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
  52 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  53
  54 #define all_ones {{~0,~0},~0}
  55 static const stateid_t one_stateid = {
  56         .si_generation = ~0,
  57         .si_opaque = all_ones,
  58 };
  59 static const stateid_t zero_stateid = {
  60         /* all fields zero */
  61 };
  62 static const stateid_t currentstateid = {
  63         .si_generation = 1,
  64 };
  65
  66 static u64 current_sessionid = 1;
  67
  68 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  69 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  70 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  71
  72 /* forward declarations */
  73 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  74 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  75
  76 /* Locking: */
  77
  78 /*
  79  * Currently used for the del_recall_lru and file hash table.  In an
  80  * effort to decrease the scope of the client_mutex, this spinlock may
  81  * eventually cover more:
  82  */
  83 static DEFINE_SPINLOCK(state_lock);
  84
  85 /*
  86  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  87  * the refcount on the open stateid to drop.
  88  */
  89 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  90
  91 static struct kmem_cache *openowner_slab;
  92 static struct kmem_cache *lockowner_slab;
  93 static struct kmem_cache *file_slab;
  94 static struct kmem_cache *stateid_slab;
  95 static struct kmem_cache *deleg_slab;
  96
  97 static void free_session(struct nfsd4_session *);
  98
  99 static struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 100
 101 static bool is_session_dead(struct nfsd4_session *ses)
 102 {
 103         return ses->se_flags & NFS4_SESSION_DEAD;
 104 }
 105
 106 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 107 {
 108         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 109                 return nfserr_jukebox;
 110         ses->se_flags |= NFS4_SESSION_DEAD;
 111         return nfs_ok;
 112 }
 113
 114 static bool is_client_expired(struct nfs4_client *clp)
 115 {
 116         return clp->cl_time == 0;
 117 }
 118
 119 static __be32 get_client_locked(struct nfs4_client *clp)
 120 {
 121         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 122
 123         lockdep_assert_held(&nn->client_lock);
 124
 125         if (is_client_expired(clp))
 126                 return nfserr_expired;
 127         atomic_inc(&clp->cl_refcount);
 128         return nfs_ok;
 129 }
 130
 131 /* must be called under the client_lock */
 132 static inline void
 133 renew_client_locked(struct nfs4_client *clp)
 134 {
 135         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 136
 137         if (is_client_expired(clp)) {
 138                 WARN_ON(1);
 139                 printk("%s: client (clientid %08x/%08x) already expired\n",
 140                         __func__,
 141                         clp->cl_clientid.cl_boot,
 142                         clp->cl_clientid.cl_id);
 143                 return;
 144         }
 145
 146         dprintk("renewing client (clientid %08x/%08x)\n",
 147                         clp->cl_clientid.cl_boot,
 148                         clp->cl_clientid.cl_id);
 149         list_move_tail(&clp->cl_lru, &nn->client_lru);
 150         clp->cl_time = get_seconds();
 151 }
 152
 153 static inline void
 154 renew_client(struct nfs4_client *clp)
 155 {
 156         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 157
 158         spin_lock(&nn->client_lock);
 159         renew_client_locked(clp);
 160         spin_unlock(&nn->client_lock);
 161 }
 162
 163 static void put_client_renew_locked(struct nfs4_client *clp)
 164 {
 165         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 166
 167         lockdep_assert_held(&nn->client_lock);
 168
 169         if (!atomic_dec_and_test(&clp->cl_refcount))
 170                 return;
 171         if (!is_client_expired(clp))
 172                 renew_client_locked(clp);
 173 }
 174
 175 static void put_client_renew(struct nfs4_client *clp)
 176 {
 177         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 178
 179         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 180                 return;
 181         if (!is_client_expired(clp))
 182                 renew_client_locked(clp);
 183         spin_unlock(&nn->client_lock);
 184 }
 185
 186 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 187 {
 188         __be32 status;
 189
 190         if (is_session_dead(ses))
 191                 return nfserr_badsession;
 192         status = get_client_locked(ses->se_client);
 193         if (status)
 194                 return status;
 195         atomic_inc(&ses->se_ref);
 196         return nfs_ok;
 197 }
 198
 199 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 200 {
 201         struct nfs4_client *clp = ses->se_client;
 202         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 203
 204         lockdep_assert_held(&nn->client_lock);
 205
 206         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 207                 free_session(ses);
 208         put_client_renew_locked(clp);
 209 }
 210
 211 static void nfsd4_put_session(struct nfsd4_session *ses)
 212 {
 213         struct nfs4_client *clp = ses->se_client;
 214         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 215
 216         spin_lock(&nn->client_lock);
 217         nfsd4_put_session_locked(ses);
 218         spin_unlock(&nn->client_lock);
 219 }
 220
 221 static inline struct nfs4_stateowner *
 222 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 223 {
 224         atomic_inc(&sop->so_count);
 225         return sop;
 226 }
 227
 228 static int
 229 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 230 {
 231         return (sop->so_owner.len == owner->len) &&
 232                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 233 }
 234
 235 static struct nfs4_openowner *
 236 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 237                         struct nfs4_client *clp)
 238 {
 239         struct nfs4_stateowner *so;
 240
 241         lockdep_assert_held(&clp->cl_lock);
 242
 243         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 244                             so_strhash) {
 245                 if (!so->so_is_open_owner)
 246                         continue;
 247                 if (same_owner_str(so, &open->op_owner))
 248                         return openowner(nfs4_get_stateowner(so));
 249         }
 250         return NULL;
 251 }
 252
 253 static struct nfs4_openowner *
 254 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 255                         struct nfs4_client *clp)
 256 {
 257         struct nfs4_openowner *oo;
 258
 259         spin_lock(&clp->cl_lock);
 260         oo = find_openstateowner_str_locked(hashval, open, clp);
 261         spin_unlock(&clp->cl_lock);
 262         return oo;
 263 }
 264
 265 static inline u32
 266 opaque_hashval(const void *ptr, int nbytes)
 267 {
 268         unsigned char *cptr = (unsigned char *) ptr;
 269
 270         u32 x = 0;
 271         while (nbytes--) {
 272                 x *= 37;
 273                 x += *cptr++;
 274         }
 275         return x;
 276 }
 277
 278 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 279 {
 280         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 281
 282         kmem_cache_free(file_slab, fp);
 283 }
 284
 285 static inline void
 286 put_nfs4_file(struct nfs4_file *fi)
 287 {
 288         might_lock(&state_lock);
 289
 290         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 291                 hlist_del_rcu(&fi->fi_hash);
 292                 spin_unlock(&state_lock);
 293                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 294                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 295         }
 296 }
 297
 298 static inline void
 299 get_nfs4_file(struct nfs4_file *fi)
 300 {
 301         atomic_inc(&fi->fi_ref);
 302 }
 303
 304 static struct file *
 305 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 306 {
 307         if (f->fi_fds[oflag])
 308                 return get_file(f->fi_fds[oflag]);
 309         return NULL;
 310 }
 311
 312 static struct file *
 313 find_writeable_file_locked(struct nfs4_file *f)
 314 {
 315         struct file *ret;
 316
 317         lockdep_assert_held(&f->fi_lock);
 318
 319         ret = __nfs4_get_fd(f, O_WRONLY);
 320         if (!ret)
 321                 ret = __nfs4_get_fd(f, O_RDWR);
 322         return ret;
 323 }
 324
 325 static struct file *
 326 find_writeable_file(struct nfs4_file *f)
 327 {
 328         struct file *ret;
 329
 330         spin_lock(&f->fi_lock);
 331         ret = find_writeable_file_locked(f);
 332         spin_unlock(&f->fi_lock);
 333
 334         return ret;
 335 }
 336
 337 static struct file *find_readable_file_locked(struct nfs4_file *f)
 338 {
 339         struct file *ret;
 340
 341         lockdep_assert_held(&f->fi_lock);
 342
 343         ret = __nfs4_get_fd(f, O_RDONLY);
 344         if (!ret)
 345                 ret = __nfs4_get_fd(f, O_RDWR);
 346         return ret;
 347 }
 348
 349 static struct file *
 350 find_readable_file(struct nfs4_file *f)
 351 {
 352         struct file *ret;
 353
 354         spin_lock(&f->fi_lock);
 355         ret = find_readable_file_locked(f);
 356         spin_unlock(&f->fi_lock);
 357
 358         return ret;
 359 }
 360
 361 static struct file *
 362 find_any_file(struct nfs4_file *f)
 363 {
 364         struct file *ret;
 365
 366         spin_lock(&f->fi_lock);
 367         ret = __nfs4_get_fd(f, O_RDWR);
 368         if (!ret) {
 369                 ret = __nfs4_get_fd(f, O_WRONLY);
 370                 if (!ret)
 371                         ret = __nfs4_get_fd(f, O_RDONLY);
 372         }
 373         spin_unlock(&f->fi_lock);
 374         return ret;
 375 }
 376
 377 static atomic_long_t num_delegations;
 378 unsigned long max_delegations;
 379
 380 /*
 381  * Open owner state (share locks)
 382  */
 383
 384 /* hash tables for lock and open owners */
 385 #define OWNER_HASH_BITS              8
 386 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 387 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 388
 389 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 390 {
 391         unsigned int ret;
 392
 393         ret = opaque_hashval(ownername->data, ownername->len);
 394         return ret & OWNER_HASH_MASK;
 395 }
 396
 397 /* hash table for nfs4_file */
 398 #define FILE_HASH_BITS                   8
 399 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 400
 401 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 402 {
 403         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 404 }
 405
 406 static unsigned int file_hashval(struct knfsd_fh *fh)
 407 {
 408         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 409 }
 410
 411 static bool nfsd_fh_match(struct knfsd_fh *fh1, struct knfsd_fh *fh2)
 412 {
 413         return fh1->fh_size == fh2->fh_size &&
 414                 !memcmp(fh1->fh_base.fh_pad,
 415                                 fh2->fh_base.fh_pad,
 416                                 fh1->fh_size);
 417 }
 418
 419 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 420
 421 static void
 422 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 423 {
 424         lockdep_assert_held(&fp->fi_lock);
 425
 426         if (access & NFS4_SHARE_ACCESS_WRITE)
 427                 atomic_inc(&fp->fi_access[O_WRONLY]);
 428         if (access & NFS4_SHARE_ACCESS_READ)
 429                 atomic_inc(&fp->fi_access[O_RDONLY]);
 430 }
 431
 432 static __be32
 433 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 434 {
 435         lockdep_assert_held(&fp->fi_lock);
 436
 437         /* Does this access mode make sense? */
 438         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 439                 return nfserr_inval;
 440
 441         /* Does it conflict with a deny mode already set? */
 442         if ((access & fp->fi_share_deny) != 0)
 443                 return nfserr_share_denied;
 444
 445         __nfs4_file_get_access(fp, access);
 446         return nfs_ok;
 447 }
 448
 449 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 450 {
 451         /* Common case is that there is no deny mode. */
 452         if (deny) {
 453                 /* Does this deny mode make sense? */
 454                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 455                         return nfserr_inval;
 456
 457                 if ((deny & NFS4_SHARE_DENY_READ) &&
 458                     atomic_read(&fp->fi_access[O_RDONLY]))
 459                         return nfserr_share_denied;
 460
 461                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 462                     atomic_read(&fp->fi_access[O_WRONLY]))
 463                         return nfserr_share_denied;
 464         }
 465         return nfs_ok;
 466 }
 467
 468 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 469 {
 470         might_lock(&fp->fi_lock);
 471
 472         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 473                 struct file *f1 = NULL;
 474                 struct file *f2 = NULL;
 475
 476                 swap(f1, fp->fi_fds[oflag]);
 477                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 478                         swap(f2, fp->fi_fds[O_RDWR]);
 479                 spin_unlock(&fp->fi_lock);
 480                 if (f1)
 481                         fput(f1);
 482                 if (f2)
 483                         fput(f2);
 484         }
 485 }
 486
 487 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 488 {
 489         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 490
 491         if (access & NFS4_SHARE_ACCESS_WRITE)
 492                 __nfs4_file_put_access(fp, O_WRONLY);
 493         if (access & NFS4_SHARE_ACCESS_READ)
 494                 __nfs4_file_put_access(fp, O_RDONLY);
 495 }
 496
 497 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
 498                                          struct kmem_cache *slab)
 499 {
 500         struct nfs4_stid *stid;
 501         int new_id;
 502
 503         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 504         if (!stid)
 505                 return NULL;
 506
 507         idr_preload(GFP_KERNEL);
 508         spin_lock(&cl->cl_lock);
 509         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 510         spin_unlock(&cl->cl_lock);
 511         idr_preload_end();
 512         if (new_id < 0)
 513                 goto out_free;
 514         stid->sc_client = cl;
 515         stid->sc_stateid.si_opaque.so_id = new_id;
 516         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 517         /* Will be incremented before return to client: */
 518         atomic_set(&stid->sc_count, 1);
 519
 520         /*
 521          * It shouldn't be a problem to reuse an opaque stateid value.
 522          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 523          * example, a stray write retransmission could be accepted by
 524          * the server when it should have been rejected.  Therefore,
 525          * adopt a trick from the sctp code to attempt to maximize the
 526          * amount of time until an id is reused, by ensuring they always
 527          * "increase" (mod INT_MAX):
 528          */
 529         return stid;
 530 out_free:
 531         kmem_cache_free(slab, stid);
 532         return NULL;
 533 }
 534
 535 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 536 {
 537         struct nfs4_stid *stid;
 538         struct nfs4_ol_stateid *stp;
 539
 540         stid = nfs4_alloc_stid(clp, stateid_slab);
 541         if (!stid)
 542                 return NULL;
 543
 544         stp = openlockstateid(stid);
 545         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 546         return stp;
 547 }
 548
 549 static void nfs4_free_deleg(struct nfs4_stid *stid)
 550 {
 551         kmem_cache_free(deleg_slab, stid);
 552         atomic_long_dec(&num_delegations);
 553 }
 554
 555 /*
 556  * When we recall a delegation, we should be careful not to hand it
 557  * out again straight away.
 558  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 559  * in which the filehandles of recalled delegations are "stored".
 560  * If a filehandle appear in either filter, a delegation is blocked.
 561  * When a delegation is recalled, the filehandle is stored in the "new"
 562  * filter.
 563  * Every 30 seconds we swap the filters and clear the "new" one,
 564  * unless both are empty of course.
 565  *
 566  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 567  * low 3 bytes as hash-table indices.
 568  *
 569  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 570  * is used to manage concurrent access.  Testing does not need the lock
 571  * except when swapping the two filters.
 572  */
 573 static DEFINE_SPINLOCK(blocked_delegations_lock);
 574 static struct bloom_pair {
 575         int     entries, old_entries;
 576         time_t  swap_time;
 577         int     new; /* index into 'set' */
 578         DECLARE_BITMAP(set[2], 256);
 579 } blocked_delegations;
 580
 581 static int delegation_blocked(struct knfsd_fh *fh)
 582 {
 583         u32 hash;
 584         struct bloom_pair *bd = &blocked_delegations;
 585
 586         if (bd->entries == 0)
 587                 return 0;
 588         if (seconds_since_boot() - bd->swap_time > 30) {
 589                 spin_lock(&blocked_delegations_lock);
 590                 if (seconds_since_boot() - bd->swap_time > 30) {
 591                         bd->entries -= bd->old_entries;
 592                         bd->old_entries = bd->entries;
 593                         memset(bd->set[bd->new], 0,
 594                                sizeof(bd->set[0]));
 595                         bd->new = 1-bd->new;
 596                         bd->swap_time = seconds_since_boot();
 597                 }
 598                 spin_unlock(&blocked_delegations_lock);
 599         }
 600         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 601         if (test_bit(hash&255, bd->set[0]) &&
 602             test_bit((hash>>8)&255, bd->set[0]) &&
 603             test_bit((hash>>16)&255, bd->set[0]))
 604                 return 1;
 605
 606         if (test_bit(hash&255, bd->set[1]) &&
 607             test_bit((hash>>8)&255, bd->set[1]) &&
 608             test_bit((hash>>16)&255, bd->set[1]))
 609                 return 1;
 610
 611         return 0;
 612 }
 613
 614 static void block_delegations(struct knfsd_fh *fh)
 615 {
 616         u32 hash;
 617         struct bloom_pair *bd = &blocked_delegations;
 618
 619         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 620
 621         spin_lock(&blocked_delegations_lock);
 622         __set_bit(hash&255, bd->set[bd->new]);
 623         __set_bit((hash>>8)&255, bd->set[bd->new]);
 624         __set_bit((hash>>16)&255, bd->set[bd->new]);
 625         if (bd->entries == 0)
 626                 bd->swap_time = seconds_since_boot();
 627         bd->entries += 1;
 628         spin_unlock(&blocked_delegations_lock);
 629 }
 630
 631 static struct nfs4_delegation *
 632 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh)
 633 {
 634         struct nfs4_delegation *dp;
 635         long n;
 636
 637         dprintk("NFSD alloc_init_deleg\n");
 638         n = atomic_long_inc_return(&num_delegations);
 639         if (n < 0 || n > max_delegations)
 640                 goto out_dec;
 641         if (delegation_blocked(&current_fh->fh_handle))
 642                 goto out_dec;
 643         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 644         if (dp == NULL)
 645                 goto out_dec;
 646
 647         dp->dl_stid.sc_free = nfs4_free_deleg;
 648         /*
 649          * delegation seqid's are never incremented.  The 4.1 special
 650          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 651          * 0 anyway just for consistency and use 1:
 652          */
 653         dp->dl_stid.sc_stateid.si_generation = 1;
 654         INIT_LIST_HEAD(&dp->dl_perfile);
 655         INIT_LIST_HEAD(&dp->dl_perclnt);
 656         INIT_LIST_HEAD(&dp->dl_recall_lru);
 657         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 658         dp->dl_retries = 1;
 659         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 660                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 661         return dp;
 662 out_dec:
 663         atomic_long_dec(&num_delegations);
 664         return NULL;
 665 }
 666
 667 void
 668 nfs4_put_stid(struct nfs4_stid *s)
 669 {
 670         struct nfs4_file *fp = s->sc_file;
 671         struct nfs4_client *clp = s->sc_client;
 672
 673         might_lock(&clp->cl_lock);
 674
 675         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 676                 wake_up_all(&close_wq);
 677                 return;
 678         }
 679         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 680         spin_unlock(&clp->cl_lock);
 681         s->sc_free(s);
 682         if (fp)
 683                 put_nfs4_file(fp);
 684 }
 685
 686 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 687 {
 688         struct file *filp = NULL;
 689
 690         spin_lock(&fp->fi_lock);
 691         if (fp->fi_deleg_file && atomic_dec_and_test(&fp->fi_delegees))
 692                 swap(filp, fp->fi_deleg_file);
 693         spin_unlock(&fp->fi_lock);
 694
 695         if (filp) {
 696                 vfs_setlease(filp, F_UNLCK, NULL, NULL);
 697                 fput(filp);
 698         }
 699 }
 700
 701 static void unhash_stid(struct nfs4_stid *s)
 702 {
 703         s->sc_type = 0;
 704 }
 705
 706 static void
 707 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 708 {
 709         lockdep_assert_held(&state_lock);
 710         lockdep_assert_held(&fp->fi_lock);
 711
 712         atomic_inc(&dp->dl_stid.sc_count);
 713         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 714         list_add(&dp->dl_perfile, &fp->fi_delegations);
 715         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
 716 }
 717
 718 static void
 719 unhash_delegation_locked(struct nfs4_delegation *dp)
 720 {
 721         struct nfs4_file *fp = dp->dl_stid.sc_file;
 722
 723         lockdep_assert_held(&state_lock);
 724
 725         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 726         /* Ensure that deleg break won't try to requeue it */
 727         ++dp->dl_time;
 728         spin_lock(&fp->fi_lock);
 729         list_del_init(&dp->dl_perclnt);
 730         list_del_init(&dp->dl_recall_lru);
 731         list_del_init(&dp->dl_perfile);
 732         spin_unlock(&fp->fi_lock);
 733 }
 734
 735 static void destroy_delegation(struct nfs4_delegation *dp)
 736 {
 737         spin_lock(&state_lock);
 738         unhash_delegation_locked(dp);
 739         spin_unlock(&state_lock);
 740         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 741         nfs4_put_stid(&dp->dl_stid);
 742 }
 743
 744 static void revoke_delegation(struct nfs4_delegation *dp)
 745 {
 746         struct nfs4_client *clp = dp->dl_stid.sc_client;
 747
 748         WARN_ON(!list_empty(&dp->dl_recall_lru));
 749
 750         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 751
 752         if (clp->cl_minorversion == 0)
 753                 nfs4_put_stid(&dp->dl_stid);
 754         else {
 755                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 756                 spin_lock(&clp->cl_lock);
 757                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 758                 spin_unlock(&clp->cl_lock);
 759         }
 760 }
 761
 762 /*
 763  * SETCLIENTID state
 764  */
 765
 766 static unsigned int clientid_hashval(u32 id)
 767 {
 768         return id & CLIENT_HASH_MASK;
 769 }
 770
 771 static unsigned int clientstr_hashval(const char *name)
 772 {
 773         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 774 }
 775
 776 /*
 777  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 778  * st_{access,deny}_bmap field of the stateid, in order to track not
 779  * only what share bits are currently in force, but also what
 780  * combinations of share bits previous opens have used.  This allows us
 781  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 782  * return an error if the client attempt to downgrade to a combination
 783  * of share bits not explicable by closing some of its previous opens.
 784  *
 785  * XXX: This enforcement is actually incomplete, since we don't keep
 786  * track of access/deny bit combinations; so, e.g., we allow:
 787  *
 788  *      OPEN allow read, deny write
 789  *      OPEN allow both, deny none
 790  *      DOWNGRADE allow read, deny none
 791  *
 792  * which we should reject.
 793  */
 794 static unsigned int
 795 bmap_to_share_mode(unsigned long bmap) {
 796         int i;
 797         unsigned int access = 0;
 798
 799         for (i = 1; i < 4; i++) {
 800                 if (test_bit(i, &bmap))
 801                         access |= i;
 802         }
 803         return access;
 804 }
 805
 806 /* set share access for a given stateid */
 807 static inline void
 808 set_access(u32 access, struct nfs4_ol_stateid *stp)
 809 {
 810         unsigned char mask = 1 << access;
 811
 812         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 813         stp->st_access_bmap |= mask;
 814 }
 815
 816 /* clear share access for a given stateid */
 817 static inline void
 818 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 819 {
 820         unsigned char mask = 1 << access;
 821
 822         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 823         stp->st_access_bmap &= ~mask;
 824 }
 825
 826 /* test whether a given stateid has access */
 827 static inline bool
 828 test_access(u32 access, struct nfs4_ol_stateid *stp)
 829 {
 830         unsigned char mask = 1 << access;
 831
 832         return (bool)(stp->st_access_bmap & mask);
 833 }
 834
 835 /* set share deny for a given stateid */
 836 static inline void
 837 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 838 {
 839         unsigned char mask = 1 << deny;
 840
 841         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 842         stp->st_deny_bmap |= mask;
 843 }
 844
 845 /* clear share deny for a given stateid */
 846 static inline void
 847 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 848 {
 849         unsigned char mask = 1 << deny;
 850
 851         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 852         stp->st_deny_bmap &= ~mask;
 853 }
 854
 855 /* test whether a given stateid is denying specific access */
 856 static inline bool
 857 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 858 {
 859         unsigned char mask = 1 << deny;
 860
 861         return (bool)(stp->st_deny_bmap & mask);
 862 }
 863
 864 static int nfs4_access_to_omode(u32 access)
 865 {
 866         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 867         case NFS4_SHARE_ACCESS_READ:
 868                 return O_RDONLY;
 869         case NFS4_SHARE_ACCESS_WRITE:
 870                 return O_WRONLY;
 871         case NFS4_SHARE_ACCESS_BOTH:
 872                 return O_RDWR;
 873         }
 874         WARN_ON_ONCE(1);
 875         return O_RDONLY;
 876 }
 877
 878 /*
 879  * A stateid that had a deny mode associated with it is being released
 880  * or downgraded. Recalculate the deny mode on the file.
