fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include <linux/jhash.h>
  45 #include "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51 #include "pnfs.h"
  52
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 #define all_ones {{~0,~0},~0}
  56 static const stateid_t one_stateid = {
  57         .si_generation = ~0,
  58         .si_opaque = all_ones,
  59 };
  60 static const stateid_t zero_stateid = {
  61         /* all fields zero */
  62 };
  63 static const stateid_t currentstateid = {
  64         .si_generation = 1,
  65 };
  66 static const stateid_t close_stateid = {
  67         .si_generation = 0xffffffffU,
  68 };
  69
  70 static u64 current_sessionid = 1;
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  75 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  76
  77 /* forward declarations */
  78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  80 void nfsd4_end_grace(struct nfsd_net *nn);
  81
  82 /* Locking: */
  83
  84 /*
  85  * Currently used for the del_recall_lru and file hash table.  In an
  86  * effort to decrease the scope of the client_mutex, this spinlock may
  87  * eventually cover more:
  88  */
  89 static DEFINE_SPINLOCK(state_lock);
  90
  91 enum nfsd4_st_mutex_lock_subclass {
  92         OPEN_STATEID_MUTEX = 0,
  93         LOCK_STATEID_MUTEX = 1,
  94 };
  95
  96 /*
  97  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  98  * the refcount on the open stateid to drop.
  99  */
 100 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 101
 102 static struct kmem_cache *client_slab;
 103 static struct kmem_cache *openowner_slab;
 104 static struct kmem_cache *lockowner_slab;
 105 static struct kmem_cache *file_slab;
 106 static struct kmem_cache *stateid_slab;
 107 static struct kmem_cache *deleg_slab;
 108 static struct kmem_cache *odstate_slab;
 109
 110 static void free_session(struct nfsd4_session *);
 111
 112 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 113 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 114
 115 static bool is_session_dead(struct nfsd4_session *ses)
 116 {
 117         return ses->se_flags & NFS4_SESSION_DEAD;
 118 }
 119
 120 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 121 {
 122         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 123                 return nfserr_jukebox;
 124         ses->se_flags |= NFS4_SESSION_DEAD;
 125         return nfs_ok;
 126 }
 127
 128 static bool is_client_expired(struct nfs4_client *clp)
 129 {
 130         return clp->cl_time == 0;
 131 }
 132
 133 static __be32 get_client_locked(struct nfs4_client *clp)
 134 {
 135         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 136
 137         lockdep_assert_held(&nn->client_lock);
 138
 139         if (is_client_expired(clp))
 140                 return nfserr_expired;
 141         atomic_inc(&clp->cl_refcount);
 142         return nfs_ok;
 143 }
 144
 145 /* must be called under the client_lock */
 146 static inline void
 147 renew_client_locked(struct nfs4_client *clp)
 148 {
 149         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 150
 151         if (is_client_expired(clp)) {
 152                 WARN_ON(1);
 153                 printk("%s: client (clientid %08x/%08x) already expired\n",
 154                         __func__,
 155                         clp->cl_clientid.cl_boot,
 156                         clp->cl_clientid.cl_id);
 157                 return;
 158         }
 159
 160         dprintk("renewing client (clientid %08x/%08x)\n",
 161                         clp->cl_clientid.cl_boot,
 162                         clp->cl_clientid.cl_id);
 163         list_move_tail(&clp->cl_lru, &nn->client_lru);
 164         clp->cl_time = get_seconds();
 165 }
 166
 167 static void put_client_renew_locked(struct nfs4_client *clp)
 168 {
 169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 170
 171         lockdep_assert_held(&nn->client_lock);
 172
 173         if (!atomic_dec_and_test(&clp->cl_refcount))
 174                 return;
 175         if (!is_client_expired(clp))
 176                 renew_client_locked(clp);
 177 }
 178
 179 static void put_client_renew(struct nfs4_client *clp)
 180 {
 181         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 182
 183         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 184                 return;
 185         if (!is_client_expired(clp))
 186                 renew_client_locked(clp);
 187         spin_unlock(&nn->client_lock);
 188 }
 189
 190 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 191 {
 192         __be32 status;
 193
 194         if (is_session_dead(ses))
 195                 return nfserr_badsession;
 196         status = get_client_locked(ses->se_client);
 197         if (status)
 198                 return status;
 199         atomic_inc(&ses->se_ref);
 200         return nfs_ok;
 201 }
 202
 203 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 204 {
 205         struct nfs4_client *clp = ses->se_client;
 206         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 207
 208         lockdep_assert_held(&nn->client_lock);
 209
 210         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 211                 free_session(ses);
 212         put_client_renew_locked(clp);
 213 }
 214
 215 static void nfsd4_put_session(struct nfsd4_session *ses)
 216 {
 217         struct nfs4_client *clp = ses->se_client;
 218         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 219
 220         spin_lock(&nn->client_lock);
 221         nfsd4_put_session_locked(ses);
 222         spin_unlock(&nn->client_lock);
 223 }
 224
 225 static struct nfsd4_blocked_lock *
 226 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 227                         struct nfsd_net *nn)
 228 {
 229         struct nfsd4_blocked_lock *cur, *found = NULL;
 230
 231         spin_lock(&nn->blocked_locks_lock);
 232         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 233                 if (fh_match(fh, &cur->nbl_fh)) {
 234                         list_del_init(&cur->nbl_list);
 235                         list_del_init(&cur->nbl_lru);
 236                         found = cur;
 237                         break;
 238                 }
 239         }
 240         spin_unlock(&nn->blocked_locks_lock);
 241         if (found)
 242                 locks_delete_block(&found->nbl_lock);
 243         return found;
 244 }
 245
 246 static struct nfsd4_blocked_lock *
 247 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 248                         struct nfsd_net *nn)
 249 {
 250         struct nfsd4_blocked_lock *nbl;
 251
 252         nbl = find_blocked_lock(lo, fh, nn);
 253         if (!nbl) {
 254                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 255                 if (nbl) {
 256                         fh_copy_shallow(&nbl->nbl_fh, fh);
 257                         locks_init_lock(&nbl->nbl_lock);
 258                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 259                                         &nfsd4_cb_notify_lock_ops,
 260                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 261                 }
 262         }
 263         return nbl;
 264 }
 265
 266 static void
 267 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 268 {
 269         locks_delete_block(&nbl->nbl_lock);
 270         locks_release_private(&nbl->nbl_lock);
 271         kfree(nbl);
 272 }
 273
 274 static void
 275 remove_blocked_locks(struct nfs4_lockowner *lo)
 276 {
 277         struct nfs4_client *clp = lo->lo_owner.so_client;
 278         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 279         struct nfsd4_blocked_lock *nbl;
 280         LIST_HEAD(reaplist);
 281
 282         /* Dequeue all blocked locks */
 283         spin_lock(&nn->blocked_locks_lock);
 284         while (!list_empty(&lo->lo_blocked)) {
 285                 nbl = list_first_entry(&lo->lo_blocked,
 286                                         struct nfsd4_blocked_lock,
 287                                         nbl_list);
 288                 list_del_init(&nbl->nbl_list);
 289                 list_move(&nbl->nbl_lru, &reaplist);
 290         }
 291         spin_unlock(&nn->blocked_locks_lock);
 292
 293         /* Now free them */
 294         while (!list_empty(&reaplist)) {
 295                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
 296                                         nbl_lru);
 297                 list_del_init(&nbl->nbl_lru);
 298                 free_blocked_lock(nbl);
 299         }
 300 }
 301
 302 static void
 303 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
 304 {
 305         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 306                                                 struct nfsd4_blocked_lock, nbl_cb);
 307         locks_delete_block(&nbl->nbl_lock);
 308 }
 309
 310 static int
 311 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 312 {
 313         /*
 314          * Since this is just an optimization, we don't try very hard if it
 315          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 316          * just quit trying on anything else.
 317          */
 318         switch (task->tk_status) {
 319         case -NFS4ERR_DELAY:
 320                 rpc_delay(task, 1 * HZ);
 321                 return 0;
 322         default:
 323                 return 1;
 324         }
 325 }
 326
 327 static void
 328 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 329 {
 330         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 331                                                 struct nfsd4_blocked_lock, nbl_cb);
 332
 333         free_blocked_lock(nbl);
 334 }
 335
 336 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 337         .prepare        = nfsd4_cb_notify_lock_prepare,
 338         .done           = nfsd4_cb_notify_lock_done,
 339         .release        = nfsd4_cb_notify_lock_release,
 340 };
 341
 342 static inline struct nfs4_stateowner *
 343 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 344 {
 345         atomic_inc(&sop->so_count);
 346         return sop;
 347 }
 348
 349 static int
 350 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 351 {
 352         return (sop->so_owner.len == owner->len) &&
 353                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 354 }
 355
 356 static struct nfs4_openowner *
 357 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 358                         struct nfs4_client *clp)
 359 {
 360         struct nfs4_stateowner *so;
 361
 362         lockdep_assert_held(&clp->cl_lock);
 363
 364         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 365                             so_strhash) {
 366                 if (!so->so_is_open_owner)
 367                         continue;
 368                 if (same_owner_str(so, &open->op_owner))
 369                         return openowner(nfs4_get_stateowner(so));
 370         }
 371         return NULL;
 372 }
 373
 374 static struct nfs4_openowner *
 375 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 376                         struct nfs4_client *clp)
 377 {
 378         struct nfs4_openowner *oo;
 379
 380         spin_lock(&clp->cl_lock);
 381         oo = find_openstateowner_str_locked(hashval, open, clp);
 382         spin_unlock(&clp->cl_lock);
 383         return oo;
 384 }
 385
 386 static inline u32
 387 opaque_hashval(const void *ptr, int nbytes)
 388 {
 389         unsigned char *cptr = (unsigned char *) ptr;
 390
 391         u32 x = 0;
 392         while (nbytes--) {
 393                 x *= 37;
 394                 x += *cptr++;
 395         }
 396         return x;
 397 }
 398
 399 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 400 {
 401         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 402
 403         kmem_cache_free(file_slab, fp);
 404 }
 405
 406 void
 407 put_nfs4_file(struct nfs4_file *fi)
 408 {
 409         might_lock(&state_lock);
 410
 411         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 412                 hlist_del_rcu(&fi->fi_hash);
 413                 spin_unlock(&state_lock);
 414                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 415                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 416                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 417         }
 418 }
 419
 420 static struct file *
 421 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 422 {
 423         if (f->fi_fds[oflag])
 424                 return get_file(f->fi_fds[oflag]);
 425         return NULL;
 426 }
 427
 428 static struct file *
 429 find_writeable_file_locked(struct nfs4_file *f)
 430 {
 431         struct file *ret;
 432
 433         lockdep_assert_held(&f->fi_lock);
 434
 435         ret = __nfs4_get_fd(f, O_WRONLY);
 436         if (!ret)
 437                 ret = __nfs4_get_fd(f, O_RDWR);
 438         return ret;
 439 }
 440
 441 static struct file *
 442 find_writeable_file(struct nfs4_file *f)
 443 {
 444         struct file *ret;
 445
 446         spin_lock(&f->fi_lock);
 447         ret = find_writeable_file_locked(f);
 448         spin_unlock(&f->fi_lock);
 449
 450         return ret;
 451 }
 452
 453 static struct file *find_readable_file_locked(struct nfs4_file *f)
 454 {
 455         struct file *ret;
 456
 457         lockdep_assert_held(&f->fi_lock);
 458
 459         ret = __nfs4_get_fd(f, O_RDONLY);
 460         if (!ret)
 461                 ret = __nfs4_get_fd(f, O_RDWR);
 462         return ret;
 463 }
 464
 465 static struct file *
 466 find_readable_file(struct nfs4_file *f)
 467 {
 468         struct file *ret;
 469
 470         spin_lock(&f->fi_lock);
 471         ret = find_readable_file_locked(f);
 472         spin_unlock(&f->fi_lock);
 473
 474         return ret;
 475 }
 476
 477 struct file *
 478 find_any_file(struct nfs4_file *f)
 479 {
 480         struct file *ret;
 481
 482         spin_lock(&f->fi_lock);
 483         ret = __nfs4_get_fd(f, O_RDWR);
 484         if (!ret) {
 485                 ret = __nfs4_get_fd(f, O_WRONLY);
 486                 if (!ret)
 487                         ret = __nfs4_get_fd(f, O_RDONLY);
 488         }
 489         spin_unlock(&f->fi_lock);
 490         return ret;
 491 }
 492
 493 static atomic_long_t num_delegations;
 494 unsigned long max_delegations;
 495
 496 /*
 497  * Open owner state (share locks)
 498  */
 499
 500 /* hash tables for lock and open owners */
 501 #define OWNER_HASH_BITS              8
 502 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 503 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 504
 505 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 506 {
 507         unsigned int ret;
 508
 509         ret = opaque_hashval(ownername->data, ownername->len);
 510         return ret & OWNER_HASH_MASK;
 511 }
 512
 513 /* hash table for nfs4_file */
 514 #define FILE_HASH_BITS                   8
 515 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 516
 517 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 518 {
 519         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 520 }
 521
 522 static unsigned int file_hashval(struct knfsd_fh *fh)
 523 {
 524         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 525 }
 526
 527 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 528
 529 static void
 530 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 531 {
 532         lockdep_assert_held(&fp->fi_lock);
 533
 534         if (access & NFS4_SHARE_ACCESS_WRITE)
 535                 atomic_inc(&fp->fi_access[O_WRONLY]);
 536         if (access & NFS4_SHARE_ACCESS_READ)
 537                 atomic_inc(&fp->fi_access[O_RDONLY]);
 538 }
 539
 540 static __be32
 541 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 542 {
 543         lockdep_assert_held(&fp->fi_lock);
 544
 545         /* Does this access mode make sense? */
 546         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 547                 return nfserr_inval;
 548
 549         /* Does it conflict with a deny mode already set? */
 550         if ((access & fp->fi_share_deny) != 0)
 551                 return nfserr_share_denied;
 552
 553         __nfs4_file_get_access(fp, access);
 554         return nfs_ok;
 555 }
 556
 557 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 558 {
 559         /* Common case is that there is no deny mode. */
 560         if (deny) {
 561                 /* Does this deny mode make sense? */
 562                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 563                         return nfserr_inval;
 564
 565                 if ((deny & NFS4_SHARE_DENY_READ) &&
 566                     atomic_read(&fp->fi_access[O_RDONLY]))
 567                         return nfserr_share_denied;
 568
 569                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 570                     atomic_read(&fp->fi_access[O_WRONLY]))
 571                         return nfserr_share_denied;
 572         }
 573         return nfs_ok;
 574 }
 575
 576 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 577 {
 578         might_lock(&fp->fi_lock);
 579
 580         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 581                 struct file *f1 = NULL;
 582                 struct file *f2 = NULL;
 583
 584                 swap(f1, fp->fi_fds[oflag]);
 585                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 586                         swap(f2, fp->fi_fds[O_RDWR]);
 587                 spin_unlock(&fp->fi_lock);
 588                 if (f1)
 589                         fput(f1);
 590                 if (f2)
 591                         fput(f2);
 592         }
 593 }
 594
 595 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 596 {
 597         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 598
 599         if (access & NFS4_SHARE_ACCESS_WRITE)
 600                 __nfs4_file_put_access(fp, O_WRONLY);
 601         if (access & NFS4_SHARE_ACCESS_READ)
 602                 __nfs4_file_put_access(fp, O_RDONLY);
 603 }
 604
 605 /*
 606  * Allocate a new open/delegation state counter. This is needed for
 607  * pNFS for proper return on close semantics.
 608  *
 609  * Note that we only allocate it for pNFS-enabled exports, otherwise
 610  * all pointers to struct nfs4_clnt_odstate are always NULL.
 611  */
 612 static struct nfs4_clnt_odstate *
 613 alloc_clnt_odstate(struct nfs4_client *clp)
 614 {
 615         struct nfs4_clnt_odstate *co;
 616
 617         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 618         if (co) {
 619                 co->co_client = clp;
 620                 refcount_set(&co->co_odcount, 1);
 621         }
 622         return co;
 623 }
 624
 625 static void
 626 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 627 {
 628         struct nfs4_file *fp = co->co_file;
 629
 630         lockdep_assert_held(&fp->fi_lock);
 631         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 632 }
 633
 634 static inline void
 635 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 636 {
 637         if (co)
 638                 refcount_inc(&co->co_odcount);
 639 }
 640
 641 static void
 642 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 643 {
 644         struct nfs4_file *fp;
 645
 646         if (!co)
 647                 return;
 648
 649         fp = co->co_file;
 650         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 651                 list_del(&co->co_perfile);
 652                 spin_unlock(&fp->fi_lock);
 653
 654                 nfsd4_return_all_file_layouts(co->co_client, fp);
 655                 kmem_cache_free(odstate_slab, co);
 656         }
 657 }
 658
 659 static struct nfs4_clnt_odstate *
 660 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 661 {
 662         struct nfs4_clnt_odstate *co;
 663         struct nfs4_client *cl;
 664
 665         if (!new)
 666                 return NULL;
 667
 668         cl = new->co_client;
 669
 670         spin_lock(&fp->fi_lock);
 671         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 672                 if (co->co_client == cl) {
 673                         get_clnt_odstate(co);
 674                         goto out;
 675                 }
 676         }
 677         co = new;
 678         co->co_file = fp;
 679         hash_clnt_odstate_locked(new);
 680 out:
 681         spin_unlock(&fp->fi_lock);
 682         return co;
 683 }
 684
 685 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 686                                   void (*sc_free)(struct nfs4_stid *))
 687 {
 688         struct nfs4_stid *stid;
 689         int new_id;
 690
 691         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 692         if (!stid)
 693                 return NULL;
 694
 695         idr_preload(GFP_KERNEL);
 696         spin_lock(&cl->cl_lock);
 697         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 698         spin_unlock(&cl->cl_lock);
 699         idr_preload_end();
 700         if (new_id < 0)
 701                 goto out_free;
 702
 703         stid->sc_free = sc_free;
 704         stid->sc_client = cl;
 705         stid->sc_stateid.si_opaque.so_id = new_id;
 706         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 707         /* Will be incremented before return to client: */
 708         refcount_set(&stid->sc_count, 1);
 709         spin_lock_init(&stid->sc_lock);
 710
 711         /*
 712          * It shouldn't be a problem to reuse an opaque stateid value.
 713          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 714          * example, a stray write retransmission could be accepted by
 715          * the server when it should have been rejected.  Therefore,
 716          * adopt a trick from the sctp code to attempt to maximize the
 717          * amount of time until an id is reused, by ensuring they always
 718          * "increase" (mod INT_MAX):
 719          */
 720         return stid;
 721 out_free:
 722         kmem_cache_free(slab, stid);
 723         return NULL;
 724 }
 725
 726 /*
 727  * Create a unique stateid_t to represent each COPY.
 728  */
 729 int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
 730 {
 731         int new_id;
 732
 733         idr_preload(GFP_KERNEL);
 734         spin_lock(&nn->s2s_cp_lock);
 735         new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, copy, 0, 0, GFP_NOWAIT);
 736         spin_unlock(&nn->s2s_cp_lock);
 737         idr_preload_end();
 738         if (new_id < 0)
 739                 return 0;
 740         copy->cp_stateid.si_opaque.so_id = new_id;
 741         copy->cp_stateid.si_opaque.so_clid.cl_boot = nn->boot_time;
 742         copy->cp_stateid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
 743         return 1;
 744 }
 745
 746 void nfs4_free_cp_state(struct nfsd4_copy *copy)
 747 {
 748         struct nfsd_net *nn;
 749
 750         nn = net_generic(copy->cp_clp->net, nfsd_net_id);
 751         spin_lock(&nn->s2s_cp_lock);
 752         idr_remove(&nn->s2s_cp_stateids, copy->cp_stateid.si_opaque.so_id);
 753         spin_unlock(&nn->s2s_cp_lock);
 754 }
 755
 756 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 757 {
 758         struct nfs4_stid *stid;
 759
 760         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 761         if (!stid)
 762                 return NULL;
 763
 764         return openlockstateid(stid);
 765 }
 766
 767 static void nfs4_free_deleg(struct nfs4_stid *stid)
 768 {
 769         kmem_cache_free(deleg_slab, stid);
 770         atomic_long_dec(&num_delegations);
 771 }
 772
 773 /*
 774  * When we recall a delegation, we should be careful not to hand it
 775  * out again straight away.
 776  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 777  * in which the filehandles of recalled delegations are "stored".
 778  * If a filehandle appear in either filter, a delegation is blocked.
 779  * When a delegation is recalled, the filehandle is stored in the "new"
 780  * filter.
 781  * Every 30 seconds we swap the filters and clear the "new" one,
 782  * unless both are empty of course.
 783  *
 784  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 785  * low 3 bytes as hash-table indices.
 786  *
 787  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 788  * is used to manage concurrent access.  Testing does not need the lock
 789  * except when swapping the two filters.
 790  */
 791 static DEFINE_SPINLOCK(blocked_delegations_lock);
 792 static struct bloom_pair {
 793         int     entries, old_entries;
 794         time_t  swap_time;
 795         int     new; /* index into 'set' */
 796         DECLARE_BITMAP(set[2], 256);
 797 } blocked_delegations;
 798
 799 static int delegation_blocked(struct knfsd_fh *fh)
 800 {
 801         u32 hash;
 802         struct bloom_pair *bd = &blocked_delegations;
 803
 804         if (bd->entries == 0)
 805                 return 0;
 806         if (seconds_since_boot() - bd->swap_time > 30) {
 807                 spin_lock(&blocked_delegations_lock);
 808                 if (seconds_since_boot() - bd->swap_time > 30) {
 809                         bd->entries -= bd->old_entries;
 810                         bd->old_entries = bd->entries;
 811                         memset(bd->set[bd->new], 0,
 812                                sizeof(bd->set[0]));
 813                         bd->new = 1-bd->new;
 814                         bd->swap_time = seconds_since_boot();
 815                 }
 816                 spin_unlock(&blocked_delegations_lock);
 817         }
 818         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 819         if (test_bit(hash&255, bd->set[0]) &&
 820             test_bit((hash>>8)&255, bd->set[0]) &&
 821             test_bit((hash>>16)&255, bd->set[0]))
 822                 return 1;
 823
 824         if (test_bit(hash&255, bd->set[1]) &&
 825             test_bit((hash>>8)&255, bd->set[1]) &&
 826             test_bit((hash>>16)&255, bd->set[1]))
 827                 return 1;
 828
 829         return 0;
 830 }
 831
 832 static void block_delegations(struct knfsd_fh *fh)
 833 {
 834         u32 hash;
 835         struct bloom_pair *bd = &blocked_delegations;
 836
 837         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 838
 839         spin_lock(&blocked_delegations_lock);
 840         __set_bit(hash&255, bd->set[bd->new]);
 841         __set_bit((hash>>8)&255, bd->set[bd->new]);
 842         __set_bit((hash>>16)&255, bd->set[bd->new]);
 843         if (bd->entries == 0)
 844                 bd->swap_time = seconds_since_boot();
 845         bd->entries += 1;
 846         spin_unlock(&blocked_delegations_lock);
 847 }
 848
 849 static struct nfs4_delegation *
 850 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
 851                  struct svc_fh *current_fh,
 852                  struct nfs4_clnt_odstate *odstate)
 853 {
 854         struct nfs4_delegation *dp;
 855         long n;
 856
 857         dprintk("NFSD alloc_init_deleg\n");
 858         n = atomic_long_inc_return(&num_delegations);
 859         if (n < 0 || n > max_delegations)
 860                 goto out_dec;
 861         if (delegation_blocked(&current_fh->fh_handle))
 862                 goto out_dec;
 863         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 864         if (dp == NULL)
 865                 goto out_dec;
 866
 867         /*
 868          * delegation seqid's are never incremented.  The 4.1 special
 869          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 870          * 0 anyway just for consistency and use 1:
 871          */
 872         dp->dl_stid.sc_stateid.si_generation = 1;
 873         INIT_LIST_HEAD(&dp->dl_perfile);
 874         INIT_LIST_HEAD(&dp->dl_perclnt);
 875         INIT_LIST_HEAD(&dp->dl_recall_lru);
 876         dp->dl_clnt_odstate = odstate;
 877         get_clnt_odstate(odstate);
 878         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 879         dp->dl_retries = 1;
 880         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 881                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 882         get_nfs4_file(fp);
 883         dp->dl_stid.sc_file = fp;
 884         return dp;
 885 out_dec:
 886         atomic_long_dec(&num_delegations);
 887         return NULL;
 888 }
 889
 890 void
 891 nfs4_put_stid(struct nfs4_stid *s)
 892 {
 893         struct nfs4_file *fp = s->sc_file;
 894         struct nfs4_client *clp = s->sc_client;
 895
 896         might_lock(&clp->cl_lock);
 897
 898         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 899                 wake_up_all(&close_wq);
 900                 return;
 901         }
 902         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 903         spin_unlock(&clp->cl_lock);
 904         s->sc_free(s);
 905         if (fp)
 906                 put_nfs4_file(fp);
 907 }
 908
 909 void
 910 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 911 {
 912         stateid_t *src = &stid->sc_stateid;
 913
 914         spin_lock(&stid->sc_lock);
 915         if (unlikely(++src->si_generation == 0))
 916                 src->si_generation = 1;
 917         memcpy(dst, src, sizeof(*dst));
 918         spin_unlock(&stid->sc_lock);
 919 }
 920
 921 static void put_deleg_file(struct nfs4_file *fp)
 922 {
 923         struct file *filp = NULL;
 924
 925         spin_lock(&fp->fi_lock);
 926         if (--fp->fi_delegees == 0)
 927                 swap(filp, fp->fi_deleg_file);
 928         spin_unlock(&fp->fi_lock);
 929
 930         if (filp)
 931                 fput(filp);
 932 }
 933
 934 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
 935 {
 936         struct nfs4_file *fp = dp->dl_stid.sc_file;
 937         struct file *filp = fp->fi_deleg_file;
 938
 939         WARN_ON_ONCE(!fp->fi_delegees);
 940
 941         vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
 942         put_deleg_file(fp);
 943 }
 944
 945 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
 946 {
 947         put_clnt_odstate(dp->dl_clnt_odstate);
 948         nfs4_unlock_deleg_lease(dp);
 949         nfs4_put_stid(&dp->dl_stid);
 950 }
 951
 952 void nfs4_unhash_stid(struct nfs4_stid *s)
 953 {
 954         s->sc_type = 0;
 955 }
 956
 957 /**
 958  * nfs4_delegation_exists - Discover if this delegation already exists
 959  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 960  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 961  *
 962  * Return:
 963  *      On success: true iff an existing delegation is found
 964  */
 965
 966 static bool
 967 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
 968 {
 969         struct nfs4_delegation *searchdp = NULL;
 970         struct nfs4_client *searchclp = NULL;
 971
 972         lockdep_assert_held(&state_lock);
 973         lockdep_assert_held(&fp->fi_lock);
 974
 975         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 976                 searchclp = searchdp->dl_stid.sc_client;
 977                 if (clp == searchclp) {
 978                         return true;
 979                 }
 980         }
 981         return false;
 982 }
 983
 984 /**
 985  * hash_delegation_locked - Add a delegation to the appropriate lists
 986  * @dp:     a pointer to the nfs4_delegation we are adding.
