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,
 637                                          struct kmem_cache *slab)
 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         stid->sc_client = cl;
 654         stid->sc_stateid.si_opaque.so_id = new_id;
 655         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 656         /* Will be incremented before return to client: */
 657         atomic_set(&stid->sc_count, 1);
 658         spin_lock_init(&stid->sc_lock);
 659
 660         /*
 661          * It shouldn't be a problem to reuse an opaque stateid value.
 662          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 663          * example, a stray write retransmission could be accepted by
 664          * the server when it should have been rejected.  Therefore,
 665          * adopt a trick from the sctp code to attempt to maximize the
 666          * amount of time until an id is reused, by ensuring they always
 667          * "increase" (mod INT_MAX):
 668          */
 669         return stid;
 670 out_free:
 671         kmem_cache_free(slab, stid);
 672         return NULL;
 673 }
 674
 675 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 676 {
 677         struct nfs4_stid *stid;
 678         struct nfs4_ol_stateid *stp;
 679
 680         stid = nfs4_alloc_stid(clp, stateid_slab);
 681         if (!stid)
 682                 return NULL;
 683
 684         stp = openlockstateid(stid);
 685         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 686         return stp;
 687 }
 688
 689 static void nfs4_free_deleg(struct nfs4_stid *stid)
 690 {
 691         kmem_cache_free(deleg_slab, stid);
 692         atomic_long_dec(&num_delegations);
 693 }
 694
 695 /*
 696  * When we recall a delegation, we should be careful not to hand it
 697  * out again straight away.
 698  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 699  * in which the filehandles of recalled delegations are "stored".
 700  * If a filehandle appear in either filter, a delegation is blocked.
 701  * When a delegation is recalled, the filehandle is stored in the "new"
 702  * filter.
 703  * Every 30 seconds we swap the filters and clear the "new" one,
 704  * unless both are empty of course.
 705  *
 706  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 707  * low 3 bytes as hash-table indices.
 708  *
 709  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 710  * is used to manage concurrent access.  Testing does not need the lock
 711  * except when swapping the two filters.
 712  */
 713 static DEFINE_SPINLOCK(blocked_delegations_lock);
 714 static struct bloom_pair {
 715         int     entries, old_entries;
 716         time_t  swap_time;
 717         int     new; /* index into 'set' */
 718         DECLARE_BITMAP(set[2], 256);
 719 } blocked_delegations;
 720
 721 static int delegation_blocked(struct knfsd_fh *fh)
 722 {
 723         u32 hash;
 724         struct bloom_pair *bd = &blocked_delegations;
 725
 726         if (bd->entries == 0)
 727                 return 0;
 728         if (seconds_since_boot() - bd->swap_time > 30) {
 729                 spin_lock(&blocked_delegations_lock);
 730                 if (seconds_since_boot() - bd->swap_time > 30) {
 731                         bd->entries -= bd->old_entries;
 732                         bd->old_entries = bd->entries;
 733                         memset(bd->set[bd->new], 0,
 734                                sizeof(bd->set[0]));
 735                         bd->new = 1-bd->new;
 736                         bd->swap_time = seconds_since_boot();
 737                 }
 738                 spin_unlock(&blocked_delegations_lock);
 739         }
 740         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 741         if (test_bit(hash&255, bd->set[0]) &&
 742             test_bit((hash>>8)&255, bd->set[0]) &&
 743             test_bit((hash>>16)&255, bd->set[0]))
 744                 return 1;
 745
 746         if (test_bit(hash&255, bd->set[1]) &&
 747             test_bit((hash>>8)&255, bd->set[1]) &&
 748             test_bit((hash>>16)&255, bd->set[1]))
 749                 return 1;
 750
 751         return 0;
 752 }
 753
 754 static void block_delegations(struct knfsd_fh *fh)
 755 {
 756         u32 hash;
 757         struct bloom_pair *bd = &blocked_delegations;
 758
 759         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 760
 761         spin_lock(&blocked_delegations_lock);
 762         __set_bit(hash&255, bd->set[bd->new]);
 763         __set_bit((hash>>8)&255, bd->set[bd->new]);
 764         __set_bit((hash>>16)&255, bd->set[bd->new]);
 765         if (bd->entries == 0)
 766                 bd->swap_time = seconds_since_boot();
 767         bd->entries += 1;
 768         spin_unlock(&blocked_delegations_lock);
 769 }
 770
 771 static struct nfs4_delegation *
 772 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
 773                  struct nfs4_clnt_odstate *odstate)
 774 {
 775         struct nfs4_delegation *dp;
 776         long n;
 777
 778         dprintk("NFSD alloc_init_deleg\n");
 779         n = atomic_long_inc_return(&num_delegations);
 780         if (n < 0 || n > max_delegations)
 781                 goto out_dec;
 782         if (delegation_blocked(&current_fh->fh_handle))
 783                 goto out_dec;
 784         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 785         if (dp == NULL)
 786                 goto out_dec;
 787
 788         dp->dl_stid.sc_free = nfs4_free_deleg;
 789         /*
 790          * delegation seqid's are never incremented.  The 4.1 special
 791          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 792          * 0 anyway just for consistency and use 1:
 793          */
 794         dp->dl_stid.sc_stateid.si_generation = 1;
 795         INIT_LIST_HEAD(&dp->dl_perfile);
 796         INIT_LIST_HEAD(&dp->dl_perclnt);
 797         INIT_LIST_HEAD(&dp->dl_recall_lru);
 798         dp->dl_clnt_odstate = odstate;
 799         get_clnt_odstate(odstate);
 800         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 801         dp->dl_retries = 1;
 802         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 803                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 804         return dp;
 805 out_dec:
 806         atomic_long_dec(&num_delegations);
 807         return NULL;
 808 }
 809
 810 void
 811 nfs4_put_stid(struct nfs4_stid *s)
 812 {
 813         struct nfs4_file *fp = s->sc_file;
 814         struct nfs4_client *clp = s->sc_client;
 815
 816         might_lock(&clp->cl_lock);
 817
 818         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 819                 wake_up_all(&close_wq);
 820                 return;
 821         }
 822         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 823         spin_unlock(&clp->cl_lock);
 824         s->sc_free(s);
 825         if (fp)
 826                 put_nfs4_file(fp);
 827 }
 828
 829 void
 830 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 831 {
 832         stateid_t *src = &stid->sc_stateid;
 833
 834         spin_lock(&stid->sc_lock);
 835         if (unlikely(++src->si_generation == 0))
 836                 src->si_generation = 1;
 837         memcpy(dst, src, sizeof(*dst));
 838         spin_unlock(&stid->sc_lock);
 839 }
 840
 841 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 842 {
 843         struct file *filp = NULL;
 844
 845         spin_lock(&fp->fi_lock);
 846         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 847                 swap(filp, fp->fi_deleg_file);
 848         spin_unlock(&fp->fi_lock);
 849
 850         if (filp) {
 851                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 852                 fput(filp);
 853         }
 854 }
 855
 856 void nfs4_unhash_stid(struct nfs4_stid *s)
 857 {
 858         s->sc_type = 0;
 859 }
 860
 861 /**
 862  * nfs4_get_existing_delegation - Discover if this delegation already exists
 863  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 864  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 865  *
 866  * Return:
 867  *      On success: NULL if an existing delegation was not found.
 868  *
 869  *      On error: -EAGAIN if one was previously granted to this nfs4_client
 870  *                 for this nfs4_file.
 871  *
 872  */
 873
 874 static int
 875 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
 876 {
 877         struct nfs4_delegation *searchdp = NULL;
 878         struct nfs4_client *searchclp = NULL;
 879
 880         lockdep_assert_held(&state_lock);
 881         lockdep_assert_held(&fp->fi_lock);
 882
 883         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 884                 searchclp = searchdp->dl_stid.sc_client;
 885                 if (clp == searchclp) {
 886                         return -EAGAIN;
 887                 }
 888         }
 889         return 0;
 890 }
 891
 892 /**
 893  * hash_delegation_locked - Add a delegation to the appropriate lists
 894  * @dp:     a pointer to the nfs4_delegation we are adding.
 895  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 896  *
 897  * Return:
 898  *      On success: NULL if the delegation was successfully hashed.
 899  *
 900  *      On error: -EAGAIN if one was previously granted to this
 901  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 902  *
 903  */
 904
 905 static int
 906 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 907 {
 908         int status;
 909         struct nfs4_client *clp = dp->dl_stid.sc_client;
 910
 911         lockdep_assert_held(&state_lock);
 912         lockdep_assert_held(&fp->fi_lock);
 913
 914         status = nfs4_get_existing_delegation(clp, fp);
 915         if (status)
 916                 return status;
 917         ++fp->fi_delegees;
 918         atomic_inc(&dp->dl_stid.sc_count);
 919         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 920         list_add(&dp->dl_perfile, &fp->fi_delegations);
 921         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 922         return 0;
 923 }
 924
 925 static bool
 926 unhash_delegation_locked(struct nfs4_delegation *dp)
 927 {
 928         struct nfs4_file *fp = dp->dl_stid.sc_file;
 929
 930         lockdep_assert_held(&state_lock);
 931
 932         if (list_empty(&dp->dl_perfile))
 933                 return false;
 934
 935         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 936         /* Ensure that deleg break won't try to requeue it */
 937         ++dp->dl_time;
 938         spin_lock(&fp->fi_lock);
 939         list_del_init(&dp->dl_perclnt);
 940         list_del_init(&dp->dl_recall_lru);
 941         list_del_init(&dp->dl_perfile);
 942         spin_unlock(&fp->fi_lock);
 943         return true;
 944 }
 945
 946 static void destroy_delegation(struct nfs4_delegation *dp)
 947 {
 948         bool unhashed;
 949
 950         spin_lock(&state_lock);
 951         unhashed = unhash_delegation_locked(dp);
 952         spin_unlock(&state_lock);
 953         if (unhashed) {
 954                 put_clnt_odstate(dp->dl_clnt_odstate);
 955                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 956                 nfs4_put_stid(&dp->dl_stid);
 957         }
 958 }
 959
 960 static void revoke_delegation(struct nfs4_delegation *dp)
 961 {
 962         struct nfs4_client *clp = dp->dl_stid.sc_client;
 963
 964         WARN_ON(!list_empty(&dp->dl_recall_lru));
 965
 966         put_clnt_odstate(dp->dl_clnt_odstate);
 967         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 968
 969         if (clp->cl_minorversion == 0)
 970                 nfs4_put_stid(&dp->dl_stid);
 971         else {
 972                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 973                 spin_lock(&clp->cl_lock);
 974                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 975                 spin_unlock(&clp->cl_lock);
 976         }
 977 }
 978
 979 /*
 980  * SETCLIENTID state
 981  */
 982
 983 static unsigned int clientid_hashval(u32 id)
 984 {
 985         return id & CLIENT_HASH_MASK;
 986 }
 987
 988 static unsigned int clientstr_hashval(const char *name)
 989 {
 990         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 991 }
 992
 993 /*
 994  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 995  * st_{access,deny}_bmap field of the stateid, in order to track not
 996  * only what share bits are currently in force, but also what
 997  * combinations of share bits previous opens have used.  This allows us
 998  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 999  * return an error if the client attempt to downgrade to a combination
1000  * of share bits not explicable by closing some of its previous opens.
1001  *
1002  * XXX: This enforcement is actually incomplete, since we don't keep
1003  * track of access/deny bit combinations; so, e.g., we allow:
1004  *
1005  *      OPEN allow read, deny write
1006  *      OPEN allow both, deny none
1007  *      DOWNGRADE allow read, deny none
1008  *
1009  * which we should reject.
1010  */
1011 static unsigned int
1012 bmap_to_share_mode(unsigned long bmap) {
1013         int i;
1014         unsigned int access = 0;
1015
1016         for (i = 1; i < 4; i++) {
1017                 if (test_bit(i, &bmap))
1018                         access |= i;
1019         }
1020         return access;
1021 }
1022
1023 /* set share access for a given stateid */
1024 static inline void
1025 set_access(u32 access, struct nfs4_ol_stateid *stp)
1026 {
1027         unsigned char mask = 1 << access;
1028
1029         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1030         stp->st_access_bmap |= mask;
1031 }
1032
1033 /* clear share access for a given stateid */
1034 static inline void
1035 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1036 {
1037         unsigned char mask = 1 << access;
1038
1039         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1040         stp->st_access_bmap &= ~mask;
1041 }
1042
1043 /* test whether a given stateid has access */
1044 static inline bool
1045 test_access(u32 access, struct nfs4_ol_stateid *stp)
1046 {
1047         unsigned char mask = 1 << access;
1048
1049         return (bool)(stp->st_access_bmap & mask);
1050 }
1051
1052 /* set share deny for a given stateid */
1053 static inline void
1054 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1055 {
1056         unsigned char mask = 1 << deny;
1057
1058         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1059         stp->st_deny_bmap |= mask;
1060 }
1061
1062 /* clear share deny for a given stateid */
1063 static inline void
1064 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1065 {
1066         unsigned char mask = 1 << deny;
1067
1068         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1069         stp->st_deny_bmap &= ~mask;
1070 }
1071
1072 /* test whether a given stateid is denying specific access */
1073 static inline bool
1074 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1075 {
1076         unsigned char mask = 1 << deny;
1077
1078         return (bool)(stp->st_deny_bmap & mask);
1079 }
1080
1081 static int nfs4_access_to_omode(u32 access)
1082 {
1083         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1084         case NFS4_SHARE_ACCESS_READ:
1085                 return O_RDONLY;
1086         case NFS4_SHARE_ACCESS_WRITE:
1087                 return O_WRONLY;
1088         case NFS4_SHARE_ACCESS_BOTH:
1089                 return O_RDWR;
1090         }
1091         WARN_ON_ONCE(1);
1092         return O_RDONLY;
1093 }
1094
1095 /*
1096  * A stateid that had a deny mode associated with it is being released
1097  * or downgraded. Recalculate the deny mode on the file.
1098  */
1099 static void
1100 recalculate_deny_mode(struct nfs4_file *fp)
1101 {
1102         struct nfs4_ol_stateid *stp;
1103
1104         spin_lock(&fp->fi_lock);
1105         fp->fi_share_deny = 0;
1106         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1107                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1108         spin_unlock(&fp->fi_lock);
1109 }
1110
1111 static void
1112 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1113 {
1114         int i;
1115         bool change = false;
1116
1117         for (i = 1; i < 4; i++) {
1118                 if ((i & deny) != i) {
1119                         change = true;
1120                         clear_deny(i, stp);
1121                 }
1122         }
1123
1124         /* Recalculate per-file deny mode if there was a change */
1125         if (change)
1126                 recalculate_deny_mode(stp->st_stid.sc_file);
1127 }
1128
1129 /* release all access and file references for a given stateid */
1130 static void
1131 release_all_access(struct nfs4_ol_stateid *stp)
1132 {
1133         int i;
1134         struct nfs4_file *fp = stp->st_stid.sc_file;
1135
1136         if (fp && stp->st_deny_bmap != 0)
1137                 recalculate_deny_mode(fp);
1138
1139         for (i = 1; i < 4; i++) {
1140                 if (test_access(i, stp))
1141                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1142                 clear_access(i, stp);
1143         }
1144 }
1145
1146 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1147 {
1148         kfree(sop->so_owner.data);
1149         sop->so_ops->so_free(sop);
1150 }
1151
1152 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1153 {
1154         struct nfs4_client *clp = sop->so_client;
1155
1156         might_lock(&clp->cl_lock);
1157
1158         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1159                 return;
1160         sop->so_ops->so_unhash(sop);
1161         spin_unlock(&clp->cl_lock);
1162         nfs4_free_stateowner(sop);
1163 }
1164
1165 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1166 {
1167         struct nfs4_file *fp = stp->st_stid.sc_file;
1168
1169         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1170
1171         if (list_empty(&stp->st_perfile))
1172                 return false;
1173
1174         spin_lock(&fp->fi_lock);
1175         list_del_init(&stp->st_perfile);
1176         spin_unlock(&fp->fi_lock);
1177         list_del(&stp->st_perstateowner);
1178         return true;
1179 }
1180
1181 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1182 {
1183         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1184
1185         put_clnt_odstate(stp->st_clnt_odstate);
1186         release_all_access(stp);
1187         if (stp->st_stateowner)
1188                 nfs4_put_stateowner(stp->st_stateowner);
1189         kmem_cache_free(stateid_slab, stid);
1190 }
1191
1192 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1193 {
1194         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1195         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1196         struct file *file;
1197
1198         file = find_any_file(stp->st_stid.sc_file);
1199         if (file)
1200                 filp_close(file, (fl_owner_t)lo);
1201         nfs4_free_ol_stateid(stid);
1202 }
1203
1204 /*
1205  * Put the persistent reference to an already unhashed generic stateid, while
1206  * holding the cl_lock. If it's the last reference, then put it onto the
1207  * reaplist for later destruction.
1208  */
1209 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1210                                        struct list_head *reaplist)
1211 {
1212         struct nfs4_stid *s = &stp->st_stid;
1213         struct nfs4_client *clp = s->sc_client;
1214
1215         lockdep_assert_held(&clp->cl_lock);
1216
1217         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1218
1219         if (!atomic_dec_and_test(&s->sc_count)) {
1220                 wake_up_all(&close_wq);
1221                 return;
1222         }
1223
1224         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1225         list_add(&stp->st_locks, reaplist);
1226 }
1227
1228 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1229 {
1230         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1231
1232         list_del_init(&stp->st_locks);
1233         nfs4_unhash_stid(&stp->st_stid);
1234         return unhash_ol_stateid(stp);
1235 }
1236
1237 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1238 {
1239         struct nfs4_client *clp = stp->st_stid.sc_client;
1240         bool unhashed;
1241
1242         spin_lock(&clp->cl_lock);
1243         unhashed = unhash_lock_stateid(stp);
1244         spin_unlock(&clp->cl_lock);
1245         if (unhashed)
1246                 nfs4_put_stid(&stp->st_stid);
1247 }
1248
1249 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1250 {
1251         struct nfs4_client *clp = lo->lo_owner.so_client;
1252
1253         lockdep_assert_held(&clp->cl_lock);
1254
1255         list_del_init(&lo->lo_owner.so_strhash);
1256 }
1257
1258 /*
1259  * Free a list of generic stateids that were collected earlier after being
1260  * fully unhashed.
1261  */
1262 static void
1263 free_ol_stateid_reaplist(struct list_head *reaplist)
1264 {
1265         struct nfs4_ol_stateid *stp;
1266         struct nfs4_file *fp;
1267
1268         might_sleep();
1269
1270         while (!list_empty(reaplist)) {
1271                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1272                                        st_locks);
1273                 list_del(&stp->st_locks);
1274                 fp = stp->st_stid.sc_file;
1275                 stp->st_stid.sc_free(&stp->st_stid);
1276                 if (fp)
1277                         put_nfs4_file(fp);
1278         }
1279 }
1280
1281 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1282                                        struct list_head *reaplist)
1283 {
1284         struct nfs4_ol_stateid *stp;
1285
1286         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1287
1288         while (!list_empty(&open_stp->st_locks)) {
1289                 stp = list_entry(open_stp->st_locks.next,
1290                                 struct nfs4_ol_stateid, st_locks);
1291                 WARN_ON(!unhash_lock_stateid(stp));
1292                 put_ol_stateid_locked(stp, reaplist);
1293         }
1294 }
1295
1296 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1297                                 struct list_head *reaplist)
1298 {
1299         bool unhashed;
1300
1301         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1302
1303         unhashed = unhash_ol_stateid(stp);
1304         release_open_stateid_locks(stp, reaplist);
1305         return unhashed;
1306 }
1307
1308 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1309 {
1310         LIST_HEAD(reaplist);
1311
1312         spin_lock(&stp->st_stid.sc_client->cl_lock);
1313         if (unhash_open_stateid(stp, &reaplist))
1314                 put_ol_stateid_locked(stp, &reaplist);
1315         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1316         free_ol_stateid_reaplist(&reaplist);
1317 }
1318
1319 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1320 {
1321         struct nfs4_client *clp = oo->oo_owner.so_client;
1322
1323         lockdep_assert_held(&clp->cl_lock);
1324
1325         list_del_init(&oo->oo_owner.so_strhash);
1326         list_del_init(&oo->oo_perclient);
1327 }
1328
1329 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1330 {
1331         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1332                                           nfsd_net_id);
1333         struct nfs4_ol_stateid *s;
1334
1335         spin_lock(&nn->client_lock);
1336         s = oo->oo_last_closed_stid;
1337         if (s) {
1338                 list_del_init(&oo->oo_close_lru);
1339                 oo->oo_last_closed_stid = NULL;
1340         }
1341         spin_unlock(&nn->client_lock);
1342         if (s)
1343                 nfs4_put_stid(&s->st_stid);
1344 }
1345
1346 static void release_openowner(struct nfs4_openowner *oo)
1347 {
1348         struct nfs4_ol_stateid *stp;
1349         struct nfs4_client *clp = oo->oo_owner.so_client;
1350         struct list_head reaplist;
1351
1352         INIT_LIST_HEAD(&reaplist);
1353
1354         spin_lock(&clp->cl_lock);
1355         unhash_openowner_locked(oo);
1356         while (!list_empty(&oo->oo_owner.so_stateids)) {
1357                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1358                                 struct nfs4_ol_stateid, st_perstateowner);
1359                 if (unhash_open_stateid(stp, &reaplist))
1360                         put_ol_stateid_locked(stp, &reaplist);
1361         }
1362         spin_unlock(&clp->cl_lock);
1363         free_ol_stateid_reaplist(&reaplist);
1364         release_last_closed_stateid(oo);
1365         nfs4_put_stateowner(&oo->oo_owner);
1366 }
1367
1368 static inline int
1369 hash_sessionid(struct nfs4_sessionid *sessionid)
1370 {
1371         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1372
1373         return sid->sequence % SESSION_HASH_SIZE;
1374 }
1375
1376 #ifdef CONFIG_SUNRPC_DEBUG
1377 static inline void
1378 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1379 {
1380         u32 *ptr = (u32 *)(&sessionid->data[0]);
1381         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1382 }
1383 #else
1384 static inline void
1385 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1386 {
1387 }
1388 #endif
1389
1390 /*
1391  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1392  * won't be used for replay.
1393  */
1394 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1395 {
1396         struct nfs4_stateowner *so = cstate->replay_owner;
1397
1398         if (nfserr == nfserr_replay_me)
1399                 return;
1400
1401         if (!seqid_mutating_err(ntohl(nfserr))) {
1402                 nfsd4_cstate_clear_replay(cstate);
1403                 return;
1404         }
1405         if (!so)
1406                 return;
1407         if (so->so_is_open_owner)
1408                 release_last_closed_stateid(openowner(so));
1409         so->so_seqid++;
1410         return;
1411 }
1412
1413 static void
1414 gen_sessionid(struct nfsd4_session *ses)
1415 {
1416         struct nfs4_client *clp = ses->se_client;
1417         struct nfsd4_sessionid *sid;
1418
1419         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1420         sid->clientid = clp->cl_clientid;
1421         sid->sequence = current_sessionid++;
1422         sid->reserved = 0;
1423 }
1424
1425 /*
1426  * The protocol defines ca_maxresponssize_cached to include the size of
1427  * the rpc header, but all we need to cache is the data starting after
1428  * the end of the initial SEQUENCE operation--the rest we regenerate
1429  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1430  * value that is the number of bytes in our cache plus a few additional
1431  * bytes.  In order to stay on the safe side, and not promise more than
1432  * we can cache, those additional bytes must be the minimum possible: 24
1433  * bytes of rpc header (xid through accept state, with AUTH_NULL
1434  * verifier), 12 for the compound header (with zero-length tag), and 44
1435  * for the SEQUENCE op response:
1436  */
1437 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1438
1439 static void
1440 free_session_slots(struct nfsd4_session *ses)
1441 {
1442         int i;
1443
1444         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1445                 kfree(ses->se_slots[i]);
1446 }
1447
1448 /*
1449  * We don't actually need to cache the rpc and session headers, so we
1450  * can allocate a little less for each slot:
1451  */
1452 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1453 {
1454         u32 size;
1455
1456         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1457                 size = 0;
1458         else
1459                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1460         return size + sizeof(struct nfsd4_slot);
1461 }
1462
1463 /*
1464  * XXX: If we run out of reserved DRC memory we could (up to a point)
1465  * re-negotiate active sessions and reduce their slot usage to make
1466  * room for new connections. For now we just fail the create session.
