2491                                 goto out;
2492                         }
2493                         if (!verfs_match) { /* case 8 */
2494                                 status = nfserr_not_same;
2495                                 goto out;
2496                         }
2497                         /* case 6 */
2498                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2499                         goto out_copy;
2500                 }
2501                 if (!creds_match) { /* case 3 */
2502                         if (client_has_state(conf)) {
2503                                 status = nfserr_clid_inuse;
2504                                 goto out;
2505                         }
2506                         goto out_new;
2507                 }
2508                 if (verfs_match) { /* case 2 */
2509                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2510                         goto out_copy;
2511                 }
2512                 /* case 5, client reboot */
2513                 conf = NULL;
2514                 goto out_new;
2515         }
2516
2517         if (update) { /* case 7 */
2518                 status = nfserr_noent;
2519                 goto out;
2520         }
2521
2522         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2523         if (unconf) /* case 4, possible retry or client restart */
2524                 unhash_client_locked(unconf);
2525
2526         /* case 1 (normal case) */
2527 out_new:
2528         if (conf) {
2529                 status = mark_client_expired_locked(conf);
2530                 if (status)
2531                         goto out;
2532         }
2533         new->cl_minorversion = cstate->minorversion;
2534         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2535         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2536
2537         gen_clid(new, nn);
2538         add_to_unconfirmed(new);
2539         swap(new, conf);
2540 out_copy:
2541         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2542         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2543
2544         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2545         nfsd4_set_ex_flags(conf, exid);
2546
2547         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2548                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2549         status = nfs_ok;
2550
2551 out:
2552         spin_unlock(&nn->client_lock);
2553 out_nolock:
2554         if (new)
2555                 expire_client(new);
2556         if (unconf)
2557                 expire_client(unconf);
2558         return status;
2559 }
2560
2561 static __be32
2562 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2563 {
2564         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2565                 slot_seqid);
2566
2567         /* The slot is in use, and no response has been sent. */
2568         if (slot_inuse) {
2569                 if (seqid == slot_seqid)
2570                         return nfserr_jukebox;
2571                 else
2572                         return nfserr_seq_misordered;
2573         }
2574         /* Note unsigned 32-bit arithmetic handles wraparound: */
2575         if (likely(seqid == slot_seqid + 1))
2576                 return nfs_ok;
2577         if (seqid == slot_seqid)
2578                 return nfserr_replay_cache;
2579         return nfserr_seq_misordered;
2580 }
2581
2582 /*
2583  * Cache the create session result into the create session single DRC
2584  * slot cache by saving the xdr structure. sl_seqid has been set.
2585  * Do this for solo or embedded create session operations.
2586  */
2587 static void
2588 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2589                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2590 {
2591         slot->sl_status = nfserr;
2592         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2593 }
2594
2595 static __be32
2596 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2597                             struct nfsd4_clid_slot *slot)
2598 {
2599         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2600         return slot->sl_status;
2601 }
2602
2603 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2604                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2605                         1 +     /* MIN tag is length with zero, only length */ \
2606                         3 +     /* version, opcount, opcode */ \
2607                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2608                                 /* seqid, slotID, slotID, cache */ \
2609                         4 ) * sizeof(__be32))
2610
2611 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2612                         2 +     /* verifier: AUTH_NULL, length 0 */\
2613                         1 +     /* status */ \
2614                         1 +     /* MIN tag is length with zero, only length */ \
2615                         3 +     /* opcount, opcode, opstatus*/ \
2616                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2617                                 /* seqid, slotID, slotID, slotID, status */ \
2618                         5 ) * sizeof(__be32))
2619
2620 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2621 {
2622         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2623
2624         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2625                 return nfserr_toosmall;
2626         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2627                 return nfserr_toosmall;
2628         ca->headerpadsz = 0;
2629         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2630         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2631         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2632         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2633                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2634         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2635         /*
2636          * Note decreasing slot size below client's request may make it
2637          * difficult for client to function correctly, whereas
2638          * decreasing the number of slots will (just?) affect
2639          * performance.  When short on memory we therefore prefer to
2640          * decrease number of slots instead of their size.  Clients that
2641          * request larger slots than they need will get poor results:
2642          */
2643         ca->maxreqs = nfsd4_get_drc_mem(ca);
2644         if (!ca->maxreqs)
2645                 return nfserr_jukebox;
2646
2647         return nfs_ok;
2648 }
2649
2650 /*
2651  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2652  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2653  */
2654 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2655         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2656
2657 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2658         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2659
2660 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2661                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2662 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2663                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2664                                  sizeof(__be32))
2665
2666 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2667 {
2668         ca->headerpadsz = 0;
2669
2670         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2671                 return nfserr_toosmall;
2672         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2673                 return nfserr_toosmall;
2674         ca->maxresp_cached = 0;
2675         if (ca->maxops < 2)
2676                 return nfserr_toosmall;
2677
2678         return nfs_ok;
2679 }
2680
2681 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2682 {
2683         switch (cbs->flavor) {
2684         case RPC_AUTH_NULL:
2685         case RPC_AUTH_UNIX:
2686                 return nfs_ok;
2687         default:
2688                 /*
2689                  * GSS case: the spec doesn't allow us to return this
2690                  * error.  But it also doesn't allow us not to support
2691                  * GSS.
2692                  * I'd rather this fail hard than return some error the
2693                  * client might think it can already handle:
2694                  */
2695                 return nfserr_encr_alg_unsupp;
2696         }
2697 }
2698
2699 __be32
2700 nfsd4_create_session(struct svc_rqst *rqstp,
2701                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2702 {
2703         struct nfsd4_create_session *cr_ses = &u->create_session;
2704         struct sockaddr *sa = svc_addr(rqstp);
2705         struct nfs4_client *conf, *unconf;
2706         struct nfs4_client *old = NULL;
2707         struct nfsd4_session *new;
2708         struct nfsd4_conn *conn;
2709         struct nfsd4_clid_slot *cs_slot = NULL;
2710         __be32 status = 0;
2711         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2712
2713         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2714                 return nfserr_inval;
2715         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2716         if (status)
2717                 return status;
2718         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2719         if (status)
2720                 return status;
2721         status = check_backchannel_attrs(&cr_ses->back_channel);
2722         if (status)
2723                 goto out_release_drc_mem;
2724         status = nfserr_jukebox;
2725         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2726         if (!new)
2727                 goto out_release_drc_mem;
2728         conn = alloc_conn_from_crses(rqstp, cr_ses);
2729         if (!conn)
2730                 goto out_free_session;
2731
2732         spin_lock(&nn->client_lock);
2733         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2734         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2735         WARN_ON_ONCE(conf && unconf);
2736
2737         if (conf) {
2738                 status = nfserr_wrong_cred;
2739                 if (!nfsd4_mach_creds_match(conf, rqstp))
2740                         goto out_free_conn;
2741                 cs_slot = &conf->cl_cs_slot;
2742                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2743                 if (status) {
2744                         if (status == nfserr_replay_cache)
2745                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2746                         goto out_free_conn;
2747                 }
2748         } else if (unconf) {
2749                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2750                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2751                         status = nfserr_clid_inuse;
2752                         goto out_free_conn;
2753                 }
2754                 status = nfserr_wrong_cred;
2755                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2756                         goto out_free_conn;
2757                 cs_slot = &unconf->cl_cs_slot;
2758                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2759                 if (status) {
2760                         /* an unconfirmed replay returns misordered */
2761                         status = nfserr_seq_misordered;
2762                         goto out_free_conn;
2763                 }
2764                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2765                 if (old) {
2766                         status = mark_client_expired_locked(old);
2767                         if (status) {
2768                                 old = NULL;
2769                                 goto out_free_conn;
2770                         }
2771                 }
2772                 move_to_confirmed(unconf);
2773                 conf = unconf;
2774         } else {
2775                 status = nfserr_stale_clientid;
2776                 goto out_free_conn;
2777         }
2778         status = nfs_ok;
2779         /* Persistent sessions are not supported */
2780         cr_ses->flags &= ~SESSION4_PERSIST;
2781         /* Upshifting from TCP to RDMA is not supported */
2782         cr_ses->flags &= ~SESSION4_RDMA;
2783
2784         init_session(rqstp, new, conf, cr_ses);
2785         nfsd4_get_session_locked(new);
2786
2787         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2788                NFS4_MAX_SESSIONID_LEN);
2789         cs_slot->sl_seqid++;
2790         cr_ses->seqid = cs_slot->sl_seqid;
2791
2792         /* cache solo and embedded create sessions under the client_lock */
2793         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2794         spin_unlock(&nn->client_lock);
2795         /* init connection and backchannel */
2796         nfsd4_init_conn(rqstp, conn, new);
2797         nfsd4_put_session(new);
2798         if (old)
2799                 expire_client(old);
2800         return status;
2801 out_free_conn:
2802         spin_unlock(&nn->client_lock);
2803         free_conn(conn);
2804         if (old)
2805                 expire_client(old);
2806 out_free_session:
2807         __free_session(new);
2808 out_release_drc_mem:
2809         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2810         return status;
2811 }
2812
2813 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2814 {
2815         switch (*dir) {
2816         case NFS4_CDFC4_FORE:
2817         case NFS4_CDFC4_BACK:
2818                 return nfs_ok;
2819         case NFS4_CDFC4_FORE_OR_BOTH:
2820         case NFS4_CDFC4_BACK_OR_BOTH:
2821                 *dir = NFS4_CDFC4_BOTH;
2822                 return nfs_ok;
2823         };
2824         return nfserr_inval;
2825 }
2826
2827 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2828                 struct nfsd4_compound_state *cstate,
2829                 union nfsd4_op_u *u)
2830 {
2831         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2832         struct nfsd4_session *session = cstate->session;
2833         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2834         __be32 status;
2835
2836         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2837         if (status)
2838                 return status;
2839         spin_lock(&nn->client_lock);
2840         session->se_cb_prog = bc->bc_cb_program;
2841         session->se_cb_sec = bc->bc_cb_sec;
2842         spin_unlock(&nn->client_lock);
2843
2844         nfsd4_probe_callback(session->se_client);
2845
2846         return nfs_ok;
2847 }
2848
2849 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2850                      struct nfsd4_compound_state *cstate,
2851                      union nfsd4_op_u *u)
2852 {
2853         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2854         __be32 status;
2855         struct nfsd4_conn *conn;
2856         struct nfsd4_session *session;
2857         struct net *net = SVC_NET(rqstp);
2858         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2859
2860         if (!nfsd4_last_compound_op(rqstp))
2861                 return nfserr_not_only_op;
2862         spin_lock(&nn->client_lock);
2863         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2864         spin_unlock(&nn->client_lock);
2865         if (!session)
2866                 goto out_no_session;
2867         status = nfserr_wrong_cred;
2868         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2869                 goto out;
2870         status = nfsd4_map_bcts_dir(&bcts->dir);
2871         if (status)
2872                 goto out;
2873         conn = alloc_conn(rqstp, bcts->dir);
2874         status = nfserr_jukebox;
2875         if (!conn)
2876                 goto out;
2877         nfsd4_init_conn(rqstp, conn, session);
2878         status = nfs_ok;
2879 out:
2880         nfsd4_put_session(session);
2881 out_no_session:
2882         return status;
2883 }
2884
2885 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2886 {
2887         if (!session)
2888                 return 0;
2889         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2890 }
2891
2892 __be32
2893 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2894                 union nfsd4_op_u *u)
2895 {
2896         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2897         struct nfsd4_session *ses;
2898         __be32 status;
2899         int ref_held_by_me = 0;
2900         struct net *net = SVC_NET(r);
2901         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2902
2903         status = nfserr_not_only_op;
2904         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2905                 if (!nfsd4_last_compound_op(r))
2906                         goto out;
2907                 ref_held_by_me++;
2908         }
2909         dump_sessionid(__func__, &sessionid->sessionid);
2910         spin_lock(&nn->client_lock);
2911         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2912         if (!ses)
2913                 goto out_client_lock;
2914         status = nfserr_wrong_cred;
2915         if (!nfsd4_mach_creds_match(ses->se_client, r))
2916                 goto out_put_session;
2917         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2918         if (status)
2919                 goto out_put_session;
2920         unhash_session(ses);
2921         spin_unlock(&nn->client_lock);
2922
2923         nfsd4_probe_callback_sync(ses->se_client);
2924
2925         spin_lock(&nn->client_lock);
2926         status = nfs_ok;
2927 out_put_session:
2928         nfsd4_put_session_locked(ses);
2929 out_client_lock:
2930         spin_unlock(&nn->client_lock);
2931 out:
2932         return status;
2933 }
2934
2935 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2936 {
2937         struct nfsd4_conn *c;
2938
2939         list_for_each_entry(c, &s->se_conns, cn_persession) {
2940                 if (c->cn_xprt == xpt) {
2941                         return c;
2942                 }
2943         }
2944         return NULL;
2945 }
2946
2947 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2948 {
2949         struct nfs4_client *clp = ses->se_client;
2950         struct nfsd4_conn *c;
2951         __be32 status = nfs_ok;
2952         int ret;
2953
2954         spin_lock(&clp->cl_lock);
2955         c = __nfsd4_find_conn(new->cn_xprt, ses);
2956         if (c)
2957                 goto out_free;
2958         status = nfserr_conn_not_bound_to_session;
2959         if (clp->cl_mach_cred)
2960                 goto out_free;
2961         __nfsd4_hash_conn(new, ses);
2962         spin_unlock(&clp->cl_lock);
2963         ret = nfsd4_register_conn(new);
2964         if (ret)
2965                 /* oops; xprt is already down: */
2966                 nfsd4_conn_lost(&new->cn_xpt_user);
2967         return nfs_ok;
2968 out_free:
2969         spin_unlock(&clp->cl_lock);
2970         free_conn(new);
2971         return status;
2972 }
2973
2974 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2975 {
2976         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2977
2978         return args->opcnt > session->se_fchannel.maxops;
2979 }
2980
2981 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2982                                   struct nfsd4_session *session)
2983 {
2984         struct xdr_buf *xb = &rqstp->rq_arg;
2985
2986         return xb->len > session->se_fchannel.maxreq_sz;
2987 }
2988
2989 __be32
2990 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2991                 union nfsd4_op_u *u)
2992 {
2993         struct nfsd4_sequence *seq = &u->sequence;
2994         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2995         struct xdr_stream *xdr = &resp->xdr;
2996         struct nfsd4_session *session;
2997         struct nfs4_client *clp;
2998         struct nfsd4_slot *slot;
2999         struct nfsd4_conn *conn;
3000         __be32 status;
3001         int buflen;
3002         struct net *net = SVC_NET(rqstp);
3003         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3004
3005         if (resp->opcnt != 1)
3006                 return nfserr_sequence_pos;
3007
3008         /*
3009          * Will be either used or freed by nfsd4_sequence_check_conn
3010          * below.
3011          */
3012         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3013         if (!conn)
3014                 return nfserr_jukebox;
3015
3016         spin_lock(&nn->client_lock);
3017         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3018         if (!session)
3019                 goto out_no_session;
3020         clp = session->se_client;
3021
3022         status = nfserr_too_many_ops;
3023         if (nfsd4_session_too_many_ops(rqstp, session))
3024                 goto out_put_session;
3025
3026         status = nfserr_req_too_big;
3027         if (nfsd4_request_too_big(rqstp, session))
3028                 goto out_put_session;
3029
3030         status = nfserr_badslot;
3031         if (seq->slotid >= session->se_fchannel.maxreqs)
3032                 goto out_put_session;
3033
3034         slot = session->se_slots[seq->slotid];
3035         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3036
3037         /* We do not negotiate the number of slots yet, so set the
3038          * maxslots to the session maxreqs which is used to encode
3039          * sr_highest_slotid and the sr_target_slot id to maxslots */
3040         seq->maxslots = session->se_fchannel.maxreqs;
3041
3042         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3043                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3044         if (status == nfserr_replay_cache) {
3045                 status = nfserr_seq_misordered;
3046                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3047                         goto out_put_session;
3048                 cstate->slot = slot;
3049                 cstate->session = session;
3050                 cstate->clp = clp;
3051                 /* Return the cached reply status and set cstate->status
3052                  * for nfsd4_proc_compound processing */
3053                 status = nfsd4_replay_cache_entry(resp, seq);
3054                 cstate->status = nfserr_replay_cache;
3055                 goto out;
3056         }
3057         if (status)
3058                 goto out_put_session;
3059
3060         status = nfsd4_sequence_check_conn(conn, session);
3061         conn = NULL;
3062         if (status)
3063                 goto out_put_session;
3064
3065         buflen = (seq->cachethis) ?
