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/hash.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 /* Currently used for almost all code touching nfsv4 state: */
  79 static DEFINE_MUTEX(client_mutex);
  80
  81 /*
  82  * Currently used for the del_recall_lru and file hash table.  In an
  83  * effort to decrease the scope of the client_mutex, this spinlock may
  84  * eventually cover more:
  85  */
  86 static DEFINE_SPINLOCK(state_lock);
  87
  88 /*
  89  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  90  * the refcount on the open stateid to drop.
  91  */
  92 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  93
  94 static struct kmem_cache *openowner_slab;
  95 static struct kmem_cache *lockowner_slab;
  96 static struct kmem_cache *file_slab;
  97 static struct kmem_cache *stateid_slab;
  98 static struct kmem_cache *deleg_slab;
  99
 100 void
 101 nfs4_lock_state(void)
 102 {
 103         mutex_lock(&client_mutex);
 104 }
 105
 106 static void free_session(struct nfsd4_session *);
 107
 108 static bool is_session_dead(struct nfsd4_session *ses)
 109 {
 110         return ses->se_flags & NFS4_SESSION_DEAD;
 111 }
 112
 113 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 114 {
 115         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 116                 return nfserr_jukebox;
 117         ses->se_flags |= NFS4_SESSION_DEAD;
 118         return nfs_ok;
 119 }
 120
 121 void
 122 nfs4_unlock_state(void)
 123 {
 124         mutex_unlock(&client_mutex);
 125 }
 126
 127 static bool is_client_expired(struct nfs4_client *clp)
 128 {
 129         return clp->cl_time == 0;
 130 }
 131
 132 static __be32 get_client_locked(struct nfs4_client *clp)
 133 {
 134         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 135
 136         lockdep_assert_held(&nn->client_lock);
 137
 138         if (is_client_expired(clp))
 139                 return nfserr_expired;
 140         atomic_inc(&clp->cl_refcount);
 141         return nfs_ok;
 142 }
 143
 144 /* must be called under the client_lock */
 145 static inline void
 146 renew_client_locked(struct nfs4_client *clp)
 147 {
 148         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 149
 150         if (is_client_expired(clp)) {
 151                 WARN_ON(1);
 152                 printk("%s: client (clientid %08x/%08x) already expired\n",
 153                         __func__,
 154                         clp->cl_clientid.cl_boot,
 155                         clp->cl_clientid.cl_id);
 156                 return;
 157         }
 158
 159         dprintk("renewing client (clientid %08x/%08x)\n",
 160                         clp->cl_clientid.cl_boot,
 161                         clp->cl_clientid.cl_id);
 162         list_move_tail(&clp->cl_lru, &nn->client_lru);
 163         clp->cl_time = get_seconds();
 164 }
 165
 166 static inline void
 167 renew_client(struct nfs4_client *clp)
 168 {
 169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 170
 171         spin_lock(&nn->client_lock);
 172         renew_client_locked(clp);
 173         spin_unlock(&nn->client_lock);
 174 }
 175
 176 static void put_client_renew_locked(struct nfs4_client *clp)
 177 {
 178         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 179
 180         lockdep_assert_held(&nn->client_lock);
 181
 182         if (!atomic_dec_and_test(&clp->cl_refcount))
 183                 return;
 184         if (!is_client_expired(clp))
 185                 renew_client_locked(clp);
 186 }
 187
 188 static void put_client_renew(struct nfs4_client *clp)
 189 {
 190         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 191
 192         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 193                 return;
 194         if (!is_client_expired(clp))
 195                 renew_client_locked(clp);
 196         spin_unlock(&nn->client_lock);
 197 }
 198
 199 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 200 {
 201         __be32 status;
 202
 203         if (is_session_dead(ses))
 204                 return nfserr_badsession;
 205         status = get_client_locked(ses->se_client);
 206         if (status)
 207                 return status;
 208         atomic_inc(&ses->se_ref);
 209         return nfs_ok;
 210 }
 211
 212 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 213 {
 214         struct nfs4_client *clp = ses->se_client;
 215         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 216
 217         lockdep_assert_held(&nn->client_lock);
 218
 219         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 220                 free_session(ses);
 221         put_client_renew_locked(clp);
 222 }
 223
 224 static void nfsd4_put_session(struct nfsd4_session *ses)
 225 {
 226         struct nfs4_client *clp = ses->se_client;
 227         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 228
 229         spin_lock(&nn->client_lock);
 230         nfsd4_put_session_locked(ses);
 231         spin_unlock(&nn->client_lock);
 232 }
 233
 234 static int
 235 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 236 {
 237         return (sop->so_owner.len == owner->len) &&
 238                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 239 }
 240
 241 static struct nfs4_openowner *
 242 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 243                         struct nfs4_client *clp)
 244 {
 245         struct nfs4_stateowner *so;
 246
 247         lockdep_assert_held(&clp->cl_lock);
 248
 249         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 250                             so_strhash) {
 251                 if (!so->so_is_open_owner)
 252                         continue;
 253                 if (same_owner_str(so, &open->op_owner)) {
 254                         atomic_inc(&so->so_count);
 255                         return openowner(so);
 256                 }
 257         }
 258         return NULL;
 259 }
 260
 261 static struct nfs4_openowner *
 262 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 263                         struct nfs4_client *clp)
 264 {
 265         struct nfs4_openowner *oo;
 266
 267         spin_lock(&clp->cl_lock);
 268         oo = find_openstateowner_str_locked(hashval, open, clp);
 269         spin_unlock(&clp->cl_lock);
 270         return oo;
 271 }
 272
 273 static inline u32
 274 opaque_hashval(const void *ptr, int nbytes)
 275 {
 276         unsigned char *cptr = (unsigned char *) ptr;
 277
 278         u32 x = 0;
 279         while (nbytes--) {
 280                 x *= 37;
 281                 x += *cptr++;
 282         }
 283         return x;
 284 }
 285
 286 static void nfsd4_free_file(struct nfs4_file *f)
 287 {
 288         kmem_cache_free(file_slab, f);
 289 }
 290
 291 static inline void
 292 put_nfs4_file(struct nfs4_file *fi)
 293 {
 294         might_lock(&state_lock);
 295
 296         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 297                 hlist_del(&fi->fi_hash);
 298                 spin_unlock(&state_lock);
 299                 nfsd4_free_file(fi);
 300         }
 301 }
 302
 303 static inline void
 304 get_nfs4_file(struct nfs4_file *fi)
 305 {
 306         atomic_inc(&fi->fi_ref);
 307 }
 308
 309 static struct file *
 310 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 311 {
 312         if (f->fi_fds[oflag])
 313                 return get_file(f->fi_fds[oflag]);
 314         return NULL;
 315 }
 316
 317 static struct file *
 318 find_writeable_file_locked(struct nfs4_file *f)
 319 {
 320         struct file *ret;
 321
 322         lockdep_assert_held(&f->fi_lock);
 323
 324         ret = __nfs4_get_fd(f, O_WRONLY);
 325         if (!ret)
 326                 ret = __nfs4_get_fd(f, O_RDWR);
 327         return ret;
 328 }
 329
 330 static struct file *
 331 find_writeable_file(struct nfs4_file *f)
 332 {
 333         struct file *ret;
 334
 335         spin_lock(&f->fi_lock);
 336         ret = find_writeable_file_locked(f);
 337         spin_unlock(&f->fi_lock);
 338
 339         return ret;
 340 }
 341
 342 static struct file *find_readable_file_locked(struct nfs4_file *f)
 343 {
 344         struct file *ret;
 345
 346         lockdep_assert_held(&f->fi_lock);
 347
 348         ret = __nfs4_get_fd(f, O_RDONLY);
 349         if (!ret)
 350                 ret = __nfs4_get_fd(f, O_RDWR);
 351         return ret;
 352 }
 353
 354 static struct file *
 355 find_readable_file(struct nfs4_file *f)
 356 {
 357         struct file *ret;
 358
 359         spin_lock(&f->fi_lock);
 360         ret = find_readable_file_locked(f);
 361         spin_unlock(&f->fi_lock);
 362
 363         return ret;
 364 }
 365
 366 static struct file *
 367 find_any_file(struct nfs4_file *f)
 368 {
 369         struct file *ret;
 370
 371         spin_lock(&f->fi_lock);
 372         ret = __nfs4_get_fd(f, O_RDWR);
 373         if (!ret) {
 374                 ret = __nfs4_get_fd(f, O_WRONLY);
 375                 if (!ret)
 376                         ret = __nfs4_get_fd(f, O_RDONLY);
 377         }
 378         spin_unlock(&f->fi_lock);
 379         return ret;
 380 }
 381
 382 static atomic_long_t num_delegations;
 383 unsigned long max_delegations;
 384
 385 /*
 386  * Open owner state (share locks)
 387  */
 388
 389 /* hash tables for lock and open owners */
 390 #define OWNER_HASH_BITS              8
 391 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 392 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 393
 394 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 395 {
 396         unsigned int ret;
 397
 398         ret = opaque_hashval(ownername->data, ownername->len);
 399         return ret & OWNER_HASH_MASK;
 400 }
 401
 402 /* hash table for nfs4_file */
 403 #define FILE_HASH_BITS                   8
 404 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 405
 406 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 407 {
 408         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 409 }
 410
 411 static unsigned int file_hashval(struct knfsd_fh *fh)
 412 {
 413         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 414 }
 415
 416 static bool nfsd_fh_match(struct knfsd_fh *fh1, struct knfsd_fh *fh2)
 417 {
 418         return fh1->fh_size == fh2->fh_size &&
 419                 !memcmp(fh1->fh_base.fh_pad,
 420                                 fh2->fh_base.fh_pad,
 421                                 fh1->fh_size);
 422 }
 423
 424 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 425
 426 static void
 427 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 428 {
 429         lockdep_assert_held(&fp->fi_lock);
 430
 431         if (access & NFS4_SHARE_ACCESS_WRITE)
 432                 atomic_inc(&fp->fi_access[O_WRONLY]);
 433         if (access & NFS4_SHARE_ACCESS_READ)
 434                 atomic_inc(&fp->fi_access[O_RDONLY]);
 435 }
 436
 437 static __be32
 438 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 439 {
 440         lockdep_assert_held(&fp->fi_lock);
 441
 442         /* Does this access mode make sense? */
 443         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 444                 return nfserr_inval;
 445
 446         /* Does it conflict with a deny mode already set? */
 447         if ((access & fp->fi_share_deny) != 0)
 448                 return nfserr_share_denied;
 449
 450         __nfs4_file_get_access(fp, access);
 451         return nfs_ok;
 452 }
 453
 454 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 455 {
 456         /* Common case is that there is no deny mode. */
 457         if (deny) {
 458                 /* Does this deny mode make sense? */
 459                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 460                         return nfserr_inval;
 461
 462                 if ((deny & NFS4_SHARE_DENY_READ) &&
 463                     atomic_read(&fp->fi_access[O_RDONLY]))
 464                         return nfserr_share_denied;
 465
 466                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 467                     atomic_read(&fp->fi_access[O_WRONLY]))
 468                         return nfserr_share_denied;
 469         }
 470         return nfs_ok;
 471 }
 472
 473 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 474 {
 475         might_lock(&fp->fi_lock);
 476
 477         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 478                 struct file *f1 = NULL;
 479                 struct file *f2 = NULL;
 480
 481                 swap(f1, fp->fi_fds[oflag]);
 482                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 483                         swap(f2, fp->fi_fds[O_RDWR]);
 484                 spin_unlock(&fp->fi_lock);
 485                 if (f1)
 486                         fput(f1);
 487                 if (f2)
 488                         fput(f2);
 489         }
 490 }
 491
 492 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 493 {
 494         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 495
 496         if (access & NFS4_SHARE_ACCESS_WRITE)
 497                 __nfs4_file_put_access(fp, O_WRONLY);
 498         if (access & NFS4_SHARE_ACCESS_READ)
 499                 __nfs4_file_put_access(fp, O_RDONLY);
 500 }
 501
 502 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
 503                                          struct kmem_cache *slab)
 504 {
 505         struct nfs4_stid *stid;
 506         int new_id;
 507
 508         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 509         if (!stid)
 510                 return NULL;
 511
 512         idr_preload(GFP_KERNEL);
 513         spin_lock(&cl->cl_lock);
 514         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 515         spin_unlock(&cl->cl_lock);
 516         idr_preload_end();
 517         if (new_id < 0)
 518                 goto out_free;
 519         stid->sc_client = cl;
 520         stid->sc_stateid.si_opaque.so_id = new_id;
 521         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 522         /* Will be incremented before return to client: */
 523         atomic_set(&stid->sc_count, 1);
 524
 525         /*
 526          * It shouldn't be a problem to reuse an opaque stateid value.
 527          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 528          * example, a stray write retransmission could be accepted by
 529          * the server when it should have been rejected.  Therefore,
 530          * adopt a trick from the sctp code to attempt to maximize the
 531          * amount of time until an id is reused, by ensuring they always
 532          * "increase" (mod INT_MAX):
 533          */
 534         return stid;
 535 out_free:
 536         kmem_cache_free(slab, stid);
 537         return NULL;
 538 }
 539
 540 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 541 {
 542         struct nfs4_stid *stid;
 543         struct nfs4_ol_stateid *stp;
 544
 545         stid = nfs4_alloc_stid(clp, stateid_slab);
 546         if (!stid)
 547                 return NULL;
 548
 549         stp = openlockstateid(stid);
 550         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 551         return stp;
 552 }
 553
 554 static void nfs4_free_deleg(struct nfs4_stid *stid)
 555 {
 556         kmem_cache_free(deleg_slab, stid);
 557         atomic_long_dec(&num_delegations);
 558 }
 559
 560 /*
 561  * When we recall a delegation, we should be careful not to hand it
 562  * out again straight away.
 563  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 564  * in which the filehandles of recalled delegations are "stored".
 565  * If a filehandle appear in either filter, a delegation is blocked.
 566  * When a delegation is recalled, the filehandle is stored in the "new"
 567  * filter.
 568  * Every 30 seconds we swap the filters and clear the "new" one,
 569  * unless both are empty of course.
 570  *
 571  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 572  * low 3 bytes as hash-table indices.
 573  *
 574  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 575  * is used to manage concurrent access.  Testing does not need the lock
 576  * except when swapping the two filters.
 577  */
 578 static DEFINE_SPINLOCK(blocked_delegations_lock);
 579 static struct bloom_pair {
 580         int     entries, old_entries;
 581         time_t  swap_time;
 582         int     new; /* index into 'set' */
 583         DECLARE_BITMAP(set[2], 256);
 584 } blocked_delegations;
 585
 586 static int delegation_blocked(struct knfsd_fh *fh)
 587 {
 588         u32 hash;
 589         struct bloom_pair *bd = &blocked_delegations;
 590
 591         if (bd->entries == 0)
 592                 return 0;
 593         if (seconds_since_boot() - bd->swap_time > 30) {
 594                 spin_lock(&blocked_delegations_lock);
 595                 if (seconds_since_boot() - bd->swap_time > 30) {
 596                         bd->entries -= bd->old_entries;
 597                         bd->old_entries = bd->entries;
 598                         memset(bd->set[bd->new], 0,
 599                                sizeof(bd->set[0]));
 600                         bd->new = 1-bd->new;
 601                         bd->swap_time = seconds_since_boot();
 602                 }
 603                 spin_unlock(&blocked_delegations_lock);
 604         }
 605         hash = arch_fast_hash(&fh->fh_base, fh->fh_size, 0);
 606         if (test_bit(hash&255, bd->set[0]) &&
 607             test_bit((hash>>8)&255, bd->set[0]) &&
 608             test_bit((hash>>16)&255, bd->set[0]))
 609                 return 1;
 610
 611         if (test_bit(hash&255, bd->set[1]) &&
 612             test_bit((hash>>8)&255, bd->set[1]) &&
 613             test_bit((hash>>16)&255, bd->set[1]))
 614                 return 1;
 615
 616         return 0;
 617 }
 618
 619 static void block_delegations(struct knfsd_fh *fh)
 620 {
 621         u32 hash;
 622         struct bloom_pair *bd = &blocked_delegations;
 623
 624         hash = arch_fast_hash(&fh->fh_base, fh->fh_size, 0);
 625
 626         spin_lock(&blocked_delegations_lock);
 627         __set_bit(hash&255, bd->set[bd->new]);
 628         __set_bit((hash>>8)&255, bd->set[bd->new]);
 629         __set_bit((hash>>16)&255, bd->set[bd->new]);
 630         if (bd->entries == 0)
 631                 bd->swap_time = seconds_since_boot();
 632         bd->entries += 1;
 633         spin_unlock(&blocked_delegations_lock);
 634 }
 635
 636 static struct nfs4_delegation *
 637 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh)
 638 {
 639         struct nfs4_delegation *dp;
 640         long n;
 641
 642         dprintk("NFSD alloc_init_deleg\n");
 643         n = atomic_long_inc_return(&num_delegations);
 644         if (n < 0 || n > max_delegations)
 645                 goto out_dec;
 646         if (delegation_blocked(&current_fh->fh_handle))
 647                 goto out_dec;
 648         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 649         if (dp == NULL)
 650                 goto out_dec;
 651
 652         dp->dl_stid.sc_free = nfs4_free_deleg;
 653         /*
 654          * delegation seqid's are never incremented.  The 4.1 special
 655          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 656          * 0 anyway just for consistency and use 1:
 657          */
 658         dp->dl_stid.sc_stateid.si_generation = 1;
 659         INIT_LIST_HEAD(&dp->dl_perfile);
 660         INIT_LIST_HEAD(&dp->dl_perclnt);
 661         INIT_LIST_HEAD(&dp->dl_recall_lru);
 662         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 663         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_run_cb_recall);
 664         return dp;
 665 out_dec:
 666         atomic_long_dec(&num_delegations);
 667         return NULL;
 668 }
 669
 670 void
 671 nfs4_put_stid(struct nfs4_stid *s)
 672 {
 673         struct nfs4_file *fp = s->sc_file;
 674         struct nfs4_client *clp = s->sc_client;
 675
 676         might_lock(&clp->cl_lock);
 677
 678         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 679                 wake_up_all(&close_wq);
 680                 return;
 681         }
 682         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 683         spin_unlock(&clp->cl_lock);
 684         s->sc_free(s);
 685         if (fp)
 686                 put_nfs4_file(fp);
 687 }
 688
 689 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 690 {
 691         lockdep_assert_held(&state_lock);
 692
 693         if (!fp->fi_lease)
 694                 return;
 695         if (atomic_dec_and_test(&fp->fi_delegees)) {
 696                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 697                 fp->fi_lease = NULL;
 698                 fput(fp->fi_deleg_file);
 699                 fp->fi_deleg_file = NULL;
 700         }
 701 }
 702
 703 static void unhash_stid(struct nfs4_stid *s)
 704 {
 705         s->sc_type = 0;
 706 }
 707
 708 static void
 709 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 710 {
 711         lockdep_assert_held(&state_lock);
 712         lockdep_assert_held(&fp->fi_lock);
 713
 714         atomic_inc(&dp->dl_stid.sc_count);
 715         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 716         list_add(&dp->dl_perfile, &fp->fi_delegations);
 717         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
 718 }
 719
 720 static void
 721 unhash_delegation_locked(struct nfs4_delegation *dp)
 722 {
 723         struct nfs4_file *fp = dp->dl_stid.sc_file;
 724
 725         lockdep_assert_held(&state_lock);
 726
 727         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 728         /* Ensure that deleg break won't try to requeue it */
 729         ++dp->dl_time;
 730         spin_lock(&fp->fi_lock);
 731         list_del_init(&dp->dl_perclnt);
 732         list_del_init(&dp->dl_recall_lru);
 733         list_del_init(&dp->dl_perfile);
 734         spin_unlock(&fp->fi_lock);
 735         if (fp)
 736                 nfs4_put_deleg_lease(fp);
 737 }
 738
 739 static void destroy_delegation(struct nfs4_delegation *dp)
 740 {
 741         spin_lock(&state_lock);
 742         unhash_delegation_locked(dp);
 743         spin_unlock(&state_lock);
 744         nfs4_put_stid(&dp->dl_stid);
 745 }
 746
 747 static void revoke_delegation(struct nfs4_delegation *dp)
 748 {
 749         struct nfs4_client *clp = dp->dl_stid.sc_client;
 750
 751         WARN_ON(!list_empty(&dp->dl_recall_lru));
 752
 753         if (clp->cl_minorversion == 0)
 754                 nfs4_put_stid(&dp->dl_stid);
 755         else {
 756                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 757                 spin_lock(&clp->cl_lock);
 758                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 759                 spin_unlock(&clp->cl_lock);
 760         }
 761 }
 762
 763 /*
 764  * SETCLIENTID state
 765  */
 766
 767 static unsigned int clientid_hashval(u32 id)
 768 {
 769         return id & CLIENT_HASH_MASK;
 770 }
 771
 772 static unsigned int clientstr_hashval(const char *name)
 773 {
 774         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 775 }
 776
 777 /*
 778  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 779  * st_{access,deny}_bmap field of the stateid, in order to track not
 780  * only what share bits are currently in force, but also what
 781  * combinations of share bits previous opens have used.  This allows us
 782  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 783  * return an error if the client attempt to downgrade to a combination
 784  * of share bits not explicable by closing some of its previous opens.
 785  *
 786  * XXX: This enforcement is actually incomplete, since we don't keep
 787  * track of access/deny bit combinations; so, e.g., we allow:
 788  *
 789  *      OPEN allow read, deny write
 790  *      OPEN allow both, deny none
 791  *      DOWNGRADE allow read, deny none
 792  *
 793  * which we should reject.
 794  */
 795 static unsigned int
 796 bmap_to_share_mode(unsigned long bmap) {
 797         int i;
 798         unsigned int access = 0;
 799
 800         for (i = 1; i < 4; i++) {
 801                 if (test_bit(i, &bmap))
 802                         access |= i;
 803         }
 804         return access;
 805 }
 806
 807 /* set share access for a given stateid */
 808 static inline void
 809 set_access(u32 access, struct nfs4_ol_stateid *stp)
 810 {
 811         unsigned char mask = 1 << access;
 812
 813         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 814         stp->st_access_bmap |= mask;
 815 }
 816
 817 /* clear share access for a given stateid */
 818 static inline void
 819 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 820 {
 821         unsigned char mask = 1 << access;
 822
 823         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 824         stp->st_access_bmap &= ~mask;
 825 }
 826
 827 /* test whether a given stateid has access */
 828 static inline bool
 829 test_access(u32 access, struct nfs4_ol_stateid *stp)
 830 {
 831         unsigned char mask = 1 << access;
 832
 833         return (bool)(stp->st_access_bmap & mask);
 834 }
 835
 836 /* set share deny for a given stateid */
 837 static inline void
 838 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 839 {
 840         unsigned char mask = 1 << deny;
 841
 842         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 843         stp->st_deny_bmap |= mask;
 844 }
 845
 846 /* clear share deny for a given stateid */
 847 static inline void
 848 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 849 {
 850         unsigned char mask = 1 << deny;
 851
 852         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 853         stp->st_deny_bmap &= ~mask;
 854 }
 855
 856 /* test whether a given stateid is denying specific access */
 857 static inline bool
 858 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 859 {
 860         unsigned char mask = 1 << deny;
 861
 862         return (bool)(stp->st_deny_bmap & mask);
 863 }
 864
 865 static int nfs4_access_to_omode(u32 access)
 866 {
 867         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 868         case NFS4_SHARE_ACCESS_READ:
 869                 return O_RDONLY;
 870         case NFS4_SHARE_ACCESS_WRITE:
 871                 return O_WRONLY;
 872         case NFS4_SHARE_ACCESS_BOTH:
 873                 return O_RDWR;
 874         }
 875         WARN_ON_ONCE(1);
 876         return O_RDONLY;
 877 }
 878
 879 /*
 880  * A stateid that had a deny mode associated with it is being released
 881  * or downgraded. Recalculate the deny mode on the file.