1467  */
1468 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1469 {
1470         u32 slotsize = slot_bytes(ca);
1471         u32 num = ca->maxreqs;
1472         int avail;
1473
1474         spin_lock(&nfsd_drc_lock);
1475         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1476                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1477         num = min_t(int, num, avail / slotsize);
1478         nfsd_drc_mem_used += num * slotsize;
1479         spin_unlock(&nfsd_drc_lock);
1480
1481         return num;
1482 }
1483
1484 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1485 {
1486         int slotsize = slot_bytes(ca);
1487
1488         spin_lock(&nfsd_drc_lock);
1489         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1490         spin_unlock(&nfsd_drc_lock);
1491 }
1492
1493 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1494                                            struct nfsd4_channel_attrs *battrs)
1495 {
1496         int numslots = fattrs->maxreqs;
1497         int slotsize = slot_bytes(fattrs);
1498         struct nfsd4_session *new;
1499         int mem, i;
1500
1501         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1502                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1503         mem = numslots * sizeof(struct nfsd4_slot *);
1504
1505         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1506         if (!new)
1507                 return NULL;
1508         /* allocate each struct nfsd4_slot and data cache in one piece */
1509         for (i = 0; i < numslots; i++) {
1510                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1511                 if (!new->se_slots[i])
1512                         goto out_free;
1513         }
1514
1515         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1516         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1517
1518         return new;
1519 out_free:
1520         while (i--)
1521                 kfree(new->se_slots[i]);
1522         kfree(new);
1523         return NULL;
1524 }
1525
1526 static void free_conn(struct nfsd4_conn *c)
1527 {
1528         svc_xprt_put(c->cn_xprt);
1529         kfree(c);
1530 }
1531
1532 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1533 {
1534         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1535         struct nfs4_client *clp = c->cn_session->se_client;
1536
1537         spin_lock(&clp->cl_lock);
1538         if (!list_empty(&c->cn_persession)) {
1539                 list_del(&c->cn_persession);
1540                 free_conn(c);
1541         }
1542         nfsd4_probe_callback(clp);
1543         spin_unlock(&clp->cl_lock);
1544 }
1545
1546 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1547 {
1548         struct nfsd4_conn *conn;
1549
1550         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1551         if (!conn)
1552                 return NULL;
1553         svc_xprt_get(rqstp->rq_xprt);
1554         conn->cn_xprt = rqstp->rq_xprt;
1555         conn->cn_flags = flags;
1556         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1557         return conn;
1558 }
1559
1560 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1561 {
1562         conn->cn_session = ses;
1563         list_add(&conn->cn_persession, &ses->se_conns);
1564 }
1565
1566 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1567 {
1568         struct nfs4_client *clp = ses->se_client;
1569
1570         spin_lock(&clp->cl_lock);
1571         __nfsd4_hash_conn(conn, ses);
1572         spin_unlock(&clp->cl_lock);
1573 }
1574
1575 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1576 {
1577         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1578         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1579 }
1580
1581 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1582 {
1583         int ret;
1584
1585         nfsd4_hash_conn(conn, ses);
1586         ret = nfsd4_register_conn(conn);
1587         if (ret)
1588                 /* oops; xprt is already down: */
1589                 nfsd4_conn_lost(&conn->cn_xpt_user);
1590         /* We may have gained or lost a callback channel: */
1591         nfsd4_probe_callback_sync(ses->se_client);
1592 }
1593
1594 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1595 {
1596         u32 dir = NFS4_CDFC4_FORE;
1597
1598         if (cses->flags & SESSION4_BACK_CHAN)
1599                 dir |= NFS4_CDFC4_BACK;
1600         return alloc_conn(rqstp, dir);
1601 }
1602
1603 /* must be called under client_lock */
1604 static void nfsd4_del_conns(struct nfsd4_session *s)
1605 {
1606         struct nfs4_client *clp = s->se_client;
1607         struct nfsd4_conn *c;
1608
1609         spin_lock(&clp->cl_lock);
1610         while (!list_empty(&s->se_conns)) {
1611                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1612                 list_del_init(&c->cn_persession);
1613                 spin_unlock(&clp->cl_lock);
1614
1615                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1616                 free_conn(c);
1617
1618                 spin_lock(&clp->cl_lock);
1619         }
1620         spin_unlock(&clp->cl_lock);
1621 }
1622
1623 static void __free_session(struct nfsd4_session *ses)
1624 {
1625         free_session_slots(ses);
1626         kfree(ses);
1627 }
1628
1629 static void free_session(struct nfsd4_session *ses)
1630 {
1631         nfsd4_del_conns(ses);
1632         nfsd4_put_drc_mem(&ses->se_fchannel);
1633         __free_session(ses);
1634 }
1635
1636 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1637 {
1638         int idx;
1639         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1640
1641         new->se_client = clp;
1642         gen_sessionid(new);
1643
1644         INIT_LIST_HEAD(&new->se_conns);
1645
1646         new->se_cb_seq_nr = 1;
1647         new->se_flags = cses->flags;
1648         new->se_cb_prog = cses->callback_prog;
1649         new->se_cb_sec = cses->cb_sec;
1650         atomic_set(&new->se_ref, 0);
1651         idx = hash_sessionid(&new->se_sessionid);
1652         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1653         spin_lock(&clp->cl_lock);
1654         list_add(&new->se_perclnt, &clp->cl_sessions);
1655         spin_unlock(&clp->cl_lock);
1656
1657         {
1658                 struct sockaddr *sa = svc_addr(rqstp);
1659                 /*
1660                  * This is a little silly; with sessions there's no real
1661                  * use for the callback address.  Use the peer address
1662                  * as a reasonable default for now, but consider fixing
1663                  * the rpc client not to require an address in the
1664                  * future:
1665                  */
1666                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1667                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1668         }
1669 }
1670
1671 /* caller must hold client_lock */
1672 static struct nfsd4_session *
1673 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1674 {
1675         struct nfsd4_session *elem;
1676         int idx;
1677         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1678
1679         lockdep_assert_held(&nn->client_lock);
1680
1681         dump_sessionid(__func__, sessionid);
1682         idx = hash_sessionid(sessionid);
1683         /* Search in the appropriate list */
1684         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1685                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1686                             NFS4_MAX_SESSIONID_LEN)) {
1687                         return elem;
1688                 }
1689         }
1690
1691         dprintk("%s: session not found\n", __func__);
1692         return NULL;
1693 }
1694
1695 static struct nfsd4_session *
1696 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1697                 __be32 *ret)
1698 {
1699         struct nfsd4_session *session;
1700         __be32 status = nfserr_badsession;
1701
1702         session = __find_in_sessionid_hashtbl(sessionid, net);
1703         if (!session)
1704                 goto out;
1705         status = nfsd4_get_session_locked(session);
1706         if (status)
1707                 session = NULL;
1708 out:
1709         *ret = status;
1710         return session;
1711 }
1712
1713 /* caller must hold client_lock */
1714 static void
1715 unhash_session(struct nfsd4_session *ses)
1716 {
1717         struct nfs4_client *clp = ses->se_client;
1718         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1719
1720         lockdep_assert_held(&nn->client_lock);
1721
1722         list_del(&ses->se_hash);
1723         spin_lock(&ses->se_client->cl_lock);
1724         list_del(&ses->se_perclnt);
1725         spin_unlock(&ses->se_client->cl_lock);
1726 }
1727
1728 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1729 static int
1730 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1731 {
1732         /*
1733          * We're assuming the clid was not given out from a boot
1734          * precisely 2^32 (about 136 years) before this one.  That seems
1735          * a safe assumption:
1736          */
1737         if (clid->cl_boot == (u32)nn->boot_time)
1738                 return 0;
1739         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1740                 clid->cl_boot, clid->cl_id, nn->boot_time);
1741         return 1;
1742 }
1743
1744 /*
1745  * XXX Should we use a slab cache ?
1746  * This type of memory management is somewhat inefficient, but we use it
1747  * anyway since SETCLIENTID is not a common operation.
1748  */
1749 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1750 {
1751         struct nfs4_client *clp;
1752         int i;
1753
1754         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1755         if (clp == NULL)
1756                 return NULL;
1757         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1758         if (clp->cl_name.data == NULL)
1759                 goto err_no_name;
1760         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1761                         OWNER_HASH_SIZE, GFP_KERNEL);
1762         if (!clp->cl_ownerstr_hashtbl)
1763                 goto err_no_hashtbl;
1764         for (i = 0; i < OWNER_HASH_SIZE; i++)
1765                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1766         clp->cl_name.len = name.len;
1767         INIT_LIST_HEAD(&clp->cl_sessions);
1768         idr_init(&clp->cl_stateids);
1769         atomic_set(&clp->cl_refcount, 0);
1770         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1771         INIT_LIST_HEAD(&clp->cl_idhash);
1772         INIT_LIST_HEAD(&clp->cl_openowners);
1773         INIT_LIST_HEAD(&clp->cl_delegations);
1774         INIT_LIST_HEAD(&clp->cl_lru);
1775         INIT_LIST_HEAD(&clp->cl_revoked);
1776 #ifdef CONFIG_NFSD_PNFS
1777         INIT_LIST_HEAD(&clp->cl_lo_states);
1778 #endif
1779         spin_lock_init(&clp->cl_lock);
1780         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1781         return clp;
1782 err_no_hashtbl:
1783         kfree(clp->cl_name.data);
1784 err_no_name:
1785         kfree(clp);
1786         return NULL;
1787 }
1788
1789 static void
1790 free_client(struct nfs4_client *clp)
1791 {
1792         while (!list_empty(&clp->cl_sessions)) {
1793                 struct nfsd4_session *ses;
1794                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1795                                 se_perclnt);
1796                 list_del(&ses->se_perclnt);
1797                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1798                 free_session(ses);
1799         }
1800         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1801         free_svc_cred(&clp->cl_cred);
1802         kfree(clp->cl_ownerstr_hashtbl);
1803         kfree(clp->cl_name.data);
1804         idr_destroy(&clp->cl_stateids);
1805         kfree(clp);
1806 }
1807
1808 /* must be called under the client_lock */
1809 static void
1810 unhash_client_locked(struct nfs4_client *clp)
1811 {
1812         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1813         struct nfsd4_session *ses;
1814
1815         lockdep_assert_held(&nn->client_lock);
1816
1817         /* Mark the client as expired! */
1818         clp->cl_time = 0;
1819         /* Make it invisible */
1820         if (!list_empty(&clp->cl_idhash)) {
1821                 list_del_init(&clp->cl_idhash);
1822                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1823                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1824                 else
1825                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1826         }
1827         list_del_init(&clp->cl_lru);
1828         spin_lock(&clp->cl_lock);
1829         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1830                 list_del_init(&ses->se_hash);
1831         spin_unlock(&clp->cl_lock);
1832 }
1833
1834 static void
1835 unhash_client(struct nfs4_client *clp)
1836 {
1837         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1838
1839         spin_lock(&nn->client_lock);
1840         unhash_client_locked(clp);
1841         spin_unlock(&nn->client_lock);
1842 }
1843
1844 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1845 {
1846         if (atomic_read(&clp->cl_refcount))
1847                 return nfserr_jukebox;
1848         unhash_client_locked(clp);
1849         return nfs_ok;
1850 }
1851
1852 static void
1853 __destroy_client(struct nfs4_client *clp)
1854 {
1855         struct nfs4_openowner *oo;
1856         struct nfs4_delegation *dp;
1857         struct list_head reaplist;
1858
1859         INIT_LIST_HEAD(&reaplist);
1860         spin_lock(&state_lock);
1861         while (!list_empty(&clp->cl_delegations)) {
1862                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1863                 WARN_ON(!unhash_delegation_locked(dp));
1864                 list_add(&dp->dl_recall_lru, &reaplist);
1865         }
1866         spin_unlock(&state_lock);
1867         while (!list_empty(&reaplist)) {
1868                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1869                 list_del_init(&dp->dl_recall_lru);
1870                 put_clnt_odstate(dp->dl_clnt_odstate);
1871                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1872                 nfs4_put_stid(&dp->dl_stid);
1873         }
1874         while (!list_empty(&clp->cl_revoked)) {
1875                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1876                 list_del_init(&dp->dl_recall_lru);
1877                 nfs4_put_stid(&dp->dl_stid);
1878         }
1879         while (!list_empty(&clp->cl_openowners)) {
1880                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1881                 nfs4_get_stateowner(&oo->oo_owner);
1882                 release_openowner(oo);
1883         }
1884         nfsd4_return_all_client_layouts(clp);
1885         nfsd4_shutdown_callback(clp);
1886         if (clp->cl_cb_conn.cb_xprt)
1887                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1888         free_client(clp);
1889 }
1890
1891 static void
1892 destroy_client(struct nfs4_client *clp)
1893 {
1894         unhash_client(clp);
1895         __destroy_client(clp);
1896 }
1897
1898 static void expire_client(struct nfs4_client *clp)
1899 {
1900         unhash_client(clp);
1901         nfsd4_client_record_remove(clp);
1902         __destroy_client(clp);
1903 }
1904
1905 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1906 {
1907         memcpy(target->cl_verifier.data, source->data,
1908                         sizeof(target->cl_verifier.data));
1909 }
1910
1911 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1912 {
1913         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1914         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1915 }
1916
1917 int strdup_if_nonnull(char **target, char *source)
1918 {
1919         if (source) {
1920                 *target = kstrdup(source, GFP_KERNEL);
1921                 if (!*target)
1922                         return -ENOMEM;
1923         } else
1924                 *target = NULL;
1925         return 0;
1926 }
1927
1928 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1929 {
1930         int ret;
1931
1932         ret = strdup_if_nonnull(&target->cr_principal, source->cr_principal);
1933         if (ret)
1934                 return ret;
1935         ret = strdup_if_nonnull(&target->cr_raw_principal,
1936                                         source->cr_raw_principal);
1937         if (ret)
1938                 return ret;
1939         target->cr_flavor = source->cr_flavor;
1940         target->cr_uid = source->cr_uid;
1941         target->cr_gid = source->cr_gid;
1942         target->cr_group_info = source->cr_group_info;
1943         get_group_info(target->cr_group_info);
1944         target->cr_gss_mech = source->cr_gss_mech;
1945         if (source->cr_gss_mech)
1946                 gss_mech_get(source->cr_gss_mech);
1947         return 0;
1948 }
1949
1950 static int
1951 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1952 {
1953         if (o1->len < o2->len)
1954                 return -1;
1955         if (o1->len > o2->len)
1956                 return 1;
1957         return memcmp(o1->data, o2->data, o1->len);
1958 }
1959
1960 static int same_name(const char *n1, const char *n2)
1961 {
1962         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1963 }
1964
1965 static int
1966 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1967 {
1968         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1969 }
1970
1971 static int
1972 same_clid(clientid_t *cl1, clientid_t *cl2)
1973 {
1974         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1975 }
1976
1977 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1978 {
1979         int i;
1980
1981         if (g1->ngroups != g2->ngroups)
1982                 return false;
1983         for (i=0; i<g1->ngroups; i++)
1984                 if (!gid_eq(g1->gid[i], g2->gid[i]))
1985                         return false;
1986         return true;
1987 }
1988
1989 /*
1990  * RFC 3530 language requires clid_inuse be returned when the
1991  * "principal" associated with a requests differs from that previously
1992  * used.  We use uid, gid's, and gss principal string as our best
1993  * approximation.  We also don't want to allow non-gss use of a client
1994  * established using gss: in theory cr_principal should catch that
1995  * change, but in practice cr_principal can be null even in the gss case
1996  * since gssd doesn't always pass down a principal string.
1997  */
1998 static bool is_gss_cred(struct svc_cred *cr)
1999 {
2000         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2001         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2002 }
2003
2004
2005 static bool
2006 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2007 {
2008         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2009                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2010                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2011                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2012                 return false;
2013         if (cr1->cr_principal == cr2->cr_principal)
2014                 return true;
2015         if (!cr1->cr_principal || !cr2->cr_principal)
2016                 return false;
2017         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2018 }
2019
2020 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2021 {
2022         struct svc_cred *cr = &rqstp->rq_cred;
2023         u32 service;
2024
2025         if (!cr->cr_gss_mech)
2026                 return false;
2027         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2028         return service == RPC_GSS_SVC_INTEGRITY ||
2029                service == RPC_GSS_SVC_PRIVACY;
2030 }
2031
2032 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2033 {
2034         struct svc_cred *cr = &rqstp->rq_cred;
2035
2036         if (!cl->cl_mach_cred)
2037                 return true;
2038         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2039                 return false;
2040         if (!svc_rqst_integrity_protected(rqstp))
2041                 return false;
2042         if (cl->cl_cred.cr_raw_principal)
2043                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2044                                                 cr->cr_raw_principal);
2045         if (!cr->cr_principal)
2046                 return false;
2047         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2048 }
2049
2050 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2051 {
2052         __be32 verf[2];
2053
2054         /*
2055          * This is opaque to client, so no need to byte-swap. Use
2056          * __force to keep sparse happy
2057          */
2058         verf[0] = (__force __be32)get_seconds();
2059         verf[1] = (__force __be32)nn->clverifier_counter++;
2060         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2061 }
2062
2063 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2064 {
2065         clp->cl_clientid.cl_boot = nn->boot_time;
2066         clp->cl_clientid.cl_id = nn->clientid_counter++;
2067         gen_confirm(clp, nn);
2068 }
2069
2070 static struct nfs4_stid *
2071 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2072 {
2073         struct nfs4_stid *ret;
2074
2075         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2076         if (!ret || !ret->sc_type)
2077                 return NULL;
2078         return ret;
2079 }
2080
2081 static struct nfs4_stid *
2082 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2083 {
2084         struct nfs4_stid *s;
2085
2086         spin_lock(&cl->cl_lock);
2087         s = find_stateid_locked(cl, t);
2088         if (s != NULL) {
2089                 if (typemask & s->sc_type)
2090                         atomic_inc(&s->sc_count);
2091                 else
2092                         s = NULL;
2093         }
2094         spin_unlock(&cl->cl_lock);
2095         return s;
2096 }
2097
2098 static struct nfs4_client *create_client(struct xdr_netobj name,
2099                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2100 {
2101         struct nfs4_client *clp;
2102         struct sockaddr *sa = svc_addr(rqstp);
2103         int ret;
2104         struct net *net = SVC_NET(rqstp);
2105
2106         clp = alloc_client(name);
2107         if (clp == NULL)
2108                 return NULL;
2109
2110         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2111         if (ret) {
2112                 free_client(clp);
2113                 return NULL;
2114         }
2115         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2116         clp->cl_time = get_seconds();
2117         clear_bit(0, &clp->cl_cb_slot_busy);
2118         copy_verf(clp, verf);
2119         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2120         clp->cl_cb_session = NULL;
2121         clp->net = net;
2122         return clp;
2123 }
2124
2125 static void
2126 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2127 {
2128         struct rb_node **new = &(root->rb_node), *parent = NULL;
2129         struct nfs4_client *clp;
2130
2131         while (*new) {
2132                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2133                 parent = *new;
2134
2135                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2136                         new = &((*new)->rb_left);
2137                 else
2138                         new = &((*new)->rb_right);
2139         }
2140
2141         rb_link_node(&new_clp->cl_namenode, parent, new);
2142         rb_insert_color(&new_clp->cl_namenode, root);
2143 }
2144
2145 static struct nfs4_client *
2146 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2147 {
2148         int cmp;
2149         struct rb_node *node = root->rb_node;
2150         struct nfs4_client *clp;
2151
2152         while (node) {
2153                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2154                 cmp = compare_blob(&clp->cl_name, name);
2155                 if (cmp > 0)
2156                         node = node->rb_left;
2157                 else if (cmp < 0)
2158                         node = node->rb_right;
2159                 else
2160                         return clp;
2161         }
2162         return NULL;
2163 }
2164
2165 static void
2166 add_to_unconfirmed(struct nfs4_client *clp)
2167 {
2168         unsigned int idhashval;
2169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2170
2171         lockdep_assert_held(&nn->client_lock);
2172
2173         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2174         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2175         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2176         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2177         renew_client_locked(clp);
2178 }
2179
2180 static void
2181 move_to_confirmed(struct nfs4_client *clp)
2182 {
2183         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2184         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2185
2186         lockdep_assert_held(&nn->client_lock);
2187
2188         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2189         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2190         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2191         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2192         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2193         renew_client_locked(clp);
2194 }
2195
2196 static struct nfs4_client *
2197 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2198 {
2199         struct nfs4_client *clp;
2200         unsigned int idhashval = clientid_hashval(clid->cl_id);
2201
2202         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2203                 if (same_clid(&clp->cl_clientid, clid)) {
2204                         if ((bool)clp->cl_minorversion != sessions)
2205                                 return NULL;
2206                         renew_client_locked(clp);
2207                         return clp;
2208                 }
2209         }
2210         return NULL;
2211 }
2212
2213 static struct nfs4_client *
2214 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2215 {
2216         struct list_head *tbl = nn->conf_id_hashtbl;
2217
2218         lockdep_assert_held(&nn->client_lock);
2219         return find_client_in_id_table(tbl, clid, sessions);
2220 }
2221
2222 static struct nfs4_client *
2223 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2224 {
2225         struct list_head *tbl = nn->unconf_id_hashtbl;
2226
2227         lockdep_assert_held(&nn->client_lock);
2228         return find_client_in_id_table(tbl, clid, sessions);
2229 }
2230
2231 static bool clp_used_exchangeid(struct nfs4_client *clp)
2232 {
2233         return clp->cl_exchange_flags != 0;
2234 }
2235
2236 static struct nfs4_client *
2237 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2238 {
2239         lockdep_assert_held(&nn->client_lock);
2240         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2241 }
2242
2243 static struct nfs4_client *
2244 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2245 {
2246         lockdep_assert_held(&nn->client_lock);
2247         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2248 }
2249
2250 static void
2251 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2252 {
2253         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2254         struct sockaddr *sa = svc_addr(rqstp);
2255         u32 scopeid = rpc_get_scope_id(sa);
2256         unsigned short expected_family;
2257
2258         /* Currently, we only support tcp and tcp6 for the callback channel */
2259         if (se->se_callback_netid_len == 3 &&
2260             !memcmp(se->se_callback_netid_val, "tcp", 3))
2261                 expected_family = AF_INET;
2262         else if (se->se_callback_netid_len == 4 &&
2263                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2264                 expected_family = AF_INET6;
2265         else
2266                 goto out_err;
2267
2268         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2269                                             se->se_callback_addr_len,
2270                                             (struct sockaddr *)&conn->cb_addr,
2271                                             sizeof(conn->cb_addr));
2272
2273         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2274                 goto out_err;
2275
2276         if (conn->cb_addr.ss_family == AF_INET6)
2277                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2278
2279         conn->cb_prog = se->se_callback_prog;
2280         conn->cb_ident = se->se_callback_ident;
2281         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2282         return;
2283 out_err:
2284         conn->cb_addr.ss_family = AF_UNSPEC;
2285         conn->cb_addrlen = 0;
2286         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2287                 "will not receive delegations\n",
2288                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2289
2290         return;
2291 }
2292
2293 /*
2294  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2295  */
2296 static void
2297 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2298 {
2299         struct xdr_buf *buf = resp->xdr.buf;
2300         struct nfsd4_slot *slot = resp->cstate.slot;
2301         unsigned int base;
2302
2303         dprintk("--> %s slot %p\n", __func__, slot);
2304
2305         slot->sl_opcnt = resp->opcnt;
2306         slot->sl_status = resp->cstate.status;
2307
2308         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2309         if (nfsd4_not_cached(resp)) {
2310                 slot->sl_datalen = 0;
2311                 return;
2312         }
2313         base = resp->cstate.data_offset;
2314         slot->sl_datalen = buf->len - base;
2315         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2316                 WARN(1, "%s: sessions DRC could not cache compound\n",
2317                      __func__);
2318         return;
2319 }
2320
2321 /*
2322  * Encode the replay sequence operation from the slot values.