3066                         session->se_fchannel.maxresp_cached :
3067                         session->se_fchannel.maxresp_sz;
3068         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3069                                     nfserr_rep_too_big;
3070         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3071                 goto out_put_session;
3072         svc_reserve(rqstp, buflen);
3073
3074         status = nfs_ok;
3075         /* Success! bump slot seqid */
3076         slot->sl_seqid = seq->seqid;
3077         slot->sl_flags |= NFSD4_SLOT_INUSE;
3078         if (seq->cachethis)
3079                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3080         else
3081                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3082
3083         cstate->slot = slot;
3084         cstate->session = session;
3085         cstate->clp = clp;
3086
3087 out:
3088         switch (clp->cl_cb_state) {
3089         case NFSD4_CB_DOWN:
3090                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3091                 break;
3092         case NFSD4_CB_FAULT:
3093                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3094                 break;
3095         default:
3096                 seq->status_flags = 0;
3097         }
3098         if (!list_empty(&clp->cl_revoked))
3099                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3100 out_no_session:
3101         if (conn)
3102                 free_conn(conn);
3103         spin_unlock(&nn->client_lock);
3104         return status;
3105 out_put_session:
3106         nfsd4_put_session_locked(session);
3107         goto out_no_session;
3108 }
3109
3110 void
3111 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3112 {
3113         struct nfsd4_compound_state *cs = &resp->cstate;
3114
3115         if (nfsd4_has_session(cs)) {
3116                 if (cs->status != nfserr_replay_cache) {
3117                         nfsd4_store_cache_entry(resp);
3118                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3119                 }
3120                 /* Drop session reference that was taken in nfsd4_sequence() */
3121                 nfsd4_put_session(cs->session);
3122         } else if (cs->clp)
3123                 put_client_renew(cs->clp);
3124 }
3125
3126 __be32
3127 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3128                 struct nfsd4_compound_state *cstate,
3129                 union nfsd4_op_u *u)
3130 {
3131         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3132         struct nfs4_client *conf, *unconf;
3133         struct nfs4_client *clp = NULL;
3134         __be32 status = 0;
3135         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3136
3137         spin_lock(&nn->client_lock);
3138         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3139         conf = find_confirmed_client(&dc->clientid, true, nn);
3140         WARN_ON_ONCE(conf && unconf);
3141
3142         if (conf) {
3143                 if (client_has_state(conf)) {
3144                         status = nfserr_clientid_busy;
3145                         goto out;
3146                 }
3147                 status = mark_client_expired_locked(conf);
3148                 if (status)
3149                         goto out;
3150                 clp = conf;
3151         } else if (unconf)
3152                 clp = unconf;
3153         else {
3154                 status = nfserr_stale_clientid;
3155                 goto out;
3156         }
3157         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3158                 clp = NULL;
3159                 status = nfserr_wrong_cred;
3160                 goto out;
3161         }
3162         unhash_client_locked(clp);
3163 out:
3164         spin_unlock(&nn->client_lock);
3165         if (clp)
3166                 expire_client(clp);
3167         return status;
3168 }
3169
3170 __be32
3171 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3172                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3173 {
3174         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3175         __be32 status = 0;
3176
3177         if (rc->rca_one_fs) {
3178                 if (!cstate->current_fh.fh_dentry)
3179                         return nfserr_nofilehandle;
3180                 /*
3181                  * We don't take advantage of the rca_one_fs case.
3182                  * That's OK, it's optional, we can safely ignore it.
3183                  */
3184                 return nfs_ok;
3185         }
3186
3187         status = nfserr_complete_already;
3188         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3189                              &cstate->session->se_client->cl_flags))
3190                 goto out;
3191
3192         status = nfserr_stale_clientid;
3193         if (is_client_expired(cstate->session->se_client))
3194                 /*
3195                  * The following error isn't really legal.
3196                  * But we only get here if the client just explicitly
3197                  * destroyed the client.  Surely it no longer cares what
3198                  * error it gets back on an operation for the dead
3199                  * client.
3200                  */
3201                 goto out;
3202
3203         status = nfs_ok;
3204         nfsd4_client_record_create(cstate->session->se_client);
3205 out:
3206         return status;
3207 }
3208
3209 __be32
3210 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3211                   union nfsd4_op_u *u)
3212 {
3213         struct nfsd4_setclientid *setclid = &u->setclientid;
3214         struct xdr_netobj       clname = setclid->se_name;
3215         nfs4_verifier           clverifier = setclid->se_verf;
3216         struct nfs4_client      *conf, *new;
3217         struct nfs4_client      *unconf = NULL;
3218         __be32                  status;
3219         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3220
3221         new = create_client(clname, rqstp, &clverifier);
3222         if (new == NULL)
3223                 return nfserr_jukebox;
3224         /* Cases below refer to rfc 3530 section 14.2.33: */
3225         spin_lock(&nn->client_lock);
3226         conf = find_confirmed_client_by_name(&clname, nn);
3227         if (conf && client_has_state(conf)) {
3228                 /* case 0: */
3229                 status = nfserr_clid_inuse;
3230                 if (clp_used_exchangeid(conf))
3231                         goto out;
3232                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3233                         char addr_str[INET6_ADDRSTRLEN];
3234                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3235                                  sizeof(addr_str));
3236                         dprintk("NFSD: setclientid: string in use by client "
3237                                 "at %s\n", addr_str);
3238                         goto out;
3239                 }
3240         }
3241         unconf = find_unconfirmed_client_by_name(&clname, nn);
3242         if (unconf)
3243                 unhash_client_locked(unconf);
3244         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3245                 /* case 1: probable callback update */
3246                 copy_clid(new, conf);
3247                 gen_confirm(new, nn);
3248         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3249                 gen_clid(new, nn);
3250         new->cl_minorversion = 0;
3251         gen_callback(new, setclid, rqstp);
3252         add_to_unconfirmed(new);
3253         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3254         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3255         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3256         new = NULL;
3257         status = nfs_ok;
3258 out:
3259         spin_unlock(&nn->client_lock);
3260         if (new)
3261                 free_client(new);
3262         if (unconf)
3263                 expire_client(unconf);
3264         return status;
3265 }
3266
3267
3268 __be32
3269 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3270                         struct nfsd4_compound_state *cstate,
3271                         union nfsd4_op_u *u)
3272 {
3273         struct nfsd4_setclientid_confirm *setclientid_confirm =
3274                         &u->setclientid_confirm;
3275         struct nfs4_client *conf, *unconf;
3276         struct nfs4_client *old = NULL;
3277         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3278         clientid_t * clid = &setclientid_confirm->sc_clientid;
3279         __be32 status;
3280         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3281
3282         if (STALE_CLIENTID(clid, nn))
3283                 return nfserr_stale_clientid;
3284
3285         spin_lock(&nn->client_lock);
3286         conf = find_confirmed_client(clid, false, nn);
3287         unconf = find_unconfirmed_client(clid, false, nn);
3288         /*
3289          * We try hard to give out unique clientid's, so if we get an
3290          * attempt to confirm the same clientid with a different cred,
3291          * the client may be buggy; this should never happen.
3292          *
3293          * Nevertheless, RFC 7530 recommends INUSE for this case:
3294          */
3295         status = nfserr_clid_inuse;
3296         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3297                 goto out;
3298         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3299                 goto out;
3300         /* cases below refer to rfc 3530 section 14.2.34: */
3301         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3302                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3303                         /* case 2: probable retransmit */
3304                         status = nfs_ok;
3305                 } else /* case 4: client hasn't noticed we rebooted yet? */
3306                         status = nfserr_stale_clientid;
3307                 goto out;
3308         }
3309         status = nfs_ok;
3310         if (conf) { /* case 1: callback update */
3311                 old = unconf;
3312                 unhash_client_locked(old);
3313                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3314         } else { /* case 3: normal case; new or rebooted client */
3315                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3316                 if (old) {
3317                         status = nfserr_clid_inuse;
3318                         if (client_has_state(old)
3319                                         && !same_creds(&unconf->cl_cred,
3320                                                         &old->cl_cred))
3321                                 goto out;
3322                         status = mark_client_expired_locked(old);
3323                         if (status) {
3324                                 old = NULL;
3325                                 goto out;
3326                         }
3327                 }
3328                 move_to_confirmed(unconf);
3329                 conf = unconf;
3330         }
3331         get_client_locked(conf);
3332         spin_unlock(&nn->client_lock);
3333         nfsd4_probe_callback(conf);
3334         spin_lock(&nn->client_lock);
3335         put_client_renew_locked(conf);
3336 out:
3337         spin_unlock(&nn->client_lock);
3338         if (old)
3339                 expire_client(old);
3340         return status;
3341 }
3342
3343 static struct nfs4_file *nfsd4_alloc_file(void)
3344 {
3345         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3346 }
3347
3348 /* OPEN Share state helper functions */
3349 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3350                                 struct nfs4_file *fp)
3351 {
3352         lockdep_assert_held(&state_lock);
3353
3354         atomic_set(&fp->fi_ref, 1);
3355         spin_lock_init(&fp->fi_lock);
3356         INIT_LIST_HEAD(&fp->fi_stateids);
3357         INIT_LIST_HEAD(&fp->fi_delegations);
3358         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3359         fh_copy_shallow(&fp->fi_fhandle, fh);
3360         fp->fi_deleg_file = NULL;
3361         fp->fi_had_conflict = false;
3362         fp->fi_share_deny = 0;
3363         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3364         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3365 #ifdef CONFIG_NFSD_PNFS
3366         INIT_LIST_HEAD(&fp->fi_lo_states);
3367         atomic_set(&fp->fi_lo_recalls, 0);
3368 #endif
3369         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3370 }
3371
3372 void
3373 nfsd4_free_slabs(void)
3374 {
3375         kmem_cache_destroy(odstate_slab);
3376         kmem_cache_destroy(openowner_slab);
3377         kmem_cache_destroy(lockowner_slab);
3378         kmem_cache_destroy(file_slab);
3379         kmem_cache_destroy(stateid_slab);
3380         kmem_cache_destroy(deleg_slab);
3381 }
3382
3383 int
3384 nfsd4_init_slabs(void)
3385 {
3386         openowner_slab = kmem_cache_create("nfsd4_openowners",
3387                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3388         if (openowner_slab == NULL)
3389                 goto out;
3390         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3391                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3392         if (lockowner_slab == NULL)
3393                 goto out_free_openowner_slab;
3394         file_slab = kmem_cache_create("nfsd4_files",
3395                         sizeof(struct nfs4_file), 0, 0, NULL);
3396         if (file_slab == NULL)
3397                 goto out_free_lockowner_slab;
3398         stateid_slab = kmem_cache_create("nfsd4_stateids",
3399                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3400         if (stateid_slab == NULL)
3401                 goto out_free_file_slab;
3402         deleg_slab = kmem_cache_create("nfsd4_delegations",
3403                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3404         if (deleg_slab == NULL)
3405                 goto out_free_stateid_slab;
3406         odstate_slab = kmem_cache_create("nfsd4_odstate",
3407                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3408         if (odstate_slab == NULL)
3409                 goto out_free_deleg_slab;
3410         return 0;
3411
3412 out_free_deleg_slab:
3413         kmem_cache_destroy(deleg_slab);
3414 out_free_stateid_slab:
3415         kmem_cache_destroy(stateid_slab);
3416 out_free_file_slab:
3417         kmem_cache_destroy(file_slab);
3418 out_free_lockowner_slab:
3419         kmem_cache_destroy(lockowner_slab);
3420 out_free_openowner_slab:
3421         kmem_cache_destroy(openowner_slab);
3422 out:
3423         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3424         return -ENOMEM;
3425 }
3426
3427 static void init_nfs4_replay(struct nfs4_replay *rp)
3428 {
3429         rp->rp_status = nfserr_serverfault;
3430         rp->rp_buflen = 0;
3431         rp->rp_buf = rp->rp_ibuf;
3432         mutex_init(&rp->rp_mutex);
3433 }
3434
3435 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3436                 struct nfs4_stateowner *so)
3437 {
3438         if (!nfsd4_has_session(cstate)) {
3439                 mutex_lock(&so->so_replay.rp_mutex);
3440                 cstate->replay_owner = nfs4_get_stateowner(so);
3441         }
3442 }
3443
3444 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3445 {
3446         struct nfs4_stateowner *so = cstate->replay_owner;
3447
3448         if (so != NULL) {
3449                 cstate->replay_owner = NULL;
3450                 mutex_unlock(&so->so_replay.rp_mutex);
3451                 nfs4_put_stateowner(so);
3452         }
3453 }
3454
3455 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3456 {
3457         struct nfs4_stateowner *sop;
3458
3459         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3460         if (!sop)
3461                 return NULL;
3462
3463         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3464         if (!sop->so_owner.data) {
3465                 kmem_cache_free(slab, sop);
3466                 return NULL;
3467         }
3468         sop->so_owner.len = owner->len;
3469
3470         INIT_LIST_HEAD(&sop->so_stateids);
3471         sop->so_client = clp;
3472         init_nfs4_replay(&sop->so_replay);
3473         atomic_set(&sop->so_count, 1);
3474         return sop;
3475 }
3476
3477 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3478 {
3479         lockdep_assert_held(&clp->cl_lock);
3480
3481         list_add(&oo->oo_owner.so_strhash,
3482                  &clp->cl_ownerstr_hashtbl[strhashval]);
3483         list_add(&oo->oo_perclient, &clp->cl_openowners);
3484 }
3485
3486 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3487 {
3488         unhash_openowner_locked(openowner(so));
3489 }
3490
3491 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3492 {
3493         struct nfs4_openowner *oo = openowner(so);
3494
3495         kmem_cache_free(openowner_slab, oo);
3496 }
3497
3498 static const struct nfs4_stateowner_operations openowner_ops = {
3499         .so_unhash =    nfs4_unhash_openowner,
3500         .so_free =      nfs4_free_openowner,
3501 };
3502
3503 static struct nfs4_ol_stateid *
3504 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3505 {
3506         struct nfs4_ol_stateid *local, *ret = NULL;
3507         struct nfs4_openowner *oo = open->op_openowner;
3508
3509         lockdep_assert_held(&fp->fi_lock);
3510
3511         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3512                 /* ignore lock owners */
3513                 if (local->st_stateowner->so_is_open_owner == 0)
3514                         continue;
3515                 if (local->st_stateowner == &oo->oo_owner) {
3516                         ret = local;
3517                         atomic_inc(&ret->st_stid.sc_count);
3518                         break;
3519                 }
3520         }
3521         return ret;
3522 }
3523
3524 static struct nfs4_openowner *
3525 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3526                            struct nfsd4_compound_state *cstate)
3527 {
3528         struct nfs4_client *clp = cstate->clp;
3529         struct nfs4_openowner *oo, *ret;
3530
3531         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3532         if (!oo)
3533                 return NULL;
3534         oo->oo_owner.so_ops = &openowner_ops;
3535         oo->oo_owner.so_is_open_owner = 1;
3536         oo->oo_owner.so_seqid = open->op_seqid;
3537         oo->oo_flags = 0;
3538         if (nfsd4_has_session(cstate))
3539                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3540         oo->oo_time = 0;
3541         oo->oo_last_closed_stid = NULL;
3542         INIT_LIST_HEAD(&oo->oo_close_lru);
3543         spin_lock(&clp->cl_lock);
3544         ret = find_openstateowner_str_locked(strhashval, open, clp);
3545         if (ret == NULL) {
3546                 hash_openowner(oo, clp, strhashval);
3547                 ret = oo;
3548         } else
3549                 nfs4_free_stateowner(&oo->oo_owner);
3550
3551         spin_unlock(&clp->cl_lock);
3552         return ret;
3553 }
3554
3555 static struct nfs4_ol_stateid *
3556 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3557 {
3558
3559         struct nfs4_openowner *oo = open->op_openowner;
3560         struct nfs4_ol_stateid *retstp = NULL;
3561         struct nfs4_ol_stateid *stp;
3562
3563         stp = open->op_stp;
3564         /* We are moving these outside of the spinlocks to avoid the warnings */
3565         mutex_init(&stp->st_mutex);
3566         mutex_lock(&stp->st_mutex);
3567
3568         spin_lock(&oo->oo_owner.so_client->cl_lock);
3569         spin_lock(&fp->fi_lock);
3570
3571         retstp = nfsd4_find_existing_open(fp, open);
3572         if (retstp)
3573                 goto out_unlock;
3574
3575         open->op_stp = NULL;
3576         atomic_inc(&stp->st_stid.sc_count);
3577         stp->st_stid.sc_type = NFS4_OPEN_STID;
3578         INIT_LIST_HEAD(&stp->st_locks);
3579         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3580         get_nfs4_file(fp);
3581         stp->st_stid.sc_file = fp;
3582         stp->st_access_bmap = 0;
3583         stp->st_deny_bmap = 0;
3584         stp->st_openstp = NULL;
3585         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3586         list_add(&stp->st_perfile, &fp->fi_stateids);
3587
3588 out_unlock:
3589         spin_unlock(&fp->fi_lock);
3590         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3591         if (retstp) {
3592                 mutex_lock(&retstp->st_mutex);
3593                 /* To keep mutex tracking happy */
3594                 mutex_unlock(&stp->st_mutex);
3595                 stp = retstp;
3596         }
3597         return stp;
3598 }
3599
3600 /*
3601  * In the 4.0 case we need to keep the owners around a little while to handle
3602  * CLOSE replay. We still do need to release any file access that is held by
3603  * them before returning however.
3604  */
3605 static void
3606 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3607 {
3608         struct nfs4_ol_stateid *last;
3609         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3610         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3611                                                 nfsd_net_id);
3612
3613         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3614
3615         /*
3616          * We know that we hold one reference via nfsd4_close, and another
3617          * "persistent" reference for the client. If the refcount is higher
3618          * than 2, then there are still calls in progress that are using this
3619          * stateid. We can't put the sc_file reference until they are finished.
3620          * Wait for the refcount to drop to 2. Since it has been unhashed,
3621          * there should be no danger of the refcount going back up again at
3622          * this point.