 882  */
 883 static void
 884 recalculate_deny_mode(struct nfs4_file *fp)
 885 {
 886         struct nfs4_ol_stateid *stp;
 887
 888         spin_lock(&fp->fi_lock);
 889         fp->fi_share_deny = 0;
 890         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
 891                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
 892         spin_unlock(&fp->fi_lock);
 893 }
 894
 895 static void
 896 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
 897 {
 898         int i;
 899         bool change = false;
 900
 901         for (i = 1; i < 4; i++) {
 902                 if ((i & deny) != i) {
 903                         change = true;
 904                         clear_deny(i, stp);
 905                 }
 906         }
 907
 908         /* Recalculate per-file deny mode if there was a change */
 909         if (change)
 910                 recalculate_deny_mode(stp->st_stid.sc_file);
 911 }
 912
 913 /* release all access and file references for a given stateid */
 914 static void
 915 release_all_access(struct nfs4_ol_stateid *stp)
 916 {
 917         int i;
 918         struct nfs4_file *fp = stp->st_stid.sc_file;
 919
 920         if (fp && stp->st_deny_bmap != 0)
 921                 recalculate_deny_mode(fp);
 922
 923         for (i = 1; i < 4; i++) {
 924                 if (test_access(i, stp))
 925                         nfs4_file_put_access(stp->st_stid.sc_file, i);
 926                 clear_access(i, stp);
 927         }
 928 }
 929
 930 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
 931 {
 932         struct nfs4_client *clp = sop->so_client;
 933
 934         might_lock(&clp->cl_lock);
 935
 936         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
 937                 return;
 938         sop->so_ops->so_unhash(sop);
 939         spin_unlock(&clp->cl_lock);
 940         kfree(sop->so_owner.data);
 941         sop->so_ops->so_free(sop);
 942 }
 943
 944 static void unhash_ol_stateid(struct nfs4_ol_stateid *stp)
 945 {
 946         struct nfs4_file *fp = stp->st_stid.sc_file;
 947
 948         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
 949
 950         spin_lock(&fp->fi_lock);
 951         list_del(&stp->st_perfile);
 952         spin_unlock(&fp->fi_lock);
 953         list_del(&stp->st_perstateowner);
 954 }
 955
 956 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
 957 {
 958         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 959
 960         release_all_access(stp);
 961         if (stp->st_stateowner)
 962                 nfs4_put_stateowner(stp->st_stateowner);
 963         kmem_cache_free(stateid_slab, stid);
 964 }
 965
 966 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
 967 {
 968         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 969         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
 970         struct file *file;
 971
 972         file = find_any_file(stp->st_stid.sc_file);
 973         if (file)
 974                 filp_close(file, (fl_owner_t)lo);
 975         nfs4_free_ol_stateid(stid);
 976 }
 977
 978 /*
 979  * Put the persistent reference to an already unhashed generic stateid, while
 980  * holding the cl_lock. If it's the last reference, then put it onto the
 981  * reaplist for later destruction.
 982  */
 983 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
 984                                        struct list_head *reaplist)
 985 {
 986         struct nfs4_stid *s = &stp->st_stid;
 987         struct nfs4_client *clp = s->sc_client;
 988
 989         lockdep_assert_held(&clp->cl_lock);
 990
 991         WARN_ON_ONCE(!list_empty(&stp->st_locks));
 992
 993         if (!atomic_dec_and_test(&s->sc_count)) {
 994                 wake_up_all(&close_wq);
 995                 return;
 996         }
 997
 998         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 999         list_add(&stp->st_locks, reaplist);
1000 }
1001
1002 static void unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1003 {
1004         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1005
1006         lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1007
1008         list_del_init(&stp->st_locks);
1009         unhash_ol_stateid(stp);
1010         unhash_stid(&stp->st_stid);
1011 }
1012
1013 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1014 {
1015         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1016
1017         spin_lock(&oo->oo_owner.so_client->cl_lock);
1018         unhash_lock_stateid(stp);
1019         spin_unlock(&oo->oo_owner.so_client->cl_lock);
1020         nfs4_put_stid(&stp->st_stid);
1021 }
1022
1023 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1024 {
1025         struct nfs4_client *clp = lo->lo_owner.so_client;
1026
1027         lockdep_assert_held(&clp->cl_lock);
1028
1029         list_del_init(&lo->lo_owner.so_strhash);
1030 }
1031
1032 /*
1033  * Free a list of generic stateids that were collected earlier after being
1034  * fully unhashed.
1035  */
1036 static void
1037 free_ol_stateid_reaplist(struct list_head *reaplist)
1038 {
1039         struct nfs4_ol_stateid *stp;
1040         struct nfs4_file *fp;
1041
1042         might_sleep();
1043
1044         while (!list_empty(reaplist)) {
1045                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1046                                        st_locks);
1047                 list_del(&stp->st_locks);
1048                 fp = stp->st_stid.sc_file;
1049                 stp->st_stid.sc_free(&stp->st_stid);
1050                 if (fp)
1051                         put_nfs4_file(fp);
1052         }
1053 }
1054
1055 static void release_lockowner(struct nfs4_lockowner *lo)
1056 {
1057         struct nfs4_client *clp = lo->lo_owner.so_client;
1058         struct nfs4_ol_stateid *stp;
1059         struct list_head reaplist;
1060
1061         INIT_LIST_HEAD(&reaplist);
1062
1063         spin_lock(&clp->cl_lock);
1064         unhash_lockowner_locked(lo);
1065         while (!list_empty(&lo->lo_owner.so_stateids)) {
1066                 stp = list_first_entry(&lo->lo_owner.so_stateids,
1067                                 struct nfs4_ol_stateid, st_perstateowner);
1068                 unhash_lock_stateid(stp);
1069                 put_ol_stateid_locked(stp, &reaplist);
1070         }
1071         spin_unlock(&clp->cl_lock);
1072         free_ol_stateid_reaplist(&reaplist);
1073         nfs4_put_stateowner(&lo->lo_owner);
1074 }
1075
1076 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1077                                        struct list_head *reaplist)
1078 {
1079         struct nfs4_ol_stateid *stp;
1080
1081         while (!list_empty(&open_stp->st_locks)) {
1082                 stp = list_entry(open_stp->st_locks.next,
1083                                 struct nfs4_ol_stateid, st_locks);
1084                 unhash_lock_stateid(stp);
1085                 put_ol_stateid_locked(stp, reaplist);
1086         }
1087 }
1088
1089 static void unhash_open_stateid(struct nfs4_ol_stateid *stp,
1090                                 struct list_head *reaplist)
1091 {
1092         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1093
1094         unhash_ol_stateid(stp);
1095         release_open_stateid_locks(stp, reaplist);
1096 }
1097
1098 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1099 {
1100         LIST_HEAD(reaplist);
1101
1102         spin_lock(&stp->st_stid.sc_client->cl_lock);
1103         unhash_open_stateid(stp, &reaplist);
1104         put_ol_stateid_locked(stp, &reaplist);
1105         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1106         free_ol_stateid_reaplist(&reaplist);
1107 }
1108
1109 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1110 {
1111         struct nfs4_client *clp = oo->oo_owner.so_client;
1112
1113         lockdep_assert_held(&clp->cl_lock);
1114
1115         list_del_init(&oo->oo_owner.so_strhash);
1116         list_del_init(&oo->oo_perclient);
1117 }
1118
1119 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1120 {
1121         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1122                                           nfsd_net_id);
1123         struct nfs4_ol_stateid *s;
1124
1125         spin_lock(&nn->client_lock);
1126         s = oo->oo_last_closed_stid;
1127         if (s) {
1128                 list_del_init(&oo->oo_close_lru);
1129                 oo->oo_last_closed_stid = NULL;
1130         }
1131         spin_unlock(&nn->client_lock);
1132         if (s)
1133                 nfs4_put_stid(&s->st_stid);
1134 }
1135
1136 static void release_openowner(struct nfs4_openowner *oo)
1137 {
1138         struct nfs4_ol_stateid *stp;
1139         struct nfs4_client *clp = oo->oo_owner.so_client;
1140         struct list_head reaplist;
1141
1142         INIT_LIST_HEAD(&reaplist);
1143
1144         spin_lock(&clp->cl_lock);
1145         unhash_openowner_locked(oo);
1146         while (!list_empty(&oo->oo_owner.so_stateids)) {
1147                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1148                                 struct nfs4_ol_stateid, st_perstateowner);
1149                 unhash_open_stateid(stp, &reaplist);
1150                 put_ol_stateid_locked(stp, &reaplist);
1151         }
1152         spin_unlock(&clp->cl_lock);
1153         free_ol_stateid_reaplist(&reaplist);
1154         release_last_closed_stateid(oo);
1155         nfs4_put_stateowner(&oo->oo_owner);
1156 }
1157
1158 static inline int
1159 hash_sessionid(struct nfs4_sessionid *sessionid)
1160 {
1161         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1162
1163         return sid->sequence % SESSION_HASH_SIZE;
1164 }
1165
1166 #ifdef NFSD_DEBUG
1167 static inline void
1168 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1169 {
1170         u32 *ptr = (u32 *)(&sessionid->data[0]);
1171         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1172 }
1173 #else
1174 static inline void
1175 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1176 {
1177 }
1178 #endif
1179
1180 /*
1181  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1182  * won't be used for replay.
1183  */
1184 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1185 {
1186         struct nfs4_stateowner *so = cstate->replay_owner;
1187
1188         if (nfserr == nfserr_replay_me)
1189                 return;
1190
1191         if (!seqid_mutating_err(ntohl(nfserr))) {
1192                 nfsd4_cstate_clear_replay(cstate);
1193                 return;
1194         }
1195         if (!so)
1196                 return;
1197         if (so->so_is_open_owner)
1198                 release_last_closed_stateid(openowner(so));
1199         so->so_seqid++;
1200         return;
1201 }
1202
1203 static void
1204 gen_sessionid(struct nfsd4_session *ses)
1205 {
1206         struct nfs4_client *clp = ses->se_client;
1207         struct nfsd4_sessionid *sid;
1208
1209         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1210         sid->clientid = clp->cl_clientid;
1211         sid->sequence = current_sessionid++;
1212         sid->reserved = 0;
1213 }
1214
1215 /*
1216  * The protocol defines ca_maxresponssize_cached to include the size of
1217  * the rpc header, but all we need to cache is the data starting after
1218  * the end of the initial SEQUENCE operation--the rest we regenerate
1219  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1220  * value that is the number of bytes in our cache plus a few additional
1221  * bytes.  In order to stay on the safe side, and not promise more than
1222  * we can cache, those additional bytes must be the minimum possible: 24
1223  * bytes of rpc header (xid through accept state, with AUTH_NULL
1224  * verifier), 12 for the compound header (with zero-length tag), and 44
1225  * for the SEQUENCE op response:
1226  */
1227 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1228
1229 static void
1230 free_session_slots(struct nfsd4_session *ses)
1231 {
1232         int i;
1233
1234         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1235                 kfree(ses->se_slots[i]);
1236 }
1237
1238 /*
1239  * We don't actually need to cache the rpc and session headers, so we
1240  * can allocate a little less for each slot:
1241  */
1242 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1243 {
1244         u32 size;
1245
1246         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1247                 size = 0;
1248         else
1249                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1250         return size + sizeof(struct nfsd4_slot);
1251 }
1252
1253 /*
1254  * XXX: If we run out of reserved DRC memory we could (up to a point)
1255  * re-negotiate active sessions and reduce their slot usage to make
1256  * room for new connections. For now we just fail the create session.
1257  */
1258 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1259 {
1260         u32 slotsize = slot_bytes(ca);
1261         u32 num = ca->maxreqs;
1262         int avail;
1263
1264         spin_lock(&nfsd_drc_lock);
1265         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1266                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1267         num = min_t(int, num, avail / slotsize);
1268         nfsd_drc_mem_used += num * slotsize;
1269         spin_unlock(&nfsd_drc_lock);
1270
1271         return num;
1272 }
1273
1274 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1275 {
1276         int slotsize = slot_bytes(ca);
1277
1278         spin_lock(&nfsd_drc_lock);
1279         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1280         spin_unlock(&nfsd_drc_lock);
1281 }
1282
1283 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1284                                            struct nfsd4_channel_attrs *battrs)
1285 {
1286         int numslots = fattrs->maxreqs;
1287         int slotsize = slot_bytes(fattrs);
1288         struct nfsd4_session *new;
1289         int mem, i;
1290
1291         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1292                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1293         mem = numslots * sizeof(struct nfsd4_slot *);
1294
1295         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1296         if (!new)
1297                 return NULL;
1298         /* allocate each struct nfsd4_slot and data cache in one piece */
1299         for (i = 0; i < numslots; i++) {
1300                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1301                 if (!new->se_slots[i])
1302                         goto out_free;
1303         }
1304
1305         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1306         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1307
1308         return new;
1309 out_free:
1310         while (i--)
1311                 kfree(new->se_slots[i]);
1312         kfree(new);
1313         return NULL;
1314 }
1315
1316 static void free_conn(struct nfsd4_conn *c)
1317 {
1318         svc_xprt_put(c->cn_xprt);
1319         kfree(c);
1320 }
1321
1322 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1323 {
1324         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1325         struct nfs4_client *clp = c->cn_session->se_client;
1326
1327         spin_lock(&clp->cl_lock);
1328         if (!list_empty(&c->cn_persession)) {
1329                 list_del(&c->cn_persession);
1330                 free_conn(c);
1331         }
1332         nfsd4_probe_callback(clp);
1333         spin_unlock(&clp->cl_lock);
1334 }
1335
1336 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1337 {
1338         struct nfsd4_conn *conn;
1339
1340         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1341         if (!conn)
1342                 return NULL;
1343         svc_xprt_get(rqstp->rq_xprt);
1344         conn->cn_xprt = rqstp->rq_xprt;
1345         conn->cn_flags = flags;
1346         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1347         return conn;
1348 }
1349
1350 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1351 {
1352         conn->cn_session = ses;
1353         list_add(&conn->cn_persession, &ses->se_conns);
1354 }
1355
1356 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1357 {
1358         struct nfs4_client *clp = ses->se_client;
1359
1360         spin_lock(&clp->cl_lock);
1361         __nfsd4_hash_conn(conn, ses);
1362         spin_unlock(&clp->cl_lock);
1363 }
1364
1365 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1366 {
1367         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1368         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1369 }
1370
1371 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1372 {
1373         int ret;
1374
1375         nfsd4_hash_conn(conn, ses);
1376         ret = nfsd4_register_conn(conn);
1377         if (ret)
1378                 /* oops; xprt is already down: */
1379                 nfsd4_conn_lost(&conn->cn_xpt_user);
1380         /* We may have gained or lost a callback channel: */
1381         nfsd4_probe_callback_sync(ses->se_client);
1382 }
1383
1384 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1385 {
1386         u32 dir = NFS4_CDFC4_FORE;
1387
1388         if (cses->flags & SESSION4_BACK_CHAN)
1389                 dir |= NFS4_CDFC4_BACK;
1390         return alloc_conn(rqstp, dir);
1391 }
1392
1393 /* must be called under client_lock */
1394 static void nfsd4_del_conns(struct nfsd4_session *s)
1395 {
1396         struct nfs4_client *clp = s->se_client;
1397         struct nfsd4_conn *c;
1398
1399         spin_lock(&clp->cl_lock);
1400         while (!list_empty(&s->se_conns)) {
1401                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1402                 list_del_init(&c->cn_persession);
1403                 spin_unlock(&clp->cl_lock);
1404
1405                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1406                 free_conn(c);
1407
1408                 spin_lock(&clp->cl_lock);
1409         }
1410         spin_unlock(&clp->cl_lock);
1411 }
1412
1413 static void __free_session(struct nfsd4_session *ses)
1414 {
1415         free_session_slots(ses);
1416         kfree(ses);
1417 }
1418
1419 static void free_session(struct nfsd4_session *ses)
1420 {
1421         nfsd4_del_conns(ses);
1422         nfsd4_put_drc_mem(&ses->se_fchannel);
1423         __free_session(ses);
1424 }
1425
1426 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1427 {
1428         int idx;
1429         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1430
1431         new->se_client = clp;
1432         gen_sessionid(new);
1433
1434         INIT_LIST_HEAD(&new->se_conns);
1435
1436         new->se_cb_seq_nr = 1;
1437         new->se_flags = cses->flags;
1438         new->se_cb_prog = cses->callback_prog;
1439         new->se_cb_sec = cses->cb_sec;
1440         atomic_set(&new->se_ref, 0);
1441         idx = hash_sessionid(&new->se_sessionid);
1442         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1443         spin_lock(&clp->cl_lock);
1444         list_add(&new->se_perclnt, &clp->cl_sessions);
1445         spin_unlock(&clp->cl_lock);
1446
1447         if (cses->flags & SESSION4_BACK_CHAN) {
1448                 struct sockaddr *sa = svc_addr(rqstp);
1449                 /*
1450                  * This is a little silly; with sessions there's no real
1451                  * use for the callback address.  Use the peer address
1452                  * as a reasonable default for now, but consider fixing
1453                  * the rpc client not to require an address in the
1454                  * future:
1455                  */
1456                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1457                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1458         }
1459 }
1460
1461 /* caller must hold client_lock */
1462 static struct nfsd4_session *
1463 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1464 {
1465         struct nfsd4_session *elem;
1466         int idx;
1467         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1468
1469         lockdep_assert_held(&nn->client_lock);
1470
1471         dump_sessionid(__func__, sessionid);
1472         idx = hash_sessionid(sessionid);
1473         /* Search in the appropriate list */
1474         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1475                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1476                             NFS4_MAX_SESSIONID_LEN)) {
1477                         return elem;
1478                 }
1479         }
1480
1481         dprintk("%s: session not found\n", __func__);
1482         return NULL;
1483 }
1484
1485 static struct nfsd4_session *
1486 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1487                 __be32 *ret)
1488 {
1489         struct nfsd4_session *session;
1490         __be32 status = nfserr_badsession;
1491
1492         session = __find_in_sessionid_hashtbl(sessionid, net);
1493         if (!session)
1494                 goto out;
1495         status = nfsd4_get_session_locked(session);
1496         if (status)
1497                 session = NULL;
1498 out:
1499         *ret = status;
1500         return session;
1501 }
1502
1503 /* caller must hold client_lock */
1504 static void
1505 unhash_session(struct nfsd4_session *ses)
1506 {
1507         struct nfs4_client *clp = ses->se_client;
1508         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1509
1510         lockdep_assert_held(&nn->client_lock);
1511
1512         list_del(&ses->se_hash);
1513         spin_lock(&ses->se_client->cl_lock);
1514         list_del(&ses->se_perclnt);
1515         spin_unlock(&ses->se_client->cl_lock);
1516 }
1517
1518 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1519 static int
1520 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1521 {
1522         if (clid->cl_boot == nn->boot_time)
1523                 return 0;
1524         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1525                 clid->cl_boot, clid->cl_id, nn->boot_time);
1526         return 1;
1527 }
1528
1529 /*
1530  * XXX Should we use a slab cache ?
1531  * This type of memory management is somewhat inefficient, but we use it
1532  * anyway since SETCLIENTID is not a common operation.
1533  */
1534 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1535 {
1536         struct nfs4_client *clp;
1537         int i;
1538
1539         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1540         if (clp == NULL)
1541                 return NULL;
1542         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1543         if (clp->cl_name.data == NULL)
1544                 goto err_no_name;
1545         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1546                         OWNER_HASH_SIZE, GFP_KERNEL);
1547         if (!clp->cl_ownerstr_hashtbl)
1548                 goto err_no_hashtbl;
1549         for (i = 0; i < OWNER_HASH_SIZE; i++)
1550                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1551         clp->cl_name.len = name.len;
1552         INIT_LIST_HEAD(&clp->cl_sessions);
1553         idr_init(&clp->cl_stateids);
1554         atomic_set(&clp->cl_refcount, 0);
1555         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1556         INIT_LIST_HEAD(&clp->cl_idhash);
1557         INIT_LIST_HEAD(&clp->cl_openowners);
1558         INIT_LIST_HEAD(&clp->cl_delegations);
1559         INIT_LIST_HEAD(&clp->cl_lru);
1560         INIT_LIST_HEAD(&clp->cl_callbacks);
1561         INIT_LIST_HEAD(&clp->cl_revoked);
1562         spin_lock_init(&clp->cl_lock);
1563         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1564         return clp;
1565 err_no_hashtbl:
1566         kfree(clp->cl_name.data);
1567 err_no_name:
1568         kfree(clp);
1569         return NULL;
1570 }
1571
1572 static void
1573 free_client(struct nfs4_client *clp)
1574 {
1575         while (!list_empty(&clp->cl_sessions)) {
1576                 struct nfsd4_session *ses;
1577                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1578                                 se_perclnt);
1579                 list_del(&ses->se_perclnt);
1580                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1581                 free_session(ses);
1582         }
1583         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1584         free_svc_cred(&clp->cl_cred);
1585         kfree(clp->cl_ownerstr_hashtbl);
1586         kfree(clp->cl_name.data);
1587         idr_destroy(&clp->cl_stateids);
1588         kfree(clp);
1589 }
1590
1591 /* must be called under the client_lock */
1592 static void
1593 unhash_client_locked(struct nfs4_client *clp)
1594 {
1595         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1596         struct nfsd4_session *ses;
1597
1598         lockdep_assert_held(&nn->client_lock);
1599
1600         /* Mark the client as expired! */
1601         clp->cl_time = 0;
1602         /* Make it invisible */
1603         if (!list_empty(&clp->cl_idhash)) {
1604                 list_del_init(&clp->cl_idhash);
1605                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1606                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1607                 else
1608                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1609         }
1610         list_del_init(&clp->cl_lru);
1611         spin_lock(&clp->cl_lock);
1612         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1613                 list_del_init(&ses->se_hash);
1614         spin_unlock(&clp->cl_lock);
1615 }
1616
1617 static void
1618 unhash_client(struct nfs4_client *clp)
1619 {
1620         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1621
1622         spin_lock(&nn->client_lock);
1623         unhash_client_locked(clp);
1624         spin_unlock(&nn->client_lock);
1625 }
1626
1627 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1628 {
1629         if (atomic_read(&clp->cl_refcount))
1630                 return nfserr_jukebox;
1631         unhash_client_locked(clp);
1632         return nfs_ok;
1633 }
1634
1635 static void
1636 __destroy_client(struct nfs4_client *clp)
1637 {
1638         struct nfs4_openowner *oo;
1639         struct nfs4_delegation *dp;
1640         struct list_head reaplist;
1641
1642         INIT_LIST_HEAD(&reaplist);
1643         spin_lock(&state_lock);
1644         while (!list_empty(&clp->cl_delegations)) {
1645                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1646                 unhash_delegation_locked(dp);
1647                 list_add(&dp->dl_recall_lru, &reaplist);
1648         }
1649         spin_unlock(&state_lock);
1650         while (!list_empty(&reaplist)) {
1651                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1652                 list_del_init(&dp->dl_recall_lru);
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                 atomic_inc(&oo->oo_owner.so_count);
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 long long
1718 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1719 {
1720         long long res;
1721
1722         res = o1->len - o2->len;
1723         if (res)
1724                 return res;
1725         return (long long)memcmp(o1->data, o2->data, o1->len);
1726 }
1727
1728 static int same_name(const char *n1, const char *n2)
1729 {
1730         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1731 }
1732
1733 static int
1734 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1735 {
1736         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1737 }
1738
1739 static int
1740 same_clid(clientid_t *cl1, clientid_t *cl2)
1741 {
1742         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1743 }
1744
1745 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1746 {
1747         int i;
1748
1749         if (g1->ngroups != g2->ngroups)
1750                 return false;
1751         for (i=0; i<g1->ngroups; i++)
1752                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1753                         return false;
1754         return true;
1755 }
1756
1757 /*
1758  * RFC 3530 language requires clid_inuse be returned when the
1759  * "principal" associated with a requests differs from that previously
1760  * used.  We use uid, gid's, and gss principal string as our best
1761  * approximation.  We also don't want to allow non-gss use of a client
1762  * established using gss: in theory cr_principal should catch that
1763  * change, but in practice cr_principal can be null even in the gss case
1764  * since gssd doesn't always pass down a principal string.
