fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include <linux/hash.h>
  45 #include "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51
  52 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  53
  54 #define all_ones {{~0,~0},~0}
  55 static const stateid_t one_stateid = {
  56         .si_generation = ~0,
  57         .si_opaque = all_ones,
  58 };
  59 static const stateid_t zero_stateid = {
  60         /* all fields zero */
  61 };
  62 static const stateid_t currentstateid = {
  63         .si_generation = 1,
  64 };
  65
  66 static u64 current_sessionid = 1;
  67
  68 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  69 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  70 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  71
  72 /* forward declarations */
  73 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  74 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  75
  76 /* Locking: */
  77
  78 /* Currently used for almost all code touching nfsv4 state: */
  79 static DEFINE_MUTEX(client_mutex);
  80
  81 /*
  82  * Currently used for the del_recall_lru and file hash table.  In an
  83  * effort to decrease the scope of the client_mutex, this spinlock may
  84  * eventually cover more:
  85  */
  86 static DEFINE_SPINLOCK(state_lock);
  87
  88 static struct kmem_cache *openowner_slab;
  89 static struct kmem_cache *lockowner_slab;
  90 static struct kmem_cache *file_slab;
  91 static struct kmem_cache *stateid_slab;
  92 static struct kmem_cache *deleg_slab;
  93
  94 void
  95 nfs4_lock_state(void)
  96 {
  97         mutex_lock(&client_mutex);
  98 }
  99
 100 static void free_session(struct nfsd4_session *);
 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 void
 116 nfs4_unlock_state(void)
 117 {
 118         mutex_unlock(&client_mutex);
 119 }
 120
 121 static bool is_client_expired(struct nfs4_client *clp)
 122 {
 123         return clp->cl_time == 0;
 124 }
 125
 126 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
 127 {
 128         if (atomic_read(&clp->cl_refcount))
 129                 return nfserr_jukebox;
 130         clp->cl_time = 0;
 131         return nfs_ok;
 132 }
 133
 134 static __be32 mark_client_expired(struct nfs4_client *clp)
 135 {
 136         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 137         __be32 ret;
 138
 139         spin_lock(&nn->client_lock);
 140         ret = mark_client_expired_locked(clp);
 141         spin_unlock(&nn->client_lock);
 142         return ret;
 143 }
 144
 145 static __be32 get_client_locked(struct nfs4_client *clp)
 146 {
 147         if (is_client_expired(clp))
 148                 return nfserr_expired;
 149         atomic_inc(&clp->cl_refcount);
 150         return nfs_ok;
 151 }
 152
 153 /* must be called under the client_lock */
 154 static inline void
 155 renew_client_locked(struct nfs4_client *clp)
 156 {
 157         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 158
 159         if (is_client_expired(clp)) {
 160                 WARN_ON(1);
 161                 printk("%s: client (clientid %08x/%08x) already expired\n",
 162                         __func__,
 163                         clp->cl_clientid.cl_boot,
 164                         clp->cl_clientid.cl_id);
 165                 return;
 166         }
 167
 168         dprintk("renewing client (clientid %08x/%08x)\n",
 169                         clp->cl_clientid.cl_boot,
 170                         clp->cl_clientid.cl_id);
 171         list_move_tail(&clp->cl_lru, &nn->client_lru);
 172         clp->cl_time = get_seconds();
 173 }
 174
 175 static inline void
 176 renew_client(struct nfs4_client *clp)
 177 {
 178         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 179
 180         spin_lock(&nn->client_lock);
 181         renew_client_locked(clp);
 182         spin_unlock(&nn->client_lock);
 183 }
 184
 185 static void put_client_renew_locked(struct nfs4_client *clp)
 186 {
 187         if (!atomic_dec_and_test(&clp->cl_refcount))
 188                 return;
 189         if (!is_client_expired(clp))
 190                 renew_client_locked(clp);
 191 }
 192
 193 static void put_client_renew(struct nfs4_client *clp)
 194 {
 195         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 196
 197         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 198                 return;
 199         if (!is_client_expired(clp))
 200                 renew_client_locked(clp);
 201         spin_unlock(&nn->client_lock);
 202 }
 203
 204 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 205 {
 206         __be32 status;
 207
 208         if (is_session_dead(ses))
 209                 return nfserr_badsession;
 210         status = get_client_locked(ses->se_client);
 211         if (status)
 212                 return status;
 213         atomic_inc(&ses->se_ref);
 214         return nfs_ok;
 215 }
 216
 217 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 218 {
 219         struct nfs4_client *clp = ses->se_client;
 220
 221         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 222                 free_session(ses);
 223         put_client_renew_locked(clp);
 224 }
 225
 226 static void nfsd4_put_session(struct nfsd4_session *ses)
 227 {
 228         struct nfs4_client *clp = ses->se_client;
 229         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 230
 231         spin_lock(&nn->client_lock);
 232         nfsd4_put_session_locked(ses);
 233         spin_unlock(&nn->client_lock);
 234 }
 235
 236
 237 static inline u32
 238 opaque_hashval(const void *ptr, int nbytes)
 239 {
 240         unsigned char *cptr = (unsigned char *) ptr;
 241
 242         u32 x = 0;
 243         while (nbytes--) {
 244                 x *= 37;
 245                 x += *cptr++;
 246         }
 247         return x;
 248 }
 249
 250 static void nfsd4_free_file(struct nfs4_file *f)
 251 {
 252         kmem_cache_free(file_slab, f);
 253 }
 254
 255 static inline void
 256 put_nfs4_file(struct nfs4_file *fi)
 257 {
 258         might_lock(&state_lock);
 259
 260         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 261                 hlist_del(&fi->fi_hash);
 262                 spin_unlock(&state_lock);
 263                 nfsd4_free_file(fi);
 264         }
 265 }
 266
 267 static inline void
 268 get_nfs4_file(struct nfs4_file *fi)
 269 {
 270         atomic_inc(&fi->fi_ref);
 271 }
 272
 273 static struct file *
 274 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 275 {
 276         if (f->fi_fds[oflag])
 277                 return get_file(f->fi_fds[oflag]);
 278         return NULL;
 279 }
 280
 281 static struct file *
 282 find_writeable_file_locked(struct nfs4_file *f)
 283 {
 284         struct file *ret;
 285
 286         lockdep_assert_held(&f->fi_lock);
 287
 288         ret = __nfs4_get_fd(f, O_WRONLY);
 289         if (!ret)
 290                 ret = __nfs4_get_fd(f, O_RDWR);
 291         return ret;
 292 }
 293
 294 static struct file *
 295 find_writeable_file(struct nfs4_file *f)
 296 {
 297         struct file *ret;
 298
 299         spin_lock(&f->fi_lock);
 300         ret = find_writeable_file_locked(f);
 301         spin_unlock(&f->fi_lock);
 302
 303         return ret;
 304 }
 305
 306 static struct file *find_readable_file_locked(struct nfs4_file *f)
 307 {
 308         struct file *ret;
 309
 310         lockdep_assert_held(&f->fi_lock);
 311
 312         ret = __nfs4_get_fd(f, O_RDONLY);
 313         if (!ret)
 314                 ret = __nfs4_get_fd(f, O_RDWR);
 315         return ret;
 316 }
 317
 318 static struct file *
 319 find_readable_file(struct nfs4_file *f)
 320 {
 321         struct file *ret;
 322
 323         spin_lock(&f->fi_lock);
 324         ret = find_readable_file_locked(f);
 325         spin_unlock(&f->fi_lock);
 326
 327         return ret;
 328 }
 329
 330 static struct file *
 331 find_any_file(struct nfs4_file *f)
 332 {
 333         struct file *ret;
 334
 335         spin_lock(&f->fi_lock);
 336         ret = __nfs4_get_fd(f, O_RDWR);
 337         if (!ret) {
 338                 ret = __nfs4_get_fd(f, O_WRONLY);
 339                 if (!ret)
 340                         ret = __nfs4_get_fd(f, O_RDONLY);
 341         }
 342         spin_unlock(&f->fi_lock);
 343         return ret;
 344 }
 345
 346 static atomic_long_t num_delegations;
 347 unsigned long max_delegations;
 348
 349 /*
 350  * Open owner state (share locks)
 351  */
 352
 353 /* hash tables for lock and open owners */
 354 #define OWNER_HASH_BITS              8
 355 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 356 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 357
 358 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 359 {
 360         unsigned int ret;
 361
 362         ret = opaque_hashval(ownername->data, ownername->len);
 363         ret += clientid;
 364         return ret & OWNER_HASH_MASK;
 365 }
 366
 367 /* hash table for nfs4_file */
 368 #define FILE_HASH_BITS                   8
 369 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 370
 371 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 372 {
 373         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 374 }
 375
 376 static unsigned int file_hashval(struct knfsd_fh *fh)
 377 {
 378         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 379 }
 380
 381 static bool nfsd_fh_match(struct knfsd_fh *fh1, struct knfsd_fh *fh2)
 382 {
 383         return fh1->fh_size == fh2->fh_size &&
 384                 !memcmp(fh1->fh_base.fh_pad,
 385                                 fh2->fh_base.fh_pad,
 386                                 fh1->fh_size);
 387 }
 388
 389 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 390
 391 static void
 392 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 393 {
 394         lockdep_assert_held(&fp->fi_lock);
 395
 396         if (access & NFS4_SHARE_ACCESS_WRITE)
 397                 atomic_inc(&fp->fi_access[O_WRONLY]);
 398         if (access & NFS4_SHARE_ACCESS_READ)
 399                 atomic_inc(&fp->fi_access[O_RDONLY]);
 400 }
 401
 402 static __be32
 403 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 404 {
 405         lockdep_assert_held(&fp->fi_lock);
 406
 407         /* Does this access mode make sense? */
 408         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 409                 return nfserr_inval;
 410
 411         /* Does it conflict with a deny mode already set? */
 412         if ((access & fp->fi_share_deny) != 0)
 413                 return nfserr_share_denied;
 414
 415         __nfs4_file_get_access(fp, access);
 416         return nfs_ok;
 417 }
 418
 419 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 420 {
 421         /* Common case is that there is no deny mode. */
 422         if (deny) {
 423                 /* Does this deny mode make sense? */
 424                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 425                         return nfserr_inval;
 426
 427                 if ((deny & NFS4_SHARE_DENY_READ) &&
 428                     atomic_read(&fp->fi_access[O_RDONLY]))
 429                         return nfserr_share_denied;
 430
 431                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 432                     atomic_read(&fp->fi_access[O_WRONLY]))
 433                         return nfserr_share_denied;
 434         }
 435         return nfs_ok;
 436 }
 437
 438 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 439 {
 440         might_lock(&fp->fi_lock);
 441
 442         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 443                 struct file *f1 = NULL;
 444                 struct file *f2 = NULL;
 445
 446                 swap(f1, fp->fi_fds[oflag]);
 447                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 448                         swap(f2, fp->fi_fds[O_RDWR]);
 449                 spin_unlock(&fp->fi_lock);
 450                 if (f1)
 451                         fput(f1);
 452                 if (f2)
 453                         fput(f2);
 454         }
 455 }
 456
 457 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 458 {
 459         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 460
 461         if (access & NFS4_SHARE_ACCESS_WRITE)
 462                 __nfs4_file_put_access(fp, O_WRONLY);
 463         if (access & NFS4_SHARE_ACCESS_READ)
 464                 __nfs4_file_put_access(fp, O_RDONLY);
 465 }
 466
 467 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
 468                                          struct kmem_cache *slab)
 469 {
 470         struct nfs4_stid *stid;
 471         int new_id;
 472
 473         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 474         if (!stid)
 475                 return NULL;
 476
 477         idr_preload(GFP_KERNEL);
 478         spin_lock(&cl->cl_lock);
 479         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 480         spin_unlock(&cl->cl_lock);
 481         idr_preload_end();
 482         if (new_id < 0)
 483                 goto out_free;
 484         stid->sc_client = cl;
 485         stid->sc_stateid.si_opaque.so_id = new_id;
 486         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 487         /* Will be incremented before return to client: */
 488         atomic_set(&stid->sc_count, 1);
 489
 490         /*
 491          * It shouldn't be a problem to reuse an opaque stateid value.
 492          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 493          * example, a stray write retransmission could be accepted by
 494          * the server when it should have been rejected.  Therefore,
 495          * adopt a trick from the sctp code to attempt to maximize the
 496          * amount of time until an id is reused, by ensuring they always
 497          * "increase" (mod INT_MAX):
 498          */
 499         return stid;
 500 out_free:
 501         kmem_cache_free(slab, stid);
 502         return NULL;
 503 }
 504
 505 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 506 {
 507         struct nfs4_stid *stid;
 508         struct nfs4_ol_stateid *stp;
 509
 510         stid = nfs4_alloc_stid(clp, stateid_slab);
 511         if (!stid)
 512                 return NULL;
 513
 514         stp = openlockstateid(stid);
 515         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 516         return stp;
 517 }
 518
 519 static void nfs4_free_deleg(struct nfs4_stid *stid)
 520 {
 521         kmem_cache_free(deleg_slab, stid);
 522         atomic_long_dec(&num_delegations);
 523 }
 524
 525 /*
 526  * When we recall a delegation, we should be careful not to hand it
 527  * out again straight away.
 528  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 529  * in which the filehandles of recalled delegations are "stored".
 530  * If a filehandle appear in either filter, a delegation is blocked.
 531  * When a delegation is recalled, the filehandle is stored in the "new"
 532  * filter.
 533  * Every 30 seconds we swap the filters and clear the "new" one,
 534  * unless both are empty of course.
 535  *
 536  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 537  * low 3 bytes as hash-table indices.
 538  *
 539  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 540  * is used to manage concurrent access.  Testing does not need the lock
 541  * except when swapping the two filters.