 987  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 988  *
 989  * Return:
 990  *      On success: NULL if the delegation was successfully hashed.
 991  *
 992  *      On error: -EAGAIN if one was previously granted to this
 993  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 994  *
 995  */
 996
 997 static int
 998 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 999 {
1000         struct nfs4_client *clp = dp->dl_stid.sc_client;
1001
1002         lockdep_assert_held(&state_lock);
1003         lockdep_assert_held(&fp->fi_lock);
1004
1005         if (nfs4_delegation_exists(clp, fp))
1006                 return -EAGAIN;
1007         refcount_inc(&dp->dl_stid.sc_count);
1008         dp->dl_stid.sc_type = NFS4_DELEG_STID;
1009         list_add(&dp->dl_perfile, &fp->fi_delegations);
1010         list_add(&dp->dl_perclnt, &clp->cl_delegations);
1011         return 0;
1012 }
1013
1014 static bool
1015 unhash_delegation_locked(struct nfs4_delegation *dp)
1016 {
1017         struct nfs4_file *fp = dp->dl_stid.sc_file;
1018
1019         lockdep_assert_held(&state_lock);
1020
1021         if (list_empty(&dp->dl_perfile))
1022                 return false;
1023
1024         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1025         /* Ensure that deleg break won't try to requeue it */
1026         ++dp->dl_time;
1027         spin_lock(&fp->fi_lock);
1028         list_del_init(&dp->dl_perclnt);
1029         list_del_init(&dp->dl_recall_lru);
1030         list_del_init(&dp->dl_perfile);
1031         spin_unlock(&fp->fi_lock);
1032         return true;
1033 }
1034
1035 static void destroy_delegation(struct nfs4_delegation *dp)
1036 {
1037         bool unhashed;
1038
1039         spin_lock(&state_lock);
1040         unhashed = unhash_delegation_locked(dp);
1041         spin_unlock(&state_lock);
1042         if (unhashed)
1043                 destroy_unhashed_deleg(dp);
1044 }
1045
1046 static void revoke_delegation(struct nfs4_delegation *dp)
1047 {
1048         struct nfs4_client *clp = dp->dl_stid.sc_client;
1049
1050         WARN_ON(!list_empty(&dp->dl_recall_lru));
1051
1052         if (clp->cl_minorversion) {
1053                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1054                 refcount_inc(&dp->dl_stid.sc_count);
1055                 spin_lock(&clp->cl_lock);
1056                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1057                 spin_unlock(&clp->cl_lock);
1058         }
1059         destroy_unhashed_deleg(dp);
1060 }
1061
1062 /*
1063  * SETCLIENTID state
1064  */
1065
1066 static unsigned int clientid_hashval(u32 id)
1067 {
1068         return id & CLIENT_HASH_MASK;
1069 }
1070
1071 static unsigned int clientstr_hashval(struct xdr_netobj name)
1072 {
1073         return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1074 }
1075
1076 /*
1077  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1078  * st_{access,deny}_bmap field of the stateid, in order to track not
1079  * only what share bits are currently in force, but also what
1080  * combinations of share bits previous opens have used.  This allows us
1081  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1082  * return an error if the client attempt to downgrade to a combination
1083  * of share bits not explicable by closing some of its previous opens.
1084  *
1085  * XXX: This enforcement is actually incomplete, since we don't keep
1086  * track of access/deny bit combinations; so, e.g., we allow:
1087  *
1088  *      OPEN allow read, deny write
1089  *      OPEN allow both, deny none
1090  *      DOWNGRADE allow read, deny none
1091  *
1092  * which we should reject.
1093  */
1094 static unsigned int
1095 bmap_to_share_mode(unsigned long bmap) {
1096         int i;
1097         unsigned int access = 0;
1098
1099         for (i = 1; i < 4; i++) {
1100                 if (test_bit(i, &bmap))
1101                         access |= i;
1102         }
1103         return access;
1104 }
1105
1106 /* set share access for a given stateid */
1107 static inline void
1108 set_access(u32 access, struct nfs4_ol_stateid *stp)
1109 {
1110         unsigned char mask = 1 << access;
1111
1112         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1113         stp->st_access_bmap |= mask;
1114 }
1115
1116 /* clear share access for a given stateid */
1117 static inline void
1118 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1119 {
1120         unsigned char mask = 1 << access;
1121
1122         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1123         stp->st_access_bmap &= ~mask;
1124 }
1125
1126 /* test whether a given stateid has access */
1127 static inline bool
1128 test_access(u32 access, struct nfs4_ol_stateid *stp)
1129 {
1130         unsigned char mask = 1 << access;
1131
1132         return (bool)(stp->st_access_bmap & mask);
1133 }
1134
1135 /* set share deny for a given stateid */
1136 static inline void
1137 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1138 {
1139         unsigned char mask = 1 << deny;
1140
1141         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1142         stp->st_deny_bmap |= mask;
1143 }
1144
1145 /* clear share deny for a given stateid */
1146 static inline void
1147 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1148 {
1149         unsigned char mask = 1 << deny;
1150
1151         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1152         stp->st_deny_bmap &= ~mask;
1153 }
1154
1155 /* test whether a given stateid is denying specific access */
1156 static inline bool
1157 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1158 {
1159         unsigned char mask = 1 << deny;
1160
1161         return (bool)(stp->st_deny_bmap & mask);
1162 }
1163
1164 static int nfs4_access_to_omode(u32 access)
1165 {
1166         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1167         case NFS4_SHARE_ACCESS_READ:
1168                 return O_RDONLY;
1169         case NFS4_SHARE_ACCESS_WRITE:
1170                 return O_WRONLY;
1171         case NFS4_SHARE_ACCESS_BOTH:
1172                 return O_RDWR;
1173         }
1174         WARN_ON_ONCE(1);
1175         return O_RDONLY;
1176 }
1177
1178 /*
1179  * A stateid that had a deny mode associated with it is being released
1180  * or downgraded. Recalculate the deny mode on the file.
1181  */
1182 static void
1183 recalculate_deny_mode(struct nfs4_file *fp)
1184 {
1185         struct nfs4_ol_stateid *stp;
1186
1187         spin_lock(&fp->fi_lock);
1188         fp->fi_share_deny = 0;
1189         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1190                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1191         spin_unlock(&fp->fi_lock);
1192 }
1193
1194 static void
1195 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1196 {
1197         int i;
1198         bool change = false;
1199
1200         for (i = 1; i < 4; i++) {
1201                 if ((i & deny) != i) {
1202                         change = true;
1203                         clear_deny(i, stp);
1204                 }
1205         }
1206
1207         /* Recalculate per-file deny mode if there was a change */
1208         if (change)
1209                 recalculate_deny_mode(stp->st_stid.sc_file);
1210 }
1211
1212 /* release all access and file references for a given stateid */
1213 static void
1214 release_all_access(struct nfs4_ol_stateid *stp)
1215 {
1216         int i;
1217         struct nfs4_file *fp = stp->st_stid.sc_file;
1218
1219         if (fp && stp->st_deny_bmap != 0)
1220                 recalculate_deny_mode(fp);
1221
1222         for (i = 1; i < 4; i++) {
1223                 if (test_access(i, stp))
1224                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1225                 clear_access(i, stp);
1226         }
1227 }
1228
1229 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1230 {
1231         kfree(sop->so_owner.data);
1232         sop->so_ops->so_free(sop);
1233 }
1234
1235 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1236 {
1237         struct nfs4_client *clp = sop->so_client;
1238
1239         might_lock(&clp->cl_lock);
1240
1241         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1242                 return;
1243         sop->so_ops->so_unhash(sop);
1244         spin_unlock(&clp->cl_lock);
1245         nfs4_free_stateowner(sop);
1246 }
1247
1248 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1249 {
1250         struct nfs4_file *fp = stp->st_stid.sc_file;
1251
1252         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1253
1254         if (list_empty(&stp->st_perfile))
1255                 return false;
1256
1257         spin_lock(&fp->fi_lock);
1258         list_del_init(&stp->st_perfile);
1259         spin_unlock(&fp->fi_lock);
1260         list_del(&stp->st_perstateowner);
1261         return true;
1262 }
1263
1264 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1265 {
1266         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1267
1268         put_clnt_odstate(stp->st_clnt_odstate);
1269         release_all_access(stp);
1270         if (stp->st_stateowner)
1271                 nfs4_put_stateowner(stp->st_stateowner);
1272         kmem_cache_free(stateid_slab, stid);
1273 }
1274
1275 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1276 {
1277         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1278         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1279         struct file *file;
1280
1281         file = find_any_file(stp->st_stid.sc_file);
1282         if (file)
1283                 filp_close(file, (fl_owner_t)lo);
1284         nfs4_free_ol_stateid(stid);
1285 }
1286
1287 /*
1288  * Put the persistent reference to an already unhashed generic stateid, while
1289  * holding the cl_lock. If it's the last reference, then put it onto the
1290  * reaplist for later destruction.
1291  */
1292 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1293                                        struct list_head *reaplist)
1294 {
1295         struct nfs4_stid *s = &stp->st_stid;
1296         struct nfs4_client *clp = s->sc_client;
1297
1298         lockdep_assert_held(&clp->cl_lock);
1299
1300         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1301
1302         if (!refcount_dec_and_test(&s->sc_count)) {
1303                 wake_up_all(&close_wq);
1304                 return;
1305         }
1306
1307         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1308         list_add(&stp->st_locks, reaplist);
1309 }
1310
1311 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1312 {
1313         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1314
1315         list_del_init(&stp->st_locks);
1316         nfs4_unhash_stid(&stp->st_stid);
1317         return unhash_ol_stateid(stp);
1318 }
1319
1320 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1321 {
1322         struct nfs4_client *clp = stp->st_stid.sc_client;
1323         bool unhashed;
1324
1325         spin_lock(&clp->cl_lock);
1326         unhashed = unhash_lock_stateid(stp);
1327         spin_unlock(&clp->cl_lock);
1328         if (unhashed)
1329                 nfs4_put_stid(&stp->st_stid);
1330 }
1331
1332 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1333 {
1334         struct nfs4_client *clp = lo->lo_owner.so_client;
1335
1336         lockdep_assert_held(&clp->cl_lock);
1337
1338         list_del_init(&lo->lo_owner.so_strhash);
1339 }
1340
1341 /*
1342  * Free a list of generic stateids that were collected earlier after being
1343  * fully unhashed.
1344  */
1345 static void
1346 free_ol_stateid_reaplist(struct list_head *reaplist)
1347 {
1348         struct nfs4_ol_stateid *stp;
1349         struct nfs4_file *fp;
1350
1351         might_sleep();
1352
1353         while (!list_empty(reaplist)) {
1354                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1355                                        st_locks);
1356                 list_del(&stp->st_locks);
1357                 fp = stp->st_stid.sc_file;
1358                 stp->st_stid.sc_free(&stp->st_stid);
1359                 if (fp)
1360                         put_nfs4_file(fp);
1361         }
1362 }
1363
1364 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1365                                        struct list_head *reaplist)
1366 {
1367         struct nfs4_ol_stateid *stp;
1368
1369         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1370
1371         while (!list_empty(&open_stp->st_locks)) {
1372                 stp = list_entry(open_stp->st_locks.next,
1373                                 struct nfs4_ol_stateid, st_locks);
1374                 WARN_ON(!unhash_lock_stateid(stp));
1375                 put_ol_stateid_locked(stp, reaplist);
1376         }
1377 }
1378
1379 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1380                                 struct list_head *reaplist)
1381 {
1382         bool unhashed;
1383
1384         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1385
1386         unhashed = unhash_ol_stateid(stp);
1387         release_open_stateid_locks(stp, reaplist);
1388         return unhashed;
1389 }
1390
1391 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1392 {
1393         LIST_HEAD(reaplist);
1394
1395         spin_lock(&stp->st_stid.sc_client->cl_lock);
1396         if (unhash_open_stateid(stp, &reaplist))
1397                 put_ol_stateid_locked(stp, &reaplist);
1398         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1399         free_ol_stateid_reaplist(&reaplist);
1400 }
1401
1402 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1403 {
1404         struct nfs4_client *clp = oo->oo_owner.so_client;
1405
1406         lockdep_assert_held(&clp->cl_lock);
1407
1408         list_del_init(&oo->oo_owner.so_strhash);
1409         list_del_init(&oo->oo_perclient);
1410 }
1411
1412 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1413 {
1414         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1415                                           nfsd_net_id);
1416         struct nfs4_ol_stateid *s;
1417
1418         spin_lock(&nn->client_lock);
1419         s = oo->oo_last_closed_stid;
1420         if (s) {
1421                 list_del_init(&oo->oo_close_lru);
1422                 oo->oo_last_closed_stid = NULL;
1423         }
1424         spin_unlock(&nn->client_lock);
1425         if (s)
1426                 nfs4_put_stid(&s->st_stid);
1427 }
1428
1429 static void release_openowner(struct nfs4_openowner *oo)
1430 {
1431         struct nfs4_ol_stateid *stp;
1432         struct nfs4_client *clp = oo->oo_owner.so_client;
1433         struct list_head reaplist;
1434
1435         INIT_LIST_HEAD(&reaplist);
1436
1437         spin_lock(&clp->cl_lock);
1438         unhash_openowner_locked(oo);
1439         while (!list_empty(&oo->oo_owner.so_stateids)) {
1440                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1441                                 struct nfs4_ol_stateid, st_perstateowner);
1442                 if (unhash_open_stateid(stp, &reaplist))
1443                         put_ol_stateid_locked(stp, &reaplist);
1444         }
1445         spin_unlock(&clp->cl_lock);
1446         free_ol_stateid_reaplist(&reaplist);
1447         release_last_closed_stateid(oo);
1448         nfs4_put_stateowner(&oo->oo_owner);
1449 }
1450
1451 static inline int
1452 hash_sessionid(struct nfs4_sessionid *sessionid)
1453 {
1454         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1455
1456         return sid->sequence % SESSION_HASH_SIZE;
1457 }
1458
1459 #ifdef CONFIG_SUNRPC_DEBUG
1460 static inline void
1461 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1462 {
1463         u32 *ptr = (u32 *)(&sessionid->data[0]);
1464         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1465 }
1466 #else
1467 static inline void
1468 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1469 {
1470 }
1471 #endif
1472
1473 /*
1474  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1475  * won't be used for replay.
1476  */
1477 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1478 {
1479         struct nfs4_stateowner *so = cstate->replay_owner;
1480
1481         if (nfserr == nfserr_replay_me)
1482                 return;
1483
1484         if (!seqid_mutating_err(ntohl(nfserr))) {
1485                 nfsd4_cstate_clear_replay(cstate);
1486                 return;
1487         }
1488         if (!so)
1489                 return;
1490         if (so->so_is_open_owner)
1491                 release_last_closed_stateid(openowner(so));
1492         so->so_seqid++;
1493         return;
1494 }
1495
1496 static void
1497 gen_sessionid(struct nfsd4_session *ses)
1498 {
1499         struct nfs4_client *clp = ses->se_client;
1500         struct nfsd4_sessionid *sid;
1501
1502         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1503         sid->clientid = clp->cl_clientid;
1504         sid->sequence = current_sessionid++;
1505         sid->reserved = 0;
1506 }
1507
1508 /*
1509  * The protocol defines ca_maxresponssize_cached to include the size of
1510  * the rpc header, but all we need to cache is the data starting after
1511  * the end of the initial SEQUENCE operation--the rest we regenerate
1512  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1513  * value that is the number of bytes in our cache plus a few additional
1514  * bytes.  In order to stay on the safe side, and not promise more than
1515  * we can cache, those additional bytes must be the minimum possible: 24
1516  * bytes of rpc header (xid through accept state, with AUTH_NULL
1517  * verifier), 12 for the compound header (with zero-length tag), and 44
1518  * for the SEQUENCE op response:
1519  */
1520 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1521
1522 static void
1523 free_session_slots(struct nfsd4_session *ses)
1524 {
1525         int i;
1526
1527         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1528                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1529                 kfree(ses->se_slots[i]);
1530         }
1531 }
1532
1533 /*
1534  * We don't actually need to cache the rpc and session headers, so we
1535  * can allocate a little less for each slot:
1536  */
1537 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1538 {
1539         u32 size;
1540
1541         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1542                 size = 0;
1543         else
1544                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1545         return size + sizeof(struct nfsd4_slot);
1546 }
1547
1548 /*
1549  * XXX: If we run out of reserved DRC memory we could (up to a point)
1550  * re-negotiate active sessions and reduce their slot usage to make
1551  * room for new connections. For now we just fail the create session.
1552  */
1553 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1554 {
1555         u32 slotsize = slot_bytes(ca);
1556         u32 num = ca->maxreqs;
1557         unsigned long avail, total_avail;
1558
1559         spin_lock(&nfsd_drc_lock);
1560         total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1561         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1562         /*
1563          * Never use more than a third of the remaining memory,
1564          * unless it's the only way to give this client a slot:
1565          */
1566         avail = clamp_t(int, avail, slotsize, total_avail/3);
1567         num = min_t(int, num, avail / slotsize);
1568         nfsd_drc_mem_used += num * slotsize;
1569         spin_unlock(&nfsd_drc_lock);
1570
1571         return num;
1572 }
1573
1574 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1575 {
1576         int slotsize = slot_bytes(ca);
1577
1578         spin_lock(&nfsd_drc_lock);
1579         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1580         spin_unlock(&nfsd_drc_lock);
1581 }
1582
1583 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1584                                            struct nfsd4_channel_attrs *battrs)
1585 {
1586         int numslots = fattrs->maxreqs;
1587         int slotsize = slot_bytes(fattrs);
1588         struct nfsd4_session *new;
1589         int mem, i;
1590
1591         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1592                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1593         mem = numslots * sizeof(struct nfsd4_slot *);
1594
1595         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1596         if (!new)
1597                 return NULL;
1598         /* allocate each struct nfsd4_slot and data cache in one piece */
1599         for (i = 0; i < numslots; i++) {
1600                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1601                 if (!new->se_slots[i])
1602                         goto out_free;
1603         }
1604
1605         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1606         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1607
1608         return new;
1609 out_free:
1610         while (i--)
1611                 kfree(new->se_slots[i]);
1612         kfree(new);
1613         return NULL;
1614 }
1615
1616 static void free_conn(struct nfsd4_conn *c)
1617 {
1618         svc_xprt_put(c->cn_xprt);
1619         kfree(c);
1620 }
1621
1622 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1623 {
1624         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1625         struct nfs4_client *clp = c->cn_session->se_client;
1626
1627         spin_lock(&clp->cl_lock);
1628         if (!list_empty(&c->cn_persession)) {
1629                 list_del(&c->cn_persession);
1630                 free_conn(c);
1631         }
1632         nfsd4_probe_callback(clp);
1633         spin_unlock(&clp->cl_lock);
1634 }
1635
1636 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1637 {
1638         struct nfsd4_conn *conn;
1639
1640         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1641         if (!conn)
1642                 return NULL;
1643         svc_xprt_get(rqstp->rq_xprt);
1644         conn->cn_xprt = rqstp->rq_xprt;
1645         conn->cn_flags = flags;
1646         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1647         return conn;
1648 }
1649
1650 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1651 {
1652         conn->cn_session = ses;
1653         list_add(&conn->cn_persession, &ses->se_conns);
1654 }
1655
1656 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1657 {
1658         struct nfs4_client *clp = ses->se_client;
1659
1660         spin_lock(&clp->cl_lock);
1661         __nfsd4_hash_conn(conn, ses);
1662         spin_unlock(&clp->cl_lock);
1663 }
1664
1665 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1666 {
1667         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1668         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1669 }
1670
1671 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1672 {
1673         int ret;
1674
1675         nfsd4_hash_conn(conn, ses);
1676         ret = nfsd4_register_conn(conn);
1677         if (ret)
1678                 /* oops; xprt is already down: */
1679                 nfsd4_conn_lost(&conn->cn_xpt_user);
1680         /* We may have gained or lost a callback channel: */
1681         nfsd4_probe_callback_sync(ses->se_client);
1682 }
1683
1684 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1685 {
1686         u32 dir = NFS4_CDFC4_FORE;
1687
1688         if (cses->flags & SESSION4_BACK_CHAN)
1689                 dir |= NFS4_CDFC4_BACK;
1690         return alloc_conn(rqstp, dir);
1691 }
1692
1693 /* must be called under client_lock */
1694 static void nfsd4_del_conns(struct nfsd4_session *s)
1695 {
1696         struct nfs4_client *clp = s->se_client;
1697         struct nfsd4_conn *c;
1698
1699         spin_lock(&clp->cl_lock);
1700         while (!list_empty(&s->se_conns)) {
1701                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1702                 list_del_init(&c->cn_persession);
1703                 spin_unlock(&clp->cl_lock);
1704
1705                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1706                 free_conn(c);
1707
1708                 spin_lock(&clp->cl_lock);
1709         }
1710         spin_unlock(&clp->cl_lock);
1711 }
1712
1713 static void __free_session(struct nfsd4_session *ses)
1714 {
1715         free_session_slots(ses);
1716         kfree(ses);
1717 }
1718
1719 static void free_session(struct nfsd4_session *ses)
1720 {
1721         nfsd4_del_conns(ses);
1722         nfsd4_put_drc_mem(&ses->se_fchannel);
1723         __free_session(ses);
1724 }
1725
1726 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1727 {
1728         int idx;
1729         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1730
1731         new->se_client = clp;
1732         gen_sessionid(new);
1733
1734         INIT_LIST_HEAD(&new->se_conns);
1735
1736         new->se_cb_seq_nr = 1;
1737         new->se_flags = cses->flags;
1738         new->se_cb_prog = cses->callback_prog;
1739         new->se_cb_sec = cses->cb_sec;
1740         atomic_set(&new->se_ref, 0);
1741         idx = hash_sessionid(&new->se_sessionid);
1742         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1743         spin_lock(&clp->cl_lock);
1744         list_add(&new->se_perclnt, &clp->cl_sessions);
1745         spin_unlock(&clp->cl_lock);
1746
1747         {
1748                 struct sockaddr *sa = svc_addr(rqstp);
1749                 /*
1750                  * This is a little silly; with sessions there's no real
1751                  * use for the callback address.  Use the peer address
1752                  * as a reasonable default for now, but consider fixing
1753                  * the rpc client not to require an address in the
1754                  * future:
1755                  */
1756                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1757                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1758         }
1759 }
1760
1761 /* caller must hold client_lock */
1762 static struct nfsd4_session *
1763 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1764 {
1765         struct nfsd4_session *elem;
1766         int idx;
1767         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1768
1769         lockdep_assert_held(&nn->client_lock);
1770
1771         dump_sessionid(__func__, sessionid);
1772         idx = hash_sessionid(sessionid);
1773         /* Search in the appropriate list */
1774         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1775                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1776                             NFS4_MAX_SESSIONID_LEN)) {
1777                         return elem;
1778                 }
1779         }
1780
1781         dprintk("%s: session not found\n", __func__);
1782         return NULL;
1783 }
1784
1785 static struct nfsd4_session *
1786 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1787                 __be32 *ret)
1788 {
1789         struct nfsd4_session *session;
1790         __be32 status = nfserr_badsession;
1791
1792         session = __find_in_sessionid_hashtbl(sessionid, net);
1793         if (!session)
1794                 goto out;
1795         status = nfsd4_get_session_locked(session);
1796         if (status)
1797                 session = NULL;
1798 out:
1799         *ret = status;
1800         return session;
1801 }
1802
1803 /* caller must hold client_lock */
1804 static void
1805 unhash_session(struct nfsd4_session *ses)
1806 {
1807         struct nfs4_client *clp = ses->se_client;
1808         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1809
1810         lockdep_assert_held(&nn->client_lock);
1811
1812         list_del(&ses->se_hash);
1813         spin_lock(&ses->se_client->cl_lock);
1814         list_del(&ses->se_perclnt);
1815         spin_unlock(&ses->se_client->cl_lock);
1816 }
1817
1818 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1819 static int
1820 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1821 {
1822         /*
1823          * We're assuming the clid was not given out from a boot
1824          * precisely 2^32 (about 136 years) before this one.  That seems
1825          * a safe assumption:
1826          */
1827         if (clid->cl_boot == (u32)nn->boot_time)
1828                 return 0;
1829         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1830                 clid->cl_boot, clid->cl_id, nn->boot_time);
1831         return 1;
1832 }
1833
1834 /*
1835  * XXX Should we use a slab cache ?
1836  * This type of memory management is somewhat inefficient, but we use it
1837  * anyway since SETCLIENTID is not a common operation.
