fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include <linux/jhash.h>
  45 #include "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51 #include "pnfs.h"
  52
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 #define all_ones {{~0,~0},~0}
  56 static const stateid_t one_stateid = {
  57         .si_generation = ~0,
  58         .si_opaque = all_ones,
  59 };
  60 static const stateid_t zero_stateid = {
  61         /* all fields zero */
  62 };
  63 static const stateid_t currentstateid = {
  64         .si_generation = 1,
  65 };
  66
  67 static u64 current_sessionid = 1;
  68
  69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  70 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  72
  73 /* forward declarations */
  74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  76
  77 /* Locking: */
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(state_lock);
  85
  86 /*
  87  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  88  * the refcount on the open stateid to drop.
  89  */
  90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  91
  92 static struct kmem_cache *openowner_slab;
  93 static struct kmem_cache *lockowner_slab;
  94 static struct kmem_cache *file_slab;
  95 static struct kmem_cache *stateid_slab;
  96 static struct kmem_cache *deleg_slab;
  97
  98 static void free_session(struct nfsd4_session *);
  99
 100 static struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 101
 102 static bool is_session_dead(struct nfsd4_session *ses)
 103 {
 104         return ses->se_flags & NFS4_SESSION_DEAD;
 105 }
 106
 107 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 108 {
 109         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 110                 return nfserr_jukebox;
 111         ses->se_flags |= NFS4_SESSION_DEAD;
 112         return nfs_ok;
 113 }
 114
 115 static bool is_client_expired(struct nfs4_client *clp)
 116 {
 117         return clp->cl_time == 0;
 118 }
 119
 120 static __be32 get_client_locked(struct nfs4_client *clp)
 121 {
 122         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 123
 124         lockdep_assert_held(&nn->client_lock);
 125
 126         if (is_client_expired(clp))
 127                 return nfserr_expired;
 128         atomic_inc(&clp->cl_refcount);
 129         return nfs_ok;
 130 }
 131
 132 /* must be called under the client_lock */
 133 static inline void
 134 renew_client_locked(struct nfs4_client *clp)
 135 {
 136         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 137
 138         if (is_client_expired(clp)) {
 139                 WARN_ON(1);
 140                 printk("%s: client (clientid %08x/%08x) already expired\n",
 141                         __func__,
 142                         clp->cl_clientid.cl_boot,
 143                         clp->cl_clientid.cl_id);
 144                 return;
 145         }
 146
 147         dprintk("renewing client (clientid %08x/%08x)\n",
 148                         clp->cl_clientid.cl_boot,
 149                         clp->cl_clientid.cl_id);
 150         list_move_tail(&clp->cl_lru, &nn->client_lru);
 151         clp->cl_time = get_seconds();
 152 }
 153
 154 static void put_client_renew_locked(struct nfs4_client *clp)
 155 {
 156         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 157
 158         lockdep_assert_held(&nn->client_lock);
 159
 160         if (!atomic_dec_and_test(&clp->cl_refcount))
 161                 return;
 162         if (!is_client_expired(clp))
 163                 renew_client_locked(clp);
 164 }
 165
 166 static void put_client_renew(struct nfs4_client *clp)
 167 {
 168         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 169
 170         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 171                 return;
 172         if (!is_client_expired(clp))
 173                 renew_client_locked(clp);
 174         spin_unlock(&nn->client_lock);
 175 }
 176
 177 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 178 {
 179         __be32 status;
 180
 181         if (is_session_dead(ses))
 182                 return nfserr_badsession;
 183         status = get_client_locked(ses->se_client);
 184         if (status)
 185                 return status;
 186         atomic_inc(&ses->se_ref);
 187         return nfs_ok;
 188 }
 189
 190 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 191 {
 192         struct nfs4_client *clp = ses->se_client;
 193         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 194
 195         lockdep_assert_held(&nn->client_lock);
 196
 197         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 198                 free_session(ses);
 199         put_client_renew_locked(clp);
 200 }
 201
 202 static void nfsd4_put_session(struct nfsd4_session *ses)
 203 {
 204         struct nfs4_client *clp = ses->se_client;
 205         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 206
 207         spin_lock(&nn->client_lock);
 208         nfsd4_put_session_locked(ses);
 209         spin_unlock(&nn->client_lock);
 210 }
 211
 212 static inline struct nfs4_stateowner *
 213 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 214 {
 215         atomic_inc(&sop->so_count);
 216         return sop;
 217 }
 218
 219 static int
 220 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 221 {
 222         return (sop->so_owner.len == owner->len) &&
 223                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 224 }
 225
 226 static struct nfs4_openowner *
 227 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 228                         struct nfs4_client *clp)
 229 {
 230         struct nfs4_stateowner *so;
 231
 232         lockdep_assert_held(&clp->cl_lock);
 233
 234         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 235                             so_strhash) {
 236                 if (!so->so_is_open_owner)
 237                         continue;
 238                 if (same_owner_str(so, &open->op_owner))
 239                         return openowner(nfs4_get_stateowner(so));
 240         }
 241         return NULL;
 242 }
 243
 244 static struct nfs4_openowner *
 245 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 246                         struct nfs4_client *clp)
 247 {
 248         struct nfs4_openowner *oo;
 249
 250         spin_lock(&clp->cl_lock);
 251         oo = find_openstateowner_str_locked(hashval, open, clp);
 252         spin_unlock(&clp->cl_lock);
 253         return oo;
 254 }
 255
 256 static inline u32
 257 opaque_hashval(const void *ptr, int nbytes)
 258 {
 259         unsigned char *cptr = (unsigned char *) ptr;
 260
 261         u32 x = 0;
 262         while (nbytes--) {
 263                 x *= 37;
 264                 x += *cptr++;
 265         }
 266         return x;
 267 }
 268
 269 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 270 {
 271         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 272
 273         kmem_cache_free(file_slab, fp);
 274 }
 275
 276 void
 277 put_nfs4_file(struct nfs4_file *fi)
 278 {
 279         might_lock(&state_lock);
 280
 281         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 282                 hlist_del_rcu(&fi->fi_hash);
 283                 spin_unlock(&state_lock);
 284                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 285                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 286         }
 287 }
 288
 289 static struct file *
 290 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 291 {
 292         if (f->fi_fds[oflag])
 293                 return get_file(f->fi_fds[oflag]);
 294         return NULL;
 295 }
 296
 297 static struct file *
 298 find_writeable_file_locked(struct nfs4_file *f)
 299 {
 300         struct file *ret;
 301
 302         lockdep_assert_held(&f->fi_lock);
 303
 304         ret = __nfs4_get_fd(f, O_WRONLY);
 305         if (!ret)
 306                 ret = __nfs4_get_fd(f, O_RDWR);
 307         return ret;
 308 }
 309
 310 static struct file *
 311 find_writeable_file(struct nfs4_file *f)
 312 {
 313         struct file *ret;
 314
 315         spin_lock(&f->fi_lock);
 316         ret = find_writeable_file_locked(f);
 317         spin_unlock(&f->fi_lock);
 318
 319         return ret;
 320 }
 321
 322 static struct file *find_readable_file_locked(struct nfs4_file *f)
 323 {
 324         struct file *ret;
 325
 326         lockdep_assert_held(&f->fi_lock);
 327
 328         ret = __nfs4_get_fd(f, O_RDONLY);
 329         if (!ret)
 330                 ret = __nfs4_get_fd(f, O_RDWR);
 331         return ret;
 332 }
 333
 334 static struct file *
 335 find_readable_file(struct nfs4_file *f)
 336 {
 337         struct file *ret;
 338
 339         spin_lock(&f->fi_lock);
 340         ret = find_readable_file_locked(f);
 341         spin_unlock(&f->fi_lock);
 342
 343         return ret;
 344 }
 345
 346 struct file *
 347 find_any_file(struct nfs4_file *f)
 348 {
 349         struct file *ret;
 350
 351         spin_lock(&f->fi_lock);
 352         ret = __nfs4_get_fd(f, O_RDWR);
 353         if (!ret) {
 354                 ret = __nfs4_get_fd(f, O_WRONLY);
 355                 if (!ret)
 356                         ret = __nfs4_get_fd(f, O_RDONLY);
 357         }
 358         spin_unlock(&f->fi_lock);
 359         return ret;
 360 }
 361
 362 static atomic_long_t num_delegations;
 363 unsigned long max_delegations;
 364
 365 /*
 366  * Open owner state (share locks)
 367  */
 368
 369 /* hash tables for lock and open owners */
 370 #define OWNER_HASH_BITS              8
 371 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 372 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 373
 374 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 375 {
 376         unsigned int ret;
 377
 378         ret = opaque_hashval(ownername->data, ownername->len);
 379         return ret & OWNER_HASH_MASK;
 380 }
 381
 382 /* hash table for nfs4_file */
 383 #define FILE_HASH_BITS                   8
 384 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 385
 386 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 387 {
 388         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 389 }
 390
 391 static unsigned int file_hashval(struct knfsd_fh *fh)
 392 {
 393         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 394 }
 395
 396 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 397
 398 static void
 399 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 400 {
 401         lockdep_assert_held(&fp->fi_lock);
 402
 403         if (access & NFS4_SHARE_ACCESS_WRITE)
 404                 atomic_inc(&fp->fi_access[O_WRONLY]);
 405         if (access & NFS4_SHARE_ACCESS_READ)
 406                 atomic_inc(&fp->fi_access[O_RDONLY]);
 407 }
 408
 409 static __be32
 410 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 411 {
 412         lockdep_assert_held(&fp->fi_lock);
 413
 414         /* Does this access mode make sense? */
 415         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 416                 return nfserr_inval;
 417
 418         /* Does it conflict with a deny mode already set? */
 419         if ((access & fp->fi_share_deny) != 0)
 420                 return nfserr_share_denied;
 421
 422         __nfs4_file_get_access(fp, access);
 423         return nfs_ok;
 424 }
 425
 426 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 427 {
 428         /* Common case is that there is no deny mode. */
 429         if (deny) {
 430                 /* Does this deny mode make sense? */
 431                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 432                         return nfserr_inval;
 433
 434                 if ((deny & NFS4_SHARE_DENY_READ) &&
 435                     atomic_read(&fp->fi_access[O_RDONLY]))
 436                         return nfserr_share_denied;
 437
 438                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 439                     atomic_read(&fp->fi_access[O_WRONLY]))
 440                         return nfserr_share_denied;
 441         }
 442         return nfs_ok;
 443 }
 444
 445 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 446 {
 447         might_lock(&fp->fi_lock);
 448
 449         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 450                 struct file *f1 = NULL;
 451                 struct file *f2 = NULL;
 452
 453                 swap(f1, fp->fi_fds[oflag]);
 454                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 455                         swap(f2, fp->fi_fds[O_RDWR]);
 456                 spin_unlock(&fp->fi_lock);
 457                 if (f1)
 458                         fput(f1);
 459                 if (f2)
 460                         fput(f2);
 461         }
 462 }
 463
 464 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 465 {
 466         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 467
 468         if (access & NFS4_SHARE_ACCESS_WRITE)
 469                 __nfs4_file_put_access(fp, O_WRONLY);
 470         if (access & NFS4_SHARE_ACCESS_READ)
 471                 __nfs4_file_put_access(fp, O_RDONLY);
 472 }
 473
 474 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
 475                                          struct kmem_cache *slab)
 476 {
 477         struct nfs4_stid *stid;
 478         int new_id;
 479
 480         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 481         if (!stid)
 482                 return NULL;
 483
 484         idr_preload(GFP_KERNEL);
 485         spin_lock(&cl->cl_lock);
 486         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 487         spin_unlock(&cl->cl_lock);
 488         idr_preload_end();
 489         if (new_id < 0)
 490                 goto out_free;
 491         stid->sc_client = cl;
 492         stid->sc_stateid.si_opaque.so_id = new_id;
 493         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 494         /* Will be incremented before return to client: */
 495         atomic_set(&stid->sc_count, 1);
 496
 497         /*
 498          * It shouldn't be a problem to reuse an opaque stateid value.
 499          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 500          * example, a stray write retransmission could be accepted by
 501          * the server when it should have been rejected.  Therefore,
 502          * adopt a trick from the sctp code to attempt to maximize the
 503          * amount of time until an id is reused, by ensuring they always
 504          * "increase" (mod INT_MAX):
 505          */
 506         return stid;
 507 out_free:
 508         kmem_cache_free(slab, stid);
 509         return NULL;
 510 }
 511
 512 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 513 {
 514         struct nfs4_stid *stid;
 515         struct nfs4_ol_stateid *stp;
 516
 517         stid = nfs4_alloc_stid(clp, stateid_slab);
 518         if (!stid)
 519                 return NULL;
 520
 521         stp = openlockstateid(stid);
 522         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 523         return stp;
 524 }
 525
 526 static void nfs4_free_deleg(struct nfs4_stid *stid)
 527 {
 528         kmem_cache_free(deleg_slab, stid);
 529         atomic_long_dec(&num_delegations);
 530 }
 531
 532 /*
 533  * When we recall a delegation, we should be careful not to hand it
 534  * out again straight away.
 535  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 536  * in which the filehandles of recalled delegations are "stored".
 537  * If a filehandle appear in either filter, a delegation is blocked.
 538  * When a delegation is recalled, the filehandle is stored in the "new"
 539  * filter.
 540  * Every 30 seconds we swap the filters and clear the "new" one,
 541  * unless both are empty of course.
 542  *
 543  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 544  * low 3 bytes as hash-table indices.
 545  *
 546  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 547  * is used to manage concurrent access.  Testing does not need the lock
 548  * except when swapping the two filters.
 549  */
 550 static DEFINE_SPINLOCK(blocked_delegations_lock);
 551 static struct bloom_pair {
 552         int     entries, old_entries;
 553         time_t  swap_time;
 554         int     new; /* index into 'set' */
 555         DECLARE_BITMAP(set[2], 256);
 556 } blocked_delegations;
 557
 558 static int delegation_blocked(struct knfsd_fh *fh)
 559 {
 560         u32 hash;
 561         struct bloom_pair *bd = &blocked_delegations;
 562
 563         if (bd->entries == 0)
 564                 return 0;
 565         if (seconds_since_boot() - bd->swap_time > 30) {
 566                 spin_lock(&blocked_delegations_lock);
 567                 if (seconds_since_boot() - bd->swap_time > 30) {
 568                         bd->entries -= bd->old_entries;
 569                         bd->old_entries = bd->entries;
 570                         memset(bd->set[bd->new], 0,
 571                                sizeof(bd->set[0]));
 572                         bd->new = 1-bd->new;
 573                         bd->swap_time = seconds_since_boot();
 574                 }
 575                 spin_unlock(&blocked_delegations_lock);
 576         }
 577         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 578         if (test_bit(hash&255, bd->set[0]) &&
 579             test_bit((hash>>8)&255, bd->set[0]) &&
 580             test_bit((hash>>16)&255, bd->set[0]))
 581                 return 1;
 582
 583         if (test_bit(hash&255, bd->set[1]) &&
 584             test_bit((hash>>8)&255, bd->set[1]) &&
 585             test_bit((hash>>16)&255, bd->set[1]))
 586                 return 1;
 587
 588         return 0;
 589 }
 590
 591 static void block_delegations(struct knfsd_fh *fh)
 592 {
 593         u32 hash;
 594         struct bloom_pair *bd = &blocked_delegations;
 595
 596         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 597
 598         spin_lock(&blocked_delegations_lock);
 599         __set_bit(hash&255, bd->set[bd->new]);
 600         __set_bit((hash>>8)&255, bd->set[bd->new]);
 601         __set_bit((hash>>16)&255, bd->set[bd->new]);
 602         if (bd->entries == 0)
 603                 bd->swap_time = seconds_since_boot();
 604         bd->entries += 1;
 605         spin_unlock(&blocked_delegations_lock);
 606 }
 607
 608 static struct nfs4_delegation *
 609 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh)
 610 {
 611         struct nfs4_delegation *dp;
 612         long n;
 613
 614         dprintk("NFSD alloc_init_deleg\n");
 615         n = atomic_long_inc_return(&num_delegations);
 616         if (n < 0 || n > max_delegations)
 617                 goto out_dec;
 618         if (delegation_blocked(&current_fh->fh_handle))
 619                 goto out_dec;
 620         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 621         if (dp == NULL)
 622                 goto out_dec;
 623
 624         dp->dl_stid.sc_free = nfs4_free_deleg;
 625         /*
 626          * delegation seqid's are never incremented.  The 4.1 special
 627          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 628          * 0 anyway just for consistency and use 1:
 629          */
 630         dp->dl_stid.sc_stateid.si_generation = 1;
 631         INIT_LIST_HEAD(&dp->dl_perfile);
 632         INIT_LIST_HEAD(&dp->dl_perclnt);
 633         INIT_LIST_HEAD(&dp->dl_recall_lru);
 634         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 635         dp->dl_retries = 1;
 636         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 637                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 638         return dp;
 639 out_dec:
 640         atomic_long_dec(&num_delegations);
 641         return NULL;
 642 }
 643
 644 void
 645 nfs4_put_stid(struct nfs4_stid *s)
 646 {
 647         struct nfs4_file *fp = s->sc_file;
 648         struct nfs4_client *clp = s->sc_client;
 649
 650         might_lock(&clp->cl_lock);
 651
 652         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 653                 wake_up_all(&close_wq);
 654                 return;
 655         }
 656         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 657         spin_unlock(&clp->cl_lock);
 658         s->sc_free(s);
 659         if (fp)
 660                 put_nfs4_file(fp);
 661 }
 662
 663 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 664 {
 665         struct file *filp = NULL;
 666
 667         spin_lock(&fp->fi_lock);
 668         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 669                 swap(filp, fp->fi_deleg_file);
 670         spin_unlock(&fp->fi_lock);
 671
 672         if (filp) {
 673                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 674                 fput(filp);
 675         }
 676 }
 677
 678 void nfs4_unhash_stid(struct nfs4_stid *s)
 679 {
 680         s->sc_type = 0;
 681 }
 682
 683 static void
 684 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 685 {
 686         lockdep_assert_held(&state_lock);
 687         lockdep_assert_held(&fp->fi_lock);
 688
 689         atomic_inc(&dp->dl_stid.sc_count);
 690         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 691         list_add(&dp->dl_perfile, &fp->fi_delegations);
 692         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
 693 }
 694
 695 static void
 696 unhash_delegation_locked(struct nfs4_delegation *dp)
 697 {
 698         struct nfs4_file *fp = dp->dl_stid.sc_file;
 699
 700         lockdep_assert_held(&state_lock);
 701
 702         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 703         /* Ensure that deleg break won't try to requeue it */
 704         ++dp->dl_time;
 705         spin_lock(&fp->fi_lock);
 706         list_del_init(&dp->dl_perclnt);
 707         list_del_init(&dp->dl_recall_lru);
 708         list_del_init(&dp->dl_perfile);
 709         spin_unlock(&fp->fi_lock);
 710 }
 711
 712 static void destroy_delegation(struct nfs4_delegation *dp)
 713 {
 714         spin_lock(&state_lock);
 715         unhash_delegation_locked(dp);
 716         spin_unlock(&state_lock);
 717         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 718         nfs4_put_stid(&dp->dl_stid);
 719 }
 720
 721 static void revoke_delegation(struct nfs4_delegation *dp)
 722 {
 723         struct nfs4_client *clp = dp->dl_stid.sc_client;
 724
 725         WARN_ON(!list_empty(&dp->dl_recall_lru));
 726
 727         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 728
 729         if (clp->cl_minorversion == 0)
 730                 nfs4_put_stid(&dp->dl_stid);
 731         else {
 732                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 733                 spin_lock(&clp->cl_lock);
 734                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 735                 spin_unlock(&clp->cl_lock);
 736         }
 737 }
 738
 739 /*
 740  * SETCLIENTID state
 741  */
 742
 743 static unsigned int clientid_hashval(u32 id)
 744 {
 745         return id & CLIENT_HASH_MASK;
 746 }
 747
 748 static unsigned int clientstr_hashval(const char *name)
 749 {
 750         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 751 }
 752
 753 /*
 754  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 755  * st_{access,deny}_bmap field of the stateid, in order to track not
 756  * only what share bits are currently in force, but also what
 757  * combinations of share bits previous opens have used.  This allows us
 758  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 759  * return an error if the client attempt to downgrade to a combination
 760  * of share bits not explicable by closing some of its previous opens.
 761  *
 762  * XXX: This enforcement is actually incomplete, since we don't keep
 763  * track of access/deny bit combinations; so, e.g., we allow:
 764  *
 765  *      OPEN allow read, deny write
 766  *      OPEN allow both, deny none
 767  *      DOWNGRADE allow read, deny none
 768  *
 769  * which we should reject.
 770  */
 771 static unsigned int
 772 bmap_to_share_mode(unsigned long bmap) {
 773         int i;
 774         unsigned int access = 0;
 775
 776         for (i = 1; i < 4; i++) {
 777                 if (test_bit(i, &bmap))
 778                         access |= i;
 779         }
 780         return access;
 781 }
 782
 783 /* set share access for a given stateid */
 784 static inline void
 785 set_access(u32 access, struct nfs4_ol_stateid *stp)
 786 {
 787         unsigned char mask = 1 << access;
 788
 789         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 790         stp->st_access_bmap |= mask;
 791 }
 792
 793 /* clear share access for a given stateid */
 794 static inline void
 795 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 796 {
 797         unsigned char mask = 1 << access;
 798
 799         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 800         stp->st_access_bmap &= ~mask;
 801 }
 802
 803 /* test whether a given stateid has access */
 804 static inline bool
 805 test_access(u32 access, struct nfs4_ol_stateid *stp)
 806 {
 807         unsigned char mask = 1 << access;
 808
 809         return (bool)(stp->st_access_bmap & mask);
 810 }
 811
 812 /* set share deny for a given stateid */
 813 static inline void
 814 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 815 {
 816         unsigned char mask = 1 << deny;
 817
 818         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 819         stp->st_deny_bmap |= mask;
 820 }
 821
 822 /* clear share deny for a given stateid */
 823 static inline void
 824 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 825 {
 826         unsigned char mask = 1 << deny;
 827
 828         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 829         stp->st_deny_bmap &= ~mask;
 830 }
 831
 832 /* test whether a given stateid is denying specific access */
 833 static inline bool
 834 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 835 {
 836         unsigned char mask = 1 << deny;
 837
 838         return (bool)(stp->st_deny_bmap & mask);
 839 }
 840
 841 static int nfs4_access_to_omode(u32 access)
 842 {
 843         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 844         case NFS4_SHARE_ACCESS_READ:
 845                 return O_RDONLY;
 846         case NFS4_SHARE_ACCESS_WRITE:
 847                 return O_WRONLY;
 848         case NFS4_SHARE_ACCESS_BOTH:
 849                 return O_RDWR;
 850         }
 851         WARN_ON_ONCE(1);
 852         return O_RDONLY;
 853 }
 854
 855 /*
 856  * A stateid that had a deny mode associated with it is being released
 857  * or downgraded. Recalculate the deny mode on the file.
