fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include <linux/jhash.h>
  45 #include "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51 #include "pnfs.h"
  52
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 #define all_ones {{~0,~0},~0}
  56 static const stateid_t one_stateid = {
  57         .si_generation = ~0,
  58         .si_opaque = all_ones,
  59 };
  60 static const stateid_t zero_stateid = {
  61         /* all fields zero */
  62 };
  63 static const stateid_t currentstateid = {
  64         .si_generation = 1,
  65 };
  66
  67 static u64 current_sessionid = 1;
  68
  69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  70 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  72
  73 /* forward declarations */
  74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  76
  77 /* Locking: */
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(state_lock);
  85
  86 /*
  87  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  88  * the refcount on the open stateid to drop.
  89  */
  90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  91
  92 static struct kmem_cache *openowner_slab;
  93 static struct kmem_cache *lockowner_slab;
  94 static struct kmem_cache *file_slab;
  95 static struct kmem_cache *stateid_slab;
  96 static struct kmem_cache *deleg_slab;
  97 static struct kmem_cache *odstate_slab;
  98
  99 static void free_session(struct nfsd4_session *);
 100
 101 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 102 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 103
 104 static bool is_session_dead(struct nfsd4_session *ses)
 105 {
 106         return ses->se_flags & NFS4_SESSION_DEAD;
 107 }
 108
 109 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 110 {
 111         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 112                 return nfserr_jukebox;
 113         ses->se_flags |= NFS4_SESSION_DEAD;
 114         return nfs_ok;
 115 }
 116
 117 static bool is_client_expired(struct nfs4_client *clp)
 118 {
 119         return clp->cl_time == 0;
 120 }
 121
 122 static __be32 get_client_locked(struct nfs4_client *clp)
 123 {
 124         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 125
 126         lockdep_assert_held(&nn->client_lock);
 127
 128         if (is_client_expired(clp))
 129                 return nfserr_expired;
 130         atomic_inc(&clp->cl_refcount);
 131         return nfs_ok;
 132 }
 133
 134 /* must be called under the client_lock */
 135 static inline void
 136 renew_client_locked(struct nfs4_client *clp)
 137 {
 138         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 139
 140         if (is_client_expired(clp)) {
 141                 WARN_ON(1);
 142                 printk("%s: client (clientid %08x/%08x) already expired\n",
 143                         __func__,
 144                         clp->cl_clientid.cl_boot,
 145                         clp->cl_clientid.cl_id);
 146                 return;
 147         }
 148
 149         dprintk("renewing client (clientid %08x/%08x)\n",
 150                         clp->cl_clientid.cl_boot,
 151                         clp->cl_clientid.cl_id);
 152         list_move_tail(&clp->cl_lru, &nn->client_lru);
 153         clp->cl_time = get_seconds();
 154 }
 155
 156 static void put_client_renew_locked(struct nfs4_client *clp)
 157 {
 158         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 159
 160         lockdep_assert_held(&nn->client_lock);
 161
 162         if (!atomic_dec_and_test(&clp->cl_refcount))
 163                 return;
 164         if (!is_client_expired(clp))
 165                 renew_client_locked(clp);
 166 }
 167
 168 static void put_client_renew(struct nfs4_client *clp)
 169 {
 170         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 171
 172         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 173                 return;
 174         if (!is_client_expired(clp))
 175                 renew_client_locked(clp);
 176         spin_unlock(&nn->client_lock);
 177 }
 178
 179 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 180 {
 181         __be32 status;
 182
 183         if (is_session_dead(ses))
 184                 return nfserr_badsession;
 185         status = get_client_locked(ses->se_client);
 186         if (status)
 187                 return status;
 188         atomic_inc(&ses->se_ref);
 189         return nfs_ok;
 190 }
 191
 192 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 193 {
 194         struct nfs4_client *clp = ses->se_client;
 195         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 196
 197         lockdep_assert_held(&nn->client_lock);
 198
 199         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 200                 free_session(ses);
 201         put_client_renew_locked(clp);
 202 }
 203
 204 static void nfsd4_put_session(struct nfsd4_session *ses)
 205 {
 206         struct nfs4_client *clp = ses->se_client;
 207         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 208
 209         spin_lock(&nn->client_lock);
 210         nfsd4_put_session_locked(ses);
 211         spin_unlock(&nn->client_lock);
 212 }
 213
 214 static struct nfsd4_blocked_lock *
 215 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 216                         struct nfsd_net *nn)
 217 {
 218         struct nfsd4_blocked_lock *cur, *found = NULL;
 219
 220         spin_lock(&nn->blocked_locks_lock);
 221         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 222                 if (fh_match(fh, &cur->nbl_fh)) {
 223                         list_del_init(&cur->nbl_list);
 224                         list_del_init(&cur->nbl_lru);
 225                         found = cur;
 226                         break;
 227                 }
 228         }
 229         spin_unlock(&nn->blocked_locks_lock);
 230         if (found)
 231                 posix_unblock_lock(&found->nbl_lock);
 232         return found;
 233 }
 234
 235 static struct nfsd4_blocked_lock *
 236 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 237                         struct nfsd_net *nn)
 238 {
 239         struct nfsd4_blocked_lock *nbl;
 240
 241         nbl = find_blocked_lock(lo, fh, nn);
 242         if (!nbl) {
 243                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 244                 if (nbl) {
 245                         fh_copy_shallow(&nbl->nbl_fh, fh);
 246                         locks_init_lock(&nbl->nbl_lock);
 247                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 248                                         &nfsd4_cb_notify_lock_ops,
 249                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 250                 }
 251         }
 252         return nbl;
 253 }
 254
 255 static void
 256 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 257 {
 258         locks_release_private(&nbl->nbl_lock);
 259         kfree(nbl);
 260 }
 261
 262 static int
 263 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 264 {
 265         /*
 266          * Since this is just an optimization, we don't try very hard if it
 267          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 268          * just quit trying on anything else.
 269          */
 270         switch (task->tk_status) {
 271         case -NFS4ERR_DELAY:
 272                 rpc_delay(task, 1 * HZ);
 273                 return 0;
 274         default:
 275                 return 1;
 276         }
 277 }
 278
 279 static void
 280 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 281 {
 282         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 283                                                 struct nfsd4_blocked_lock, nbl_cb);
 284
 285         free_blocked_lock(nbl);
 286 }
 287
 288 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 289         .done           = nfsd4_cb_notify_lock_done,
 290         .release        = nfsd4_cb_notify_lock_release,
 291 };
 292
 293 static inline struct nfs4_stateowner *
 294 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 295 {
 296         atomic_inc(&sop->so_count);
 297         return sop;
 298 }
 299
 300 static int
 301 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 302 {
 303         return (sop->so_owner.len == owner->len) &&
 304                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 305 }
 306
 307 static struct nfs4_openowner *
 308 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 309                         struct nfs4_client *clp)
 310 {
 311         struct nfs4_stateowner *so;
 312
 313         lockdep_assert_held(&clp->cl_lock);
 314
 315         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 316                             so_strhash) {
 317                 if (!so->so_is_open_owner)
 318                         continue;
 319                 if (same_owner_str(so, &open->op_owner))
 320                         return openowner(nfs4_get_stateowner(so));
 321         }
 322         return NULL;
 323 }
 324
 325 static struct nfs4_openowner *
 326 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 327                         struct nfs4_client *clp)
 328 {
 329         struct nfs4_openowner *oo;
 330
 331         spin_lock(&clp->cl_lock);
 332         oo = find_openstateowner_str_locked(hashval, open, clp);
 333         spin_unlock(&clp->cl_lock);
 334         return oo;
 335 }
 336
 337 static inline u32
 338 opaque_hashval(const void *ptr, int nbytes)
 339 {
 340         unsigned char *cptr = (unsigned char *) ptr;
 341
 342         u32 x = 0;
 343         while (nbytes--) {
 344                 x *= 37;
 345                 x += *cptr++;
 346         }
 347         return x;
 348 }
 349
 350 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 351 {
 352         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 353
 354         kmem_cache_free(file_slab, fp);
 355 }
 356
 357 void
 358 put_nfs4_file(struct nfs4_file *fi)
 359 {
 360         might_lock(&state_lock);
 361
 362         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 363                 hlist_del_rcu(&fi->fi_hash);
 364                 spin_unlock(&state_lock);
 365                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 366                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 367                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 368         }
 369 }
 370
 371 static struct file *
 372 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 373 {
 374         if (f->fi_fds[oflag])
 375                 return get_file(f->fi_fds[oflag]);
 376         return NULL;
 377 }
 378
 379 static struct file *
 380 find_writeable_file_locked(struct nfs4_file *f)
 381 {
 382         struct file *ret;
 383
 384         lockdep_assert_held(&f->fi_lock);
 385
 386         ret = __nfs4_get_fd(f, O_WRONLY);
 387         if (!ret)
 388                 ret = __nfs4_get_fd(f, O_RDWR);
 389         return ret;
 390 }
 391
 392 static struct file *
 393 find_writeable_file(struct nfs4_file *f)
 394 {
 395         struct file *ret;
 396
 397         spin_lock(&f->fi_lock);
 398         ret = find_writeable_file_locked(f);
 399         spin_unlock(&f->fi_lock);
 400
 401         return ret;
 402 }
 403
 404 static struct file *find_readable_file_locked(struct nfs4_file *f)
 405 {
 406         struct file *ret;
 407
 408         lockdep_assert_held(&f->fi_lock);
 409
 410         ret = __nfs4_get_fd(f, O_RDONLY);
 411         if (!ret)
 412                 ret = __nfs4_get_fd(f, O_RDWR);
 413         return ret;
 414 }
 415
 416 static struct file *
 417 find_readable_file(struct nfs4_file *f)
 418 {
 419         struct file *ret;
 420
 421         spin_lock(&f->fi_lock);
 422         ret = find_readable_file_locked(f);
 423         spin_unlock(&f->fi_lock);
 424
 425         return ret;
 426 }
 427
 428 struct file *
 429 find_any_file(struct nfs4_file *f)
 430 {
 431         struct file *ret;
 432
 433         spin_lock(&f->fi_lock);
 434         ret = __nfs4_get_fd(f, O_RDWR);
 435         if (!ret) {
 436                 ret = __nfs4_get_fd(f, O_WRONLY);
 437                 if (!ret)
 438                         ret = __nfs4_get_fd(f, O_RDONLY);
 439         }
 440         spin_unlock(&f->fi_lock);
 441         return ret;
 442 }
 443
 444 static atomic_long_t num_delegations;
 445 unsigned long max_delegations;
 446
 447 /*
 448  * Open owner state (share locks)
 449  */
 450
 451 /* hash tables for lock and open owners */
 452 #define OWNER_HASH_BITS              8
 453 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 454 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 455
 456 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 457 {
 458         unsigned int ret;
 459
 460         ret = opaque_hashval(ownername->data, ownername->len);
 461         return ret & OWNER_HASH_MASK;
 462 }
 463
 464 /* hash table for nfs4_file */
 465 #define FILE_HASH_BITS                   8
 466 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 467
 468 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 469 {
 470         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 471 }
 472
 473 static unsigned int file_hashval(struct knfsd_fh *fh)
 474 {
 475         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 476 }
 477
 478 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 479
 480 static void
 481 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 482 {
 483         lockdep_assert_held(&fp->fi_lock);
 484
 485         if (access & NFS4_SHARE_ACCESS_WRITE)
 486                 atomic_inc(&fp->fi_access[O_WRONLY]);
 487         if (access & NFS4_SHARE_ACCESS_READ)
 488                 atomic_inc(&fp->fi_access[O_RDONLY]);
 489 }
 490
 491 static __be32
 492 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 493 {
 494         lockdep_assert_held(&fp->fi_lock);
 495
 496         /* Does this access mode make sense? */
 497         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 498                 return nfserr_inval;
 499
 500         /* Does it conflict with a deny mode already set? */
 501         if ((access & fp->fi_share_deny) != 0)
 502                 return nfserr_share_denied;
 503
 504         __nfs4_file_get_access(fp, access);
 505         return nfs_ok;
 506 }
 507
 508 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 509 {
 510         /* Common case is that there is no deny mode. */
 511         if (deny) {
 512                 /* Does this deny mode make sense? */
 513                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 514                         return nfserr_inval;
 515
 516                 if ((deny & NFS4_SHARE_DENY_READ) &&
 517                     atomic_read(&fp->fi_access[O_RDONLY]))
 518                         return nfserr_share_denied;
 519
 520                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 521                     atomic_read(&fp->fi_access[O_WRONLY]))
 522                         return nfserr_share_denied;
 523         }
 524         return nfs_ok;
 525 }
 526
 527 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 528 {
 529         might_lock(&fp->fi_lock);
 530
 531         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 532                 struct file *f1 = NULL;
 533                 struct file *f2 = NULL;
 534
 535                 swap(f1, fp->fi_fds[oflag]);
 536                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 537                         swap(f2, fp->fi_fds[O_RDWR]);
 538                 spin_unlock(&fp->fi_lock);
 539                 if (f1)
 540                         fput(f1);
 541                 if (f2)
 542                         fput(f2);
 543         }
 544 }
 545
 546 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 547 {
 548         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 549
 550         if (access & NFS4_SHARE_ACCESS_WRITE)
 551                 __nfs4_file_put_access(fp, O_WRONLY);
 552         if (access & NFS4_SHARE_ACCESS_READ)
 553                 __nfs4_file_put_access(fp, O_RDONLY);
 554 }
 555
 556 /*
 557  * Allocate a new open/delegation state counter. This is needed for
 558  * pNFS for proper return on close semantics.
 559  *
 560  * Note that we only allocate it for pNFS-enabled exports, otherwise
 561  * all pointers to struct nfs4_clnt_odstate are always NULL.
 562  */
 563 static struct nfs4_clnt_odstate *
 564 alloc_clnt_odstate(struct nfs4_client *clp)
 565 {
 566         struct nfs4_clnt_odstate *co;
 567
 568         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 569         if (co) {
 570                 co->co_client = clp;
 571                 atomic_set(&co->co_odcount, 1);
 572         }
 573         return co;
 574 }
 575
 576 static void
 577 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 578 {
 579         struct nfs4_file *fp = co->co_file;
 580
 581         lockdep_assert_held(&fp->fi_lock);
 582         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 583 }
 584
 585 static inline void
 586 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 587 {
 588         if (co)
 589                 atomic_inc(&co->co_odcount);
 590 }
 591
 592 static void
 593 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 594 {
 595         struct nfs4_file *fp;
 596
 597         if (!co)
 598                 return;
 599
 600         fp = co->co_file;
 601         if (atomic_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 602                 list_del(&co->co_perfile);
 603                 spin_unlock(&fp->fi_lock);
 604
 605                 nfsd4_return_all_file_layouts(co->co_client, fp);
 606                 kmem_cache_free(odstate_slab, co);
 607         }
 608 }
 609
 610 static struct nfs4_clnt_odstate *
 611 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 612 {
 613         struct nfs4_clnt_odstate *co;
 614         struct nfs4_client *cl;
 615
 616         if (!new)
 617                 return NULL;
 618
 619         cl = new->co_client;
 620
 621         spin_lock(&fp->fi_lock);
 622         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 623                 if (co->co_client == cl) {
 624                         get_clnt_odstate(co);
 625                         goto out;
 626                 }
 627         }
 628         co = new;
 629         co->co_file = fp;
 630         hash_clnt_odstate_locked(new);
 631 out:
 632         spin_unlock(&fp->fi_lock);
 633         return co;
 634 }
 635
 636 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 637                                   void (*sc_free)(struct nfs4_stid *))
 638 {
 639         struct nfs4_stid *stid;
 640         int new_id;
 641
 642         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 643         if (!stid)
 644                 return NULL;
 645
 646         idr_preload(GFP_KERNEL);
 647         spin_lock(&cl->cl_lock);
 648         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 649         spin_unlock(&cl->cl_lock);
 650         idr_preload_end();
 651         if (new_id < 0)
 652                 goto out_free;
 653
 654         stid->sc_free = sc_free;
 655         stid->sc_client = cl;
 656         stid->sc_stateid.si_opaque.so_id = new_id;
 657         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 658         /* Will be incremented before return to client: */
 659         atomic_set(&stid->sc_count, 1);
 660         spin_lock_init(&stid->sc_lock);
 661
 662         /*
 663          * It shouldn't be a problem to reuse an opaque stateid value.
 664          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 665          * example, a stray write retransmission could be accepted by
 666          * the server when it should have been rejected.  Therefore,
 667          * adopt a trick from the sctp code to attempt to maximize the
 668          * amount of time until an id is reused, by ensuring they always
 669          * "increase" (mod INT_MAX):
 670          */
 671         return stid;
 672 out_free:
 673         kmem_cache_free(slab, stid);
 674         return NULL;
 675 }
 676
 677 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 678 {
 679         struct nfs4_stid *stid;
 680
 681         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 682         if (!stid)
 683                 return NULL;
 684
 685         return openlockstateid(stid);
 686 }
 687
 688 static void nfs4_free_deleg(struct nfs4_stid *stid)
 689 {
 690         kmem_cache_free(deleg_slab, stid);
 691         atomic_long_dec(&num_delegations);
 692 }
 693
 694 /*
 695  * When we recall a delegation, we should be careful not to hand it
 696  * out again straight away.
 697  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 698  * in which the filehandles of recalled delegations are "stored".
 699  * If a filehandle appear in either filter, a delegation is blocked.
 700  * When a delegation is recalled, the filehandle is stored in the "new"
 701  * filter.
 702  * Every 30 seconds we swap the filters and clear the "new" one,
 703  * unless both are empty of course.
 704  *
 705  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 706  * low 3 bytes as hash-table indices.
 707  *
 708  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 709  * is used to manage concurrent access.  Testing does not need the lock
 710  * except when swapping the two filters.
 711  */
 712 static DEFINE_SPINLOCK(blocked_delegations_lock);
 713 static struct bloom_pair {
 714         int     entries, old_entries;
 715         time_t  swap_time;
 716         int     new; /* index into 'set' */
 717         DECLARE_BITMAP(set[2], 256);
 718 } blocked_delegations;
 719
 720 static int delegation_blocked(struct knfsd_fh *fh)
 721 {
 722         u32 hash;
 723         struct bloom_pair *bd = &blocked_delegations;
 724
 725         if (bd->entries == 0)
 726                 return 0;
 727         if (seconds_since_boot() - bd->swap_time > 30) {
 728                 spin_lock(&blocked_delegations_lock);
 729                 if (seconds_since_boot() - bd->swap_time > 30) {
 730                         bd->entries -= bd->old_entries;
 731                         bd->old_entries = bd->entries;
 732                         memset(bd->set[bd->new], 0,
 733                                sizeof(bd->set[0]));
 734                         bd->new = 1-bd->new;
 735                         bd->swap_time = seconds_since_boot();
 736                 }
 737                 spin_unlock(&blocked_delegations_lock);
 738         }
 739         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 740         if (test_bit(hash&255, bd->set[0]) &&
 741             test_bit((hash>>8)&255, bd->set[0]) &&
 742             test_bit((hash>>16)&255, bd->set[0]))
 743                 return 1;
 744
 745         if (test_bit(hash&255, bd->set[1]) &&
 746             test_bit((hash>>8)&255, bd->set[1]) &&
 747             test_bit((hash>>16)&255, bd->set[1]))
 748                 return 1;
 749
 750         return 0;
 751 }
 752
 753 static void block_delegations(struct knfsd_fh *fh)
 754 {
 755         u32 hash;
 756         struct bloom_pair *bd = &blocked_delegations;
 757
 758         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 759
 760         spin_lock(&blocked_delegations_lock);
 761         __set_bit(hash&255, bd->set[bd->new]);
 762         __set_bit((hash>>8)&255, bd->set[bd->new]);
 763         __set_bit((hash>>16)&255, bd->set[bd->new]);
 764         if (bd->entries == 0)
 765                 bd->swap_time = seconds_since_boot();
 766         bd->entries += 1;
 767         spin_unlock(&blocked_delegations_lock);
 768 }
 769
 770 static struct nfs4_delegation *
 771 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
 772                  struct nfs4_clnt_odstate *odstate)
 773 {
 774         struct nfs4_delegation *dp;
 775         long n;
 776
 777         dprintk("NFSD alloc_init_deleg\n");
 778         n = atomic_long_inc_return(&num_delegations);
 779         if (n < 0 || n > max_delegations)
 780                 goto out_dec;
 781         if (delegation_blocked(&current_fh->fh_handle))
 782                 goto out_dec;
 783         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 784         if (dp == NULL)
 785                 goto out_dec;
 786
 787         /*
 788          * delegation seqid's are never incremented.  The 4.1 special
 789          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 790          * 0 anyway just for consistency and use 1:
 791          */
 792         dp->dl_stid.sc_stateid.si_generation = 1;
 793         INIT_LIST_HEAD(&dp->dl_perfile);
 794         INIT_LIST_HEAD(&dp->dl_perclnt);
 795         INIT_LIST_HEAD(&dp->dl_recall_lru);
 796         dp->dl_clnt_odstate = odstate;
 797         get_clnt_odstate(odstate);
 798         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 799         dp->dl_retries = 1;
 800         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 801                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 802         return dp;
 803 out_dec:
 804         atomic_long_dec(&num_delegations);
 805         return NULL;
 806 }
 807
 808 void
 809 nfs4_put_stid(struct nfs4_stid *s)
 810 {
 811         struct nfs4_file *fp = s->sc_file;
 812         struct nfs4_client *clp = s->sc_client;
 813
 814         might_lock(&clp->cl_lock);
 815
 816         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 817                 wake_up_all(&close_wq);
 818                 return;
 819         }
 820         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 821         spin_unlock(&clp->cl_lock);
 822         s->sc_free(s);
 823         if (fp)
 824                 put_nfs4_file(fp);
 825 }
 826
 827 void
 828 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 829 {
 830         stateid_t *src = &stid->sc_stateid;
 831
 832         spin_lock(&stid->sc_lock);
 833         if (unlikely(++src->si_generation == 0))
 834                 src->si_generation = 1;
 835         memcpy(dst, src, sizeof(*dst));
 836         spin_unlock(&stid->sc_lock);
 837 }
 838
 839 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 840 {
 841         struct file *filp = NULL;
 842
 843         spin_lock(&fp->fi_lock);
 844         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 845                 swap(filp, fp->fi_deleg_file);
 846         spin_unlock(&fp->fi_lock);
 847
 848         if (filp) {
 849                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 850                 fput(filp);
 851         }
 852 }
 853
 854 void nfs4_unhash_stid(struct nfs4_stid *s)
 855 {
 856         s->sc_type = 0;
 857 }
 858
 859 /**
 860  * nfs4_get_existing_delegation - Discover if this delegation already exists
 861  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 862  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 863  *
 864  * Return:
 865  *      On success: NULL if an existing delegation was not found.
 866  *
 867  *      On error: -EAGAIN if one was previously granted to this nfs4_client
 868  *                 for this nfs4_file.
 869  *
 870  */
 871
 872 static int
 873 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
 874 {
 875         struct nfs4_delegation *searchdp = NULL;
 876         struct nfs4_client *searchclp = NULL;
 877
 878         lockdep_assert_held(&state_lock);
 879         lockdep_assert_held(&fp->fi_lock);
 880
 881         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 882                 searchclp = searchdp->dl_stid.sc_client;
 883                 if (clp == searchclp) {
 884                         return -EAGAIN;
 885                 }
 886         }
 887         return 0;
 888 }
 889
 890 /**
 891  * hash_delegation_locked - Add a delegation to the appropriate lists
 892  * @dp:     a pointer to the nfs4_delegation we are adding.
 893  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 894  *
 895  * Return:
 896  *      On success: NULL if the delegation was successfully hashed.
 897  *
 898  *      On error: -EAGAIN if one was previously granted to this
 899  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 900  *
 901  */
 902
 903 static int
 904 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 905 {
 906         int status;
 907         struct nfs4_client *clp = dp->dl_stid.sc_client;
 908
 909         lockdep_assert_held(&state_lock);
 910         lockdep_assert_held(&fp->fi_lock);
 911
 912         status = nfs4_get_existing_delegation(clp, fp);
 913         if (status)
 914                 return status;
 915         ++fp->fi_delegees;
 916         atomic_inc(&dp->dl_stid.sc_count);
 917         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 918         list_add(&dp->dl_perfile, &fp->fi_delegations);
 919         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 920         return 0;
 921 }
 922
 923 static bool
 924 unhash_delegation_locked(struct nfs4_delegation *dp)
 925 {
 926         struct nfs4_file *fp = dp->dl_stid.sc_file;
 927
 928         lockdep_assert_held(&state_lock);
 929
 930         if (list_empty(&dp->dl_perfile))
 931                 return false;
 932
 933         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 934         /* Ensure that deleg break won't try to requeue it */
 935         ++dp->dl_time;
 936         spin_lock(&fp->fi_lock);
 937         list_del_init(&dp->dl_perclnt);
 938         list_del_init(&dp->dl_recall_lru);
 939         list_del_init(&dp->dl_perfile);
 940         spin_unlock(&fp->fi_lock);
 941         return true;
 942 }
 943
 944 static void destroy_delegation(struct nfs4_delegation *dp)
 945 {
 946         bool unhashed;
 947
 948         spin_lock(&state_lock);
 949         unhashed = unhash_delegation_locked(dp);
 950         spin_unlock(&state_lock);
 951         if (unhashed) {
 952                 put_clnt_odstate(dp->dl_clnt_odstate);
 953                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 954                 nfs4_put_stid(&dp->dl_stid);
 955         }
 956 }
 957
 958 static void revoke_delegation(struct nfs4_delegation *dp)
 959 {
 960         struct nfs4_client *clp = dp->dl_stid.sc_client;
 961
 962         WARN_ON(!list_empty(&dp->dl_recall_lru));
 963
 964         put_clnt_odstate(dp->dl_clnt_odstate);
 965         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 966
 967         if (clp->cl_minorversion == 0)
 968                 nfs4_put_stid(&dp->dl_stid);
 969         else {
 970                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 971                 spin_lock(&clp->cl_lock);
 972                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 973                 spin_unlock(&clp->cl_lock);
 974         }
 975 }
 976
 977 /*
 978  * SETCLIENTID state
 979  */
 980
 981 static unsigned int clientid_hashval(u32 id)
 982 {
 983         return id & CLIENT_HASH_MASK;
 984 }
 985
 986 static unsigned int clientstr_hashval(const char *name)
 987 {
 988         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 989 }
 990
 991 /*
 992  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 993  * st_{access,deny}_bmap field of the stateid, in order to track not
 994  * only what share bits are currently in force, but also what
 995  * combinations of share bits previous opens have used.  This allows us
 996  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 997  * return an error if the client attempt to downgrade to a combination
 998  * of share bits not explicable by closing some of its previous opens.