3623          */
3624         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3625
3626         release_all_access(s);
3627         if (s->st_stid.sc_file) {
3628                 put_nfs4_file(s->st_stid.sc_file);
3629                 s->st_stid.sc_file = NULL;
3630         }
3631
3632         spin_lock(&nn->client_lock);
3633         last = oo->oo_last_closed_stid;
3634         oo->oo_last_closed_stid = s;
3635         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3636         oo->oo_time = get_seconds();
3637         spin_unlock(&nn->client_lock);
3638         if (last)
3639                 nfs4_put_stid(&last->st_stid);
3640 }
3641
3642 /* search file_hashtbl[] for file */
3643 static struct nfs4_file *
3644 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3645 {
3646         struct nfs4_file *fp;
3647
3648         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3649                 if (fh_match(&fp->fi_fhandle, fh)) {
3650                         if (atomic_inc_not_zero(&fp->fi_ref))
3651                                 return fp;
3652                 }
3653         }
3654         return NULL;
3655 }
3656
3657 struct nfs4_file *
3658 find_file(struct knfsd_fh *fh)
3659 {
3660         struct nfs4_file *fp;
3661         unsigned int hashval = file_hashval(fh);
3662
3663         rcu_read_lock();
3664         fp = find_file_locked(fh, hashval);
3665         rcu_read_unlock();
3666         return fp;
3667 }
3668
3669 static struct nfs4_file *
3670 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3671 {
3672         struct nfs4_file *fp;
3673         unsigned int hashval = file_hashval(fh);
3674
3675         rcu_read_lock();
3676         fp = find_file_locked(fh, hashval);
3677         rcu_read_unlock();
3678         if (fp)
3679                 return fp;
3680
3681         spin_lock(&state_lock);
3682         fp = find_file_locked(fh, hashval);
3683         if (likely(fp == NULL)) {
3684                 nfsd4_init_file(fh, hashval, new);
3685                 fp = new;
3686         }
3687         spin_unlock(&state_lock);
3688
3689         return fp;
3690 }
3691
3692 /*
3693  * Called to check deny when READ with all zero stateid or
3694  * WRITE with all zero or all one stateid
3695  */
3696 static __be32
3697 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3698 {
3699         struct nfs4_file *fp;
3700         __be32 ret = nfs_ok;
3701
3702         fp = find_file(&current_fh->fh_handle);
3703         if (!fp)
3704                 return ret;
3705         /* Check for conflicting share reservations */
3706         spin_lock(&fp->fi_lock);
3707         if (fp->fi_share_deny & deny_type)
3708                 ret = nfserr_locked;
3709         spin_unlock(&fp->fi_lock);
3710         put_nfs4_file(fp);
3711         return ret;
3712 }
3713
3714 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3715 {
3716         struct nfs4_delegation *dp = cb_to_delegation(cb);
3717         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3718                                           nfsd_net_id);
3719
3720         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3721
3722         /*
3723          * We can't do this in nfsd_break_deleg_cb because it is
3724          * already holding inode->i_lock.
3725          *
3726          * If the dl_time != 0, then we know that it has already been
3727          * queued for a lease break. Don't queue it again.
3728          */
3729         spin_lock(&state_lock);
3730         if (dp->dl_time == 0) {
3731                 dp->dl_time = get_seconds();
3732                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3733         }
3734         spin_unlock(&state_lock);
3735 }
3736
3737 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3738                 struct rpc_task *task)
3739 {
3740         struct nfs4_delegation *dp = cb_to_delegation(cb);
3741
3742         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3743                 return 1;
3744
3745         switch (task->tk_status) {
3746         case 0:
3747                 return 1;
3748         case -EBADHANDLE:
3749         case -NFS4ERR_BAD_STATEID:
3750                 /*
3751                  * Race: client probably got cb_recall before open reply
3752                  * granting delegation.
3753                  */
3754                 if (dp->dl_retries--) {
3755                         rpc_delay(task, 2 * HZ);
3756                         return 0;
3757                 }
3758                 /*FALLTHRU*/
3759         default:
3760                 return -1;
3761         }
3762 }
3763
3764 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3765 {
3766         struct nfs4_delegation *dp = cb_to_delegation(cb);
3767
3768         nfs4_put_stid(&dp->dl_stid);
3769 }
3770
3771 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3772         .prepare        = nfsd4_cb_recall_prepare,
3773         .done           = nfsd4_cb_recall_done,
3774         .release        = nfsd4_cb_recall_release,
3775 };
3776
3777 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3778 {
3779         /*
3780          * We're assuming the state code never drops its reference
3781          * without first removing the lease.  Since we're in this lease
3782          * callback (and since the lease code is serialized by the kernel
3783          * lock) we know the server hasn't removed the lease yet, we know
3784          * it's safe to take a reference.
3785          */
3786         atomic_inc(&dp->dl_stid.sc_count);
3787         nfsd4_run_cb(&dp->dl_recall);
3788 }
3789
3790 /* Called from break_lease() with i_lock held. */
3791 static bool
3792 nfsd_break_deleg_cb(struct file_lock *fl)
3793 {
3794         bool ret = false;
3795         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3796         struct nfs4_delegation *dp;
3797
3798         if (!fp) {
3799                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3800                 return ret;
3801         }
3802         if (fp->fi_had_conflict) {
3803                 WARN(1, "duplicate break on %p\n", fp);
3804                 return ret;
3805         }
3806         /*
3807          * We don't want the locks code to timeout the lease for us;
3808          * we'll remove it ourself if a delegation isn't returned
3809          * in time:
3810          */
3811         fl->fl_break_time = 0;
3812
3813         spin_lock(&fp->fi_lock);
3814         fp->fi_had_conflict = true;
3815         /*
3816          * If there are no delegations on the list, then return true
3817          * so that the lease code will go ahead and delete it.
3818          */
3819         if (list_empty(&fp->fi_delegations))
3820                 ret = true;
3821         else
3822                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3823                         nfsd_break_one_deleg(dp);
3824         spin_unlock(&fp->fi_lock);
3825         return ret;
3826 }
3827
3828 static int
3829 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3830                      struct list_head *dispose)
3831 {
3832         if (arg & F_UNLCK)
3833                 return lease_modify(onlist, arg, dispose);
3834         else
3835                 return -EAGAIN;
3836 }
3837
3838 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3839         .lm_break = nfsd_break_deleg_cb,
3840         .lm_change = nfsd_change_deleg_cb,
3841 };
3842
3843 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3844 {
3845         if (nfsd4_has_session(cstate))
3846                 return nfs_ok;
3847         if (seqid == so->so_seqid - 1)
3848                 return nfserr_replay_me;
3849         if (seqid == so->so_seqid)
3850                 return nfs_ok;
3851         return nfserr_bad_seqid;
3852 }
3853
3854 static __be32 lookup_clientid(clientid_t *clid,
3855                 struct nfsd4_compound_state *cstate,
3856                 struct nfsd_net *nn)
3857 {
3858         struct nfs4_client *found;
3859
3860         if (cstate->clp) {
3861                 found = cstate->clp;
3862                 if (!same_clid(&found->cl_clientid, clid))
3863                         return nfserr_stale_clientid;
3864                 return nfs_ok;
3865         }
3866
3867         if (STALE_CLIENTID(clid, nn))
3868                 return nfserr_stale_clientid;
3869
3870         /*
3871          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3872          * cached already then we know this is for is for v4.0 and "sessions"
3873          * will be false.
3874          */
3875         WARN_ON_ONCE(cstate->session);
3876         spin_lock(&nn->client_lock);
3877         found = find_confirmed_client(clid, false, nn);
3878         if (!found) {
3879                 spin_unlock(&nn->client_lock);
3880                 return nfserr_expired;
3881         }
3882         atomic_inc(&found->cl_refcount);
3883         spin_unlock(&nn->client_lock);
3884
3885         /* Cache the nfs4_client in cstate! */
3886         cstate->clp = found;
3887         return nfs_ok;
3888 }
3889
3890 __be32
3891 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3892                     struct nfsd4_open *open, struct nfsd_net *nn)
3893 {
3894         clientid_t *clientid = &open->op_clientid;
3895         struct nfs4_client *clp = NULL;
3896         unsigned int strhashval;
3897         struct nfs4_openowner *oo = NULL;
3898         __be32 status;
3899
3900         if (STALE_CLIENTID(&open->op_clientid, nn))
3901                 return nfserr_stale_clientid;
3902         /*
3903          * In case we need it later, after we've already created the
3904          * file and don't want to risk a further failure:
3905          */
3906         open->op_file = nfsd4_alloc_file();
3907         if (open->op_file == NULL)
3908                 return nfserr_jukebox;
3909
3910         status = lookup_clientid(clientid, cstate, nn);
3911         if (status)
3912                 return status;
3913         clp = cstate->clp;
3914
3915         strhashval = ownerstr_hashval(&open->op_owner);
3916         oo = find_openstateowner_str(strhashval, open, clp);
3917         open->op_openowner = oo;
3918         if (!oo) {
3919                 goto new_owner;
3920         }
3921         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3922                 /* Replace unconfirmed owners without checking for replay. */
3923                 release_openowner(oo);
3924                 open->op_openowner = NULL;
3925                 goto new_owner;
3926         }
3927         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3928         if (status)
3929                 return status;
3930         goto alloc_stateid;
3931 new_owner:
3932         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3933         if (oo == NULL)
3934                 return nfserr_jukebox;
3935         open->op_openowner = oo;
3936 alloc_stateid:
3937         open->op_stp = nfs4_alloc_open_stateid(clp);
3938         if (!open->op_stp)
3939                 return nfserr_jukebox;
3940
3941         if (nfsd4_has_session(cstate) &&
3942             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3943                 open->op_odstate = alloc_clnt_odstate(clp);
3944                 if (!open->op_odstate)
3945                         return nfserr_jukebox;
3946         }
3947
3948         return nfs_ok;
3949 }
3950
3951 static inline __be32
3952 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3953 {
3954         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3955                 return nfserr_openmode;
3956         else
3957                 return nfs_ok;
3958 }
3959
3960 static int share_access_to_flags(u32 share_access)
3961 {
3962         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3963 }
3964
3965 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3966 {
3967         struct nfs4_stid *ret;
3968
3969         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3970         if (!ret)
3971                 return NULL;
3972         return delegstateid(ret);
3973 }
3974
3975 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3976 {
3977         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3978                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3979 }
3980
3981 static __be32
3982 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3983                 struct nfs4_delegation **dp)
3984 {
3985         int flags;
3986         __be32 status = nfserr_bad_stateid;
3987         struct nfs4_delegation *deleg;
3988
3989         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3990         if (deleg == NULL)
3991                 goto out;
3992         flags = share_access_to_flags(open->op_share_access);
3993         status = nfs4_check_delegmode(deleg, flags);
3994         if (status) {
3995                 nfs4_put_stid(&deleg->dl_stid);
3996                 goto out;
3997         }
3998         *dp = deleg;
3999 out:
4000         if (!nfsd4_is_deleg_cur(open))
4001                 return nfs_ok;
4002         if (status)
4003                 return status;
4004         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4005         return nfs_ok;
4006 }
4007
4008 static inline int nfs4_access_to_access(u32 nfs4_access)
4009 {
4010         int flags = 0;
4011
4012         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4013                 flags |= NFSD_MAY_READ;
4014         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4015                 flags |= NFSD_MAY_WRITE;
4016         return flags;
4017 }
4018
4019 static inline __be32
4020 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4021                 struct nfsd4_open *open)
4022 {
4023         struct iattr iattr = {
4024                 .ia_valid = ATTR_SIZE,
4025                 .ia_size = 0,
4026         };
4027         if (!open->op_truncate)
4028                 return 0;
4029         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4030                 return nfserr_inval;
4031         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4032 }
4033
4034 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4035                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4036                 struct nfsd4_open *open)
4037 {
4038         struct file *filp = NULL;
4039         __be32 status;
4040         int oflag = nfs4_access_to_omode(open->op_share_access);
4041         int access = nfs4_access_to_access(open->op_share_access);
4042         unsigned char old_access_bmap, old_deny_bmap;
4043
4044         spin_lock(&fp->fi_lock);
4045
4046         /*
4047          * Are we trying to set a deny mode that would conflict with
4048          * current access?
4049          */
4050         status = nfs4_file_check_deny(fp, open->op_share_deny);
4051         if (status != nfs_ok) {
4052                 spin_unlock(&fp->fi_lock);
4053                 goto out;
4054         }
4055
4056         /* set access to the file */
4057         status = nfs4_file_get_access(fp, open->op_share_access);
4058         if (status != nfs_ok) {
4059                 spin_unlock(&fp->fi_lock);
4060                 goto out;
4061         }
4062
4063         /* Set access bits in stateid */
4064         old_access_bmap = stp->st_access_bmap;
4065         set_access(open->op_share_access, stp);
4066
4067         /* Set new deny mask */
4068         old_deny_bmap = stp->st_deny_bmap;
4069         set_deny(open->op_share_deny, stp);
4070         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4071
4072         if (!fp->fi_fds[oflag]) {
4073                 spin_unlock(&fp->fi_lock);
4074                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4075                 if (status)
4076                         goto out_put_access;
4077                 spin_lock(&fp->fi_lock);
4078                 if (!fp->fi_fds[oflag]) {
4079                         fp->fi_fds[oflag] = filp;
4080                         filp = NULL;
4081                 }
4082         }
4083         spin_unlock(&fp->fi_lock);
4084         if (filp)
4085                 fput(filp);
4086
4087         status = nfsd4_truncate(rqstp, cur_fh, open);
4088         if (status)
4089                 goto out_put_access;
4090 out:
4091         return status;
4092 out_put_access:
4093         stp->st_access_bmap = old_access_bmap;
4094         nfs4_file_put_access(fp, open->op_share_access);
4095         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4096         goto out;
4097 }
4098
4099 static __be32
4100 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)
4101 {
4102         __be32 status;
4103         unsigned char old_deny_bmap = stp->st_deny_bmap;
4104
4105         if (!test_access(open->op_share_access, stp))
4106                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4107
4108         /* test and set deny mode */
4109         spin_lock(&fp->fi_lock);
4110         status = nfs4_file_check_deny(fp, open->op_share_deny);
4111         if (status == nfs_ok) {
4112                 set_deny(open->op_share_deny, stp);
4113                 fp->fi_share_deny |=
4114                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4115         }
4116         spin_unlock(&fp->fi_lock);
4117
4118         if (status != nfs_ok)
4119                 return status;
4120
4121         status = nfsd4_truncate(rqstp, cur_fh, open);
4122         if (status != nfs_ok)
4123                 reset_union_bmap_deny(old_deny_bmap, stp);
4124         return status;
4125 }
4126
4127 /* Should we give out recallable state?: */
4128 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4129 {
4130         if (clp->cl_cb_state == NFSD4_CB_UP)
4131                 return true;
4132         /*
4133          * In the sessions case, since we don't have to establish a
4134          * separate connection for callbacks, we assume it's OK
4135          * until we hear otherwise:
4136          */
4137         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4138 }
4139
4140 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4141 {
4142         struct file_lock *fl;
4143
4144         fl = locks_alloc_lock();
4145         if (!fl)
4146                 return NULL;
4147         fl->fl_lmops = &nfsd_lease_mng_ops;
4148         fl->fl_flags = FL_DELEG;
4149         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4150         fl->fl_end = OFFSET_MAX;
4151         fl->fl_owner = (fl_owner_t)fp;
4152         fl->fl_pid = current->tgid;
4153         return fl;
4154 }
4155
4156 /**
4157  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4158  * @dp:   a pointer to the nfs4_delegation we're adding.
4159  *
4160  * Return:
4161  *      On success: Return code will be 0 on success.
4162  *
4163  *      On error: -EAGAIN if there was an existing delegation.
4164  *                 nonzero if there is an error in other cases.