2323  * If cachethis is FALSE encode the uncached rep error on the next
2324  * operation which sets resp->p and increments resp->opcnt for
2325  * nfs4svc_encode_compoundres.
2326  *
2327  */
2328 static __be32
2329 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2330                           struct nfsd4_compoundres *resp)
2331 {
2332         struct nfsd4_op *op;
2333         struct nfsd4_slot *slot = resp->cstate.slot;
2334
2335         /* Encode the replayed sequence operation */
2336         op = &args->ops[resp->opcnt - 1];
2337         nfsd4_encode_operation(resp, op);
2338
2339         /* Return nfserr_retry_uncached_rep in next operation. */
2340         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2341                 op = &args->ops[resp->opcnt++];
2342                 op->status = nfserr_retry_uncached_rep;
2343                 nfsd4_encode_operation(resp, op);
2344         }
2345         return op->status;
2346 }
2347
2348 /*
2349  * The sequence operation is not cached because we can use the slot and
2350  * session values.
2351  */
2352 static __be32
2353 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2354                          struct nfsd4_sequence *seq)
2355 {
2356         struct nfsd4_slot *slot = resp->cstate.slot;
2357         struct xdr_stream *xdr = &resp->xdr;
2358         __be32 *p;
2359         __be32 status;
2360
2361         dprintk("--> %s slot %p\n", __func__, slot);
2362
2363         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2364         if (status)
2365                 return status;
2366
2367         p = xdr_reserve_space(xdr, slot->sl_datalen);
2368         if (!p) {
2369                 WARN_ON_ONCE(1);
2370                 return nfserr_serverfault;
2371         }
2372         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2373         xdr_commit_encode(xdr);
2374
2375         resp->opcnt = slot->sl_opcnt;
2376         return slot->sl_status;
2377 }
2378
2379 /*
2380  * Set the exchange_id flags returned by the server.
2381  */
2382 static void
2383 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2384 {
2385 #ifdef CONFIG_NFSD_PNFS
2386         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2387 #else
2388         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2389 #endif
2390
2391         /* Referrals are supported, Migration is not. */
2392         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2393
2394         /* set the wire flags to return to client. */
2395         clid->flags = new->cl_exchange_flags;
2396 }
2397
2398 static bool client_has_openowners(struct nfs4_client *clp)
2399 {
2400         struct nfs4_openowner *oo;
2401
2402         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2403                 if (!list_empty(&oo->oo_owner.so_stateids))
2404                         return true;
2405         }
2406         return false;
2407 }
2408
2409 static bool client_has_state(struct nfs4_client *clp)
2410 {
2411         return client_has_openowners(clp)
2412 #ifdef CONFIG_NFSD_PNFS
2413                 || !list_empty(&clp->cl_lo_states)
2414 #endif
2415                 || !list_empty(&clp->cl_delegations)
2416                 || !list_empty(&clp->cl_sessions);
2417 }
2418
2419 __be32
2420 nfsd4_exchange_id(struct svc_rqst *rqstp,
2421                   struct nfsd4_compound_state *cstate,
2422                   struct nfsd4_exchange_id *exid)
2423 {
2424         struct nfs4_client *conf, *new;
2425         struct nfs4_client *unconf = NULL;
2426         __be32 status;
2427         char                    addr_str[INET6_ADDRSTRLEN];
2428         nfs4_verifier           verf = exid->verifier;
2429         struct sockaddr         *sa = svc_addr(rqstp);
2430         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2431         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2432
2433         rpc_ntop(sa, addr_str, sizeof(addr_str));
2434         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2435                 "ip_addr=%s flags %x, spa_how %d\n",
2436                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2437                 addr_str, exid->flags, exid->spa_how);
2438
2439         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2440                 return nfserr_inval;
2441
2442         new = create_client(exid->clname, rqstp, &verf);
2443         if (new == NULL)
2444                 return nfserr_jukebox;
2445
2446         switch (exid->spa_how) {
2447         case SP4_MACH_CRED:
2448                 exid->spo_must_enforce[0] = 0;
2449                 exid->spo_must_enforce[1] = (
2450                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2451                         1 << (OP_EXCHANGE_ID - 32) |
2452                         1 << (OP_CREATE_SESSION - 32) |
2453                         1 << (OP_DESTROY_SESSION - 32) |
2454                         1 << (OP_DESTROY_CLIENTID - 32));
2455
2456                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2457                                         1 << (OP_OPEN_DOWNGRADE) |
2458                                         1 << (OP_LOCKU) |
2459                                         1 << (OP_DELEGRETURN));
2460
2461                 exid->spo_must_allow[1] &= (
2462                                         1 << (OP_TEST_STATEID - 32) |
2463                                         1 << (OP_FREE_STATEID - 32));
2464                 if (!svc_rqst_integrity_protected(rqstp)) {
2465                         status = nfserr_inval;
2466                         goto out_nolock;
2467                 }
2468                 /*
2469                  * Sometimes userspace doesn't give us a principal.
2470                  * Which is a bug, really.  Anyway, we can't enforce
2471                  * MACH_CRED in that case, better to give up now:
2472                  */
2473                 if (!new->cl_cred.cr_principal &&
2474                                         !new->cl_cred.cr_raw_principal) {
2475                         status = nfserr_serverfault;
2476                         goto out_nolock;
2477                 }
2478                 new->cl_mach_cred = true;
2479         case SP4_NONE:
2480                 break;
2481         default:                                /* checked by xdr code */
2482                 WARN_ON_ONCE(1);
2483         case SP4_SSV:
2484                 status = nfserr_encr_alg_unsupp;
2485                 goto out_nolock;
2486         }
2487
2488         /* Cases below refer to rfc 5661 section 18.35.4: */
2489         spin_lock(&nn->client_lock);
2490         conf = find_confirmed_client_by_name(&exid->clname, nn);
2491         if (conf) {
2492                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2493                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2494
2495                 if (update) {
2496                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2497                                 status = nfserr_inval;
2498                                 goto out;
2499                         }
2500                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2501                                 status = nfserr_wrong_cred;
2502                                 goto out;
2503                         }
2504                         if (!creds_match) { /* case 9 */
2505                                 status = nfserr_perm;
2506                                 goto out;
2507                         }
2508                         if (!verfs_match) { /* case 8 */
2509                                 status = nfserr_not_same;
2510                                 goto out;
2511                         }
2512                         /* case 6 */
2513                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2514                         goto out_copy;
2515                 }
2516                 if (!creds_match) { /* case 3 */
2517                         if (client_has_state(conf)) {
2518                                 status = nfserr_clid_inuse;
2519                                 goto out;
2520                         }
2521                         goto out_new;
2522                 }
2523                 if (verfs_match) { /* case 2 */
2524                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2525                         goto out_copy;
2526                 }
2527                 /* case 5, client reboot */
2528                 conf = NULL;
2529                 goto out_new;
2530         }
2531
2532         if (update) { /* case 7 */
2533                 status = nfserr_noent;
2534                 goto out;
2535         }
2536
2537         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2538         if (unconf) /* case 4, possible retry or client restart */
2539                 unhash_client_locked(unconf);
2540
2541         /* case 1 (normal case) */
2542 out_new:
2543         if (conf) {
2544                 status = mark_client_expired_locked(conf);
2545                 if (status)
2546                         goto out;
2547         }
2548         new->cl_minorversion = cstate->minorversion;
2549         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2550         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2551
2552         gen_clid(new, nn);
2553         add_to_unconfirmed(new);
2554         swap(new, conf);
2555 out_copy:
2556         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2557         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2558
2559         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2560         nfsd4_set_ex_flags(conf, exid);
2561
2562         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2563                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2564         status = nfs_ok;
2565
2566 out:
2567         spin_unlock(&nn->client_lock);
2568 out_nolock:
2569         if (new)
2570                 expire_client(new);
2571         if (unconf)
2572                 expire_client(unconf);
2573         return status;
2574 }
2575
2576 static __be32
2577 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2578 {
2579         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2580                 slot_seqid);
2581
2582         /* The slot is in use, and no response has been sent. */
2583         if (slot_inuse) {
2584                 if (seqid == slot_seqid)
2585                         return nfserr_jukebox;
2586                 else
2587                         return nfserr_seq_misordered;
2588         }
2589         /* Note unsigned 32-bit arithmetic handles wraparound: */
2590         if (likely(seqid == slot_seqid + 1))
2591                 return nfs_ok;
2592         if (seqid == slot_seqid)
2593                 return nfserr_replay_cache;
2594         return nfserr_seq_misordered;
2595 }
2596
2597 /*
2598  * Cache the create session result into the create session single DRC
2599  * slot cache by saving the xdr structure. sl_seqid has been set.
2600  * Do this for solo or embedded create session operations.
2601  */
2602 static void
2603 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2604                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2605 {
2606         slot->sl_status = nfserr;
2607         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2608 }
2609
2610 static __be32
2611 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2612                             struct nfsd4_clid_slot *slot)
2613 {
2614         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2615         return slot->sl_status;
2616 }
2617
2618 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2619                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2620                         1 +     /* MIN tag is length with zero, only length */ \
2621                         3 +     /* version, opcount, opcode */ \
2622                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2623                                 /* seqid, slotID, slotID, cache */ \
2624                         4 ) * sizeof(__be32))
2625
2626 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2627                         2 +     /* verifier: AUTH_NULL, length 0 */\
2628                         1 +     /* status */ \
2629                         1 +     /* MIN tag is length with zero, only length */ \
2630                         3 +     /* opcount, opcode, opstatus*/ \
2631                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2632                                 /* seqid, slotID, slotID, slotID, status */ \
2633                         5 ) * sizeof(__be32))
2634
2635 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2636 {
2637         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2638
2639         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2640                 return nfserr_toosmall;
2641         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2642                 return nfserr_toosmall;
2643         ca->headerpadsz = 0;
2644         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2645         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2646         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2647         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2648                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2649         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2650         /*
2651          * Note decreasing slot size below client's request may make it
2652          * difficult for client to function correctly, whereas
2653          * decreasing the number of slots will (just?) affect
2654          * performance.  When short on memory we therefore prefer to
2655          * decrease number of slots instead of their size.  Clients that
2656          * request larger slots than they need will get poor results:
2657          */
2658         ca->maxreqs = nfsd4_get_drc_mem(ca);
2659         if (!ca->maxreqs)
2660                 return nfserr_jukebox;
2661
2662         return nfs_ok;
2663 }
2664
2665 /*
2666  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2667  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2668  */
2669 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2670         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2671
2672 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2673         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2674
2675 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2676                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2677 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2678                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2679                                  sizeof(__be32))
2680
2681 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2682 {
2683         ca->headerpadsz = 0;
2684
2685         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2686                 return nfserr_toosmall;
2687         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2688                 return nfserr_toosmall;
2689         ca->maxresp_cached = 0;
2690         if (ca->maxops < 2)
2691                 return nfserr_toosmall;
2692
2693         return nfs_ok;
2694 }
2695
2696 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2697 {
2698         switch (cbs->flavor) {
2699         case RPC_AUTH_NULL:
2700         case RPC_AUTH_UNIX:
2701                 return nfs_ok;
2702         default:
2703                 /*
2704                  * GSS case: the spec doesn't allow us to return this
2705                  * error.  But it also doesn't allow us not to support
2706                  * GSS.
2707                  * I'd rather this fail hard than return some error the
2708                  * client might think it can already handle:
2709                  */
2710                 return nfserr_encr_alg_unsupp;
2711         }
2712 }
2713
2714 __be32
2715 nfsd4_create_session(struct svc_rqst *rqstp,
2716                      struct nfsd4_compound_state *cstate,
2717                      struct nfsd4_create_session *cr_ses)
2718 {
2719         struct sockaddr *sa = svc_addr(rqstp);
2720         struct nfs4_client *conf, *unconf;
2721         struct nfs4_client *old = NULL;
2722         struct nfsd4_session *new;
2723         struct nfsd4_conn *conn;
2724         struct nfsd4_clid_slot *cs_slot = NULL;
2725         __be32 status = 0;
2726         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2727
2728         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2729                 return nfserr_inval;
2730         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2731         if (status)
2732                 return status;
2733         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2734         if (status)
2735                 return status;
2736         status = check_backchannel_attrs(&cr_ses->back_channel);
2737         if (status)
2738                 goto out_release_drc_mem;
2739         status = nfserr_jukebox;
2740         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2741         if (!new)
2742                 goto out_release_drc_mem;
2743         conn = alloc_conn_from_crses(rqstp, cr_ses);
2744         if (!conn)
2745                 goto out_free_session;
2746
2747         spin_lock(&nn->client_lock);
2748         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2749         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2750         WARN_ON_ONCE(conf && unconf);
2751
2752         if (conf) {
2753                 status = nfserr_wrong_cred;
2754                 if (!nfsd4_mach_creds_match(conf, rqstp))
2755                         goto out_free_conn;
2756                 cs_slot = &conf->cl_cs_slot;
2757                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2758                 if (status) {
2759                         if (status == nfserr_replay_cache)
2760                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2761                         goto out_free_conn;
2762                 }
2763         } else if (unconf) {
2764                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2765                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2766                         status = nfserr_clid_inuse;
2767                         goto out_free_conn;
2768                 }
2769                 status = nfserr_wrong_cred;
2770                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2771                         goto out_free_conn;
2772                 cs_slot = &unconf->cl_cs_slot;
2773                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2774                 if (status) {
2775                         /* an unconfirmed replay returns misordered */
2776                         status = nfserr_seq_misordered;
2777                         goto out_free_conn;
2778                 }
2779                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2780                 if (old) {
2781                         status = mark_client_expired_locked(old);
2782                         if (status) {
2783                                 old = NULL;
2784                                 goto out_free_conn;
2785                         }
2786                 }
2787                 move_to_confirmed(unconf);
2788                 conf = unconf;
2789         } else {
2790                 status = nfserr_stale_clientid;
2791                 goto out_free_conn;
2792         }
2793         status = nfs_ok;
2794         /* Persistent sessions are not supported */
2795         cr_ses->flags &= ~SESSION4_PERSIST;
2796         /* Upshifting from TCP to RDMA is not supported */
2797         cr_ses->flags &= ~SESSION4_RDMA;
2798
2799         init_session(rqstp, new, conf, cr_ses);
2800         nfsd4_get_session_locked(new);
2801
2802         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2803                NFS4_MAX_SESSIONID_LEN);
2804         cs_slot->sl_seqid++;
2805         cr_ses->seqid = cs_slot->sl_seqid;
2806
2807         /* cache solo and embedded create sessions under the client_lock */
2808         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2809         spin_unlock(&nn->client_lock);
2810         /* init connection and backchannel */
2811         nfsd4_init_conn(rqstp, conn, new);
2812         nfsd4_put_session(new);
2813         if (old)
2814                 expire_client(old);
2815         return status;
2816 out_free_conn:
2817         spin_unlock(&nn->client_lock);
2818         free_conn(conn);
2819         if (old)
2820                 expire_client(old);
2821 out_free_session:
2822         __free_session(new);
2823 out_release_drc_mem:
2824         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2825         return status;
2826 }
2827
2828 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2829 {
2830         switch (*dir) {
2831         case NFS4_CDFC4_FORE:
2832         case NFS4_CDFC4_BACK:
2833                 return nfs_ok;
2834         case NFS4_CDFC4_FORE_OR_BOTH:
2835         case NFS4_CDFC4_BACK_OR_BOTH:
2836                 *dir = NFS4_CDFC4_BOTH;
2837                 return nfs_ok;
2838         };
2839         return nfserr_inval;
2840 }
2841
2842 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2843 {
2844         struct nfsd4_session *session = cstate->session;
2845         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2846         __be32 status;
2847
2848         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2849         if (status)
2850                 return status;
2851         spin_lock(&nn->client_lock);
2852         session->se_cb_prog = bc->bc_cb_program;
2853         session->se_cb_sec = bc->bc_cb_sec;
2854         spin_unlock(&nn->client_lock);
2855
2856         nfsd4_probe_callback(session->se_client);
2857
2858         return nfs_ok;
2859 }
2860
2861 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2862                      struct nfsd4_compound_state *cstate,
2863                      struct nfsd4_bind_conn_to_session *bcts)
2864 {
2865         __be32 status;
2866         struct nfsd4_conn *conn;
2867         struct nfsd4_session *session;
2868         struct net *net = SVC_NET(rqstp);
2869         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2870
2871         if (!nfsd4_last_compound_op(rqstp))
2872                 return nfserr_not_only_op;
2873         spin_lock(&nn->client_lock);
2874         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2875         spin_unlock(&nn->client_lock);
2876         if (!session)
2877                 goto out_no_session;
2878         status = nfserr_wrong_cred;
2879         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2880                 goto out;
2881         status = nfsd4_map_bcts_dir(&bcts->dir);
2882         if (status)
2883                 goto out;
2884         conn = alloc_conn(rqstp, bcts->dir);
2885         status = nfserr_jukebox;
2886         if (!conn)
2887                 goto out;
2888         nfsd4_init_conn(rqstp, conn, session);
2889         status = nfs_ok;
2890 out:
2891         nfsd4_put_session(session);
2892 out_no_session:
2893         return status;
2894 }
2895
2896 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2897 {
2898         if (!session)
2899                 return 0;
2900         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2901 }
2902
2903 __be32
2904 nfsd4_destroy_session(struct svc_rqst *r,
2905                       struct nfsd4_compound_state *cstate,
2906                       struct nfsd4_destroy_session *sessionid)
2907 {
2908         struct nfsd4_session *ses;
2909         __be32 status;
2910         int ref_held_by_me = 0;
2911         struct net *net = SVC_NET(r);
2912         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2913
2914         status = nfserr_not_only_op;
2915         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2916                 if (!nfsd4_last_compound_op(r))
2917                         goto out;
2918                 ref_held_by_me++;
2919         }
2920         dump_sessionid(__func__, &sessionid->sessionid);
2921         spin_lock(&nn->client_lock);
2922         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2923         if (!ses)
2924                 goto out_client_lock;
2925         status = nfserr_wrong_cred;
2926         if (!nfsd4_mach_creds_match(ses->se_client, r))
2927                 goto out_put_session;
2928         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2929         if (status)
2930                 goto out_put_session;
2931         unhash_session(ses);
2932         spin_unlock(&nn->client_lock);
2933
2934         nfsd4_probe_callback_sync(ses->se_client);
2935
2936         spin_lock(&nn->client_lock);
2937         status = nfs_ok;
2938 out_put_session:
2939         nfsd4_put_session_locked(ses);
2940 out_client_lock:
2941         spin_unlock(&nn->client_lock);
2942 out:
2943         return status;
2944 }
2945
2946 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2947 {
2948         struct nfsd4_conn *c;
2949
2950         list_for_each_entry(c, &s->se_conns, cn_persession) {
2951                 if (c->cn_xprt == xpt) {
2952                         return c;
2953                 }
2954         }
2955         return NULL;
2956 }
2957
2958 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2959 {
2960         struct nfs4_client *clp = ses->se_client;
2961         struct nfsd4_conn *c;
2962         __be32 status = nfs_ok;
2963         int ret;
2964
2965         spin_lock(&clp->cl_lock);
2966         c = __nfsd4_find_conn(new->cn_xprt, ses);
2967         if (c)
2968                 goto out_free;
2969         status = nfserr_conn_not_bound_to_session;
2970         if (clp->cl_mach_cred)
2971                 goto out_free;
2972         __nfsd4_hash_conn(new, ses);
2973         spin_unlock(&clp->cl_lock);
2974         ret = nfsd4_register_conn(new);
2975         if (ret)
2976                 /* oops; xprt is already down: */
2977                 nfsd4_conn_lost(&new->cn_xpt_user);
2978         return nfs_ok;
2979 out_free:
2980         spin_unlock(&clp->cl_lock);
2981         free_conn(new);
2982         return status;
2983 }
2984
2985 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2986 {
2987         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2988
2989         return args->opcnt > session->se_fchannel.maxops;
2990 }
2991
2992 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2993                                   struct nfsd4_session *session)
2994 {
2995         struct xdr_buf *xb = &rqstp->rq_arg;
2996
2997         return xb->len > session->se_fchannel.maxreq_sz;
2998 }
2999
3000 __be32
3001 nfsd4_sequence(struct svc_rqst *rqstp,
3002                struct nfsd4_compound_state *cstate,
3003                struct nfsd4_sequence *seq)
3004 {
3005         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3006         struct xdr_stream *xdr = &resp->xdr;
3007         struct nfsd4_session *session;
3008         struct nfs4_client *clp;
3009         struct nfsd4_slot *slot;
3010         struct nfsd4_conn *conn;
3011         __be32 status;
3012         int buflen;
3013         struct net *net = SVC_NET(rqstp);
3014         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3015
3016         if (resp->opcnt != 1)
3017                 return nfserr_sequence_pos;
3018
3019         /*
3020          * Will be either used or freed by nfsd4_sequence_check_conn
3021          * below.
3022          */
3023         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3024         if (!conn)
3025                 return nfserr_jukebox;
3026
3027         spin_lock(&nn->client_lock);
3028         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3029         if (!session)
3030                 goto out_no_session;
3031         clp = session->se_client;
3032
3033         status = nfserr_too_many_ops;
3034         if (nfsd4_session_too_many_ops(rqstp, session))
3035                 goto out_put_session;
3036
3037         status = nfserr_req_too_big;
3038         if (nfsd4_request_too_big(rqstp, session))
3039                 goto out_put_session;
3040
3041         status = nfserr_badslot;
3042         if (seq->slotid >= session->se_fchannel.maxreqs)
3043                 goto out_put_session;
3044
3045         slot = session->se_slots[seq->slotid];
3046         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3047
3048         /* We do not negotiate the number of slots yet, so set the
3049          * maxslots to the session maxreqs which is used to encode
3050          * sr_highest_slotid and the sr_target_slot id to maxslots */
3051         seq->maxslots = session->se_fchannel.maxreqs;
3052
3053         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3054                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3055         if (status == nfserr_replay_cache) {
3056                 status = nfserr_seq_misordered;
3057                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3058                         goto out_put_session;
3059                 cstate->slot = slot;
3060                 cstate->session = session;
3061                 cstate->clp = clp;
3062                 /* Return the cached reply status and set cstate->status
3063                  * for nfsd4_proc_compound processing */
3064                 status = nfsd4_replay_cache_entry(resp, seq);
3065                 cstate->status = nfserr_replay_cache;
3066                 goto out;
3067         }
3068         if (status)
3069                 goto out_put_session;
3070
3071         status = nfsd4_sequence_check_conn(conn, session);
3072         conn = NULL;
3073         if (status)
3074                 goto out_put_session;
3075
3076         buflen = (seq->cachethis) ?