 999  *
1000  * XXX: This enforcement is actually incomplete, since we don't keep
1001  * track of access/deny bit combinations; so, e.g., we allow:
1002  *
1003  *      OPEN allow read, deny write
1004  *      OPEN allow both, deny none
1005  *      DOWNGRADE allow read, deny none
1006  *
1007  * which we should reject.
1008  */
1009 static unsigned int
1010 bmap_to_share_mode(unsigned long bmap) {
1011         int i;
1012         unsigned int access = 0;
1013
1014         for (i = 1; i < 4; i++) {
1015                 if (test_bit(i, &bmap))
1016                         access |= i;
1017         }
1018         return access;
1019 }
1020
1021 /* set share access for a given stateid */
1022 static inline void
1023 set_access(u32 access, struct nfs4_ol_stateid *stp)
1024 {
1025         unsigned char mask = 1 << access;
1026
1027         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1028         stp->st_access_bmap |= mask;
1029 }
1030
1031 /* clear share access for a given stateid */
1032 static inline void
1033 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1034 {
1035         unsigned char mask = 1 << access;
1036
1037         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1038         stp->st_access_bmap &= ~mask;
1039 }
1040
1041 /* test whether a given stateid has access */
1042 static inline bool
1043 test_access(u32 access, struct nfs4_ol_stateid *stp)
1044 {
1045         unsigned char mask = 1 << access;
1046
1047         return (bool)(stp->st_access_bmap & mask);
1048 }
1049
1050 /* set share deny for a given stateid */
1051 static inline void
1052 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1053 {
1054         unsigned char mask = 1 << deny;
1055
1056         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1057         stp->st_deny_bmap |= mask;
1058 }
1059
1060 /* clear share deny for a given stateid */
1061 static inline void
1062 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1063 {
1064         unsigned char mask = 1 << deny;
1065
1066         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1067         stp->st_deny_bmap &= ~mask;
1068 }
1069
1070 /* test whether a given stateid is denying specific access */
1071 static inline bool
1072 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1073 {
1074         unsigned char mask = 1 << deny;
1075
1076         return (bool)(stp->st_deny_bmap & mask);
1077 }
1078
1079 static int nfs4_access_to_omode(u32 access)
1080 {
1081         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1082         case NFS4_SHARE_ACCESS_READ:
1083                 return O_RDONLY;
1084         case NFS4_SHARE_ACCESS_WRITE:
1085                 return O_WRONLY;
1086         case NFS4_SHARE_ACCESS_BOTH:
1087                 return O_RDWR;
1088         }
1089         WARN_ON_ONCE(1);
1090         return O_RDONLY;
1091 }
1092
1093 /*
1094  * A stateid that had a deny mode associated with it is being released
1095  * or downgraded. Recalculate the deny mode on the file.
1096  */
1097 static void
1098 recalculate_deny_mode(struct nfs4_file *fp)
1099 {
1100         struct nfs4_ol_stateid *stp;
1101
1102         spin_lock(&fp->fi_lock);
1103         fp->fi_share_deny = 0;
1104         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1105                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1106         spin_unlock(&fp->fi_lock);
1107 }
1108
1109 static void
1110 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1111 {
1112         int i;
1113         bool change = false;
1114
1115         for (i = 1; i < 4; i++) {
1116                 if ((i & deny) != i) {
1117                         change = true;
1118                         clear_deny(i, stp);
1119                 }
1120         }
1121
1122         /* Recalculate per-file deny mode if there was a change */
1123         if (change)
1124                 recalculate_deny_mode(stp->st_stid.sc_file);
1125 }
1126
1127 /* release all access and file references for a given stateid */
1128 static void
1129 release_all_access(struct nfs4_ol_stateid *stp)
1130 {
1131         int i;
1132         struct nfs4_file *fp = stp->st_stid.sc_file;
1133
1134         if (fp && stp->st_deny_bmap != 0)
1135                 recalculate_deny_mode(fp);
1136
1137         for (i = 1; i < 4; i++) {
1138                 if (test_access(i, stp))
1139                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1140                 clear_access(i, stp);
1141         }
1142 }
1143
1144 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1145 {
1146         kfree(sop->so_owner.data);
1147         sop->so_ops->so_free(sop);
1148 }
1149
1150 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1151 {
1152         struct nfs4_client *clp = sop->so_client;
1153
1154         might_lock(&clp->cl_lock);
1155
1156         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1157                 return;
1158         sop->so_ops->so_unhash(sop);
1159         spin_unlock(&clp->cl_lock);
1160         nfs4_free_stateowner(sop);
1161 }
1162
1163 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1164 {
1165         struct nfs4_file *fp = stp->st_stid.sc_file;
1166
1167         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1168
1169         if (list_empty(&stp->st_perfile))
1170                 return false;
1171
1172         spin_lock(&fp->fi_lock);
1173         list_del_init(&stp->st_perfile);
1174         spin_unlock(&fp->fi_lock);
1175         list_del(&stp->st_perstateowner);
1176         return true;
1177 }
1178
1179 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1180 {
1181         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1182
1183         put_clnt_odstate(stp->st_clnt_odstate);
1184         release_all_access(stp);
1185         if (stp->st_stateowner)
1186                 nfs4_put_stateowner(stp->st_stateowner);
1187         kmem_cache_free(stateid_slab, stid);
1188 }
1189
1190 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1191 {
1192         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1193         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1194         struct file *file;
1195
1196         file = find_any_file(stp->st_stid.sc_file);
1197         if (file)
1198                 filp_close(file, (fl_owner_t)lo);
1199         nfs4_free_ol_stateid(stid);
1200 }
1201
1202 /*
1203  * Put the persistent reference to an already unhashed generic stateid, while
1204  * holding the cl_lock. If it's the last reference, then put it onto the
1205  * reaplist for later destruction.
1206  */
1207 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1208                                        struct list_head *reaplist)
1209 {
1210         struct nfs4_stid *s = &stp->st_stid;
1211         struct nfs4_client *clp = s->sc_client;
1212
1213         lockdep_assert_held(&clp->cl_lock);
1214
1215         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1216
1217         if (!atomic_dec_and_test(&s->sc_count)) {
1218                 wake_up_all(&close_wq);
1219                 return;
1220         }
1221
1222         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1223         list_add(&stp->st_locks, reaplist);
1224 }
1225
1226 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1227 {
1228         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1229
1230         list_del_init(&stp->st_locks);
1231         nfs4_unhash_stid(&stp->st_stid);
1232         return unhash_ol_stateid(stp);
1233 }
1234
1235 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1236 {
1237         struct nfs4_client *clp = stp->st_stid.sc_client;
1238         bool unhashed;
1239
1240         spin_lock(&clp->cl_lock);
1241         unhashed = unhash_lock_stateid(stp);
1242         spin_unlock(&clp->cl_lock);
1243         if (unhashed)
1244                 nfs4_put_stid(&stp->st_stid);
1245 }
1246
1247 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1248 {
1249         struct nfs4_client *clp = lo->lo_owner.so_client;
1250
1251         lockdep_assert_held(&clp->cl_lock);
1252
1253         list_del_init(&lo->lo_owner.so_strhash);
1254 }
1255
1256 /*
1257  * Free a list of generic stateids that were collected earlier after being
1258  * fully unhashed.
1259  */
1260 static void
1261 free_ol_stateid_reaplist(struct list_head *reaplist)
1262 {
1263         struct nfs4_ol_stateid *stp;
1264         struct nfs4_file *fp;
1265
1266         might_sleep();
1267
1268         while (!list_empty(reaplist)) {
1269                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1270                                        st_locks);
1271                 list_del(&stp->st_locks);
1272                 fp = stp->st_stid.sc_file;
1273                 stp->st_stid.sc_free(&stp->st_stid);
1274                 if (fp)
1275                         put_nfs4_file(fp);
1276         }
1277 }
1278
1279 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1280                                        struct list_head *reaplist)
1281 {
1282         struct nfs4_ol_stateid *stp;
1283
1284         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1285
1286         while (!list_empty(&open_stp->st_locks)) {
1287                 stp = list_entry(open_stp->st_locks.next,
1288                                 struct nfs4_ol_stateid, st_locks);
1289                 WARN_ON(!unhash_lock_stateid(stp));
1290                 put_ol_stateid_locked(stp, reaplist);
1291         }
1292 }
1293
1294 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1295                                 struct list_head *reaplist)
1296 {
1297         bool unhashed;
1298
1299         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1300
1301         unhashed = unhash_ol_stateid(stp);
1302         release_open_stateid_locks(stp, reaplist);
1303         return unhashed;
1304 }
1305
1306 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1307 {
1308         LIST_HEAD(reaplist);
1309
1310         spin_lock(&stp->st_stid.sc_client->cl_lock);
1311         if (unhash_open_stateid(stp, &reaplist))
1312                 put_ol_stateid_locked(stp, &reaplist);
1313         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1314         free_ol_stateid_reaplist(&reaplist);
1315 }
1316
1317 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1318 {
1319         struct nfs4_client *clp = oo->oo_owner.so_client;
1320
1321         lockdep_assert_held(&clp->cl_lock);
1322
1323         list_del_init(&oo->oo_owner.so_strhash);
1324         list_del_init(&oo->oo_perclient);
1325 }
1326
1327 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1328 {
1329         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1330                                           nfsd_net_id);
1331         struct nfs4_ol_stateid *s;
1332
1333         spin_lock(&nn->client_lock);
1334         s = oo->oo_last_closed_stid;
1335         if (s) {
1336                 list_del_init(&oo->oo_close_lru);
1337                 oo->oo_last_closed_stid = NULL;
1338         }
1339         spin_unlock(&nn->client_lock);
1340         if (s)
1341                 nfs4_put_stid(&s->st_stid);
1342 }
1343
1344 static void release_openowner(struct nfs4_openowner *oo)
1345 {
1346         struct nfs4_ol_stateid *stp;
1347         struct nfs4_client *clp = oo->oo_owner.so_client;
1348         struct list_head reaplist;
1349
1350         INIT_LIST_HEAD(&reaplist);
1351
1352         spin_lock(&clp->cl_lock);
1353         unhash_openowner_locked(oo);
1354         while (!list_empty(&oo->oo_owner.so_stateids)) {
1355                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1356                                 struct nfs4_ol_stateid, st_perstateowner);
1357                 if (unhash_open_stateid(stp, &reaplist))
1358                         put_ol_stateid_locked(stp, &reaplist);
1359         }
1360         spin_unlock(&clp->cl_lock);
1361         free_ol_stateid_reaplist(&reaplist);
1362         release_last_closed_stateid(oo);
1363         nfs4_put_stateowner(&oo->oo_owner);
1364 }
1365
1366 static inline int
1367 hash_sessionid(struct nfs4_sessionid *sessionid)
1368 {
1369         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1370
1371         return sid->sequence % SESSION_HASH_SIZE;
1372 }
1373
1374 #ifdef CONFIG_SUNRPC_DEBUG
1375 static inline void
1376 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1377 {
1378         u32 *ptr = (u32 *)(&sessionid->data[0]);
1379         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1380 }
1381 #else
1382 static inline void
1383 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1384 {
1385 }
1386 #endif
1387
1388 /*
1389  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1390  * won't be used for replay.
1391  */
1392 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1393 {
1394         struct nfs4_stateowner *so = cstate->replay_owner;
1395
1396         if (nfserr == nfserr_replay_me)
1397                 return;
1398
1399         if (!seqid_mutating_err(ntohl(nfserr))) {
1400                 nfsd4_cstate_clear_replay(cstate);
1401                 return;
1402         }
1403         if (!so)
1404                 return;
1405         if (so->so_is_open_owner)
1406                 release_last_closed_stateid(openowner(so));
1407         so->so_seqid++;
1408         return;
1409 }
1410
1411 static void
1412 gen_sessionid(struct nfsd4_session *ses)
1413 {
1414         struct nfs4_client *clp = ses->se_client;
1415         struct nfsd4_sessionid *sid;
1416
1417         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1418         sid->clientid = clp->cl_clientid;
1419         sid->sequence = current_sessionid++;
1420         sid->reserved = 0;
1421 }
1422
1423 /*
1424  * The protocol defines ca_maxresponssize_cached to include the size of
1425  * the rpc header, but all we need to cache is the data starting after
1426  * the end of the initial SEQUENCE operation--the rest we regenerate
1427  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1428  * value that is the number of bytes in our cache plus a few additional
1429  * bytes.  In order to stay on the safe side, and not promise more than
1430  * we can cache, those additional bytes must be the minimum possible: 24
1431  * bytes of rpc header (xid through accept state, with AUTH_NULL
1432  * verifier), 12 for the compound header (with zero-length tag), and 44
1433  * for the SEQUENCE op response:
1434  */
1435 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1436
1437 static void
1438 free_session_slots(struct nfsd4_session *ses)
1439 {
1440         int i;
1441
1442         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1443                 kfree(ses->se_slots[i]);
1444 }
1445
1446 /*
1447  * We don't actually need to cache the rpc and session headers, so we
1448  * can allocate a little less for each slot:
1449  */
1450 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1451 {
1452         u32 size;
1453
1454         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1455                 size = 0;
1456         else
1457                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1458         return size + sizeof(struct nfsd4_slot);
1459 }
1460
1461 /*
1462  * XXX: If we run out of reserved DRC memory we could (up to a point)
1463  * re-negotiate active sessions and reduce their slot usage to make
1464  * room for new connections. For now we just fail the create session.
1465  */
1466 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1467 {
1468         u32 slotsize = slot_bytes(ca);
1469         u32 num = ca->maxreqs;
1470         int avail;
1471
1472         spin_lock(&nfsd_drc_lock);
1473         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1474                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1475         num = min_t(int, num, avail / slotsize);
1476         nfsd_drc_mem_used += num * slotsize;
1477         spin_unlock(&nfsd_drc_lock);
1478
1479         return num;
1480 }
1481
1482 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1483 {
1484         int slotsize = slot_bytes(ca);
1485
1486         spin_lock(&nfsd_drc_lock);
1487         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1488         spin_unlock(&nfsd_drc_lock);
1489 }
1490
1491 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1492                                            struct nfsd4_channel_attrs *battrs)
1493 {
1494         int numslots = fattrs->maxreqs;
1495         int slotsize = slot_bytes(fattrs);
1496         struct nfsd4_session *new;
1497         int mem, i;
1498
1499         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1500                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1501         mem = numslots * sizeof(struct nfsd4_slot *);
1502
1503         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1504         if (!new)
1505                 return NULL;
1506         /* allocate each struct nfsd4_slot and data cache in one piece */
1507         for (i = 0; i < numslots; i++) {
1508                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1509                 if (!new->se_slots[i])
1510                         goto out_free;
1511         }
1512
1513         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1514         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1515
1516         return new;
1517 out_free:
1518         while (i--)
1519                 kfree(new->se_slots[i]);
1520         kfree(new);
1521         return NULL;
1522 }
1523
1524 static void free_conn(struct nfsd4_conn *c)
1525 {
1526         svc_xprt_put(c->cn_xprt);
1527         kfree(c);
1528 }
1529
1530 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1531 {
1532         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1533         struct nfs4_client *clp = c->cn_session->se_client;
1534
1535         spin_lock(&clp->cl_lock);
1536         if (!list_empty(&c->cn_persession)) {
1537                 list_del(&c->cn_persession);
1538                 free_conn(c);
1539         }
1540         nfsd4_probe_callback(clp);
1541         spin_unlock(&clp->cl_lock);
1542 }
1543
1544 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1545 {
1546         struct nfsd4_conn *conn;
1547
1548         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1549         if (!conn)
1550                 return NULL;
1551         svc_xprt_get(rqstp->rq_xprt);
1552         conn->cn_xprt = rqstp->rq_xprt;
1553         conn->cn_flags = flags;
1554         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1555         return conn;
1556 }
1557
1558 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1559 {
1560         conn->cn_session = ses;
1561         list_add(&conn->cn_persession, &ses->se_conns);
1562 }
1563
1564 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1565 {
1566         struct nfs4_client *clp = ses->se_client;
1567
1568         spin_lock(&clp->cl_lock);
1569         __nfsd4_hash_conn(conn, ses);
1570         spin_unlock(&clp->cl_lock);
1571 }
1572
1573 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1574 {
1575         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1576         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1577 }
1578
1579 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1580 {
1581         int ret;
1582
1583         nfsd4_hash_conn(conn, ses);
1584         ret = nfsd4_register_conn(conn);
1585         if (ret)
1586                 /* oops; xprt is already down: */
1587                 nfsd4_conn_lost(&conn->cn_xpt_user);
1588         /* We may have gained or lost a callback channel: */
1589         nfsd4_probe_callback_sync(ses->se_client);
1590 }
1591
1592 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1593 {
1594         u32 dir = NFS4_CDFC4_FORE;
1595
1596         if (cses->flags & SESSION4_BACK_CHAN)
1597                 dir |= NFS4_CDFC4_BACK;
1598         return alloc_conn(rqstp, dir);
1599 }
1600
1601 /* must be called under client_lock */
1602 static void nfsd4_del_conns(struct nfsd4_session *s)
1603 {
1604         struct nfs4_client *clp = s->se_client;
1605         struct nfsd4_conn *c;
1606
1607         spin_lock(&clp->cl_lock);
1608         while (!list_empty(&s->se_conns)) {
1609                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1610                 list_del_init(&c->cn_persession);
1611                 spin_unlock(&clp->cl_lock);
1612
1613                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1614                 free_conn(c);
1615
1616                 spin_lock(&clp->cl_lock);
1617         }
1618         spin_unlock(&clp->cl_lock);
1619 }
1620
1621 static void __free_session(struct nfsd4_session *ses)
1622 {
1623         free_session_slots(ses);
1624         kfree(ses);
1625 }
1626
1627 static void free_session(struct nfsd4_session *ses)
1628 {
1629         nfsd4_del_conns(ses);
1630         nfsd4_put_drc_mem(&ses->se_fchannel);
1631         __free_session(ses);
1632 }
1633
1634 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1635 {
1636         int idx;
1637         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1638
1639         new->se_client = clp;
1640         gen_sessionid(new);
1641
1642         INIT_LIST_HEAD(&new->se_conns);
1643
1644         new->se_cb_seq_nr = 1;
1645         new->se_flags = cses->flags;
1646         new->se_cb_prog = cses->callback_prog;
1647         new->se_cb_sec = cses->cb_sec;
1648         atomic_set(&new->se_ref, 0);
1649         idx = hash_sessionid(&new->se_sessionid);
1650         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1651         spin_lock(&clp->cl_lock);
1652         list_add(&new->se_perclnt, &clp->cl_sessions);
1653         spin_unlock(&clp->cl_lock);
1654
1655         {
1656                 struct sockaddr *sa = svc_addr(rqstp);
1657                 /*
1658                  * This is a little silly; with sessions there's no real
1659                  * use for the callback address.  Use the peer address
1660                  * as a reasonable default for now, but consider fixing
1661                  * the rpc client not to require an address in the
1662                  * future:
1663                  */
1664                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1665                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1666         }
1667 }
1668
1669 /* caller must hold client_lock */
1670 static struct nfsd4_session *
1671 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1672 {
1673         struct nfsd4_session *elem;
1674         int idx;
1675         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1676
1677         lockdep_assert_held(&nn->client_lock);
1678
1679         dump_sessionid(__func__, sessionid);
1680         idx = hash_sessionid(sessionid);
1681         /* Search in the appropriate list */
1682         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1683                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1684                             NFS4_MAX_SESSIONID_LEN)) {
1685                         return elem;
1686                 }
1687         }
1688
1689         dprintk("%s: session not found\n", __func__);
1690         return NULL;
1691 }
1692
1693 static struct nfsd4_session *
1694 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1695                 __be32 *ret)
1696 {
1697         struct nfsd4_session *session;
1698         __be32 status = nfserr_badsession;
1699
1700         session = __find_in_sessionid_hashtbl(sessionid, net);
1701         if (!session)
1702                 goto out;
1703         status = nfsd4_get_session_locked(session);
1704         if (status)
1705                 session = NULL;
1706 out:
1707         *ret = status;
1708         return session;
1709 }
1710
1711 /* caller must hold client_lock */
1712 static void
1713 unhash_session(struct nfsd4_session *ses)
1714 {
1715         struct nfs4_client *clp = ses->se_client;
1716         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1717
1718         lockdep_assert_held(&nn->client_lock);
1719
1720         list_del(&ses->se_hash);
1721         spin_lock(&ses->se_client->cl_lock);
1722         list_del(&ses->se_perclnt);
1723         spin_unlock(&ses->se_client->cl_lock);
1724 }
1725
1726 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1727 static int
1728 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1729 {
1730         /*
1731          * We're assuming the clid was not given out from a boot
1732          * precisely 2^32 (about 136 years) before this one.  That seems
1733          * a safe assumption:
1734          */
1735         if (clid->cl_boot == (u32)nn->boot_time)
1736                 return 0;
1737         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1738                 clid->cl_boot, clid->cl_id, nn->boot_time);
1739         return 1;
1740 }
1741
1742 /*
1743  * XXX Should we use a slab cache ?
1744  * This type of memory management is somewhat inefficient, but we use it
1745  * anyway since SETCLIENTID is not a common operation.
1746  */
1747 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1748 {
1749         struct nfs4_client *clp;
1750         int i;
1751
1752         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1753         if (clp == NULL)
1754                 return NULL;
1755         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1756         if (clp->cl_name.data == NULL)
1757                 goto err_no_name;
1758         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1759                         OWNER_HASH_SIZE, GFP_KERNEL);
1760         if (!clp->cl_ownerstr_hashtbl)
1761                 goto err_no_hashtbl;
1762         for (i = 0; i < OWNER_HASH_SIZE; i++)
1763                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1764         clp->cl_name.len = name.len;
1765         INIT_LIST_HEAD(&clp->cl_sessions);
1766         idr_init(&clp->cl_stateids);
1767         atomic_set(&clp->cl_refcount, 0);
1768         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1769         INIT_LIST_HEAD(&clp->cl_idhash);
1770         INIT_LIST_HEAD(&clp->cl_openowners);
1771         INIT_LIST_HEAD(&clp->cl_delegations);
1772         INIT_LIST_HEAD(&clp->cl_lru);
1773         INIT_LIST_HEAD(&clp->cl_revoked);
1774 #ifdef CONFIG_NFSD_PNFS
1775         INIT_LIST_HEAD(&clp->cl_lo_states);
1776 #endif
1777         spin_lock_init(&clp->cl_lock);
1778         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1779         return clp;
1780 err_no_hashtbl:
1781         kfree(clp->cl_name.data);
1782 err_no_name:
1783         kfree(clp);
1784         return NULL;
1785 }
1786
1787 static void
1788 free_client(struct nfs4_client *clp)
1789 {
1790         while (!list_empty(&clp->cl_sessions)) {
1791                 struct nfsd4_session *ses;
1792                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1793                                 se_perclnt);
1794                 list_del(&ses->se_perclnt);
1795                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1796                 free_session(ses);
1797         }
1798         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1799         free_svc_cred(&clp->cl_cred);
1800         kfree(clp->cl_ownerstr_hashtbl);
1801         kfree(clp->cl_name.data);
1802         idr_destroy(&clp->cl_stateids);
1803         kfree(clp);
1804 }
1805
1806 /* must be called under the client_lock */
1807 static void
1808 unhash_client_locked(struct nfs4_client *clp)
1809 {
1810         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1811         struct nfsd4_session *ses;
1812
1813         lockdep_assert_held(&nn->client_lock);
1814
1815         /* Mark the client as expired! */
1816         clp->cl_time = 0;
1817         /* Make it invisible */
1818         if (!list_empty(&clp->cl_idhash)) {
1819                 list_del_init(&clp->cl_idhash);
1820                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1821                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1822                 else
1823                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1824         }
1825         list_del_init(&clp->cl_lru);
1826         spin_lock(&clp->cl_lock);
1827         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1828                 list_del_init(&ses->se_hash);
1829         spin_unlock(&clp->cl_lock);
1830 }
1831
1832 static void
1833 unhash_client(struct nfs4_client *clp)
1834 {
1835         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1836
1837         spin_lock(&nn->client_lock);
1838         unhash_client_locked(clp);
1839         spin_unlock(&nn->client_lock);
1840 }
1841
1842 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1843 {
1844         if (atomic_read(&clp->cl_refcount))
1845                 return nfserr_jukebox;
1846         unhash_client_locked(clp);
1847         return nfs_ok;
1848 }
1849
1850 static void
1851 __destroy_client(struct nfs4_client *clp)
1852 {
1853         struct nfs4_openowner *oo;
1854         struct nfs4_delegation *dp;
1855         struct list_head reaplist;
1856
1857         INIT_LIST_HEAD(&reaplist);
1858         spin_lock(&state_lock);
1859         while (!list_empty(&clp->cl_delegations)) {
1860                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1861                 WARN_ON(!unhash_delegation_locked(dp));
1862                 list_add(&dp->dl_recall_lru, &reaplist);
1863         }
1864         spin_unlock(&state_lock);
1865         while (!list_empty(&reaplist)) {
1866                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1867                 list_del_init(&dp->dl_recall_lru);
1868                 put_clnt_odstate(dp->dl_clnt_odstate);
1869                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1870                 nfs4_put_stid(&dp->dl_stid);
1871         }
1872         while (!list_empty(&clp->cl_revoked)) {
1873                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1874                 list_del_init(&dp->dl_recall_lru);
1875                 nfs4_put_stid(&dp->dl_stid);
1876         }
1877         while (!list_empty(&clp->cl_openowners)) {
1878                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1879                 nfs4_get_stateowner(&oo->oo_owner);
1880                 release_openowner(oo);
1881         }
1882         nfsd4_return_all_client_layouts(clp);
1883         nfsd4_shutdown_callback(clp);
1884         if (clp->cl_cb_conn.cb_xprt)
1885                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1886         free_client(clp);
1887 }
1888
1889 static void
1890 destroy_client(struct nfs4_client *clp)
1891 {
1892         unhash_client(clp);
1893         __destroy_client(clp);
1894 }
1895
1896 static void expire_client(struct nfs4_client *clp)
1897 {
1898         unhash_client(clp);
1899         nfsd4_client_record_remove(clp);
1900         __destroy_client(clp);
1901 }
1902
1903 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1904 {
1905         memcpy(target->cl_verifier.data, source->data,
1906                         sizeof(target->cl_verifier.data));
1907 }
1908
1909 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1910 {
1911         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1912         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1913 }
1914
1915 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1916 {
1917         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1918         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1919                                                                 GFP_KERNEL);
1920         if ((source->cr_principal && ! target->cr_principal) ||
1921             (source->cr_raw_principal && ! target->cr_raw_principal))
1922                 return -ENOMEM;
1923
1924         target->cr_flavor = source->cr_flavor;
1925         target->cr_uid = source->cr_uid;
1926         target->cr_gid = source->cr_gid;
1927         target->cr_group_info = source->cr_group_info;
1928         get_group_info(target->cr_group_info);
1929         target->cr_gss_mech = source->cr_gss_mech;
1930         if (source->cr_gss_mech)
1931                 gss_mech_get(source->cr_gss_mech);
1932         return 0;
1933 }
1934
1935 static int
1936 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1937 {
1938         if (o1->len < o2->len)
1939                 return -1;
1940         if (o1->len > o2->len)
1941                 return 1;
1942         return memcmp(o1->data, o2->data, o1->len);
1943 }
1944
1945 static int same_name(const char *n1, const char *n2)
1946 {
1947         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1948 }
1949
1950 static int
1951 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1952 {
1953         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1954 }
1955
1956 static int
1957 same_clid(clientid_t *cl1, clientid_t *cl2)
1958 {
1959         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1960 }
1961
1962 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1963 {
1964         int i;
1965
1966         if (g1->ngroups != g2->ngroups)
1967                 return false;
1968         for (i=0; i<g1->ngroups; i++)
1969                 if (!gid_eq(g1->gid[i], g2->gid[i]))
1970                         return false;
1971         return true;
1972 }
1973
1974 /*
1975  * RFC 3530 language requires clid_inuse be returned when the
1976  * "principal" associated with a requests differs from that previously
1977  * used.  We use uid, gid's, and gss principal string as our best
1978  * approximation.  We also don't want to allow non-gss use of a client
1979  * established using gss: in theory cr_principal should catch that
1980  * change, but in practice cr_principal can be null even in the gss case
1981  * since gssd doesn't always pass down a principal string.