1765  */
1766 static bool is_gss_cred(struct svc_cred *cr)
1767 {
1768         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1769         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1770 }
1771
1772
1773 static bool
1774 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1775 {
1776         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1777                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1778                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1779                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1780                 return false;
1781         if (cr1->cr_principal == cr2->cr_principal)
1782                 return true;
1783         if (!cr1->cr_principal || !cr2->cr_principal)
1784                 return false;
1785         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1786 }
1787
1788 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1789 {
1790         struct svc_cred *cr = &rqstp->rq_cred;
1791         u32 service;
1792
1793         if (!cr->cr_gss_mech)
1794                 return false;
1795         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1796         return service == RPC_GSS_SVC_INTEGRITY ||
1797                service == RPC_GSS_SVC_PRIVACY;
1798 }
1799
1800 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1801 {
1802         struct svc_cred *cr = &rqstp->rq_cred;
1803
1804         if (!cl->cl_mach_cred)
1805                 return true;
1806         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1807                 return false;
1808         if (!svc_rqst_integrity_protected(rqstp))
1809                 return false;
1810         if (!cr->cr_principal)
1811                 return false;
1812         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1813 }
1814
1815 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1816 {
1817         __be32 verf[2];
1818
1819         /*
1820          * This is opaque to client, so no need to byte-swap. Use
1821          * __force to keep sparse happy
1822          */
1823         verf[0] = (__force __be32)get_seconds();
1824         verf[1] = (__force __be32)nn->clientid_counter;
1825         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1826 }
1827
1828 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1829 {
1830         clp->cl_clientid.cl_boot = nn->boot_time;
1831         clp->cl_clientid.cl_id = nn->clientid_counter++;
1832         gen_confirm(clp, nn);
1833 }
1834
1835 static struct nfs4_stid *
1836 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
1837 {
1838         struct nfs4_stid *ret;
1839
1840         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1841         if (!ret || !ret->sc_type)
1842                 return NULL;
1843         return ret;
1844 }
1845
1846 static struct nfs4_stid *
1847 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1848 {
1849         struct nfs4_stid *s;
1850
1851         spin_lock(&cl->cl_lock);
1852         s = find_stateid_locked(cl, t);
1853         if (s != NULL) {
1854                 if (typemask & s->sc_type)
1855                         atomic_inc(&s->sc_count);
1856                 else
1857                         s = NULL;
1858         }
1859         spin_unlock(&cl->cl_lock);
1860         return s;
1861 }
1862
1863 static struct nfs4_client *create_client(struct xdr_netobj name,
1864                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1865 {
1866         struct nfs4_client *clp;
1867         struct sockaddr *sa = svc_addr(rqstp);
1868         int ret;
1869         struct net *net = SVC_NET(rqstp);
1870
1871         clp = alloc_client(name);
1872         if (clp == NULL)
1873                 return NULL;
1874
1875         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1876         if (ret) {
1877                 free_client(clp);
1878                 return NULL;
1879         }
1880         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_run_cb_null);
1881         clp->cl_time = get_seconds();
1882         clear_bit(0, &clp->cl_cb_slot_busy);
1883         copy_verf(clp, verf);
1884         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1885         clp->cl_cb_session = NULL;
1886         clp->net = net;
1887         return clp;
1888 }
1889
1890 static void
1891 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1892 {
1893         struct rb_node **new = &(root->rb_node), *parent = NULL;
1894         struct nfs4_client *clp;
1895
1896         while (*new) {
1897                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1898                 parent = *new;
1899
1900                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1901                         new = &((*new)->rb_left);
1902                 else
1903                         new = &((*new)->rb_right);
1904         }
1905
1906         rb_link_node(&new_clp->cl_namenode, parent, new);
1907         rb_insert_color(&new_clp->cl_namenode, root);
1908 }
1909
1910 static struct nfs4_client *
1911 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1912 {
1913         long long cmp;
1914         struct rb_node *node = root->rb_node;
1915         struct nfs4_client *clp;
1916
1917         while (node) {
1918                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1919                 cmp = compare_blob(&clp->cl_name, name);
1920                 if (cmp > 0)
1921                         node = node->rb_left;
1922                 else if (cmp < 0)
1923                         node = node->rb_right;
1924                 else
1925                         return clp;
1926         }
1927         return NULL;
1928 }
1929
1930 static void
1931 add_to_unconfirmed(struct nfs4_client *clp)
1932 {
1933         unsigned int idhashval;
1934         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1935
1936         lockdep_assert_held(&nn->client_lock);
1937
1938         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1939         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1940         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1941         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1942         renew_client_locked(clp);
1943 }
1944
1945 static void
1946 move_to_confirmed(struct nfs4_client *clp)
1947 {
1948         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1949         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1950
1951         lockdep_assert_held(&nn->client_lock);
1952
1953         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1954         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1955         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1956         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1957         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1958         renew_client_locked(clp);
1959 }
1960
1961 static struct nfs4_client *
1962 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1963 {
1964         struct nfs4_client *clp;
1965         unsigned int idhashval = clientid_hashval(clid->cl_id);
1966
1967         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1968                 if (same_clid(&clp->cl_clientid, clid)) {
1969                         if ((bool)clp->cl_minorversion != sessions)
1970                                 return NULL;
1971                         renew_client_locked(clp);
1972                         return clp;
1973                 }
1974         }
1975         return NULL;
1976 }
1977
1978 static struct nfs4_client *
1979 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1980 {
1981         struct list_head *tbl = nn->conf_id_hashtbl;
1982
1983         lockdep_assert_held(&nn->client_lock);
1984         return find_client_in_id_table(tbl, clid, sessions);
1985 }
1986
1987 static struct nfs4_client *
1988 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1989 {
1990         struct list_head *tbl = nn->unconf_id_hashtbl;
1991
1992         lockdep_assert_held(&nn->client_lock);
1993         return find_client_in_id_table(tbl, clid, sessions);
1994 }
1995
1996 static bool clp_used_exchangeid(struct nfs4_client *clp)
1997 {
1998         return clp->cl_exchange_flags != 0;
1999 }
2000
2001 static struct nfs4_client *
2002 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2003 {
2004         lockdep_assert_held(&nn->client_lock);
2005         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2006 }
2007
2008 static struct nfs4_client *
2009 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2010 {
2011         lockdep_assert_held(&nn->client_lock);
2012         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2013 }
2014
2015 static void
2016 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2017 {
2018         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2019         struct sockaddr *sa = svc_addr(rqstp);
2020         u32 scopeid = rpc_get_scope_id(sa);
2021         unsigned short expected_family;
2022
2023         /* Currently, we only support tcp and tcp6 for the callback channel */
2024         if (se->se_callback_netid_len == 3 &&
2025             !memcmp(se->se_callback_netid_val, "tcp", 3))
2026                 expected_family = AF_INET;
2027         else if (se->se_callback_netid_len == 4 &&
2028                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2029                 expected_family = AF_INET6;
2030         else
2031                 goto out_err;
2032
2033         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2034                                             se->se_callback_addr_len,
2035                                             (struct sockaddr *)&conn->cb_addr,
2036                                             sizeof(conn->cb_addr));
2037
2038         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2039                 goto out_err;
2040
2041         if (conn->cb_addr.ss_family == AF_INET6)
2042                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2043
2044         conn->cb_prog = se->se_callback_prog;
2045         conn->cb_ident = se->se_callback_ident;
2046         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2047         return;
2048 out_err:
2049         conn->cb_addr.ss_family = AF_UNSPEC;
2050         conn->cb_addrlen = 0;
2051         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2052                 "will not receive delegations\n",
2053                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2054
2055         return;
2056 }
2057
2058 /*
2059  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2060  */
2061 static void
2062 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2063 {
2064         struct xdr_buf *buf = resp->xdr.buf;
2065         struct nfsd4_slot *slot = resp->cstate.slot;
2066         unsigned int base;
2067
2068         dprintk("--> %s slot %p\n", __func__, slot);
2069
2070         slot->sl_opcnt = resp->opcnt;
2071         slot->sl_status = resp->cstate.status;
2072
2073         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2074         if (nfsd4_not_cached(resp)) {
2075                 slot->sl_datalen = 0;
2076                 return;
2077         }
2078         base = resp->cstate.data_offset;
2079         slot->sl_datalen = buf->len - base;
2080         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2081                 WARN("%s: sessions DRC could not cache compound\n", __func__);
2082         return;
2083 }
2084
2085 /*
2086  * Encode the replay sequence operation from the slot values.
2087  * If cachethis is FALSE encode the uncached rep error on the next
2088  * operation which sets resp->p and increments resp->opcnt for
2089  * nfs4svc_encode_compoundres.
2090  *
2091  */
2092 static __be32
2093 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2094                           struct nfsd4_compoundres *resp)
2095 {
2096         struct nfsd4_op *op;
2097         struct nfsd4_slot *slot = resp->cstate.slot;
2098
2099         /* Encode the replayed sequence operation */
2100         op = &args->ops[resp->opcnt - 1];
2101         nfsd4_encode_operation(resp, op);
2102
2103         /* Return nfserr_retry_uncached_rep in next operation. */
2104         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2105                 op = &args->ops[resp->opcnt++];
2106                 op->status = nfserr_retry_uncached_rep;
2107                 nfsd4_encode_operation(resp, op);
2108         }
2109         return op->status;
2110 }
2111
2112 /*
2113  * The sequence operation is not cached because we can use the slot and
2114  * session values.
2115  */
2116 static __be32
2117 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2118                          struct nfsd4_sequence *seq)
2119 {
2120         struct nfsd4_slot *slot = resp->cstate.slot;
2121         struct xdr_stream *xdr = &resp->xdr;
2122         __be32 *p;
2123         __be32 status;
2124
2125         dprintk("--> %s slot %p\n", __func__, slot);
2126
2127         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2128         if (status)
2129                 return status;
2130
2131         p = xdr_reserve_space(xdr, slot->sl_datalen);
2132         if (!p) {
2133                 WARN_ON_ONCE(1);
2134                 return nfserr_serverfault;
2135         }
2136         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2137         xdr_commit_encode(xdr);
2138
2139         resp->opcnt = slot->sl_opcnt;
2140         return slot->sl_status;
2141 }
2142
2143 /*
2144  * Set the exchange_id flags returned by the server.
2145  */
2146 static void
2147 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2148 {
2149         /* pNFS is not supported */
2150         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2151
2152         /* Referrals are supported, Migration is not. */
2153         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2154
2155         /* set the wire flags to return to client. */
2156         clid->flags = new->cl_exchange_flags;
2157 }
2158
2159 static bool client_has_state(struct nfs4_client *clp)
2160 {
2161         /*
2162          * Note clp->cl_openowners check isn't quite right: there's no
2163          * need to count owners without stateid's.
2164          *
2165          * Also note we should probably be using this in 4.0 case too.
2166          */
2167         return !list_empty(&clp->cl_openowners)
2168                 || !list_empty(&clp->cl_delegations)
2169                 || !list_empty(&clp->cl_sessions);
2170 }
2171
2172 __be32
2173 nfsd4_exchange_id(struct svc_rqst *rqstp,
2174                   struct nfsd4_compound_state *cstate,
2175                   struct nfsd4_exchange_id *exid)
2176 {
2177         struct nfs4_client *conf, *new;
2178         struct nfs4_client *unconf = NULL;
2179         __be32 status;
2180         char                    addr_str[INET6_ADDRSTRLEN];
2181         nfs4_verifier           verf = exid->verifier;
2182         struct sockaddr         *sa = svc_addr(rqstp);
2183         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2184         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2185
2186         rpc_ntop(sa, addr_str, sizeof(addr_str));
2187         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2188                 "ip_addr=%s flags %x, spa_how %d\n",
2189                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2190                 addr_str, exid->flags, exid->spa_how);
2191
2192         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2193                 return nfserr_inval;
2194
2195         switch (exid->spa_how) {
2196         case SP4_MACH_CRED:
2197                 if (!svc_rqst_integrity_protected(rqstp))
2198                         return nfserr_inval;
2199         case SP4_NONE:
2200                 break;
2201         default:                                /* checked by xdr code */
2202                 WARN_ON_ONCE(1);
2203         case SP4_SSV:
2204                 return nfserr_encr_alg_unsupp;
2205         }
2206
2207         new = create_client(exid->clname, rqstp, &verf);
2208         if (new == NULL)
2209                 return nfserr_jukebox;
2210
2211         /* Cases below refer to rfc 5661 section 18.35.4: */
2212         nfs4_lock_state();
2213         spin_lock(&nn->client_lock);
2214         conf = find_confirmed_client_by_name(&exid->clname, nn);
2215         if (conf) {
2216                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2217                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2218
2219                 if (update) {
2220                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2221                                 status = nfserr_inval;
2222                                 goto out;
2223                         }
2224                         if (!mach_creds_match(conf, rqstp)) {
2225                                 status = nfserr_wrong_cred;
2226                                 goto out;
2227                         }
2228                         if (!creds_match) { /* case 9 */
2229                                 status = nfserr_perm;
2230                                 goto out;
2231                         }
2232                         if (!verfs_match) { /* case 8 */
2233                                 status = nfserr_not_same;
2234                                 goto out;
2235                         }
2236                         /* case 6 */
2237                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2238                         goto out_copy;
2239                 }
2240                 if (!creds_match) { /* case 3 */
2241                         if (client_has_state(conf)) {
2242                                 status = nfserr_clid_inuse;
2243                                 goto out;
2244                         }
2245                         goto out_new;
2246                 }
2247                 if (verfs_match) { /* case 2 */
2248                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2249                         goto out_copy;
2250                 }
2251                 /* case 5, client reboot */
2252                 conf = NULL;
2253                 goto out_new;
2254         }
2255
2256         if (update) { /* case 7 */
2257                 status = nfserr_noent;
2258                 goto out;
2259         }
2260
2261         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2262         if (unconf) /* case 4, possible retry or client restart */
2263                 unhash_client_locked(unconf);
2264
2265         /* case 1 (normal case) */
2266 out_new:
2267         if (conf) {
2268                 status = mark_client_expired_locked(conf);
2269                 if (status)
2270                         goto out;
2271         }
2272         new->cl_minorversion = cstate->minorversion;
2273         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
2274
2275         gen_clid(new, nn);
2276         add_to_unconfirmed(new);
2277         swap(new, conf);
2278 out_copy:
2279         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2280         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2281
2282         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2283         nfsd4_set_ex_flags(conf, exid);
2284
2285         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2286                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2287         status = nfs_ok;
2288
2289 out:
2290         spin_unlock(&nn->client_lock);
2291         nfs4_unlock_state();
2292         if (new)
2293                 expire_client(new);
2294         if (unconf)
2295                 expire_client(unconf);
2296         return status;
2297 }
2298
2299 static __be32
2300 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2301 {
2302         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2303                 slot_seqid);
2304
2305         /* The slot is in use, and no response has been sent. */
2306         if (slot_inuse) {
2307                 if (seqid == slot_seqid)
2308                         return nfserr_jukebox;
2309                 else
2310                         return nfserr_seq_misordered;
2311         }
2312         /* Note unsigned 32-bit arithmetic handles wraparound: */
2313         if (likely(seqid == slot_seqid + 1))
2314                 return nfs_ok;
2315         if (seqid == slot_seqid)
2316                 return nfserr_replay_cache;
2317         return nfserr_seq_misordered;
2318 }
2319
2320 /*
2321  * Cache the create session result into the create session single DRC
2322  * slot cache by saving the xdr structure. sl_seqid has been set.
2323  * Do this for solo or embedded create session operations.
2324  */
2325 static void
2326 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2327                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2328 {
2329         slot->sl_status = nfserr;
2330         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2331 }
2332
2333 static __be32
2334 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2335                             struct nfsd4_clid_slot *slot)
2336 {
2337         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2338         return slot->sl_status;
2339 }
2340
2341 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2342                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2343                         1 +     /* MIN tag is length with zero, only length */ \
2344                         3 +     /* version, opcount, opcode */ \
2345                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2346                                 /* seqid, slotID, slotID, cache */ \
2347                         4 ) * sizeof(__be32))
2348
2349 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2350                         2 +     /* verifier: AUTH_NULL, length 0 */\
2351                         1 +     /* status */ \
2352                         1 +     /* MIN tag is length with zero, only length */ \
2353                         3 +     /* opcount, opcode, opstatus*/ \
2354                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2355                                 /* seqid, slotID, slotID, slotID, status */ \
2356                         5 ) * sizeof(__be32))
2357
2358 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2359 {
2360         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2361
2362         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2363                 return nfserr_toosmall;
2364         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2365                 return nfserr_toosmall;
2366         ca->headerpadsz = 0;
2367         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2368         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2369         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2370         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2371                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2372         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2373         /*
2374          * Note decreasing slot size below client's request may make it
2375          * difficult for client to function correctly, whereas
2376          * decreasing the number of slots will (just?) affect
2377          * performance.  When short on memory we therefore prefer to
2378          * decrease number of slots instead of their size.  Clients that
2379          * request larger slots than they need will get poor results:
2380          */
2381         ca->maxreqs = nfsd4_get_drc_mem(ca);
2382         if (!ca->maxreqs)
2383                 return nfserr_jukebox;
2384
2385         return nfs_ok;
2386 }
2387
2388 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2389                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2390 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2391                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2392
2393 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2394 {
2395         ca->headerpadsz = 0;
2396
2397         /*
2398          * These RPC_MAX_HEADER macros are overkill, especially since we
2399          * don't even do gss on the backchannel yet.  But this is still
2400          * less than 1k.  Tighten up this estimate in the unlikely event
2401          * it turns out to be a problem for some client:
2402          */
2403         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2404                 return nfserr_toosmall;
2405         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2406                 return nfserr_toosmall;
2407         ca->maxresp_cached = 0;
2408         if (ca->maxops < 2)
2409                 return nfserr_toosmall;
2410
2411         return nfs_ok;
2412 }
2413
2414 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2415 {
2416         switch (cbs->flavor) {
2417         case RPC_AUTH_NULL:
2418         case RPC_AUTH_UNIX:
2419                 return nfs_ok;
2420         default:
2421                 /*
2422                  * GSS case: the spec doesn't allow us to return this
2423                  * error.  But it also doesn't allow us not to support
2424                  * GSS.