 881  */
 882 static void
 883 recalculate_deny_mode(struct nfs4_file *fp)
 884 {
 885         struct nfs4_ol_stateid *stp;
 886
 887         spin_lock(&fp->fi_lock);
 888         fp->fi_share_deny = 0;
 889         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
 890                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
 891         spin_unlock(&fp->fi_lock);
 892 }
 893
 894 static void
 895 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
 896 {
 897         int i;
 898         bool change = false;
 899
 900         for (i = 1; i < 4; i++) {
 901                 if ((i & deny) != i) {
 902                         change = true;
 903                         clear_deny(i, stp);
 904                 }
 905         }
 906
 907         /* Recalculate per-file deny mode if there was a change */
 908         if (change)
 909                 recalculate_deny_mode(stp->st_stid.sc_file);
 910 }
 911
 912 /* release all access and file references for a given stateid */
 913 static void
 914 release_all_access(struct nfs4_ol_stateid *stp)
 915 {
 916         int i;
 917         struct nfs4_file *fp = stp->st_stid.sc_file;
 918
 919         if (fp && stp->st_deny_bmap != 0)
 920                 recalculate_deny_mode(fp);
 921
 922         for (i = 1; i < 4; i++) {
 923                 if (test_access(i, stp))
 924                         nfs4_file_put_access(stp->st_stid.sc_file, i);
 925                 clear_access(i, stp);
 926         }
 927 }
 928
 929 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
 930 {
 931         struct nfs4_client *clp = sop->so_client;
 932
 933         might_lock(&clp->cl_lock);
 934
 935         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
 936                 return;
 937         sop->so_ops->so_unhash(sop);
 938         spin_unlock(&clp->cl_lock);
 939         kfree(sop->so_owner.data);
 940         sop->so_ops->so_free(sop);
 941 }
 942
 943 static void unhash_ol_stateid(struct nfs4_ol_stateid *stp)
 944 {
 945         struct nfs4_file *fp = stp->st_stid.sc_file;
 946
 947         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
 948
 949         spin_lock(&fp->fi_lock);
 950         list_del(&stp->st_perfile);
 951         spin_unlock(&fp->fi_lock);
 952         list_del(&stp->st_perstateowner);
 953 }
 954
 955 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
 956 {
 957         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 958
 959         release_all_access(stp);
 960         if (stp->st_stateowner)
 961                 nfs4_put_stateowner(stp->st_stateowner);
 962         kmem_cache_free(stateid_slab, stid);
 963 }
 964
 965 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
 966 {
 967         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 968         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
 969         struct file *file;
 970
 971         file = find_any_file(stp->st_stid.sc_file);
 972         if (file)
 973                 filp_close(file, (fl_owner_t)lo);
 974         nfs4_free_ol_stateid(stid);
 975 }
 976
 977 /*
 978  * Put the persistent reference to an already unhashed generic stateid, while
 979  * holding the cl_lock. If it's the last reference, then put it onto the
 980  * reaplist for later destruction.
 981  */
 982 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
 983                                        struct list_head *reaplist)
 984 {
 985         struct nfs4_stid *s = &stp->st_stid;
 986         struct nfs4_client *clp = s->sc_client;
 987
 988         lockdep_assert_held(&clp->cl_lock);
 989
 990         WARN_ON_ONCE(!list_empty(&stp->st_locks));
 991
 992         if (!atomic_dec_and_test(&s->sc_count)) {
 993                 wake_up_all(&close_wq);
 994                 return;
 995         }
 996
 997         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 998         list_add(&stp->st_locks, reaplist);
 999 }
1000
1001 static void unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1002 {
1003         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1004
1005         lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1006
1007         list_del_init(&stp->st_locks);
1008         unhash_ol_stateid(stp);
1009         unhash_stid(&stp->st_stid);
1010 }
1011
1012 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1013 {
1014         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1015
1016         spin_lock(&oo->oo_owner.so_client->cl_lock);
1017         unhash_lock_stateid(stp);
1018         spin_unlock(&oo->oo_owner.so_client->cl_lock);
1019         nfs4_put_stid(&stp->st_stid);
1020 }
1021
1022 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1023 {
1024         struct nfs4_client *clp = lo->lo_owner.so_client;
1025
1026         lockdep_assert_held(&clp->cl_lock);
1027
1028         list_del_init(&lo->lo_owner.so_strhash);
1029 }
1030
1031 /*
1032  * Free a list of generic stateids that were collected earlier after being
1033  * fully unhashed.
1034  */
1035 static void
1036 free_ol_stateid_reaplist(struct list_head *reaplist)
1037 {
1038         struct nfs4_ol_stateid *stp;
1039         struct nfs4_file *fp;
1040
1041         might_sleep();
1042
1043         while (!list_empty(reaplist)) {
1044                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1045                                        st_locks);
1046                 list_del(&stp->st_locks);
1047                 fp = stp->st_stid.sc_file;
1048                 stp->st_stid.sc_free(&stp->st_stid);
1049                 if (fp)
1050                         put_nfs4_file(fp);
1051         }
1052 }
1053
1054 static void release_lockowner(struct nfs4_lockowner *lo)
1055 {
1056         struct nfs4_client *clp = lo->lo_owner.so_client;
1057         struct nfs4_ol_stateid *stp;
1058         struct list_head reaplist;
1059
1060         INIT_LIST_HEAD(&reaplist);
1061
1062         spin_lock(&clp->cl_lock);
1063         unhash_lockowner_locked(lo);
1064         while (!list_empty(&lo->lo_owner.so_stateids)) {
1065                 stp = list_first_entry(&lo->lo_owner.so_stateids,
1066                                 struct nfs4_ol_stateid, st_perstateowner);
1067                 unhash_lock_stateid(stp);
1068                 put_ol_stateid_locked(stp, &reaplist);
1069         }
1070         spin_unlock(&clp->cl_lock);
1071         free_ol_stateid_reaplist(&reaplist);
1072         nfs4_put_stateowner(&lo->lo_owner);
1073 }
1074
1075 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1076                                        struct list_head *reaplist)
1077 {
1078         struct nfs4_ol_stateid *stp;
1079
1080         while (!list_empty(&open_stp->st_locks)) {
1081                 stp = list_entry(open_stp->st_locks.next,
1082                                 struct nfs4_ol_stateid, st_locks);
1083                 unhash_lock_stateid(stp);
1084                 put_ol_stateid_locked(stp, reaplist);
1085         }
1086 }
1087
1088 static void unhash_open_stateid(struct nfs4_ol_stateid *stp,
1089                                 struct list_head *reaplist)
1090 {
1091         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1092
1093         unhash_ol_stateid(stp);
1094         release_open_stateid_locks(stp, reaplist);
1095 }
1096
1097 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1098 {
1099         LIST_HEAD(reaplist);
1100
1101         spin_lock(&stp->st_stid.sc_client->cl_lock);
1102         unhash_open_stateid(stp, &reaplist);
1103         put_ol_stateid_locked(stp, &reaplist);
1104         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1105         free_ol_stateid_reaplist(&reaplist);
1106 }
1107
1108 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1109 {
1110         struct nfs4_client *clp = oo->oo_owner.so_client;
1111
1112         lockdep_assert_held(&clp->cl_lock);
1113
1114         list_del_init(&oo->oo_owner.so_strhash);
1115         list_del_init(&oo->oo_perclient);
1116 }
1117
1118 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1119 {
1120         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1121                                           nfsd_net_id);
1122         struct nfs4_ol_stateid *s;
1123
1124         spin_lock(&nn->client_lock);
1125         s = oo->oo_last_closed_stid;
1126         if (s) {
1127                 list_del_init(&oo->oo_close_lru);
1128                 oo->oo_last_closed_stid = NULL;
1129         }
1130         spin_unlock(&nn->client_lock);
1131         if (s)
1132                 nfs4_put_stid(&s->st_stid);
1133 }
1134
1135 static void release_openowner(struct nfs4_openowner *oo)
1136 {
1137         struct nfs4_ol_stateid *stp;
1138         struct nfs4_client *clp = oo->oo_owner.so_client;
1139         struct list_head reaplist;
1140
1141         INIT_LIST_HEAD(&reaplist);
1142
1143         spin_lock(&clp->cl_lock);
1144         unhash_openowner_locked(oo);
1145         while (!list_empty(&oo->oo_owner.so_stateids)) {
1146                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1147                                 struct nfs4_ol_stateid, st_perstateowner);
1148                 unhash_open_stateid(stp, &reaplist);
1149                 put_ol_stateid_locked(stp, &reaplist);
1150         }
1151         spin_unlock(&clp->cl_lock);
1152         free_ol_stateid_reaplist(&reaplist);
1153         release_last_closed_stateid(oo);
1154         nfs4_put_stateowner(&oo->oo_owner);
1155 }
1156
1157 static inline int
1158 hash_sessionid(struct nfs4_sessionid *sessionid)
1159 {
1160         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1161
1162         return sid->sequence % SESSION_HASH_SIZE;
1163 }
1164
1165 #ifdef NFSD_DEBUG
1166 static inline void
1167 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1168 {
1169         u32 *ptr = (u32 *)(&sessionid->data[0]);
1170         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1171 }
1172 #else
1173 static inline void
1174 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1175 {
1176 }
1177 #endif
1178
1179 /*
1180  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1181  * won't be used for replay.
1182  */
1183 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1184 {
1185         struct nfs4_stateowner *so = cstate->replay_owner;
1186
1187         if (nfserr == nfserr_replay_me)
1188                 return;
1189
1190         if (!seqid_mutating_err(ntohl(nfserr))) {
1191                 nfsd4_cstate_clear_replay(cstate);
1192                 return;
1193         }
1194         if (!so)
1195                 return;
1196         if (so->so_is_open_owner)
1197                 release_last_closed_stateid(openowner(so));
1198         so->so_seqid++;
1199         return;
1200 }
1201
1202 static void
1203 gen_sessionid(struct nfsd4_session *ses)
1204 {
1205         struct nfs4_client *clp = ses->se_client;
1206         struct nfsd4_sessionid *sid;
1207
1208         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1209         sid->clientid = clp->cl_clientid;
1210         sid->sequence = current_sessionid++;
1211         sid->reserved = 0;
1212 }
1213
1214 /*
1215  * The protocol defines ca_maxresponssize_cached to include the size of
1216  * the rpc header, but all we need to cache is the data starting after
1217  * the end of the initial SEQUENCE operation--the rest we regenerate
1218  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1219  * value that is the number of bytes in our cache plus a few additional
1220  * bytes.  In order to stay on the safe side, and not promise more than
1221  * we can cache, those additional bytes must be the minimum possible: 24
1222  * bytes of rpc header (xid through accept state, with AUTH_NULL
1223  * verifier), 12 for the compound header (with zero-length tag), and 44
1224  * for the SEQUENCE op response:
1225  */
1226 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1227
1228 static void
1229 free_session_slots(struct nfsd4_session *ses)
1230 {
1231         int i;
1232
1233         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1234                 kfree(ses->se_slots[i]);
1235 }
1236
1237 /*
1238  * We don't actually need to cache the rpc and session headers, so we
1239  * can allocate a little less for each slot:
1240  */
1241 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1242 {
1243         u32 size;
1244
1245         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1246                 size = 0;
1247         else
1248                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1249         return size + sizeof(struct nfsd4_slot);
1250 }
1251
1252 /*
1253  * XXX: If we run out of reserved DRC memory we could (up to a point)
1254  * re-negotiate active sessions and reduce their slot usage to make
1255  * room for new connections. For now we just fail the create session.
1256  */
1257 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1258 {
1259         u32 slotsize = slot_bytes(ca);
1260         u32 num = ca->maxreqs;
1261         int avail;
1262
1263         spin_lock(&nfsd_drc_lock);
1264         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1265                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1266         num = min_t(int, num, avail / slotsize);
1267         nfsd_drc_mem_used += num * slotsize;
1268         spin_unlock(&nfsd_drc_lock);
1269
1270         return num;
1271 }
1272
1273 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1274 {
1275         int slotsize = slot_bytes(ca);
1276
1277         spin_lock(&nfsd_drc_lock);
1278         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1279         spin_unlock(&nfsd_drc_lock);
1280 }
1281
1282 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1283                                            struct nfsd4_channel_attrs *battrs)
1284 {
1285         int numslots = fattrs->maxreqs;
1286         int slotsize = slot_bytes(fattrs);
1287         struct nfsd4_session *new;
1288         int mem, i;
1289
1290         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1291                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1292         mem = numslots * sizeof(struct nfsd4_slot *);
1293
1294         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1295         if (!new)
1296                 return NULL;
1297         /* allocate each struct nfsd4_slot and data cache in one piece */
1298         for (i = 0; i < numslots; i++) {
1299                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1300                 if (!new->se_slots[i])
1301                         goto out_free;
1302         }
1303
1304         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1305         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1306
1307         return new;
1308 out_free:
1309         while (i--)
1310                 kfree(new->se_slots[i]);
1311         kfree(new);
1312         return NULL;
1313 }
1314
1315 static void free_conn(struct nfsd4_conn *c)
1316 {
1317         svc_xprt_put(c->cn_xprt);
1318         kfree(c);
1319 }
1320
1321 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1322 {
1323         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1324         struct nfs4_client *clp = c->cn_session->se_client;
1325
1326         spin_lock(&clp->cl_lock);
1327         if (!list_empty(&c->cn_persession)) {
1328                 list_del(&c->cn_persession);
1329                 free_conn(c);
1330         }
1331         nfsd4_probe_callback(clp);
1332         spin_unlock(&clp->cl_lock);
1333 }
1334
1335 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1336 {
1337         struct nfsd4_conn *conn;
1338
1339         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1340         if (!conn)
1341                 return NULL;
1342         svc_xprt_get(rqstp->rq_xprt);
1343         conn->cn_xprt = rqstp->rq_xprt;
1344         conn->cn_flags = flags;
1345         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1346         return conn;
1347 }
1348
1349 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1350 {
1351         conn->cn_session = ses;
1352         list_add(&conn->cn_persession, &ses->se_conns);
1353 }
1354
1355 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1356 {
1357         struct nfs4_client *clp = ses->se_client;
1358
1359         spin_lock(&clp->cl_lock);
1360         __nfsd4_hash_conn(conn, ses);
1361         spin_unlock(&clp->cl_lock);
1362 }
1363
1364 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1365 {
1366         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1367         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1368 }
1369
1370 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1371 {
1372         int ret;
1373
1374         nfsd4_hash_conn(conn, ses);
1375         ret = nfsd4_register_conn(conn);
1376         if (ret)
1377                 /* oops; xprt is already down: */
1378                 nfsd4_conn_lost(&conn->cn_xpt_user);
1379         /* We may have gained or lost a callback channel: */
1380         nfsd4_probe_callback_sync(ses->se_client);
1381 }
1382
1383 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1384 {
1385         u32 dir = NFS4_CDFC4_FORE;
1386
1387         if (cses->flags & SESSION4_BACK_CHAN)
1388                 dir |= NFS4_CDFC4_BACK;
1389         return alloc_conn(rqstp, dir);
1390 }
1391
1392 /* must be called under client_lock */
1393 static void nfsd4_del_conns(struct nfsd4_session *s)
1394 {
1395         struct nfs4_client *clp = s->se_client;
1396         struct nfsd4_conn *c;
1397
1398         spin_lock(&clp->cl_lock);
1399         while (!list_empty(&s->se_conns)) {
1400                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1401                 list_del_init(&c->cn_persession);
1402                 spin_unlock(&clp->cl_lock);
1403
1404                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1405                 free_conn(c);
1406
1407                 spin_lock(&clp->cl_lock);
1408         }
1409         spin_unlock(&clp->cl_lock);
1410 }
1411
1412 static void __free_session(struct nfsd4_session *ses)
1413 {
1414         free_session_slots(ses);
1415         kfree(ses);
1416 }
1417
1418 static void free_session(struct nfsd4_session *ses)
1419 {
1420         nfsd4_del_conns(ses);
1421         nfsd4_put_drc_mem(&ses->se_fchannel);
1422         __free_session(ses);
1423 }
1424
1425 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1426 {
1427         int idx;
1428         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1429
1430         new->se_client = clp;
1431         gen_sessionid(new);
1432
1433         INIT_LIST_HEAD(&new->se_conns);
1434
1435         new->se_cb_seq_nr = 1;
1436         new->se_flags = cses->flags;
1437         new->se_cb_prog = cses->callback_prog;
1438         new->se_cb_sec = cses->cb_sec;
1439         atomic_set(&new->se_ref, 0);
1440         idx = hash_sessionid(&new->se_sessionid);
1441         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1442         spin_lock(&clp->cl_lock);
1443         list_add(&new->se_perclnt, &clp->cl_sessions);
1444         spin_unlock(&clp->cl_lock);
1445
1446         {
1447                 struct sockaddr *sa = svc_addr(rqstp);
1448                 /*
1449                  * This is a little silly; with sessions there's no real
1450                  * use for the callback address.  Use the peer address
1451                  * as a reasonable default for now, but consider fixing
1452                  * the rpc client not to require an address in the
1453                  * future:
1454                  */
1455                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1456                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1457         }
1458 }
1459
1460 /* caller must hold client_lock */
1461 static struct nfsd4_session *
1462 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1463 {
1464         struct nfsd4_session *elem;
1465         int idx;
1466         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1467
1468         lockdep_assert_held(&nn->client_lock);
1469
1470         dump_sessionid(__func__, sessionid);
1471         idx = hash_sessionid(sessionid);
1472         /* Search in the appropriate list */
1473         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1474                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1475                             NFS4_MAX_SESSIONID_LEN)) {
1476                         return elem;
1477                 }
1478         }
1479
1480         dprintk("%s: session not found\n", __func__);
1481         return NULL;
1482 }
1483
1484 static struct nfsd4_session *
1485 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1486                 __be32 *ret)
1487 {
1488         struct nfsd4_session *session;
1489         __be32 status = nfserr_badsession;
1490
1491         session = __find_in_sessionid_hashtbl(sessionid, net);
1492         if (!session)
1493                 goto out;
1494         status = nfsd4_get_session_locked(session);
1495         if (status)
1496                 session = NULL;
1497 out:
1498         *ret = status;
1499         return session;
1500 }
1501
1502 /* caller must hold client_lock */
1503 static void
1504 unhash_session(struct nfsd4_session *ses)
1505 {
1506         struct nfs4_client *clp = ses->se_client;
1507         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1508
1509         lockdep_assert_held(&nn->client_lock);
1510
1511         list_del(&ses->se_hash);
1512         spin_lock(&ses->se_client->cl_lock);
1513         list_del(&ses->se_perclnt);
1514         spin_unlock(&ses->se_client->cl_lock);
1515 }
1516
1517 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1518 static int
1519 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1520 {
1521         if (clid->cl_boot == nn->boot_time)
1522                 return 0;
1523         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1524                 clid->cl_boot, clid->cl_id, nn->boot_time);
1525         return 1;
1526 }
1527
1528 /*
1529  * XXX Should we use a slab cache ?
1530  * This type of memory management is somewhat inefficient, but we use it
1531  * anyway since SETCLIENTID is not a common operation.
1532  */
1533 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1534 {
1535         struct nfs4_client *clp;
1536         int i;
1537
1538         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1539         if (clp == NULL)
1540                 return NULL;
1541         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1542         if (clp->cl_name.data == NULL)
1543                 goto err_no_name;
1544         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1545                         OWNER_HASH_SIZE, GFP_KERNEL);
1546         if (!clp->cl_ownerstr_hashtbl)
1547                 goto err_no_hashtbl;
1548         for (i = 0; i < OWNER_HASH_SIZE; i++)
1549                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1550         clp->cl_name.len = name.len;
1551         INIT_LIST_HEAD(&clp->cl_sessions);
1552         idr_init(&clp->cl_stateids);
1553         atomic_set(&clp->cl_refcount, 0);
1554         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1555         INIT_LIST_HEAD(&clp->cl_idhash);
1556         INIT_LIST_HEAD(&clp->cl_openowners);
1557         INIT_LIST_HEAD(&clp->cl_delegations);
1558         INIT_LIST_HEAD(&clp->cl_lru);
1559         INIT_LIST_HEAD(&clp->cl_callbacks);
1560         INIT_LIST_HEAD(&clp->cl_revoked);
1561         spin_lock_init(&clp->cl_lock);
1562         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1563         return clp;
1564 err_no_hashtbl:
1565         kfree(clp->cl_name.data);
1566 err_no_name:
1567         kfree(clp);
1568         return NULL;
1569 }
1570
1571 static void
1572 free_client(struct nfs4_client *clp)
1573 {
1574         while (!list_empty(&clp->cl_sessions)) {
1575                 struct nfsd4_session *ses;
1576                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1577                                 se_perclnt);
1578                 list_del(&ses->se_perclnt);
1579                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1580                 free_session(ses);
1581         }
1582         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1583         free_svc_cred(&clp->cl_cred);
1584         kfree(clp->cl_ownerstr_hashtbl);
1585         kfree(clp->cl_name.data);
1586         idr_destroy(&clp->cl_stateids);
1587         kfree(clp);
1588 }
1589
1590 /* must be called under the client_lock */
1591 static void
1592 unhash_client_locked(struct nfs4_client *clp)
1593 {
1594         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1595         struct nfsd4_session *ses;
1596
1597         lockdep_assert_held(&nn->client_lock);
1598
1599         /* Mark the client as expired! */
1600         clp->cl_time = 0;
1601         /* Make it invisible */
1602         if (!list_empty(&clp->cl_idhash)) {
1603                 list_del_init(&clp->cl_idhash);
1604                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1605                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1606                 else
1607                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1608         }
1609         list_del_init(&clp->cl_lru);
1610         spin_lock(&clp->cl_lock);
1611         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1612                 list_del_init(&ses->se_hash);
1613         spin_unlock(&clp->cl_lock);
1614 }
1615
1616 static void
1617 unhash_client(struct nfs4_client *clp)
1618 {
1619         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1620
1621         spin_lock(&nn->client_lock);
1622         unhash_client_locked(clp);
1623         spin_unlock(&nn->client_lock);
1624 }
1625
1626 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1627 {
1628         if (atomic_read(&clp->cl_refcount))
1629                 return nfserr_jukebox;
1630         unhash_client_locked(clp);
1631         return nfs_ok;
1632 }
1633
1634 static void
1635 __destroy_client(struct nfs4_client *clp)
1636 {
1637         struct nfs4_openowner *oo;
1638         struct nfs4_delegation *dp;
1639         struct list_head reaplist;
1640
1641         INIT_LIST_HEAD(&reaplist);
1642         spin_lock(&state_lock);
1643         while (!list_empty(&clp->cl_delegations)) {
1644                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1645                 unhash_delegation_locked(dp);
1646                 list_add(&dp->dl_recall_lru, &reaplist);
1647         }
1648         spin_unlock(&state_lock);
1649         while (!list_empty(&reaplist)) {
1650                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1651                 list_del_init(&dp->dl_recall_lru);
1652                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1653                 nfs4_put_stid(&dp->dl_stid);
1654         }
1655         while (!list_empty(&clp->cl_revoked)) {
1656                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1657                 list_del_init(&dp->dl_recall_lru);
1658                 nfs4_put_stid(&dp->dl_stid);
1659         }
1660         while (!list_empty(&clp->cl_openowners)) {
1661                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1662                 nfs4_get_stateowner(&oo->oo_owner);
1663                 release_openowner(oo);
1664         }
1665         nfsd4_shutdown_callback(clp);
1666         if (clp->cl_cb_conn.cb_xprt)
1667                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1668         free_client(clp);
1669 }
1670
1671 static void
1672 destroy_client(struct nfs4_client *clp)
1673 {
1674         unhash_client(clp);
1675         __destroy_client(clp);
1676 }
1677
1678 static void expire_client(struct nfs4_client *clp)
1679 {
1680         unhash_client(clp);
1681         nfsd4_client_record_remove(clp);
1682         __destroy_client(clp);
1683 }
1684
1685 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1686 {
1687         memcpy(target->cl_verifier.data, source->data,
1688                         sizeof(target->cl_verifier.data));
1689 }
1690
1691 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1692 {
1693         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1694         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1695 }
1696
1697 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1698 {
1699         if (source->cr_principal) {
1700                 target->cr_principal =
1701                                 kstrdup(source->cr_principal, GFP_KERNEL);
1702                 if (target->cr_principal == NULL)
1703                         return -ENOMEM;
1704         } else
1705                 target->cr_principal = NULL;
1706         target->cr_flavor = source->cr_flavor;
1707         target->cr_uid = source->cr_uid;
1708         target->cr_gid = source->cr_gid;
1709         target->cr_group_info = source->cr_group_info;
1710         get_group_info(target->cr_group_info);
1711         target->cr_gss_mech = source->cr_gss_mech;
1712         if (source->cr_gss_mech)
1713                 gss_mech_get(source->cr_gss_mech);
1714         return 0;
1715 }
1716
1717 static int
1718 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1719 {
1720         if (o1->len < o2->len)
1721                 return -1;
1722         if (o1->len > o2->len)
1723                 return 1;
1724         return memcmp(o1->data, o2->data, o1->len);
1725 }
1726
1727 static int same_name(const char *n1, const char *n2)
1728 {
1729         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1730 }
1731
1732 static int
1733 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1734 {
1735         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1736 }
1737
1738 static int
1739 same_clid(clientid_t *cl1, clientid_t *cl2)
1740 {
1741         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1742 }
1743
1744 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1745 {
1746         int i;
1747
1748         if (g1->ngroups != g2->ngroups)
1749                 return false;
1750         for (i=0; i<g1->ngroups; i++)
1751                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1752                         return false;
1753         return true;
1754 }
1755
1756 /*
1757  * RFC 3530 language requires clid_inuse be returned when the
1758  * "principal" associated with a requests differs from that previously
1759  * used.  We use uid, gid's, and gss principal string as our best
1760  * approximation.  We also don't want to allow non-gss use of a client
1761  * established using gss: in theory cr_principal should catch that
1762  * change, but in practice cr_principal can be null even in the gss case
1763  * since gssd doesn't always pass down a principal string.