1982  */
1983 static bool is_gss_cred(struct svc_cred *cr)
1984 {
1985         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1986         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1987 }
1988
1989
1990 static bool
1991 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1992 {
1993         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1994                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1995                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1996                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1997                 return false;
1998         if (cr1->cr_principal == cr2->cr_principal)
1999                 return true;
2000         if (!cr1->cr_principal || !cr2->cr_principal)
2001                 return false;
2002         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2003 }
2004
2005 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2006 {
2007         struct svc_cred *cr = &rqstp->rq_cred;
2008         u32 service;
2009
2010         if (!cr->cr_gss_mech)
2011                 return false;
2012         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2013         return service == RPC_GSS_SVC_INTEGRITY ||
2014                service == RPC_GSS_SVC_PRIVACY;
2015 }
2016
2017 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2018 {
2019         struct svc_cred *cr = &rqstp->rq_cred;
2020
2021         if (!cl->cl_mach_cred)
2022                 return true;
2023         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2024                 return false;
2025         if (!svc_rqst_integrity_protected(rqstp))
2026                 return false;
2027         if (cl->cl_cred.cr_raw_principal)
2028                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2029                                                 cr->cr_raw_principal);
2030         if (!cr->cr_principal)
2031                 return false;
2032         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2033 }
2034
2035 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2036 {
2037         __be32 verf[2];
2038
2039         /*
2040          * This is opaque to client, so no need to byte-swap. Use
2041          * __force to keep sparse happy
2042          */
2043         verf[0] = (__force __be32)get_seconds();
2044         verf[1] = (__force __be32)nn->clverifier_counter++;
2045         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2046 }
2047
2048 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2049 {
2050         clp->cl_clientid.cl_boot = nn->boot_time;
2051         clp->cl_clientid.cl_id = nn->clientid_counter++;
2052         gen_confirm(clp, nn);
2053 }
2054
2055 static struct nfs4_stid *
2056 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2057 {
2058         struct nfs4_stid *ret;
2059
2060         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2061         if (!ret || !ret->sc_type)
2062                 return NULL;
2063         return ret;
2064 }
2065
2066 static struct nfs4_stid *
2067 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2068 {
2069         struct nfs4_stid *s;
2070
2071         spin_lock(&cl->cl_lock);
2072         s = find_stateid_locked(cl, t);
2073         if (s != NULL) {
2074                 if (typemask & s->sc_type)
2075                         atomic_inc(&s->sc_count);
2076                 else
2077                         s = NULL;
2078         }
2079         spin_unlock(&cl->cl_lock);
2080         return s;
2081 }
2082
2083 static struct nfs4_client *create_client(struct xdr_netobj name,
2084                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2085 {
2086         struct nfs4_client *clp;
2087         struct sockaddr *sa = svc_addr(rqstp);
2088         int ret;
2089         struct net *net = SVC_NET(rqstp);
2090
2091         clp = alloc_client(name);
2092         if (clp == NULL)
2093                 return NULL;
2094
2095         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2096         if (ret) {
2097                 free_client(clp);
2098                 return NULL;
2099         }
2100         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2101         clp->cl_time = get_seconds();
2102         clear_bit(0, &clp->cl_cb_slot_busy);
2103         copy_verf(clp, verf);
2104         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2105         clp->cl_cb_session = NULL;
2106         clp->net = net;
2107         return clp;
2108 }
2109
2110 static void
2111 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2112 {
2113         struct rb_node **new = &(root->rb_node), *parent = NULL;
2114         struct nfs4_client *clp;
2115
2116         while (*new) {
2117                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2118                 parent = *new;
2119
2120                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2121                         new = &((*new)->rb_left);
2122                 else
2123                         new = &((*new)->rb_right);
2124         }
2125
2126         rb_link_node(&new_clp->cl_namenode, parent, new);
2127         rb_insert_color(&new_clp->cl_namenode, root);
2128 }
2129
2130 static struct nfs4_client *
2131 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2132 {
2133         int cmp;
2134         struct rb_node *node = root->rb_node;
2135         struct nfs4_client *clp;
2136
2137         while (node) {
2138                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2139                 cmp = compare_blob(&clp->cl_name, name);
2140                 if (cmp > 0)
2141                         node = node->rb_left;
2142                 else if (cmp < 0)
2143                         node = node->rb_right;
2144                 else
2145                         return clp;
2146         }
2147         return NULL;
2148 }
2149
2150 static void
2151 add_to_unconfirmed(struct nfs4_client *clp)
2152 {
2153         unsigned int idhashval;
2154         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2155
2156         lockdep_assert_held(&nn->client_lock);
2157
2158         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2159         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2160         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2161         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2162         renew_client_locked(clp);
2163 }
2164
2165 static void
2166 move_to_confirmed(struct nfs4_client *clp)
2167 {
2168         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2170
2171         lockdep_assert_held(&nn->client_lock);
2172
2173         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2174         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2175         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2176         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2177         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2178         renew_client_locked(clp);
2179 }
2180
2181 static struct nfs4_client *
2182 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2183 {
2184         struct nfs4_client *clp;
2185         unsigned int idhashval = clientid_hashval(clid->cl_id);
2186
2187         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2188                 if (same_clid(&clp->cl_clientid, clid)) {
2189                         if ((bool)clp->cl_minorversion != sessions)
2190                                 return NULL;
2191                         renew_client_locked(clp);
2192                         return clp;
2193                 }
2194         }
2195         return NULL;
2196 }
2197
2198 static struct nfs4_client *
2199 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2200 {
2201         struct list_head *tbl = nn->conf_id_hashtbl;
2202
2203         lockdep_assert_held(&nn->client_lock);
2204         return find_client_in_id_table(tbl, clid, sessions);
2205 }
2206
2207 static struct nfs4_client *
2208 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2209 {
2210         struct list_head *tbl = nn->unconf_id_hashtbl;
2211
2212         lockdep_assert_held(&nn->client_lock);
2213         return find_client_in_id_table(tbl, clid, sessions);
2214 }
2215
2216 static bool clp_used_exchangeid(struct nfs4_client *clp)
2217 {
2218         return clp->cl_exchange_flags != 0;
2219 }
2220
2221 static struct nfs4_client *
2222 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2223 {
2224         lockdep_assert_held(&nn->client_lock);
2225         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2226 }
2227
2228 static struct nfs4_client *
2229 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2230 {
2231         lockdep_assert_held(&nn->client_lock);
2232         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2233 }
2234
2235 static void
2236 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2237 {
2238         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2239         struct sockaddr *sa = svc_addr(rqstp);
2240         u32 scopeid = rpc_get_scope_id(sa);
2241         unsigned short expected_family;
2242
2243         /* Currently, we only support tcp and tcp6 for the callback channel */
2244         if (se->se_callback_netid_len == 3 &&
2245             !memcmp(se->se_callback_netid_val, "tcp", 3))
2246                 expected_family = AF_INET;
2247         else if (se->se_callback_netid_len == 4 &&
2248                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2249                 expected_family = AF_INET6;
2250         else
2251                 goto out_err;
2252
2253         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2254                                             se->se_callback_addr_len,
2255                                             (struct sockaddr *)&conn->cb_addr,
2256                                             sizeof(conn->cb_addr));
2257
2258         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2259                 goto out_err;
2260
2261         if (conn->cb_addr.ss_family == AF_INET6)
2262                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2263
2264         conn->cb_prog = se->se_callback_prog;
2265         conn->cb_ident = se->se_callback_ident;
2266         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2267         return;
2268 out_err:
2269         conn->cb_addr.ss_family = AF_UNSPEC;
2270         conn->cb_addrlen = 0;
2271         dprintk("NFSD: this client (clientid %08x/%08x) "
2272                 "will not receive delegations\n",
2273                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2274
2275         return;
2276 }
2277
2278 /*
2279  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2280  */
2281 static void
2282 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2283 {
2284         struct xdr_buf *buf = resp->xdr.buf;
2285         struct nfsd4_slot *slot = resp->cstate.slot;
2286         unsigned int base;
2287
2288         dprintk("--> %s slot %p\n", __func__, slot);
2289
2290         slot->sl_opcnt = resp->opcnt;
2291         slot->sl_status = resp->cstate.status;
2292
2293         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2294         if (nfsd4_not_cached(resp)) {
2295                 slot->sl_datalen = 0;
2296                 return;
2297         }
2298         base = resp->cstate.data_offset;
2299         slot->sl_datalen = buf->len - base;
2300         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2301                 WARN(1, "%s: sessions DRC could not cache compound\n",
2302                      __func__);
2303         return;
2304 }
2305
2306 /*
2307  * Encode the replay sequence operation from the slot values.
2308  * If cachethis is FALSE encode the uncached rep error on the next
2309  * operation which sets resp->p and increments resp->opcnt for
2310  * nfs4svc_encode_compoundres.
2311  *
2312  */
2313 static __be32
2314 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2315                           struct nfsd4_compoundres *resp)
2316 {
2317         struct nfsd4_op *op;
2318         struct nfsd4_slot *slot = resp->cstate.slot;
2319
2320         /* Encode the replayed sequence operation */
2321         op = &args->ops[resp->opcnt - 1];
2322         nfsd4_encode_operation(resp, op);
2323
2324         /* Return nfserr_retry_uncached_rep in next operation. */
2325         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2326                 op = &args->ops[resp->opcnt++];
2327                 op->status = nfserr_retry_uncached_rep;
2328                 nfsd4_encode_operation(resp, op);
2329         }
2330         return op->status;
2331 }
2332
2333 /*
2334  * The sequence operation is not cached because we can use the slot and
2335  * session values.
2336  */
2337 static __be32
2338 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2339                          struct nfsd4_sequence *seq)
2340 {
2341         struct nfsd4_slot *slot = resp->cstate.slot;
2342         struct xdr_stream *xdr = &resp->xdr;
2343         __be32 *p;
2344         __be32 status;
2345
2346         dprintk("--> %s slot %p\n", __func__, slot);
2347
2348         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2349         if (status)
2350                 return status;
2351
2352         p = xdr_reserve_space(xdr, slot->sl_datalen);
2353         if (!p) {
2354                 WARN_ON_ONCE(1);
2355                 return nfserr_serverfault;
2356         }
2357         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2358         xdr_commit_encode(xdr);
2359
2360         resp->opcnt = slot->sl_opcnt;
2361         return slot->sl_status;
2362 }
2363
2364 /*
2365  * Set the exchange_id flags returned by the server.
2366  */
2367 static void
2368 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2369 {
2370 #ifdef CONFIG_NFSD_PNFS
2371         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2372 #else
2373         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2374 #endif
2375
2376         /* Referrals are supported, Migration is not. */
2377         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2378
2379         /* set the wire flags to return to client. */
2380         clid->flags = new->cl_exchange_flags;
2381 }
2382
2383 static bool client_has_openowners(struct nfs4_client *clp)
2384 {
2385         struct nfs4_openowner *oo;
2386
2387         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2388                 if (!list_empty(&oo->oo_owner.so_stateids))
2389                         return true;
2390         }
2391         return false;
2392 }
2393
2394 static bool client_has_state(struct nfs4_client *clp)
2395 {
2396         return client_has_openowners(clp)
2397 #ifdef CONFIG_NFSD_PNFS
2398                 || !list_empty(&clp->cl_lo_states)
2399 #endif
2400                 || !list_empty(&clp->cl_delegations)
2401                 || !list_empty(&clp->cl_sessions);
2402 }
2403
2404 __be32
2405 nfsd4_exchange_id(struct svc_rqst *rqstp,
2406                   struct nfsd4_compound_state *cstate,
2407                   struct nfsd4_exchange_id *exid)
2408 {
2409         struct nfs4_client *conf, *new;
2410         struct nfs4_client *unconf = NULL;
2411         __be32 status;
2412         char                    addr_str[INET6_ADDRSTRLEN];
2413         nfs4_verifier           verf = exid->verifier;
2414         struct sockaddr         *sa = svc_addr(rqstp);
2415         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2416         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2417
2418         rpc_ntop(sa, addr_str, sizeof(addr_str));
2419         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2420                 "ip_addr=%s flags %x, spa_how %d\n",
2421                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2422                 addr_str, exid->flags, exid->spa_how);
2423
2424         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2425                 return nfserr_inval;
2426
2427         new = create_client(exid->clname, rqstp, &verf);
2428         if (new == NULL)
2429                 return nfserr_jukebox;
2430
2431         switch (exid->spa_how) {
2432         case SP4_MACH_CRED:
2433                 exid->spo_must_enforce[0] = 0;
2434                 exid->spo_must_enforce[1] = (
2435                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2436                         1 << (OP_EXCHANGE_ID - 32) |
2437                         1 << (OP_CREATE_SESSION - 32) |
2438                         1 << (OP_DESTROY_SESSION - 32) |
2439                         1 << (OP_DESTROY_CLIENTID - 32));
2440
2441                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2442                                         1 << (OP_OPEN_DOWNGRADE) |
2443                                         1 << (OP_LOCKU) |
2444                                         1 << (OP_DELEGRETURN));
2445
2446                 exid->spo_must_allow[1] &= (
2447                                         1 << (OP_TEST_STATEID - 32) |
2448                                         1 << (OP_FREE_STATEID - 32));
2449                 if (!svc_rqst_integrity_protected(rqstp)) {
2450                         status = nfserr_inval;
2451                         goto out_nolock;
2452                 }
2453                 /*
2454                  * Sometimes userspace doesn't give us a principal.
2455                  * Which is a bug, really.  Anyway, we can't enforce
2456                  * MACH_CRED in that case, better to give up now:
2457                  */
2458                 if (!new->cl_cred.cr_principal &&
2459                                         !new->cl_cred.cr_raw_principal) {
2460                         status = nfserr_serverfault;
2461                         goto out_nolock;
2462                 }
2463                 new->cl_mach_cred = true;
2464         case SP4_NONE:
2465                 break;
2466         default:                                /* checked by xdr code */
2467                 WARN_ON_ONCE(1);
2468         case SP4_SSV:
2469                 status = nfserr_encr_alg_unsupp;
2470                 goto out_nolock;
2471         }
2472
2473         /* Cases below refer to rfc 5661 section 18.35.4: */
2474         spin_lock(&nn->client_lock);
2475         conf = find_confirmed_client_by_name(&exid->clname, nn);
2476         if (conf) {
2477                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2478                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2479
2480                 if (update) {
2481                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2482                                 status = nfserr_inval;
2483                                 goto out;
2484                         }
2485                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2486                                 status = nfserr_wrong_cred;
2487                                 goto out;
2488                         }
2489                         if (!creds_match) { /* case 9 */
2490                                 status = nfserr_perm;
2491                                 goto out;
2492                         }
2493                         if (!verfs_match) { /* case 8 */
2494                                 status = nfserr_not_same;
2495                                 goto out;
2496                         }
2497                         /* case 6 */
2498                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2499                         goto out_copy;
2500                 }
2501                 if (!creds_match) { /* case 3 */
2502                         if (client_has_state(conf)) {
2503                                 status = nfserr_clid_inuse;
2504                                 goto out;
2505                         }
2506                         goto out_new;
2507                 }
2508                 if (verfs_match) { /* case 2 */
2509                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2510                         goto out_copy;
2511                 }
2512                 /* case 5, client reboot */
2513                 conf = NULL;
2514                 goto out_new;
2515         }
2516
2517         if (update) { /* case 7 */
2518                 status = nfserr_noent;
2519                 goto out;
2520         }
2521
2522         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2523         if (unconf) /* case 4, possible retry or client restart */
2524                 unhash_client_locked(unconf);
2525
2526         /* case 1 (normal case) */
2527 out_new:
2528         if (conf) {
2529                 status = mark_client_expired_locked(conf);
2530                 if (status)
2531                         goto out;
2532         }
2533         new->cl_minorversion = cstate->minorversion;
2534         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2535         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2536
2537         gen_clid(new, nn);
2538         add_to_unconfirmed(new);
2539         swap(new, conf);
2540 out_copy:
2541         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2542         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2543
2544         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2545         nfsd4_set_ex_flags(conf, exid);
2546
2547         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2548                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2549         status = nfs_ok;
2550
2551 out:
2552         spin_unlock(&nn->client_lock);
2553 out_nolock:
2554         if (new)
2555                 expire_client(new);
2556         if (unconf)
2557                 expire_client(unconf);
2558         return status;
2559 }
2560
2561 static __be32
2562 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2563 {
2564         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2565                 slot_seqid);
2566
2567         /* The slot is in use, and no response has been sent. */
2568         if (slot_inuse) {
2569                 if (seqid == slot_seqid)
2570                         return nfserr_jukebox;
2571                 else
2572                         return nfserr_seq_misordered;
2573         }
2574         /* Note unsigned 32-bit arithmetic handles wraparound: */
2575         if (likely(seqid == slot_seqid + 1))
2576                 return nfs_ok;
2577         if (seqid == slot_seqid)
2578                 return nfserr_replay_cache;
2579         return nfserr_seq_misordered;
2580 }
2581
2582 /*
2583  * Cache the create session result into the create session single DRC
2584  * slot cache by saving the xdr structure. sl_seqid has been set.
2585  * Do this for solo or embedded create session operations.
2586  */
2587 static void
2588 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2589                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2590 {
2591         slot->sl_status = nfserr;
2592         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2593 }
2594
2595 static __be32
2596 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2597                             struct nfsd4_clid_slot *slot)
2598 {
2599         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2600         return slot->sl_status;
2601 }
2602
2603 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2604                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2605                         1 +     /* MIN tag is length with zero, only length */ \
2606                         3 +     /* version, opcount, opcode */ \
2607                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2608                                 /* seqid, slotID, slotID, cache */ \
2609                         4 ) * sizeof(__be32))
2610
2611 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2612                         2 +     /* verifier: AUTH_NULL, length 0 */\
2613                         1 +     /* status */ \
2614                         1 +     /* MIN tag is length with zero, only length */ \
2615                         3 +     /* opcount, opcode, opstatus*/ \
2616                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2617                                 /* seqid, slotID, slotID, slotID, status */ \
2618                         5 ) * sizeof(__be32))
2619
2620 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2621 {
2622         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2623
2624         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2625                 return nfserr_toosmall;
2626         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2627                 return nfserr_toosmall;
2628         ca->headerpadsz = 0;
2629         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2630         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2631         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2632         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2633                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2634         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2635         /*
2636          * Note decreasing slot size below client's request may make it
2637          * difficult for client to function correctly, whereas
2638          * decreasing the number of slots will (just?) affect
2639          * performance.  When short on memory we therefore prefer to
2640          * decrease number of slots instead of their size.  Clients that
2641          * request larger slots than they need will get poor results:
2642          */
2643         ca->maxreqs = nfsd4_get_drc_mem(ca);
2644         if (!ca->maxreqs)
2645                 return nfserr_jukebox;
2646
2647         return nfs_ok;
2648 }
2649
2650 /*
2651  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2652  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2653  */
2654 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2655         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2656
2657 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2658         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2659
2660 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2661                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2662 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2663                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2664                                  sizeof(__be32))
2665
2666 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2667 {
2668         ca->headerpadsz = 0;
2669
2670         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2671                 return nfserr_toosmall;
2672         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2673                 return nfserr_toosmall;
2674         ca->maxresp_cached = 0;
2675         if (ca->maxops < 2)
2676                 return nfserr_toosmall;
2677
2678         return nfs_ok;
2679 }
2680
2681 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2682 {
2683         switch (cbs->flavor) {
2684         case RPC_AUTH_NULL:
2685         case RPC_AUTH_UNIX:
2686                 return nfs_ok;
2687         default:
2688                 /*
2689                  * GSS case: the spec doesn't allow us to return this
2690                  * error.  But it also doesn't allow us not to support
2691                  * GSS.