3077                         session->se_fchannel.maxresp_cached :
3078                         session->se_fchannel.maxresp_sz;
3079         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3080                                     nfserr_rep_too_big;
3081         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3082                 goto out_put_session;
3083         svc_reserve(rqstp, buflen);
3084
3085         status = nfs_ok;
3086         /* Success! bump slot seqid */
3087         slot->sl_seqid = seq->seqid;
3088         slot->sl_flags |= NFSD4_SLOT_INUSE;
3089         if (seq->cachethis)
3090                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3091         else
3092                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3093
3094         cstate->slot = slot;
3095         cstate->session = session;
3096         cstate->clp = clp;
3097
3098 out:
3099         switch (clp->cl_cb_state) {
3100         case NFSD4_CB_DOWN:
3101                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3102                 break;
3103         case NFSD4_CB_FAULT:
3104                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3105                 break;
3106         default:
3107                 seq->status_flags = 0;
3108         }
3109         if (!list_empty(&clp->cl_revoked))
3110                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3111 out_no_session:
3112         if (conn)
3113                 free_conn(conn);
3114         spin_unlock(&nn->client_lock);
3115         return status;
3116 out_put_session:
3117         nfsd4_put_session_locked(session);
3118         goto out_no_session;
3119 }
3120
3121 void
3122 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3123 {
3124         struct nfsd4_compound_state *cs = &resp->cstate;
3125
3126         if (nfsd4_has_session(cs)) {
3127                 if (cs->status != nfserr_replay_cache) {
3128                         nfsd4_store_cache_entry(resp);
3129                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3130                 }
3131                 /* Drop session reference that was taken in nfsd4_sequence() */
3132                 nfsd4_put_session(cs->session);
3133         } else if (cs->clp)
3134                 put_client_renew(cs->clp);
3135 }
3136
3137 __be32
3138 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3139 {
3140         struct nfs4_client *conf, *unconf;
3141         struct nfs4_client *clp = NULL;
3142         __be32 status = 0;
3143         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3144
3145         spin_lock(&nn->client_lock);
3146         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3147         conf = find_confirmed_client(&dc->clientid, true, nn);
3148         WARN_ON_ONCE(conf && unconf);
3149
3150         if (conf) {
3151                 if (client_has_state(conf)) {
3152                         status = nfserr_clientid_busy;
3153                         goto out;
3154                 }
3155                 status = mark_client_expired_locked(conf);
3156                 if (status)
3157                         goto out;
3158                 clp = conf;
3159         } else if (unconf)
3160                 clp = unconf;
3161         else {
3162                 status = nfserr_stale_clientid;
3163                 goto out;
3164         }
3165         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3166                 clp = NULL;
3167                 status = nfserr_wrong_cred;
3168                 goto out;
3169         }
3170         unhash_client_locked(clp);
3171 out:
3172         spin_unlock(&nn->client_lock);
3173         if (clp)
3174                 expire_client(clp);
3175         return status;
3176 }
3177
3178 __be32
3179 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3180 {
3181         __be32 status = 0;
3182
3183         if (rc->rca_one_fs) {
3184                 if (!cstate->current_fh.fh_dentry)
3185                         return nfserr_nofilehandle;
3186                 /*
3187                  * We don't take advantage of the rca_one_fs case.
3188                  * That's OK, it's optional, we can safely ignore it.
3189                  */
3190                 return nfs_ok;
3191         }
3192
3193         status = nfserr_complete_already;
3194         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3195                              &cstate->session->se_client->cl_flags))
3196                 goto out;
3197
3198         status = nfserr_stale_clientid;
3199         if (is_client_expired(cstate->session->se_client))
3200                 /*
3201                  * The following error isn't really legal.
3202                  * But we only get here if the client just explicitly
3203                  * destroyed the client.  Surely it no longer cares what
3204                  * error it gets back on an operation for the dead
3205                  * client.
3206                  */
3207                 goto out;
3208
3209         status = nfs_ok;
3210         nfsd4_client_record_create(cstate->session->se_client);
3211 out:
3212         return status;
3213 }
3214
3215 __be32
3216 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3217                   struct nfsd4_setclientid *setclid)
3218 {
3219         struct xdr_netobj       clname = setclid->se_name;
3220         nfs4_verifier           clverifier = setclid->se_verf;
3221         struct nfs4_client      *conf, *new;
3222         struct nfs4_client      *unconf = NULL;
3223         __be32                  status;
3224         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3225
3226         new = create_client(clname, rqstp, &clverifier);
3227         if (new == NULL)
3228                 return nfserr_jukebox;
3229         /* Cases below refer to rfc 3530 section 14.2.33: */
3230         spin_lock(&nn->client_lock);
3231         conf = find_confirmed_client_by_name(&clname, nn);
3232         if (conf && client_has_state(conf)) {
3233                 /* case 0: */
3234                 status = nfserr_clid_inuse;
3235                 if (clp_used_exchangeid(conf))
3236                         goto out;
3237                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3238                         char addr_str[INET6_ADDRSTRLEN];
3239                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3240                                  sizeof(addr_str));
3241                         dprintk("NFSD: setclientid: string in use by client "
3242                                 "at %s\n", addr_str);
3243                         goto out;
3244                 }
3245         }
3246         unconf = find_unconfirmed_client_by_name(&clname, nn);
3247         if (unconf)
3248                 unhash_client_locked(unconf);
3249         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3250                 /* case 1: probable callback update */
3251                 copy_clid(new, conf);
3252                 gen_confirm(new, nn);
3253         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3254                 gen_clid(new, nn);
3255         new->cl_minorversion = 0;
3256         gen_callback(new, setclid, rqstp);
3257         add_to_unconfirmed(new);
3258         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3259         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3260         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3261         new = NULL;
3262         status = nfs_ok;
3263 out:
3264         spin_unlock(&nn->client_lock);
3265         if (new)
3266                 free_client(new);
3267         if (unconf)
3268                 expire_client(unconf);
3269         return status;
3270 }
3271
3272
3273 __be32
3274 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3275                          struct nfsd4_compound_state *cstate,
3276                          struct nfsd4_setclientid_confirm *setclientid_confirm)
3277 {
3278         struct nfs4_client *conf, *unconf;
3279         struct nfs4_client *old = NULL;
3280         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3281         clientid_t * clid = &setclientid_confirm->sc_clientid;
3282         __be32 status;
3283         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3284
3285         if (STALE_CLIENTID(clid, nn))
3286                 return nfserr_stale_clientid;
3287
3288         spin_lock(&nn->client_lock);
3289         conf = find_confirmed_client(clid, false, nn);
3290         unconf = find_unconfirmed_client(clid, false, nn);
3291         /*
3292          * We try hard to give out unique clientid's, so if we get an
3293          * attempt to confirm the same clientid with a different cred,
3294          * the client may be buggy; this should never happen.
3295          *
3296          * Nevertheless, RFC 7530 recommends INUSE for this case:
3297          */
3298         status = nfserr_clid_inuse;
3299         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3300                 goto out;
3301         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3302                 goto out;
3303         /* cases below refer to rfc 3530 section 14.2.34: */
3304         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3305                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3306                         /* case 2: probable retransmit */
3307                         status = nfs_ok;
3308                 } else /* case 4: client hasn't noticed we rebooted yet? */
3309                         status = nfserr_stale_clientid;
3310                 goto out;
3311         }
3312         status = nfs_ok;
3313         if (conf) { /* case 1: callback update */
3314                 old = unconf;
3315                 unhash_client_locked(old);
3316                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3317         } else { /* case 3: normal case; new or rebooted client */
3318                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3319                 if (old) {
3320                         status = nfserr_clid_inuse;
3321                         if (client_has_state(old)
3322                                         && !same_creds(&unconf->cl_cred,
3323                                                         &old->cl_cred))
3324                                 goto out;
3325                         status = mark_client_expired_locked(old);
3326                         if (status) {
3327                                 old = NULL;
3328                                 goto out;
3329                         }
3330                 }
3331                 move_to_confirmed(unconf);
3332                 conf = unconf;
3333         }
3334         get_client_locked(conf);
3335         spin_unlock(&nn->client_lock);
3336         nfsd4_probe_callback(conf);
3337         spin_lock(&nn->client_lock);
3338         put_client_renew_locked(conf);
3339 out:
3340         spin_unlock(&nn->client_lock);
3341         if (old)
3342                 expire_client(old);
3343         return status;
3344 }
3345
3346 static struct nfs4_file *nfsd4_alloc_file(void)
3347 {
3348         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3349 }
3350
3351 /* OPEN Share state helper functions */
3352 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3353                                 struct nfs4_file *fp)
3354 {
3355         lockdep_assert_held(&state_lock);
3356
3357         atomic_set(&fp->fi_ref, 1);
3358         spin_lock_init(&fp->fi_lock);
3359         INIT_LIST_HEAD(&fp->fi_stateids);
3360         INIT_LIST_HEAD(&fp->fi_delegations);
3361         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3362         fh_copy_shallow(&fp->fi_fhandle, fh);
3363         fp->fi_deleg_file = NULL;
3364         fp->fi_had_conflict = false;
3365         fp->fi_share_deny = 0;
3366         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3367         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3368 #ifdef CONFIG_NFSD_PNFS
3369         INIT_LIST_HEAD(&fp->fi_lo_states);
3370         atomic_set(&fp->fi_lo_recalls, 0);
3371 #endif
3372         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3373 }
3374
3375 void
3376 nfsd4_free_slabs(void)
3377 {
3378         kmem_cache_destroy(odstate_slab);
3379         kmem_cache_destroy(openowner_slab);
3380         kmem_cache_destroy(lockowner_slab);
3381         kmem_cache_destroy(file_slab);
3382         kmem_cache_destroy(stateid_slab);
3383         kmem_cache_destroy(deleg_slab);
3384 }
3385
3386 int
3387 nfsd4_init_slabs(void)
3388 {
3389         openowner_slab = kmem_cache_create("nfsd4_openowners",
3390                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3391         if (openowner_slab == NULL)
3392                 goto out;
3393         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3394                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3395         if (lockowner_slab == NULL)
3396                 goto out_free_openowner_slab;
3397         file_slab = kmem_cache_create("nfsd4_files",
3398                         sizeof(struct nfs4_file), 0, 0, NULL);
3399         if (file_slab == NULL)
3400                 goto out_free_lockowner_slab;
3401         stateid_slab = kmem_cache_create("nfsd4_stateids",
3402                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3403         if (stateid_slab == NULL)
3404                 goto out_free_file_slab;
3405         deleg_slab = kmem_cache_create("nfsd4_delegations",
3406                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3407         if (deleg_slab == NULL)
3408                 goto out_free_stateid_slab;
3409         odstate_slab = kmem_cache_create("nfsd4_odstate",
3410                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3411         if (odstate_slab == NULL)
3412                 goto out_free_deleg_slab;
3413         return 0;
3414
3415 out_free_deleg_slab:
3416         kmem_cache_destroy(deleg_slab);
3417 out_free_stateid_slab:
3418         kmem_cache_destroy(stateid_slab);
3419 out_free_file_slab:
3420         kmem_cache_destroy(file_slab);
3421 out_free_lockowner_slab:
3422         kmem_cache_destroy(lockowner_slab);
3423 out_free_openowner_slab:
3424         kmem_cache_destroy(openowner_slab);
3425 out:
3426         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3427         return -ENOMEM;
3428 }
3429
3430 static void init_nfs4_replay(struct nfs4_replay *rp)
3431 {
3432         rp->rp_status = nfserr_serverfault;
3433         rp->rp_buflen = 0;
3434         rp->rp_buf = rp->rp_ibuf;
3435         mutex_init(&rp->rp_mutex);
3436 }
3437
3438 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3439                 struct nfs4_stateowner *so)
3440 {
3441         if (!nfsd4_has_session(cstate)) {
3442                 mutex_lock(&so->so_replay.rp_mutex);
3443                 cstate->replay_owner = nfs4_get_stateowner(so);
3444         }
3445 }
3446
3447 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3448 {
3449         struct nfs4_stateowner *so = cstate->replay_owner;
3450
3451         if (so != NULL) {
3452                 cstate->replay_owner = NULL;
3453                 mutex_unlock(&so->so_replay.rp_mutex);
3454                 nfs4_put_stateowner(so);
3455         }
3456 }
3457
3458 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3459 {
3460         struct nfs4_stateowner *sop;
3461
3462         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3463         if (!sop)
3464                 return NULL;
3465
3466         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3467         if (!sop->so_owner.data) {
3468                 kmem_cache_free(slab, sop);
3469                 return NULL;
3470         }
3471         sop->so_owner.len = owner->len;
3472
3473         INIT_LIST_HEAD(&sop->so_stateids);
3474         sop->so_client = clp;
3475         init_nfs4_replay(&sop->so_replay);
3476         atomic_set(&sop->so_count, 1);
3477         return sop;
3478 }
3479
3480 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3481 {
3482         lockdep_assert_held(&clp->cl_lock);
3483
3484         list_add(&oo->oo_owner.so_strhash,
3485                  &clp->cl_ownerstr_hashtbl[strhashval]);
3486         list_add(&oo->oo_perclient, &clp->cl_openowners);
3487 }
3488
3489 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3490 {
3491         unhash_openowner_locked(openowner(so));
3492 }
3493
3494 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3495 {
3496         struct nfs4_openowner *oo = openowner(so);
3497
3498         kmem_cache_free(openowner_slab, oo);
3499 }
3500
3501 static const struct nfs4_stateowner_operations openowner_ops = {
3502         .so_unhash =    nfs4_unhash_openowner,
3503         .so_free =      nfs4_free_openowner,
3504 };
3505
3506 static struct nfs4_ol_stateid *
3507 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3508 {
3509         struct nfs4_ol_stateid *local, *ret = NULL;
3510         struct nfs4_openowner *oo = open->op_openowner;
3511
3512         lockdep_assert_held(&fp->fi_lock);
3513
3514         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3515                 /* ignore lock owners */
3516                 if (local->st_stateowner->so_is_open_owner == 0)
3517                         continue;
3518                 if (local->st_stateowner == &oo->oo_owner) {
3519                         ret = local;
3520                         atomic_inc(&ret->st_stid.sc_count);
3521                         break;
3522                 }
3523         }
3524         return ret;
3525 }
3526
3527 static struct nfs4_openowner *
3528 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3529                            struct nfsd4_compound_state *cstate)
3530 {
3531         struct nfs4_client *clp = cstate->clp;
3532         struct nfs4_openowner *oo, *ret;
3533
3534         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3535         if (!oo)
3536                 return NULL;
3537         oo->oo_owner.so_ops = &openowner_ops;
3538         oo->oo_owner.so_is_open_owner = 1;
3539         oo->oo_owner.so_seqid = open->op_seqid;
3540         oo->oo_flags = 0;
3541         if (nfsd4_has_session(cstate))
3542                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3543         oo->oo_time = 0;
3544         oo->oo_last_closed_stid = NULL;
3545         INIT_LIST_HEAD(&oo->oo_close_lru);
3546         spin_lock(&clp->cl_lock);
3547         ret = find_openstateowner_str_locked(strhashval, open, clp);
3548         if (ret == NULL) {
3549                 hash_openowner(oo, clp, strhashval);
3550                 ret = oo;
3551         } else
3552                 nfs4_free_stateowner(&oo->oo_owner);
3553
3554         spin_unlock(&clp->cl_lock);
3555         return ret;
3556 }
3557
3558 static struct nfs4_ol_stateid *
3559 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3560 {
3561
3562         struct nfs4_openowner *oo = open->op_openowner;
3563         struct nfs4_ol_stateid *retstp = NULL;
3564         struct nfs4_ol_stateid *stp;
3565
3566         stp = open->op_stp;
3567         /* We are moving these outside of the spinlocks to avoid the warnings */
3568         mutex_init(&stp->st_mutex);
3569         mutex_lock(&stp->st_mutex);
3570
3571         spin_lock(&oo->oo_owner.so_client->cl_lock);
3572         spin_lock(&fp->fi_lock);
3573
3574         retstp = nfsd4_find_existing_open(fp, open);
3575         if (retstp)
3576                 goto out_unlock;
3577
3578         open->op_stp = NULL;
3579         atomic_inc(&stp->st_stid.sc_count);
3580         stp->st_stid.sc_type = NFS4_OPEN_STID;
3581         INIT_LIST_HEAD(&stp->st_locks);
3582         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3583         get_nfs4_file(fp);
3584         stp->st_stid.sc_file = fp;
3585         stp->st_access_bmap = 0;
3586         stp->st_deny_bmap = 0;
3587         stp->st_openstp = NULL;
3588         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3589         list_add(&stp->st_perfile, &fp->fi_stateids);
3590
3591 out_unlock:
3592         spin_unlock(&fp->fi_lock);
3593         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3594         if (retstp) {
3595                 mutex_lock(&retstp->st_mutex);
3596                 /* To keep mutex tracking happy */
3597                 mutex_unlock(&stp->st_mutex);
3598                 stp = retstp;
3599         }
3600         return stp;
3601 }
3602
3603 /*
3604  * In the 4.0 case we need to keep the owners around a little while to handle
3605  * CLOSE replay. We still do need to release any file access that is held by
3606  * them before returning however.
3607  */
3608 static void
3609 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3610 {
3611         struct nfs4_ol_stateid *last;
3612         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3613         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3614                                                 nfsd_net_id);
3615
3616         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3617
3618         /*
3619          * We know that we hold one reference via nfsd4_close, and another
3620          * "persistent" reference for the client. If the refcount is higher
3621          * than 2, then there are still calls in progress that are using this
3622          * stateid. We can't put the sc_file reference until they are finished.
3623          * Wait for the refcount to drop to 2. Since it has been unhashed,
3624          * there should be no danger of the refcount going back up again at
3625          * this point.
3626          */
3627         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3628
3629         release_all_access(s);
3630         if (s->st_stid.sc_file) {
3631                 put_nfs4_file(s->st_stid.sc_file);
3632                 s->st_stid.sc_file = NULL;
3633         }
3634
3635         spin_lock(&nn->client_lock);
3636         last = oo->oo_last_closed_stid;
3637         oo->oo_last_closed_stid = s;
3638         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3639         oo->oo_time = get_seconds();
3640         spin_unlock(&nn->client_lock);
3641         if (last)
3642                 nfs4_put_stid(&last->st_stid);
3643 }
3644
3645 /* search file_hashtbl[] for file */
3646 static struct nfs4_file *
3647 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3648 {
3649         struct nfs4_file *fp;
3650
3651         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3652                 if (fh_match(&fp->fi_fhandle, fh)) {
3653                         if (atomic_inc_not_zero(&fp->fi_ref))
3654                                 return fp;
3655                 }
3656         }
3657         return NULL;
3658 }
3659
3660 struct nfs4_file *
3661 find_file(struct knfsd_fh *fh)
3662 {
3663         struct nfs4_file *fp;
3664         unsigned int hashval = file_hashval(fh);
3665
3666         rcu_read_lock();
3667         fp = find_file_locked(fh, hashval);
3668         rcu_read_unlock();
3669         return fp;
3670 }
3671
3672 static struct nfs4_file *
3673 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3674 {
3675         struct nfs4_file *fp;
3676         unsigned int hashval = file_hashval(fh);
3677
3678         rcu_read_lock();
3679         fp = find_file_locked(fh, hashval);
3680         rcu_read_unlock();
3681         if (fp)
3682                 return fp;
3683
3684         spin_lock(&state_lock);
3685         fp = find_file_locked(fh, hashval);
3686         if (likely(fp == NULL)) {
3687                 nfsd4_init_file(fh, hashval, new);
3688                 fp = new;
3689         }
3690         spin_unlock(&state_lock);
3691
3692         return fp;
3693 }
3694
3695 /*
3696  * Called to check deny when READ with all zero stateid or
3697  * WRITE with all zero or all one stateid
3698  */
3699 static __be32
3700 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3701 {
3702         struct nfs4_file *fp;
3703         __be32 ret = nfs_ok;
3704
3705         fp = find_file(&current_fh->fh_handle);
3706         if (!fp)
3707                 return ret;
3708         /* Check for conflicting share reservations */
3709         spin_lock(&fp->fi_lock);
3710         if (fp->fi_share_deny & deny_type)
3711                 ret = nfserr_locked;
3712         spin_unlock(&fp->fi_lock);
3713         put_nfs4_file(fp);
3714         return ret;
3715 }
3716
3717 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3718 {
3719         struct nfs4_delegation *dp = cb_to_delegation(cb);
3720         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3721                                           nfsd_net_id);
3722
3723         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3724
3725         /*
3726          * We can't do this in nfsd_break_deleg_cb because it is
3727          * already holding inode->i_lock.
3728          *
3729          * If the dl_time != 0, then we know that it has already been
3730          * queued for a lease break. Don't queue it again.
3731          */
3732         spin_lock(&state_lock);
3733         if (dp->dl_time == 0) {
3734                 dp->dl_time = get_seconds();
3735                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3736         }
3737         spin_unlock(&state_lock);
3738 }
3739
3740 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3741                 struct rpc_task *task)
3742 {
3743         struct nfs4_delegation *dp = cb_to_delegation(cb);
3744
3745         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3746                 return 1;
3747
3748         switch (task->tk_status) {
3749         case 0:
3750                 return 1;
3751         case -EBADHANDLE:
3752         case -NFS4ERR_BAD_STATEID:
3753                 /*
3754                  * Race: client probably got cb_recall before open reply
3755                  * granting delegation.
3756                  */
3757                 if (dp->dl_retries--) {
3758                         rpc_delay(task, 2 * HZ);
3759                         return 0;
3760                 }
3761                 /*FALLTHRU*/
3762         default:
3763                 return -1;
3764         }
3765 }
3766
3767 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3768 {
3769         struct nfs4_delegation *dp = cb_to_delegation(cb);
3770
3771         nfs4_put_stid(&dp->dl_stid);
3772 }
3773
3774 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3775         .prepare        = nfsd4_cb_recall_prepare,
3776         .done           = nfsd4_cb_recall_done,
3777         .release        = nfsd4_cb_recall_release,
3778 };
3779
3780 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3781 {
3782         /*
3783          * We're assuming the state code never drops its reference
3784          * without first removing the lease.  Since we're in this lease
3785          * callback (and since the lease code is serialized by the kernel
3786          * lock) we know the server hasn't removed the lease yet, we know
3787          * it's safe to take a reference.
3788          */
3789         atomic_inc(&dp->dl_stid.sc_count);
3790         nfsd4_run_cb(&dp->dl_recall);
3791 }
3792
3793 /* Called from break_lease() with i_lock held. */
3794 static bool
3795 nfsd_break_deleg_cb(struct file_lock *fl)
3796 {
3797         bool ret = false;
3798         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3799         struct nfs4_delegation *dp;
3800
3801         if (!fp) {
3802                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3803                 return ret;
3804         }
3805         if (fp->fi_had_conflict) {
3806                 WARN(1, "duplicate break on %p\n", fp);
3807                 return ret;
3808         }
3809         /*
3810          * We don't want the locks code to timeout the lease for us;
3811          * we'll remove it ourself if a delegation isn't returned
3812          * in time:
3813          */
3814         fl->fl_break_time = 0;
3815
3816         spin_lock(&fp->fi_lock);
3817         fp->fi_had_conflict = true;
3818         /*
3819          * If there are no delegations on the list, then return true
3820          * so that the lease code will go ahead and delete it.