 542  */
 543 static DEFINE_SPINLOCK(blocked_delegations_lock);
 544 static struct bloom_pair {
 545         int     entries, old_entries;
 546         time_t  swap_time;
 547         int     new; /* index into 'set' */
 548         DECLARE_BITMAP(set[2], 256);
 549 } blocked_delegations;
 550
 551 static int delegation_blocked(struct knfsd_fh *fh)
 552 {
 553         u32 hash;
 554         struct bloom_pair *bd = &blocked_delegations;
 555
 556         if (bd->entries == 0)
 557                 return 0;
 558         if (seconds_since_boot() - bd->swap_time > 30) {
 559                 spin_lock(&blocked_delegations_lock);
 560                 if (seconds_since_boot() - bd->swap_time > 30) {
 561                         bd->entries -= bd->old_entries;
 562                         bd->old_entries = bd->entries;
 563                         memset(bd->set[bd->new], 0,
 564                                sizeof(bd->set[0]));
 565                         bd->new = 1-bd->new;
 566                         bd->swap_time = seconds_since_boot();
 567                 }
 568                 spin_unlock(&blocked_delegations_lock);
 569         }
 570         hash = arch_fast_hash(&fh->fh_base, fh->fh_size, 0);
 571         if (test_bit(hash&255, bd->set[0]) &&
 572             test_bit((hash>>8)&255, bd->set[0]) &&
 573             test_bit((hash>>16)&255, bd->set[0]))
 574                 return 1;
 575
 576         if (test_bit(hash&255, bd->set[1]) &&
 577             test_bit((hash>>8)&255, bd->set[1]) &&
 578             test_bit((hash>>16)&255, bd->set[1]))
 579                 return 1;
 580
 581         return 0;
 582 }
 583
 584 static void block_delegations(struct knfsd_fh *fh)
 585 {
 586         u32 hash;
 587         struct bloom_pair *bd = &blocked_delegations;
 588
 589         hash = arch_fast_hash(&fh->fh_base, fh->fh_size, 0);
 590
 591         spin_lock(&blocked_delegations_lock);
 592         __set_bit(hash&255, bd->set[bd->new]);
 593         __set_bit((hash>>8)&255, bd->set[bd->new]);
 594         __set_bit((hash>>16)&255, bd->set[bd->new]);
 595         if (bd->entries == 0)
 596                 bd->swap_time = seconds_since_boot();
 597         bd->entries += 1;
 598         spin_unlock(&blocked_delegations_lock);
 599 }
 600
 601 static struct nfs4_delegation *
 602 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh)
 603 {
 604         struct nfs4_delegation *dp;
 605         long n;
 606
 607         dprintk("NFSD alloc_init_deleg\n");
 608         n = atomic_long_inc_return(&num_delegations);
 609         if (n < 0 || n > max_delegations)
 610                 goto out_dec;
 611         if (delegation_blocked(&current_fh->fh_handle))
 612                 goto out_dec;
 613         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 614         if (dp == NULL)
 615                 goto out_dec;
 616
 617         dp->dl_stid.sc_free = nfs4_free_deleg;
 618         /*
 619          * delegation seqid's are never incremented.  The 4.1 special
 620          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 621          * 0 anyway just for consistency and use 1:
 622          */
 623         dp->dl_stid.sc_stateid.si_generation = 1;
 624         INIT_LIST_HEAD(&dp->dl_perfile);
 625         INIT_LIST_HEAD(&dp->dl_perclnt);
 626         INIT_LIST_HEAD(&dp->dl_recall_lru);
 627         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 628         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_run_cb_recall);
 629         return dp;
 630 out_dec:
 631         atomic_long_dec(&num_delegations);
 632         return NULL;
 633 }
 634
 635 void
 636 nfs4_put_stid(struct nfs4_stid *s)
 637 {
 638         struct nfs4_file *fp = s->sc_file;
 639         struct nfs4_client *clp = s->sc_client;
 640
 641         might_lock(&clp->cl_lock);
 642
 643         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock))
 644                 return;
 645         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 646         spin_unlock(&clp->cl_lock);
 647         s->sc_free(s);
 648         if (fp)
 649                 put_nfs4_file(fp);
 650 }
 651
 652 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 653 {
 654         lockdep_assert_held(&state_lock);
 655
 656         if (!fp->fi_lease)
 657                 return;
 658         if (atomic_dec_and_test(&fp->fi_delegees)) {
 659                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 660                 fp->fi_lease = NULL;
 661                 fput(fp->fi_deleg_file);
 662                 fp->fi_deleg_file = NULL;
 663         }
 664 }
 665
 666 static void unhash_stid(struct nfs4_stid *s)
 667 {
 668         s->sc_type = 0;
 669 }
 670
 671 static void
 672 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 673 {
 674         lockdep_assert_held(&state_lock);
 675         lockdep_assert_held(&fp->fi_lock);
 676
 677         atomic_inc(&dp->dl_stid.sc_count);
 678         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 679         list_add(&dp->dl_perfile, &fp->fi_delegations);
 680         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
 681 }
 682
 683 static void
 684 unhash_delegation_locked(struct nfs4_delegation *dp)
 685 {
 686         struct nfs4_file *fp = dp->dl_stid.sc_file;
 687
 688         lockdep_assert_held(&state_lock);
 689
 690         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 691         /* Ensure that deleg break won't try to requeue it */
 692         ++dp->dl_time;
 693         spin_lock(&fp->fi_lock);
 694         list_del_init(&dp->dl_perclnt);
 695         list_del_init(&dp->dl_recall_lru);
 696         list_del_init(&dp->dl_perfile);
 697         spin_unlock(&fp->fi_lock);
 698         if (fp)
 699                 nfs4_put_deleg_lease(fp);
 700 }
 701
 702 static void destroy_delegation(struct nfs4_delegation *dp)
 703 {
 704         spin_lock(&state_lock);
 705         unhash_delegation_locked(dp);
 706         spin_unlock(&state_lock);
 707         nfs4_put_stid(&dp->dl_stid);
 708 }
 709
 710 static void revoke_delegation(struct nfs4_delegation *dp)
 711 {
 712         struct nfs4_client *clp = dp->dl_stid.sc_client;
 713
 714         WARN_ON(!list_empty(&dp->dl_recall_lru));
 715
 716         if (clp->cl_minorversion == 0)
 717                 nfs4_put_stid(&dp->dl_stid);
 718         else {
 719                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 720                 spin_lock(&clp->cl_lock);
 721                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 722                 spin_unlock(&clp->cl_lock);
 723         }
 724 }
 725
 726 /*
 727  * SETCLIENTID state
 728  */
 729
 730 static unsigned int clientid_hashval(u32 id)
 731 {
 732         return id & CLIENT_HASH_MASK;
 733 }
 734
 735 static unsigned int clientstr_hashval(const char *name)
 736 {
 737         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 738 }
 739
 740 /*
 741  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 742  * st_{access,deny}_bmap field of the stateid, in order to track not
 743  * only what share bits are currently in force, but also what
 744  * combinations of share bits previous opens have used.  This allows us
 745  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 746  * return an error if the client attempt to downgrade to a combination
 747  * of share bits not explicable by closing some of its previous opens.
 748  *
 749  * XXX: This enforcement is actually incomplete, since we don't keep
 750  * track of access/deny bit combinations; so, e.g., we allow:
 751  *
 752  *      OPEN allow read, deny write
 753  *      OPEN allow both, deny none
 754  *      DOWNGRADE allow read, deny none
 755  *
 756  * which we should reject.
 757  */
 758 static unsigned int
 759 bmap_to_share_mode(unsigned long bmap) {
 760         int i;
 761         unsigned int access = 0;
 762
 763         for (i = 1; i < 4; i++) {
 764                 if (test_bit(i, &bmap))
 765                         access |= i;
 766         }
 767         return access;
 768 }
 769
 770 /* set share access for a given stateid */
 771 static inline void
 772 set_access(u32 access, struct nfs4_ol_stateid *stp)
 773 {
 774         unsigned char mask = 1 << access;
 775
 776         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 777         stp->st_access_bmap |= mask;
 778 }
 779
 780 /* clear share access for a given stateid */
 781 static inline void
 782 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 783 {
 784         unsigned char mask = 1 << access;
 785
 786         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 787         stp->st_access_bmap &= ~mask;
 788 }
 789
 790 /* test whether a given stateid has access */
 791 static inline bool
 792 test_access(u32 access, struct nfs4_ol_stateid *stp)
 793 {
 794         unsigned char mask = 1 << access;
 795
 796         return (bool)(stp->st_access_bmap & mask);
 797 }
 798
 799 /* set share deny for a given stateid */
 800 static inline void
 801 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 802 {
 803         unsigned char mask = 1 << deny;
 804
 805         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 806         stp->st_deny_bmap |= mask;
 807 }
 808
 809 /* clear share deny for a given stateid */
 810 static inline void
 811 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 812 {
 813         unsigned char mask = 1 << deny;
 814
 815         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 816         stp->st_deny_bmap &= ~mask;
 817 }
 818
 819 /* test whether a given stateid is denying specific access */
 820 static inline bool
 821 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 822 {
 823         unsigned char mask = 1 << deny;
 824
 825         return (bool)(stp->st_deny_bmap & mask);
 826 }
 827
 828 static int nfs4_access_to_omode(u32 access)
 829 {
 830         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 831         case NFS4_SHARE_ACCESS_READ:
 832                 return O_RDONLY;
 833         case NFS4_SHARE_ACCESS_WRITE:
 834                 return O_WRONLY;
 835         case NFS4_SHARE_ACCESS_BOTH:
 836                 return O_RDWR;
 837         }
 838         WARN_ON_ONCE(1);
 839         return O_RDONLY;
 840 }
 841
 842 /*
 843  * A stateid that had a deny mode associated with it is being released
 844  * or downgraded. Recalculate the deny mode on the file.
 845  */
 846 static void
 847 recalculate_deny_mode(struct nfs4_file *fp)
 848 {
 849         struct nfs4_ol_stateid *stp;
 850
 851         spin_lock(&fp->fi_lock);
 852         fp->fi_share_deny = 0;
 853         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
 854                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
 855         spin_unlock(&fp->fi_lock);
 856 }
 857
 858 static void
 859 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
 860 {
 861         int i;
 862         bool change = false;
 863
 864         for (i = 1; i < 4; i++) {
 865                 if ((i & deny) != i) {
 866                         change = true;
 867                         clear_deny(i, stp);
 868                 }
 869         }
 870
 871         /* Recalculate per-file deny mode if there was a change */
 872         if (change)
 873                 recalculate_deny_mode(stp->st_stid.sc_file);
 874 }
 875
 876 /* release all access and file references for a given stateid */
 877 static void
 878 release_all_access(struct nfs4_ol_stateid *stp)
 879 {
 880         int i;
 881         struct nfs4_file *fp = stp->st_stid.sc_file;
 882
 883         if (fp && stp->st_deny_bmap != 0)
 884                 recalculate_deny_mode(fp);
 885
 886         for (i = 1; i < 4; i++) {
 887                 if (test_access(i, stp))
 888                         nfs4_file_put_access(stp->st_stid.sc_file, i);
 889                 clear_access(i, stp);
 890         }
 891 }
 892
 893 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 894 {
 895         struct nfs4_file *fp = stp->st_stid.sc_file;
 896
 897         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
 898
 899         spin_lock(&fp->fi_lock);
 900         list_del(&stp->st_perfile);
 901         spin_unlock(&fp->fi_lock);
 902         list_del(&stp->st_perstateowner);
 903 }
 904
 905 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
 906 {
 907         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 908
 909         release_all_access(stp);
 910         kmem_cache_free(stateid_slab, stid);
 911 }
 912
 913 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
 914 {
 915         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 916         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
 917         struct file *file;
 918
 919         file = find_any_file(stp->st_stid.sc_file);
 920         if (file)
 921                 filp_close(file, (fl_owner_t)lo);
 922         nfs4_free_ol_stateid(stid);
 923 }
 924
 925 static void __release_lock_stateid(struct nfs4_ol_stateid *stp)
 926 {
 927         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
 928
 929         spin_lock(&oo->oo_owner.so_client->cl_lock);
 930         list_del(&stp->st_locks);
 931         unhash_generic_stateid(stp);
 932         unhash_stid(&stp->st_stid);
 933         spin_unlock(&oo->oo_owner.so_client->cl_lock);
 934         nfs4_put_stid(&stp->st_stid);
 935 }
 936
 937 static void unhash_lockowner(struct nfs4_lockowner *lo)
 938 {
 939         struct nfs4_ol_stateid *stp;
 940
 941         list_del(&lo->lo_owner.so_strhash);
 942         while (!list_empty(&lo->lo_owner.so_stateids)) {
 943                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 944                                 struct nfs4_ol_stateid, st_perstateowner);
 945                 __release_lock_stateid(stp);
 946         }
 947 }
 948
 949 static void nfs4_free_lockowner(struct nfs4_lockowner *lo)
 950 {
 951         kfree(lo->lo_owner.so_owner.data);
 952         kmem_cache_free(lockowner_slab, lo);
 953 }
 954
 955 static void release_lockowner(struct nfs4_lockowner *lo)
 956 {
 957         unhash_lockowner(lo);
 958         nfs4_free_lockowner(lo);
 959 }
 960
 961 static void release_lockowner_if_empty(struct nfs4_lockowner *lo)
 962 {
 963         if (list_empty(&lo->lo_owner.so_stateids))
 964                 release_lockowner(lo);
 965 }
 966
 967 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 968 {
 969         struct nfs4_lockowner *lo;
 970
 971         lo = lockowner(stp->st_stateowner);
 972         __release_lock_stateid(stp);
 973         release_lockowner_if_empty(lo);
 974 }
 975
 976 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp)
 977         __releases(&open_stp->st_stateowner->so_client->cl_lock)
 978         __acquires(&open_stp->st_stateowner->so_client->cl_lock)
 979 {
 980         struct nfs4_ol_stateid *stp;
 981
 982         while (!list_empty(&open_stp->st_locks)) {
 983                 stp = list_entry(open_stp->st_locks.next,
 984                                 struct nfs4_ol_stateid, st_locks);
 985                 spin_unlock(&open_stp->st_stateowner->so_client->cl_lock);
 986                 release_lock_stateid(stp);
 987                 spin_lock(&open_stp->st_stateowner->so_client->cl_lock);
 988         }
 989 }
 990
 991 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 992 {
 993         spin_lock(&stp->st_stateowner->so_client->cl_lock);
 994         unhash_generic_stateid(stp);
 995         release_open_stateid_locks(stp);
 996         spin_unlock(&stp->st_stateowner->so_client->cl_lock);
 997 }
 998
 999 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1000 {
1001         unhash_open_stateid(stp);
1002         nfs4_put_stid(&stp->st_stid);
1003 }
1004
1005 static void unhash_openowner(struct nfs4_openowner *oo)
1006 {
1007         struct nfs4_ol_stateid *stp;
1008
1009         list_del(&oo->oo_owner.so_strhash);
1010         list_del(&oo->oo_perclient);
1011         while (!list_empty(&oo->oo_owner.so_stateids)) {
1012                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1013                                 struct nfs4_ol_stateid, st_perstateowner);
1014                 release_open_stateid(stp);
1015         }
1016 }
1017
1018 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1019 {
1020         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
1021
1022         if (s) {
1023                 nfs4_put_stid(&s->st_stid);
1024                 oo->oo_last_closed_stid = NULL;
1025         }
1026 }
1027
1028 static void nfs4_free_openowner(struct nfs4_openowner *oo)
1029 {
1030         kfree(oo->oo_owner.so_owner.data);
1031         kmem_cache_free(openowner_slab, oo);
1032 }
1033
1034 static void release_openowner(struct nfs4_openowner *oo)
1035 {
1036         unhash_openowner(oo);
1037         list_del(&oo->oo_close_lru);
1038         release_last_closed_stateid(oo);
1039         nfs4_free_openowner(oo);
1040 }
1041
1042 static inline int
1043 hash_sessionid(struct nfs4_sessionid *sessionid)
1044 {
1045         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1046
1047         return sid->sequence % SESSION_HASH_SIZE;
1048 }
1049
1050 #ifdef NFSD_DEBUG
1051 static inline void
1052 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1053 {
1054         u32 *ptr = (u32 *)(&sessionid->data[0]);
1055         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1056 }
1057 #else
1058 static inline void
1059 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1060 {
1061 }
1062 #endif
1063
1064 /*
1065  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1066  * won't be used for replay.
1067  */
1068 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1069 {
1070         struct nfs4_stateowner *so = cstate->replay_owner;
1071
1072         if (nfserr == nfserr_replay_me)
1073                 return;
1074
1075         if (!seqid_mutating_err(ntohl(nfserr))) {
1076                 cstate->replay_owner = NULL;
1077                 return;
1078         }
1079         if (!so)
1080                 return;
1081         if (so->so_is_open_owner)
1082                 release_last_closed_stateid(openowner(so));
1083         so->so_seqid++;
1084         return;
1085 }
1086
1087 static void
1088 gen_sessionid(struct nfsd4_session *ses)
1089 {
1090         struct nfs4_client *clp = ses->se_client;
1091         struct nfsd4_sessionid *sid;
1092
1093         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1094         sid->clientid = clp->cl_clientid;
1095         sid->sequence = current_sessionid++;
1096         sid->reserved = 0;
1097 }
1098
1099 /*
1100  * The protocol defines ca_maxresponssize_cached to include the size of
1101  * the rpc header, but all we need to cache is the data starting after
1102  * the end of the initial SEQUENCE operation--the rest we regenerate
1103  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1104  * value that is the number of bytes in our cache plus a few additional
1105  * bytes.  In order to stay on the safe side, and not promise more than
1106  * we can cache, those additional bytes must be the minimum possible: 24
1107  * bytes of rpc header (xid through accept state, with AUTH_NULL
1108  * verifier), 12 for the compound header (with zero-length tag), and 44
1109  * for the SEQUENCE op response:
1110  */
1111 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1112
1113 static void
1114 free_session_slots(struct nfsd4_session *ses)
1115 {
1116         int i;
1117
1118         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1119                 kfree(ses->se_slots[i]);
1120 }
1121
1122 /*
1123  * We don't actually need to cache the rpc and session headers, so we
1124  * can allocate a little less for each slot:
1125  */
1126 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1127 {
1128         u32 size;
1129
1130         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1131                 size = 0;
1132         else
1133                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1134         return size + sizeof(struct nfsd4_slot);
1135 }
1136
1137 /*
1138  * XXX: If we run out of reserved DRC memory we could (up to a point)
1139  * re-negotiate active sessions and reduce their slot usage to make
1140  * room for new connections. For now we just fail the create session.