2692                  * I'd rather this fail hard than return some error the
2693                  * client might think it can already handle:
2694                  */
2695                 return nfserr_encr_alg_unsupp;
2696         }
2697 }
2698
2699 __be32
2700 nfsd4_create_session(struct svc_rqst *rqstp,
2701                      struct nfsd4_compound_state *cstate,
2702                      struct nfsd4_create_session *cr_ses)
2703 {
2704         struct sockaddr *sa = svc_addr(rqstp);
2705         struct nfs4_client *conf, *unconf;
2706         struct nfs4_client *old = NULL;
2707         struct nfsd4_session *new;
2708         struct nfsd4_conn *conn;
2709         struct nfsd4_clid_slot *cs_slot = NULL;
2710         __be32 status = 0;
2711         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2712
2713         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2714                 return nfserr_inval;
2715         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2716         if (status)
2717                 return status;
2718         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2719         if (status)
2720                 return status;
2721         status = check_backchannel_attrs(&cr_ses->back_channel);
2722         if (status)
2723                 goto out_release_drc_mem;
2724         status = nfserr_jukebox;
2725         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2726         if (!new)
2727                 goto out_release_drc_mem;
2728         conn = alloc_conn_from_crses(rqstp, cr_ses);
2729         if (!conn)
2730                 goto out_free_session;
2731
2732         spin_lock(&nn->client_lock);
2733         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2734         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2735         WARN_ON_ONCE(conf && unconf);
2736
2737         if (conf) {
2738                 status = nfserr_wrong_cred;
2739                 if (!nfsd4_mach_creds_match(conf, rqstp))
2740                         goto out_free_conn;
2741                 cs_slot = &conf->cl_cs_slot;
2742                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2743                 if (status) {
2744                         if (status == nfserr_replay_cache)
2745                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2746                         goto out_free_conn;
2747                 }
2748         } else if (unconf) {
2749                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2750                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2751                         status = nfserr_clid_inuse;
2752                         goto out_free_conn;
2753                 }
2754                 status = nfserr_wrong_cred;
2755                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2756                         goto out_free_conn;
2757                 cs_slot = &unconf->cl_cs_slot;
2758                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2759                 if (status) {
2760                         /* an unconfirmed replay returns misordered */
2761                         status = nfserr_seq_misordered;
2762                         goto out_free_conn;
2763                 }
2764                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2765                 if (old) {
2766                         status = mark_client_expired_locked(old);
2767                         if (status) {
2768                                 old = NULL;
2769                                 goto out_free_conn;
2770                         }
2771                 }
2772                 move_to_confirmed(unconf);
2773                 conf = unconf;
2774         } else {
2775                 status = nfserr_stale_clientid;
2776                 goto out_free_conn;
2777         }
2778         status = nfs_ok;
2779         /* Persistent sessions are not supported */
2780         cr_ses->flags &= ~SESSION4_PERSIST;
2781         /* Upshifting from TCP to RDMA is not supported */
2782         cr_ses->flags &= ~SESSION4_RDMA;
2783
2784         init_session(rqstp, new, conf, cr_ses);
2785         nfsd4_get_session_locked(new);
2786
2787         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2788                NFS4_MAX_SESSIONID_LEN);
2789         cs_slot->sl_seqid++;
2790         cr_ses->seqid = cs_slot->sl_seqid;
2791
2792         /* cache solo and embedded create sessions under the client_lock */
2793         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2794         spin_unlock(&nn->client_lock);
2795         /* init connection and backchannel */
2796         nfsd4_init_conn(rqstp, conn, new);
2797         nfsd4_put_session(new);
2798         if (old)
2799                 expire_client(old);
2800         return status;
2801 out_free_conn:
2802         spin_unlock(&nn->client_lock);
2803         free_conn(conn);
2804         if (old)
2805                 expire_client(old);
2806 out_free_session:
2807         __free_session(new);
2808 out_release_drc_mem:
2809         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2810         return status;
2811 }
2812
2813 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2814 {
2815         switch (*dir) {
2816         case NFS4_CDFC4_FORE:
2817         case NFS4_CDFC4_BACK:
2818                 return nfs_ok;
2819         case NFS4_CDFC4_FORE_OR_BOTH:
2820         case NFS4_CDFC4_BACK_OR_BOTH:
2821                 *dir = NFS4_CDFC4_BOTH;
2822                 return nfs_ok;
2823         };
2824         return nfserr_inval;
2825 }
2826
2827 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2828 {
2829         struct nfsd4_session *session = cstate->session;
2830         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2831         __be32 status;
2832
2833         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2834         if (status)
2835                 return status;
2836         spin_lock(&nn->client_lock);
2837         session->se_cb_prog = bc->bc_cb_program;
2838         session->se_cb_sec = bc->bc_cb_sec;
2839         spin_unlock(&nn->client_lock);
2840
2841         nfsd4_probe_callback(session->se_client);
2842
2843         return nfs_ok;
2844 }
2845
2846 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2847                      struct nfsd4_compound_state *cstate,
2848                      struct nfsd4_bind_conn_to_session *bcts)
2849 {
2850         __be32 status;
2851         struct nfsd4_conn *conn;
2852         struct nfsd4_session *session;
2853         struct net *net = SVC_NET(rqstp);
2854         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2855
2856         if (!nfsd4_last_compound_op(rqstp))
2857                 return nfserr_not_only_op;
2858         spin_lock(&nn->client_lock);
2859         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2860         spin_unlock(&nn->client_lock);
2861         if (!session)
2862                 goto out_no_session;
2863         status = nfserr_wrong_cred;
2864         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2865                 goto out;
2866         status = nfsd4_map_bcts_dir(&bcts->dir);
2867         if (status)
2868                 goto out;
2869         conn = alloc_conn(rqstp, bcts->dir);
2870         status = nfserr_jukebox;
2871         if (!conn)
2872                 goto out;
2873         nfsd4_init_conn(rqstp, conn, session);
2874         status = nfs_ok;
2875 out:
2876         nfsd4_put_session(session);
2877 out_no_session:
2878         return status;
2879 }
2880
2881 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2882 {
2883         if (!session)
2884                 return 0;
2885         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2886 }
2887
2888 __be32
2889 nfsd4_destroy_session(struct svc_rqst *r,
2890                       struct nfsd4_compound_state *cstate,
2891                       struct nfsd4_destroy_session *sessionid)
2892 {
2893         struct nfsd4_session *ses;
2894         __be32 status;
2895         int ref_held_by_me = 0;
2896         struct net *net = SVC_NET(r);
2897         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2898
2899         status = nfserr_not_only_op;
2900         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2901                 if (!nfsd4_last_compound_op(r))
2902                         goto out;
2903                 ref_held_by_me++;
2904         }
2905         dump_sessionid(__func__, &sessionid->sessionid);
2906         spin_lock(&nn->client_lock);
2907         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2908         if (!ses)
2909                 goto out_client_lock;
2910         status = nfserr_wrong_cred;
2911         if (!nfsd4_mach_creds_match(ses->se_client, r))
2912                 goto out_put_session;
2913         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2914         if (status)
2915                 goto out_put_session;
2916         unhash_session(ses);
2917         spin_unlock(&nn->client_lock);
2918
2919         nfsd4_probe_callback_sync(ses->se_client);
2920
2921         spin_lock(&nn->client_lock);
2922         status = nfs_ok;
2923 out_put_session:
2924         nfsd4_put_session_locked(ses);
2925 out_client_lock:
2926         spin_unlock(&nn->client_lock);
2927 out:
2928         return status;
2929 }
2930
2931 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2932 {
2933         struct nfsd4_conn *c;
2934
2935         list_for_each_entry(c, &s->se_conns, cn_persession) {
2936                 if (c->cn_xprt == xpt) {
2937                         return c;
2938                 }
2939         }
2940         return NULL;
2941 }
2942
2943 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2944 {
2945         struct nfs4_client *clp = ses->se_client;
2946         struct nfsd4_conn *c;
2947         __be32 status = nfs_ok;
2948         int ret;
2949
2950         spin_lock(&clp->cl_lock);
2951         c = __nfsd4_find_conn(new->cn_xprt, ses);
2952         if (c)
2953                 goto out_free;
2954         status = nfserr_conn_not_bound_to_session;
2955         if (clp->cl_mach_cred)
2956                 goto out_free;
2957         __nfsd4_hash_conn(new, ses);
2958         spin_unlock(&clp->cl_lock);
2959         ret = nfsd4_register_conn(new);
2960         if (ret)
2961                 /* oops; xprt is already down: */
2962                 nfsd4_conn_lost(&new->cn_xpt_user);
2963         return nfs_ok;
2964 out_free:
2965         spin_unlock(&clp->cl_lock);
2966         free_conn(new);
2967         return status;
2968 }
2969
2970 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2971 {
2972         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2973
2974         return args->opcnt > session->se_fchannel.maxops;
2975 }
2976
2977 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2978                                   struct nfsd4_session *session)
2979 {
2980         struct xdr_buf *xb = &rqstp->rq_arg;
2981
2982         return xb->len > session->se_fchannel.maxreq_sz;
2983 }
2984
2985 __be32
2986 nfsd4_sequence(struct svc_rqst *rqstp,
2987                struct nfsd4_compound_state *cstate,
2988                struct nfsd4_sequence *seq)
2989 {
2990         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2991         struct xdr_stream *xdr = &resp->xdr;
2992         struct nfsd4_session *session;
2993         struct nfs4_client *clp;
2994         struct nfsd4_slot *slot;
2995         struct nfsd4_conn *conn;
2996         __be32 status;
2997         int buflen;
2998         struct net *net = SVC_NET(rqstp);
2999         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3000
3001         if (resp->opcnt != 1)
3002                 return nfserr_sequence_pos;
3003
3004         /*
3005          * Will be either used or freed by nfsd4_sequence_check_conn
3006          * below.
3007          */
3008         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3009         if (!conn)
3010                 return nfserr_jukebox;
3011
3012         spin_lock(&nn->client_lock);
3013         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3014         if (!session)
3015                 goto out_no_session;
3016         clp = session->se_client;
3017
3018         status = nfserr_too_many_ops;
3019         if (nfsd4_session_too_many_ops(rqstp, session))
3020                 goto out_put_session;
3021
3022         status = nfserr_req_too_big;
3023         if (nfsd4_request_too_big(rqstp, session))
3024                 goto out_put_session;
3025
3026         status = nfserr_badslot;
3027         if (seq->slotid >= session->se_fchannel.maxreqs)
3028                 goto out_put_session;
3029
3030         slot = session->se_slots[seq->slotid];
3031         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3032
3033         /* We do not negotiate the number of slots yet, so set the
3034          * maxslots to the session maxreqs which is used to encode
3035          * sr_highest_slotid and the sr_target_slot id to maxslots */
3036         seq->maxslots = session->se_fchannel.maxreqs;
3037
3038         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3039                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3040         if (status == nfserr_replay_cache) {
3041                 status = nfserr_seq_misordered;
3042                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3043                         goto out_put_session;
3044                 cstate->slot = slot;
3045                 cstate->session = session;
3046                 cstate->clp = clp;
3047                 /* Return the cached reply status and set cstate->status
3048                  * for nfsd4_proc_compound processing */
3049                 status = nfsd4_replay_cache_entry(resp, seq);
3050                 cstate->status = nfserr_replay_cache;
3051                 goto out;
3052         }
3053         if (status)
3054                 goto out_put_session;
3055
3056         status = nfsd4_sequence_check_conn(conn, session);
3057         conn = NULL;
3058         if (status)
3059                 goto out_put_session;
3060
3061         buflen = (seq->cachethis) ?
3062                         session->se_fchannel.maxresp_cached :
3063                         session->se_fchannel.maxresp_sz;
3064         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3065                                     nfserr_rep_too_big;
3066         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3067                 goto out_put_session;
3068         svc_reserve(rqstp, buflen);
3069
3070         status = nfs_ok;
3071         /* Success! bump slot seqid */
3072         slot->sl_seqid = seq->seqid;
3073         slot->sl_flags |= NFSD4_SLOT_INUSE;
3074         if (seq->cachethis)
3075                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3076         else
3077                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3078
3079         cstate->slot = slot;
3080         cstate->session = session;
3081         cstate->clp = clp;
3082
3083 out:
3084         switch (clp->cl_cb_state) {
3085         case NFSD4_CB_DOWN:
3086                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3087                 break;
3088         case NFSD4_CB_FAULT:
3089                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3090                 break;
3091         default:
3092                 seq->status_flags = 0;
3093         }
3094         if (!list_empty(&clp->cl_revoked))
3095                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3096 out_no_session:
3097         if (conn)
3098                 free_conn(conn);
3099         spin_unlock(&nn->client_lock);
3100         return status;
3101 out_put_session:
3102         nfsd4_put_session_locked(session);
3103         goto out_no_session;
3104 }
3105
3106 void
3107 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3108 {
3109         struct nfsd4_compound_state *cs = &resp->cstate;
3110
3111         if (nfsd4_has_session(cs)) {
3112                 if (cs->status != nfserr_replay_cache) {
3113                         nfsd4_store_cache_entry(resp);
3114                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3115                 }
3116                 /* Drop session reference that was taken in nfsd4_sequence() */
3117                 nfsd4_put_session(cs->session);
3118         } else if (cs->clp)
3119                 put_client_renew(cs->clp);
3120 }
3121
3122 __be32
3123 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3124 {
3125         struct nfs4_client *conf, *unconf;
3126         struct nfs4_client *clp = NULL;
3127         __be32 status = 0;
3128         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3129
3130         spin_lock(&nn->client_lock);
3131         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3132         conf = find_confirmed_client(&dc->clientid, true, nn);
3133         WARN_ON_ONCE(conf && unconf);
3134
3135         if (conf) {
3136                 if (client_has_state(conf)) {
3137                         status = nfserr_clientid_busy;
3138                         goto out;
3139                 }
3140                 status = mark_client_expired_locked(conf);
3141                 if (status)
3142                         goto out;
3143                 clp = conf;
3144         } else if (unconf)
3145                 clp = unconf;
3146         else {
3147                 status = nfserr_stale_clientid;
3148                 goto out;
3149         }
3150         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3151                 clp = NULL;
3152                 status = nfserr_wrong_cred;
3153                 goto out;
3154         }
3155         unhash_client_locked(clp);
3156 out:
3157         spin_unlock(&nn->client_lock);
3158         if (clp)
3159                 expire_client(clp);
3160         return status;
3161 }
3162
3163 __be32
3164 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3165 {
3166         __be32 status = 0;
3167
3168         if (rc->rca_one_fs) {
3169                 if (!cstate->current_fh.fh_dentry)
3170                         return nfserr_nofilehandle;
3171                 /*
3172                  * We don't take advantage of the rca_one_fs case.
3173                  * That's OK, it's optional, we can safely ignore it.
3174                  */
3175                 return nfs_ok;
3176         }
3177
3178         status = nfserr_complete_already;
3179         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3180                              &cstate->session->se_client->cl_flags))
3181                 goto out;
3182
3183         status = nfserr_stale_clientid;
3184         if (is_client_expired(cstate->session->se_client))
3185                 /*
3186                  * The following error isn't really legal.
3187                  * But we only get here if the client just explicitly
3188                  * destroyed the client.  Surely it no longer cares what
3189                  * error it gets back on an operation for the dead
3190                  * client.
3191                  */
3192                 goto out;
3193
3194         status = nfs_ok;
3195         nfsd4_client_record_create(cstate->session->se_client);
3196 out:
3197         return status;
3198 }
3199
3200 __be32
3201 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3202                   struct nfsd4_setclientid *setclid)
3203 {
3204         struct xdr_netobj       clname = setclid->se_name;
3205         nfs4_verifier           clverifier = setclid->se_verf;
3206         struct nfs4_client      *conf, *new;
3207         struct nfs4_client      *unconf = NULL;
3208         __be32                  status;
3209         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3210
3211         new = create_client(clname, rqstp, &clverifier);
3212         if (new == NULL)
3213                 return nfserr_jukebox;
3214         /* Cases below refer to rfc 3530 section 14.2.33: */
3215         spin_lock(&nn->client_lock);
3216         conf = find_confirmed_client_by_name(&clname, nn);
3217         if (conf && client_has_state(conf)) {
3218                 /* case 0: */
3219                 status = nfserr_clid_inuse;
3220                 if (clp_used_exchangeid(conf))
3221                         goto out;
3222                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3223                         char addr_str[INET6_ADDRSTRLEN];
3224                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3225                                  sizeof(addr_str));
3226                         dprintk("NFSD: setclientid: string in use by client "
3227                                 "at %s\n", addr_str);
3228                         goto out;
3229                 }
3230         }
3231         unconf = find_unconfirmed_client_by_name(&clname, nn);
3232         if (unconf)
3233                 unhash_client_locked(unconf);
3234         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3235                 /* case 1: probable callback update */
3236                 copy_clid(new, conf);
3237                 gen_confirm(new, nn);
3238         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3239                 gen_clid(new, nn);
3240         new->cl_minorversion = 0;
3241         gen_callback(new, setclid, rqstp);
3242         add_to_unconfirmed(new);
3243         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3244         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3245         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3246         new = NULL;
3247         status = nfs_ok;
3248 out:
3249         spin_unlock(&nn->client_lock);
3250         if (new)
3251                 free_client(new);
3252         if (unconf)
3253                 expire_client(unconf);
3254         return status;
3255 }
3256
3257
3258 __be32
3259 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3260                          struct nfsd4_compound_state *cstate,
3261                          struct nfsd4_setclientid_confirm *setclientid_confirm)
3262 {
3263         struct nfs4_client *conf, *unconf;
3264         struct nfs4_client *old = NULL;
3265         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3266         clientid_t * clid = &setclientid_confirm->sc_clientid;
3267         __be32 status;
3268         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3269
3270         if (STALE_CLIENTID(clid, nn))
3271                 return nfserr_stale_clientid;
3272
3273         spin_lock(&nn->client_lock);
3274         conf = find_confirmed_client(clid, false, nn);
3275         unconf = find_unconfirmed_client(clid, false, nn);
3276         /*
3277          * We try hard to give out unique clientid's, so if we get an
3278          * attempt to confirm the same clientid with a different cred,
3279          * the client may be buggy; this should never happen.
3280          *
3281          * Nevertheless, RFC 7530 recommends INUSE for this case:
3282          */
3283         status = nfserr_clid_inuse;
3284         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3285                 goto out;
3286         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3287                 goto out;
3288         /* cases below refer to rfc 3530 section 14.2.34: */
3289         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3290                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3291                         /* case 2: probable retransmit */
3292                         status = nfs_ok;
3293                 } else /* case 4: client hasn't noticed we rebooted yet? */
3294                         status = nfserr_stale_clientid;
3295                 goto out;
3296         }
3297         status = nfs_ok;
3298         if (conf) { /* case 1: callback update */
3299                 old = unconf;
3300                 unhash_client_locked(old);
3301                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3302         } else { /* case 3: normal case; new or rebooted client */
3303                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3304                 if (old) {
3305                         status = nfserr_clid_inuse;
3306                         if (client_has_state(old)
3307                                         && !same_creds(&unconf->cl_cred,
3308                                                         &old->cl_cred))
3309                                 goto out;
3310                         status = mark_client_expired_locked(old);
3311                         if (status) {
3312                                 old = NULL;
3313                                 goto out;
3314                         }
3315                 }
3316                 move_to_confirmed(unconf);
3317                 conf = unconf;
3318         }
3319         get_client_locked(conf);
3320         spin_unlock(&nn->client_lock);
3321         nfsd4_probe_callback(conf);
3322         spin_lock(&nn->client_lock);
3323         put_client_renew_locked(conf);
3324 out:
3325         spin_unlock(&nn->client_lock);
3326         if (old)
3327                 expire_client(old);
3328         return status;
3329 }
3330
3331 static struct nfs4_file *nfsd4_alloc_file(void)
3332 {
3333         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3334 }
3335
3336 /* OPEN Share state helper functions */
3337 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3338                                 struct nfs4_file *fp)
3339 {
3340         lockdep_assert_held(&state_lock);
3341
3342         atomic_set(&fp->fi_ref, 1);
3343         spin_lock_init(&fp->fi_lock);
3344         INIT_LIST_HEAD(&fp->fi_stateids);
3345         INIT_LIST_HEAD(&fp->fi_delegations);
3346         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3347         fh_copy_shallow(&fp->fi_fhandle, fh);
3348         fp->fi_deleg_file = NULL;
3349         fp->fi_had_conflict = false;
3350         fp->fi_share_deny = 0;
3351         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3352         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3353 #ifdef CONFIG_NFSD_PNFS
3354         INIT_LIST_HEAD(&fp->fi_lo_states);
3355         atomic_set(&fp->fi_lo_recalls, 0);
3356 #endif
3357         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3358 }
3359
3360 void
3361 nfsd4_free_slabs(void)
3362 {
3363         kmem_cache_destroy(odstate_slab);
3364         kmem_cache_destroy(openowner_slab);
3365         kmem_cache_destroy(lockowner_slab);
3366         kmem_cache_destroy(file_slab);
3367         kmem_cache_destroy(stateid_slab);
3368         kmem_cache_destroy(deleg_slab);
3369 }
3370
3371 int
3372 nfsd4_init_slabs(void)
3373 {
3374         openowner_slab = kmem_cache_create("nfsd4_openowners",
3375                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3376         if (openowner_slab == NULL)
3377                 goto out;
3378         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3379                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3380         if (lockowner_slab == NULL)
3381                 goto out_free_openowner_slab;
3382         file_slab = kmem_cache_create("nfsd4_files",
3383                         sizeof(struct nfs4_file), 0, 0, NULL);
3384         if (file_slab == NULL)
3385                 goto out_free_lockowner_slab;
3386         stateid_slab = kmem_cache_create("nfsd4_stateids",
3387                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3388         if (stateid_slab == NULL)
3389                 goto out_free_file_slab;
3390         deleg_slab = kmem_cache_create("nfsd4_delegations",
3391                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3392         if (deleg_slab == NULL)
3393                 goto out_free_stateid_slab;
3394         odstate_slab = kmem_cache_create("nfsd4_odstate",
3395                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3396         if (odstate_slab == NULL)
3397                 goto out_free_deleg_slab;
3398         return 0;
3399
3400 out_free_deleg_slab:
3401         kmem_cache_destroy(deleg_slab);
3402 out_free_stateid_slab:
3403         kmem_cache_destroy(stateid_slab);
3404 out_free_file_slab:
3405         kmem_cache_destroy(file_slab);
3406 out_free_lockowner_slab:
3407         kmem_cache_destroy(lockowner_slab);
3408 out_free_openowner_slab:
3409         kmem_cache_destroy(openowner_slab);
3410 out:
3411         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3412         return -ENOMEM;
3413 }
3414
3415 static void init_nfs4_replay(struct nfs4_replay *rp)
3416 {
3417         rp->rp_status = nfserr_serverfault;
3418         rp->rp_buflen = 0;
3419         rp->rp_buf = rp->rp_ibuf;
3420         mutex_init(&rp->rp_mutex);
3421 }
3422
3423 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3424                 struct nfs4_stateowner *so)
3425 {
3426         if (!nfsd4_has_session(cstate)) {
3427                 mutex_lock(&so->so_replay.rp_mutex);
3428                 cstate->replay_owner = nfs4_get_stateowner(so);
3429         }
3430 }
3431
3432 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3433 {
3434         struct nfs4_stateowner *so = cstate->replay_owner;
3435
3436         if (so != NULL) {
3437                 cstate->replay_owner = NULL;
3438                 mutex_unlock(&so->so_replay.rp_mutex);
3439                 nfs4_put_stateowner(so);
3440         }
3441 }
3442
3443 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3444 {
3445         struct nfs4_stateowner *sop;
3446
3447         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3448         if (!sop)
3449                 return NULL;
3450
3451         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3452         if (!sop->so_owner.data) {
3453                 kmem_cache_free(slab, sop);
3454                 return NULL;
3455         }
3456         sop->so_owner.len = owner->len;
3457
3458         INIT_LIST_HEAD(&sop->so_stateids);
3459         sop->so_client = clp;
3460         init_nfs4_replay(&sop->so_replay);
3461         atomic_set(&sop->so_count, 1);
3462         return sop;
3463 }
3464
3465 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3466 {
3467         lockdep_assert_held(&clp->cl_lock);
3468
3469         list_add(&oo->oo_owner.so_strhash,
3470                  &clp->cl_ownerstr_hashtbl[strhashval]);
3471         list_add(&oo->oo_perclient, &clp->cl_openowners);
3472 }
3473
3474 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3475 {
3476         unhash_openowner_locked(openowner(so));
3477 }
3478
3479 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3480 {
3481         struct nfs4_openowner *oo = openowner(so);
3482
3483         kmem_cache_free(openowner_slab, oo);
3484 }
3485
3486 static const struct nfs4_stateowner_operations openowner_ops = {
3487         .so_unhash =    nfs4_unhash_openowner,
3488         .so_free =      nfs4_free_openowner,
3489 };
3490
3491 static struct nfs4_ol_stateid *
3492 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3493 {
3494         struct nfs4_ol_stateid *local, *ret = NULL;
3495         struct nfs4_openowner *oo = open->op_openowner;
3496
3497         lockdep_assert_held(&fp->fi_lock);
3498
3499         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3500                 /* ignore lock owners */
3501                 if (local->st_stateowner->so_is_open_owner == 0)
3502                         continue;
3503                 if (local->st_stateowner == &oo->oo_owner) {
3504                         ret = local;
3505                         atomic_inc(&ret->st_stid.sc_count);
3506                         break;
3507                 }
3508         }
3509         return ret;
3510 }
3511
3512 static struct nfs4_openowner *
3513 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3514                            struct nfsd4_compound_state *cstate)
3515 {
3516         struct nfs4_client *clp = cstate->clp;
3517         struct nfs4_openowner *oo, *ret;
3518
3519         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3520         if (!oo)
3521                 return NULL;
3522         oo->oo_owner.so_ops = &openowner_ops;
3523         oo->oo_owner.so_is_open_owner = 1;
3524         oo->oo_owner.so_seqid = open->op_seqid;
3525         oo->oo_flags = 0;
3526         if (nfsd4_has_session(cstate))
3527                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3528         oo->oo_time = 0;
3529         oo->oo_last_closed_stid = NULL;
3530         INIT_LIST_HEAD(&oo->oo_close_lru);
3531         spin_lock(&clp->cl_lock);
3532         ret = find_openstateowner_str_locked(strhashval, open, clp);
3533         if (ret == NULL) {
3534                 hash_openowner(oo, clp, strhashval);
3535                 ret = oo;
3536         } else
3537                 nfs4_free_stateowner(&oo->oo_owner);
3538
3539         spin_unlock(&clp->cl_lock);
3540         return ret;
3541 }
3542
3543 static struct nfs4_ol_stateid *
3544 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3545 {
3546
3547         struct nfs4_openowner *oo = open->op_openowner;
3548         struct nfs4_ol_stateid *retstp = NULL;
3549         struct nfs4_ol_stateid *stp;
3550
3551         stp = open->op_stp;
3552         /* We are moving these outside of the spinlocks to avoid the warnings */
3553         mutex_init(&stp->st_mutex);
3554         mutex_lock(&stp->st_mutex);
3555
3556         spin_lock(&oo->oo_owner.so_client->cl_lock);
3557         spin_lock(&fp->fi_lock);
3558
3559         retstp = nfsd4_find_existing_open(fp, open);
3560         if (retstp)
3561                 goto out_unlock;
3562
3563         open->op_stp = NULL;
3564         atomic_inc(&stp->st_stid.sc_count);
3565         stp->st_stid.sc_type = NFS4_OPEN_STID;
3566         INIT_LIST_HEAD(&stp->st_locks);
3567         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3568         get_nfs4_file(fp);
3569         stp->st_stid.sc_file = fp;
3570         stp->st_access_bmap = 0;
3571         stp->st_deny_bmap = 0;
3572         stp->st_openstp = NULL;
3573         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3574         list_add(&stp->st_perfile, &fp->fi_stateids);
3575
3576 out_unlock:
3577         spin_unlock(&fp->fi_lock);
3578         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3579         if (retstp) {
3580                 mutex_lock(&retstp->st_mutex);
3581                 /* To keep mutex tracking happy */
3582                 mutex_unlock(&stp->st_mutex);
3583                 stp = retstp;
3584         }
3585         return stp;
3586 }
3587
3588 /*
3589  * In the 4.0 case we need to keep the owners around a little while to handle
3590  * CLOSE replay. We still do need to release any file access that is held by
3591  * them before returning however.
3592  */
3593 static void
3594 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3595 {
3596         struct nfs4_ol_stateid *last;
3597         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3598         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3599                                                 nfsd_net_id);
3600
3601         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3602
3603         /*
3604          * We know that we hold one reference via nfsd4_close, and another
3605          * "persistent" reference for the client. If the refcount is higher
3606          * than 2, then there are still calls in progress that are using this
3607          * stateid. We can't put the sc_file reference until they are finished.
3608          * Wait for the refcount to drop to 2. Since it has been unhashed,
3609          * there should be no danger of the refcount going back up again at
3610          * this point.