4165  *
4166  */
4167
4168 static int nfs4_setlease(struct nfs4_delegation *dp)
4169 {
4170         struct nfs4_file *fp = dp->dl_stid.sc_file;
4171         struct file_lock *fl;
4172         struct file *filp;
4173         int status = 0;
4174
4175         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4176         if (!fl)
4177                 return -ENOMEM;
4178         filp = find_readable_file(fp);
4179         if (!filp) {
4180                 /* We should always have a readable file here */
4181                 WARN_ON_ONCE(1);
4182                 locks_free_lock(fl);
4183                 return -EBADF;
4184         }
4185         fl->fl_file = filp;
4186         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4187         if (fl)
4188                 locks_free_lock(fl);
4189         if (status)
4190                 goto out_fput;
4191         spin_lock(&state_lock);
4192         spin_lock(&fp->fi_lock);
4193         /* Did the lease get broken before we took the lock? */
4194         status = -EAGAIN;
4195         if (fp->fi_had_conflict)
4196                 goto out_unlock;
4197         /* Race breaker */
4198         if (fp->fi_deleg_file) {
4199                 status = hash_delegation_locked(dp, fp);
4200                 goto out_unlock;
4201         }
4202         fp->fi_deleg_file = filp;
4203         fp->fi_delegees = 0;
4204         status = hash_delegation_locked(dp, fp);
4205         spin_unlock(&fp->fi_lock);
4206         spin_unlock(&state_lock);
4207         if (status) {
4208                 /* Should never happen, this is a new fi_deleg_file  */
4209                 WARN_ON_ONCE(1);
4210                 goto out_fput;
4211         }
4212         return 0;
4213 out_unlock:
4214         spin_unlock(&fp->fi_lock);
4215         spin_unlock(&state_lock);
4216 out_fput:
4217         fput(filp);
4218         return status;
4219 }
4220
4221 static struct nfs4_delegation *
4222 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4223                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4224 {
4225         int status;
4226         struct nfs4_delegation *dp;
4227
4228         if (fp->fi_had_conflict)
4229                 return ERR_PTR(-EAGAIN);
4230
4231         spin_lock(&state_lock);
4232         spin_lock(&fp->fi_lock);
4233         status = nfs4_get_existing_delegation(clp, fp);
4234         spin_unlock(&fp->fi_lock);
4235         spin_unlock(&state_lock);
4236
4237         if (status)
4238                 return ERR_PTR(status);
4239
4240         dp = alloc_init_deleg(clp, fh, odstate);
4241         if (!dp)
4242                 return ERR_PTR(-ENOMEM);
4243
4244         get_nfs4_file(fp);
4245         spin_lock(&state_lock);
4246         spin_lock(&fp->fi_lock);
4247         dp->dl_stid.sc_file = fp;
4248         if (!fp->fi_deleg_file) {
4249                 spin_unlock(&fp->fi_lock);
4250                 spin_unlock(&state_lock);
4251                 status = nfs4_setlease(dp);
4252                 goto out;
4253         }
4254         if (fp->fi_had_conflict) {
4255                 status = -EAGAIN;
4256                 goto out_unlock;
4257         }
4258         status = hash_delegation_locked(dp, fp);
4259 out_unlock:
4260         spin_unlock(&fp->fi_lock);
4261         spin_unlock(&state_lock);
4262 out:
4263         if (status) {
4264                 put_clnt_odstate(dp->dl_clnt_odstate);
4265                 nfs4_put_stid(&dp->dl_stid);
4266                 return ERR_PTR(status);
4267         }
4268         return dp;
4269 }
4270
4271 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4272 {
4273         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4274         if (status == -EAGAIN)
4275                 open->op_why_no_deleg = WND4_CONTENTION;
4276         else {
4277                 open->op_why_no_deleg = WND4_RESOURCE;
4278                 switch (open->op_deleg_want) {
4279                 case NFS4_SHARE_WANT_READ_DELEG:
4280                 case NFS4_SHARE_WANT_WRITE_DELEG:
4281                 case NFS4_SHARE_WANT_ANY_DELEG:
4282                         break;
4283                 case NFS4_SHARE_WANT_CANCEL:
4284                         open->op_why_no_deleg = WND4_CANCELLED;
4285                         break;
4286                 case NFS4_SHARE_WANT_NO_DELEG:
4287                         WARN_ON_ONCE(1);
4288                 }
4289         }
4290 }
4291
4292 /*
4293  * Attempt to hand out a delegation.
4294  *
4295  * Note we don't support write delegations, and won't until the vfs has
4296  * proper support for them.
4297  */
4298 static void
4299 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4300                         struct nfs4_ol_stateid *stp)
4301 {
4302         struct nfs4_delegation *dp;
4303         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4304         struct nfs4_client *clp = stp->st_stid.sc_client;
4305         int cb_up;
4306         int status = 0;
4307
4308         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4309         open->op_recall = 0;
4310         switch (open->op_claim_type) {
4311                 case NFS4_OPEN_CLAIM_PREVIOUS:
4312                         if (!cb_up)
4313                                 open->op_recall = 1;
4314                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4315                                 goto out_no_deleg;
4316                         break;
4317                 case NFS4_OPEN_CLAIM_NULL:
4318                 case NFS4_OPEN_CLAIM_FH:
4319                         /*
4320                          * Let's not give out any delegations till everyone's
4321                          * had the chance to reclaim theirs, *and* until
4322                          * NLM locks have all been reclaimed:
4323                          */
4324                         if (locks_in_grace(clp->net))
4325                                 goto out_no_deleg;
4326                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4327                                 goto out_no_deleg;
4328                         /*
4329                          * Also, if the file was opened for write or
4330                          * create, there's a good chance the client's
4331                          * about to write to it, resulting in an
4332                          * immediate recall (since we don't support
4333                          * write delegations):
4334                          */
4335                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4336                                 goto out_no_deleg;
4337                         if (open->op_create == NFS4_OPEN_CREATE)
4338                                 goto out_no_deleg;
4339                         break;
4340                 default:
4341                         goto out_no_deleg;
4342         }
4343         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4344         if (IS_ERR(dp))
4345                 goto out_no_deleg;
4346
4347         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4348
4349         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4350                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4351         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4352         nfs4_put_stid(&dp->dl_stid);
4353         return;
4354 out_no_deleg:
4355         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4356         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4357             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4358                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4359                 open->op_recall = 1;
4360         }
4361
4362         /* 4.1 client asking for a delegation? */
4363         if (open->op_deleg_want)
4364                 nfsd4_open_deleg_none_ext(open, status);
4365         return;
4366 }
4367
4368 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4369                                         struct nfs4_delegation *dp)
4370 {
4371         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4372             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4373                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4374                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4375         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4376                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4377                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4378                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4379         }
4380         /* Otherwise the client must be confused wanting a delegation
4381          * it already has, therefore we don't return
4382          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4383          */
4384 }
4385
4386 __be32
4387 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4388 {
4389         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4390         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4391         struct nfs4_file *fp = NULL;
4392         struct nfs4_ol_stateid *stp = NULL;
4393         struct nfs4_delegation *dp = NULL;
4394         __be32 status;
4395
4396         /*
4397          * Lookup file; if found, lookup stateid and check open request,
4398          * and check for delegations in the process of being recalled.
4399          * If not found, create the nfs4_file struct
4400          */
4401         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4402         if (fp != open->op_file) {
4403                 status = nfs4_check_deleg(cl, open, &dp);
4404                 if (status)
4405                         goto out;
4406                 spin_lock(&fp->fi_lock);
4407                 stp = nfsd4_find_existing_open(fp, open);
4408                 spin_unlock(&fp->fi_lock);
4409         } else {
4410                 open->op_file = NULL;
4411                 status = nfserr_bad_stateid;
4412                 if (nfsd4_is_deleg_cur(open))
4413                         goto out;
4414         }
4415
4416         /*
4417          * OPEN the file, or upgrade an existing OPEN.
4418          * If truncate fails, the OPEN fails.
4419          */
4420         if (stp) {
4421                 /* Stateid was found, this is an OPEN upgrade */
4422                 mutex_lock(&stp->st_mutex);
4423                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4424                 if (status) {
4425                         mutex_unlock(&stp->st_mutex);
4426                         goto out;
4427                 }
4428         } else {
4429                 /* stp is returned locked. */
4430                 stp = init_open_stateid(fp, open);
4431                 /* See if we lost the race to some other thread */
4432                 if (stp->st_access_bmap != 0) {
4433                         status = nfs4_upgrade_open(rqstp, fp, current_fh,
4434                                                 stp, open);
4435                         if (status) {
4436                                 mutex_unlock(&stp->st_mutex);
4437                                 goto out;
4438                         }
4439                         goto upgrade_out;
4440                 }
4441                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4442                 if (status) {
4443                         mutex_unlock(&stp->st_mutex);
4444                         release_open_stateid(stp);
4445                         goto out;
4446                 }
4447
4448                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4449                                                         open->op_odstate);
4450                 if (stp->st_clnt_odstate == open->op_odstate)
4451                         open->op_odstate = NULL;
4452         }
4453 upgrade_out:
4454         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4455         mutex_unlock(&stp->st_mutex);
4456
4457         if (nfsd4_has_session(&resp->cstate)) {
4458                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4459                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4460                         open->op_why_no_deleg = WND4_NOT_WANTED;
4461                         goto nodeleg;
4462                 }
4463         }
4464
4465         /*
4466         * Attempt to hand out a delegation. No error return, because the
4467         * OPEN succeeds even if we fail.
4468         */
4469         nfs4_open_delegation(current_fh, open, stp);
4470 nodeleg:
4471         status = nfs_ok;
4472
4473         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4474                 STATEID_VAL(&stp->st_stid.sc_stateid));
4475 out:
4476         /* 4.1 client trying to upgrade/downgrade delegation? */
4477         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4478             open->op_deleg_want)
4479                 nfsd4_deleg_xgrade_none_ext(open, dp);
4480
4481         if (fp)
4482                 put_nfs4_file(fp);
4483         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4484                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4485         /*
4486         * To finish the open response, we just need to set the rflags.
4487         */
4488         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4489         if (nfsd4_has_session(&resp->cstate))
4490                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4491         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4492                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4493
4494         if (dp)
4495                 nfs4_put_stid(&dp->dl_stid);
4496         if (stp)
4497                 nfs4_put_stid(&stp->st_stid);
4498
4499         return status;
4500 }
4501
4502 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4503                               struct nfsd4_open *open)
4504 {
4505         if (open->op_openowner) {
4506                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4507
4508                 nfsd4_cstate_assign_replay(cstate, so);
4509                 nfs4_put_stateowner(so);
4510         }
4511         if (open->op_file)
4512                 kmem_cache_free(file_slab, open->op_file);
4513         if (open->op_stp)
4514                 nfs4_put_stid(&open->op_stp->st_stid);
4515         if (open->op_odstate)
4516                 kmem_cache_free(odstate_slab, open->op_odstate);
4517 }
4518
4519 __be32
4520 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4521             union nfsd4_op_u *u)
4522 {
4523         clientid_t *clid = &u->renew;
4524         struct nfs4_client *clp;
4525         __be32 status;
4526         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4527
4528         dprintk("process_renew(%08x/%08x): starting\n",
4529                         clid->cl_boot, clid->cl_id);
4530         status = lookup_clientid(clid, cstate, nn);
4531         if (status)
4532                 goto out;
4533         clp = cstate->clp;
4534         status = nfserr_cb_path_down;
4535         if (!list_empty(&clp->cl_delegations)
4536                         && clp->cl_cb_state != NFSD4_CB_UP)
4537                 goto out;
4538         status = nfs_ok;
4539 out:
4540         return status;
4541 }
4542
4543 void
4544 nfsd4_end_grace(struct nfsd_net *nn)
4545 {
4546         /* do nothing if grace period already ended */
4547         if (nn->grace_ended)
4548                 return;
4549
4550         dprintk("NFSD: end of grace period\n");
4551         nn->grace_ended = true;
4552         /*
4553          * If the server goes down again right now, an NFSv4
4554          * client will still be allowed to reclaim after it comes back up,
4555          * even if it hasn't yet had a chance to reclaim state this time.
4556          *
4557          */
4558         nfsd4_record_grace_done(nn);
4559         /*
4560          * At this point, NFSv4 clients can still reclaim.  But if the
4561          * server crashes, any that have not yet reclaimed will be out
4562          * of luck on the next boot.
4563          *
4564          * (NFSv4.1+ clients are considered to have reclaimed once they
4565          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4566          * have reclaimed after their first OPEN.)
4567          */
4568         locks_end_grace(&nn->nfsd4_manager);
4569         /*
4570          * At this point, and once lockd and/or any other containers
4571          * exit their grace period, further reclaims will fail and
4572          * regular locking can resume.
4573          */
4574 }
4575
4576 static time_t
4577 nfs4_laundromat(struct nfsd_net *nn)
4578 {
4579         struct nfs4_client *clp;
4580         struct nfs4_openowner *oo;
4581         struct nfs4_delegation *dp;
4582         struct nfs4_ol_stateid *stp;
4583         struct nfsd4_blocked_lock *nbl;
4584         struct list_head *pos, *next, reaplist;
4585         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4586         time_t t, new_timeo = nn->nfsd4_lease;
4587
4588         dprintk("NFSD: laundromat service - starting\n");
4589         nfsd4_end_grace(nn);
4590         INIT_LIST_HEAD(&reaplist);
4591         spin_lock(&nn->client_lock);
4592         list_for_each_safe(pos, next, &nn->client_lru) {
4593                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4594                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4595                         t = clp->cl_time - cutoff;
4596                         new_timeo = min(new_timeo, t);
4597                         break;
4598                 }
4599                 if (mark_client_expired_locked(clp)) {
4600                         dprintk("NFSD: client in use (clientid %08x)\n",
4601                                 clp->cl_clientid.cl_id);
4602                         continue;
4603                 }
4604                 list_add(&clp->cl_lru, &reaplist);
4605         }
4606         spin_unlock(&nn->client_lock);
4607         list_for_each_safe(pos, next, &reaplist) {
4608                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4609                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4610                         clp->cl_clientid.cl_id);
4611                 list_del_init(&clp->cl_lru);
4612                 expire_client(clp);
4613         }
4614         spin_lock(&state_lock);
4615         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4616                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4617                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4618                         t = dp->dl_time - cutoff;
4619                         new_timeo = min(new_timeo, t);
4620                         break;
4621                 }
4622                 WARN_ON(!unhash_delegation_locked(dp));
4623                 list_add(&dp->dl_recall_lru, &reaplist);
4624         }
4625         spin_unlock(&state_lock);
4626         while (!list_empty(&reaplist)) {
4627                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4628                                         dl_recall_lru);
4629                 list_del_init(&dp->dl_recall_lru);
4630                 revoke_delegation(dp);
4631         }
4632
4633         spin_lock(&nn->client_lock);
4634         while (!list_empty(&nn->close_lru)) {
4635                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4636                                         oo_close_lru);
4637                 if (time_after((unsigned long)oo->oo_time,
4638                                (unsigned long)cutoff)) {
4639                         t = oo->oo_time - cutoff;
4640                         new_timeo = min(new_timeo, t);
4641                         break;
4642                 }
4643                 list_del_init(&oo->oo_close_lru);
4644                 stp = oo->oo_last_closed_stid;
4645                 oo->oo_last_closed_stid = NULL;
4646                 spin_unlock(&nn->client_lock);
4647                 nfs4_put_stid(&stp->st_stid);
4648                 spin_lock(&nn->client_lock);
4649         }
4650         spin_unlock(&nn->client_lock);
4651
4652         /*
4653          * It's possible for a client to try and acquire an already held lock
4654          * that is being held for a long time, and then lose interest in it.
4655          * So, we clean out any un-revisited request after a lease period
4656          * under the assumption that the client is no longer interested.
4657          *
4658          * RFC5661, sec. 9.6 states that the client must not rely on getting
4659          * notifications and must continue to poll for locks, even when the
4660          * server supports them. Thus this shouldn't lead to clients blocking
4661          * indefinitely once the lock does become free.
4662          */
4663         BUG_ON(!list_empty(&reaplist));
4664         spin_lock(&nn->blocked_locks_lock);
4665         while (!list_empty(&nn->blocked_locks_lru)) {
4666                 nbl = list_first_entry(&nn->blocked_locks_lru,
4667                                         struct nfsd4_blocked_lock, nbl_lru);
4668                 if (time_after((unsigned long)nbl->nbl_time,
4669                                (unsigned long)cutoff)) {
4670                         t = nbl->nbl_time - cutoff;
4671                         new_timeo = min(new_timeo, t);
4672                         break;
4673                 }
4674                 list_move(&nbl->nbl_lru, &reaplist);
4675                 list_del_init(&nbl->nbl_list);
4676         }
4677         spin_unlock(&nn->blocked_locks_lock);
4678
4679         while (!list_empty(&reaplist)) {
4680                 nbl = list_first_entry(&nn->blocked_locks_lru,
4681                                         struct nfsd4_blocked_lock, nbl_lru);
4682                 list_del_init(&nbl->nbl_lru);
4683                 posix_unblock_lock(&nbl->nbl_lock);
4684                 free_blocked_lock(nbl);
4685         }
4686
4687         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4688         return new_timeo;
4689 }
4690
4691 static struct workqueue_struct *laundry_wq;
4692 static void laundromat_main(struct work_struct *);
4693
4694 static void
4695 laundromat_main(struct work_struct *laundry)
4696 {
4697         time_t t;
4698         struct delayed_work *dwork = to_delayed_work(laundry);
4699         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4700                                            laundromat_work);
4701
4702         t = nfs4_laundromat(nn);
4703         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4704         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4705 }
4706
4707 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4708 {
4709         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4710                 return nfserr_bad_stateid;
4711         return nfs_ok;
4712 }
4713
4714 static inline int
4715 access_permit_read(struct nfs4_ol_stateid *stp)
4716 {
4717         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4718                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4719                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4720 }
4721
4722 static inline int
4723 access_permit_write(struct nfs4_ol_stateid *stp)
4724 {
4725         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4726                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4727 }
4728
4729 static
4730 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4731 {
4732         __be32 status = nfserr_openmode;
4733
4734         /* For lock stateid's, we test the parent open, not the lock: */
4735         if (stp->st_openstp)
4736                 stp = stp->st_openstp;
4737         if ((flags & WR_STATE) && !access_permit_write(stp))
4738                 goto out;
4739         if ((flags & RD_STATE) && !access_permit_read(stp))
4740                 goto out;
4741         status = nfs_ok;
4742 out:
4743         return status;
4744 }
4745
4746 static inline __be32
4747 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4748 {
4749         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4750                 return nfs_ok;
4751         else if (opens_in_grace(net)) {
4752                 /* Answer in remaining cases depends on existence of
4753                  * conflicting state; so we must wait out the grace period. */
4754                 return nfserr_grace;
4755         } else if (flags & WR_STATE)
4756                 return nfs4_share_conflict(current_fh,
4757                                 NFS4_SHARE_DENY_WRITE);
4758         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4759                 return nfs4_share_conflict(current_fh,
4760                                 NFS4_SHARE_DENY_READ);
4761 }
4762
4763 /*
4764  * Allow READ/WRITE during grace period on recovered state only for files
4765  * that are not able to provide mandatory locking.
4766  */
4767 static inline int
4768 grace_disallows_io(struct net *net, struct inode *inode)
4769 {
4770         return opens_in_grace(net) && mandatory_lock(inode);
4771 }
4772
4773 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4774 {
4775         /*
4776          * When sessions are used the stateid generation number is ignored
4777          * when it is zero.