3821          */
3822         if (list_empty(&fp->fi_delegations))
3823                 ret = true;
3824         else
3825                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3826                         nfsd_break_one_deleg(dp);
3827         spin_unlock(&fp->fi_lock);
3828         return ret;
3829 }
3830
3831 static int
3832 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3833                      struct list_head *dispose)
3834 {
3835         if (arg & F_UNLCK)
3836                 return lease_modify(onlist, arg, dispose);
3837         else
3838                 return -EAGAIN;
3839 }
3840
3841 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3842         .lm_break = nfsd_break_deleg_cb,
3843         .lm_change = nfsd_change_deleg_cb,
3844 };
3845
3846 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3847 {
3848         if (nfsd4_has_session(cstate))
3849                 return nfs_ok;
3850         if (seqid == so->so_seqid - 1)
3851                 return nfserr_replay_me;
3852         if (seqid == so->so_seqid)
3853                 return nfs_ok;
3854         return nfserr_bad_seqid;
3855 }
3856
3857 static __be32 lookup_clientid(clientid_t *clid,
3858                 struct nfsd4_compound_state *cstate,
3859                 struct nfsd_net *nn)
3860 {
3861         struct nfs4_client *found;
3862
3863         if (cstate->clp) {
3864                 found = cstate->clp;
3865                 if (!same_clid(&found->cl_clientid, clid))
3866                         return nfserr_stale_clientid;
3867                 return nfs_ok;
3868         }
3869
3870         if (STALE_CLIENTID(clid, nn))
3871                 return nfserr_stale_clientid;
3872
3873         /*
3874          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3875          * cached already then we know this is for is for v4.0 and "sessions"
3876          * will be false.
3877          */
3878         WARN_ON_ONCE(cstate->session);
3879         spin_lock(&nn->client_lock);
3880         found = find_confirmed_client(clid, false, nn);
3881         if (!found) {
3882                 spin_unlock(&nn->client_lock);
3883                 return nfserr_expired;
3884         }
3885         atomic_inc(&found->cl_refcount);
3886         spin_unlock(&nn->client_lock);
3887
3888         /* Cache the nfs4_client in cstate! */
3889         cstate->clp = found;
3890         return nfs_ok;
3891 }
3892
3893 __be32
3894 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3895                     struct nfsd4_open *open, struct nfsd_net *nn)
3896 {
3897         clientid_t *clientid = &open->op_clientid;
3898         struct nfs4_client *clp = NULL;
3899         unsigned int strhashval;
3900         struct nfs4_openowner *oo = NULL;
3901         __be32 status;
3902
3903         if (STALE_CLIENTID(&open->op_clientid, nn))
3904                 return nfserr_stale_clientid;
3905         /*
3906          * In case we need it later, after we've already created the
3907          * file and don't want to risk a further failure:
3908          */
3909         open->op_file = nfsd4_alloc_file();
3910         if (open->op_file == NULL)
3911                 return nfserr_jukebox;
3912
3913         status = lookup_clientid(clientid, cstate, nn);
3914         if (status)
3915                 return status;
3916         clp = cstate->clp;
3917
3918         strhashval = ownerstr_hashval(&open->op_owner);
3919         oo = find_openstateowner_str(strhashval, open, clp);
3920         open->op_openowner = oo;
3921         if (!oo) {
3922                 goto new_owner;
3923         }
3924         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3925                 /* Replace unconfirmed owners without checking for replay. */
3926                 release_openowner(oo);
3927                 open->op_openowner = NULL;
3928                 goto new_owner;
3929         }
3930         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3931         if (status)
3932                 return status;
3933         goto alloc_stateid;
3934 new_owner:
3935         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3936         if (oo == NULL)
3937                 return nfserr_jukebox;
3938         open->op_openowner = oo;
3939 alloc_stateid:
3940         open->op_stp = nfs4_alloc_open_stateid(clp);
3941         if (!open->op_stp)
3942                 return nfserr_jukebox;
3943
3944         if (nfsd4_has_session(cstate) &&
3945             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3946                 open->op_odstate = alloc_clnt_odstate(clp);
3947                 if (!open->op_odstate)
3948                         return nfserr_jukebox;
3949         }
3950
3951         return nfs_ok;
3952 }
3953
3954 static inline __be32
3955 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3956 {
3957         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3958                 return nfserr_openmode;
3959         else
3960                 return nfs_ok;
3961 }
3962
3963 static int share_access_to_flags(u32 share_access)
3964 {
3965         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3966 }
3967
3968 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3969 {
3970         struct nfs4_stid *ret;
3971
3972         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3973         if (!ret)
3974                 return NULL;
3975         return delegstateid(ret);
3976 }
3977
3978 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3979 {
3980         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3981                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3982 }
3983
3984 static __be32
3985 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3986                 struct nfs4_delegation **dp)
3987 {
3988         int flags;
3989         __be32 status = nfserr_bad_stateid;
3990         struct nfs4_delegation *deleg;
3991
3992         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3993         if (deleg == NULL)
3994                 goto out;
3995         flags = share_access_to_flags(open->op_share_access);
3996         status = nfs4_check_delegmode(deleg, flags);
3997         if (status) {
3998                 nfs4_put_stid(&deleg->dl_stid);
3999                 goto out;
4000         }
4001         *dp = deleg;
4002 out:
4003         if (!nfsd4_is_deleg_cur(open))
4004                 return nfs_ok;
4005         if (status)
4006                 return status;
4007         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4008         return nfs_ok;
4009 }
4010
4011 static inline int nfs4_access_to_access(u32 nfs4_access)
4012 {
4013         int flags = 0;
4014
4015         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4016                 flags |= NFSD_MAY_READ;
4017         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4018                 flags |= NFSD_MAY_WRITE;
4019         return flags;
4020 }
4021
4022 static inline __be32
4023 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4024                 struct nfsd4_open *open)
4025 {
4026         struct iattr iattr = {
4027                 .ia_valid = ATTR_SIZE,
4028                 .ia_size = 0,
4029         };
4030         if (!open->op_truncate)
4031                 return 0;
4032         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4033                 return nfserr_inval;
4034         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4035 }
4036
4037 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4038                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4039                 struct nfsd4_open *open)
4040 {
4041         struct file *filp = NULL;
4042         __be32 status;
4043         int oflag = nfs4_access_to_omode(open->op_share_access);
4044         int access = nfs4_access_to_access(open->op_share_access);
4045         unsigned char old_access_bmap, old_deny_bmap;
4046
4047         spin_lock(&fp->fi_lock);
4048
4049         /*
4050          * Are we trying to set a deny mode that would conflict with
4051          * current access?
4052          */
4053         status = nfs4_file_check_deny(fp, open->op_share_deny);
4054         if (status != nfs_ok) {
4055                 spin_unlock(&fp->fi_lock);
4056                 goto out;
4057         }
4058
4059         /* set access to the file */
4060         status = nfs4_file_get_access(fp, open->op_share_access);
4061         if (status != nfs_ok) {
4062                 spin_unlock(&fp->fi_lock);
4063                 goto out;
4064         }
4065
4066         /* Set access bits in stateid */
4067         old_access_bmap = stp->st_access_bmap;
4068         set_access(open->op_share_access, stp);
4069
4070         /* Set new deny mask */
4071         old_deny_bmap = stp->st_deny_bmap;
4072         set_deny(open->op_share_deny, stp);
4073         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4074
4075         if (!fp->fi_fds[oflag]) {
4076                 spin_unlock(&fp->fi_lock);
4077                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4078                 if (status)
4079                         goto out_put_access;
4080                 spin_lock(&fp->fi_lock);
4081                 if (!fp->fi_fds[oflag]) {
4082                         fp->fi_fds[oflag] = filp;
4083                         filp = NULL;
4084                 }
4085         }
4086         spin_unlock(&fp->fi_lock);
4087         if (filp)
4088                 fput(filp);
4089
4090         status = nfsd4_truncate(rqstp, cur_fh, open);
4091         if (status)
4092                 goto out_put_access;
4093 out:
4094         return status;
4095 out_put_access:
4096         stp->st_access_bmap = old_access_bmap;
4097         nfs4_file_put_access(fp, open->op_share_access);
4098         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4099         goto out;
4100 }
4101
4102 static __be32
4103 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)
4104 {
4105         __be32 status;
4106         unsigned char old_deny_bmap = stp->st_deny_bmap;
4107
4108         if (!test_access(open->op_share_access, stp))
4109                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4110
4111         /* test and set deny mode */
4112         spin_lock(&fp->fi_lock);
4113         status = nfs4_file_check_deny(fp, open->op_share_deny);
4114         if (status == nfs_ok) {
4115                 set_deny(open->op_share_deny, stp);
4116                 fp->fi_share_deny |=
4117                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4118         }
4119         spin_unlock(&fp->fi_lock);
4120
4121         if (status != nfs_ok)
4122                 return status;
4123
4124         status = nfsd4_truncate(rqstp, cur_fh, open);
4125         if (status != nfs_ok)
4126                 reset_union_bmap_deny(old_deny_bmap, stp);
4127         return status;
4128 }
4129
4130 /* Should we give out recallable state?: */
4131 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4132 {
4133         if (clp->cl_cb_state == NFSD4_CB_UP)
4134                 return true;
4135         /*
4136          * In the sessions case, since we don't have to establish a
4137          * separate connection for callbacks, we assume it's OK
4138          * until we hear otherwise:
4139          */
4140         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4141 }
4142
4143 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4144 {
4145         struct file_lock *fl;
4146
4147         fl = locks_alloc_lock();
4148         if (!fl)
4149                 return NULL;
4150         fl->fl_lmops = &nfsd_lease_mng_ops;
4151         fl->fl_flags = FL_DELEG;
4152         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4153         fl->fl_end = OFFSET_MAX;
4154         fl->fl_owner = (fl_owner_t)fp;
4155         fl->fl_pid = current->tgid;
4156         return fl;
4157 }
4158
4159 /**
4160  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4161  * @dp:   a pointer to the nfs4_delegation we're adding.
4162  *
4163  * Return:
4164  *      On success: Return code will be 0 on success.
4165  *
4166  *      On error: -EAGAIN if there was an existing delegation.
4167  *                 nonzero if there is an error in other cases.
4168  *
4169  */
4170
4171 static int nfs4_setlease(struct nfs4_delegation *dp)
4172 {
4173         struct nfs4_file *fp = dp->dl_stid.sc_file;
4174         struct file_lock *fl;
4175         struct file *filp;
4176         int status = 0;
4177
4178         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4179         if (!fl)
4180                 return -ENOMEM;
4181         filp = find_readable_file(fp);
4182         if (!filp) {
4183                 /* We should always have a readable file here */
4184                 WARN_ON_ONCE(1);
4185                 locks_free_lock(fl);
4186                 return -EBADF;
4187         }
4188         fl->fl_file = filp;
4189         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4190         if (fl)
4191                 locks_free_lock(fl);
4192         if (status)
4193                 goto out_fput;
4194         spin_lock(&state_lock);
4195         spin_lock(&fp->fi_lock);
4196         /* Did the lease get broken before we took the lock? */
4197         status = -EAGAIN;
4198         if (fp->fi_had_conflict)
4199                 goto out_unlock;
4200         /* Race breaker */
4201         if (fp->fi_deleg_file) {
4202                 status = hash_delegation_locked(dp, fp);
4203                 goto out_unlock;
4204         }
4205         fp->fi_deleg_file = filp;
4206         fp->fi_delegees = 0;
4207         status = hash_delegation_locked(dp, fp);
4208         spin_unlock(&fp->fi_lock);
4209         spin_unlock(&state_lock);
4210         if (status) {
4211                 /* Should never happen, this is a new fi_deleg_file  */
4212                 WARN_ON_ONCE(1);
4213                 goto out_fput;
4214         }
4215         return 0;
4216 out_unlock:
4217         spin_unlock(&fp->fi_lock);
4218         spin_unlock(&state_lock);
4219 out_fput:
4220         fput(filp);
4221         return status;
4222 }
4223
4224 static struct nfs4_delegation *
4225 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4226                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4227 {
4228         int status;
4229         struct nfs4_delegation *dp;
4230
4231         if (fp->fi_had_conflict)
4232                 return ERR_PTR(-EAGAIN);
4233
4234         spin_lock(&state_lock);
4235         spin_lock(&fp->fi_lock);
4236         status = nfs4_get_existing_delegation(clp, fp);
4237         spin_unlock(&fp->fi_lock);
4238         spin_unlock(&state_lock);
4239
4240         if (status)
4241                 return ERR_PTR(status);
4242
4243         dp = alloc_init_deleg(clp, fh, odstate);
4244         if (!dp)
4245                 return ERR_PTR(-ENOMEM);
4246
4247         get_nfs4_file(fp);
4248         spin_lock(&state_lock);
4249         spin_lock(&fp->fi_lock);
4250         dp->dl_stid.sc_file = fp;
4251         if (!fp->fi_deleg_file) {
4252                 spin_unlock(&fp->fi_lock);
4253                 spin_unlock(&state_lock);
4254                 status = nfs4_setlease(dp);
4255                 goto out;
4256         }
4257         if (fp->fi_had_conflict) {
4258                 status = -EAGAIN;
4259                 goto out_unlock;
4260         }
4261         status = hash_delegation_locked(dp, fp);
4262 out_unlock:
4263         spin_unlock(&fp->fi_lock);
4264         spin_unlock(&state_lock);
4265 out:
4266         if (status) {
4267                 put_clnt_odstate(dp->dl_clnt_odstate);
4268                 nfs4_put_stid(&dp->dl_stid);
4269                 return ERR_PTR(status);
4270         }
4271         return dp;
4272 }
4273
4274 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4275 {
4276         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4277         if (status == -EAGAIN)
4278                 open->op_why_no_deleg = WND4_CONTENTION;
4279         else {
4280                 open->op_why_no_deleg = WND4_RESOURCE;
4281                 switch (open->op_deleg_want) {
4282                 case NFS4_SHARE_WANT_READ_DELEG:
4283                 case NFS4_SHARE_WANT_WRITE_DELEG:
4284                 case NFS4_SHARE_WANT_ANY_DELEG:
4285                         break;
4286                 case NFS4_SHARE_WANT_CANCEL:
4287                         open->op_why_no_deleg = WND4_CANCELLED;
4288                         break;
4289                 case NFS4_SHARE_WANT_NO_DELEG:
4290                         WARN_ON_ONCE(1);
4291                 }
4292         }
4293 }
4294
4295 /*
4296  * Attempt to hand out a delegation.
4297  *
4298  * Note we don't support write delegations, and won't until the vfs has
4299  * proper support for them.
4300  */
4301 static void
4302 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4303                         struct nfs4_ol_stateid *stp)
4304 {
4305         struct nfs4_delegation *dp;
4306         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4307         struct nfs4_client *clp = stp->st_stid.sc_client;
4308         int cb_up;
4309         int status = 0;
4310
4311         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4312         open->op_recall = 0;
4313         switch (open->op_claim_type) {
4314                 case NFS4_OPEN_CLAIM_PREVIOUS:
4315                         if (!cb_up)
4316                                 open->op_recall = 1;
4317                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4318                                 goto out_no_deleg;
4319                         break;
4320                 case NFS4_OPEN_CLAIM_NULL:
4321                 case NFS4_OPEN_CLAIM_FH:
4322                         /*
4323                          * Let's not give out any delegations till everyone's
4324                          * had the chance to reclaim theirs, *and* until
4325                          * NLM locks have all been reclaimed:
4326                          */
4327                         if (locks_in_grace(clp->net))
4328                                 goto out_no_deleg;
4329                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4330                                 goto out_no_deleg;
4331                         /*
4332                          * Also, if the file was opened for write or
4333                          * create, there's a good chance the client's
4334                          * about to write to it, resulting in an
4335                          * immediate recall (since we don't support
4336                          * write delegations):
4337                          */
4338                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4339                                 goto out_no_deleg;
4340                         if (open->op_create == NFS4_OPEN_CREATE)
4341                                 goto out_no_deleg;
4342                         break;
4343                 default:
4344                         goto out_no_deleg;
4345         }
4346         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4347         if (IS_ERR(dp))
4348                 goto out_no_deleg;
4349
4350         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4351
4352         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4353                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4354         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4355         nfs4_put_stid(&dp->dl_stid);
4356         return;
4357 out_no_deleg:
4358         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4359         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4360             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4361                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4362                 open->op_recall = 1;
4363         }
4364
4365         /* 4.1 client asking for a delegation? */
4366         if (open->op_deleg_want)
4367                 nfsd4_open_deleg_none_ext(open, status);
4368         return;
4369 }
4370
4371 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4372                                         struct nfs4_delegation *dp)
4373 {
4374         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4375             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4376                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4377                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4378         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4379                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4380                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4381                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4382         }
4383         /* Otherwise the client must be confused wanting a delegation
4384          * it already has, therefore we don't return
4385          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4386          */
4387 }
4388
4389 __be32
4390 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4391 {
4392         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4393         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4394         struct nfs4_file *fp = NULL;
4395         struct nfs4_ol_stateid *stp = NULL;
4396         struct nfs4_delegation *dp = NULL;
4397         __be32 status;
4398
4399         /*
4400          * Lookup file; if found, lookup stateid and check open request,
4401          * and check for delegations in the process of being recalled.
4402          * If not found, create the nfs4_file struct
4403          */
4404         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4405         if (fp != open->op_file) {
4406                 status = nfs4_check_deleg(cl, open, &dp);
4407                 if (status)
4408                         goto out;
4409                 spin_lock(&fp->fi_lock);
4410                 stp = nfsd4_find_existing_open(fp, open);
4411                 spin_unlock(&fp->fi_lock);
4412         } else {
4413                 open->op_file = NULL;
4414                 status = nfserr_bad_stateid;
4415                 if (nfsd4_is_deleg_cur(open))
4416                         goto out;
4417         }
4418
4419         /*
4420          * OPEN the file, or upgrade an existing OPEN.
4421          * If truncate fails, the OPEN fails.
4422          */
4423         if (stp) {
4424                 /* Stateid was found, this is an OPEN upgrade */
4425                 mutex_lock(&stp->st_mutex);
4426                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4427                 if (status) {
4428                         mutex_unlock(&stp->st_mutex);
4429                         goto out;
4430                 }
4431         } else {
4432                 /* stp is returned locked. */
4433                 stp = init_open_stateid(fp, open);
4434                 /* See if we lost the race to some other thread */
4435                 if (stp->st_access_bmap != 0) {
4436                         status = nfs4_upgrade_open(rqstp, fp, current_fh,
4437                                                 stp, open);
4438                         if (status) {
4439                                 mutex_unlock(&stp->st_mutex);
4440                                 goto out;
4441                         }
4442                         goto upgrade_out;
4443                 }
4444                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4445                 if (status) {
4446                         mutex_unlock(&stp->st_mutex);
4447                         release_open_stateid(stp);
4448                         goto out;
4449                 }
4450
4451                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4452                                                         open->op_odstate);
4453                 if (stp->st_clnt_odstate == open->op_odstate)
4454                         open->op_odstate = NULL;
4455         }
4456 upgrade_out:
4457         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4458         mutex_unlock(&stp->st_mutex);
4459
4460         if (nfsd4_has_session(&resp->cstate)) {
4461                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4462                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4463                         open->op_why_no_deleg = WND4_NOT_WANTED;
4464                         goto nodeleg;
4465                 }
4466         }
4467
4468         /*
4469         * Attempt to hand out a delegation. No error return, because the
4470         * OPEN succeeds even if we fail.
4471         */
4472         nfs4_open_delegation(current_fh, open, stp);
4473 nodeleg:
4474         status = nfs_ok;
4475
4476         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4477                 STATEID_VAL(&stp->st_stid.sc_stateid));
4478 out:
4479         /* 4.1 client trying to upgrade/downgrade delegation? */
4480         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4481             open->op_deleg_want)
4482                 nfsd4_deleg_xgrade_none_ext(open, dp);
4483
4484         if (fp)
4485                 put_nfs4_file(fp);
4486         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4487                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4488         /*
4489         * To finish the open response, we just need to set the rflags.
4490         */
4491         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4492         if (nfsd4_has_session(&resp->cstate))
4493                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4494         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4495                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4496
4497         if (dp)
4498                 nfs4_put_stid(&dp->dl_stid);
4499         if (stp)
4500                 nfs4_put_stid(&stp->st_stid);
4501
4502         return status;
4503 }
4504
4505 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4506                               struct nfsd4_open *open)
4507 {
4508         if (open->op_openowner) {
4509                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4510
4511                 nfsd4_cstate_assign_replay(cstate, so);
4512                 nfs4_put_stateowner(so);
4513         }
4514         if (open->op_file)
4515                 kmem_cache_free(file_slab, open->op_file);
4516         if (open->op_stp)
4517                 nfs4_put_stid(&open->op_stp->st_stid);
4518         if (open->op_odstate)
4519                 kmem_cache_free(odstate_slab, open->op_odstate);
4520 }
4521
4522 __be32
4523 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4524             clientid_t *clid)
4525 {
4526         struct nfs4_client *clp;
4527         __be32 status;
4528         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4529
4530         dprintk("process_renew(%08x/%08x): starting\n",
4531                         clid->cl_boot, clid->cl_id);
4532         status = lookup_clientid(clid, cstate, nn);
4533         if (status)
4534                 goto out;
4535         clp = cstate->clp;
4536         status = nfserr_cb_path_down;
4537         if (!list_empty(&clp->cl_delegations)
4538                         && clp->cl_cb_state != NFSD4_CB_UP)
4539                 goto out;
4540         status = nfs_ok;
4541 out:
4542         return status;
4543 }
4544
4545 void
4546 nfsd4_end_grace(struct nfsd_net *nn)
4547 {
4548         /* do nothing if grace period already ended */
4549         if (nn->grace_ended)
4550                 return;
4551
4552         dprintk("NFSD: end of grace period\n");
4553         nn->grace_ended = true;
4554         /*
4555          * If the server goes down again right now, an NFSv4
4556          * client will still be allowed to reclaim after it comes back up,
4557          * even if it hasn't yet had a chance to reclaim state this time.
4558          *
4559          */
4560         nfsd4_record_grace_done(nn);
4561         /*
4562          * At this point, NFSv4 clients can still reclaim.  But if the
4563          * server crashes, any that have not yet reclaimed will be out
4564          * of luck on the next boot.
4565          *
4566          * (NFSv4.1+ clients are considered to have reclaimed once they
4567          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4568          * have reclaimed after their first OPEN.)
4569          */
4570         locks_end_grace(&nn->nfsd4_manager);
4571         /*
4572          * At this point, and once lockd and/or any other containers
4573          * exit their grace period, further reclaims will fail and
4574          * regular locking can resume.