3611          */
3612         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3613
3614         release_all_access(s);
3615         if (s->st_stid.sc_file) {
3616                 put_nfs4_file(s->st_stid.sc_file);
3617                 s->st_stid.sc_file = NULL;
3618         }
3619
3620         spin_lock(&nn->client_lock);
3621         last = oo->oo_last_closed_stid;
3622         oo->oo_last_closed_stid = s;
3623         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3624         oo->oo_time = get_seconds();
3625         spin_unlock(&nn->client_lock);
3626         if (last)
3627                 nfs4_put_stid(&last->st_stid);
3628 }
3629
3630 /* search file_hashtbl[] for file */
3631 static struct nfs4_file *
3632 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3633 {
3634         struct nfs4_file *fp;
3635
3636         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3637                 if (fh_match(&fp->fi_fhandle, fh)) {
3638                         if (atomic_inc_not_zero(&fp->fi_ref))
3639                                 return fp;
3640                 }
3641         }
3642         return NULL;
3643 }
3644
3645 struct nfs4_file *
3646 find_file(struct knfsd_fh *fh)
3647 {
3648         struct nfs4_file *fp;
3649         unsigned int hashval = file_hashval(fh);
3650
3651         rcu_read_lock();
3652         fp = find_file_locked(fh, hashval);
3653         rcu_read_unlock();
3654         return fp;
3655 }
3656
3657 static struct nfs4_file *
3658 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3659 {
3660         struct nfs4_file *fp;
3661         unsigned int hashval = file_hashval(fh);
3662
3663         rcu_read_lock();
3664         fp = find_file_locked(fh, hashval);
3665         rcu_read_unlock();
3666         if (fp)
3667                 return fp;
3668
3669         spin_lock(&state_lock);
3670         fp = find_file_locked(fh, hashval);
3671         if (likely(fp == NULL)) {
3672                 nfsd4_init_file(fh, hashval, new);
3673                 fp = new;
3674         }
3675         spin_unlock(&state_lock);
3676
3677         return fp;
3678 }
3679
3680 /*
3681  * Called to check deny when READ with all zero stateid or
3682  * WRITE with all zero or all one stateid
3683  */
3684 static __be32
3685 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3686 {
3687         struct nfs4_file *fp;
3688         __be32 ret = nfs_ok;
3689
3690         fp = find_file(&current_fh->fh_handle);
3691         if (!fp)
3692                 return ret;
3693         /* Check for conflicting share reservations */
3694         spin_lock(&fp->fi_lock);
3695         if (fp->fi_share_deny & deny_type)
3696                 ret = nfserr_locked;
3697         spin_unlock(&fp->fi_lock);
3698         put_nfs4_file(fp);
3699         return ret;
3700 }
3701
3702 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3703 {
3704         struct nfs4_delegation *dp = cb_to_delegation(cb);
3705         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3706                                           nfsd_net_id);
3707
3708         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3709
3710         /*
3711          * We can't do this in nfsd_break_deleg_cb because it is
3712          * already holding inode->i_lock.
3713          *
3714          * If the dl_time != 0, then we know that it has already been
3715          * queued for a lease break. Don't queue it again.
3716          */
3717         spin_lock(&state_lock);
3718         if (dp->dl_time == 0) {
3719                 dp->dl_time = get_seconds();
3720                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3721         }
3722         spin_unlock(&state_lock);
3723 }
3724
3725 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3726                 struct rpc_task *task)
3727 {
3728         struct nfs4_delegation *dp = cb_to_delegation(cb);
3729
3730         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3731                 return 1;
3732
3733         switch (task->tk_status) {
3734         case 0:
3735                 return 1;
3736         case -EBADHANDLE:
3737         case -NFS4ERR_BAD_STATEID:
3738                 /*
3739                  * Race: client probably got cb_recall before open reply
3740                  * granting delegation.
3741                  */
3742                 if (dp->dl_retries--) {
3743                         rpc_delay(task, 2 * HZ);
3744                         return 0;
3745                 }
3746                 /*FALLTHRU*/
3747         default:
3748                 return -1;
3749         }
3750 }
3751
3752 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3753 {
3754         struct nfs4_delegation *dp = cb_to_delegation(cb);
3755
3756         nfs4_put_stid(&dp->dl_stid);
3757 }
3758
3759 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3760         .prepare        = nfsd4_cb_recall_prepare,
3761         .done           = nfsd4_cb_recall_done,
3762         .release        = nfsd4_cb_recall_release,
3763 };
3764
3765 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3766 {
3767         /*
3768          * We're assuming the state code never drops its reference
3769          * without first removing the lease.  Since we're in this lease
3770          * callback (and since the lease code is serialized by the kernel
3771          * lock) we know the server hasn't removed the lease yet, we know
3772          * it's safe to take a reference.
3773          */
3774         atomic_inc(&dp->dl_stid.sc_count);
3775         nfsd4_run_cb(&dp->dl_recall);
3776 }
3777
3778 /* Called from break_lease() with i_lock held. */
3779 static bool
3780 nfsd_break_deleg_cb(struct file_lock *fl)
3781 {
3782         bool ret = false;
3783         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3784         struct nfs4_delegation *dp;
3785
3786         if (!fp) {
3787                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3788                 return ret;
3789         }
3790         if (fp->fi_had_conflict) {
3791                 WARN(1, "duplicate break on %p\n", fp);
3792                 return ret;
3793         }
3794         /*
3795          * We don't want the locks code to timeout the lease for us;
3796          * we'll remove it ourself if a delegation isn't returned
3797          * in time:
3798          */
3799         fl->fl_break_time = 0;
3800
3801         spin_lock(&fp->fi_lock);
3802         fp->fi_had_conflict = true;
3803         /*
3804          * If there are no delegations on the list, then return true
3805          * so that the lease code will go ahead and delete it.
3806          */
3807         if (list_empty(&fp->fi_delegations))
3808                 ret = true;
3809         else
3810                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3811                         nfsd_break_one_deleg(dp);
3812         spin_unlock(&fp->fi_lock);
3813         return ret;
3814 }
3815
3816 static int
3817 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3818                      struct list_head *dispose)
3819 {
3820         if (arg & F_UNLCK)
3821                 return lease_modify(onlist, arg, dispose);
3822         else
3823                 return -EAGAIN;
3824 }
3825
3826 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3827         .lm_break = nfsd_break_deleg_cb,
3828         .lm_change = nfsd_change_deleg_cb,
3829 };
3830
3831 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3832 {
3833         if (nfsd4_has_session(cstate))
3834                 return nfs_ok;
3835         if (seqid == so->so_seqid - 1)
3836                 return nfserr_replay_me;
3837         if (seqid == so->so_seqid)
3838                 return nfs_ok;
3839         return nfserr_bad_seqid;
3840 }
3841
3842 static __be32 lookup_clientid(clientid_t *clid,
3843                 struct nfsd4_compound_state *cstate,
3844                 struct nfsd_net *nn)
3845 {
3846         struct nfs4_client *found;
3847
3848         if (cstate->clp) {
3849                 found = cstate->clp;
3850                 if (!same_clid(&found->cl_clientid, clid))
3851                         return nfserr_stale_clientid;
3852                 return nfs_ok;
3853         }
3854
3855         if (STALE_CLIENTID(clid, nn))
3856                 return nfserr_stale_clientid;
3857
3858         /*
3859          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3860          * cached already then we know this is for is for v4.0 and "sessions"
3861          * will be false.
3862          */
3863         WARN_ON_ONCE(cstate->session);
3864         spin_lock(&nn->client_lock);
3865         found = find_confirmed_client(clid, false, nn);
3866         if (!found) {
3867                 spin_unlock(&nn->client_lock);
3868                 return nfserr_expired;
3869         }
3870         atomic_inc(&found->cl_refcount);
3871         spin_unlock(&nn->client_lock);
3872
3873         /* Cache the nfs4_client in cstate! */
3874         cstate->clp = found;
3875         return nfs_ok;
3876 }
3877
3878 __be32
3879 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3880                     struct nfsd4_open *open, struct nfsd_net *nn)
3881 {
3882         clientid_t *clientid = &open->op_clientid;
3883         struct nfs4_client *clp = NULL;
3884         unsigned int strhashval;
3885         struct nfs4_openowner *oo = NULL;
3886         __be32 status;
3887
3888         if (STALE_CLIENTID(&open->op_clientid, nn))
3889                 return nfserr_stale_clientid;
3890         /*
3891          * In case we need it later, after we've already created the
3892          * file and don't want to risk a further failure:
3893          */
3894         open->op_file = nfsd4_alloc_file();
3895         if (open->op_file == NULL)
3896                 return nfserr_jukebox;
3897
3898         status = lookup_clientid(clientid, cstate, nn);
3899         if (status)
3900                 return status;
3901         clp = cstate->clp;
3902
3903         strhashval = ownerstr_hashval(&open->op_owner);
3904         oo = find_openstateowner_str(strhashval, open, clp);
3905         open->op_openowner = oo;
3906         if (!oo) {
3907                 goto new_owner;
3908         }
3909         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3910                 /* Replace unconfirmed owners without checking for replay. */
3911                 release_openowner(oo);
3912                 open->op_openowner = NULL;
3913                 goto new_owner;
3914         }
3915         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3916         if (status)
3917                 return status;
3918         goto alloc_stateid;
3919 new_owner:
3920         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3921         if (oo == NULL)
3922                 return nfserr_jukebox;
3923         open->op_openowner = oo;
3924 alloc_stateid:
3925         open->op_stp = nfs4_alloc_open_stateid(clp);
3926         if (!open->op_stp)
3927                 return nfserr_jukebox;
3928
3929         if (nfsd4_has_session(cstate) &&
3930             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3931                 open->op_odstate = alloc_clnt_odstate(clp);
3932                 if (!open->op_odstate)
3933                         return nfserr_jukebox;
3934         }
3935
3936         return nfs_ok;
3937 }
3938
3939 static inline __be32
3940 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3941 {
3942         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3943                 return nfserr_openmode;
3944         else
3945                 return nfs_ok;
3946 }
3947
3948 static int share_access_to_flags(u32 share_access)
3949 {
3950         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3951 }
3952
3953 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3954 {
3955         struct nfs4_stid *ret;
3956
3957         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3958         if (!ret)
3959                 return NULL;
3960         return delegstateid(ret);
3961 }
3962
3963 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3964 {
3965         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3966                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3967 }
3968
3969 static __be32
3970 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3971                 struct nfs4_delegation **dp)
3972 {
3973         int flags;
3974         __be32 status = nfserr_bad_stateid;
3975         struct nfs4_delegation *deleg;
3976
3977         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3978         if (deleg == NULL)
3979                 goto out;
3980         flags = share_access_to_flags(open->op_share_access);
3981         status = nfs4_check_delegmode(deleg, flags);
3982         if (status) {
3983                 nfs4_put_stid(&deleg->dl_stid);
3984                 goto out;
3985         }
3986         *dp = deleg;
3987 out:
3988         if (!nfsd4_is_deleg_cur(open))
3989                 return nfs_ok;
3990         if (status)
3991                 return status;
3992         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3993         return nfs_ok;
3994 }
3995
3996 static inline int nfs4_access_to_access(u32 nfs4_access)
3997 {
3998         int flags = 0;
3999
4000         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4001                 flags |= NFSD_MAY_READ;
4002         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4003                 flags |= NFSD_MAY_WRITE;
4004         return flags;
4005 }
4006
4007 static inline __be32
4008 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4009                 struct nfsd4_open *open)
4010 {
4011         struct iattr iattr = {
4012                 .ia_valid = ATTR_SIZE,
4013                 .ia_size = 0,
4014         };
4015         if (!open->op_truncate)
4016                 return 0;
4017         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4018                 return nfserr_inval;
4019         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4020 }
4021
4022 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4023                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4024                 struct nfsd4_open *open)
4025 {
4026         struct file *filp = NULL;
4027         __be32 status;
4028         int oflag = nfs4_access_to_omode(open->op_share_access);
4029         int access = nfs4_access_to_access(open->op_share_access);
4030         unsigned char old_access_bmap, old_deny_bmap;
4031
4032         spin_lock(&fp->fi_lock);
4033
4034         /*
4035          * Are we trying to set a deny mode that would conflict with
4036          * current access?
4037          */
4038         status = nfs4_file_check_deny(fp, open->op_share_deny);
4039         if (status != nfs_ok) {
4040                 spin_unlock(&fp->fi_lock);
4041                 goto out;
4042         }
4043
4044         /* set access to the file */
4045         status = nfs4_file_get_access(fp, open->op_share_access);
4046         if (status != nfs_ok) {
4047                 spin_unlock(&fp->fi_lock);
4048                 goto out;
4049         }
4050
4051         /* Set access bits in stateid */
4052         old_access_bmap = stp->st_access_bmap;
4053         set_access(open->op_share_access, stp);
4054
4055         /* Set new deny mask */
4056         old_deny_bmap = stp->st_deny_bmap;
4057         set_deny(open->op_share_deny, stp);
4058         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4059
4060         if (!fp->fi_fds[oflag]) {
4061                 spin_unlock(&fp->fi_lock);
4062                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4063                 if (status)
4064                         goto out_put_access;
4065                 spin_lock(&fp->fi_lock);
4066                 if (!fp->fi_fds[oflag]) {
4067                         fp->fi_fds[oflag] = filp;
4068                         filp = NULL;
4069                 }
4070         }
4071         spin_unlock(&fp->fi_lock);
4072         if (filp)
4073                 fput(filp);
4074
4075         status = nfsd4_truncate(rqstp, cur_fh, open);
4076         if (status)
4077                 goto out_put_access;
4078 out:
4079         return status;
4080 out_put_access:
4081         stp->st_access_bmap = old_access_bmap;
4082         nfs4_file_put_access(fp, open->op_share_access);
4083         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4084         goto out;
4085 }
4086
4087 static __be32
4088 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)
4089 {
4090         __be32 status;
4091         unsigned char old_deny_bmap = stp->st_deny_bmap;
4092
4093         if (!test_access(open->op_share_access, stp))
4094                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4095
4096         /* test and set deny mode */
4097         spin_lock(&fp->fi_lock);
4098         status = nfs4_file_check_deny(fp, open->op_share_deny);
4099         if (status == nfs_ok) {
4100                 set_deny(open->op_share_deny, stp);
4101                 fp->fi_share_deny |=
4102                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4103         }
4104         spin_unlock(&fp->fi_lock);
4105
4106         if (status != nfs_ok)
4107                 return status;
4108
4109         status = nfsd4_truncate(rqstp, cur_fh, open);
4110         if (status != nfs_ok)
4111                 reset_union_bmap_deny(old_deny_bmap, stp);
4112         return status;
4113 }
4114
4115 /* Should we give out recallable state?: */
4116 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4117 {
4118         if (clp->cl_cb_state == NFSD4_CB_UP)
4119                 return true;
4120         /*
4121          * In the sessions case, since we don't have to establish a
4122          * separate connection for callbacks, we assume it's OK
4123          * until we hear otherwise:
4124          */
4125         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4126 }
4127
4128 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4129 {
4130         struct file_lock *fl;
4131
4132         fl = locks_alloc_lock();
4133         if (!fl)
4134                 return NULL;
4135         fl->fl_lmops = &nfsd_lease_mng_ops;
4136         fl->fl_flags = FL_DELEG;
4137         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4138         fl->fl_end = OFFSET_MAX;
4139         fl->fl_owner = (fl_owner_t)fp;
4140         fl->fl_pid = current->tgid;
4141         return fl;
4142 }
4143
4144 /**
4145  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4146  * @dp:   a pointer to the nfs4_delegation we're adding.
4147  *
4148  * Return:
4149  *      On success: Return code will be 0 on success.
4150  *
4151  *      On error: -EAGAIN if there was an existing delegation.
4152  *                 nonzero if there is an error in other cases.
4153  *
4154  */
4155
4156 static int nfs4_setlease(struct nfs4_delegation *dp)
4157 {
4158         struct nfs4_file *fp = dp->dl_stid.sc_file;
4159         struct file_lock *fl;
4160         struct file *filp;
4161         int status = 0;
4162
4163         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4164         if (!fl)
4165                 return -ENOMEM;
4166         filp = find_readable_file(fp);
4167         if (!filp) {
4168                 /* We should always have a readable file here */
4169                 WARN_ON_ONCE(1);
4170                 locks_free_lock(fl);
4171                 return -EBADF;
4172         }
4173         fl->fl_file = filp;
4174         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4175         if (fl)
4176                 locks_free_lock(fl);
4177         if (status)
4178                 goto out_fput;
4179         spin_lock(&state_lock);
4180         spin_lock(&fp->fi_lock);
4181         /* Did the lease get broken before we took the lock? */
4182         status = -EAGAIN;
4183         if (fp->fi_had_conflict)
4184                 goto out_unlock;
4185         /* Race breaker */
4186         if (fp->fi_deleg_file) {
4187                 status = hash_delegation_locked(dp, fp);
4188                 goto out_unlock;
4189         }
4190         fp->fi_deleg_file = filp;
4191         fp->fi_delegees = 0;
4192         status = hash_delegation_locked(dp, fp);
4193         spin_unlock(&fp->fi_lock);
4194         spin_unlock(&state_lock);
4195         if (status) {
4196                 /* Should never happen, this is a new fi_deleg_file  */
4197                 WARN_ON_ONCE(1);
4198                 goto out_fput;
4199         }
4200         return 0;
4201 out_unlock:
4202         spin_unlock(&fp->fi_lock);
4203         spin_unlock(&state_lock);
4204 out_fput:
4205         fput(filp);
4206         return status;
4207 }
4208
4209 static struct nfs4_delegation *
4210 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4211                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4212 {
4213         int status;
4214         struct nfs4_delegation *dp;
4215
4216         if (fp->fi_had_conflict)
4217                 return ERR_PTR(-EAGAIN);
4218
4219         spin_lock(&state_lock);
4220         spin_lock(&fp->fi_lock);
4221         status = nfs4_get_existing_delegation(clp, fp);
4222         spin_unlock(&fp->fi_lock);
4223         spin_unlock(&state_lock);
4224
4225         if (status)
4226                 return ERR_PTR(status);
4227
4228         dp = alloc_init_deleg(clp, fh, odstate);
4229         if (!dp)
4230                 return ERR_PTR(-ENOMEM);
4231
4232         get_nfs4_file(fp);
4233         spin_lock(&state_lock);
4234         spin_lock(&fp->fi_lock);
4235         dp->dl_stid.sc_file = fp;
4236         if (!fp->fi_deleg_file) {
4237                 spin_unlock(&fp->fi_lock);
4238                 spin_unlock(&state_lock);
4239                 status = nfs4_setlease(dp);
4240                 goto out;
4241         }
4242         if (fp->fi_had_conflict) {
4243                 status = -EAGAIN;
4244                 goto out_unlock;
4245         }
4246         status = hash_delegation_locked(dp, fp);
4247 out_unlock:
4248         spin_unlock(&fp->fi_lock);
4249         spin_unlock(&state_lock);
4250 out:
4251         if (status) {
4252                 put_clnt_odstate(dp->dl_clnt_odstate);
4253                 nfs4_put_stid(&dp->dl_stid);
4254                 return ERR_PTR(status);
4255         }
4256         return dp;
4257 }
4258
4259 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4260 {
4261         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4262         if (status == -EAGAIN)
4263                 open->op_why_no_deleg = WND4_CONTENTION;
4264         else {
4265                 open->op_why_no_deleg = WND4_RESOURCE;
4266                 switch (open->op_deleg_want) {
4267                 case NFS4_SHARE_WANT_READ_DELEG:
4268                 case NFS4_SHARE_WANT_WRITE_DELEG:
4269                 case NFS4_SHARE_WANT_ANY_DELEG:
4270                         break;
4271                 case NFS4_SHARE_WANT_CANCEL:
4272                         open->op_why_no_deleg = WND4_CANCELLED;
4273                         break;
4274                 case NFS4_SHARE_WANT_NO_DELEG:
4275                         WARN_ON_ONCE(1);
4276                 }
4277         }
4278 }
4279
4280 /*
4281  * Attempt to hand out a delegation.
4282  *
4283  * Note we don't support write delegations, and won't until the vfs has
4284  * proper support for them.
4285  */
4286 static void
4287 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4288                         struct nfs4_ol_stateid *stp)
4289 {
4290         struct nfs4_delegation *dp;
4291         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4292         struct nfs4_client *clp = stp->st_stid.sc_client;
4293         int cb_up;
4294         int status = 0;
4295
4296         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4297         open->op_recall = 0;
4298         switch (open->op_claim_type) {
4299                 case NFS4_OPEN_CLAIM_PREVIOUS:
4300                         if (!cb_up)
4301                                 open->op_recall = 1;
4302                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4303                                 goto out_no_deleg;
4304                         break;
4305                 case NFS4_OPEN_CLAIM_NULL:
4306                 case NFS4_OPEN_CLAIM_FH:
4307                         /*
4308                          * Let's not give out any delegations till everyone's
4309                          * had the chance to reclaim theirs, *and* until
4310                          * NLM locks have all been reclaimed:
4311                          */
4312                         if (locks_in_grace(clp->net))
4313                                 goto out_no_deleg;
4314                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4315                                 goto out_no_deleg;
4316                         /*
4317                          * Also, if the file was opened for write or
4318                          * create, there's a good chance the client's
4319                          * about to write to it, resulting in an
4320                          * immediate recall (since we don't support
4321                          * write delegations):
4322                          */
4323                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4324                                 goto out_no_deleg;
4325                         if (open->op_create == NFS4_OPEN_CREATE)
4326                                 goto out_no_deleg;
4327                         break;
4328                 default:
4329                         goto out_no_deleg;
4330         }
4331         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4332         if (IS_ERR(dp))
4333                 goto out_no_deleg;
4334
4335         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4336
4337         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4338                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4339         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4340         nfs4_put_stid(&dp->dl_stid);
4341         return;
4342 out_no_deleg:
4343         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4344         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4345             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4346                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4347                 open->op_recall = 1;
4348         }
4349
4350         /* 4.1 client asking for a delegation? */
4351         if (open->op_deleg_want)
4352                 nfsd4_open_deleg_none_ext(open, status);
4353         return;
4354 }
4355
4356 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4357                                         struct nfs4_delegation *dp)
4358 {
4359         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4360             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4361                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4362                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4363         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4364                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4365                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4366                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4367         }
4368         /* Otherwise the client must be confused wanting a delegation
4369          * it already has, therefore we don't return
4370          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4371          */
4372 }
4373
4374 __be32
4375 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4376 {
4377         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4378         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4379         struct nfs4_file *fp = NULL;
4380         struct nfs4_ol_stateid *stp = NULL;
4381         struct nfs4_delegation *dp = NULL;
4382         __be32 status;
4383
4384         /*
4385          * Lookup file; if found, lookup stateid and check open request,
4386          * and check for delegations in the process of being recalled.
4387          * If not found, create the nfs4_file struct
4388          */
4389         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4390         if (fp != open->op_file) {
4391                 status = nfs4_check_deleg(cl, open, &dp);
4392                 if (status)
4393                         goto out;
4394                 spin_lock(&fp->fi_lock);
4395                 stp = nfsd4_find_existing_open(fp, open);
4396                 spin_unlock(&fp->fi_lock);
4397         } else {
4398                 open->op_file = NULL;
4399                 status = nfserr_bad_stateid;
4400                 if (nfsd4_is_deleg_cur(open))
4401                         goto out;
4402         }
4403
4404         /*
4405          * OPEN the file, or upgrade an existing OPEN.
4406          * If truncate fails, the OPEN fails.
4407          */
4408         if (stp) {
4409                 /* Stateid was found, this is an OPEN upgrade */
4410                 mutex_lock(&stp->st_mutex);
4411                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4412                 if (status) {
4413                         mutex_unlock(&stp->st_mutex);
4414                         goto out;
4415                 }
4416         } else {
4417                 /* stp is returned locked. */
4418                 stp = init_open_stateid(fp, open);
4419                 /* See if we lost the race to some other thread */
4420                 if (stp->st_access_bmap != 0) {
4421                         status = nfs4_upgrade_open(rqstp, fp, current_fh,
4422                                                 stp, open);
4423                         if (status) {
4424                                 mutex_unlock(&stp->st_mutex);
4425                                 goto out;
4426                         }
4427                         goto upgrade_out;
4428                 }
4429                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4430                 if (status) {
4431                         mutex_unlock(&stp->st_mutex);
4432                         release_open_stateid(stp);
4433                         goto out;
4434                 }
4435
4436                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4437                                                         open->op_odstate);
4438                 if (stp->st_clnt_odstate == open->op_odstate)
4439                         open->op_odstate = NULL;
4440         }
4441 upgrade_out:
4442         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4443         mutex_unlock(&stp->st_mutex);
4444
4445         if (nfsd4_has_session(&resp->cstate)) {
4446                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4447                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4448                         open->op_why_no_deleg = WND4_NOT_WANTED;
4449                         goto nodeleg;
4450                 }
4451         }
4452
4453         /*
4454         * Attempt to hand out a delegation. No error return, because the
4455         * OPEN succeeds even if we fail.
4456         */
4457         nfs4_open_delegation(current_fh, open, stp);
4458 nodeleg:
4459         status = nfs_ok;
4460
4461         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4462                 STATEID_VAL(&stp->st_stid.sc_stateid));
4463 out:
4464         /* 4.1 client trying to upgrade/downgrade delegation? */
4465         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4466             open->op_deleg_want)
4467                 nfsd4_deleg_xgrade_none_ext(open, dp);
4468
4469         if (fp)
4470                 put_nfs4_file(fp);
4471         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4472                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4473         /*
4474         * To finish the open response, we just need to set the rflags.