1141  */
1142 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1143 {
1144         u32 slotsize = slot_bytes(ca);
1145         u32 num = ca->maxreqs;
1146         int avail;
1147
1148         spin_lock(&nfsd_drc_lock);
1149         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1150                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1151         num = min_t(int, num, avail / slotsize);
1152         nfsd_drc_mem_used += num * slotsize;
1153         spin_unlock(&nfsd_drc_lock);
1154
1155         return num;
1156 }
1157
1158 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1159 {
1160         int slotsize = slot_bytes(ca);
1161
1162         spin_lock(&nfsd_drc_lock);
1163         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1164         spin_unlock(&nfsd_drc_lock);
1165 }
1166
1167 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1168                                            struct nfsd4_channel_attrs *battrs)
1169 {
1170         int numslots = fattrs->maxreqs;
1171         int slotsize = slot_bytes(fattrs);
1172         struct nfsd4_session *new;
1173         int mem, i;
1174
1175         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1176                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1177         mem = numslots * sizeof(struct nfsd4_slot *);
1178
1179         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1180         if (!new)
1181                 return NULL;
1182         /* allocate each struct nfsd4_slot and data cache in one piece */
1183         for (i = 0; i < numslots; i++) {
1184                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1185                 if (!new->se_slots[i])
1186                         goto out_free;
1187         }
1188
1189         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1190         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1191
1192         return new;
1193 out_free:
1194         while (i--)
1195                 kfree(new->se_slots[i]);
1196         kfree(new);
1197         return NULL;
1198 }
1199
1200 static void free_conn(struct nfsd4_conn *c)
1201 {
1202         svc_xprt_put(c->cn_xprt);
1203         kfree(c);
1204 }
1205
1206 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1207 {
1208         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1209         struct nfs4_client *clp = c->cn_session->se_client;
1210
1211         spin_lock(&clp->cl_lock);
1212         if (!list_empty(&c->cn_persession)) {
1213                 list_del(&c->cn_persession);
1214                 free_conn(c);
1215         }
1216         nfsd4_probe_callback(clp);
1217         spin_unlock(&clp->cl_lock);
1218 }
1219
1220 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1221 {
1222         struct nfsd4_conn *conn;
1223
1224         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1225         if (!conn)
1226                 return NULL;
1227         svc_xprt_get(rqstp->rq_xprt);
1228         conn->cn_xprt = rqstp->rq_xprt;
1229         conn->cn_flags = flags;
1230         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1231         return conn;
1232 }
1233
1234 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1235 {
1236         conn->cn_session = ses;
1237         list_add(&conn->cn_persession, &ses->se_conns);
1238 }
1239
1240 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1241 {
1242         struct nfs4_client *clp = ses->se_client;
1243
1244         spin_lock(&clp->cl_lock);
1245         __nfsd4_hash_conn(conn, ses);
1246         spin_unlock(&clp->cl_lock);
1247 }
1248
1249 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1250 {
1251         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1252         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1253 }
1254
1255 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1256 {
1257         int ret;
1258
1259         nfsd4_hash_conn(conn, ses);
1260         ret = nfsd4_register_conn(conn);
1261         if (ret)
1262                 /* oops; xprt is already down: */
1263                 nfsd4_conn_lost(&conn->cn_xpt_user);
1264         /* We may have gained or lost a callback channel: */
1265         nfsd4_probe_callback_sync(ses->se_client);
1266 }
1267
1268 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1269 {
1270         u32 dir = NFS4_CDFC4_FORE;
1271
1272         if (cses->flags & SESSION4_BACK_CHAN)
1273                 dir |= NFS4_CDFC4_BACK;
1274         return alloc_conn(rqstp, dir);
1275 }
1276
1277 /* must be called under client_lock */
1278 static void nfsd4_del_conns(struct nfsd4_session *s)
1279 {
1280         struct nfs4_client *clp = s->se_client;
1281         struct nfsd4_conn *c;
1282
1283         spin_lock(&clp->cl_lock);
1284         while (!list_empty(&s->se_conns)) {
1285                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1286                 list_del_init(&c->cn_persession);
1287                 spin_unlock(&clp->cl_lock);
1288
1289                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1290                 free_conn(c);
1291
1292                 spin_lock(&clp->cl_lock);
1293         }
1294         spin_unlock(&clp->cl_lock);
1295 }
1296
1297 static void __free_session(struct nfsd4_session *ses)
1298 {
1299         free_session_slots(ses);
1300         kfree(ses);
1301 }
1302
1303 static void free_session(struct nfsd4_session *ses)
1304 {
1305         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
1306
1307         lockdep_assert_held(&nn->client_lock);
1308         nfsd4_del_conns(ses);
1309         nfsd4_put_drc_mem(&ses->se_fchannel);
1310         __free_session(ses);
1311 }
1312
1313 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1314 {
1315         int idx;
1316         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1317
1318         new->se_client = clp;
1319         gen_sessionid(new);
1320
1321         INIT_LIST_HEAD(&new->se_conns);
1322
1323         new->se_cb_seq_nr = 1;
1324         new->se_flags = cses->flags;
1325         new->se_cb_prog = cses->callback_prog;
1326         new->se_cb_sec = cses->cb_sec;
1327         atomic_set(&new->se_ref, 0);
1328         idx = hash_sessionid(&new->se_sessionid);
1329         spin_lock(&nn->client_lock);
1330         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1331         spin_lock(&clp->cl_lock);
1332         list_add(&new->se_perclnt, &clp->cl_sessions);
1333         spin_unlock(&clp->cl_lock);
1334         spin_unlock(&nn->client_lock);
1335
1336         if (cses->flags & SESSION4_BACK_CHAN) {
1337                 struct sockaddr *sa = svc_addr(rqstp);
1338                 /*
1339                  * This is a little silly; with sessions there's no real
1340                  * use for the callback address.  Use the peer address
1341                  * as a reasonable default for now, but consider fixing
1342                  * the rpc client not to require an address in the
1343                  * future:
1344                  */
1345                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1346                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1347         }
1348 }
1349
1350 /* caller must hold client_lock */
1351 static struct nfsd4_session *
1352 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1353 {
1354         struct nfsd4_session *elem;
1355         int idx;
1356         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1357
1358         dump_sessionid(__func__, sessionid);
1359         idx = hash_sessionid(sessionid);
1360         /* Search in the appropriate list */
1361         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1362                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1363                             NFS4_MAX_SESSIONID_LEN)) {
1364                         return elem;
1365                 }
1366         }
1367
1368         dprintk("%s: session not found\n", __func__);
1369         return NULL;
1370 }
1371
1372 static struct nfsd4_session *
1373 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1374                 __be32 *ret)
1375 {
1376         struct nfsd4_session *session;
1377         __be32 status = nfserr_badsession;
1378
1379         session = __find_in_sessionid_hashtbl(sessionid, net);
1380         if (!session)
1381                 goto out;
1382         status = nfsd4_get_session_locked(session);
1383         if (status)
1384                 session = NULL;
1385 out:
1386         *ret = status;
1387         return session;
1388 }
1389
1390 /* caller must hold client_lock */
1391 static void
1392 unhash_session(struct nfsd4_session *ses)
1393 {
1394         list_del(&ses->se_hash);
1395         spin_lock(&ses->se_client->cl_lock);
1396         list_del(&ses->se_perclnt);
1397         spin_unlock(&ses->se_client->cl_lock);
1398 }
1399
1400 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1401 static int
1402 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1403 {
1404         if (clid->cl_boot == nn->boot_time)
1405                 return 0;
1406         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1407                 clid->cl_boot, clid->cl_id, nn->boot_time);
1408         return 1;
1409 }
1410
1411 /*
1412  * XXX Should we use a slab cache ?
1413  * This type of memory management is somewhat inefficient, but we use it
1414  * anyway since SETCLIENTID is not a common operation.
1415  */
1416 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1417 {
1418         struct nfs4_client *clp;
1419
1420         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1421         if (clp == NULL)
1422                 return NULL;
1423         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1424         if (clp->cl_name.data == NULL) {
1425                 kfree(clp);
1426                 return NULL;
1427         }
1428         clp->cl_name.len = name.len;
1429         INIT_LIST_HEAD(&clp->cl_sessions);
1430         idr_init(&clp->cl_stateids);
1431         atomic_set(&clp->cl_refcount, 0);
1432         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1433         INIT_LIST_HEAD(&clp->cl_idhash);
1434         INIT_LIST_HEAD(&clp->cl_openowners);
1435         INIT_LIST_HEAD(&clp->cl_delegations);
1436         INIT_LIST_HEAD(&clp->cl_lru);
1437         INIT_LIST_HEAD(&clp->cl_callbacks);
1438         INIT_LIST_HEAD(&clp->cl_revoked);
1439         spin_lock_init(&clp->cl_lock);
1440         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1441         return clp;
1442 }
1443
1444 static void
1445 free_client(struct nfs4_client *clp)
1446 {
1447         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1448
1449         lockdep_assert_held(&nn->client_lock);
1450         while (!list_empty(&clp->cl_sessions)) {
1451                 struct nfsd4_session *ses;
1452                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1453                                 se_perclnt);
1454                 list_del(&ses->se_perclnt);
1455                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1456                 free_session(ses);
1457         }
1458         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1459         free_svc_cred(&clp->cl_cred);
1460         kfree(clp->cl_name.data);
1461         idr_destroy(&clp->cl_stateids);
1462         kfree(clp);
1463 }
1464
1465 /* must be called under the client_lock */
1466 static inline void
1467 unhash_client_locked(struct nfs4_client *clp)
1468 {
1469         struct nfsd4_session *ses;
1470
1471         list_del(&clp->cl_lru);
1472         spin_lock(&clp->cl_lock);
1473         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1474                 list_del_init(&ses->se_hash);
1475         spin_unlock(&clp->cl_lock);
1476 }
1477
1478 static void
1479 destroy_client(struct nfs4_client *clp)
1480 {
1481         struct nfs4_openowner *oo;
1482         struct nfs4_delegation *dp;
1483         struct list_head reaplist;
1484         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1485
1486         INIT_LIST_HEAD(&reaplist);
1487         spin_lock(&state_lock);
1488         while (!list_empty(&clp->cl_delegations)) {
1489                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1490                 unhash_delegation_locked(dp);
1491                 list_add(&dp->dl_recall_lru, &reaplist);
1492         }
1493         spin_unlock(&state_lock);
1494         while (!list_empty(&reaplist)) {
1495                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1496                 list_del_init(&dp->dl_recall_lru);
1497                 nfs4_put_stid(&dp->dl_stid);
1498         }
1499         while (!list_empty(&clp->cl_revoked)) {
1500                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1501                 list_del_init(&dp->dl_recall_lru);
1502                 nfs4_put_stid(&dp->dl_stid);
1503         }
1504         while (!list_empty(&clp->cl_openowners)) {
1505                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1506                 release_openowner(oo);
1507         }
1508         nfsd4_shutdown_callback(clp);
1509         if (clp->cl_cb_conn.cb_xprt)
1510                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1511         list_del(&clp->cl_idhash);
1512         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1513                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1514         else
1515                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1516         spin_lock(&nn->client_lock);
1517         unhash_client_locked(clp);
1518         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1519         free_client(clp);
1520         spin_unlock(&nn->client_lock);
1521 }
1522
1523 static void expire_client(struct nfs4_client *clp)
1524 {
1525         nfsd4_client_record_remove(clp);
1526         destroy_client(clp);
1527 }
1528
1529 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1530 {
1531         memcpy(target->cl_verifier.data, source->data,
1532                         sizeof(target->cl_verifier.data));
1533 }
1534
1535 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1536 {
1537         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1538         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1539 }
1540
1541 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1542 {
1543         if (source->cr_principal) {
1544                 target->cr_principal =
1545                                 kstrdup(source->cr_principal, GFP_KERNEL);
1546                 if (target->cr_principal == NULL)
1547                         return -ENOMEM;
1548         } else
1549                 target->cr_principal = NULL;
1550         target->cr_flavor = source->cr_flavor;
1551         target->cr_uid = source->cr_uid;
1552         target->cr_gid = source->cr_gid;
1553         target->cr_group_info = source->cr_group_info;
1554         get_group_info(target->cr_group_info);
1555         target->cr_gss_mech = source->cr_gss_mech;
1556         if (source->cr_gss_mech)
1557                 gss_mech_get(source->cr_gss_mech);
1558         return 0;
1559 }
1560
1561 static long long
1562 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1563 {
1564         long long res;
1565
1566         res = o1->len - o2->len;
1567         if (res)
1568                 return res;
1569         return (long long)memcmp(o1->data, o2->data, o1->len);
1570 }
1571
1572 static int same_name(const char *n1, const char *n2)
1573 {
1574         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1575 }
1576
1577 static int
1578 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1579 {
1580         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1581 }
1582
1583 static int
1584 same_clid(clientid_t *cl1, clientid_t *cl2)
1585 {
1586         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1587 }
1588
1589 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1590 {
1591         int i;
1592
1593         if (g1->ngroups != g2->ngroups)
1594                 return false;
1595         for (i=0; i<g1->ngroups; i++)
1596                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1597                         return false;
1598         return true;
1599 }
1600
1601 /*
1602  * RFC 3530 language requires clid_inuse be returned when the
1603  * "principal" associated with a requests differs from that previously
1604  * used.  We use uid, gid's, and gss principal string as our best
1605  * approximation.  We also don't want to allow non-gss use of a client
1606  * established using gss: in theory cr_principal should catch that
1607  * change, but in practice cr_principal can be null even in the gss case
1608  * since gssd doesn't always pass down a principal string.
1609  */
1610 static bool is_gss_cred(struct svc_cred *cr)
1611 {
1612         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1613         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1614 }
1615
1616
1617 static bool
1618 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1619 {
1620         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1621                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1622                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1623                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1624                 return false;
1625         if (cr1->cr_principal == cr2->cr_principal)
1626                 return true;
1627         if (!cr1->cr_principal || !cr2->cr_principal)
1628                 return false;
1629         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1630 }
1631
1632 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1633 {
1634         struct svc_cred *cr = &rqstp->rq_cred;
1635         u32 service;
1636
1637         if (!cr->cr_gss_mech)
1638                 return false;
1639         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1640         return service == RPC_GSS_SVC_INTEGRITY ||
1641                service == RPC_GSS_SVC_PRIVACY;
1642 }
1643
1644 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1645 {
1646         struct svc_cred *cr = &rqstp->rq_cred;
1647
1648         if (!cl->cl_mach_cred)
1649                 return true;
1650         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1651                 return false;
1652         if (!svc_rqst_integrity_protected(rqstp))
1653                 return false;
1654         if (!cr->cr_principal)
1655                 return false;
1656         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1657 }
1658
1659 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1660 {
1661         static u32 current_clientid = 1;
1662
1663         clp->cl_clientid.cl_boot = nn->boot_time;
1664         clp->cl_clientid.cl_id = current_clientid++;
1665 }
1666
1667 static void gen_confirm(struct nfs4_client *clp)
1668 {
1669         __be32 verf[2];
1670         static u32 i;
1671
1672         /*
1673          * This is opaque to client, so no need to byte-swap. Use
1674          * __force to keep sparse happy
1675          */
1676         verf[0] = (__force __be32)get_seconds();
1677         verf[1] = (__force __be32)i++;
1678         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1679 }
1680
1681 static struct nfs4_stid *
1682 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
1683 {
1684         struct nfs4_stid *ret;
1685
1686         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1687         if (!ret || !ret->sc_type)
1688                 return NULL;
1689         return ret;
1690 }
1691
1692 static struct nfs4_stid *
1693 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1694 {
1695         struct nfs4_stid *s;
1696
1697         spin_lock(&cl->cl_lock);
1698         s = find_stateid_locked(cl, t);
1699         if (s != NULL && !(typemask & s->sc_type))
1700                 s = NULL;
1701         spin_unlock(&cl->cl_lock);
1702         return s;
1703 }
1704
1705 static struct nfs4_client *create_client(struct xdr_netobj name,
1706                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1707 {
1708         struct nfs4_client *clp;
1709         struct sockaddr *sa = svc_addr(rqstp);
1710         int ret;
1711         struct net *net = SVC_NET(rqstp);
1712         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1713
1714         clp = alloc_client(name);
1715         if (clp == NULL)
1716                 return NULL;
1717
1718         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1719         if (ret) {
1720                 spin_lock(&nn->client_lock);
1721                 free_client(clp);
1722                 spin_unlock(&nn->client_lock);
1723                 return NULL;
1724         }
1725         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_run_cb_null);
1726         clp->cl_time = get_seconds();
1727         clear_bit(0, &clp->cl_cb_slot_busy);
1728         copy_verf(clp, verf);
1729         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1730         gen_confirm(clp);
1731         clp->cl_cb_session = NULL;
1732         clp->net = net;
1733         return clp;
1734 }
1735
1736 static void
1737 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1738 {
1739         struct rb_node **new = &(root->rb_node), *parent = NULL;
1740         struct nfs4_client *clp;
1741
1742         while (*new) {
1743                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1744                 parent = *new;
1745
1746                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1747                         new = &((*new)->rb_left);
1748                 else
1749                         new = &((*new)->rb_right);
1750         }
1751
1752         rb_link_node(&new_clp->cl_namenode, parent, new);
1753         rb_insert_color(&new_clp->cl_namenode, root);
1754 }
1755
1756 static struct nfs4_client *
1757 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1758 {
1759         long long cmp;
1760         struct rb_node *node = root->rb_node;
1761         struct nfs4_client *clp;
1762
1763         while (node) {
1764                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1765                 cmp = compare_blob(&clp->cl_name, name);
1766                 if (cmp > 0)
1767                         node = node->rb_left;
1768                 else if (cmp < 0)
1769                         node = node->rb_right;
1770                 else
1771                         return clp;
1772         }
1773         return NULL;
1774 }
1775
1776 static void
1777 add_to_unconfirmed(struct nfs4_client *clp)
1778 {
1779         unsigned int idhashval;
1780         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1781
1782         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1783         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1784         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1785         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1786         renew_client(clp);
1787 }
1788
1789 static void
1790 move_to_confirmed(struct nfs4_client *clp)
1791 {
1792         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1793         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1794
1795         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1796         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1797         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1798         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1799         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1800         renew_client(clp);
1801 }
1802
1803 static struct nfs4_client *
1804 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1805 {
1806         struct nfs4_client *clp;
1807         unsigned int idhashval = clientid_hashval(clid->cl_id);
1808
1809         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1810                 if (same_clid(&clp->cl_clientid, clid)) {
1811                         if ((bool)clp->cl_minorversion != sessions)
1812                                 return NULL;
1813                         renew_client(clp);
1814                         return clp;
1815                 }
1816         }
1817         return NULL;
1818 }
1819
1820 static struct nfs4_client *
1821 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1822 {
1823         struct list_head *tbl = nn->conf_id_hashtbl;
1824
1825         return find_client_in_id_table(tbl, clid, sessions);
1826 }
1827
1828 static struct nfs4_client *
1829 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1830 {
1831         struct list_head *tbl = nn->unconf_id_hashtbl;
1832
1833         return find_client_in_id_table(tbl, clid, sessions);
1834 }
1835
1836 static bool clp_used_exchangeid(struct nfs4_client *clp)
1837 {
1838         return clp->cl_exchange_flags != 0;
1839 }
1840
1841 static struct nfs4_client *
1842 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1843 {
1844         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1845 }
1846
1847 static struct nfs4_client *
1848 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1849 {
1850         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1851 }
1852
1853 static void
1854 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1855 {
1856         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1857         struct sockaddr *sa = svc_addr(rqstp);
1858         u32 scopeid = rpc_get_scope_id(sa);
1859         unsigned short expected_family;
1860
1861         /* Currently, we only support tcp and tcp6 for the callback channel */
1862         if (se->se_callback_netid_len == 3 &&
1863             !memcmp(se->se_callback_netid_val, "tcp", 3))
1864                 expected_family = AF_INET;
1865         else if (se->se_callback_netid_len == 4 &&
1866                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1867                 expected_family = AF_INET6;
1868         else
1869                 goto out_err;
1870
1871         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1872                                             se->se_callback_addr_len,
1873                                             (struct sockaddr *)&conn->cb_addr,
1874                                             sizeof(conn->cb_addr));
1875
1876         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1877                 goto out_err;
1878
1879         if (conn->cb_addr.ss_family == AF_INET6)
1880                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1881
1882         conn->cb_prog = se->se_callback_prog;
1883         conn->cb_ident = se->se_callback_ident;
1884         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1885         return;
1886 out_err:
1887         conn->cb_addr.ss_family = AF_UNSPEC;
1888         conn->cb_addrlen = 0;
1889         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1890                 "will not receive delegations\n",
1891                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1892
1893         return;
1894 }
1895
1896 /*
1897  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
1898  */
1899 static void
1900 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1901 {
1902         struct xdr_buf *buf = resp->xdr.buf;
1903         struct nfsd4_slot *slot = resp->cstate.slot;
1904         unsigned int base;
1905
1906         dprintk("--> %s slot %p\n", __func__, slot);
1907
1908         slot->sl_opcnt = resp->opcnt;
1909         slot->sl_status = resp->cstate.status;
1910
1911         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1912         if (nfsd4_not_cached(resp)) {
1913                 slot->sl_datalen = 0;
1914                 return;
1915         }
1916         base = resp->cstate.data_offset;
1917         slot->sl_datalen = buf->len - base;
1918         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
1919                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1920         return;
1921 }
1922
1923 /*
1924  * Encode the replay sequence operation from the slot values.