4778          */
4779         if (has_session && in->si_generation == 0)
4780                 return nfs_ok;
4781
4782         if (in->si_generation == ref->si_generation)
4783                 return nfs_ok;
4784
4785         /* If the client sends us a stateid from the future, it's buggy: */
4786         if (nfsd4_stateid_generation_after(in, ref))
4787                 return nfserr_bad_stateid;
4788         /*
4789          * However, we could see a stateid from the past, even from a
4790          * non-buggy client.  For example, if the client sends a lock
4791          * while some IO is outstanding, the lock may bump si_generation
4792          * while the IO is still in flight.  The client could avoid that
4793          * situation by waiting for responses on all the IO requests,
4794          * but better performance may result in retrying IO that
4795          * receives an old_stateid error if requests are rarely
4796          * reordered in flight:
4797          */
4798         return nfserr_old_stateid;
4799 }
4800
4801 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4802 {
4803         if (ols->st_stateowner->so_is_open_owner &&
4804             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4805                 return nfserr_bad_stateid;
4806         return nfs_ok;
4807 }
4808
4809 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4810 {
4811         struct nfs4_stid *s;
4812         __be32 status = nfserr_bad_stateid;
4813
4814         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4815                 return status;
4816         /* Client debugging aid. */
4817         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4818                 char addr_str[INET6_ADDRSTRLEN];
4819                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4820                                  sizeof(addr_str));
4821                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4822                                         "with incorrect client ID\n", addr_str);
4823                 return status;
4824         }
4825         spin_lock(&cl->cl_lock);
4826         s = find_stateid_locked(cl, stateid);
4827         if (!s)
4828                 goto out_unlock;
4829         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4830         if (status)
4831                 goto out_unlock;
4832         switch (s->sc_type) {
4833         case NFS4_DELEG_STID:
4834                 status = nfs_ok;
4835                 break;
4836         case NFS4_REVOKED_DELEG_STID:
4837                 status = nfserr_deleg_revoked;
4838                 break;
4839         case NFS4_OPEN_STID:
4840         case NFS4_LOCK_STID:
4841                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4842                 break;
4843         default:
4844                 printk("unknown stateid type %x\n", s->sc_type);
4845                 /* Fallthrough */
4846         case NFS4_CLOSED_STID:
4847         case NFS4_CLOSED_DELEG_STID:
4848                 status = nfserr_bad_stateid;
4849         }
4850 out_unlock:
4851         spin_unlock(&cl->cl_lock);
4852         return status;
4853 }
4854
4855 __be32
4856 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4857                      stateid_t *stateid, unsigned char typemask,
4858                      struct nfs4_stid **s, struct nfsd_net *nn)
4859 {
4860         __be32 status;
4861
4862         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4863                 return nfserr_bad_stateid;
4864         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4865         if (status == nfserr_stale_clientid) {
4866                 if (cstate->session)
4867                         return nfserr_bad_stateid;
4868                 return nfserr_stale_stateid;
4869         }
4870         if (status)
4871                 return status;
4872         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4873         if (!*s)
4874                 return nfserr_bad_stateid;
4875         return nfs_ok;
4876 }
4877
4878 static struct file *
4879 nfs4_find_file(struct nfs4_stid *s, int flags)
4880 {
4881         if (!s)
4882                 return NULL;
4883
4884         switch (s->sc_type) {
4885         case NFS4_DELEG_STID:
4886                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4887                         return NULL;
4888                 return get_file(s->sc_file->fi_deleg_file);
4889         case NFS4_OPEN_STID:
4890         case NFS4_LOCK_STID:
4891                 if (flags & RD_STATE)
4892                         return find_readable_file(s->sc_file);
4893                 else
4894                         return find_writeable_file(s->sc_file);
4895                 break;
4896         }
4897
4898         return NULL;
4899 }
4900
4901 static __be32
4902 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4903 {
4904         __be32 status;
4905
4906         status = nfsd4_check_openowner_confirmed(ols);
4907         if (status)
4908                 return status;
4909         return nfs4_check_openmode(ols, flags);
4910 }
4911
4912 static __be32
4913 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4914                 struct file **filpp, bool *tmp_file, int flags)
4915 {
4916         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4917         struct file *file;
4918         __be32 status;
4919
4920         file = nfs4_find_file(s, flags);
4921         if (file) {
4922                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4923                                 acc | NFSD_MAY_OWNER_OVERRIDE);
4924                 if (status) {
4925                         fput(file);
4926                         return status;
4927                 }
4928
4929                 *filpp = file;
4930         } else {
4931                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4932                 if (status)
4933                         return status;
4934
4935                 if (tmp_file)
4936                         *tmp_file = true;
4937         }
4938
4939         return 0;
4940 }
4941
4942 /*
4943  * Checks for stateid operations
4944  */
4945 __be32
4946 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4947                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4948                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4949 {
4950         struct inode *ino = d_inode(fhp->fh_dentry);
4951         struct net *net = SVC_NET(rqstp);
4952         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4953         struct nfs4_stid *s = NULL;
4954         __be32 status;
4955
4956         if (filpp)
4957                 *filpp = NULL;
4958         if (tmp_file)
4959                 *tmp_file = false;
4960
4961         if (grace_disallows_io(net, ino))
4962                 return nfserr_grace;
4963
4964         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4965                 status = check_special_stateids(net, fhp, stateid, flags);
4966                 goto done;
4967         }
4968
4969         status = nfsd4_lookup_stateid(cstate, stateid,
4970                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4971                                 &s, nn);
4972         if (status)
4973                 return status;
4974         status = check_stateid_generation(stateid, &s->sc_stateid,
4975                         nfsd4_has_session(cstate));
4976         if (status)
4977                 goto out;
4978
4979         switch (s->sc_type) {
4980         case NFS4_DELEG_STID:
4981                 status = nfs4_check_delegmode(delegstateid(s), flags);
4982                 break;
4983         case NFS4_OPEN_STID:
4984         case NFS4_LOCK_STID:
4985                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4986                 break;
4987         default:
4988                 status = nfserr_bad_stateid;
4989                 break;
4990         }
4991         if (status)
4992                 goto out;
4993         status = nfs4_check_fh(fhp, s);
4994
4995 done:
4996         if (!status && filpp)
4997                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
4998 out:
4999         if (s)
5000                 nfs4_put_stid(s);
5001         return status;
5002 }
5003
5004 /*
5005  * Test if the stateid is valid
5006  */
5007 __be32
5008 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5009                    union nfsd4_op_u *u)
5010 {
5011         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5012         struct nfsd4_test_stateid_id *stateid;
5013         struct nfs4_client *cl = cstate->session->se_client;
5014
5015         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5016                 stateid->ts_id_status =
5017                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5018
5019         return nfs_ok;
5020 }
5021
5022 static __be32
5023 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5024 {
5025         struct nfs4_ol_stateid *stp = openlockstateid(s);
5026         __be32 ret;
5027
5028         mutex_lock(&stp->st_mutex);
5029
5030         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5031         if (ret)
5032                 goto out;
5033
5034         ret = nfserr_locks_held;
5035         if (check_for_locks(stp->st_stid.sc_file,
5036                             lockowner(stp->st_stateowner)))
5037                 goto out;
5038
5039         release_lock_stateid(stp);
5040         ret = nfs_ok;
5041
5042 out:
5043         mutex_unlock(&stp->st_mutex);
5044         nfs4_put_stid(s);
5045         return ret;
5046 }
5047
5048 __be32
5049 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5050                    union nfsd4_op_u *u)
5051 {
5052         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5053         stateid_t *stateid = &free_stateid->fr_stateid;
5054         struct nfs4_stid *s;
5055         struct nfs4_delegation *dp;
5056         struct nfs4_client *cl = cstate->session->se_client;
5057         __be32 ret = nfserr_bad_stateid;
5058
5059         spin_lock(&cl->cl_lock);
5060         s = find_stateid_locked(cl, stateid);
5061         if (!s)
5062                 goto out_unlock;
5063         switch (s->sc_type) {
5064         case NFS4_DELEG_STID:
5065                 ret = nfserr_locks_held;
5066                 break;
5067         case NFS4_OPEN_STID:
5068                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5069                 if (ret)
5070                         break;
5071                 ret = nfserr_locks_held;
5072                 break;
5073         case NFS4_LOCK_STID:
5074                 atomic_inc(&s->sc_count);
5075                 spin_unlock(&cl->cl_lock);
5076                 ret = nfsd4_free_lock_stateid(stateid, s);
5077                 goto out;
5078         case NFS4_REVOKED_DELEG_STID:
5079                 dp = delegstateid(s);
5080                 list_del_init(&dp->dl_recall_lru);
5081                 spin_unlock(&cl->cl_lock);
5082                 nfs4_put_stid(s);
5083                 ret = nfs_ok;
5084                 goto out;
5085         /* Default falls through and returns nfserr_bad_stateid */
5086         }
5087 out_unlock:
5088         spin_unlock(&cl->cl_lock);
5089 out:
5090         return ret;
5091 }
5092
5093 static inline int
5094 setlkflg (int type)
5095 {
5096         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5097                 RD_STATE : WR_STATE;
5098 }
5099
5100 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5101 {
5102         struct svc_fh *current_fh = &cstate->current_fh;
5103         struct nfs4_stateowner *sop = stp->st_stateowner;
5104         __be32 status;
5105
5106         status = nfsd4_check_seqid(cstate, sop, seqid);
5107         if (status)
5108                 return status;
5109         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
5110                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
5111                 /*
5112                  * "Closed" stateid's exist *only* to return
5113                  * nfserr_replay_me from the previous step, and
5114                  * revoked delegations are kept only for free_stateid.
5115                  */
5116                 return nfserr_bad_stateid;
5117         mutex_lock(&stp->st_mutex);
5118         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5119         if (status == nfs_ok)
5120                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5121         if (status != nfs_ok)
5122                 mutex_unlock(&stp->st_mutex);
5123         return status;
5124 }
5125
5126 /*
5127  * Checks for sequence id mutating operations.
5128  */
5129 static __be32
5130 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5131                          stateid_t *stateid, char typemask,
5132                          struct nfs4_ol_stateid **stpp,
5133                          struct nfsd_net *nn)
5134 {
5135         __be32 status;
5136         struct nfs4_stid *s;
5137         struct nfs4_ol_stateid *stp = NULL;
5138
5139         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5140                 seqid, STATEID_VAL(stateid));
5141
5142         *stpp = NULL;
5143         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5144         if (status)
5145                 return status;
5146         stp = openlockstateid(s);
5147         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5148
5149         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5150         if (!status)
5151                 *stpp = stp;
5152         else
5153                 nfs4_put_stid(&stp->st_stid);
5154         return status;
5155 }
5156
5157 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5158                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5159 {
5160         __be32 status;
5161         struct nfs4_openowner *oo;
5162         struct nfs4_ol_stateid *stp;
5163
5164         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5165                                                 NFS4_OPEN_STID, &stp, nn);
5166         if (status)
5167                 return status;
5168         oo = openowner(stp->st_stateowner);
5169         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5170                 mutex_unlock(&stp->st_mutex);
5171                 nfs4_put_stid(&stp->st_stid);
5172                 return nfserr_bad_stateid;
5173         }
5174         *stpp = stp;
5175         return nfs_ok;
5176 }
5177
5178 __be32
5179 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5180                    union nfsd4_op_u *u)
5181 {
5182         struct nfsd4_open_confirm *oc = &u->open_confirm;
5183         __be32 status;
5184         struct nfs4_openowner *oo;
5185         struct nfs4_ol_stateid *stp;
5186         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5187
5188         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5189                         cstate->current_fh.fh_dentry);
5190
5191         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5192         if (status)
5193                 return status;
5194
5195         status = nfs4_preprocess_seqid_op(cstate,
5196                                         oc->oc_seqid, &oc->oc_req_stateid,
5197                                         NFS4_OPEN_STID, &stp, nn);
5198         if (status)
5199                 goto out;
5200         oo = openowner(stp->st_stateowner);
5201         status = nfserr_bad_stateid;
5202         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5203                 mutex_unlock(&stp->st_mutex);
5204                 goto put_stateid;
5205         }
5206         oo->oo_flags |= NFS4_OO_CONFIRMED;
5207         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5208         mutex_unlock(&stp->st_mutex);
5209         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5210                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5211
5212         nfsd4_client_record_create(oo->oo_owner.so_client);
5213         status = nfs_ok;
5214 put_stateid:
5215         nfs4_put_stid(&stp->st_stid);
5216 out:
5217         nfsd4_bump_seqid(cstate, status);
5218         return status;
5219 }
5220
5221 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5222 {
5223         if (!test_access(access, stp))
5224                 return;
5225         nfs4_file_put_access(stp->st_stid.sc_file, access);
5226         clear_access(access, stp);
5227 }
5228
5229 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5230 {
5231         switch (to_access) {
5232         case NFS4_SHARE_ACCESS_READ:
5233                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5234                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5235                 break;
5236         case NFS4_SHARE_ACCESS_WRITE:
5237                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5238                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5239                 break;
5240         case NFS4_SHARE_ACCESS_BOTH:
5241                 break;
5242         default:
5243                 WARN_ON_ONCE(1);
5244         }
5245 }
5246
5247 __be32
5248 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5249                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5250 {
5251         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5252         __be32 status;
5253         struct nfs4_ol_stateid *stp;
5254         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5255
5256         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5257                         cstate->current_fh.fh_dentry);
5258
5259         /* We don't yet support WANT bits: */
5260         if (od->od_deleg_want)
5261                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5262                         od->od_deleg_want);
5263
5264         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5265                                         &od->od_stateid, &stp, nn);
5266         if (status)
5267                 goto out;
5268         status = nfserr_inval;
5269         if (!test_access(od->od_share_access, stp)) {
5270                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5271                         stp->st_access_bmap, od->od_share_access);
5272                 goto put_stateid;
5273         }
5274         if (!test_deny(od->od_share_deny, stp)) {
5275                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5276                         stp->st_deny_bmap, od->od_share_deny);
5277                 goto put_stateid;
5278         }
5279         nfs4_stateid_downgrade(stp, od->od_share_access);
5280         reset_union_bmap_deny(od->od_share_deny, stp);
5281         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5282         status = nfs_ok;
5283 put_stateid:
5284         mutex_unlock(&stp->st_mutex);
5285         nfs4_put_stid(&stp->st_stid);
5286 out:
5287         nfsd4_bump_seqid(cstate, status);
5288         return status;
5289 }
5290
5291 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5292 {
5293         struct nfs4_client *clp = s->st_stid.sc_client;
5294         bool unhashed;
5295         LIST_HEAD(reaplist);
5296
5297         s->st_stid.sc_type = NFS4_CLOSED_STID;
5298         spin_lock(&clp->cl_lock);
5299         unhashed = unhash_open_stateid(s, &reaplist);
5300
5301         if (clp->cl_minorversion) {
5302                 if (unhashed)
5303                         put_ol_stateid_locked(s, &reaplist);
5304                 spin_unlock(&clp->cl_lock);
5305                 free_ol_stateid_reaplist(&reaplist);
5306         } else {
5307                 spin_unlock(&clp->cl_lock);
5308                 free_ol_stateid_reaplist(&reaplist);
5309                 if (unhashed)
5310                         move_to_close_lru(s, clp->net);
5311         }
5312 }
5313
5314 /*
5315  * nfs4_unlock_state() called after encode
5316  */
5317 __be32
5318 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5319                 union nfsd4_op_u *u)
5320 {
5321         struct nfsd4_close *close = &u->close;
5322         __be32 status;
5323         struct nfs4_ol_stateid *stp;
5324         struct net *net = SVC_NET(rqstp);
5325         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5326
5327         dprintk("NFSD: nfsd4_close on file %pd\n",
5328                         cstate->current_fh.fh_dentry);
5329
5330         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5331                                         &close->cl_stateid,
5332                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5333                                         &stp, nn);
5334         nfsd4_bump_seqid(cstate, status);
5335         if (status)
5336                 goto out;
5337         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5338         mutex_unlock(&stp->st_mutex);
5339
5340         nfsd4_close_open_stateid(stp);
5341
5342         /* put reference from nfs4_preprocess_seqid_op */
5343         nfs4_put_stid(&stp->st_stid);
5344 out:
5345         return status;
5346 }
5347
5348 __be32
5349 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5350                   union nfsd4_op_u *u)
5351 {
5352         struct nfsd4_delegreturn *dr = &u->delegreturn;
5353         struct nfs4_delegation *dp;
5354         stateid_t *stateid = &dr->dr_stateid;
5355         struct nfs4_stid *s;
5356         __be32 status;
5357         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5358
5359         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5360                 return status;
5361
5362         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5363         if (status)
5364                 goto out;
5365         dp = delegstateid(s);
5366         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5367         if (status)
5368                 goto put_stateid;
5369
5370         destroy_delegation(dp);
5371 put_stateid:
5372         nfs4_put_stid(&dp->dl_stid);
5373 out:
5374         return status;
5375 }
5376
5377 static inline u64
5378 end_offset(u64 start, u64 len)
5379 {
5380         u64 end;
5381
5382         end = start + len;
5383         return end >= start ? end: NFS4_MAX_UINT64;
5384 }
5385
5386 /* last octet in a range */
5387 static inline u64
5388 last_byte_offset(u64 start, u64 len)
5389 {
5390         u64 end;
5391
5392         WARN_ON_ONCE(!len);
5393         end = start + len;
5394         return end > start ? end - 1: NFS4_MAX_UINT64;
5395 }
5396
5397 /*
5398  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5399  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5400  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5401  * locking, this prevents us from being completely protocol-compliant.  The
5402  * real solution to this problem is to start using unsigned file offsets in
5403  * the VFS, but this is a very deep change!