2425                  * I'd rather this fail hard than return some error the
2426                  * client might think it can already handle:
2427                  */
2428                 return nfserr_encr_alg_unsupp;
2429         }
2430 }
2431
2432 __be32
2433 nfsd4_create_session(struct svc_rqst *rqstp,
2434                      struct nfsd4_compound_state *cstate,
2435                      struct nfsd4_create_session *cr_ses)
2436 {
2437         struct sockaddr *sa = svc_addr(rqstp);
2438         struct nfs4_client *conf, *unconf;
2439         struct nfs4_client *old = NULL;
2440         struct nfsd4_session *new;
2441         struct nfsd4_conn *conn;
2442         struct nfsd4_clid_slot *cs_slot = NULL;
2443         __be32 status = 0;
2444         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2445
2446         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2447                 return nfserr_inval;
2448         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2449         if (status)
2450                 return status;
2451         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2452         if (status)
2453                 return status;
2454         status = check_backchannel_attrs(&cr_ses->back_channel);
2455         if (status)
2456                 goto out_release_drc_mem;
2457         status = nfserr_jukebox;
2458         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2459         if (!new)
2460                 goto out_release_drc_mem;
2461         conn = alloc_conn_from_crses(rqstp, cr_ses);
2462         if (!conn)
2463                 goto out_free_session;
2464
2465         nfs4_lock_state();
2466         spin_lock(&nn->client_lock);
2467         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2468         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2469         WARN_ON_ONCE(conf && unconf);
2470
2471         if (conf) {
2472                 status = nfserr_wrong_cred;
2473                 if (!mach_creds_match(conf, rqstp))
2474                         goto out_free_conn;
2475                 cs_slot = &conf->cl_cs_slot;
2476                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2477                 if (status == nfserr_replay_cache) {
2478                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
2479                         goto out_free_conn;
2480                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
2481                         status = nfserr_seq_misordered;
2482                         goto out_free_conn;
2483                 }
2484         } else if (unconf) {
2485                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2486                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2487                         status = nfserr_clid_inuse;
2488                         goto out_free_conn;
2489                 }
2490                 status = nfserr_wrong_cred;
2491                 if (!mach_creds_match(unconf, rqstp))
2492                         goto out_free_conn;
2493                 cs_slot = &unconf->cl_cs_slot;
2494                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2495                 if (status) {
2496                         /* an unconfirmed replay returns misordered */
2497                         status = nfserr_seq_misordered;
2498                         goto out_free_conn;
2499                 }
2500                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2501                 if (old) {
2502                         status = mark_client_expired_locked(old);
2503                         if (status) {
2504                                 old = NULL;
2505                                 goto out_free_conn;
2506                         }
2507                 }
2508                 move_to_confirmed(unconf);
2509                 conf = unconf;
2510         } else {
2511                 status = nfserr_stale_clientid;
2512                 goto out_free_conn;
2513         }
2514         status = nfs_ok;
2515         /*
2516          * We do not support RDMA or persistent sessions
2517          */
2518         cr_ses->flags &= ~SESSION4_PERSIST;
2519         cr_ses->flags &= ~SESSION4_RDMA;
2520
2521         init_session(rqstp, new, conf, cr_ses);
2522         nfsd4_get_session_locked(new);
2523
2524         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2525                NFS4_MAX_SESSIONID_LEN);
2526         cs_slot->sl_seqid++;
2527         cr_ses->seqid = cs_slot->sl_seqid;
2528
2529         /* cache solo and embedded create sessions under the client_lock */
2530         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2531         spin_unlock(&nn->client_lock);
2532         /* init connection and backchannel */
2533         nfsd4_init_conn(rqstp, conn, new);
2534         nfsd4_put_session(new);
2535         nfs4_unlock_state();
2536         if (old)
2537                 expire_client(old);
2538         return status;
2539 out_free_conn:
2540         spin_unlock(&nn->client_lock);
2541         nfs4_unlock_state();
2542         free_conn(conn);
2543         if (old)
2544                 expire_client(old);
2545 out_free_session:
2546         __free_session(new);
2547 out_release_drc_mem:
2548         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2549         return status;
2550 }
2551
2552 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2553 {
2554         switch (*dir) {
2555         case NFS4_CDFC4_FORE:
2556         case NFS4_CDFC4_BACK:
2557                 return nfs_ok;
2558         case NFS4_CDFC4_FORE_OR_BOTH:
2559         case NFS4_CDFC4_BACK_OR_BOTH:
2560                 *dir = NFS4_CDFC4_BOTH;
2561                 return nfs_ok;
2562         };
2563         return nfserr_inval;
2564 }
2565
2566 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2567 {
2568         struct nfsd4_session *session = cstate->session;
2569         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2570         __be32 status;
2571
2572         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2573         if (status)
2574                 return status;
2575         spin_lock(&nn->client_lock);
2576         session->se_cb_prog = bc->bc_cb_program;
2577         session->se_cb_sec = bc->bc_cb_sec;
2578         spin_unlock(&nn->client_lock);
2579
2580         nfsd4_probe_callback(session->se_client);
2581
2582         return nfs_ok;
2583 }
2584
2585 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2586                      struct nfsd4_compound_state *cstate,
2587                      struct nfsd4_bind_conn_to_session *bcts)
2588 {
2589         __be32 status;
2590         struct nfsd4_conn *conn;
2591         struct nfsd4_session *session;
2592         struct net *net = SVC_NET(rqstp);
2593         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2594
2595         if (!nfsd4_last_compound_op(rqstp))
2596                 return nfserr_not_only_op;
2597         nfs4_lock_state();
2598         spin_lock(&nn->client_lock);
2599         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2600         spin_unlock(&nn->client_lock);
2601         if (!session)
2602                 goto out_no_session;
2603         status = nfserr_wrong_cred;
2604         if (!mach_creds_match(session->se_client, rqstp))
2605                 goto out;
2606         status = nfsd4_map_bcts_dir(&bcts->dir);
2607         if (status)
2608                 goto out;
2609         conn = alloc_conn(rqstp, bcts->dir);
2610         status = nfserr_jukebox;
2611         if (!conn)
2612                 goto out;
2613         nfsd4_init_conn(rqstp, conn, session);
2614         status = nfs_ok;
2615 out:
2616         nfsd4_put_session(session);
2617 out_no_session:
2618         nfs4_unlock_state();
2619         return status;
2620 }
2621
2622 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2623 {
2624         if (!session)
2625                 return 0;
2626         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2627 }
2628
2629 __be32
2630 nfsd4_destroy_session(struct svc_rqst *r,
2631                       struct nfsd4_compound_state *cstate,
2632                       struct nfsd4_destroy_session *sessionid)
2633 {
2634         struct nfsd4_session *ses;
2635         __be32 status;
2636         int ref_held_by_me = 0;
2637         struct net *net = SVC_NET(r);
2638         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2639
2640         nfs4_lock_state();
2641         status = nfserr_not_only_op;
2642         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2643                 if (!nfsd4_last_compound_op(r))
2644                         goto out;
2645                 ref_held_by_me++;
2646         }
2647         dump_sessionid(__func__, &sessionid->sessionid);
2648         spin_lock(&nn->client_lock);
2649         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2650         if (!ses)
2651                 goto out_client_lock;
2652         status = nfserr_wrong_cred;
2653         if (!mach_creds_match(ses->se_client, r))
2654                 goto out_put_session;
2655         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2656         if (status)
2657                 goto out_put_session;
2658         unhash_session(ses);
2659         spin_unlock(&nn->client_lock);
2660
2661         nfsd4_probe_callback_sync(ses->se_client);
2662
2663         spin_lock(&nn->client_lock);
2664         status = nfs_ok;
2665 out_put_session:
2666         nfsd4_put_session_locked(ses);
2667 out_client_lock:
2668         spin_unlock(&nn->client_lock);
2669 out:
2670         nfs4_unlock_state();
2671         return status;
2672 }
2673
2674 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2675 {
2676         struct nfsd4_conn *c;
2677
2678         list_for_each_entry(c, &s->se_conns, cn_persession) {
2679                 if (c->cn_xprt == xpt) {
2680                         return c;
2681                 }
2682         }
2683         return NULL;
2684 }
2685
2686 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2687 {
2688         struct nfs4_client *clp = ses->se_client;
2689         struct nfsd4_conn *c;
2690         __be32 status = nfs_ok;
2691         int ret;
2692
2693         spin_lock(&clp->cl_lock);
2694         c = __nfsd4_find_conn(new->cn_xprt, ses);
2695         if (c)
2696                 goto out_free;
2697         status = nfserr_conn_not_bound_to_session;
2698         if (clp->cl_mach_cred)
2699                 goto out_free;
2700         __nfsd4_hash_conn(new, ses);
2701         spin_unlock(&clp->cl_lock);
2702         ret = nfsd4_register_conn(new);
2703         if (ret)
2704                 /* oops; xprt is already down: */
2705                 nfsd4_conn_lost(&new->cn_xpt_user);
2706         return nfs_ok;
2707 out_free:
2708         spin_unlock(&clp->cl_lock);
2709         free_conn(new);
2710         return status;
2711 }
2712
2713 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2714 {
2715         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2716
2717         return args->opcnt > session->se_fchannel.maxops;
2718 }
2719
2720 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2721                                   struct nfsd4_session *session)
2722 {
2723         struct xdr_buf *xb = &rqstp->rq_arg;
2724
2725         return xb->len > session->se_fchannel.maxreq_sz;
2726 }
2727
2728 __be32
2729 nfsd4_sequence(struct svc_rqst *rqstp,
2730                struct nfsd4_compound_state *cstate,
2731                struct nfsd4_sequence *seq)
2732 {
2733         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2734         struct xdr_stream *xdr = &resp->xdr;
2735         struct nfsd4_session *session;
2736         struct nfs4_client *clp;
2737         struct nfsd4_slot *slot;
2738         struct nfsd4_conn *conn;
2739         __be32 status;
2740         int buflen;
2741         struct net *net = SVC_NET(rqstp);
2742         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2743
2744         if (resp->opcnt != 1)
2745                 return nfserr_sequence_pos;
2746
2747         /*
2748          * Will be either used or freed by nfsd4_sequence_check_conn
2749          * below.
2750          */
2751         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2752         if (!conn)
2753                 return nfserr_jukebox;
2754
2755         spin_lock(&nn->client_lock);
2756         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2757         if (!session)
2758                 goto out_no_session;
2759         clp = session->se_client;
2760
2761         status = nfserr_too_many_ops;
2762         if (nfsd4_session_too_many_ops(rqstp, session))
2763                 goto out_put_session;
2764
2765         status = nfserr_req_too_big;
2766         if (nfsd4_request_too_big(rqstp, session))
2767                 goto out_put_session;
2768
2769         status = nfserr_badslot;
2770         if (seq->slotid >= session->se_fchannel.maxreqs)
2771                 goto out_put_session;
2772
2773         slot = session->se_slots[seq->slotid];
2774         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2775
2776         /* We do not negotiate the number of slots yet, so set the
2777          * maxslots to the session maxreqs which is used to encode
2778          * sr_highest_slotid and the sr_target_slot id to maxslots */
2779         seq->maxslots = session->se_fchannel.maxreqs;
2780
2781         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2782                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2783         if (status == nfserr_replay_cache) {
2784                 status = nfserr_seq_misordered;
2785                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2786                         goto out_put_session;
2787                 cstate->slot = slot;
2788                 cstate->session = session;
2789                 cstate->clp = clp;
2790                 /* Return the cached reply status and set cstate->status
2791                  * for nfsd4_proc_compound processing */
2792                 status = nfsd4_replay_cache_entry(resp, seq);
2793                 cstate->status = nfserr_replay_cache;
2794                 goto out;
2795         }
2796         if (status)
2797                 goto out_put_session;
2798
2799         status = nfsd4_sequence_check_conn(conn, session);
2800         conn = NULL;
2801         if (status)
2802                 goto out_put_session;
2803
2804         buflen = (seq->cachethis) ?
2805                         session->se_fchannel.maxresp_cached :
2806                         session->se_fchannel.maxresp_sz;
2807         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
2808                                     nfserr_rep_too_big;
2809         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
2810                 goto out_put_session;
2811         svc_reserve(rqstp, buflen);
2812
2813         status = nfs_ok;
2814         /* Success! bump slot seqid */
2815         slot->sl_seqid = seq->seqid;
2816         slot->sl_flags |= NFSD4_SLOT_INUSE;
2817         if (seq->cachethis)
2818                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2819         else
2820                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2821
2822         cstate->slot = slot;
2823         cstate->session = session;
2824         cstate->clp = clp;
2825
2826 out:
2827         switch (clp->cl_cb_state) {
2828         case NFSD4_CB_DOWN:
2829                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2830                 break;
2831         case NFSD4_CB_FAULT:
2832                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2833                 break;
2834         default:
2835                 seq->status_flags = 0;
2836         }
2837         if (!list_empty(&clp->cl_revoked))
2838                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2839 out_no_session:
2840         if (conn)
2841                 free_conn(conn);
2842         spin_unlock(&nn->client_lock);
2843         return status;
2844 out_put_session:
2845         nfsd4_put_session_locked(session);
2846         goto out_no_session;
2847 }
2848
2849 void
2850 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
2851 {
2852         struct nfsd4_compound_state *cs = &resp->cstate;
2853
2854         if (nfsd4_has_session(cs)) {
2855                 if (cs->status != nfserr_replay_cache) {
2856                         nfsd4_store_cache_entry(resp);
2857                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
2858                 }
2859                 /* Drop session reference that was taken in nfsd4_sequence() */
2860                 nfsd4_put_session(cs->session);
2861         } else if (cs->clp)
2862                 put_client_renew(cs->clp);
2863 }
2864
2865 __be32
2866 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2867 {
2868         struct nfs4_client *conf, *unconf;
2869         struct nfs4_client *clp = NULL;
2870         __be32 status = 0;
2871         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2872
2873         nfs4_lock_state();
2874         spin_lock(&nn->client_lock);
2875         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2876         conf = find_confirmed_client(&dc->clientid, true, nn);
2877         WARN_ON_ONCE(conf && unconf);
2878
2879         if (conf) {
2880                 if (client_has_state(conf)) {
2881                         status = nfserr_clientid_busy;
2882                         goto out;
2883                 }
2884                 status = mark_client_expired_locked(conf);
2885                 if (status)
2886                         goto out;
2887                 clp = conf;
2888         } else if (unconf)
2889                 clp = unconf;
2890         else {
2891                 status = nfserr_stale_clientid;
2892                 goto out;
2893         }
2894         if (!mach_creds_match(clp, rqstp)) {
2895                 clp = NULL;
2896                 status = nfserr_wrong_cred;
2897                 goto out;
2898         }
2899         unhash_client_locked(clp);
2900 out:
2901         spin_unlock(&nn->client_lock);
2902         nfs4_unlock_state();
2903         if (clp)
2904                 expire_client(clp);
2905         return status;
2906 }
2907
2908 __be32
2909 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2910 {
2911         __be32 status = 0;
2912
2913         if (rc->rca_one_fs) {
2914                 if (!cstate->current_fh.fh_dentry)
2915                         return nfserr_nofilehandle;
2916                 /*
2917                  * We don't take advantage of the rca_one_fs case.
2918                  * That's OK, it's optional, we can safely ignore it.
2919                  */
2920                  return nfs_ok;
2921         }
2922
2923         nfs4_lock_state();
2924         status = nfserr_complete_already;
2925         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2926                              &cstate->session->se_client->cl_flags))
2927                 goto out;
2928
2929         status = nfserr_stale_clientid;
2930         if (is_client_expired(cstate->session->se_client))
2931                 /*
2932                  * The following error isn't really legal.
2933                  * But we only get here if the client just explicitly
2934                  * destroyed the client.  Surely it no longer cares what
2935                  * error it gets back on an operation for the dead
2936                  * client.
2937                  */
2938                 goto out;
2939
2940         status = nfs_ok;
2941         nfsd4_client_record_create(cstate->session->se_client);
2942 out:
2943         nfs4_unlock_state();
2944         return status;
2945 }
2946
2947 __be32
2948 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2949                   struct nfsd4_setclientid *setclid)
2950 {
2951         struct xdr_netobj       clname = setclid->se_name;
2952         nfs4_verifier           clverifier = setclid->se_verf;
2953         struct nfs4_client      *conf, *new;
2954         struct nfs4_client      *unconf = NULL;
2955         __be32                  status;
2956         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2957
2958         new = create_client(clname, rqstp, &clverifier);
2959         if (new == NULL)
2960                 return nfserr_jukebox;
2961         /* Cases below refer to rfc 3530 section 14.2.33: */
2962         nfs4_lock_state();
2963         spin_lock(&nn->client_lock);
2964         conf = find_confirmed_client_by_name(&clname, nn);
2965         if (conf) {
2966                 /* case 0: */
2967                 status = nfserr_clid_inuse;
2968                 if (clp_used_exchangeid(conf))
2969                         goto out;
2970                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2971                         char addr_str[INET6_ADDRSTRLEN];
2972                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2973                                  sizeof(addr_str));
2974                         dprintk("NFSD: setclientid: string in use by client "
2975                                 "at %s\n", addr_str);
2976                         goto out;
2977                 }
2978         }
2979         unconf = find_unconfirmed_client_by_name(&clname, nn);
2980         if (unconf)
2981                 unhash_client_locked(unconf);
2982         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2983                 /* case 1: probable callback update */
2984                 copy_clid(new, conf);
2985         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2986                 gen_clid(new, nn);
2987         new->cl_minorversion = 0;
2988         gen_callback(new, setclid, rqstp);
2989         add_to_unconfirmed(new);
2990         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2991         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2992         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2993         new = NULL;
2994         status = nfs_ok;
2995 out:
2996         spin_unlock(&nn->client_lock);
2997         nfs4_unlock_state();
2998         if (new)
2999                 free_client(new);
3000         if (unconf)
3001                 expire_client(unconf);
3002         return status;
3003 }
3004
3005
3006 __be32
3007 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3008                          struct nfsd4_compound_state *cstate,
3009                          struct nfsd4_setclientid_confirm *setclientid_confirm)
3010 {
3011         struct nfs4_client *conf, *unconf;
3012         struct nfs4_client *old = NULL;
3013         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3014         clientid_t * clid = &setclientid_confirm->sc_clientid;
3015         __be32 status;
3016         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3017
3018         if (STALE_CLIENTID(clid, nn))
3019                 return nfserr_stale_clientid;
3020         nfs4_lock_state();
3021
3022         spin_lock(&nn->client_lock);
3023         conf = find_confirmed_client(clid, false, nn);
3024         unconf = find_unconfirmed_client(clid, false, nn);
3025         /*
3026          * We try hard to give out unique clientid's, so if we get an
3027          * attempt to confirm the same clientid with a different cred,
3028          * there's a bug somewhere.  Let's charitably assume it's our
3029          * bug.
3030          */
3031         status = nfserr_serverfault;
3032         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3033                 goto out;
3034         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3035                 goto out;
3036         /* cases below refer to rfc 3530 section 14.2.34: */
3037         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3038                 if (conf && !unconf) /* case 2: probable retransmit */
3039                         status = nfs_ok;
3040                 else /* case 4: client hasn't noticed we rebooted yet? */
3041                         status = nfserr_stale_clientid;
3042                 goto out;
3043         }
3044         status = nfs_ok;
3045         if (conf) { /* case 1: callback update */
3046                 old = unconf;
3047                 unhash_client_locked(old);
3048                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3049         } else { /* case 3: normal case; new or rebooted client */
3050                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3051                 if (old) {
3052                         status = mark_client_expired_locked(old);
3053                         if (status) {
3054                                 old = NULL;
3055                                 goto out;
3056                         }
3057                 }
3058                 move_to_confirmed(unconf);
3059                 conf = unconf;
3060         }
3061         get_client_locked(conf);
3062         spin_unlock(&nn->client_lock);
3063         nfsd4_probe_callback(conf);
3064         spin_lock(&nn->client_lock);
3065         put_client_renew_locked(conf);
3066 out:
3067         spin_unlock(&nn->client_lock);
3068         if (old)
3069                 expire_client(old);
3070         nfs4_unlock_state();
3071         return status;
3072 }
3073
3074 static struct nfs4_file *nfsd4_alloc_file(void)
3075 {
3076         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3077 }
3078
3079 /* OPEN Share state helper functions */
3080 static void nfsd4_init_file(struct nfs4_file *fp, struct knfsd_fh *fh)
3081 {
3082         unsigned int hashval = file_hashval(fh);
3083
3084         lockdep_assert_held(&state_lock);
3085
3086         atomic_set(&fp->fi_ref, 1);
3087         spin_lock_init(&fp->fi_lock);
3088         INIT_LIST_HEAD(&fp->fi_stateids);
3089         INIT_LIST_HEAD(&fp->fi_delegations);
3090         fh_copy_shallow(&fp->fi_fhandle, fh);
3091         fp->fi_had_conflict = false;
3092         fp->fi_lease = NULL;
3093         fp->fi_share_deny = 0;
3094         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3095         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3096         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
3097 }
3098
3099 void
3100 nfsd4_free_slabs(void)
3101 {
3102         kmem_cache_destroy(openowner_slab);
3103         kmem_cache_destroy(lockowner_slab);
3104         kmem_cache_destroy(file_slab);
3105         kmem_cache_destroy(stateid_slab);
3106         kmem_cache_destroy(deleg_slab);
3107 }
3108
3109 int
3110 nfsd4_init_slabs(void)
3111 {
3112         openowner_slab = kmem_cache_create("nfsd4_openowners",
3113                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3114         if (openowner_slab == NULL)
3115                 goto out;
3116         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3117                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3118         if (lockowner_slab == NULL)
3119                 goto out_free_openowner_slab;
3120         file_slab = kmem_cache_create("nfsd4_files",
3121                         sizeof(struct nfs4_file), 0, 0, NULL);
3122         if (file_slab == NULL)
3123                 goto out_free_lockowner_slab;
3124         stateid_slab = kmem_cache_create("nfsd4_stateids",
3125                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3126         if (stateid_slab == NULL)
3127                 goto out_free_file_slab;
3128         deleg_slab = kmem_cache_create("nfsd4_delegations",
3129                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3130         if (deleg_slab == NULL)
3131                 goto out_free_stateid_slab;
3132         return 0;
3133
3134 out_free_stateid_slab:
3135         kmem_cache_destroy(stateid_slab);
3136 out_free_file_slab:
3137         kmem_cache_destroy(file_slab);
3138 out_free_lockowner_slab:
3139         kmem_cache_destroy(lockowner_slab);
3140 out_free_openowner_slab:
3141         kmem_cache_destroy(openowner_slab);
3142 out:
3143         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3144         return -ENOMEM;
3145 }
3146
3147 static void init_nfs4_replay(struct nfs4_replay *rp)
3148 {
3149         rp->rp_status = nfserr_serverfault;
3150         rp->rp_buflen = 0;
3151         rp->rp_buf = rp->rp_ibuf;
3152         mutex_init(&rp->rp_mutex);
3153 }
3154
3155 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3156                 struct nfs4_stateowner *so)
3157 {
3158         if (!nfsd4_has_session(cstate)) {
3159                 mutex_lock(&so->so_replay.rp_mutex);
3160                 cstate->replay_owner = so;
3161                 atomic_inc(&so->so_count);
3162         }
3163 }
3164
3165 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3166 {
3167         struct nfs4_stateowner *so = cstate->replay_owner;
3168
3169         if (so != NULL) {
3170                 cstate->replay_owner = NULL;
3171                 mutex_unlock(&so->so_replay.rp_mutex);
3172                 nfs4_put_stateowner(so);
3173         }
3174 }
3175
3176 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3177 {
3178         struct nfs4_stateowner *sop;
3179
3180         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3181         if (!sop)
3182                 return NULL;
3183
3184         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3185         if (!sop->so_owner.data) {
3186                 kmem_cache_free(slab, sop);
3187                 return NULL;
3188         }
3189         sop->so_owner.len = owner->len;
3190
3191         INIT_LIST_HEAD(&sop->so_stateids);
3192         sop->so_client = clp;
3193         init_nfs4_replay(&sop->so_replay);
3194         atomic_set(&sop->so_count, 1);
3195         return sop;
3196 }
3197
3198 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3199 {
3200         lockdep_assert_held(&clp->cl_lock);
3201
3202         list_add(&oo->oo_owner.so_strhash,
3203                  &clp->cl_ownerstr_hashtbl[strhashval]);
3204         list_add(&oo->oo_perclient, &clp->cl_openowners);
3205 }
3206
3207 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3208 {
3209         unhash_openowner_locked(openowner(so));
3210 }
3211
3212 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3213 {
3214         struct nfs4_openowner *oo = openowner(so);
3215
3216         kmem_cache_free(openowner_slab, oo);
3217 }
3218
3219 static const struct nfs4_stateowner_operations openowner_ops = {
3220         .so_unhash =    nfs4_unhash_openowner,
3221         .so_free =      nfs4_free_openowner,
3222 };
3223
3224 static struct nfs4_openowner *
3225 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3226                            struct nfsd4_compound_state *cstate)
3227 {
3228         struct nfs4_client *clp = cstate->clp;
3229         struct nfs4_openowner *oo, *ret;
3230
3231         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3232         if (!oo)
3233                 return NULL;
3234         oo->oo_owner.so_ops = &openowner_ops;
3235         oo->oo_owner.so_is_open_owner = 1;
3236         oo->oo_owner.so_seqid = open->op_seqid;
3237         oo->oo_flags = 0;
3238         if (nfsd4_has_session(cstate))
3239                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3240         oo->oo_time = 0;
3241         oo->oo_last_closed_stid = NULL;
3242         INIT_LIST_HEAD(&oo->oo_close_lru);
3243         spin_lock(&clp->cl_lock);
3244         ret = find_openstateowner_str_locked(strhashval, open, clp);
3245         if (ret == NULL) {
3246                 hash_openowner(oo, clp, strhashval);
3247                 ret = oo;
3248         } else
3249                 nfs4_free_openowner(&oo->oo_owner);
3250         spin_unlock(&clp->cl_lock);
3251         return oo;
3252 }
3253
3254 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
3255         struct nfs4_openowner *oo = open->op_openowner;
3256
3257         atomic_inc(&stp->st_stid.sc_count);
3258         stp->st_stid.sc_type = NFS4_OPEN_STID;
3259         INIT_LIST_HEAD(&stp->st_locks);
3260         stp->st_stateowner = &oo->oo_owner;
3261         atomic_inc(&stp->st_stateowner->so_count);
3262         get_nfs4_file(fp);
3263         stp->st_stid.sc_file = fp;
3264         stp->st_access_bmap = 0;
3265         stp->st_deny_bmap = 0;
3266         stp->st_openstp = NULL;
3267         spin_lock(&oo->oo_owner.so_client->cl_lock);
3268         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3269         spin_lock(&fp->fi_lock);
3270         list_add(&stp->st_perfile, &fp->fi_stateids);
3271         spin_unlock(&fp->fi_lock);
3272         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3273 }
3274
3275 /*
3276  * In the 4.0 case we need to keep the owners around a little while to handle
3277  * CLOSE replay. We still do need to release any file access that is held by
3278  * them before returning however.
3279  */
3280 static void
3281 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3282 {
3283         struct nfs4_ol_stateid *last;
3284         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3285         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3286                                                 nfsd_net_id);
3287
3288         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3289
3290         /*
3291          * We know that we hold one reference via nfsd4_close, and another
3292          * "persistent" reference for the client. If the refcount is higher
3293          * than 2, then there are still calls in progress that are using this
3294          * stateid. We can't put the sc_file reference until they are finished.
3295          * Wait for the refcount to drop to 2. Since it has been unhashed,
3296          * there should be no danger of the refcount going back up again at
3297          * this point.