4575          */
4576 }
4577
4578 static time_t
4579 nfs4_laundromat(struct nfsd_net *nn)
4580 {
4581         struct nfs4_client *clp;
4582         struct nfs4_openowner *oo;
4583         struct nfs4_delegation *dp;
4584         struct nfs4_ol_stateid *stp;
4585         struct nfsd4_blocked_lock *nbl;
4586         struct list_head *pos, *next, reaplist;
4587         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4588         time_t t, new_timeo = nn->nfsd4_lease;
4589
4590         dprintk("NFSD: laundromat service - starting\n");
4591         nfsd4_end_grace(nn);
4592         INIT_LIST_HEAD(&reaplist);
4593         spin_lock(&nn->client_lock);
4594         list_for_each_safe(pos, next, &nn->client_lru) {
4595                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4596                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4597                         t = clp->cl_time - cutoff;
4598                         new_timeo = min(new_timeo, t);
4599                         break;
4600                 }
4601                 if (mark_client_expired_locked(clp)) {
4602                         dprintk("NFSD: client in use (clientid %08x)\n",
4603                                 clp->cl_clientid.cl_id);
4604                         continue;
4605                 }
4606                 list_add(&clp->cl_lru, &reaplist);
4607         }
4608         spin_unlock(&nn->client_lock);
4609         list_for_each_safe(pos, next, &reaplist) {
4610                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4611                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4612                         clp->cl_clientid.cl_id);
4613                 list_del_init(&clp->cl_lru);
4614                 expire_client(clp);
4615         }
4616         spin_lock(&state_lock);
4617         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4618                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4619                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4620                         t = dp->dl_time - cutoff;
4621                         new_timeo = min(new_timeo, t);
4622                         break;
4623                 }
4624                 WARN_ON(!unhash_delegation_locked(dp));
4625                 list_add(&dp->dl_recall_lru, &reaplist);
4626         }
4627         spin_unlock(&state_lock);
4628         while (!list_empty(&reaplist)) {
4629                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4630                                         dl_recall_lru);
4631                 list_del_init(&dp->dl_recall_lru);
4632                 revoke_delegation(dp);
4633         }
4634
4635         spin_lock(&nn->client_lock);
4636         while (!list_empty(&nn->close_lru)) {
4637                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4638                                         oo_close_lru);
4639                 if (time_after((unsigned long)oo->oo_time,
4640                                (unsigned long)cutoff)) {
4641                         t = oo->oo_time - cutoff;
4642                         new_timeo = min(new_timeo, t);
4643                         break;
4644                 }
4645                 list_del_init(&oo->oo_close_lru);
4646                 stp = oo->oo_last_closed_stid;
4647                 oo->oo_last_closed_stid = NULL;
4648                 spin_unlock(&nn->client_lock);
4649                 nfs4_put_stid(&stp->st_stid);
4650                 spin_lock(&nn->client_lock);
4651         }
4652         spin_unlock(&nn->client_lock);
4653
4654         /*
4655          * It's possible for a client to try and acquire an already held lock
4656          * that is being held for a long time, and then lose interest in it.
4657          * So, we clean out any un-revisited request after a lease period
4658          * under the assumption that the client is no longer interested.
4659          *
4660          * RFC5661, sec. 9.6 states that the client must not rely on getting
4661          * notifications and must continue to poll for locks, even when the
4662          * server supports them. Thus this shouldn't lead to clients blocking
4663          * indefinitely once the lock does become free.
4664          */
4665         BUG_ON(!list_empty(&reaplist));
4666         spin_lock(&nn->blocked_locks_lock);
4667         while (!list_empty(&nn->blocked_locks_lru)) {
4668                 nbl = list_first_entry(&nn->blocked_locks_lru,
4669                                         struct nfsd4_blocked_lock, nbl_lru);
4670                 if (time_after((unsigned long)nbl->nbl_time,
4671                                (unsigned long)cutoff)) {
4672                         t = nbl->nbl_time - cutoff;
4673                         new_timeo = min(new_timeo, t);
4674                         break;
4675                 }
4676                 list_move(&nbl->nbl_lru, &reaplist);
4677                 list_del_init(&nbl->nbl_list);
4678         }
4679         spin_unlock(&nn->blocked_locks_lock);
4680
4681         while (!list_empty(&reaplist)) {
4682                 nbl = list_first_entry(&nn->blocked_locks_lru,
4683                                         struct nfsd4_blocked_lock, nbl_lru);
4684                 list_del_init(&nbl->nbl_lru);
4685                 posix_unblock_lock(&nbl->nbl_lock);
4686                 free_blocked_lock(nbl);
4687         }
4688
4689         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4690         return new_timeo;
4691 }
4692
4693 static struct workqueue_struct *laundry_wq;
4694 static void laundromat_main(struct work_struct *);
4695
4696 static void
4697 laundromat_main(struct work_struct *laundry)
4698 {
4699         time_t t;
4700         struct delayed_work *dwork = to_delayed_work(laundry);
4701         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4702                                            laundromat_work);
4703
4704         t = nfs4_laundromat(nn);
4705         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4706         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4707 }
4708
4709 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4710 {
4711         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4712                 return nfserr_bad_stateid;
4713         return nfs_ok;
4714 }
4715
4716 static inline int
4717 access_permit_read(struct nfs4_ol_stateid *stp)
4718 {
4719         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4720                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4721                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4722 }
4723
4724 static inline int
4725 access_permit_write(struct nfs4_ol_stateid *stp)
4726 {
4727         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4728                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4729 }
4730
4731 static
4732 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4733 {
4734         __be32 status = nfserr_openmode;
4735
4736         /* For lock stateid's, we test the parent open, not the lock: */
4737         if (stp->st_openstp)
4738                 stp = stp->st_openstp;
4739         if ((flags & WR_STATE) && !access_permit_write(stp))
4740                 goto out;
4741         if ((flags & RD_STATE) && !access_permit_read(stp))
4742                 goto out;
4743         status = nfs_ok;
4744 out:
4745         return status;
4746 }
4747
4748 static inline __be32
4749 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4750 {
4751         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4752                 return nfs_ok;
4753         else if (opens_in_grace(net)) {
4754                 /* Answer in remaining cases depends on existence of
4755                  * conflicting state; so we must wait out the grace period. */
4756                 return nfserr_grace;
4757         } else if (flags & WR_STATE)
4758                 return nfs4_share_conflict(current_fh,
4759                                 NFS4_SHARE_DENY_WRITE);
4760         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4761                 return nfs4_share_conflict(current_fh,
4762                                 NFS4_SHARE_DENY_READ);
4763 }
4764
4765 /*
4766  * Allow READ/WRITE during grace period on recovered state only for files
4767  * that are not able to provide mandatory locking.
4768  */
4769 static inline int
4770 grace_disallows_io(struct net *net, struct inode *inode)
4771 {
4772         return opens_in_grace(net) && mandatory_lock(inode);
4773 }
4774
4775 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4776 {
4777         /*
4778          * When sessions are used the stateid generation number is ignored
4779          * when it is zero.
4780          */
4781         if (has_session && in->si_generation == 0)
4782                 return nfs_ok;
4783
4784         if (in->si_generation == ref->si_generation)
4785                 return nfs_ok;
4786
4787         /* If the client sends us a stateid from the future, it's buggy: */
4788         if (nfsd4_stateid_generation_after(in, ref))
4789                 return nfserr_bad_stateid;
4790         /*
4791          * However, we could see a stateid from the past, even from a
4792          * non-buggy client.  For example, if the client sends a lock
4793          * while some IO is outstanding, the lock may bump si_generation
4794          * while the IO is still in flight.  The client could avoid that
4795          * situation by waiting for responses on all the IO requests,
4796          * but better performance may result in retrying IO that
4797          * receives an old_stateid error if requests are rarely
4798          * reordered in flight:
4799          */
4800         return nfserr_old_stateid;
4801 }
4802
4803 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4804 {
4805         if (ols->st_stateowner->so_is_open_owner &&
4806             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4807                 return nfserr_bad_stateid;
4808         return nfs_ok;
4809 }
4810
4811 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4812 {
4813         struct nfs4_stid *s;
4814         __be32 status = nfserr_bad_stateid;
4815
4816         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4817                 return status;
4818         /* Client debugging aid. */
4819         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4820                 char addr_str[INET6_ADDRSTRLEN];
4821                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4822                                  sizeof(addr_str));
4823                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4824                                         "with incorrect client ID\n", addr_str);
4825                 return status;
4826         }
4827         spin_lock(&cl->cl_lock);
4828         s = find_stateid_locked(cl, stateid);
4829         if (!s)
4830                 goto out_unlock;
4831         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4832         if (status)
4833                 goto out_unlock;
4834         switch (s->sc_type) {
4835         case NFS4_DELEG_STID:
4836                 status = nfs_ok;
4837                 break;
4838         case NFS4_REVOKED_DELEG_STID:
4839                 status = nfserr_deleg_revoked;
4840                 break;
4841         case NFS4_OPEN_STID:
4842         case NFS4_LOCK_STID:
4843                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4844                 break;
4845         default:
4846                 printk("unknown stateid type %x\n", s->sc_type);
4847                 /* Fallthrough */
4848         case NFS4_CLOSED_STID:
4849         case NFS4_CLOSED_DELEG_STID:
4850                 status = nfserr_bad_stateid;
4851         }
4852 out_unlock:
4853         spin_unlock(&cl->cl_lock);
4854         return status;
4855 }
4856
4857 __be32
4858 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4859                      stateid_t *stateid, unsigned char typemask,
4860                      struct nfs4_stid **s, struct nfsd_net *nn)
4861 {
4862         __be32 status;
4863
4864         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4865                 return nfserr_bad_stateid;
4866         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4867         if (status == nfserr_stale_clientid) {
4868                 if (cstate->session)
4869                         return nfserr_bad_stateid;
4870                 return nfserr_stale_stateid;
4871         }
4872         if (status)
4873                 return status;
4874         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4875         if (!*s)
4876                 return nfserr_bad_stateid;
4877         return nfs_ok;
4878 }
4879
4880 static struct file *
4881 nfs4_find_file(struct nfs4_stid *s, int flags)
4882 {
4883         if (!s)
4884                 return NULL;
4885
4886         switch (s->sc_type) {
4887         case NFS4_DELEG_STID:
4888                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4889                         return NULL;
4890                 return get_file(s->sc_file->fi_deleg_file);
4891         case NFS4_OPEN_STID:
4892         case NFS4_LOCK_STID:
4893                 if (flags & RD_STATE)
4894                         return find_readable_file(s->sc_file);
4895                 else
4896                         return find_writeable_file(s->sc_file);
4897                 break;
4898         }
4899
4900         return NULL;
4901 }
4902
4903 static __be32
4904 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4905 {
4906         __be32 status;
4907
4908         status = nfsd4_check_openowner_confirmed(ols);
4909         if (status)
4910                 return status;
4911         return nfs4_check_openmode(ols, flags);
4912 }
4913
4914 static __be32
4915 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4916                 struct file **filpp, bool *tmp_file, int flags)
4917 {
4918         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4919         struct file *file;
4920         __be32 status;
4921
4922         file = nfs4_find_file(s, flags);
4923         if (file) {
4924                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4925                                 acc | NFSD_MAY_OWNER_OVERRIDE);
4926                 if (status) {
4927                         fput(file);
4928                         return status;
4929                 }
4930
4931                 *filpp = file;
4932         } else {
4933                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4934                 if (status)
4935                         return status;
4936
4937                 if (tmp_file)
4938                         *tmp_file = true;
4939         }
4940
4941         return 0;
4942 }
4943
4944 /*
4945  * Checks for stateid operations
4946  */
4947 __be32
4948 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4949                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4950                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4951 {
4952         struct inode *ino = d_inode(fhp->fh_dentry);
4953         struct net *net = SVC_NET(rqstp);
4954         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4955         struct nfs4_stid *s = NULL;
4956         __be32 status;
4957
4958         if (filpp)
4959                 *filpp = NULL;
4960         if (tmp_file)
4961                 *tmp_file = false;
4962
4963         if (grace_disallows_io(net, ino))
4964                 return nfserr_grace;
4965
4966         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4967                 status = check_special_stateids(net, fhp, stateid, flags);
4968                 goto done;
4969         }
4970
4971         status = nfsd4_lookup_stateid(cstate, stateid,
4972                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4973                                 &s, nn);
4974         if (status)
4975                 return status;
4976         status = check_stateid_generation(stateid, &s->sc_stateid,
4977                         nfsd4_has_session(cstate));
4978         if (status)
4979                 goto out;
4980
4981         switch (s->sc_type) {
4982         case NFS4_DELEG_STID:
4983                 status = nfs4_check_delegmode(delegstateid(s), flags);
4984                 break;
4985         case NFS4_OPEN_STID:
4986         case NFS4_LOCK_STID:
4987                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4988                 break;
4989         default:
4990                 status = nfserr_bad_stateid;
4991                 break;
4992         }
4993         if (status)
4994                 goto out;
4995         status = nfs4_check_fh(fhp, s);
4996
4997 done:
4998         if (!status && filpp)
4999                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5000 out:
5001         if (s)
5002                 nfs4_put_stid(s);
5003         return status;
5004 }
5005
5006 /*
5007  * Test if the stateid is valid
5008  */
5009 __be32
5010 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5011                    struct nfsd4_test_stateid *test_stateid)
5012 {
5013         struct nfsd4_test_stateid_id *stateid;
5014         struct nfs4_client *cl = cstate->session->se_client;
5015
5016         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5017                 stateid->ts_id_status =
5018                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5019
5020         return nfs_ok;
5021 }
5022
5023 static __be32
5024 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5025 {
5026         struct nfs4_ol_stateid *stp = openlockstateid(s);
5027         __be32 ret;
5028
5029         mutex_lock(&stp->st_mutex);
5030
5031         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5032         if (ret)
5033                 goto out;
5034
5035         ret = nfserr_locks_held;
5036         if (check_for_locks(stp->st_stid.sc_file,
5037                             lockowner(stp->st_stateowner)))
5038                 goto out;
5039
5040         release_lock_stateid(stp);
5041         ret = nfs_ok;
5042
5043 out:
5044         mutex_unlock(&stp->st_mutex);
5045         nfs4_put_stid(s);
5046         return ret;
5047 }
5048
5049 __be32
5050 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5051                    struct nfsd4_free_stateid *free_stateid)
5052 {
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                    struct nfsd4_open_confirm *oc)
5181 {
5182         __be32 status;
5183         struct nfs4_openowner *oo;
5184         struct nfs4_ol_stateid *stp;
5185         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5186
5187         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5188                         cstate->current_fh.fh_dentry);
5189
5190         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5191         if (status)
5192                 return status;
5193
5194         status = nfs4_preprocess_seqid_op(cstate,
5195                                         oc->oc_seqid, &oc->oc_req_stateid,
5196                                         NFS4_OPEN_STID, &stp, nn);
5197         if (status)
5198                 goto out;
5199         oo = openowner(stp->st_stateowner);
5200         status = nfserr_bad_stateid;
5201         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5202                 mutex_unlock(&stp->st_mutex);
5203                 goto put_stateid;
5204         }
5205         oo->oo_flags |= NFS4_OO_CONFIRMED;
5206         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5207         mutex_unlock(&stp->st_mutex);
5208         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5209                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5210
5211         nfsd4_client_record_create(oo->oo_owner.so_client);
5212         status = nfs_ok;
5213 put_stateid:
5214         nfs4_put_stid(&stp->st_stid);
5215 out:
5216         nfsd4_bump_seqid(cstate, status);
5217         return status;
5218 }
5219
5220 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5221 {
5222         if (!test_access(access, stp))
5223                 return;
5224         nfs4_file_put_access(stp->st_stid.sc_file, access);
5225         clear_access(access, stp);
5226 }
5227
5228 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5229 {
5230         switch (to_access) {
5231         case NFS4_SHARE_ACCESS_READ:
5232                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5233                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5234                 break;
5235         case NFS4_SHARE_ACCESS_WRITE:
5236                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5237                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5238                 break;
5239         case NFS4_SHARE_ACCESS_BOTH:
5240                 break;
5241         default:
5242                 WARN_ON_ONCE(1);
5243         }
5244 }
5245
5246 __be32
5247 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5248                      struct nfsd4_compound_state *cstate,
5249                      struct nfsd4_open_downgrade *od)
5250 {
5251         __be32 status;
5252         struct nfs4_ol_stateid *stp;
5253         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5254
5255         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5256                         cstate->current_fh.fh_dentry);
5257
5258         /* We don't yet support WANT bits: */
5259         if (od->od_deleg_want)
5260                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5261                         od->od_deleg_want);
5262
5263         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5264                                         &od->od_stateid, &stp, nn);
5265         if (status)
5266                 goto out;
5267         status = nfserr_inval;
5268         if (!test_access(od->od_share_access, stp)) {
5269                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5270                         stp->st_access_bmap, od->od_share_access);
5271                 goto put_stateid;
5272         }
5273         if (!test_deny(od->od_share_deny, stp)) {
5274                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5275                         stp->st_deny_bmap, od->od_share_deny);
5276                 goto put_stateid;
5277         }
5278         nfs4_stateid_downgrade(stp, od->od_share_access);
5279         reset_union_bmap_deny(od->od_share_deny, stp);
5280         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5281         status = nfs_ok;
5282 put_stateid:
5283         mutex_unlock(&stp->st_mutex);
5284         nfs4_put_stid(&stp->st_stid);
5285 out:
5286         nfsd4_bump_seqid(cstate, status);
5287         return status;
5288 }
5289
5290 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5291 {
5292         struct nfs4_client *clp = s->st_stid.sc_client;
5293         bool unhashed;
5294         LIST_HEAD(reaplist);
5295
5296         s->st_stid.sc_type = NFS4_CLOSED_STID;
5297         spin_lock(&clp->cl_lock);
5298         unhashed = unhash_open_stateid(s, &reaplist);
5299
5300         if (clp->cl_minorversion) {
5301                 if (unhashed)
5302                         put_ol_stateid_locked(s, &reaplist);
5303                 spin_unlock(&clp->cl_lock);
5304                 free_ol_stateid_reaplist(&reaplist);
5305         } else {
5306                 spin_unlock(&clp->cl_lock);
5307                 free_ol_stateid_reaplist(&reaplist);
5308                 if (unhashed)
5309                         move_to_close_lru(s, clp->net);
5310         }
5311 }
5312
5313 /*
5314  * nfs4_unlock_state() called after encode
5315  */
5316 __be32
5317 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5318             struct nfsd4_close *close)
5319 {
5320         __be32 status;
5321         struct nfs4_ol_stateid *stp;
5322         struct net *net = SVC_NET(rqstp);
5323         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5324
5325         dprintk("NFSD: nfsd4_close on file %pd\n",
5326                         cstate->current_fh.fh_dentry);
5327
5328         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5329                                         &close->cl_stateid,
5330                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5331                                         &stp, nn);
5332         nfsd4_bump_seqid(cstate, status);
5333         if (status)
5334                 goto out;
5335         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5336         mutex_unlock(&stp->st_mutex);
5337
5338         nfsd4_close_open_stateid(stp);
5339
5340         /* put reference from nfs4_preprocess_seqid_op */
5341         nfs4_put_stid(&stp->st_stid);
5342 out:
5343         return status;
5344 }
5345
5346 __be32
5347 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5348                   struct nfsd4_delegreturn *dr)
5349 {
5350         struct nfs4_delegation *dp;
5351         stateid_t *stateid = &dr->dr_stateid;
5352         struct nfs4_stid *s;
5353         __be32 status;
5354         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5355
5356         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5357                 return status;
5358
5359         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5360         if (status)
5361                 goto out;
5362         dp = delegstateid(s);
5363         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5364         if (status)
5365                 goto put_stateid;
5366
5367         destroy_delegation(dp);
5368 put_stateid:
5369         nfs4_put_stid(&dp->dl_stid);
5370 out:
5371         return status;
5372 }
5373
5374 static inline u64
5375 end_offset(u64 start, u64 len)
5376 {
5377         u64 end;
5378
5379         end = start + len;
5380         return end >= start ? end: NFS4_MAX_UINT64;
5381 }
5382
5383 /* last octet in a range */
5384 static inline u64
5385 last_byte_offset(u64 start, u64 len)
5386 {
5387         u64 end;
5388
5389         WARN_ON_ONCE(!len);
5390         end = start + len;
5391         return end > start ? end - 1: NFS4_MAX_UINT64;
5392 }
5393
5394 /*
5395  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5396  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5397  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5398  * locking, this prevents us from being completely protocol-compliant.  The
5399  * real solution to this problem is to start using unsigned file offsets in
5400  * the VFS, but this is a very deep change!
5401  */
5402 static inline void
5403 nfs4_transform_lock_offset(struct file_lock *lock)
5404 {
5405         if (lock->fl_start < 0)
5406                 lock->fl_start = OFFSET_MAX;
5407         if (lock->fl_end < 0)
5408                 lock->fl_end = OFFSET_MAX;
5409 }
5410
5411 static fl_owner_t
5412 nfsd4_fl_get_owner(fl_owner_t owner)
5413 {
5414         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5415
5416         nfs4_get_stateowner(&lo->lo_owner);
5417         return owner;
5418 }
5419
5420 static void
5421 nfsd4_fl_put_owner(fl_owner_t owner)
5422 {
5423         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5424
5425         if (lo)
5426                 nfs4_put_stateowner(&lo->lo_owner);
5427 }
5428
5429 static void
5430 nfsd4_lm_notify(struct file_lock *fl)
5431 {
5432         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5433         struct net                      *net = lo->lo_owner.so_client->net;
5434         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5435         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5436                                                 struct nfsd4_blocked_lock, nbl_lock);
5437         bool queue = false;
5438
5439         /* An empty list means that something else is going to be using it */
5440         spin_lock(&nn->blocked_locks_lock);
5441         if (!list_empty(&nbl->nbl_list)) {
5442                 list_del_init(&nbl->nbl_list);
5443                 list_del_init(&nbl->nbl_lru);
5444                 queue = true;
5445         }
5446         spin_unlock(&nn->blocked_locks_lock);
5447
5448         if (queue)
5449                 nfsd4_run_cb(&nbl->nbl_cb);
5450 }
5451
5452 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5453         .lm_notify = nfsd4_lm_notify,
5454         .lm_get_owner = nfsd4_fl_get_owner,
5455         .lm_put_owner = nfsd4_fl_put_owner,
5456 };
5457
5458 static inline void
5459 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5460 {
5461         struct nfs4_lockowner *lo;
5462
5463         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5464                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5465                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5466                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5467                 if (!deny->ld_owner.data)
5468                         /* We just don't care that much */
5469                         goto nevermind;
5470                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5471                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5472         } else {
5473 nevermind:
5474                 deny->ld_owner.len = 0;
5475                 deny->ld_owner.data = NULL;
5476                 deny->ld_clientid.cl_boot = 0;
5477                 deny->ld_clientid.cl_id = 0;
5478         }
5479         deny->ld_start = fl->fl_start;
5480         deny->ld_length = NFS4_MAX_UINT64;
5481         if (fl->fl_end != NFS4_MAX_UINT64)
5482                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5483         deny->ld_type = NFS4_READ_LT;
5484         if (fl->fl_type != F_RDLCK)
5485                 deny->ld_type = NFS4_WRITE_LT;
5486 }
5487
5488 static struct nfs4_lockowner *
5489 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5490 {
5491         unsigned int strhashval = ownerstr_hashval(owner);
5492         struct nfs4_stateowner *so;
5493
5494         lockdep_assert_held(&clp->cl_lock);
5495
5496         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5497                             so_strhash) {
5498                 if (so->so_is_open_owner)
5499                         continue;
5500                 if (same_owner_str(so, owner))
5501                         return lockowner(nfs4_get_stateowner(so));
5502         }
5503         return NULL;
5504 }
5505
5506 static struct nfs4_lockowner *
5507 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5508 {
5509         struct nfs4_lockowner *lo;
5510
5511         spin_lock(&clp->cl_lock);
5512         lo = find_lockowner_str_locked(clp, owner);
5513         spin_unlock(&clp->cl_lock);
5514         return lo;
5515 }
5516
5517 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5518 {
5519         unhash_lockowner_locked(lockowner(sop));
5520 }
5521
5522 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5523 {
5524         struct nfs4_lockowner *lo = lockowner(sop);
5525
5526         kmem_cache_free(lockowner_slab, lo);
5527 }
5528
5529 static const struct nfs4_stateowner_operations lockowner_ops = {
5530         .so_unhash =    nfs4_unhash_lockowner,
5531         .so_free =      nfs4_free_lockowner,
5532 };
5533
5534 /*
5535  * Alloc a lock owner structure.
5536  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5537  * occurred.