4475         */
4476         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4477         if (nfsd4_has_session(&resp->cstate))
4478                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4479         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4480                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4481
4482         if (dp)
4483                 nfs4_put_stid(&dp->dl_stid);
4484         if (stp)
4485                 nfs4_put_stid(&stp->st_stid);
4486
4487         return status;
4488 }
4489
4490 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4491                               struct nfsd4_open *open)
4492 {
4493         if (open->op_openowner) {
4494                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4495
4496                 nfsd4_cstate_assign_replay(cstate, so);
4497                 nfs4_put_stateowner(so);
4498         }
4499         if (open->op_file)
4500                 kmem_cache_free(file_slab, open->op_file);
4501         if (open->op_stp)
4502                 nfs4_put_stid(&open->op_stp->st_stid);
4503         if (open->op_odstate)
4504                 kmem_cache_free(odstate_slab, open->op_odstate);
4505 }
4506
4507 __be32
4508 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4509             clientid_t *clid)
4510 {
4511         struct nfs4_client *clp;
4512         __be32 status;
4513         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4514
4515         dprintk("process_renew(%08x/%08x): starting\n",
4516                         clid->cl_boot, clid->cl_id);
4517         status = lookup_clientid(clid, cstate, nn);
4518         if (status)
4519                 goto out;
4520         clp = cstate->clp;
4521         status = nfserr_cb_path_down;
4522         if (!list_empty(&clp->cl_delegations)
4523                         && clp->cl_cb_state != NFSD4_CB_UP)
4524                 goto out;
4525         status = nfs_ok;
4526 out:
4527         return status;
4528 }
4529
4530 void
4531 nfsd4_end_grace(struct nfsd_net *nn)
4532 {
4533         /* do nothing if grace period already ended */
4534         if (nn->grace_ended)
4535                 return;
4536
4537         dprintk("NFSD: end of grace period\n");
4538         nn->grace_ended = true;
4539         /*
4540          * If the server goes down again right now, an NFSv4
4541          * client will still be allowed to reclaim after it comes back up,
4542          * even if it hasn't yet had a chance to reclaim state this time.
4543          *
4544          */
4545         nfsd4_record_grace_done(nn);
4546         /*
4547          * At this point, NFSv4 clients can still reclaim.  But if the
4548          * server crashes, any that have not yet reclaimed will be out
4549          * of luck on the next boot.
4550          *
4551          * (NFSv4.1+ clients are considered to have reclaimed once they
4552          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4553          * have reclaimed after their first OPEN.)
4554          */
4555         locks_end_grace(&nn->nfsd4_manager);
4556         /*
4557          * At this point, and once lockd and/or any other containers
4558          * exit their grace period, further reclaims will fail and
4559          * regular locking can resume.
4560          */
4561 }
4562
4563 static time_t
4564 nfs4_laundromat(struct nfsd_net *nn)
4565 {
4566         struct nfs4_client *clp;
4567         struct nfs4_openowner *oo;
4568         struct nfs4_delegation *dp;
4569         struct nfs4_ol_stateid *stp;
4570         struct nfsd4_blocked_lock *nbl;
4571         struct list_head *pos, *next, reaplist;
4572         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4573         time_t t, new_timeo = nn->nfsd4_lease;
4574
4575         dprintk("NFSD: laundromat service - starting\n");
4576         nfsd4_end_grace(nn);
4577         INIT_LIST_HEAD(&reaplist);
4578         spin_lock(&nn->client_lock);
4579         list_for_each_safe(pos, next, &nn->client_lru) {
4580                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4581                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4582                         t = clp->cl_time - cutoff;
4583                         new_timeo = min(new_timeo, t);
4584                         break;
4585                 }
4586                 if (mark_client_expired_locked(clp)) {
4587                         dprintk("NFSD: client in use (clientid %08x)\n",
4588                                 clp->cl_clientid.cl_id);
4589                         continue;
4590                 }
4591                 list_add(&clp->cl_lru, &reaplist);
4592         }
4593         spin_unlock(&nn->client_lock);
4594         list_for_each_safe(pos, next, &reaplist) {
4595                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4596                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4597                         clp->cl_clientid.cl_id);
4598                 list_del_init(&clp->cl_lru);
4599                 expire_client(clp);
4600         }
4601         spin_lock(&state_lock);
4602         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4603                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4604                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4605                         t = dp->dl_time - cutoff;
4606                         new_timeo = min(new_timeo, t);
4607                         break;
4608                 }
4609                 WARN_ON(!unhash_delegation_locked(dp));
4610                 list_add(&dp->dl_recall_lru, &reaplist);
4611         }
4612         spin_unlock(&state_lock);
4613         while (!list_empty(&reaplist)) {
4614                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4615                                         dl_recall_lru);
4616                 list_del_init(&dp->dl_recall_lru);
4617                 revoke_delegation(dp);
4618         }
4619
4620         spin_lock(&nn->client_lock);
4621         while (!list_empty(&nn->close_lru)) {
4622                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4623                                         oo_close_lru);
4624                 if (time_after((unsigned long)oo->oo_time,
4625                                (unsigned long)cutoff)) {
4626                         t = oo->oo_time - cutoff;
4627                         new_timeo = min(new_timeo, t);
4628                         break;
4629                 }
4630                 list_del_init(&oo->oo_close_lru);
4631                 stp = oo->oo_last_closed_stid;
4632                 oo->oo_last_closed_stid = NULL;
4633                 spin_unlock(&nn->client_lock);
4634                 nfs4_put_stid(&stp->st_stid);
4635                 spin_lock(&nn->client_lock);
4636         }
4637         spin_unlock(&nn->client_lock);
4638
4639         /*
4640          * It's possible for a client to try and acquire an already held lock
4641          * that is being held for a long time, and then lose interest in it.
4642          * So, we clean out any un-revisited request after a lease period
4643          * under the assumption that the client is no longer interested.
4644          *
4645          * RFC5661, sec. 9.6 states that the client must not rely on getting
4646          * notifications and must continue to poll for locks, even when the
4647          * server supports them. Thus this shouldn't lead to clients blocking
4648          * indefinitely once the lock does become free.
4649          */
4650         BUG_ON(!list_empty(&reaplist));
4651         spin_lock(&nn->blocked_locks_lock);
4652         while (!list_empty(&nn->blocked_locks_lru)) {
4653                 nbl = list_first_entry(&nn->blocked_locks_lru,
4654                                         struct nfsd4_blocked_lock, nbl_lru);
4655                 if (time_after((unsigned long)nbl->nbl_time,
4656                                (unsigned long)cutoff)) {
4657                         t = nbl->nbl_time - cutoff;
4658                         new_timeo = min(new_timeo, t);
4659                         break;
4660                 }
4661                 list_move(&nbl->nbl_lru, &reaplist);
4662                 list_del_init(&nbl->nbl_list);
4663         }
4664         spin_unlock(&nn->blocked_locks_lock);
4665
4666         while (!list_empty(&reaplist)) {
4667                 nbl = list_first_entry(&nn->blocked_locks_lru,
4668                                         struct nfsd4_blocked_lock, nbl_lru);
4669                 list_del_init(&nbl->nbl_lru);
4670                 posix_unblock_lock(&nbl->nbl_lock);
4671                 free_blocked_lock(nbl);
4672         }
4673
4674         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4675         return new_timeo;
4676 }
4677
4678 static struct workqueue_struct *laundry_wq;
4679 static void laundromat_main(struct work_struct *);
4680
4681 static void
4682 laundromat_main(struct work_struct *laundry)
4683 {
4684         time_t t;
4685         struct delayed_work *dwork = to_delayed_work(laundry);
4686         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4687                                            laundromat_work);
4688
4689         t = nfs4_laundromat(nn);
4690         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4691         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4692 }
4693
4694 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4695 {
4696         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4697                 return nfserr_bad_stateid;
4698         return nfs_ok;
4699 }
4700
4701 static inline int
4702 access_permit_read(struct nfs4_ol_stateid *stp)
4703 {
4704         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4705                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4706                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4707 }
4708
4709 static inline int
4710 access_permit_write(struct nfs4_ol_stateid *stp)
4711 {
4712         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4713                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4714 }
4715
4716 static
4717 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4718 {
4719         __be32 status = nfserr_openmode;
4720
4721         /* For lock stateid's, we test the parent open, not the lock: */
4722         if (stp->st_openstp)
4723                 stp = stp->st_openstp;
4724         if ((flags & WR_STATE) && !access_permit_write(stp))
4725                 goto out;
4726         if ((flags & RD_STATE) && !access_permit_read(stp))
4727                 goto out;
4728         status = nfs_ok;
4729 out:
4730         return status;
4731 }
4732
4733 static inline __be32
4734 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4735 {
4736         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4737                 return nfs_ok;
4738         else if (opens_in_grace(net)) {
4739                 /* Answer in remaining cases depends on existence of
4740                  * conflicting state; so we must wait out the grace period. */
4741                 return nfserr_grace;
4742         } else if (flags & WR_STATE)
4743                 return nfs4_share_conflict(current_fh,
4744                                 NFS4_SHARE_DENY_WRITE);
4745         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4746                 return nfs4_share_conflict(current_fh,
4747                                 NFS4_SHARE_DENY_READ);
4748 }
4749
4750 /*
4751  * Allow READ/WRITE during grace period on recovered state only for files
4752  * that are not able to provide mandatory locking.
4753  */
4754 static inline int
4755 grace_disallows_io(struct net *net, struct inode *inode)
4756 {
4757         return opens_in_grace(net) && mandatory_lock(inode);
4758 }
4759
4760 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4761 {
4762         /*
4763          * When sessions are used the stateid generation number is ignored
4764          * when it is zero.
4765          */
4766         if (has_session && in->si_generation == 0)
4767                 return nfs_ok;
4768
4769         if (in->si_generation == ref->si_generation)
4770                 return nfs_ok;
4771
4772         /* If the client sends us a stateid from the future, it's buggy: */
4773         if (nfsd4_stateid_generation_after(in, ref))
4774                 return nfserr_bad_stateid;
4775         /*
4776          * However, we could see a stateid from the past, even from a
4777          * non-buggy client.  For example, if the client sends a lock
4778          * while some IO is outstanding, the lock may bump si_generation
4779          * while the IO is still in flight.  The client could avoid that
4780          * situation by waiting for responses on all the IO requests,
4781          * but better performance may result in retrying IO that
4782          * receives an old_stateid error if requests are rarely
4783          * reordered in flight:
4784          */
4785         return nfserr_old_stateid;
4786 }
4787
4788 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4789 {
4790         if (ols->st_stateowner->so_is_open_owner &&
4791             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4792                 return nfserr_bad_stateid;
4793         return nfs_ok;
4794 }
4795
4796 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4797 {
4798         struct nfs4_stid *s;
4799         __be32 status = nfserr_bad_stateid;
4800
4801         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4802                 return status;
4803         /* Client debugging aid. */
4804         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4805                 char addr_str[INET6_ADDRSTRLEN];
4806                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4807                                  sizeof(addr_str));
4808                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4809                                         "with incorrect client ID\n", addr_str);
4810                 return status;
4811         }
4812         spin_lock(&cl->cl_lock);
4813         s = find_stateid_locked(cl, stateid);
4814         if (!s)
4815                 goto out_unlock;
4816         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4817         if (status)
4818                 goto out_unlock;
4819         switch (s->sc_type) {
4820         case NFS4_DELEG_STID:
4821                 status = nfs_ok;
4822                 break;
4823         case NFS4_REVOKED_DELEG_STID:
4824                 status = nfserr_deleg_revoked;
4825                 break;
4826         case NFS4_OPEN_STID:
4827         case NFS4_LOCK_STID:
4828                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4829                 break;
4830         default:
4831                 printk("unknown stateid type %x\n", s->sc_type);
4832                 /* Fallthrough */
4833         case NFS4_CLOSED_STID:
4834         case NFS4_CLOSED_DELEG_STID:
4835                 status = nfserr_bad_stateid;
4836         }
4837 out_unlock:
4838         spin_unlock(&cl->cl_lock);
4839         return status;
4840 }
4841
4842 __be32
4843 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4844                      stateid_t *stateid, unsigned char typemask,
4845                      struct nfs4_stid **s, struct nfsd_net *nn)
4846 {
4847         __be32 status;
4848
4849         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4850                 return nfserr_bad_stateid;
4851         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4852         if (status == nfserr_stale_clientid) {
4853                 if (cstate->session)
4854                         return nfserr_bad_stateid;
4855                 return nfserr_stale_stateid;
4856         }
4857         if (status)
4858                 return status;
4859         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4860         if (!*s)
4861                 return nfserr_bad_stateid;
4862         return nfs_ok;
4863 }
4864
4865 static struct file *
4866 nfs4_find_file(struct nfs4_stid *s, int flags)
4867 {
4868         if (!s)
4869                 return NULL;
4870
4871         switch (s->sc_type) {
4872         case NFS4_DELEG_STID:
4873                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4874                         return NULL;
4875                 return get_file(s->sc_file->fi_deleg_file);
4876         case NFS4_OPEN_STID:
4877         case NFS4_LOCK_STID:
4878                 if (flags & RD_STATE)
4879                         return find_readable_file(s->sc_file);
4880                 else
4881                         return find_writeable_file(s->sc_file);
4882                 break;
4883         }
4884
4885         return NULL;
4886 }
4887
4888 static __be32
4889 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4890 {
4891         __be32 status;
4892
4893         status = nfsd4_check_openowner_confirmed(ols);
4894         if (status)
4895                 return status;
4896         return nfs4_check_openmode(ols, flags);
4897 }
4898
4899 static __be32
4900 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4901                 struct file **filpp, bool *tmp_file, int flags)
4902 {
4903         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4904         struct file *file;
4905         __be32 status;
4906
4907         file = nfs4_find_file(s, flags);
4908         if (file) {
4909                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4910                                 acc | NFSD_MAY_OWNER_OVERRIDE);
4911                 if (status) {
4912                         fput(file);
4913                         return status;
4914                 }
4915
4916                 *filpp = file;
4917         } else {
4918                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4919                 if (status)
4920                         return status;
4921
4922                 if (tmp_file)
4923                         *tmp_file = true;
4924         }
4925
4926         return 0;
4927 }
4928
4929 /*
4930  * Checks for stateid operations
4931  */
4932 __be32
4933 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4934                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
4935                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
4936 {
4937         struct inode *ino = d_inode(fhp->fh_dentry);
4938         struct net *net = SVC_NET(rqstp);
4939         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4940         struct nfs4_stid *s = NULL;
4941         __be32 status;
4942
4943         if (filpp)
4944                 *filpp = NULL;
4945         if (tmp_file)
4946                 *tmp_file = false;
4947
4948         if (grace_disallows_io(net, ino))
4949                 return nfserr_grace;
4950
4951         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4952                 status = check_special_stateids(net, fhp, stateid, flags);
4953                 goto done;
4954         }
4955
4956         status = nfsd4_lookup_stateid(cstate, stateid,
4957                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4958                                 &s, nn);
4959         if (status)
4960                 return status;
4961         status = check_stateid_generation(stateid, &s->sc_stateid,
4962                         nfsd4_has_session(cstate));
4963         if (status)
4964                 goto out;
4965
4966         switch (s->sc_type) {
4967         case NFS4_DELEG_STID:
4968                 status = nfs4_check_delegmode(delegstateid(s), flags);
4969                 break;
4970         case NFS4_OPEN_STID:
4971         case NFS4_LOCK_STID:
4972                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4973                 break;
4974         default:
4975                 status = nfserr_bad_stateid;
4976                 break;
4977         }
4978         if (status)
4979                 goto out;
4980         status = nfs4_check_fh(fhp, s);
4981
4982 done:
4983         if (!status && filpp)
4984                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
4985 out:
4986         if (s)
4987                 nfs4_put_stid(s);
4988         return status;
4989 }
4990
4991 /*
4992  * Test if the stateid is valid
4993  */
4994 __be32
4995 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4996                    struct nfsd4_test_stateid *test_stateid)
4997 {
4998         struct nfsd4_test_stateid_id *stateid;
4999         struct nfs4_client *cl = cstate->session->se_client;
5000
5001         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5002                 stateid->ts_id_status =
5003                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5004
5005         return nfs_ok;
5006 }
5007
5008 static __be32
5009 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5010 {
5011         struct nfs4_ol_stateid *stp = openlockstateid(s);
5012         __be32 ret;
5013
5014         mutex_lock(&stp->st_mutex);
5015
5016         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5017         if (ret)
5018                 goto out;
5019
5020         ret = nfserr_locks_held;
5021         if (check_for_locks(stp->st_stid.sc_file,
5022                             lockowner(stp->st_stateowner)))
5023                 goto out;
5024
5025         release_lock_stateid(stp);
5026         ret = nfs_ok;
5027
5028 out:
5029         mutex_unlock(&stp->st_mutex);
5030         nfs4_put_stid(s);
5031         return ret;
5032 }
5033
5034 __be32
5035 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5036                    struct nfsd4_free_stateid *free_stateid)
5037 {
5038         stateid_t *stateid = &free_stateid->fr_stateid;
5039         struct nfs4_stid *s;
5040         struct nfs4_delegation *dp;
5041         struct nfs4_client *cl = cstate->session->se_client;
5042         __be32 ret = nfserr_bad_stateid;
5043
5044         spin_lock(&cl->cl_lock);
5045         s = find_stateid_locked(cl, stateid);
5046         if (!s)
5047                 goto out_unlock;
5048         switch (s->sc_type) {
5049         case NFS4_DELEG_STID:
5050                 ret = nfserr_locks_held;
5051                 break;
5052         case NFS4_OPEN_STID:
5053                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5054                 if (ret)
5055                         break;
5056                 ret = nfserr_locks_held;
5057                 break;
5058         case NFS4_LOCK_STID:
5059                 atomic_inc(&s->sc_count);
5060                 spin_unlock(&cl->cl_lock);
5061                 ret = nfsd4_free_lock_stateid(stateid, s);
5062                 goto out;
5063         case NFS4_REVOKED_DELEG_STID:
5064                 dp = delegstateid(s);
5065                 list_del_init(&dp->dl_recall_lru);
5066                 spin_unlock(&cl->cl_lock);
5067                 nfs4_put_stid(s);
5068                 ret = nfs_ok;
5069                 goto out;
5070         /* Default falls through and returns nfserr_bad_stateid */
5071         }
5072 out_unlock:
5073         spin_unlock(&cl->cl_lock);
5074 out:
5075         return ret;
5076 }
5077
5078 static inline int
5079 setlkflg (int type)
5080 {
5081         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5082                 RD_STATE : WR_STATE;
5083 }
5084
5085 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5086 {
5087         struct svc_fh *current_fh = &cstate->current_fh;
5088         struct nfs4_stateowner *sop = stp->st_stateowner;
5089         __be32 status;
5090
5091         status = nfsd4_check_seqid(cstate, sop, seqid);
5092         if (status)
5093                 return status;
5094         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
5095                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
5096                 /*
5097                  * "Closed" stateid's exist *only* to return
5098                  * nfserr_replay_me from the previous step, and
5099                  * revoked delegations are kept only for free_stateid.
5100                  */
5101                 return nfserr_bad_stateid;
5102         mutex_lock(&stp->st_mutex);
5103         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5104         if (status == nfs_ok)
5105                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5106         if (status != nfs_ok)
5107                 mutex_unlock(&stp->st_mutex);
5108         return status;
5109 }
5110
5111 /*
5112  * Checks for sequence id mutating operations.
5113  */
5114 static __be32
5115 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5116                          stateid_t *stateid, char typemask,
5117                          struct nfs4_ol_stateid **stpp,
5118                          struct nfsd_net *nn)
5119 {
5120         __be32 status;
5121         struct nfs4_stid *s;
5122         struct nfs4_ol_stateid *stp = NULL;
5123
5124         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5125                 seqid, STATEID_VAL(stateid));
5126
5127         *stpp = NULL;
5128         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5129         if (status)
5130                 return status;
5131         stp = openlockstateid(s);
5132         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5133
5134         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5135         if (!status)
5136                 *stpp = stp;
5137         else
5138                 nfs4_put_stid(&stp->st_stid);
5139         return status;
5140 }
5141
5142 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5143                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5144 {
5145         __be32 status;
5146         struct nfs4_openowner *oo;
5147         struct nfs4_ol_stateid *stp;
5148
5149         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5150                                                 NFS4_OPEN_STID, &stp, nn);
5151         if (status)
5152                 return status;
5153         oo = openowner(stp->st_stateowner);
5154         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5155                 mutex_unlock(&stp->st_mutex);
5156                 nfs4_put_stid(&stp->st_stid);
5157                 return nfserr_bad_stateid;
5158         }
5159         *stpp = stp;
5160         return nfs_ok;
5161 }
5162
5163 __be32
5164 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5165                    struct nfsd4_open_confirm *oc)
5166 {
5167         __be32 status;
5168         struct nfs4_openowner *oo;
5169         struct nfs4_ol_stateid *stp;
5170         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5171
5172         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5173                         cstate->current_fh.fh_dentry);
5174
5175         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5176         if (status)
5177                 return status;
5178
5179         status = nfs4_preprocess_seqid_op(cstate,
5180                                         oc->oc_seqid, &oc->oc_req_stateid,
5181                                         NFS4_OPEN_STID, &stp, nn);
5182         if (status)
5183                 goto out;
5184         oo = openowner(stp->st_stateowner);
5185         status = nfserr_bad_stateid;
5186         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5187                 mutex_unlock(&stp->st_mutex);
5188                 goto put_stateid;
5189         }
5190         oo->oo_flags |= NFS4_OO_CONFIRMED;
5191         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5192         mutex_unlock(&stp->st_mutex);
5193         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5194                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5195
5196         nfsd4_client_record_create(oo->oo_owner.so_client);
5197         status = nfs_ok;
5198 put_stateid:
5199         nfs4_put_stid(&stp->st_stid);
5200 out:
5201         nfsd4_bump_seqid(cstate, status);
5202         return status;
5203 }
5204
5205 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5206 {
5207         if (!test_access(access, stp))
5208                 return;
5209         nfs4_file_put_access(stp->st_stid.sc_file, access);
5210         clear_access(access, stp);
5211 }
5212
5213 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5214 {
5215         switch (to_access) {
5216         case NFS4_SHARE_ACCESS_READ:
5217                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5218                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5219                 break;
5220         case NFS4_SHARE_ACCESS_WRITE:
5221                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5222                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5223                 break;
5224         case NFS4_SHARE_ACCESS_BOTH:
5225                 break;
5226         default:
5227                 WARN_ON_ONCE(1);
5228         }
5229 }
5230
5231 __be32
5232 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5233                      struct nfsd4_compound_state *cstate,
5234                      struct nfsd4_open_downgrade *od)
5235 {
5236         __be32 status;
5237         struct nfs4_ol_stateid *stp;
5238         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5239
5240         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5241                         cstate->current_fh.fh_dentry);
5242
5243         /* We don't yet support WANT bits: */
5244         if (od->od_deleg_want)
5245                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5246                         od->od_deleg_want);
5247
5248         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5249                                         &od->od_stateid, &stp, nn);
5250         if (status)
5251                 goto out;
5252         status = nfserr_inval;
5253         if (!test_access(od->od_share_access, stp)) {
5254                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5255                         stp->st_access_bmap, od->od_share_access);
5256                 goto put_stateid;
5257         }
5258         if (!test_deny(od->od_share_deny, stp)) {
5259                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5260                         stp->st_deny_bmap, od->od_share_deny);
5261                 goto put_stateid;
5262         }
5263         nfs4_stateid_downgrade(stp, od->od_share_access);
5264         reset_union_bmap_deny(od->od_share_deny, stp);
5265         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5266         status = nfs_ok;
5267 put_stateid:
5268         mutex_unlock(&stp->st_mutex);
5269         nfs4_put_stid(&stp->st_stid);
5270 out:
5271         nfsd4_bump_seqid(cstate, status);
5272         return status;
5273 }
5274
5275 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5276 {
5277         struct nfs4_client *clp = s->st_stid.sc_client;
5278         bool unhashed;
5279         LIST_HEAD(reaplist);
5280
5281         s->st_stid.sc_type = NFS4_CLOSED_STID;
5282         spin_lock(&clp->cl_lock);
5283         unhashed = unhash_open_stateid(s, &reaplist);
5284
5285         if (clp->cl_minorversion) {
5286                 if (unhashed)
5287                         put_ol_stateid_locked(s, &reaplist);
5288                 spin_unlock(&clp->cl_lock);
5289                 free_ol_stateid_reaplist(&reaplist);
5290         } else {
5291                 spin_unlock(&clp->cl_lock);
5292                 free_ol_stateid_reaplist(&reaplist);
5293                 if (unhashed)
5294                         move_to_close_lru(s, clp->net);
5295         }
5296 }
5297
5298 /*
5299  * nfs4_unlock_state() called after encode
5300  */
5301 __be32
5302 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5303             struct nfsd4_close *close)
5304 {
5305         __be32 status;
5306         struct nfs4_ol_stateid *stp;
5307         struct net *net = SVC_NET(rqstp);
5308         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5309
5310         dprintk("NFSD: nfsd4_close on file %pd\n",
5311                         cstate->current_fh.fh_dentry);
5312
5313         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5314                                         &close->cl_stateid,
5315                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5316                                         &stp, nn);
5317         nfsd4_bump_seqid(cstate, status);
5318         if (status)
5319                 goto out;
5320         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5321         mutex_unlock(&stp->st_mutex);
5322
5323         nfsd4_close_open_stateid(stp);
5324
5325         /* put reference from nfs4_preprocess_seqid_op */
5326         nfs4_put_stid(&stp->st_stid);
5327 out:
5328         return status;
5329 }
5330
5331 __be32
5332 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5333                   struct nfsd4_delegreturn *dr)
5334 {
5335         struct nfs4_delegation *dp;
5336         stateid_t *stateid = &dr->dr_stateid;
5337         struct nfs4_stid *s;
5338         __be32 status;
5339         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5340
5341         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5342                 return status;
5343
5344         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5345         if (status)
5346                 goto out;
5347         dp = delegstateid(s);
5348         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5349         if (status)
5350                 goto put_stateid;
5351
5352         destroy_delegation(dp);
5353 put_stateid:
5354         nfs4_put_stid(&dp->dl_stid);
5355 out:
5356         return status;
5357 }
5358
5359 static inline u64
5360 end_offset(u64 start, u64 len)
5361 {
5362         u64 end;
5363
5364         end = start + len;
5365         return end >= start ? end: NFS4_MAX_UINT64;
5366 }
5367
5368 /* last octet in a range */
5369 static inline u64
5370 last_byte_offset(u64 start, u64 len)
5371 {
5372         u64 end;
5373
5374         WARN_ON_ONCE(!len);
5375         end = start + len;
5376         return end > start ? end - 1: NFS4_MAX_UINT64;
5377 }
5378
5379 /*
5380  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5381  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5382  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5383  * locking, this prevents us from being completely protocol-compliant.  The
5384  * real solution to this problem is to start using unsigned file offsets in
5385  * the VFS, but this is a very deep change!