3298          */
3299         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3300
3301         release_all_access(s);
3302         if (s->st_stid.sc_file) {
3303                 put_nfs4_file(s->st_stid.sc_file);
3304                 s->st_stid.sc_file = NULL;
3305         }
3306
3307         spin_lock(&nn->client_lock);
3308         last = oo->oo_last_closed_stid;
3309         oo->oo_last_closed_stid = s;
3310         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3311         oo->oo_time = get_seconds();
3312         spin_unlock(&nn->client_lock);
3313         if (last)
3314                 nfs4_put_stid(&last->st_stid);
3315 }
3316
3317 /* search file_hashtbl[] for file */
3318 static struct nfs4_file *
3319 find_file_locked(struct knfsd_fh *fh)
3320 {
3321         unsigned int hashval = file_hashval(fh);
3322         struct nfs4_file *fp;
3323
3324         lockdep_assert_held(&state_lock);
3325
3326         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
3327                 if (nfsd_fh_match(&fp->fi_fhandle, fh)) {
3328                         get_nfs4_file(fp);
3329                         return fp;
3330                 }
3331         }
3332         return NULL;
3333 }
3334
3335 static struct nfs4_file *
3336 find_file(struct knfsd_fh *fh)
3337 {
3338         struct nfs4_file *fp;
3339
3340         spin_lock(&state_lock);
3341         fp = find_file_locked(fh);
3342         spin_unlock(&state_lock);
3343         return fp;
3344 }
3345
3346 static struct nfs4_file *
3347 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3348 {
3349         struct nfs4_file *fp;
3350
3351         spin_lock(&state_lock);
3352         fp = find_file_locked(fh);
3353         if (fp == NULL) {
3354                 nfsd4_init_file(new, fh);
3355                 fp = new;
3356         }
3357         spin_unlock(&state_lock);
3358
3359         return fp;
3360 }
3361
3362 /*
3363  * Called to check deny when READ with all zero stateid or
3364  * WRITE with all zero or all one stateid
3365  */
3366 static __be32
3367 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3368 {
3369         struct nfs4_file *fp;
3370         __be32 ret = nfs_ok;
3371
3372         fp = find_file(&current_fh->fh_handle);
3373         if (!fp)
3374                 return ret;
3375         /* Check for conflicting share reservations */
3376         spin_lock(&fp->fi_lock);
3377         if (fp->fi_share_deny & deny_type)
3378                 ret = nfserr_locked;
3379         spin_unlock(&fp->fi_lock);
3380         put_nfs4_file(fp);
3381         return ret;
3382 }
3383
3384 void nfsd4_prepare_cb_recall(struct nfs4_delegation *dp)
3385 {
3386         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3387                                           nfsd_net_id);
3388
3389         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3390
3391         /*
3392          * We can't do this in nfsd_break_deleg_cb because it is
3393          * already holding inode->i_lock.
3394          *
3395          * If the dl_time != 0, then we know that it has already been
3396          * queued for a lease break. Don't queue it again.
3397          */
3398         spin_lock(&state_lock);
3399         if (dp->dl_time == 0) {
3400                 dp->dl_time = get_seconds();
3401                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3402         }
3403         spin_unlock(&state_lock);
3404 }
3405
3406 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3407 {
3408         /*
3409          * We're assuming the state code never drops its reference
3410          * without first removing the lease.  Since we're in this lease
3411          * callback (and since the lease code is serialized by the kernel
3412          * lock) we know the server hasn't removed the lease yet, we know
3413          * it's safe to take a reference.
3414          */
3415         atomic_inc(&dp->dl_stid.sc_count);
3416         nfsd4_cb_recall(dp);
3417 }
3418
3419 /* Called from break_lease() with i_lock held. */
3420 static void nfsd_break_deleg_cb(struct file_lock *fl)
3421 {
3422         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3423         struct nfs4_delegation *dp;
3424
3425         if (!fp) {
3426                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3427                 return;
3428         }
3429         if (fp->fi_had_conflict) {
3430                 WARN(1, "duplicate break on %p\n", fp);
3431                 return;
3432         }
3433         /*
3434          * We don't want the locks code to timeout the lease for us;
3435          * we'll remove it ourself if a delegation isn't returned
3436          * in time:
3437          */
3438         fl->fl_break_time = 0;
3439
3440         spin_lock(&fp->fi_lock);
3441         fp->fi_had_conflict = true;
3442         /*
3443          * If there are no delegations on the list, then we can't count on this
3444          * lease ever being cleaned up. Set the fl_break_time to jiffies so that
3445          * time_out_leases will do it ASAP. The fact that fi_had_conflict is now
3446          * true should keep any new delegations from being hashed.
3447          */
3448         if (list_empty(&fp->fi_delegations))
3449                 fl->fl_break_time = jiffies;
3450         else
3451                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3452                         nfsd_break_one_deleg(dp);
3453         spin_unlock(&fp->fi_lock);
3454 }
3455
3456 static
3457 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
3458 {
3459         if (arg & F_UNLCK)
3460                 return lease_modify(onlist, arg);
3461         else
3462                 return -EAGAIN;
3463 }
3464
3465 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3466         .lm_break = nfsd_break_deleg_cb,
3467         .lm_change = nfsd_change_deleg_cb,
3468 };
3469
3470 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3471 {
3472         if (nfsd4_has_session(cstate))
3473                 return nfs_ok;
3474         if (seqid == so->so_seqid - 1)
3475                 return nfserr_replay_me;
3476         if (seqid == so->so_seqid)
3477                 return nfs_ok;
3478         return nfserr_bad_seqid;
3479 }
3480
3481 static __be32 lookup_clientid(clientid_t *clid,
3482                 struct nfsd4_compound_state *cstate,
3483                 struct nfsd_net *nn)
3484 {
3485         struct nfs4_client *found;
3486
3487         if (cstate->clp) {
3488                 found = cstate->clp;
3489                 if (!same_clid(&found->cl_clientid, clid))
3490                         return nfserr_stale_clientid;
3491                 return nfs_ok;
3492         }
3493
3494         if (STALE_CLIENTID(clid, nn))
3495                 return nfserr_stale_clientid;
3496
3497         /*
3498          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3499          * cached already then we know this is for is for v4.0 and "sessions"
3500          * will be false.
3501          */
3502         WARN_ON_ONCE(cstate->session);
3503         spin_lock(&nn->client_lock);
3504         found = find_confirmed_client(clid, false, nn);
3505         if (!found) {
3506                 spin_unlock(&nn->client_lock);
3507                 return nfserr_expired;
3508         }
3509         atomic_inc(&found->cl_refcount);
3510         spin_unlock(&nn->client_lock);
3511
3512         /* Cache the nfs4_client in cstate! */
3513         cstate->clp = found;
3514         return nfs_ok;
3515 }
3516
3517 __be32
3518 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3519                     struct nfsd4_open *open, struct nfsd_net *nn)
3520 {
3521         clientid_t *clientid = &open->op_clientid;
3522         struct nfs4_client *clp = NULL;
3523         unsigned int strhashval;
3524         struct nfs4_openowner *oo = NULL;
3525         __be32 status;
3526
3527         if (STALE_CLIENTID(&open->op_clientid, nn))
3528                 return nfserr_stale_clientid;
3529         /*
3530          * In case we need it later, after we've already created the
3531          * file and don't want to risk a further failure:
3532          */
3533         open->op_file = nfsd4_alloc_file();
3534         if (open->op_file == NULL)
3535                 return nfserr_jukebox;
3536
3537         status = lookup_clientid(clientid, cstate, nn);
3538         if (status)
3539                 return status;
3540         clp = cstate->clp;
3541
3542         strhashval = ownerstr_hashval(&open->op_owner);
3543         oo = find_openstateowner_str(strhashval, open, clp);
3544         open->op_openowner = oo;
3545         if (!oo) {
3546                 goto new_owner;
3547         }
3548         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3549                 /* Replace unconfirmed owners without checking for replay. */
3550                 release_openowner(oo);
3551                 open->op_openowner = NULL;
3552                 goto new_owner;
3553         }
3554         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3555         if (status)
3556                 return status;
3557         goto alloc_stateid;
3558 new_owner:
3559         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3560         if (oo == NULL)
3561                 return nfserr_jukebox;
3562         open->op_openowner = oo;
3563 alloc_stateid:
3564         open->op_stp = nfs4_alloc_open_stateid(clp);
3565         if (!open->op_stp)
3566                 return nfserr_jukebox;
3567         return nfs_ok;
3568 }
3569
3570 static inline __be32
3571 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3572 {
3573         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3574                 return nfserr_openmode;
3575         else
3576                 return nfs_ok;
3577 }
3578
3579 static int share_access_to_flags(u32 share_access)
3580 {
3581         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3582 }
3583
3584 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3585 {
3586         struct nfs4_stid *ret;
3587
3588         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3589         if (!ret)
3590                 return NULL;
3591         return delegstateid(ret);
3592 }
3593
3594 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3595 {
3596         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3597                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3598 }
3599
3600 static __be32
3601 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3602                 struct nfs4_delegation **dp)
3603 {
3604         int flags;
3605         __be32 status = nfserr_bad_stateid;
3606         struct nfs4_delegation *deleg;
3607
3608         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3609         if (deleg == NULL)
3610                 goto out;
3611         flags = share_access_to_flags(open->op_share_access);
3612         status = nfs4_check_delegmode(deleg, flags);
3613         if (status) {
3614                 nfs4_put_stid(&deleg->dl_stid);
3615                 goto out;
3616         }
3617         *dp = deleg;
3618 out:
3619         if (!nfsd4_is_deleg_cur(open))
3620                 return nfs_ok;
3621         if (status)
3622                 return status;
3623         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3624         return nfs_ok;
3625 }
3626
3627 static struct nfs4_ol_stateid *
3628 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3629 {
3630         struct nfs4_ol_stateid *local, *ret = NULL;
3631         struct nfs4_openowner *oo = open->op_openowner;
3632
3633         spin_lock(&fp->fi_lock);
3634         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3635                 /* ignore lock owners */
3636                 if (local->st_stateowner->so_is_open_owner == 0)
3637                         continue;
3638                 if (local->st_stateowner == &oo->oo_owner) {
3639                         ret = local;
3640                         atomic_inc(&ret->st_stid.sc_count);
3641                         break;
3642                 }
3643         }
3644         spin_unlock(&fp->fi_lock);
3645         return ret;
3646 }
3647
3648 static inline int nfs4_access_to_access(u32 nfs4_access)
3649 {
3650         int flags = 0;
3651
3652         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3653                 flags |= NFSD_MAY_READ;
3654         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3655                 flags |= NFSD_MAY_WRITE;
3656         return flags;
3657 }
3658
3659 static inline __be32
3660 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3661                 struct nfsd4_open *open)
3662 {
3663         struct iattr iattr = {
3664                 .ia_valid = ATTR_SIZE,
3665                 .ia_size = 0,
3666         };
3667         if (!open->op_truncate)
3668                 return 0;
3669         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3670                 return nfserr_inval;
3671         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3672 }
3673
3674 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3675                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3676                 struct nfsd4_open *open)
3677 {
3678         struct file *filp = NULL;
3679         __be32 status;
3680         int oflag = nfs4_access_to_omode(open->op_share_access);
3681         int access = nfs4_access_to_access(open->op_share_access);
3682         unsigned char old_access_bmap, old_deny_bmap;
3683
3684         spin_lock(&fp->fi_lock);
3685
3686         /*
3687          * Are we trying to set a deny mode that would conflict with
3688          * current access?
3689          */
3690         status = nfs4_file_check_deny(fp, open->op_share_deny);
3691         if (status != nfs_ok) {
3692                 spin_unlock(&fp->fi_lock);
3693                 goto out;
3694         }
3695
3696         /* set access to the file */
3697         status = nfs4_file_get_access(fp, open->op_share_access);
3698         if (status != nfs_ok) {
3699                 spin_unlock(&fp->fi_lock);
3700                 goto out;
3701         }
3702
3703         /* Set access bits in stateid */
3704         old_access_bmap = stp->st_access_bmap;
3705         set_access(open->op_share_access, stp);
3706
3707         /* Set new deny mask */
3708         old_deny_bmap = stp->st_deny_bmap;
3709         set_deny(open->op_share_deny, stp);
3710         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3711
3712         if (!fp->fi_fds[oflag]) {
3713                 spin_unlock(&fp->fi_lock);
3714                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3715                 if (status)
3716                         goto out_put_access;
3717                 spin_lock(&fp->fi_lock);
3718                 if (!fp->fi_fds[oflag]) {
3719                         fp->fi_fds[oflag] = filp;
3720                         filp = NULL;
3721                 }
3722         }
3723         spin_unlock(&fp->fi_lock);
3724         if (filp)
3725                 fput(filp);
3726
3727         status = nfsd4_truncate(rqstp, cur_fh, open);
3728         if (status)
3729                 goto out_put_access;
3730 out:
3731         return status;
3732 out_put_access:
3733         stp->st_access_bmap = old_access_bmap;
3734         nfs4_file_put_access(fp, open->op_share_access);
3735         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3736         goto out;
3737 }
3738
3739 static __be32
3740 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)
3741 {
3742         __be32 status;
3743         unsigned char old_deny_bmap;
3744
3745         if (!test_access(open->op_share_access, stp))
3746                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3747
3748         /* test and set deny mode */
3749         spin_lock(&fp->fi_lock);
3750         status = nfs4_file_check_deny(fp, open->op_share_deny);
3751         if (status == nfs_ok) {
3752                 old_deny_bmap = stp->st_deny_bmap;
3753                 set_deny(open->op_share_deny, stp);
3754                 fp->fi_share_deny |=
3755                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3756         }
3757         spin_unlock(&fp->fi_lock);
3758
3759         if (status != nfs_ok)
3760                 return status;
3761
3762         status = nfsd4_truncate(rqstp, cur_fh, open);
3763         if (status != nfs_ok)
3764                 reset_union_bmap_deny(old_deny_bmap, stp);
3765         return status;
3766 }
3767
3768 static void
3769 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3770 {
3771         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3772 }
3773
3774 /* Should we give out recallable state?: */
3775 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3776 {
3777         if (clp->cl_cb_state == NFSD4_CB_UP)
3778                 return true;
3779         /*
3780          * In the sessions case, since we don't have to establish a
3781          * separate connection for callbacks, we assume it's OK
3782          * until we hear otherwise:
3783          */
3784         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3785 }
3786
3787 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
3788 {
3789         struct file_lock *fl;
3790
3791         fl = locks_alloc_lock();
3792         if (!fl)
3793                 return NULL;
3794         locks_init_lock(fl);
3795         fl->fl_lmops = &nfsd_lease_mng_ops;
3796         fl->fl_flags = FL_DELEG;
3797         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3798         fl->fl_end = OFFSET_MAX;
3799         fl->fl_owner = (fl_owner_t)fp;
3800         fl->fl_pid = current->tgid;
3801         return fl;
3802 }
3803
3804 static int nfs4_setlease(struct nfs4_delegation *dp)
3805 {
3806         struct nfs4_file *fp = dp->dl_stid.sc_file;
3807         struct file_lock *fl;
3808         struct file *filp;
3809         int status = 0;
3810
3811         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
3812         if (!fl)
3813                 return -ENOMEM;
3814         filp = find_readable_file(fp);
3815         if (!filp) {
3816                 /* We should always have a readable file here */
3817                 WARN_ON_ONCE(1);
3818                 return -EBADF;
3819         }
3820         fl->fl_file = filp;
3821         status = vfs_setlease(filp, fl->fl_type, &fl);
3822         if (status) {
3823                 locks_free_lock(fl);
3824                 goto out_fput;
3825         }
3826         spin_lock(&state_lock);
3827         spin_lock(&fp->fi_lock);
3828         /* Did the lease get broken before we took the lock? */
3829         status = -EAGAIN;
3830         if (fp->fi_had_conflict)
3831                 goto out_unlock;
3832         /* Race breaker */
3833         if (fp->fi_lease) {
3834                 status = 0;
3835                 atomic_inc(&fp->fi_delegees);
3836                 hash_delegation_locked(dp, fp);
3837                 goto out_unlock;
3838         }
3839         fp->fi_lease = fl;
3840         fp->fi_deleg_file = filp;
3841         atomic_set(&fp->fi_delegees, 1);
3842         hash_delegation_locked(dp, fp);
3843         spin_unlock(&fp->fi_lock);
3844         spin_unlock(&state_lock);
3845         return 0;
3846 out_unlock:
3847         spin_unlock(&fp->fi_lock);
3848         spin_unlock(&state_lock);
3849 out_fput:
3850         fput(filp);
3851         return status;
3852 }
3853
3854 static struct nfs4_delegation *
3855 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
3856                     struct nfs4_file *fp)
3857 {
3858         int status;
3859         struct nfs4_delegation *dp;
3860
3861         if (fp->fi_had_conflict)
3862                 return ERR_PTR(-EAGAIN);
3863
3864         dp = alloc_init_deleg(clp, fh);
3865         if (!dp)
3866                 return ERR_PTR(-ENOMEM);
3867
3868         get_nfs4_file(fp);
3869         spin_lock(&state_lock);
3870         spin_lock(&fp->fi_lock);
3871         dp->dl_stid.sc_file = fp;
3872         if (!fp->fi_lease) {
3873                 spin_unlock(&fp->fi_lock);
3874                 spin_unlock(&state_lock);
3875                 status = nfs4_setlease(dp);
3876                 goto out;
3877         }
3878         atomic_inc(&fp->fi_delegees);
3879         if (fp->fi_had_conflict) {
3880                 status = -EAGAIN;
3881                 goto out_unlock;
3882         }
3883         hash_delegation_locked(dp, fp);
3884         status = 0;
3885 out_unlock:
3886         spin_unlock(&fp->fi_lock);
3887         spin_unlock(&state_lock);
3888 out:
3889         if (status) {
3890                 nfs4_put_stid(&dp->dl_stid);
3891                 return ERR_PTR(status);
3892         }
3893         return dp;
3894 }
3895
3896 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3897 {
3898         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3899         if (status == -EAGAIN)
3900                 open->op_why_no_deleg = WND4_CONTENTION;
3901         else {
3902                 open->op_why_no_deleg = WND4_RESOURCE;
3903                 switch (open->op_deleg_want) {
3904                 case NFS4_SHARE_WANT_READ_DELEG:
3905                 case NFS4_SHARE_WANT_WRITE_DELEG:
3906                 case NFS4_SHARE_WANT_ANY_DELEG:
3907                         break;
3908                 case NFS4_SHARE_WANT_CANCEL:
3909                         open->op_why_no_deleg = WND4_CANCELLED;
3910                         break;
3911                 case NFS4_SHARE_WANT_NO_DELEG:
3912                         WARN_ON_ONCE(1);
3913                 }
3914         }
3915 }
3916
3917 /*
3918  * Attempt to hand out a delegation.
3919  *
3920  * Note we don't support write delegations, and won't until the vfs has
3921  * proper support for them.
3922  */
3923 static void
3924 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
3925                         struct nfs4_ol_stateid *stp)
3926 {
3927         struct nfs4_delegation *dp;
3928         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
3929         struct nfs4_client *clp = stp->st_stid.sc_client;
3930         int cb_up;
3931         int status = 0;
3932
3933         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3934         open->op_recall = 0;
3935         switch (open->op_claim_type) {
3936                 case NFS4_OPEN_CLAIM_PREVIOUS:
3937                         if (!cb_up)
3938                                 open->op_recall = 1;
3939                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3940                                 goto out_no_deleg;
3941                         break;
3942                 case NFS4_OPEN_CLAIM_NULL:
3943                 case NFS4_OPEN_CLAIM_FH:
3944                         /*
3945                          * Let's not give out any delegations till everyone's
3946                          * had the chance to reclaim theirs....
3947                          */
3948                         if (locks_in_grace(clp->net))
3949                                 goto out_no_deleg;
3950                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3951                                 goto out_no_deleg;
3952                         /*
3953                          * Also, if the file was opened for write or
3954                          * create, there's a good chance the client's
3955                          * about to write to it, resulting in an
3956                          * immediate recall (since we don't support
3957                          * write delegations):
3958                          */
3959                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3960                                 goto out_no_deleg;
3961                         if (open->op_create == NFS4_OPEN_CREATE)
3962                                 goto out_no_deleg;
3963                         break;
3964                 default:
3965                         goto out_no_deleg;
3966         }
3967         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file);
3968         if (IS_ERR(dp))
3969                 goto out_no_deleg;
3970
3971         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3972
3973         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3974                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3975         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3976         nfs4_put_stid(&dp->dl_stid);
3977         return;
3978 out_no_deleg:
3979         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3980         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3981             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3982                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3983                 open->op_recall = 1;
3984         }
3985
3986         /* 4.1 client asking for a delegation? */
3987         if (open->op_deleg_want)
3988                 nfsd4_open_deleg_none_ext(open, status);
3989         return;
3990 }
3991
3992 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3993                                         struct nfs4_delegation *dp)
3994 {
3995         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3996             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3997                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3998                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3999         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4000                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4001                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4002                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4003         }
4004         /* Otherwise the client must be confused wanting a delegation
4005          * it already has, therefore we don't return
4006          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4007          */
4008 }
4009
4010 /*
4011  * called with nfs4_lock_state() held.
4012  */
4013 __be32
4014 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4015 {
4016         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4017         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4018         struct nfs4_file *fp = NULL;
4019         struct nfs4_ol_stateid *stp = NULL;
4020         struct nfs4_delegation *dp = NULL;
4021         __be32 status;
4022
4023         /*
4024          * Lookup file; if found, lookup stateid and check open request,
4025          * and check for delegations in the process of being recalled.
4026          * If not found, create the nfs4_file struct
4027          */
4028         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4029         if (fp != open->op_file) {
4030                 status = nfs4_check_deleg(cl, open, &dp);
4031                 if (status)
4032                         goto out;
4033                 stp = nfsd4_find_existing_open(fp, open);
4034         } else {
4035                 open->op_file = NULL;
4036                 status = nfserr_bad_stateid;
4037                 if (nfsd4_is_deleg_cur(open))
4038                         goto out;
4039                 status = nfserr_jukebox;
4040         }
4041
4042         /*
4043          * OPEN the file, or upgrade an existing OPEN.
4044          * If truncate fails, the OPEN fails.
4045          */
4046         if (stp) {
4047                 /* Stateid was found, this is an OPEN upgrade */
4048                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4049                 if (status)
4050                         goto out;
4051         } else {
4052                 stp = open->op_stp;
4053                 open->op_stp = NULL;
4054                 init_open_stateid(stp, fp, open);
4055                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4056                 if (status) {
4057                         release_open_stateid(stp);
4058                         goto out;
4059                 }
4060         }
4061         update_stateid(&stp->st_stid.sc_stateid);
4062         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4063
4064         if (nfsd4_has_session(&resp->cstate)) {
4065                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4066                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4067                         open->op_why_no_deleg = WND4_NOT_WANTED;
4068                         goto nodeleg;
4069                 }
4070         }
4071
4072         /*
4073         * Attempt to hand out a delegation. No error return, because the
4074         * OPEN succeeds even if we fail.
4075         */
4076         nfs4_open_delegation(current_fh, open, stp);
4077 nodeleg:
4078         status = nfs_ok;
4079
4080         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4081                 STATEID_VAL(&stp->st_stid.sc_stateid));
4082 out:
4083         /* 4.1 client trying to upgrade/downgrade delegation? */
4084         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4085             open->op_deleg_want)
4086                 nfsd4_deleg_xgrade_none_ext(open, dp);
4087
4088         if (fp)
4089                 put_nfs4_file(fp);
4090         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4091                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
4092         /*
4093         * To finish the open response, we just need to set the rflags.