1764  */
1765 static bool is_gss_cred(struct svc_cred *cr)
1766 {
1767         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1768         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1769 }
1770
1771
1772 static bool
1773 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1774 {
1775         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1776                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1777                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1778                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1779                 return false;
1780         if (cr1->cr_principal == cr2->cr_principal)
1781                 return true;
1782         if (!cr1->cr_principal || !cr2->cr_principal)
1783                 return false;
1784         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1785 }
1786
1787 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1788 {
1789         struct svc_cred *cr = &rqstp->rq_cred;
1790         u32 service;
1791
1792         if (!cr->cr_gss_mech)
1793                 return false;
1794         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1795         return service == RPC_GSS_SVC_INTEGRITY ||
1796                service == RPC_GSS_SVC_PRIVACY;
1797 }
1798
1799 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1800 {
1801         struct svc_cred *cr = &rqstp->rq_cred;
1802
1803         if (!cl->cl_mach_cred)
1804                 return true;
1805         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1806                 return false;
1807         if (!svc_rqst_integrity_protected(rqstp))
1808                 return false;
1809         if (!cr->cr_principal)
1810                 return false;
1811         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1812 }
1813
1814 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1815 {
1816         __be32 verf[2];
1817
1818         /*
1819          * This is opaque to client, so no need to byte-swap. Use
1820          * __force to keep sparse happy
1821          */
1822         verf[0] = (__force __be32)get_seconds();
1823         verf[1] = (__force __be32)nn->clientid_counter;
1824         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1825 }
1826
1827 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1828 {
1829         clp->cl_clientid.cl_boot = nn->boot_time;
1830         clp->cl_clientid.cl_id = nn->clientid_counter++;
1831         gen_confirm(clp, nn);
1832 }
1833
1834 static struct nfs4_stid *
1835 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
1836 {
1837         struct nfs4_stid *ret;
1838
1839         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1840         if (!ret || !ret->sc_type)
1841                 return NULL;
1842         return ret;
1843 }
1844
1845 static struct nfs4_stid *
1846 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1847 {
1848         struct nfs4_stid *s;
1849
1850         spin_lock(&cl->cl_lock);
1851         s = find_stateid_locked(cl, t);
1852         if (s != NULL) {
1853                 if (typemask & s->sc_type)
1854                         atomic_inc(&s->sc_count);
1855                 else
1856                         s = NULL;
1857         }
1858         spin_unlock(&cl->cl_lock);
1859         return s;
1860 }
1861
1862 static struct nfs4_client *create_client(struct xdr_netobj name,
1863                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1864 {
1865         struct nfs4_client *clp;
1866         struct sockaddr *sa = svc_addr(rqstp);
1867         int ret;
1868         struct net *net = SVC_NET(rqstp);
1869
1870         clp = alloc_client(name);
1871         if (clp == NULL)
1872                 return NULL;
1873
1874         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1875         if (ret) {
1876                 free_client(clp);
1877                 return NULL;
1878         }
1879         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
1880         clp->cl_time = get_seconds();
1881         clear_bit(0, &clp->cl_cb_slot_busy);
1882         copy_verf(clp, verf);
1883         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1884         clp->cl_cb_session = NULL;
1885         clp->net = net;
1886         return clp;
1887 }
1888
1889 static void
1890 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1891 {
1892         struct rb_node **new = &(root->rb_node), *parent = NULL;
1893         struct nfs4_client *clp;
1894
1895         while (*new) {
1896                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1897                 parent = *new;
1898
1899                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1900                         new = &((*new)->rb_left);
1901                 else
1902                         new = &((*new)->rb_right);
1903         }
1904
1905         rb_link_node(&new_clp->cl_namenode, parent, new);
1906         rb_insert_color(&new_clp->cl_namenode, root);
1907 }
1908
1909 static struct nfs4_client *
1910 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1911 {
1912         int cmp;
1913         struct rb_node *node = root->rb_node;
1914         struct nfs4_client *clp;
1915
1916         while (node) {
1917                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1918                 cmp = compare_blob(&clp->cl_name, name);
1919                 if (cmp > 0)
1920                         node = node->rb_left;
1921                 else if (cmp < 0)
1922                         node = node->rb_right;
1923                 else
1924                         return clp;
1925         }
1926         return NULL;
1927 }
1928
1929 static void
1930 add_to_unconfirmed(struct nfs4_client *clp)
1931 {
1932         unsigned int idhashval;
1933         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1934
1935         lockdep_assert_held(&nn->client_lock);
1936
1937         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1938         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1939         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1940         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1941         renew_client_locked(clp);
1942 }
1943
1944 static void
1945 move_to_confirmed(struct nfs4_client *clp)
1946 {
1947         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1948         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1949
1950         lockdep_assert_held(&nn->client_lock);
1951
1952         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1953         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1954         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1955         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1956         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1957         renew_client_locked(clp);
1958 }
1959
1960 static struct nfs4_client *
1961 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1962 {
1963         struct nfs4_client *clp;
1964         unsigned int idhashval = clientid_hashval(clid->cl_id);
1965
1966         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1967                 if (same_clid(&clp->cl_clientid, clid)) {
1968                         if ((bool)clp->cl_minorversion != sessions)
1969                                 return NULL;
1970                         renew_client_locked(clp);
1971                         return clp;
1972                 }
1973         }
1974         return NULL;
1975 }
1976
1977 static struct nfs4_client *
1978 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1979 {
1980         struct list_head *tbl = nn->conf_id_hashtbl;
1981
1982         lockdep_assert_held(&nn->client_lock);
1983         return find_client_in_id_table(tbl, clid, sessions);
1984 }
1985
1986 static struct nfs4_client *
1987 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1988 {
1989         struct list_head *tbl = nn->unconf_id_hashtbl;
1990
1991         lockdep_assert_held(&nn->client_lock);
1992         return find_client_in_id_table(tbl, clid, sessions);
1993 }
1994
1995 static bool clp_used_exchangeid(struct nfs4_client *clp)
1996 {
1997         return clp->cl_exchange_flags != 0;
1998 }
1999
2000 static struct nfs4_client *
2001 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2002 {
2003         lockdep_assert_held(&nn->client_lock);
2004         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2005 }
2006
2007 static struct nfs4_client *
2008 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2009 {
2010         lockdep_assert_held(&nn->client_lock);
2011         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2012 }
2013
2014 static void
2015 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2016 {
2017         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2018         struct sockaddr *sa = svc_addr(rqstp);
2019         u32 scopeid = rpc_get_scope_id(sa);
2020         unsigned short expected_family;
2021
2022         /* Currently, we only support tcp and tcp6 for the callback channel */
2023         if (se->se_callback_netid_len == 3 &&
2024             !memcmp(se->se_callback_netid_val, "tcp", 3))
2025                 expected_family = AF_INET;
2026         else if (se->se_callback_netid_len == 4 &&
2027                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2028                 expected_family = AF_INET6;
2029         else
2030                 goto out_err;
2031
2032         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2033                                             se->se_callback_addr_len,
2034                                             (struct sockaddr *)&conn->cb_addr,
2035                                             sizeof(conn->cb_addr));
2036
2037         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2038                 goto out_err;
2039
2040         if (conn->cb_addr.ss_family == AF_INET6)
2041                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2042
2043         conn->cb_prog = se->se_callback_prog;
2044         conn->cb_ident = se->se_callback_ident;
2045         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2046         return;
2047 out_err:
2048         conn->cb_addr.ss_family = AF_UNSPEC;
2049         conn->cb_addrlen = 0;
2050         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2051                 "will not receive delegations\n",
2052                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2053
2054         return;
2055 }
2056
2057 /*
2058  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2059  */
2060 static void
2061 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2062 {
2063         struct xdr_buf *buf = resp->xdr.buf;
2064         struct nfsd4_slot *slot = resp->cstate.slot;
2065         unsigned int base;
2066
2067         dprintk("--> %s slot %p\n", __func__, slot);
2068
2069         slot->sl_opcnt = resp->opcnt;
2070         slot->sl_status = resp->cstate.status;
2071
2072         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2073         if (nfsd4_not_cached(resp)) {
2074                 slot->sl_datalen = 0;
2075                 return;
2076         }
2077         base = resp->cstate.data_offset;
2078         slot->sl_datalen = buf->len - base;
2079         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2080                 WARN("%s: sessions DRC could not cache compound\n", __func__);
2081         return;
2082 }
2083
2084 /*
2085  * Encode the replay sequence operation from the slot values.
2086  * If cachethis is FALSE encode the uncached rep error on the next
2087  * operation which sets resp->p and increments resp->opcnt for
2088  * nfs4svc_encode_compoundres.
2089  *
2090  */
2091 static __be32
2092 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2093                           struct nfsd4_compoundres *resp)
2094 {
2095         struct nfsd4_op *op;
2096         struct nfsd4_slot *slot = resp->cstate.slot;
2097
2098         /* Encode the replayed sequence operation */
2099         op = &args->ops[resp->opcnt - 1];
2100         nfsd4_encode_operation(resp, op);
2101
2102         /* Return nfserr_retry_uncached_rep in next operation. */
2103         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2104                 op = &args->ops[resp->opcnt++];
2105                 op->status = nfserr_retry_uncached_rep;
2106                 nfsd4_encode_operation(resp, op);
2107         }
2108         return op->status;
2109 }
2110
2111 /*
2112  * The sequence operation is not cached because we can use the slot and
2113  * session values.
2114  */
2115 static __be32
2116 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2117                          struct nfsd4_sequence *seq)
2118 {
2119         struct nfsd4_slot *slot = resp->cstate.slot;
2120         struct xdr_stream *xdr = &resp->xdr;
2121         __be32 *p;
2122         __be32 status;
2123
2124         dprintk("--> %s slot %p\n", __func__, slot);
2125
2126         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2127         if (status)
2128                 return status;
2129
2130         p = xdr_reserve_space(xdr, slot->sl_datalen);
2131         if (!p) {
2132                 WARN_ON_ONCE(1);
2133                 return nfserr_serverfault;
2134         }
2135         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2136         xdr_commit_encode(xdr);
2137
2138         resp->opcnt = slot->sl_opcnt;
2139         return slot->sl_status;
2140 }
2141
2142 /*
2143  * Set the exchange_id flags returned by the server.
2144  */
2145 static void
2146 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2147 {
2148         /* pNFS is not supported */
2149         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2150
2151         /* Referrals are supported, Migration is not. */
2152         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2153
2154         /* set the wire flags to return to client. */
2155         clid->flags = new->cl_exchange_flags;
2156 }
2157
2158 static bool client_has_state(struct nfs4_client *clp)
2159 {
2160         /*
2161          * Note clp->cl_openowners check isn't quite right: there's no
2162          * need to count owners without stateid's.
2163          *
2164          * Also note we should probably be using this in 4.0 case too.
2165          */
2166         return !list_empty(&clp->cl_openowners)
2167                 || !list_empty(&clp->cl_delegations)
2168                 || !list_empty(&clp->cl_sessions);
2169 }
2170
2171 __be32
2172 nfsd4_exchange_id(struct svc_rqst *rqstp,
2173                   struct nfsd4_compound_state *cstate,
2174                   struct nfsd4_exchange_id *exid)
2175 {
2176         struct nfs4_client *conf, *new;
2177         struct nfs4_client *unconf = NULL;
2178         __be32 status;
2179         char                    addr_str[INET6_ADDRSTRLEN];
2180         nfs4_verifier           verf = exid->verifier;
2181         struct sockaddr         *sa = svc_addr(rqstp);
2182         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2183         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2184
2185         rpc_ntop(sa, addr_str, sizeof(addr_str));
2186         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2187                 "ip_addr=%s flags %x, spa_how %d\n",
2188                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2189                 addr_str, exid->flags, exid->spa_how);
2190
2191         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2192                 return nfserr_inval;
2193
2194         switch (exid->spa_how) {
2195         case SP4_MACH_CRED:
2196                 if (!svc_rqst_integrity_protected(rqstp))
2197                         return nfserr_inval;
2198         case SP4_NONE:
2199                 break;
2200         default:                                /* checked by xdr code */
2201                 WARN_ON_ONCE(1);
2202         case SP4_SSV:
2203                 return nfserr_encr_alg_unsupp;
2204         }
2205
2206         new = create_client(exid->clname, rqstp, &verf);
2207         if (new == NULL)
2208                 return nfserr_jukebox;
2209
2210         /* Cases below refer to rfc 5661 section 18.35.4: */
2211         spin_lock(&nn->client_lock);
2212         conf = find_confirmed_client_by_name(&exid->clname, nn);
2213         if (conf) {
2214                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2215                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2216
2217                 if (update) {
2218                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2219                                 status = nfserr_inval;
2220                                 goto out;
2221                         }
2222                         if (!mach_creds_match(conf, rqstp)) {
2223                                 status = nfserr_wrong_cred;
2224                                 goto out;
2225                         }
2226                         if (!creds_match) { /* case 9 */
2227                                 status = nfserr_perm;
2228                                 goto out;
2229                         }
2230                         if (!verfs_match) { /* case 8 */
2231                                 status = nfserr_not_same;
2232                                 goto out;
2233                         }
2234                         /* case 6 */
2235                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2236                         goto out_copy;
2237                 }
2238                 if (!creds_match) { /* case 3 */
2239                         if (client_has_state(conf)) {
2240                                 status = nfserr_clid_inuse;
2241                                 goto out;
2242                         }
2243                         goto out_new;
2244                 }
2245                 if (verfs_match) { /* case 2 */
2246                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2247                         goto out_copy;
2248                 }
2249                 /* case 5, client reboot */
2250                 conf = NULL;
2251                 goto out_new;
2252         }
2253
2254         if (update) { /* case 7 */
2255                 status = nfserr_noent;
2256                 goto out;
2257         }
2258
2259         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2260         if (unconf) /* case 4, possible retry or client restart */
2261                 unhash_client_locked(unconf);
2262
2263         /* case 1 (normal case) */
2264 out_new:
2265         if (conf) {
2266                 status = mark_client_expired_locked(conf);
2267                 if (status)
2268                         goto out;
2269         }
2270         new->cl_minorversion = cstate->minorversion;
2271         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
2272
2273         gen_clid(new, nn);
2274         add_to_unconfirmed(new);
2275         swap(new, conf);
2276 out_copy:
2277         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2278         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2279
2280         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2281         nfsd4_set_ex_flags(conf, exid);
2282
2283         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2284                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2285         status = nfs_ok;
2286
2287 out:
2288         spin_unlock(&nn->client_lock);
2289         if (new)
2290                 expire_client(new);
2291         if (unconf)
2292                 expire_client(unconf);
2293         return status;
2294 }
2295
2296 static __be32
2297 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2298 {
2299         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2300                 slot_seqid);
2301
2302         /* The slot is in use, and no response has been sent. */
2303         if (slot_inuse) {
2304                 if (seqid == slot_seqid)
2305                         return nfserr_jukebox;
2306                 else
2307                         return nfserr_seq_misordered;
2308         }
2309         /* Note unsigned 32-bit arithmetic handles wraparound: */
2310         if (likely(seqid == slot_seqid + 1))
2311                 return nfs_ok;
2312         if (seqid == slot_seqid)
2313                 return nfserr_replay_cache;
2314         return nfserr_seq_misordered;
2315 }
2316
2317 /*
2318  * Cache the create session result into the create session single DRC
2319  * slot cache by saving the xdr structure. sl_seqid has been set.
2320  * Do this for solo or embedded create session operations.
2321  */
2322 static void
2323 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2324                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2325 {
2326         slot->sl_status = nfserr;
2327         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2328 }
2329
2330 static __be32
2331 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2332                             struct nfsd4_clid_slot *slot)
2333 {
2334         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2335         return slot->sl_status;
2336 }
2337
2338 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2339                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2340                         1 +     /* MIN tag is length with zero, only length */ \
2341                         3 +     /* version, opcount, opcode */ \
2342                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2343                                 /* seqid, slotID, slotID, cache */ \
2344                         4 ) * sizeof(__be32))
2345
2346 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2347                         2 +     /* verifier: AUTH_NULL, length 0 */\
2348                         1 +     /* status */ \
2349                         1 +     /* MIN tag is length with zero, only length */ \
2350                         3 +     /* opcount, opcode, opstatus*/ \
2351                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2352                                 /* seqid, slotID, slotID, slotID, status */ \
2353                         5 ) * sizeof(__be32))
2354
2355 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2356 {
2357         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2358
2359         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2360                 return nfserr_toosmall;
2361         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2362                 return nfserr_toosmall;
2363         ca->headerpadsz = 0;
2364         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2365         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2366         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2367         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2368                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2369         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2370         /*
2371          * Note decreasing slot size below client's request may make it
2372          * difficult for client to function correctly, whereas
2373          * decreasing the number of slots will (just?) affect
2374          * performance.  When short on memory we therefore prefer to
2375          * decrease number of slots instead of their size.  Clients that
2376          * request larger slots than they need will get poor results:
2377          */
2378         ca->maxreqs = nfsd4_get_drc_mem(ca);
2379         if (!ca->maxreqs)
2380                 return nfserr_jukebox;
2381
2382         return nfs_ok;
2383 }
2384
2385 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2386                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2387 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2388                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2389
2390 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2391 {
2392         ca->headerpadsz = 0;
2393
2394         /*
2395          * These RPC_MAX_HEADER macros are overkill, especially since we
2396          * don't even do gss on the backchannel yet.  But this is still
2397          * less than 1k.  Tighten up this estimate in the unlikely event
2398          * it turns out to be a problem for some client:
2399          */
2400         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2401                 return nfserr_toosmall;
2402         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2403                 return nfserr_toosmall;
2404         ca->maxresp_cached = 0;
2405         if (ca->maxops < 2)
2406                 return nfserr_toosmall;
2407
2408         return nfs_ok;
2409 }
2410
2411 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2412 {
2413         switch (cbs->flavor) {
2414         case RPC_AUTH_NULL:
2415         case RPC_AUTH_UNIX:
2416                 return nfs_ok;
2417         default:
2418                 /*
2419                  * GSS case: the spec doesn't allow us to return this
2420                  * error.  But it also doesn't allow us not to support
2421                  * GSS.