 858  */
 859 static void
 860 recalculate_deny_mode(struct nfs4_file *fp)
 861 {
 862         struct nfs4_ol_stateid *stp;
 863
 864         spin_lock(&fp->fi_lock);
 865         fp->fi_share_deny = 0;
 866         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
 867                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
 868         spin_unlock(&fp->fi_lock);
 869 }
 870
 871 static void
 872 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
 873 {
 874         int i;
 875         bool change = false;
 876
 877         for (i = 1; i < 4; i++) {
 878                 if ((i & deny) != i) {
 879                         change = true;
 880                         clear_deny(i, stp);
 881                 }
 882         }
 883
 884         /* Recalculate per-file deny mode if there was a change */
 885         if (change)
 886                 recalculate_deny_mode(stp->st_stid.sc_file);
 887 }
 888
 889 /* release all access and file references for a given stateid */
 890 static void
 891 release_all_access(struct nfs4_ol_stateid *stp)
 892 {
 893         int i;
 894         struct nfs4_file *fp = stp->st_stid.sc_file;
 895
 896         if (fp && stp->st_deny_bmap != 0)
 897                 recalculate_deny_mode(fp);
 898
 899         for (i = 1; i < 4; i++) {
 900                 if (test_access(i, stp))
 901                         nfs4_file_put_access(stp->st_stid.sc_file, i);
 902                 clear_access(i, stp);
 903         }
 904 }
 905
 906 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
 907 {
 908         struct nfs4_client *clp = sop->so_client;
 909
 910         might_lock(&clp->cl_lock);
 911
 912         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
 913                 return;
 914         sop->so_ops->so_unhash(sop);
 915         spin_unlock(&clp->cl_lock);
 916         kfree(sop->so_owner.data);
 917         sop->so_ops->so_free(sop);
 918 }
 919
 920 static void unhash_ol_stateid(struct nfs4_ol_stateid *stp)
 921 {
 922         struct nfs4_file *fp = stp->st_stid.sc_file;
 923
 924         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
 925
 926         spin_lock(&fp->fi_lock);
 927         list_del(&stp->st_perfile);
 928         spin_unlock(&fp->fi_lock);
 929         list_del(&stp->st_perstateowner);
 930 }
 931
 932 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
 933 {
 934         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 935
 936         release_all_access(stp);
 937         if (stp->st_stateowner)
 938                 nfs4_put_stateowner(stp->st_stateowner);
 939         kmem_cache_free(stateid_slab, stid);
 940 }
 941
 942 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
 943 {
 944         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 945         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
 946         struct file *file;
 947
 948         file = find_any_file(stp->st_stid.sc_file);
 949         if (file)
 950                 filp_close(file, (fl_owner_t)lo);
 951         nfs4_free_ol_stateid(stid);
 952 }
 953
 954 /*
 955  * Put the persistent reference to an already unhashed generic stateid, while
 956  * holding the cl_lock. If it's the last reference, then put it onto the
 957  * reaplist for later destruction.
 958  */
 959 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
 960                                        struct list_head *reaplist)
 961 {
 962         struct nfs4_stid *s = &stp->st_stid;
 963         struct nfs4_client *clp = s->sc_client;
 964
 965         lockdep_assert_held(&clp->cl_lock);
 966
 967         WARN_ON_ONCE(!list_empty(&stp->st_locks));
 968
 969         if (!atomic_dec_and_test(&s->sc_count)) {
 970                 wake_up_all(&close_wq);
 971                 return;
 972         }
 973
 974         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 975         list_add(&stp->st_locks, reaplist);
 976 }
 977
 978 static void unhash_lock_stateid(struct nfs4_ol_stateid *stp)
 979 {
 980         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
 981
 982         lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
 983
 984         list_del_init(&stp->st_locks);
 985         unhash_ol_stateid(stp);
 986         nfs4_unhash_stid(&stp->st_stid);
 987 }
 988
 989 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 990 {
 991         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
 992
 993         spin_lock(&oo->oo_owner.so_client->cl_lock);
 994         unhash_lock_stateid(stp);
 995         spin_unlock(&oo->oo_owner.so_client->cl_lock);
 996         nfs4_put_stid(&stp->st_stid);
 997 }
 998
 999 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1000 {
1001         struct nfs4_client *clp = lo->lo_owner.so_client;
1002
1003         lockdep_assert_held(&clp->cl_lock);
1004
1005         list_del_init(&lo->lo_owner.so_strhash);
1006 }
1007
1008 /*
1009  * Free a list of generic stateids that were collected earlier after being
1010  * fully unhashed.
1011  */
1012 static void
1013 free_ol_stateid_reaplist(struct list_head *reaplist)
1014 {
1015         struct nfs4_ol_stateid *stp;
1016         struct nfs4_file *fp;
1017
1018         might_sleep();
1019
1020         while (!list_empty(reaplist)) {
1021                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1022                                        st_locks);
1023                 list_del(&stp->st_locks);
1024                 fp = stp->st_stid.sc_file;
1025                 stp->st_stid.sc_free(&stp->st_stid);
1026                 if (fp)
1027                         put_nfs4_file(fp);
1028         }
1029 }
1030
1031 static void release_lockowner(struct nfs4_lockowner *lo)
1032 {
1033         struct nfs4_client *clp = lo->lo_owner.so_client;
1034         struct nfs4_ol_stateid *stp;
1035         struct list_head reaplist;
1036
1037         INIT_LIST_HEAD(&reaplist);
1038
1039         spin_lock(&clp->cl_lock);
1040         unhash_lockowner_locked(lo);
1041         while (!list_empty(&lo->lo_owner.so_stateids)) {
1042                 stp = list_first_entry(&lo->lo_owner.so_stateids,
1043                                 struct nfs4_ol_stateid, st_perstateowner);
1044                 unhash_lock_stateid(stp);
1045                 put_ol_stateid_locked(stp, &reaplist);
1046         }
1047         spin_unlock(&clp->cl_lock);
1048         free_ol_stateid_reaplist(&reaplist);
1049         nfs4_put_stateowner(&lo->lo_owner);
1050 }
1051
1052 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1053                                        struct list_head *reaplist)
1054 {
1055         struct nfs4_ol_stateid *stp;
1056
1057         while (!list_empty(&open_stp->st_locks)) {
1058                 stp = list_entry(open_stp->st_locks.next,
1059                                 struct nfs4_ol_stateid, st_locks);
1060                 unhash_lock_stateid(stp);
1061                 put_ol_stateid_locked(stp, reaplist);
1062         }
1063 }
1064
1065 static void unhash_open_stateid(struct nfs4_ol_stateid *stp,
1066                                 struct list_head *reaplist)
1067 {
1068         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1069
1070         unhash_ol_stateid(stp);
1071         release_open_stateid_locks(stp, reaplist);
1072 }
1073
1074 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1075 {
1076         LIST_HEAD(reaplist);
1077
1078         spin_lock(&stp->st_stid.sc_client->cl_lock);
1079         unhash_open_stateid(stp, &reaplist);
1080         put_ol_stateid_locked(stp, &reaplist);
1081         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1082         free_ol_stateid_reaplist(&reaplist);
1083 }
1084
1085 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1086 {
1087         struct nfs4_client *clp = oo->oo_owner.so_client;
1088
1089         lockdep_assert_held(&clp->cl_lock);
1090
1091         list_del_init(&oo->oo_owner.so_strhash);
1092         list_del_init(&oo->oo_perclient);
1093 }
1094
1095 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1096 {
1097         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1098                                           nfsd_net_id);
1099         struct nfs4_ol_stateid *s;
1100
1101         spin_lock(&nn->client_lock);
1102         s = oo->oo_last_closed_stid;
1103         if (s) {
1104                 list_del_init(&oo->oo_close_lru);
1105                 oo->oo_last_closed_stid = NULL;
1106         }
1107         spin_unlock(&nn->client_lock);
1108         if (s)
1109                 nfs4_put_stid(&s->st_stid);
1110 }
1111
1112 static void release_openowner(struct nfs4_openowner *oo)
1113 {
1114         struct nfs4_ol_stateid *stp;
1115         struct nfs4_client *clp = oo->oo_owner.so_client;
1116         struct list_head reaplist;
1117
1118         INIT_LIST_HEAD(&reaplist);
1119
1120         spin_lock(&clp->cl_lock);
1121         unhash_openowner_locked(oo);
1122         while (!list_empty(&oo->oo_owner.so_stateids)) {
1123                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1124                                 struct nfs4_ol_stateid, st_perstateowner);
1125                 unhash_open_stateid(stp, &reaplist);
1126                 put_ol_stateid_locked(stp, &reaplist);
1127         }
1128         spin_unlock(&clp->cl_lock);
1129         free_ol_stateid_reaplist(&reaplist);
1130         release_last_closed_stateid(oo);
1131         nfs4_put_stateowner(&oo->oo_owner);
1132 }
1133
1134 static inline int
1135 hash_sessionid(struct nfs4_sessionid *sessionid)
1136 {
1137         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1138
1139         return sid->sequence % SESSION_HASH_SIZE;
1140 }
1141
1142 #ifdef CONFIG_SUNRPC_DEBUG
1143 static inline void
1144 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1145 {
1146         u32 *ptr = (u32 *)(&sessionid->data[0]);
1147         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1148 }
1149 #else
1150 static inline void
1151 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1152 {
1153 }
1154 #endif
1155
1156 /*
1157  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1158  * won't be used for replay.
1159  */
1160 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1161 {
1162         struct nfs4_stateowner *so = cstate->replay_owner;
1163
1164         if (nfserr == nfserr_replay_me)
1165                 return;
1166
1167         if (!seqid_mutating_err(ntohl(nfserr))) {
1168                 nfsd4_cstate_clear_replay(cstate);
1169                 return;
1170         }
1171         if (!so)
1172                 return;
1173         if (so->so_is_open_owner)
1174                 release_last_closed_stateid(openowner(so));
1175         so->so_seqid++;
1176         return;
1177 }
1178
1179 static void
1180 gen_sessionid(struct nfsd4_session *ses)
1181 {
1182         struct nfs4_client *clp = ses->se_client;
1183         struct nfsd4_sessionid *sid;
1184
1185         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1186         sid->clientid = clp->cl_clientid;
1187         sid->sequence = current_sessionid++;
1188         sid->reserved = 0;
1189 }
1190
1191 /*
1192  * The protocol defines ca_maxresponssize_cached to include the size of
1193  * the rpc header, but all we need to cache is the data starting after
1194  * the end of the initial SEQUENCE operation--the rest we regenerate
1195  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1196  * value that is the number of bytes in our cache plus a few additional
1197  * bytes.  In order to stay on the safe side, and not promise more than
1198  * we can cache, those additional bytes must be the minimum possible: 24
1199  * bytes of rpc header (xid through accept state, with AUTH_NULL
1200  * verifier), 12 for the compound header (with zero-length tag), and 44
1201  * for the SEQUENCE op response:
1202  */
1203 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1204
1205 static void
1206 free_session_slots(struct nfsd4_session *ses)
1207 {
1208         int i;
1209
1210         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1211                 kfree(ses->se_slots[i]);
1212 }
1213
1214 /*
1215  * We don't actually need to cache the rpc and session headers, so we
1216  * can allocate a little less for each slot:
1217  */
1218 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1219 {
1220         u32 size;
1221
1222         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1223                 size = 0;
1224         else
1225                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1226         return size + sizeof(struct nfsd4_slot);
1227 }
1228
1229 /*
1230  * XXX: If we run out of reserved DRC memory we could (up to a point)
1231  * re-negotiate active sessions and reduce their slot usage to make
1232  * room for new connections. For now we just fail the create session.
1233  */
1234 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1235 {
1236         u32 slotsize = slot_bytes(ca);
1237         u32 num = ca->maxreqs;
1238         int avail;
1239
1240         spin_lock(&nfsd_drc_lock);
1241         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1242                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1243         num = min_t(int, num, avail / slotsize);
1244         nfsd_drc_mem_used += num * slotsize;
1245         spin_unlock(&nfsd_drc_lock);
1246
1247         return num;
1248 }
1249
1250 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1251 {
1252         int slotsize = slot_bytes(ca);
1253
1254         spin_lock(&nfsd_drc_lock);
1255         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1256         spin_unlock(&nfsd_drc_lock);
1257 }
1258
1259 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1260                                            struct nfsd4_channel_attrs *battrs)
1261 {
1262         int numslots = fattrs->maxreqs;
1263         int slotsize = slot_bytes(fattrs);
1264         struct nfsd4_session *new;
1265         int mem, i;
1266
1267         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1268                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1269         mem = numslots * sizeof(struct nfsd4_slot *);
1270
1271         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1272         if (!new)
1273                 return NULL;
1274         /* allocate each struct nfsd4_slot and data cache in one piece */
1275         for (i = 0; i < numslots; i++) {
1276                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1277                 if (!new->se_slots[i])
1278                         goto out_free;
1279         }
1280
1281         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1282         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1283
1284         return new;
1285 out_free:
1286         while (i--)
1287                 kfree(new->se_slots[i]);
1288         kfree(new);
1289         return NULL;
1290 }
1291
1292 static void free_conn(struct nfsd4_conn *c)
1293 {
1294         svc_xprt_put(c->cn_xprt);
1295         kfree(c);
1296 }
1297
1298 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1299 {
1300         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1301         struct nfs4_client *clp = c->cn_session->se_client;
1302
1303         spin_lock(&clp->cl_lock);
1304         if (!list_empty(&c->cn_persession)) {
1305                 list_del(&c->cn_persession);
1306                 free_conn(c);
1307         }
1308         nfsd4_probe_callback(clp);
1309         spin_unlock(&clp->cl_lock);
1310 }
1311
1312 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1313 {
1314         struct nfsd4_conn *conn;
1315
1316         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1317         if (!conn)
1318                 return NULL;
1319         svc_xprt_get(rqstp->rq_xprt);
1320         conn->cn_xprt = rqstp->rq_xprt;
1321         conn->cn_flags = flags;
1322         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1323         return conn;
1324 }
1325
1326 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1327 {
1328         conn->cn_session = ses;
1329         list_add(&conn->cn_persession, &ses->se_conns);
1330 }
1331
1332 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1333 {
1334         struct nfs4_client *clp = ses->se_client;
1335
1336         spin_lock(&clp->cl_lock);
1337         __nfsd4_hash_conn(conn, ses);
1338         spin_unlock(&clp->cl_lock);
1339 }
1340
1341 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1342 {
1343         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1344         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1345 }
1346
1347 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1348 {
1349         int ret;
1350
1351         nfsd4_hash_conn(conn, ses);
1352         ret = nfsd4_register_conn(conn);
1353         if (ret)
1354                 /* oops; xprt is already down: */
1355                 nfsd4_conn_lost(&conn->cn_xpt_user);
1356         /* We may have gained or lost a callback channel: */
1357         nfsd4_probe_callback_sync(ses->se_client);
1358 }
1359
1360 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1361 {
1362         u32 dir = NFS4_CDFC4_FORE;
1363
1364         if (cses->flags & SESSION4_BACK_CHAN)
1365                 dir |= NFS4_CDFC4_BACK;
1366         return alloc_conn(rqstp, dir);
1367 }
1368
1369 /* must be called under client_lock */
1370 static void nfsd4_del_conns(struct nfsd4_session *s)
1371 {
1372         struct nfs4_client *clp = s->se_client;
1373         struct nfsd4_conn *c;
1374
1375         spin_lock(&clp->cl_lock);
1376         while (!list_empty(&s->se_conns)) {
1377                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1378                 list_del_init(&c->cn_persession);
1379                 spin_unlock(&clp->cl_lock);
1380
1381                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1382                 free_conn(c);
1383
1384                 spin_lock(&clp->cl_lock);
1385         }
1386         spin_unlock(&clp->cl_lock);
1387 }
1388
1389 static void __free_session(struct nfsd4_session *ses)
1390 {
1391         free_session_slots(ses);
1392         kfree(ses);
1393 }
1394
1395 static void free_session(struct nfsd4_session *ses)
1396 {
1397         nfsd4_del_conns(ses);
1398         nfsd4_put_drc_mem(&ses->se_fchannel);
1399         __free_session(ses);
1400 }
1401
1402 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1403 {
1404         int idx;
1405         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1406
1407         new->se_client = clp;
1408         gen_sessionid(new);
1409
1410         INIT_LIST_HEAD(&new->se_conns);
1411
1412         new->se_cb_seq_nr = 1;
1413         new->se_flags = cses->flags;
1414         new->se_cb_prog = cses->callback_prog;
1415         new->se_cb_sec = cses->cb_sec;
1416         atomic_set(&new->se_ref, 0);
1417         idx = hash_sessionid(&new->se_sessionid);
1418         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1419         spin_lock(&clp->cl_lock);
1420         list_add(&new->se_perclnt, &clp->cl_sessions);
1421         spin_unlock(&clp->cl_lock);
1422
1423         {
1424                 struct sockaddr *sa = svc_addr(rqstp);
1425                 /*
1426                  * This is a little silly; with sessions there's no real
1427                  * use for the callback address.  Use the peer address
1428                  * as a reasonable default for now, but consider fixing
1429                  * the rpc client not to require an address in the
1430                  * future:
1431                  */
1432                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1433                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1434         }
1435 }
1436
1437 /* caller must hold client_lock */
1438 static struct nfsd4_session *
1439 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1440 {
1441         struct nfsd4_session *elem;
1442         int idx;
1443         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1444
1445         lockdep_assert_held(&nn->client_lock);
1446
1447         dump_sessionid(__func__, sessionid);
1448         idx = hash_sessionid(sessionid);
1449         /* Search in the appropriate list */
1450         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1451                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1452                             NFS4_MAX_SESSIONID_LEN)) {
1453                         return elem;
1454                 }
1455         }
1456
1457         dprintk("%s: session not found\n", __func__);
1458         return NULL;
1459 }
1460
1461 static struct nfsd4_session *
1462 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1463                 __be32 *ret)
1464 {
1465         struct nfsd4_session *session;
1466         __be32 status = nfserr_badsession;
1467
1468         session = __find_in_sessionid_hashtbl(sessionid, net);
1469         if (!session)
1470                 goto out;
1471         status = nfsd4_get_session_locked(session);
1472         if (status)
1473                 session = NULL;
1474 out:
1475         *ret = status;
1476         return session;
1477 }
1478
1479 /* caller must hold client_lock */
1480 static void
1481 unhash_session(struct nfsd4_session *ses)
1482 {
1483         struct nfs4_client *clp = ses->se_client;
1484         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1485
1486         lockdep_assert_held(&nn->client_lock);
1487
1488         list_del(&ses->se_hash);
1489         spin_lock(&ses->se_client->cl_lock);
1490         list_del(&ses->se_perclnt);
1491         spin_unlock(&ses->se_client->cl_lock);
1492 }
1493
1494 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1495 static int
1496 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1497 {
1498         /*
1499          * We're assuming the clid was not given out from a boot
1500          * precisely 2^32 (about 136 years) before this one.  That seems
1501          * a safe assumption:
1502          */
1503         if (clid->cl_boot == (u32)nn->boot_time)
1504                 return 0;
1505         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1506                 clid->cl_boot, clid->cl_id, nn->boot_time);
1507         return 1;
1508 }
1509
1510 /*
1511  * XXX Should we use a slab cache ?
1512  * This type of memory management is somewhat inefficient, but we use it
1513  * anyway since SETCLIENTID is not a common operation.
1514  */
1515 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1516 {
1517         struct nfs4_client *clp;
1518         int i;
1519
1520         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1521         if (clp == NULL)
1522                 return NULL;
1523         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1524         if (clp->cl_name.data == NULL)
1525                 goto err_no_name;
1526         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1527                         OWNER_HASH_SIZE, GFP_KERNEL);
1528         if (!clp->cl_ownerstr_hashtbl)
1529                 goto err_no_hashtbl;
1530         for (i = 0; i < OWNER_HASH_SIZE; i++)
1531                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1532         clp->cl_name.len = name.len;
1533         INIT_LIST_HEAD(&clp->cl_sessions);
1534         idr_init(&clp->cl_stateids);
1535         atomic_set(&clp->cl_refcount, 0);
1536         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1537         INIT_LIST_HEAD(&clp->cl_idhash);
1538         INIT_LIST_HEAD(&clp->cl_openowners);
1539         INIT_LIST_HEAD(&clp->cl_delegations);
1540         INIT_LIST_HEAD(&clp->cl_lru);
1541         INIT_LIST_HEAD(&clp->cl_callbacks);
1542         INIT_LIST_HEAD(&clp->cl_revoked);
1543 #ifdef CONFIG_NFSD_PNFS
1544         INIT_LIST_HEAD(&clp->cl_lo_states);
1545 #endif
1546         spin_lock_init(&clp->cl_lock);
1547         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1548         return clp;
1549 err_no_hashtbl:
1550         kfree(clp->cl_name.data);
1551 err_no_name:
1552         kfree(clp);
1553         return NULL;
1554 }
1555
1556 static void
1557 free_client(struct nfs4_client *clp)
1558 {
1559         while (!list_empty(&clp->cl_sessions)) {
1560                 struct nfsd4_session *ses;
1561                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1562                                 se_perclnt);
1563                 list_del(&ses->se_perclnt);
1564                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1565                 free_session(ses);
1566         }
1567         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1568         free_svc_cred(&clp->cl_cred);
1569         kfree(clp->cl_ownerstr_hashtbl);
1570         kfree(clp->cl_name.data);
1571         idr_destroy(&clp->cl_stateids);
1572         kfree(clp);
1573 }
1574
1575 /* must be called under the client_lock */
1576 static void
1577 unhash_client_locked(struct nfs4_client *clp)
1578 {
1579         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1580         struct nfsd4_session *ses;
1581
1582         lockdep_assert_held(&nn->client_lock);
1583
1584         /* Mark the client as expired! */
1585         clp->cl_time = 0;
1586         /* Make it invisible */
1587         if (!list_empty(&clp->cl_idhash)) {
1588                 list_del_init(&clp->cl_idhash);
1589                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1590                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1591                 else
1592                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1593         }
1594         list_del_init(&clp->cl_lru);
1595         spin_lock(&clp->cl_lock);
1596         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1597                 list_del_init(&ses->se_hash);
1598         spin_unlock(&clp->cl_lock);
1599 }
1600
1601 static void
1602 unhash_client(struct nfs4_client *clp)
1603 {
1604         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1605
1606         spin_lock(&nn->client_lock);
1607         unhash_client_locked(clp);
1608         spin_unlock(&nn->client_lock);
1609 }
1610
1611 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1612 {
1613         if (atomic_read(&clp->cl_refcount))
1614                 return nfserr_jukebox;
1615         unhash_client_locked(clp);
1616         return nfs_ok;
1617 }
1618
1619 static void
1620 __destroy_client(struct nfs4_client *clp)
1621 {
1622         struct nfs4_openowner *oo;
1623         struct nfs4_delegation *dp;
1624         struct list_head reaplist;
1625
1626         INIT_LIST_HEAD(&reaplist);
1627         spin_lock(&state_lock);
1628         while (!list_empty(&clp->cl_delegations)) {
1629                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1630                 unhash_delegation_locked(dp);
1631                 list_add(&dp->dl_recall_lru, &reaplist);
1632         }
1633         spin_unlock(&state_lock);
1634         while (!list_empty(&reaplist)) {
1635                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1636                 list_del_init(&dp->dl_recall_lru);
1637                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1638                 nfs4_put_stid(&dp->dl_stid);
1639         }
1640         while (!list_empty(&clp->cl_revoked)) {
1641                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1642                 list_del_init(&dp->dl_recall_lru);
1643                 nfs4_put_stid(&dp->dl_stid);
1644         }
1645         while (!list_empty(&clp->cl_openowners)) {
1646                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1647                 nfs4_get_stateowner(&oo->oo_owner);
1648                 release_openowner(oo);
1649         }
1650         nfsd4_return_all_client_layouts(clp);
1651         nfsd4_shutdown_callback(clp);
1652         if (clp->cl_cb_conn.cb_xprt)
1653                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1654         free_client(clp);
1655 }
1656
1657 static void
1658 destroy_client(struct nfs4_client *clp)
1659 {
1660         unhash_client(clp);
1661         __destroy_client(clp);
1662 }
1663
1664 static void expire_client(struct nfs4_client *clp)
1665 {
1666         unhash_client(clp);
1667         nfsd4_client_record_remove(clp);
1668         __destroy_client(clp);
1669 }
1670
1671 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1672 {
1673         memcpy(target->cl_verifier.data, source->data,
1674                         sizeof(target->cl_verifier.data));
1675 }
1676
1677 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1678 {
1679         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1680         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1681 }
1682
1683 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1684 {
1685         if (source->cr_principal) {
1686                 target->cr_principal =
1687                                 kstrdup(source->cr_principal, GFP_KERNEL);
1688                 if (target->cr_principal == NULL)
1689                         return -ENOMEM;
1690         } else
1691                 target->cr_principal = NULL;
1692         target->cr_flavor = source->cr_flavor;
1693         target->cr_uid = source->cr_uid;
1694         target->cr_gid = source->cr_gid;
1695         target->cr_group_info = source->cr_group_info;
1696         get_group_info(target->cr_group_info);
1697         target->cr_gss_mech = source->cr_gss_mech;
1698         if (source->cr_gss_mech)
1699                 gss_mech_get(source->cr_gss_mech);
1700         return 0;
1701 }
1702
1703 static int
1704 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1705 {
1706         if (o1->len < o2->len)
1707                 return -1;
1708         if (o1->len > o2->len)
1709                 return 1;
1710         return memcmp(o1->data, o2->data, o1->len);
1711 }
1712
1713 static int same_name(const char *n1, const char *n2)
1714 {
1715         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1716 }
1717
1718 static int
1719 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1720 {
1721         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1722 }
1723
1724 static int
1725 same_clid(clientid_t *cl1, clientid_t *cl2)
1726 {
1727         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1728 }
1729
1730 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1731 {
1732         int i;
1733
1734         if (g1->ngroups != g2->ngroups)
1735                 return false;
1736         for (i=0; i<g1->ngroups; i++)
1737                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1738                         return false;
1739         return true;
1740 }
1741
1742 /*
1743  * RFC 3530 language requires clid_inuse be returned when the
1744  * "principal" associated with a requests differs from that previously
1745  * used.  We use uid, gid's, and gss principal string as our best
1746  * approximation.  We also don't want to allow non-gss use of a client
1747  * established using gss: in theory cr_principal should catch that
1748  * change, but in practice cr_principal can be null even in the gss case
1749  * since gssd doesn't always pass down a principal string.