4094         */
4095         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4096         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4097             !nfsd4_has_session(&resp->cstate))
4098                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4099         if (dp)
4100                 nfs4_put_stid(&dp->dl_stid);
4101         if (stp)
4102                 nfs4_put_stid(&stp->st_stid);
4103
4104         return status;
4105 }
4106
4107 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4108                               struct nfsd4_open *open, __be32 status)
4109 {
4110         if (open->op_openowner) {
4111                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4112
4113                 nfsd4_cstate_assign_replay(cstate, so);
4114                 nfs4_put_stateowner(so);
4115         }
4116         if (open->op_file)
4117                 nfsd4_free_file(open->op_file);
4118         if (open->op_stp)
4119                 nfs4_put_stid(&open->op_stp->st_stid);
4120 }
4121
4122 __be32
4123 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4124             clientid_t *clid)
4125 {
4126         struct nfs4_client *clp;
4127         __be32 status;
4128         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4129
4130         nfs4_lock_state();
4131         dprintk("process_renew(%08x/%08x): starting\n",
4132                         clid->cl_boot, clid->cl_id);
4133         status = lookup_clientid(clid, cstate, nn);
4134         if (status)
4135                 goto out;
4136         clp = cstate->clp;
4137         status = nfserr_cb_path_down;
4138         if (!list_empty(&clp->cl_delegations)
4139                         && clp->cl_cb_state != NFSD4_CB_UP)
4140                 goto out;
4141         status = nfs_ok;
4142 out:
4143         nfs4_unlock_state();
4144         return status;
4145 }
4146
4147 static void
4148 nfsd4_end_grace(struct nfsd_net *nn)
4149 {
4150         /* do nothing if grace period already ended */
4151         if (nn->grace_ended)
4152                 return;
4153
4154         dprintk("NFSD: end of grace period\n");
4155         nn->grace_ended = true;
4156         nfsd4_record_grace_done(nn, nn->boot_time);
4157         locks_end_grace(&nn->nfsd4_manager);
4158         /*
4159          * Now that every NFSv4 client has had the chance to recover and
4160          * to see the (possibly new, possibly shorter) lease time, we
4161          * can safely set the next grace time to the current lease time:
4162          */
4163         nn->nfsd4_grace = nn->nfsd4_lease;
4164 }
4165
4166 static time_t
4167 nfs4_laundromat(struct nfsd_net *nn)
4168 {
4169         struct nfs4_client *clp;
4170         struct nfs4_openowner *oo;
4171         struct nfs4_delegation *dp;
4172         struct nfs4_ol_stateid *stp;
4173         struct list_head *pos, *next, reaplist;
4174         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4175         time_t t, new_timeo = nn->nfsd4_lease;
4176
4177         nfs4_lock_state();
4178
4179         dprintk("NFSD: laundromat service - starting\n");
4180         nfsd4_end_grace(nn);
4181         INIT_LIST_HEAD(&reaplist);
4182         spin_lock(&nn->client_lock);
4183         list_for_each_safe(pos, next, &nn->client_lru) {
4184                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4185                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4186                         t = clp->cl_time - cutoff;
4187                         new_timeo = min(new_timeo, t);
4188                         break;
4189                 }
4190                 if (mark_client_expired_locked(clp)) {
4191                         dprintk("NFSD: client in use (clientid %08x)\n",
4192                                 clp->cl_clientid.cl_id);
4193                         continue;
4194                 }
4195                 list_add(&clp->cl_lru, &reaplist);
4196         }
4197         spin_unlock(&nn->client_lock);
4198         list_for_each_safe(pos, next, &reaplist) {
4199                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4200                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4201                         clp->cl_clientid.cl_id);
4202                 list_del_init(&clp->cl_lru);
4203                 expire_client(clp);
4204         }
4205         spin_lock(&state_lock);
4206         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4207                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4208                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
4209                         continue;
4210                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4211                         t = dp->dl_time - cutoff;
4212                         new_timeo = min(new_timeo, t);
4213                         break;
4214                 }
4215                 unhash_delegation_locked(dp);
4216                 list_add(&dp->dl_recall_lru, &reaplist);
4217         }
4218         spin_unlock(&state_lock);
4219         while (!list_empty(&reaplist)) {
4220                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4221                                         dl_recall_lru);
4222                 list_del_init(&dp->dl_recall_lru);
4223                 revoke_delegation(dp);
4224         }
4225
4226         spin_lock(&nn->client_lock);
4227         while (!list_empty(&nn->close_lru)) {
4228                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4229                                         oo_close_lru);
4230                 if (time_after((unsigned long)oo->oo_time,
4231                                (unsigned long)cutoff)) {
4232                         t = oo->oo_time - cutoff;
4233                         new_timeo = min(new_timeo, t);
4234                         break;
4235                 }
4236                 list_del_init(&oo->oo_close_lru);
4237                 stp = oo->oo_last_closed_stid;
4238                 oo->oo_last_closed_stid = NULL;
4239                 spin_unlock(&nn->client_lock);
4240                 nfs4_put_stid(&stp->st_stid);
4241                 spin_lock(&nn->client_lock);
4242         }
4243         spin_unlock(&nn->client_lock);
4244
4245         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4246         nfs4_unlock_state();
4247         return new_timeo;
4248 }
4249
4250 static struct workqueue_struct *laundry_wq;
4251 static void laundromat_main(struct work_struct *);
4252
4253 static void
4254 laundromat_main(struct work_struct *laundry)
4255 {
4256         time_t t;
4257         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
4258                                                   work);
4259         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4260                                            laundromat_work);
4261
4262         t = nfs4_laundromat(nn);
4263         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4264         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4265 }
4266
4267 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
4268 {
4269         if (!nfsd_fh_match(&fhp->fh_handle, &stp->st_stid.sc_file->fi_fhandle))
4270                 return nfserr_bad_stateid;
4271         return nfs_ok;
4272 }
4273
4274 static inline int
4275 access_permit_read(struct nfs4_ol_stateid *stp)
4276 {
4277         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4278                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4279                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4280 }
4281
4282 static inline int
4283 access_permit_write(struct nfs4_ol_stateid *stp)
4284 {
4285         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4286                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4287 }
4288
4289 static
4290 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4291 {
4292         __be32 status = nfserr_openmode;
4293
4294         /* For lock stateid's, we test the parent open, not the lock: */
4295         if (stp->st_openstp)
4296                 stp = stp->st_openstp;
4297         if ((flags & WR_STATE) && !access_permit_write(stp))
4298                 goto out;
4299         if ((flags & RD_STATE) && !access_permit_read(stp))
4300                 goto out;
4301         status = nfs_ok;
4302 out:
4303         return status;
4304 }
4305
4306 static inline __be32
4307 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4308 {
4309         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4310                 return nfs_ok;
4311         else if (locks_in_grace(net)) {
4312                 /* Answer in remaining cases depends on existence of
4313                  * conflicting state; so we must wait out the grace period. */
4314                 return nfserr_grace;
4315         } else if (flags & WR_STATE)
4316                 return nfs4_share_conflict(current_fh,
4317                                 NFS4_SHARE_DENY_WRITE);
4318         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4319                 return nfs4_share_conflict(current_fh,
4320                                 NFS4_SHARE_DENY_READ);
4321 }
4322
4323 /*
4324  * Allow READ/WRITE during grace period on recovered state only for files
4325  * that are not able to provide mandatory locking.
4326  */
4327 static inline int
4328 grace_disallows_io(struct net *net, struct inode *inode)
4329 {
4330         return locks_in_grace(net) && mandatory_lock(inode);
4331 }
4332
4333 /* Returns true iff a is later than b: */
4334 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4335 {
4336         return (s32)(a->si_generation - b->si_generation) > 0;
4337 }
4338
4339 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4340 {
4341         /*
4342          * When sessions are used the stateid generation number is ignored
4343          * when it is zero.
4344          */
4345         if (has_session && in->si_generation == 0)
4346                 return nfs_ok;
4347
4348         if (in->si_generation == ref->si_generation)
4349                 return nfs_ok;
4350
4351         /* If the client sends us a stateid from the future, it's buggy: */
4352         if (stateid_generation_after(in, ref))
4353                 return nfserr_bad_stateid;
4354         /*
4355          * However, we could see a stateid from the past, even from a
4356          * non-buggy client.  For example, if the client sends a lock
4357          * while some IO is outstanding, the lock may bump si_generation
4358          * while the IO is still in flight.  The client could avoid that
4359          * situation by waiting for responses on all the IO requests,
4360          * but better performance may result in retrying IO that
4361          * receives an old_stateid error if requests are rarely
4362          * reordered in flight:
4363          */
4364         return nfserr_old_stateid;
4365 }
4366
4367 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4368 {
4369         struct nfs4_stid *s;
4370         struct nfs4_ol_stateid *ols;
4371         __be32 status = nfserr_bad_stateid;
4372
4373         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4374                 return status;
4375         /* Client debugging aid. */
4376         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4377                 char addr_str[INET6_ADDRSTRLEN];
4378                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4379                                  sizeof(addr_str));
4380                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4381                                         "with incorrect client ID\n", addr_str);
4382                 return status;
4383         }
4384         spin_lock(&cl->cl_lock);
4385         s = find_stateid_locked(cl, stateid);
4386         if (!s)
4387                 goto out_unlock;
4388         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4389         if (status)
4390                 goto out_unlock;
4391         switch (s->sc_type) {
4392         case NFS4_DELEG_STID:
4393                 status = nfs_ok;
4394                 break;
4395         case NFS4_REVOKED_DELEG_STID:
4396                 status = nfserr_deleg_revoked;
4397                 break;
4398         case NFS4_OPEN_STID:
4399         case NFS4_LOCK_STID:
4400                 ols = openlockstateid(s);
4401                 if (ols->st_stateowner->so_is_open_owner
4402                                 && !(openowner(ols->st_stateowner)->oo_flags
4403                                                 & NFS4_OO_CONFIRMED))
4404                         status = nfserr_bad_stateid;
4405                 else
4406                         status = nfs_ok;
4407                 break;
4408         default:
4409                 printk("unknown stateid type %x\n", s->sc_type);
4410                 /* Fallthrough */
4411         case NFS4_CLOSED_STID:
4412         case NFS4_CLOSED_DELEG_STID:
4413                 status = nfserr_bad_stateid;
4414         }
4415 out_unlock:
4416         spin_unlock(&cl->cl_lock);
4417         return status;
4418 }
4419
4420 static __be32
4421 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4422                      stateid_t *stateid, unsigned char typemask,
4423                      struct nfs4_stid **s, struct nfsd_net *nn)
4424 {
4425         __be32 status;
4426
4427         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4428                 return nfserr_bad_stateid;
4429         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4430         if (status == nfserr_stale_clientid) {
4431                 if (cstate->session)
4432                         return nfserr_bad_stateid;
4433                 return nfserr_stale_stateid;
4434         }
4435         if (status)
4436                 return status;
4437         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4438         if (!*s)
4439                 return nfserr_bad_stateid;
4440         return nfs_ok;
4441 }
4442
4443 /*
4444 * Checks for stateid operations
4445 */
4446 __be32
4447 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
4448                            stateid_t *stateid, int flags, struct file **filpp)
4449 {
4450         struct nfs4_stid *s;
4451         struct nfs4_ol_stateid *stp = NULL;
4452         struct nfs4_delegation *dp = NULL;
4453         struct svc_fh *current_fh = &cstate->current_fh;
4454         struct inode *ino = current_fh->fh_dentry->d_inode;
4455         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4456         struct file *file = NULL;
4457         __be32 status;
4458
4459         if (filpp)
4460                 *filpp = NULL;
4461
4462         if (grace_disallows_io(net, ino))
4463                 return nfserr_grace;
4464
4465         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4466                 return check_special_stateids(net, current_fh, stateid, flags);
4467
4468         status = nfsd4_lookup_stateid(cstate, stateid,
4469                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4470                                 &s, nn);
4471         if (status)
4472                 return status;
4473         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
4474         if (status)
4475                 goto out;
4476         switch (s->sc_type) {
4477         case NFS4_DELEG_STID:
4478                 dp = delegstateid(s);
4479                 status = nfs4_check_delegmode(dp, flags);
4480                 if (status)
4481                         goto out;
4482                 if (filpp) {
4483                         file = dp->dl_stid.sc_file->fi_deleg_file;
4484                         if (!file) {
4485                                 WARN_ON_ONCE(1);
4486                                 status = nfserr_serverfault;
4487                                 goto out;
4488                         }
4489                         get_file(file);
4490                 }
4491                 break;
4492         case NFS4_OPEN_STID:
4493         case NFS4_LOCK_STID:
4494                 stp = openlockstateid(s);
4495                 status = nfs4_check_fh(current_fh, stp);
4496                 if (status)
4497                         goto out;
4498                 if (stp->st_stateowner->so_is_open_owner
4499                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4500                         goto out;
4501                 status = nfs4_check_openmode(stp, flags);
4502                 if (status)
4503                         goto out;
4504                 if (filpp) {
4505                         struct nfs4_file *fp = stp->st_stid.sc_file;
4506
4507                         if (flags & RD_STATE)
4508                                 file = find_readable_file(fp);
4509                         else
4510                                 file = find_writeable_file(fp);
4511                 }
4512                 break;
4513         default:
4514                 status = nfserr_bad_stateid;
4515                 goto out;
4516         }
4517         status = nfs_ok;
4518         if (file)
4519                 *filpp = file;
4520 out:
4521         nfs4_put_stid(s);
4522         return status;
4523 }
4524
4525 /*
4526  * Test if the stateid is valid
4527  */
4528 __be32
4529 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4530                    struct nfsd4_test_stateid *test_stateid)
4531 {
4532         struct nfsd4_test_stateid_id *stateid;
4533         struct nfs4_client *cl = cstate->session->se_client;
4534
4535         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4536                 stateid->ts_id_status =
4537                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4538
4539         return nfs_ok;
4540 }
4541
4542 __be32
4543 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4544                    struct nfsd4_free_stateid *free_stateid)
4545 {
4546         stateid_t *stateid = &free_stateid->fr_stateid;
4547         struct nfs4_stid *s;
4548         struct nfs4_delegation *dp;
4549         struct nfs4_ol_stateid *stp;
4550         struct nfs4_client *cl = cstate->session->se_client;
4551         __be32 ret = nfserr_bad_stateid;
4552
4553         spin_lock(&cl->cl_lock);
4554         s = find_stateid_locked(cl, stateid);
4555         if (!s)
4556                 goto out_unlock;
4557         switch (s->sc_type) {
4558         case NFS4_DELEG_STID:
4559                 ret = nfserr_locks_held;
4560                 break;
4561         case NFS4_OPEN_STID:
4562                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4563                 if (ret)
4564                         break;
4565                 ret = nfserr_locks_held;
4566                 break;
4567         case NFS4_LOCK_STID:
4568                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4569                 if (ret)
4570                         break;
4571                 stp = openlockstateid(s);
4572                 ret = nfserr_locks_held;
4573                 if (check_for_locks(stp->st_stid.sc_file,
4574                                     lockowner(stp->st_stateowner)))
4575                         break;
4576                 unhash_lock_stateid(stp);
4577                 spin_unlock(&cl->cl_lock);
4578                 nfs4_put_stid(s);
4579                 ret = nfs_ok;
4580                 goto out;
4581         case NFS4_REVOKED_DELEG_STID:
4582                 dp = delegstateid(s);
4583                 list_del_init(&dp->dl_recall_lru);
4584                 spin_unlock(&cl->cl_lock);
4585                 nfs4_put_stid(s);
4586                 ret = nfs_ok;
4587                 goto out;
4588         /* Default falls through and returns nfserr_bad_stateid */
4589         }
4590 out_unlock:
4591         spin_unlock(&cl->cl_lock);
4592 out:
4593         return ret;
4594 }
4595
4596 static inline int
4597 setlkflg (int type)
4598 {
4599         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4600                 RD_STATE : WR_STATE;
4601 }
4602
4603 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4604 {
4605         struct svc_fh *current_fh = &cstate->current_fh;
4606         struct nfs4_stateowner *sop = stp->st_stateowner;
4607         __be32 status;
4608
4609         status = nfsd4_check_seqid(cstate, sop, seqid);
4610         if (status)
4611                 return status;
4612         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4613                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4614                 /*
4615                  * "Closed" stateid's exist *only* to return
4616                  * nfserr_replay_me from the previous step, and
4617                  * revoked delegations are kept only for free_stateid.
4618                  */
4619                 return nfserr_bad_stateid;
4620         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4621         if (status)
4622                 return status;
4623         return nfs4_check_fh(current_fh, stp);
4624 }
4625
4626 /*
4627  * Checks for sequence id mutating operations.
4628  */
4629 static __be32
4630 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4631                          stateid_t *stateid, char typemask,
4632                          struct nfs4_ol_stateid **stpp,
4633                          struct nfsd_net *nn)
4634 {
4635         __be32 status;
4636         struct nfs4_stid *s;
4637         struct nfs4_ol_stateid *stp = NULL;
4638
4639         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4640                 seqid, STATEID_VAL(stateid));
4641
4642         *stpp = NULL;
4643         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4644         if (status)
4645                 return status;
4646         stp = openlockstateid(s);
4647         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
4648
4649         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4650         if (!status)
4651                 *stpp = stp;
4652         else
4653                 nfs4_put_stid(&stp->st_stid);
4654         return status;
4655 }
4656
4657 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4658                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4659 {
4660         __be32 status;
4661         struct nfs4_openowner *oo;
4662         struct nfs4_ol_stateid *stp;
4663
4664         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4665                                                 NFS4_OPEN_STID, &stp, nn);
4666         if (status)
4667                 return status;
4668         oo = openowner(stp->st_stateowner);
4669         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4670                 nfs4_put_stid(&stp->st_stid);
4671                 return nfserr_bad_stateid;
4672         }
4673         *stpp = stp;
4674         return nfs_ok;
4675 }
4676
4677 __be32
4678 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4679                    struct nfsd4_open_confirm *oc)
4680 {
4681         __be32 status;
4682         struct nfs4_openowner *oo;
4683         struct nfs4_ol_stateid *stp;
4684         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4685
4686         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4687                         cstate->current_fh.fh_dentry);
4688
4689         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4690         if (status)
4691                 return status;
4692
4693         nfs4_lock_state();
4694
4695         status = nfs4_preprocess_seqid_op(cstate,
4696                                         oc->oc_seqid, &oc->oc_req_stateid,
4697                                         NFS4_OPEN_STID, &stp, nn);
4698         if (status)
4699                 goto out;
4700         oo = openowner(stp->st_stateowner);
4701         status = nfserr_bad_stateid;
4702         if (oo->oo_flags & NFS4_OO_CONFIRMED)
4703                 goto put_stateid;
4704         oo->oo_flags |= NFS4_OO_CONFIRMED;
4705         update_stateid(&stp->st_stid.sc_stateid);
4706         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4707         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
4708                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
4709
4710         nfsd4_client_record_create(oo->oo_owner.so_client);
4711         status = nfs_ok;
4712 put_stateid:
4713         nfs4_put_stid(&stp->st_stid);
4714 out:
4715         nfsd4_bump_seqid(cstate, status);
4716         nfs4_unlock_state();
4717         return status;
4718 }
4719
4720 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
4721 {
4722         if (!test_access(access, stp))
4723                 return;
4724         nfs4_file_put_access(stp->st_stid.sc_file, access);
4725         clear_access(access, stp);
4726 }
4727
4728 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
4729 {
4730         switch (to_access) {
4731         case NFS4_SHARE_ACCESS_READ:
4732                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
4733                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4734                 break;
4735         case NFS4_SHARE_ACCESS_WRITE:
4736                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
4737                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4738                 break;
4739         case NFS4_SHARE_ACCESS_BOTH:
4740                 break;
4741         default:
4742                 WARN_ON_ONCE(1);
4743         }
4744 }
4745
4746 __be32
4747 nfsd4_open_downgrade(struct svc_rqst *rqstp,
4748                      struct nfsd4_compound_state *cstate,
4749                      struct nfsd4_open_downgrade *od)
4750 {
4751         __be32 status;
4752         struct nfs4_ol_stateid *stp;
4753         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4754
4755         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
4756                         cstate->current_fh.fh_dentry);
4757
4758         /* We don't yet support WANT bits: */
4759         if (od->od_deleg_want)
4760                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
4761                         od->od_deleg_want);
4762
4763         nfs4_lock_state();
4764         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
4765                                         &od->od_stateid, &stp, nn);
4766         if (status)
4767                 goto out;
4768         status = nfserr_inval;
4769         if (!test_access(od->od_share_access, stp)) {
4770                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
4771                         stp->st_access_bmap, od->od_share_access);
4772                 goto put_stateid;
4773         }
4774         if (!test_deny(od->od_share_deny, stp)) {
4775                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
4776                         stp->st_deny_bmap, od->od_share_deny);
4777                 goto put_stateid;
4778         }
4779         nfs4_stateid_downgrade(stp, od->od_share_access);
4780
4781         reset_union_bmap_deny(od->od_share_deny, stp);
4782
4783         update_stateid(&stp->st_stid.sc_stateid);
4784         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4785         status = nfs_ok;
4786 put_stateid:
4787         nfs4_put_stid(&stp->st_stid);
4788 out:
4789         nfsd4_bump_seqid(cstate, status);
4790         nfs4_unlock_state();
4791         return status;
4792 }
4793
4794 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
4795 {
4796         struct nfs4_client *clp = s->st_stid.sc_client;
4797         LIST_HEAD(reaplist);
4798
4799         s->st_stid.sc_type = NFS4_CLOSED_STID;
4800         spin_lock(&clp->cl_lock);
4801         unhash_open_stateid(s, &reaplist);
4802
4803         if (clp->cl_minorversion) {
4804                 put_ol_stateid_locked(s, &reaplist);
4805                 spin_unlock(&clp->cl_lock);
4806                 free_ol_stateid_reaplist(&reaplist);
4807         } else {
4808                 spin_unlock(&clp->cl_lock);
4809                 free_ol_stateid_reaplist(&reaplist);
4810                 move_to_close_lru(s, clp->net);
4811         }
4812 }
4813
4814 /*
4815  * nfs4_unlock_state() called after encode
4816  */
4817 __be32
4818 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4819             struct nfsd4_close *close)
4820 {
4821         __be32 status;
4822         struct nfs4_ol_stateid *stp;
4823         struct net *net = SVC_NET(rqstp);
4824         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4825
4826         dprintk("NFSD: nfsd4_close on file %pd\n",
4827                         cstate->current_fh.fh_dentry);
4828
4829         nfs4_lock_state();
4830         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4831                                         &close->cl_stateid,
4832                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4833                                         &stp, nn);
4834         nfsd4_bump_seqid(cstate, status);
4835         if (status)
4836                 goto out;
4837         update_stateid(&stp->st_stid.sc_stateid);
4838         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4839
4840         nfsd4_close_open_stateid(stp);
4841
4842         /* put reference from nfs4_preprocess_seqid_op */
4843         nfs4_put_stid(&stp->st_stid);
4844 out:
4845         nfs4_unlock_state();
4846         return status;
4847 }
4848
4849 __be32
4850 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4851                   struct nfsd4_delegreturn *dr)
4852 {
4853         struct nfs4_delegation *dp;
4854         stateid_t *stateid = &dr->dr_stateid;
4855         struct nfs4_stid *s;
4856         __be32 status;
4857         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4858
4859         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4860                 return status;
4861
4862         nfs4_lock_state();
4863         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
4864         if (status)
4865                 goto out;
4866         dp = delegstateid(s);
4867         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4868         if (status)
4869                 goto put_stateid;
4870
4871         destroy_delegation(dp);
4872 put_stateid:
4873         nfs4_put_stid(&dp->dl_stid);
4874 out:
4875         nfs4_unlock_state();
4876
4877         return status;
4878 }
4879
4880
4881 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4882
4883 static inline u64
4884 end_offset(u64 start, u64 len)
4885 {
4886         u64 end;
4887
4888         end = start + len;
4889         return end >= start ? end: NFS4_MAX_UINT64;
4890 }
4891
4892 /* last octet in a range */
4893 static inline u64
4894 last_byte_offset(u64 start, u64 len)
4895 {
4896         u64 end;
4897
4898         WARN_ON_ONCE(!len);
4899         end = start + len;
4900         return end > start ? end - 1: NFS4_MAX_UINT64;
4901 }
4902
4903 /*
4904  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4905  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4906  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4907  * locking, this prevents us from being completely protocol-compliant.  The
4908  * real solution to this problem is to start using unsigned file offsets in
4909  * the VFS, but this is a very deep change!