5404  */
5405 static inline void
5406 nfs4_transform_lock_offset(struct file_lock *lock)
5407 {
5408         if (lock->fl_start < 0)
5409                 lock->fl_start = OFFSET_MAX;
5410         if (lock->fl_end < 0)
5411                 lock->fl_end = OFFSET_MAX;
5412 }
5413
5414 static fl_owner_t
5415 nfsd4_fl_get_owner(fl_owner_t owner)
5416 {
5417         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5418
5419         nfs4_get_stateowner(&lo->lo_owner);
5420         return owner;
5421 }
5422
5423 static void
5424 nfsd4_fl_put_owner(fl_owner_t owner)
5425 {
5426         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5427
5428         if (lo)
5429                 nfs4_put_stateowner(&lo->lo_owner);
5430 }
5431
5432 static void
5433 nfsd4_lm_notify(struct file_lock *fl)
5434 {
5435         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5436         struct net                      *net = lo->lo_owner.so_client->net;
5437         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5438         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5439                                                 struct nfsd4_blocked_lock, nbl_lock);
5440         bool queue = false;
5441
5442         /* An empty list means that something else is going to be using it */
5443         spin_lock(&nn->blocked_locks_lock);
5444         if (!list_empty(&nbl->nbl_list)) {
5445                 list_del_init(&nbl->nbl_list);
5446                 list_del_init(&nbl->nbl_lru);
5447                 queue = true;
5448         }
5449         spin_unlock(&nn->blocked_locks_lock);
5450
5451         if (queue)
5452                 nfsd4_run_cb(&nbl->nbl_cb);
5453 }
5454
5455 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5456         .lm_notify = nfsd4_lm_notify,
5457         .lm_get_owner = nfsd4_fl_get_owner,
5458         .lm_put_owner = nfsd4_fl_put_owner,
5459 };
5460
5461 static inline void
5462 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5463 {
5464         struct nfs4_lockowner *lo;
5465
5466         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5467                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5468                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5469                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5470                 if (!deny->ld_owner.data)
5471                         /* We just don't care that much */
5472                         goto nevermind;
5473                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5474                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5475         } else {
5476 nevermind:
5477                 deny->ld_owner.len = 0;
5478                 deny->ld_owner.data = NULL;
5479                 deny->ld_clientid.cl_boot = 0;
5480                 deny->ld_clientid.cl_id = 0;
5481         }
5482         deny->ld_start = fl->fl_start;
5483         deny->ld_length = NFS4_MAX_UINT64;
5484         if (fl->fl_end != NFS4_MAX_UINT64)
5485                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5486         deny->ld_type = NFS4_READ_LT;
5487         if (fl->fl_type != F_RDLCK)
5488                 deny->ld_type = NFS4_WRITE_LT;
5489 }
5490
5491 static struct nfs4_lockowner *
5492 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5493 {
5494         unsigned int strhashval = ownerstr_hashval(owner);
5495         struct nfs4_stateowner *so;
5496
5497         lockdep_assert_held(&clp->cl_lock);
5498
5499         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5500                             so_strhash) {
5501                 if (so->so_is_open_owner)
5502                         continue;
5503                 if (same_owner_str(so, owner))
5504                         return lockowner(nfs4_get_stateowner(so));
5505         }
5506         return NULL;
5507 }
5508
5509 static struct nfs4_lockowner *
5510 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5511 {
5512         struct nfs4_lockowner *lo;
5513
5514         spin_lock(&clp->cl_lock);
5515         lo = find_lockowner_str_locked(clp, owner);
5516         spin_unlock(&clp->cl_lock);
5517         return lo;
5518 }
5519
5520 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5521 {
5522         unhash_lockowner_locked(lockowner(sop));
5523 }
5524
5525 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5526 {
5527         struct nfs4_lockowner *lo = lockowner(sop);
5528
5529         kmem_cache_free(lockowner_slab, lo);
5530 }
5531
5532 static const struct nfs4_stateowner_operations lockowner_ops = {
5533         .so_unhash =    nfs4_unhash_lockowner,
5534         .so_free =      nfs4_free_lockowner,
5535 };
5536
5537 /*
5538  * Alloc a lock owner structure.
5539  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5540  * occurred.
5541  *
5542  * strhashval = ownerstr_hashval
5543  */
5544 static struct nfs4_lockowner *
5545 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5546                            struct nfs4_ol_stateid *open_stp,
5547                            struct nfsd4_lock *lock)
5548 {
5549         struct nfs4_lockowner *lo, *ret;
5550
5551         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5552         if (!lo)
5553                 return NULL;
5554         INIT_LIST_HEAD(&lo->lo_blocked);
5555         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5556         lo->lo_owner.so_is_open_owner = 0;
5557         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5558         lo->lo_owner.so_ops = &lockowner_ops;
5559         spin_lock(&clp->cl_lock);
5560         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5561         if (ret == NULL) {
5562                 list_add(&lo->lo_owner.so_strhash,
5563                          &clp->cl_ownerstr_hashtbl[strhashval]);
5564                 ret = lo;
5565         } else
5566                 nfs4_free_stateowner(&lo->lo_owner);
5567
5568         spin_unlock(&clp->cl_lock);
5569         return ret;
5570 }
5571
5572 static void
5573 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5574                   struct nfs4_file *fp, struct inode *inode,
5575                   struct nfs4_ol_stateid *open_stp)
5576 {
5577         struct nfs4_client *clp = lo->lo_owner.so_client;
5578
5579         lockdep_assert_held(&clp->cl_lock);
5580
5581         atomic_inc(&stp->st_stid.sc_count);
5582         stp->st_stid.sc_type = NFS4_LOCK_STID;
5583         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5584         get_nfs4_file(fp);
5585         stp->st_stid.sc_file = fp;
5586         stp->st_access_bmap = 0;
5587         stp->st_deny_bmap = open_stp->st_deny_bmap;
5588         stp->st_openstp = open_stp;
5589         mutex_init(&stp->st_mutex);
5590         list_add(&stp->st_locks, &open_stp->st_locks);
5591         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5592         spin_lock(&fp->fi_lock);
5593         list_add(&stp->st_perfile, &fp->fi_stateids);
5594         spin_unlock(&fp->fi_lock);
5595 }
5596
5597 static struct nfs4_ol_stateid *
5598 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5599 {
5600         struct nfs4_ol_stateid *lst;
5601         struct nfs4_client *clp = lo->lo_owner.so_client;
5602
5603         lockdep_assert_held(&clp->cl_lock);
5604
5605         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5606                 if (lst->st_stid.sc_file == fp) {
5607                         atomic_inc(&lst->st_stid.sc_count);
5608                         return lst;
5609                 }
5610         }
5611         return NULL;
5612 }
5613
5614 static struct nfs4_ol_stateid *
5615 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5616                             struct inode *inode, struct nfs4_ol_stateid *ost,
5617                             bool *new)
5618 {
5619         struct nfs4_stid *ns = NULL;
5620         struct nfs4_ol_stateid *lst;
5621         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5622         struct nfs4_client *clp = oo->oo_owner.so_client;
5623
5624         spin_lock(&clp->cl_lock);
5625         lst = find_lock_stateid(lo, fi);
5626         if (lst == NULL) {
5627                 spin_unlock(&clp->cl_lock);
5628                 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5629                 if (ns == NULL)
5630                         return NULL;
5631
5632                 spin_lock(&clp->cl_lock);
5633                 lst = find_lock_stateid(lo, fi);
5634                 if (likely(!lst)) {
5635                         lst = openlockstateid(ns);
5636                         init_lock_stateid(lst, lo, fi, inode, ost);
5637                         ns = NULL;
5638                         *new = true;
5639                 }
5640         }
5641         spin_unlock(&clp->cl_lock);
5642         if (ns)
5643                 nfs4_put_stid(ns);
5644         return lst;
5645 }
5646
5647 static int
5648 check_lock_length(u64 offset, u64 length)
5649 {
5650         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5651                 (length > ~offset)));
5652 }
5653
5654 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5655 {
5656         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5657
5658         lockdep_assert_held(&fp->fi_lock);
5659
5660         if (test_access(access, lock_stp))
5661                 return;
5662         __nfs4_file_get_access(fp, access);
5663         set_access(access, lock_stp);
5664 }
5665
5666 static __be32
5667 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5668                             struct nfs4_ol_stateid *ost,
5669                             struct nfsd4_lock *lock,
5670                             struct nfs4_ol_stateid **plst, bool *new)
5671 {
5672         __be32 status;
5673         struct nfs4_file *fi = ost->st_stid.sc_file;
5674         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5675         struct nfs4_client *cl = oo->oo_owner.so_client;
5676         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5677         struct nfs4_lockowner *lo;
5678         struct nfs4_ol_stateid *lst;
5679         unsigned int strhashval;
5680         bool hashed;
5681
5682         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5683         if (!lo) {
5684                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5685                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5686                 if (lo == NULL)
5687                         return nfserr_jukebox;
5688         } else {
5689                 /* with an existing lockowner, seqids must be the same */
5690                 status = nfserr_bad_seqid;
5691                 if (!cstate->minorversion &&
5692                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5693                         goto out;
5694         }
5695
5696 retry:
5697         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5698         if (lst == NULL) {
5699                 status = nfserr_jukebox;
5700                 goto out;
5701         }
5702
5703         mutex_lock(&lst->st_mutex);
5704
5705         /* See if it's still hashed to avoid race with FREE_STATEID */
5706         spin_lock(&cl->cl_lock);
5707         hashed = !list_empty(&lst->st_perfile);
5708         spin_unlock(&cl->cl_lock);
5709
5710         if (!hashed) {
5711                 mutex_unlock(&lst->st_mutex);
5712                 nfs4_put_stid(&lst->st_stid);
5713                 goto retry;
5714         }
5715         status = nfs_ok;
5716         *plst = lst;
5717 out:
5718         nfs4_put_stateowner(&lo->lo_owner);
5719         return status;
5720 }
5721
5722 /*
5723  *  LOCK operation
5724  */
5725 __be32
5726 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5727            union nfsd4_op_u *u)
5728 {
5729         struct nfsd4_lock *lock = &u->lock;
5730         struct nfs4_openowner *open_sop = NULL;
5731         struct nfs4_lockowner *lock_sop = NULL;
5732         struct nfs4_ol_stateid *lock_stp = NULL;
5733         struct nfs4_ol_stateid *open_stp = NULL;
5734         struct nfs4_file *fp;
5735         struct file *filp = NULL;
5736         struct nfsd4_blocked_lock *nbl = NULL;
5737         struct file_lock *file_lock = NULL;
5738         struct file_lock *conflock = NULL;
5739         __be32 status = 0;
5740         int lkflg;
5741         int err;
5742         bool new = false;
5743         unsigned char fl_type;
5744         unsigned int fl_flags = FL_POSIX;
5745         struct net *net = SVC_NET(rqstp);
5746         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5747
5748         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5749                 (long long) lock->lk_offset,
5750                 (long long) lock->lk_length);
5751
5752         if (check_lock_length(lock->lk_offset, lock->lk_length))
5753                  return nfserr_inval;
5754
5755         if ((status = fh_verify(rqstp, &cstate->current_fh,
5756                                 S_IFREG, NFSD_MAY_LOCK))) {
5757                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5758                 return status;
5759         }
5760
5761         if (lock->lk_is_new) {
5762                 if (nfsd4_has_session(cstate))
5763                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5764                         memcpy(&lock->lk_new_clientid,
5765                                 &cstate->session->se_client->cl_clientid,
5766                                 sizeof(clientid_t));
5767
5768                 status = nfserr_stale_clientid;
5769                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5770                         goto out;
5771
5772                 /* validate and update open stateid and open seqid */
5773                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5774                                         lock->lk_new_open_seqid,
5775                                         &lock->lk_new_open_stateid,
5776                                         &open_stp, nn);
5777                 if (status)
5778                         goto out;
5779                 mutex_unlock(&open_stp->st_mutex);
5780                 open_sop = openowner(open_stp->st_stateowner);
5781                 status = nfserr_bad_stateid;
5782                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5783                                                 &lock->lk_new_clientid))
5784                         goto out;
5785                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5786                                                         &lock_stp, &new);
5787         } else {
5788                 status = nfs4_preprocess_seqid_op(cstate,
5789                                        lock->lk_old_lock_seqid,
5790                                        &lock->lk_old_lock_stateid,
5791                                        NFS4_LOCK_STID, &lock_stp, nn);
5792         }
5793         if (status)
5794                 goto out;
5795         lock_sop = lockowner(lock_stp->st_stateowner);
5796
5797         lkflg = setlkflg(lock->lk_type);
5798         status = nfs4_check_openmode(lock_stp, lkflg);
5799         if (status)
5800                 goto out;
5801
5802         status = nfserr_grace;
5803         if (locks_in_grace(net) && !lock->lk_reclaim)
5804                 goto out;
5805         status = nfserr_no_grace;
5806         if (!locks_in_grace(net) && lock->lk_reclaim)
5807                 goto out;
5808
5809         fp = lock_stp->st_stid.sc_file;
5810         switch (lock->lk_type) {
5811                 case NFS4_READW_LT:
5812                         if (nfsd4_has_session(cstate))
5813                                 fl_flags |= FL_SLEEP;
5814                         /* Fallthrough */
5815                 case NFS4_READ_LT:
5816                         spin_lock(&fp->fi_lock);
5817                         filp = find_readable_file_locked(fp);
5818                         if (filp)
5819                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5820                         spin_unlock(&fp->fi_lock);
5821                         fl_type = F_RDLCK;
5822                         break;
5823                 case NFS4_WRITEW_LT:
5824                         if (nfsd4_has_session(cstate))
5825                                 fl_flags |= FL_SLEEP;
5826                         /* Fallthrough */
5827                 case NFS4_WRITE_LT:
5828                         spin_lock(&fp->fi_lock);
5829                         filp = find_writeable_file_locked(fp);
5830                         if (filp)
5831                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5832                         spin_unlock(&fp->fi_lock);
5833                         fl_type = F_WRLCK;
5834                         break;
5835                 default:
5836                         status = nfserr_inval;
5837                 goto out;
5838         }
5839
5840         if (!filp) {
5841                 status = nfserr_openmode;
5842                 goto out;
5843         }
5844
5845         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
5846         if (!nbl) {
5847                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
5848                 status = nfserr_jukebox;
5849                 goto out;
5850         }
5851
5852         file_lock = &nbl->nbl_lock;
5853         file_lock->fl_type = fl_type;
5854         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5855         file_lock->fl_pid = current->tgid;
5856         file_lock->fl_file = filp;
5857         file_lock->fl_flags = fl_flags;
5858         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5859         file_lock->fl_start = lock->lk_offset;
5860         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5861         nfs4_transform_lock_offset(file_lock);
5862
5863         conflock = locks_alloc_lock();
5864         if (!conflock) {
5865                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5866                 status = nfserr_jukebox;
5867                 goto out;
5868         }
5869
5870         if (fl_flags & FL_SLEEP) {
5871                 nbl->nbl_time = jiffies;
5872                 spin_lock(&nn->blocked_locks_lock);
5873                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
5874                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
5875                 spin_unlock(&nn->blocked_locks_lock);
5876         }
5877
5878         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5879         switch (err) {
5880         case 0: /* success! */
5881                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5882                 status = 0;
5883                 break;
5884         case FILE_LOCK_DEFERRED:
5885                 nbl = NULL;
5886                 /* Fallthrough */
5887         case -EAGAIN:           /* conflock holds conflicting lock */
5888                 status = nfserr_denied;
5889                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5890                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5891                 break;
5892         case -EDEADLK:
5893                 status = nfserr_deadlock;
5894                 break;
5895         default:
5896                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5897                 status = nfserrno(err);
5898                 break;
5899         }
5900 out:
5901         if (nbl) {
5902                 /* dequeue it if we queued it before */
5903                 if (fl_flags & FL_SLEEP) {
5904                         spin_lock(&nn->blocked_locks_lock);
5905                         list_del_init(&nbl->nbl_list);
5906                         list_del_init(&nbl->nbl_lru);
5907                         spin_unlock(&nn->blocked_locks_lock);
5908                 }
5909                 free_blocked_lock(nbl);
5910         }
5911         if (filp)
5912                 fput(filp);
5913         if (lock_stp) {
5914                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5915                 if (cstate->replay_owner &&
5916                     cstate->replay_owner != &lock_sop->lo_owner &&
5917                     seqid_mutating_err(ntohl(status)))
5918                         lock_sop->lo_owner.so_seqid++;
5919
5920                 mutex_unlock(&lock_stp->st_mutex);
5921
5922                 /*
5923                  * If this is a new, never-before-used stateid, and we are
5924                  * returning an error, then just go ahead and release it.
5925                  */
5926                 if (status && new)
5927                         release_lock_stateid(lock_stp);
5928
5929                 nfs4_put_stid(&lock_stp->st_stid);
5930         }
5931         if (open_stp)
5932                 nfs4_put_stid(&open_stp->st_stid);
5933         nfsd4_bump_seqid(cstate, status);
5934         if (conflock)
5935                 locks_free_lock(conflock);
5936         return status;
5937 }
5938
5939 /*
5940  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5941  * so we do a temporary open here just to get an open file to pass to
5942  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5943  * inode operation.)