1750  */
1751 static bool is_gss_cred(struct svc_cred *cr)
1752 {
1753         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1754         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1755 }
1756
1757
1758 static bool
1759 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1760 {
1761         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1762                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1763                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1764                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1765                 return false;
1766         if (cr1->cr_principal == cr2->cr_principal)
1767                 return true;
1768         if (!cr1->cr_principal || !cr2->cr_principal)
1769                 return false;
1770         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1771 }
1772
1773 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1774 {
1775         struct svc_cred *cr = &rqstp->rq_cred;
1776         u32 service;
1777
1778         if (!cr->cr_gss_mech)
1779                 return false;
1780         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1781         return service == RPC_GSS_SVC_INTEGRITY ||
1782                service == RPC_GSS_SVC_PRIVACY;
1783 }
1784
1785 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1786 {
1787         struct svc_cred *cr = &rqstp->rq_cred;
1788
1789         if (!cl->cl_mach_cred)
1790                 return true;
1791         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1792                 return false;
1793         if (!svc_rqst_integrity_protected(rqstp))
1794                 return false;
1795         if (!cr->cr_principal)
1796                 return false;
1797         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1798 }
1799
1800 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1801 {
1802         __be32 verf[2];
1803
1804         /*
1805          * This is opaque to client, so no need to byte-swap. Use
1806          * __force to keep sparse happy
1807          */
1808         verf[0] = (__force __be32)get_seconds();
1809         verf[1] = (__force __be32)nn->clientid_counter;
1810         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1811 }
1812
1813 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1814 {
1815         clp->cl_clientid.cl_boot = nn->boot_time;
1816         clp->cl_clientid.cl_id = nn->clientid_counter++;
1817         gen_confirm(clp, nn);
1818 }
1819
1820 static struct nfs4_stid *
1821 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
1822 {
1823         struct nfs4_stid *ret;
1824
1825         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1826         if (!ret || !ret->sc_type)
1827                 return NULL;
1828         return ret;
1829 }
1830
1831 static struct nfs4_stid *
1832 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1833 {
1834         struct nfs4_stid *s;
1835
1836         spin_lock(&cl->cl_lock);
1837         s = find_stateid_locked(cl, t);
1838         if (s != NULL) {
1839                 if (typemask & s->sc_type)
1840                         atomic_inc(&s->sc_count);
1841                 else
1842                         s = NULL;
1843         }
1844         spin_unlock(&cl->cl_lock);
1845         return s;
1846 }
1847
1848 static struct nfs4_client *create_client(struct xdr_netobj name,
1849                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1850 {
1851         struct nfs4_client *clp;
1852         struct sockaddr *sa = svc_addr(rqstp);
1853         int ret;
1854         struct net *net = SVC_NET(rqstp);
1855
1856         clp = alloc_client(name);
1857         if (clp == NULL)
1858                 return NULL;
1859
1860         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1861         if (ret) {
1862                 free_client(clp);
1863                 return NULL;
1864         }
1865         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
1866         clp->cl_time = get_seconds();
1867         clear_bit(0, &clp->cl_cb_slot_busy);
1868         copy_verf(clp, verf);
1869         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1870         clp->cl_cb_session = NULL;
1871         clp->net = net;
1872         return clp;
1873 }
1874
1875 static void
1876 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1877 {
1878         struct rb_node **new = &(root->rb_node), *parent = NULL;
1879         struct nfs4_client *clp;
1880
1881         while (*new) {
1882                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1883                 parent = *new;
1884
1885                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1886                         new = &((*new)->rb_left);
1887                 else
1888                         new = &((*new)->rb_right);
1889         }
1890
1891         rb_link_node(&new_clp->cl_namenode, parent, new);
1892         rb_insert_color(&new_clp->cl_namenode, root);
1893 }
1894
1895 static struct nfs4_client *
1896 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1897 {
1898         int cmp;
1899         struct rb_node *node = root->rb_node;
1900         struct nfs4_client *clp;
1901
1902         while (node) {
1903                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1904                 cmp = compare_blob(&clp->cl_name, name);
1905                 if (cmp > 0)
1906                         node = node->rb_left;
1907                 else if (cmp < 0)
1908                         node = node->rb_right;
1909                 else
1910                         return clp;
1911         }
1912         return NULL;
1913 }
1914
1915 static void
1916 add_to_unconfirmed(struct nfs4_client *clp)
1917 {
1918         unsigned int idhashval;
1919         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1920
1921         lockdep_assert_held(&nn->client_lock);
1922
1923         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1924         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1925         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1926         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1927         renew_client_locked(clp);
1928 }
1929
1930 static void
1931 move_to_confirmed(struct nfs4_client *clp)
1932 {
1933         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1934         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1935
1936         lockdep_assert_held(&nn->client_lock);
1937
1938         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1939         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1940         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1941         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1942         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1943         renew_client_locked(clp);
1944 }
1945
1946 static struct nfs4_client *
1947 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1948 {
1949         struct nfs4_client *clp;
1950         unsigned int idhashval = clientid_hashval(clid->cl_id);
1951
1952         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1953                 if (same_clid(&clp->cl_clientid, clid)) {
1954                         if ((bool)clp->cl_minorversion != sessions)
1955                                 return NULL;
1956                         renew_client_locked(clp);
1957                         return clp;
1958                 }
1959         }
1960         return NULL;
1961 }
1962
1963 static struct nfs4_client *
1964 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1965 {
1966         struct list_head *tbl = nn->conf_id_hashtbl;
1967
1968         lockdep_assert_held(&nn->client_lock);
1969         return find_client_in_id_table(tbl, clid, sessions);
1970 }
1971
1972 static struct nfs4_client *
1973 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1974 {
1975         struct list_head *tbl = nn->unconf_id_hashtbl;
1976
1977         lockdep_assert_held(&nn->client_lock);
1978         return find_client_in_id_table(tbl, clid, sessions);
1979 }
1980
1981 static bool clp_used_exchangeid(struct nfs4_client *clp)
1982 {
1983         return clp->cl_exchange_flags != 0;
1984 }
1985
1986 static struct nfs4_client *
1987 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1988 {
1989         lockdep_assert_held(&nn->client_lock);
1990         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1991 }
1992
1993 static struct nfs4_client *
1994 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1995 {
1996         lockdep_assert_held(&nn->client_lock);
1997         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1998 }
1999
2000 static void
2001 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2002 {
2003         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2004         struct sockaddr *sa = svc_addr(rqstp);
2005         u32 scopeid = rpc_get_scope_id(sa);
2006         unsigned short expected_family;
2007
2008         /* Currently, we only support tcp and tcp6 for the callback channel */
2009         if (se->se_callback_netid_len == 3 &&
2010             !memcmp(se->se_callback_netid_val, "tcp", 3))
2011                 expected_family = AF_INET;
2012         else if (se->se_callback_netid_len == 4 &&
2013                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2014                 expected_family = AF_INET6;
2015         else
2016                 goto out_err;
2017
2018         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2019                                             se->se_callback_addr_len,
2020                                             (struct sockaddr *)&conn->cb_addr,
2021                                             sizeof(conn->cb_addr));
2022
2023         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2024                 goto out_err;
2025
2026         if (conn->cb_addr.ss_family == AF_INET6)
2027                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2028
2029         conn->cb_prog = se->se_callback_prog;
2030         conn->cb_ident = se->se_callback_ident;
2031         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2032         return;
2033 out_err:
2034         conn->cb_addr.ss_family = AF_UNSPEC;
2035         conn->cb_addrlen = 0;
2036         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2037                 "will not receive delegations\n",
2038                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2039
2040         return;
2041 }
2042
2043 /*
2044  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2045  */
2046 static void
2047 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2048 {
2049         struct xdr_buf *buf = resp->xdr.buf;
2050         struct nfsd4_slot *slot = resp->cstate.slot;
2051         unsigned int base;
2052
2053         dprintk("--> %s slot %p\n", __func__, slot);
2054
2055         slot->sl_opcnt = resp->opcnt;
2056         slot->sl_status = resp->cstate.status;
2057
2058         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2059         if (nfsd4_not_cached(resp)) {
2060                 slot->sl_datalen = 0;
2061                 return;
2062         }
2063         base = resp->cstate.data_offset;
2064         slot->sl_datalen = buf->len - base;
2065         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2066                 WARN("%s: sessions DRC could not cache compound\n", __func__);
2067         return;
2068 }
2069
2070 /*
2071  * Encode the replay sequence operation from the slot values.
2072  * If cachethis is FALSE encode the uncached rep error on the next
2073  * operation which sets resp->p and increments resp->opcnt for
2074  * nfs4svc_encode_compoundres.
2075  *
2076  */
2077 static __be32
2078 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2079                           struct nfsd4_compoundres *resp)
2080 {
2081         struct nfsd4_op *op;
2082         struct nfsd4_slot *slot = resp->cstate.slot;
2083
2084         /* Encode the replayed sequence operation */
2085         op = &args->ops[resp->opcnt - 1];
2086         nfsd4_encode_operation(resp, op);
2087
2088         /* Return nfserr_retry_uncached_rep in next operation. */
2089         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2090                 op = &args->ops[resp->opcnt++];
2091                 op->status = nfserr_retry_uncached_rep;
2092                 nfsd4_encode_operation(resp, op);
2093         }
2094         return op->status;
2095 }
2096
2097 /*
2098  * The sequence operation is not cached because we can use the slot and
2099  * session values.
2100  */
2101 static __be32
2102 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2103                          struct nfsd4_sequence *seq)
2104 {
2105         struct nfsd4_slot *slot = resp->cstate.slot;
2106         struct xdr_stream *xdr = &resp->xdr;
2107         __be32 *p;
2108         __be32 status;
2109
2110         dprintk("--> %s slot %p\n", __func__, slot);
2111
2112         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2113         if (status)
2114                 return status;
2115
2116         p = xdr_reserve_space(xdr, slot->sl_datalen);
2117         if (!p) {
2118                 WARN_ON_ONCE(1);
2119                 return nfserr_serverfault;
2120         }
2121         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2122         xdr_commit_encode(xdr);
2123
2124         resp->opcnt = slot->sl_opcnt;
2125         return slot->sl_status;
2126 }
2127
2128 /*
2129  * Set the exchange_id flags returned by the server.
2130  */
2131 static void
2132 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2133 {
2134 #ifdef CONFIG_NFSD_PNFS
2135         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2136 #else
2137         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2138 #endif
2139
2140         /* Referrals are supported, Migration is not. */
2141         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2142
2143         /* set the wire flags to return to client. */
2144         clid->flags = new->cl_exchange_flags;
2145 }
2146
2147 static bool client_has_state(struct nfs4_client *clp)
2148 {
2149         /*
2150          * Note clp->cl_openowners check isn't quite right: there's no
2151          * need to count owners without stateid's.
2152          *
2153          * Also note we should probably be using this in 4.0 case too.
2154          */
2155         return !list_empty(&clp->cl_openowners)
2156                 || !list_empty(&clp->cl_delegations)
2157                 || !list_empty(&clp->cl_sessions);
2158 }
2159
2160 __be32
2161 nfsd4_exchange_id(struct svc_rqst *rqstp,
2162                   struct nfsd4_compound_state *cstate,
2163                   struct nfsd4_exchange_id *exid)
2164 {
2165         struct nfs4_client *conf, *new;
2166         struct nfs4_client *unconf = NULL;
2167         __be32 status;
2168         char                    addr_str[INET6_ADDRSTRLEN];
2169         nfs4_verifier           verf = exid->verifier;
2170         struct sockaddr         *sa = svc_addr(rqstp);
2171         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2172         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2173
2174         rpc_ntop(sa, addr_str, sizeof(addr_str));
2175         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2176                 "ip_addr=%s flags %x, spa_how %d\n",
2177                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2178                 addr_str, exid->flags, exid->spa_how);
2179
2180         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2181                 return nfserr_inval;
2182
2183         switch (exid->spa_how) {
2184         case SP4_MACH_CRED:
2185                 if (!svc_rqst_integrity_protected(rqstp))
2186                         return nfserr_inval;
2187         case SP4_NONE:
2188                 break;
2189         default:                                /* checked by xdr code */
2190                 WARN_ON_ONCE(1);
2191         case SP4_SSV:
2192                 return nfserr_encr_alg_unsupp;
2193         }
2194
2195         new = create_client(exid->clname, rqstp, &verf);
2196         if (new == NULL)
2197                 return nfserr_jukebox;
2198
2199         /* Cases below refer to rfc 5661 section 18.35.4: */
2200         spin_lock(&nn->client_lock);
2201         conf = find_confirmed_client_by_name(&exid->clname, nn);
2202         if (conf) {
2203                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2204                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2205
2206                 if (update) {
2207                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2208                                 status = nfserr_inval;
2209                                 goto out;
2210                         }
2211                         if (!mach_creds_match(conf, rqstp)) {
2212                                 status = nfserr_wrong_cred;
2213                                 goto out;
2214                         }
2215                         if (!creds_match) { /* case 9 */
2216                                 status = nfserr_perm;
2217                                 goto out;
2218                         }
2219                         if (!verfs_match) { /* case 8 */
2220                                 status = nfserr_not_same;
2221                                 goto out;
2222                         }
2223                         /* case 6 */
2224                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2225                         goto out_copy;
2226                 }
2227                 if (!creds_match) { /* case 3 */
2228                         if (client_has_state(conf)) {
2229                                 status = nfserr_clid_inuse;
2230                                 goto out;
2231                         }
2232                         goto out_new;
2233                 }
2234                 if (verfs_match) { /* case 2 */
2235                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2236                         goto out_copy;
2237                 }
2238                 /* case 5, client reboot */
2239                 conf = NULL;
2240                 goto out_new;
2241         }
2242
2243         if (update) { /* case 7 */
2244                 status = nfserr_noent;
2245                 goto out;
2246         }
2247
2248         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2249         if (unconf) /* case 4, possible retry or client restart */
2250                 unhash_client_locked(unconf);
2251
2252         /* case 1 (normal case) */
2253 out_new:
2254         if (conf) {
2255                 status = mark_client_expired_locked(conf);
2256                 if (status)
2257                         goto out;
2258         }
2259         new->cl_minorversion = cstate->minorversion;
2260         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
2261
2262         gen_clid(new, nn);
2263         add_to_unconfirmed(new);
2264         swap(new, conf);
2265 out_copy:
2266         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2267         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2268
2269         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2270         nfsd4_set_ex_flags(conf, exid);
2271
2272         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2273                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2274         status = nfs_ok;
2275
2276 out:
2277         spin_unlock(&nn->client_lock);
2278         if (new)
2279                 expire_client(new);
2280         if (unconf)
2281                 expire_client(unconf);
2282         return status;
2283 }
2284
2285 static __be32
2286 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2287 {
2288         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2289                 slot_seqid);
2290
2291         /* The slot is in use, and no response has been sent. */
2292         if (slot_inuse) {
2293                 if (seqid == slot_seqid)
2294                         return nfserr_jukebox;
2295                 else
2296                         return nfserr_seq_misordered;
2297         }
2298         /* Note unsigned 32-bit arithmetic handles wraparound: */
2299         if (likely(seqid == slot_seqid + 1))
2300                 return nfs_ok;
2301         if (seqid == slot_seqid)
2302                 return nfserr_replay_cache;
2303         return nfserr_seq_misordered;
2304 }
2305
2306 /*
2307  * Cache the create session result into the create session single DRC
2308  * slot cache by saving the xdr structure. sl_seqid has been set.
2309  * Do this for solo or embedded create session operations.
2310  */
2311 static void
2312 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2313                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2314 {
2315         slot->sl_status = nfserr;
2316         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2317 }
2318
2319 static __be32
2320 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2321                             struct nfsd4_clid_slot *slot)
2322 {
2323         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2324         return slot->sl_status;
2325 }
2326
2327 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2328                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2329                         1 +     /* MIN tag is length with zero, only length */ \
2330                         3 +     /* version, opcount, opcode */ \
2331                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2332                                 /* seqid, slotID, slotID, cache */ \
2333                         4 ) * sizeof(__be32))
2334
2335 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2336                         2 +     /* verifier: AUTH_NULL, length 0 */\
2337                         1 +     /* status */ \
2338                         1 +     /* MIN tag is length with zero, only length */ \
2339                         3 +     /* opcount, opcode, opstatus*/ \
2340                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2341                                 /* seqid, slotID, slotID, slotID, status */ \
2342                         5 ) * sizeof(__be32))
2343
2344 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2345 {
2346         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2347
2348         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2349                 return nfserr_toosmall;
2350         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2351                 return nfserr_toosmall;
2352         ca->headerpadsz = 0;
2353         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2354         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2355         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2356         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2357                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2358         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2359         /*
2360          * Note decreasing slot size below client's request may make it
2361          * difficult for client to function correctly, whereas
2362          * decreasing the number of slots will (just?) affect
2363          * performance.  When short on memory we therefore prefer to
2364          * decrease number of slots instead of their size.  Clients that
2365          * request larger slots than they need will get poor results:
2366          */
2367         ca->maxreqs = nfsd4_get_drc_mem(ca);
2368         if (!ca->maxreqs)
2369                 return nfserr_jukebox;
2370
2371         return nfs_ok;
2372 }
2373
2374 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2375                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2376 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2377                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2378
2379 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2380 {
2381         ca->headerpadsz = 0;
2382
2383         /*
2384          * These RPC_MAX_HEADER macros are overkill, especially since we
2385          * don't even do gss on the backchannel yet.  But this is still
2386          * less than 1k.  Tighten up this estimate in the unlikely event
2387          * it turns out to be a problem for some client:
2388          */
2389         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2390                 return nfserr_toosmall;
2391         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2392                 return nfserr_toosmall;
2393         ca->maxresp_cached = 0;
2394         if (ca->maxops < 2)
2395                 return nfserr_toosmall;
2396
2397         return nfs_ok;
2398 }
2399
2400 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2401 {
2402         switch (cbs->flavor) {
2403         case RPC_AUTH_NULL:
2404         case RPC_AUTH_UNIX:
2405                 return nfs_ok;
2406         default:
2407                 /*
2408                  * GSS case: the spec doesn't allow us to return this
2409                  * error.  But it also doesn't allow us not to support
2410                  * GSS.