1925  * If cachethis is FALSE encode the uncached rep error on the next
1926  * operation which sets resp->p and increments resp->opcnt for
1927  * nfs4svc_encode_compoundres.
1928  *
1929  */
1930 static __be32
1931 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1932                           struct nfsd4_compoundres *resp)
1933 {
1934         struct nfsd4_op *op;
1935         struct nfsd4_slot *slot = resp->cstate.slot;
1936
1937         /* Encode the replayed sequence operation */
1938         op = &args->ops[resp->opcnt - 1];
1939         nfsd4_encode_operation(resp, op);
1940
1941         /* Return nfserr_retry_uncached_rep in next operation. */
1942         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1943                 op = &args->ops[resp->opcnt++];
1944                 op->status = nfserr_retry_uncached_rep;
1945                 nfsd4_encode_operation(resp, op);
1946         }
1947         return op->status;
1948 }
1949
1950 /*
1951  * The sequence operation is not cached because we can use the slot and
1952  * session values.
1953  */
1954 static __be32
1955 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1956                          struct nfsd4_sequence *seq)
1957 {
1958         struct nfsd4_slot *slot = resp->cstate.slot;
1959         struct xdr_stream *xdr = &resp->xdr;
1960         __be32 *p;
1961         __be32 status;
1962
1963         dprintk("--> %s slot %p\n", __func__, slot);
1964
1965         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1966         if (status)
1967                 return status;
1968
1969         p = xdr_reserve_space(xdr, slot->sl_datalen);
1970         if (!p) {
1971                 WARN_ON_ONCE(1);
1972                 return nfserr_serverfault;
1973         }
1974         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
1975         xdr_commit_encode(xdr);
1976
1977         resp->opcnt = slot->sl_opcnt;
1978         return slot->sl_status;
1979 }
1980
1981 /*
1982  * Set the exchange_id flags returned by the server.
1983  */
1984 static void
1985 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1986 {
1987         /* pNFS is not supported */
1988         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1989
1990         /* Referrals are supported, Migration is not. */
1991         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1992
1993         /* set the wire flags to return to client. */
1994         clid->flags = new->cl_exchange_flags;
1995 }
1996
1997 static bool client_has_state(struct nfs4_client *clp)
1998 {
1999         /*
2000          * Note clp->cl_openowners check isn't quite right: there's no
2001          * need to count owners without stateid's.
2002          *
2003          * Also note we should probably be using this in 4.0 case too.
2004          */
2005         return !list_empty(&clp->cl_openowners)
2006                 || !list_empty(&clp->cl_delegations)
2007                 || !list_empty(&clp->cl_sessions);
2008 }
2009
2010 __be32
2011 nfsd4_exchange_id(struct svc_rqst *rqstp,
2012                   struct nfsd4_compound_state *cstate,
2013                   struct nfsd4_exchange_id *exid)
2014 {
2015         struct nfs4_client *unconf, *conf, *new;
2016         __be32 status;
2017         char                    addr_str[INET6_ADDRSTRLEN];
2018         nfs4_verifier           verf = exid->verifier;
2019         struct sockaddr         *sa = svc_addr(rqstp);
2020         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2021         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2022
2023         rpc_ntop(sa, addr_str, sizeof(addr_str));
2024         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2025                 "ip_addr=%s flags %x, spa_how %d\n",
2026                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2027                 addr_str, exid->flags, exid->spa_how);
2028
2029         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2030                 return nfserr_inval;
2031
2032         switch (exid->spa_how) {
2033         case SP4_MACH_CRED:
2034                 if (!svc_rqst_integrity_protected(rqstp))
2035                         return nfserr_inval;
2036         case SP4_NONE:
2037                 break;
2038         default:                                /* checked by xdr code */
2039                 WARN_ON_ONCE(1);
2040         case SP4_SSV:
2041                 return nfserr_encr_alg_unsupp;
2042         }
2043
2044         /* Cases below refer to rfc 5661 section 18.35.4: */
2045         nfs4_lock_state();
2046         conf = find_confirmed_client_by_name(&exid->clname, nn);
2047         if (conf) {
2048                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2049                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2050
2051                 if (update) {
2052                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2053                                 status = nfserr_inval;
2054                                 goto out;
2055                         }
2056                         if (!mach_creds_match(conf, rqstp)) {
2057                                 status = nfserr_wrong_cred;
2058                                 goto out;
2059                         }
2060                         if (!creds_match) { /* case 9 */
2061                                 status = nfserr_perm;
2062                                 goto out;
2063                         }
2064                         if (!verfs_match) { /* case 8 */
2065                                 status = nfserr_not_same;
2066                                 goto out;
2067                         }
2068                         /* case 6 */
2069                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2070                         new = conf;
2071                         goto out_copy;
2072                 }
2073                 if (!creds_match) { /* case 3 */
2074                         if (client_has_state(conf)) {
2075                                 status = nfserr_clid_inuse;
2076                                 goto out;
2077                         }
2078                         expire_client(conf);
2079                         goto out_new;
2080                 }
2081                 if (verfs_match) { /* case 2 */
2082                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2083                         new = conf;
2084                         goto out_copy;
2085                 }
2086                 /* case 5, client reboot */
2087                 goto out_new;
2088         }
2089
2090         if (update) { /* case 7 */
2091                 status = nfserr_noent;
2092                 goto out;
2093         }
2094
2095         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2096         if (unconf) /* case 4, possible retry or client restart */
2097                 expire_client(unconf);
2098
2099         /* case 1 (normal case) */
2100 out_new:
2101         new = create_client(exid->clname, rqstp, &verf);
2102         if (new == NULL) {
2103                 status = nfserr_jukebox;
2104                 goto out;
2105         }
2106         new->cl_minorversion = cstate->minorversion;
2107         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
2108
2109         gen_clid(new, nn);
2110         add_to_unconfirmed(new);
2111 out_copy:
2112         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
2113         exid->clientid.cl_id = new->cl_clientid.cl_id;
2114
2115         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
2116         nfsd4_set_ex_flags(new, exid);
2117
2118         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2119                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
2120         status = nfs_ok;
2121
2122 out:
2123         nfs4_unlock_state();
2124         return status;
2125 }
2126
2127 static __be32
2128 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2129 {
2130         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2131                 slot_seqid);
2132
2133         /* The slot is in use, and no response has been sent. */
2134         if (slot_inuse) {
2135                 if (seqid == slot_seqid)
2136                         return nfserr_jukebox;
2137                 else
2138                         return nfserr_seq_misordered;
2139         }
2140         /* Note unsigned 32-bit arithmetic handles wraparound: */
2141         if (likely(seqid == slot_seqid + 1))
2142                 return nfs_ok;
2143         if (seqid == slot_seqid)
2144                 return nfserr_replay_cache;
2145         return nfserr_seq_misordered;
2146 }
2147
2148 /*
2149  * Cache the create session result into the create session single DRC
2150  * slot cache by saving the xdr structure. sl_seqid has been set.
2151  * Do this for solo or embedded create session operations.
2152  */
2153 static void
2154 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2155                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2156 {
2157         slot->sl_status = nfserr;
2158         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2159 }
2160
2161 static __be32
2162 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2163                             struct nfsd4_clid_slot *slot)
2164 {
2165         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2166         return slot->sl_status;
2167 }
2168
2169 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2170                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2171                         1 +     /* MIN tag is length with zero, only length */ \
2172                         3 +     /* version, opcount, opcode */ \
2173                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2174                                 /* seqid, slotID, slotID, cache */ \
2175                         4 ) * sizeof(__be32))
2176
2177 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2178                         2 +     /* verifier: AUTH_NULL, length 0 */\
2179                         1 +     /* status */ \
2180                         1 +     /* MIN tag is length with zero, only length */ \
2181                         3 +     /* opcount, opcode, opstatus*/ \
2182                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2183                                 /* seqid, slotID, slotID, slotID, status */ \
2184                         5 ) * sizeof(__be32))
2185
2186 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2187 {
2188         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2189
2190         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2191                 return nfserr_toosmall;
2192         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2193                 return nfserr_toosmall;
2194         ca->headerpadsz = 0;
2195         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2196         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2197         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2198         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2199                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2200         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2201         /*
2202          * Note decreasing slot size below client's request may make it
2203          * difficult for client to function correctly, whereas
2204          * decreasing the number of slots will (just?) affect
2205          * performance.  When short on memory we therefore prefer to
2206          * decrease number of slots instead of their size.  Clients that
2207          * request larger slots than they need will get poor results:
2208          */
2209         ca->maxreqs = nfsd4_get_drc_mem(ca);
2210         if (!ca->maxreqs)
2211                 return nfserr_jukebox;
2212
2213         return nfs_ok;
2214 }
2215
2216 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2217                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2218 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2219                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2220
2221 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2222 {
2223         ca->headerpadsz = 0;
2224
2225         /*
2226          * These RPC_MAX_HEADER macros are overkill, especially since we
2227          * don't even do gss on the backchannel yet.  But this is still
2228          * less than 1k.  Tighten up this estimate in the unlikely event
2229          * it turns out to be a problem for some client:
2230          */
2231         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2232                 return nfserr_toosmall;
2233         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2234                 return nfserr_toosmall;
2235         ca->maxresp_cached = 0;
2236         if (ca->maxops < 2)
2237                 return nfserr_toosmall;
2238
2239         return nfs_ok;
2240 }
2241
2242 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2243 {
2244         switch (cbs->flavor) {
2245         case RPC_AUTH_NULL:
2246         case RPC_AUTH_UNIX:
2247                 return nfs_ok;
2248         default:
2249                 /*
2250                  * GSS case: the spec doesn't allow us to return this
2251                  * error.  But it also doesn't allow us not to support
2252                  * GSS.