2422                  * I'd rather this fail hard than return some error the
2423                  * client might think it can already handle:
2424                  */
2425                 return nfserr_encr_alg_unsupp;
2426         }
2427 }
2428
2429 __be32
2430 nfsd4_create_session(struct svc_rqst *rqstp,
2431                      struct nfsd4_compound_state *cstate,
2432                      struct nfsd4_create_session *cr_ses)
2433 {
2434         struct sockaddr *sa = svc_addr(rqstp);
2435         struct nfs4_client *conf, *unconf;
2436         struct nfs4_client *old = NULL;
2437         struct nfsd4_session *new;
2438         struct nfsd4_conn *conn;
2439         struct nfsd4_clid_slot *cs_slot = NULL;
2440         __be32 status = 0;
2441         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2442
2443         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2444                 return nfserr_inval;
2445         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2446         if (status)
2447                 return status;
2448         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2449         if (status)
2450                 return status;
2451         status = check_backchannel_attrs(&cr_ses->back_channel);
2452         if (status)
2453                 goto out_release_drc_mem;
2454         status = nfserr_jukebox;
2455         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2456         if (!new)
2457                 goto out_release_drc_mem;
2458         conn = alloc_conn_from_crses(rqstp, cr_ses);
2459         if (!conn)
2460                 goto out_free_session;
2461
2462         spin_lock(&nn->client_lock);
2463         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2464         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2465         WARN_ON_ONCE(conf && unconf);
2466
2467         if (conf) {
2468                 status = nfserr_wrong_cred;
2469                 if (!mach_creds_match(conf, rqstp))
2470                         goto out_free_conn;
2471                 cs_slot = &conf->cl_cs_slot;
2472                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2473                 if (status == nfserr_replay_cache) {
2474                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
2475                         goto out_free_conn;
2476                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
2477                         status = nfserr_seq_misordered;
2478                         goto out_free_conn;
2479                 }
2480         } else if (unconf) {
2481                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2482                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2483                         status = nfserr_clid_inuse;
2484                         goto out_free_conn;
2485                 }
2486                 status = nfserr_wrong_cred;
2487                 if (!mach_creds_match(unconf, rqstp))
2488                         goto out_free_conn;
2489                 cs_slot = &unconf->cl_cs_slot;
2490                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2491                 if (status) {
2492                         /* an unconfirmed replay returns misordered */
2493                         status = nfserr_seq_misordered;
2494                         goto out_free_conn;
2495                 }
2496                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2497                 if (old) {
2498                         status = mark_client_expired_locked(old);
2499                         if (status) {
2500                                 old = NULL;
2501                                 goto out_free_conn;
2502                         }
2503                 }
2504                 move_to_confirmed(unconf);
2505                 conf = unconf;
2506         } else {
2507                 status = nfserr_stale_clientid;
2508                 goto out_free_conn;
2509         }
2510         status = nfs_ok;
2511         /*
2512          * We do not support RDMA or persistent sessions
2513          */
2514         cr_ses->flags &= ~SESSION4_PERSIST;
2515         cr_ses->flags &= ~SESSION4_RDMA;
2516
2517         init_session(rqstp, new, conf, cr_ses);
2518         nfsd4_get_session_locked(new);
2519
2520         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2521                NFS4_MAX_SESSIONID_LEN);
2522         cs_slot->sl_seqid++;
2523         cr_ses->seqid = cs_slot->sl_seqid;
2524
2525         /* cache solo and embedded create sessions under the client_lock */
2526         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2527         spin_unlock(&nn->client_lock);
2528         /* init connection and backchannel */
2529         nfsd4_init_conn(rqstp, conn, new);
2530         nfsd4_put_session(new);
2531         if (old)
2532                 expire_client(old);
2533         return status;
2534 out_free_conn:
2535         spin_unlock(&nn->client_lock);
2536         free_conn(conn);
2537         if (old)
2538                 expire_client(old);
2539 out_free_session:
2540         __free_session(new);
2541 out_release_drc_mem:
2542         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2543         return status;
2544 }
2545
2546 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2547 {
2548         switch (*dir) {
2549         case NFS4_CDFC4_FORE:
2550         case NFS4_CDFC4_BACK:
2551                 return nfs_ok;
2552         case NFS4_CDFC4_FORE_OR_BOTH:
2553         case NFS4_CDFC4_BACK_OR_BOTH:
2554                 *dir = NFS4_CDFC4_BOTH;
2555                 return nfs_ok;
2556         };
2557         return nfserr_inval;
2558 }
2559
2560 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2561 {
2562         struct nfsd4_session *session = cstate->session;
2563         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2564         __be32 status;
2565
2566         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2567         if (status)
2568                 return status;
2569         spin_lock(&nn->client_lock);
2570         session->se_cb_prog = bc->bc_cb_program;
2571         session->se_cb_sec = bc->bc_cb_sec;
2572         spin_unlock(&nn->client_lock);
2573
2574         nfsd4_probe_callback(session->se_client);
2575
2576         return nfs_ok;
2577 }
2578
2579 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2580                      struct nfsd4_compound_state *cstate,
2581                      struct nfsd4_bind_conn_to_session *bcts)
2582 {
2583         __be32 status;
2584         struct nfsd4_conn *conn;
2585         struct nfsd4_session *session;
2586         struct net *net = SVC_NET(rqstp);
2587         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2588
2589         if (!nfsd4_last_compound_op(rqstp))
2590                 return nfserr_not_only_op;
2591         spin_lock(&nn->client_lock);
2592         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2593         spin_unlock(&nn->client_lock);
2594         if (!session)
2595                 goto out_no_session;
2596         status = nfserr_wrong_cred;
2597         if (!mach_creds_match(session->se_client, rqstp))
2598                 goto out;
2599         status = nfsd4_map_bcts_dir(&bcts->dir);
2600         if (status)
2601                 goto out;
2602         conn = alloc_conn(rqstp, bcts->dir);
2603         status = nfserr_jukebox;
2604         if (!conn)
2605                 goto out;
2606         nfsd4_init_conn(rqstp, conn, session);
2607         status = nfs_ok;
2608 out:
2609         nfsd4_put_session(session);
2610 out_no_session:
2611         return status;
2612 }
2613
2614 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2615 {
2616         if (!session)
2617                 return 0;
2618         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2619 }
2620
2621 __be32
2622 nfsd4_destroy_session(struct svc_rqst *r,
2623                       struct nfsd4_compound_state *cstate,
2624                       struct nfsd4_destroy_session *sessionid)
2625 {
2626         struct nfsd4_session *ses;
2627         __be32 status;
2628         int ref_held_by_me = 0;
2629         struct net *net = SVC_NET(r);
2630         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2631
2632         status = nfserr_not_only_op;
2633         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2634                 if (!nfsd4_last_compound_op(r))
2635                         goto out;
2636                 ref_held_by_me++;
2637         }
2638         dump_sessionid(__func__, &sessionid->sessionid);
2639         spin_lock(&nn->client_lock);
2640         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2641         if (!ses)
2642                 goto out_client_lock;
2643         status = nfserr_wrong_cred;
2644         if (!mach_creds_match(ses->se_client, r))
2645                 goto out_put_session;
2646         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2647         if (status)
2648                 goto out_put_session;
2649         unhash_session(ses);
2650         spin_unlock(&nn->client_lock);
2651
2652         nfsd4_probe_callback_sync(ses->se_client);
2653
2654         spin_lock(&nn->client_lock);
2655         status = nfs_ok;
2656 out_put_session:
2657         nfsd4_put_session_locked(ses);
2658 out_client_lock:
2659         spin_unlock(&nn->client_lock);
2660 out:
2661         return status;
2662 }
2663
2664 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2665 {
2666         struct nfsd4_conn *c;
2667
2668         list_for_each_entry(c, &s->se_conns, cn_persession) {
2669                 if (c->cn_xprt == xpt) {
2670                         return c;
2671                 }
2672         }
2673         return NULL;
2674 }
2675
2676 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2677 {
2678         struct nfs4_client *clp = ses->se_client;
2679         struct nfsd4_conn *c;
2680         __be32 status = nfs_ok;
2681         int ret;
2682
2683         spin_lock(&clp->cl_lock);
2684         c = __nfsd4_find_conn(new->cn_xprt, ses);
2685         if (c)
2686                 goto out_free;
2687         status = nfserr_conn_not_bound_to_session;
2688         if (clp->cl_mach_cred)
2689                 goto out_free;
2690         __nfsd4_hash_conn(new, ses);
2691         spin_unlock(&clp->cl_lock);
2692         ret = nfsd4_register_conn(new);
2693         if (ret)
2694                 /* oops; xprt is already down: */
2695                 nfsd4_conn_lost(&new->cn_xpt_user);
2696         return nfs_ok;
2697 out_free:
2698         spin_unlock(&clp->cl_lock);
2699         free_conn(new);
2700         return status;
2701 }
2702
2703 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2704 {
2705         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2706
2707         return args->opcnt > session->se_fchannel.maxops;
2708 }
2709
2710 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2711                                   struct nfsd4_session *session)
2712 {
2713         struct xdr_buf *xb = &rqstp->rq_arg;
2714
2715         return xb->len > session->se_fchannel.maxreq_sz;
2716 }
2717
2718 __be32
2719 nfsd4_sequence(struct svc_rqst *rqstp,
2720                struct nfsd4_compound_state *cstate,
2721                struct nfsd4_sequence *seq)
2722 {
2723         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2724         struct xdr_stream *xdr = &resp->xdr;
2725         struct nfsd4_session *session;
2726         struct nfs4_client *clp;
2727         struct nfsd4_slot *slot;
2728         struct nfsd4_conn *conn;
2729         __be32 status;
2730         int buflen;
2731         struct net *net = SVC_NET(rqstp);
2732         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2733
2734         if (resp->opcnt != 1)
2735                 return nfserr_sequence_pos;
2736
2737         /*
2738          * Will be either used or freed by nfsd4_sequence_check_conn
2739          * below.
2740          */
2741         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2742         if (!conn)
2743                 return nfserr_jukebox;
2744
2745         spin_lock(&nn->client_lock);
2746         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2747         if (!session)
2748                 goto out_no_session;
2749         clp = session->se_client;
2750
2751         status = nfserr_too_many_ops;
2752         if (nfsd4_session_too_many_ops(rqstp, session))
2753                 goto out_put_session;
2754
2755         status = nfserr_req_too_big;
2756         if (nfsd4_request_too_big(rqstp, session))
2757                 goto out_put_session;
2758
2759         status = nfserr_badslot;
2760         if (seq->slotid >= session->se_fchannel.maxreqs)
2761                 goto out_put_session;
2762
2763         slot = session->se_slots[seq->slotid];
2764         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2765
2766         /* We do not negotiate the number of slots yet, so set the
2767          * maxslots to the session maxreqs which is used to encode
2768          * sr_highest_slotid and the sr_target_slot id to maxslots */
2769         seq->maxslots = session->se_fchannel.maxreqs;
2770
2771         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2772                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2773         if (status == nfserr_replay_cache) {
2774                 status = nfserr_seq_misordered;
2775                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2776                         goto out_put_session;
2777                 cstate->slot = slot;
2778                 cstate->session = session;
2779                 cstate->clp = clp;
2780                 /* Return the cached reply status and set cstate->status
2781                  * for nfsd4_proc_compound processing */
2782                 status = nfsd4_replay_cache_entry(resp, seq);
2783                 cstate->status = nfserr_replay_cache;
2784                 goto out;
2785         }
2786         if (status)
2787                 goto out_put_session;
2788
2789         status = nfsd4_sequence_check_conn(conn, session);
2790         conn = NULL;
2791         if (status)
2792                 goto out_put_session;
2793
2794         buflen = (seq->cachethis) ?
2795                         session->se_fchannel.maxresp_cached :
2796                         session->se_fchannel.maxresp_sz;
2797         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
2798                                     nfserr_rep_too_big;
2799         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
2800                 goto out_put_session;
2801         svc_reserve(rqstp, buflen);
2802
2803         status = nfs_ok;
2804         /* Success! bump slot seqid */
2805         slot->sl_seqid = seq->seqid;
2806         slot->sl_flags |= NFSD4_SLOT_INUSE;
2807         if (seq->cachethis)
2808                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2809         else
2810                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2811
2812         cstate->slot = slot;
2813         cstate->session = session;
2814         cstate->clp = clp;
2815
2816 out:
2817         switch (clp->cl_cb_state) {
2818         case NFSD4_CB_DOWN:
2819                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2820                 break;
2821         case NFSD4_CB_FAULT:
2822                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2823                 break;
2824         default:
2825                 seq->status_flags = 0;
2826         }
2827         if (!list_empty(&clp->cl_revoked))
2828                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2829 out_no_session:
2830         if (conn)
2831                 free_conn(conn);
2832         spin_unlock(&nn->client_lock);
2833         return status;
2834 out_put_session:
2835         nfsd4_put_session_locked(session);
2836         goto out_no_session;
2837 }
2838
2839 void
2840 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
2841 {
2842         struct nfsd4_compound_state *cs = &resp->cstate;
2843
2844         if (nfsd4_has_session(cs)) {
2845                 if (cs->status != nfserr_replay_cache) {
2846                         nfsd4_store_cache_entry(resp);
2847                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
2848                 }
2849                 /* Drop session reference that was taken in nfsd4_sequence() */
2850                 nfsd4_put_session(cs->session);
2851         } else if (cs->clp)
2852                 put_client_renew(cs->clp);
2853 }
2854
2855 __be32
2856 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2857 {
2858         struct nfs4_client *conf, *unconf;
2859         struct nfs4_client *clp = NULL;
2860         __be32 status = 0;
2861         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2862
2863         spin_lock(&nn->client_lock);
2864         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2865         conf = find_confirmed_client(&dc->clientid, true, nn);
2866         WARN_ON_ONCE(conf && unconf);
2867
2868         if (conf) {
2869                 if (client_has_state(conf)) {
2870                         status = nfserr_clientid_busy;
2871                         goto out;
2872                 }
2873                 status = mark_client_expired_locked(conf);
2874                 if (status)
2875                         goto out;
2876                 clp = conf;
2877         } else if (unconf)
2878                 clp = unconf;
2879         else {
2880                 status = nfserr_stale_clientid;
2881                 goto out;
2882         }
2883         if (!mach_creds_match(clp, rqstp)) {
2884                 clp = NULL;
2885                 status = nfserr_wrong_cred;
2886                 goto out;
2887         }
2888         unhash_client_locked(clp);
2889 out:
2890         spin_unlock(&nn->client_lock);
2891         if (clp)
2892                 expire_client(clp);
2893         return status;
2894 }
2895
2896 __be32
2897 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2898 {
2899         __be32 status = 0;
2900
2901         if (rc->rca_one_fs) {
2902                 if (!cstate->current_fh.fh_dentry)
2903                         return nfserr_nofilehandle;
2904                 /*
2905                  * We don't take advantage of the rca_one_fs case.
2906                  * That's OK, it's optional, we can safely ignore it.
2907                  */
2908                  return nfs_ok;
2909         }
2910
2911         status = nfserr_complete_already;
2912         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2913                              &cstate->session->se_client->cl_flags))
2914                 goto out;
2915
2916         status = nfserr_stale_clientid;
2917         if (is_client_expired(cstate->session->se_client))
2918                 /*
2919                  * The following error isn't really legal.
2920                  * But we only get here if the client just explicitly
2921                  * destroyed the client.  Surely it no longer cares what
2922                  * error it gets back on an operation for the dead
2923                  * client.
2924                  */
2925                 goto out;
2926
2927         status = nfs_ok;
2928         nfsd4_client_record_create(cstate->session->se_client);
2929 out:
2930         return status;
2931 }
2932
2933 __be32
2934 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2935                   struct nfsd4_setclientid *setclid)
2936 {
2937         struct xdr_netobj       clname = setclid->se_name;
2938         nfs4_verifier           clverifier = setclid->se_verf;
2939         struct nfs4_client      *conf, *new;
2940         struct nfs4_client      *unconf = NULL;
2941         __be32                  status;
2942         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2943
2944         new = create_client(clname, rqstp, &clverifier);
2945         if (new == NULL)
2946                 return nfserr_jukebox;
2947         /* Cases below refer to rfc 3530 section 14.2.33: */
2948         spin_lock(&nn->client_lock);
2949         conf = find_confirmed_client_by_name(&clname, nn);
2950         if (conf) {
2951                 /* case 0: */
2952                 status = nfserr_clid_inuse;
2953                 if (clp_used_exchangeid(conf))
2954                         goto out;
2955                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2956                         char addr_str[INET6_ADDRSTRLEN];
2957                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2958                                  sizeof(addr_str));
2959                         dprintk("NFSD: setclientid: string in use by client "
2960                                 "at %s\n", addr_str);
2961                         goto out;
2962                 }
2963         }
2964         unconf = find_unconfirmed_client_by_name(&clname, nn);
2965         if (unconf)
2966                 unhash_client_locked(unconf);
2967         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2968                 /* case 1: probable callback update */
2969                 copy_clid(new, conf);
2970         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2971                 gen_clid(new, nn);
2972         new->cl_minorversion = 0;
2973         gen_callback(new, setclid, rqstp);
2974         add_to_unconfirmed(new);
2975         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2976         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2977         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2978         new = NULL;
2979         status = nfs_ok;
2980 out:
2981         spin_unlock(&nn->client_lock);
2982         if (new)
2983                 free_client(new);
2984         if (unconf)
2985                 expire_client(unconf);
2986         return status;
2987 }
2988
2989
2990 __be32
2991 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2992                          struct nfsd4_compound_state *cstate,
2993                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2994 {
2995         struct nfs4_client *conf, *unconf;
2996         struct nfs4_client *old = NULL;
2997         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2998         clientid_t * clid = &setclientid_confirm->sc_clientid;
2999         __be32 status;
3000         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3001
3002         if (STALE_CLIENTID(clid, nn))
3003                 return nfserr_stale_clientid;
3004
3005         spin_lock(&nn->client_lock);
3006         conf = find_confirmed_client(clid, false, nn);
3007         unconf = find_unconfirmed_client(clid, false, nn);
3008         /*
3009          * We try hard to give out unique clientid's, so if we get an
3010          * attempt to confirm the same clientid with a different cred,
3011          * there's a bug somewhere.  Let's charitably assume it's our
3012          * bug.
3013          */
3014         status = nfserr_serverfault;
3015         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3016                 goto out;
3017         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3018                 goto out;
3019         /* cases below refer to rfc 3530 section 14.2.34: */
3020         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3021                 if (conf && !unconf) /* case 2: probable retransmit */
3022                         status = nfs_ok;
3023                 else /* case 4: client hasn't noticed we rebooted yet? */
3024                         status = nfserr_stale_clientid;
3025                 goto out;
3026         }
3027         status = nfs_ok;
3028         if (conf) { /* case 1: callback update */
3029                 old = unconf;
3030                 unhash_client_locked(old);
3031                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3032         } else { /* case 3: normal case; new or rebooted client */
3033                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3034                 if (old) {
3035                         status = mark_client_expired_locked(old);
3036                         if (status) {
3037                                 old = NULL;
3038                                 goto out;
3039                         }
3040                 }
3041                 move_to_confirmed(unconf);
3042                 conf = unconf;
3043         }
3044         get_client_locked(conf);
3045         spin_unlock(&nn->client_lock);
3046         nfsd4_probe_callback(conf);
3047         spin_lock(&nn->client_lock);
3048         put_client_renew_locked(conf);
3049 out:
3050         spin_unlock(&nn->client_lock);
3051         if (old)
3052                 expire_client(old);
3053         return status;
3054 }
3055
3056 static struct nfs4_file *nfsd4_alloc_file(void)
3057 {
3058         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3059 }
3060
3061 /* OPEN Share state helper functions */
3062 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3063                                 struct nfs4_file *fp)
3064 {
3065         lockdep_assert_held(&state_lock);
3066
3067         atomic_set(&fp->fi_ref, 1);
3068         spin_lock_init(&fp->fi_lock);
3069         INIT_LIST_HEAD(&fp->fi_stateids);
3070         INIT_LIST_HEAD(&fp->fi_delegations);
3071         fh_copy_shallow(&fp->fi_fhandle, fh);
3072         fp->fi_deleg_file = NULL;
3073         fp->fi_had_conflict = false;
3074         fp->fi_share_deny = 0;
3075         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3076         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3077         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3078 }
3079
3080 void
3081 nfsd4_free_slabs(void)
3082 {
3083         kmem_cache_destroy(openowner_slab);
3084         kmem_cache_destroy(lockowner_slab);
3085         kmem_cache_destroy(file_slab);
3086         kmem_cache_destroy(stateid_slab);
3087         kmem_cache_destroy(deleg_slab);
3088 }
3089
3090 int
3091 nfsd4_init_slabs(void)
3092 {
3093         openowner_slab = kmem_cache_create("nfsd4_openowners",
3094                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3095         if (openowner_slab == NULL)
3096                 goto out;
3097         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3098                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3099         if (lockowner_slab == NULL)
3100                 goto out_free_openowner_slab;
3101         file_slab = kmem_cache_create("nfsd4_files",
3102                         sizeof(struct nfs4_file), 0, 0, NULL);
3103         if (file_slab == NULL)
3104                 goto out_free_lockowner_slab;
3105         stateid_slab = kmem_cache_create("nfsd4_stateids",
3106                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3107         if (stateid_slab == NULL)
3108                 goto out_free_file_slab;
3109         deleg_slab = kmem_cache_create("nfsd4_delegations",
3110                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3111         if (deleg_slab == NULL)
3112                 goto out_free_stateid_slab;
3113         return 0;
3114
3115 out_free_stateid_slab:
3116         kmem_cache_destroy(stateid_slab);
3117 out_free_file_slab:
3118         kmem_cache_destroy(file_slab);
3119 out_free_lockowner_slab:
3120         kmem_cache_destroy(lockowner_slab);
3121 out_free_openowner_slab:
3122         kmem_cache_destroy(openowner_slab);
3123 out:
3124         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3125         return -ENOMEM;
3126 }
3127
3128 static void init_nfs4_replay(struct nfs4_replay *rp)
3129 {
3130         rp->rp_status = nfserr_serverfault;
3131         rp->rp_buflen = 0;
3132         rp->rp_buf = rp->rp_ibuf;
3133         mutex_init(&rp->rp_mutex);
3134 }
3135
3136 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3137                 struct nfs4_stateowner *so)
3138 {
3139         if (!nfsd4_has_session(cstate)) {
3140                 mutex_lock(&so->so_replay.rp_mutex);
3141                 cstate->replay_owner = nfs4_get_stateowner(so);
3142         }
3143 }
3144
3145 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3146 {
3147         struct nfs4_stateowner *so = cstate->replay_owner;
3148
3149         if (so != NULL) {
3150                 cstate->replay_owner = NULL;
3151                 mutex_unlock(&so->so_replay.rp_mutex);
3152                 nfs4_put_stateowner(so);
3153         }
3154 }
3155
3156 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3157 {
3158         struct nfs4_stateowner *sop;
3159
3160         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3161         if (!sop)
3162                 return NULL;
3163
3164         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3165         if (!sop->so_owner.data) {
3166                 kmem_cache_free(slab, sop);
3167                 return NULL;
3168         }
3169         sop->so_owner.len = owner->len;
3170
3171         INIT_LIST_HEAD(&sop->so_stateids);
3172         sop->so_client = clp;
3173         init_nfs4_replay(&sop->so_replay);
3174         atomic_set(&sop->so_count, 1);
3175         return sop;
3176 }
3177
3178 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3179 {
3180         lockdep_assert_held(&clp->cl_lock);
3181
3182         list_add(&oo->oo_owner.so_strhash,
3183                  &clp->cl_ownerstr_hashtbl[strhashval]);
3184         list_add(&oo->oo_perclient, &clp->cl_openowners);
3185 }
3186
3187 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3188 {
3189         unhash_openowner_locked(openowner(so));
3190 }
3191
3192 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3193 {
3194         struct nfs4_openowner *oo = openowner(so);
3195
3196         kmem_cache_free(openowner_slab, oo);
3197 }
3198
3199 static const struct nfs4_stateowner_operations openowner_ops = {
3200         .so_unhash =    nfs4_unhash_openowner,
3201         .so_free =      nfs4_free_openowner,
3202 };
3203
3204 static struct nfs4_openowner *
3205 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3206                            struct nfsd4_compound_state *cstate)
3207 {
3208         struct nfs4_client *clp = cstate->clp;
3209         struct nfs4_openowner *oo, *ret;
3210
3211         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3212         if (!oo)
3213                 return NULL;
3214         oo->oo_owner.so_ops = &openowner_ops;
3215         oo->oo_owner.so_is_open_owner = 1;
3216         oo->oo_owner.so_seqid = open->op_seqid;
3217         oo->oo_flags = 0;
3218         if (nfsd4_has_session(cstate))
3219                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3220         oo->oo_time = 0;
3221         oo->oo_last_closed_stid = NULL;
3222         INIT_LIST_HEAD(&oo->oo_close_lru);
3223         spin_lock(&clp->cl_lock);
3224         ret = find_openstateowner_str_locked(strhashval, open, clp);
3225         if (ret == NULL) {
3226                 hash_openowner(oo, clp, strhashval);
3227                 ret = oo;
3228         } else
3229                 nfs4_free_openowner(&oo->oo_owner);
3230         spin_unlock(&clp->cl_lock);
3231         return oo;
3232 }
3233
3234 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
3235         struct nfs4_openowner *oo = open->op_openowner;
3236
3237         atomic_inc(&stp->st_stid.sc_count);
3238         stp->st_stid.sc_type = NFS4_OPEN_STID;
3239         INIT_LIST_HEAD(&stp->st_locks);
3240         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3241         get_nfs4_file(fp);
3242         stp->st_stid.sc_file = fp;
3243         stp->st_access_bmap = 0;
3244         stp->st_deny_bmap = 0;
3245         stp->st_openstp = NULL;
3246         spin_lock(&oo->oo_owner.so_client->cl_lock);
3247         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3248         spin_lock(&fp->fi_lock);
3249         list_add(&stp->st_perfile, &fp->fi_stateids);
3250         spin_unlock(&fp->fi_lock);
3251         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3252 }
3253
3254 /*
3255  * In the 4.0 case we need to keep the owners around a little while to handle
3256  * CLOSE replay. We still do need to release any file access that is held by
3257  * them before returning however.
3258  */
3259 static void
3260 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3261 {
3262         struct nfs4_ol_stateid *last;
3263         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3264         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3265                                                 nfsd_net_id);
3266
3267         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3268
3269         /*
3270          * We know that we hold one reference via nfsd4_close, and another
3271          * "persistent" reference for the client. If the refcount is higher
3272          * than 2, then there are still calls in progress that are using this
3273          * stateid. We can't put the sc_file reference until they are finished.
3274          * Wait for the refcount to drop to 2. Since it has been unhashed,
3275          * there should be no danger of the refcount going back up again at
3276          * this point.