5386  */
5387 static inline void
5388 nfs4_transform_lock_offset(struct file_lock *lock)
5389 {
5390         if (lock->fl_start < 0)
5391                 lock->fl_start = OFFSET_MAX;
5392         if (lock->fl_end < 0)
5393                 lock->fl_end = OFFSET_MAX;
5394 }
5395
5396 static fl_owner_t
5397 nfsd4_fl_get_owner(fl_owner_t owner)
5398 {
5399         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5400
5401         nfs4_get_stateowner(&lo->lo_owner);
5402         return owner;
5403 }
5404
5405 static void
5406 nfsd4_fl_put_owner(fl_owner_t owner)
5407 {
5408         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5409
5410         if (lo)
5411                 nfs4_put_stateowner(&lo->lo_owner);
5412 }
5413
5414 static void
5415 nfsd4_lm_notify(struct file_lock *fl)
5416 {
5417         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5418         struct net                      *net = lo->lo_owner.so_client->net;
5419         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5420         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5421                                                 struct nfsd4_blocked_lock, nbl_lock);
5422         bool queue = false;
5423
5424         /* An empty list means that something else is going to be using it */
5425         spin_lock(&nn->blocked_locks_lock);
5426         if (!list_empty(&nbl->nbl_list)) {
5427                 list_del_init(&nbl->nbl_list);
5428                 list_del_init(&nbl->nbl_lru);
5429                 queue = true;
5430         }
5431         spin_unlock(&nn->blocked_locks_lock);
5432
5433         if (queue)
5434                 nfsd4_run_cb(&nbl->nbl_cb);
5435 }
5436
5437 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5438         .lm_notify = nfsd4_lm_notify,
5439         .lm_get_owner = nfsd4_fl_get_owner,
5440         .lm_put_owner = nfsd4_fl_put_owner,
5441 };
5442
5443 static inline void
5444 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5445 {
5446         struct nfs4_lockowner *lo;
5447
5448         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5449                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5450                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5451                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5452                 if (!deny->ld_owner.data)
5453                         /* We just don't care that much */
5454                         goto nevermind;
5455                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5456                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5457         } else {
5458 nevermind:
5459                 deny->ld_owner.len = 0;
5460                 deny->ld_owner.data = NULL;
5461                 deny->ld_clientid.cl_boot = 0;
5462                 deny->ld_clientid.cl_id = 0;
5463         }
5464         deny->ld_start = fl->fl_start;
5465         deny->ld_length = NFS4_MAX_UINT64;
5466         if (fl->fl_end != NFS4_MAX_UINT64)
5467                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5468         deny->ld_type = NFS4_READ_LT;
5469         if (fl->fl_type != F_RDLCK)
5470                 deny->ld_type = NFS4_WRITE_LT;
5471 }
5472
5473 static struct nfs4_lockowner *
5474 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5475 {
5476         unsigned int strhashval = ownerstr_hashval(owner);
5477         struct nfs4_stateowner *so;
5478
5479         lockdep_assert_held(&clp->cl_lock);
5480
5481         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5482                             so_strhash) {
5483                 if (so->so_is_open_owner)
5484                         continue;
5485                 if (same_owner_str(so, owner))
5486                         return lockowner(nfs4_get_stateowner(so));
5487         }
5488         return NULL;
5489 }
5490
5491 static struct nfs4_lockowner *
5492 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5493 {
5494         struct nfs4_lockowner *lo;
5495
5496         spin_lock(&clp->cl_lock);
5497         lo = find_lockowner_str_locked(clp, owner);
5498         spin_unlock(&clp->cl_lock);
5499         return lo;
5500 }
5501
5502 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5503 {
5504         unhash_lockowner_locked(lockowner(sop));
5505 }
5506
5507 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5508 {
5509         struct nfs4_lockowner *lo = lockowner(sop);
5510
5511         kmem_cache_free(lockowner_slab, lo);
5512 }
5513
5514 static const struct nfs4_stateowner_operations lockowner_ops = {
5515         .so_unhash =    nfs4_unhash_lockowner,
5516         .so_free =      nfs4_free_lockowner,
5517 };
5518
5519 /*
5520  * Alloc a lock owner structure.
5521  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5522  * occurred.
5523  *
5524  * strhashval = ownerstr_hashval
5525  */
5526 static struct nfs4_lockowner *
5527 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5528                            struct nfs4_ol_stateid *open_stp,
5529                            struct nfsd4_lock *lock)
5530 {
5531         struct nfs4_lockowner *lo, *ret;
5532
5533         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5534         if (!lo)
5535                 return NULL;
5536         INIT_LIST_HEAD(&lo->lo_blocked);
5537         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5538         lo->lo_owner.so_is_open_owner = 0;
5539         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5540         lo->lo_owner.so_ops = &lockowner_ops;
5541         spin_lock(&clp->cl_lock);
5542         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5543         if (ret == NULL) {
5544                 list_add(&lo->lo_owner.so_strhash,
5545                          &clp->cl_ownerstr_hashtbl[strhashval]);
5546                 ret = lo;
5547         } else
5548                 nfs4_free_stateowner(&lo->lo_owner);
5549
5550         spin_unlock(&clp->cl_lock);
5551         return ret;
5552 }
5553
5554 static void
5555 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5556                   struct nfs4_file *fp, struct inode *inode,
5557                   struct nfs4_ol_stateid *open_stp)
5558 {
5559         struct nfs4_client *clp = lo->lo_owner.so_client;
5560
5561         lockdep_assert_held(&clp->cl_lock);
5562
5563         atomic_inc(&stp->st_stid.sc_count);
5564         stp->st_stid.sc_type = NFS4_LOCK_STID;
5565         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5566         get_nfs4_file(fp);
5567         stp->st_stid.sc_file = fp;
5568         stp->st_access_bmap = 0;
5569         stp->st_deny_bmap = open_stp->st_deny_bmap;
5570         stp->st_openstp = open_stp;
5571         mutex_init(&stp->st_mutex);
5572         list_add(&stp->st_locks, &open_stp->st_locks);
5573         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5574         spin_lock(&fp->fi_lock);
5575         list_add(&stp->st_perfile, &fp->fi_stateids);
5576         spin_unlock(&fp->fi_lock);
5577 }
5578
5579 static struct nfs4_ol_stateid *
5580 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5581 {
5582         struct nfs4_ol_stateid *lst;
5583         struct nfs4_client *clp = lo->lo_owner.so_client;
5584
5585         lockdep_assert_held(&clp->cl_lock);
5586
5587         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5588                 if (lst->st_stid.sc_file == fp) {
5589                         atomic_inc(&lst->st_stid.sc_count);
5590                         return lst;
5591                 }
5592         }
5593         return NULL;
5594 }
5595
5596 static struct nfs4_ol_stateid *
5597 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5598                             struct inode *inode, struct nfs4_ol_stateid *ost,
5599                             bool *new)
5600 {
5601         struct nfs4_stid *ns = NULL;
5602         struct nfs4_ol_stateid *lst;
5603         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5604         struct nfs4_client *clp = oo->oo_owner.so_client;
5605
5606         spin_lock(&clp->cl_lock);
5607         lst = find_lock_stateid(lo, fi);
5608         if (lst == NULL) {
5609                 spin_unlock(&clp->cl_lock);
5610                 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5611                 if (ns == NULL)
5612                         return NULL;
5613
5614                 spin_lock(&clp->cl_lock);
5615                 lst = find_lock_stateid(lo, fi);
5616                 if (likely(!lst)) {
5617                         lst = openlockstateid(ns);
5618                         init_lock_stateid(lst, lo, fi, inode, ost);
5619                         ns = NULL;
5620                         *new = true;
5621                 }
5622         }
5623         spin_unlock(&clp->cl_lock);
5624         if (ns)
5625                 nfs4_put_stid(ns);
5626         return lst;
5627 }
5628
5629 static int
5630 check_lock_length(u64 offset, u64 length)
5631 {
5632         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5633                 (length > ~offset)));
5634 }
5635
5636 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5637 {
5638         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5639
5640         lockdep_assert_held(&fp->fi_lock);
5641
5642         if (test_access(access, lock_stp))
5643                 return;
5644         __nfs4_file_get_access(fp, access);
5645         set_access(access, lock_stp);
5646 }
5647
5648 static __be32
5649 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5650                             struct nfs4_ol_stateid *ost,
5651                             struct nfsd4_lock *lock,
5652                             struct nfs4_ol_stateid **plst, bool *new)
5653 {
5654         __be32 status;
5655         struct nfs4_file *fi = ost->st_stid.sc_file;
5656         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5657         struct nfs4_client *cl = oo->oo_owner.so_client;
5658         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5659         struct nfs4_lockowner *lo;
5660         struct nfs4_ol_stateid *lst;
5661         unsigned int strhashval;
5662         bool hashed;
5663
5664         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5665         if (!lo) {
5666                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5667                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5668                 if (lo == NULL)
5669                         return nfserr_jukebox;
5670         } else {
5671                 /* with an existing lockowner, seqids must be the same */
5672                 status = nfserr_bad_seqid;
5673                 if (!cstate->minorversion &&
5674                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5675                         goto out;
5676         }
5677
5678 retry:
5679         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5680         if (lst == NULL) {
5681                 status = nfserr_jukebox;
5682                 goto out;
5683         }
5684
5685         mutex_lock(&lst->st_mutex);
5686
5687         /* See if it's still hashed to avoid race with FREE_STATEID */
5688         spin_lock(&cl->cl_lock);
5689         hashed = !list_empty(&lst->st_perfile);
5690         spin_unlock(&cl->cl_lock);
5691
5692         if (!hashed) {
5693                 mutex_unlock(&lst->st_mutex);
5694                 nfs4_put_stid(&lst->st_stid);
5695                 goto retry;
5696         }
5697         status = nfs_ok;
5698         *plst = lst;
5699 out:
5700         nfs4_put_stateowner(&lo->lo_owner);
5701         return status;
5702 }
5703
5704 /*
5705  *  LOCK operation
5706  */
5707 __be32
5708 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5709            struct nfsd4_lock *lock)
5710 {
5711         struct nfs4_openowner *open_sop = NULL;
5712         struct nfs4_lockowner *lock_sop = NULL;
5713         struct nfs4_ol_stateid *lock_stp = NULL;
5714         struct nfs4_ol_stateid *open_stp = NULL;
5715         struct nfs4_file *fp;
5716         struct file *filp = NULL;
5717         struct nfsd4_blocked_lock *nbl = NULL;
5718         struct file_lock *file_lock = NULL;
5719         struct file_lock *conflock = NULL;
5720         __be32 status = 0;
5721         int lkflg;
5722         int err;
5723         bool new = false;
5724         unsigned char fl_type;
5725         unsigned int fl_flags = FL_POSIX;
5726         struct net *net = SVC_NET(rqstp);
5727         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5728
5729         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5730                 (long long) lock->lk_offset,
5731                 (long long) lock->lk_length);
5732
5733         if (check_lock_length(lock->lk_offset, lock->lk_length))
5734                  return nfserr_inval;
5735
5736         if ((status = fh_verify(rqstp, &cstate->current_fh,
5737                                 S_IFREG, NFSD_MAY_LOCK))) {
5738                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5739                 return status;
5740         }
5741
5742         if (lock->lk_is_new) {
5743                 if (nfsd4_has_session(cstate))
5744                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5745                         memcpy(&lock->lk_new_clientid,
5746                                 &cstate->session->se_client->cl_clientid,
5747                                 sizeof(clientid_t));
5748
5749                 status = nfserr_stale_clientid;
5750                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5751                         goto out;
5752
5753                 /* validate and update open stateid and open seqid */
5754                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5755                                         lock->lk_new_open_seqid,
5756                                         &lock->lk_new_open_stateid,
5757                                         &open_stp, nn);
5758                 if (status)
5759                         goto out;
5760                 mutex_unlock(&open_stp->st_mutex);
5761                 open_sop = openowner(open_stp->st_stateowner);
5762                 status = nfserr_bad_stateid;
5763                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5764                                                 &lock->lk_new_clientid))
5765                         goto out;
5766                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5767                                                         &lock_stp, &new);
5768         } else {
5769                 status = nfs4_preprocess_seqid_op(cstate,
5770                                        lock->lk_old_lock_seqid,
5771                                        &lock->lk_old_lock_stateid,
5772                                        NFS4_LOCK_STID, &lock_stp, nn);
5773         }
5774         if (status)
5775                 goto out;
5776         lock_sop = lockowner(lock_stp->st_stateowner);
5777
5778         lkflg = setlkflg(lock->lk_type);
5779         status = nfs4_check_openmode(lock_stp, lkflg);
5780         if (status)
5781                 goto out;
5782
5783         status = nfserr_grace;
5784         if (locks_in_grace(net) && !lock->lk_reclaim)
5785                 goto out;
5786         status = nfserr_no_grace;
5787         if (!locks_in_grace(net) && lock->lk_reclaim)
5788                 goto out;
5789
5790         fp = lock_stp->st_stid.sc_file;
5791         switch (lock->lk_type) {
5792                 case NFS4_READW_LT:
5793                         if (nfsd4_has_session(cstate))
5794                                 fl_flags |= FL_SLEEP;
5795                         /* Fallthrough */
5796                 case NFS4_READ_LT:
5797                         spin_lock(&fp->fi_lock);
5798                         filp = find_readable_file_locked(fp);
5799                         if (filp)
5800                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5801                         spin_unlock(&fp->fi_lock);
5802                         fl_type = F_RDLCK;
5803                         break;
5804                 case NFS4_WRITEW_LT:
5805                         if (nfsd4_has_session(cstate))
5806                                 fl_flags |= FL_SLEEP;
5807                         /* Fallthrough */
5808                 case NFS4_WRITE_LT:
5809                         spin_lock(&fp->fi_lock);
5810                         filp = find_writeable_file_locked(fp);
5811                         if (filp)
5812                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5813                         spin_unlock(&fp->fi_lock);
5814                         fl_type = F_WRLCK;
5815                         break;
5816                 default:
5817                         status = nfserr_inval;
5818                 goto out;
5819         }
5820
5821         if (!filp) {
5822                 status = nfserr_openmode;
5823                 goto out;
5824         }
5825
5826         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
5827         if (!nbl) {
5828                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
5829                 status = nfserr_jukebox;
5830                 goto out;
5831         }
5832
5833         file_lock = &nbl->nbl_lock;
5834         file_lock->fl_type = fl_type;
5835         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5836         file_lock->fl_pid = current->tgid;
5837         file_lock->fl_file = filp;
5838         file_lock->fl_flags = fl_flags;
5839         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5840         file_lock->fl_start = lock->lk_offset;
5841         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5842         nfs4_transform_lock_offset(file_lock);
5843
5844         conflock = locks_alloc_lock();
5845         if (!conflock) {
5846                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5847                 status = nfserr_jukebox;
5848                 goto out;
5849         }
5850
5851         if (fl_flags & FL_SLEEP) {
5852                 nbl->nbl_time = jiffies;
5853                 spin_lock(&nn->blocked_locks_lock);
5854                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
5855                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
5856                 spin_unlock(&nn->blocked_locks_lock);
5857         }
5858
5859         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5860         switch (err) {
5861         case 0: /* success! */
5862                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5863                 status = 0;
5864                 break;
5865         case FILE_LOCK_DEFERRED:
5866                 nbl = NULL;
5867                 /* Fallthrough */
5868         case -EAGAIN:           /* conflock holds conflicting lock */
5869                 status = nfserr_denied;
5870                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5871                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5872                 break;
5873         case -EDEADLK:
5874                 status = nfserr_deadlock;
5875                 break;
5876         default:
5877                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5878                 status = nfserrno(err);
5879                 break;
5880         }
5881 out:
5882         if (nbl) {
5883                 /* dequeue it if we queued it before */
5884                 if (fl_flags & FL_SLEEP) {
5885                         spin_lock(&nn->blocked_locks_lock);
5886                         list_del_init(&nbl->nbl_list);
5887                         list_del_init(&nbl->nbl_lru);
5888                         spin_unlock(&nn->blocked_locks_lock);
5889                 }
5890                 free_blocked_lock(nbl);
5891         }
5892         if (filp)
5893                 fput(filp);
5894         if (lock_stp) {
5895                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5896                 if (cstate->replay_owner &&
5897                     cstate->replay_owner != &lock_sop->lo_owner &&
5898                     seqid_mutating_err(ntohl(status)))
5899                         lock_sop->lo_owner.so_seqid++;
5900
5901                 mutex_unlock(&lock_stp->st_mutex);
5902
5903                 /*
5904                  * If this is a new, never-before-used stateid, and we are
5905                  * returning an error, then just go ahead and release it.
5906                  */
5907                 if (status && new)
5908                         release_lock_stateid(lock_stp);
5909
5910                 nfs4_put_stid(&lock_stp->st_stid);
5911         }
5912         if (open_stp)
5913                 nfs4_put_stid(&open_stp->st_stid);
5914         nfsd4_bump_seqid(cstate, status);
5915         if (conflock)
5916                 locks_free_lock(conflock);
5917         return status;
5918 }
5919
5920 /*
5921  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5922  * so we do a temporary open here just to get an open file to pass to
5923  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5924  * inode operation.)
5925  */
5926 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5927 {
5928         struct file *file;
5929         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5930         if (!err) {
5931                 err = nfserrno(vfs_test_lock(file, lock));
5932                 fput(file);
5933         }
5934         return err;
5935 }
5936
5937 /*
5938  * LOCKT operation
5939  */
5940 __be32
5941 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5942             struct nfsd4_lockt *lockt)
5943 {
5944         struct file_lock *file_lock = NULL;
5945         struct nfs4_lockowner *lo = NULL;
5946         __be32 status;
5947         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5948
5949         if (locks_in_grace(SVC_NET(rqstp)))
5950                 return nfserr_grace;
5951
5952         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5953                  return nfserr_inval;
5954
5955         if (!nfsd4_has_session(cstate)) {
5956                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5957                 if (status)
5958                         goto out;
5959         }
5960
5961         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5962                 goto out;
5963
5964         file_lock = locks_alloc_lock();
5965         if (!file_lock) {
5966                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5967                 status = nfserr_jukebox;
5968                 goto out;
5969         }
5970
5971         switch (lockt->lt_type) {
5972                 case NFS4_READ_LT:
5973                 case NFS4_READW_LT:
5974                         file_lock->fl_type = F_RDLCK;
5975                 break;
5976                 case NFS4_WRITE_LT:
5977                 case NFS4_WRITEW_LT:
5978                         file_lock->fl_type = F_WRLCK;
5979                 break;
5980                 default:
5981                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5982                         status = nfserr_inval;
5983                 goto out;
5984         }
5985
5986         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
5987         if (lo)
5988                 file_lock->fl_owner = (fl_owner_t)lo;
5989         file_lock->fl_pid = current->tgid;
5990         file_lock->fl_flags = FL_POSIX;
5991
5992         file_lock->fl_start = lockt->lt_offset;
5993         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5994
5995         nfs4_transform_lock_offset(file_lock);
5996
5997         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5998         if (status)
5999                 goto out;
6000
6001         if (file_lock->fl_type != F_UNLCK) {
6002                 status = nfserr_denied;
6003                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6004         }
6005 out:
6006         if (lo)
6007                 nfs4_put_stateowner(&lo->lo_owner);
6008         if (file_lock)
6009                 locks_free_lock(file_lock);
6010         return status;
6011 }
6012
6013 __be32
6014 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6015             struct nfsd4_locku *locku)
6016 {
6017         struct nfs4_ol_stateid *stp;
6018         struct file *filp = NULL;
6019         struct file_lock *file_lock = NULL;
6020         __be32 status;
6021         int err;
6022         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6023
6024         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6025                 (long long) locku->lu_offset,
6026                 (long long) locku->lu_length);
6027
6028         if (check_lock_length(locku->lu_offset, locku->lu_length))
6029                  return nfserr_inval;
6030
6031         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6032                                         &locku->lu_stateid, NFS4_LOCK_STID,
6033                                         &stp, nn);
6034         if (status)
6035                 goto out;
6036         filp = find_any_file(stp->st_stid.sc_file);
6037         if (!filp) {
6038                 status = nfserr_lock_range;
6039                 goto put_stateid;
6040         }
6041         file_lock = locks_alloc_lock();
6042         if (!file_lock) {
6043                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6044                 status = nfserr_jukebox;
6045                 goto fput;
6046         }
6047
6048         file_lock->fl_type = F_UNLCK;
6049         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6050         file_lock->fl_pid = current->tgid;
6051         file_lock->fl_file = filp;
6052         file_lock->fl_flags = FL_POSIX;
6053         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6054         file_lock->fl_start = locku->lu_offset;
6055
6056         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6057                                                 locku->lu_length);
6058         nfs4_transform_lock_offset(file_lock);
6059
6060         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6061         if (err) {
6062                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6063                 goto out_nfserr;
6064         }
6065         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6066 fput:
6067         fput(filp);
6068 put_stateid:
6069         mutex_unlock(&stp->st_mutex);
6070         nfs4_put_stid(&stp->st_stid);
6071 out:
6072         nfsd4_bump_seqid(cstate, status);
6073         if (file_lock)
6074                 locks_free_lock(file_lock);
6075         return status;
6076
6077 out_nfserr:
6078         status = nfserrno(err);
6079         goto fput;
6080 }
6081
6082 /*
6083  * returns
6084  *      true:  locks held by lockowner
6085  *      false: no locks held by lockowner
6086  */
6087 static bool
6088 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6089 {
6090         struct file_lock *fl;
6091         int status = false;
6092         struct file *filp = find_any_file(fp);
6093         struct inode *inode;
6094         struct file_lock_context *flctx;
6095
6096         if (!filp) {
6097                 /* Any valid lock stateid should have some sort of access */
6098                 WARN_ON_ONCE(1);
6099                 return status;
6100         }
6101
6102         inode = file_inode(filp);
6103         flctx = inode->i_flctx;
6104
6105         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6106                 spin_lock(&flctx->flc_lock);
6107                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6108                         if (fl->fl_owner == (fl_owner_t)lowner) {
6109                                 status = true;
6110                                 break;
6111                         }
6112                 }
6113                 spin_unlock(&flctx->flc_lock);
6114         }
6115         fput(filp);
6116         return status;
6117 }
6118
6119 __be32
6120 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6121                         struct nfsd4_compound_state *cstate,
6122                         struct nfsd4_release_lockowner *rlockowner)
6123 {
6124         clientid_t *clid = &rlockowner->rl_clientid;
6125         struct nfs4_stateowner *sop;
6126         struct nfs4_lockowner *lo = NULL;
6127         struct nfs4_ol_stateid *stp;
6128         struct xdr_netobj *owner = &rlockowner->rl_owner;
6129         unsigned int hashval = ownerstr_hashval(owner);
6130         __be32 status;
6131         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6132         struct nfs4_client *clp;
6133         LIST_HEAD (reaplist);
6134
6135         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6136                 clid->cl_boot, clid->cl_id);
6137
6138         status = lookup_clientid(clid, cstate, nn);
6139         if (status)
6140                 return status;
6141
6142         clp = cstate->clp;
6143         /* Find the matching lock stateowner */
6144         spin_lock(&clp->cl_lock);
6145         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6146                             so_strhash) {
6147
6148                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6149                         continue;
6150
6151                 /* see if there are still any locks associated with it */
6152                 lo = lockowner(sop);
6153                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6154                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6155                                 status = nfserr_locks_held;
6156                                 spin_unlock(&clp->cl_lock);
6157                                 return status;
6158                         }
6159                 }
6160
6161                 nfs4_get_stateowner(sop);
6162                 break;
6163         }
6164         if (!lo) {
6165                 spin_unlock(&clp->cl_lock);
6166                 return status;
6167         }
6168
6169         unhash_lockowner_locked(lo);
6170         while (!list_empty(&lo->lo_owner.so_stateids)) {
6171                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6172                                        struct nfs4_ol_stateid,
6173                                        st_perstateowner);
6174                 WARN_ON(!unhash_lock_stateid(stp));
6175                 put_ol_stateid_locked(stp, &reaplist);
6176         }
6177         spin_unlock(&clp->cl_lock);
6178         free_ol_stateid_reaplist(&reaplist);
6179         nfs4_put_stateowner(&lo->lo_owner);
6180
6181         return status;
6182 }
6183
6184 static inline struct nfs4_client_reclaim *
6185 alloc_reclaim(void)
6186 {
6187         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6188 }
6189
6190 bool
6191 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6192 {
6193         struct nfs4_client_reclaim *crp;
6194
6195         crp = nfsd4_find_reclaim_client(name, nn);
6196         return (crp && crp->cr_clp);
6197 }
6198
6199 /*
6200  * failure => all reset bets are off, nfserr_no_grace...