2253                  * I'd rather this fail hard than return some error the
2254                  * client might think it can already handle:
2255                  */
2256                 return nfserr_encr_alg_unsupp;
2257         }
2258 }
2259
2260 __be32
2261 nfsd4_create_session(struct svc_rqst *rqstp,
2262                      struct nfsd4_compound_state *cstate,
2263                      struct nfsd4_create_session *cr_ses)
2264 {
2265         struct sockaddr *sa = svc_addr(rqstp);
2266         struct nfs4_client *conf, *unconf;
2267         struct nfsd4_session *new;
2268         struct nfsd4_conn *conn;
2269         struct nfsd4_clid_slot *cs_slot = NULL;
2270         __be32 status = 0;
2271         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2272
2273         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2274                 return nfserr_inval;
2275         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2276         if (status)
2277                 return status;
2278         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2279         if (status)
2280                 return status;
2281         status = check_backchannel_attrs(&cr_ses->back_channel);
2282         if (status)
2283                 goto out_release_drc_mem;
2284         status = nfserr_jukebox;
2285         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2286         if (!new)
2287                 goto out_release_drc_mem;
2288         conn = alloc_conn_from_crses(rqstp, cr_ses);
2289         if (!conn)
2290                 goto out_free_session;
2291
2292         nfs4_lock_state();
2293         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2294         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2295         WARN_ON_ONCE(conf && unconf);
2296
2297         if (conf) {
2298                 status = nfserr_wrong_cred;
2299                 if (!mach_creds_match(conf, rqstp))
2300                         goto out_free_conn;
2301                 cs_slot = &conf->cl_cs_slot;
2302                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2303                 if (status == nfserr_replay_cache) {
2304                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
2305                         goto out_free_conn;
2306                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
2307                         status = nfserr_seq_misordered;
2308                         goto out_free_conn;
2309                 }
2310         } else if (unconf) {
2311                 struct nfs4_client *old;
2312                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2313                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2314                         status = nfserr_clid_inuse;
2315                         goto out_free_conn;
2316                 }
2317                 status = nfserr_wrong_cred;
2318                 if (!mach_creds_match(unconf, rqstp))
2319                         goto out_free_conn;
2320                 cs_slot = &unconf->cl_cs_slot;
2321                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2322                 if (status) {
2323                         /* an unconfirmed replay returns misordered */
2324                         status = nfserr_seq_misordered;
2325                         goto out_free_conn;
2326                 }
2327                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2328                 if (old) {
2329                         status = mark_client_expired(old);
2330                         if (status)
2331                                 goto out_free_conn;
2332                         expire_client(old);
2333                 }
2334                 move_to_confirmed(unconf);
2335                 conf = unconf;
2336         } else {
2337                 status = nfserr_stale_clientid;
2338                 goto out_free_conn;
2339         }
2340         status = nfs_ok;
2341         /*
2342          * We do not support RDMA or persistent sessions
2343          */
2344         cr_ses->flags &= ~SESSION4_PERSIST;
2345         cr_ses->flags &= ~SESSION4_RDMA;
2346
2347         init_session(rqstp, new, conf, cr_ses);
2348         nfsd4_init_conn(rqstp, conn, new);
2349
2350         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2351                NFS4_MAX_SESSIONID_LEN);
2352         cs_slot->sl_seqid++;
2353         cr_ses->seqid = cs_slot->sl_seqid;
2354
2355         /* cache solo and embedded create sessions under the state lock */
2356         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2357         nfs4_unlock_state();
2358         return status;
2359 out_free_conn:
2360         nfs4_unlock_state();
2361         free_conn(conn);
2362 out_free_session:
2363         __free_session(new);
2364 out_release_drc_mem:
2365         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2366         return status;
2367 }
2368
2369 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2370 {
2371         switch (*dir) {
2372         case NFS4_CDFC4_FORE:
2373         case NFS4_CDFC4_BACK:
2374                 return nfs_ok;
2375         case NFS4_CDFC4_FORE_OR_BOTH:
2376         case NFS4_CDFC4_BACK_OR_BOTH:
2377                 *dir = NFS4_CDFC4_BOTH;
2378                 return nfs_ok;
2379         };
2380         return nfserr_inval;
2381 }
2382
2383 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2384 {
2385         struct nfsd4_session *session = cstate->session;
2386         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2387         __be32 status;
2388
2389         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2390         if (status)
2391                 return status;
2392         spin_lock(&nn->client_lock);
2393         session->se_cb_prog = bc->bc_cb_program;
2394         session->se_cb_sec = bc->bc_cb_sec;
2395         spin_unlock(&nn->client_lock);
2396
2397         nfsd4_probe_callback(session->se_client);
2398
2399         return nfs_ok;
2400 }
2401
2402 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2403                      struct nfsd4_compound_state *cstate,
2404                      struct nfsd4_bind_conn_to_session *bcts)
2405 {
2406         __be32 status;
2407         struct nfsd4_conn *conn;
2408         struct nfsd4_session *session;
2409         struct net *net = SVC_NET(rqstp);
2410         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2411
2412         if (!nfsd4_last_compound_op(rqstp))
2413                 return nfserr_not_only_op;
2414         nfs4_lock_state();
2415         spin_lock(&nn->client_lock);
2416         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2417         spin_unlock(&nn->client_lock);
2418         if (!session)
2419                 goto out_no_session;
2420         status = nfserr_wrong_cred;
2421         if (!mach_creds_match(session->se_client, rqstp))
2422                 goto out;
2423         status = nfsd4_map_bcts_dir(&bcts->dir);
2424         if (status)
2425                 goto out;
2426         conn = alloc_conn(rqstp, bcts->dir);
2427         status = nfserr_jukebox;
2428         if (!conn)
2429                 goto out;
2430         nfsd4_init_conn(rqstp, conn, session);
2431         status = nfs_ok;
2432 out:
2433         nfsd4_put_session(session);
2434 out_no_session:
2435         nfs4_unlock_state();
2436         return status;
2437 }
2438
2439 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2440 {
2441         if (!session)
2442                 return 0;
2443         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2444 }
2445
2446 __be32
2447 nfsd4_destroy_session(struct svc_rqst *r,
2448                       struct nfsd4_compound_state *cstate,
2449                       struct nfsd4_destroy_session *sessionid)
2450 {
2451         struct nfsd4_session *ses;
2452         __be32 status;
2453         int ref_held_by_me = 0;
2454         struct net *net = SVC_NET(r);
2455         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2456
2457         nfs4_lock_state();
2458         status = nfserr_not_only_op;
2459         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2460                 if (!nfsd4_last_compound_op(r))
2461                         goto out;
2462                 ref_held_by_me++;
2463         }
2464         dump_sessionid(__func__, &sessionid->sessionid);
2465         spin_lock(&nn->client_lock);
2466         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2467         if (!ses)
2468                 goto out_client_lock;
2469         status = nfserr_wrong_cred;
2470         if (!mach_creds_match(ses->se_client, r))
2471                 goto out_put_session;
2472         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2473         if (status)
2474                 goto out_put_session;
2475         unhash_session(ses);
2476         spin_unlock(&nn->client_lock);
2477
2478         nfsd4_probe_callback_sync(ses->se_client);
2479
2480         spin_lock(&nn->client_lock);
2481         status = nfs_ok;
2482 out_put_session:
2483         nfsd4_put_session_locked(ses);
2484 out_client_lock:
2485         spin_unlock(&nn->client_lock);
2486 out:
2487         nfs4_unlock_state();
2488         return status;
2489 }
2490
2491 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2492 {
2493         struct nfsd4_conn *c;
2494
2495         list_for_each_entry(c, &s->se_conns, cn_persession) {
2496                 if (c->cn_xprt == xpt) {
2497                         return c;
2498                 }
2499         }
2500         return NULL;
2501 }
2502
2503 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2504 {
2505         struct nfs4_client *clp = ses->se_client;
2506         struct nfsd4_conn *c;
2507         __be32 status = nfs_ok;
2508         int ret;
2509
2510         spin_lock(&clp->cl_lock);
2511         c = __nfsd4_find_conn(new->cn_xprt, ses);
2512         if (c)
2513                 goto out_free;
2514         status = nfserr_conn_not_bound_to_session;
2515         if (clp->cl_mach_cred)
2516                 goto out_free;
2517         __nfsd4_hash_conn(new, ses);
2518         spin_unlock(&clp->cl_lock);
2519         ret = nfsd4_register_conn(new);
2520         if (ret)
2521                 /* oops; xprt is already down: */
2522                 nfsd4_conn_lost(&new->cn_xpt_user);
2523         return nfs_ok;
2524 out_free:
2525         spin_unlock(&clp->cl_lock);
2526         free_conn(new);
2527         return status;
2528 }
2529
2530 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2531 {
2532         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2533
2534         return args->opcnt > session->se_fchannel.maxops;
2535 }
2536
2537 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2538                                   struct nfsd4_session *session)
2539 {
2540         struct xdr_buf *xb = &rqstp->rq_arg;
2541
2542         return xb->len > session->se_fchannel.maxreq_sz;
2543 }
2544
2545 __be32
2546 nfsd4_sequence(struct svc_rqst *rqstp,
2547                struct nfsd4_compound_state *cstate,
2548                struct nfsd4_sequence *seq)
2549 {
2550         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2551         struct xdr_stream *xdr = &resp->xdr;
2552         struct nfsd4_session *session;
2553         struct nfs4_client *clp;
2554         struct nfsd4_slot *slot;
2555         struct nfsd4_conn *conn;
2556         __be32 status;
2557         int buflen;
2558         struct net *net = SVC_NET(rqstp);
2559         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2560
2561         if (resp->opcnt != 1)
2562                 return nfserr_sequence_pos;
2563
2564         /*
2565          * Will be either used or freed by nfsd4_sequence_check_conn
2566          * below.
2567          */
2568         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2569         if (!conn)
2570                 return nfserr_jukebox;
2571
2572         spin_lock(&nn->client_lock);
2573         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2574         if (!session)
2575                 goto out_no_session;
2576         clp = session->se_client;
2577
2578         status = nfserr_too_many_ops;
2579         if (nfsd4_session_too_many_ops(rqstp, session))
2580                 goto out_put_session;
2581
2582         status = nfserr_req_too_big;
2583         if (nfsd4_request_too_big(rqstp, session))
2584                 goto out_put_session;
2585
2586         status = nfserr_badslot;
2587         if (seq->slotid >= session->se_fchannel.maxreqs)
2588                 goto out_put_session;
2589
2590         slot = session->se_slots[seq->slotid];
2591         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2592
2593         /* We do not negotiate the number of slots yet, so set the
2594          * maxslots to the session maxreqs which is used to encode
2595          * sr_highest_slotid and the sr_target_slot id to maxslots */
2596         seq->maxslots = session->se_fchannel.maxreqs;
2597
2598         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2599                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2600         if (status == nfserr_replay_cache) {
2601                 status = nfserr_seq_misordered;
2602                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2603                         goto out_put_session;
2604                 cstate->slot = slot;
2605                 cstate->session = session;
2606                 cstate->clp = clp;
2607                 /* Return the cached reply status and set cstate->status
2608                  * for nfsd4_proc_compound processing */
2609                 status = nfsd4_replay_cache_entry(resp, seq);
2610                 cstate->status = nfserr_replay_cache;
2611                 goto out;
2612         }
2613         if (status)
2614                 goto out_put_session;
2615
2616         status = nfsd4_sequence_check_conn(conn, session);
2617         conn = NULL;
2618         if (status)
2619                 goto out_put_session;
2620
2621         buflen = (seq->cachethis) ?
2622                         session->se_fchannel.maxresp_cached :
2623                         session->se_fchannel.maxresp_sz;
2624         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
2625                                     nfserr_rep_too_big;
2626         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
2627                 goto out_put_session;
2628         svc_reserve(rqstp, buflen);
2629
2630         status = nfs_ok;
2631         /* Success! bump slot seqid */
2632         slot->sl_seqid = seq->seqid;
2633         slot->sl_flags |= NFSD4_SLOT_INUSE;
2634         if (seq->cachethis)
2635                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2636         else
2637                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2638
2639         cstate->slot = slot;
2640         cstate->session = session;
2641         cstate->clp = clp;
2642
2643 out:
2644         switch (clp->cl_cb_state) {
2645         case NFSD4_CB_DOWN:
2646                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2647                 break;
2648         case NFSD4_CB_FAULT:
2649                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2650                 break;
2651         default:
2652                 seq->status_flags = 0;
2653         }
2654         if (!list_empty(&clp->cl_revoked))
2655                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2656 out_no_session:
2657         if (conn)
2658                 free_conn(conn);
2659         spin_unlock(&nn->client_lock);
2660         return status;
2661 out_put_session:
2662         nfsd4_put_session_locked(session);
2663         goto out_no_session;
2664 }
2665
2666 void
2667 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
2668 {
2669         struct nfsd4_compound_state *cs = &resp->cstate;
2670
2671         if (nfsd4_has_session(cs)) {
2672                 if (cs->status != nfserr_replay_cache) {
2673                         nfsd4_store_cache_entry(resp);
2674                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
2675                 }
2676                 /* Drop session reference that was taken in nfsd4_sequence() */
2677                 nfsd4_put_session(cs->session);
2678         } else if (cs->clp)
2679                 put_client_renew(cs->clp);
2680 }
2681
2682 __be32
2683 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2684 {
2685         struct nfs4_client *conf, *unconf, *clp;
2686         __be32 status = 0;
2687         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2688
2689         nfs4_lock_state();
2690         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2691         conf = find_confirmed_client(&dc->clientid, true, nn);
2692         WARN_ON_ONCE(conf && unconf);
2693
2694         if (conf) {
2695                 clp = conf;
2696
2697                 if (client_has_state(conf)) {
2698                         status = nfserr_clientid_busy;
2699                         goto out;
2700                 }
2701         } else if (unconf)
2702                 clp = unconf;
2703         else {
2704                 status = nfserr_stale_clientid;
2705                 goto out;
2706         }
2707         if (!mach_creds_match(clp, rqstp)) {
2708                 status = nfserr_wrong_cred;
2709                 goto out;
2710         }
2711         expire_client(clp);
2712 out:
2713         nfs4_unlock_state();
2714         return status;
2715 }
2716
2717 __be32
2718 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2719 {
2720         __be32 status = 0;
2721
2722         if (rc->rca_one_fs) {
2723                 if (!cstate->current_fh.fh_dentry)
2724                         return nfserr_nofilehandle;
2725                 /*
2726                  * We don't take advantage of the rca_one_fs case.
2727                  * That's OK, it's optional, we can safely ignore it.
2728                  */
2729                  return nfs_ok;
2730         }
2731
2732         nfs4_lock_state();
2733         status = nfserr_complete_already;
2734         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2735                              &cstate->session->se_client->cl_flags))
2736                 goto out;
2737
2738         status = nfserr_stale_clientid;
2739         if (is_client_expired(cstate->session->se_client))
2740                 /*
2741                  * The following error isn't really legal.
2742                  * But we only get here if the client just explicitly
2743                  * destroyed the client.  Surely it no longer cares what
2744                  * error it gets back on an operation for the dead
2745                  * client.
2746                  */
2747                 goto out;
2748
2749         status = nfs_ok;
2750         nfsd4_client_record_create(cstate->session->se_client);
2751 out:
2752         nfs4_unlock_state();
2753         return status;
2754 }
2755
2756 __be32
2757 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2758                   struct nfsd4_setclientid *setclid)
2759 {
2760         struct xdr_netobj       clname = setclid->se_name;
2761         nfs4_verifier           clverifier = setclid->se_verf;
2762         struct nfs4_client      *conf, *unconf, *new;
2763         __be32                  status;
2764         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2765
2766         /* Cases below refer to rfc 3530 section 14.2.33: */
2767         nfs4_lock_state();
2768         conf = find_confirmed_client_by_name(&clname, nn);
2769         if (conf) {
2770                 /* case 0: */
2771                 status = nfserr_clid_inuse;
2772                 if (clp_used_exchangeid(conf))
2773                         goto out;
2774                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2775                         char addr_str[INET6_ADDRSTRLEN];
2776                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2777                                  sizeof(addr_str));
2778                         dprintk("NFSD: setclientid: string in use by client "
2779                                 "at %s\n", addr_str);
2780                         goto out;
2781                 }
2782         }
2783         unconf = find_unconfirmed_client_by_name(&clname, nn);
2784         if (unconf)
2785                 expire_client(unconf);
2786         status = nfserr_jukebox;
2787         new = create_client(clname, rqstp, &clverifier);
2788         if (new == NULL)
2789                 goto out;
2790         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2791                 /* case 1: probable callback update */
2792                 copy_clid(new, conf);
2793         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2794                 gen_clid(new, nn);
2795         new->cl_minorversion = 0;
2796         gen_callback(new, setclid, rqstp);
2797         add_to_unconfirmed(new);
2798         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2799         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2800         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2801         status = nfs_ok;
2802 out:
2803         nfs4_unlock_state();
2804         return status;
2805 }
2806
2807
2808 __be32
2809 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2810                          struct nfsd4_compound_state *cstate,
2811                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2812 {
2813         struct nfs4_client *conf, *unconf;
2814         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2815         clientid_t * clid = &setclientid_confirm->sc_clientid;
2816         __be32 status;
2817         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2818
2819         if (STALE_CLIENTID(clid, nn))
2820                 return nfserr_stale_clientid;
2821         nfs4_lock_state();
2822
2823         conf = find_confirmed_client(clid, false, nn);
2824         unconf = find_unconfirmed_client(clid, false, nn);
2825         /*
2826          * We try hard to give out unique clientid's, so if we get an
2827          * attempt to confirm the same clientid with a different cred,
2828          * there's a bug somewhere.  Let's charitably assume it's our
2829          * bug.