4910  */
4911 static inline void
4912 nfs4_transform_lock_offset(struct file_lock *lock)
4913 {
4914         if (lock->fl_start < 0)
4915                 lock->fl_start = OFFSET_MAX;
4916         if (lock->fl_end < 0)
4917                 lock->fl_end = OFFSET_MAX;
4918 }
4919
4920 /* Hack!: For now, we're defining this just so we can use a pointer to it
4921  * as a unique cookie to identify our (NFSv4's) posix locks. */
4922 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4923 };
4924
4925 static inline void
4926 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4927 {
4928         struct nfs4_lockowner *lo;
4929
4930         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4931                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4932                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4933                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4934                 if (!deny->ld_owner.data)
4935                         /* We just don't care that much */
4936                         goto nevermind;
4937                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4938                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4939         } else {
4940 nevermind:
4941                 deny->ld_owner.len = 0;
4942                 deny->ld_owner.data = NULL;
4943                 deny->ld_clientid.cl_boot = 0;
4944                 deny->ld_clientid.cl_id = 0;
4945         }
4946         deny->ld_start = fl->fl_start;
4947         deny->ld_length = NFS4_MAX_UINT64;
4948         if (fl->fl_end != NFS4_MAX_UINT64)
4949                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4950         deny->ld_type = NFS4_READ_LT;
4951         if (fl->fl_type != F_RDLCK)
4952                 deny->ld_type = NFS4_WRITE_LT;
4953 }
4954
4955 static struct nfs4_lockowner *
4956 find_lockowner_str_locked(clientid_t *clid, struct xdr_netobj *owner,
4957                 struct nfs4_client *clp)
4958 {
4959         unsigned int strhashval = ownerstr_hashval(owner);
4960         struct nfs4_stateowner *so;
4961
4962         lockdep_assert_held(&clp->cl_lock);
4963
4964         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
4965                             so_strhash) {
4966                 if (so->so_is_open_owner)
4967                         continue;
4968                 if (!same_owner_str(so, owner))
4969                         continue;
4970                 atomic_inc(&so->so_count);
4971                 return lockowner(so);
4972         }
4973         return NULL;
4974 }
4975
4976 static struct nfs4_lockowner *
4977 find_lockowner_str(clientid_t *clid, struct xdr_netobj *owner,
4978                 struct nfs4_client *clp)
4979 {
4980         struct nfs4_lockowner *lo;
4981
4982         spin_lock(&clp->cl_lock);
4983         lo = find_lockowner_str_locked(clid, owner, clp);
4984         spin_unlock(&clp->cl_lock);
4985         return lo;
4986 }
4987
4988 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
4989 {
4990         unhash_lockowner_locked(lockowner(sop));
4991 }
4992
4993 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
4994 {
4995         struct nfs4_lockowner *lo = lockowner(sop);
4996
4997         kmem_cache_free(lockowner_slab, lo);
4998 }
4999
5000 static const struct nfs4_stateowner_operations lockowner_ops = {
5001         .so_unhash =    nfs4_unhash_lockowner,
5002         .so_free =      nfs4_free_lockowner,
5003 };
5004
5005 /*
5006  * Alloc a lock owner structure.
5007  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5008  * occurred.
5009  *
5010  * strhashval = ownerstr_hashval
5011  */
5012 static struct nfs4_lockowner *
5013 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5014                            struct nfs4_ol_stateid *open_stp,
5015                            struct nfsd4_lock *lock)
5016 {
5017         struct nfs4_lockowner *lo, *ret;
5018
5019         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5020         if (!lo)
5021                 return NULL;
5022         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5023         lo->lo_owner.so_is_open_owner = 0;
5024         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5025         lo->lo_owner.so_ops = &lockowner_ops;
5026         spin_lock(&clp->cl_lock);
5027         ret = find_lockowner_str_locked(&clp->cl_clientid,
5028                         &lock->lk_new_owner, clp);
5029         if (ret == NULL) {
5030                 list_add(&lo->lo_owner.so_strhash,
5031                          &clp->cl_ownerstr_hashtbl[strhashval]);
5032                 ret = lo;
5033         } else
5034                 nfs4_free_lockowner(&lo->lo_owner);
5035         spin_unlock(&clp->cl_lock);
5036         return lo;
5037 }
5038
5039 static void
5040 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5041                   struct nfs4_file *fp, struct inode *inode,
5042                   struct nfs4_ol_stateid *open_stp)
5043 {
5044         struct nfs4_client *clp = lo->lo_owner.so_client;
5045
5046         lockdep_assert_held(&clp->cl_lock);
5047
5048         atomic_inc(&stp->st_stid.sc_count);
5049         stp->st_stid.sc_type = NFS4_LOCK_STID;
5050         stp->st_stateowner = &lo->lo_owner;
5051         atomic_inc(&lo->lo_owner.so_count);
5052         get_nfs4_file(fp);
5053         stp->st_stid.sc_file = fp;
5054         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5055         stp->st_access_bmap = 0;
5056         stp->st_deny_bmap = open_stp->st_deny_bmap;
5057         stp->st_openstp = open_stp;
5058         list_add(&stp->st_locks, &open_stp->st_locks);
5059         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5060         spin_lock(&fp->fi_lock);
5061         list_add(&stp->st_perfile, &fp->fi_stateids);
5062         spin_unlock(&fp->fi_lock);
5063 }
5064
5065 static struct nfs4_ol_stateid *
5066 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5067 {
5068         struct nfs4_ol_stateid *lst;
5069         struct nfs4_client *clp = lo->lo_owner.so_client;
5070
5071         lockdep_assert_held(&clp->cl_lock);
5072
5073         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5074                 if (lst->st_stid.sc_file == fp) {
5075                         atomic_inc(&lst->st_stid.sc_count);
5076                         return lst;
5077                 }
5078         }
5079         return NULL;
5080 }
5081
5082 static struct nfs4_ol_stateid *
5083 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5084                             struct inode *inode, struct nfs4_ol_stateid *ost,
5085                             bool *new)
5086 {
5087         struct nfs4_stid *ns = NULL;
5088         struct nfs4_ol_stateid *lst;
5089         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5090         struct nfs4_client *clp = oo->oo_owner.so_client;
5091
5092         spin_lock(&clp->cl_lock);
5093         lst = find_lock_stateid(lo, fi);
5094         if (lst == NULL) {
5095                 spin_unlock(&clp->cl_lock);
5096                 ns = nfs4_alloc_stid(clp, stateid_slab);
5097                 if (ns == NULL)
5098                         return NULL;
5099
5100                 spin_lock(&clp->cl_lock);
5101                 lst = find_lock_stateid(lo, fi);
5102                 if (likely(!lst)) {
5103                         lst = openlockstateid(ns);
5104                         init_lock_stateid(lst, lo, fi, inode, ost);
5105                         ns = NULL;
5106                         *new = true;
5107                 }
5108         }
5109         spin_unlock(&clp->cl_lock);
5110         if (ns)
5111                 nfs4_put_stid(ns);
5112         return lst;
5113 }
5114
5115 static int
5116 check_lock_length(u64 offset, u64 length)
5117 {
5118         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
5119              LOFF_OVERFLOW(offset, length)));
5120 }
5121
5122 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5123 {
5124         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5125
5126         lockdep_assert_held(&fp->fi_lock);
5127
5128         if (test_access(access, lock_stp))
5129                 return;
5130         __nfs4_file_get_access(fp, access);
5131         set_access(access, lock_stp);
5132 }
5133
5134 static __be32
5135 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5136                             struct nfs4_ol_stateid *ost,
5137                             struct nfsd4_lock *lock,
5138                             struct nfs4_ol_stateid **lst, bool *new)
5139 {
5140         __be32 status;
5141         struct nfs4_file *fi = ost->st_stid.sc_file;
5142         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5143         struct nfs4_client *cl = oo->oo_owner.so_client;
5144         struct inode *inode = cstate->current_fh.fh_dentry->d_inode;
5145         struct nfs4_lockowner *lo;
5146         unsigned int strhashval;
5147
5148         lo = find_lockowner_str(&cl->cl_clientid, &lock->v.new.owner, cl);
5149         if (!lo) {
5150                 strhashval = ownerstr_hashval(&lock->v.new.owner);
5151                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5152                 if (lo == NULL)
5153                         return nfserr_jukebox;
5154         } else {
5155                 /* with an existing lockowner, seqids must be the same */
5156                 status = nfserr_bad_seqid;
5157                 if (!cstate->minorversion &&
5158                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5159                         goto out;
5160         }
5161
5162         *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5163         if (*lst == NULL) {
5164                 status = nfserr_jukebox;
5165                 goto out;
5166         }
5167         status = nfs_ok;
5168 out:
5169         nfs4_put_stateowner(&lo->lo_owner);
5170         return status;
5171 }
5172
5173 /*
5174  *  LOCK operation
5175  */
5176 __be32
5177 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5178            struct nfsd4_lock *lock)
5179 {
5180         struct nfs4_openowner *open_sop = NULL;
5181         struct nfs4_lockowner *lock_sop = NULL;
5182         struct nfs4_ol_stateid *lock_stp = NULL;
5183         struct nfs4_ol_stateid *open_stp = NULL;
5184         struct nfs4_file *fp;
5185         struct file *filp = NULL;
5186         struct file_lock *file_lock = NULL;
5187         struct file_lock *conflock = NULL;
5188         __be32 status = 0;
5189         int lkflg;
5190         int err;
5191         bool new = false;
5192         struct net *net = SVC_NET(rqstp);
5193         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5194
5195         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5196                 (long long) lock->lk_offset,
5197                 (long long) lock->lk_length);
5198
5199         if (check_lock_length(lock->lk_offset, lock->lk_length))
5200                  return nfserr_inval;
5201
5202         if ((status = fh_verify(rqstp, &cstate->current_fh,
5203                                 S_IFREG, NFSD_MAY_LOCK))) {
5204                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5205                 return status;
5206         }
5207
5208         nfs4_lock_state();
5209
5210         if (lock->lk_is_new) {
5211                 if (nfsd4_has_session(cstate))
5212                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5213                         memcpy(&lock->v.new.clientid,
5214                                 &cstate->session->se_client->cl_clientid,
5215                                 sizeof(clientid_t));
5216
5217                 status = nfserr_stale_clientid;
5218                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5219                         goto out;
5220
5221                 /* validate and update open stateid and open seqid */
5222                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5223                                         lock->lk_new_open_seqid,
5224                                         &lock->lk_new_open_stateid,
5225                                         &open_stp, nn);
5226                 if (status)
5227                         goto out;
5228                 open_sop = openowner(open_stp->st_stateowner);
5229                 status = nfserr_bad_stateid;
5230                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5231                                                 &lock->v.new.clientid))
5232                         goto out;
5233                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5234                                                         &lock_stp, &new);
5235         } else {
5236                 status = nfs4_preprocess_seqid_op(cstate,
5237                                        lock->lk_old_lock_seqid,
5238                                        &lock->lk_old_lock_stateid,
5239                                        NFS4_LOCK_STID, &lock_stp, nn);
5240         }
5241         if (status)
5242                 goto out;
5243         lock_sop = lockowner(lock_stp->st_stateowner);
5244
5245         lkflg = setlkflg(lock->lk_type);
5246         status = nfs4_check_openmode(lock_stp, lkflg);
5247         if (status)
5248                 goto out;
5249
5250         status = nfserr_grace;
5251         if (locks_in_grace(net) && !lock->lk_reclaim)
5252                 goto out;
5253         status = nfserr_no_grace;
5254         if (!locks_in_grace(net) && lock->lk_reclaim)
5255                 goto out;
5256
5257         file_lock = locks_alloc_lock();
5258         if (!file_lock) {
5259                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5260                 status = nfserr_jukebox;
5261                 goto out;
5262         }
5263
5264         fp = lock_stp->st_stid.sc_file;
5265         locks_init_lock(file_lock);
5266         switch (lock->lk_type) {
5267                 case NFS4_READ_LT:
5268                 case NFS4_READW_LT:
5269                         spin_lock(&fp->fi_lock);
5270                         filp = find_readable_file_locked(fp);
5271                         if (filp)
5272                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5273                         spin_unlock(&fp->fi_lock);
5274                         file_lock->fl_type = F_RDLCK;
5275                         break;
5276                 case NFS4_WRITE_LT:
5277                 case NFS4_WRITEW_LT:
5278                         spin_lock(&fp->fi_lock);
5279                         filp = find_writeable_file_locked(fp);
5280                         if (filp)
5281                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5282                         spin_unlock(&fp->fi_lock);
5283                         file_lock->fl_type = F_WRLCK;
5284                         break;
5285                 default:
5286                         status = nfserr_inval;
5287                 goto out;
5288         }
5289         if (!filp) {
5290                 status = nfserr_openmode;
5291                 goto out;
5292         }
5293         file_lock->fl_owner = (fl_owner_t)lock_sop;
5294         file_lock->fl_pid = current->tgid;
5295         file_lock->fl_file = filp;
5296         file_lock->fl_flags = FL_POSIX;
5297         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5298         file_lock->fl_start = lock->lk_offset;
5299         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5300         nfs4_transform_lock_offset(file_lock);
5301
5302         conflock = locks_alloc_lock();
5303         if (!conflock) {
5304                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5305                 status = nfserr_jukebox;
5306                 goto out;
5307         }
5308
5309         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5310         switch (-err) {
5311         case 0: /* success! */
5312                 update_stateid(&lock_stp->st_stid.sc_stateid);
5313                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
5314                                 sizeof(stateid_t));
5315                 status = 0;
5316                 break;
5317         case (EAGAIN):          /* conflock holds conflicting lock */
5318                 status = nfserr_denied;
5319                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5320                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5321                 break;
5322         case (EDEADLK):
5323                 status = nfserr_deadlock;
5324                 break;
5325         default:
5326                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5327                 status = nfserrno(err);
5328                 break;
5329         }
5330 out:
5331         if (filp)
5332                 fput(filp);
5333         if (lock_stp) {
5334                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5335                 if (cstate->replay_owner &&
5336                     cstate->replay_owner != &lock_sop->lo_owner &&
5337                     seqid_mutating_err(ntohl(status)))
5338                         lock_sop->lo_owner.so_seqid++;
5339
5340                 /*
5341                  * If this is a new, never-before-used stateid, and we are
5342                  * returning an error, then just go ahead and release it.
5343                  */
5344                 if (status && new)
5345                         release_lock_stateid(lock_stp);
5346
5347                 nfs4_put_stid(&lock_stp->st_stid);
5348         }
5349         if (open_stp)
5350                 nfs4_put_stid(&open_stp->st_stid);
5351         nfsd4_bump_seqid(cstate, status);
5352         nfs4_unlock_state();
5353         if (file_lock)
5354                 locks_free_lock(file_lock);
5355         if (conflock)
5356                 locks_free_lock(conflock);
5357         return status;
5358 }
5359
5360 /*
5361  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5362  * so we do a temporary open here just to get an open file to pass to
5363  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5364  * inode operation.)
5365  */
5366 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5367 {
5368         struct file *file;
5369         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5370         if (!err) {
5371                 err = nfserrno(vfs_test_lock(file, lock));
5372                 nfsd_close(file);
5373         }
5374         return err;
5375 }
5376
5377 /*
5378  * LOCKT operation
5379  */
5380 __be32
5381 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5382             struct nfsd4_lockt *lockt)
5383 {
5384         struct file_lock *file_lock = NULL;
5385         struct nfs4_lockowner *lo = NULL;
5386         __be32 status;
5387         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5388
5389         if (locks_in_grace(SVC_NET(rqstp)))
5390                 return nfserr_grace;
5391
5392         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5393                  return nfserr_inval;
5394
5395         nfs4_lock_state();
5396
5397         if (!nfsd4_has_session(cstate)) {
5398                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5399                 if (status)
5400                         goto out;
5401         }
5402
5403         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5404                 goto out;
5405
5406         file_lock = locks_alloc_lock();
5407         if (!file_lock) {
5408                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5409                 status = nfserr_jukebox;
5410                 goto out;
5411         }
5412         locks_init_lock(file_lock);
5413         switch (lockt->lt_type) {
5414                 case NFS4_READ_LT:
5415                 case NFS4_READW_LT:
5416                         file_lock->fl_type = F_RDLCK;
5417                 break;
5418                 case NFS4_WRITE_LT:
5419                 case NFS4_WRITEW_LT:
5420                         file_lock->fl_type = F_WRLCK;
5421                 break;
5422                 default:
5423                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5424                         status = nfserr_inval;
5425                 goto out;
5426         }
5427
5428         lo = find_lockowner_str(&lockt->lt_clientid, &lockt->lt_owner,
5429                                 cstate->clp);
5430         if (lo)
5431                 file_lock->fl_owner = (fl_owner_t)lo;
5432         file_lock->fl_pid = current->tgid;
5433         file_lock->fl_flags = FL_POSIX;
5434
5435         file_lock->fl_start = lockt->lt_offset;
5436         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5437
5438         nfs4_transform_lock_offset(file_lock);
5439
5440         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5441         if (status)
5442                 goto out;
5443
5444         if (file_lock->fl_type != F_UNLCK) {
5445                 status = nfserr_denied;
5446                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5447         }
5448 out:
5449         if (lo)
5450                 nfs4_put_stateowner(&lo->lo_owner);
5451         nfs4_unlock_state();
5452         if (file_lock)
5453                 locks_free_lock(file_lock);
5454         return status;
5455 }
5456
5457 __be32
5458 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5459             struct nfsd4_locku *locku)
5460 {
5461         struct nfs4_ol_stateid *stp;
5462         struct file *filp = NULL;
5463         struct file_lock *file_lock = NULL;
5464         __be32 status;
5465         int err;
5466         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5467
5468         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5469                 (long long) locku->lu_offset,
5470                 (long long) locku->lu_length);
5471
5472         if (check_lock_length(locku->lu_offset, locku->lu_length))
5473                  return nfserr_inval;
5474
5475         nfs4_lock_state();
5476
5477         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5478                                         &locku->lu_stateid, NFS4_LOCK_STID,
5479                                         &stp, nn);
5480         if (status)
5481                 goto out;
5482         filp = find_any_file(stp->st_stid.sc_file);
5483         if (!filp) {
5484                 status = nfserr_lock_range;
5485                 goto put_stateid;
5486         }
5487         file_lock = locks_alloc_lock();
5488         if (!file_lock) {
5489                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5490                 status = nfserr_jukebox;
5491                 goto fput;
5492         }
5493         locks_init_lock(file_lock);
5494         file_lock->fl_type = F_UNLCK;
5495         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
5496         file_lock->fl_pid = current->tgid;
5497         file_lock->fl_file = filp;
5498         file_lock->fl_flags = FL_POSIX;
5499         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5500         file_lock->fl_start = locku->lu_offset;
5501
5502         file_lock->fl_end = last_byte_offset(locku->lu_offset,
5503                                                 locku->lu_length);
5504         nfs4_transform_lock_offset(file_lock);
5505
5506         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5507         if (err) {
5508                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5509                 goto out_nfserr;
5510         }
5511         update_stateid(&stp->st_stid.sc_stateid);
5512         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
5513 fput:
5514         fput(filp);
5515 put_stateid:
5516         nfs4_put_stid(&stp->st_stid);
5517 out:
5518         nfsd4_bump_seqid(cstate, status);
5519         nfs4_unlock_state();
5520         if (file_lock)
5521                 locks_free_lock(file_lock);
5522         return status;
5523
5524 out_nfserr:
5525         status = nfserrno(err);
5526         goto fput;
5527 }
5528
5529 /*
5530  * returns
5531  *      true:  locks held by lockowner
5532  *      false: no locks held by lockowner
5533  */
5534 static bool
5535 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5536 {
5537         struct file_lock **flpp;
5538         int status = false;
5539         struct file *filp = find_any_file(fp);
5540         struct inode *inode;
5541
5542         if (!filp) {
5543                 /* Any valid lock stateid should have some sort of access */
5544                 WARN_ON_ONCE(1);
5545                 return status;
5546         }
5547
5548         inode = file_inode(filp);
5549
5550         spin_lock(&inode->i_lock);
5551         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
5552                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
5553                         status = true;
5554                         break;
5555                 }
5556         }
5557         spin_unlock(&inode->i_lock);
5558         fput(filp);
5559         return status;
5560 }
5561
5562 __be32
5563 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5564                         struct nfsd4_compound_state *cstate,
5565                         struct nfsd4_release_lockowner *rlockowner)
5566 {
5567         clientid_t *clid = &rlockowner->rl_clientid;
5568         struct nfs4_stateowner *sop;
5569         struct nfs4_lockowner *lo = NULL;
5570         struct nfs4_ol_stateid *stp;
5571         struct xdr_netobj *owner = &rlockowner->rl_owner;
5572         unsigned int hashval = ownerstr_hashval(owner);
5573         __be32 status;
5574         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5575         struct nfs4_client *clp;
5576
5577         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5578                 clid->cl_boot, clid->cl_id);
5579
5580         status = lookup_clientid(clid, cstate, nn);
5581         if (status)
5582                 return status;
5583
5584         clp = cstate->clp;
5585         /* Find the matching lock stateowner */
5586         spin_lock(&clp->cl_lock);
5587         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5588                             so_strhash) {
5589
5590                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5591                         continue;
5592
5593                 /* see if there are still any locks associated with it */
5594                 lo = lockowner(sop);
5595                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5596                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
5597                                 status = nfserr_locks_held;
5598                                 spin_unlock(&clp->cl_lock);
5599                                 return status;
5600                         }
5601                 }
5602
5603                 atomic_inc(&sop->so_count);
5604                 break;
5605         }
5606         spin_unlock(&clp->cl_lock);
5607         if (lo)
5608                 release_lockowner(lo);
5609         return status;
5610 }
5611
5612 static inline struct nfs4_client_reclaim *
5613 alloc_reclaim(void)
5614 {
5615         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5616 }
5617
5618 bool
5619 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5620 {
5621         struct nfs4_client_reclaim *crp;
5622
5623         crp = nfsd4_find_reclaim_client(name, nn);
5624         return (crp && crp->cr_clp);
5625 }
5626
5627 /*
5628  * failure => all reset bets are off, nfserr_no_grace...
5629  */
5630 struct nfs4_client_reclaim *
5631 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5632 {
5633         unsigned int strhashval;
5634         struct nfs4_client_reclaim *crp;
5635
5636         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5637         crp = alloc_reclaim();
5638         if (crp) {
5639                 strhashval = clientstr_hashval(name);
5640                 INIT_LIST_HEAD(&crp->cr_strhash);
5641                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5642                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5643                 crp->cr_clp = NULL;
5644                 nn->reclaim_str_hashtbl_size++;
5645         }
5646         return crp;
5647 }
5648
5649 void
5650 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5651 {
5652         list_del(&crp->cr_strhash);
5653         kfree(crp);
5654         nn->reclaim_str_hashtbl_size--;
5655 }
5656
5657 void
5658 nfs4_release_reclaim(struct nfsd_net *nn)
5659 {
5660         struct nfs4_client_reclaim *crp = NULL;
5661         int i;
5662
5663         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5664                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5665                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5666                                         struct nfs4_client_reclaim, cr_strhash);
5667                         nfs4_remove_reclaim_record(crp, nn);
5668                 }
5669         }
5670         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5671 }
5672
5673 /*
5674  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5675 struct nfs4_client_reclaim *
5676 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5677 {
5678         unsigned int strhashval;
5679         struct nfs4_client_reclaim *crp = NULL;
5680
5681         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5682
5683         strhashval = clientstr_hashval(recdir);
5684         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5685                 if (same_name(crp->cr_recdir, recdir)) {
5686                         return crp;
5687                 }
5688         }
5689         return NULL;
5690 }
5691
5692 /*
5693 * Called from OPEN. Look for clientid in reclaim list.