2411                  * I'd rather this fail hard than return some error the
2412                  * client might think it can already handle:
2413                  */
2414                 return nfserr_encr_alg_unsupp;
2415         }
2416 }
2417
2418 __be32
2419 nfsd4_create_session(struct svc_rqst *rqstp,
2420                      struct nfsd4_compound_state *cstate,
2421                      struct nfsd4_create_session *cr_ses)
2422 {
2423         struct sockaddr *sa = svc_addr(rqstp);
2424         struct nfs4_client *conf, *unconf;
2425         struct nfs4_client *old = NULL;
2426         struct nfsd4_session *new;
2427         struct nfsd4_conn *conn;
2428         struct nfsd4_clid_slot *cs_slot = NULL;
2429         __be32 status = 0;
2430         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2431
2432         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2433                 return nfserr_inval;
2434         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2435         if (status)
2436                 return status;
2437         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2438         if (status)
2439                 return status;
2440         status = check_backchannel_attrs(&cr_ses->back_channel);
2441         if (status)
2442                 goto out_release_drc_mem;
2443         status = nfserr_jukebox;
2444         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2445         if (!new)
2446                 goto out_release_drc_mem;
2447         conn = alloc_conn_from_crses(rqstp, cr_ses);
2448         if (!conn)
2449                 goto out_free_session;
2450
2451         spin_lock(&nn->client_lock);
2452         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2453         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2454         WARN_ON_ONCE(conf && unconf);
2455
2456         if (conf) {
2457                 status = nfserr_wrong_cred;
2458                 if (!mach_creds_match(conf, rqstp))
2459                         goto out_free_conn;
2460                 cs_slot = &conf->cl_cs_slot;
2461                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2462                 if (status == nfserr_replay_cache) {
2463                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
2464                         goto out_free_conn;
2465                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
2466                         status = nfserr_seq_misordered;
2467                         goto out_free_conn;
2468                 }
2469         } else if (unconf) {
2470                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2471                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2472                         status = nfserr_clid_inuse;
2473                         goto out_free_conn;
2474                 }
2475                 status = nfserr_wrong_cred;
2476                 if (!mach_creds_match(unconf, rqstp))
2477                         goto out_free_conn;
2478                 cs_slot = &unconf->cl_cs_slot;
2479                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2480                 if (status) {
2481                         /* an unconfirmed replay returns misordered */
2482                         status = nfserr_seq_misordered;
2483                         goto out_free_conn;
2484                 }
2485                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2486                 if (old) {
2487                         status = mark_client_expired_locked(old);
2488                         if (status) {
2489                                 old = NULL;
2490                                 goto out_free_conn;
2491                         }
2492                 }
2493                 move_to_confirmed(unconf);
2494                 conf = unconf;
2495         } else {
2496                 status = nfserr_stale_clientid;
2497                 goto out_free_conn;
2498         }
2499         status = nfs_ok;
2500         /*
2501          * We do not support RDMA or persistent sessions
2502          */
2503         cr_ses->flags &= ~SESSION4_PERSIST;
2504         cr_ses->flags &= ~SESSION4_RDMA;
2505
2506         init_session(rqstp, new, conf, cr_ses);
2507         nfsd4_get_session_locked(new);
2508
2509         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2510                NFS4_MAX_SESSIONID_LEN);
2511         cs_slot->sl_seqid++;
2512         cr_ses->seqid = cs_slot->sl_seqid;
2513
2514         /* cache solo and embedded create sessions under the client_lock */
2515         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2516         spin_unlock(&nn->client_lock);
2517         /* init connection and backchannel */
2518         nfsd4_init_conn(rqstp, conn, new);
2519         nfsd4_put_session(new);
2520         if (old)
2521                 expire_client(old);
2522         return status;
2523 out_free_conn:
2524         spin_unlock(&nn->client_lock);
2525         free_conn(conn);
2526         if (old)
2527                 expire_client(old);
2528 out_free_session:
2529         __free_session(new);
2530 out_release_drc_mem:
2531         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2532         return status;
2533 }
2534
2535 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2536 {
2537         switch (*dir) {
2538         case NFS4_CDFC4_FORE:
2539         case NFS4_CDFC4_BACK:
2540                 return nfs_ok;
2541         case NFS4_CDFC4_FORE_OR_BOTH:
2542         case NFS4_CDFC4_BACK_OR_BOTH:
2543                 *dir = NFS4_CDFC4_BOTH;
2544                 return nfs_ok;
2545         };
2546         return nfserr_inval;
2547 }
2548
2549 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2550 {
2551         struct nfsd4_session *session = cstate->session;
2552         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2553         __be32 status;
2554
2555         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2556         if (status)
2557                 return status;
2558         spin_lock(&nn->client_lock);
2559         session->se_cb_prog = bc->bc_cb_program;
2560         session->se_cb_sec = bc->bc_cb_sec;
2561         spin_unlock(&nn->client_lock);
2562
2563         nfsd4_probe_callback(session->se_client);
2564
2565         return nfs_ok;
2566 }
2567
2568 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2569                      struct nfsd4_compound_state *cstate,
2570                      struct nfsd4_bind_conn_to_session *bcts)
2571 {
2572         __be32 status;
2573         struct nfsd4_conn *conn;
2574         struct nfsd4_session *session;
2575         struct net *net = SVC_NET(rqstp);
2576         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2577
2578         if (!nfsd4_last_compound_op(rqstp))
2579                 return nfserr_not_only_op;
2580         spin_lock(&nn->client_lock);
2581         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2582         spin_unlock(&nn->client_lock);
2583         if (!session)
2584                 goto out_no_session;
2585         status = nfserr_wrong_cred;
2586         if (!mach_creds_match(session->se_client, rqstp))
2587                 goto out;
2588         status = nfsd4_map_bcts_dir(&bcts->dir);
2589         if (status)
2590                 goto out;
2591         conn = alloc_conn(rqstp, bcts->dir);
2592         status = nfserr_jukebox;
2593         if (!conn)
2594                 goto out;
2595         nfsd4_init_conn(rqstp, conn, session);
2596         status = nfs_ok;
2597 out:
2598         nfsd4_put_session(session);
2599 out_no_session:
2600         return status;
2601 }
2602
2603 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2604 {
2605         if (!session)
2606                 return 0;
2607         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2608 }
2609
2610 __be32
2611 nfsd4_destroy_session(struct svc_rqst *r,
2612                       struct nfsd4_compound_state *cstate,
2613                       struct nfsd4_destroy_session *sessionid)
2614 {
2615         struct nfsd4_session *ses;
2616         __be32 status;
2617         int ref_held_by_me = 0;
2618         struct net *net = SVC_NET(r);
2619         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2620
2621         status = nfserr_not_only_op;
2622         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2623                 if (!nfsd4_last_compound_op(r))
2624                         goto out;
2625                 ref_held_by_me++;
2626         }
2627         dump_sessionid(__func__, &sessionid->sessionid);
2628         spin_lock(&nn->client_lock);
2629         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2630         if (!ses)
2631                 goto out_client_lock;
2632         status = nfserr_wrong_cred;
2633         if (!mach_creds_match(ses->se_client, r))
2634                 goto out_put_session;
2635         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2636         if (status)
2637                 goto out_put_session;
2638         unhash_session(ses);
2639         spin_unlock(&nn->client_lock);
2640
2641         nfsd4_probe_callback_sync(ses->se_client);
2642
2643         spin_lock(&nn->client_lock);
2644         status = nfs_ok;
2645 out_put_session:
2646         nfsd4_put_session_locked(ses);
2647 out_client_lock:
2648         spin_unlock(&nn->client_lock);
2649 out:
2650         return status;
2651 }
2652
2653 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2654 {
2655         struct nfsd4_conn *c;
2656
2657         list_for_each_entry(c, &s->se_conns, cn_persession) {
2658                 if (c->cn_xprt == xpt) {
2659                         return c;
2660                 }
2661         }
2662         return NULL;
2663 }
2664
2665 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2666 {
2667         struct nfs4_client *clp = ses->se_client;
2668         struct nfsd4_conn *c;
2669         __be32 status = nfs_ok;
2670         int ret;
2671
2672         spin_lock(&clp->cl_lock);
2673         c = __nfsd4_find_conn(new->cn_xprt, ses);
2674         if (c)
2675                 goto out_free;
2676         status = nfserr_conn_not_bound_to_session;
2677         if (clp->cl_mach_cred)
2678                 goto out_free;
2679         __nfsd4_hash_conn(new, ses);
2680         spin_unlock(&clp->cl_lock);
2681         ret = nfsd4_register_conn(new);
2682         if (ret)
2683                 /* oops; xprt is already down: */
2684                 nfsd4_conn_lost(&new->cn_xpt_user);
2685         return nfs_ok;
2686 out_free:
2687         spin_unlock(&clp->cl_lock);
2688         free_conn(new);
2689         return status;
2690 }
2691
2692 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2693 {
2694         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2695
2696         return args->opcnt > session->se_fchannel.maxops;
2697 }
2698
2699 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2700                                   struct nfsd4_session *session)
2701 {
2702         struct xdr_buf *xb = &rqstp->rq_arg;
2703
2704         return xb->len > session->se_fchannel.maxreq_sz;
2705 }
2706
2707 __be32
2708 nfsd4_sequence(struct svc_rqst *rqstp,
2709                struct nfsd4_compound_state *cstate,
2710                struct nfsd4_sequence *seq)
2711 {
2712         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2713         struct xdr_stream *xdr = &resp->xdr;
2714         struct nfsd4_session *session;
2715         struct nfs4_client *clp;
2716         struct nfsd4_slot *slot;
2717         struct nfsd4_conn *conn;
2718         __be32 status;
2719         int buflen;
2720         struct net *net = SVC_NET(rqstp);
2721         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2722
2723         if (resp->opcnt != 1)
2724                 return nfserr_sequence_pos;
2725
2726         /*
2727          * Will be either used or freed by nfsd4_sequence_check_conn
2728          * below.
2729          */
2730         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2731         if (!conn)
2732                 return nfserr_jukebox;
2733
2734         spin_lock(&nn->client_lock);
2735         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2736         if (!session)
2737                 goto out_no_session;
2738         clp = session->se_client;
2739
2740         status = nfserr_too_many_ops;
2741         if (nfsd4_session_too_many_ops(rqstp, session))
2742                 goto out_put_session;
2743
2744         status = nfserr_req_too_big;
2745         if (nfsd4_request_too_big(rqstp, session))
2746                 goto out_put_session;
2747
2748         status = nfserr_badslot;
2749         if (seq->slotid >= session->se_fchannel.maxreqs)
2750                 goto out_put_session;
2751
2752         slot = session->se_slots[seq->slotid];
2753         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2754
2755         /* We do not negotiate the number of slots yet, so set the
2756          * maxslots to the session maxreqs which is used to encode
2757          * sr_highest_slotid and the sr_target_slot id to maxslots */
2758         seq->maxslots = session->se_fchannel.maxreqs;
2759
2760         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2761                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2762         if (status == nfserr_replay_cache) {
2763                 status = nfserr_seq_misordered;
2764                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2765                         goto out_put_session;
2766                 cstate->slot = slot;
2767                 cstate->session = session;
2768                 cstate->clp = clp;
2769                 /* Return the cached reply status and set cstate->status
2770                  * for nfsd4_proc_compound processing */
2771                 status = nfsd4_replay_cache_entry(resp, seq);
2772                 cstate->status = nfserr_replay_cache;
2773                 goto out;
2774         }
2775         if (status)
2776                 goto out_put_session;
2777
2778         status = nfsd4_sequence_check_conn(conn, session);
2779         conn = NULL;
2780         if (status)
2781                 goto out_put_session;
2782
2783         buflen = (seq->cachethis) ?
2784                         session->se_fchannel.maxresp_cached :
2785                         session->se_fchannel.maxresp_sz;
2786         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
2787                                     nfserr_rep_too_big;
2788         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
2789                 goto out_put_session;
2790         svc_reserve(rqstp, buflen);
2791
2792         status = nfs_ok;
2793         /* Success! bump slot seqid */
2794         slot->sl_seqid = seq->seqid;
2795         slot->sl_flags |= NFSD4_SLOT_INUSE;
2796         if (seq->cachethis)
2797                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2798         else
2799                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2800
2801         cstate->slot = slot;
2802         cstate->session = session;
2803         cstate->clp = clp;
2804
2805 out:
2806         switch (clp->cl_cb_state) {
2807         case NFSD4_CB_DOWN:
2808                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2809                 break;
2810         case NFSD4_CB_FAULT:
2811                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2812                 break;
2813         default:
2814                 seq->status_flags = 0;
2815         }
2816         if (!list_empty(&clp->cl_revoked))
2817                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2818 out_no_session:
2819         if (conn)
2820                 free_conn(conn);
2821         spin_unlock(&nn->client_lock);
2822         return status;
2823 out_put_session:
2824         nfsd4_put_session_locked(session);
2825         goto out_no_session;
2826 }
2827
2828 void
2829 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
2830 {
2831         struct nfsd4_compound_state *cs = &resp->cstate;
2832
2833         if (nfsd4_has_session(cs)) {
2834                 if (cs->status != nfserr_replay_cache) {
2835                         nfsd4_store_cache_entry(resp);
2836                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
2837                 }
2838                 /* Drop session reference that was taken in nfsd4_sequence() */
2839                 nfsd4_put_session(cs->session);
2840         } else if (cs->clp)
2841                 put_client_renew(cs->clp);
2842 }
2843
2844 __be32
2845 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2846 {
2847         struct nfs4_client *conf, *unconf;
2848         struct nfs4_client *clp = NULL;
2849         __be32 status = 0;
2850         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2851
2852         spin_lock(&nn->client_lock);
2853         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2854         conf = find_confirmed_client(&dc->clientid, true, nn);
2855         WARN_ON_ONCE(conf && unconf);
2856
2857         if (conf) {
2858                 if (client_has_state(conf)) {
2859                         status = nfserr_clientid_busy;
2860                         goto out;
2861                 }
2862                 status = mark_client_expired_locked(conf);
2863                 if (status)
2864                         goto out;
2865                 clp = conf;
2866         } else if (unconf)
2867                 clp = unconf;
2868         else {
2869                 status = nfserr_stale_clientid;
2870                 goto out;
2871         }
2872         if (!mach_creds_match(clp, rqstp)) {
2873                 clp = NULL;
2874                 status = nfserr_wrong_cred;
2875                 goto out;
2876         }
2877         unhash_client_locked(clp);
2878 out:
2879         spin_unlock(&nn->client_lock);
2880         if (clp)
2881                 expire_client(clp);
2882         return status;
2883 }
2884
2885 __be32
2886 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2887 {
2888         __be32 status = 0;
2889
2890         if (rc->rca_one_fs) {
2891                 if (!cstate->current_fh.fh_dentry)
2892                         return nfserr_nofilehandle;
2893                 /*
2894                  * We don't take advantage of the rca_one_fs case.
2895                  * That's OK, it's optional, we can safely ignore it.
2896                  */
2897                  return nfs_ok;
2898         }
2899
2900         status = nfserr_complete_already;
2901         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2902                              &cstate->session->se_client->cl_flags))
2903                 goto out;
2904
2905         status = nfserr_stale_clientid;
2906         if (is_client_expired(cstate->session->se_client))
2907                 /*
2908                  * The following error isn't really legal.
2909                  * But we only get here if the client just explicitly
2910                  * destroyed the client.  Surely it no longer cares what
2911                  * error it gets back on an operation for the dead
2912                  * client.
2913                  */
2914                 goto out;
2915
2916         status = nfs_ok;
2917         nfsd4_client_record_create(cstate->session->se_client);
2918 out:
2919         return status;
2920 }
2921
2922 __be32
2923 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2924                   struct nfsd4_setclientid *setclid)
2925 {
2926         struct xdr_netobj       clname = setclid->se_name;
2927         nfs4_verifier           clverifier = setclid->se_verf;
2928         struct nfs4_client      *conf, *new;
2929         struct nfs4_client      *unconf = NULL;
2930         __be32                  status;
2931         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2932
2933         new = create_client(clname, rqstp, &clverifier);
2934         if (new == NULL)
2935                 return nfserr_jukebox;
2936         /* Cases below refer to rfc 3530 section 14.2.33: */
2937         spin_lock(&nn->client_lock);
2938         conf = find_confirmed_client_by_name(&clname, nn);
2939         if (conf) {
2940                 /* case 0: */
2941                 status = nfserr_clid_inuse;
2942                 if (clp_used_exchangeid(conf))
2943                         goto out;
2944                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2945                         char addr_str[INET6_ADDRSTRLEN];
2946                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2947                                  sizeof(addr_str));
2948                         dprintk("NFSD: setclientid: string in use by client "
2949                                 "at %s\n", addr_str);
2950                         goto out;
2951                 }
2952         }
2953         unconf = find_unconfirmed_client_by_name(&clname, nn);
2954         if (unconf)
2955                 unhash_client_locked(unconf);
2956         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2957                 /* case 1: probable callback update */
2958                 copy_clid(new, conf);
2959         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2960                 gen_clid(new, nn);
2961         new->cl_minorversion = 0;
2962         gen_callback(new, setclid, rqstp);
2963         add_to_unconfirmed(new);
2964         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2965         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2966         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2967         new = NULL;
2968         status = nfs_ok;
2969 out:
2970         spin_unlock(&nn->client_lock);
2971         if (new)
2972                 free_client(new);
2973         if (unconf)
2974                 expire_client(unconf);
2975         return status;
2976 }
2977
2978
2979 __be32
2980 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2981                          struct nfsd4_compound_state *cstate,
2982                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2983 {
2984         struct nfs4_client *conf, *unconf;
2985         struct nfs4_client *old = NULL;
2986         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2987         clientid_t * clid = &setclientid_confirm->sc_clientid;
2988         __be32 status;
2989         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2990
2991         if (STALE_CLIENTID(clid, nn))
2992                 return nfserr_stale_clientid;
2993
2994         spin_lock(&nn->client_lock);
2995         conf = find_confirmed_client(clid, false, nn);
2996         unconf = find_unconfirmed_client(clid, false, nn);
2997         /*
2998          * We try hard to give out unique clientid's, so if we get an
2999          * attempt to confirm the same clientid with a different cred,
3000          * there's a bug somewhere.  Let's charitably assume it's our
3001          * bug.
3002          */
3003         status = nfserr_serverfault;
3004         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3005                 goto out;
3006         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3007                 goto out;
3008         /* cases below refer to rfc 3530 section 14.2.34: */
3009         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3010                 if (conf && !unconf) /* case 2: probable retransmit */
3011                         status = nfs_ok;
3012                 else /* case 4: client hasn't noticed we rebooted yet? */
3013                         status = nfserr_stale_clientid;
3014                 goto out;
3015         }
3016         status = nfs_ok;
3017         if (conf) { /* case 1: callback update */
3018                 old = unconf;
3019                 unhash_client_locked(old);
3020                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3021         } else { /* case 3: normal case; new or rebooted client */
3022                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3023                 if (old) {
3024                         status = mark_client_expired_locked(old);
3025                         if (status) {
3026                                 old = NULL;
3027                                 goto out;
3028                         }
3029                 }
3030                 move_to_confirmed(unconf);
3031                 conf = unconf;
3032         }
3033         get_client_locked(conf);
3034         spin_unlock(&nn->client_lock);
3035         nfsd4_probe_callback(conf);
3036         spin_lock(&nn->client_lock);
3037         put_client_renew_locked(conf);
3038 out:
3039         spin_unlock(&nn->client_lock);
3040         if (old)
3041                 expire_client(old);
3042         return status;
3043 }
3044
3045 static struct nfs4_file *nfsd4_alloc_file(void)
3046 {
3047         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3048 }
3049
3050 /* OPEN Share state helper functions */
3051 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3052                                 struct nfs4_file *fp)
3053 {
3054         lockdep_assert_held(&state_lock);
3055
3056         atomic_set(&fp->fi_ref, 1);
3057         spin_lock_init(&fp->fi_lock);
3058         INIT_LIST_HEAD(&fp->fi_stateids);
3059         INIT_LIST_HEAD(&fp->fi_delegations);
3060         fh_copy_shallow(&fp->fi_fhandle, fh);
3061         fp->fi_deleg_file = NULL;
3062         fp->fi_had_conflict = false;
3063         fp->fi_share_deny = 0;
3064         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3065         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3066 #ifdef CONFIG_NFSD_PNFS
3067         INIT_LIST_HEAD(&fp->fi_lo_states);
3068         atomic_set(&fp->fi_lo_recalls, 0);
3069 #endif
3070         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3071 }
3072
3073 void
3074 nfsd4_free_slabs(void)
3075 {
3076         kmem_cache_destroy(openowner_slab);
3077         kmem_cache_destroy(lockowner_slab);
3078         kmem_cache_destroy(file_slab);
3079         kmem_cache_destroy(stateid_slab);
3080         kmem_cache_destroy(deleg_slab);
3081 }
3082
3083 int
3084 nfsd4_init_slabs(void)
3085 {
3086         openowner_slab = kmem_cache_create("nfsd4_openowners",
3087                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3088         if (openowner_slab == NULL)
3089                 goto out;
3090         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3091                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3092         if (lockowner_slab == NULL)
3093                 goto out_free_openowner_slab;
3094         file_slab = kmem_cache_create("nfsd4_files",
3095                         sizeof(struct nfs4_file), 0, 0, NULL);
3096         if (file_slab == NULL)
3097                 goto out_free_lockowner_slab;
3098         stateid_slab = kmem_cache_create("nfsd4_stateids",
3099                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3100         if (stateid_slab == NULL)
3101                 goto out_free_file_slab;
3102         deleg_slab = kmem_cache_create("nfsd4_delegations",
3103                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3104         if (deleg_slab == NULL)
3105                 goto out_free_stateid_slab;
3106         return 0;
3107
3108 out_free_stateid_slab:
3109         kmem_cache_destroy(stateid_slab);
3110 out_free_file_slab:
3111         kmem_cache_destroy(file_slab);
3112 out_free_lockowner_slab:
3113         kmem_cache_destroy(lockowner_slab);
3114 out_free_openowner_slab:
3115         kmem_cache_destroy(openowner_slab);
3116 out:
3117         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3118         return -ENOMEM;
3119 }
3120
3121 static void init_nfs4_replay(struct nfs4_replay *rp)
3122 {
3123         rp->rp_status = nfserr_serverfault;
3124         rp->rp_buflen = 0;
3125         rp->rp_buf = rp->rp_ibuf;
3126         mutex_init(&rp->rp_mutex);
3127 }
3128
3129 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3130                 struct nfs4_stateowner *so)
3131 {
3132         if (!nfsd4_has_session(cstate)) {
3133                 mutex_lock(&so->so_replay.rp_mutex);
3134                 cstate->replay_owner = nfs4_get_stateowner(so);
3135         }
3136 }
3137
3138 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3139 {
3140         struct nfs4_stateowner *so = cstate->replay_owner;
3141
3142         if (so != NULL) {
3143                 cstate->replay_owner = NULL;
3144                 mutex_unlock(&so->so_replay.rp_mutex);
3145                 nfs4_put_stateowner(so);
3146         }
3147 }
3148
3149 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3150 {
3151         struct nfs4_stateowner *sop;
3152
3153         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3154         if (!sop)
3155                 return NULL;
3156
3157         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3158         if (!sop->so_owner.data) {
3159                 kmem_cache_free(slab, sop);
3160                 return NULL;
3161         }
3162         sop->so_owner.len = owner->len;
3163
3164         INIT_LIST_HEAD(&sop->so_stateids);
3165         sop->so_client = clp;
3166         init_nfs4_replay(&sop->so_replay);
3167         atomic_set(&sop->so_count, 1);
3168         return sop;
3169 }
3170
3171 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3172 {
3173         lockdep_assert_held(&clp->cl_lock);
3174
3175         list_add(&oo->oo_owner.so_strhash,
3176                  &clp->cl_ownerstr_hashtbl[strhashval]);
3177         list_add(&oo->oo_perclient, &clp->cl_openowners);
3178 }
3179
3180 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3181 {
3182         unhash_openowner_locked(openowner(so));
3183 }
3184
3185 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3186 {
3187         struct nfs4_openowner *oo = openowner(so);
3188
3189         kmem_cache_free(openowner_slab, oo);
3190 }
3191
3192 static const struct nfs4_stateowner_operations openowner_ops = {
3193         .so_unhash =    nfs4_unhash_openowner,
3194         .so_free =      nfs4_free_openowner,
3195 };
3196
3197 static struct nfs4_openowner *
3198 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3199                            struct nfsd4_compound_state *cstate)
3200 {
3201         struct nfs4_client *clp = cstate->clp;
3202         struct nfs4_openowner *oo, *ret;
3203
3204         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3205         if (!oo)
3206                 return NULL;
3207         oo->oo_owner.so_ops = &openowner_ops;
3208         oo->oo_owner.so_is_open_owner = 1;
3209         oo->oo_owner.so_seqid = open->op_seqid;
3210         oo->oo_flags = 0;
3211         if (nfsd4_has_session(cstate))
3212                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3213         oo->oo_time = 0;
3214         oo->oo_last_closed_stid = NULL;
3215         INIT_LIST_HEAD(&oo->oo_close_lru);
3216         spin_lock(&clp->cl_lock);
3217         ret = find_openstateowner_str_locked(strhashval, open, clp);
3218         if (ret == NULL) {
3219                 hash_openowner(oo, clp, strhashval);
3220                 ret = oo;
3221         } else
3222                 nfs4_free_openowner(&oo->oo_owner);
3223         spin_unlock(&clp->cl_lock);
3224         return ret;
3225 }
3226
3227 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
3228         struct nfs4_openowner *oo = open->op_openowner;
3229
3230         atomic_inc(&stp->st_stid.sc_count);
3231         stp->st_stid.sc_type = NFS4_OPEN_STID;
3232         INIT_LIST_HEAD(&stp->st_locks);
3233         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3234         get_nfs4_file(fp);
3235         stp->st_stid.sc_file = fp;
3236         stp->st_access_bmap = 0;
3237         stp->st_deny_bmap = 0;
3238         stp->st_openstp = NULL;
3239         spin_lock(&oo->oo_owner.so_client->cl_lock);
3240         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3241         spin_lock(&fp->fi_lock);
3242         list_add(&stp->st_perfile, &fp->fi_stateids);
3243         spin_unlock(&fp->fi_lock);
3244         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3245 }
3246
3247 /*
3248  * In the 4.0 case we need to keep the owners around a little while to handle
3249  * CLOSE replay. We still do need to release any file access that is held by
3250  * them before returning however.
3251  */
3252 static void
3253 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3254 {
3255         struct nfs4_ol_stateid *last;
3256         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3257         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3258                                                 nfsd_net_id);
3259
3260         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3261
3262         /*
3263          * We know that we hold one reference via nfsd4_close, and another
3264          * "persistent" reference for the client. If the refcount is higher
3265          * than 2, then there are still calls in progress that are using this
3266          * stateid. We can't put the sc_file reference until they are finished.
3267          * Wait for the refcount to drop to 2. Since it has been unhashed,
3268          * there should be no danger of the refcount going back up again at
3269          * this point.