3277          */
3278         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3279
3280         release_all_access(s);
3281         if (s->st_stid.sc_file) {
3282                 put_nfs4_file(s->st_stid.sc_file);
3283                 s->st_stid.sc_file = NULL;
3284         }
3285
3286         spin_lock(&nn->client_lock);
3287         last = oo->oo_last_closed_stid;
3288         oo->oo_last_closed_stid = s;
3289         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3290         oo->oo_time = get_seconds();
3291         spin_unlock(&nn->client_lock);
3292         if (last)
3293                 nfs4_put_stid(&last->st_stid);
3294 }
3295
3296 /* search file_hashtbl[] for file */
3297 static struct nfs4_file *
3298 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3299 {
3300         struct nfs4_file *fp;
3301
3302         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3303                 if (nfsd_fh_match(&fp->fi_fhandle, fh)) {
3304                         if (atomic_inc_not_zero(&fp->fi_ref))
3305                                 return fp;
3306                 }
3307         }
3308         return NULL;
3309 }
3310
3311 static struct nfs4_file *
3312 find_file(struct knfsd_fh *fh)
3313 {
3314         struct nfs4_file *fp;
3315         unsigned int hashval = file_hashval(fh);
3316
3317         rcu_read_lock();
3318         fp = find_file_locked(fh, hashval);
3319         rcu_read_unlock();
3320         return fp;
3321 }
3322
3323 static struct nfs4_file *
3324 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3325 {
3326         struct nfs4_file *fp;
3327         unsigned int hashval = file_hashval(fh);
3328
3329         rcu_read_lock();
3330         fp = find_file_locked(fh, hashval);
3331         rcu_read_unlock();
3332         if (fp)
3333                 return fp;
3334
3335         spin_lock(&state_lock);
3336         fp = find_file_locked(fh, hashval);
3337         if (likely(fp == NULL)) {
3338                 nfsd4_init_file(fh, hashval, new);
3339                 fp = new;
3340         }
3341         spin_unlock(&state_lock);
3342
3343         return fp;
3344 }
3345
3346 /*
3347  * Called to check deny when READ with all zero stateid or
3348  * WRITE with all zero or all one stateid
3349  */
3350 static __be32
3351 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3352 {
3353         struct nfs4_file *fp;
3354         __be32 ret = nfs_ok;
3355
3356         fp = find_file(&current_fh->fh_handle);
3357         if (!fp)
3358                 return ret;
3359         /* Check for conflicting share reservations */
3360         spin_lock(&fp->fi_lock);
3361         if (fp->fi_share_deny & deny_type)
3362                 ret = nfserr_locked;
3363         spin_unlock(&fp->fi_lock);
3364         put_nfs4_file(fp);
3365         return ret;
3366 }
3367
3368 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3369 {
3370         struct nfs4_delegation *dp = cb_to_delegation(cb);
3371         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3372                                           nfsd_net_id);
3373
3374         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3375
3376         /*
3377          * We can't do this in nfsd_break_deleg_cb because it is
3378          * already holding inode->i_lock.
3379          *
3380          * If the dl_time != 0, then we know that it has already been
3381          * queued for a lease break. Don't queue it again.
3382          */
3383         spin_lock(&state_lock);
3384         if (dp->dl_time == 0) {
3385                 dp->dl_time = get_seconds();
3386                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3387         }
3388         spin_unlock(&state_lock);
3389 }
3390
3391 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3392                 struct rpc_task *task)
3393 {
3394         struct nfs4_delegation *dp = cb_to_delegation(cb);
3395
3396         switch (task->tk_status) {
3397         case 0:
3398                 return 1;
3399         case -EBADHANDLE:
3400         case -NFS4ERR_BAD_STATEID:
3401                 /*
3402                  * Race: client probably got cb_recall before open reply
3403                  * granting delegation.
3404                  */
3405                 if (dp->dl_retries--) {
3406                         rpc_delay(task, 2 * HZ);
3407                         return 0;
3408                 }
3409                 /*FALLTHRU*/
3410         default:
3411                 return -1;
3412         }
3413 }
3414
3415 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3416 {
3417         struct nfs4_delegation *dp = cb_to_delegation(cb);
3418
3419         nfs4_put_stid(&dp->dl_stid);
3420 }
3421
3422 static struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3423         .prepare        = nfsd4_cb_recall_prepare,
3424         .done           = nfsd4_cb_recall_done,
3425         .release        = nfsd4_cb_recall_release,
3426 };
3427
3428 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3429 {
3430         /*
3431          * We're assuming the state code never drops its reference
3432          * without first removing the lease.  Since we're in this lease
3433          * callback (and since the lease code is serialized by the kernel
3434          * lock) we know the server hasn't removed the lease yet, we know
3435          * it's safe to take a reference.
3436          */
3437         atomic_inc(&dp->dl_stid.sc_count);
3438         nfsd4_run_cb(&dp->dl_recall);
3439 }
3440
3441 /* Called from break_lease() with i_lock held. */
3442 static bool
3443 nfsd_break_deleg_cb(struct file_lock *fl)
3444 {
3445         bool ret = false;
3446         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3447         struct nfs4_delegation *dp;
3448
3449         if (!fp) {
3450                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3451                 return ret;
3452         }
3453         if (fp->fi_had_conflict) {
3454                 WARN(1, "duplicate break on %p\n", fp);
3455                 return ret;
3456         }
3457         /*
3458          * We don't want the locks code to timeout the lease for us;
3459          * we'll remove it ourself if a delegation isn't returned
3460          * in time:
3461          */
3462         fl->fl_break_time = 0;
3463
3464         spin_lock(&fp->fi_lock);
3465         fp->fi_had_conflict = true;
3466         /*
3467          * If there are no delegations on the list, then return true
3468          * so that the lease code will go ahead and delete it.
3469          */
3470         if (list_empty(&fp->fi_delegations))
3471                 ret = true;
3472         else
3473                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3474                         nfsd_break_one_deleg(dp);
3475         spin_unlock(&fp->fi_lock);
3476         return ret;
3477 }
3478
3479 static int
3480 nfsd_change_deleg_cb(struct file_lock **onlist, int arg, struct list_head *dispose)
3481 {
3482         if (arg & F_UNLCK)
3483                 return lease_modify(onlist, arg, dispose);
3484         else
3485                 return -EAGAIN;
3486 }
3487
3488 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3489         .lm_break = nfsd_break_deleg_cb,
3490         .lm_change = nfsd_change_deleg_cb,
3491 };
3492
3493 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3494 {
3495         if (nfsd4_has_session(cstate))
3496                 return nfs_ok;
3497         if (seqid == so->so_seqid - 1)
3498                 return nfserr_replay_me;
3499         if (seqid == so->so_seqid)
3500                 return nfs_ok;
3501         return nfserr_bad_seqid;
3502 }
3503
3504 static __be32 lookup_clientid(clientid_t *clid,
3505                 struct nfsd4_compound_state *cstate,
3506                 struct nfsd_net *nn)
3507 {
3508         struct nfs4_client *found;
3509
3510         if (cstate->clp) {
3511                 found = cstate->clp;
3512                 if (!same_clid(&found->cl_clientid, clid))
3513                         return nfserr_stale_clientid;
3514                 return nfs_ok;
3515         }
3516
3517         if (STALE_CLIENTID(clid, nn))
3518                 return nfserr_stale_clientid;
3519
3520         /*
3521          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3522          * cached already then we know this is for is for v4.0 and "sessions"
3523          * will be false.
3524          */
3525         WARN_ON_ONCE(cstate->session);
3526         spin_lock(&nn->client_lock);
3527         found = find_confirmed_client(clid, false, nn);
3528         if (!found) {
3529                 spin_unlock(&nn->client_lock);
3530                 return nfserr_expired;
3531         }
3532         atomic_inc(&found->cl_refcount);
3533         spin_unlock(&nn->client_lock);
3534
3535         /* Cache the nfs4_client in cstate! */
3536         cstate->clp = found;
3537         return nfs_ok;
3538 }
3539
3540 __be32
3541 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3542                     struct nfsd4_open *open, struct nfsd_net *nn)
3543 {
3544         clientid_t *clientid = &open->op_clientid;
3545         struct nfs4_client *clp = NULL;
3546         unsigned int strhashval;
3547         struct nfs4_openowner *oo = NULL;
3548         __be32 status;
3549
3550         if (STALE_CLIENTID(&open->op_clientid, nn))
3551                 return nfserr_stale_clientid;
3552         /*
3553          * In case we need it later, after we've already created the
3554          * file and don't want to risk a further failure:
3555          */
3556         open->op_file = nfsd4_alloc_file();
3557         if (open->op_file == NULL)
3558                 return nfserr_jukebox;
3559
3560         status = lookup_clientid(clientid, cstate, nn);
3561         if (status)
3562                 return status;
3563         clp = cstate->clp;
3564
3565         strhashval = ownerstr_hashval(&open->op_owner);
3566         oo = find_openstateowner_str(strhashval, open, clp);
3567         open->op_openowner = oo;
3568         if (!oo) {
3569                 goto new_owner;
3570         }
3571         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3572                 /* Replace unconfirmed owners without checking for replay. */
3573                 release_openowner(oo);
3574                 open->op_openowner = NULL;
3575                 goto new_owner;
3576         }
3577         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3578         if (status)
3579                 return status;
3580         goto alloc_stateid;
3581 new_owner:
3582         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3583         if (oo == NULL)
3584                 return nfserr_jukebox;
3585         open->op_openowner = oo;
3586 alloc_stateid:
3587         open->op_stp = nfs4_alloc_open_stateid(clp);
3588         if (!open->op_stp)
3589                 return nfserr_jukebox;
3590         return nfs_ok;
3591 }
3592
3593 static inline __be32
3594 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3595 {
3596         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3597                 return nfserr_openmode;
3598         else
3599                 return nfs_ok;
3600 }
3601
3602 static int share_access_to_flags(u32 share_access)
3603 {
3604         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3605 }
3606
3607 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3608 {
3609         struct nfs4_stid *ret;
3610
3611         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3612         if (!ret)
3613                 return NULL;
3614         return delegstateid(ret);
3615 }
3616
3617 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3618 {
3619         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3620                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3621 }
3622
3623 static __be32
3624 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3625                 struct nfs4_delegation **dp)
3626 {
3627         int flags;
3628         __be32 status = nfserr_bad_stateid;
3629         struct nfs4_delegation *deleg;
3630
3631         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3632         if (deleg == NULL)
3633                 goto out;
3634         flags = share_access_to_flags(open->op_share_access);
3635         status = nfs4_check_delegmode(deleg, flags);
3636         if (status) {
3637                 nfs4_put_stid(&deleg->dl_stid);
3638                 goto out;
3639         }
3640         *dp = deleg;
3641 out:
3642         if (!nfsd4_is_deleg_cur(open))
3643                 return nfs_ok;
3644         if (status)
3645                 return status;
3646         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3647         return nfs_ok;
3648 }
3649
3650 static struct nfs4_ol_stateid *
3651 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3652 {
3653         struct nfs4_ol_stateid *local, *ret = NULL;
3654         struct nfs4_openowner *oo = open->op_openowner;
3655
3656         spin_lock(&fp->fi_lock);
3657         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3658                 /* ignore lock owners */
3659                 if (local->st_stateowner->so_is_open_owner == 0)
3660                         continue;
3661                 if (local->st_stateowner == &oo->oo_owner) {
3662                         ret = local;
3663                         atomic_inc(&ret->st_stid.sc_count);
3664                         break;
3665                 }
3666         }
3667         spin_unlock(&fp->fi_lock);
3668         return ret;
3669 }
3670
3671 static inline int nfs4_access_to_access(u32 nfs4_access)
3672 {
3673         int flags = 0;
3674
3675         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3676                 flags |= NFSD_MAY_READ;
3677         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3678                 flags |= NFSD_MAY_WRITE;
3679         return flags;
3680 }
3681
3682 static inline __be32
3683 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3684                 struct nfsd4_open *open)
3685 {
3686         struct iattr iattr = {
3687                 .ia_valid = ATTR_SIZE,
3688                 .ia_size = 0,
3689         };
3690         if (!open->op_truncate)
3691                 return 0;
3692         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3693                 return nfserr_inval;
3694         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3695 }
3696
3697 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3698                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3699                 struct nfsd4_open *open)
3700 {
3701         struct file *filp = NULL;
3702         __be32 status;
3703         int oflag = nfs4_access_to_omode(open->op_share_access);
3704         int access = nfs4_access_to_access(open->op_share_access);
3705         unsigned char old_access_bmap, old_deny_bmap;
3706
3707         spin_lock(&fp->fi_lock);
3708
3709         /*
3710          * Are we trying to set a deny mode that would conflict with
3711          * current access?
3712          */
3713         status = nfs4_file_check_deny(fp, open->op_share_deny);
3714         if (status != nfs_ok) {
3715                 spin_unlock(&fp->fi_lock);
3716                 goto out;
3717         }
3718
3719         /* set access to the file */
3720         status = nfs4_file_get_access(fp, open->op_share_access);
3721         if (status != nfs_ok) {
3722                 spin_unlock(&fp->fi_lock);
3723                 goto out;
3724         }
3725
3726         /* Set access bits in stateid */
3727         old_access_bmap = stp->st_access_bmap;
3728         set_access(open->op_share_access, stp);
3729
3730         /* Set new deny mask */
3731         old_deny_bmap = stp->st_deny_bmap;
3732         set_deny(open->op_share_deny, stp);
3733         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3734
3735         if (!fp->fi_fds[oflag]) {
3736                 spin_unlock(&fp->fi_lock);
3737                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3738                 if (status)
3739                         goto out_put_access;
3740                 spin_lock(&fp->fi_lock);
3741                 if (!fp->fi_fds[oflag]) {
3742                         fp->fi_fds[oflag] = filp;
3743                         filp = NULL;
3744                 }
3745         }
3746         spin_unlock(&fp->fi_lock);
3747         if (filp)
3748                 fput(filp);
3749
3750         status = nfsd4_truncate(rqstp, cur_fh, open);
3751         if (status)
3752                 goto out_put_access;
3753 out:
3754         return status;
3755 out_put_access:
3756         stp->st_access_bmap = old_access_bmap;
3757         nfs4_file_put_access(fp, open->op_share_access);
3758         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3759         goto out;
3760 }
3761
3762 static __be32
3763 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)
3764 {
3765         __be32 status;
3766         unsigned char old_deny_bmap;
3767
3768         if (!test_access(open->op_share_access, stp))
3769                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3770
3771         /* test and set deny mode */
3772         spin_lock(&fp->fi_lock);
3773         status = nfs4_file_check_deny(fp, open->op_share_deny);
3774         if (status == nfs_ok) {
3775                 old_deny_bmap = stp->st_deny_bmap;
3776                 set_deny(open->op_share_deny, stp);
3777                 fp->fi_share_deny |=
3778                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3779         }
3780         spin_unlock(&fp->fi_lock);
3781
3782         if (status != nfs_ok)
3783                 return status;
3784
3785         status = nfsd4_truncate(rqstp, cur_fh, open);
3786         if (status != nfs_ok)
3787                 reset_union_bmap_deny(old_deny_bmap, stp);
3788         return status;
3789 }
3790
3791 static void
3792 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3793 {
3794         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3795 }
3796
3797 /* Should we give out recallable state?: */
3798 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3799 {
3800         if (clp->cl_cb_state == NFSD4_CB_UP)
3801                 return true;
3802         /*
3803          * In the sessions case, since we don't have to establish a
3804          * separate connection for callbacks, we assume it's OK
3805          * until we hear otherwise:
3806          */
3807         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3808 }
3809
3810 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
3811 {
3812         struct file_lock *fl;
3813
3814         fl = locks_alloc_lock();
3815         if (!fl)
3816                 return NULL;
3817         fl->fl_lmops = &nfsd_lease_mng_ops;
3818         fl->fl_flags = FL_DELEG;
3819         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3820         fl->fl_end = OFFSET_MAX;
3821         fl->fl_owner = (fl_owner_t)fp;
3822         fl->fl_pid = current->tgid;
3823         return fl;
3824 }
3825
3826 static int nfs4_setlease(struct nfs4_delegation *dp)
3827 {
3828         struct nfs4_file *fp = dp->dl_stid.sc_file;
3829         struct file_lock *fl, *ret;
3830         struct file *filp;
3831         int status = 0;
3832
3833         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
3834         if (!fl)
3835                 return -ENOMEM;
3836         filp = find_readable_file(fp);
3837         if (!filp) {
3838                 /* We should always have a readable file here */
3839                 WARN_ON_ONCE(1);
3840                 return -EBADF;
3841         }
3842         fl->fl_file = filp;
3843         ret = fl;
3844         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
3845         if (fl)
3846                 locks_free_lock(fl);
3847         if (status)
3848                 goto out_fput;
3849         spin_lock(&state_lock);
3850         spin_lock(&fp->fi_lock);
3851         /* Did the lease get broken before we took the lock? */
3852         status = -EAGAIN;
3853         if (fp->fi_had_conflict)
3854                 goto out_unlock;
3855         /* Race breaker */
3856         if (fp->fi_deleg_file) {
3857                 status = 0;
3858                 atomic_inc(&fp->fi_delegees);
3859                 hash_delegation_locked(dp, fp);
3860                 goto out_unlock;
3861         }
3862         fp->fi_deleg_file = filp;
3863         atomic_set(&fp->fi_delegees, 1);
3864         hash_delegation_locked(dp, fp);
3865         spin_unlock(&fp->fi_lock);
3866         spin_unlock(&state_lock);
3867         return 0;
3868 out_unlock:
3869         spin_unlock(&fp->fi_lock);
3870         spin_unlock(&state_lock);
3871 out_fput:
3872         fput(filp);
3873         return status;
3874 }
3875
3876 static struct nfs4_delegation *
3877 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
3878                     struct nfs4_file *fp)
3879 {
3880         int status;
3881         struct nfs4_delegation *dp;
3882
3883         if (fp->fi_had_conflict)
3884                 return ERR_PTR(-EAGAIN);
3885
3886         dp = alloc_init_deleg(clp, fh);
3887         if (!dp)
3888                 return ERR_PTR(-ENOMEM);
3889
3890         get_nfs4_file(fp);
3891         spin_lock(&state_lock);
3892         spin_lock(&fp->fi_lock);
3893         dp->dl_stid.sc_file = fp;
3894         if (!fp->fi_deleg_file) {
3895                 spin_unlock(&fp->fi_lock);
3896                 spin_unlock(&state_lock);
3897                 status = nfs4_setlease(dp);
3898                 goto out;
3899         }
3900         atomic_inc(&fp->fi_delegees);
3901         if (fp->fi_had_conflict) {
3902                 status = -EAGAIN;
3903                 goto out_unlock;
3904         }
3905         hash_delegation_locked(dp, fp);
3906         status = 0;
3907 out_unlock:
3908         spin_unlock(&fp->fi_lock);
3909         spin_unlock(&state_lock);
3910 out:
3911         if (status) {
3912                 nfs4_put_stid(&dp->dl_stid);
3913                 return ERR_PTR(status);
3914         }
3915         return dp;
3916 }
3917
3918 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3919 {
3920         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3921         if (status == -EAGAIN)
3922                 open->op_why_no_deleg = WND4_CONTENTION;
3923         else {
3924                 open->op_why_no_deleg = WND4_RESOURCE;
3925                 switch (open->op_deleg_want) {
3926                 case NFS4_SHARE_WANT_READ_DELEG:
3927                 case NFS4_SHARE_WANT_WRITE_DELEG:
3928                 case NFS4_SHARE_WANT_ANY_DELEG:
3929                         break;
3930                 case NFS4_SHARE_WANT_CANCEL:
3931                         open->op_why_no_deleg = WND4_CANCELLED;
3932                         break;
3933                 case NFS4_SHARE_WANT_NO_DELEG:
3934                         WARN_ON_ONCE(1);
3935                 }
3936         }
3937 }
3938
3939 /*
3940  * Attempt to hand out a delegation.
3941  *
3942  * Note we don't support write delegations, and won't until the vfs has
3943  * proper support for them.
3944  */
3945 static void
3946 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
3947                         struct nfs4_ol_stateid *stp)
3948 {
3949         struct nfs4_delegation *dp;
3950         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
3951         struct nfs4_client *clp = stp->st_stid.sc_client;
3952         int cb_up;
3953         int status = 0;
3954
3955         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3956         open->op_recall = 0;
3957         switch (open->op_claim_type) {
3958                 case NFS4_OPEN_CLAIM_PREVIOUS:
3959                         if (!cb_up)
3960                                 open->op_recall = 1;
3961                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3962                                 goto out_no_deleg;
3963                         break;
3964                 case NFS4_OPEN_CLAIM_NULL:
3965                 case NFS4_OPEN_CLAIM_FH:
3966                         /*
3967                          * Let's not give out any delegations till everyone's
3968                          * had the chance to reclaim theirs....
3969                          */
3970                         if (locks_in_grace(clp->net))
3971                                 goto out_no_deleg;
3972                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3973                                 goto out_no_deleg;
3974                         /*
3975                          * Also, if the file was opened for write or
3976                          * create, there's a good chance the client's
3977                          * about to write to it, resulting in an
3978                          * immediate recall (since we don't support
3979                          * write delegations):
3980                          */
3981                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3982                                 goto out_no_deleg;
3983                         if (open->op_create == NFS4_OPEN_CREATE)
3984                                 goto out_no_deleg;
3985                         break;
3986                 default:
3987                         goto out_no_deleg;
3988         }
3989         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file);
3990         if (IS_ERR(dp))
3991                 goto out_no_deleg;
3992
3993         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3994
3995         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3996                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3997         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3998         nfs4_put_stid(&dp->dl_stid);
3999         return;
4000 out_no_deleg:
4001         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4002         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4003             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4004                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4005                 open->op_recall = 1;
4006         }
4007
4008         /* 4.1 client asking for a delegation? */
4009         if (open->op_deleg_want)
4010                 nfsd4_open_deleg_none_ext(open, status);
4011         return;
4012 }
4013
4014 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4015                                         struct nfs4_delegation *dp)
4016 {
4017         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4018             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4019                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4020                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4021         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4022                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4023                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4024                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4025         }
4026         /* Otherwise the client must be confused wanting a delegation
4027          * it already has, therefore we don't return
4028          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4029          */
4030 }
4031
4032 __be32
4033 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4034 {
4035         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4036         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4037         struct nfs4_file *fp = NULL;
4038         struct nfs4_ol_stateid *stp = NULL;
4039         struct nfs4_delegation *dp = NULL;
4040         __be32 status;
4041
4042         /*
4043          * Lookup file; if found, lookup stateid and check open request,
4044          * and check for delegations in the process of being recalled.
4045          * If not found, create the nfs4_file struct
4046          */
4047         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4048         if (fp != open->op_file) {
4049                 status = nfs4_check_deleg(cl, open, &dp);
4050                 if (status)
4051                         goto out;
4052                 stp = nfsd4_find_existing_open(fp, open);
4053         } else {
4054                 open->op_file = NULL;
4055                 status = nfserr_bad_stateid;
4056                 if (nfsd4_is_deleg_cur(open))
4057                         goto out;
4058                 status = nfserr_jukebox;
4059         }
4060
4061         /*
4062          * OPEN the file, or upgrade an existing OPEN.
4063          * If truncate fails, the OPEN fails.
4064          */
4065         if (stp) {
4066                 /* Stateid was found, this is an OPEN upgrade */
4067                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4068                 if (status)
4069                         goto out;
4070         } else {
4071                 stp = open->op_stp;
4072                 open->op_stp = NULL;
4073                 init_open_stateid(stp, fp, open);
4074                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4075                 if (status) {
4076                         release_open_stateid(stp);
4077                         goto out;
4078                 }
4079         }
4080         update_stateid(&stp->st_stid.sc_stateid);
4081         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4082
4083         if (nfsd4_has_session(&resp->cstate)) {
4084                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4085                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4086                         open->op_why_no_deleg = WND4_NOT_WANTED;
4087                         goto nodeleg;
4088                 }
4089         }
4090
4091         /*
4092         * Attempt to hand out a delegation. No error return, because the
4093         * OPEN succeeds even if we fail.
4094         */
4095         nfs4_open_delegation(current_fh, open, stp);
4096 nodeleg:
4097         status = nfs_ok;
4098
4099         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4100                 STATEID_VAL(&stp->st_stid.sc_stateid));
4101 out:
4102         /* 4.1 client trying to upgrade/downgrade delegation? */
4103         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4104             open->op_deleg_want)
4105                 nfsd4_deleg_xgrade_none_ext(open, dp);
4106
4107         if (fp)
4108                 put_nfs4_file(fp);
4109         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4110                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
4111         /*
4112         * To finish the open response, we just need to set the rflags.