5538  *
5539  * strhashval = ownerstr_hashval
5540  */
5541 static struct nfs4_lockowner *
5542 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5543                            struct nfs4_ol_stateid *open_stp,
5544                            struct nfsd4_lock *lock)
5545 {
5546         struct nfs4_lockowner *lo, *ret;
5547
5548         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5549         if (!lo)
5550                 return NULL;
5551         INIT_LIST_HEAD(&lo->lo_blocked);
5552         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5553         lo->lo_owner.so_is_open_owner = 0;
5554         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5555         lo->lo_owner.so_ops = &lockowner_ops;
5556         spin_lock(&clp->cl_lock);
5557         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5558         if (ret == NULL) {
5559                 list_add(&lo->lo_owner.so_strhash,
5560                          &clp->cl_ownerstr_hashtbl[strhashval]);
5561                 ret = lo;
5562         } else
5563                 nfs4_free_stateowner(&lo->lo_owner);
5564
5565         spin_unlock(&clp->cl_lock);
5566         return ret;
5567 }
5568
5569 static void
5570 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5571                   struct nfs4_file *fp, struct inode *inode,
5572                   struct nfs4_ol_stateid *open_stp)
5573 {
5574         struct nfs4_client *clp = lo->lo_owner.so_client;
5575
5576         lockdep_assert_held(&clp->cl_lock);
5577
5578         atomic_inc(&stp->st_stid.sc_count);
5579         stp->st_stid.sc_type = NFS4_LOCK_STID;
5580         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5581         get_nfs4_file(fp);
5582         stp->st_stid.sc_file = fp;
5583         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5584         stp->st_access_bmap = 0;
5585         stp->st_deny_bmap = open_stp->st_deny_bmap;
5586         stp->st_openstp = open_stp;
5587         mutex_init(&stp->st_mutex);
5588         list_add(&stp->st_locks, &open_stp->st_locks);
5589         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5590         spin_lock(&fp->fi_lock);
5591         list_add(&stp->st_perfile, &fp->fi_stateids);
5592         spin_unlock(&fp->fi_lock);
5593 }
5594
5595 static struct nfs4_ol_stateid *
5596 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5597 {
5598         struct nfs4_ol_stateid *lst;
5599         struct nfs4_client *clp = lo->lo_owner.so_client;
5600
5601         lockdep_assert_held(&clp->cl_lock);
5602
5603         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5604                 if (lst->st_stid.sc_file == fp) {
5605                         atomic_inc(&lst->st_stid.sc_count);
5606                         return lst;
5607                 }
5608         }
5609         return NULL;
5610 }
5611
5612 static struct nfs4_ol_stateid *
5613 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5614                             struct inode *inode, struct nfs4_ol_stateid *ost,
5615                             bool *new)
5616 {
5617         struct nfs4_stid *ns = NULL;
5618         struct nfs4_ol_stateid *lst;
5619         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5620         struct nfs4_client *clp = oo->oo_owner.so_client;
5621
5622         spin_lock(&clp->cl_lock);
5623         lst = find_lock_stateid(lo, fi);
5624         if (lst == NULL) {
5625                 spin_unlock(&clp->cl_lock);
5626                 ns = nfs4_alloc_stid(clp, stateid_slab);
5627                 if (ns == NULL)
5628                         return NULL;
5629
5630                 spin_lock(&clp->cl_lock);
5631                 lst = find_lock_stateid(lo, fi);
5632                 if (likely(!lst)) {
5633                         lst = openlockstateid(ns);
5634                         init_lock_stateid(lst, lo, fi, inode, ost);
5635                         ns = NULL;
5636                         *new = true;
5637                 }
5638         }
5639         spin_unlock(&clp->cl_lock);
5640         if (ns)
5641                 nfs4_put_stid(ns);
5642         return lst;
5643 }
5644
5645 static int
5646 check_lock_length(u64 offset, u64 length)
5647 {
5648         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5649                 (length > ~offset)));
5650 }
5651
5652 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5653 {
5654         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5655
5656         lockdep_assert_held(&fp->fi_lock);
5657
5658         if (test_access(access, lock_stp))
5659                 return;
5660         __nfs4_file_get_access(fp, access);
5661         set_access(access, lock_stp);
5662 }
5663
5664 static __be32
5665 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5666                             struct nfs4_ol_stateid *ost,
5667                             struct nfsd4_lock *lock,
5668                             struct nfs4_ol_stateid **plst, bool *new)
5669 {
5670         __be32 status;
5671         struct nfs4_file *fi = ost->st_stid.sc_file;
5672         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5673         struct nfs4_client *cl = oo->oo_owner.so_client;
5674         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5675         struct nfs4_lockowner *lo;
5676         struct nfs4_ol_stateid *lst;
5677         unsigned int strhashval;
5678         bool hashed;
5679
5680         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5681         if (!lo) {
5682                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5683                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5684                 if (lo == NULL)
5685                         return nfserr_jukebox;
5686         } else {
5687                 /* with an existing lockowner, seqids must be the same */
5688                 status = nfserr_bad_seqid;
5689                 if (!cstate->minorversion &&
5690                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5691                         goto out;
5692         }
5693
5694 retry:
5695         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5696         if (lst == NULL) {
5697                 status = nfserr_jukebox;
5698                 goto out;
5699         }
5700
5701         mutex_lock(&lst->st_mutex);
5702
5703         /* See if it's still hashed to avoid race with FREE_STATEID */
5704         spin_lock(&cl->cl_lock);
5705         hashed = !list_empty(&lst->st_perfile);
5706         spin_unlock(&cl->cl_lock);
5707
5708         if (!hashed) {
5709                 mutex_unlock(&lst->st_mutex);
5710                 nfs4_put_stid(&lst->st_stid);
5711                 goto retry;
5712         }
5713         status = nfs_ok;
5714         *plst = lst;
5715 out:
5716         nfs4_put_stateowner(&lo->lo_owner);
5717         return status;
5718 }
5719
5720 /*
5721  *  LOCK operation
5722  */
5723 __be32
5724 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5725            struct nfsd4_lock *lock)
5726 {
5727         struct nfs4_openowner *open_sop = NULL;
5728         struct nfs4_lockowner *lock_sop = NULL;
5729         struct nfs4_ol_stateid *lock_stp = NULL;
5730         struct nfs4_ol_stateid *open_stp = NULL;
5731         struct nfs4_file *fp;
5732         struct file *filp = NULL;
5733         struct nfsd4_blocked_lock *nbl = NULL;
5734         struct file_lock *file_lock = NULL;
5735         struct file_lock *conflock = NULL;
5736         __be32 status = 0;
5737         int lkflg;
5738         int err;
5739         bool new = false;
5740         unsigned char fl_type;
5741         unsigned int fl_flags = FL_POSIX;
5742         struct net *net = SVC_NET(rqstp);
5743         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5744
5745         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5746                 (long long) lock->lk_offset,
5747                 (long long) lock->lk_length);
5748
5749         if (check_lock_length(lock->lk_offset, lock->lk_length))
5750                  return nfserr_inval;
5751
5752         if ((status = fh_verify(rqstp, &cstate->current_fh,
5753                                 S_IFREG, NFSD_MAY_LOCK))) {
5754                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5755                 return status;
5756         }
5757
5758         if (lock->lk_is_new) {
5759                 if (nfsd4_has_session(cstate))
5760                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5761                         memcpy(&lock->lk_new_clientid,
5762                                 &cstate->session->se_client->cl_clientid,
5763                                 sizeof(clientid_t));
5764
5765                 status = nfserr_stale_clientid;
5766                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5767                         goto out;
5768
5769                 /* validate and update open stateid and open seqid */
5770                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5771                                         lock->lk_new_open_seqid,
5772                                         &lock->lk_new_open_stateid,
5773                                         &open_stp, nn);
5774                 if (status)
5775                         goto out;
5776                 mutex_unlock(&open_stp->st_mutex);
5777                 open_sop = openowner(open_stp->st_stateowner);
5778                 status = nfserr_bad_stateid;
5779                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5780                                                 &lock->lk_new_clientid))
5781                         goto out;
5782                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5783                                                         &lock_stp, &new);
5784         } else {
5785                 status = nfs4_preprocess_seqid_op(cstate,
5786                                        lock->lk_old_lock_seqid,
5787                                        &lock->lk_old_lock_stateid,
5788                                        NFS4_LOCK_STID, &lock_stp, nn);
5789         }
5790         if (status)
5791                 goto out;
5792         lock_sop = lockowner(lock_stp->st_stateowner);
5793
5794         lkflg = setlkflg(lock->lk_type);
5795         status = nfs4_check_openmode(lock_stp, lkflg);
5796         if (status)
5797                 goto out;
5798
5799         status = nfserr_grace;
5800         if (locks_in_grace(net) && !lock->lk_reclaim)
5801                 goto out;
5802         status = nfserr_no_grace;
5803         if (!locks_in_grace(net) && lock->lk_reclaim)
5804                 goto out;
5805
5806         fp = lock_stp->st_stid.sc_file;
5807         switch (lock->lk_type) {
5808                 case NFS4_READW_LT:
5809                         if (nfsd4_has_session(cstate))
5810                                 fl_flags |= FL_SLEEP;
5811                         /* Fallthrough */
5812                 case NFS4_READ_LT:
5813                         spin_lock(&fp->fi_lock);
5814                         filp = find_readable_file_locked(fp);
5815                         if (filp)
5816                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5817                         spin_unlock(&fp->fi_lock);
5818                         fl_type = F_RDLCK;
5819                         break;
5820                 case NFS4_WRITEW_LT:
5821                         if (nfsd4_has_session(cstate))
5822                                 fl_flags |= FL_SLEEP;
5823                         /* Fallthrough */
5824                 case NFS4_WRITE_LT:
5825                         spin_lock(&fp->fi_lock);
5826                         filp = find_writeable_file_locked(fp);
5827                         if (filp)
5828                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5829                         spin_unlock(&fp->fi_lock);
5830                         fl_type = F_WRLCK;
5831                         break;
5832                 default:
5833                         status = nfserr_inval;
5834                 goto out;
5835         }
5836
5837         if (!filp) {
5838                 status = nfserr_openmode;
5839                 goto out;
5840         }
5841
5842         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
5843         if (!nbl) {
5844                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
5845                 status = nfserr_jukebox;
5846                 goto out;
5847         }
5848
5849         file_lock = &nbl->nbl_lock;
5850         file_lock->fl_type = fl_type;
5851         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5852         file_lock->fl_pid = current->tgid;
5853         file_lock->fl_file = filp;
5854         file_lock->fl_flags = fl_flags;
5855         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5856         file_lock->fl_start = lock->lk_offset;
5857         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5858         nfs4_transform_lock_offset(file_lock);
5859
5860         conflock = locks_alloc_lock();
5861         if (!conflock) {
5862                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5863                 status = nfserr_jukebox;
5864                 goto out;
5865         }
5866
5867         if (fl_flags & FL_SLEEP) {
5868                 nbl->nbl_time = jiffies;
5869                 spin_lock(&nn->blocked_locks_lock);
5870                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
5871                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
5872                 spin_unlock(&nn->blocked_locks_lock);
5873         }
5874
5875         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5876         switch (err) {
5877         case 0: /* success! */
5878                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5879                 status = 0;
5880                 break;
5881         case FILE_LOCK_DEFERRED:
5882                 nbl = NULL;
5883                 /* Fallthrough */
5884         case -EAGAIN:           /* conflock holds conflicting lock */
5885                 status = nfserr_denied;
5886                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5887                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5888                 break;
5889         case -EDEADLK:
5890                 status = nfserr_deadlock;
5891                 break;
5892         default:
5893                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5894                 status = nfserrno(err);
5895                 break;
5896         }
5897 out:
5898         if (nbl) {
5899                 /* dequeue it if we queued it before */
5900                 if (fl_flags & FL_SLEEP) {
5901                         spin_lock(&nn->blocked_locks_lock);
5902                         list_del_init(&nbl->nbl_list);
5903                         list_del_init(&nbl->nbl_lru);
5904                         spin_unlock(&nn->blocked_locks_lock);
5905                 }
5906                 free_blocked_lock(nbl);
5907         }
5908         if (filp)
5909                 fput(filp);
5910         if (lock_stp) {
5911                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5912                 if (cstate->replay_owner &&
5913                     cstate->replay_owner != &lock_sop->lo_owner &&
5914                     seqid_mutating_err(ntohl(status)))
5915                         lock_sop->lo_owner.so_seqid++;
5916
5917                 mutex_unlock(&lock_stp->st_mutex);
5918
5919                 /*
5920                  * If this is a new, never-before-used stateid, and we are
5921                  * returning an error, then just go ahead and release it.
5922                  */
5923                 if (status && new)
5924                         release_lock_stateid(lock_stp);
5925
5926                 nfs4_put_stid(&lock_stp->st_stid);
5927         }
5928         if (open_stp)
5929                 nfs4_put_stid(&open_stp->st_stid);
5930         nfsd4_bump_seqid(cstate, status);
5931         if (conflock)
5932                 locks_free_lock(conflock);
5933         return status;
5934 }
5935
5936 /*
5937  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5938  * so we do a temporary open here just to get an open file to pass to
5939  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5940  * inode operation.)
5941  */
5942 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5943 {
5944         struct file *file;
5945         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5946         if (!err) {
5947                 err = nfserrno(vfs_test_lock(file, lock));
5948                 fput(file);
5949         }
5950         return err;
5951 }
5952
5953 /*
5954  * LOCKT operation
5955  */
5956 __be32
5957 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5958             struct nfsd4_lockt *lockt)
5959 {
5960         struct file_lock *file_lock = NULL;
5961         struct nfs4_lockowner *lo = NULL;
5962         __be32 status;
5963         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5964
5965         if (locks_in_grace(SVC_NET(rqstp)))
5966                 return nfserr_grace;
5967
5968         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5969                  return nfserr_inval;
5970
5971         if (!nfsd4_has_session(cstate)) {
5972                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5973                 if (status)
5974                         goto out;
5975         }
5976
5977         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5978                 goto out;
5979
5980         file_lock = locks_alloc_lock();
5981         if (!file_lock) {
5982                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5983                 status = nfserr_jukebox;
5984                 goto out;
5985         }
5986
5987         switch (lockt->lt_type) {
5988                 case NFS4_READ_LT:
5989                 case NFS4_READW_LT:
5990                         file_lock->fl_type = F_RDLCK;
5991                 break;
5992                 case NFS4_WRITE_LT:
5993                 case NFS4_WRITEW_LT:
5994                         file_lock->fl_type = F_WRLCK;
5995                 break;
5996                 default:
5997                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5998                         status = nfserr_inval;
5999                 goto out;
6000         }
6001
6002         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6003         if (lo)
6004                 file_lock->fl_owner = (fl_owner_t)lo;
6005         file_lock->fl_pid = current->tgid;
6006         file_lock->fl_flags = FL_POSIX;
6007
6008         file_lock->fl_start = lockt->lt_offset;
6009         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6010
6011         nfs4_transform_lock_offset(file_lock);
6012
6013         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6014         if (status)
6015                 goto out;
6016
6017         if (file_lock->fl_type != F_UNLCK) {
6018                 status = nfserr_denied;
6019                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6020         }
6021 out:
6022         if (lo)
6023                 nfs4_put_stateowner(&lo->lo_owner);
6024         if (file_lock)
6025                 locks_free_lock(file_lock);
6026         return status;
6027 }
6028
6029 __be32
6030 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6031             struct nfsd4_locku *locku)
6032 {
6033         struct nfs4_ol_stateid *stp;
6034         struct file *filp = NULL;
6035         struct file_lock *file_lock = NULL;
6036         __be32 status;
6037         int err;
6038         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6039
6040         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6041                 (long long) locku->lu_offset,
6042                 (long long) locku->lu_length);
6043
6044         if (check_lock_length(locku->lu_offset, locku->lu_length))
6045                  return nfserr_inval;
6046
6047         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6048                                         &locku->lu_stateid, NFS4_LOCK_STID,
6049                                         &stp, nn);
6050         if (status)
6051                 goto out;
6052         filp = find_any_file(stp->st_stid.sc_file);
6053         if (!filp) {
6054                 status = nfserr_lock_range;
6055                 goto put_stateid;
6056         }
6057         file_lock = locks_alloc_lock();
6058         if (!file_lock) {
6059                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6060                 status = nfserr_jukebox;
6061                 goto fput;
6062         }
6063
6064         file_lock->fl_type = F_UNLCK;
6065         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6066         file_lock->fl_pid = current->tgid;
6067         file_lock->fl_file = filp;
6068         file_lock->fl_flags = FL_POSIX;
6069         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6070         file_lock->fl_start = locku->lu_offset;
6071
6072         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6073                                                 locku->lu_length);
6074         nfs4_transform_lock_offset(file_lock);
6075
6076         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6077         if (err) {
6078                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6079                 goto out_nfserr;
6080         }
6081         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6082 fput:
6083         fput(filp);
6084 put_stateid:
6085         mutex_unlock(&stp->st_mutex);
6086         nfs4_put_stid(&stp->st_stid);
6087 out:
6088         nfsd4_bump_seqid(cstate, status);
6089         if (file_lock)
6090                 locks_free_lock(file_lock);
6091         return status;
6092
6093 out_nfserr:
6094         status = nfserrno(err);
6095         goto fput;
6096 }
6097
6098 /*
6099  * returns
6100  *      true:  locks held by lockowner
6101  *      false: no locks held by lockowner
6102  */
6103 static bool
6104 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6105 {
6106         struct file_lock *fl;
6107         int status = false;
6108         struct file *filp = find_any_file(fp);
6109         struct inode *inode;
6110         struct file_lock_context *flctx;
6111
6112         if (!filp) {
6113                 /* Any valid lock stateid should have some sort of access */
6114                 WARN_ON_ONCE(1);
6115                 return status;
6116         }
6117
6118         inode = file_inode(filp);
6119         flctx = inode->i_flctx;
6120
6121         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6122                 spin_lock(&flctx->flc_lock);
6123                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6124                         if (fl->fl_owner == (fl_owner_t)lowner) {
6125                                 status = true;
6126                                 break;
6127                         }
6128                 }
6129                 spin_unlock(&flctx->flc_lock);
6130         }
6131         fput(filp);
6132         return status;
6133 }
6134
6135 __be32
6136 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6137                         struct nfsd4_compound_state *cstate,
6138                         struct nfsd4_release_lockowner *rlockowner)
6139 {
6140         clientid_t *clid = &rlockowner->rl_clientid;
6141         struct nfs4_stateowner *sop;
6142         struct nfs4_lockowner *lo = NULL;
6143         struct nfs4_ol_stateid *stp;
6144         struct xdr_netobj *owner = &rlockowner->rl_owner;
6145         unsigned int hashval = ownerstr_hashval(owner);
6146         __be32 status;
6147         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6148         struct nfs4_client *clp;
6149         LIST_HEAD (reaplist);
6150
6151         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6152                 clid->cl_boot, clid->cl_id);
6153
6154         status = lookup_clientid(clid, cstate, nn);
6155         if (status)
6156                 return status;
6157
6158         clp = cstate->clp;
6159         /* Find the matching lock stateowner */
6160         spin_lock(&clp->cl_lock);
6161         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6162                             so_strhash) {
6163
6164                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6165                         continue;
6166
6167                 /* see if there are still any locks associated with it */
6168                 lo = lockowner(sop);
6169                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6170                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6171                                 status = nfserr_locks_held;
6172                                 spin_unlock(&clp->cl_lock);
6173                                 return status;
6174                         }
6175                 }
6176
6177                 nfs4_get_stateowner(sop);
6178                 break;
6179         }
6180         if (!lo) {
6181                 spin_unlock(&clp->cl_lock);
6182                 return status;
6183         }
6184
6185         unhash_lockowner_locked(lo);
6186         while (!list_empty(&lo->lo_owner.so_stateids)) {
6187                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6188                                        struct nfs4_ol_stateid,
6189                                        st_perstateowner);
6190                 WARN_ON(!unhash_lock_stateid(stp));
6191                 put_ol_stateid_locked(stp, &reaplist);
6192         }
6193         spin_unlock(&clp->cl_lock);
6194         free_ol_stateid_reaplist(&reaplist);
6195         nfs4_put_stateowner(&lo->lo_owner);
6196
6197         return status;
6198 }
6199
6200 static inline struct nfs4_client_reclaim *
6201 alloc_reclaim(void)
6202 {
6203         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6204 }
6205
6206 bool
6207 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6208 {
6209         struct nfs4_client_reclaim *crp;
6210
6211         crp = nfsd4_find_reclaim_client(name, nn);
6212         return (crp && crp->cr_clp);
6213 }
6214
6215 /*
6216  * failure => all reset bets are off, nfserr_no_grace...
6217  */
6218 struct nfs4_client_reclaim *
6219 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6220 {
6221         unsigned int strhashval;
6222         struct nfs4_client_reclaim *crp;
6223
6224         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6225         crp = alloc_reclaim();
6226         if (crp) {
6227                 strhashval = clientstr_hashval(name);
6228                 INIT_LIST_HEAD(&crp->cr_strhash);
6229                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6230                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6231                 crp->cr_clp = NULL;
6232                 nn->reclaim_str_hashtbl_size++;
6233         }
6234         return crp;
6235 }
6236
6237 void
6238 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6239 {
6240         list_del(&crp->cr_strhash);
6241         kfree(crp);
6242         nn->reclaim_str_hashtbl_size--;
6243 }
6244
6245 void
6246 nfs4_release_reclaim(struct nfsd_net *nn)
6247 {
6248         struct nfs4_client_reclaim *crp = NULL;
6249         int i;
6250
6251         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6252                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6253                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6254                                         struct nfs4_client_reclaim, cr_strhash);
6255                         nfs4_remove_reclaim_record(crp, nn);
6256                 }
6257         }
6258         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6259 }
6260
6261 /*
6262  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6263 struct nfs4_client_reclaim *
6264 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6265 {
6266         unsigned int strhashval;
6267         struct nfs4_client_reclaim *crp = NULL;
6268
6269         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6270
6271         strhashval = clientstr_hashval(recdir);
6272         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6273                 if (same_name(crp->cr_recdir, recdir)) {
6274                         return crp;
6275                 }
6276         }
6277         return NULL;
6278 }
6279
6280 /*
6281 * Called from OPEN. Look for clientid in reclaim list.