2830          */
2831         status = nfserr_serverfault;
2832         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2833                 goto out;
2834         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2835                 goto out;
2836         /* cases below refer to rfc 3530 section 14.2.34: */
2837         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2838                 if (conf && !unconf) /* case 2: probable retransmit */
2839                         status = nfs_ok;
2840                 else /* case 4: client hasn't noticed we rebooted yet? */
2841                         status = nfserr_stale_clientid;
2842                 goto out;
2843         }
2844         status = nfs_ok;
2845         if (conf) { /* case 1: callback update */
2846                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2847                 nfsd4_probe_callback(conf);
2848                 expire_client(unconf);
2849         } else { /* case 3: normal case; new or rebooted client */
2850                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2851                 if (conf) {
2852                         status = mark_client_expired(conf);
2853                         if (status)
2854                                 goto out;
2855                         expire_client(conf);
2856                 }
2857                 move_to_confirmed(unconf);
2858                 nfsd4_probe_callback(unconf);
2859         }
2860 out:
2861         nfs4_unlock_state();
2862         return status;
2863 }
2864
2865 static struct nfs4_file *nfsd4_alloc_file(void)
2866 {
2867         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2868 }
2869
2870 /* OPEN Share state helper functions */
2871 static void nfsd4_init_file(struct nfs4_file *fp, struct knfsd_fh *fh)
2872 {
2873         unsigned int hashval = file_hashval(fh);
2874
2875         lockdep_assert_held(&state_lock);
2876
2877         atomic_set(&fp->fi_ref, 1);
2878         spin_lock_init(&fp->fi_lock);
2879         INIT_LIST_HEAD(&fp->fi_stateids);
2880         INIT_LIST_HEAD(&fp->fi_delegations);
2881         fh_copy_shallow(&fp->fi_fhandle, fh);
2882         fp->fi_had_conflict = false;
2883         fp->fi_lease = NULL;
2884         fp->fi_share_deny = 0;
2885         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2886         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2887         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2888 }
2889
2890 void
2891 nfsd4_free_slabs(void)
2892 {
2893         kmem_cache_destroy(openowner_slab);
2894         kmem_cache_destroy(lockowner_slab);
2895         kmem_cache_destroy(file_slab);
2896         kmem_cache_destroy(stateid_slab);
2897         kmem_cache_destroy(deleg_slab);
2898 }
2899
2900 int
2901 nfsd4_init_slabs(void)
2902 {
2903         openowner_slab = kmem_cache_create("nfsd4_openowners",
2904                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2905         if (openowner_slab == NULL)
2906                 goto out;
2907         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2908                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2909         if (lockowner_slab == NULL)
2910                 goto out_free_openowner_slab;
2911         file_slab = kmem_cache_create("nfsd4_files",
2912                         sizeof(struct nfs4_file), 0, 0, NULL);
2913         if (file_slab == NULL)
2914                 goto out_free_lockowner_slab;
2915         stateid_slab = kmem_cache_create("nfsd4_stateids",
2916                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2917         if (stateid_slab == NULL)
2918                 goto out_free_file_slab;
2919         deleg_slab = kmem_cache_create("nfsd4_delegations",
2920                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2921         if (deleg_slab == NULL)
2922                 goto out_free_stateid_slab;
2923         return 0;
2924
2925 out_free_stateid_slab:
2926         kmem_cache_destroy(stateid_slab);
2927 out_free_file_slab:
2928         kmem_cache_destroy(file_slab);
2929 out_free_lockowner_slab:
2930         kmem_cache_destroy(lockowner_slab);
2931 out_free_openowner_slab:
2932         kmem_cache_destroy(openowner_slab);
2933 out:
2934         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2935         return -ENOMEM;
2936 }
2937
2938 static void init_nfs4_replay(struct nfs4_replay *rp)
2939 {
2940         rp->rp_status = nfserr_serverfault;
2941         rp->rp_buflen = 0;
2942         rp->rp_buf = rp->rp_ibuf;
2943 }
2944
2945 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2946 {
2947         struct nfs4_stateowner *sop;
2948
2949         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2950         if (!sop)
2951                 return NULL;
2952
2953         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2954         if (!sop->so_owner.data) {
2955                 kmem_cache_free(slab, sop);
2956                 return NULL;
2957         }
2958         sop->so_owner.len = owner->len;
2959
2960         INIT_LIST_HEAD(&sop->so_stateids);
2961         sop->so_client = clp;
2962         init_nfs4_replay(&sop->so_replay);
2963         return sop;
2964 }
2965
2966 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2967 {
2968         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2969
2970         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2971         list_add(&oo->oo_perclient, &clp->cl_openowners);
2972 }
2973
2974 static struct nfs4_openowner *
2975 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
2976                            struct nfsd4_compound_state *cstate)
2977 {
2978         struct nfs4_client *clp = cstate->clp;
2979         struct nfs4_openowner *oo;
2980
2981         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2982         if (!oo)
2983                 return NULL;
2984         oo->oo_owner.so_is_open_owner = 1;
2985         oo->oo_owner.so_seqid = open->op_seqid;
2986         oo->oo_flags = NFS4_OO_NEW;
2987         if (nfsd4_has_session(cstate))
2988                 oo->oo_flags |= NFS4_OO_CONFIRMED;
2989         oo->oo_time = 0;
2990         oo->oo_last_closed_stid = NULL;
2991         INIT_LIST_HEAD(&oo->oo_close_lru);
2992         hash_openowner(oo, clp, strhashval);
2993         return oo;
2994 }
2995
2996 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2997         struct nfs4_openowner *oo = open->op_openowner;
2998
2999         stp->st_stid.sc_type = NFS4_OPEN_STID;
3000         INIT_LIST_HEAD(&stp->st_locks);
3001         stp->st_stateowner = &oo->oo_owner;
3002         get_nfs4_file(fp);
3003         stp->st_stid.sc_file = fp;
3004         stp->st_access_bmap = 0;
3005         stp->st_deny_bmap = 0;
3006         stp->st_openstp = NULL;
3007         spin_lock(&oo->oo_owner.so_client->cl_lock);
3008         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3009         spin_lock(&fp->fi_lock);
3010         list_add(&stp->st_perfile, &fp->fi_stateids);
3011         spin_unlock(&fp->fi_lock);
3012         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3013 }
3014
3015 static void
3016 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
3017 {
3018         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3019
3020         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3021
3022         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3023         oo->oo_time = get_seconds();
3024 }
3025
3026 static int
3027 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
3028                                                         clientid_t *clid)
3029 {
3030         return (sop->so_owner.len == owner->len) &&
3031                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
3032                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
3033 }
3034
3035 static struct nfs4_openowner *
3036 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
3037                         bool sessions, struct nfsd_net *nn)
3038 {
3039         struct nfs4_stateowner *so;
3040         struct nfs4_openowner *oo;
3041         struct nfs4_client *clp;
3042
3043         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
3044                 if (!so->so_is_open_owner)
3045                         continue;
3046                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
3047                         oo = openowner(so);
3048                         clp = oo->oo_owner.so_client;
3049                         if ((bool)clp->cl_minorversion != sessions)
3050                                 return NULL;
3051                         renew_client(oo->oo_owner.so_client);
3052                         return oo;
3053                 }
3054         }
3055         return NULL;
3056 }
3057
3058 /* search file_hashtbl[] for file */
3059 static struct nfs4_file *
3060 find_file_locked(struct knfsd_fh *fh)
3061 {
3062         unsigned int hashval = file_hashval(fh);
3063         struct nfs4_file *fp;
3064
3065         lockdep_assert_held(&state_lock);
3066
3067         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
3068                 if (nfsd_fh_match(&fp->fi_fhandle, fh)) {
3069                         get_nfs4_file(fp);
3070                         return fp;
3071                 }
3072         }
3073         return NULL;
3074 }
3075
3076 static struct nfs4_file *
3077 find_file(struct knfsd_fh *fh)
3078 {
3079         struct nfs4_file *fp;
3080
3081         spin_lock(&state_lock);
3082         fp = find_file_locked(fh);
3083         spin_unlock(&state_lock);
3084         return fp;
3085 }
3086
3087 static struct nfs4_file *
3088 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3089 {
3090         struct nfs4_file *fp;
3091
3092         spin_lock(&state_lock);
3093         fp = find_file_locked(fh);
3094         if (fp == NULL) {
3095                 nfsd4_init_file(new, fh);
3096                 fp = new;
3097         }
3098         spin_unlock(&state_lock);
3099
3100         return fp;
3101 }
3102
3103 /*
3104  * Called to check deny when READ with all zero stateid or
3105  * WRITE with all zero or all one stateid
3106  */
3107 static __be32
3108 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3109 {
3110         struct nfs4_file *fp;
3111         __be32 ret = nfs_ok;
3112
3113         fp = find_file(&current_fh->fh_handle);
3114         if (!fp)
3115                 return ret;
3116         /* Check for conflicting share reservations */
3117         spin_lock(&fp->fi_lock);
3118         if (fp->fi_share_deny & deny_type)
3119                 ret = nfserr_locked;
3120         spin_unlock(&fp->fi_lock);
3121         put_nfs4_file(fp);
3122         return ret;
3123 }
3124
3125 void nfsd4_prepare_cb_recall(struct nfs4_delegation *dp)
3126 {
3127         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3128                                           nfsd_net_id);
3129
3130         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3131
3132         /*
3133          * We can't do this in nfsd_break_deleg_cb because it is
3134          * already holding inode->i_lock.
3135          *
3136          * If the dl_time != 0, then we know that it has already been
3137          * queued for a lease break. Don't queue it again.
3138          */
3139         spin_lock(&state_lock);
3140         if (dp->dl_time == 0) {
3141                 dp->dl_time = get_seconds();
3142                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3143         }
3144         spin_unlock(&state_lock);
3145 }
3146
3147 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3148 {
3149         /*
3150          * We're assuming the state code never drops its reference
3151          * without first removing the lease.  Since we're in this lease
3152          * callback (and since the lease code is serialized by the kernel
3153          * lock) we know the server hasn't removed the lease yet, we know
3154          * it's safe to take a reference.
3155          */
3156         atomic_inc(&dp->dl_stid.sc_count);
3157         nfsd4_cb_recall(dp);
3158 }
3159
3160 /* Called from break_lease() with i_lock held. */
3161 static void nfsd_break_deleg_cb(struct file_lock *fl)
3162 {
3163         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3164         struct nfs4_delegation *dp;
3165
3166         if (!fp) {
3167                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3168                 return;
3169         }
3170         if (fp->fi_had_conflict) {
3171                 WARN(1, "duplicate break on %p\n", fp);
3172                 return;
3173         }
3174         /*
3175          * We don't want the locks code to timeout the lease for us;
3176          * we'll remove it ourself if a delegation isn't returned
3177          * in time:
3178          */
3179         fl->fl_break_time = 0;
3180
3181         spin_lock(&fp->fi_lock);
3182         fp->fi_had_conflict = true;
3183         /*
3184          * If there are no delegations on the list, then we can't count on this
3185          * lease ever being cleaned up. Set the fl_break_time to jiffies so that
3186          * time_out_leases will do it ASAP. The fact that fi_had_conflict is now
3187          * true should keep any new delegations from being hashed.
3188          */
3189         if (list_empty(&fp->fi_delegations))
3190                 fl->fl_break_time = jiffies;
3191         else
3192                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3193                         nfsd_break_one_deleg(dp);
3194         spin_unlock(&fp->fi_lock);
3195 }
3196
3197 static
3198 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
3199 {
3200         if (arg & F_UNLCK)
3201                 return lease_modify(onlist, arg);
3202         else
3203                 return -EAGAIN;
3204 }
3205
3206 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3207         .lm_break = nfsd_break_deleg_cb,
3208         .lm_change = nfsd_change_deleg_cb,
3209 };
3210
3211 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3212 {
3213         if (nfsd4_has_session(cstate))
3214                 return nfs_ok;
3215         if (seqid == so->so_seqid - 1)
3216                 return nfserr_replay_me;
3217         if (seqid == so->so_seqid)
3218                 return nfs_ok;
3219         return nfserr_bad_seqid;
3220 }
3221
3222 static __be32 lookup_clientid(clientid_t *clid,
3223                 struct nfsd4_compound_state *cstate,
3224                 struct nfsd_net *nn)
3225 {
3226         struct nfs4_client *found;
3227
3228         if (cstate->clp) {
3229                 found = cstate->clp;
3230                 if (!same_clid(&found->cl_clientid, clid))
3231                         return nfserr_stale_clientid;
3232                 return nfs_ok;
3233         }
3234
3235         if (STALE_CLIENTID(clid, nn))
3236                 return nfserr_stale_clientid;
3237
3238         /*
3239          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3240          * cached already then we know this is for is for v4.0 and "sessions"
3241          * will be false.
3242          */
3243         WARN_ON_ONCE(cstate->session);
3244         found = find_confirmed_client(clid, false, nn);
3245         if (!found)
3246                 return nfserr_expired;
3247
3248         /* Cache the nfs4_client in cstate! */
3249         cstate->clp = found;
3250         atomic_inc(&found->cl_refcount);
3251         return nfs_ok;
3252 }
3253
3254 __be32
3255 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3256                     struct nfsd4_open *open, struct nfsd_net *nn)
3257 {
3258         clientid_t *clientid = &open->op_clientid;
3259         struct nfs4_client *clp = NULL;
3260         unsigned int strhashval;
3261         struct nfs4_openowner *oo = NULL;
3262         __be32 status;
3263
3264         if (STALE_CLIENTID(&open->op_clientid, nn))
3265                 return nfserr_stale_clientid;
3266         /*
3267          * In case we need it later, after we've already created the
3268          * file and don't want to risk a further failure:
3269          */
3270         open->op_file = nfsd4_alloc_file();
3271         if (open->op_file == NULL)
3272                 return nfserr_jukebox;
3273
3274         status = lookup_clientid(clientid, cstate, nn);
3275         if (status)
3276                 return status;
3277         clp = cstate->clp;
3278
3279         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
3280         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
3281         open->op_openowner = oo;
3282         if (!oo) {
3283                 goto new_owner;
3284         }
3285         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3286                 /* Replace unconfirmed owners without checking for replay. */
3287                 release_openowner(oo);
3288                 open->op_openowner = NULL;
3289                 goto new_owner;
3290         }
3291         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3292         if (status)
3293                 return status;
3294         goto alloc_stateid;
3295 new_owner:
3296         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3297         if (oo == NULL)
3298                 return nfserr_jukebox;
3299         open->op_openowner = oo;
3300 alloc_stateid:
3301         open->op_stp = nfs4_alloc_open_stateid(clp);
3302         if (!open->op_stp)
3303                 return nfserr_jukebox;
3304         return nfs_ok;
3305 }
3306
3307 static inline __be32
3308 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3309 {
3310         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3311                 return nfserr_openmode;
3312         else
3313                 return nfs_ok;
3314 }
3315
3316 static int share_access_to_flags(u32 share_access)
3317 {
3318         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3319 }
3320
3321 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3322 {
3323         struct nfs4_stid *ret;
3324
3325         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3326         if (!ret)
3327                 return NULL;
3328         /* FIXME: move into find_stateid_by_type */
3329         atomic_inc(&ret->sc_count);
3330         return delegstateid(ret);
3331 }
3332
3333 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3334 {
3335         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3336                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3337 }
3338
3339 static __be32
3340 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3341                 struct nfs4_delegation **dp)
3342 {
3343         int flags;
3344         __be32 status = nfserr_bad_stateid;
3345         struct nfs4_delegation *deleg;
3346
3347         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3348         if (deleg == NULL)
3349                 goto out;
3350         flags = share_access_to_flags(open->op_share_access);
3351         status = nfs4_check_delegmode(deleg, flags);
3352         if (status) {
3353                 nfs4_put_stid(&deleg->dl_stid);
3354                 goto out;
3355         }
3356         *dp = deleg;
3357 out:
3358         if (!nfsd4_is_deleg_cur(open))
3359                 return nfs_ok;
3360         if (status)
3361                 return status;
3362         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3363         return nfs_ok;
3364 }
3365
3366 static struct nfs4_ol_stateid *
3367 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3368 {
3369         struct nfs4_ol_stateid *local, *ret = NULL;
3370         struct nfs4_openowner *oo = open->op_openowner;
3371
3372         spin_lock(&fp->fi_lock);
3373         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3374                 /* ignore lock owners */
3375                 if (local->st_stateowner->so_is_open_owner == 0)
3376                         continue;
3377                 if (local->st_stateowner == &oo->oo_owner) {
3378                         ret = local;
3379                         break;
3380                 }
3381         }
3382         spin_unlock(&fp->fi_lock);
3383         return ret;
3384 }
3385
3386 static inline int nfs4_access_to_access(u32 nfs4_access)
3387 {
3388         int flags = 0;
3389
3390         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3391                 flags |= NFSD_MAY_READ;
3392         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3393                 flags |= NFSD_MAY_WRITE;
3394         return flags;
3395 }
3396
3397 static inline __be32
3398 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3399                 struct nfsd4_open *open)
3400 {
3401         struct iattr iattr = {
3402                 .ia_valid = ATTR_SIZE,
3403                 .ia_size = 0,
3404         };
3405         if (!open->op_truncate)
3406                 return 0;
3407         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3408                 return nfserr_inval;
3409         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3410 }
3411
3412 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3413                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3414                 struct nfsd4_open *open)
3415 {
3416         struct file *filp = NULL;
3417         __be32 status;
3418         int oflag = nfs4_access_to_omode(open->op_share_access);
3419         int access = nfs4_access_to_access(open->op_share_access);
3420         unsigned char old_access_bmap, old_deny_bmap;
3421
3422         spin_lock(&fp->fi_lock);
3423
3424         /*
3425          * Are we trying to set a deny mode that would conflict with
3426          * current access?