1838  */
1839 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1840 {
1841         struct nfs4_client *clp;
1842         int i;
1843
1844         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1845         if (clp == NULL)
1846                 return NULL;
1847         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1848         if (clp->cl_name.data == NULL)
1849                 goto err_no_name;
1850         clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1851                                                  sizeof(struct list_head),
1852                                                  GFP_KERNEL);
1853         if (!clp->cl_ownerstr_hashtbl)
1854                 goto err_no_hashtbl;
1855         for (i = 0; i < OWNER_HASH_SIZE; i++)
1856                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1857         clp->cl_name.len = name.len;
1858         INIT_LIST_HEAD(&clp->cl_sessions);
1859         idr_init(&clp->cl_stateids);
1860         atomic_set(&clp->cl_refcount, 0);
1861         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1862         INIT_LIST_HEAD(&clp->cl_idhash);
1863         INIT_LIST_HEAD(&clp->cl_openowners);
1864         INIT_LIST_HEAD(&clp->cl_delegations);
1865         INIT_LIST_HEAD(&clp->cl_lru);
1866         INIT_LIST_HEAD(&clp->cl_revoked);
1867 #ifdef CONFIG_NFSD_PNFS
1868         INIT_LIST_HEAD(&clp->cl_lo_states);
1869 #endif
1870         INIT_LIST_HEAD(&clp->async_copies);
1871         spin_lock_init(&clp->async_lock);
1872         spin_lock_init(&clp->cl_lock);
1873         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1874         return clp;
1875 err_no_hashtbl:
1876         kfree(clp->cl_name.data);
1877 err_no_name:
1878         kmem_cache_free(client_slab, clp);
1879         return NULL;
1880 }
1881
1882 static void
1883 free_client(struct nfs4_client *clp)
1884 {
1885         while (!list_empty(&clp->cl_sessions)) {
1886                 struct nfsd4_session *ses;
1887                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1888                                 se_perclnt);
1889                 list_del(&ses->se_perclnt);
1890                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1891                 free_session(ses);
1892         }
1893         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1894         free_svc_cred(&clp->cl_cred);
1895         kfree(clp->cl_ownerstr_hashtbl);
1896         kfree(clp->cl_name.data);
1897         idr_destroy(&clp->cl_stateids);
1898         kmem_cache_free(client_slab, clp);
1899 }
1900
1901 /* must be called under the client_lock */
1902 static void
1903 unhash_client_locked(struct nfs4_client *clp)
1904 {
1905         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1906         struct nfsd4_session *ses;
1907
1908         lockdep_assert_held(&nn->client_lock);
1909
1910         /* Mark the client as expired! */
1911         clp->cl_time = 0;
1912         /* Make it invisible */
1913         if (!list_empty(&clp->cl_idhash)) {
1914                 list_del_init(&clp->cl_idhash);
1915                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1916                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1917                 else
1918                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1919         }
1920         list_del_init(&clp->cl_lru);
1921         spin_lock(&clp->cl_lock);
1922         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1923                 list_del_init(&ses->se_hash);
1924         spin_unlock(&clp->cl_lock);
1925 }
1926
1927 static void
1928 unhash_client(struct nfs4_client *clp)
1929 {
1930         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1931
1932         spin_lock(&nn->client_lock);
1933         unhash_client_locked(clp);
1934         spin_unlock(&nn->client_lock);
1935 }
1936
1937 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1938 {
1939         if (atomic_read(&clp->cl_refcount))
1940                 return nfserr_jukebox;
1941         unhash_client_locked(clp);
1942         return nfs_ok;
1943 }
1944
1945 static void
1946 __destroy_client(struct nfs4_client *clp)
1947 {
1948         int i;
1949         struct nfs4_openowner *oo;
1950         struct nfs4_delegation *dp;
1951         struct list_head reaplist;
1952
1953         INIT_LIST_HEAD(&reaplist);
1954         spin_lock(&state_lock);
1955         while (!list_empty(&clp->cl_delegations)) {
1956                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1957                 WARN_ON(!unhash_delegation_locked(dp));
1958                 list_add(&dp->dl_recall_lru, &reaplist);
1959         }
1960         spin_unlock(&state_lock);
1961         while (!list_empty(&reaplist)) {
1962                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1963                 list_del_init(&dp->dl_recall_lru);
1964                 destroy_unhashed_deleg(dp);
1965         }
1966         while (!list_empty(&clp->cl_revoked)) {
1967                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1968                 list_del_init(&dp->dl_recall_lru);
1969                 nfs4_put_stid(&dp->dl_stid);
1970         }
1971         while (!list_empty(&clp->cl_openowners)) {
1972                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1973                 nfs4_get_stateowner(&oo->oo_owner);
1974                 release_openowner(oo);
1975         }
1976         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1977                 struct nfs4_stateowner *so, *tmp;
1978
1979                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1980                                          so_strhash) {
1981                         /* Should be no openowners at this point */
1982                         WARN_ON_ONCE(so->so_is_open_owner);
1983                         remove_blocked_locks(lockowner(so));
1984                 }
1985         }
1986         nfsd4_return_all_client_layouts(clp);
1987         nfsd4_shutdown_copy(clp);
1988         nfsd4_shutdown_callback(clp);
1989         if (clp->cl_cb_conn.cb_xprt)
1990                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1991         free_client(clp);
1992 }
1993
1994 static void
1995 destroy_client(struct nfs4_client *clp)
1996 {
1997         unhash_client(clp);
1998         __destroy_client(clp);
1999 }
2000
2001 static void inc_reclaim_complete(struct nfs4_client *clp)
2002 {
2003         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2004
2005         if (!nn->track_reclaim_completes)
2006                 return;
2007         if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2008                 return;
2009         if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2010                         nn->reclaim_str_hashtbl_size) {
2011                 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2012                                 clp->net->ns.inum);
2013                 nfsd4_end_grace(nn);
2014         }
2015 }
2016
2017 static void expire_client(struct nfs4_client *clp)
2018 {
2019         unhash_client(clp);
2020         nfsd4_client_record_remove(clp);
2021         __destroy_client(clp);
2022 }
2023
2024 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2025 {
2026         memcpy(target->cl_verifier.data, source->data,
2027                         sizeof(target->cl_verifier.data));
2028 }
2029
2030 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2031 {
2032         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2033         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2034 }
2035
2036 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2037 {
2038         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2039         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2040                                                                 GFP_KERNEL);
2041         target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2042         if ((source->cr_principal && !target->cr_principal) ||
2043             (source->cr_raw_principal && !target->cr_raw_principal) ||
2044             (source->cr_targ_princ && !target->cr_targ_princ))
2045                 return -ENOMEM;
2046
2047         target->cr_flavor = source->cr_flavor;
2048         target->cr_uid = source->cr_uid;
2049         target->cr_gid = source->cr_gid;
2050         target->cr_group_info = source->cr_group_info;
2051         get_group_info(target->cr_group_info);
2052         target->cr_gss_mech = source->cr_gss_mech;
2053         if (source->cr_gss_mech)
2054                 gss_mech_get(source->cr_gss_mech);
2055         return 0;
2056 }
2057
2058 static int
2059 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2060 {
2061         if (o1->len < o2->len)
2062                 return -1;
2063         if (o1->len > o2->len)
2064                 return 1;
2065         return memcmp(o1->data, o2->data, o1->len);
2066 }
2067
2068 static int
2069 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2070 {
2071         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2072 }
2073
2074 static int
2075 same_clid(clientid_t *cl1, clientid_t *cl2)
2076 {
2077         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2078 }
2079
2080 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2081 {
2082         int i;
2083
2084         if (g1->ngroups != g2->ngroups)
2085                 return false;
2086         for (i=0; i<g1->ngroups; i++)
2087                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2088                         return false;
2089         return true;
2090 }
2091
2092 /*
2093  * RFC 3530 language requires clid_inuse be returned when the
2094  * "principal" associated with a requests differs from that previously
2095  * used.  We use uid, gid's, and gss principal string as our best
2096  * approximation.  We also don't want to allow non-gss use of a client
2097  * established using gss: in theory cr_principal should catch that
2098  * change, but in practice cr_principal can be null even in the gss case
2099  * since gssd doesn't always pass down a principal string.
2100  */
2101 static bool is_gss_cred(struct svc_cred *cr)
2102 {
2103         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2104         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2105 }
2106
2107
2108 static bool
2109 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2110 {
2111         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2112                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2113                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2114                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2115                 return false;
2116         /* XXX: check that cr_targ_princ fields match ? */
2117         if (cr1->cr_principal == cr2->cr_principal)
2118                 return true;
2119         if (!cr1->cr_principal || !cr2->cr_principal)
2120                 return false;
2121         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2122 }
2123
2124 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2125 {
2126         struct svc_cred *cr = &rqstp->rq_cred;
2127         u32 service;
2128
2129         if (!cr->cr_gss_mech)
2130                 return false;
2131         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2132         return service == RPC_GSS_SVC_INTEGRITY ||
2133                service == RPC_GSS_SVC_PRIVACY;
2134 }
2135
2136 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2137 {
2138         struct svc_cred *cr = &rqstp->rq_cred;
2139
2140         if (!cl->cl_mach_cred)
2141                 return true;
2142         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2143                 return false;
2144         if (!svc_rqst_integrity_protected(rqstp))
2145                 return false;
2146         if (cl->cl_cred.cr_raw_principal)
2147                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2148                                                 cr->cr_raw_principal);
2149         if (!cr->cr_principal)
2150                 return false;
2151         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2152 }
2153
2154 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2155 {
2156         __be32 verf[2];
2157
2158         /*
2159          * This is opaque to client, so no need to byte-swap. Use
2160          * __force to keep sparse happy
2161          */
2162         verf[0] = (__force __be32)get_seconds();
2163         verf[1] = (__force __be32)nn->clverifier_counter++;
2164         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2165 }
2166
2167 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2168 {
2169         clp->cl_clientid.cl_boot = nn->boot_time;
2170         clp->cl_clientid.cl_id = nn->clientid_counter++;
2171         gen_confirm(clp, nn);
2172 }
2173
2174 static struct nfs4_stid *
2175 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2176 {
2177         struct nfs4_stid *ret;
2178
2179         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2180         if (!ret || !ret->sc_type)
2181                 return NULL;
2182         return ret;
2183 }
2184
2185 static struct nfs4_stid *
2186 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2187 {
2188         struct nfs4_stid *s;
2189
2190         spin_lock(&cl->cl_lock);
2191         s = find_stateid_locked(cl, t);
2192         if (s != NULL) {
2193                 if (typemask & s->sc_type)
2194                         refcount_inc(&s->sc_count);
2195                 else
2196                         s = NULL;
2197         }
2198         spin_unlock(&cl->cl_lock);
2199         return s;
2200 }
2201
2202 static struct nfs4_client *create_client(struct xdr_netobj name,
2203                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2204 {
2205         struct nfs4_client *clp;
2206         struct sockaddr *sa = svc_addr(rqstp);
2207         int ret;
2208         struct net *net = SVC_NET(rqstp);
2209
2210         clp = alloc_client(name);
2211         if (clp == NULL)
2212                 return NULL;
2213
2214         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2215         if (ret) {
2216                 free_client(clp);
2217                 return NULL;
2218         }
2219         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2220         clp->cl_time = get_seconds();
2221         clear_bit(0, &clp->cl_cb_slot_busy);
2222         copy_verf(clp, verf);
2223         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2224         clp->cl_cb_session = NULL;
2225         clp->net = net;
2226         return clp;
2227 }
2228
2229 static void
2230 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2231 {
2232         struct rb_node **new = &(root->rb_node), *parent = NULL;
2233         struct nfs4_client *clp;
2234
2235         while (*new) {
2236                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2237                 parent = *new;
2238
2239                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2240                         new = &((*new)->rb_left);
2241                 else
2242                         new = &((*new)->rb_right);
2243         }
2244
2245         rb_link_node(&new_clp->cl_namenode, parent, new);
2246         rb_insert_color(&new_clp->cl_namenode, root);
2247 }
2248
2249 static struct nfs4_client *
2250 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2251 {
2252         int cmp;
2253         struct rb_node *node = root->rb_node;
2254         struct nfs4_client *clp;
2255
2256         while (node) {
2257                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2258                 cmp = compare_blob(&clp->cl_name, name);
2259                 if (cmp > 0)
2260                         node = node->rb_left;
2261                 else if (cmp < 0)
2262                         node = node->rb_right;
2263                 else
2264                         return clp;
2265         }
2266         return NULL;
2267 }
2268
2269 static void
2270 add_to_unconfirmed(struct nfs4_client *clp)
2271 {
2272         unsigned int idhashval;
2273         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2274
2275         lockdep_assert_held(&nn->client_lock);
2276
2277         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2278         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2279         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2280         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2281         renew_client_locked(clp);
2282 }
2283
2284 static void
2285 move_to_confirmed(struct nfs4_client *clp)
2286 {
2287         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2288         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2289
2290         lockdep_assert_held(&nn->client_lock);
2291
2292         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2293         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2294         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2295         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2296         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2297         renew_client_locked(clp);
2298 }
2299
2300 static struct nfs4_client *
2301 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2302 {
2303         struct nfs4_client *clp;
2304         unsigned int idhashval = clientid_hashval(clid->cl_id);
2305
2306         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2307                 if (same_clid(&clp->cl_clientid, clid)) {
2308                         if ((bool)clp->cl_minorversion != sessions)
2309                                 return NULL;
2310                         renew_client_locked(clp);
2311                         return clp;
2312                 }
2313         }
2314         return NULL;
2315 }
2316
2317 static struct nfs4_client *
2318 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2319 {
2320         struct list_head *tbl = nn->conf_id_hashtbl;
2321
2322         lockdep_assert_held(&nn->client_lock);
2323         return find_client_in_id_table(tbl, clid, sessions);
2324 }
2325
2326 static struct nfs4_client *
2327 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2328 {
2329         struct list_head *tbl = nn->unconf_id_hashtbl;
2330
2331         lockdep_assert_held(&nn->client_lock);
2332         return find_client_in_id_table(tbl, clid, sessions);
2333 }
2334
2335 static bool clp_used_exchangeid(struct nfs4_client *clp)
2336 {
2337         return clp->cl_exchange_flags != 0;
2338 }
2339
2340 static struct nfs4_client *
2341 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2342 {
2343         lockdep_assert_held(&nn->client_lock);
2344         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2345 }
2346
2347 static struct nfs4_client *
2348 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2349 {
2350         lockdep_assert_held(&nn->client_lock);
2351         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2352 }
2353
2354 static void
2355 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2356 {
2357         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2358         struct sockaddr *sa = svc_addr(rqstp);
2359         u32 scopeid = rpc_get_scope_id(sa);
2360         unsigned short expected_family;
2361
2362         /* Currently, we only support tcp and tcp6 for the callback channel */
2363         if (se->se_callback_netid_len == 3 &&
2364             !memcmp(se->se_callback_netid_val, "tcp", 3))
2365                 expected_family = AF_INET;
2366         else if (se->se_callback_netid_len == 4 &&
2367                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2368                 expected_family = AF_INET6;
2369         else
2370                 goto out_err;
2371
2372         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2373                                             se->se_callback_addr_len,
2374                                             (struct sockaddr *)&conn->cb_addr,
2375                                             sizeof(conn->cb_addr));
2376
2377         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2378                 goto out_err;
2379
2380         if (conn->cb_addr.ss_family == AF_INET6)
2381                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2382
2383         conn->cb_prog = se->se_callback_prog;
2384         conn->cb_ident = se->se_callback_ident;
2385         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2386         return;
2387 out_err:
2388         conn->cb_addr.ss_family = AF_UNSPEC;
2389         conn->cb_addrlen = 0;
2390         dprintk("NFSD: this client (clientid %08x/%08x) "
2391                 "will not receive delegations\n",
2392                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2393
2394         return;
2395 }
2396
2397 /*
2398  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2399  */
2400 static void
2401 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2402 {
2403         struct xdr_buf *buf = resp->xdr.buf;
2404         struct nfsd4_slot *slot = resp->cstate.slot;
2405         unsigned int base;
2406
2407         dprintk("--> %s slot %p\n", __func__, slot);
2408
2409         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2410         slot->sl_opcnt = resp->opcnt;
2411         slot->sl_status = resp->cstate.status;
2412         free_svc_cred(&slot->sl_cred);
2413         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2414
2415         if (!nfsd4_cache_this(resp)) {
2416                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2417                 return;
2418         }
2419         slot->sl_flags |= NFSD4_SLOT_CACHED;
2420
2421         base = resp->cstate.data_offset;
2422         slot->sl_datalen = buf->len - base;
2423         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2424                 WARN(1, "%s: sessions DRC could not cache compound\n",
2425                      __func__);
2426         return;
2427 }
2428
2429 /*
2430  * Encode the replay sequence operation from the slot values.
2431  * If cachethis is FALSE encode the uncached rep error on the next
2432  * operation which sets resp->p and increments resp->opcnt for
2433  * nfs4svc_encode_compoundres.
2434  *
2435  */
2436 static __be32
2437 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2438                           struct nfsd4_compoundres *resp)
2439 {
2440         struct nfsd4_op *op;
2441         struct nfsd4_slot *slot = resp->cstate.slot;
2442
2443         /* Encode the replayed sequence operation */
2444         op = &args->ops[resp->opcnt - 1];
2445         nfsd4_encode_operation(resp, op);
2446
2447         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2448                 return op->status;
2449         if (args->opcnt == 1) {
2450                 /*
2451                  * The original operation wasn't a solo sequence--we
2452                  * always cache those--so this retry must not match the
2453                  * original:
2454                  */
2455                 op->status = nfserr_seq_false_retry;
2456         } else {
2457                 op = &args->ops[resp->opcnt++];
2458                 op->status = nfserr_retry_uncached_rep;
2459                 nfsd4_encode_operation(resp, op);
2460         }
2461         return op->status;
2462 }
2463
2464 /*
2465  * The sequence operation is not cached because we can use the slot and
2466  * session values.
2467  */
2468 static __be32
2469 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2470                          struct nfsd4_sequence *seq)
2471 {
2472         struct nfsd4_slot *slot = resp->cstate.slot;
2473         struct xdr_stream *xdr = &resp->xdr;
2474         __be32 *p;
2475         __be32 status;
2476
2477         dprintk("--> %s slot %p\n", __func__, slot);
2478
2479         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2480         if (status)
2481                 return status;
2482
2483         p = xdr_reserve_space(xdr, slot->sl_datalen);
2484         if (!p) {
2485                 WARN_ON_ONCE(1);
2486                 return nfserr_serverfault;
2487         }
2488         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2489         xdr_commit_encode(xdr);
2490
2491         resp->opcnt = slot->sl_opcnt;
2492         return slot->sl_status;
2493 }
2494
2495 /*
2496  * Set the exchange_id flags returned by the server.
2497  */
2498 static void
2499 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2500 {
2501 #ifdef CONFIG_NFSD_PNFS
2502         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2503 #else
2504         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2505 #endif
2506
2507         /* Referrals are supported, Migration is not. */
2508         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2509
2510         /* set the wire flags to return to client. */
2511         clid->flags = new->cl_exchange_flags;
2512 }
2513
2514 static bool client_has_openowners(struct nfs4_client *clp)
2515 {
2516         struct nfs4_openowner *oo;
2517
2518         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2519                 if (!list_empty(&oo->oo_owner.so_stateids))
2520                         return true;
2521         }
2522         return false;
2523 }
2524
2525 static bool client_has_state(struct nfs4_client *clp)
2526 {
2527         return client_has_openowners(clp)
2528 #ifdef CONFIG_NFSD_PNFS
2529                 || !list_empty(&clp->cl_lo_states)
2530 #endif
2531                 || !list_empty(&clp->cl_delegations)
2532                 || !list_empty(&clp->cl_sessions)
2533                 || !list_empty(&clp->async_copies);
2534 }
2535
2536 __be32
2537 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2538                 union nfsd4_op_u *u)
2539 {
2540         struct nfsd4_exchange_id *exid = &u->exchange_id;
2541         struct nfs4_client *conf, *new;
2542         struct nfs4_client *unconf = NULL;
2543         __be32 status;
2544         char                    addr_str[INET6_ADDRSTRLEN];
2545         nfs4_verifier           verf = exid->verifier;
2546         struct sockaddr         *sa = svc_addr(rqstp);
2547         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2548         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2549
2550         rpc_ntop(sa, addr_str, sizeof(addr_str));
2551         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2552                 "ip_addr=%s flags %x, spa_how %d\n",
2553                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2554                 addr_str, exid->flags, exid->spa_how);
2555
2556         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2557                 return nfserr_inval;
2558
2559         new = create_client(exid->clname, rqstp, &verf);
2560         if (new == NULL)
2561                 return nfserr_jukebox;
2562
2563         switch (exid->spa_how) {
2564         case SP4_MACH_CRED:
2565                 exid->spo_must_enforce[0] = 0;
2566                 exid->spo_must_enforce[1] = (
2567                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2568                         1 << (OP_EXCHANGE_ID - 32) |
2569                         1 << (OP_CREATE_SESSION - 32) |
2570                         1 << (OP_DESTROY_SESSION - 32) |
2571                         1 << (OP_DESTROY_CLIENTID - 32));
2572
2573                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2574                                         1 << (OP_OPEN_DOWNGRADE) |
2575                                         1 << (OP_LOCKU) |
2576                                         1 << (OP_DELEGRETURN));
2577
2578                 exid->spo_must_allow[1] &= (
2579                                         1 << (OP_TEST_STATEID - 32) |
2580                                         1 << (OP_FREE_STATEID - 32));
2581                 if (!svc_rqst_integrity_protected(rqstp)) {
2582                         status = nfserr_inval;
2583                         goto out_nolock;
2584                 }
2585                 /*
2586                  * Sometimes userspace doesn't give us a principal.
2587                  * Which is a bug, really.  Anyway, we can't enforce
2588                  * MACH_CRED in that case, better to give up now:
2589                  */
2590                 if (!new->cl_cred.cr_principal &&
2591                                         !new->cl_cred.cr_raw_principal) {
2592                         status = nfserr_serverfault;
2593                         goto out_nolock;
2594                 }
2595                 new->cl_mach_cred = true;
2596         case SP4_NONE:
2597                 break;
2598         default:                                /* checked by xdr code */
2599                 WARN_ON_ONCE(1);
2600                 /* fall through */
2601         case SP4_SSV:
2602                 status = nfserr_encr_alg_unsupp;
2603                 goto out_nolock;
2604         }
2605
2606         /* Cases below refer to rfc 5661 section 18.35.4: */
2607         spin_lock(&nn->client_lock);
2608         conf = find_confirmed_client_by_name(&exid->clname, nn);
2609         if (conf) {
2610                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2611                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2612
2613                 if (update) {
2614                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2615                                 status = nfserr_inval;
2616                                 goto out;
2617                         }
2618                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2619                                 status = nfserr_wrong_cred;
2620                                 goto out;
2621                         }
2622                         if (!creds_match) { /* case 9 */
2623                                 status = nfserr_perm;
2624                                 goto out;
2625                         }
2626                         if (!verfs_match) { /* case 8 */
2627                                 status = nfserr_not_same;
2628                                 goto out;
2629                         }
2630                         /* case 6 */
2631                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2632                         goto out_copy;
2633                 }
2634                 if (!creds_match) { /* case 3 */
2635                         if (client_has_state(conf)) {
2636                                 status = nfserr_clid_inuse;
2637                                 goto out;
2638                         }
2639                         goto out_new;
2640                 }
2641                 if (verfs_match) { /* case 2 */
2642                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2643                         goto out_copy;
2644                 }
2645                 /* case 5, client reboot */
2646                 conf = NULL;
2647                 goto out_new;
2648         }
2649
2650         if (update) { /* case 7 */
2651                 status = nfserr_noent;
2652                 goto out;
2653         }
2654
2655         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2656         if (unconf) /* case 4, possible retry or client restart */
2657                 unhash_client_locked(unconf);
2658
2659         /* case 1 (normal case) */
2660 out_new:
2661         if (conf) {
2662                 status = mark_client_expired_locked(conf);
2663                 if (status)
2664                         goto out;
2665         }
2666         new->cl_minorversion = cstate->minorversion;
2667         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2668         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2669
2670         gen_clid(new, nn);
2671         add_to_unconfirmed(new);
2672         swap(new, conf);
2673 out_copy:
2674         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2675         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2676
2677         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2678         nfsd4_set_ex_flags(conf, exid);
2679
2680         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2681                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2682         status = nfs_ok;
2683
2684 out:
2685         spin_unlock(&nn->client_lock);
2686 out_nolock:
2687         if (new)
2688                 expire_client(new);
2689         if (unconf)
2690                 expire_client(unconf);
2691         return status;
2692 }
2693
2694 static __be32
2695 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2696 {
2697         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2698                 slot_seqid);
2699
2700         /* The slot is in use, and no response has been sent. */
2701         if (slot_inuse) {
2702                 if (seqid == slot_seqid)
2703                         return nfserr_jukebox;
2704                 else
2705                         return nfserr_seq_misordered;
2706         }
2707         /* Note unsigned 32-bit arithmetic handles wraparound: */
2708         if (likely(seqid == slot_seqid + 1))
2709                 return nfs_ok;
2710         if (seqid == slot_seqid)
2711                 return nfserr_replay_cache;
2712         return nfserr_seq_misordered;
2713 }
2714
2715 /*
2716  * Cache the create session result into the create session single DRC
2717  * slot cache by saving the xdr structure. sl_seqid has been set.
2718  * Do this for solo or embedded create session operations.
2719  */
2720 static void
2721 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2722                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2723 {
2724         slot->sl_status = nfserr;
2725         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2726 }
2727
2728 static __be32
2729 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2730                             struct nfsd4_clid_slot *slot)
2731 {
2732         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2733         return slot->sl_status;
2734 }
2735
2736 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2737                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2738                         1 +     /* MIN tag is length with zero, only length */ \
2739                         3 +     /* version, opcount, opcode */ \
2740                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2741                                 /* seqid, slotID, slotID, cache */ \
2742                         4 ) * sizeof(__be32))
2743
2744 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2745                         2 +     /* verifier: AUTH_NULL, length 0 */\
2746                         1 +     /* status */ \
2747                         1 +     /* MIN tag is length with zero, only length */ \
2748                         3 +     /* opcount, opcode, opstatus*/ \
2749                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2750                                 /* seqid, slotID, slotID, slotID, status */ \
2751                         5 ) * sizeof(__be32))
2752
2753 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2754 {
2755         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2756
2757         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2758                 return nfserr_toosmall;
2759         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2760                 return nfserr_toosmall;
2761         ca->headerpadsz = 0;
2762         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2763         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2764         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2765         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2766                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2767         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2768         /*
2769          * Note decreasing slot size below client's request may make it
2770          * difficult for client to function correctly, whereas
2771          * decreasing the number of slots will (just?) affect
2772          * performance.  When short on memory we therefore prefer to
2773          * decrease number of slots instead of their size.  Clients that
2774          * request larger slots than they need will get poor results:
2775          */
2776         ca->maxreqs = nfsd4_get_drc_mem(ca);
2777         if (!ca->maxreqs)
2778                 return nfserr_jukebox;
2779
2780         return nfs_ok;
2781 }
2782
2783 /*
2784  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2785  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2786  */
2787 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2788         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2789
2790 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2791         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2792
2793 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2794                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2795 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2796                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2797                                  sizeof(__be32))
2798
2799 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2800 {
2801         ca->headerpadsz = 0;
2802
2803         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2804                 return nfserr_toosmall;
2805         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2806                 return nfserr_toosmall;
2807         ca->maxresp_cached = 0;
2808         if (ca->maxops < 2)
2809                 return nfserr_toosmall;
2810
2811         return nfs_ok;
2812 }
2813
2814 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2815 {
2816         switch (cbs->flavor) {
2817         case RPC_AUTH_NULL:
2818         case RPC_AUTH_UNIX:
2819                 return nfs_ok;
2820         default:
2821                 /*
2822                  * GSS case: the spec doesn't allow us to return this
2823                  * error.  But it also doesn't allow us not to support
2824                  * GSS.