6282 */
6283 __be32
6284 nfs4_check_open_reclaim(clientid_t *clid,
6285                 struct nfsd4_compound_state *cstate,
6286                 struct nfsd_net *nn)
6287 {
6288         __be32 status;
6289
6290         /* find clientid in conf_id_hashtbl */
6291         status = lookup_clientid(clid, cstate, nn);
6292         if (status)
6293                 return nfserr_reclaim_bad;
6294
6295         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6296                 return nfserr_no_grace;
6297
6298         if (nfsd4_client_record_check(cstate->clp))
6299                 return nfserr_reclaim_bad;
6300
6301         return nfs_ok;
6302 }
6303
6304 #ifdef CONFIG_NFSD_FAULT_INJECTION
6305 static inline void
6306 put_client(struct nfs4_client *clp)
6307 {
6308         atomic_dec(&clp->cl_refcount);
6309 }
6310
6311 static struct nfs4_client *
6312 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6313 {
6314         struct nfs4_client *clp;
6315         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6316                                           nfsd_net_id);
6317
6318         if (!nfsd_netns_ready(nn))
6319                 return NULL;
6320
6321         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6322                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6323                         return clp;
6324         }
6325         return NULL;
6326 }
6327
6328 u64
6329 nfsd_inject_print_clients(void)
6330 {
6331         struct nfs4_client *clp;
6332         u64 count = 0;
6333         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6334                                           nfsd_net_id);
6335         char buf[INET6_ADDRSTRLEN];
6336
6337         if (!nfsd_netns_ready(nn))
6338                 return 0;
6339
6340         spin_lock(&nn->client_lock);
6341         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6342                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6343                 pr_info("NFS Client: %s\n", buf);
6344                 ++count;
6345         }
6346         spin_unlock(&nn->client_lock);
6347
6348         return count;
6349 }
6350
6351 u64
6352 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6353 {
6354         u64 count = 0;
6355         struct nfs4_client *clp;
6356         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6357                                           nfsd_net_id);
6358
6359         if (!nfsd_netns_ready(nn))
6360                 return count;
6361
6362         spin_lock(&nn->client_lock);
6363         clp = nfsd_find_client(addr, addr_size);
6364         if (clp) {
6365                 if (mark_client_expired_locked(clp) == nfs_ok)
6366                         ++count;
6367                 else
6368                         clp = NULL;
6369         }
6370         spin_unlock(&nn->client_lock);
6371
6372         if (clp)
6373                 expire_client(clp);
6374
6375         return count;
6376 }
6377
6378 u64
6379 nfsd_inject_forget_clients(u64 max)
6380 {
6381         u64 count = 0;
6382         struct nfs4_client *clp, *next;
6383         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6384                                                 nfsd_net_id);
6385         LIST_HEAD(reaplist);
6386
6387         if (!nfsd_netns_ready(nn))
6388                 return count;
6389
6390         spin_lock(&nn->client_lock);
6391         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6392                 if (mark_client_expired_locked(clp) == nfs_ok) {
6393                         list_add(&clp->cl_lru, &reaplist);
6394                         if (max != 0 && ++count >= max)
6395                                 break;
6396                 }
6397         }
6398         spin_unlock(&nn->client_lock);
6399
6400         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6401                 expire_client(clp);
6402
6403         return count;
6404 }
6405
6406 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6407                              const char *type)
6408 {
6409         char buf[INET6_ADDRSTRLEN];
6410         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6411         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6412 }
6413
6414 static void
6415 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6416                              struct list_head *collect)
6417 {
6418         struct nfs4_client *clp = lst->st_stid.sc_client;
6419         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6420                                           nfsd_net_id);
6421
6422         if (!collect)
6423                 return;
6424
6425         lockdep_assert_held(&nn->client_lock);
6426         atomic_inc(&clp->cl_refcount);
6427         list_add(&lst->st_locks, collect);
6428 }
6429
6430 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6431                                     struct list_head *collect,
6432                                     bool (*func)(struct nfs4_ol_stateid *))
6433 {
6434         struct nfs4_openowner *oop;
6435         struct nfs4_ol_stateid *stp, *st_next;
6436         struct nfs4_ol_stateid *lst, *lst_next;
6437         u64 count = 0;
6438
6439         spin_lock(&clp->cl_lock);
6440         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6441                 list_for_each_entry_safe(stp, st_next,
6442                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6443                         list_for_each_entry_safe(lst, lst_next,
6444                                         &stp->st_locks, st_locks) {
6445                                 if (func) {
6446                                         if (func(lst))
6447                                                 nfsd_inject_add_lock_to_list(lst,
6448                                                                         collect);
6449                                 }
6450                                 ++count;
6451                                 /*
6452                                  * Despite the fact that these functions deal
6453                                  * with 64-bit integers for "count", we must
6454                                  * ensure that it doesn't blow up the
6455                                  * clp->cl_refcount. Throw a warning if we
6456                                  * start to approach INT_MAX here.
6457                                  */
6458                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6459                                 if (count == max)
6460                                         goto out;
6461                         }
6462                 }
6463         }
6464 out:
6465         spin_unlock(&clp->cl_lock);
6466
6467         return count;
6468 }
6469
6470 static u64
6471 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6472                           u64 max)
6473 {
6474         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6475 }
6476
6477 static u64
6478 nfsd_print_client_locks(struct nfs4_client *clp)
6479 {
6480         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6481         nfsd_print_count(clp, count, "locked files");
6482         return count;
6483 }
6484
6485 u64
6486 nfsd_inject_print_locks(void)
6487 {
6488         struct nfs4_client *clp;
6489         u64 count = 0;
6490         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6491                                                 nfsd_net_id);
6492
6493         if (!nfsd_netns_ready(nn))
6494                 return 0;
6495
6496         spin_lock(&nn->client_lock);
6497         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6498                 count += nfsd_print_client_locks(clp);
6499         spin_unlock(&nn->client_lock);
6500
6501         return count;
6502 }
6503
6504 static void
6505 nfsd_reap_locks(struct list_head *reaplist)
6506 {
6507         struct nfs4_client *clp;
6508         struct nfs4_ol_stateid *stp, *next;
6509
6510         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6511                 list_del_init(&stp->st_locks);
6512                 clp = stp->st_stid.sc_client;
6513                 nfs4_put_stid(&stp->st_stid);
6514                 put_client(clp);
6515         }
6516 }
6517
6518 u64
6519 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6520 {
6521         unsigned int count = 0;
6522         struct nfs4_client *clp;
6523         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6524                                                 nfsd_net_id);
6525         LIST_HEAD(reaplist);
6526
6527         if (!nfsd_netns_ready(nn))
6528                 return count;
6529
6530         spin_lock(&nn->client_lock);
6531         clp = nfsd_find_client(addr, addr_size);
6532         if (clp)
6533                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6534         spin_unlock(&nn->client_lock);
6535         nfsd_reap_locks(&reaplist);
6536         return count;
6537 }
6538
6539 u64
6540 nfsd_inject_forget_locks(u64 max)
6541 {
6542         u64 count = 0;
6543         struct nfs4_client *clp;
6544         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6545                                                 nfsd_net_id);
6546         LIST_HEAD(reaplist);
6547
6548         if (!nfsd_netns_ready(nn))
6549                 return count;
6550
6551         spin_lock(&nn->client_lock);
6552         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6553                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6554                 if (max != 0 && count >= max)
6555                         break;
6556         }
6557         spin_unlock(&nn->client_lock);
6558         nfsd_reap_locks(&reaplist);
6559         return count;
6560 }
6561
6562 static u64
6563 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6564                               struct list_head *collect,
6565                               void (*func)(struct nfs4_openowner *))
6566 {
6567         struct nfs4_openowner *oop, *next;
6568         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6569                                                 nfsd_net_id);
6570         u64 count = 0;
6571
6572         lockdep_assert_held(&nn->client_lock);
6573
6574         spin_lock(&clp->cl_lock);
6575         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6576                 if (func) {
6577                         func(oop);
6578                         if (collect) {
6579                                 atomic_inc(&clp->cl_refcount);
6580                                 list_add(&oop->oo_perclient, collect);
6581                         }
6582                 }
6583                 ++count;
6584                 /*
6585                  * Despite the fact that these functions deal with
6586                  * 64-bit integers for "count", we must ensure that
6587                  * it doesn't blow up the clp->cl_refcount. Throw a
6588                  * warning if we start to approach INT_MAX here.
6589                  */
6590                 WARN_ON_ONCE(count == (INT_MAX / 2));
6591                 if (count == max)
6592                         break;
6593         }
6594         spin_unlock(&clp->cl_lock);
6595
6596         return count;
6597 }
6598
6599 static u64
6600 nfsd_print_client_openowners(struct nfs4_client *clp)
6601 {
6602         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6603
6604         nfsd_print_count(clp, count, "openowners");
6605         return count;
6606 }
6607
6608 static u64
6609 nfsd_collect_client_openowners(struct nfs4_client *clp,
6610                                struct list_head *collect, u64 max)
6611 {
6612         return nfsd_foreach_client_openowner(clp, max, collect,
6613                                                 unhash_openowner_locked);
6614 }
6615
6616 u64
6617 nfsd_inject_print_openowners(void)
6618 {
6619         struct nfs4_client *clp;
6620         u64 count = 0;
6621         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6622                                                 nfsd_net_id);
6623
6624         if (!nfsd_netns_ready(nn))
6625                 return 0;
6626
6627         spin_lock(&nn->client_lock);
6628         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6629                 count += nfsd_print_client_openowners(clp);
6630         spin_unlock(&nn->client_lock);
6631
6632         return count;
6633 }
6634
6635 static void
6636 nfsd_reap_openowners(struct list_head *reaplist)
6637 {
6638         struct nfs4_client *clp;
6639         struct nfs4_openowner *oop, *next;
6640
6641         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6642                 list_del_init(&oop->oo_perclient);
6643                 clp = oop->oo_owner.so_client;
6644                 release_openowner(oop);
6645                 put_client(clp);
6646         }
6647 }
6648
6649 u64
6650 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6651                                      size_t addr_size)
6652 {
6653         unsigned int count = 0;
6654         struct nfs4_client *clp;
6655         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6656                                                 nfsd_net_id);
6657         LIST_HEAD(reaplist);
6658
6659         if (!nfsd_netns_ready(nn))
6660                 return count;
6661
6662         spin_lock(&nn->client_lock);
6663         clp = nfsd_find_client(addr, addr_size);
6664         if (clp)
6665                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6666         spin_unlock(&nn->client_lock);
6667         nfsd_reap_openowners(&reaplist);
6668         return count;
6669 }
6670
6671 u64
6672 nfsd_inject_forget_openowners(u64 max)
6673 {
6674         u64 count = 0;
6675         struct nfs4_client *clp;
6676         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6677                                                 nfsd_net_id);
6678         LIST_HEAD(reaplist);
6679
6680         if (!nfsd_netns_ready(nn))
6681                 return count;
6682
6683         spin_lock(&nn->client_lock);
6684         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6685                 count += nfsd_collect_client_openowners(clp, &reaplist,
6686                                                         max - count);
6687                 if (max != 0 && count >= max)
6688                         break;
6689         }
6690         spin_unlock(&nn->client_lock);
6691         nfsd_reap_openowners(&reaplist);
6692         return count;
6693 }
6694
6695 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6696                                      struct list_head *victims)
6697 {
6698         struct nfs4_delegation *dp, *next;
6699         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6700                                                 nfsd_net_id);
6701         u64 count = 0;
6702
6703         lockdep_assert_held(&nn->client_lock);
6704
6705         spin_lock(&state_lock);
6706         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6707                 if (victims) {
6708                         /*
6709                          * It's not safe to mess with delegations that have a
6710                          * non-zero dl_time. They might have already been broken
6711                          * and could be processed by the laundromat outside of
6712                          * the state_lock. Just leave them be.
6713                          */
6714                         if (dp->dl_time != 0)
6715                                 continue;
6716
6717                         atomic_inc(&clp->cl_refcount);
6718                         WARN_ON(!unhash_delegation_locked(dp));
6719                         list_add(&dp->dl_recall_lru, victims);
6720                 }
6721                 ++count;
6722                 /*
6723                  * Despite the fact that these functions deal with
6724                  * 64-bit integers for "count", we must ensure that
6725                  * it doesn't blow up the clp->cl_refcount. Throw a
6726                  * warning if we start to approach INT_MAX here.
6727                  */
6728                 WARN_ON_ONCE(count == (INT_MAX / 2));
6729                 if (count == max)
6730                         break;
6731         }
6732         spin_unlock(&state_lock);
6733         return count;
6734 }
6735
6736 static u64
6737 nfsd_print_client_delegations(struct nfs4_client *clp)
6738 {
6739         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6740
6741         nfsd_print_count(clp, count, "delegations");
6742         return count;
6743 }
6744
6745 u64
6746 nfsd_inject_print_delegations(void)
6747 {
6748         struct nfs4_client *clp;
6749         u64 count = 0;
6750         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6751                                                 nfsd_net_id);
6752
6753         if (!nfsd_netns_ready(nn))
6754                 return 0;
6755
6756         spin_lock(&nn->client_lock);
6757         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6758                 count += nfsd_print_client_delegations(clp);
6759         spin_unlock(&nn->client_lock);
6760
6761         return count;
6762 }
6763
6764 static void
6765 nfsd_forget_delegations(struct list_head *reaplist)
6766 {
6767         struct nfs4_client *clp;
6768         struct nfs4_delegation *dp, *next;
6769
6770         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6771                 list_del_init(&dp->dl_recall_lru);
6772                 clp = dp->dl_stid.sc_client;
6773                 revoke_delegation(dp);
6774                 put_client(clp);
6775         }
6776 }
6777
6778 u64
6779 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6780                                       size_t addr_size)
6781 {
6782         u64 count = 0;
6783         struct nfs4_client *clp;
6784         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6785                                                 nfsd_net_id);
6786         LIST_HEAD(reaplist);
6787
6788         if (!nfsd_netns_ready(nn))
6789                 return count;
6790
6791         spin_lock(&nn->client_lock);
6792         clp = nfsd_find_client(addr, addr_size);
6793         if (clp)
6794                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6795         spin_unlock(&nn->client_lock);
6796
6797         nfsd_forget_delegations(&reaplist);
6798         return count;
6799 }
6800
6801 u64
6802 nfsd_inject_forget_delegations(u64 max)
6803 {
6804         u64 count = 0;
6805         struct nfs4_client *clp;
6806         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6807                                                 nfsd_net_id);
6808         LIST_HEAD(reaplist);
6809
6810         if (!nfsd_netns_ready(nn))
6811                 return count;
6812
6813         spin_lock(&nn->client_lock);
6814         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6815                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6816                 if (max != 0 && count >= max)
6817                         break;
6818         }
6819         spin_unlock(&nn->client_lock);
6820         nfsd_forget_delegations(&reaplist);
6821         return count;
6822 }
6823
6824 static void
6825 nfsd_recall_delegations(struct list_head *reaplist)
6826 {
6827         struct nfs4_client *clp;
6828         struct nfs4_delegation *dp, *next;
6829
6830         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6831                 list_del_init(&dp->dl_recall_lru);
6832                 clp = dp->dl_stid.sc_client;
6833                 /*
6834                  * We skipped all entries that had a zero dl_time before,
6835                  * so we can now reset the dl_time back to 0. If a delegation
6836                  * break comes in now, then it won't make any difference since
6837                  * we're recalling it either way.
6838                  */
6839                 spin_lock(&state_lock);
6840                 dp->dl_time = 0;
6841                 spin_unlock(&state_lock);
6842                 nfsd_break_one_deleg(dp);
6843                 put_client(clp);
6844         }
6845 }
6846
6847 u64
6848 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6849                                       size_t addr_size)
6850 {
6851         u64 count = 0;
6852         struct nfs4_client *clp;
6853         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6854                                                 nfsd_net_id);
6855         LIST_HEAD(reaplist);
6856
6857         if (!nfsd_netns_ready(nn))
6858                 return count;
6859
6860         spin_lock(&nn->client_lock);
6861         clp = nfsd_find_client(addr, addr_size);
6862         if (clp)
6863                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6864         spin_unlock(&nn->client_lock);
6865
6866         nfsd_recall_delegations(&reaplist);
6867         return count;
6868 }
6869
6870 u64
6871 nfsd_inject_recall_delegations(u64 max)
6872 {
6873         u64 count = 0;
6874         struct nfs4_client *clp, *next;
6875         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6876                                                 nfsd_net_id);
6877         LIST_HEAD(reaplist);
6878
6879         if (!nfsd_netns_ready(nn))
6880                 return count;
6881
6882         spin_lock(&nn->client_lock);
6883         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6884                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6885                 if (max != 0 && ++count >= max)
6886                         break;
6887         }
6888         spin_unlock(&nn->client_lock);
6889         nfsd_recall_delegations(&reaplist);
6890         return count;
6891 }
6892 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6893
6894 /*
6895  * Since the lifetime of a delegation isn't limited to that of an open, a
6896  * client may quite reasonably hang on to a delegation as long as it has
6897  * the inode cached.  This becomes an obvious problem the first time a
6898  * client's inode cache approaches the size of the server's total memory.
6899  *
6900  * For now we avoid this problem by imposing a hard limit on the number
6901  * of delegations, which varies according to the server's memory size.
6902  */
6903 static void
6904 set_max_delegations(void)
6905 {
6906         /*
6907          * Allow at most 4 delegations per megabyte of RAM.  Quick
6908          * estimates suggest that in the worst case (where every delegation
6909          * is for a different inode), a delegation could take about 1.5K,
6910          * giving a worst case usage of about 6% of memory.
6911          */
6912         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6913 }
6914
6915 static int nfs4_state_create_net(struct net *net)
6916 {
6917         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6918         int i;
6919
6920         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6921                         CLIENT_HASH_SIZE, GFP_KERNEL);
6922         if (!nn->conf_id_hashtbl)
6923                 goto err;
6924         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6925                         CLIENT_HASH_SIZE, GFP_KERNEL);
6926         if (!nn->unconf_id_hashtbl)
6927                 goto err_unconf_id;
6928         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6929                         SESSION_HASH_SIZE, GFP_KERNEL);
6930         if (!nn->sessionid_hashtbl)
6931                 goto err_sessionid;
6932
6933         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6934                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6935                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6936         }
6937         for (i = 0; i < SESSION_HASH_SIZE; i++)
6938                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6939         nn->conf_name_tree = RB_ROOT;
6940         nn->unconf_name_tree = RB_ROOT;
6941         INIT_LIST_HEAD(&nn->client_lru);
6942         INIT_LIST_HEAD(&nn->close_lru);
6943         INIT_LIST_HEAD(&nn->del_recall_lru);
6944         spin_lock_init(&nn->client_lock);
6945
6946         spin_lock_init(&nn->blocked_locks_lock);
6947         INIT_LIST_HEAD(&nn->blocked_locks_lru);
6948
6949         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6950         get_net(net);
6951
6952         return 0;
6953
6954 err_sessionid:
6955         kfree(nn->unconf_id_hashtbl);
6956 err_unconf_id:
6957         kfree(nn->conf_id_hashtbl);
6958 err:
6959         return -ENOMEM;
6960 }
6961
6962 static void
6963 nfs4_state_destroy_net(struct net *net)
6964 {
6965         int i;
6966         struct nfs4_client *clp = NULL;
6967         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6968
6969         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6970                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6971                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6972                         destroy_client(clp);
6973                 }
6974         }
6975
6976         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6977                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6978                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6979                         destroy_client(clp);
6980                 }
6981         }
6982
6983         kfree(nn->sessionid_hashtbl);
6984         kfree(nn->unconf_id_hashtbl);
6985         kfree(nn->conf_id_hashtbl);
6986         put_net(net);
6987 }
6988
6989 int
6990 nfs4_state_start_net(struct net *net)
6991 {
6992         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6993         int ret;
6994
6995         ret = nfs4_state_create_net(net);
6996         if (ret)
6997                 return ret;
6998         nn->boot_time = get_seconds();
6999         nn->grace_ended = false;
7000         nn->nfsd4_manager.block_opens = true;
7001         locks_start_grace(net, &nn->nfsd4_manager);
7002         nfsd4_client_tracking_init(net);
7003         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
7004                nn->nfsd4_grace, net);
7005         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7006         return 0;
7007 }
7008
7009 /* initialization to perform when the nfsd service is started: */
7010
7011 int
7012 nfs4_state_start(void)
7013 {
7014         int ret;
7015
7016         ret = set_callback_cred();
7017         if (ret)
7018                 return -ENOMEM;
7019         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7020         if (laundry_wq == NULL) {
7021                 ret = -ENOMEM;
7022                 goto out_recovery;
7023         }
7024         ret = nfsd4_create_callback_queue();
7025         if (ret)
7026                 goto out_free_laundry;
7027
7028         set_max_delegations();
7029
7030         return 0;
7031
7032 out_free_laundry:
7033         destroy_workqueue(laundry_wq);
7034 out_recovery:
7035         return ret;
7036 }
7037
7038 void
7039 nfs4_state_shutdown_net(struct net *net)
7040 {
7041         struct nfs4_delegation *dp = NULL;
7042         struct list_head *pos, *next, reaplist;
7043         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7044         struct nfsd4_blocked_lock *nbl;
7045
7046         cancel_delayed_work_sync(&nn->laundromat_work);
7047         locks_end_grace(&nn->nfsd4_manager);
7048
7049         INIT_LIST_HEAD(&reaplist);
7050         spin_lock(&state_lock);
7051         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7052                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7053                 WARN_ON(!unhash_delegation_locked(dp));
7054                 list_add(&dp->dl_recall_lru, &reaplist);
7055         }
7056         spin_unlock(&state_lock);
7057         list_for_each_safe(pos, next, &reaplist) {
7058                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7059                 list_del_init(&dp->dl_recall_lru);
7060                 put_clnt_odstate(dp->dl_clnt_odstate);
7061                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7062                 nfs4_put_stid(&dp->dl_stid);
7063         }
7064
7065         BUG_ON(!list_empty(&reaplist));
7066         spin_lock(&nn->blocked_locks_lock);
7067         while (!list_empty(&nn->blocked_locks_lru)) {
7068                 nbl = list_first_entry(&nn->blocked_locks_lru,
7069                                         struct nfsd4_blocked_lock, nbl_lru);
7070                 list_move(&nbl->nbl_lru, &reaplist);
7071                 list_del_init(&nbl->nbl_list);
7072         }
7073         spin_unlock(&nn->blocked_locks_lock);
7074
7075         while (!list_empty(&reaplist)) {
7076                 nbl = list_first_entry(&nn->blocked_locks_lru,
7077                                         struct nfsd4_blocked_lock, nbl_lru);
7078                 list_del_init(&nbl->nbl_lru);
7079                 posix_unblock_lock(&nbl->nbl_lock);
7080                 free_blocked_lock(nbl);
7081         }
7082
7083         nfsd4_client_tracking_exit(net);
7084         nfs4_state_destroy_net(net);
7085 }
7086
7087 void
7088 nfs4_state_shutdown(void)
7089 {
7090         destroy_workqueue(laundry_wq);
7091         nfsd4_destroy_callback_queue();
7092 }
7093
7094 static void
7095 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7096 {
7097         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7098                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7099 }
7100
7101 static void
7102 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7103 {
7104         if (cstate->minorversion) {
7105                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7106                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7107         }
7108 }
7109
7110 void
7111 clear_current_stateid(struct nfsd4_compound_state *cstate)
7112 {
7113         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7114 }
7115
7116 /*
7117  * functions to set current state id
7118  */
7119 void
7120 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7121 {
7122         put_stateid(cstate, &odp->od_stateid);
7123 }
7124
7125 void
7126 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
7127 {
7128         put_stateid(cstate, &open->op_stateid);
7129 }
7130
7131 void
7132 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7133 {
7134         put_stateid(cstate, &close->cl_stateid);
7135 }
7136
7137 void
7138 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
7139 {
7140         put_stateid(cstate, &lock->lk_resp_stateid);
7141 }
7142
7143 /*
7144  * functions to consume current state id
7145  */
7146
7147 void
7148 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7149 {
7150         get_stateid(cstate, &odp->od_stateid);
7151 }
7152
7153 void
7154 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
7155 {
7156         get_stateid(cstate, &drp->dr_stateid);
7157 }
7158
7159 void
7160 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
7161 {
7162         get_stateid(cstate, &fsp->fr_stateid);
7163 }
7164
7165 void
7166 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
7167 {
7168         get_stateid(cstate, &setattr->sa_stateid);
7169 }
7170
7171 void
7172 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7173 {
7174         get_stateid(cstate, &close->cl_stateid);
7175 }
7176
7177 void
7178 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
7179 {
7180         get_stateid(cstate, &locku->lu_stateid);
7181 }
7182
7183 void
7184 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
7185 {
7186         get_stateid(cstate, &read->rd_stateid);
7187 }
7188
7189 void
7190 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
7191 {
7192         get_stateid(cstate, &write->wr_stateid);
7193 }