3427          */
3428         status = nfs4_file_check_deny(fp, open->op_share_deny);
3429         if (status != nfs_ok) {
3430                 spin_unlock(&fp->fi_lock);
3431                 goto out;
3432         }
3433
3434         /* set access to the file */
3435         status = nfs4_file_get_access(fp, open->op_share_access);
3436         if (status != nfs_ok) {
3437                 spin_unlock(&fp->fi_lock);
3438                 goto out;
3439         }
3440
3441         /* Set access bits in stateid */
3442         old_access_bmap = stp->st_access_bmap;
3443         set_access(open->op_share_access, stp);
3444
3445         /* Set new deny mask */
3446         old_deny_bmap = stp->st_deny_bmap;
3447         set_deny(open->op_share_deny, stp);
3448         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3449
3450         if (!fp->fi_fds[oflag]) {
3451                 spin_unlock(&fp->fi_lock);
3452                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3453                 if (status)
3454                         goto out_put_access;
3455                 spin_lock(&fp->fi_lock);
3456                 if (!fp->fi_fds[oflag]) {
3457                         fp->fi_fds[oflag] = filp;
3458                         filp = NULL;
3459                 }
3460         }
3461         spin_unlock(&fp->fi_lock);
3462         if (filp)
3463                 fput(filp);
3464
3465         status = nfsd4_truncate(rqstp, cur_fh, open);
3466         if (status)
3467                 goto out_put_access;
3468 out:
3469         return status;
3470 out_put_access:
3471         stp->st_access_bmap = old_access_bmap;
3472         nfs4_file_put_access(fp, open->op_share_access);
3473         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3474         goto out;
3475 }
3476
3477 static __be32
3478 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)
3479 {
3480         __be32 status;
3481         unsigned char old_deny_bmap;
3482
3483         if (!test_access(open->op_share_access, stp))
3484                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3485
3486         /* test and set deny mode */
3487         spin_lock(&fp->fi_lock);
3488         status = nfs4_file_check_deny(fp, open->op_share_deny);
3489         if (status == nfs_ok) {
3490                 old_deny_bmap = stp->st_deny_bmap;
3491                 set_deny(open->op_share_deny, stp);
3492                 fp->fi_share_deny |=
3493                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3494         }
3495         spin_unlock(&fp->fi_lock);
3496
3497         if (status != nfs_ok)
3498                 return status;
3499
3500         status = nfsd4_truncate(rqstp, cur_fh, open);
3501         if (status != nfs_ok)
3502                 reset_union_bmap_deny(old_deny_bmap, stp);
3503         return status;
3504 }
3505
3506 static void
3507 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3508 {
3509         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3510 }
3511
3512 /* Should we give out recallable state?: */
3513 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3514 {
3515         if (clp->cl_cb_state == NFSD4_CB_UP)
3516                 return true;
3517         /*
3518          * In the sessions case, since we don't have to establish a
3519          * separate connection for callbacks, we assume it's OK
3520          * until we hear otherwise:
3521          */
3522         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3523 }
3524
3525 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
3526 {
3527         struct file_lock *fl;
3528
3529         fl = locks_alloc_lock();
3530         if (!fl)
3531                 return NULL;
3532         locks_init_lock(fl);
3533         fl->fl_lmops = &nfsd_lease_mng_ops;
3534         fl->fl_flags = FL_DELEG;
3535         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3536         fl->fl_end = OFFSET_MAX;
3537         fl->fl_owner = (fl_owner_t)fp;
3538         fl->fl_pid = current->tgid;
3539         return fl;
3540 }
3541
3542 static int nfs4_setlease(struct nfs4_delegation *dp)
3543 {
3544         struct nfs4_file *fp = dp->dl_stid.sc_file;
3545         struct file_lock *fl;
3546         struct file *filp;
3547         int status = 0;
3548
3549         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
3550         if (!fl)
3551                 return -ENOMEM;
3552         filp = find_readable_file(fp);
3553         if (!filp) {
3554                 /* We should always have a readable file here */
3555                 WARN_ON_ONCE(1);
3556                 return -EBADF;
3557         }
3558         fl->fl_file = filp;
3559         status = vfs_setlease(filp, fl->fl_type, &fl);
3560         if (status) {
3561                 locks_free_lock(fl);
3562                 goto out_fput;
3563         }
3564         spin_lock(&state_lock);
3565         spin_lock(&fp->fi_lock);
3566         /* Did the lease get broken before we took the lock? */
3567         status = -EAGAIN;
3568         if (fp->fi_had_conflict)
3569                 goto out_unlock;
3570         /* Race breaker */
3571         if (fp->fi_lease) {
3572                 status = 0;
3573                 atomic_inc(&fp->fi_delegees);
3574                 hash_delegation_locked(dp, fp);
3575                 goto out_unlock;
3576         }
3577         fp->fi_lease = fl;
3578         fp->fi_deleg_file = filp;
3579         atomic_set(&fp->fi_delegees, 1);
3580         hash_delegation_locked(dp, fp);
3581         spin_unlock(&fp->fi_lock);
3582         spin_unlock(&state_lock);
3583         return 0;
3584 out_unlock:
3585         spin_unlock(&fp->fi_lock);
3586         spin_unlock(&state_lock);
3587 out_fput:
3588         fput(filp);
3589         return status;
3590 }
3591
3592 static struct nfs4_delegation *
3593 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
3594                     struct nfs4_file *fp)
3595 {
3596         int status;
3597         struct nfs4_delegation *dp;
3598
3599         if (fp->fi_had_conflict)
3600                 return ERR_PTR(-EAGAIN);
3601
3602         dp = alloc_init_deleg(clp, fh);
3603         if (!dp)
3604                 return ERR_PTR(-ENOMEM);
3605
3606         get_nfs4_file(fp);
3607         spin_lock(&state_lock);
3608         spin_lock(&fp->fi_lock);
3609         dp->dl_stid.sc_file = fp;
3610         if (!fp->fi_lease) {
3611                 spin_unlock(&fp->fi_lock);
3612                 spin_unlock(&state_lock);
3613                 status = nfs4_setlease(dp);
3614                 goto out;
3615         }
3616         atomic_inc(&fp->fi_delegees);
3617         if (fp->fi_had_conflict) {
3618                 status = -EAGAIN;
3619                 goto out_unlock;
3620         }
3621         hash_delegation_locked(dp, fp);
3622         status = 0;
3623 out_unlock:
3624         spin_unlock(&fp->fi_lock);
3625         spin_unlock(&state_lock);
3626 out:
3627         if (status) {
3628                 nfs4_put_stid(&dp->dl_stid);
3629                 return ERR_PTR(status);
3630         }
3631         return dp;
3632 }
3633
3634 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3635 {
3636         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3637         if (status == -EAGAIN)
3638                 open->op_why_no_deleg = WND4_CONTENTION;
3639         else {
3640                 open->op_why_no_deleg = WND4_RESOURCE;
3641                 switch (open->op_deleg_want) {
3642                 case NFS4_SHARE_WANT_READ_DELEG:
3643                 case NFS4_SHARE_WANT_WRITE_DELEG:
3644                 case NFS4_SHARE_WANT_ANY_DELEG:
3645                         break;
3646                 case NFS4_SHARE_WANT_CANCEL:
3647                         open->op_why_no_deleg = WND4_CANCELLED;
3648                         break;
3649                 case NFS4_SHARE_WANT_NO_DELEG:
3650                         WARN_ON_ONCE(1);
3651                 }
3652         }
3653 }
3654
3655 /*
3656  * Attempt to hand out a delegation.
3657  *
3658  * Note we don't support write delegations, and won't until the vfs has
3659  * proper support for them.
3660  */
3661 static void
3662 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
3663                         struct nfs4_ol_stateid *stp)
3664 {
3665         struct nfs4_delegation *dp;
3666         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
3667         struct nfs4_client *clp = stp->st_stid.sc_client;
3668         int cb_up;
3669         int status = 0;
3670
3671         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3672         open->op_recall = 0;
3673         switch (open->op_claim_type) {
3674                 case NFS4_OPEN_CLAIM_PREVIOUS:
3675                         if (!cb_up)
3676                                 open->op_recall = 1;
3677                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3678                                 goto out_no_deleg;
3679                         break;
3680                 case NFS4_OPEN_CLAIM_NULL:
3681                 case NFS4_OPEN_CLAIM_FH:
3682                         /*
3683                          * Let's not give out any delegations till everyone's
3684                          * had the chance to reclaim theirs....
3685                          */
3686                         if (locks_in_grace(clp->net))
3687                                 goto out_no_deleg;
3688                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3689                                 goto out_no_deleg;
3690                         /*
3691                          * Also, if the file was opened for write or
3692                          * create, there's a good chance the client's
3693                          * about to write to it, resulting in an
3694                          * immediate recall (since we don't support
3695                          * write delegations):
3696                          */
3697                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3698                                 goto out_no_deleg;
3699                         if (open->op_create == NFS4_OPEN_CREATE)
3700                                 goto out_no_deleg;
3701                         break;
3702                 default:
3703                         goto out_no_deleg;
3704         }
3705         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file);
3706         if (IS_ERR(dp))
3707                 goto out_no_deleg;
3708
3709         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3710
3711         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3712                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3713         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3714         nfs4_put_stid(&dp->dl_stid);
3715         return;
3716 out_no_deleg:
3717         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3718         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3719             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3720                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3721                 open->op_recall = 1;
3722         }
3723
3724         /* 4.1 client asking for a delegation? */
3725         if (open->op_deleg_want)
3726                 nfsd4_open_deleg_none_ext(open, status);
3727         return;
3728 }
3729
3730 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3731                                         struct nfs4_delegation *dp)
3732 {
3733         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3734             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3735                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3736                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3737         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3738                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3739                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3740                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3741         }
3742         /* Otherwise the client must be confused wanting a delegation
3743          * it already has, therefore we don't return
3744          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3745          */
3746 }
3747
3748 /*
3749  * called with nfs4_lock_state() held.
3750  */
3751 __be32
3752 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3753 {
3754         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3755         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3756         struct nfs4_file *fp = NULL;
3757         struct nfs4_ol_stateid *stp = NULL;
3758         struct nfs4_delegation *dp = NULL;
3759         __be32 status;
3760
3761         /*
3762          * Lookup file; if found, lookup stateid and check open request,
3763          * and check for delegations in the process of being recalled.
3764          * If not found, create the nfs4_file struct
3765          */
3766         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
3767         if (fp != open->op_file) {
3768                 status = nfs4_check_deleg(cl, open, &dp);
3769                 if (status)
3770                         goto out;
3771                 stp = nfsd4_find_existing_open(fp, open);
3772         } else {
3773                 open->op_file = NULL;
3774                 status = nfserr_bad_stateid;
3775                 if (nfsd4_is_deleg_cur(open))
3776                         goto out;
3777                 status = nfserr_jukebox;
3778         }
3779
3780         /*
3781          * OPEN the file, or upgrade an existing OPEN.
3782          * If truncate fails, the OPEN fails.
3783          */
3784         if (stp) {
3785                 /* Stateid was found, this is an OPEN upgrade */
3786                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3787                 if (status)
3788                         goto out;
3789         } else {
3790                 stp = open->op_stp;
3791                 open->op_stp = NULL;
3792                 init_open_stateid(stp, fp, open);
3793                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
3794                 if (status) {
3795                         release_open_stateid(stp);
3796                         goto out;
3797                 }
3798         }
3799         update_stateid(&stp->st_stid.sc_stateid);
3800         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3801
3802         if (nfsd4_has_session(&resp->cstate)) {
3803                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3804                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3805                         open->op_why_no_deleg = WND4_NOT_WANTED;
3806                         goto nodeleg;
3807                 }
3808         }
3809
3810         /*
3811         * Attempt to hand out a delegation. No error return, because the
3812         * OPEN succeeds even if we fail.
3813         */
3814         nfs4_open_delegation(current_fh, open, stp);
3815 nodeleg:
3816         status = nfs_ok;
3817
3818         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3819                 STATEID_VAL(&stp->st_stid.sc_stateid));
3820 out:
3821         /* 4.1 client trying to upgrade/downgrade delegation? */
3822         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3823             open->op_deleg_want)
3824                 nfsd4_deleg_xgrade_none_ext(open, dp);
3825
3826         if (fp)
3827                 put_nfs4_file(fp);
3828         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3829                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3830         /*
3831         * To finish the open response, we just need to set the rflags.
3832         */
3833         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3834         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3835             !nfsd4_has_session(&resp->cstate))
3836                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3837         if (dp)
3838                 nfs4_put_stid(&dp->dl_stid);
3839
3840         return status;
3841 }
3842
3843 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3844 {
3845         if (open->op_openowner) {
3846                 struct nfs4_openowner *oo = open->op_openowner;
3847
3848                 if (!list_empty(&oo->oo_owner.so_stateids))
3849                         list_del_init(&oo->oo_close_lru);
3850                 if (oo->oo_flags & NFS4_OO_NEW) {
3851                         if (status) {
3852                                 release_openowner(oo);
3853                                 open->op_openowner = NULL;
3854                         } else
3855                                 oo->oo_flags &= ~NFS4_OO_NEW;
3856                 }
3857         }
3858         if (open->op_file)
3859                 nfsd4_free_file(open->op_file);
3860         if (open->op_stp)
3861                 nfs4_put_stid(&open->op_stp->st_stid);
3862 }
3863
3864 __be32
3865 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3866             clientid_t *clid)
3867 {
3868         struct nfs4_client *clp;
3869         __be32 status;
3870         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3871
3872         nfs4_lock_state();
3873         dprintk("process_renew(%08x/%08x): starting\n",
3874                         clid->cl_boot, clid->cl_id);
3875         status = lookup_clientid(clid, cstate, nn);
3876         if (status)
3877                 goto out;
3878         clp = cstate->clp;
3879         status = nfserr_cb_path_down;
3880         if (!list_empty(&clp->cl_delegations)
3881                         && clp->cl_cb_state != NFSD4_CB_UP)
3882                 goto out;
3883         status = nfs_ok;
3884 out:
3885         nfs4_unlock_state();
3886         return status;
3887 }
3888
3889 static void
3890 nfsd4_end_grace(struct nfsd_net *nn)
3891 {
3892         /* do nothing if grace period already ended */
3893         if (nn->grace_ended)
3894                 return;
3895
3896         dprintk("NFSD: end of grace period\n");
3897         nn->grace_ended = true;
3898         nfsd4_record_grace_done(nn, nn->boot_time);
3899         locks_end_grace(&nn->nfsd4_manager);
3900         /*
3901          * Now that every NFSv4 client has had the chance to recover and
3902          * to see the (possibly new, possibly shorter) lease time, we
3903          * can safely set the next grace time to the current lease time:
3904          */
3905         nn->nfsd4_grace = nn->nfsd4_lease;
3906 }
3907
3908 static time_t
3909 nfs4_laundromat(struct nfsd_net *nn)
3910 {
3911         struct nfs4_client *clp;
3912         struct nfs4_openowner *oo;
3913         struct nfs4_delegation *dp;
3914         struct list_head *pos, *next, reaplist;
3915         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3916         time_t t, new_timeo = nn->nfsd4_lease;
3917
3918         nfs4_lock_state();
3919
3920         dprintk("NFSD: laundromat service - starting\n");
3921         nfsd4_end_grace(nn);
3922         INIT_LIST_HEAD(&reaplist);
3923         spin_lock(&nn->client_lock);
3924         list_for_each_safe(pos, next, &nn->client_lru) {
3925                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3926                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3927                         t = clp->cl_time - cutoff;
3928                         new_timeo = min(new_timeo, t);
3929                         break;
3930                 }
3931                 if (mark_client_expired_locked(clp)) {
3932                         dprintk("NFSD: client in use (clientid %08x)\n",
3933                                 clp->cl_clientid.cl_id);
3934                         continue;
3935                 }
3936                 list_move(&clp->cl_lru, &reaplist);
3937         }
3938         spin_unlock(&nn->client_lock);
3939         list_for_each_safe(pos, next, &reaplist) {
3940                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3941                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3942                         clp->cl_clientid.cl_id);
3943                 expire_client(clp);
3944         }
3945         spin_lock(&state_lock);
3946         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3947                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3948                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3949                         continue;
3950                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3951                         t = dp->dl_time - cutoff;
3952                         new_timeo = min(new_timeo, t);
3953                         break;
3954                 }
3955                 unhash_delegation_locked(dp);
3956                 list_add(&dp->dl_recall_lru, &reaplist);
3957         }
3958         spin_unlock(&state_lock);
3959         while (!list_empty(&reaplist)) {
3960                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
3961                                         dl_recall_lru);
3962                 list_del_init(&dp->dl_recall_lru);
3963                 revoke_delegation(dp);
3964         }
3965         list_for_each_safe(pos, next, &nn->close_lru) {
3966                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3967                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3968                         t = oo->oo_time - cutoff;
3969                         new_timeo = min(new_timeo, t);
3970                         break;
3971                 }
3972                 release_openowner(oo);
3973         }
3974         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
3975         nfs4_unlock_state();
3976         return new_timeo;
3977 }
3978
3979 static struct workqueue_struct *laundry_wq;
3980 static void laundromat_main(struct work_struct *);
3981
3982 static void
3983 laundromat_main(struct work_struct *laundry)
3984 {
3985         time_t t;
3986         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3987                                                   work);
3988         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3989                                            laundromat_work);
3990
3991         t = nfs4_laundromat(nn);
3992         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3993         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3994 }
3995
3996 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3997 {
3998         if (!nfsd_fh_match(&fhp->fh_handle, &stp->st_stid.sc_file->fi_fhandle))
3999                 return nfserr_bad_stateid;
4000         return nfs_ok;
4001 }
4002
4003 static inline int
4004 access_permit_read(struct nfs4_ol_stateid *stp)
4005 {
4006         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4007                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4008                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4009 }
4010
4011 static inline int
4012 access_permit_write(struct nfs4_ol_stateid *stp)
4013 {
4014         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4015                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4016 }
4017
4018 static
4019 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4020 {
4021         __be32 status = nfserr_openmode;
4022
4023         /* For lock stateid's, we test the parent open, not the lock: */
4024         if (stp->st_openstp)
4025                 stp = stp->st_openstp;
4026         if ((flags & WR_STATE) && !access_permit_write(stp))
4027                 goto out;
4028         if ((flags & RD_STATE) && !access_permit_read(stp))
4029                 goto out;
4030         status = nfs_ok;
4031 out:
4032         return status;
4033 }
4034
4035 static inline __be32
4036 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4037 {
4038         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4039                 return nfs_ok;
4040         else if (locks_in_grace(net)) {
4041                 /* Answer in remaining cases depends on existence of
4042                  * conflicting state; so we must wait out the grace period. */
4043                 return nfserr_grace;
4044         } else if (flags & WR_STATE)
4045                 return nfs4_share_conflict(current_fh,
4046                                 NFS4_SHARE_DENY_WRITE);
4047         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4048                 return nfs4_share_conflict(current_fh,
4049                                 NFS4_SHARE_DENY_READ);
4050 }
4051
4052 /*
4053  * Allow READ/WRITE during grace period on recovered state only for files
4054  * that are not able to provide mandatory locking.