2825                  * I'd rather this fail hard than return some error the
2826                  * client might think it can already handle:
2827                  */
2828                 return nfserr_encr_alg_unsupp;
2829         }
2830 }
2831
2832 __be32
2833 nfsd4_create_session(struct svc_rqst *rqstp,
2834                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2835 {
2836         struct nfsd4_create_session *cr_ses = &u->create_session;
2837         struct sockaddr *sa = svc_addr(rqstp);
2838         struct nfs4_client *conf, *unconf;
2839         struct nfs4_client *old = NULL;
2840         struct nfsd4_session *new;
2841         struct nfsd4_conn *conn;
2842         struct nfsd4_clid_slot *cs_slot = NULL;
2843         __be32 status = 0;
2844         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2845
2846         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2847                 return nfserr_inval;
2848         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2849         if (status)
2850                 return status;
2851         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2852         if (status)
2853                 return status;
2854         status = check_backchannel_attrs(&cr_ses->back_channel);
2855         if (status)
2856                 goto out_release_drc_mem;
2857         status = nfserr_jukebox;
2858         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2859         if (!new)
2860                 goto out_release_drc_mem;
2861         conn = alloc_conn_from_crses(rqstp, cr_ses);
2862         if (!conn)
2863                 goto out_free_session;
2864
2865         spin_lock(&nn->client_lock);
2866         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2867         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2868         WARN_ON_ONCE(conf && unconf);
2869
2870         if (conf) {
2871                 status = nfserr_wrong_cred;
2872                 if (!nfsd4_mach_creds_match(conf, rqstp))
2873                         goto out_free_conn;
2874                 cs_slot = &conf->cl_cs_slot;
2875                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2876                 if (status) {
2877                         if (status == nfserr_replay_cache)
2878                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2879                         goto out_free_conn;
2880                 }
2881         } else if (unconf) {
2882                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2883                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2884                         status = nfserr_clid_inuse;
2885                         goto out_free_conn;
2886                 }
2887                 status = nfserr_wrong_cred;
2888                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2889                         goto out_free_conn;
2890                 cs_slot = &unconf->cl_cs_slot;
2891                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2892                 if (status) {
2893                         /* an unconfirmed replay returns misordered */
2894                         status = nfserr_seq_misordered;
2895                         goto out_free_conn;
2896                 }
2897                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2898                 if (old) {
2899                         status = mark_client_expired_locked(old);
2900                         if (status) {
2901                                 old = NULL;
2902                                 goto out_free_conn;
2903                         }
2904                 }
2905                 move_to_confirmed(unconf);
2906                 conf = unconf;
2907         } else {
2908                 status = nfserr_stale_clientid;
2909                 goto out_free_conn;
2910         }
2911         status = nfs_ok;
2912         /* Persistent sessions are not supported */
2913         cr_ses->flags &= ~SESSION4_PERSIST;
2914         /* Upshifting from TCP to RDMA is not supported */
2915         cr_ses->flags &= ~SESSION4_RDMA;
2916
2917         init_session(rqstp, new, conf, cr_ses);
2918         nfsd4_get_session_locked(new);
2919
2920         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2921                NFS4_MAX_SESSIONID_LEN);
2922         cs_slot->sl_seqid++;
2923         cr_ses->seqid = cs_slot->sl_seqid;
2924
2925         /* cache solo and embedded create sessions under the client_lock */
2926         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2927         spin_unlock(&nn->client_lock);
2928         /* init connection and backchannel */
2929         nfsd4_init_conn(rqstp, conn, new);
2930         nfsd4_put_session(new);
2931         if (old)
2932                 expire_client(old);
2933         return status;
2934 out_free_conn:
2935         spin_unlock(&nn->client_lock);
2936         free_conn(conn);
2937         if (old)
2938                 expire_client(old);
2939 out_free_session:
2940         __free_session(new);
2941 out_release_drc_mem:
2942         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2943         return status;
2944 }
2945
2946 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2947 {
2948         switch (*dir) {
2949         case NFS4_CDFC4_FORE:
2950         case NFS4_CDFC4_BACK:
2951                 return nfs_ok;
2952         case NFS4_CDFC4_FORE_OR_BOTH:
2953         case NFS4_CDFC4_BACK_OR_BOTH:
2954                 *dir = NFS4_CDFC4_BOTH;
2955                 return nfs_ok;
2956         };
2957         return nfserr_inval;
2958 }
2959
2960 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2961                 struct nfsd4_compound_state *cstate,
2962                 union nfsd4_op_u *u)
2963 {
2964         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2965         struct nfsd4_session *session = cstate->session;
2966         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2967         __be32 status;
2968
2969         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2970         if (status)
2971                 return status;
2972         spin_lock(&nn->client_lock);
2973         session->se_cb_prog = bc->bc_cb_program;
2974         session->se_cb_sec = bc->bc_cb_sec;
2975         spin_unlock(&nn->client_lock);
2976
2977         nfsd4_probe_callback(session->se_client);
2978
2979         return nfs_ok;
2980 }
2981
2982 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2983                      struct nfsd4_compound_state *cstate,
2984                      union nfsd4_op_u *u)
2985 {
2986         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2987         __be32 status;
2988         struct nfsd4_conn *conn;
2989         struct nfsd4_session *session;
2990         struct net *net = SVC_NET(rqstp);
2991         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2992
2993         if (!nfsd4_last_compound_op(rqstp))
2994                 return nfserr_not_only_op;
2995         spin_lock(&nn->client_lock);
2996         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2997         spin_unlock(&nn->client_lock);
2998         if (!session)
2999                 goto out_no_session;
3000         status = nfserr_wrong_cred;
3001         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3002                 goto out;
3003         status = nfsd4_map_bcts_dir(&bcts->dir);
3004         if (status)
3005                 goto out;
3006         conn = alloc_conn(rqstp, bcts->dir);
3007         status = nfserr_jukebox;
3008         if (!conn)
3009                 goto out;
3010         nfsd4_init_conn(rqstp, conn, session);
3011         status = nfs_ok;
3012 out:
3013         nfsd4_put_session(session);
3014 out_no_session:
3015         return status;
3016 }
3017
3018 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3019 {
3020         if (!cstate->session)
3021                 return false;
3022         return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3023 }
3024
3025 __be32
3026 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3027                 union nfsd4_op_u *u)
3028 {
3029         struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3030         struct nfsd4_session *ses;
3031         __be32 status;
3032         int ref_held_by_me = 0;
3033         struct net *net = SVC_NET(r);
3034         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3035
3036         status = nfserr_not_only_op;
3037         if (nfsd4_compound_in_session(cstate, sessionid)) {
3038                 if (!nfsd4_last_compound_op(r))
3039                         goto out;
3040                 ref_held_by_me++;
3041         }
3042         dump_sessionid(__func__, sessionid);
3043         spin_lock(&nn->client_lock);
3044         ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3045         if (!ses)
3046                 goto out_client_lock;
3047         status = nfserr_wrong_cred;
3048         if (!nfsd4_mach_creds_match(ses->se_client, r))
3049                 goto out_put_session;
3050         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3051         if (status)
3052                 goto out_put_session;
3053         unhash_session(ses);
3054         spin_unlock(&nn->client_lock);
3055
3056         nfsd4_probe_callback_sync(ses->se_client);
3057
3058         spin_lock(&nn->client_lock);
3059         status = nfs_ok;
3060 out_put_session:
3061         nfsd4_put_session_locked(ses);
3062 out_client_lock:
3063         spin_unlock(&nn->client_lock);
3064 out:
3065         return status;
3066 }
3067
3068 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3069 {
3070         struct nfsd4_conn *c;
3071
3072         list_for_each_entry(c, &s->se_conns, cn_persession) {
3073                 if (c->cn_xprt == xpt) {
3074                         return c;
3075                 }
3076         }
3077         return NULL;
3078 }
3079
3080 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3081 {
3082         struct nfs4_client *clp = ses->se_client;
3083         struct nfsd4_conn *c;
3084         __be32 status = nfs_ok;
3085         int ret;
3086
3087         spin_lock(&clp->cl_lock);
3088         c = __nfsd4_find_conn(new->cn_xprt, ses);
3089         if (c)
3090                 goto out_free;
3091         status = nfserr_conn_not_bound_to_session;
3092         if (clp->cl_mach_cred)
3093                 goto out_free;
3094         __nfsd4_hash_conn(new, ses);
3095         spin_unlock(&clp->cl_lock);
3096         ret = nfsd4_register_conn(new);
3097         if (ret)
3098                 /* oops; xprt is already down: */
3099                 nfsd4_conn_lost(&new->cn_xpt_user);
3100         return nfs_ok;
3101 out_free:
3102         spin_unlock(&clp->cl_lock);
3103         free_conn(new);
3104         return status;
3105 }
3106
3107 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3108 {
3109         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3110
3111         return args->opcnt > session->se_fchannel.maxops;
3112 }
3113
3114 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3115                                   struct nfsd4_session *session)
3116 {
3117         struct xdr_buf *xb = &rqstp->rq_arg;
3118
3119         return xb->len > session->se_fchannel.maxreq_sz;
3120 }
3121
3122 static bool replay_matches_cache(struct svc_rqst *rqstp,
3123                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3124 {
3125         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3126
3127         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3128             (bool)seq->cachethis)
3129                 return false;
3130         /*
3131          * If there's an error than the reply can have fewer ops than
3132          * the call.  But if we cached a reply with *more* ops than the
3133          * call you're sending us now, then this new call is clearly not
3134          * really a replay of the old one:
3135          */
3136         if (slot->sl_opcnt < argp->opcnt)
3137                 return false;
3138         /* This is the only check explicitly called by spec: */
3139         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3140                 return false;
3141         /*
3142          * There may be more comparisons we could actually do, but the
3143          * spec doesn't require us to catch every case where the calls
3144          * don't match (that would require caching the call as well as
3145          * the reply), so we don't bother.
3146          */
3147         return true;
3148 }
3149
3150 __be32
3151 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3152                 union nfsd4_op_u *u)
3153 {
3154         struct nfsd4_sequence *seq = &u->sequence;
3155         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3156         struct xdr_stream *xdr = &resp->xdr;
3157         struct nfsd4_session *session;
3158         struct nfs4_client *clp;
3159         struct nfsd4_slot *slot;
3160         struct nfsd4_conn *conn;
3161         __be32 status;
3162         int buflen;
3163         struct net *net = SVC_NET(rqstp);
3164         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3165
3166         if (resp->opcnt != 1)
3167                 return nfserr_sequence_pos;
3168
3169         /*
3170          * Will be either used or freed by nfsd4_sequence_check_conn
3171          * below.
3172          */
3173         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3174         if (!conn)
3175                 return nfserr_jukebox;
3176
3177         spin_lock(&nn->client_lock);
3178         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3179         if (!session)
3180                 goto out_no_session;
3181         clp = session->se_client;
3182
3183         status = nfserr_too_many_ops;
3184         if (nfsd4_session_too_many_ops(rqstp, session))
3185                 goto out_put_session;
3186
3187         status = nfserr_req_too_big;
3188         if (nfsd4_request_too_big(rqstp, session))
3189                 goto out_put_session;
3190
3191         status = nfserr_badslot;
3192         if (seq->slotid >= session->se_fchannel.maxreqs)
3193                 goto out_put_session;
3194
3195         slot = session->se_slots[seq->slotid];
3196         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3197
3198         /* We do not negotiate the number of slots yet, so set the
3199          * maxslots to the session maxreqs which is used to encode
3200          * sr_highest_slotid and the sr_target_slot id to maxslots */
3201         seq->maxslots = session->se_fchannel.maxreqs;
3202
3203         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3204                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3205         if (status == nfserr_replay_cache) {
3206                 status = nfserr_seq_misordered;
3207                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3208                         goto out_put_session;
3209                 status = nfserr_seq_false_retry;
3210                 if (!replay_matches_cache(rqstp, seq, slot))
3211                         goto out_put_session;
3212                 cstate->slot = slot;
3213                 cstate->session = session;
3214                 cstate->clp = clp;
3215                 /* Return the cached reply status and set cstate->status
3216                  * for nfsd4_proc_compound processing */
3217                 status = nfsd4_replay_cache_entry(resp, seq);
3218                 cstate->status = nfserr_replay_cache;
3219                 goto out;
3220         }
3221         if (status)
3222                 goto out_put_session;
3223
3224         status = nfsd4_sequence_check_conn(conn, session);
3225         conn = NULL;
3226         if (status)
3227                 goto out_put_session;
3228
3229         buflen = (seq->cachethis) ?
3230                         session->se_fchannel.maxresp_cached :
3231                         session->se_fchannel.maxresp_sz;
3232         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3233                                     nfserr_rep_too_big;
3234         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3235                 goto out_put_session;
3236         svc_reserve(rqstp, buflen);
3237
3238         status = nfs_ok;
3239         /* Success! bump slot seqid */
3240         slot->sl_seqid = seq->seqid;
3241         slot->sl_flags |= NFSD4_SLOT_INUSE;
3242         if (seq->cachethis)
3243                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3244         else
3245                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3246
3247         cstate->slot = slot;
3248         cstate->session = session;
3249         cstate->clp = clp;
3250
3251 out:
3252         switch (clp->cl_cb_state) {
3253         case NFSD4_CB_DOWN:
3254                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3255                 break;
3256         case NFSD4_CB_FAULT:
3257                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3258                 break;
3259         default:
3260                 seq->status_flags = 0;
3261         }
3262         if (!list_empty(&clp->cl_revoked))
3263                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3264 out_no_session:
3265         if (conn)
3266                 free_conn(conn);
3267         spin_unlock(&nn->client_lock);
3268         return status;
3269 out_put_session:
3270         nfsd4_put_session_locked(session);
3271         goto out_no_session;
3272 }
3273
3274 void
3275 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3276 {
3277         struct nfsd4_compound_state *cs = &resp->cstate;
3278
3279         if (nfsd4_has_session(cs)) {
3280                 if (cs->status != nfserr_replay_cache) {
3281                         nfsd4_store_cache_entry(resp);
3282                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3283                 }
3284                 /* Drop session reference that was taken in nfsd4_sequence() */
3285                 nfsd4_put_session(cs->session);
3286         } else if (cs->clp)
3287                 put_client_renew(cs->clp);
3288 }
3289
3290 __be32
3291 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3292                 struct nfsd4_compound_state *cstate,
3293                 union nfsd4_op_u *u)
3294 {
3295         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3296         struct nfs4_client *conf, *unconf;
3297         struct nfs4_client *clp = NULL;
3298         __be32 status = 0;
3299         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3300
3301         spin_lock(&nn->client_lock);
3302         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3303         conf = find_confirmed_client(&dc->clientid, true, nn);
3304         WARN_ON_ONCE(conf && unconf);
3305
3306         if (conf) {
3307                 if (client_has_state(conf)) {
3308                         status = nfserr_clientid_busy;
3309                         goto out;
3310                 }
3311                 status = mark_client_expired_locked(conf);
3312                 if (status)
3313                         goto out;
3314                 clp = conf;
3315         } else if (unconf)
3316                 clp = unconf;
3317         else {
3318                 status = nfserr_stale_clientid;
3319                 goto out;
3320         }
3321         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3322                 clp = NULL;
3323                 status = nfserr_wrong_cred;
3324                 goto out;
3325         }
3326         unhash_client_locked(clp);
3327 out:
3328         spin_unlock(&nn->client_lock);
3329         if (clp)
3330                 expire_client(clp);
3331         return status;
3332 }
3333
3334 __be32
3335 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3336                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3337 {
3338         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3339         __be32 status = 0;
3340
3341         if (rc->rca_one_fs) {
3342                 if (!cstate->current_fh.fh_dentry)
3343                         return nfserr_nofilehandle;
3344                 /*
3345                  * We don't take advantage of the rca_one_fs case.
3346                  * That's OK, it's optional, we can safely ignore it.
3347                  */
3348                 return nfs_ok;
3349         }
3350
3351         status = nfserr_complete_already;
3352         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3353                              &cstate->session->se_client->cl_flags))
3354                 goto out;
3355
3356         status = nfserr_stale_clientid;
3357         if (is_client_expired(cstate->session->se_client))
3358                 /*
3359                  * The following error isn't really legal.
3360                  * But we only get here if the client just explicitly
3361                  * destroyed the client.  Surely it no longer cares what
3362                  * error it gets back on an operation for the dead
3363                  * client.
3364                  */
3365                 goto out;
3366
3367         status = nfs_ok;
3368         nfsd4_client_record_create(cstate->session->se_client);
3369         inc_reclaim_complete(cstate->session->se_client);
3370 out:
3371         return status;
3372 }
3373
3374 __be32
3375 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3376                   union nfsd4_op_u *u)
3377 {
3378         struct nfsd4_setclientid *setclid = &u->setclientid;
3379         struct xdr_netobj       clname = setclid->se_name;
3380         nfs4_verifier           clverifier = setclid->se_verf;
3381         struct nfs4_client      *conf, *new;
3382         struct nfs4_client      *unconf = NULL;
3383         __be32                  status;
3384         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3385
3386         new = create_client(clname, rqstp, &clverifier);
3387         if (new == NULL)
3388                 return nfserr_jukebox;
3389         /* Cases below refer to rfc 3530 section 14.2.33: */
3390         spin_lock(&nn->client_lock);
3391         conf = find_confirmed_client_by_name(&clname, nn);
3392         if (conf && client_has_state(conf)) {
3393                 /* case 0: */
3394                 status = nfserr_clid_inuse;
3395                 if (clp_used_exchangeid(conf))
3396                         goto out;
3397                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3398                         char addr_str[INET6_ADDRSTRLEN];
3399                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3400                                  sizeof(addr_str));
3401                         dprintk("NFSD: setclientid: string in use by client "
3402                                 "at %s\n", addr_str);
3403                         goto out;
3404                 }
3405         }
3406         unconf = find_unconfirmed_client_by_name(&clname, nn);
3407         if (unconf)
3408                 unhash_client_locked(unconf);
3409         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3410                 /* case 1: probable callback update */
3411                 copy_clid(new, conf);
3412                 gen_confirm(new, nn);
3413         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3414                 gen_clid(new, nn);
3415         new->cl_minorversion = 0;
3416         gen_callback(new, setclid, rqstp);
3417         add_to_unconfirmed(new);
3418         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3419         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3420         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3421         new = NULL;
3422         status = nfs_ok;
3423 out:
3424         spin_unlock(&nn->client_lock);
3425         if (new)
3426                 free_client(new);
3427         if (unconf)
3428                 expire_client(unconf);
3429         return status;
3430 }
3431
3432
3433 __be32
3434 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3435                         struct nfsd4_compound_state *cstate,
3436                         union nfsd4_op_u *u)
3437 {
3438         struct nfsd4_setclientid_confirm *setclientid_confirm =
3439                         &u->setclientid_confirm;
3440         struct nfs4_client *conf, *unconf;
3441         struct nfs4_client *old = NULL;
3442         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3443         clientid_t * clid = &setclientid_confirm->sc_clientid;
3444         __be32 status;
3445         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3446
3447         if (STALE_CLIENTID(clid, nn))
3448                 return nfserr_stale_clientid;
3449
3450         spin_lock(&nn->client_lock);
3451         conf = find_confirmed_client(clid, false, nn);
3452         unconf = find_unconfirmed_client(clid, false, nn);
3453         /*
3454          * We try hard to give out unique clientid's, so if we get an
3455          * attempt to confirm the same clientid with a different cred,
3456          * the client may be buggy; this should never happen.
3457          *
3458          * Nevertheless, RFC 7530 recommends INUSE for this case:
3459          */
3460         status = nfserr_clid_inuse;
3461         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3462                 goto out;
3463         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3464                 goto out;
3465         /* cases below refer to rfc 3530 section 14.2.34: */
3466         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3467                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3468                         /* case 2: probable retransmit */
3469                         status = nfs_ok;
3470                 } else /* case 4: client hasn't noticed we rebooted yet? */
3471                         status = nfserr_stale_clientid;
3472                 goto out;
3473         }
3474         status = nfs_ok;
3475         if (conf) { /* case 1: callback update */
3476                 old = unconf;
3477                 unhash_client_locked(old);
3478                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3479         } else { /* case 3: normal case; new or rebooted client */
3480                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3481                 if (old) {
3482                         status = nfserr_clid_inuse;
3483                         if (client_has_state(old)
3484                                         && !same_creds(&unconf->cl_cred,
3485                                                         &old->cl_cred))
3486                                 goto out;
3487                         status = mark_client_expired_locked(old);
3488                         if (status) {
3489                                 old = NULL;
3490                                 goto out;
3491                         }
3492                 }
3493                 move_to_confirmed(unconf);
3494                 conf = unconf;
3495         }
3496         get_client_locked(conf);
3497         spin_unlock(&nn->client_lock);
3498         nfsd4_probe_callback(conf);
3499         spin_lock(&nn->client_lock);
3500         put_client_renew_locked(conf);
3501 out:
3502         spin_unlock(&nn->client_lock);
3503         if (old)
3504                 expire_client(old);
3505         return status;
3506 }
3507
3508 static struct nfs4_file *nfsd4_alloc_file(void)
3509 {
3510         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3511 }
3512
3513 /* OPEN Share state helper functions */
3514 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3515                                 struct nfs4_file *fp)
3516 {
3517         lockdep_assert_held(&state_lock);
3518
3519         refcount_set(&fp->fi_ref, 1);
3520         spin_lock_init(&fp->fi_lock);
3521         INIT_LIST_HEAD(&fp->fi_stateids);
3522         INIT_LIST_HEAD(&fp->fi_delegations);
3523         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3524         fh_copy_shallow(&fp->fi_fhandle, fh);
3525         fp->fi_deleg_file = NULL;
3526         fp->fi_had_conflict = false;
3527         fp->fi_share_deny = 0;
3528         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3529         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3530 #ifdef CONFIG_NFSD_PNFS
3531         INIT_LIST_HEAD(&fp->fi_lo_states);
3532         atomic_set(&fp->fi_lo_recalls, 0);
3533 #endif
3534         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3535 }
3536
3537 void
3538 nfsd4_free_slabs(void)
3539 {
3540         kmem_cache_destroy(client_slab);
3541         kmem_cache_destroy(openowner_slab);
3542         kmem_cache_destroy(lockowner_slab);
3543         kmem_cache_destroy(file_slab);
3544         kmem_cache_destroy(stateid_slab);
3545         kmem_cache_destroy(deleg_slab);
3546         kmem_cache_destroy(odstate_slab);
3547 }
3548
3549 int
3550 nfsd4_init_slabs(void)
3551 {
3552         client_slab = kmem_cache_create("nfsd4_clients",
3553                         sizeof(struct nfs4_client), 0, 0, NULL);
3554         if (client_slab == NULL)
3555                 goto out;
3556         openowner_slab = kmem_cache_create("nfsd4_openowners",
3557                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3558         if (openowner_slab == NULL)
3559                 goto out_free_client_slab;
3560         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3561                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3562         if (lockowner_slab == NULL)
3563                 goto out_free_openowner_slab;
3564         file_slab = kmem_cache_create("nfsd4_files",
3565                         sizeof(struct nfs4_file), 0, 0, NULL);
3566         if (file_slab == NULL)
3567                 goto out_free_lockowner_slab;
3568         stateid_slab = kmem_cache_create("nfsd4_stateids",
3569                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3570         if (stateid_slab == NULL)
3571                 goto out_free_file_slab;
3572         deleg_slab = kmem_cache_create("nfsd4_delegations",
3573                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3574         if (deleg_slab == NULL)
3575                 goto out_free_stateid_slab;
3576         odstate_slab = kmem_cache_create("nfsd4_odstate",
3577                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3578         if (odstate_slab == NULL)
3579                 goto out_free_deleg_slab;
3580         return 0;
3581
3582 out_free_deleg_slab:
3583         kmem_cache_destroy(deleg_slab);
3584 out_free_stateid_slab:
3585         kmem_cache_destroy(stateid_slab);
3586 out_free_file_slab:
3587         kmem_cache_destroy(file_slab);
3588 out_free_lockowner_slab:
3589         kmem_cache_destroy(lockowner_slab);
3590 out_free_openowner_slab:
3591         kmem_cache_destroy(openowner_slab);
3592 out_free_client_slab:
3593         kmem_cache_destroy(client_slab);
3594 out:
3595         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3596         return -ENOMEM;
3597 }
3598
3599 static void init_nfs4_replay(struct nfs4_replay *rp)
3600 {
3601         rp->rp_status = nfserr_serverfault;
3602         rp->rp_buflen = 0;
3603         rp->rp_buf = rp->rp_ibuf;
3604         mutex_init(&rp->rp_mutex);
3605 }
3606
3607 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3608                 struct nfs4_stateowner *so)
3609 {
3610         if (!nfsd4_has_session(cstate)) {
3611                 mutex_lock(&so->so_replay.rp_mutex);
3612                 cstate->replay_owner = nfs4_get_stateowner(so);
3613         }
3614 }
3615
3616 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3617 {
3618         struct nfs4_stateowner *so = cstate->replay_owner;
3619
3620         if (so != NULL) {
3621                 cstate->replay_owner = NULL;
3622                 mutex_unlock(&so->so_replay.rp_mutex);
3623                 nfs4_put_stateowner(so);
3624         }
3625 }
3626
3627 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3628 {
3629         struct nfs4_stateowner *sop;
3630
3631         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3632         if (!sop)
3633                 return NULL;
3634
3635         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3636         if (!sop->so_owner.data) {
3637                 kmem_cache_free(slab, sop);
3638                 return NULL;
3639         }
3640         sop->so_owner.len = owner->len;
3641
3642         INIT_LIST_HEAD(&sop->so_stateids);
3643         sop->so_client = clp;
3644         init_nfs4_replay(&sop->so_replay);
3645         atomic_set(&sop->so_count, 1);
3646         return sop;
3647 }
3648
3649 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3650 {
3651         lockdep_assert_held(&clp->cl_lock);
3652
3653         list_add(&oo->oo_owner.so_strhash,
3654                  &clp->cl_ownerstr_hashtbl[strhashval]);
3655         list_add(&oo->oo_perclient, &clp->cl_openowners);
3656 }
3657
3658 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3659 {
3660         unhash_openowner_locked(openowner(so));
3661 }
3662
3663 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3664 {
3665         struct nfs4_openowner *oo = openowner(so);
3666
3667         kmem_cache_free(openowner_slab, oo);
3668 }
3669
3670 static const struct nfs4_stateowner_operations openowner_ops = {
3671         .so_unhash =    nfs4_unhash_openowner,
3672         .so_free =      nfs4_free_openowner,
3673 };
3674
3675 static struct nfs4_ol_stateid *
3676 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3677 {
3678         struct nfs4_ol_stateid *local, *ret = NULL;
3679         struct nfs4_openowner *oo = open->op_openowner;
3680
3681         lockdep_assert_held(&fp->fi_lock);
3682
3683         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3684                 /* ignore lock owners */
3685                 if (local->st_stateowner->so_is_open_owner == 0)
3686                         continue;
3687                 if (local->st_stateowner != &oo->oo_owner)
3688                         continue;
3689                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3690                         ret = local;
3691                         refcount_inc(&ret->st_stid.sc_count);
3692                         break;
3693                 }
3694         }
3695         return ret;
3696 }
3697
3698 static __be32
3699 nfsd4_verify_open_stid(struct nfs4_stid *s)
3700 {
3701         __be32 ret = nfs_ok;
3702
3703         switch (s->sc_type) {
3704         default:
3705                 break;
3706         case 0:
3707         case NFS4_CLOSED_STID:
3708         case NFS4_CLOSED_DELEG_STID:
3709                 ret = nfserr_bad_stateid;
3710                 break;
3711         case NFS4_REVOKED_DELEG_STID:
3712                 ret = nfserr_deleg_revoked;
3713         }
3714         return ret;
3715 }
3716
3717 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3718 static __be32
3719 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3720 {
3721         __be32 ret;
3722
3723         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3724         ret = nfsd4_verify_open_stid(&stp->st_stid);
3725         if (ret != nfs_ok)
3726                 mutex_unlock(&stp->st_mutex);
3727         return ret;
3728 }
3729
3730 static struct nfs4_ol_stateid *
3731 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3732 {
3733         struct nfs4_ol_stateid *stp;
3734         for (;;) {
3735                 spin_lock(&fp->fi_lock);
3736                 stp = nfsd4_find_existing_open(fp, open);
3737                 spin_unlock(&fp->fi_lock);
3738                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3739                         break;
3740                 nfs4_put_stid(&stp->st_stid);
3741         }
3742         return stp;
3743 }
3744
3745 static struct nfs4_openowner *
3746 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3747                            struct nfsd4_compound_state *cstate)
3748 {
3749         struct nfs4_client *clp = cstate->clp;
3750         struct nfs4_openowner *oo, *ret;
3751
3752         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3753         if (!oo)
3754                 return NULL;
3755         oo->oo_owner.so_ops = &openowner_ops;
3756         oo->oo_owner.so_is_open_owner = 1;
3757         oo->oo_owner.so_seqid = open->op_seqid;
3758         oo->oo_flags = 0;
3759         if (nfsd4_has_session(cstate))
3760                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3761         oo->oo_time = 0;
3762         oo->oo_last_closed_stid = NULL;
3763         INIT_LIST_HEAD(&oo->oo_close_lru);
3764         spin_lock(&clp->cl_lock);
3765         ret = find_openstateowner_str_locked(strhashval, open, clp);
3766         if (ret == NULL) {
3767                 hash_openowner(oo, clp, strhashval);
3768                 ret = oo;
3769         } else
3770                 nfs4_free_stateowner(&oo->oo_owner);
3771
3772         spin_unlock(&clp->cl_lock);
3773         return ret;
3774 }
3775
3776 static struct nfs4_ol_stateid *
3777 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3778 {
3779
3780         struct nfs4_openowner *oo = open->op_openowner;
3781         struct nfs4_ol_stateid *retstp = NULL;
3782         struct nfs4_ol_stateid *stp;
3783
3784         stp = open->op_stp;
3785         /* We are moving these outside of the spinlocks to avoid the warnings */
3786         mutex_init(&stp->st_mutex);
3787         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3788
3789 retry:
3790         spin_lock(&oo->oo_owner.so_client->cl_lock);
3791         spin_lock(&fp->fi_lock);
3792
3793         retstp = nfsd4_find_existing_open(fp, open);
3794         if (retstp)
3795                 goto out_unlock;
3796
3797         open->op_stp = NULL;
3798         refcount_inc(&stp->st_stid.sc_count);
3799         stp->st_stid.sc_type = NFS4_OPEN_STID;
3800         INIT_LIST_HEAD(&stp->st_locks);
3801         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3802         get_nfs4_file(fp);
3803         stp->st_stid.sc_file = fp;
3804         stp->st_access_bmap = 0;
3805         stp->st_deny_bmap = 0;
3806         stp->st_openstp = NULL;
3807         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3808         list_add(&stp->st_perfile, &fp->fi_stateids);
3809
3810 out_unlock:
3811         spin_unlock(&fp->fi_lock);
3812         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3813         if (retstp) {
3814                 /* Handle races with CLOSE */
3815                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3816                         nfs4_put_stid(&retstp->st_stid);
3817                         goto retry;
3818                 }
3819                 /* To keep mutex tracking happy */
3820                 mutex_unlock(&stp->st_mutex);
3821                 stp = retstp;
3822         }
3823         return stp;
3824 }
3825
3826 /*
3827  * In the 4.0 case we need to keep the owners around a little while to handle
3828  * CLOSE replay. We still do need to release any file access that is held by
3829  * them before returning however.