3270          */
3271         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3272
3273         release_all_access(s);
3274         if (s->st_stid.sc_file) {
3275                 put_nfs4_file(s->st_stid.sc_file);
3276                 s->st_stid.sc_file = NULL;
3277         }
3278
3279         spin_lock(&nn->client_lock);
3280         last = oo->oo_last_closed_stid;
3281         oo->oo_last_closed_stid = s;
3282         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3283         oo->oo_time = get_seconds();
3284         spin_unlock(&nn->client_lock);
3285         if (last)
3286                 nfs4_put_stid(&last->st_stid);
3287 }
3288
3289 /* search file_hashtbl[] for file */
3290 static struct nfs4_file *
3291 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3292 {
3293         struct nfs4_file *fp;
3294
3295         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3296                 if (fh_match(&fp->fi_fhandle, fh)) {
3297                         if (atomic_inc_not_zero(&fp->fi_ref))
3298                                 return fp;
3299                 }
3300         }
3301         return NULL;
3302 }
3303
3304 struct nfs4_file *
3305 find_file(struct knfsd_fh *fh)
3306 {
3307         struct nfs4_file *fp;
3308         unsigned int hashval = file_hashval(fh);
3309
3310         rcu_read_lock();
3311         fp = find_file_locked(fh, hashval);
3312         rcu_read_unlock();
3313         return fp;
3314 }
3315
3316 static struct nfs4_file *
3317 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3318 {
3319         struct nfs4_file *fp;
3320         unsigned int hashval = file_hashval(fh);
3321
3322         rcu_read_lock();
3323         fp = find_file_locked(fh, hashval);
3324         rcu_read_unlock();
3325         if (fp)
3326                 return fp;
3327
3328         spin_lock(&state_lock);
3329         fp = find_file_locked(fh, hashval);
3330         if (likely(fp == NULL)) {
3331                 nfsd4_init_file(fh, hashval, new);
3332                 fp = new;
3333         }
3334         spin_unlock(&state_lock);
3335
3336         return fp;
3337 }
3338
3339 /*
3340  * Called to check deny when READ with all zero stateid or
3341  * WRITE with all zero or all one stateid
3342  */
3343 static __be32
3344 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3345 {
3346         struct nfs4_file *fp;
3347         __be32 ret = nfs_ok;
3348
3349         fp = find_file(&current_fh->fh_handle);
3350         if (!fp)
3351                 return ret;
3352         /* Check for conflicting share reservations */
3353         spin_lock(&fp->fi_lock);
3354         if (fp->fi_share_deny & deny_type)
3355                 ret = nfserr_locked;
3356         spin_unlock(&fp->fi_lock);
3357         put_nfs4_file(fp);
3358         return ret;
3359 }
3360
3361 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3362 {
3363         struct nfs4_delegation *dp = cb_to_delegation(cb);
3364         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3365                                           nfsd_net_id);
3366
3367         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3368
3369         /*
3370          * We can't do this in nfsd_break_deleg_cb because it is
3371          * already holding inode->i_lock.
3372          *
3373          * If the dl_time != 0, then we know that it has already been
3374          * queued for a lease break. Don't queue it again.
3375          */
3376         spin_lock(&state_lock);
3377         if (dp->dl_time == 0) {
3378                 dp->dl_time = get_seconds();
3379                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3380         }
3381         spin_unlock(&state_lock);
3382 }
3383
3384 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3385                 struct rpc_task *task)
3386 {
3387         struct nfs4_delegation *dp = cb_to_delegation(cb);
3388
3389         switch (task->tk_status) {
3390         case 0:
3391                 return 1;
3392         case -EBADHANDLE:
3393         case -NFS4ERR_BAD_STATEID:
3394                 /*
3395                  * Race: client probably got cb_recall before open reply
3396                  * granting delegation.
3397                  */
3398                 if (dp->dl_retries--) {
3399                         rpc_delay(task, 2 * HZ);
3400                         return 0;
3401                 }
3402                 /*FALLTHRU*/
3403         default:
3404                 return -1;
3405         }
3406 }
3407
3408 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3409 {
3410         struct nfs4_delegation *dp = cb_to_delegation(cb);
3411
3412         nfs4_put_stid(&dp->dl_stid);
3413 }
3414
3415 static struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3416         .prepare        = nfsd4_cb_recall_prepare,
3417         .done           = nfsd4_cb_recall_done,
3418         .release        = nfsd4_cb_recall_release,
3419 };
3420
3421 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3422 {
3423         /*
3424          * We're assuming the state code never drops its reference
3425          * without first removing the lease.  Since we're in this lease
3426          * callback (and since the lease code is serialized by the kernel
3427          * lock) we know the server hasn't removed the lease yet, we know
3428          * it's safe to take a reference.
3429          */
3430         atomic_inc(&dp->dl_stid.sc_count);
3431         nfsd4_run_cb(&dp->dl_recall);
3432 }
3433
3434 /* Called from break_lease() with i_lock held. */
3435 static bool
3436 nfsd_break_deleg_cb(struct file_lock *fl)
3437 {
3438         bool ret = false;
3439         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3440         struct nfs4_delegation *dp;
3441
3442         if (!fp) {
3443                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3444                 return ret;
3445         }
3446         if (fp->fi_had_conflict) {
3447                 WARN(1, "duplicate break on %p\n", fp);
3448                 return ret;
3449         }
3450         /*
3451          * We don't want the locks code to timeout the lease for us;
3452          * we'll remove it ourself if a delegation isn't returned
3453          * in time:
3454          */
3455         fl->fl_break_time = 0;
3456
3457         spin_lock(&fp->fi_lock);
3458         fp->fi_had_conflict = true;
3459         /*
3460          * If there are no delegations on the list, then return true
3461          * so that the lease code will go ahead and delete it.
3462          */
3463         if (list_empty(&fp->fi_delegations))
3464                 ret = true;
3465         else
3466                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3467                         nfsd_break_one_deleg(dp);
3468         spin_unlock(&fp->fi_lock);
3469         return ret;
3470 }
3471
3472 static int
3473 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3474                      struct list_head *dispose)
3475 {
3476         if (arg & F_UNLCK)
3477                 return lease_modify(onlist, arg, dispose);
3478         else
3479                 return -EAGAIN;
3480 }
3481
3482 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3483         .lm_break = nfsd_break_deleg_cb,
3484         .lm_change = nfsd_change_deleg_cb,
3485 };
3486
3487 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3488 {
3489         if (nfsd4_has_session(cstate))
3490                 return nfs_ok;
3491         if (seqid == so->so_seqid - 1)
3492                 return nfserr_replay_me;
3493         if (seqid == so->so_seqid)
3494                 return nfs_ok;
3495         return nfserr_bad_seqid;
3496 }
3497
3498 static __be32 lookup_clientid(clientid_t *clid,
3499                 struct nfsd4_compound_state *cstate,
3500                 struct nfsd_net *nn)
3501 {
3502         struct nfs4_client *found;
3503
3504         if (cstate->clp) {
3505                 found = cstate->clp;
3506                 if (!same_clid(&found->cl_clientid, clid))
3507                         return nfserr_stale_clientid;
3508                 return nfs_ok;
3509         }
3510
3511         if (STALE_CLIENTID(clid, nn))
3512                 return nfserr_stale_clientid;
3513
3514         /*
3515          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3516          * cached already then we know this is for is for v4.0 and "sessions"
3517          * will be false.
3518          */
3519         WARN_ON_ONCE(cstate->session);
3520         spin_lock(&nn->client_lock);
3521         found = find_confirmed_client(clid, false, nn);
3522         if (!found) {
3523                 spin_unlock(&nn->client_lock);
3524                 return nfserr_expired;
3525         }
3526         atomic_inc(&found->cl_refcount);
3527         spin_unlock(&nn->client_lock);
3528
3529         /* Cache the nfs4_client in cstate! */
3530         cstate->clp = found;
3531         return nfs_ok;
3532 }
3533
3534 __be32
3535 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3536                     struct nfsd4_open *open, struct nfsd_net *nn)
3537 {
3538         clientid_t *clientid = &open->op_clientid;
3539         struct nfs4_client *clp = NULL;
3540         unsigned int strhashval;
3541         struct nfs4_openowner *oo = NULL;
3542         __be32 status;
3543
3544         if (STALE_CLIENTID(&open->op_clientid, nn))
3545                 return nfserr_stale_clientid;
3546         /*
3547          * In case we need it later, after we've already created the
3548          * file and don't want to risk a further failure:
3549          */
3550         open->op_file = nfsd4_alloc_file();
3551         if (open->op_file == NULL)
3552                 return nfserr_jukebox;
3553
3554         status = lookup_clientid(clientid, cstate, nn);
3555         if (status)
3556                 return status;
3557         clp = cstate->clp;
3558
3559         strhashval = ownerstr_hashval(&open->op_owner);
3560         oo = find_openstateowner_str(strhashval, open, clp);
3561         open->op_openowner = oo;
3562         if (!oo) {
3563                 goto new_owner;
3564         }
3565         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3566                 /* Replace unconfirmed owners without checking for replay. */
3567                 release_openowner(oo);
3568                 open->op_openowner = NULL;
3569                 goto new_owner;
3570         }
3571         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3572         if (status)
3573                 return status;
3574         goto alloc_stateid;
3575 new_owner:
3576         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3577         if (oo == NULL)
3578                 return nfserr_jukebox;
3579         open->op_openowner = oo;
3580 alloc_stateid:
3581         open->op_stp = nfs4_alloc_open_stateid(clp);
3582         if (!open->op_stp)
3583                 return nfserr_jukebox;
3584         return nfs_ok;
3585 }
3586
3587 static inline __be32
3588 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3589 {
3590         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3591                 return nfserr_openmode;
3592         else
3593                 return nfs_ok;
3594 }
3595
3596 static int share_access_to_flags(u32 share_access)
3597 {
3598         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3599 }
3600
3601 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3602 {
3603         struct nfs4_stid *ret;
3604
3605         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3606         if (!ret)
3607                 return NULL;
3608         return delegstateid(ret);
3609 }
3610
3611 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3612 {
3613         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3614                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3615 }
3616
3617 static __be32
3618 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3619                 struct nfs4_delegation **dp)
3620 {
3621         int flags;
3622         __be32 status = nfserr_bad_stateid;
3623         struct nfs4_delegation *deleg;
3624
3625         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3626         if (deleg == NULL)
3627                 goto out;
3628         flags = share_access_to_flags(open->op_share_access);
3629         status = nfs4_check_delegmode(deleg, flags);
3630         if (status) {
3631                 nfs4_put_stid(&deleg->dl_stid);
3632                 goto out;
3633         }
3634         *dp = deleg;
3635 out:
3636         if (!nfsd4_is_deleg_cur(open))
3637                 return nfs_ok;
3638         if (status)
3639                 return status;
3640         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3641         return nfs_ok;
3642 }
3643
3644 static struct nfs4_ol_stateid *
3645 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3646 {
3647         struct nfs4_ol_stateid *local, *ret = NULL;
3648         struct nfs4_openowner *oo = open->op_openowner;
3649
3650         spin_lock(&fp->fi_lock);
3651         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3652                 /* ignore lock owners */
3653                 if (local->st_stateowner->so_is_open_owner == 0)
3654                         continue;
3655                 if (local->st_stateowner == &oo->oo_owner) {
3656                         ret = local;
3657                         atomic_inc(&ret->st_stid.sc_count);
3658                         break;
3659                 }
3660         }
3661         spin_unlock(&fp->fi_lock);
3662         return ret;
3663 }
3664
3665 static inline int nfs4_access_to_access(u32 nfs4_access)
3666 {
3667         int flags = 0;
3668
3669         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3670                 flags |= NFSD_MAY_READ;
3671         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3672                 flags |= NFSD_MAY_WRITE;
3673         return flags;
3674 }
3675
3676 static inline __be32
3677 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3678                 struct nfsd4_open *open)
3679 {
3680         struct iattr iattr = {
3681                 .ia_valid = ATTR_SIZE,
3682                 .ia_size = 0,
3683         };
3684         if (!open->op_truncate)
3685                 return 0;
3686         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3687                 return nfserr_inval;
3688         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3689 }
3690
3691 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3692                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3693                 struct nfsd4_open *open)
3694 {
3695         struct file *filp = NULL;
3696         __be32 status;
3697         int oflag = nfs4_access_to_omode(open->op_share_access);
3698         int access = nfs4_access_to_access(open->op_share_access);
3699         unsigned char old_access_bmap, old_deny_bmap;
3700
3701         spin_lock(&fp->fi_lock);
3702
3703         /*
3704          * Are we trying to set a deny mode that would conflict with
3705          * current access?
3706          */
3707         status = nfs4_file_check_deny(fp, open->op_share_deny);
3708         if (status != nfs_ok) {
3709                 spin_unlock(&fp->fi_lock);
3710                 goto out;
3711         }
3712
3713         /* set access to the file */
3714         status = nfs4_file_get_access(fp, open->op_share_access);
3715         if (status != nfs_ok) {
3716                 spin_unlock(&fp->fi_lock);
3717                 goto out;
3718         }
3719
3720         /* Set access bits in stateid */
3721         old_access_bmap = stp->st_access_bmap;
3722         set_access(open->op_share_access, stp);
3723
3724         /* Set new deny mask */
3725         old_deny_bmap = stp->st_deny_bmap;
3726         set_deny(open->op_share_deny, stp);
3727         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3728
3729         if (!fp->fi_fds[oflag]) {
3730                 spin_unlock(&fp->fi_lock);
3731                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3732                 if (status)
3733                         goto out_put_access;
3734                 spin_lock(&fp->fi_lock);
3735                 if (!fp->fi_fds[oflag]) {
3736                         fp->fi_fds[oflag] = filp;
3737                         filp = NULL;
3738                 }
3739         }
3740         spin_unlock(&fp->fi_lock);
3741         if (filp)
3742                 fput(filp);
3743
3744         status = nfsd4_truncate(rqstp, cur_fh, open);
3745         if (status)
3746                 goto out_put_access;
3747 out:
3748         return status;
3749 out_put_access:
3750         stp->st_access_bmap = old_access_bmap;
3751         nfs4_file_put_access(fp, open->op_share_access);
3752         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3753         goto out;
3754 }
3755
3756 static __be32
3757 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)
3758 {
3759         __be32 status;
3760         unsigned char old_deny_bmap;
3761
3762         if (!test_access(open->op_share_access, stp))
3763                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3764
3765         /* test and set deny mode */
3766         spin_lock(&fp->fi_lock);
3767         status = nfs4_file_check_deny(fp, open->op_share_deny);
3768         if (status == nfs_ok) {
3769                 old_deny_bmap = stp->st_deny_bmap;
3770                 set_deny(open->op_share_deny, stp);
3771                 fp->fi_share_deny |=
3772                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3773         }
3774         spin_unlock(&fp->fi_lock);
3775
3776         if (status != nfs_ok)
3777                 return status;
3778
3779         status = nfsd4_truncate(rqstp, cur_fh, open);
3780         if (status != nfs_ok)
3781                 reset_union_bmap_deny(old_deny_bmap, stp);
3782         return status;
3783 }
3784
3785 static void
3786 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3787 {
3788         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3789 }
3790
3791 /* Should we give out recallable state?: */
3792 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3793 {
3794         if (clp->cl_cb_state == NFSD4_CB_UP)
3795                 return true;
3796         /*
3797          * In the sessions case, since we don't have to establish a
3798          * separate connection for callbacks, we assume it's OK
3799          * until we hear otherwise:
3800          */
3801         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3802 }
3803
3804 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
3805 {
3806         struct file_lock *fl;
3807
3808         fl = locks_alloc_lock();
3809         if (!fl)
3810                 return NULL;
3811         fl->fl_lmops = &nfsd_lease_mng_ops;
3812         fl->fl_flags = FL_DELEG;
3813         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3814         fl->fl_end = OFFSET_MAX;
3815         fl->fl_owner = (fl_owner_t)fp;
3816         fl->fl_pid = current->tgid;
3817         return fl;
3818 }
3819
3820 static int nfs4_setlease(struct nfs4_delegation *dp)
3821 {
3822         struct nfs4_file *fp = dp->dl_stid.sc_file;
3823         struct file_lock *fl, *ret;
3824         struct file *filp;
3825         int status = 0;
3826
3827         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
3828         if (!fl)
3829                 return -ENOMEM;
3830         filp = find_readable_file(fp);
3831         if (!filp) {
3832                 /* We should always have a readable file here */
3833                 WARN_ON_ONCE(1);
3834                 return -EBADF;
3835         }
3836         fl->fl_file = filp;
3837         ret = fl;
3838         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
3839         if (fl)
3840                 locks_free_lock(fl);
3841         if (status)
3842                 goto out_fput;
3843         spin_lock(&state_lock);
3844         spin_lock(&fp->fi_lock);
3845         /* Did the lease get broken before we took the lock? */
3846         status = -EAGAIN;
3847         if (fp->fi_had_conflict)
3848                 goto out_unlock;
3849         /* Race breaker */
3850         if (fp->fi_deleg_file) {
3851                 status = 0;
3852                 ++fp->fi_delegees;
3853                 hash_delegation_locked(dp, fp);
3854                 goto out_unlock;
3855         }
3856         fp->fi_deleg_file = filp;
3857         fp->fi_delegees = 1;
3858         hash_delegation_locked(dp, fp);
3859         spin_unlock(&fp->fi_lock);
3860         spin_unlock(&state_lock);
3861         return 0;
3862 out_unlock:
3863         spin_unlock(&fp->fi_lock);
3864         spin_unlock(&state_lock);
3865 out_fput:
3866         fput(filp);
3867         return status;
3868 }
3869
3870 static struct nfs4_delegation *
3871 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
3872                     struct nfs4_file *fp)
3873 {
3874         int status;
3875         struct nfs4_delegation *dp;
3876
3877         if (fp->fi_had_conflict)
3878                 return ERR_PTR(-EAGAIN);
3879
3880         dp = alloc_init_deleg(clp, fh);
3881         if (!dp)
3882                 return ERR_PTR(-ENOMEM);
3883
3884         get_nfs4_file(fp);
3885         spin_lock(&state_lock);
3886         spin_lock(&fp->fi_lock);
3887         dp->dl_stid.sc_file = fp;
3888         if (!fp->fi_deleg_file) {
3889                 spin_unlock(&fp->fi_lock);
3890                 spin_unlock(&state_lock);
3891                 status = nfs4_setlease(dp);
3892                 goto out;
3893         }
3894         if (fp->fi_had_conflict) {
3895                 status = -EAGAIN;
3896                 goto out_unlock;
3897         }
3898         ++fp->fi_delegees;
3899         hash_delegation_locked(dp, fp);
3900         status = 0;
3901 out_unlock:
3902         spin_unlock(&fp->fi_lock);
3903         spin_unlock(&state_lock);
3904 out:
3905         if (status) {
3906                 nfs4_put_stid(&dp->dl_stid);
3907                 return ERR_PTR(status);
3908         }
3909         return dp;
3910 }
3911
3912 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3913 {
3914         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3915         if (status == -EAGAIN)
3916                 open->op_why_no_deleg = WND4_CONTENTION;
3917         else {
3918                 open->op_why_no_deleg = WND4_RESOURCE;
3919                 switch (open->op_deleg_want) {
3920                 case NFS4_SHARE_WANT_READ_DELEG:
3921                 case NFS4_SHARE_WANT_WRITE_DELEG:
3922                 case NFS4_SHARE_WANT_ANY_DELEG:
3923                         break;
3924                 case NFS4_SHARE_WANT_CANCEL:
3925                         open->op_why_no_deleg = WND4_CANCELLED;
3926                         break;
3927                 case NFS4_SHARE_WANT_NO_DELEG:
3928                         WARN_ON_ONCE(1);
3929                 }
3930         }
3931 }
3932
3933 /*
3934  * Attempt to hand out a delegation.
3935  *
3936  * Note we don't support write delegations, and won't until the vfs has
3937  * proper support for them.
3938  */
3939 static void
3940 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
3941                         struct nfs4_ol_stateid *stp)
3942 {
3943         struct nfs4_delegation *dp;
3944         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
3945         struct nfs4_client *clp = stp->st_stid.sc_client;
3946         int cb_up;
3947         int status = 0;
3948
3949         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3950         open->op_recall = 0;
3951         switch (open->op_claim_type) {
3952                 case NFS4_OPEN_CLAIM_PREVIOUS:
3953                         if (!cb_up)
3954                                 open->op_recall = 1;
3955                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3956                                 goto out_no_deleg;
3957                         break;
3958                 case NFS4_OPEN_CLAIM_NULL:
3959                 case NFS4_OPEN_CLAIM_FH:
3960                         /*
3961                          * Let's not give out any delegations till everyone's
3962                          * had the chance to reclaim theirs....
3963                          */
3964                         if (locks_in_grace(clp->net))
3965                                 goto out_no_deleg;
3966                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3967                                 goto out_no_deleg;
3968                         /*
3969                          * Also, if the file was opened for write or
3970                          * create, there's a good chance the client's
3971                          * about to write to it, resulting in an
3972                          * immediate recall (since we don't support
3973                          * write delegations):
3974                          */
3975                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3976                                 goto out_no_deleg;
3977                         if (open->op_create == NFS4_OPEN_CREATE)
3978                                 goto out_no_deleg;
3979                         break;
3980                 default:
3981                         goto out_no_deleg;
3982         }
3983         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file);
3984         if (IS_ERR(dp))
3985                 goto out_no_deleg;
3986
3987         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3988
3989         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3990                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3991         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3992         nfs4_put_stid(&dp->dl_stid);
3993         return;
3994 out_no_deleg:
3995         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3996         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3997             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3998                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3999                 open->op_recall = 1;
4000         }
4001
4002         /* 4.1 client asking for a delegation? */
4003         if (open->op_deleg_want)
4004                 nfsd4_open_deleg_none_ext(open, status);
4005         return;
4006 }
4007
4008 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4009                                         struct nfs4_delegation *dp)
4010 {
4011         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4012             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4013                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4014                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4015         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4016                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4017                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4018                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4019         }
4020         /* Otherwise the client must be confused wanting a delegation
4021          * it already has, therefore we don't return
4022          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4023          */
4024 }
4025
4026 __be32
4027 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4028 {
4029         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4030         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4031         struct nfs4_file *fp = NULL;
4032         struct nfs4_ol_stateid *stp = NULL;
4033         struct nfs4_delegation *dp = NULL;
4034         __be32 status;
4035
4036         /*
4037          * Lookup file; if found, lookup stateid and check open request,
4038          * and check for delegations in the process of being recalled.
4039          * If not found, create the nfs4_file struct
4040          */
4041         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4042         if (fp != open->op_file) {
4043                 status = nfs4_check_deleg(cl, open, &dp);
4044                 if (status)
4045                         goto out;
4046                 stp = nfsd4_find_existing_open(fp, open);
4047         } else {
4048                 open->op_file = NULL;
4049                 status = nfserr_bad_stateid;
4050                 if (nfsd4_is_deleg_cur(open))
4051                         goto out;
4052         }
4053
4054         /*
4055          * OPEN the file, or upgrade an existing OPEN.
4056          * If truncate fails, the OPEN fails.
4057          */
4058         if (stp) {
4059                 /* Stateid was found, this is an OPEN upgrade */
4060                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4061                 if (status)
4062                         goto out;
4063         } else {
4064                 stp = open->op_stp;
4065                 open->op_stp = NULL;
4066                 init_open_stateid(stp, fp, open);
4067                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4068                 if (status) {
4069                         release_open_stateid(stp);
4070                         goto out;
4071                 }
4072         }
4073         update_stateid(&stp->st_stid.sc_stateid);
4074         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4075
4076         if (nfsd4_has_session(&resp->cstate)) {
4077                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4078                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4079                         open->op_why_no_deleg = WND4_NOT_WANTED;
4080                         goto nodeleg;
4081                 }
4082         }
4083
4084         /*
4085         * Attempt to hand out a delegation. No error return, because the
4086         * OPEN succeeds even if we fail.
4087         */
4088         nfs4_open_delegation(current_fh, open, stp);
4089 nodeleg:
4090         status = nfs_ok;
4091
4092         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4093                 STATEID_VAL(&stp->st_stid.sc_stateid));
4094 out:
4095         /* 4.1 client trying to upgrade/downgrade delegation? */
4096         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4097             open->op_deleg_want)
4098                 nfsd4_deleg_xgrade_none_ext(open, dp);
4099
4100         if (fp)
4101                 put_nfs4_file(fp);
4102         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4103                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
4104         /*
4105         * To finish the open response, we just need to set the rflags.