4113         */
4114         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4115         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4116             !nfsd4_has_session(&resp->cstate))
4117                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4118         if (dp)
4119                 nfs4_put_stid(&dp->dl_stid);
4120         if (stp)
4121                 nfs4_put_stid(&stp->st_stid);
4122
4123         return status;
4124 }
4125
4126 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4127                               struct nfsd4_open *open, __be32 status)
4128 {
4129         if (open->op_openowner) {
4130                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4131
4132                 nfsd4_cstate_assign_replay(cstate, so);
4133                 nfs4_put_stateowner(so);
4134         }
4135         if (open->op_file)
4136                 kmem_cache_free(file_slab, open->op_file);
4137         if (open->op_stp)
4138                 nfs4_put_stid(&open->op_stp->st_stid);
4139 }
4140
4141 __be32
4142 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4143             clientid_t *clid)
4144 {
4145         struct nfs4_client *clp;
4146         __be32 status;
4147         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4148
4149         dprintk("process_renew(%08x/%08x): starting\n",
4150                         clid->cl_boot, clid->cl_id);
4151         status = lookup_clientid(clid, cstate, nn);
4152         if (status)
4153                 goto out;
4154         clp = cstate->clp;
4155         status = nfserr_cb_path_down;
4156         if (!list_empty(&clp->cl_delegations)
4157                         && clp->cl_cb_state != NFSD4_CB_UP)
4158                 goto out;
4159         status = nfs_ok;
4160 out:
4161         return status;
4162 }
4163
4164 void
4165 nfsd4_end_grace(struct nfsd_net *nn)
4166 {
4167         /* do nothing if grace period already ended */
4168         if (nn->grace_ended)
4169                 return;
4170
4171         dprintk("NFSD: end of grace period\n");
4172         nn->grace_ended = true;
4173         /*
4174          * If the server goes down again right now, an NFSv4
4175          * client will still be allowed to reclaim after it comes back up,
4176          * even if it hasn't yet had a chance to reclaim state this time.
4177          *
4178          */
4179         nfsd4_record_grace_done(nn);
4180         /*
4181          * At this point, NFSv4 clients can still reclaim.  But if the
4182          * server crashes, any that have not yet reclaimed will be out
4183          * of luck on the next boot.
4184          *
4185          * (NFSv4.1+ clients are considered to have reclaimed once they
4186          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4187          * have reclaimed after their first OPEN.)
4188          */
4189         locks_end_grace(&nn->nfsd4_manager);
4190         /*
4191          * At this point, and once lockd and/or any other containers
4192          * exit their grace period, further reclaims will fail and
4193          * regular locking can resume.
4194          */
4195 }
4196
4197 static time_t
4198 nfs4_laundromat(struct nfsd_net *nn)
4199 {
4200         struct nfs4_client *clp;
4201         struct nfs4_openowner *oo;
4202         struct nfs4_delegation *dp;
4203         struct nfs4_ol_stateid *stp;
4204         struct list_head *pos, *next, reaplist;
4205         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4206         time_t t, new_timeo = nn->nfsd4_lease;
4207
4208         dprintk("NFSD: laundromat service - starting\n");
4209         nfsd4_end_grace(nn);
4210         INIT_LIST_HEAD(&reaplist);
4211         spin_lock(&nn->client_lock);
4212         list_for_each_safe(pos, next, &nn->client_lru) {
4213                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4214                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4215                         t = clp->cl_time - cutoff;
4216                         new_timeo = min(new_timeo, t);
4217                         break;
4218                 }
4219                 if (mark_client_expired_locked(clp)) {
4220                         dprintk("NFSD: client in use (clientid %08x)\n",
4221                                 clp->cl_clientid.cl_id);
4222                         continue;
4223                 }
4224                 list_add(&clp->cl_lru, &reaplist);
4225         }
4226         spin_unlock(&nn->client_lock);
4227         list_for_each_safe(pos, next, &reaplist) {
4228                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4229                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4230                         clp->cl_clientid.cl_id);
4231                 list_del_init(&clp->cl_lru);
4232                 expire_client(clp);
4233         }
4234         spin_lock(&state_lock);
4235         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4236                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4237                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
4238                         continue;
4239                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4240                         t = dp->dl_time - cutoff;
4241                         new_timeo = min(new_timeo, t);
4242                         break;
4243                 }
4244                 unhash_delegation_locked(dp);
4245                 list_add(&dp->dl_recall_lru, &reaplist);
4246         }
4247         spin_unlock(&state_lock);
4248         while (!list_empty(&reaplist)) {
4249                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4250                                         dl_recall_lru);
4251                 list_del_init(&dp->dl_recall_lru);
4252                 revoke_delegation(dp);
4253         }
4254
4255         spin_lock(&nn->client_lock);
4256         while (!list_empty(&nn->close_lru)) {
4257                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4258                                         oo_close_lru);
4259                 if (time_after((unsigned long)oo->oo_time,
4260                                (unsigned long)cutoff)) {
4261                         t = oo->oo_time - cutoff;
4262                         new_timeo = min(new_timeo, t);
4263                         break;
4264                 }
4265                 list_del_init(&oo->oo_close_lru);
4266                 stp = oo->oo_last_closed_stid;
4267                 oo->oo_last_closed_stid = NULL;
4268                 spin_unlock(&nn->client_lock);
4269                 nfs4_put_stid(&stp->st_stid);
4270                 spin_lock(&nn->client_lock);
4271         }
4272         spin_unlock(&nn->client_lock);
4273
4274         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4275         return new_timeo;
4276 }
4277
4278 static struct workqueue_struct *laundry_wq;
4279 static void laundromat_main(struct work_struct *);
4280
4281 static void
4282 laundromat_main(struct work_struct *laundry)
4283 {
4284         time_t t;
4285         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
4286                                                   work);
4287         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4288                                            laundromat_work);
4289
4290         t = nfs4_laundromat(nn);
4291         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4292         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4293 }
4294
4295 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
4296 {
4297         if (!nfsd_fh_match(&fhp->fh_handle, &stp->st_stid.sc_file->fi_fhandle))
4298                 return nfserr_bad_stateid;
4299         return nfs_ok;
4300 }
4301
4302 static inline int
4303 access_permit_read(struct nfs4_ol_stateid *stp)
4304 {
4305         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4306                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4307                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4308 }
4309
4310 static inline int
4311 access_permit_write(struct nfs4_ol_stateid *stp)
4312 {
4313         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4314                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4315 }
4316
4317 static
4318 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4319 {
4320         __be32 status = nfserr_openmode;
4321
4322         /* For lock stateid's, we test the parent open, not the lock: */
4323         if (stp->st_openstp)
4324                 stp = stp->st_openstp;
4325         if ((flags & WR_STATE) && !access_permit_write(stp))
4326                 goto out;
4327         if ((flags & RD_STATE) && !access_permit_read(stp))
4328                 goto out;
4329         status = nfs_ok;
4330 out:
4331         return status;
4332 }
4333
4334 static inline __be32
4335 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4336 {
4337         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4338                 return nfs_ok;
4339         else if (locks_in_grace(net)) {
4340                 /* Answer in remaining cases depends on existence of
4341                  * conflicting state; so we must wait out the grace period. */
4342                 return nfserr_grace;
4343         } else if (flags & WR_STATE)
4344                 return nfs4_share_conflict(current_fh,
4345                                 NFS4_SHARE_DENY_WRITE);
4346         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4347                 return nfs4_share_conflict(current_fh,
4348                                 NFS4_SHARE_DENY_READ);
4349 }
4350
4351 /*
4352  * Allow READ/WRITE during grace period on recovered state only for files
4353  * that are not able to provide mandatory locking.
4354  */
4355 static inline int
4356 grace_disallows_io(struct net *net, struct inode *inode)
4357 {
4358         return locks_in_grace(net) && mandatory_lock(inode);
4359 }
4360
4361 /* Returns true iff a is later than b: */
4362 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4363 {
4364         return (s32)(a->si_generation - b->si_generation) > 0;
4365 }
4366
4367 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4368 {
4369         /*
4370          * When sessions are used the stateid generation number is ignored
4371          * when it is zero.
4372          */
4373         if (has_session && in->si_generation == 0)
4374                 return nfs_ok;
4375
4376         if (in->si_generation == ref->si_generation)
4377                 return nfs_ok;
4378
4379         /* If the client sends us a stateid from the future, it's buggy: */
4380         if (stateid_generation_after(in, ref))
4381                 return nfserr_bad_stateid;
4382         /*
4383          * However, we could see a stateid from the past, even from a
4384          * non-buggy client.  For example, if the client sends a lock
4385          * while some IO is outstanding, the lock may bump si_generation
4386          * while the IO is still in flight.  The client could avoid that
4387          * situation by waiting for responses on all the IO requests,
4388          * but better performance may result in retrying IO that
4389          * receives an old_stateid error if requests are rarely
4390          * reordered in flight:
4391          */
4392         return nfserr_old_stateid;
4393 }
4394
4395 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4396 {
4397         struct nfs4_stid *s;
4398         struct nfs4_ol_stateid *ols;
4399         __be32 status = nfserr_bad_stateid;
4400
4401         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4402                 return status;
4403         /* Client debugging aid. */
4404         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4405                 char addr_str[INET6_ADDRSTRLEN];
4406                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4407                                  sizeof(addr_str));
4408                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4409                                         "with incorrect client ID\n", addr_str);
4410                 return status;
4411         }
4412         spin_lock(&cl->cl_lock);
4413         s = find_stateid_locked(cl, stateid);
4414         if (!s)
4415                 goto out_unlock;
4416         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4417         if (status)
4418                 goto out_unlock;
4419         switch (s->sc_type) {
4420         case NFS4_DELEG_STID:
4421                 status = nfs_ok;
4422                 break;
4423         case NFS4_REVOKED_DELEG_STID:
4424                 status = nfserr_deleg_revoked;
4425                 break;
4426         case NFS4_OPEN_STID:
4427         case NFS4_LOCK_STID:
4428                 ols = openlockstateid(s);
4429                 if (ols->st_stateowner->so_is_open_owner
4430                                 && !(openowner(ols->st_stateowner)->oo_flags
4431                                                 & NFS4_OO_CONFIRMED))
4432                         status = nfserr_bad_stateid;
4433                 else
4434                         status = nfs_ok;
4435                 break;
4436         default:
4437                 printk("unknown stateid type %x\n", s->sc_type);
4438                 /* Fallthrough */
4439         case NFS4_CLOSED_STID:
4440         case NFS4_CLOSED_DELEG_STID:
4441                 status = nfserr_bad_stateid;
4442         }
4443 out_unlock:
4444         spin_unlock(&cl->cl_lock);
4445         return status;
4446 }
4447
4448 static __be32
4449 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4450                      stateid_t *stateid, unsigned char typemask,
4451                      struct nfs4_stid **s, struct nfsd_net *nn)
4452 {
4453         __be32 status;
4454
4455         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4456                 return nfserr_bad_stateid;
4457         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4458         if (status == nfserr_stale_clientid) {
4459                 if (cstate->session)
4460                         return nfserr_bad_stateid;
4461                 return nfserr_stale_stateid;
4462         }
4463         if (status)
4464                 return status;
4465         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4466         if (!*s)
4467                 return nfserr_bad_stateid;
4468         return nfs_ok;
4469 }
4470
4471 /*
4472 * Checks for stateid operations
4473 */
4474 __be32
4475 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
4476                            stateid_t *stateid, int flags, struct file **filpp)
4477 {
4478         struct nfs4_stid *s;
4479         struct nfs4_ol_stateid *stp = NULL;
4480         struct nfs4_delegation *dp = NULL;
4481         struct svc_fh *current_fh = &cstate->current_fh;
4482         struct inode *ino = current_fh->fh_dentry->d_inode;
4483         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4484         struct file *file = NULL;
4485         __be32 status;
4486
4487         if (filpp)
4488                 *filpp = NULL;
4489
4490         if (grace_disallows_io(net, ino))
4491                 return nfserr_grace;
4492
4493         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4494                 return check_special_stateids(net, current_fh, stateid, flags);
4495
4496         status = nfsd4_lookup_stateid(cstate, stateid,
4497                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4498                                 &s, nn);
4499         if (status)
4500                 return status;
4501         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
4502         if (status)
4503                 goto out;
4504         switch (s->sc_type) {
4505         case NFS4_DELEG_STID:
4506                 dp = delegstateid(s);
4507                 status = nfs4_check_delegmode(dp, flags);
4508                 if (status)
4509                         goto out;
4510                 if (filpp) {
4511                         file = dp->dl_stid.sc_file->fi_deleg_file;
4512                         if (!file) {
4513                                 WARN_ON_ONCE(1);
4514                                 status = nfserr_serverfault;
4515                                 goto out;
4516                         }
4517                         get_file(file);
4518                 }
4519                 break;
4520         case NFS4_OPEN_STID:
4521         case NFS4_LOCK_STID:
4522                 stp = openlockstateid(s);
4523                 status = nfs4_check_fh(current_fh, stp);
4524                 if (status)
4525                         goto out;
4526                 if (stp->st_stateowner->so_is_open_owner
4527                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4528                         goto out;
4529                 status = nfs4_check_openmode(stp, flags);
4530                 if (status)
4531                         goto out;
4532                 if (filpp) {
4533                         struct nfs4_file *fp = stp->st_stid.sc_file;
4534
4535                         if (flags & RD_STATE)
4536                                 file = find_readable_file(fp);
4537                         else
4538                                 file = find_writeable_file(fp);
4539                 }
4540                 break;
4541         default:
4542                 status = nfserr_bad_stateid;
4543                 goto out;
4544         }
4545         status = nfs_ok;
4546         if (file)
4547                 *filpp = file;
4548 out:
4549         nfs4_put_stid(s);
4550         return status;
4551 }
4552
4553 /*
4554  * Test if the stateid is valid
4555  */
4556 __be32
4557 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4558                    struct nfsd4_test_stateid *test_stateid)
4559 {
4560         struct nfsd4_test_stateid_id *stateid;
4561         struct nfs4_client *cl = cstate->session->se_client;
4562
4563         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4564                 stateid->ts_id_status =
4565                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4566
4567         return nfs_ok;
4568 }
4569
4570 __be32
4571 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4572                    struct nfsd4_free_stateid *free_stateid)
4573 {
4574         stateid_t *stateid = &free_stateid->fr_stateid;
4575         struct nfs4_stid *s;
4576         struct nfs4_delegation *dp;
4577         struct nfs4_ol_stateid *stp;
4578         struct nfs4_client *cl = cstate->session->se_client;
4579         __be32 ret = nfserr_bad_stateid;
4580
4581         spin_lock(&cl->cl_lock);
4582         s = find_stateid_locked(cl, stateid);
4583         if (!s)
4584                 goto out_unlock;
4585         switch (s->sc_type) {
4586         case NFS4_DELEG_STID:
4587                 ret = nfserr_locks_held;
4588                 break;
4589         case NFS4_OPEN_STID:
4590                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4591                 if (ret)
4592                         break;
4593                 ret = nfserr_locks_held;
4594                 break;
4595         case NFS4_LOCK_STID:
4596                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4597                 if (ret)
4598                         break;
4599                 stp = openlockstateid(s);
4600                 ret = nfserr_locks_held;
4601                 if (check_for_locks(stp->st_stid.sc_file,
4602                                     lockowner(stp->st_stateowner)))
4603                         break;
4604                 unhash_lock_stateid(stp);
4605                 spin_unlock(&cl->cl_lock);
4606                 nfs4_put_stid(s);
4607                 ret = nfs_ok;
4608                 goto out;
4609         case NFS4_REVOKED_DELEG_STID:
4610                 dp = delegstateid(s);
4611                 list_del_init(&dp->dl_recall_lru);
4612                 spin_unlock(&cl->cl_lock);
4613                 nfs4_put_stid(s);
4614                 ret = nfs_ok;
4615                 goto out;
4616         /* Default falls through and returns nfserr_bad_stateid */
4617         }
4618 out_unlock:
4619         spin_unlock(&cl->cl_lock);
4620 out:
4621         return ret;
4622 }
4623
4624 static inline int
4625 setlkflg (int type)
4626 {
4627         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4628                 RD_STATE : WR_STATE;
4629 }
4630
4631 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4632 {
4633         struct svc_fh *current_fh = &cstate->current_fh;
4634         struct nfs4_stateowner *sop = stp->st_stateowner;
4635         __be32 status;
4636
4637         status = nfsd4_check_seqid(cstate, sop, seqid);
4638         if (status)
4639                 return status;
4640         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4641                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4642                 /*
4643                  * "Closed" stateid's exist *only* to return
4644                  * nfserr_replay_me from the previous step, and
4645                  * revoked delegations are kept only for free_stateid.
4646                  */
4647                 return nfserr_bad_stateid;
4648         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4649         if (status)
4650                 return status;
4651         return nfs4_check_fh(current_fh, stp);
4652 }
4653
4654 /*
4655  * Checks for sequence id mutating operations.
4656  */
4657 static __be32
4658 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4659                          stateid_t *stateid, char typemask,
4660                          struct nfs4_ol_stateid **stpp,
4661                          struct nfsd_net *nn)
4662 {
4663         __be32 status;
4664         struct nfs4_stid *s;
4665         struct nfs4_ol_stateid *stp = NULL;
4666
4667         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4668                 seqid, STATEID_VAL(stateid));
4669
4670         *stpp = NULL;
4671         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4672         if (status)
4673                 return status;
4674         stp = openlockstateid(s);
4675         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
4676
4677         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4678         if (!status)
4679                 *stpp = stp;
4680         else
4681                 nfs4_put_stid(&stp->st_stid);
4682         return status;
4683 }
4684
4685 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4686                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4687 {
4688         __be32 status;
4689         struct nfs4_openowner *oo;
4690         struct nfs4_ol_stateid *stp;
4691
4692         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4693                                                 NFS4_OPEN_STID, &stp, nn);
4694         if (status)
4695                 return status;
4696         oo = openowner(stp->st_stateowner);
4697         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4698                 nfs4_put_stid(&stp->st_stid);
4699                 return nfserr_bad_stateid;
4700         }
4701         *stpp = stp;
4702         return nfs_ok;
4703 }
4704
4705 __be32
4706 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4707                    struct nfsd4_open_confirm *oc)
4708 {
4709         __be32 status;
4710         struct nfs4_openowner *oo;
4711         struct nfs4_ol_stateid *stp;
4712         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4713
4714         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4715                         cstate->current_fh.fh_dentry);
4716
4717         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4718         if (status)
4719                 return status;
4720
4721         status = nfs4_preprocess_seqid_op(cstate,
4722                                         oc->oc_seqid, &oc->oc_req_stateid,
4723                                         NFS4_OPEN_STID, &stp, nn);
4724         if (status)
4725                 goto out;
4726         oo = openowner(stp->st_stateowner);
4727         status = nfserr_bad_stateid;
4728         if (oo->oo_flags & NFS4_OO_CONFIRMED)
4729                 goto put_stateid;
4730         oo->oo_flags |= NFS4_OO_CONFIRMED;
4731         update_stateid(&stp->st_stid.sc_stateid);
4732         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4733         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
4734                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
4735
4736         nfsd4_client_record_create(oo->oo_owner.so_client);
4737         status = nfs_ok;
4738 put_stateid:
4739         nfs4_put_stid(&stp->st_stid);
4740 out:
4741         nfsd4_bump_seqid(cstate, status);
4742         return status;
4743 }
4744
4745 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
4746 {
4747         if (!test_access(access, stp))
4748                 return;
4749         nfs4_file_put_access(stp->st_stid.sc_file, access);
4750         clear_access(access, stp);
4751 }
4752
4753 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
4754 {
4755         switch (to_access) {
4756         case NFS4_SHARE_ACCESS_READ:
4757                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
4758                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4759                 break;
4760         case NFS4_SHARE_ACCESS_WRITE:
4761                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
4762                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4763                 break;
4764         case NFS4_SHARE_ACCESS_BOTH:
4765                 break;
4766         default:
4767                 WARN_ON_ONCE(1);
4768         }
4769 }
4770
4771 __be32
4772 nfsd4_open_downgrade(struct svc_rqst *rqstp,
4773                      struct nfsd4_compound_state *cstate,
4774                      struct nfsd4_open_downgrade *od)
4775 {
4776         __be32 status;
4777         struct nfs4_ol_stateid *stp;
4778         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4779
4780         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
4781                         cstate->current_fh.fh_dentry);
4782
4783         /* We don't yet support WANT bits: */
4784         if (od->od_deleg_want)
4785                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
4786                         od->od_deleg_want);
4787
4788         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
4789                                         &od->od_stateid, &stp, nn);
4790         if (status)
4791                 goto out;
4792         status = nfserr_inval;
4793         if (!test_access(od->od_share_access, stp)) {
4794                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
4795                         stp->st_access_bmap, od->od_share_access);
4796                 goto put_stateid;
4797         }
4798         if (!test_deny(od->od_share_deny, stp)) {
4799                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
4800                         stp->st_deny_bmap, od->od_share_deny);
4801                 goto put_stateid;
4802         }
4803         nfs4_stateid_downgrade(stp, od->od_share_access);
4804
4805         reset_union_bmap_deny(od->od_share_deny, stp);
4806
4807         update_stateid(&stp->st_stid.sc_stateid);
4808         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4809         status = nfs_ok;
4810 put_stateid:
4811         nfs4_put_stid(&stp->st_stid);
4812 out:
4813         nfsd4_bump_seqid(cstate, status);
4814         return status;
4815 }
4816
4817 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
4818 {
4819         struct nfs4_client *clp = s->st_stid.sc_client;
4820         LIST_HEAD(reaplist);
4821
4822         s->st_stid.sc_type = NFS4_CLOSED_STID;
4823         spin_lock(&clp->cl_lock);
4824         unhash_open_stateid(s, &reaplist);
4825
4826         if (clp->cl_minorversion) {
4827                 put_ol_stateid_locked(s, &reaplist);
4828                 spin_unlock(&clp->cl_lock);
4829                 free_ol_stateid_reaplist(&reaplist);
4830         } else {
4831                 spin_unlock(&clp->cl_lock);
4832                 free_ol_stateid_reaplist(&reaplist);
4833                 move_to_close_lru(s, clp->net);
4834         }
4835 }
4836
4837 /*
4838  * nfs4_unlock_state() called after encode
4839  */
4840 __be32
4841 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4842             struct nfsd4_close *close)
4843 {
4844         __be32 status;
4845         struct nfs4_ol_stateid *stp;
4846         struct net *net = SVC_NET(rqstp);
4847         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4848
4849         dprintk("NFSD: nfsd4_close on file %pd\n",
4850                         cstate->current_fh.fh_dentry);
4851
4852         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4853                                         &close->cl_stateid,
4854                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4855                                         &stp, nn);
4856         nfsd4_bump_seqid(cstate, status);
4857         if (status)
4858                 goto out;
4859         update_stateid(&stp->st_stid.sc_stateid);
4860         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4861
4862         nfsd4_close_open_stateid(stp);
4863
4864         /* put reference from nfs4_preprocess_seqid_op */
4865         nfs4_put_stid(&stp->st_stid);
4866 out:
4867         return status;
4868 }
4869
4870 __be32
4871 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4872                   struct nfsd4_delegreturn *dr)
4873 {
4874         struct nfs4_delegation *dp;
4875         stateid_t *stateid = &dr->dr_stateid;
4876         struct nfs4_stid *s;
4877         __be32 status;
4878         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4879
4880         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4881                 return status;
4882
4883         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
4884         if (status)
4885                 goto out;
4886         dp = delegstateid(s);
4887         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4888         if (status)
4889                 goto put_stateid;
4890
4891         destroy_delegation(dp);
4892 put_stateid:
4893         nfs4_put_stid(&dp->dl_stid);
4894 out:
4895         return status;
4896 }
4897
4898
4899 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4900
4901 static inline u64
4902 end_offset(u64 start, u64 len)
4903 {
4904         u64 end;
4905
4906         end = start + len;
4907         return end >= start ? end: NFS4_MAX_UINT64;
4908 }
4909
4910 /* last octet in a range */
4911 static inline u64
4912 last_byte_offset(u64 start, u64 len)
4913 {
4914         u64 end;
4915
4916         WARN_ON_ONCE(!len);
4917         end = start + len;
4918         return end > start ? end - 1: NFS4_MAX_UINT64;
4919 }
4920
4921 /*
4922  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4923  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4924  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4925  * locking, this prevents us from being completely protocol-compliant.  The
4926  * real solution to this problem is to start using unsigned file offsets in
4927  * the VFS, but this is a very deep change!