3830  */
3831 static void
3832 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3833 {
3834         struct nfs4_ol_stateid *last;
3835         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3836         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3837                                                 nfsd_net_id);
3838
3839         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3840
3841         /*
3842          * We know that we hold one reference via nfsd4_close, and another
3843          * "persistent" reference for the client. If the refcount is higher
3844          * than 2, then there are still calls in progress that are using this
3845          * stateid. We can't put the sc_file reference until they are finished.
3846          * Wait for the refcount to drop to 2. Since it has been unhashed,
3847          * there should be no danger of the refcount going back up again at
3848          * this point.
3849          */
3850         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3851
3852         release_all_access(s);
3853         if (s->st_stid.sc_file) {
3854                 put_nfs4_file(s->st_stid.sc_file);
3855                 s->st_stid.sc_file = NULL;
3856         }
3857
3858         spin_lock(&nn->client_lock);
3859         last = oo->oo_last_closed_stid;
3860         oo->oo_last_closed_stid = s;
3861         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3862         oo->oo_time = get_seconds();
3863         spin_unlock(&nn->client_lock);
3864         if (last)
3865                 nfs4_put_stid(&last->st_stid);
3866 }
3867
3868 /* search file_hashtbl[] for file */
3869 static struct nfs4_file *
3870 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3871 {
3872         struct nfs4_file *fp;
3873
3874         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3875                 if (fh_match(&fp->fi_fhandle, fh)) {
3876                         if (refcount_inc_not_zero(&fp->fi_ref))
3877                                 return fp;
3878                 }
3879         }
3880         return NULL;
3881 }
3882
3883 struct nfs4_file *
3884 find_file(struct knfsd_fh *fh)
3885 {
3886         struct nfs4_file *fp;
3887         unsigned int hashval = file_hashval(fh);
3888
3889         rcu_read_lock();
3890         fp = find_file_locked(fh, hashval);
3891         rcu_read_unlock();
3892         return fp;
3893 }
3894
3895 static struct nfs4_file *
3896 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3897 {
3898         struct nfs4_file *fp;
3899         unsigned int hashval = file_hashval(fh);
3900
3901         rcu_read_lock();
3902         fp = find_file_locked(fh, hashval);
3903         rcu_read_unlock();
3904         if (fp)
3905                 return fp;
3906
3907         spin_lock(&state_lock);
3908         fp = find_file_locked(fh, hashval);
3909         if (likely(fp == NULL)) {
3910                 nfsd4_init_file(fh, hashval, new);
3911                 fp = new;
3912         }
3913         spin_unlock(&state_lock);
3914
3915         return fp;
3916 }
3917
3918 /*
3919  * Called to check deny when READ with all zero stateid or
3920  * WRITE with all zero or all one stateid
3921  */
3922 static __be32
3923 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3924 {
3925         struct nfs4_file *fp;
3926         __be32 ret = nfs_ok;
3927
3928         fp = find_file(&current_fh->fh_handle);
3929         if (!fp)
3930                 return ret;
3931         /* Check for conflicting share reservations */
3932         spin_lock(&fp->fi_lock);
3933         if (fp->fi_share_deny & deny_type)
3934                 ret = nfserr_locked;
3935         spin_unlock(&fp->fi_lock);
3936         put_nfs4_file(fp);
3937         return ret;
3938 }
3939
3940 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3941 {
3942         struct nfs4_delegation *dp = cb_to_delegation(cb);
3943         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3944                                           nfsd_net_id);
3945
3946         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3947
3948         /*
3949          * We can't do this in nfsd_break_deleg_cb because it is
3950          * already holding inode->i_lock.
3951          *
3952          * If the dl_time != 0, then we know that it has already been
3953          * queued for a lease break. Don't queue it again.
3954          */
3955         spin_lock(&state_lock);
3956         if (dp->dl_time == 0) {
3957                 dp->dl_time = get_seconds();
3958                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3959         }
3960         spin_unlock(&state_lock);
3961 }
3962
3963 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3964                 struct rpc_task *task)
3965 {
3966         struct nfs4_delegation *dp = cb_to_delegation(cb);
3967
3968         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3969                 return 1;
3970
3971         switch (task->tk_status) {
3972         case 0:
3973                 return 1;
3974         case -NFS4ERR_DELAY:
3975                 rpc_delay(task, 2 * HZ);
3976                 return 0;
3977         case -EBADHANDLE:
3978         case -NFS4ERR_BAD_STATEID:
3979                 /*
3980                  * Race: client probably got cb_recall before open reply
3981                  * granting delegation.
3982                  */
3983                 if (dp->dl_retries--) {
3984                         rpc_delay(task, 2 * HZ);
3985                         return 0;
3986                 }
3987                 /*FALLTHRU*/
3988         default:
3989                 return 1;
3990         }
3991 }
3992
3993 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3994 {
3995         struct nfs4_delegation *dp = cb_to_delegation(cb);
3996
3997         nfs4_put_stid(&dp->dl_stid);
3998 }
3999
4000 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4001         .prepare        = nfsd4_cb_recall_prepare,
4002         .done           = nfsd4_cb_recall_done,
4003         .release        = nfsd4_cb_recall_release,
4004 };
4005
4006 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4007 {
4008         /*
4009          * We're assuming the state code never drops its reference
4010          * without first removing the lease.  Since we're in this lease
4011          * callback (and since the lease code is serialized by the
4012          * i_lock) we know the server hasn't removed the lease yet, and
4013          * we know it's safe to take a reference.
4014          */
4015         refcount_inc(&dp->dl_stid.sc_count);
4016         nfsd4_run_cb(&dp->dl_recall);
4017 }
4018
4019 /* Called from break_lease() with i_lock held. */
4020 static bool
4021 nfsd_break_deleg_cb(struct file_lock *fl)
4022 {
4023         bool ret = false;
4024         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4025         struct nfs4_file *fp = dp->dl_stid.sc_file;
4026
4027         /*
4028          * We don't want the locks code to timeout the lease for us;
4029          * we'll remove it ourself if a delegation isn't returned
4030          * in time:
4031          */
4032         fl->fl_break_time = 0;
4033
4034         spin_lock(&fp->fi_lock);
4035         fp->fi_had_conflict = true;
4036         nfsd_break_one_deleg(dp);
4037         spin_unlock(&fp->fi_lock);
4038         return ret;
4039 }
4040
4041 static int
4042 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4043                      struct list_head *dispose)
4044 {
4045         if (arg & F_UNLCK)
4046                 return lease_modify(onlist, arg, dispose);
4047         else
4048                 return -EAGAIN;
4049 }
4050
4051 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4052         .lm_break = nfsd_break_deleg_cb,
4053         .lm_change = nfsd_change_deleg_cb,
4054 };
4055
4056 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4057 {
4058         if (nfsd4_has_session(cstate))
4059                 return nfs_ok;
4060         if (seqid == so->so_seqid - 1)
4061                 return nfserr_replay_me;
4062         if (seqid == so->so_seqid)
4063                 return nfs_ok;
4064         return nfserr_bad_seqid;
4065 }
4066
4067 static __be32 lookup_clientid(clientid_t *clid,
4068                 struct nfsd4_compound_state *cstate,
4069                 struct nfsd_net *nn)
4070 {
4071         struct nfs4_client *found;
4072
4073         if (cstate->clp) {
4074                 found = cstate->clp;
4075                 if (!same_clid(&found->cl_clientid, clid))
4076                         return nfserr_stale_clientid;
4077                 return nfs_ok;
4078         }
4079
4080         if (STALE_CLIENTID(clid, nn))
4081                 return nfserr_stale_clientid;
4082
4083         /*
4084          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4085          * cached already then we know this is for is for v4.0 and "sessions"
4086          * will be false.
4087          */
4088         WARN_ON_ONCE(cstate->session);
4089         spin_lock(&nn->client_lock);
4090         found = find_confirmed_client(clid, false, nn);
4091         if (!found) {
4092                 spin_unlock(&nn->client_lock);
4093                 return nfserr_expired;
4094         }
4095         atomic_inc(&found->cl_refcount);
4096         spin_unlock(&nn->client_lock);
4097
4098         /* Cache the nfs4_client in cstate! */
4099         cstate->clp = found;
4100         return nfs_ok;
4101 }
4102
4103 __be32
4104 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4105                     struct nfsd4_open *open, struct nfsd_net *nn)
4106 {
4107         clientid_t *clientid = &open->op_clientid;
4108         struct nfs4_client *clp = NULL;
4109         unsigned int strhashval;
4110         struct nfs4_openowner *oo = NULL;
4111         __be32 status;
4112
4113         if (STALE_CLIENTID(&open->op_clientid, nn))
4114                 return nfserr_stale_clientid;
4115         /*
4116          * In case we need it later, after we've already created the
4117          * file and don't want to risk a further failure:
4118          */
4119         open->op_file = nfsd4_alloc_file();
4120         if (open->op_file == NULL)
4121                 return nfserr_jukebox;
4122
4123         status = lookup_clientid(clientid, cstate, nn);
4124         if (status)
4125                 return status;
4126         clp = cstate->clp;
4127
4128         strhashval = ownerstr_hashval(&open->op_owner);
4129         oo = find_openstateowner_str(strhashval, open, clp);
4130         open->op_openowner = oo;
4131         if (!oo) {
4132                 goto new_owner;
4133         }
4134         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4135                 /* Replace unconfirmed owners without checking for replay. */
4136                 release_openowner(oo);
4137                 open->op_openowner = NULL;
4138                 goto new_owner;
4139         }
4140         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4141         if (status)
4142                 return status;
4143         goto alloc_stateid;
4144 new_owner:
4145         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4146         if (oo == NULL)
4147                 return nfserr_jukebox;
4148         open->op_openowner = oo;
4149 alloc_stateid:
4150         open->op_stp = nfs4_alloc_open_stateid(clp);
4151         if (!open->op_stp)
4152                 return nfserr_jukebox;
4153
4154         if (nfsd4_has_session(cstate) &&
4155             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4156                 open->op_odstate = alloc_clnt_odstate(clp);
4157                 if (!open->op_odstate)
4158                         return nfserr_jukebox;
4159         }
4160
4161         return nfs_ok;
4162 }
4163
4164 static inline __be32
4165 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4166 {
4167         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4168                 return nfserr_openmode;
4169         else
4170                 return nfs_ok;
4171 }
4172
4173 static int share_access_to_flags(u32 share_access)
4174 {
4175         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4176 }
4177
4178 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4179 {
4180         struct nfs4_stid *ret;
4181
4182         ret = find_stateid_by_type(cl, s,
4183                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4184         if (!ret)
4185                 return NULL;
4186         return delegstateid(ret);
4187 }
4188
4189 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4190 {
4191         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4192                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4193 }
4194
4195 static __be32
4196 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4197                 struct nfs4_delegation **dp)
4198 {
4199         int flags;
4200         __be32 status = nfserr_bad_stateid;
4201         struct nfs4_delegation *deleg;
4202
4203         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4204         if (deleg == NULL)
4205                 goto out;
4206         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4207                 nfs4_put_stid(&deleg->dl_stid);
4208                 if (cl->cl_minorversion)
4209                         status = nfserr_deleg_revoked;
4210                 goto out;
4211         }
4212         flags = share_access_to_flags(open->op_share_access);
4213         status = nfs4_check_delegmode(deleg, flags);
4214         if (status) {
4215                 nfs4_put_stid(&deleg->dl_stid);
4216                 goto out;
4217         }
4218         *dp = deleg;
4219 out:
4220         if (!nfsd4_is_deleg_cur(open))
4221                 return nfs_ok;
4222         if (status)
4223                 return status;
4224         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4225         return nfs_ok;
4226 }
4227
4228 static inline int nfs4_access_to_access(u32 nfs4_access)
4229 {
4230         int flags = 0;
4231
4232         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4233                 flags |= NFSD_MAY_READ;
4234         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4235                 flags |= NFSD_MAY_WRITE;
4236         return flags;
4237 }
4238
4239 static inline __be32
4240 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4241                 struct nfsd4_open *open)
4242 {
4243         struct iattr iattr = {
4244                 .ia_valid = ATTR_SIZE,
4245                 .ia_size = 0,
4246         };
4247         if (!open->op_truncate)
4248                 return 0;
4249         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4250                 return nfserr_inval;
4251         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4252 }
4253
4254 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4255                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4256                 struct nfsd4_open *open)
4257 {
4258         struct file *filp = NULL;
4259         __be32 status;
4260         int oflag = nfs4_access_to_omode(open->op_share_access);
4261         int access = nfs4_access_to_access(open->op_share_access);
4262         unsigned char old_access_bmap, old_deny_bmap;
4263
4264         spin_lock(&fp->fi_lock);
4265
4266         /*
4267          * Are we trying to set a deny mode that would conflict with
4268          * current access?
4269          */
4270         status = nfs4_file_check_deny(fp, open->op_share_deny);
4271         if (status != nfs_ok) {
4272                 spin_unlock(&fp->fi_lock);
4273                 goto out;
4274         }
4275
4276         /* set access to the file */
4277         status = nfs4_file_get_access(fp, open->op_share_access);
4278         if (status != nfs_ok) {
4279                 spin_unlock(&fp->fi_lock);
4280                 goto out;
4281         }
4282
4283         /* Set access bits in stateid */
4284         old_access_bmap = stp->st_access_bmap;
4285         set_access(open->op_share_access, stp);
4286
4287         /* Set new deny mask */
4288         old_deny_bmap = stp->st_deny_bmap;
4289         set_deny(open->op_share_deny, stp);
4290         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4291
4292         if (!fp->fi_fds[oflag]) {
4293                 spin_unlock(&fp->fi_lock);
4294                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4295                 if (status)
4296                         goto out_put_access;
4297                 spin_lock(&fp->fi_lock);
4298                 if (!fp->fi_fds[oflag]) {
4299                         fp->fi_fds[oflag] = filp;
4300                         filp = NULL;
4301                 }
4302         }
4303         spin_unlock(&fp->fi_lock);
4304         if (filp)
4305                 fput(filp);
4306
4307         status = nfsd4_truncate(rqstp, cur_fh, open);
4308         if (status)
4309                 goto out_put_access;
4310 out:
4311         return status;
4312 out_put_access:
4313         stp->st_access_bmap = old_access_bmap;
4314         nfs4_file_put_access(fp, open->op_share_access);
4315         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4316         goto out;
4317 }
4318
4319 static __be32
4320 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)
4321 {
4322         __be32 status;
4323         unsigned char old_deny_bmap = stp->st_deny_bmap;
4324
4325         if (!test_access(open->op_share_access, stp))
4326                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4327
4328         /* test and set deny mode */
4329         spin_lock(&fp->fi_lock);
4330         status = nfs4_file_check_deny(fp, open->op_share_deny);
4331         if (status == nfs_ok) {
4332                 set_deny(open->op_share_deny, stp);
4333                 fp->fi_share_deny |=
4334                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4335         }
4336         spin_unlock(&fp->fi_lock);
4337
4338         if (status != nfs_ok)
4339                 return status;
4340
4341         status = nfsd4_truncate(rqstp, cur_fh, open);
4342         if (status != nfs_ok)
4343                 reset_union_bmap_deny(old_deny_bmap, stp);
4344         return status;
4345 }
4346
4347 /* Should we give out recallable state?: */
4348 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4349 {
4350         if (clp->cl_cb_state == NFSD4_CB_UP)
4351                 return true;
4352         /*
4353          * In the sessions case, since we don't have to establish a
4354          * separate connection for callbacks, we assume it's OK
4355          * until we hear otherwise:
4356          */
4357         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4358 }
4359
4360 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4361                                                 int flag)
4362 {
4363         struct file_lock *fl;
4364
4365         fl = locks_alloc_lock();
4366         if (!fl)
4367                 return NULL;
4368         fl->fl_lmops = &nfsd_lease_mng_ops;
4369         fl->fl_flags = FL_DELEG;
4370         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4371         fl->fl_end = OFFSET_MAX;
4372         fl->fl_owner = (fl_owner_t)dp;
4373         fl->fl_pid = current->tgid;
4374         fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4375         return fl;
4376 }
4377
4378 static struct nfs4_delegation *
4379 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4380                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4381 {
4382         int status = 0;
4383         struct nfs4_delegation *dp;
4384         struct file *filp;
4385         struct file_lock *fl;
4386
4387         /*
4388          * The fi_had_conflict and nfs_get_existing_delegation checks
4389          * here are just optimizations; we'll need to recheck them at
4390          * the end:
4391          */
4392         if (fp->fi_had_conflict)
4393                 return ERR_PTR(-EAGAIN);
4394
4395         filp = find_readable_file(fp);
4396         if (!filp) {
4397                 /* We should always have a readable file here */
4398                 WARN_ON_ONCE(1);
4399                 return ERR_PTR(-EBADF);
4400         }
4401         spin_lock(&state_lock);
4402         spin_lock(&fp->fi_lock);
4403         if (nfs4_delegation_exists(clp, fp))
4404                 status = -EAGAIN;
4405         else if (!fp->fi_deleg_file) {
4406                 fp->fi_deleg_file = filp;
4407                 /* increment early to prevent fi_deleg_file from being
4408                  * cleared */
4409                 fp->fi_delegees = 1;
4410                 filp = NULL;
4411         } else
4412                 fp->fi_delegees++;
4413         spin_unlock(&fp->fi_lock);
4414         spin_unlock(&state_lock);
4415         if (filp)
4416                 fput(filp);
4417         if (status)
4418                 return ERR_PTR(status);
4419
4420         status = -ENOMEM;
4421         dp = alloc_init_deleg(clp, fp, fh, odstate);
4422         if (!dp)
4423                 goto out_delegees;
4424
4425         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4426         if (!fl)
4427                 goto out_clnt_odstate;
4428
4429         status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4430         if (fl)
4431                 locks_free_lock(fl);
4432         if (status)
4433                 goto out_clnt_odstate;
4434
4435         spin_lock(&state_lock);
4436         spin_lock(&fp->fi_lock);
4437         if (fp->fi_had_conflict)
4438                 status = -EAGAIN;
4439         else
4440                 status = hash_delegation_locked(dp, fp);
4441         spin_unlock(&fp->fi_lock);
4442         spin_unlock(&state_lock);
4443
4444         if (status)
4445                 goto out_unlock;
4446
4447         return dp;
4448 out_unlock:
4449         vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4450 out_clnt_odstate:
4451         put_clnt_odstate(dp->dl_clnt_odstate);
4452         nfs4_put_stid(&dp->dl_stid);
4453 out_delegees:
4454         put_deleg_file(fp);
4455         return ERR_PTR(status);
4456 }
4457
4458 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4459 {
4460         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4461         if (status == -EAGAIN)
4462                 open->op_why_no_deleg = WND4_CONTENTION;
4463         else {
4464                 open->op_why_no_deleg = WND4_RESOURCE;
4465                 switch (open->op_deleg_want) {
4466                 case NFS4_SHARE_WANT_READ_DELEG:
4467                 case NFS4_SHARE_WANT_WRITE_DELEG:
4468                 case NFS4_SHARE_WANT_ANY_DELEG:
4469                         break;
4470                 case NFS4_SHARE_WANT_CANCEL:
4471                         open->op_why_no_deleg = WND4_CANCELLED;
4472                         break;
4473                 case NFS4_SHARE_WANT_NO_DELEG:
4474                         WARN_ON_ONCE(1);
4475                 }
4476         }
4477 }
4478
4479 /*
4480  * Attempt to hand out a delegation.
4481  *
4482  * Note we don't support write delegations, and won't until the vfs has
4483  * proper support for them.
4484  */
4485 static void
4486 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4487                         struct nfs4_ol_stateid *stp)
4488 {
4489         struct nfs4_delegation *dp;
4490         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4491         struct nfs4_client *clp = stp->st_stid.sc_client;
4492         int cb_up;
4493         int status = 0;
4494
4495         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4496         open->op_recall = 0;
4497         switch (open->op_claim_type) {
4498                 case NFS4_OPEN_CLAIM_PREVIOUS:
4499                         if (!cb_up)
4500                                 open->op_recall = 1;
4501                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4502                                 goto out_no_deleg;
4503                         break;
4504                 case NFS4_OPEN_CLAIM_NULL:
4505                 case NFS4_OPEN_CLAIM_FH:
4506                         /*
4507                          * Let's not give out any delegations till everyone's
4508                          * had the chance to reclaim theirs, *and* until
4509                          * NLM locks have all been reclaimed:
4510                          */
4511                         if (locks_in_grace(clp->net))
4512                                 goto out_no_deleg;
4513                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4514                                 goto out_no_deleg;
4515                         /*
4516                          * Also, if the file was opened for write or
4517                          * create, there's a good chance the client's
4518                          * about to write to it, resulting in an
4519                          * immediate recall (since we don't support
4520                          * write delegations):
4521                          */
4522                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4523                                 goto out_no_deleg;
4524                         if (open->op_create == NFS4_OPEN_CREATE)
4525                                 goto out_no_deleg;
4526                         break;
4527                 default:
4528                         goto out_no_deleg;
4529         }
4530         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4531         if (IS_ERR(dp))
4532                 goto out_no_deleg;
4533
4534         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4535
4536         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4537                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4538         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4539         nfs4_put_stid(&dp->dl_stid);
4540         return;
4541 out_no_deleg:
4542         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4543         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4544             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4545                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4546                 open->op_recall = 1;
4547         }
4548
4549         /* 4.1 client asking for a delegation? */
4550         if (open->op_deleg_want)
4551                 nfsd4_open_deleg_none_ext(open, status);
4552         return;
4553 }
4554
4555 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4556                                         struct nfs4_delegation *dp)
4557 {
4558         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4559             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4560                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4561                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4562         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4563                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4564                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4565                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4566         }
4567         /* Otherwise the client must be confused wanting a delegation
4568          * it already has, therefore we don't return
4569          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4570          */
4571 }
4572
4573 __be32
4574 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4575 {
4576         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4577         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4578         struct nfs4_file *fp = NULL;
4579         struct nfs4_ol_stateid *stp = NULL;
4580         struct nfs4_delegation *dp = NULL;
4581         __be32 status;
4582         bool new_stp = false;
4583
4584         /*
4585          * Lookup file; if found, lookup stateid and check open request,
4586          * and check for delegations in the process of being recalled.
4587          * If not found, create the nfs4_file struct
4588          */
4589         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4590         if (fp != open->op_file) {
4591                 status = nfs4_check_deleg(cl, open, &dp);
4592                 if (status)
4593                         goto out;
4594                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4595         } else {
4596                 open->op_file = NULL;
4597                 status = nfserr_bad_stateid;
4598                 if (nfsd4_is_deleg_cur(open))
4599                         goto out;
4600         }
4601
4602         if (!stp) {
4603                 stp = init_open_stateid(fp, open);
4604                 if (!open->op_stp)
4605                         new_stp = true;
4606         }
4607
4608         /*
4609          * OPEN the file, or upgrade an existing OPEN.
4610          * If truncate fails, the OPEN fails.
4611          *
4612          * stp is already locked.
4613          */
4614         if (!new_stp) {
4615                 /* Stateid was found, this is an OPEN upgrade */
4616                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4617                 if (status) {
4618                         mutex_unlock(&stp->st_mutex);
4619                         goto out;
4620                 }
4621         } else {
4622                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4623                 if (status) {
4624                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4625                         release_open_stateid(stp);
4626                         mutex_unlock(&stp->st_mutex);
4627                         goto out;
4628                 }
4629
4630                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4631                                                         open->op_odstate);
4632                 if (stp->st_clnt_odstate == open->op_odstate)
4633                         open->op_odstate = NULL;
4634         }
4635
4636         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4637         mutex_unlock(&stp->st_mutex);
4638
4639         if (nfsd4_has_session(&resp->cstate)) {
4640                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4641                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4642                         open->op_why_no_deleg = WND4_NOT_WANTED;
4643                         goto nodeleg;
4644                 }
4645         }
4646
4647         /*
4648         * Attempt to hand out a delegation. No error return, because the
4649         * OPEN succeeds even if we fail.
4650         */
4651         nfs4_open_delegation(current_fh, open, stp);
4652 nodeleg:
4653         status = nfs_ok;
4654
4655         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4656                 STATEID_VAL(&stp->st_stid.sc_stateid));
4657 out:
4658         /* 4.1 client trying to upgrade/downgrade delegation? */
4659         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4660             open->op_deleg_want)
4661                 nfsd4_deleg_xgrade_none_ext(open, dp);
4662
4663         if (fp)
4664                 put_nfs4_file(fp);
4665         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4666                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4667         /*
4668         * To finish the open response, we just need to set the rflags.