5694 */
5695 __be32
5696 nfs4_check_open_reclaim(clientid_t *clid,
5697                 struct nfsd4_compound_state *cstate,
5698                 struct nfsd_net *nn)
5699 {
5700         __be32 status;
5701
5702         /* find clientid in conf_id_hashtbl */
5703         status = lookup_clientid(clid, cstate, nn);
5704         if (status)
5705                 return nfserr_reclaim_bad;
5706
5707         if (nfsd4_client_record_check(cstate->clp))
5708                 return nfserr_reclaim_bad;
5709
5710         return nfs_ok;
5711 }
5712
5713 #ifdef CONFIG_NFSD_FAULT_INJECTION
5714 static inline void
5715 put_client(struct nfs4_client *clp)
5716 {
5717         atomic_dec(&clp->cl_refcount);
5718 }
5719
5720 static struct nfs4_client *
5721 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
5722 {
5723         struct nfs4_client *clp;
5724         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5725                                           nfsd_net_id);
5726
5727         if (!nfsd_netns_ready(nn))
5728                 return NULL;
5729
5730         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5731                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
5732                         return clp;
5733         }
5734         return NULL;
5735 }
5736
5737 u64
5738 nfsd_inject_print_clients(void)
5739 {
5740         struct nfs4_client *clp;
5741         u64 count = 0;
5742         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5743                                           nfsd_net_id);
5744         char buf[INET6_ADDRSTRLEN];
5745
5746         if (!nfsd_netns_ready(nn))
5747                 return 0;
5748
5749         spin_lock(&nn->client_lock);
5750         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5751                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5752                 pr_info("NFS Client: %s\n", buf);
5753                 ++count;
5754         }
5755         spin_unlock(&nn->client_lock);
5756
5757         return count;
5758 }
5759
5760 u64
5761 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
5762 {
5763         u64 count = 0;
5764         struct nfs4_client *clp;
5765         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5766                                           nfsd_net_id);
5767
5768         if (!nfsd_netns_ready(nn))
5769                 return count;
5770
5771         spin_lock(&nn->client_lock);
5772         clp = nfsd_find_client(addr, addr_size);
5773         if (clp) {
5774                 if (mark_client_expired_locked(clp) == nfs_ok)
5775                         ++count;
5776                 else
5777                         clp = NULL;
5778         }
5779         spin_unlock(&nn->client_lock);
5780
5781         if (clp)
5782                 expire_client(clp);
5783
5784         return count;
5785 }
5786
5787 u64
5788 nfsd_inject_forget_clients(u64 max)
5789 {
5790         u64 count = 0;
5791         struct nfs4_client *clp, *next;
5792         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5793                                                 nfsd_net_id);
5794         LIST_HEAD(reaplist);
5795
5796         if (!nfsd_netns_ready(nn))
5797                 return count;
5798
5799         spin_lock(&nn->client_lock);
5800         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
5801                 if (mark_client_expired_locked(clp) == nfs_ok) {
5802                         list_add(&clp->cl_lru, &reaplist);
5803                         if (max != 0 && ++count >= max)
5804                                 break;
5805                 }
5806         }
5807         spin_unlock(&nn->client_lock);
5808
5809         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
5810                 expire_client(clp);
5811
5812         return count;
5813 }
5814
5815 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
5816                              const char *type)
5817 {
5818         char buf[INET6_ADDRSTRLEN];
5819         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5820         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
5821 }
5822
5823 static void
5824 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
5825                              struct list_head *collect)
5826 {
5827         struct nfs4_client *clp = lst->st_stid.sc_client;
5828         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5829                                           nfsd_net_id);
5830
5831         if (!collect)
5832                 return;
5833
5834         lockdep_assert_held(&nn->client_lock);
5835         atomic_inc(&clp->cl_refcount);
5836         list_add(&lst->st_locks, collect);
5837 }
5838
5839 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
5840                                     struct list_head *collect,
5841                                     void (*func)(struct nfs4_ol_stateid *))
5842 {
5843         struct nfs4_openowner *oop;
5844         struct nfs4_ol_stateid *stp, *st_next;
5845         struct nfs4_ol_stateid *lst, *lst_next;
5846         u64 count = 0;
5847
5848         spin_lock(&clp->cl_lock);
5849         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
5850                 list_for_each_entry_safe(stp, st_next,
5851                                 &oop->oo_owner.so_stateids, st_perstateowner) {
5852                         list_for_each_entry_safe(lst, lst_next,
5853                                         &stp->st_locks, st_locks) {
5854                                 if (func) {
5855                                         func(lst);
5856                                         nfsd_inject_add_lock_to_list(lst,
5857                                                                 collect);
5858                                 }
5859                                 ++count;
5860                                 /*
5861                                  * Despite the fact that these functions deal
5862                                  * with 64-bit integers for "count", we must
5863                                  * ensure that it doesn't blow up the
5864                                  * clp->cl_refcount. Throw a warning if we
5865                                  * start to approach INT_MAX here.
5866                                  */
5867                                 WARN_ON_ONCE(count == (INT_MAX / 2));
5868                                 if (count == max)
5869                                         goto out;
5870                         }
5871                 }
5872         }
5873 out:
5874         spin_unlock(&clp->cl_lock);
5875
5876         return count;
5877 }
5878
5879 static u64
5880 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
5881                           u64 max)
5882 {
5883         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
5884 }
5885
5886 static u64
5887 nfsd_print_client_locks(struct nfs4_client *clp)
5888 {
5889         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
5890         nfsd_print_count(clp, count, "locked files");
5891         return count;
5892 }
5893
5894 u64
5895 nfsd_inject_print_locks(void)
5896 {
5897         struct nfs4_client *clp;
5898         u64 count = 0;
5899         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5900                                                 nfsd_net_id);
5901
5902         if (!nfsd_netns_ready(nn))
5903                 return 0;
5904
5905         spin_lock(&nn->client_lock);
5906         list_for_each_entry(clp, &nn->client_lru, cl_lru)
5907                 count += nfsd_print_client_locks(clp);
5908         spin_unlock(&nn->client_lock);
5909
5910         return count;
5911 }
5912
5913 static void
5914 nfsd_reap_locks(struct list_head *reaplist)
5915 {
5916         struct nfs4_client *clp;
5917         struct nfs4_ol_stateid *stp, *next;
5918
5919         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
5920                 list_del_init(&stp->st_locks);
5921                 clp = stp->st_stid.sc_client;
5922                 nfs4_put_stid(&stp->st_stid);
5923                 put_client(clp);
5924         }
5925 }
5926
5927 u64
5928 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
5929 {
5930         unsigned int count = 0;
5931         struct nfs4_client *clp;
5932         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5933                                                 nfsd_net_id);
5934         LIST_HEAD(reaplist);
5935
5936         if (!nfsd_netns_ready(nn))
5937                 return count;
5938
5939         spin_lock(&nn->client_lock);
5940         clp = nfsd_find_client(addr, addr_size);
5941         if (clp)
5942                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
5943         spin_unlock(&nn->client_lock);
5944         nfsd_reap_locks(&reaplist);
5945         return count;
5946 }
5947
5948 u64
5949 nfsd_inject_forget_locks(u64 max)
5950 {
5951         u64 count = 0;
5952         struct nfs4_client *clp;
5953         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5954                                                 nfsd_net_id);
5955         LIST_HEAD(reaplist);
5956
5957         if (!nfsd_netns_ready(nn))
5958                 return count;
5959
5960         spin_lock(&nn->client_lock);
5961         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5962                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
5963                 if (max != 0 && count >= max)
5964                         break;
5965         }
5966         spin_unlock(&nn->client_lock);
5967         nfsd_reap_locks(&reaplist);
5968         return count;
5969 }
5970
5971 static u64
5972 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
5973                               struct list_head *collect,
5974                               void (*func)(struct nfs4_openowner *))
5975 {
5976         struct nfs4_openowner *oop, *next;
5977         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5978                                                 nfsd_net_id);
5979         u64 count = 0;
5980
5981         lockdep_assert_held(&nn->client_lock);
5982
5983         spin_lock(&clp->cl_lock);
5984         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
5985                 if (func) {
5986                         func(oop);
5987                         if (collect) {
5988                                 atomic_inc(&clp->cl_refcount);
5989                                 list_add(&oop->oo_perclient, collect);
5990                         }
5991                 }
5992                 ++count;
5993                 /*
5994                  * Despite the fact that these functions deal with
5995                  * 64-bit integers for "count", we must ensure that
5996                  * it doesn't blow up the clp->cl_refcount. Throw a
5997                  * warning if we start to approach INT_MAX here.
5998                  */
5999                 WARN_ON_ONCE(count == (INT_MAX / 2));
6000                 if (count == max)
6001                         break;
6002         }
6003         spin_unlock(&clp->cl_lock);
6004
6005         return count;
6006 }
6007
6008 static u64
6009 nfsd_print_client_openowners(struct nfs4_client *clp)
6010 {
6011         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6012
6013         nfsd_print_count(clp, count, "openowners");
6014         return count;
6015 }
6016
6017 static u64
6018 nfsd_collect_client_openowners(struct nfs4_client *clp,
6019                                struct list_head *collect, u64 max)
6020 {
6021         return nfsd_foreach_client_openowner(clp, max, collect,
6022                                                 unhash_openowner_locked);
6023 }
6024
6025 u64
6026 nfsd_inject_print_openowners(void)
6027 {
6028         struct nfs4_client *clp;
6029         u64 count = 0;
6030         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6031                                                 nfsd_net_id);
6032
6033         if (!nfsd_netns_ready(nn))
6034                 return 0;
6035
6036         spin_lock(&nn->client_lock);
6037         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6038                 count += nfsd_print_client_openowners(clp);
6039         spin_unlock(&nn->client_lock);
6040
6041         return count;
6042 }
6043
6044 static void
6045 nfsd_reap_openowners(struct list_head *reaplist)
6046 {
6047         struct nfs4_client *clp;
6048         struct nfs4_openowner *oop, *next;
6049
6050         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6051                 list_del_init(&oop->oo_perclient);
6052                 clp = oop->oo_owner.so_client;
6053                 release_openowner(oop);
6054                 put_client(clp);
6055         }
6056 }
6057
6058 u64
6059 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6060                                      size_t addr_size)
6061 {
6062         unsigned int count = 0;
6063         struct nfs4_client *clp;
6064         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6065                                                 nfsd_net_id);
6066         LIST_HEAD(reaplist);
6067
6068         if (!nfsd_netns_ready(nn))
6069                 return count;
6070
6071         spin_lock(&nn->client_lock);
6072         clp = nfsd_find_client(addr, addr_size);
6073         if (clp)
6074                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6075         spin_unlock(&nn->client_lock);
6076         nfsd_reap_openowners(&reaplist);
6077         return count;
6078 }
6079
6080 u64
6081 nfsd_inject_forget_openowners(u64 max)
6082 {
6083         u64 count = 0;
6084         struct nfs4_client *clp;
6085         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6086                                                 nfsd_net_id);
6087         LIST_HEAD(reaplist);
6088
6089         if (!nfsd_netns_ready(nn))
6090                 return count;
6091
6092         spin_lock(&nn->client_lock);
6093         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6094                 count += nfsd_collect_client_openowners(clp, &reaplist,
6095                                                         max - count);
6096                 if (max != 0 && count >= max)
6097                         break;
6098         }
6099         spin_unlock(&nn->client_lock);
6100         nfsd_reap_openowners(&reaplist);
6101         return count;
6102 }
6103
6104 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6105                                      struct list_head *victims)
6106 {
6107         struct nfs4_delegation *dp, *next;
6108         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6109                                                 nfsd_net_id);
6110         u64 count = 0;
6111
6112         lockdep_assert_held(&nn->client_lock);
6113
6114         spin_lock(&state_lock);
6115         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6116                 if (victims) {
6117                         /*
6118                          * It's not safe to mess with delegations that have a
6119                          * non-zero dl_time. They might have already been broken
6120                          * and could be processed by the laundromat outside of
6121                          * the state_lock. Just leave them be.
6122                          */
6123                         if (dp->dl_time != 0)
6124                                 continue;
6125
6126                         atomic_inc(&clp->cl_refcount);
6127                         unhash_delegation_locked(dp);
6128                         list_add(&dp->dl_recall_lru, victims);
6129                 }
6130                 ++count;
6131                 /*
6132                  * Despite the fact that these functions deal with
6133                  * 64-bit integers for "count", we must ensure that
6134                  * it doesn't blow up the clp->cl_refcount. Throw a
6135                  * warning if we start to approach INT_MAX here.
6136                  */
6137                 WARN_ON_ONCE(count == (INT_MAX / 2));
6138                 if (count == max)
6139                         break;
6140         }
6141         spin_unlock(&state_lock);
6142         return count;
6143 }
6144
6145 static u64
6146 nfsd_print_client_delegations(struct nfs4_client *clp)
6147 {
6148         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6149
6150         nfsd_print_count(clp, count, "delegations");
6151         return count;
6152 }
6153
6154 u64
6155 nfsd_inject_print_delegations(void)
6156 {
6157         struct nfs4_client *clp;
6158         u64 count = 0;
6159         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6160                                                 nfsd_net_id);
6161
6162         if (!nfsd_netns_ready(nn))
6163                 return 0;
6164
6165         spin_lock(&nn->client_lock);
6166         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6167                 count += nfsd_print_client_delegations(clp);
6168         spin_unlock(&nn->client_lock);
6169
6170         return count;
6171 }
6172
6173 static void
6174 nfsd_forget_delegations(struct list_head *reaplist)
6175 {
6176         struct nfs4_client *clp;
6177         struct nfs4_delegation *dp, *next;
6178
6179         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6180                 list_del_init(&dp->dl_recall_lru);
6181                 clp = dp->dl_stid.sc_client;
6182                 revoke_delegation(dp);
6183                 put_client(clp);
6184         }
6185 }
6186
6187 u64
6188 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6189                                       size_t addr_size)
6190 {
6191         u64 count = 0;
6192         struct nfs4_client *clp;
6193         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6194                                                 nfsd_net_id);
6195         LIST_HEAD(reaplist);
6196
6197         if (!nfsd_netns_ready(nn))
6198                 return count;
6199
6200         spin_lock(&nn->client_lock);
6201         clp = nfsd_find_client(addr, addr_size);
6202         if (clp)
6203                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6204         spin_unlock(&nn->client_lock);
6205
6206         nfsd_forget_delegations(&reaplist);
6207         return count;
6208 }
6209
6210 u64
6211 nfsd_inject_forget_delegations(u64 max)
6212 {
6213         u64 count = 0;
6214         struct nfs4_client *clp;
6215         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6216                                                 nfsd_net_id);
6217         LIST_HEAD(reaplist);
6218
6219         if (!nfsd_netns_ready(nn))
6220                 return count;
6221
6222         spin_lock(&nn->client_lock);
6223         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6224                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6225                 if (max != 0 && count >= max)
6226                         break;
6227         }
6228         spin_unlock(&nn->client_lock);
6229         nfsd_forget_delegations(&reaplist);
6230         return count;
6231 }
6232
6233 static void
6234 nfsd_recall_delegations(struct list_head *reaplist)
6235 {
6236         struct nfs4_client *clp;
6237         struct nfs4_delegation *dp, *next;
6238
6239         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6240                 list_del_init(&dp->dl_recall_lru);
6241                 clp = dp->dl_stid.sc_client;
6242                 /*
6243                  * We skipped all entries that had a zero dl_time before,
6244                  * so we can now reset the dl_time back to 0. If a delegation
6245                  * break comes in now, then it won't make any difference since
6246                  * we're recalling it either way.
6247                  */
6248                 spin_lock(&state_lock);
6249                 dp->dl_time = 0;
6250                 spin_unlock(&state_lock);
6251                 nfsd_break_one_deleg(dp);
6252                 put_client(clp);
6253         }
6254 }
6255
6256 u64
6257 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6258                                       size_t addr_size)
6259 {
6260         u64 count = 0;
6261         struct nfs4_client *clp;
6262         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6263                                                 nfsd_net_id);
6264         LIST_HEAD(reaplist);
6265
6266         if (!nfsd_netns_ready(nn))
6267                 return count;
6268
6269         spin_lock(&nn->client_lock);
6270         clp = nfsd_find_client(addr, addr_size);
6271         if (clp)
6272                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6273         spin_unlock(&nn->client_lock);
6274
6275         nfsd_recall_delegations(&reaplist);
6276         return count;
6277 }
6278
6279 u64
6280 nfsd_inject_recall_delegations(u64 max)
6281 {
6282         u64 count = 0;
6283         struct nfs4_client *clp, *next;
6284         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6285                                                 nfsd_net_id);
6286         LIST_HEAD(reaplist);
6287
6288         if (!nfsd_netns_ready(nn))
6289                 return count;
6290
6291         spin_lock(&nn->client_lock);
6292         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6293                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6294                 if (max != 0 && ++count >= max)
6295                         break;
6296         }
6297         spin_unlock(&nn->client_lock);
6298         nfsd_recall_delegations(&reaplist);
6299         return count;
6300 }
6301 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6302
6303 /*
6304  * Since the lifetime of a delegation isn't limited to that of an open, a
6305  * client may quite reasonably hang on to a delegation as long as it has
6306  * the inode cached.  This becomes an obvious problem the first time a
6307  * client's inode cache approaches the size of the server's total memory.
6308  *
6309  * For now we avoid this problem by imposing a hard limit on the number
6310  * of delegations, which varies according to the server's memory size.
6311  */
6312 static void
6313 set_max_delegations(void)
6314 {
6315         /*
6316          * Allow at most 4 delegations per megabyte of RAM.  Quick
6317          * estimates suggest that in the worst case (where every delegation
6318          * is for a different inode), a delegation could take about 1.5K,
6319          * giving a worst case usage of about 6% of memory.
6320          */
6321         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6322 }
6323
6324 static int nfs4_state_create_net(struct net *net)
6325 {
6326         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6327         int i;
6328
6329         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6330                         CLIENT_HASH_SIZE, GFP_KERNEL);
6331         if (!nn->conf_id_hashtbl)
6332                 goto err;
6333         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6334                         CLIENT_HASH_SIZE, GFP_KERNEL);
6335         if (!nn->unconf_id_hashtbl)
6336                 goto err_unconf_id;
6337         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6338                         SESSION_HASH_SIZE, GFP_KERNEL);
6339         if (!nn->sessionid_hashtbl)
6340                 goto err_sessionid;
6341
6342         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6343                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6344                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6345         }
6346         for (i = 0; i < SESSION_HASH_SIZE; i++)
6347                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6348         nn->conf_name_tree = RB_ROOT;
6349         nn->unconf_name_tree = RB_ROOT;
6350         INIT_LIST_HEAD(&nn->client_lru);
6351         INIT_LIST_HEAD(&nn->close_lru);
6352         INIT_LIST_HEAD(&nn->del_recall_lru);
6353         spin_lock_init(&nn->client_lock);
6354
6355         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6356         get_net(net);
6357
6358         return 0;
6359
6360 err_sessionid:
6361         kfree(nn->unconf_id_hashtbl);
6362 err_unconf_id:
6363         kfree(nn->conf_id_hashtbl);
6364 err:
6365         return -ENOMEM;
6366 }
6367
6368 static void
6369 nfs4_state_destroy_net(struct net *net)
6370 {
6371         int i;
6372         struct nfs4_client *clp = NULL;
6373         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6374
6375         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6376                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6377                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6378                         destroy_client(clp);
6379                 }
6380         }
6381
6382         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6383                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6384                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6385                         destroy_client(clp);
6386                 }
6387         }
6388
6389         kfree(nn->sessionid_hashtbl);
6390         kfree(nn->unconf_id_hashtbl);
6391         kfree(nn->conf_id_hashtbl);
6392         put_net(net);
6393 }
6394
6395 int
6396 nfs4_state_start_net(struct net *net)
6397 {
6398         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6399         int ret;
6400
6401         ret = nfs4_state_create_net(net);
6402         if (ret)
6403                 return ret;
6404         nfsd4_client_tracking_init(net);
6405         nn->boot_time = get_seconds();
6406         locks_start_grace(net, &nn->nfsd4_manager);
6407         nn->grace_ended = false;
6408         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6409                nn->nfsd4_grace, net);
6410         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6411         return 0;
6412 }
6413
6414 /* initialization to perform when the nfsd service is started: */
6415
6416 int
6417 nfs4_state_start(void)
6418 {
6419         int ret;
6420
6421         ret = set_callback_cred();
6422         if (ret)
6423                 return -ENOMEM;
6424         laundry_wq = create_singlethread_workqueue("nfsd4");
6425         if (laundry_wq == NULL) {
6426                 ret = -ENOMEM;
6427                 goto out_recovery;
6428         }
6429         ret = nfsd4_create_callback_queue();
6430         if (ret)
6431                 goto out_free_laundry;
6432
6433         set_max_delegations();
6434
6435         return 0;
6436
6437 out_free_laundry:
6438         destroy_workqueue(laundry_wq);
6439 out_recovery:
6440         return ret;
6441 }
6442
6443 void
6444 nfs4_state_shutdown_net(struct net *net)
6445 {
6446         struct nfs4_delegation *dp = NULL;
6447         struct list_head *pos, *next, reaplist;
6448         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6449
6450         cancel_delayed_work_sync(&nn->laundromat_work);
6451         locks_end_grace(&nn->nfsd4_manager);
6452
6453         nfs4_lock_state();
6454         INIT_LIST_HEAD(&reaplist);
6455         spin_lock(&state_lock);
6456         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6457                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6458                 unhash_delegation_locked(dp);
6459                 list_add(&dp->dl_recall_lru, &reaplist);
6460         }
6461         spin_unlock(&state_lock);
6462         list_for_each_safe(pos, next, &reaplist) {
6463                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6464                 list_del_init(&dp->dl_recall_lru);
6465                 nfs4_put_stid(&dp->dl_stid);
6466         }
6467
6468         nfsd4_client_tracking_exit(net);
6469         nfs4_state_destroy_net(net);
6470         nfs4_unlock_state();
6471 }
6472
6473 void
6474 nfs4_state_shutdown(void)
6475 {
6476         destroy_workqueue(laundry_wq);
6477         nfsd4_destroy_callback_queue();
6478 }
6479
6480 static void
6481 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6482 {
6483         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6484                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6485 }
6486
6487 static void
6488 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6489 {
6490         if (cstate->minorversion) {
6491                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6492                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6493         }
6494 }
6495
6496 void
6497 clear_current_stateid(struct nfsd4_compound_state *cstate)
6498 {
6499         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6500 }
6501
6502 /*
6503  * functions to set current state id
6504  */
6505 void
6506 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6507 {
6508         put_stateid(cstate, &odp->od_stateid);
6509 }
6510
6511 void
6512 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6513 {
6514         put_stateid(cstate, &open->op_stateid);
6515 }
6516
6517 void
6518 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6519 {
6520         put_stateid(cstate, &close->cl_stateid);
6521 }
6522
6523 void
6524 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6525 {
6526         put_stateid(cstate, &lock->lk_resp_stateid);
6527 }
6528
6529 /*
6530  * functions to consume current state id
6531  */
6532
6533 void
6534 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6535 {
6536         get_stateid(cstate, &odp->od_stateid);
6537 }
6538
6539 void
6540 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6541 {
6542         get_stateid(cstate, &drp->dr_stateid);
6543 }
6544
6545 void
6546 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6547 {
6548         get_stateid(cstate, &fsp->fr_stateid);
6549 }
6550
6551 void
6552 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6553 {
6554         get_stateid(cstate, &setattr->sa_stateid);
6555 }
6556
6557 void
6558 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6559 {
6560         get_stateid(cstate, &close->cl_stateid);
6561 }
6562
6563 void
6564 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6565 {
6566         get_stateid(cstate, &locku->lu_stateid);
6567 }
6568
6569 void
6570 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6571 {
6572         get_stateid(cstate, &read->rd_stateid);
6573 }
6574
6575 void
6576 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6577 {
6578         get_stateid(cstate, &write->wr_stateid);
6579 }