4106         */
4107         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4108         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4109             !nfsd4_has_session(&resp->cstate))
4110                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4111         if (dp)
4112                 nfs4_put_stid(&dp->dl_stid);
4113         if (stp)
4114                 nfs4_put_stid(&stp->st_stid);
4115
4116         return status;
4117 }
4118
4119 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4120                               struct nfsd4_open *open)
4121 {
4122         if (open->op_openowner) {
4123                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4124
4125                 nfsd4_cstate_assign_replay(cstate, so);
4126                 nfs4_put_stateowner(so);
4127         }
4128         if (open->op_file)
4129                 kmem_cache_free(file_slab, open->op_file);
4130         if (open->op_stp)
4131                 nfs4_put_stid(&open->op_stp->st_stid);
4132 }
4133
4134 __be32
4135 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4136             clientid_t *clid)
4137 {
4138         struct nfs4_client *clp;
4139         __be32 status;
4140         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4141
4142         dprintk("process_renew(%08x/%08x): starting\n",
4143                         clid->cl_boot, clid->cl_id);
4144         status = lookup_clientid(clid, cstate, nn);
4145         if (status)
4146                 goto out;
4147         clp = cstate->clp;
4148         status = nfserr_cb_path_down;
4149         if (!list_empty(&clp->cl_delegations)
4150                         && clp->cl_cb_state != NFSD4_CB_UP)
4151                 goto out;
4152         status = nfs_ok;
4153 out:
4154         return status;
4155 }
4156
4157 void
4158 nfsd4_end_grace(struct nfsd_net *nn)
4159 {
4160         /* do nothing if grace period already ended */
4161         if (nn->grace_ended)
4162                 return;
4163
4164         dprintk("NFSD: end of grace period\n");
4165         nn->grace_ended = true;
4166         /*
4167          * If the server goes down again right now, an NFSv4
4168          * client will still be allowed to reclaim after it comes back up,
4169          * even if it hasn't yet had a chance to reclaim state this time.
4170          *
4171          */
4172         nfsd4_record_grace_done(nn);
4173         /*
4174          * At this point, NFSv4 clients can still reclaim.  But if the
4175          * server crashes, any that have not yet reclaimed will be out
4176          * of luck on the next boot.
4177          *
4178          * (NFSv4.1+ clients are considered to have reclaimed once they
4179          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4180          * have reclaimed after their first OPEN.)
4181          */
4182         locks_end_grace(&nn->nfsd4_manager);
4183         /*
4184          * At this point, and once lockd and/or any other containers
4185          * exit their grace period, further reclaims will fail and
4186          * regular locking can resume.
4187          */
4188 }
4189
4190 static time_t
4191 nfs4_laundromat(struct nfsd_net *nn)
4192 {
4193         struct nfs4_client *clp;
4194         struct nfs4_openowner *oo;
4195         struct nfs4_delegation *dp;
4196         struct nfs4_ol_stateid *stp;
4197         struct list_head *pos, *next, reaplist;
4198         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4199         time_t t, new_timeo = nn->nfsd4_lease;
4200
4201         dprintk("NFSD: laundromat service - starting\n");
4202         nfsd4_end_grace(nn);
4203         INIT_LIST_HEAD(&reaplist);
4204         spin_lock(&nn->client_lock);
4205         list_for_each_safe(pos, next, &nn->client_lru) {
4206                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4207                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4208                         t = clp->cl_time - cutoff;
4209                         new_timeo = min(new_timeo, t);
4210                         break;
4211                 }
4212                 if (mark_client_expired_locked(clp)) {
4213                         dprintk("NFSD: client in use (clientid %08x)\n",
4214                                 clp->cl_clientid.cl_id);
4215                         continue;
4216                 }
4217                 list_add(&clp->cl_lru, &reaplist);
4218         }
4219         spin_unlock(&nn->client_lock);
4220         list_for_each_safe(pos, next, &reaplist) {
4221                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4222                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4223                         clp->cl_clientid.cl_id);
4224                 list_del_init(&clp->cl_lru);
4225                 expire_client(clp);
4226         }
4227         spin_lock(&state_lock);
4228         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4229                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4230                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
4231                         continue;
4232                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4233                         t = dp->dl_time - cutoff;
4234                         new_timeo = min(new_timeo, t);
4235                         break;
4236                 }
4237                 unhash_delegation_locked(dp);
4238                 list_add(&dp->dl_recall_lru, &reaplist);
4239         }
4240         spin_unlock(&state_lock);
4241         while (!list_empty(&reaplist)) {
4242                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4243                                         dl_recall_lru);
4244                 list_del_init(&dp->dl_recall_lru);
4245                 revoke_delegation(dp);
4246         }
4247
4248         spin_lock(&nn->client_lock);
4249         while (!list_empty(&nn->close_lru)) {
4250                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4251                                         oo_close_lru);
4252                 if (time_after((unsigned long)oo->oo_time,
4253                                (unsigned long)cutoff)) {
4254                         t = oo->oo_time - cutoff;
4255                         new_timeo = min(new_timeo, t);
4256                         break;
4257                 }
4258                 list_del_init(&oo->oo_close_lru);
4259                 stp = oo->oo_last_closed_stid;
4260                 oo->oo_last_closed_stid = NULL;
4261                 spin_unlock(&nn->client_lock);
4262                 nfs4_put_stid(&stp->st_stid);
4263                 spin_lock(&nn->client_lock);
4264         }
4265         spin_unlock(&nn->client_lock);
4266
4267         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4268         return new_timeo;
4269 }
4270
4271 static struct workqueue_struct *laundry_wq;
4272 static void laundromat_main(struct work_struct *);
4273
4274 static void
4275 laundromat_main(struct work_struct *laundry)
4276 {
4277         time_t t;
4278         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
4279                                                   work);
4280         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4281                                            laundromat_work);
4282
4283         t = nfs4_laundromat(nn);
4284         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4285         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4286 }
4287
4288 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
4289 {
4290         if (!fh_match(&fhp->fh_handle, &stp->st_stid.sc_file->fi_fhandle))
4291                 return nfserr_bad_stateid;
4292         return nfs_ok;
4293 }
4294
4295 static inline int
4296 access_permit_read(struct nfs4_ol_stateid *stp)
4297 {
4298         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4299                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4300                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4301 }
4302
4303 static inline int
4304 access_permit_write(struct nfs4_ol_stateid *stp)
4305 {
4306         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4307                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4308 }
4309
4310 static
4311 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4312 {
4313         __be32 status = nfserr_openmode;
4314
4315         /* For lock stateid's, we test the parent open, not the lock: */
4316         if (stp->st_openstp)
4317                 stp = stp->st_openstp;
4318         if ((flags & WR_STATE) && !access_permit_write(stp))
4319                 goto out;
4320         if ((flags & RD_STATE) && !access_permit_read(stp))
4321                 goto out;
4322         status = nfs_ok;
4323 out:
4324         return status;
4325 }
4326
4327 static inline __be32
4328 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4329 {
4330         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4331                 return nfs_ok;
4332         else if (locks_in_grace(net)) {
4333                 /* Answer in remaining cases depends on existence of
4334                  * conflicting state; so we must wait out the grace period. */
4335                 return nfserr_grace;
4336         } else if (flags & WR_STATE)
4337                 return nfs4_share_conflict(current_fh,
4338                                 NFS4_SHARE_DENY_WRITE);
4339         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4340                 return nfs4_share_conflict(current_fh,
4341                                 NFS4_SHARE_DENY_READ);
4342 }
4343
4344 /*
4345  * Allow READ/WRITE during grace period on recovered state only for files
4346  * that are not able to provide mandatory locking.
4347  */
4348 static inline int
4349 grace_disallows_io(struct net *net, struct inode *inode)
4350 {
4351         return locks_in_grace(net) && mandatory_lock(inode);
4352 }
4353
4354 /* Returns true iff a is later than b: */
4355 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4356 {
4357         return (s32)(a->si_generation - b->si_generation) > 0;
4358 }
4359
4360 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4361 {
4362         /*
4363          * When sessions are used the stateid generation number is ignored
4364          * when it is zero.
4365          */
4366         if (has_session && in->si_generation == 0)
4367                 return nfs_ok;
4368
4369         if (in->si_generation == ref->si_generation)
4370                 return nfs_ok;
4371
4372         /* If the client sends us a stateid from the future, it's buggy: */
4373         if (stateid_generation_after(in, ref))
4374                 return nfserr_bad_stateid;
4375         /*
4376          * However, we could see a stateid from the past, even from a
4377          * non-buggy client.  For example, if the client sends a lock
4378          * while some IO is outstanding, the lock may bump si_generation
4379          * while the IO is still in flight.  The client could avoid that
4380          * situation by waiting for responses on all the IO requests,
4381          * but better performance may result in retrying IO that
4382          * receives an old_stateid error if requests are rarely
4383          * reordered in flight:
4384          */
4385         return nfserr_old_stateid;
4386 }
4387
4388 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4389 {
4390         struct nfs4_stid *s;
4391         struct nfs4_ol_stateid *ols;
4392         __be32 status = nfserr_bad_stateid;
4393
4394         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4395                 return status;
4396         /* Client debugging aid. */
4397         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4398                 char addr_str[INET6_ADDRSTRLEN];
4399                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4400                                  sizeof(addr_str));
4401                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4402                                         "with incorrect client ID\n", addr_str);
4403                 return status;
4404         }
4405         spin_lock(&cl->cl_lock);
4406         s = find_stateid_locked(cl, stateid);
4407         if (!s)
4408                 goto out_unlock;
4409         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4410         if (status)
4411                 goto out_unlock;
4412         switch (s->sc_type) {
4413         case NFS4_DELEG_STID:
4414                 status = nfs_ok;
4415                 break;
4416         case NFS4_REVOKED_DELEG_STID:
4417                 status = nfserr_deleg_revoked;
4418                 break;
4419         case NFS4_OPEN_STID:
4420         case NFS4_LOCK_STID:
4421                 ols = openlockstateid(s);
4422                 if (ols->st_stateowner->so_is_open_owner
4423                                 && !(openowner(ols->st_stateowner)->oo_flags
4424                                                 & NFS4_OO_CONFIRMED))
4425                         status = nfserr_bad_stateid;
4426                 else
4427                         status = nfs_ok;
4428                 break;
4429         default:
4430                 printk("unknown stateid type %x\n", s->sc_type);
4431                 /* Fallthrough */
4432         case NFS4_CLOSED_STID:
4433         case NFS4_CLOSED_DELEG_STID:
4434                 status = nfserr_bad_stateid;
4435         }
4436 out_unlock:
4437         spin_unlock(&cl->cl_lock);
4438         return status;
4439 }
4440
4441 __be32
4442 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4443                      stateid_t *stateid, unsigned char typemask,
4444                      struct nfs4_stid **s, struct nfsd_net *nn)
4445 {
4446         __be32 status;
4447
4448         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4449                 return nfserr_bad_stateid;
4450         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4451         if (status == nfserr_stale_clientid) {
4452                 if (cstate->session)
4453                         return nfserr_bad_stateid;
4454                 return nfserr_stale_stateid;
4455         }
4456         if (status)
4457                 return status;
4458         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4459         if (!*s)
4460                 return nfserr_bad_stateid;
4461         return nfs_ok;
4462 }
4463
4464 /*
4465 * Checks for stateid operations
4466 */
4467 __be32
4468 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
4469                            stateid_t *stateid, int flags, struct file **filpp)
4470 {
4471         struct nfs4_stid *s;
4472         struct nfs4_ol_stateid *stp = NULL;
4473         struct nfs4_delegation *dp = NULL;
4474         struct svc_fh *current_fh = &cstate->current_fh;
4475         struct inode *ino = current_fh->fh_dentry->d_inode;
4476         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4477         struct file *file = NULL;
4478         __be32 status;
4479
4480         if (filpp)
4481                 *filpp = NULL;
4482
4483         if (grace_disallows_io(net, ino))
4484                 return nfserr_grace;
4485
4486         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4487                 return check_special_stateids(net, current_fh, stateid, flags);
4488
4489         status = nfsd4_lookup_stateid(cstate, stateid,
4490                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4491                                 &s, nn);
4492         if (status)
4493                 return status;
4494         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
4495         if (status)
4496                 goto out;
4497         switch (s->sc_type) {
4498         case NFS4_DELEG_STID:
4499                 dp = delegstateid(s);
4500                 status = nfs4_check_delegmode(dp, flags);
4501                 if (status)
4502                         goto out;
4503                 if (filpp) {
4504                         file = dp->dl_stid.sc_file->fi_deleg_file;
4505                         if (!file) {
4506                                 WARN_ON_ONCE(1);
4507                                 status = nfserr_serverfault;
4508                                 goto out;
4509                         }
4510                         get_file(file);
4511                 }
4512                 break;
4513         case NFS4_OPEN_STID:
4514         case NFS4_LOCK_STID:
4515                 stp = openlockstateid(s);
4516                 status = nfs4_check_fh(current_fh, stp);
4517                 if (status)
4518                         goto out;
4519                 if (stp->st_stateowner->so_is_open_owner
4520                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4521                         goto out;
4522                 status = nfs4_check_openmode(stp, flags);
4523                 if (status)
4524                         goto out;
4525                 if (filpp) {
4526                         struct nfs4_file *fp = stp->st_stid.sc_file;
4527
4528                         if (flags & RD_STATE)
4529                                 file = find_readable_file(fp);
4530                         else
4531                                 file = find_writeable_file(fp);
4532                 }
4533                 break;
4534         default:
4535                 status = nfserr_bad_stateid;
4536                 goto out;
4537         }
4538         status = nfs_ok;
4539         if (file)
4540                 *filpp = file;
4541 out:
4542         nfs4_put_stid(s);
4543         return status;
4544 }
4545
4546 /*
4547  * Test if the stateid is valid
4548  */
4549 __be32
4550 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4551                    struct nfsd4_test_stateid *test_stateid)
4552 {
4553         struct nfsd4_test_stateid_id *stateid;
4554         struct nfs4_client *cl = cstate->session->se_client;
4555
4556         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4557                 stateid->ts_id_status =
4558                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4559
4560         return nfs_ok;
4561 }
4562
4563 __be32
4564 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4565                    struct nfsd4_free_stateid *free_stateid)
4566 {
4567         stateid_t *stateid = &free_stateid->fr_stateid;
4568         struct nfs4_stid *s;
4569         struct nfs4_delegation *dp;
4570         struct nfs4_ol_stateid *stp;
4571         struct nfs4_client *cl = cstate->session->se_client;
4572         __be32 ret = nfserr_bad_stateid;
4573
4574         spin_lock(&cl->cl_lock);
4575         s = find_stateid_locked(cl, stateid);
4576         if (!s)
4577                 goto out_unlock;
4578         switch (s->sc_type) {
4579         case NFS4_DELEG_STID:
4580                 ret = nfserr_locks_held;
4581                 break;
4582         case NFS4_OPEN_STID:
4583                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4584                 if (ret)
4585                         break;
4586                 ret = nfserr_locks_held;
4587                 break;
4588         case NFS4_LOCK_STID:
4589                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4590                 if (ret)
4591                         break;
4592                 stp = openlockstateid(s);
4593                 ret = nfserr_locks_held;
4594                 if (check_for_locks(stp->st_stid.sc_file,
4595                                     lockowner(stp->st_stateowner)))
4596                         break;
4597                 unhash_lock_stateid(stp);
4598                 spin_unlock(&cl->cl_lock);
4599                 nfs4_put_stid(s);
4600                 ret = nfs_ok;
4601                 goto out;
4602         case NFS4_REVOKED_DELEG_STID:
4603                 dp = delegstateid(s);
4604                 list_del_init(&dp->dl_recall_lru);
4605                 spin_unlock(&cl->cl_lock);
4606                 nfs4_put_stid(s);
4607                 ret = nfs_ok;
4608                 goto out;
4609         /* Default falls through and returns nfserr_bad_stateid */
4610         }
4611 out_unlock:
4612         spin_unlock(&cl->cl_lock);
4613 out:
4614         return ret;
4615 }
4616
4617 static inline int
4618 setlkflg (int type)
4619 {
4620         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4621                 RD_STATE : WR_STATE;
4622 }
4623
4624 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4625 {
4626         struct svc_fh *current_fh = &cstate->current_fh;
4627         struct nfs4_stateowner *sop = stp->st_stateowner;
4628         __be32 status;
4629
4630         status = nfsd4_check_seqid(cstate, sop, seqid);
4631         if (status)
4632                 return status;
4633         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4634                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4635                 /*
4636                  * "Closed" stateid's exist *only* to return
4637                  * nfserr_replay_me from the previous step, and
4638                  * revoked delegations are kept only for free_stateid.
4639                  */
4640                 return nfserr_bad_stateid;
4641         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4642         if (status)
4643                 return status;
4644         return nfs4_check_fh(current_fh, stp);
4645 }
4646
4647 /*
4648  * Checks for sequence id mutating operations.
4649  */
4650 static __be32
4651 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4652                          stateid_t *stateid, char typemask,
4653                          struct nfs4_ol_stateid **stpp,
4654                          struct nfsd_net *nn)
4655 {
4656         __be32 status;
4657         struct nfs4_stid *s;
4658         struct nfs4_ol_stateid *stp = NULL;
4659
4660         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4661                 seqid, STATEID_VAL(stateid));
4662
4663         *stpp = NULL;
4664         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4665         if (status)
4666                 return status;
4667         stp = openlockstateid(s);
4668         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
4669
4670         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4671         if (!status)
4672                 *stpp = stp;
4673         else
4674                 nfs4_put_stid(&stp->st_stid);
4675         return status;
4676 }
4677
4678 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4679                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4680 {
4681         __be32 status;
4682         struct nfs4_openowner *oo;
4683         struct nfs4_ol_stateid *stp;
4684
4685         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4686                                                 NFS4_OPEN_STID, &stp, nn);
4687         if (status)
4688                 return status;
4689         oo = openowner(stp->st_stateowner);
4690         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4691                 nfs4_put_stid(&stp->st_stid);
4692                 return nfserr_bad_stateid;
4693         }
4694         *stpp = stp;
4695         return nfs_ok;
4696 }
4697
4698 __be32
4699 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4700                    struct nfsd4_open_confirm *oc)
4701 {
4702         __be32 status;
4703         struct nfs4_openowner *oo;
4704         struct nfs4_ol_stateid *stp;
4705         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4706
4707         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4708                         cstate->current_fh.fh_dentry);
4709
4710         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4711         if (status)
4712                 return status;
4713
4714         status = nfs4_preprocess_seqid_op(cstate,
4715                                         oc->oc_seqid, &oc->oc_req_stateid,
4716                                         NFS4_OPEN_STID, &stp, nn);
4717         if (status)
4718                 goto out;
4719         oo = openowner(stp->st_stateowner);
4720         status = nfserr_bad_stateid;
4721         if (oo->oo_flags & NFS4_OO_CONFIRMED)
4722                 goto put_stateid;
4723         oo->oo_flags |= NFS4_OO_CONFIRMED;
4724         update_stateid(&stp->st_stid.sc_stateid);
4725         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4726         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
4727                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
4728
4729         nfsd4_client_record_create(oo->oo_owner.so_client);
4730         status = nfs_ok;
4731 put_stateid:
4732         nfs4_put_stid(&stp->st_stid);
4733 out:
4734         nfsd4_bump_seqid(cstate, status);
4735         return status;
4736 }
4737
4738 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
4739 {
4740         if (!test_access(access, stp))
4741                 return;
4742         nfs4_file_put_access(stp->st_stid.sc_file, access);
4743         clear_access(access, stp);
4744 }
4745
4746 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
4747 {
4748         switch (to_access) {
4749         case NFS4_SHARE_ACCESS_READ:
4750                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
4751                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4752                 break;
4753         case NFS4_SHARE_ACCESS_WRITE:
4754                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
4755                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4756                 break;
4757         case NFS4_SHARE_ACCESS_BOTH:
4758                 break;
4759         default:
4760                 WARN_ON_ONCE(1);
4761         }
4762 }
4763
4764 __be32
4765 nfsd4_open_downgrade(struct svc_rqst *rqstp,
4766                      struct nfsd4_compound_state *cstate,
4767                      struct nfsd4_open_downgrade *od)
4768 {
4769         __be32 status;
4770         struct nfs4_ol_stateid *stp;
4771         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4772
4773         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
4774                         cstate->current_fh.fh_dentry);
4775
4776         /* We don't yet support WANT bits: */
4777         if (od->od_deleg_want)
4778                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
4779                         od->od_deleg_want);
4780
4781         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
4782                                         &od->od_stateid, &stp, nn);
4783         if (status)
4784                 goto out;
4785         status = nfserr_inval;
4786         if (!test_access(od->od_share_access, stp)) {
4787                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
4788                         stp->st_access_bmap, od->od_share_access);
4789                 goto put_stateid;
4790         }
4791         if (!test_deny(od->od_share_deny, stp)) {
4792                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
4793                         stp->st_deny_bmap, od->od_share_deny);
4794                 goto put_stateid;
4795         }
4796         nfs4_stateid_downgrade(stp, od->od_share_access);
4797
4798         reset_union_bmap_deny(od->od_share_deny, stp);
4799
4800         update_stateid(&stp->st_stid.sc_stateid);
4801         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4802         status = nfs_ok;
4803 put_stateid:
4804         nfs4_put_stid(&stp->st_stid);
4805 out:
4806         nfsd4_bump_seqid(cstate, status);
4807         return status;
4808 }
4809
4810 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
4811 {
4812         struct nfs4_client *clp = s->st_stid.sc_client;
4813         LIST_HEAD(reaplist);
4814
4815         s->st_stid.sc_type = NFS4_CLOSED_STID;
4816         spin_lock(&clp->cl_lock);
4817         unhash_open_stateid(s, &reaplist);
4818
4819         if (clp->cl_minorversion) {
4820                 put_ol_stateid_locked(s, &reaplist);
4821                 spin_unlock(&clp->cl_lock);
4822                 free_ol_stateid_reaplist(&reaplist);
4823         } else {
4824                 spin_unlock(&clp->cl_lock);
4825                 free_ol_stateid_reaplist(&reaplist);
4826                 move_to_close_lru(s, clp->net);
4827         }
4828 }
4829
4830 /*
4831  * nfs4_unlock_state() called after encode
4832  */
4833 __be32
4834 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4835             struct nfsd4_close *close)
4836 {
4837         __be32 status;
4838         struct nfs4_ol_stateid *stp;
4839         struct net *net = SVC_NET(rqstp);
4840         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4841
4842         dprintk("NFSD: nfsd4_close on file %pd\n",
4843                         cstate->current_fh.fh_dentry);
4844
4845         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4846                                         &close->cl_stateid,
4847                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4848                                         &stp, nn);
4849         nfsd4_bump_seqid(cstate, status);
4850         if (status)
4851                 goto out;
4852         update_stateid(&stp->st_stid.sc_stateid);
4853         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4854
4855         nfsd4_return_all_file_layouts(stp->st_stateowner->so_client,
4856                                       stp->st_stid.sc_file);
4857
4858         nfsd4_close_open_stateid(stp);
4859
4860         /* put reference from nfs4_preprocess_seqid_op */
4861         nfs4_put_stid(&stp->st_stid);
4862 out:
4863         return status;
4864 }
4865
4866 __be32
4867 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4868                   struct nfsd4_delegreturn *dr)
4869 {
4870         struct nfs4_delegation *dp;
4871         stateid_t *stateid = &dr->dr_stateid;
4872         struct nfs4_stid *s;
4873         __be32 status;
4874         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4875
4876         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4877                 return status;
4878
4879         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
4880         if (status)
4881                 goto out;
4882         dp = delegstateid(s);
4883         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4884         if (status)
4885                 goto put_stateid;
4886
4887         destroy_delegation(dp);
4888 put_stateid:
4889         nfs4_put_stid(&dp->dl_stid);
4890 out:
4891         return status;
4892 }
4893
4894
4895 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4896
4897 static inline u64
4898 end_offset(u64 start, u64 len)
4899 {
4900         u64 end;
4901
4902         end = start + len;
4903         return end >= start ? end: NFS4_MAX_UINT64;
4904 }
4905
4906 /* last octet in a range */
4907 static inline u64
4908 last_byte_offset(u64 start, u64 len)
4909 {
4910         u64 end;
4911
4912         WARN_ON_ONCE(!len);
4913         end = start + len;
4914         return end > start ? end - 1: NFS4_MAX_UINT64;
4915 }
4916
4917 /*
4918  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4919  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4920  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4921  * locking, this prevents us from being completely protocol-compliant.  The
4922  * real solution to this problem is to start using unsigned file offsets in
4923  * the VFS, but this is a very deep change!