4669         */
4670         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4671         if (nfsd4_has_session(&resp->cstate))
4672                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4673         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4674                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4675
4676         if (dp)
4677                 nfs4_put_stid(&dp->dl_stid);
4678         if (stp)
4679                 nfs4_put_stid(&stp->st_stid);
4680
4681         return status;
4682 }
4683
4684 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4685                               struct nfsd4_open *open)
4686 {
4687         if (open->op_openowner) {
4688                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4689
4690                 nfsd4_cstate_assign_replay(cstate, so);
4691                 nfs4_put_stateowner(so);
4692         }
4693         if (open->op_file)
4694                 kmem_cache_free(file_slab, open->op_file);
4695         if (open->op_stp)
4696                 nfs4_put_stid(&open->op_stp->st_stid);
4697         if (open->op_odstate)
4698                 kmem_cache_free(odstate_slab, open->op_odstate);
4699 }
4700
4701 __be32
4702 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4703             union nfsd4_op_u *u)
4704 {
4705         clientid_t *clid = &u->renew;
4706         struct nfs4_client *clp;
4707         __be32 status;
4708         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4709
4710         dprintk("process_renew(%08x/%08x): starting\n",
4711                         clid->cl_boot, clid->cl_id);
4712         status = lookup_clientid(clid, cstate, nn);
4713         if (status)
4714                 goto out;
4715         clp = cstate->clp;
4716         status = nfserr_cb_path_down;
4717         if (!list_empty(&clp->cl_delegations)
4718                         && clp->cl_cb_state != NFSD4_CB_UP)
4719                 goto out;
4720         status = nfs_ok;
4721 out:
4722         return status;
4723 }
4724
4725 void
4726 nfsd4_end_grace(struct nfsd_net *nn)
4727 {
4728         /* do nothing if grace period already ended */
4729         if (nn->grace_ended)
4730                 return;
4731
4732         nn->grace_ended = true;
4733         /*
4734          * If the server goes down again right now, an NFSv4
4735          * client will still be allowed to reclaim after it comes back up,
4736          * even if it hasn't yet had a chance to reclaim state this time.
4737          *
4738          */
4739         nfsd4_record_grace_done(nn);
4740         /*
4741          * At this point, NFSv4 clients can still reclaim.  But if the
4742          * server crashes, any that have not yet reclaimed will be out
4743          * of luck on the next boot.
4744          *
4745          * (NFSv4.1+ clients are considered to have reclaimed once they
4746          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4747          * have reclaimed after their first OPEN.)
4748          */
4749         locks_end_grace(&nn->nfsd4_manager);
4750         /*
4751          * At this point, and once lockd and/or any other containers
4752          * exit their grace period, further reclaims will fail and
4753          * regular locking can resume.
4754          */
4755 }
4756
4757 /*
4758  * If we've waited a lease period but there are still clients trying to
4759  * reclaim, wait a little longer to give them a chance to finish.
4760  */
4761 static bool clients_still_reclaiming(struct nfsd_net *nn)
4762 {
4763         unsigned long now = get_seconds();
4764         unsigned long double_grace_period_end = nn->boot_time +
4765                                                 2 * nn->nfsd4_lease;
4766
4767         if (nn->track_reclaim_completes &&
4768                         atomic_read(&nn->nr_reclaim_complete) ==
4769                         nn->reclaim_str_hashtbl_size)
4770                 return false;
4771         if (!nn->somebody_reclaimed)
4772                 return false;
4773         nn->somebody_reclaimed = false;
4774         /*
4775          * If we've given them *two* lease times to reclaim, and they're
4776          * still not done, give up:
4777          */
4778         if (time_after(now, double_grace_period_end))
4779                 return false;
4780         return true;
4781 }
4782
4783 static time_t
4784 nfs4_laundromat(struct nfsd_net *nn)
4785 {
4786         struct nfs4_client *clp;
4787         struct nfs4_openowner *oo;
4788         struct nfs4_delegation *dp;
4789         struct nfs4_ol_stateid *stp;
4790         struct nfsd4_blocked_lock *nbl;
4791         struct list_head *pos, *next, reaplist;
4792         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4793         time_t t, new_timeo = nn->nfsd4_lease;
4794
4795         dprintk("NFSD: laundromat service - starting\n");
4796
4797         if (clients_still_reclaiming(nn)) {
4798                 new_timeo = 0;
4799                 goto out;
4800         }
4801         dprintk("NFSD: end of grace period\n");
4802         nfsd4_end_grace(nn);
4803         INIT_LIST_HEAD(&reaplist);
4804         spin_lock(&nn->client_lock);
4805         list_for_each_safe(pos, next, &nn->client_lru) {
4806                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4807                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4808                         t = clp->cl_time - cutoff;
4809                         new_timeo = min(new_timeo, t);
4810                         break;
4811                 }
4812                 if (mark_client_expired_locked(clp)) {
4813                         dprintk("NFSD: client in use (clientid %08x)\n",
4814                                 clp->cl_clientid.cl_id);
4815                         continue;
4816                 }
4817                 list_add(&clp->cl_lru, &reaplist);
4818         }
4819         spin_unlock(&nn->client_lock);
4820         list_for_each_safe(pos, next, &reaplist) {
4821                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4822                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4823                         clp->cl_clientid.cl_id);
4824                 list_del_init(&clp->cl_lru);
4825                 expire_client(clp);
4826         }
4827         spin_lock(&state_lock);
4828         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4829                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4830                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4831                         t = dp->dl_time - cutoff;
4832                         new_timeo = min(new_timeo, t);
4833                         break;
4834                 }
4835                 WARN_ON(!unhash_delegation_locked(dp));
4836                 list_add(&dp->dl_recall_lru, &reaplist);
4837         }
4838         spin_unlock(&state_lock);
4839         while (!list_empty(&reaplist)) {
4840                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4841                                         dl_recall_lru);
4842                 list_del_init(&dp->dl_recall_lru);
4843                 revoke_delegation(dp);
4844         }
4845
4846         spin_lock(&nn->client_lock);
4847         while (!list_empty(&nn->close_lru)) {
4848                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4849                                         oo_close_lru);
4850                 if (time_after((unsigned long)oo->oo_time,
4851                                (unsigned long)cutoff)) {
4852                         t = oo->oo_time - cutoff;
4853                         new_timeo = min(new_timeo, t);
4854                         break;
4855                 }
4856                 list_del_init(&oo->oo_close_lru);
4857                 stp = oo->oo_last_closed_stid;
4858                 oo->oo_last_closed_stid = NULL;
4859                 spin_unlock(&nn->client_lock);
4860                 nfs4_put_stid(&stp->st_stid);
4861                 spin_lock(&nn->client_lock);
4862         }
4863         spin_unlock(&nn->client_lock);
4864
4865         /*
4866          * It's possible for a client to try and acquire an already held lock
4867          * that is being held for a long time, and then lose interest in it.
4868          * So, we clean out any un-revisited request after a lease period
4869          * under the assumption that the client is no longer interested.
4870          *
4871          * RFC5661, sec. 9.6 states that the client must not rely on getting
4872          * notifications and must continue to poll for locks, even when the
4873          * server supports them. Thus this shouldn't lead to clients blocking
4874          * indefinitely once the lock does become free.
4875          */
4876         BUG_ON(!list_empty(&reaplist));
4877         spin_lock(&nn->blocked_locks_lock);
4878         while (!list_empty(&nn->blocked_locks_lru)) {
4879                 nbl = list_first_entry(&nn->blocked_locks_lru,
4880                                         struct nfsd4_blocked_lock, nbl_lru);
4881                 if (time_after((unsigned long)nbl->nbl_time,
4882                                (unsigned long)cutoff)) {
4883                         t = nbl->nbl_time - cutoff;
4884                         new_timeo = min(new_timeo, t);
4885                         break;
4886                 }
4887                 list_move(&nbl->nbl_lru, &reaplist);
4888                 list_del_init(&nbl->nbl_list);
4889         }
4890         spin_unlock(&nn->blocked_locks_lock);
4891
4892         while (!list_empty(&reaplist)) {
4893                 nbl = list_first_entry(&reaplist,
4894                                         struct nfsd4_blocked_lock, nbl_lru);
4895                 list_del_init(&nbl->nbl_lru);
4896                 free_blocked_lock(nbl);
4897         }
4898 out:
4899         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4900         return new_timeo;
4901 }
4902
4903 static struct workqueue_struct *laundry_wq;
4904 static void laundromat_main(struct work_struct *);
4905
4906 static void
4907 laundromat_main(struct work_struct *laundry)
4908 {
4909         time_t t;
4910         struct delayed_work *dwork = to_delayed_work(laundry);
4911         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4912                                            laundromat_work);
4913
4914         t = nfs4_laundromat(nn);
4915         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4916         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4917 }
4918
4919 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4920 {
4921         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4922                 return nfserr_bad_stateid;
4923         return nfs_ok;
4924 }
4925
4926 static inline int
4927 access_permit_read(struct nfs4_ol_stateid *stp)
4928 {
4929         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4930                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4931                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4932 }
4933
4934 static inline int
4935 access_permit_write(struct nfs4_ol_stateid *stp)
4936 {
4937         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4938                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4939 }
4940
4941 static
4942 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4943 {
4944         __be32 status = nfserr_openmode;
4945
4946         /* For lock stateid's, we test the parent open, not the lock: */
4947         if (stp->st_openstp)
4948                 stp = stp->st_openstp;
4949         if ((flags & WR_STATE) && !access_permit_write(stp))
4950                 goto out;
4951         if ((flags & RD_STATE) && !access_permit_read(stp))
4952                 goto out;
4953         status = nfs_ok;
4954 out:
4955         return status;
4956 }
4957
4958 static inline __be32
4959 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4960 {
4961         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4962                 return nfs_ok;
4963         else if (opens_in_grace(net)) {
4964                 /* Answer in remaining cases depends on existence of
4965                  * conflicting state; so we must wait out the grace period. */
4966                 return nfserr_grace;
4967         } else if (flags & WR_STATE)
4968                 return nfs4_share_conflict(current_fh,
4969                                 NFS4_SHARE_DENY_WRITE);
4970         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4971                 return nfs4_share_conflict(current_fh,
4972                                 NFS4_SHARE_DENY_READ);
4973 }
4974
4975 /*
4976  * Allow READ/WRITE during grace period on recovered state only for files
4977  * that are not able to provide mandatory locking.
4978  */
4979 static inline int
4980 grace_disallows_io(struct net *net, struct inode *inode)
4981 {
4982         return opens_in_grace(net) && mandatory_lock(inode);
4983 }
4984
4985 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4986 {
4987         /*
4988          * When sessions are used the stateid generation number is ignored
4989          * when it is zero.
4990          */
4991         if (has_session && in->si_generation == 0)
4992                 return nfs_ok;
4993
4994         if (in->si_generation == ref->si_generation)
4995                 return nfs_ok;
4996
4997         /* If the client sends us a stateid from the future, it's buggy: */
4998         if (nfsd4_stateid_generation_after(in, ref))
4999                 return nfserr_bad_stateid;
5000         /*
5001          * However, we could see a stateid from the past, even from a
5002          * non-buggy client.  For example, if the client sends a lock
5003          * while some IO is outstanding, the lock may bump si_generation
5004          * while the IO is still in flight.  The client could avoid that
5005          * situation by waiting for responses on all the IO requests,
5006          * but better performance may result in retrying IO that
5007          * receives an old_stateid error if requests are rarely
5008          * reordered in flight:
5009          */
5010         return nfserr_old_stateid;
5011 }
5012
5013 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5014 {
5015         __be32 ret;
5016
5017         spin_lock(&s->sc_lock);
5018         ret = nfsd4_verify_open_stid(s);
5019         if (ret == nfs_ok)
5020                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5021         spin_unlock(&s->sc_lock);
5022         return ret;
5023 }
5024
5025 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5026 {
5027         if (ols->st_stateowner->so_is_open_owner &&
5028             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5029                 return nfserr_bad_stateid;
5030         return nfs_ok;
5031 }
5032
5033 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5034 {
5035         struct nfs4_stid *s;
5036         __be32 status = nfserr_bad_stateid;
5037
5038         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5039                 CLOSE_STATEID(stateid))
5040                 return status;
5041         /* Client debugging aid. */
5042         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5043                 char addr_str[INET6_ADDRSTRLEN];
5044                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5045                                  sizeof(addr_str));
5046                 pr_warn_ratelimited("NFSD: client %s testing state ID "
5047                                         "with incorrect client ID\n", addr_str);
5048                 return status;
5049         }
5050         spin_lock(&cl->cl_lock);
5051         s = find_stateid_locked(cl, stateid);
5052         if (!s)
5053                 goto out_unlock;
5054         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5055         if (status)
5056                 goto out_unlock;
5057         switch (s->sc_type) {
5058         case NFS4_DELEG_STID:
5059                 status = nfs_ok;
5060                 break;
5061         case NFS4_REVOKED_DELEG_STID:
5062                 status = nfserr_deleg_revoked;
5063                 break;
5064         case NFS4_OPEN_STID:
5065         case NFS4_LOCK_STID:
5066                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5067                 break;
5068         default:
5069                 printk("unknown stateid type %x\n", s->sc_type);
5070                 /* Fallthrough */
5071         case NFS4_CLOSED_STID:
5072         case NFS4_CLOSED_DELEG_STID:
5073                 status = nfserr_bad_stateid;
5074         }
5075 out_unlock:
5076         spin_unlock(&cl->cl_lock);
5077         return status;
5078 }
5079
5080 __be32
5081 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5082                      stateid_t *stateid, unsigned char typemask,
5083                      struct nfs4_stid **s, struct nfsd_net *nn)
5084 {
5085         __be32 status;
5086         bool return_revoked = false;
5087
5088         /*
5089          *  only return revoked delegations if explicitly asked.
5090          *  otherwise we report revoked or bad_stateid status.
5091          */
5092         if (typemask & NFS4_REVOKED_DELEG_STID)
5093                 return_revoked = true;
5094         else if (typemask & NFS4_DELEG_STID)
5095                 typemask |= NFS4_REVOKED_DELEG_STID;
5096
5097         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5098                 CLOSE_STATEID(stateid))
5099                 return nfserr_bad_stateid;
5100         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5101         if (status == nfserr_stale_clientid) {
5102                 if (cstate->session)
5103                         return nfserr_bad_stateid;
5104                 return nfserr_stale_stateid;
5105         }
5106         if (status)
5107                 return status;
5108         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5109         if (!*s)
5110                 return nfserr_bad_stateid;
5111         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5112                 nfs4_put_stid(*s);
5113                 if (cstate->minorversion)
5114                         return nfserr_deleg_revoked;
5115                 return nfserr_bad_stateid;
5116         }
5117         return nfs_ok;
5118 }
5119
5120 static struct file *
5121 nfs4_find_file(struct nfs4_stid *s, int flags)
5122 {
5123         if (!s)
5124                 return NULL;
5125
5126         switch (s->sc_type) {
5127         case NFS4_DELEG_STID:
5128                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5129                         return NULL;
5130                 return get_file(s->sc_file->fi_deleg_file);
5131         case NFS4_OPEN_STID:
5132         case NFS4_LOCK_STID:
5133                 if (flags & RD_STATE)
5134                         return find_readable_file(s->sc_file);
5135                 else
5136                         return find_writeable_file(s->sc_file);
5137                 break;
5138         }
5139
5140         return NULL;
5141 }
5142
5143 static __be32
5144 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5145 {
5146         __be32 status;
5147
5148         status = nfsd4_check_openowner_confirmed(ols);
5149         if (status)
5150                 return status;
5151         return nfs4_check_openmode(ols, flags);
5152 }
5153
5154 static __be32
5155 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5156                 struct file **filpp, bool *tmp_file, int flags)
5157 {
5158         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5159         struct file *file;
5160         __be32 status;
5161
5162         file = nfs4_find_file(s, flags);
5163         if (file) {
5164                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5165                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5166                 if (status) {
5167                         fput(file);
5168                         return status;
5169                 }
5170
5171                 *filpp = file;
5172         } else {
5173                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5174                 if (status)
5175                         return status;
5176
5177                 if (tmp_file)
5178                         *tmp_file = true;
5179         }
5180
5181         return 0;
5182 }
5183
5184 /*
5185  * Checks for stateid operations
5186  */
5187 __be32
5188 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5189                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5190                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5191 {
5192         struct inode *ino = d_inode(fhp->fh_dentry);
5193         struct net *net = SVC_NET(rqstp);
5194         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5195         struct nfs4_stid *s = NULL;
5196         __be32 status;
5197
5198         if (filpp)
5199                 *filpp = NULL;
5200         if (tmp_file)
5201                 *tmp_file = false;
5202
5203         if (grace_disallows_io(net, ino))
5204                 return nfserr_grace;
5205
5206         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5207                 status = check_special_stateids(net, fhp, stateid, flags);
5208                 goto done;
5209         }
5210
5211         status = nfsd4_lookup_stateid(cstate, stateid,
5212                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5213                                 &s, nn);
5214         if (status)
5215                 return status;
5216         status = nfsd4_stid_check_stateid_generation(stateid, s,
5217                         nfsd4_has_session(cstate));
5218         if (status)
5219                 goto out;
5220
5221         switch (s->sc_type) {
5222         case NFS4_DELEG_STID:
5223                 status = nfs4_check_delegmode(delegstateid(s), flags);
5224                 break;
5225         case NFS4_OPEN_STID:
5226         case NFS4_LOCK_STID:
5227                 status = nfs4_check_olstateid(openlockstateid(s), flags);
5228                 break;
5229         default:
5230                 status = nfserr_bad_stateid;
5231                 break;
5232         }
5233         if (status)
5234                 goto out;
5235         status = nfs4_check_fh(fhp, s);
5236
5237 done:
5238         if (!status && filpp)
5239                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5240 out:
5241         if (s)
5242                 nfs4_put_stid(s);
5243         return status;
5244 }
5245
5246 /*
5247  * Test if the stateid is valid
5248  */
5249 __be32
5250 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5251                    union nfsd4_op_u *u)
5252 {
5253         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5254         struct nfsd4_test_stateid_id *stateid;
5255         struct nfs4_client *cl = cstate->session->se_client;
5256
5257         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5258                 stateid->ts_id_status =
5259                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5260
5261         return nfs_ok;
5262 }
5263
5264 static __be32
5265 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5266 {
5267         struct nfs4_ol_stateid *stp = openlockstateid(s);
5268         __be32 ret;
5269
5270         ret = nfsd4_lock_ol_stateid(stp);
5271         if (ret)
5272                 goto out_put_stid;
5273
5274         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5275         if (ret)
5276                 goto out;
5277
5278         ret = nfserr_locks_held;
5279         if (check_for_locks(stp->st_stid.sc_file,
5280                             lockowner(stp->st_stateowner)))
5281                 goto out;
5282
5283         release_lock_stateid(stp);
5284         ret = nfs_ok;
5285
5286 out:
5287         mutex_unlock(&stp->st_mutex);
5288 out_put_stid:
5289         nfs4_put_stid(s);
5290         return ret;
5291 }
5292
5293 __be32
5294 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5295                    union nfsd4_op_u *u)
5296 {
5297         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5298         stateid_t *stateid = &free_stateid->fr_stateid;
5299         struct nfs4_stid *s;
5300         struct nfs4_delegation *dp;
5301         struct nfs4_client *cl = cstate->session->se_client;
5302         __be32 ret = nfserr_bad_stateid;
5303
5304         spin_lock(&cl->cl_lock);
5305         s = find_stateid_locked(cl, stateid);
5306         if (!s)
5307                 goto out_unlock;
5308         spin_lock(&s->sc_lock);
5309         switch (s->sc_type) {
5310         case NFS4_DELEG_STID:
5311                 ret = nfserr_locks_held;
5312                 break;
5313         case NFS4_OPEN_STID:
5314                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5315                 if (ret)
5316                         break;
5317                 ret = nfserr_locks_held;
5318                 break;
5319         case NFS4_LOCK_STID:
5320                 spin_unlock(&s->sc_lock);
5321                 refcount_inc(&s->sc_count);
5322                 spin_unlock(&cl->cl_lock);
5323                 ret = nfsd4_free_lock_stateid(stateid, s);
5324                 goto out;
5325         case NFS4_REVOKED_DELEG_STID:
5326                 spin_unlock(&s->sc_lock);
5327                 dp = delegstateid(s);
5328                 list_del_init(&dp->dl_recall_lru);
5329                 spin_unlock(&cl->cl_lock);
5330                 nfs4_put_stid(s);
5331                 ret = nfs_ok;
5332                 goto out;
5333         /* Default falls through and returns nfserr_bad_stateid */
5334         }
5335         spin_unlock(&s->sc_lock);
5336 out_unlock:
5337         spin_unlock(&cl->cl_lock);
5338 out:
5339         return ret;
5340 }
5341
5342 static inline int
5343 setlkflg (int type)
5344 {
5345         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5346                 RD_STATE : WR_STATE;
5347 }
5348
5349 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5350 {
5351         struct svc_fh *current_fh = &cstate->current_fh;
5352         struct nfs4_stateowner *sop = stp->st_stateowner;
5353         __be32 status;
5354
5355         status = nfsd4_check_seqid(cstate, sop, seqid);
5356         if (status)
5357                 return status;
5358         status = nfsd4_lock_ol_stateid(stp);
5359         if (status != nfs_ok)
5360                 return status;
5361         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5362         if (status == nfs_ok)
5363                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5364         if (status != nfs_ok)
5365                 mutex_unlock(&stp->st_mutex);
5366         return status;
5367 }
5368
5369 /*
5370  * Checks for sequence id mutating operations.
5371  */
5372 static __be32
5373 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5374                          stateid_t *stateid, char typemask,
5375                          struct nfs4_ol_stateid **stpp,
5376                          struct nfsd_net *nn)
5377 {
5378         __be32 status;
5379         struct nfs4_stid *s;
5380         struct nfs4_ol_stateid *stp = NULL;
5381
5382         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5383                 seqid, STATEID_VAL(stateid));
5384
5385         *stpp = NULL;
5386         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5387         if (status)
5388                 return status;
5389         stp = openlockstateid(s);
5390         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5391
5392         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5393         if (!status)
5394                 *stpp = stp;
5395         else
5396                 nfs4_put_stid(&stp->st_stid);
5397         return status;
5398 }
5399
5400 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5401                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5402 {
5403         __be32 status;
5404         struct nfs4_openowner *oo;
5405         struct nfs4_ol_stateid *stp;
5406
5407         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5408                                                 NFS4_OPEN_STID, &stp, nn);
5409         if (status)
5410                 return status;
5411         oo = openowner(stp->st_stateowner);
5412         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5413                 mutex_unlock(&stp->st_mutex);
5414                 nfs4_put_stid(&stp->st_stid);
5415                 return nfserr_bad_stateid;
5416         }
5417         *stpp = stp;
5418         return nfs_ok;
5419 }
5420
5421 __be32
5422 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5423                    union nfsd4_op_u *u)
5424 {
5425         struct nfsd4_open_confirm *oc = &u->open_confirm;
5426         __be32 status;
5427         struct nfs4_openowner *oo;
5428         struct nfs4_ol_stateid *stp;
5429         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5430
5431         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5432                         cstate->current_fh.fh_dentry);
5433
5434         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5435         if (status)
5436                 return status;
5437
5438         status = nfs4_preprocess_seqid_op(cstate,
5439                                         oc->oc_seqid, &oc->oc_req_stateid,
5440                                         NFS4_OPEN_STID, &stp, nn);
5441         if (status)
5442                 goto out;
5443         oo = openowner(stp->st_stateowner);
5444         status = nfserr_bad_stateid;
5445         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5446                 mutex_unlock(&stp->st_mutex);
5447                 goto put_stateid;
5448         }
5449         oo->oo_flags |= NFS4_OO_CONFIRMED;
5450         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5451         mutex_unlock(&stp->st_mutex);
5452         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5453                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5454
5455         nfsd4_client_record_create(oo->oo_owner.so_client);
5456         status = nfs_ok;
5457 put_stateid:
5458         nfs4_put_stid(&stp->st_stid);
5459 out:
5460         nfsd4_bump_seqid(cstate, status);
5461         return status;
5462 }
5463
5464 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5465 {
5466         if (!test_access(access, stp))
5467                 return;
5468         nfs4_file_put_access(stp->st_stid.sc_file, access);
5469         clear_access(access, stp);
5470 }
5471
5472 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5473 {
5474         switch (to_access) {
5475         case NFS4_SHARE_ACCESS_READ:
5476                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5477                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5478                 break;
5479         case NFS4_SHARE_ACCESS_WRITE:
5480                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5481                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5482                 break;
5483         case NFS4_SHARE_ACCESS_BOTH:
5484                 break;
5485         default:
5486                 WARN_ON_ONCE(1);
5487         }
5488 }
5489
5490 __be32
5491 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5492                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5493 {
5494         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5495         __be32 status;
5496         struct nfs4_ol_stateid *stp;
5497         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5498
5499         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5500                         cstate->current_fh.fh_dentry);
5501
5502         /* We don't yet support WANT bits: */
5503         if (od->od_deleg_want)
5504                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5505                         od->od_deleg_want);
5506
5507         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5508                                         &od->od_stateid, &stp, nn);
5509         if (status)
5510                 goto out;
5511         status = nfserr_inval;
5512         if (!test_access(od->od_share_access, stp)) {
5513                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5514                         stp->st_access_bmap, od->od_share_access);
5515                 goto put_stateid;
5516         }
5517         if (!test_deny(od->od_share_deny, stp)) {
5518                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5519                         stp->st_deny_bmap, od->od_share_deny);
5520                 goto put_stateid;
5521         }
5522         nfs4_stateid_downgrade(stp, od->od_share_access);
5523         reset_union_bmap_deny(od->od_share_deny, stp);
5524         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5525         status = nfs_ok;
5526 put_stateid:
5527         mutex_unlock(&stp->st_mutex);
5528         nfs4_put_stid(&stp->st_stid);
5529 out:
5530         nfsd4_bump_seqid(cstate, status);
5531         return status;
5532 }
5533
5534 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5535 {
5536         struct nfs4_client *clp = s->st_stid.sc_client;
5537         bool unhashed;
5538         LIST_HEAD(reaplist);
5539
5540         spin_lock(&clp->cl_lock);
5541         unhashed = unhash_open_stateid(s, &reaplist);
5542
5543         if (clp->cl_minorversion) {
5544                 if (unhashed)
5545                         put_ol_stateid_locked(s, &reaplist);
5546                 spin_unlock(&clp->cl_lock);
5547                 free_ol_stateid_reaplist(&reaplist);
5548         } else {
5549                 spin_unlock(&clp->cl_lock);
5550                 free_ol_stateid_reaplist(&reaplist);
5551                 if (unhashed)
5552                         move_to_close_lru(s, clp->net);
5553         }
5554 }
5555
5556 /*
5557  * nfs4_unlock_state() called after encode
5558  */
5559 __be32
5560 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5561                 union nfsd4_op_u *u)
5562 {
5563         struct nfsd4_close *close = &u->close;
5564         __be32 status;
5565         struct nfs4_ol_stateid *stp;
5566         struct net *net = SVC_NET(rqstp);
5567         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5568
5569         dprintk("NFSD: nfsd4_close on file %pd\n",
5570                         cstate->current_fh.fh_dentry);
5571
5572         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5573                                         &close->cl_stateid,
5574                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5575                                         &stp, nn);
5576         nfsd4_bump_seqid(cstate, status);
5577         if (status)
5578                 goto out;
5579
5580         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5581
5582         /*
5583          * Technically we don't _really_ have to increment or copy it, since
5584          * it should just be gone after this operation and we clobber the
5585          * copied value below, but we continue to do so here just to ensure
5586          * that racing ops see that there was a state change.