5944  */
5945 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5946 {
5947         struct file *file;
5948         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5949         if (!err) {
5950                 err = nfserrno(vfs_test_lock(file, lock));
5951                 fput(file);
5952         }
5953         return err;
5954 }
5955
5956 /*
5957  * LOCKT operation
5958  */
5959 __be32
5960 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5961             union nfsd4_op_u *u)
5962 {
5963         struct nfsd4_lockt *lockt = &u->lockt;
5964         struct file_lock *file_lock = NULL;
5965         struct nfs4_lockowner *lo = NULL;
5966         __be32 status;
5967         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5968
5969         if (locks_in_grace(SVC_NET(rqstp)))
5970                 return nfserr_grace;
5971
5972         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5973                  return nfserr_inval;
5974
5975         if (!nfsd4_has_session(cstate)) {
5976                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5977                 if (status)
5978                         goto out;
5979         }
5980
5981         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5982                 goto out;
5983
5984         file_lock = locks_alloc_lock();
5985         if (!file_lock) {
5986                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5987                 status = nfserr_jukebox;
5988                 goto out;
5989         }
5990
5991         switch (lockt->lt_type) {
5992                 case NFS4_READ_LT:
5993                 case NFS4_READW_LT:
5994                         file_lock->fl_type = F_RDLCK;
5995                 break;
5996                 case NFS4_WRITE_LT:
5997                 case NFS4_WRITEW_LT:
5998                         file_lock->fl_type = F_WRLCK;
5999                 break;
6000                 default:
6001                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6002                         status = nfserr_inval;
6003                 goto out;
6004         }
6005
6006         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6007         if (lo)
6008                 file_lock->fl_owner = (fl_owner_t)lo;
6009         file_lock->fl_pid = current->tgid;
6010         file_lock->fl_flags = FL_POSIX;
6011
6012         file_lock->fl_start = lockt->lt_offset;
6013         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6014
6015         nfs4_transform_lock_offset(file_lock);
6016
6017         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6018         if (status)
6019                 goto out;
6020
6021         if (file_lock->fl_type != F_UNLCK) {
6022                 status = nfserr_denied;
6023                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6024         }
6025 out:
6026         if (lo)
6027                 nfs4_put_stateowner(&lo->lo_owner);
6028         if (file_lock)
6029                 locks_free_lock(file_lock);
6030         return status;
6031 }
6032
6033 __be32
6034 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6035             union nfsd4_op_u *u)
6036 {
6037         struct nfsd4_locku *locku = &u->locku;
6038         struct nfs4_ol_stateid *stp;
6039         struct file *filp = NULL;
6040         struct file_lock *file_lock = NULL;
6041         __be32 status;
6042         int err;
6043         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6044
6045         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6046                 (long long) locku->lu_offset,
6047                 (long long) locku->lu_length);
6048
6049         if (check_lock_length(locku->lu_offset, locku->lu_length))
6050                  return nfserr_inval;
6051
6052         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6053                                         &locku->lu_stateid, NFS4_LOCK_STID,
6054                                         &stp, nn);
6055         if (status)
6056                 goto out;
6057         filp = find_any_file(stp->st_stid.sc_file);
6058         if (!filp) {
6059                 status = nfserr_lock_range;
6060                 goto put_stateid;
6061         }
6062         file_lock = locks_alloc_lock();
6063         if (!file_lock) {
6064                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6065                 status = nfserr_jukebox;
6066                 goto fput;
6067         }
6068
6069         file_lock->fl_type = F_UNLCK;
6070         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6071         file_lock->fl_pid = current->tgid;
6072         file_lock->fl_file = filp;
6073         file_lock->fl_flags = FL_POSIX;
6074         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6075         file_lock->fl_start = locku->lu_offset;
6076
6077         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6078                                                 locku->lu_length);
6079         nfs4_transform_lock_offset(file_lock);
6080
6081         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6082         if (err) {
6083                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6084                 goto out_nfserr;
6085         }
6086         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6087 fput:
6088         fput(filp);
6089 put_stateid:
6090         mutex_unlock(&stp->st_mutex);
6091         nfs4_put_stid(&stp->st_stid);
6092 out:
6093         nfsd4_bump_seqid(cstate, status);
6094         if (file_lock)
6095                 locks_free_lock(file_lock);
6096         return status;
6097
6098 out_nfserr:
6099         status = nfserrno(err);
6100         goto fput;
6101 }
6102
6103 /*
6104  * returns
6105  *      true:  locks held by lockowner
6106  *      false: no locks held by lockowner
6107  */
6108 static bool
6109 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6110 {
6111         struct file_lock *fl;
6112         int status = false;
6113         struct file *filp = find_any_file(fp);
6114         struct inode *inode;
6115         struct file_lock_context *flctx;
6116
6117         if (!filp) {
6118                 /* Any valid lock stateid should have some sort of access */
6119                 WARN_ON_ONCE(1);
6120                 return status;
6121         }
6122
6123         inode = file_inode(filp);
6124         flctx = inode->i_flctx;
6125
6126         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6127                 spin_lock(&flctx->flc_lock);
6128                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6129                         if (fl->fl_owner == (fl_owner_t)lowner) {
6130                                 status = true;
6131                                 break;
6132                         }
6133                 }
6134                 spin_unlock(&flctx->flc_lock);
6135         }
6136         fput(filp);
6137         return status;
6138 }
6139
6140 __be32
6141 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6142                         struct nfsd4_compound_state *cstate,
6143                         union nfsd4_op_u *u)
6144 {
6145         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6146         clientid_t *clid = &rlockowner->rl_clientid;
6147         struct nfs4_stateowner *sop;
6148         struct nfs4_lockowner *lo = NULL;
6149         struct nfs4_ol_stateid *stp;
6150         struct xdr_netobj *owner = &rlockowner->rl_owner;
6151         unsigned int hashval = ownerstr_hashval(owner);
6152         __be32 status;
6153         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6154         struct nfs4_client *clp;
6155         LIST_HEAD (reaplist);
6156
6157         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6158                 clid->cl_boot, clid->cl_id);
6159
6160         status = lookup_clientid(clid, cstate, nn);
6161         if (status)
6162                 return status;
6163
6164         clp = cstate->clp;
6165         /* Find the matching lock stateowner */
6166         spin_lock(&clp->cl_lock);
6167         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6168                             so_strhash) {
6169
6170                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6171                         continue;
6172
6173                 /* see if there are still any locks associated with it */
6174                 lo = lockowner(sop);
6175                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6176                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6177                                 status = nfserr_locks_held;
6178                                 spin_unlock(&clp->cl_lock);
6179                                 return status;
6180                         }
6181                 }
6182
6183                 nfs4_get_stateowner(sop);
6184                 break;
6185         }
6186         if (!lo) {
6187                 spin_unlock(&clp->cl_lock);
6188                 return status;
6189         }
6190
6191         unhash_lockowner_locked(lo);
6192         while (!list_empty(&lo->lo_owner.so_stateids)) {
6193                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6194                                        struct nfs4_ol_stateid,
6195                                        st_perstateowner);
6196                 WARN_ON(!unhash_lock_stateid(stp));
6197                 put_ol_stateid_locked(stp, &reaplist);
6198         }
6199         spin_unlock(&clp->cl_lock);
6200         free_ol_stateid_reaplist(&reaplist);
6201         nfs4_put_stateowner(&lo->lo_owner);
6202
6203         return status;
6204 }
6205
6206 static inline struct nfs4_client_reclaim *
6207 alloc_reclaim(void)
6208 {
6209         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6210 }
6211
6212 bool
6213 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6214 {
6215         struct nfs4_client_reclaim *crp;
6216
6217         crp = nfsd4_find_reclaim_client(name, nn);
6218         return (crp && crp->cr_clp);
6219 }
6220
6221 /*
6222  * failure => all reset bets are off, nfserr_no_grace...
6223  */
6224 struct nfs4_client_reclaim *
6225 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6226 {
6227         unsigned int strhashval;
6228         struct nfs4_client_reclaim *crp;
6229
6230         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6231         crp = alloc_reclaim();
6232         if (crp) {
6233                 strhashval = clientstr_hashval(name);
6234                 INIT_LIST_HEAD(&crp->cr_strhash);
6235                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6236                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6237                 crp->cr_clp = NULL;
6238                 nn->reclaim_str_hashtbl_size++;
6239         }
6240         return crp;
6241 }
6242
6243 void
6244 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6245 {
6246         list_del(&crp->cr_strhash);
6247         kfree(crp);
6248         nn->reclaim_str_hashtbl_size--;
6249 }
6250
6251 void
6252 nfs4_release_reclaim(struct nfsd_net *nn)
6253 {
6254         struct nfs4_client_reclaim *crp = NULL;
6255         int i;
6256
6257         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6258                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6259                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6260                                         struct nfs4_client_reclaim, cr_strhash);
6261                         nfs4_remove_reclaim_record(crp, nn);
6262                 }
6263         }
6264         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6265 }
6266
6267 /*
6268  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6269 struct nfs4_client_reclaim *
6270 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6271 {
6272         unsigned int strhashval;
6273         struct nfs4_client_reclaim *crp = NULL;
6274
6275         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6276
6277         strhashval = clientstr_hashval(recdir);
6278         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6279                 if (same_name(crp->cr_recdir, recdir)) {
6280                         return crp;
6281                 }
6282         }
6283         return NULL;
6284 }
6285
6286 /*
6287 * Called from OPEN. Look for clientid in reclaim list.
6288 */
6289 __be32
6290 nfs4_check_open_reclaim(clientid_t *clid,
6291                 struct nfsd4_compound_state *cstate,
6292                 struct nfsd_net *nn)
6293 {
6294         __be32 status;
6295
6296         /* find clientid in conf_id_hashtbl */
6297         status = lookup_clientid(clid, cstate, nn);
6298         if (status)
6299                 return nfserr_reclaim_bad;
6300
6301         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6302                 return nfserr_no_grace;
6303
6304         if (nfsd4_client_record_check(cstate->clp))
6305                 return nfserr_reclaim_bad;
6306
6307         return nfs_ok;
6308 }
6309
6310 #ifdef CONFIG_NFSD_FAULT_INJECTION
6311 static inline void
6312 put_client(struct nfs4_client *clp)
6313 {
6314         atomic_dec(&clp->cl_refcount);
6315 }
6316
6317 static struct nfs4_client *
6318 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6319 {
6320         struct nfs4_client *clp;
6321         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6322                                           nfsd_net_id);
6323
6324         if (!nfsd_netns_ready(nn))
6325                 return NULL;
6326
6327         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6328                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6329                         return clp;
6330         }
6331         return NULL;
6332 }
6333
6334 u64
6335 nfsd_inject_print_clients(void)
6336 {
6337         struct nfs4_client *clp;
6338         u64 count = 0;
6339         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6340                                           nfsd_net_id);
6341         char buf[INET6_ADDRSTRLEN];
6342
6343         if (!nfsd_netns_ready(nn))
6344                 return 0;
6345
6346         spin_lock(&nn->client_lock);
6347         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6348                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6349                 pr_info("NFS Client: %s\n", buf);
6350                 ++count;
6351         }
6352         spin_unlock(&nn->client_lock);
6353
6354         return count;
6355 }
6356
6357 u64
6358 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6359 {
6360         u64 count = 0;
6361         struct nfs4_client *clp;
6362         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6363                                           nfsd_net_id);
6364
6365         if (!nfsd_netns_ready(nn))
6366                 return count;
6367
6368         spin_lock(&nn->client_lock);
6369         clp = nfsd_find_client(addr, addr_size);
6370         if (clp) {
6371                 if (mark_client_expired_locked(clp) == nfs_ok)
6372                         ++count;
6373                 else
6374                         clp = NULL;
6375         }
6376         spin_unlock(&nn->client_lock);
6377
6378         if (clp)
6379                 expire_client(clp);
6380
6381         return count;
6382 }
6383
6384 u64
6385 nfsd_inject_forget_clients(u64 max)
6386 {
6387         u64 count = 0;
6388         struct nfs4_client *clp, *next;
6389         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6390                                                 nfsd_net_id);
6391         LIST_HEAD(reaplist);
6392
6393         if (!nfsd_netns_ready(nn))
6394                 return count;
6395
6396         spin_lock(&nn->client_lock);
6397         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6398                 if (mark_client_expired_locked(clp) == nfs_ok) {
6399                         list_add(&clp->cl_lru, &reaplist);
6400                         if (max != 0 && ++count >= max)
6401                                 break;
6402                 }
6403         }
6404         spin_unlock(&nn->client_lock);
6405
6406         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6407                 expire_client(clp);
6408
6409         return count;
6410 }
6411
6412 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6413                              const char *type)
6414 {
6415         char buf[INET6_ADDRSTRLEN];
6416         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6417         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6418 }
6419
6420 static void
6421 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6422                              struct list_head *collect)
6423 {
6424         struct nfs4_client *clp = lst->st_stid.sc_client;
6425         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6426                                           nfsd_net_id);
6427
6428         if (!collect)
6429                 return;
6430
6431         lockdep_assert_held(&nn->client_lock);
6432         atomic_inc(&clp->cl_refcount);
6433         list_add(&lst->st_locks, collect);
6434 }
6435
6436 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6437                                     struct list_head *collect,
6438                                     bool (*func)(struct nfs4_ol_stateid *))
6439 {
6440         struct nfs4_openowner *oop;
6441         struct nfs4_ol_stateid *stp, *st_next;
6442         struct nfs4_ol_stateid *lst, *lst_next;
6443         u64 count = 0;
6444
6445         spin_lock(&clp->cl_lock);
6446         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6447                 list_for_each_entry_safe(stp, st_next,
6448                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6449                         list_for_each_entry_safe(lst, lst_next,
6450                                         &stp->st_locks, st_locks) {
6451                                 if (func) {
6452                                         if (func(lst))
6453                                                 nfsd_inject_add_lock_to_list(lst,
6454                                                                         collect);
6455                                 }
6456                                 ++count;
6457                                 /*
6458                                  * Despite the fact that these functions deal
6459                                  * with 64-bit integers for "count", we must
6460                                  * ensure that it doesn't blow up the
6461                                  * clp->cl_refcount. Throw a warning if we
6462                                  * start to approach INT_MAX here.
6463                                  */
6464                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6465                                 if (count == max)
6466                                         goto out;
6467                         }
6468                 }
6469         }
6470 out:
6471         spin_unlock(&clp->cl_lock);
6472
6473         return count;
6474 }
6475
6476 static u64
6477 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6478                           u64 max)
6479 {
6480         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6481 }
6482
6483 static u64
6484 nfsd_print_client_locks(struct nfs4_client *clp)
6485 {
6486         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6487         nfsd_print_count(clp, count, "locked files");
6488         return count;
6489 }
6490
6491 u64
6492 nfsd_inject_print_locks(void)
6493 {
6494         struct nfs4_client *clp;
6495         u64 count = 0;
6496         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6497                                                 nfsd_net_id);
6498
6499         if (!nfsd_netns_ready(nn))
6500                 return 0;
6501
6502         spin_lock(&nn->client_lock);
6503         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6504                 count += nfsd_print_client_locks(clp);
6505         spin_unlock(&nn->client_lock);
6506
6507         return count;
6508 }
6509
6510 static void
6511 nfsd_reap_locks(struct list_head *reaplist)
6512 {
6513         struct nfs4_client *clp;
6514         struct nfs4_ol_stateid *stp, *next;
6515
6516         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6517                 list_del_init(&stp->st_locks);
6518                 clp = stp->st_stid.sc_client;
6519                 nfs4_put_stid(&stp->st_stid);
6520                 put_client(clp);
6521         }
6522 }
6523
6524 u64
6525 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6526 {
6527         unsigned int count = 0;
6528         struct nfs4_client *clp;
6529         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6530                                                 nfsd_net_id);
6531         LIST_HEAD(reaplist);
6532
6533         if (!nfsd_netns_ready(nn))
6534                 return count;
6535
6536         spin_lock(&nn->client_lock);
6537         clp = nfsd_find_client(addr, addr_size);
6538         if (clp)
6539                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6540         spin_unlock(&nn->client_lock);
6541         nfsd_reap_locks(&reaplist);
6542         return count;
6543 }
6544
6545 u64
6546 nfsd_inject_forget_locks(u64 max)
6547 {
6548         u64 count = 0;
6549         struct nfs4_client *clp;
6550         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6551                                                 nfsd_net_id);
6552         LIST_HEAD(reaplist);
6553
6554         if (!nfsd_netns_ready(nn))
6555                 return count;
6556
6557         spin_lock(&nn->client_lock);
6558         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6559                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6560                 if (max != 0 && count >= max)
6561                         break;
6562         }
6563         spin_unlock(&nn->client_lock);
6564         nfsd_reap_locks(&reaplist);
6565         return count;
6566 }
6567
6568 static u64
6569 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6570                               struct list_head *collect,
6571                               void (*func)(struct nfs4_openowner *))
6572 {
6573         struct nfs4_openowner *oop, *next;
6574         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6575                                                 nfsd_net_id);
6576         u64 count = 0;
6577
6578         lockdep_assert_held(&nn->client_lock);
6579
6580         spin_lock(&clp->cl_lock);
6581         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6582                 if (func) {
6583                         func(oop);
6584                         if (collect) {
6585                                 atomic_inc(&clp->cl_refcount);
6586                                 list_add(&oop->oo_perclient, collect);
6587                         }
6588                 }
6589                 ++count;
6590                 /*
6591                  * Despite the fact that these functions deal with
6592                  * 64-bit integers for "count", we must ensure that
6593                  * it doesn't blow up the clp->cl_refcount. Throw a
6594                  * warning if we start to approach INT_MAX here.