4924  */
4925 static inline void
4926 nfs4_transform_lock_offset(struct file_lock *lock)
4927 {
4928         if (lock->fl_start < 0)
4929                 lock->fl_start = OFFSET_MAX;
4930         if (lock->fl_end < 0)
4931                 lock->fl_end = OFFSET_MAX;
4932 }
4933
4934 static void nfsd4_fl_get_owner(struct file_lock *dst, struct file_lock *src)
4935 {
4936         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)src->fl_owner;
4937         dst->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lo->lo_owner));
4938 }
4939
4940 static void nfsd4_fl_put_owner(struct file_lock *fl)
4941 {
4942         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
4943
4944         if (lo) {
4945                 nfs4_put_stateowner(&lo->lo_owner);
4946                 fl->fl_owner = NULL;
4947         }
4948 }
4949
4950 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4951         .lm_get_owner = nfsd4_fl_get_owner,
4952         .lm_put_owner = nfsd4_fl_put_owner,
4953 };
4954
4955 static inline void
4956 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4957 {
4958         struct nfs4_lockowner *lo;
4959
4960         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4961                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4962                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4963                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4964                 if (!deny->ld_owner.data)
4965                         /* We just don't care that much */
4966                         goto nevermind;
4967                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4968                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4969         } else {
4970 nevermind:
4971                 deny->ld_owner.len = 0;
4972                 deny->ld_owner.data = NULL;
4973                 deny->ld_clientid.cl_boot = 0;
4974                 deny->ld_clientid.cl_id = 0;
4975         }
4976         deny->ld_start = fl->fl_start;
4977         deny->ld_length = NFS4_MAX_UINT64;
4978         if (fl->fl_end != NFS4_MAX_UINT64)
4979                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4980         deny->ld_type = NFS4_READ_LT;
4981         if (fl->fl_type != F_RDLCK)
4982                 deny->ld_type = NFS4_WRITE_LT;
4983 }
4984
4985 static struct nfs4_lockowner *
4986 find_lockowner_str_locked(clientid_t *clid, struct xdr_netobj *owner,
4987                 struct nfs4_client *clp)
4988 {
4989         unsigned int strhashval = ownerstr_hashval(owner);
4990         struct nfs4_stateowner *so;
4991
4992         lockdep_assert_held(&clp->cl_lock);
4993
4994         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
4995                             so_strhash) {
4996                 if (so->so_is_open_owner)
4997                         continue;
4998                 if (same_owner_str(so, owner))
4999                         return lockowner(nfs4_get_stateowner(so));
5000         }
5001         return NULL;
5002 }
5003
5004 static struct nfs4_lockowner *
5005 find_lockowner_str(clientid_t *clid, struct xdr_netobj *owner,
5006                 struct nfs4_client *clp)
5007 {
5008         struct nfs4_lockowner *lo;
5009
5010         spin_lock(&clp->cl_lock);
5011         lo = find_lockowner_str_locked(clid, owner, clp);
5012         spin_unlock(&clp->cl_lock);
5013         return lo;
5014 }
5015
5016 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5017 {
5018         unhash_lockowner_locked(lockowner(sop));
5019 }
5020
5021 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5022 {
5023         struct nfs4_lockowner *lo = lockowner(sop);
5024
5025         kmem_cache_free(lockowner_slab, lo);
5026 }
5027
5028 static const struct nfs4_stateowner_operations lockowner_ops = {
5029         .so_unhash =    nfs4_unhash_lockowner,
5030         .so_free =      nfs4_free_lockowner,
5031 };
5032
5033 /*
5034  * Alloc a lock owner structure.
5035  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5036  * occurred.
5037  *
5038  * strhashval = ownerstr_hashval
5039  */
5040 static struct nfs4_lockowner *
5041 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5042                            struct nfs4_ol_stateid *open_stp,
5043                            struct nfsd4_lock *lock)
5044 {
5045         struct nfs4_lockowner *lo, *ret;
5046
5047         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5048         if (!lo)
5049                 return NULL;
5050         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5051         lo->lo_owner.so_is_open_owner = 0;
5052         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5053         lo->lo_owner.so_ops = &lockowner_ops;
5054         spin_lock(&clp->cl_lock);
5055         ret = find_lockowner_str_locked(&clp->cl_clientid,
5056                         &lock->lk_new_owner, clp);
5057         if (ret == NULL) {
5058                 list_add(&lo->lo_owner.so_strhash,
5059                          &clp->cl_ownerstr_hashtbl[strhashval]);
5060                 ret = lo;
5061         } else
5062                 nfs4_free_lockowner(&lo->lo_owner);
5063         spin_unlock(&clp->cl_lock);
5064         return ret;
5065 }
5066
5067 static void
5068 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5069                   struct nfs4_file *fp, struct inode *inode,
5070                   struct nfs4_ol_stateid *open_stp)
5071 {
5072         struct nfs4_client *clp = lo->lo_owner.so_client;
5073
5074         lockdep_assert_held(&clp->cl_lock);
5075
5076         atomic_inc(&stp->st_stid.sc_count);
5077         stp->st_stid.sc_type = NFS4_LOCK_STID;
5078         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5079         get_nfs4_file(fp);
5080         stp->st_stid.sc_file = fp;
5081         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5082         stp->st_access_bmap = 0;
5083         stp->st_deny_bmap = open_stp->st_deny_bmap;
5084         stp->st_openstp = open_stp;
5085         list_add(&stp->st_locks, &open_stp->st_locks);
5086         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5087         spin_lock(&fp->fi_lock);
5088         list_add(&stp->st_perfile, &fp->fi_stateids);
5089         spin_unlock(&fp->fi_lock);
5090 }
5091
5092 static struct nfs4_ol_stateid *
5093 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5094 {
5095         struct nfs4_ol_stateid *lst;
5096         struct nfs4_client *clp = lo->lo_owner.so_client;
5097
5098         lockdep_assert_held(&clp->cl_lock);
5099
5100         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5101                 if (lst->st_stid.sc_file == fp) {
5102                         atomic_inc(&lst->st_stid.sc_count);
5103                         return lst;
5104                 }
5105         }
5106         return NULL;
5107 }
5108
5109 static struct nfs4_ol_stateid *
5110 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5111                             struct inode *inode, struct nfs4_ol_stateid *ost,
5112                             bool *new)
5113 {
5114         struct nfs4_stid *ns = NULL;
5115         struct nfs4_ol_stateid *lst;
5116         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5117         struct nfs4_client *clp = oo->oo_owner.so_client;
5118
5119         spin_lock(&clp->cl_lock);
5120         lst = find_lock_stateid(lo, fi);
5121         if (lst == NULL) {
5122                 spin_unlock(&clp->cl_lock);
5123                 ns = nfs4_alloc_stid(clp, stateid_slab);
5124                 if (ns == NULL)
5125                         return NULL;
5126
5127                 spin_lock(&clp->cl_lock);
5128                 lst = find_lock_stateid(lo, fi);
5129                 if (likely(!lst)) {
5130                         lst = openlockstateid(ns);
5131                         init_lock_stateid(lst, lo, fi, inode, ost);
5132                         ns = NULL;
5133                         *new = true;
5134                 }
5135         }
5136         spin_unlock(&clp->cl_lock);
5137         if (ns)
5138                 nfs4_put_stid(ns);
5139         return lst;
5140 }
5141
5142 static int
5143 check_lock_length(u64 offset, u64 length)
5144 {
5145         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
5146              LOFF_OVERFLOW(offset, length)));
5147 }
5148
5149 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5150 {
5151         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5152
5153         lockdep_assert_held(&fp->fi_lock);
5154
5155         if (test_access(access, lock_stp))
5156                 return;
5157         __nfs4_file_get_access(fp, access);
5158         set_access(access, lock_stp);
5159 }
5160
5161 static __be32
5162 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5163                             struct nfs4_ol_stateid *ost,
5164                             struct nfsd4_lock *lock,
5165                             struct nfs4_ol_stateid **lst, bool *new)
5166 {
5167         __be32 status;
5168         struct nfs4_file *fi = ost->st_stid.sc_file;
5169         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5170         struct nfs4_client *cl = oo->oo_owner.so_client;
5171         struct inode *inode = cstate->current_fh.fh_dentry->d_inode;
5172         struct nfs4_lockowner *lo;
5173         unsigned int strhashval;
5174
5175         lo = find_lockowner_str(&cl->cl_clientid, &lock->v.new.owner, cl);
5176         if (!lo) {
5177                 strhashval = ownerstr_hashval(&lock->v.new.owner);
5178                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5179                 if (lo == NULL)
5180                         return nfserr_jukebox;
5181         } else {
5182                 /* with an existing lockowner, seqids must be the same */
5183                 status = nfserr_bad_seqid;
5184                 if (!cstate->minorversion &&
5185                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5186                         goto out;
5187         }
5188
5189         *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5190         if (*lst == NULL) {
5191                 status = nfserr_jukebox;
5192                 goto out;
5193         }
5194         status = nfs_ok;
5195 out:
5196         nfs4_put_stateowner(&lo->lo_owner);
5197         return status;
5198 }
5199
5200 /*
5201  *  LOCK operation
5202  */
5203 __be32
5204 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5205            struct nfsd4_lock *lock)
5206 {
5207         struct nfs4_openowner *open_sop = NULL;
5208         struct nfs4_lockowner *lock_sop = NULL;
5209         struct nfs4_ol_stateid *lock_stp = NULL;
5210         struct nfs4_ol_stateid *open_stp = NULL;
5211         struct nfs4_file *fp;
5212         struct file *filp = NULL;
5213         struct file_lock *file_lock = NULL;
5214         struct file_lock *conflock = NULL;
5215         __be32 status = 0;
5216         int lkflg;
5217         int err;
5218         bool new = false;
5219         struct net *net = SVC_NET(rqstp);
5220         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5221
5222         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5223                 (long long) lock->lk_offset,
5224                 (long long) lock->lk_length);
5225
5226         if (check_lock_length(lock->lk_offset, lock->lk_length))
5227                  return nfserr_inval;
5228
5229         if ((status = fh_verify(rqstp, &cstate->current_fh,
5230                                 S_IFREG, NFSD_MAY_LOCK))) {
5231                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5232                 return status;
5233         }
5234
5235         if (lock->lk_is_new) {
5236                 if (nfsd4_has_session(cstate))
5237                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5238                         memcpy(&lock->v.new.clientid,
5239                                 &cstate->session->se_client->cl_clientid,
5240                                 sizeof(clientid_t));
5241
5242                 status = nfserr_stale_clientid;
5243                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5244                         goto out;
5245
5246                 /* validate and update open stateid and open seqid */
5247                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5248                                         lock->lk_new_open_seqid,
5249                                         &lock->lk_new_open_stateid,
5250                                         &open_stp, nn);
5251                 if (status)
5252                         goto out;
5253                 open_sop = openowner(open_stp->st_stateowner);
5254                 status = nfserr_bad_stateid;
5255                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5256                                                 &lock->v.new.clientid))
5257                         goto out;
5258                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5259                                                         &lock_stp, &new);
5260         } else {
5261                 status = nfs4_preprocess_seqid_op(cstate,
5262                                        lock->lk_old_lock_seqid,
5263                                        &lock->lk_old_lock_stateid,
5264                                        NFS4_LOCK_STID, &lock_stp, nn);
5265         }
5266         if (status)
5267                 goto out;
5268         lock_sop = lockowner(lock_stp->st_stateowner);
5269
5270         lkflg = setlkflg(lock->lk_type);
5271         status = nfs4_check_openmode(lock_stp, lkflg);
5272         if (status)
5273                 goto out;
5274
5275         status = nfserr_grace;
5276         if (locks_in_grace(net) && !lock->lk_reclaim)
5277                 goto out;
5278         status = nfserr_no_grace;
5279         if (!locks_in_grace(net) && lock->lk_reclaim)
5280                 goto out;
5281
5282         file_lock = locks_alloc_lock();
5283         if (!file_lock) {
5284                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5285                 status = nfserr_jukebox;
5286                 goto out;
5287         }
5288
5289         fp = lock_stp->st_stid.sc_file;
5290         switch (lock->lk_type) {
5291                 case NFS4_READ_LT:
5292                 case NFS4_READW_LT:
5293                         spin_lock(&fp->fi_lock);
5294                         filp = find_readable_file_locked(fp);
5295                         if (filp)
5296                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5297                         spin_unlock(&fp->fi_lock);
5298                         file_lock->fl_type = F_RDLCK;
5299                         break;
5300                 case NFS4_WRITE_LT:
5301                 case NFS4_WRITEW_LT:
5302                         spin_lock(&fp->fi_lock);
5303                         filp = find_writeable_file_locked(fp);
5304                         if (filp)
5305                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5306                         spin_unlock(&fp->fi_lock);
5307                         file_lock->fl_type = F_WRLCK;
5308                         break;
5309                 default:
5310                         status = nfserr_inval;
5311                 goto out;
5312         }
5313         if (!filp) {
5314                 status = nfserr_openmode;
5315                 goto out;
5316         }
5317
5318         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5319         file_lock->fl_pid = current->tgid;
5320         file_lock->fl_file = filp;
5321         file_lock->fl_flags = FL_POSIX;
5322         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5323         file_lock->fl_start = lock->lk_offset;
5324         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5325         nfs4_transform_lock_offset(file_lock);
5326
5327         conflock = locks_alloc_lock();
5328         if (!conflock) {
5329                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5330                 status = nfserr_jukebox;
5331                 goto out;
5332         }
5333
5334         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5335         switch (-err) {
5336         case 0: /* success! */
5337                 update_stateid(&lock_stp->st_stid.sc_stateid);
5338                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
5339                                 sizeof(stateid_t));
5340                 status = 0;
5341                 break;
5342         case (EAGAIN):          /* conflock holds conflicting lock */
5343                 status = nfserr_denied;
5344                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5345                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5346                 break;
5347         case (EDEADLK):
5348                 status = nfserr_deadlock;
5349                 break;
5350         default:
5351                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5352                 status = nfserrno(err);
5353                 break;
5354         }
5355 out:
5356         if (filp)
5357                 fput(filp);
5358         if (lock_stp) {
5359                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5360                 if (cstate->replay_owner &&
5361                     cstate->replay_owner != &lock_sop->lo_owner &&
5362                     seqid_mutating_err(ntohl(status)))
5363                         lock_sop->lo_owner.so_seqid++;
5364
5365                 /*
5366                  * If this is a new, never-before-used stateid, and we are
5367                  * returning an error, then just go ahead and release it.
5368                  */
5369                 if (status && new)
5370                         release_lock_stateid(lock_stp);
5371
5372                 nfs4_put_stid(&lock_stp->st_stid);
5373         }
5374         if (open_stp)
5375                 nfs4_put_stid(&open_stp->st_stid);
5376         nfsd4_bump_seqid(cstate, status);
5377         if (file_lock)
5378                 locks_free_lock(file_lock);
5379         if (conflock)
5380                 locks_free_lock(conflock);
5381         return status;
5382 }
5383
5384 /*
5385  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5386  * so we do a temporary open here just to get an open file to pass to
5387  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5388  * inode operation.)
5389  */
5390 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5391 {
5392         struct file *file;
5393         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5394         if (!err) {
5395                 err = nfserrno(vfs_test_lock(file, lock));
5396                 nfsd_close(file);
5397         }
5398         return err;
5399 }
5400
5401 /*
5402  * LOCKT operation
5403  */
5404 __be32
5405 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5406             struct nfsd4_lockt *lockt)
5407 {
5408         struct file_lock *file_lock = NULL;
5409         struct nfs4_lockowner *lo = NULL;
5410         __be32 status;
5411         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5412
5413         if (locks_in_grace(SVC_NET(rqstp)))
5414                 return nfserr_grace;
5415
5416         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5417                  return nfserr_inval;
5418
5419         if (!nfsd4_has_session(cstate)) {
5420                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5421                 if (status)
5422                         goto out;
5423         }
5424
5425         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5426                 goto out;
5427
5428         file_lock = locks_alloc_lock();
5429         if (!file_lock) {
5430                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5431                 status = nfserr_jukebox;
5432                 goto out;
5433         }
5434
5435         switch (lockt->lt_type) {
5436                 case NFS4_READ_LT:
5437                 case NFS4_READW_LT:
5438                         file_lock->fl_type = F_RDLCK;
5439                 break;
5440                 case NFS4_WRITE_LT:
5441                 case NFS4_WRITEW_LT:
5442                         file_lock->fl_type = F_WRLCK;
5443                 break;
5444                 default:
5445                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5446                         status = nfserr_inval;
5447                 goto out;
5448         }
5449
5450         lo = find_lockowner_str(&lockt->lt_clientid, &lockt->lt_owner,
5451                                 cstate->clp);
5452         if (lo)
5453                 file_lock->fl_owner = (fl_owner_t)lo;
5454         file_lock->fl_pid = current->tgid;
5455         file_lock->fl_flags = FL_POSIX;
5456
5457         file_lock->fl_start = lockt->lt_offset;
5458         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5459
5460         nfs4_transform_lock_offset(file_lock);
5461
5462         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5463         if (status)
5464                 goto out;
5465
5466         if (file_lock->fl_type != F_UNLCK) {
5467                 status = nfserr_denied;
5468                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5469         }
5470 out:
5471         if (lo)
5472                 nfs4_put_stateowner(&lo->lo_owner);
5473         if (file_lock)
5474                 locks_free_lock(file_lock);
5475         return status;
5476 }
5477
5478 __be32
5479 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5480             struct nfsd4_locku *locku)
5481 {
5482         struct nfs4_ol_stateid *stp;
5483         struct file *filp = NULL;
5484         struct file_lock *file_lock = NULL;
5485         __be32 status;
5486         int err;
5487         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5488
5489         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5490                 (long long) locku->lu_offset,
5491                 (long long) locku->lu_length);
5492
5493         if (check_lock_length(locku->lu_offset, locku->lu_length))
5494                  return nfserr_inval;
5495
5496         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5497                                         &locku->lu_stateid, NFS4_LOCK_STID,
5498                                         &stp, nn);
5499         if (status)
5500                 goto out;
5501         filp = find_any_file(stp->st_stid.sc_file);
5502         if (!filp) {
5503                 status = nfserr_lock_range;
5504                 goto put_stateid;
5505         }
5506         file_lock = locks_alloc_lock();
5507         if (!file_lock) {
5508                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5509                 status = nfserr_jukebox;
5510                 goto fput;
5511         }
5512
5513         file_lock->fl_type = F_UNLCK;
5514         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5515         file_lock->fl_pid = current->tgid;
5516         file_lock->fl_file = filp;
5517         file_lock->fl_flags = FL_POSIX;
5518         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5519         file_lock->fl_start = locku->lu_offset;
5520
5521         file_lock->fl_end = last_byte_offset(locku->lu_offset,
5522                                                 locku->lu_length);
5523         nfs4_transform_lock_offset(file_lock);
5524
5525         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5526         if (err) {
5527                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5528                 goto out_nfserr;
5529         }
5530         update_stateid(&stp->st_stid.sc_stateid);
5531         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
5532 fput:
5533         fput(filp);
5534 put_stateid:
5535         nfs4_put_stid(&stp->st_stid);
5536 out:
5537         nfsd4_bump_seqid(cstate, status);
5538         if (file_lock)
5539                 locks_free_lock(file_lock);
5540         return status;
5541
5542 out_nfserr:
5543         status = nfserrno(err);
5544         goto fput;
5545 }
5546
5547 /*
5548  * returns
5549  *      true:  locks held by lockowner
5550  *      false: no locks held by lockowner
5551  */
5552 static bool
5553 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5554 {
5555         struct file_lock *fl;
5556         int status = false;
5557         struct file *filp = find_any_file(fp);
5558         struct inode *inode;
5559         struct file_lock_context *flctx;
5560
5561         if (!filp) {
5562                 /* Any valid lock stateid should have some sort of access */
5563                 WARN_ON_ONCE(1);
5564                 return status;
5565         }
5566
5567         inode = file_inode(filp);
5568         flctx = inode->i_flctx;
5569
5570         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
5571                 spin_lock(&flctx->flc_lock);
5572                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
5573                         if (fl->fl_owner == (fl_owner_t)lowner) {
5574                                 status = true;
5575                                 break;
5576                         }
5577                 }
5578                 spin_unlock(&flctx->flc_lock);
5579         }
5580         fput(filp);
5581         return status;
5582 }
5583
5584 __be32
5585 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5586                         struct nfsd4_compound_state *cstate,
5587                         struct nfsd4_release_lockowner *rlockowner)
5588 {
5589         clientid_t *clid = &rlockowner->rl_clientid;
5590         struct nfs4_stateowner *sop;
5591         struct nfs4_lockowner *lo = NULL;
5592         struct nfs4_ol_stateid *stp;
5593         struct xdr_netobj *owner = &rlockowner->rl_owner;
5594         unsigned int hashval = ownerstr_hashval(owner);
5595         __be32 status;
5596         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5597         struct nfs4_client *clp;
5598
5599         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5600                 clid->cl_boot, clid->cl_id);
5601
5602         status = lookup_clientid(clid, cstate, nn);
5603         if (status)
5604                 return status;
5605
5606         clp = cstate->clp;
5607         /* Find the matching lock stateowner */
5608         spin_lock(&clp->cl_lock);
5609         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5610                             so_strhash) {
5611
5612                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5613                         continue;
5614
5615                 /* see if there are still any locks associated with it */
5616                 lo = lockowner(sop);
5617                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5618                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
5619                                 status = nfserr_locks_held;
5620                                 spin_unlock(&clp->cl_lock);
5621                                 return status;
5622                         }
5623                 }
5624
5625                 nfs4_get_stateowner(sop);
5626                 break;
5627         }
5628         spin_unlock(&clp->cl_lock);
5629         if (lo)
5630                 release_lockowner(lo);
5631         return status;
5632 }
5633
5634 static inline struct nfs4_client_reclaim *
5635 alloc_reclaim(void)
5636 {
5637         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5638 }
5639
5640 bool
5641 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5642 {
5643         struct nfs4_client_reclaim *crp;
5644
5645         crp = nfsd4_find_reclaim_client(name, nn);
5646         return (crp && crp->cr_clp);
5647 }
5648
5649 /*
5650  * failure => all reset bets are off, nfserr_no_grace...
5651  */
5652 struct nfs4_client_reclaim *
5653 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5654 {
5655         unsigned int strhashval;
5656         struct nfs4_client_reclaim *crp;
5657
5658         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5659         crp = alloc_reclaim();
5660         if (crp) {
5661                 strhashval = clientstr_hashval(name);
5662                 INIT_LIST_HEAD(&crp->cr_strhash);
5663                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5664                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5665                 crp->cr_clp = NULL;
5666                 nn->reclaim_str_hashtbl_size++;
5667         }
5668         return crp;
5669 }
5670
5671 void
5672 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5673 {
5674         list_del(&crp->cr_strhash);
5675         kfree(crp);
5676         nn->reclaim_str_hashtbl_size--;
5677 }
5678
5679 void
5680 nfs4_release_reclaim(struct nfsd_net *nn)
5681 {
5682         struct nfs4_client_reclaim *crp = NULL;
5683         int i;
5684
5685         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5686                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5687                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5688                                         struct nfs4_client_reclaim, cr_strhash);
5689                         nfs4_remove_reclaim_record(crp, nn);
5690                 }
5691         }
5692         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5693 }
5694
5695 /*
5696  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5697 struct nfs4_client_reclaim *
5698 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5699 {
5700         unsigned int strhashval;
5701         struct nfs4_client_reclaim *crp = NULL;
5702
5703         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5704
5705         strhashval = clientstr_hashval(recdir);
5706         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5707                 if (same_name(crp->cr_recdir, recdir)) {
5708                         return crp;
5709                 }
5710         }
5711         return NULL;
5712 }
5713
5714 /*
5715 * Called from OPEN. Look for clientid in reclaim list.