5587          */
5588         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5589
5590         nfsd4_close_open_stateid(stp);
5591         mutex_unlock(&stp->st_mutex);
5592
5593         /* v4.1+ suggests that we send a special stateid in here, since the
5594          * clients should just ignore this anyway. Since this is not useful
5595          * for v4.0 clients either, we set it to the special close_stateid
5596          * universally.
5597          *
5598          * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5599          */
5600         memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5601
5602         /* put reference from nfs4_preprocess_seqid_op */
5603         nfs4_put_stid(&stp->st_stid);
5604 out:
5605         return status;
5606 }
5607
5608 __be32
5609 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5610                   union nfsd4_op_u *u)
5611 {
5612         struct nfsd4_delegreturn *dr = &u->delegreturn;
5613         struct nfs4_delegation *dp;
5614         stateid_t *stateid = &dr->dr_stateid;
5615         struct nfs4_stid *s;
5616         __be32 status;
5617         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5618
5619         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5620                 return status;
5621
5622         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5623         if (status)
5624                 goto out;
5625         dp = delegstateid(s);
5626         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5627         if (status)
5628                 goto put_stateid;
5629
5630         destroy_delegation(dp);
5631 put_stateid:
5632         nfs4_put_stid(&dp->dl_stid);
5633 out:
5634         return status;
5635 }
5636
5637 static inline u64
5638 end_offset(u64 start, u64 len)
5639 {
5640         u64 end;
5641
5642         end = start + len;
5643         return end >= start ? end: NFS4_MAX_UINT64;
5644 }
5645
5646 /* last octet in a range */
5647 static inline u64
5648 last_byte_offset(u64 start, u64 len)
5649 {
5650         u64 end;
5651
5652         WARN_ON_ONCE(!len);
5653         end = start + len;
5654         return end > start ? end - 1: NFS4_MAX_UINT64;
5655 }
5656
5657 /*
5658  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5659  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5660  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5661  * locking, this prevents us from being completely protocol-compliant.  The
5662  * real solution to this problem is to start using unsigned file offsets in
5663  * the VFS, but this is a very deep change!
5664  */
5665 static inline void
5666 nfs4_transform_lock_offset(struct file_lock *lock)
5667 {
5668         if (lock->fl_start < 0)
5669                 lock->fl_start = OFFSET_MAX;
5670         if (lock->fl_end < 0)
5671                 lock->fl_end = OFFSET_MAX;
5672 }
5673
5674 static fl_owner_t
5675 nfsd4_fl_get_owner(fl_owner_t owner)
5676 {
5677         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5678
5679         nfs4_get_stateowner(&lo->lo_owner);
5680         return owner;
5681 }
5682
5683 static void
5684 nfsd4_fl_put_owner(fl_owner_t owner)
5685 {
5686         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5687
5688         if (lo)
5689                 nfs4_put_stateowner(&lo->lo_owner);
5690 }
5691
5692 static void
5693 nfsd4_lm_notify(struct file_lock *fl)
5694 {
5695         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5696         struct net                      *net = lo->lo_owner.so_client->net;
5697         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5698         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5699                                                 struct nfsd4_blocked_lock, nbl_lock);
5700         bool queue = false;
5701
5702         /* An empty list means that something else is going to be using it */
5703         spin_lock(&nn->blocked_locks_lock);
5704         if (!list_empty(&nbl->nbl_list)) {
5705                 list_del_init(&nbl->nbl_list);
5706                 list_del_init(&nbl->nbl_lru);
5707                 queue = true;
5708         }
5709         spin_unlock(&nn->blocked_locks_lock);
5710
5711         if (queue)
5712                 nfsd4_run_cb(&nbl->nbl_cb);
5713 }
5714
5715 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5716         .lm_notify = nfsd4_lm_notify,
5717         .lm_get_owner = nfsd4_fl_get_owner,
5718         .lm_put_owner = nfsd4_fl_put_owner,
5719 };
5720
5721 static inline void
5722 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5723 {
5724         struct nfs4_lockowner *lo;
5725
5726         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5727                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5728                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5729                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5730                 if (!deny->ld_owner.data)
5731                         /* We just don't care that much */
5732                         goto nevermind;
5733                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5734                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5735         } else {
5736 nevermind:
5737                 deny->ld_owner.len = 0;
5738                 deny->ld_owner.data = NULL;
5739                 deny->ld_clientid.cl_boot = 0;
5740                 deny->ld_clientid.cl_id = 0;
5741         }
5742         deny->ld_start = fl->fl_start;
5743         deny->ld_length = NFS4_MAX_UINT64;
5744         if (fl->fl_end != NFS4_MAX_UINT64)
5745                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5746         deny->ld_type = NFS4_READ_LT;
5747         if (fl->fl_type != F_RDLCK)
5748                 deny->ld_type = NFS4_WRITE_LT;
5749 }
5750
5751 static struct nfs4_lockowner *
5752 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5753 {
5754         unsigned int strhashval = ownerstr_hashval(owner);
5755         struct nfs4_stateowner *so;
5756
5757         lockdep_assert_held(&clp->cl_lock);
5758
5759         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5760                             so_strhash) {
5761                 if (so->so_is_open_owner)
5762                         continue;
5763                 if (same_owner_str(so, owner))
5764                         return lockowner(nfs4_get_stateowner(so));
5765         }
5766         return NULL;
5767 }
5768
5769 static struct nfs4_lockowner *
5770 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5771 {
5772         struct nfs4_lockowner *lo;
5773
5774         spin_lock(&clp->cl_lock);
5775         lo = find_lockowner_str_locked(clp, owner);
5776         spin_unlock(&clp->cl_lock);
5777         return lo;
5778 }
5779
5780 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5781 {
5782         unhash_lockowner_locked(lockowner(sop));
5783 }
5784
5785 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5786 {
5787         struct nfs4_lockowner *lo = lockowner(sop);
5788
5789         kmem_cache_free(lockowner_slab, lo);
5790 }
5791
5792 static const struct nfs4_stateowner_operations lockowner_ops = {
5793         .so_unhash =    nfs4_unhash_lockowner,
5794         .so_free =      nfs4_free_lockowner,
5795 };
5796
5797 /*
5798  * Alloc a lock owner structure.
5799  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5800  * occurred.
5801  *
5802  * strhashval = ownerstr_hashval
5803  */
5804 static struct nfs4_lockowner *
5805 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5806                            struct nfs4_ol_stateid *open_stp,
5807                            struct nfsd4_lock *lock)
5808 {
5809         struct nfs4_lockowner *lo, *ret;
5810
5811         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5812         if (!lo)
5813                 return NULL;
5814         INIT_LIST_HEAD(&lo->lo_blocked);
5815         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5816         lo->lo_owner.so_is_open_owner = 0;
5817         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5818         lo->lo_owner.so_ops = &lockowner_ops;
5819         spin_lock(&clp->cl_lock);
5820         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5821         if (ret == NULL) {
5822                 list_add(&lo->lo_owner.so_strhash,
5823                          &clp->cl_ownerstr_hashtbl[strhashval]);
5824                 ret = lo;
5825         } else
5826                 nfs4_free_stateowner(&lo->lo_owner);
5827
5828         spin_unlock(&clp->cl_lock);
5829         return ret;
5830 }
5831
5832 static struct nfs4_ol_stateid *
5833 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5834 {
5835         struct nfs4_ol_stateid *lst;
5836         struct nfs4_client *clp = lo->lo_owner.so_client;
5837
5838         lockdep_assert_held(&clp->cl_lock);
5839
5840         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5841                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5842                         continue;
5843                 if (lst->st_stid.sc_file == fp) {
5844                         refcount_inc(&lst->st_stid.sc_count);
5845                         return lst;
5846                 }
5847         }
5848         return NULL;
5849 }
5850
5851 static struct nfs4_ol_stateid *
5852 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5853                   struct nfs4_file *fp, struct inode *inode,
5854                   struct nfs4_ol_stateid *open_stp)
5855 {
5856         struct nfs4_client *clp = lo->lo_owner.so_client;
5857         struct nfs4_ol_stateid *retstp;
5858
5859         mutex_init(&stp->st_mutex);
5860         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5861 retry:
5862         spin_lock(&clp->cl_lock);
5863         spin_lock(&fp->fi_lock);
5864         retstp = find_lock_stateid(lo, fp);
5865         if (retstp)
5866                 goto out_unlock;
5867
5868         refcount_inc(&stp->st_stid.sc_count);
5869         stp->st_stid.sc_type = NFS4_LOCK_STID;
5870         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5871         get_nfs4_file(fp);
5872         stp->st_stid.sc_file = fp;
5873         stp->st_access_bmap = 0;
5874         stp->st_deny_bmap = open_stp->st_deny_bmap;
5875         stp->st_openstp = open_stp;
5876         list_add(&stp->st_locks, &open_stp->st_locks);
5877         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5878         list_add(&stp->st_perfile, &fp->fi_stateids);
5879 out_unlock:
5880         spin_unlock(&fp->fi_lock);
5881         spin_unlock(&clp->cl_lock);
5882         if (retstp) {
5883                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5884                         nfs4_put_stid(&retstp->st_stid);
5885                         goto retry;
5886                 }
5887                 /* To keep mutex tracking happy */
5888                 mutex_unlock(&stp->st_mutex);
5889                 stp = retstp;
5890         }
5891         return stp;
5892 }
5893
5894 static struct nfs4_ol_stateid *
5895 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5896                             struct inode *inode, struct nfs4_ol_stateid *ost,
5897                             bool *new)
5898 {
5899         struct nfs4_stid *ns = NULL;
5900         struct nfs4_ol_stateid *lst;
5901         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5902         struct nfs4_client *clp = oo->oo_owner.so_client;
5903
5904         *new = false;
5905         spin_lock(&clp->cl_lock);
5906         lst = find_lock_stateid(lo, fi);
5907         spin_unlock(&clp->cl_lock);
5908         if (lst != NULL) {
5909                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5910                         goto out;
5911                 nfs4_put_stid(&lst->st_stid);
5912         }
5913         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5914         if (ns == NULL)
5915                 return NULL;
5916
5917         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5918         if (lst == openlockstateid(ns))
5919                 *new = true;
5920         else
5921                 nfs4_put_stid(ns);
5922 out:
5923         return lst;
5924 }
5925
5926 static int
5927 check_lock_length(u64 offset, u64 length)
5928 {
5929         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5930                 (length > ~offset)));
5931 }
5932
5933 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5934 {
5935         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5936
5937         lockdep_assert_held(&fp->fi_lock);
5938
5939         if (test_access(access, lock_stp))
5940                 return;
5941         __nfs4_file_get_access(fp, access);
5942         set_access(access, lock_stp);
5943 }
5944
5945 static __be32
5946 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5947                             struct nfs4_ol_stateid *ost,
5948                             struct nfsd4_lock *lock,
5949                             struct nfs4_ol_stateid **plst, bool *new)
5950 {
5951         __be32 status;
5952         struct nfs4_file *fi = ost->st_stid.sc_file;
5953         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5954         struct nfs4_client *cl = oo->oo_owner.so_client;
5955         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5956         struct nfs4_lockowner *lo;
5957         struct nfs4_ol_stateid *lst;
5958         unsigned int strhashval;
5959
5960         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5961         if (!lo) {
5962                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5963                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5964                 if (lo == NULL)
5965                         return nfserr_jukebox;
5966         } else {
5967                 /* with an existing lockowner, seqids must be the same */
5968                 status = nfserr_bad_seqid;
5969                 if (!cstate->minorversion &&
5970                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5971                         goto out;
5972         }
5973
5974         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5975         if (lst == NULL) {
5976                 status = nfserr_jukebox;
5977                 goto out;
5978         }
5979
5980         status = nfs_ok;
5981         *plst = lst;
5982 out:
5983         nfs4_put_stateowner(&lo->lo_owner);
5984         return status;
5985 }
5986
5987 /*
5988  *  LOCK operation
5989  */
5990 __be32
5991 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5992            union nfsd4_op_u *u)
5993 {
5994         struct nfsd4_lock *lock = &u->lock;
5995         struct nfs4_openowner *open_sop = NULL;
5996         struct nfs4_lockowner *lock_sop = NULL;
5997         struct nfs4_ol_stateid *lock_stp = NULL;
5998         struct nfs4_ol_stateid *open_stp = NULL;
5999         struct nfs4_file *fp;
6000         struct file *filp = NULL;
6001         struct nfsd4_blocked_lock *nbl = NULL;
6002         struct file_lock *file_lock = NULL;
6003         struct file_lock *conflock = NULL;
6004         __be32 status = 0;
6005         int lkflg;
6006         int err;
6007         bool new = false;
6008         unsigned char fl_type;
6009         unsigned int fl_flags = FL_POSIX;
6010         struct net *net = SVC_NET(rqstp);
6011         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6012
6013         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6014                 (long long) lock->lk_offset,
6015                 (long long) lock->lk_length);
6016
6017         if (check_lock_length(lock->lk_offset, lock->lk_length))
6018                  return nfserr_inval;
6019
6020         if ((status = fh_verify(rqstp, &cstate->current_fh,
6021                                 S_IFREG, NFSD_MAY_LOCK))) {
6022                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6023                 return status;
6024         }
6025
6026         if (lock->lk_is_new) {
6027                 if (nfsd4_has_session(cstate))
6028                         /* See rfc 5661 18.10.3: given clientid is ignored: */
6029                         memcpy(&lock->lk_new_clientid,
6030                                 &cstate->session->se_client->cl_clientid,
6031                                 sizeof(clientid_t));
6032
6033                 status = nfserr_stale_clientid;
6034                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6035                         goto out;
6036
6037                 /* validate and update open stateid and open seqid */
6038                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6039                                         lock->lk_new_open_seqid,
6040                                         &lock->lk_new_open_stateid,
6041                                         &open_stp, nn);
6042                 if (status)
6043                         goto out;
6044                 mutex_unlock(&open_stp->st_mutex);
6045                 open_sop = openowner(open_stp->st_stateowner);
6046                 status = nfserr_bad_stateid;
6047                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6048                                                 &lock->lk_new_clientid))
6049                         goto out;
6050                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6051                                                         &lock_stp, &new);
6052         } else {
6053                 status = nfs4_preprocess_seqid_op(cstate,
6054                                        lock->lk_old_lock_seqid,
6055                                        &lock->lk_old_lock_stateid,
6056                                        NFS4_LOCK_STID, &lock_stp, nn);
6057         }
6058         if (status)
6059                 goto out;
6060         lock_sop = lockowner(lock_stp->st_stateowner);
6061
6062         lkflg = setlkflg(lock->lk_type);
6063         status = nfs4_check_openmode(lock_stp, lkflg);
6064         if (status)
6065                 goto out;
6066
6067         status = nfserr_grace;
6068         if (locks_in_grace(net) && !lock->lk_reclaim)
6069                 goto out;
6070         status = nfserr_no_grace;
6071         if (!locks_in_grace(net) && lock->lk_reclaim)
6072                 goto out;
6073
6074         fp = lock_stp->st_stid.sc_file;
6075         switch (lock->lk_type) {
6076                 case NFS4_READW_LT:
6077                         if (nfsd4_has_session(cstate))
6078                                 fl_flags |= FL_SLEEP;
6079                         /* Fallthrough */
6080                 case NFS4_READ_LT:
6081                         spin_lock(&fp->fi_lock);
6082                         filp = find_readable_file_locked(fp);
6083                         if (filp)
6084                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6085                         spin_unlock(&fp->fi_lock);
6086                         fl_type = F_RDLCK;
6087                         break;
6088                 case NFS4_WRITEW_LT:
6089                         if (nfsd4_has_session(cstate))
6090                                 fl_flags |= FL_SLEEP;
6091                         /* Fallthrough */
6092                 case NFS4_WRITE_LT:
6093                         spin_lock(&fp->fi_lock);
6094                         filp = find_writeable_file_locked(fp);
6095                         if (filp)
6096                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6097                         spin_unlock(&fp->fi_lock);
6098                         fl_type = F_WRLCK;
6099                         break;
6100                 default:
6101                         status = nfserr_inval;
6102                 goto out;
6103         }
6104
6105         if (!filp) {
6106                 status = nfserr_openmode;
6107                 goto out;
6108         }
6109
6110         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6111         if (!nbl) {
6112                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6113                 status = nfserr_jukebox;
6114                 goto out;
6115         }
6116
6117         file_lock = &nbl->nbl_lock;
6118         file_lock->fl_type = fl_type;
6119         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6120         file_lock->fl_pid = current->tgid;
6121         file_lock->fl_file = filp;
6122         file_lock->fl_flags = fl_flags;
6123         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6124         file_lock->fl_start = lock->lk_offset;
6125         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6126         nfs4_transform_lock_offset(file_lock);
6127
6128         conflock = locks_alloc_lock();
6129         if (!conflock) {
6130                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6131                 status = nfserr_jukebox;
6132                 goto out;
6133         }
6134
6135         if (fl_flags & FL_SLEEP) {
6136                 nbl->nbl_time = jiffies;
6137                 spin_lock(&nn->blocked_locks_lock);
6138                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6139                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6140                 spin_unlock(&nn->blocked_locks_lock);
6141         }
6142
6143         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6144         switch (err) {
6145         case 0: /* success! */
6146                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6147                 status = 0;
6148                 if (lock->lk_reclaim)
6149                         nn->somebody_reclaimed = true;
6150                 break;
6151         case FILE_LOCK_DEFERRED:
6152                 nbl = NULL;
6153                 /* Fallthrough */
6154         case -EAGAIN:           /* conflock holds conflicting lock */
6155                 status = nfserr_denied;
6156                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6157                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6158                 break;
6159         case -EDEADLK:
6160                 status = nfserr_deadlock;
6161                 break;
6162         default:
6163                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6164                 status = nfserrno(err);
6165                 break;
6166         }
6167 out:
6168         if (nbl) {
6169                 /* dequeue it if we queued it before */
6170                 if (fl_flags & FL_SLEEP) {
6171                         spin_lock(&nn->blocked_locks_lock);
6172                         list_del_init(&nbl->nbl_list);
6173                         list_del_init(&nbl->nbl_lru);
6174                         spin_unlock(&nn->blocked_locks_lock);
6175                 }
6176                 free_blocked_lock(nbl);
6177         }
6178         if (filp)
6179                 fput(filp);
6180         if (lock_stp) {
6181                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6182                 if (cstate->replay_owner &&
6183                     cstate->replay_owner != &lock_sop->lo_owner &&
6184                     seqid_mutating_err(ntohl(status)))
6185                         lock_sop->lo_owner.so_seqid++;
6186
6187                 /*
6188                  * If this is a new, never-before-used stateid, and we are
6189                  * returning an error, then just go ahead and release it.
6190                  */
6191                 if (status && new)
6192                         release_lock_stateid(lock_stp);
6193
6194                 mutex_unlock(&lock_stp->st_mutex);
6195
6196                 nfs4_put_stid(&lock_stp->st_stid);
6197         }
6198         if (open_stp)
6199                 nfs4_put_stid(&open_stp->st_stid);
6200         nfsd4_bump_seqid(cstate, status);
6201         if (conflock)
6202                 locks_free_lock(conflock);
6203         return status;
6204 }
6205
6206 /*
6207  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6208  * so we do a temporary open here just to get an open file to pass to
6209  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6210  * inode operation.)
6211  */
6212 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6213 {
6214         struct file *file;
6215         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6216         if (!err) {
6217                 err = nfserrno(vfs_test_lock(file, lock));
6218                 fput(file);
6219         }
6220         return err;
6221 }
6222
6223 /*
6224  * LOCKT operation
6225  */
6226 __be32
6227 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6228             union nfsd4_op_u *u)
6229 {
6230         struct nfsd4_lockt *lockt = &u->lockt;
6231         struct file_lock *file_lock = NULL;
6232         struct nfs4_lockowner *lo = NULL;
6233         __be32 status;
6234         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6235
6236         if (locks_in_grace(SVC_NET(rqstp)))
6237                 return nfserr_grace;
6238
6239         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6240                  return nfserr_inval;
6241
6242         if (!nfsd4_has_session(cstate)) {
6243                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6244                 if (status)
6245                         goto out;
6246         }
6247
6248         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6249                 goto out;
6250
6251         file_lock = locks_alloc_lock();
6252         if (!file_lock) {
6253                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6254                 status = nfserr_jukebox;
6255                 goto out;
6256         }
6257
6258         switch (lockt->lt_type) {
6259                 case NFS4_READ_LT:
6260                 case NFS4_READW_LT:
6261                         file_lock->fl_type = F_RDLCK;
6262                         break;
6263                 case NFS4_WRITE_LT:
6264                 case NFS4_WRITEW_LT:
6265                         file_lock->fl_type = F_WRLCK;
6266                         break;
6267                 default:
6268                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6269                         status = nfserr_inval;
6270                         goto out;
6271         }
6272
6273         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6274         if (lo)
6275                 file_lock->fl_owner = (fl_owner_t)lo;
6276         file_lock->fl_pid = current->tgid;
6277         file_lock->fl_flags = FL_POSIX;
6278
6279         file_lock->fl_start = lockt->lt_offset;
6280         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6281
6282         nfs4_transform_lock_offset(file_lock);
6283
6284         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6285         if (status)
6286                 goto out;
6287
6288         if (file_lock->fl_type != F_UNLCK) {
6289                 status = nfserr_denied;
6290                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6291         }
6292 out:
6293         if (lo)
6294                 nfs4_put_stateowner(&lo->lo_owner);
6295         if (file_lock)
6296                 locks_free_lock(file_lock);
6297         return status;
6298 }
6299
6300 __be32
6301 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6302             union nfsd4_op_u *u)
6303 {
6304         struct nfsd4_locku *locku = &u->locku;
6305         struct nfs4_ol_stateid *stp;
6306         struct file *filp = NULL;
6307         struct file_lock *file_lock = NULL;
6308         __be32 status;
6309         int err;
6310         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6311
6312         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6313                 (long long) locku->lu_offset,
6314                 (long long) locku->lu_length);
6315
6316         if (check_lock_length(locku->lu_offset, locku->lu_length))
6317                  return nfserr_inval;
6318
6319         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6320                                         &locku->lu_stateid, NFS4_LOCK_STID,
6321                                         &stp, nn);
6322         if (status)
6323                 goto out;
6324         filp = find_any_file(stp->st_stid.sc_file);
6325         if (!filp) {
6326                 status = nfserr_lock_range;
6327                 goto put_stateid;
6328         }
6329         file_lock = locks_alloc_lock();
6330         if (!file_lock) {
6331                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6332                 status = nfserr_jukebox;
6333                 goto fput;
6334         }
6335
6336         file_lock->fl_type = F_UNLCK;
6337         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6338         file_lock->fl_pid = current->tgid;
6339         file_lock->fl_file = filp;
6340         file_lock->fl_flags = FL_POSIX;
6341         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6342         file_lock->fl_start = locku->lu_offset;
6343
6344         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6345                                                 locku->lu_length);
6346         nfs4_transform_lock_offset(file_lock);
6347
6348         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6349         if (err) {
6350                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6351                 goto out_nfserr;
6352         }
6353         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6354 fput:
6355         fput(filp);
6356 put_stateid:
6357         mutex_unlock(&stp->st_mutex);
6358         nfs4_put_stid(&stp->st_stid);
6359 out:
6360         nfsd4_bump_seqid(cstate, status);
6361         if (file_lock)
6362                 locks_free_lock(file_lock);
6363         return status;
6364
6365 out_nfserr:
6366         status = nfserrno(err);
6367         goto fput;
6368 }
6369
6370 /*
6371  * returns
6372  *      true:  locks held by lockowner
6373  *      false: no locks held by lockowner
6374  */
6375 static bool
6376 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6377 {
6378         struct file_lock *fl;
6379         int status = false;
6380         struct file *filp = find_any_file(fp);
6381         struct inode *inode;
6382         struct file_lock_context *flctx;
6383
6384         if (!filp) {
6385                 /* Any valid lock stateid should have some sort of access */
6386                 WARN_ON_ONCE(1);
6387                 return status;
6388         }
6389
6390         inode = locks_inode(filp);
6391         flctx = inode->i_flctx;
6392
6393         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6394                 spin_lock(&flctx->flc_lock);
6395                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6396                         if (fl->fl_owner == (fl_owner_t)lowner) {
6397                                 status = true;
6398                                 break;
6399                         }
6400                 }
6401                 spin_unlock(&flctx->flc_lock);
6402         }
6403         fput(filp);
6404         return status;
6405 }
6406
6407 __be32
6408 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6409                         struct nfsd4_compound_state *cstate,
6410                         union nfsd4_op_u *u)
6411 {
6412         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6413         clientid_t *clid = &rlockowner->rl_clientid;
6414         struct nfs4_stateowner *sop;
6415         struct nfs4_lockowner *lo = NULL;
6416         struct nfs4_ol_stateid *stp;
6417         struct xdr_netobj *owner = &rlockowner->rl_owner;
6418         unsigned int hashval = ownerstr_hashval(owner);
6419         __be32 status;
6420         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6421         struct nfs4_client *clp;
6422         LIST_HEAD (reaplist);
6423
6424         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6425                 clid->cl_boot, clid->cl_id);
6426
6427         status = lookup_clientid(clid, cstate, nn);
6428         if (status)
6429                 return status;
6430
6431         clp = cstate->clp;
6432         /* Find the matching lock stateowner */
6433         spin_lock(&clp->cl_lock);
6434         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6435                             so_strhash) {
6436
6437                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6438                         continue;
6439
6440                 /* see if there are still any locks associated with it */
6441                 lo = lockowner(sop);
6442                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6443                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6444                                 status = nfserr_locks_held;
6445                                 spin_unlock(&clp->cl_lock);
6446                                 return status;
6447                         }
6448                 }
6449
6450                 nfs4_get_stateowner(sop);
6451                 break;
6452         }
6453         if (!lo) {
6454                 spin_unlock(&clp->cl_lock);
6455                 return status;
6456         }
6457
6458         unhash_lockowner_locked(lo);
6459         while (!list_empty(&lo->lo_owner.so_stateids)) {
6460                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6461                                        struct nfs4_ol_stateid,
6462                                        st_perstateowner);
6463                 WARN_ON(!unhash_lock_stateid(stp));
6464                 put_ol_stateid_locked(stp, &reaplist);
6465         }
6466         spin_unlock(&clp->cl_lock);
6467         free_ol_stateid_reaplist(&reaplist);
6468         remove_blocked_locks(lo);
6469         nfs4_put_stateowner(&lo->lo_owner);
6470
6471         return status;
6472 }
6473
6474 static inline struct nfs4_client_reclaim *
6475 alloc_reclaim(void)
6476 {
6477         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6478 }
6479
6480 bool
6481 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
6482 {
6483         struct nfs4_client_reclaim *crp;
6484
6485         crp = nfsd4_find_reclaim_client(name, nn);
6486         return (crp && crp->cr_clp);
6487 }
6488
6489 /*
6490  * failure => all reset bets are off, nfserr_no_grace...