6201  */
6202 struct nfs4_client_reclaim *
6203 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6204 {
6205         unsigned int strhashval;
6206         struct nfs4_client_reclaim *crp;
6207
6208         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6209         crp = alloc_reclaim();
6210         if (crp) {
6211                 strhashval = clientstr_hashval(name);
6212                 INIT_LIST_HEAD(&crp->cr_strhash);
6213                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6214                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6215                 crp->cr_clp = NULL;
6216                 nn->reclaim_str_hashtbl_size++;
6217         }
6218         return crp;
6219 }
6220
6221 void
6222 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6223 {
6224         list_del(&crp->cr_strhash);
6225         kfree(crp);
6226         nn->reclaim_str_hashtbl_size--;
6227 }
6228
6229 void
6230 nfs4_release_reclaim(struct nfsd_net *nn)
6231 {
6232         struct nfs4_client_reclaim *crp = NULL;
6233         int i;
6234
6235         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6236                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6237                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6238                                         struct nfs4_client_reclaim, cr_strhash);
6239                         nfs4_remove_reclaim_record(crp, nn);
6240                 }
6241         }
6242         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6243 }
6244
6245 /*
6246  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6247 struct nfs4_client_reclaim *
6248 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6249 {
6250         unsigned int strhashval;
6251         struct nfs4_client_reclaim *crp = NULL;
6252
6253         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6254
6255         strhashval = clientstr_hashval(recdir);
6256         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6257                 if (same_name(crp->cr_recdir, recdir)) {
6258                         return crp;
6259                 }
6260         }
6261         return NULL;
6262 }
6263
6264 /*
6265 * Called from OPEN. Look for clientid in reclaim list.
6266 */
6267 __be32
6268 nfs4_check_open_reclaim(clientid_t *clid,
6269                 struct nfsd4_compound_state *cstate,
6270                 struct nfsd_net *nn)
6271 {
6272         __be32 status;
6273
6274         /* find clientid in conf_id_hashtbl */
6275         status = lookup_clientid(clid, cstate, nn);
6276         if (status)
6277                 return nfserr_reclaim_bad;
6278
6279         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6280                 return nfserr_no_grace;
6281
6282         if (nfsd4_client_record_check(cstate->clp))
6283                 return nfserr_reclaim_bad;
6284
6285         return nfs_ok;
6286 }
6287
6288 #ifdef CONFIG_NFSD_FAULT_INJECTION
6289 static inline void
6290 put_client(struct nfs4_client *clp)
6291 {
6292         atomic_dec(&clp->cl_refcount);
6293 }
6294
6295 static struct nfs4_client *
6296 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6297 {
6298         struct nfs4_client *clp;
6299         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6300                                           nfsd_net_id);
6301
6302         if (!nfsd_netns_ready(nn))
6303                 return NULL;
6304
6305         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6306                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6307                         return clp;
6308         }
6309         return NULL;
6310 }
6311
6312 u64
6313 nfsd_inject_print_clients(void)
6314 {
6315         struct nfs4_client *clp;
6316         u64 count = 0;
6317         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6318                                           nfsd_net_id);
6319         char buf[INET6_ADDRSTRLEN];
6320
6321         if (!nfsd_netns_ready(nn))
6322                 return 0;
6323
6324         spin_lock(&nn->client_lock);
6325         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6326                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6327                 pr_info("NFS Client: %s\n", buf);
6328                 ++count;
6329         }
6330         spin_unlock(&nn->client_lock);
6331
6332         return count;
6333 }
6334
6335 u64
6336 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6337 {
6338         u64 count = 0;
6339         struct nfs4_client *clp;
6340         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6341                                           nfsd_net_id);
6342
6343         if (!nfsd_netns_ready(nn))
6344                 return count;
6345
6346         spin_lock(&nn->client_lock);
6347         clp = nfsd_find_client(addr, addr_size);
6348         if (clp) {
6349                 if (mark_client_expired_locked(clp) == nfs_ok)
6350                         ++count;
6351                 else
6352                         clp = NULL;
6353         }
6354         spin_unlock(&nn->client_lock);
6355
6356         if (clp)
6357                 expire_client(clp);
6358
6359         return count;
6360 }
6361
6362 u64
6363 nfsd_inject_forget_clients(u64 max)
6364 {
6365         u64 count = 0;
6366         struct nfs4_client *clp, *next;
6367         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6368                                                 nfsd_net_id);
6369         LIST_HEAD(reaplist);
6370
6371         if (!nfsd_netns_ready(nn))
6372                 return count;
6373
6374         spin_lock(&nn->client_lock);
6375         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6376                 if (mark_client_expired_locked(clp) == nfs_ok) {
6377                         list_add(&clp->cl_lru, &reaplist);
6378                         if (max != 0 && ++count >= max)
6379                                 break;
6380                 }
6381         }
6382         spin_unlock(&nn->client_lock);
6383
6384         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6385                 expire_client(clp);
6386
6387         return count;
6388 }
6389
6390 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6391                              const char *type)
6392 {
6393         char buf[INET6_ADDRSTRLEN];
6394         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6395         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6396 }
6397
6398 static void
6399 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6400                              struct list_head *collect)
6401 {
6402         struct nfs4_client *clp = lst->st_stid.sc_client;
6403         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6404                                           nfsd_net_id);
6405
6406         if (!collect)
6407                 return;
6408
6409         lockdep_assert_held(&nn->client_lock);
6410         atomic_inc(&clp->cl_refcount);
6411         list_add(&lst->st_locks, collect);
6412 }
6413
6414 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6415                                     struct list_head *collect,
6416                                     bool (*func)(struct nfs4_ol_stateid *))
6417 {
6418         struct nfs4_openowner *oop;
6419         struct nfs4_ol_stateid *stp, *st_next;
6420         struct nfs4_ol_stateid *lst, *lst_next;
6421         u64 count = 0;
6422
6423         spin_lock(&clp->cl_lock);
6424         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6425                 list_for_each_entry_safe(stp, st_next,
6426                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6427                         list_for_each_entry_safe(lst, lst_next,
6428                                         &stp->st_locks, st_locks) {
6429                                 if (func) {
6430                                         if (func(lst))
6431                                                 nfsd_inject_add_lock_to_list(lst,
6432                                                                         collect);
6433                                 }
6434                                 ++count;
6435                                 /*
6436                                  * Despite the fact that these functions deal
6437                                  * with 64-bit integers for "count", we must
6438                                  * ensure that it doesn't blow up the
6439                                  * clp->cl_refcount. Throw a warning if we
6440                                  * start to approach INT_MAX here.
6441                                  */
6442                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6443                                 if (count == max)
6444                                         goto out;
6445                         }
6446                 }
6447         }
6448 out:
6449         spin_unlock(&clp->cl_lock);
6450
6451         return count;
6452 }
6453
6454 static u64
6455 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6456                           u64 max)
6457 {
6458         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6459 }
6460
6461 static u64
6462 nfsd_print_client_locks(struct nfs4_client *clp)
6463 {
6464         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6465         nfsd_print_count(clp, count, "locked files");
6466         return count;
6467 }
6468
6469 u64
6470 nfsd_inject_print_locks(void)
6471 {
6472         struct nfs4_client *clp;
6473         u64 count = 0;
6474         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6475                                                 nfsd_net_id);
6476
6477         if (!nfsd_netns_ready(nn))
6478                 return 0;
6479
6480         spin_lock(&nn->client_lock);
6481         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6482                 count += nfsd_print_client_locks(clp);
6483         spin_unlock(&nn->client_lock);
6484
6485         return count;
6486 }
6487
6488 static void
6489 nfsd_reap_locks(struct list_head *reaplist)
6490 {
6491         struct nfs4_client *clp;
6492         struct nfs4_ol_stateid *stp, *next;
6493
6494         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6495                 list_del_init(&stp->st_locks);
6496                 clp = stp->st_stid.sc_client;
6497                 nfs4_put_stid(&stp->st_stid);
6498                 put_client(clp);
6499         }
6500 }
6501
6502 u64
6503 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6504 {
6505         unsigned int count = 0;
6506         struct nfs4_client *clp;
6507         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6508                                                 nfsd_net_id);
6509         LIST_HEAD(reaplist);
6510
6511         if (!nfsd_netns_ready(nn))
6512                 return count;
6513
6514         spin_lock(&nn->client_lock);
6515         clp = nfsd_find_client(addr, addr_size);
6516         if (clp)
6517                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6518         spin_unlock(&nn->client_lock);
6519         nfsd_reap_locks(&reaplist);
6520         return count;
6521 }
6522
6523 u64
6524 nfsd_inject_forget_locks(u64 max)
6525 {
6526         u64 count = 0;
6527         struct nfs4_client *clp;
6528         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6529                                                 nfsd_net_id);
6530         LIST_HEAD(reaplist);
6531
6532         if (!nfsd_netns_ready(nn))
6533                 return count;
6534
6535         spin_lock(&nn->client_lock);
6536         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6537                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6538                 if (max != 0 && count >= max)
6539                         break;
6540         }
6541         spin_unlock(&nn->client_lock);
6542         nfsd_reap_locks(&reaplist);
6543         return count;
6544 }
6545
6546 static u64
6547 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6548                               struct list_head *collect,
6549                               void (*func)(struct nfs4_openowner *))
6550 {
6551         struct nfs4_openowner *oop, *next;
6552         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6553                                                 nfsd_net_id);
6554         u64 count = 0;
6555
6556         lockdep_assert_held(&nn->client_lock);
6557
6558         spin_lock(&clp->cl_lock);
6559         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6560                 if (func) {
6561                         func(oop);
6562                         if (collect) {
6563                                 atomic_inc(&clp->cl_refcount);
6564                                 list_add(&oop->oo_perclient, collect);
6565                         }
6566                 }
6567                 ++count;
6568                 /*
6569                  * Despite the fact that these functions deal with
6570                  * 64-bit integers for "count", we must ensure that
6571                  * it doesn't blow up the clp->cl_refcount. Throw a
6572                  * warning if we start to approach INT_MAX here.
6573                  */
6574                 WARN_ON_ONCE(count == (INT_MAX / 2));
6575                 if (count == max)
6576                         break;
6577         }
6578         spin_unlock(&clp->cl_lock);
6579
6580         return count;
6581 }
6582
6583 static u64
6584 nfsd_print_client_openowners(struct nfs4_client *clp)
6585 {
6586         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6587
6588         nfsd_print_count(clp, count, "openowners");
6589         return count;
6590 }
6591
6592 static u64
6593 nfsd_collect_client_openowners(struct nfs4_client *clp,
6594                                struct list_head *collect, u64 max)
6595 {
6596         return nfsd_foreach_client_openowner(clp, max, collect,
6597                                                 unhash_openowner_locked);
6598 }
6599
6600 u64
6601 nfsd_inject_print_openowners(void)
6602 {
6603         struct nfs4_client *clp;
6604         u64 count = 0;
6605         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6606                                                 nfsd_net_id);
6607
6608         if (!nfsd_netns_ready(nn))
6609                 return 0;
6610
6611         spin_lock(&nn->client_lock);
6612         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6613                 count += nfsd_print_client_openowners(clp);
6614         spin_unlock(&nn->client_lock);
6615
6616         return count;
6617 }
6618
6619 static void
6620 nfsd_reap_openowners(struct list_head *reaplist)
6621 {
6622         struct nfs4_client *clp;
6623         struct nfs4_openowner *oop, *next;
6624
6625         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6626                 list_del_init(&oop->oo_perclient);
6627                 clp = oop->oo_owner.so_client;
6628                 release_openowner(oop);
6629                 put_client(clp);
6630         }
6631 }
6632
6633 u64
6634 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6635                                      size_t addr_size)
6636 {
6637         unsigned int count = 0;
6638         struct nfs4_client *clp;
6639         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6640                                                 nfsd_net_id);
6641         LIST_HEAD(reaplist);
6642
6643         if (!nfsd_netns_ready(nn))
6644                 return count;
6645
6646         spin_lock(&nn->client_lock);
6647         clp = nfsd_find_client(addr, addr_size);
6648         if (clp)
6649                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6650         spin_unlock(&nn->client_lock);
6651         nfsd_reap_openowners(&reaplist);
6652         return count;
6653 }
6654
6655 u64
6656 nfsd_inject_forget_openowners(u64 max)
6657 {
6658         u64 count = 0;
6659         struct nfs4_client *clp;
6660         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6661                                                 nfsd_net_id);
6662         LIST_HEAD(reaplist);
6663
6664         if (!nfsd_netns_ready(nn))
6665                 return count;
6666
6667         spin_lock(&nn->client_lock);
6668         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6669                 count += nfsd_collect_client_openowners(clp, &reaplist,
6670                                                         max - count);
6671                 if (max != 0 && count >= max)
6672                         break;
6673         }
6674         spin_unlock(&nn->client_lock);
6675         nfsd_reap_openowners(&reaplist);
6676         return count;
6677 }
6678
6679 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6680                                      struct list_head *victims)
6681 {
6682         struct nfs4_delegation *dp, *next;
6683         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6684                                                 nfsd_net_id);
6685         u64 count = 0;
6686
6687         lockdep_assert_held(&nn->client_lock);
6688
6689         spin_lock(&state_lock);
6690         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6691                 if (victims) {
6692                         /*
6693                          * It's not safe to mess with delegations that have a
6694                          * non-zero dl_time. They might have already been broken
6695                          * and could be processed by the laundromat outside of
6696                          * the state_lock. Just leave them be.
6697                          */
6698                         if (dp->dl_time != 0)
6699                                 continue;
6700
6701                         atomic_inc(&clp->cl_refcount);
6702                         WARN_ON(!unhash_delegation_locked(dp));
6703                         list_add(&dp->dl_recall_lru, victims);
6704                 }
6705                 ++count;
6706                 /*
6707                  * Despite the fact that these functions deal with
6708                  * 64-bit integers for "count", we must ensure that
6709                  * it doesn't blow up the clp->cl_refcount. Throw a
6710                  * warning if we start to approach INT_MAX here.
6711                  */
6712                 WARN_ON_ONCE(count == (INT_MAX / 2));
6713                 if (count == max)
6714                         break;
6715         }
6716         spin_unlock(&state_lock);
6717         return count;
6718 }
6719
6720 static u64
6721 nfsd_print_client_delegations(struct nfs4_client *clp)
6722 {
6723         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6724
6725         nfsd_print_count(clp, count, "delegations");
6726         return count;
6727 }
6728
6729 u64
6730 nfsd_inject_print_delegations(void)
6731 {
6732         struct nfs4_client *clp;
6733         u64 count = 0;
6734         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6735                                                 nfsd_net_id);
6736
6737         if (!nfsd_netns_ready(nn))
6738                 return 0;
6739
6740         spin_lock(&nn->client_lock);
6741         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6742                 count += nfsd_print_client_delegations(clp);
6743         spin_unlock(&nn->client_lock);
6744
6745         return count;
6746 }
6747
6748 static void
6749 nfsd_forget_delegations(struct list_head *reaplist)
6750 {
6751         struct nfs4_client *clp;
6752         struct nfs4_delegation *dp, *next;
6753
6754         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6755                 list_del_init(&dp->dl_recall_lru);
6756                 clp = dp->dl_stid.sc_client;
6757                 revoke_delegation(dp);
6758                 put_client(clp);
6759         }
6760 }
6761
6762 u64
6763 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6764                                       size_t addr_size)
6765 {
6766         u64 count = 0;
6767         struct nfs4_client *clp;
6768         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6769                                                 nfsd_net_id);
6770         LIST_HEAD(reaplist);
6771
6772         if (!nfsd_netns_ready(nn))
6773                 return count;
6774
6775         spin_lock(&nn->client_lock);
6776         clp = nfsd_find_client(addr, addr_size);
6777         if (clp)
6778                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6779         spin_unlock(&nn->client_lock);
6780
6781         nfsd_forget_delegations(&reaplist);
6782         return count;
6783 }
6784
6785 u64
6786 nfsd_inject_forget_delegations(u64 max)
6787 {
6788         u64 count = 0;
6789         struct nfs4_client *clp;
6790         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6791                                                 nfsd_net_id);
6792         LIST_HEAD(reaplist);
6793
6794         if (!nfsd_netns_ready(nn))
6795                 return count;
6796
6797         spin_lock(&nn->client_lock);
6798         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6799                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6800                 if (max != 0 && count >= max)
6801                         break;
6802         }
6803         spin_unlock(&nn->client_lock);
6804         nfsd_forget_delegations(&reaplist);
6805         return count;
6806 }
6807
6808 static void
6809 nfsd_recall_delegations(struct list_head *reaplist)
6810 {
6811         struct nfs4_client *clp;
6812         struct nfs4_delegation *dp, *next;
6813
6814         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6815                 list_del_init(&dp->dl_recall_lru);
6816                 clp = dp->dl_stid.sc_client;
6817                 /*
6818                  * We skipped all entries that had a zero dl_time before,
6819                  * so we can now reset the dl_time back to 0. If a delegation
6820                  * break comes in now, then it won't make any difference since
6821                  * we're recalling it either way.
6822                  */
6823                 spin_lock(&state_lock);
6824                 dp->dl_time = 0;
6825                 spin_unlock(&state_lock);
6826                 nfsd_break_one_deleg(dp);
6827                 put_client(clp);
6828         }
6829 }
6830
6831 u64
6832 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6833                                       size_t addr_size)
6834 {
6835         u64 count = 0;
6836         struct nfs4_client *clp;
6837         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6838                                                 nfsd_net_id);
6839         LIST_HEAD(reaplist);
6840
6841         if (!nfsd_netns_ready(nn))
6842                 return count;
6843
6844         spin_lock(&nn->client_lock);
6845         clp = nfsd_find_client(addr, addr_size);
6846         if (clp)
6847                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6848         spin_unlock(&nn->client_lock);
6849
6850         nfsd_recall_delegations(&reaplist);
6851         return count;
6852 }
6853
6854 u64
6855 nfsd_inject_recall_delegations(u64 max)
6856 {
6857         u64 count = 0;
6858         struct nfs4_client *clp, *next;
6859         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6860                                                 nfsd_net_id);
6861         LIST_HEAD(reaplist);
6862
6863         if (!nfsd_netns_ready(nn))
6864                 return count;
6865
6866         spin_lock(&nn->client_lock);
6867         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6868                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6869                 if (max != 0 && ++count >= max)
6870                         break;
6871         }
6872         spin_unlock(&nn->client_lock);
6873         nfsd_recall_delegations(&reaplist);
6874         return count;
6875 }
6876 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6877
6878 /*
6879  * Since the lifetime of a delegation isn't limited to that of an open, a
6880  * client may quite reasonably hang on to a delegation as long as it has
6881  * the inode cached.  This becomes an obvious problem the first time a
6882  * client's inode cache approaches the size of the server's total memory.
6883  *
6884  * For now we avoid this problem by imposing a hard limit on the number
6885  * of delegations, which varies according to the server's memory size.
6886  */
6887 static void
6888 set_max_delegations(void)
6889 {
6890         /*
6891          * Allow at most 4 delegations per megabyte of RAM.  Quick
6892          * estimates suggest that in the worst case (where every delegation
6893          * is for a different inode), a delegation could take about 1.5K,
6894          * giving a worst case usage of about 6% of memory.
6895          */
6896         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6897 }
6898
6899 static int nfs4_state_create_net(struct net *net)
6900 {
6901         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6902         int i;
6903
6904         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6905                         CLIENT_HASH_SIZE, GFP_KERNEL);
6906         if (!nn->conf_id_hashtbl)
6907                 goto err;
6908         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6909                         CLIENT_HASH_SIZE, GFP_KERNEL);
6910         if (!nn->unconf_id_hashtbl)
6911                 goto err_unconf_id;
6912         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6913                         SESSION_HASH_SIZE, GFP_KERNEL);
6914         if (!nn->sessionid_hashtbl)
6915                 goto err_sessionid;
6916
6917         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6918                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6919                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6920         }
6921         for (i = 0; i < SESSION_HASH_SIZE; i++)
6922                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6923         nn->conf_name_tree = RB_ROOT;
6924         nn->unconf_name_tree = RB_ROOT;
6925         INIT_LIST_HEAD(&nn->client_lru);
6926         INIT_LIST_HEAD(&nn->close_lru);
6927         INIT_LIST_HEAD(&nn->del_recall_lru);
6928         spin_lock_init(&nn->client_lock);
6929
6930         spin_lock_init(&nn->blocked_locks_lock);
6931         INIT_LIST_HEAD(&nn->blocked_locks_lru);
6932
6933         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6934         get_net(net);
6935
6936         return 0;
6937
6938 err_sessionid:
6939         kfree(nn->unconf_id_hashtbl);
6940 err_unconf_id:
6941         kfree(nn->conf_id_hashtbl);
6942 err:
6943         return -ENOMEM;
6944 }
6945
6946 static void
6947 nfs4_state_destroy_net(struct net *net)
6948 {
6949         int i;
6950         struct nfs4_client *clp = NULL;
6951         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6952
6953         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6954                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6955                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6956                         destroy_client(clp);
6957                 }
6958         }
6959
6960         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6961                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6962                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6963                         destroy_client(clp);
6964                 }
6965         }
6966
6967         kfree(nn->sessionid_hashtbl);
6968         kfree(nn->unconf_id_hashtbl);
6969         kfree(nn->conf_id_hashtbl);
6970         put_net(net);
6971 }
6972
6973 int
6974 nfs4_state_start_net(struct net *net)
6975 {
6976         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6977         int ret;
6978
6979         ret = nfs4_state_create_net(net);
6980         if (ret)
6981                 return ret;
6982         nn->boot_time = get_seconds();
6983         nn->grace_ended = false;
6984         nn->nfsd4_manager.block_opens = true;
6985         locks_start_grace(net, &nn->nfsd4_manager);
6986         nfsd4_client_tracking_init(net);
6987         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6988                nn->nfsd4_grace, net);
6989         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6990         return 0;
6991 }
6992
6993 /* initialization to perform when the nfsd service is started: */
6994
6995 int
6996 nfs4_state_start(void)
6997 {
6998         int ret;
6999
7000         ret = set_callback_cred();
7001         if (ret)
7002                 return ret;
7003
7004         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7005         if (laundry_wq == NULL) {
7006                 ret = -ENOMEM;
7007                 goto out_cleanup_cred;
7008         }
7009         ret = nfsd4_create_callback_queue();
7010         if (ret)
7011                 goto out_free_laundry;
7012
7013         set_max_delegations();
7014         return 0;
7015
7016 out_free_laundry:
7017         destroy_workqueue(laundry_wq);
7018 out_cleanup_cred:
7019         cleanup_callback_cred();
7020         return ret;
7021 }
7022
7023 void
7024 nfs4_state_shutdown_net(struct net *net)
7025 {
7026         struct nfs4_delegation *dp = NULL;
7027         struct list_head *pos, *next, reaplist;
7028         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7029         struct nfsd4_blocked_lock *nbl;
7030
7031         cancel_delayed_work_sync(&nn->laundromat_work);
7032         locks_end_grace(&nn->nfsd4_manager);
7033
7034         INIT_LIST_HEAD(&reaplist);
7035         spin_lock(&state_lock);
7036         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7037                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7038                 WARN_ON(!unhash_delegation_locked(dp));
7039                 list_add(&dp->dl_recall_lru, &reaplist);
7040         }
7041         spin_unlock(&state_lock);
7042         list_for_each_safe(pos, next, &reaplist) {
7043                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7044                 list_del_init(&dp->dl_recall_lru);
7045                 put_clnt_odstate(dp->dl_clnt_odstate);
7046                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7047                 nfs4_put_stid(&dp->dl_stid);
7048         }
7049
7050         BUG_ON(!list_empty(&reaplist));
7051         spin_lock(&nn->blocked_locks_lock);
7052         while (!list_empty(&nn->blocked_locks_lru)) {
7053                 nbl = list_first_entry(&nn->blocked_locks_lru,
7054                                         struct nfsd4_blocked_lock, nbl_lru);
7055                 list_move(&nbl->nbl_lru, &reaplist);
7056                 list_del_init(&nbl->nbl_list);
7057         }
7058         spin_unlock(&nn->blocked_locks_lock);
7059
7060         while (!list_empty(&reaplist)) {
7061                 nbl = list_first_entry(&nn->blocked_locks_lru,
7062                                         struct nfsd4_blocked_lock, nbl_lru);
7063                 list_del_init(&nbl->nbl_lru);
7064                 posix_unblock_lock(&nbl->nbl_lock);
7065                 free_blocked_lock(nbl);
7066         }
7067
7068         nfsd4_client_tracking_exit(net);
7069         nfs4_state_destroy_net(net);
7070 }
7071
7072 void
7073 nfs4_state_shutdown(void)
7074 {
7075         destroy_workqueue(laundry_wq);
7076         nfsd4_destroy_callback_queue();
7077         cleanup_callback_cred();
7078 }
7079
7080 static void
7081 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7082 {
7083         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7084                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7085 }
7086
7087 static void
7088 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7089 {
7090         if (cstate->minorversion) {
7091                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7092                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7093         }
7094 }
7095
7096 void
7097 clear_current_stateid(struct nfsd4_compound_state *cstate)
7098 {
7099         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7100 }
7101
7102 /*
7103  * functions to set current state id
7104  */
7105 void
7106 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7107 {
7108         put_stateid(cstate, &odp->od_stateid);
7109 }
7110
7111 void
7112 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
7113 {
7114         put_stateid(cstate, &open->op_stateid);
7115 }
7116
7117 void
7118 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7119 {
7120         put_stateid(cstate, &close->cl_stateid);
7121 }
7122
7123 void
7124 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
7125 {
7126         put_stateid(cstate, &lock->lk_resp_stateid);
7127 }
7128
7129 /*
7130  * functions to consume current state id
7131  */
7132
7133 void
7134 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7135 {
7136         get_stateid(cstate, &odp->od_stateid);
7137 }
7138
7139 void
7140 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
7141 {
7142         get_stateid(cstate, &drp->dr_stateid);
7143 }
7144
7145 void
7146 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
7147 {
7148         get_stateid(cstate, &fsp->fr_stateid);
7149 }
7150
7151 void
7152 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
7153 {
7154         get_stateid(cstate, &setattr->sa_stateid);
7155 }
7156
7157 void
7158 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7159 {
7160         get_stateid(cstate, &close->cl_stateid);
7161 }
7162
7163 void
7164 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
7165 {
7166         get_stateid(cstate, &locku->lu_stateid);
7167 }
7168
7169 void
7170 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
7171 {
7172         get_stateid(cstate, &read->rd_stateid);
7173 }
7174
7175 void
7176 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
7177 {
7178         get_stateid(cstate, &write->wr_stateid);
7179 }