4928  */
4929 static inline void
4930 nfs4_transform_lock_offset(struct file_lock *lock)
4931 {
4932         if (lock->fl_start < 0)
4933                 lock->fl_start = OFFSET_MAX;
4934         if (lock->fl_end < 0)
4935                 lock->fl_end = OFFSET_MAX;
4936 }
4937
4938 static void nfsd4_fl_get_owner(struct file_lock *dst, struct file_lock *src)
4939 {
4940         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)src->fl_owner;
4941         dst->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lo->lo_owner));
4942 }
4943
4944 static void nfsd4_fl_put_owner(struct file_lock *fl)
4945 {
4946         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
4947
4948         if (lo) {
4949                 nfs4_put_stateowner(&lo->lo_owner);
4950                 fl->fl_owner = NULL;
4951         }
4952 }
4953
4954 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4955         .lm_get_owner = nfsd4_fl_get_owner,
4956         .lm_put_owner = nfsd4_fl_put_owner,
4957 };
4958
4959 static inline void
4960 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4961 {
4962         struct nfs4_lockowner *lo;
4963
4964         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4965                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4966                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4967                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4968                 if (!deny->ld_owner.data)
4969                         /* We just don't care that much */
4970                         goto nevermind;
4971                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4972                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4973         } else {
4974 nevermind:
4975                 deny->ld_owner.len = 0;
4976                 deny->ld_owner.data = NULL;
4977                 deny->ld_clientid.cl_boot = 0;
4978                 deny->ld_clientid.cl_id = 0;
4979         }
4980         deny->ld_start = fl->fl_start;
4981         deny->ld_length = NFS4_MAX_UINT64;
4982         if (fl->fl_end != NFS4_MAX_UINT64)
4983                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4984         deny->ld_type = NFS4_READ_LT;
4985         if (fl->fl_type != F_RDLCK)
4986                 deny->ld_type = NFS4_WRITE_LT;
4987 }
4988
4989 static struct nfs4_lockowner *
4990 find_lockowner_str_locked(clientid_t *clid, struct xdr_netobj *owner,
4991                 struct nfs4_client *clp)
4992 {
4993         unsigned int strhashval = ownerstr_hashval(owner);
4994         struct nfs4_stateowner *so;
4995
4996         lockdep_assert_held(&clp->cl_lock);
4997
4998         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
4999                             so_strhash) {
5000                 if (so->so_is_open_owner)
5001                         continue;
5002                 if (same_owner_str(so, owner))
5003                         return lockowner(nfs4_get_stateowner(so));
5004         }
5005         return NULL;
5006 }
5007
5008 static struct nfs4_lockowner *
5009 find_lockowner_str(clientid_t *clid, struct xdr_netobj *owner,
5010                 struct nfs4_client *clp)
5011 {
5012         struct nfs4_lockowner *lo;
5013
5014         spin_lock(&clp->cl_lock);
5015         lo = find_lockowner_str_locked(clid, owner, clp);
5016         spin_unlock(&clp->cl_lock);
5017         return lo;
5018 }
5019
5020 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5021 {
5022         unhash_lockowner_locked(lockowner(sop));
5023 }
5024
5025 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5026 {
5027         struct nfs4_lockowner *lo = lockowner(sop);
5028
5029         kmem_cache_free(lockowner_slab, lo);
5030 }
5031
5032 static const struct nfs4_stateowner_operations lockowner_ops = {
5033         .so_unhash =    nfs4_unhash_lockowner,
5034         .so_free =      nfs4_free_lockowner,
5035 };
5036
5037 /*
5038  * Alloc a lock owner structure.
5039  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5040  * occurred.
5041  *
5042  * strhashval = ownerstr_hashval
5043  */
5044 static struct nfs4_lockowner *
5045 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5046                            struct nfs4_ol_stateid *open_stp,
5047                            struct nfsd4_lock *lock)
5048 {
5049         struct nfs4_lockowner *lo, *ret;
5050
5051         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5052         if (!lo)
5053                 return NULL;
5054         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5055         lo->lo_owner.so_is_open_owner = 0;
5056         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5057         lo->lo_owner.so_ops = &lockowner_ops;
5058         spin_lock(&clp->cl_lock);
5059         ret = find_lockowner_str_locked(&clp->cl_clientid,
5060                         &lock->lk_new_owner, clp);
5061         if (ret == NULL) {
5062                 list_add(&lo->lo_owner.so_strhash,
5063                          &clp->cl_ownerstr_hashtbl[strhashval]);
5064                 ret = lo;
5065         } else
5066                 nfs4_free_lockowner(&lo->lo_owner);
5067         spin_unlock(&clp->cl_lock);
5068         return lo;
5069 }
5070
5071 static void
5072 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5073                   struct nfs4_file *fp, struct inode *inode,
5074                   struct nfs4_ol_stateid *open_stp)
5075 {
5076         struct nfs4_client *clp = lo->lo_owner.so_client;
5077
5078         lockdep_assert_held(&clp->cl_lock);
5079
5080         atomic_inc(&stp->st_stid.sc_count);
5081         stp->st_stid.sc_type = NFS4_LOCK_STID;
5082         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5083         get_nfs4_file(fp);
5084         stp->st_stid.sc_file = fp;
5085         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5086         stp->st_access_bmap = 0;
5087         stp->st_deny_bmap = open_stp->st_deny_bmap;
5088         stp->st_openstp = open_stp;
5089         list_add(&stp->st_locks, &open_stp->st_locks);
5090         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5091         spin_lock(&fp->fi_lock);
5092         list_add(&stp->st_perfile, &fp->fi_stateids);
5093         spin_unlock(&fp->fi_lock);
5094 }
5095
5096 static struct nfs4_ol_stateid *
5097 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5098 {
5099         struct nfs4_ol_stateid *lst;
5100         struct nfs4_client *clp = lo->lo_owner.so_client;
5101
5102         lockdep_assert_held(&clp->cl_lock);
5103
5104         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5105                 if (lst->st_stid.sc_file == fp) {
5106                         atomic_inc(&lst->st_stid.sc_count);
5107                         return lst;
5108                 }
5109         }
5110         return NULL;
5111 }
5112
5113 static struct nfs4_ol_stateid *
5114 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5115                             struct inode *inode, struct nfs4_ol_stateid *ost,
5116                             bool *new)
5117 {
5118         struct nfs4_stid *ns = NULL;
5119         struct nfs4_ol_stateid *lst;
5120         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5121         struct nfs4_client *clp = oo->oo_owner.so_client;
5122
5123         spin_lock(&clp->cl_lock);
5124         lst = find_lock_stateid(lo, fi);
5125         if (lst == NULL) {
5126                 spin_unlock(&clp->cl_lock);
5127                 ns = nfs4_alloc_stid(clp, stateid_slab);
5128                 if (ns == NULL)
5129                         return NULL;
5130
5131                 spin_lock(&clp->cl_lock);
5132                 lst = find_lock_stateid(lo, fi);
5133                 if (likely(!lst)) {
5134                         lst = openlockstateid(ns);
5135                         init_lock_stateid(lst, lo, fi, inode, ost);
5136                         ns = NULL;
5137                         *new = true;
5138                 }
5139         }
5140         spin_unlock(&clp->cl_lock);
5141         if (ns)
5142                 nfs4_put_stid(ns);
5143         return lst;
5144 }
5145
5146 static int
5147 check_lock_length(u64 offset, u64 length)
5148 {
5149         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
5150              LOFF_OVERFLOW(offset, length)));
5151 }
5152
5153 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5154 {
5155         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5156
5157         lockdep_assert_held(&fp->fi_lock);
5158
5159         if (test_access(access, lock_stp))
5160                 return;
5161         __nfs4_file_get_access(fp, access);
5162         set_access(access, lock_stp);
5163 }
5164
5165 static __be32
5166 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5167                             struct nfs4_ol_stateid *ost,
5168                             struct nfsd4_lock *lock,
5169                             struct nfs4_ol_stateid **lst, bool *new)
5170 {
5171         __be32 status;
5172         struct nfs4_file *fi = ost->st_stid.sc_file;
5173         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5174         struct nfs4_client *cl = oo->oo_owner.so_client;
5175         struct inode *inode = cstate->current_fh.fh_dentry->d_inode;
5176         struct nfs4_lockowner *lo;
5177         unsigned int strhashval;
5178
5179         lo = find_lockowner_str(&cl->cl_clientid, &lock->v.new.owner, cl);
5180         if (!lo) {
5181                 strhashval = ownerstr_hashval(&lock->v.new.owner);
5182                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5183                 if (lo == NULL)
5184                         return nfserr_jukebox;
5185         } else {
5186                 /* with an existing lockowner, seqids must be the same */
5187                 status = nfserr_bad_seqid;
5188                 if (!cstate->minorversion &&
5189                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5190                         goto out;
5191         }
5192
5193         *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5194         if (*lst == NULL) {
5195                 status = nfserr_jukebox;
5196                 goto out;
5197         }
5198         status = nfs_ok;
5199 out:
5200         nfs4_put_stateowner(&lo->lo_owner);
5201         return status;
5202 }
5203
5204 /*
5205  *  LOCK operation
5206  */
5207 __be32
5208 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5209            struct nfsd4_lock *lock)
5210 {
5211         struct nfs4_openowner *open_sop = NULL;
5212         struct nfs4_lockowner *lock_sop = NULL;
5213         struct nfs4_ol_stateid *lock_stp = NULL;
5214         struct nfs4_ol_stateid *open_stp = NULL;
5215         struct nfs4_file *fp;
5216         struct file *filp = NULL;
5217         struct file_lock *file_lock = NULL;
5218         struct file_lock *conflock = NULL;
5219         __be32 status = 0;
5220         int lkflg;
5221         int err;
5222         bool new = false;
5223         struct net *net = SVC_NET(rqstp);
5224         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5225
5226         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5227                 (long long) lock->lk_offset,
5228                 (long long) lock->lk_length);
5229
5230         if (check_lock_length(lock->lk_offset, lock->lk_length))
5231                  return nfserr_inval;
5232
5233         if ((status = fh_verify(rqstp, &cstate->current_fh,
5234                                 S_IFREG, NFSD_MAY_LOCK))) {
5235                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5236                 return status;
5237         }
5238
5239         if (lock->lk_is_new) {
5240                 if (nfsd4_has_session(cstate))
5241                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5242                         memcpy(&lock->v.new.clientid,
5243                                 &cstate->session->se_client->cl_clientid,
5244                                 sizeof(clientid_t));
5245
5246                 status = nfserr_stale_clientid;
5247                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5248                         goto out;
5249
5250                 /* validate and update open stateid and open seqid */
5251                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5252                                         lock->lk_new_open_seqid,
5253                                         &lock->lk_new_open_stateid,
5254                                         &open_stp, nn);
5255                 if (status)
5256                         goto out;
5257                 open_sop = openowner(open_stp->st_stateowner);
5258                 status = nfserr_bad_stateid;
5259                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5260                                                 &lock->v.new.clientid))
5261                         goto out;
5262                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5263                                                         &lock_stp, &new);
5264         } else {
5265                 status = nfs4_preprocess_seqid_op(cstate,
5266                                        lock->lk_old_lock_seqid,
5267                                        &lock->lk_old_lock_stateid,
5268                                        NFS4_LOCK_STID, &lock_stp, nn);
5269         }
5270         if (status)
5271                 goto out;
5272         lock_sop = lockowner(lock_stp->st_stateowner);
5273
5274         lkflg = setlkflg(lock->lk_type);
5275         status = nfs4_check_openmode(lock_stp, lkflg);
5276         if (status)
5277                 goto out;
5278
5279         status = nfserr_grace;
5280         if (locks_in_grace(net) && !lock->lk_reclaim)
5281                 goto out;
5282         status = nfserr_no_grace;
5283         if (!locks_in_grace(net) && lock->lk_reclaim)
5284                 goto out;
5285
5286         file_lock = locks_alloc_lock();
5287         if (!file_lock) {
5288                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5289                 status = nfserr_jukebox;
5290                 goto out;
5291         }
5292
5293         fp = lock_stp->st_stid.sc_file;
5294         switch (lock->lk_type) {
5295                 case NFS4_READ_LT:
5296                 case NFS4_READW_LT:
5297                         spin_lock(&fp->fi_lock);
5298                         filp = find_readable_file_locked(fp);
5299                         if (filp)
5300                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5301                         spin_unlock(&fp->fi_lock);
5302                         file_lock->fl_type = F_RDLCK;
5303                         break;
5304                 case NFS4_WRITE_LT:
5305                 case NFS4_WRITEW_LT:
5306                         spin_lock(&fp->fi_lock);
5307                         filp = find_writeable_file_locked(fp);
5308                         if (filp)
5309                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5310                         spin_unlock(&fp->fi_lock);
5311                         file_lock->fl_type = F_WRLCK;
5312                         break;
5313                 default:
5314                         status = nfserr_inval;
5315                 goto out;
5316         }
5317         if (!filp) {
5318                 status = nfserr_openmode;
5319                 goto out;
5320         }
5321
5322         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5323         file_lock->fl_pid = current->tgid;
5324         file_lock->fl_file = filp;
5325         file_lock->fl_flags = FL_POSIX;
5326         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5327         file_lock->fl_start = lock->lk_offset;
5328         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5329         nfs4_transform_lock_offset(file_lock);
5330
5331         conflock = locks_alloc_lock();
5332         if (!conflock) {
5333                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5334                 status = nfserr_jukebox;
5335                 goto out;
5336         }
5337
5338         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5339         switch (-err) {
5340         case 0: /* success! */
5341                 update_stateid(&lock_stp->st_stid.sc_stateid);
5342                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
5343                                 sizeof(stateid_t));
5344                 status = 0;
5345                 break;
5346         case (EAGAIN):          /* conflock holds conflicting lock */
5347                 status = nfserr_denied;
5348                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5349                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5350                 break;
5351         case (EDEADLK):
5352                 status = nfserr_deadlock;
5353                 break;
5354         default:
5355                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5356                 status = nfserrno(err);
5357                 break;
5358         }
5359 out:
5360         if (filp)
5361                 fput(filp);
5362         if (lock_stp) {
5363                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5364                 if (cstate->replay_owner &&
5365                     cstate->replay_owner != &lock_sop->lo_owner &&
5366                     seqid_mutating_err(ntohl(status)))
5367                         lock_sop->lo_owner.so_seqid++;
5368
5369                 /*
5370                  * If this is a new, never-before-used stateid, and we are
5371                  * returning an error, then just go ahead and release it.
5372                  */
5373                 if (status && new)
5374                         release_lock_stateid(lock_stp);
5375
5376                 nfs4_put_stid(&lock_stp->st_stid);
5377         }
5378         if (open_stp)
5379                 nfs4_put_stid(&open_stp->st_stid);
5380         nfsd4_bump_seqid(cstate, status);
5381         if (file_lock)
5382                 locks_free_lock(file_lock);
5383         if (conflock)
5384                 locks_free_lock(conflock);
5385         return status;
5386 }
5387
5388 /*
5389  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5390  * so we do a temporary open here just to get an open file to pass to
5391  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5392  * inode operation.)
5393  */
5394 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5395 {
5396         struct file *file;
5397         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5398         if (!err) {
5399                 err = nfserrno(vfs_test_lock(file, lock));
5400                 nfsd_close(file);
5401         }
5402         return err;
5403 }
5404
5405 /*
5406  * LOCKT operation
5407  */
5408 __be32
5409 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5410             struct nfsd4_lockt *lockt)
5411 {
5412         struct file_lock *file_lock = NULL;
5413         struct nfs4_lockowner *lo = NULL;
5414         __be32 status;
5415         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5416
5417         if (locks_in_grace(SVC_NET(rqstp)))
5418                 return nfserr_grace;
5419
5420         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5421                  return nfserr_inval;
5422
5423         if (!nfsd4_has_session(cstate)) {
5424                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5425                 if (status)
5426                         goto out;
5427         }
5428
5429         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5430                 goto out;
5431
5432         file_lock = locks_alloc_lock();
5433         if (!file_lock) {
5434                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5435                 status = nfserr_jukebox;
5436                 goto out;
5437         }
5438
5439         switch (lockt->lt_type) {
5440                 case NFS4_READ_LT:
5441                 case NFS4_READW_LT:
5442                         file_lock->fl_type = F_RDLCK;
5443                 break;
5444                 case NFS4_WRITE_LT:
5445                 case NFS4_WRITEW_LT:
5446                         file_lock->fl_type = F_WRLCK;
5447                 break;
5448                 default:
5449                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5450                         status = nfserr_inval;
5451                 goto out;
5452         }
5453
5454         lo = find_lockowner_str(&lockt->lt_clientid, &lockt->lt_owner,
5455                                 cstate->clp);
5456         if (lo)
5457                 file_lock->fl_owner = (fl_owner_t)lo;
5458         file_lock->fl_pid = current->tgid;
5459         file_lock->fl_flags = FL_POSIX;
5460
5461         file_lock->fl_start = lockt->lt_offset;
5462         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5463
5464         nfs4_transform_lock_offset(file_lock);
5465
5466         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5467         if (status)
5468                 goto out;
5469
5470         if (file_lock->fl_type != F_UNLCK) {
5471                 status = nfserr_denied;
5472                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5473         }
5474 out:
5475         if (lo)
5476                 nfs4_put_stateowner(&lo->lo_owner);
5477         if (file_lock)
5478                 locks_free_lock(file_lock);
5479         return status;
5480 }
5481
5482 __be32
5483 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5484             struct nfsd4_locku *locku)
5485 {
5486         struct nfs4_ol_stateid *stp;
5487         struct file *filp = NULL;
5488         struct file_lock *file_lock = NULL;
5489         __be32 status;
5490         int err;
5491         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5492
5493         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5494                 (long long) locku->lu_offset,
5495                 (long long) locku->lu_length);
5496
5497         if (check_lock_length(locku->lu_offset, locku->lu_length))
5498                  return nfserr_inval;
5499
5500         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5501                                         &locku->lu_stateid, NFS4_LOCK_STID,
5502                                         &stp, nn);
5503         if (status)
5504                 goto out;
5505         filp = find_any_file(stp->st_stid.sc_file);
5506         if (!filp) {
5507                 status = nfserr_lock_range;
5508                 goto put_stateid;
5509         }
5510         file_lock = locks_alloc_lock();
5511         if (!file_lock) {
5512                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5513                 status = nfserr_jukebox;
5514                 goto fput;
5515         }
5516
5517         file_lock->fl_type = F_UNLCK;
5518         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5519         file_lock->fl_pid = current->tgid;
5520         file_lock->fl_file = filp;
5521         file_lock->fl_flags = FL_POSIX;
5522         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5523         file_lock->fl_start = locku->lu_offset;
5524
5525         file_lock->fl_end = last_byte_offset(locku->lu_offset,
5526                                                 locku->lu_length);
5527         nfs4_transform_lock_offset(file_lock);
5528
5529         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5530         if (err) {
5531                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5532                 goto out_nfserr;
5533         }
5534         update_stateid(&stp->st_stid.sc_stateid);
5535         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
5536 fput:
5537         fput(filp);
5538 put_stateid:
5539         nfs4_put_stid(&stp->st_stid);
5540 out:
5541         nfsd4_bump_seqid(cstate, status);
5542         if (file_lock)
5543                 locks_free_lock(file_lock);
5544         return status;
5545
5546 out_nfserr:
5547         status = nfserrno(err);
5548         goto fput;
5549 }
5550
5551 /*
5552  * returns
5553  *      true:  locks held by lockowner
5554  *      false: no locks held by lockowner
5555  */
5556 static bool
5557 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5558 {
5559         struct file_lock **flpp;
5560         int status = false;
5561         struct file *filp = find_any_file(fp);
5562         struct inode *inode;
5563
5564         if (!filp) {
5565                 /* Any valid lock stateid should have some sort of access */
5566                 WARN_ON_ONCE(1);
5567                 return status;
5568         }
5569
5570         inode = file_inode(filp);
5571
5572         spin_lock(&inode->i_lock);
5573         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
5574                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
5575                         status = true;
5576                         break;
5577                 }
5578         }
5579         spin_unlock(&inode->i_lock);
5580         fput(filp);
5581         return status;
5582 }
5583
5584 __be32
5585 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5586                         struct nfsd4_compound_state *cstate,
5587                         struct nfsd4_release_lockowner *rlockowner)
5588 {
5589         clientid_t *clid = &rlockowner->rl_clientid;
5590         struct nfs4_stateowner *sop;
5591         struct nfs4_lockowner *lo = NULL;
5592         struct nfs4_ol_stateid *stp;
5593         struct xdr_netobj *owner = &rlockowner->rl_owner;
5594         unsigned int hashval = ownerstr_hashval(owner);
5595         __be32 status;
5596         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5597         struct nfs4_client *clp;
5598
5599         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5600                 clid->cl_boot, clid->cl_id);
5601
5602         status = lookup_clientid(clid, cstate, nn);
5603         if (status)
5604                 return status;
5605
5606         clp = cstate->clp;
5607         /* Find the matching lock stateowner */
5608         spin_lock(&clp->cl_lock);
5609         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5610                             so_strhash) {
5611
5612                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5613                         continue;
5614
5615                 /* see if there are still any locks associated with it */
5616                 lo = lockowner(sop);
5617                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5618                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
5619                                 status = nfserr_locks_held;
5620                                 spin_unlock(&clp->cl_lock);
5621                                 return status;
5622                         }
5623                 }
5624
5625                 nfs4_get_stateowner(sop);
5626                 break;
5627         }
5628         spin_unlock(&clp->cl_lock);
5629         if (lo)
5630                 release_lockowner(lo);
5631         return status;
5632 }
5633
5634 static inline struct nfs4_client_reclaim *
5635 alloc_reclaim(void)
5636 {
5637         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5638 }
5639
5640 bool
5641 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5642 {
5643         struct nfs4_client_reclaim *crp;
5644
5645         crp = nfsd4_find_reclaim_client(name, nn);
5646         return (crp && crp->cr_clp);
5647 }
5648
5649 /*
5650  * failure => all reset bets are off, nfserr_no_grace...
5651  */
5652 struct nfs4_client_reclaim *
5653 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5654 {
5655         unsigned int strhashval;
5656         struct nfs4_client_reclaim *crp;
5657
5658         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5659         crp = alloc_reclaim();
5660         if (crp) {
5661                 strhashval = clientstr_hashval(name);
5662                 INIT_LIST_HEAD(&crp->cr_strhash);
5663                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5664                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5665                 crp->cr_clp = NULL;
5666                 nn->reclaim_str_hashtbl_size++;
5667         }
5668         return crp;
5669 }
5670
5671 void
5672 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5673 {
5674         list_del(&crp->cr_strhash);
5675         kfree(crp);
5676         nn->reclaim_str_hashtbl_size--;
5677 }
5678
5679 void
5680 nfs4_release_reclaim(struct nfsd_net *nn)
5681 {
5682         struct nfs4_client_reclaim *crp = NULL;
5683         int i;
5684
5685         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5686                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5687                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5688                                         struct nfs4_client_reclaim, cr_strhash);
5689                         nfs4_remove_reclaim_record(crp, nn);
5690                 }
5691         }
5692         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5693 }
5694
5695 /*
5696  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5697 struct nfs4_client_reclaim *
5698 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5699 {
5700         unsigned int strhashval;
5701         struct nfs4_client_reclaim *crp = NULL;
5702
5703         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5704
5705         strhashval = clientstr_hashval(recdir);
5706         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5707                 if (same_name(crp->cr_recdir, recdir)) {
5708                         return crp;
5709                 }
5710         }
5711         return NULL;
5712 }
5713
5714 /*
5715 * Called from OPEN. Look for clientid in reclaim list.