5716 */
5717 __be32
5718 nfs4_check_open_reclaim(clientid_t *clid,
5719                 struct nfsd4_compound_state *cstate,
5720                 struct nfsd_net *nn)
5721 {
5722         __be32 status;
5723
5724         /* find clientid in conf_id_hashtbl */
5725         status = lookup_clientid(clid, cstate, nn);
5726         if (status)
5727                 return nfserr_reclaim_bad;
5728
5729         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
5730                 return nfserr_no_grace;
5731
5732         if (nfsd4_client_record_check(cstate->clp))
5733                 return nfserr_reclaim_bad;
5734
5735         return nfs_ok;
5736 }
5737
5738 #ifdef CONFIG_NFSD_FAULT_INJECTION
5739 static inline void
5740 put_client(struct nfs4_client *clp)
5741 {
5742         atomic_dec(&clp->cl_refcount);
5743 }
5744
5745 static struct nfs4_client *
5746 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
5747 {
5748         struct nfs4_client *clp;
5749         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5750                                           nfsd_net_id);
5751
5752         if (!nfsd_netns_ready(nn))
5753                 return NULL;
5754
5755         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5756                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
5757                         return clp;
5758         }
5759         return NULL;
5760 }
5761
5762 u64
5763 nfsd_inject_print_clients(void)
5764 {
5765         struct nfs4_client *clp;
5766         u64 count = 0;
5767         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5768                                           nfsd_net_id);
5769         char buf[INET6_ADDRSTRLEN];
5770
5771         if (!nfsd_netns_ready(nn))
5772                 return 0;
5773
5774         spin_lock(&nn->client_lock);
5775         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5776                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5777                 pr_info("NFS Client: %s\n", buf);
5778                 ++count;
5779         }
5780         spin_unlock(&nn->client_lock);
5781
5782         return count;
5783 }
5784
5785 u64
5786 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
5787 {
5788         u64 count = 0;
5789         struct nfs4_client *clp;
5790         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5791                                           nfsd_net_id);
5792
5793         if (!nfsd_netns_ready(nn))
5794                 return count;
5795
5796         spin_lock(&nn->client_lock);
5797         clp = nfsd_find_client(addr, addr_size);
5798         if (clp) {
5799                 if (mark_client_expired_locked(clp) == nfs_ok)
5800                         ++count;
5801                 else
5802                         clp = NULL;
5803         }
5804         spin_unlock(&nn->client_lock);
5805
5806         if (clp)
5807                 expire_client(clp);
5808
5809         return count;
5810 }
5811
5812 u64
5813 nfsd_inject_forget_clients(u64 max)
5814 {
5815         u64 count = 0;
5816         struct nfs4_client *clp, *next;
5817         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5818                                                 nfsd_net_id);
5819         LIST_HEAD(reaplist);
5820
5821         if (!nfsd_netns_ready(nn))
5822                 return count;
5823
5824         spin_lock(&nn->client_lock);
5825         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
5826                 if (mark_client_expired_locked(clp) == nfs_ok) {
5827                         list_add(&clp->cl_lru, &reaplist);
5828                         if (max != 0 && ++count >= max)
5829                                 break;
5830                 }
5831         }
5832         spin_unlock(&nn->client_lock);
5833
5834         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
5835                 expire_client(clp);
5836
5837         return count;
5838 }
5839
5840 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
5841                              const char *type)
5842 {
5843         char buf[INET6_ADDRSTRLEN];
5844         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5845         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
5846 }
5847
5848 static void
5849 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
5850                              struct list_head *collect)
5851 {
5852         struct nfs4_client *clp = lst->st_stid.sc_client;
5853         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5854                                           nfsd_net_id);
5855
5856         if (!collect)
5857                 return;
5858
5859         lockdep_assert_held(&nn->client_lock);
5860         atomic_inc(&clp->cl_refcount);
5861         list_add(&lst->st_locks, collect);
5862 }
5863
5864 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
5865                                     struct list_head *collect,
5866                                     void (*func)(struct nfs4_ol_stateid *))
5867 {
5868         struct nfs4_openowner *oop;
5869         struct nfs4_ol_stateid *stp, *st_next;
5870         struct nfs4_ol_stateid *lst, *lst_next;
5871         u64 count = 0;
5872
5873         spin_lock(&clp->cl_lock);
5874         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
5875                 list_for_each_entry_safe(stp, st_next,
5876                                 &oop->oo_owner.so_stateids, st_perstateowner) {
5877                         list_for_each_entry_safe(lst, lst_next,
5878                                         &stp->st_locks, st_locks) {
5879                                 if (func) {
5880                                         func(lst);
5881                                         nfsd_inject_add_lock_to_list(lst,
5882                                                                 collect);
5883                                 }
5884                                 ++count;
5885                                 /*
5886                                  * Despite the fact that these functions deal
5887                                  * with 64-bit integers for "count", we must
5888                                  * ensure that it doesn't blow up the
5889                                  * clp->cl_refcount. Throw a warning if we
5890                                  * start to approach INT_MAX here.
5891                                  */
5892                                 WARN_ON_ONCE(count == (INT_MAX / 2));
5893                                 if (count == max)
5894                                         goto out;
5895                         }
5896                 }
5897         }
5898 out:
5899         spin_unlock(&clp->cl_lock);
5900
5901         return count;
5902 }
5903
5904 static u64
5905 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
5906                           u64 max)
5907 {
5908         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
5909 }
5910
5911 static u64
5912 nfsd_print_client_locks(struct nfs4_client *clp)
5913 {
5914         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
5915         nfsd_print_count(clp, count, "locked files");
5916         return count;
5917 }
5918
5919 u64
5920 nfsd_inject_print_locks(void)
5921 {
5922         struct nfs4_client *clp;
5923         u64 count = 0;
5924         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5925                                                 nfsd_net_id);
5926
5927         if (!nfsd_netns_ready(nn))
5928                 return 0;
5929
5930         spin_lock(&nn->client_lock);
5931         list_for_each_entry(clp, &nn->client_lru, cl_lru)
5932                 count += nfsd_print_client_locks(clp);
5933         spin_unlock(&nn->client_lock);
5934
5935         return count;
5936 }
5937
5938 static void
5939 nfsd_reap_locks(struct list_head *reaplist)
5940 {
5941         struct nfs4_client *clp;
5942         struct nfs4_ol_stateid *stp, *next;
5943
5944         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
5945                 list_del_init(&stp->st_locks);
5946                 clp = stp->st_stid.sc_client;
5947                 nfs4_put_stid(&stp->st_stid);
5948                 put_client(clp);
5949         }
5950 }
5951
5952 u64
5953 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
5954 {
5955         unsigned int count = 0;
5956         struct nfs4_client *clp;
5957         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5958                                                 nfsd_net_id);
5959         LIST_HEAD(reaplist);
5960
5961         if (!nfsd_netns_ready(nn))
5962                 return count;
5963
5964         spin_lock(&nn->client_lock);
5965         clp = nfsd_find_client(addr, addr_size);
5966         if (clp)
5967                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
5968         spin_unlock(&nn->client_lock);
5969         nfsd_reap_locks(&reaplist);
5970         return count;
5971 }
5972
5973 u64
5974 nfsd_inject_forget_locks(u64 max)
5975 {
5976         u64 count = 0;
5977         struct nfs4_client *clp;
5978         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5979                                                 nfsd_net_id);
5980         LIST_HEAD(reaplist);
5981
5982         if (!nfsd_netns_ready(nn))
5983                 return count;
5984
5985         spin_lock(&nn->client_lock);
5986         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5987                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
5988                 if (max != 0 && count >= max)
5989                         break;
5990         }
5991         spin_unlock(&nn->client_lock);
5992         nfsd_reap_locks(&reaplist);
5993         return count;
5994 }
5995
5996 static u64
5997 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
5998                               struct list_head *collect,
5999                               void (*func)(struct nfs4_openowner *))
6000 {
6001         struct nfs4_openowner *oop, *next;
6002         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6003                                                 nfsd_net_id);
6004         u64 count = 0;
6005
6006         lockdep_assert_held(&nn->client_lock);
6007
6008         spin_lock(&clp->cl_lock);
6009         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6010                 if (func) {
6011                         func(oop);
6012                         if (collect) {
6013                                 atomic_inc(&clp->cl_refcount);
6014                                 list_add(&oop->oo_perclient, collect);
6015                         }
6016                 }
6017                 ++count;
6018                 /*
6019                  * Despite the fact that these functions deal with
6020                  * 64-bit integers for "count", we must ensure that
6021                  * it doesn't blow up the clp->cl_refcount. Throw a
6022                  * warning if we start to approach INT_MAX here.
6023                  */
6024                 WARN_ON_ONCE(count == (INT_MAX / 2));
6025                 if (count == max)
6026                         break;
6027         }
6028         spin_unlock(&clp->cl_lock);
6029
6030         return count;
6031 }
6032
6033 static u64
6034 nfsd_print_client_openowners(struct nfs4_client *clp)
6035 {
6036         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6037
6038         nfsd_print_count(clp, count, "openowners");
6039         return count;
6040 }
6041
6042 static u64
6043 nfsd_collect_client_openowners(struct nfs4_client *clp,
6044                                struct list_head *collect, u64 max)
6045 {
6046         return nfsd_foreach_client_openowner(clp, max, collect,
6047                                                 unhash_openowner_locked);
6048 }
6049
6050 u64
6051 nfsd_inject_print_openowners(void)
6052 {
6053         struct nfs4_client *clp;
6054         u64 count = 0;
6055         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6056                                                 nfsd_net_id);
6057
6058         if (!nfsd_netns_ready(nn))
6059                 return 0;
6060
6061         spin_lock(&nn->client_lock);
6062         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6063                 count += nfsd_print_client_openowners(clp);
6064         spin_unlock(&nn->client_lock);
6065
6066         return count;
6067 }
6068
6069 static void
6070 nfsd_reap_openowners(struct list_head *reaplist)
6071 {
6072         struct nfs4_client *clp;
6073         struct nfs4_openowner *oop, *next;
6074
6075         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6076                 list_del_init(&oop->oo_perclient);
6077                 clp = oop->oo_owner.so_client;
6078                 release_openowner(oop);
6079                 put_client(clp);
6080         }
6081 }
6082
6083 u64
6084 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6085                                      size_t addr_size)
6086 {
6087         unsigned int count = 0;
6088         struct nfs4_client *clp;
6089         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6090                                                 nfsd_net_id);
6091         LIST_HEAD(reaplist);
6092
6093         if (!nfsd_netns_ready(nn))
6094                 return count;
6095
6096         spin_lock(&nn->client_lock);
6097         clp = nfsd_find_client(addr, addr_size);
6098         if (clp)
6099                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6100         spin_unlock(&nn->client_lock);
6101         nfsd_reap_openowners(&reaplist);
6102         return count;
6103 }
6104
6105 u64
6106 nfsd_inject_forget_openowners(u64 max)
6107 {
6108         u64 count = 0;
6109         struct nfs4_client *clp;
6110         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6111                                                 nfsd_net_id);
6112         LIST_HEAD(reaplist);
6113
6114         if (!nfsd_netns_ready(nn))
6115                 return count;
6116
6117         spin_lock(&nn->client_lock);
6118         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6119                 count += nfsd_collect_client_openowners(clp, &reaplist,
6120                                                         max - count);
6121                 if (max != 0 && count >= max)
6122                         break;
6123         }
6124         spin_unlock(&nn->client_lock);
6125         nfsd_reap_openowners(&reaplist);
6126         return count;
6127 }
6128
6129 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6130                                      struct list_head *victims)
6131 {
6132         struct nfs4_delegation *dp, *next;
6133         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6134                                                 nfsd_net_id);
6135         u64 count = 0;
6136
6137         lockdep_assert_held(&nn->client_lock);
6138
6139         spin_lock(&state_lock);
6140         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6141                 if (victims) {
6142                         /*
6143                          * It's not safe to mess with delegations that have a
6144                          * non-zero dl_time. They might have already been broken
6145                          * and could be processed by the laundromat outside of
6146                          * the state_lock. Just leave them be.
6147                          */
6148                         if (dp->dl_time != 0)
6149                                 continue;
6150
6151                         atomic_inc(&clp->cl_refcount);
6152                         unhash_delegation_locked(dp);
6153                         list_add(&dp->dl_recall_lru, victims);
6154                 }
6155                 ++count;
6156                 /*
6157                  * Despite the fact that these functions deal with
6158                  * 64-bit integers for "count", we must ensure that
6159                  * it doesn't blow up the clp->cl_refcount. Throw a
6160                  * warning if we start to approach INT_MAX here.
6161                  */
6162                 WARN_ON_ONCE(count == (INT_MAX / 2));
6163                 if (count == max)
6164                         break;
6165         }
6166         spin_unlock(&state_lock);
6167         return count;
6168 }
6169
6170 static u64
6171 nfsd_print_client_delegations(struct nfs4_client *clp)
6172 {
6173         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6174
6175         nfsd_print_count(clp, count, "delegations");
6176         return count;
6177 }
6178
6179 u64
6180 nfsd_inject_print_delegations(void)
6181 {
6182         struct nfs4_client *clp;
6183         u64 count = 0;
6184         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6185                                                 nfsd_net_id);
6186
6187         if (!nfsd_netns_ready(nn))
6188                 return 0;
6189
6190         spin_lock(&nn->client_lock);
6191         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6192                 count += nfsd_print_client_delegations(clp);
6193         spin_unlock(&nn->client_lock);
6194
6195         return count;
6196 }
6197
6198 static void
6199 nfsd_forget_delegations(struct list_head *reaplist)
6200 {
6201         struct nfs4_client *clp;
6202         struct nfs4_delegation *dp, *next;
6203
6204         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6205                 list_del_init(&dp->dl_recall_lru);
6206                 clp = dp->dl_stid.sc_client;
6207                 revoke_delegation(dp);
6208                 put_client(clp);
6209         }
6210 }
6211
6212 u64
6213 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6214                                       size_t addr_size)
6215 {
6216         u64 count = 0;
6217         struct nfs4_client *clp;
6218         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6219                                                 nfsd_net_id);
6220         LIST_HEAD(reaplist);
6221
6222         if (!nfsd_netns_ready(nn))
6223                 return count;
6224
6225         spin_lock(&nn->client_lock);
6226         clp = nfsd_find_client(addr, addr_size);
6227         if (clp)
6228                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6229         spin_unlock(&nn->client_lock);
6230
6231         nfsd_forget_delegations(&reaplist);
6232         return count;
6233 }
6234
6235 u64
6236 nfsd_inject_forget_delegations(u64 max)
6237 {
6238         u64 count = 0;
6239         struct nfs4_client *clp;
6240         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6241                                                 nfsd_net_id);
6242         LIST_HEAD(reaplist);
6243
6244         if (!nfsd_netns_ready(nn))
6245                 return count;
6246
6247         spin_lock(&nn->client_lock);
6248         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6249                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6250                 if (max != 0 && count >= max)
6251                         break;
6252         }
6253         spin_unlock(&nn->client_lock);
6254         nfsd_forget_delegations(&reaplist);
6255         return count;
6256 }
6257
6258 static void
6259 nfsd_recall_delegations(struct list_head *reaplist)
6260 {
6261         struct nfs4_client *clp;
6262         struct nfs4_delegation *dp, *next;
6263
6264         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6265                 list_del_init(&dp->dl_recall_lru);
6266                 clp = dp->dl_stid.sc_client;
6267                 /*
6268                  * We skipped all entries that had a zero dl_time before,
6269                  * so we can now reset the dl_time back to 0. If a delegation
6270                  * break comes in now, then it won't make any difference since
6271                  * we're recalling it either way.
6272                  */
6273                 spin_lock(&state_lock);
6274                 dp->dl_time = 0;
6275                 spin_unlock(&state_lock);
6276                 nfsd_break_one_deleg(dp);
6277                 put_client(clp);
6278         }
6279 }
6280
6281 u64
6282 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6283                                       size_t addr_size)
6284 {
6285         u64 count = 0;
6286         struct nfs4_client *clp;
6287         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6288                                                 nfsd_net_id);
6289         LIST_HEAD(reaplist);
6290
6291         if (!nfsd_netns_ready(nn))
6292                 return count;
6293
6294         spin_lock(&nn->client_lock);
6295         clp = nfsd_find_client(addr, addr_size);
6296         if (clp)
6297                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6298         spin_unlock(&nn->client_lock);
6299
6300         nfsd_recall_delegations(&reaplist);
6301         return count;
6302 }
6303
6304 u64
6305 nfsd_inject_recall_delegations(u64 max)
6306 {
6307         u64 count = 0;
6308         struct nfs4_client *clp, *next;
6309         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6310                                                 nfsd_net_id);
6311         LIST_HEAD(reaplist);
6312
6313         if (!nfsd_netns_ready(nn))
6314                 return count;
6315
6316         spin_lock(&nn->client_lock);
6317         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6318                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6319                 if (max != 0 && ++count >= max)
6320                         break;
6321         }
6322         spin_unlock(&nn->client_lock);
6323         nfsd_recall_delegations(&reaplist);
6324         return count;
6325 }
6326 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6327
6328 /*
6329  * Since the lifetime of a delegation isn't limited to that of an open, a
6330  * client may quite reasonably hang on to a delegation as long as it has
6331  * the inode cached.  This becomes an obvious problem the first time a
6332  * client's inode cache approaches the size of the server's total memory.
6333  *
6334  * For now we avoid this problem by imposing a hard limit on the number
6335  * of delegations, which varies according to the server's memory size.
6336  */
6337 static void
6338 set_max_delegations(void)
6339 {
6340         /*
6341          * Allow at most 4 delegations per megabyte of RAM.  Quick
6342          * estimates suggest that in the worst case (where every delegation
6343          * is for a different inode), a delegation could take about 1.5K,
6344          * giving a worst case usage of about 6% of memory.
6345          */
6346         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6347 }
6348
6349 static int nfs4_state_create_net(struct net *net)
6350 {
6351         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6352         int i;
6353
6354         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6355                         CLIENT_HASH_SIZE, GFP_KERNEL);
6356         if (!nn->conf_id_hashtbl)
6357                 goto err;
6358         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6359                         CLIENT_HASH_SIZE, GFP_KERNEL);
6360         if (!nn->unconf_id_hashtbl)
6361                 goto err_unconf_id;
6362         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6363                         SESSION_HASH_SIZE, GFP_KERNEL);
6364         if (!nn->sessionid_hashtbl)
6365                 goto err_sessionid;
6366
6367         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6368                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6369                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6370         }
6371         for (i = 0; i < SESSION_HASH_SIZE; i++)
6372                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6373         nn->conf_name_tree = RB_ROOT;
6374         nn->unconf_name_tree = RB_ROOT;
6375         INIT_LIST_HEAD(&nn->client_lru);
6376         INIT_LIST_HEAD(&nn->close_lru);
6377         INIT_LIST_HEAD(&nn->del_recall_lru);
6378         spin_lock_init(&nn->client_lock);
6379
6380         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6381         get_net(net);
6382
6383         return 0;
6384
6385 err_sessionid:
6386         kfree(nn->unconf_id_hashtbl);
6387 err_unconf_id:
6388         kfree(nn->conf_id_hashtbl);
6389 err:
6390         return -ENOMEM;
6391 }
6392
6393 static void
6394 nfs4_state_destroy_net(struct net *net)
6395 {
6396         int i;
6397         struct nfs4_client *clp = NULL;
6398         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6399
6400         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6401                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6402                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6403                         destroy_client(clp);
6404                 }
6405         }
6406
6407         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6408                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6409                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6410                         destroy_client(clp);
6411                 }
6412         }
6413
6414         kfree(nn->sessionid_hashtbl);
6415         kfree(nn->unconf_id_hashtbl);
6416         kfree(nn->conf_id_hashtbl);
6417         put_net(net);
6418 }
6419
6420 int
6421 nfs4_state_start_net(struct net *net)
6422 {
6423         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6424         int ret;
6425
6426         ret = nfs4_state_create_net(net);
6427         if (ret)
6428                 return ret;
6429         nn->boot_time = get_seconds();
6430         nn->grace_ended = false;
6431         locks_start_grace(net, &nn->nfsd4_manager);
6432         nfsd4_client_tracking_init(net);
6433         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6434                nn->nfsd4_grace, net);
6435         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6436         return 0;
6437 }
6438
6439 /* initialization to perform when the nfsd service is started: */
6440
6441 int
6442 nfs4_state_start(void)
6443 {
6444         int ret;
6445
6446         ret = set_callback_cred();
6447         if (ret)
6448                 return -ENOMEM;
6449         laundry_wq = create_singlethread_workqueue("nfsd4");
6450         if (laundry_wq == NULL) {
6451                 ret = -ENOMEM;
6452                 goto out_recovery;
6453         }
6454         ret = nfsd4_create_callback_queue();
6455         if (ret)
6456                 goto out_free_laundry;
6457
6458         set_max_delegations();
6459
6460         return 0;
6461
6462 out_free_laundry:
6463         destroy_workqueue(laundry_wq);
6464 out_recovery:
6465         return ret;
6466 }
6467
6468 void
6469 nfs4_state_shutdown_net(struct net *net)
6470 {
6471         struct nfs4_delegation *dp = NULL;
6472         struct list_head *pos, *next, reaplist;
6473         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6474
6475         cancel_delayed_work_sync(&nn->laundromat_work);
6476         locks_end_grace(&nn->nfsd4_manager);
6477
6478         INIT_LIST_HEAD(&reaplist);
6479         spin_lock(&state_lock);
6480         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6481                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6482                 unhash_delegation_locked(dp);
6483                 list_add(&dp->dl_recall_lru, &reaplist);
6484         }
6485         spin_unlock(&state_lock);
6486         list_for_each_safe(pos, next, &reaplist) {
6487                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6488                 list_del_init(&dp->dl_recall_lru);
6489                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6490                 nfs4_put_stid(&dp->dl_stid);
6491         }
6492
6493         nfsd4_client_tracking_exit(net);
6494         nfs4_state_destroy_net(net);
6495 }
6496
6497 void
6498 nfs4_state_shutdown(void)
6499 {
6500         destroy_workqueue(laundry_wq);
6501         nfsd4_destroy_callback_queue();
6502 }
6503
6504 static void
6505 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6506 {
6507         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6508                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6509 }
6510
6511 static void
6512 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6513 {
6514         if (cstate->minorversion) {
6515                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6516                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6517         }
6518 }
6519
6520 void
6521 clear_current_stateid(struct nfsd4_compound_state *cstate)
6522 {
6523         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6524 }
6525
6526 /*
6527  * functions to set current state id
6528  */
6529 void
6530 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6531 {
6532         put_stateid(cstate, &odp->od_stateid);
6533 }
6534
6535 void
6536 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6537 {
6538         put_stateid(cstate, &open->op_stateid);
6539 }
6540
6541 void
6542 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6543 {
6544         put_stateid(cstate, &close->cl_stateid);
6545 }
6546
6547 void
6548 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6549 {
6550         put_stateid(cstate, &lock->lk_resp_stateid);
6551 }
6552
6553 /*
6554  * functions to consume current state id
6555  */
6556
6557 void
6558 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6559 {
6560         get_stateid(cstate, &odp->od_stateid);
6561 }
6562
6563 void
6564 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6565 {
6566         get_stateid(cstate, &drp->dr_stateid);
6567 }
6568
6569 void
6570 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6571 {
6572         get_stateid(cstate, &fsp->fr_stateid);
6573 }
6574
6575 void
6576 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6577 {
6578         get_stateid(cstate, &setattr->sa_stateid);
6579 }
6580
6581 void
6582 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6583 {
6584         get_stateid(cstate, &close->cl_stateid);
6585 }
6586
6587 void
6588 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6589 {
6590         get_stateid(cstate, &locku->lu_stateid);
6591 }
6592
6593 void
6594 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6595 {
6596         get_stateid(cstate, &read->rd_stateid);
6597 }
6598
6599 void
6600 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6601 {
6602         get_stateid(cstate, &write->wr_stateid);
6603 }