6595                  */
6596                 WARN_ON_ONCE(count == (INT_MAX / 2));
6597                 if (count == max)
6598                         break;
6599         }
6600         spin_unlock(&clp->cl_lock);
6601
6602         return count;
6603 }
6604
6605 static u64
6606 nfsd_print_client_openowners(struct nfs4_client *clp)
6607 {
6608         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6609
6610         nfsd_print_count(clp, count, "openowners");
6611         return count;
6612 }
6613
6614 static u64
6615 nfsd_collect_client_openowners(struct nfs4_client *clp,
6616                                struct list_head *collect, u64 max)
6617 {
6618         return nfsd_foreach_client_openowner(clp, max, collect,
6619                                                 unhash_openowner_locked);
6620 }
6621
6622 u64
6623 nfsd_inject_print_openowners(void)
6624 {
6625         struct nfs4_client *clp;
6626         u64 count = 0;
6627         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6628                                                 nfsd_net_id);
6629
6630         if (!nfsd_netns_ready(nn))
6631                 return 0;
6632
6633         spin_lock(&nn->client_lock);
6634         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6635                 count += nfsd_print_client_openowners(clp);
6636         spin_unlock(&nn->client_lock);
6637
6638         return count;
6639 }
6640
6641 static void
6642 nfsd_reap_openowners(struct list_head *reaplist)
6643 {
6644         struct nfs4_client *clp;
6645         struct nfs4_openowner *oop, *next;
6646
6647         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6648                 list_del_init(&oop->oo_perclient);
6649                 clp = oop->oo_owner.so_client;
6650                 release_openowner(oop);
6651                 put_client(clp);
6652         }
6653 }
6654
6655 u64
6656 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6657                                      size_t addr_size)
6658 {
6659         unsigned int count = 0;
6660         struct nfs4_client *clp;
6661         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6662                                                 nfsd_net_id);
6663         LIST_HEAD(reaplist);
6664
6665         if (!nfsd_netns_ready(nn))
6666                 return count;
6667
6668         spin_lock(&nn->client_lock);
6669         clp = nfsd_find_client(addr, addr_size);
6670         if (clp)
6671                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6672         spin_unlock(&nn->client_lock);
6673         nfsd_reap_openowners(&reaplist);
6674         return count;
6675 }
6676
6677 u64
6678 nfsd_inject_forget_openowners(u64 max)
6679 {
6680         u64 count = 0;
6681         struct nfs4_client *clp;
6682         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6683                                                 nfsd_net_id);
6684         LIST_HEAD(reaplist);
6685
6686         if (!nfsd_netns_ready(nn))
6687                 return count;
6688
6689         spin_lock(&nn->client_lock);
6690         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6691                 count += nfsd_collect_client_openowners(clp, &reaplist,
6692                                                         max - count);
6693                 if (max != 0 && count >= max)
6694                         break;
6695         }
6696         spin_unlock(&nn->client_lock);
6697         nfsd_reap_openowners(&reaplist);
6698         return count;
6699 }
6700
6701 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6702                                      struct list_head *victims)
6703 {
6704         struct nfs4_delegation *dp, *next;
6705         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6706                                                 nfsd_net_id);
6707         u64 count = 0;
6708
6709         lockdep_assert_held(&nn->client_lock);
6710
6711         spin_lock(&state_lock);
6712         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6713                 if (victims) {
6714                         /*
6715                          * It's not safe to mess with delegations that have a
6716                          * non-zero dl_time. They might have already been broken
6717                          * and could be processed by the laundromat outside of
6718                          * the state_lock. Just leave them be.
6719                          */
6720                         if (dp->dl_time != 0)
6721                                 continue;
6722
6723                         atomic_inc(&clp->cl_refcount);
6724                         WARN_ON(!unhash_delegation_locked(dp));
6725                         list_add(&dp->dl_recall_lru, victims);
6726                 }
6727                 ++count;
6728                 /*
6729                  * Despite the fact that these functions deal with
6730                  * 64-bit integers for "count", we must ensure that
6731                  * it doesn't blow up the clp->cl_refcount. Throw a
6732                  * warning if we start to approach INT_MAX here.
6733                  */
6734                 WARN_ON_ONCE(count == (INT_MAX / 2));
6735                 if (count == max)
6736                         break;
6737         }
6738         spin_unlock(&state_lock);
6739         return count;
6740 }
6741
6742 static u64
6743 nfsd_print_client_delegations(struct nfs4_client *clp)
6744 {
6745         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6746
6747         nfsd_print_count(clp, count, "delegations");
6748         return count;
6749 }
6750
6751 u64
6752 nfsd_inject_print_delegations(void)
6753 {
6754         struct nfs4_client *clp;
6755         u64 count = 0;
6756         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6757                                                 nfsd_net_id);
6758
6759         if (!nfsd_netns_ready(nn))
6760                 return 0;
6761
6762         spin_lock(&nn->client_lock);
6763         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6764                 count += nfsd_print_client_delegations(clp);
6765         spin_unlock(&nn->client_lock);
6766
6767         return count;
6768 }
6769
6770 static void
6771 nfsd_forget_delegations(struct list_head *reaplist)
6772 {
6773         struct nfs4_client *clp;
6774         struct nfs4_delegation *dp, *next;
6775
6776         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6777                 list_del_init(&dp->dl_recall_lru);
6778                 clp = dp->dl_stid.sc_client;
6779                 revoke_delegation(dp);
6780                 put_client(clp);
6781         }
6782 }
6783
6784 u64
6785 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6786                                       size_t addr_size)
6787 {
6788         u64 count = 0;
6789         struct nfs4_client *clp;
6790         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6791                                                 nfsd_net_id);
6792         LIST_HEAD(reaplist);
6793
6794         if (!nfsd_netns_ready(nn))
6795                 return count;
6796
6797         spin_lock(&nn->client_lock);
6798         clp = nfsd_find_client(addr, addr_size);
6799         if (clp)
6800                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6801         spin_unlock(&nn->client_lock);
6802
6803         nfsd_forget_delegations(&reaplist);
6804         return count;
6805 }
6806
6807 u64
6808 nfsd_inject_forget_delegations(u64 max)
6809 {
6810         u64 count = 0;
6811         struct nfs4_client *clp;
6812         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6813                                                 nfsd_net_id);
6814         LIST_HEAD(reaplist);
6815
6816         if (!nfsd_netns_ready(nn))
6817                 return count;
6818
6819         spin_lock(&nn->client_lock);
6820         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6821                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6822                 if (max != 0 && count >= max)
6823                         break;
6824         }
6825         spin_unlock(&nn->client_lock);
6826         nfsd_forget_delegations(&reaplist);
6827         return count;
6828 }
6829
6830 static void
6831 nfsd_recall_delegations(struct list_head *reaplist)
6832 {
6833         struct nfs4_client *clp;
6834         struct nfs4_delegation *dp, *next;
6835
6836         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6837                 list_del_init(&dp->dl_recall_lru);
6838                 clp = dp->dl_stid.sc_client;
6839                 /*
6840                  * We skipped all entries that had a zero dl_time before,
6841                  * so we can now reset the dl_time back to 0. If a delegation
6842                  * break comes in now, then it won't make any difference since
6843                  * we're recalling it either way.
6844                  */
6845                 spin_lock(&state_lock);
6846                 dp->dl_time = 0;
6847                 spin_unlock(&state_lock);
6848                 nfsd_break_one_deleg(dp);
6849                 put_client(clp);
6850         }
6851 }
6852
6853 u64
6854 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6855                                       size_t addr_size)
6856 {
6857         u64 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_find_all_delegations(clp, 0, &reaplist);
6870         spin_unlock(&nn->client_lock);
6871
6872         nfsd_recall_delegations(&reaplist);
6873         return count;
6874 }
6875
6876 u64
6877 nfsd_inject_recall_delegations(u64 max)
6878 {
6879         u64 count = 0;
6880         struct nfs4_client *clp, *next;
6881         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6882                                                 nfsd_net_id);
6883         LIST_HEAD(reaplist);
6884
6885         if (!nfsd_netns_ready(nn))
6886                 return count;
6887
6888         spin_lock(&nn->client_lock);
6889         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6890                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6891                 if (max != 0 && ++count >= max)
6892                         break;
6893         }
6894         spin_unlock(&nn->client_lock);
6895         nfsd_recall_delegations(&reaplist);
6896         return count;
6897 }
6898 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6899
6900 /*
6901  * Since the lifetime of a delegation isn't limited to that of an open, a
6902  * client may quite reasonably hang on to a delegation as long as it has
6903  * the inode cached.  This becomes an obvious problem the first time a
6904  * client's inode cache approaches the size of the server's total memory.
6905  *
6906  * For now we avoid this problem by imposing a hard limit on the number
6907  * of delegations, which varies according to the server's memory size.
6908  */
6909 static void
6910 set_max_delegations(void)
6911 {
6912         /*
6913          * Allow at most 4 delegations per megabyte of RAM.  Quick
6914          * estimates suggest that in the worst case (where every delegation
6915          * is for a different inode), a delegation could take about 1.5K,
6916          * giving a worst case usage of about 6% of memory.
6917          */
6918         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6919 }
6920
6921 static int nfs4_state_create_net(struct net *net)
6922 {
6923         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6924         int i;
6925
6926         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6927                         CLIENT_HASH_SIZE, GFP_KERNEL);
6928         if (!nn->conf_id_hashtbl)
6929                 goto err;
6930         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6931                         CLIENT_HASH_SIZE, GFP_KERNEL);
6932         if (!nn->unconf_id_hashtbl)
6933                 goto err_unconf_id;
6934         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6935                         SESSION_HASH_SIZE, GFP_KERNEL);
6936         if (!nn->sessionid_hashtbl)
6937                 goto err_sessionid;
6938
6939         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6940                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6941                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6942         }
6943         for (i = 0; i < SESSION_HASH_SIZE; i++)
6944                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6945         nn->conf_name_tree = RB_ROOT;
6946         nn->unconf_name_tree = RB_ROOT;
6947         INIT_LIST_HEAD(&nn->client_lru);
6948         INIT_LIST_HEAD(&nn->close_lru);
6949         INIT_LIST_HEAD(&nn->del_recall_lru);
6950         spin_lock_init(&nn->client_lock);
6951
6952         spin_lock_init(&nn->blocked_locks_lock);
6953         INIT_LIST_HEAD(&nn->blocked_locks_lru);
6954
6955         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6956         get_net(net);
6957
6958         return 0;
6959
6960 err_sessionid:
6961         kfree(nn->unconf_id_hashtbl);
6962 err_unconf_id:
6963         kfree(nn->conf_id_hashtbl);
6964 err:
6965         return -ENOMEM;
6966 }
6967
6968 static void
6969 nfs4_state_destroy_net(struct net *net)
6970 {
6971         int i;
6972         struct nfs4_client *clp = NULL;
6973         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6974
6975         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6976                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6977                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6978                         destroy_client(clp);
6979                 }
6980         }
6981
6982         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6983                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6984                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6985                         destroy_client(clp);
6986                 }
6987         }
6988
6989         kfree(nn->sessionid_hashtbl);
6990         kfree(nn->unconf_id_hashtbl);
6991         kfree(nn->conf_id_hashtbl);
6992         put_net(net);
6993 }
6994
6995 int
6996 nfs4_state_start_net(struct net *net)
6997 {
6998         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6999         int ret;
7000
7001         ret = nfs4_state_create_net(net);
7002         if (ret)
7003                 return ret;
7004         nn->boot_time = get_seconds();
7005         nn->grace_ended = false;
7006         nn->nfsd4_manager.block_opens = true;
7007         locks_start_grace(net, &nn->nfsd4_manager);
7008         nfsd4_client_tracking_init(net);
7009         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
7010                nn->nfsd4_grace, net);
7011         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7012         return 0;
7013 }
7014
7015 /* initialization to perform when the nfsd service is started: */
7016
7017 int
7018 nfs4_state_start(void)
7019 {
7020         int ret;
7021
7022         ret = set_callback_cred();
7023         if (ret)
7024                 return ret;
7025
7026         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7027         if (laundry_wq == NULL) {
7028                 ret = -ENOMEM;
7029                 goto out_cleanup_cred;
7030         }
7031         ret = nfsd4_create_callback_queue();
7032         if (ret)
7033                 goto out_free_laundry;
7034
7035         set_max_delegations();
7036         return 0;
7037
7038 out_free_laundry:
7039         destroy_workqueue(laundry_wq);
7040 out_cleanup_cred:
7041         cleanup_callback_cred();
7042         return ret;
7043 }
7044
7045 void
7046 nfs4_state_shutdown_net(struct net *net)
7047 {
7048         struct nfs4_delegation *dp = NULL;
7049         struct list_head *pos, *next, reaplist;
7050         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7051         struct nfsd4_blocked_lock *nbl;
7052
7053         cancel_delayed_work_sync(&nn->laundromat_work);
7054         locks_end_grace(&nn->nfsd4_manager);
7055
7056         INIT_LIST_HEAD(&reaplist);
7057         spin_lock(&state_lock);
7058         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7059                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7060                 WARN_ON(!unhash_delegation_locked(dp));
7061                 list_add(&dp->dl_recall_lru, &reaplist);
7062         }
7063         spin_unlock(&state_lock);
7064         list_for_each_safe(pos, next, &reaplist) {
7065                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7066                 list_del_init(&dp->dl_recall_lru);
7067                 put_clnt_odstate(dp->dl_clnt_odstate);
7068                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7069                 nfs4_put_stid(&dp->dl_stid);
7070         }
7071
7072         BUG_ON(!list_empty(&reaplist));
7073         spin_lock(&nn->blocked_locks_lock);
7074         while (!list_empty(&nn->blocked_locks_lru)) {
7075                 nbl = list_first_entry(&nn->blocked_locks_lru,
7076                                         struct nfsd4_blocked_lock, nbl_lru);
7077                 list_move(&nbl->nbl_lru, &reaplist);
7078                 list_del_init(&nbl->nbl_list);
7079         }
7080         spin_unlock(&nn->blocked_locks_lock);
7081
7082         while (!list_empty(&reaplist)) {
7083                 nbl = list_first_entry(&nn->blocked_locks_lru,
7084                                         struct nfsd4_blocked_lock, nbl_lru);
7085                 list_del_init(&nbl->nbl_lru);
7086                 posix_unblock_lock(&nbl->nbl_lock);
7087                 free_blocked_lock(nbl);
7088         }
7089
7090         nfsd4_client_tracking_exit(net);
7091         nfs4_state_destroy_net(net);
7092 }
7093
7094 void
7095 nfs4_state_shutdown(void)
7096 {
7097         destroy_workqueue(laundry_wq);
7098         nfsd4_destroy_callback_queue();
7099         cleanup_callback_cred();
7100 }
7101
7102 static void
7103 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7104 {
7105         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7106                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7107 }
7108
7109 static void
7110 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7111 {
7112         if (cstate->minorversion) {
7113                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7114                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7115         }
7116 }
7117
7118 void
7119 clear_current_stateid(struct nfsd4_compound_state *cstate)
7120 {
7121         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7122 }
7123
7124 /*
7125  * functions to set current state id
7126  */
7127 void
7128 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7129                 union nfsd4_op_u *u)
7130 {
7131         put_stateid(cstate, &u->open_downgrade.od_stateid);
7132 }
7133
7134 void
7135 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7136                 union nfsd4_op_u *u)
7137 {
7138         put_stateid(cstate, &u->open.op_stateid);
7139 }
7140
7141 void
7142 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7143                 union nfsd4_op_u *u)
7144 {
7145         put_stateid(cstate, &u->close.cl_stateid);
7146 }
7147
7148 void
7149 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7150                 union nfsd4_op_u *u)
7151 {
7152         put_stateid(cstate, &u->lock.lk_resp_stateid);
7153 }
7154
7155 /*
7156  * functions to consume current state id
7157  */
7158
7159 void
7160 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7161                 union nfsd4_op_u *u)
7162 {
7163         get_stateid(cstate, &u->open_downgrade.od_stateid);
7164 }
7165
7166 void
7167 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7168                 union nfsd4_op_u *u)
7169 {
7170         get_stateid(cstate, &u->delegreturn.dr_stateid);
7171 }
7172
7173 void
7174 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7175                 union nfsd4_op_u *u)
7176 {
7177         get_stateid(cstate, &u->free_stateid.fr_stateid);
7178 }
7179
7180 void
7181 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7182                 union nfsd4_op_u *u)
7183 {
7184         get_stateid(cstate, &u->setattr.sa_stateid);
7185 }
7186
7187 void
7188 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7189                 union nfsd4_op_u *u)
7190 {
7191         get_stateid(cstate, &u->close.cl_stateid);
7192 }
7193
7194 void
7195 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7196                 union nfsd4_op_u *u)
7197 {
7198         get_stateid(cstate, &u->locku.lu_stateid);
7199 }
7200
7201 void
7202 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7203                 union nfsd4_op_u *u)
7204 {
7205         get_stateid(cstate, &u->read.rd_stateid);
7206 }
7207
7208 void
7209 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7210                 union nfsd4_op_u *u)
7211 {
7212         get_stateid(cstate, &u->write.wr_stateid);
7213 }