5716 */
5717 __be32
5718 nfs4_check_open_reclaim(clientid_t *clid,
5719                 struct nfsd4_compound_state *cstate,
5720                 struct nfsd_net *nn)
5721 {
5722         __be32 status;
5723
5724         /* find clientid in conf_id_hashtbl */
5725         status = lookup_clientid(clid, cstate, nn);
5726         if (status)
5727                 return nfserr_reclaim_bad;
5728
5729         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
5730                 return nfserr_no_grace;
5731
5732         if (nfsd4_client_record_check(cstate->clp))
5733                 return nfserr_reclaim_bad;
5734
5735         return nfs_ok;
5736 }
5737
5738 #ifdef CONFIG_NFSD_FAULT_INJECTION
5739 static inline void
5740 put_client(struct nfs4_client *clp)
5741 {
5742         atomic_dec(&clp->cl_refcount);
5743 }
5744
5745 static struct nfs4_client *
5746 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
5747 {
5748         struct nfs4_client *clp;
5749         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5750                                           nfsd_net_id);
5751
5752         if (!nfsd_netns_ready(nn))
5753                 return NULL;
5754
5755         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5756                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
5757                         return clp;
5758         }
5759         return NULL;
5760 }
5761
5762 u64
5763 nfsd_inject_print_clients(void)
5764 {
5765         struct nfs4_client *clp;
5766         u64 count = 0;
5767         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5768                                           nfsd_net_id);
5769         char buf[INET6_ADDRSTRLEN];
5770
5771         if (!nfsd_netns_ready(nn))
5772                 return 0;
5773
5774         spin_lock(&nn->client_lock);
5775         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5776                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5777                 pr_info("NFS Client: %s\n", buf);
5778                 ++count;
5779         }
5780         spin_unlock(&nn->client_lock);
5781
5782         return count;
5783 }
5784
5785 u64
5786 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
5787 {
5788         u64 count = 0;
5789         struct nfs4_client *clp;
5790         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5791                                           nfsd_net_id);
5792
5793         if (!nfsd_netns_ready(nn))
5794                 return count;
5795
5796         spin_lock(&nn->client_lock);
5797         clp = nfsd_find_client(addr, addr_size);
5798         if (clp) {
5799                 if (mark_client_expired_locked(clp) == nfs_ok)
5800                         ++count;
5801                 else
5802                         clp = NULL;
5803         }
5804         spin_unlock(&nn->client_lock);
5805
5806         if (clp)
5807                 expire_client(clp);
5808
5809         return count;
5810 }
5811
5812 u64
5813 nfsd_inject_forget_clients(u64 max)
5814 {
5815         u64 count = 0;
5816         struct nfs4_client *clp, *next;
5817         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5818                                                 nfsd_net_id);
5819         LIST_HEAD(reaplist);
5820
5821         if (!nfsd_netns_ready(nn))
5822                 return count;
5823
5824         spin_lock(&nn->client_lock);
5825         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
5826                 if (mark_client_expired_locked(clp) == nfs_ok) {
5827                         list_add(&clp->cl_lru, &reaplist);
5828                         if (max != 0 && ++count >= max)
5829                                 break;
5830                 }
5831         }
5832         spin_unlock(&nn->client_lock);
5833
5834         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
5835                 expire_client(clp);
5836
5837         return count;
5838 }
5839
5840 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
5841                              const char *type)
5842 {
5843         char buf[INET6_ADDRSTRLEN];
5844         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5845         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
5846 }
5847
5848 static void
5849 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
5850                              struct list_head *collect)
5851 {
5852         struct nfs4_client *clp = lst->st_stid.sc_client;
5853         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5854                                           nfsd_net_id);
5855
5856         if (!collect)
5857                 return;
5858
5859         lockdep_assert_held(&nn->client_lock);
5860         atomic_inc(&clp->cl_refcount);
5861         list_add(&lst->st_locks, collect);
5862 }
5863
5864 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
5865                                     struct list_head *collect,
5866                                     void (*func)(struct nfs4_ol_stateid *))
5867 {
5868         struct nfs4_openowner *oop;
5869         struct nfs4_ol_stateid *stp, *st_next;
5870         struct nfs4_ol_stateid *lst, *lst_next;
5871         u64 count = 0;
5872
5873         spin_lock(&clp->cl_lock);
5874         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
5875                 list_for_each_entry_safe(stp, st_next,
5876                                 &oop->oo_owner.so_stateids, st_perstateowner) {
5877                         list_for_each_entry_safe(lst, lst_next,
5878                                         &stp->st_locks, st_locks) {
5879                                 if (func) {
5880                                         func(lst);
5881                                         nfsd_inject_add_lock_to_list(lst,
5882                                                                 collect);
5883                                 }
5884                                 ++count;
5885                                 /*
5886                                  * Despite the fact that these functions deal
5887                                  * with 64-bit integers for "count", we must
5888                                  * ensure that it doesn't blow up the
5889                                  * clp->cl_refcount. Throw a warning if we
5890                                  * start to approach INT_MAX here.
5891                                  */
5892                                 WARN_ON_ONCE(count == (INT_MAX / 2));
5893                                 if (count == max)
5894                                         goto out;
5895                         }
5896                 }
5897         }
5898 out:
5899         spin_unlock(&clp->cl_lock);
5900
5901         return count;
5902 }
5903
5904 static u64
5905 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
5906                           u64 max)
5907 {
5908         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
5909 }
5910
5911 static u64
5912 nfsd_print_client_locks(struct nfs4_client *clp)
5913 {
5914         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
5915         nfsd_print_count(clp, count, "locked files");
5916         return count;
5917 }
5918
5919 u64
5920 nfsd_inject_print_locks(void)
5921 {
5922         struct nfs4_client *clp;
5923         u64 count = 0;
5924         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5925                                                 nfsd_net_id);
5926
5927         if (!nfsd_netns_ready(nn))
5928                 return 0;
5929
5930         spin_lock(&nn->client_lock);
5931         list_for_each_entry(clp, &nn->client_lru, cl_lru)
5932                 count += nfsd_print_client_locks(clp);
5933         spin_unlock(&nn->client_lock);
5934
5935         return count;
5936 }
5937
5938 static void
5939 nfsd_reap_locks(struct list_head *reaplist)
5940 {
5941         struct nfs4_client *clp;
5942         struct nfs4_ol_stateid *stp, *next;
5943
5944         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
5945                 list_del_init(&stp->st_locks);
5946                 clp = stp->st_stid.sc_client;
5947                 nfs4_put_stid(&stp->st_stid);
5948                 put_client(clp);
5949         }
5950 }
5951
5952 u64
5953 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
5954 {
5955         unsigned int count = 0;
5956         struct nfs4_client *clp;
5957         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5958                                                 nfsd_net_id);
5959         LIST_HEAD(reaplist);
5960
5961         if (!nfsd_netns_ready(nn))
5962                 return count;
5963
5964         spin_lock(&nn->client_lock);
5965         clp = nfsd_find_client(addr, addr_size);
5966         if (clp)
5967                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
5968         spin_unlock(&nn->client_lock);
5969         nfsd_reap_locks(&reaplist);
5970         return count;
5971 }
5972
5973 u64
5974 nfsd_inject_forget_locks(u64 max)
5975 {
5976         u64 count = 0;
5977         struct nfs4_client *clp;
5978         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5979                                                 nfsd_net_id);
5980         LIST_HEAD(reaplist);
5981
5982         if (!nfsd_netns_ready(nn))
5983                 return count;
5984
5985         spin_lock(&nn->client_lock);
5986         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5987                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
5988                 if (max != 0 && count >= max)
5989                         break;
5990         }
5991         spin_unlock(&nn->client_lock);
5992         nfsd_reap_locks(&reaplist);
5993         return count;
5994 }
5995
5996 static u64
5997 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
5998                               struct list_head *collect,
5999                               void (*func)(struct nfs4_openowner *))
6000 {
6001         struct nfs4_openowner *oop, *next;
6002         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6003                                                 nfsd_net_id);
6004         u64 count = 0;
6005
6006         lockdep_assert_held(&nn->client_lock);
6007
6008         spin_lock(&clp->cl_lock);
6009         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6010                 if (func) {
6011                         func(oop);
6012                         if (collect) {
6013                                 atomic_inc(&clp->cl_refcount);
6014                                 list_add(&oop->oo_perclient, collect);
6015                         }
6016                 }
6017                 ++count;
6018                 /*
6019                  * Despite the fact that these functions deal with
6020                  * 64-bit integers for "count", we must ensure that
6021                  * it doesn't blow up the clp->cl_refcount. Throw a
6022                  * warning if we start to approach INT_MAX here.
6023                  */
6024                 WARN_ON_ONCE(count == (INT_MAX / 2));
6025                 if (count == max)
6026                         break;
6027         }
6028         spin_unlock(&clp->cl_lock);
6029
6030         return count;
6031 }
6032
6033 static u64
6034 nfsd_print_client_openowners(struct nfs4_client *clp)
6035 {
6036         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6037
6038         nfsd_print_count(clp, count, "openowners");
6039         return count;
6040 }
6041
6042 static u64
6043 nfsd_collect_client_openowners(struct nfs4_client *clp,
6044                                struct list_head *collect, u64 max)
6045 {
6046         return nfsd_foreach_client_openowner(clp, max, collect,
6047                                                 unhash_openowner_locked);
6048 }
6049
6050 u64
6051 nfsd_inject_print_openowners(void)
6052 {
6053         struct nfs4_client *clp;
6054         u64 count = 0;
6055         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6056                                                 nfsd_net_id);
6057
6058         if (!nfsd_netns_ready(nn))
6059                 return 0;
6060
6061         spin_lock(&nn->client_lock);
6062         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6063                 count += nfsd_print_client_openowners(clp);
6064         spin_unlock(&nn->client_lock);
6065
6066         return count;
6067 }
6068
6069 static void
6070 nfsd_reap_openowners(struct list_head *reaplist)
6071 {
6072         struct nfs4_client *clp;
6073         struct nfs4_openowner *oop, *next;
6074
6075         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6076                 list_del_init(&oop->oo_perclient);
6077                 clp = oop->oo_owner.so_client;
6078                 release_openowner(oop);
6079                 put_client(clp);
6080         }
6081 }
6082
6083 u64
6084 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6085                                      size_t addr_size)
6086 {
6087         unsigned int count = 0;
6088         struct nfs4_client *clp;
6089         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6090                                                 nfsd_net_id);
6091         LIST_HEAD(reaplist);
6092
6093         if (!nfsd_netns_ready(nn))
6094                 return count;
6095
6096         spin_lock(&nn->client_lock);
6097         clp = nfsd_find_client(addr, addr_size);
6098         if (clp)
6099                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6100         spin_unlock(&nn->client_lock);
6101         nfsd_reap_openowners(&reaplist);
6102         return count;
6103 }
6104
6105 u64
6106 nfsd_inject_forget_openowners(u64 max)
6107 {
6108         u64 count = 0;
6109         struct nfs4_client *clp;
6110         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6111                                                 nfsd_net_id);
6112         LIST_HEAD(reaplist);
6113
6114         if (!nfsd_netns_ready(nn))
6115                 return count;
6116
6117         spin_lock(&nn->client_lock);
6118         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6119                 count += nfsd_collect_client_openowners(clp, &reaplist,
6120                                                         max - count);
6121                 if (max != 0 && count >= max)
6122                         break;
6123         }
6124         spin_unlock(&nn->client_lock);
6125         nfsd_reap_openowners(&reaplist);
6126         return count;
6127 }
6128
6129 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6130                                      struct list_head *victims)
6131 {
6132         struct nfs4_delegation *dp, *next;
6133         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6134                                                 nfsd_net_id);
6135         u64 count = 0;
6136
6137         lockdep_assert_held(&nn->client_lock);
6138
6139         spin_lock(&state_lock);
6140         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6141                 if (victims) {
6142                         /*
6143                          * It's not safe to mess with delegations that have a
6144                          * non-zero dl_time. They might have already been broken
6145                          * and could be processed by the laundromat outside of
6146                          * the state_lock. Just leave them be.
6147                          */
6148                         if (dp->dl_time != 0)
6149                                 continue;
6150
6151                         atomic_inc(&clp->cl_refcount);
6152                         unhash_delegation_locked(dp);
6153                         list_add(&dp->dl_recall_lru, victims);
6154                 }
6155                 ++count;
6156                 /*
6157                  * Despite the fact that these functions deal with
6158                  * 64-bit integers for "count", we must ensure that
6159                  * it doesn't blow up the clp->cl_refcount. Throw a
6160                  * warning if we start to approach INT_MAX here.
6161                  */
6162                 WARN_ON_ONCE(count == (INT_MAX / 2));
6163                 if (count == max)
6164                         break;
6165         }
6166         spin_unlock(&state_lock);
6167         return count;
6168 }
6169
6170 static u64
6171 nfsd_print_client_delegations(struct nfs4_client *clp)
6172 {
6173         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6174
6175         nfsd_print_count(clp, count, "delegations");
6176         return count;
6177 }
6178
6179 u64
6180 nfsd_inject_print_delegations(void)
6181 {
6182         struct nfs4_client *clp;
6183         u64 count = 0;
6184         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6185                                                 nfsd_net_id);
6186
6187         if (!nfsd_netns_ready(nn))
6188                 return 0;
6189
6190         spin_lock(&nn->client_lock);
6191         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6192                 count += nfsd_print_client_delegations(clp);
6193         spin_unlock(&nn->client_lock);
6194
6195         return count;
6196 }
6197
6198 static void
6199 nfsd_forget_delegations(struct list_head *reaplist)
6200 {
6201         struct nfs4_client *clp;
6202         struct nfs4_delegation *dp, *next;
6203
6204         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6205                 list_del_init(&dp->dl_recall_lru);
6206                 clp = dp->dl_stid.sc_client;
6207                 revoke_delegation(dp);
6208                 put_client(clp);
6209         }
6210 }
6211
6212 u64
6213 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6214                                       size_t addr_size)
6215 {
6216         u64 count = 0;
6217         struct nfs4_client *clp;
6218         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6219                                                 nfsd_net_id);
6220         LIST_HEAD(reaplist);
6221
6222         if (!nfsd_netns_ready(nn))
6223                 return count;
6224
6225         spin_lock(&nn->client_lock);
6226         clp = nfsd_find_client(addr, addr_size);
6227         if (clp)
6228                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6229         spin_unlock(&nn->client_lock);
6230
6231         nfsd_forget_delegations(&reaplist);
6232         return count;
6233 }
6234
6235 u64
6236 nfsd_inject_forget_delegations(u64 max)
6237 {
6238         u64 count = 0;
6239         struct nfs4_client *clp;
6240         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6241                                                 nfsd_net_id);
6242         LIST_HEAD(reaplist);
6243
6244         if (!nfsd_netns_ready(nn))
6245                 return count;
6246
6247         spin_lock(&nn->client_lock);
6248         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6249                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6250                 if (max != 0 && count >= max)
6251                         break;
6252         }
6253         spin_unlock(&nn->client_lock);
6254         nfsd_forget_delegations(&reaplist);
6255         return count;
6256 }
6257
6258 static void
6259 nfsd_recall_delegations(struct list_head *reaplist)
6260 {
6261         struct nfs4_client *clp;
6262         struct nfs4_delegation *dp, *next;
6263
6264         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6265                 list_del_init(&dp->dl_recall_lru);
6266                 clp = dp->dl_stid.sc_client;
6267                 /*
6268                  * We skipped all entries that had a zero dl_time before,
6269                  * so we can now reset the dl_time back to 0. If a delegation
6270                  * break comes in now, then it won't make any difference since
6271                  * we're recalling it either way.
6272                  */
6273                 spin_lock(&state_lock);
6274                 dp->dl_time = 0;
6275                 spin_unlock(&state_lock);
6276                 nfsd_break_one_deleg(dp);
6277                 put_client(clp);
6278         }
6279 }
6280
6281 u64
6282 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6283                                       size_t addr_size)
6284 {
6285         u64 count = 0;
6286         struct nfs4_client *clp;
6287         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6288                                                 nfsd_net_id);
6289         LIST_HEAD(reaplist);
6290
6291         if (!nfsd_netns_ready(nn))
6292                 return count;
6293
6294         spin_lock(&nn->client_lock);
6295         clp = nfsd_find_client(addr, addr_size);
6296         if (clp)
6297                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6298         spin_unlock(&nn->client_lock);
6299
6300         nfsd_recall_delegations(&reaplist);
6301         return count;
6302 }
6303
6304 u64
6305 nfsd_inject_recall_delegations(u64 max)
6306 {
6307         u64 count = 0;
6308         struct nfs4_client *clp, *next;
6309         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6310                                                 nfsd_net_id);
6311         LIST_HEAD(reaplist);
6312
6313         if (!nfsd_netns_ready(nn))
6314                 return count;
6315
6316         spin_lock(&nn->client_lock);
6317         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6318                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6319                 if (max != 0 && ++count >= max)
6320                         break;
6321         }
6322         spin_unlock(&nn->client_lock);
6323         nfsd_recall_delegations(&reaplist);
6324         return count;
6325 }
6326 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6327
6328 /*
6329  * Since the lifetime of a delegation isn't limited to that of an open, a
6330  * client may quite reasonably hang on to a delegation as long as it has
6331  * the inode cached.  This becomes an obvious problem the first time a
6332  * client's inode cache approaches the size of the server's total memory.
6333  *
6334  * For now we avoid this problem by imposing a hard limit on the number
6335  * of delegations, which varies according to the server's memory size.
6336  */
6337 static void
6338 set_max_delegations(void)
6339 {
6340         /*
6341          * Allow at most 4 delegations per megabyte of RAM.  Quick
6342          * estimates suggest that in the worst case (where every delegation
6343          * is for a different inode), a delegation could take about 1.5K,
6344          * giving a worst case usage of about 6% of memory.
6345          */
6346         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6347 }
6348
6349 static int nfs4_state_create_net(struct net *net)
6350 {
6351         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6352         int i;
6353
6354         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6355                         CLIENT_HASH_SIZE, GFP_KERNEL);
6356         if (!nn->conf_id_hashtbl)
6357                 goto err;
6358         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6359                         CLIENT_HASH_SIZE, GFP_KERNEL);
6360         if (!nn->unconf_id_hashtbl)
6361                 goto err_unconf_id;
6362         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6363                         SESSION_HASH_SIZE, GFP_KERNEL);
6364         if (!nn->sessionid_hashtbl)
6365                 goto err_sessionid;
6366
6367         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6368                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6369                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6370         }
6371         for (i = 0; i < SESSION_HASH_SIZE; i++)
6372                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6373         nn->conf_name_tree = RB_ROOT;
6374         nn->unconf_name_tree = RB_ROOT;
6375         INIT_LIST_HEAD(&nn->client_lru);
6376         INIT_LIST_HEAD(&nn->close_lru);
6377         INIT_LIST_HEAD(&nn->del_recall_lru);
6378         spin_lock_init(&nn->client_lock);
6379
6380         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6381         get_net(net);
6382
6383         return 0;
6384
6385 err_sessionid:
6386         kfree(nn->unconf_id_hashtbl);
6387 err_unconf_id:
6388         kfree(nn->conf_id_hashtbl);
6389 err:
6390         return -ENOMEM;
6391 }
6392
6393 static void
6394 nfs4_state_destroy_net(struct net *net)
6395 {
6396         int i;
6397         struct nfs4_client *clp = NULL;
6398         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6399
6400         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6401                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6402                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6403                         destroy_client(clp);
6404                 }
6405         }
6406
6407         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6408                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6409                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6410                         destroy_client(clp);
6411                 }
6412         }
6413
6414         kfree(nn->sessionid_hashtbl);
6415         kfree(nn->unconf_id_hashtbl);
6416         kfree(nn->conf_id_hashtbl);
6417         put_net(net);
6418 }
6419
6420 int
6421 nfs4_state_start_net(struct net *net)
6422 {
6423         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6424         int ret;
6425
6426         ret = nfs4_state_create_net(net);
6427         if (ret)
6428                 return ret;
6429         nn->boot_time = get_seconds();
6430         nn->grace_ended = false;
6431         locks_start_grace(net, &nn->nfsd4_manager);
6432         nfsd4_client_tracking_init(net);
6433         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6434                nn->nfsd4_grace, net);
6435         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6436         return 0;
6437 }
6438
6439 /* initialization to perform when the nfsd service is started: */
6440
6441 int
6442 nfs4_state_start(void)
6443 {
6444         int ret;
6445
6446         ret = set_callback_cred();
6447         if (ret)
6448                 return -ENOMEM;
6449         laundry_wq = create_singlethread_workqueue("nfsd4");
6450         if (laundry_wq == NULL) {
6451                 ret = -ENOMEM;
6452                 goto out_recovery;
6453         }
6454         ret = nfsd4_create_callback_queue();
6455         if (ret)
6456                 goto out_free_laundry;
6457
6458         set_max_delegations();
6459
6460         return 0;
6461
6462 out_free_laundry:
6463         destroy_workqueue(laundry_wq);
6464 out_recovery:
6465         return ret;
6466 }
6467
6468 void
6469 nfs4_state_shutdown_net(struct net *net)
6470 {
6471         struct nfs4_delegation *dp = NULL;
6472         struct list_head *pos, *next, reaplist;
6473         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6474
6475         cancel_delayed_work_sync(&nn->laundromat_work);
6476         locks_end_grace(&nn->nfsd4_manager);
6477
6478         INIT_LIST_HEAD(&reaplist);
6479         spin_lock(&state_lock);
6480         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6481                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6482                 unhash_delegation_locked(dp);
6483                 list_add(&dp->dl_recall_lru, &reaplist);
6484         }
6485         spin_unlock(&state_lock);
6486         list_for_each_safe(pos, next, &reaplist) {
6487                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6488                 list_del_init(&dp->dl_recall_lru);
6489                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6490                 nfs4_put_stid(&dp->dl_stid);
6491         }
6492
6493         nfsd4_client_tracking_exit(net);
6494         nfs4_state_destroy_net(net);
6495 }
6496
6497 void
6498 nfs4_state_shutdown(void)
6499 {
6500         destroy_workqueue(laundry_wq);
6501         nfsd4_destroy_callback_queue();
6502 }
6503
6504 static void
6505 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6506 {
6507         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6508                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6509 }
6510
6511 static void
6512 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6513 {
6514         if (cstate->minorversion) {
6515                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6516                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6517         }
6518 }
6519
6520 void
6521 clear_current_stateid(struct nfsd4_compound_state *cstate)
6522 {
6523         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6524 }
6525
6526 /*
6527  * functions to set current state id
6528  */
6529 void
6530 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6531 {
6532         put_stateid(cstate, &odp->od_stateid);
6533 }
6534
6535 void
6536 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6537 {
6538         put_stateid(cstate, &open->op_stateid);
6539 }
6540
6541 void
6542 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6543 {
6544         put_stateid(cstate, &close->cl_stateid);
6545 }
6546
6547 void
6548 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6549 {
6550         put_stateid(cstate, &lock->lk_resp_stateid);
6551 }
6552
6553 /*
6554  * functions to consume current state id
6555  */
6556
6557 void
6558 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6559 {
6560         get_stateid(cstate, &odp->od_stateid);
6561 }
6562
6563 void
6564 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6565 {
6566         get_stateid(cstate, &drp->dr_stateid);
6567 }
6568
6569 void
6570 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6571 {
6572         get_stateid(cstate, &fsp->fr_stateid);
6573 }
6574
6575 void
6576 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6577 {
6578         get_stateid(cstate, &setattr->sa_stateid);
6579 }
6580
6581 void
6582 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6583 {
6584         get_stateid(cstate, &close->cl_stateid);
6585 }
6586
6587 void
6588 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6589 {
6590         get_stateid(cstate, &locku->lu_stateid);
6591 }
6592
6593 void
6594 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6595 {
6596         get_stateid(cstate, &read->rd_stateid);
6597 }
6598
6599 void
6600 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6601 {
6602         get_stateid(cstate, &write->wr_stateid);
6603 }