6491  *
6492  * The caller is responsible for freeing name.data if NULL is returned (it
6493  * will be freed in nfs4_remove_reclaim_record in the normal case).
6494  */
6495 struct nfs4_client_reclaim *
6496 nfs4_client_to_reclaim(struct xdr_netobj name, struct nfsd_net *nn)
6497 {
6498         unsigned int strhashval;
6499         struct nfs4_client_reclaim *crp;
6500
6501         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
6502         crp = alloc_reclaim();
6503         if (crp) {
6504                 strhashval = clientstr_hashval(name);
6505                 INIT_LIST_HEAD(&crp->cr_strhash);
6506                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6507                 crp->cr_name.data = name.data;
6508                 crp->cr_name.len = name.len;
6509                 crp->cr_clp = NULL;
6510                 nn->reclaim_str_hashtbl_size++;
6511         }
6512         return crp;
6513 }
6514
6515 void
6516 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6517 {
6518         list_del(&crp->cr_strhash);
6519         kfree(crp->cr_name.data);
6520         kfree(crp);
6521         nn->reclaim_str_hashtbl_size--;
6522 }
6523
6524 void
6525 nfs4_release_reclaim(struct nfsd_net *nn)
6526 {
6527         struct nfs4_client_reclaim *crp = NULL;
6528         int i;
6529
6530         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6531                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6532                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6533                                         struct nfs4_client_reclaim, cr_strhash);
6534                         nfs4_remove_reclaim_record(crp, nn);
6535                 }
6536         }
6537         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6538 }
6539
6540 /*
6541  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6542 struct nfs4_client_reclaim *
6543 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
6544 {
6545         unsigned int strhashval;
6546         struct nfs4_client_reclaim *crp = NULL;
6547
6548         dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
6549
6550         strhashval = clientstr_hashval(name);
6551         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6552                 if (compare_blob(&crp->cr_name, &name) == 0) {
6553                         return crp;
6554                 }
6555         }
6556         return NULL;
6557 }
6558
6559 /*
6560 * Called from OPEN. Look for clientid in reclaim list.
6561 */
6562 __be32
6563 nfs4_check_open_reclaim(clientid_t *clid,
6564                 struct nfsd4_compound_state *cstate,
6565                 struct nfsd_net *nn)
6566 {
6567         __be32 status;
6568
6569         /* find clientid in conf_id_hashtbl */
6570         status = lookup_clientid(clid, cstate, nn);
6571         if (status)
6572                 return nfserr_reclaim_bad;
6573
6574         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6575                 return nfserr_no_grace;
6576
6577         if (nfsd4_client_record_check(cstate->clp))
6578                 return nfserr_reclaim_bad;
6579
6580         return nfs_ok;
6581 }
6582
6583 #ifdef CONFIG_NFSD_FAULT_INJECTION
6584 static inline void
6585 put_client(struct nfs4_client *clp)
6586 {
6587         atomic_dec(&clp->cl_refcount);
6588 }
6589
6590 static struct nfs4_client *
6591 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6592 {
6593         struct nfs4_client *clp;
6594         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6595                                           nfsd_net_id);
6596
6597         if (!nfsd_netns_ready(nn))
6598                 return NULL;
6599
6600         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6601                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6602                         return clp;
6603         }
6604         return NULL;
6605 }
6606
6607 u64
6608 nfsd_inject_print_clients(void)
6609 {
6610         struct nfs4_client *clp;
6611         u64 count = 0;
6612         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6613                                           nfsd_net_id);
6614         char buf[INET6_ADDRSTRLEN];
6615
6616         if (!nfsd_netns_ready(nn))
6617                 return 0;
6618
6619         spin_lock(&nn->client_lock);
6620         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6621                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6622                 pr_info("NFS Client: %s\n", buf);
6623                 ++count;
6624         }
6625         spin_unlock(&nn->client_lock);
6626
6627         return count;
6628 }
6629
6630 u64
6631 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6632 {
6633         u64 count = 0;
6634         struct nfs4_client *clp;
6635         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6636                                           nfsd_net_id);
6637
6638         if (!nfsd_netns_ready(nn))
6639                 return count;
6640
6641         spin_lock(&nn->client_lock);
6642         clp = nfsd_find_client(addr, addr_size);
6643         if (clp) {
6644                 if (mark_client_expired_locked(clp) == nfs_ok)
6645                         ++count;
6646                 else
6647                         clp = NULL;
6648         }
6649         spin_unlock(&nn->client_lock);
6650
6651         if (clp)
6652                 expire_client(clp);
6653
6654         return count;
6655 }
6656
6657 u64
6658 nfsd_inject_forget_clients(u64 max)
6659 {
6660         u64 count = 0;
6661         struct nfs4_client *clp, *next;
6662         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6663                                                 nfsd_net_id);
6664         LIST_HEAD(reaplist);
6665
6666         if (!nfsd_netns_ready(nn))
6667                 return count;
6668
6669         spin_lock(&nn->client_lock);
6670         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6671                 if (mark_client_expired_locked(clp) == nfs_ok) {
6672                         list_add(&clp->cl_lru, &reaplist);
6673                         if (max != 0 && ++count >= max)
6674                                 break;
6675                 }
6676         }
6677         spin_unlock(&nn->client_lock);
6678
6679         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6680                 expire_client(clp);
6681
6682         return count;
6683 }
6684
6685 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6686                              const char *type)
6687 {
6688         char buf[INET6_ADDRSTRLEN];
6689         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6690         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6691 }
6692
6693 static void
6694 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6695                              struct list_head *collect)
6696 {
6697         struct nfs4_client *clp = lst->st_stid.sc_client;
6698         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6699                                           nfsd_net_id);
6700
6701         if (!collect)
6702                 return;
6703
6704         lockdep_assert_held(&nn->client_lock);
6705         atomic_inc(&clp->cl_refcount);
6706         list_add(&lst->st_locks, collect);
6707 }
6708
6709 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6710                                     struct list_head *collect,
6711                                     bool (*func)(struct nfs4_ol_stateid *))
6712 {
6713         struct nfs4_openowner *oop;
6714         struct nfs4_ol_stateid *stp, *st_next;
6715         struct nfs4_ol_stateid *lst, *lst_next;
6716         u64 count = 0;
6717
6718         spin_lock(&clp->cl_lock);
6719         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6720                 list_for_each_entry_safe(stp, st_next,
6721                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6722                         list_for_each_entry_safe(lst, lst_next,
6723                                         &stp->st_locks, st_locks) {
6724                                 if (func) {
6725                                         if (func(lst))
6726                                                 nfsd_inject_add_lock_to_list(lst,
6727                                                                         collect);
6728                                 }
6729                                 ++count;
6730                                 /*
6731                                  * Despite the fact that these functions deal
6732                                  * with 64-bit integers for "count", we must
6733                                  * ensure that it doesn't blow up the
6734                                  * clp->cl_refcount. Throw a warning if we
6735                                  * start to approach INT_MAX here.
6736                                  */
6737                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6738                                 if (count == max)
6739                                         goto out;
6740                         }
6741                 }
6742         }
6743 out:
6744         spin_unlock(&clp->cl_lock);
6745
6746         return count;
6747 }
6748
6749 static u64
6750 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6751                           u64 max)
6752 {
6753         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6754 }
6755
6756 static u64
6757 nfsd_print_client_locks(struct nfs4_client *clp)
6758 {
6759         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6760         nfsd_print_count(clp, count, "locked files");
6761         return count;
6762 }
6763
6764 u64
6765 nfsd_inject_print_locks(void)
6766 {
6767         struct nfs4_client *clp;
6768         u64 count = 0;
6769         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6770                                                 nfsd_net_id);
6771
6772         if (!nfsd_netns_ready(nn))
6773                 return 0;
6774
6775         spin_lock(&nn->client_lock);
6776         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6777                 count += nfsd_print_client_locks(clp);
6778         spin_unlock(&nn->client_lock);
6779
6780         return count;
6781 }
6782
6783 static void
6784 nfsd_reap_locks(struct list_head *reaplist)
6785 {
6786         struct nfs4_client *clp;
6787         struct nfs4_ol_stateid *stp, *next;
6788
6789         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6790                 list_del_init(&stp->st_locks);
6791                 clp = stp->st_stid.sc_client;
6792                 nfs4_put_stid(&stp->st_stid);
6793                 put_client(clp);
6794         }
6795 }
6796
6797 u64
6798 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6799 {
6800         unsigned int count = 0;
6801         struct nfs4_client *clp;
6802         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6803                                                 nfsd_net_id);
6804         LIST_HEAD(reaplist);
6805
6806         if (!nfsd_netns_ready(nn))
6807                 return count;
6808
6809         spin_lock(&nn->client_lock);
6810         clp = nfsd_find_client(addr, addr_size);
6811         if (clp)
6812                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6813         spin_unlock(&nn->client_lock);
6814         nfsd_reap_locks(&reaplist);
6815         return count;
6816 }
6817
6818 u64
6819 nfsd_inject_forget_locks(u64 max)
6820 {
6821         u64 count = 0;
6822         struct nfs4_client *clp;
6823         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6824                                                 nfsd_net_id);
6825         LIST_HEAD(reaplist);
6826
6827         if (!nfsd_netns_ready(nn))
6828                 return count;
6829
6830         spin_lock(&nn->client_lock);
6831         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6832                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6833                 if (max != 0 && count >= max)
6834                         break;
6835         }
6836         spin_unlock(&nn->client_lock);
6837         nfsd_reap_locks(&reaplist);
6838         return count;
6839 }
6840
6841 static u64
6842 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6843                               struct list_head *collect,
6844                               void (*func)(struct nfs4_openowner *))
6845 {
6846         struct nfs4_openowner *oop, *next;
6847         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6848                                                 nfsd_net_id);
6849         u64 count = 0;
6850
6851         lockdep_assert_held(&nn->client_lock);
6852
6853         spin_lock(&clp->cl_lock);
6854         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6855                 if (func) {
6856                         func(oop);
6857                         if (collect) {
6858                                 atomic_inc(&clp->cl_refcount);
6859                                 list_add(&oop->oo_perclient, collect);
6860                         }
6861                 }
6862                 ++count;
6863                 /*
6864                  * Despite the fact that these functions deal with
6865                  * 64-bit integers for "count", we must ensure that
6866                  * it doesn't blow up the clp->cl_refcount. Throw a
6867                  * warning if we start to approach INT_MAX here.
6868                  */
6869                 WARN_ON_ONCE(count == (INT_MAX / 2));
6870                 if (count == max)
6871                         break;
6872         }
6873         spin_unlock(&clp->cl_lock);
6874
6875         return count;
6876 }
6877
6878 static u64
6879 nfsd_print_client_openowners(struct nfs4_client *clp)
6880 {
6881         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6882
6883         nfsd_print_count(clp, count, "openowners");
6884         return count;
6885 }
6886
6887 static u64
6888 nfsd_collect_client_openowners(struct nfs4_client *clp,
6889                                struct list_head *collect, u64 max)
6890 {
6891         return nfsd_foreach_client_openowner(clp, max, collect,
6892                                                 unhash_openowner_locked);
6893 }
6894
6895 u64
6896 nfsd_inject_print_openowners(void)
6897 {
6898         struct nfs4_client *clp;
6899         u64 count = 0;
6900         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6901                                                 nfsd_net_id);
6902
6903         if (!nfsd_netns_ready(nn))
6904                 return 0;
6905
6906         spin_lock(&nn->client_lock);
6907         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6908                 count += nfsd_print_client_openowners(clp);
6909         spin_unlock(&nn->client_lock);
6910
6911         return count;
6912 }
6913
6914 static void
6915 nfsd_reap_openowners(struct list_head *reaplist)
6916 {
6917         struct nfs4_client *clp;
6918         struct nfs4_openowner *oop, *next;
6919
6920         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6921                 list_del_init(&oop->oo_perclient);
6922                 clp = oop->oo_owner.so_client;
6923                 release_openowner(oop);
6924                 put_client(clp);
6925         }
6926 }
6927
6928 u64
6929 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6930                                      size_t addr_size)
6931 {
6932         unsigned int count = 0;
6933         struct nfs4_client *clp;
6934         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6935                                                 nfsd_net_id);
6936         LIST_HEAD(reaplist);
6937
6938         if (!nfsd_netns_ready(nn))
6939                 return count;
6940
6941         spin_lock(&nn->client_lock);
6942         clp = nfsd_find_client(addr, addr_size);
6943         if (clp)
6944                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6945         spin_unlock(&nn->client_lock);
6946         nfsd_reap_openowners(&reaplist);
6947         return count;
6948 }
6949
6950 u64
6951 nfsd_inject_forget_openowners(u64 max)
6952 {
6953         u64 count = 0;
6954         struct nfs4_client *clp;
6955         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6956                                                 nfsd_net_id);
6957         LIST_HEAD(reaplist);
6958
6959         if (!nfsd_netns_ready(nn))
6960                 return count;
6961
6962         spin_lock(&nn->client_lock);
6963         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6964                 count += nfsd_collect_client_openowners(clp, &reaplist,
6965                                                         max - count);
6966                 if (max != 0 && count >= max)
6967                         break;
6968         }
6969         spin_unlock(&nn->client_lock);
6970         nfsd_reap_openowners(&reaplist);
6971         return count;
6972 }
6973
6974 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6975                                      struct list_head *victims)
6976 {
6977         struct nfs4_delegation *dp, *next;
6978         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6979                                                 nfsd_net_id);
6980         u64 count = 0;
6981
6982         lockdep_assert_held(&nn->client_lock);
6983
6984         spin_lock(&state_lock);
6985         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6986                 if (victims) {
6987                         /*
6988                          * It's not safe to mess with delegations that have a
6989                          * non-zero dl_time. They might have already been broken
6990                          * and could be processed by the laundromat outside of
6991                          * the state_lock. Just leave them be.
6992                          */
6993                         if (dp->dl_time != 0)
6994                                 continue;
6995
6996                         atomic_inc(&clp->cl_refcount);
6997                         WARN_ON(!unhash_delegation_locked(dp));
6998                         list_add(&dp->dl_recall_lru, victims);
6999                 }
7000                 ++count;
7001                 /*
7002                  * Despite the fact that these functions deal with
7003                  * 64-bit integers for "count", we must ensure that
7004                  * it doesn't blow up the clp->cl_refcount. Throw a
7005                  * warning if we start to approach INT_MAX here.
7006                  */
7007                 WARN_ON_ONCE(count == (INT_MAX / 2));
7008                 if (count == max)
7009                         break;
7010         }
7011         spin_unlock(&state_lock);
7012         return count;
7013 }
7014
7015 static u64
7016 nfsd_print_client_delegations(struct nfs4_client *clp)
7017 {
7018         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7019
7020         nfsd_print_count(clp, count, "delegations");
7021         return count;
7022 }
7023
7024 u64
7025 nfsd_inject_print_delegations(void)
7026 {
7027         struct nfs4_client *clp;
7028         u64 count = 0;
7029         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7030                                                 nfsd_net_id);
7031
7032         if (!nfsd_netns_ready(nn))
7033                 return 0;
7034
7035         spin_lock(&nn->client_lock);
7036         list_for_each_entry(clp, &nn->client_lru, cl_lru)
7037                 count += nfsd_print_client_delegations(clp);
7038         spin_unlock(&nn->client_lock);
7039
7040         return count;
7041 }
7042
7043 static void
7044 nfsd_forget_delegations(struct list_head *reaplist)
7045 {
7046         struct nfs4_client *clp;
7047         struct nfs4_delegation *dp, *next;
7048
7049         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7050                 list_del_init(&dp->dl_recall_lru);
7051                 clp = dp->dl_stid.sc_client;
7052                 revoke_delegation(dp);
7053                 put_client(clp);
7054         }
7055 }
7056
7057 u64
7058 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7059                                       size_t addr_size)
7060 {
7061         u64 count = 0;
7062         struct nfs4_client *clp;
7063         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7064                                                 nfsd_net_id);
7065         LIST_HEAD(reaplist);
7066
7067         if (!nfsd_netns_ready(nn))
7068                 return count;
7069
7070         spin_lock(&nn->client_lock);
7071         clp = nfsd_find_client(addr, addr_size);
7072         if (clp)
7073                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7074         spin_unlock(&nn->client_lock);
7075
7076         nfsd_forget_delegations(&reaplist);
7077         return count;
7078 }
7079
7080 u64
7081 nfsd_inject_forget_delegations(u64 max)
7082 {
7083         u64 count = 0;
7084         struct nfs4_client *clp;
7085         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7086                                                 nfsd_net_id);
7087         LIST_HEAD(reaplist);
7088
7089         if (!nfsd_netns_ready(nn))
7090                 return count;
7091
7092         spin_lock(&nn->client_lock);
7093         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7094                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7095                 if (max != 0 && count >= max)
7096                         break;
7097         }
7098         spin_unlock(&nn->client_lock);
7099         nfsd_forget_delegations(&reaplist);
7100         return count;
7101 }
7102
7103 static void
7104 nfsd_recall_delegations(struct list_head *reaplist)
7105 {
7106         struct nfs4_client *clp;
7107         struct nfs4_delegation *dp, *next;
7108
7109         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7110                 list_del_init(&dp->dl_recall_lru);
7111                 clp = dp->dl_stid.sc_client;
7112                 /*
7113                  * We skipped all entries that had a zero dl_time before,
7114                  * so we can now reset the dl_time back to 0. If a delegation
7115                  * break comes in now, then it won't make any difference since
7116                  * we're recalling it either way.
7117                  */
7118                 spin_lock(&state_lock);
7119                 dp->dl_time = 0;
7120                 spin_unlock(&state_lock);
7121                 nfsd_break_one_deleg(dp);
7122                 put_client(clp);
7123         }
7124 }
7125
7126 u64
7127 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7128                                       size_t addr_size)
7129 {
7130         u64 count = 0;
7131         struct nfs4_client *clp;
7132         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7133                                                 nfsd_net_id);
7134         LIST_HEAD(reaplist);
7135
7136         if (!nfsd_netns_ready(nn))
7137                 return count;
7138
7139         spin_lock(&nn->client_lock);
7140         clp = nfsd_find_client(addr, addr_size);
7141         if (clp)
7142                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7143         spin_unlock(&nn->client_lock);
7144
7145         nfsd_recall_delegations(&reaplist);
7146         return count;
7147 }
7148
7149 u64
7150 nfsd_inject_recall_delegations(u64 max)
7151 {
7152         u64 count = 0;
7153         struct nfs4_client *clp, *next;
7154         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7155                                                 nfsd_net_id);
7156         LIST_HEAD(reaplist);
7157
7158         if (!nfsd_netns_ready(nn))
7159                 return count;
7160
7161         spin_lock(&nn->client_lock);
7162         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7163                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7164                 if (max != 0 && ++count >= max)
7165                         break;
7166         }
7167         spin_unlock(&nn->client_lock);
7168         nfsd_recall_delegations(&reaplist);
7169         return count;
7170 }
7171 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7172
7173 /*
7174  * Since the lifetime of a delegation isn't limited to that of an open, a
7175  * client may quite reasonably hang on to a delegation as long as it has
7176  * the inode cached.  This becomes an obvious problem the first time a
7177  * client's inode cache approaches the size of the server's total memory.
7178  *
7179  * For now we avoid this problem by imposing a hard limit on the number
7180  * of delegations, which varies according to the server's memory size.
7181  */
7182 static void
7183 set_max_delegations(void)
7184 {
7185         /*
7186          * Allow at most 4 delegations per megabyte of RAM.  Quick
7187          * estimates suggest that in the worst case (where every delegation
7188          * is for a different inode), a delegation could take about 1.5K,
7189          * giving a worst case usage of about 6% of memory.
7190          */
7191         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7192 }
7193
7194 static int nfs4_state_create_net(struct net *net)
7195 {
7196         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7197         int i;
7198
7199         nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7200                                             sizeof(struct list_head),
7201                                             GFP_KERNEL);
7202         if (!nn->conf_id_hashtbl)
7203                 goto err;
7204         nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7205                                               sizeof(struct list_head),
7206                                               GFP_KERNEL);
7207         if (!nn->unconf_id_hashtbl)
7208                 goto err_unconf_id;
7209         nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7210                                               sizeof(struct list_head),
7211                                               GFP_KERNEL);
7212         if (!nn->sessionid_hashtbl)
7213                 goto err_sessionid;
7214
7215         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7216                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7217                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7218         }
7219         for (i = 0; i < SESSION_HASH_SIZE; i++)
7220                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7221         nn->conf_name_tree = RB_ROOT;
7222         nn->unconf_name_tree = RB_ROOT;
7223         nn->boot_time = get_seconds();
7224         nn->grace_ended = false;
7225         nn->nfsd4_manager.block_opens = true;
7226         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7227         INIT_LIST_HEAD(&nn->client_lru);
7228         INIT_LIST_HEAD(&nn->close_lru);
7229         INIT_LIST_HEAD(&nn->del_recall_lru);
7230         spin_lock_init(&nn->client_lock);
7231         spin_lock_init(&nn->s2s_cp_lock);
7232         idr_init(&nn->s2s_cp_stateids);
7233
7234         spin_lock_init(&nn->blocked_locks_lock);
7235         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7236
7237         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7238         get_net(net);
7239
7240         return 0;
7241
7242 err_sessionid:
7243         kfree(nn->unconf_id_hashtbl);
7244 err_unconf_id:
7245         kfree(nn->conf_id_hashtbl);
7246 err:
7247         return -ENOMEM;
7248 }
7249
7250 static void
7251 nfs4_state_destroy_net(struct net *net)
7252 {
7253         int i;
7254         struct nfs4_client *clp = NULL;
7255         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7256
7257         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7258                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7259                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7260                         destroy_client(clp);
7261                 }
7262         }
7263
7264         WARN_ON(!list_empty(&nn->blocked_locks_lru));
7265
7266         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7267                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7268                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7269                         destroy_client(clp);
7270                 }
7271         }
7272
7273         kfree(nn->sessionid_hashtbl);
7274         kfree(nn->unconf_id_hashtbl);
7275         kfree(nn->conf_id_hashtbl);
7276         put_net(net);
7277 }
7278
7279 int
7280 nfs4_state_start_net(struct net *net)
7281 {
7282         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7283         int ret;
7284
7285         ret = nfs4_state_create_net(net);
7286         if (ret)
7287                 return ret;
7288         locks_start_grace(net, &nn->nfsd4_manager);
7289         nfsd4_client_tracking_init(net);
7290         if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7291                 goto skip_grace;
7292         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7293                nn->nfsd4_grace, net->ns.inum);
7294         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7295         return 0;
7296
7297 skip_grace:
7298         printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7299                         net->ns.inum);
7300         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7301         nfsd4_end_grace(nn);
7302         return 0;
7303 }
7304
7305 /* initialization to perform when the nfsd service is started: */
7306
7307 int
7308 nfs4_state_start(void)
7309 {
7310         int ret;
7311
7312         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7313         if (laundry_wq == NULL) {
7314                 ret = -ENOMEM;
7315                 goto out;
7316         }
7317         ret = nfsd4_create_callback_queue();
7318         if (ret)
7319                 goto out_free_laundry;
7320
7321         set_max_delegations();
7322         return 0;
7323
7324 out_free_laundry:
7325         destroy_workqueue(laundry_wq);
7326 out:
7327         return ret;
7328 }
7329
7330 void
7331 nfs4_state_shutdown_net(struct net *net)
7332 {
7333         struct nfs4_delegation *dp = NULL;
7334         struct list_head *pos, *next, reaplist;
7335         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7336
7337         cancel_delayed_work_sync(&nn->laundromat_work);
7338         locks_end_grace(&nn->nfsd4_manager);
7339
7340         INIT_LIST_HEAD(&reaplist);
7341         spin_lock(&state_lock);
7342         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7343                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7344                 WARN_ON(!unhash_delegation_locked(dp));
7345                 list_add(&dp->dl_recall_lru, &reaplist);
7346         }
7347         spin_unlock(&state_lock);
7348         list_for_each_safe(pos, next, &reaplist) {
7349                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7350                 list_del_init(&dp->dl_recall_lru);
7351                 destroy_unhashed_deleg(dp);
7352         }
7353
7354         nfsd4_client_tracking_exit(net);
7355         nfs4_state_destroy_net(net);
7356 }
7357
7358 void
7359 nfs4_state_shutdown(void)
7360 {
7361         destroy_workqueue(laundry_wq);
7362         nfsd4_destroy_callback_queue();
7363 }
7364
7365 static void
7366 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7367 {
7368         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7369                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7370 }
7371
7372 static void
7373 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7374 {
7375         if (cstate->minorversion) {
7376                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7377                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7378         }
7379 }
7380
7381 void
7382 clear_current_stateid(struct nfsd4_compound_state *cstate)
7383 {
7384         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7385 }
7386
7387 /*
7388  * functions to set current state id
7389  */
7390 void
7391 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7392                 union nfsd4_op_u *u)
7393 {
7394         put_stateid(cstate, &u->open_downgrade.od_stateid);
7395 }
7396
7397 void
7398 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7399                 union nfsd4_op_u *u)
7400 {
7401         put_stateid(cstate, &u->open.op_stateid);
7402 }
7403
7404 void
7405 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7406                 union nfsd4_op_u *u)
7407 {
7408         put_stateid(cstate, &u->close.cl_stateid);
7409 }
7410
7411 void
7412 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7413                 union nfsd4_op_u *u)
7414 {
7415         put_stateid(cstate, &u->lock.lk_resp_stateid);
7416 }
7417
7418 /*
7419  * functions to consume current state id
7420  */
7421
7422 void
7423 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7424                 union nfsd4_op_u *u)
7425 {
7426         get_stateid(cstate, &u->open_downgrade.od_stateid);
7427 }
7428
7429 void
7430 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7431                 union nfsd4_op_u *u)
7432 {
7433         get_stateid(cstate, &u->delegreturn.dr_stateid);
7434 }
7435
7436 void
7437 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7438                 union nfsd4_op_u *u)
7439 {
7440         get_stateid(cstate, &u->free_stateid.fr_stateid);
7441 }
7442
7443 void
7444 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7445                 union nfsd4_op_u *u)
7446 {
7447         get_stateid(cstate, &u->setattr.sa_stateid);
7448 }
7449
7450 void
7451 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7452                 union nfsd4_op_u *u)
7453 {
7454         get_stateid(cstate, &u->close.cl_stateid);
7455 }
7456
7457 void
7458 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7459                 union nfsd4_op_u *u)
7460 {
7461         get_stateid(cstate, &u->locku.lu_stateid);
7462 }
7463
7464 void
7465 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7466                 union nfsd4_op_u *u)
7467 {
7468         get_stateid(cstate, &u->read.rd_stateid);
7469 }
7470
7471 void
7472 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7473                 union nfsd4_op_u *u)
7474 {
7475         get_stateid(cstate, &u->write.wr_stateid);
7476 }