4055  */
4056 static inline int
4057 grace_disallows_io(struct net *net, struct inode *inode)
4058 {
4059         return locks_in_grace(net) && mandatory_lock(inode);
4060 }
4061
4062 /* Returns true iff a is later than b: */
4063 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4064 {
4065         return (s32)(a->si_generation - b->si_generation) > 0;
4066 }
4067
4068 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4069 {
4070         /*
4071          * When sessions are used the stateid generation number is ignored
4072          * when it is zero.
4073          */
4074         if (has_session && in->si_generation == 0)
4075                 return nfs_ok;
4076
4077         if (in->si_generation == ref->si_generation)
4078                 return nfs_ok;
4079
4080         /* If the client sends us a stateid from the future, it's buggy: */
4081         if (stateid_generation_after(in, ref))
4082                 return nfserr_bad_stateid;
4083         /*
4084          * However, we could see a stateid from the past, even from a
4085          * non-buggy client.  For example, if the client sends a lock
4086          * while some IO is outstanding, the lock may bump si_generation
4087          * while the IO is still in flight.  The client could avoid that
4088          * situation by waiting for responses on all the IO requests,
4089          * but better performance may result in retrying IO that
4090          * receives an old_stateid error if requests are rarely
4091          * reordered in flight:
4092          */
4093         return nfserr_old_stateid;
4094 }
4095
4096 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4097 {
4098         struct nfs4_stid *s;
4099         struct nfs4_ol_stateid *ols;
4100         __be32 status = nfserr_bad_stateid;
4101
4102         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4103                 return status;
4104         /* Client debugging aid. */
4105         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4106                 char addr_str[INET6_ADDRSTRLEN];
4107                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4108                                  sizeof(addr_str));
4109                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4110                                         "with incorrect client ID\n", addr_str);
4111                 return status;
4112         }
4113         spin_lock(&cl->cl_lock);
4114         s = find_stateid_locked(cl, stateid);
4115         if (!s)
4116                 goto out_unlock;
4117         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4118         if (status)
4119                 goto out_unlock;
4120         switch (s->sc_type) {
4121         case NFS4_DELEG_STID:
4122                 status = nfs_ok;
4123                 break;
4124         case NFS4_REVOKED_DELEG_STID:
4125                 status = nfserr_deleg_revoked;
4126                 break;
4127         case NFS4_OPEN_STID:
4128         case NFS4_LOCK_STID:
4129                 ols = openlockstateid(s);
4130                 if (ols->st_stateowner->so_is_open_owner
4131                                 && !(openowner(ols->st_stateowner)->oo_flags
4132                                                 & NFS4_OO_CONFIRMED))
4133                         status = nfserr_bad_stateid;
4134                 else
4135                         status = nfs_ok;
4136                 break;
4137         default:
4138                 printk("unknown stateid type %x\n", s->sc_type);
4139                 /* Fallthrough */
4140         case NFS4_CLOSED_STID:
4141         case NFS4_CLOSED_DELEG_STID:
4142                 status = nfserr_bad_stateid;
4143         }
4144 out_unlock:
4145         spin_unlock(&cl->cl_lock);
4146         return status;
4147 }
4148
4149 static __be32
4150 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4151                      stateid_t *stateid, unsigned char typemask,
4152                      struct nfs4_stid **s, struct nfsd_net *nn)
4153 {
4154         __be32 status;
4155
4156         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4157                 return nfserr_bad_stateid;
4158         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4159         if (status == nfserr_stale_clientid) {
4160                 if (cstate->session)
4161                         return nfserr_bad_stateid;
4162                 return nfserr_stale_stateid;
4163         }
4164         if (status)
4165                 return status;
4166         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4167         if (!*s)
4168                 return nfserr_bad_stateid;
4169         return nfs_ok;
4170 }
4171
4172 /*
4173 * Checks for stateid operations
4174 */
4175 __be32
4176 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
4177                            stateid_t *stateid, int flags, struct file **filpp)
4178 {
4179         struct nfs4_stid *s;
4180         struct nfs4_ol_stateid *stp = NULL;
4181         struct nfs4_delegation *dp = NULL;
4182         struct svc_fh *current_fh = &cstate->current_fh;
4183         struct inode *ino = current_fh->fh_dentry->d_inode;
4184         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4185         struct file *file = NULL;
4186         __be32 status;
4187
4188         if (filpp)
4189                 *filpp = NULL;
4190
4191         if (grace_disallows_io(net, ino))
4192                 return nfserr_grace;
4193
4194         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4195                 return check_special_stateids(net, current_fh, stateid, flags);
4196
4197         nfs4_lock_state();
4198
4199         status = nfsd4_lookup_stateid(cstate, stateid,
4200                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4201                                 &s, nn);
4202         if (status)
4203                 goto out;
4204         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
4205         if (status)
4206                 goto out;
4207         switch (s->sc_type) {
4208         case NFS4_DELEG_STID:
4209                 dp = delegstateid(s);
4210                 status = nfs4_check_delegmode(dp, flags);
4211                 if (status)
4212                         goto out;
4213                 if (filpp) {
4214                         file = dp->dl_stid.sc_file->fi_deleg_file;
4215                         if (!file) {
4216                                 WARN_ON_ONCE(1);
4217                                 status = nfserr_serverfault;
4218                                 goto out;
4219                         }
4220                         get_file(file);
4221                 }
4222                 break;
4223         case NFS4_OPEN_STID:
4224         case NFS4_LOCK_STID:
4225                 stp = openlockstateid(s);
4226                 status = nfs4_check_fh(current_fh, stp);
4227                 if (status)
4228                         goto out;
4229                 if (stp->st_stateowner->so_is_open_owner
4230                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4231                         goto out;
4232                 status = nfs4_check_openmode(stp, flags);
4233                 if (status)
4234                         goto out;
4235                 if (filpp) {
4236                         struct nfs4_file *fp = stp->st_stid.sc_file;
4237
4238                         if (flags & RD_STATE)
4239                                 file = find_readable_file(fp);
4240                         else
4241                                 file = find_writeable_file(fp);
4242                 }
4243                 break;
4244         default:
4245                 status = nfserr_bad_stateid;
4246                 goto out;
4247         }
4248         status = nfs_ok;
4249         if (file)
4250                 *filpp = file;
4251 out:
4252         nfs4_unlock_state();
4253         return status;
4254 }
4255
4256 static __be32
4257 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
4258 {
4259         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
4260
4261         if (check_for_locks(stp->st_stid.sc_file, lo))
4262                 return nfserr_locks_held;
4263         release_lockowner_if_empty(lo);
4264         return nfs_ok;
4265 }
4266
4267 /*
4268  * Test if the stateid is valid
4269  */
4270 __be32
4271 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4272                    struct nfsd4_test_stateid *test_stateid)
4273 {
4274         struct nfsd4_test_stateid_id *stateid;
4275         struct nfs4_client *cl = cstate->session->se_client;
4276
4277         nfs4_lock_state();
4278         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4279                 stateid->ts_id_status =
4280                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4281         nfs4_unlock_state();
4282
4283         return nfs_ok;
4284 }
4285
4286 __be32
4287 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4288                    struct nfsd4_free_stateid *free_stateid)
4289 {
4290         stateid_t *stateid = &free_stateid->fr_stateid;
4291         struct nfs4_stid *s;
4292         struct nfs4_delegation *dp;
4293         struct nfs4_client *cl = cstate->session->se_client;
4294         __be32 ret = nfserr_bad_stateid;
4295
4296         nfs4_lock_state();
4297         spin_lock(&cl->cl_lock);
4298         s = find_stateid_locked(cl, stateid);
4299         if (!s)
4300                 goto out_unlock;
4301         switch (s->sc_type) {
4302         case NFS4_DELEG_STID:
4303                 ret = nfserr_locks_held;
4304                 break;
4305         case NFS4_OPEN_STID:
4306                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4307                 if (ret)
4308                         break;
4309                 ret = nfserr_locks_held;
4310                 break;
4311         case NFS4_LOCK_STID:
4312                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4313                 if (ret)
4314                         break;
4315                 spin_unlock(&cl->cl_lock);
4316                 ret = nfsd4_free_lock_stateid(openlockstateid(s));
4317                 goto out;
4318         case NFS4_REVOKED_DELEG_STID:
4319                 dp = delegstateid(s);
4320                 list_del_init(&dp->dl_recall_lru);
4321                 spin_unlock(&cl->cl_lock);
4322                 nfs4_put_stid(s);
4323                 ret = nfs_ok;
4324                 goto out;
4325         /* Default falls through and returns nfserr_bad_stateid */
4326         }
4327 out_unlock:
4328         spin_unlock(&cl->cl_lock);
4329 out:
4330         nfs4_unlock_state();
4331         return ret;
4332 }
4333
4334 static inline int
4335 setlkflg (int type)
4336 {
4337         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4338                 RD_STATE : WR_STATE;
4339 }
4340
4341 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4342 {
4343         struct svc_fh *current_fh = &cstate->current_fh;
4344         struct nfs4_stateowner *sop = stp->st_stateowner;
4345         __be32 status;
4346
4347         status = nfsd4_check_seqid(cstate, sop, seqid);
4348         if (status)
4349                 return status;
4350         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4351                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4352                 /*
4353                  * "Closed" stateid's exist *only* to return
4354                  * nfserr_replay_me from the previous step, and
4355                  * revoked delegations are kept only for free_stateid.
4356                  */
4357                 return nfserr_bad_stateid;
4358         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4359         if (status)
4360                 return status;
4361         return nfs4_check_fh(current_fh, stp);
4362 }
4363
4364 /*
4365  * Checks for sequence id mutating operations.
4366  */
4367 static __be32
4368 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4369                          stateid_t *stateid, char typemask,
4370                          struct nfs4_ol_stateid **stpp,
4371                          struct nfsd_net *nn)
4372 {
4373         __be32 status;
4374         struct nfs4_stid *s;
4375         struct nfs4_ol_stateid *stp = NULL;
4376
4377         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4378                 seqid, STATEID_VAL(stateid));
4379
4380         *stpp = NULL;
4381         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4382         if (status)
4383                 return status;
4384         stp = openlockstateid(s);
4385         if (!nfsd4_has_session(cstate))
4386                 cstate->replay_owner = stp->st_stateowner;
4387
4388         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4389         if (!status)
4390                 *stpp = stp;
4391         return status;
4392 }
4393
4394 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4395                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4396 {
4397         __be32 status;
4398         struct nfs4_openowner *oo;
4399
4400         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4401                                                 NFS4_OPEN_STID, stpp, nn);
4402         if (status)
4403                 return status;
4404         oo = openowner((*stpp)->st_stateowner);
4405         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
4406                 return nfserr_bad_stateid;
4407         return nfs_ok;
4408 }
4409
4410 __be32
4411 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4412                    struct nfsd4_open_confirm *oc)
4413 {
4414         __be32 status;
4415         struct nfs4_openowner *oo;
4416         struct nfs4_ol_stateid *stp;
4417         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4418
4419         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4420                         cstate->current_fh.fh_dentry);
4421
4422         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4423         if (status)
4424                 return status;
4425
4426         nfs4_lock_state();
4427
4428         status = nfs4_preprocess_seqid_op(cstate,
4429                                         oc->oc_seqid, &oc->oc_req_stateid,
4430                                         NFS4_OPEN_STID, &stp, nn);
4431         if (status)
4432                 goto out;
4433         oo = openowner(stp->st_stateowner);
4434         status = nfserr_bad_stateid;
4435         if (oo->oo_flags & NFS4_OO_CONFIRMED)
4436                 goto out;
4437         oo->oo_flags |= NFS4_OO_CONFIRMED;
4438         update_stateid(&stp->st_stid.sc_stateid);
4439         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4440         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
4441                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
4442
4443         nfsd4_client_record_create(oo->oo_owner.so_client);
4444         status = nfs_ok;
4445 out:
4446         nfsd4_bump_seqid(cstate, status);
4447         if (!cstate->replay_owner)
4448                 nfs4_unlock_state();
4449         return status;
4450 }
4451
4452 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
4453 {
4454         if (!test_access(access, stp))
4455                 return;
4456         nfs4_file_put_access(stp->st_stid.sc_file, access);
4457         clear_access(access, stp);
4458 }
4459
4460 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
4461 {
4462         switch (to_access) {
4463         case NFS4_SHARE_ACCESS_READ:
4464                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
4465                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4466                 break;
4467         case NFS4_SHARE_ACCESS_WRITE:
4468                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
4469                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4470                 break;
4471         case NFS4_SHARE_ACCESS_BOTH:
4472                 break;
4473         default:
4474                 WARN_ON_ONCE(1);
4475         }
4476 }
4477
4478 __be32
4479 nfsd4_open_downgrade(struct svc_rqst *rqstp,
4480                      struct nfsd4_compound_state *cstate,
4481                      struct nfsd4_open_downgrade *od)
4482 {
4483         __be32 status;
4484         struct nfs4_ol_stateid *stp;
4485         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4486
4487         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
4488                         cstate->current_fh.fh_dentry);
4489
4490         /* We don't yet support WANT bits: */
4491         if (od->od_deleg_want)
4492                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
4493                         od->od_deleg_want);
4494
4495         nfs4_lock_state();
4496         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
4497                                         &od->od_stateid, &stp, nn);
4498         if (status)
4499                 goto out;
4500         status = nfserr_inval;
4501         if (!test_access(od->od_share_access, stp)) {
4502                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
4503                         stp->st_access_bmap, od->od_share_access);
4504                 goto out;
4505         }
4506         if (!test_deny(od->od_share_deny, stp)) {
4507                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
4508                         stp->st_deny_bmap, od->od_share_deny);
4509                 goto out;
4510         }
4511         nfs4_stateid_downgrade(stp, od->od_share_access);
4512
4513         reset_union_bmap_deny(od->od_share_deny, stp);
4514
4515         update_stateid(&stp->st_stid.sc_stateid);
4516         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4517         status = nfs_ok;
4518 out:
4519         nfsd4_bump_seqid(cstate, status);
4520         if (!cstate->replay_owner)
4521                 nfs4_unlock_state();
4522         return status;
4523 }
4524
4525 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
4526 {
4527         struct nfs4_client *clp = s->st_stid.sc_client;
4528         struct nfs4_openowner *oo = openowner(s->st_stateowner);
4529
4530         s->st_stid.sc_type = NFS4_CLOSED_STID;
4531         unhash_open_stateid(s);
4532
4533         if (clp->cl_minorversion) {
4534                 if (list_empty(&oo->oo_owner.so_stateids))
4535                         release_openowner(oo);
4536                 nfs4_put_stid(&s->st_stid);
4537         } else {
4538                 /*
4539                  * In the 4.0 case we need to keep the owners around a
4540                  * little while to handle CLOSE replay. We still do need
4541                  * to release any file access that is held by them
4542                  * before returning however.
4543                  */
4544                 release_all_access(s);
4545                 if (s->st_stid.sc_file) {
4546                         put_nfs4_file(s->st_stid.sc_file);
4547                         s->st_stid.sc_file = NULL;
4548                 }
4549                 oo->oo_last_closed_stid = s;
4550                 if (list_empty(&oo->oo_owner.so_stateids))
4551                         move_to_close_lru(oo, clp->net);
4552         }
4553 }
4554
4555 /*
4556  * nfs4_unlock_state() called after encode
4557  */
4558 __be32
4559 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4560             struct nfsd4_close *close)
4561 {
4562         __be32 status;
4563         struct nfs4_ol_stateid *stp;
4564         struct net *net = SVC_NET(rqstp);
4565         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4566
4567         dprintk("NFSD: nfsd4_close on file %pd\n",
4568                         cstate->current_fh.fh_dentry);
4569
4570         nfs4_lock_state();
4571         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4572                                         &close->cl_stateid,
4573                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4574                                         &stp, nn);
4575         nfsd4_bump_seqid(cstate, status);
4576         if (status)
4577                 goto out;
4578         update_stateid(&stp->st_stid.sc_stateid);
4579         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4580
4581         nfsd4_close_open_stateid(stp);
4582 out:
4583         if (!cstate->replay_owner)
4584                 nfs4_unlock_state();
4585         return status;
4586 }
4587
4588 __be32
4589 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4590                   struct nfsd4_delegreturn *dr)
4591 {
4592         struct nfs4_delegation *dp;
4593         stateid_t *stateid = &dr->dr_stateid;
4594         struct nfs4_stid *s;
4595         __be32 status;
4596         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4597
4598         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4599                 return status;
4600
4601         nfs4_lock_state();
4602         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
4603         if (status)
4604                 goto out;
4605         dp = delegstateid(s);
4606         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4607         if (status)
4608                 goto out;
4609
4610         destroy_delegation(dp);
4611 out:
4612         nfs4_unlock_state();
4613
4614         return status;
4615 }
4616
4617
4618 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4619
4620 static inline u64
4621 end_offset(u64 start, u64 len)
4622 {
4623         u64 end;
4624
4625         end = start + len;
4626         return end >= start ? end: NFS4_MAX_UINT64;
4627 }
4628
4629 /* last octet in a range */
4630 static inline u64
4631 last_byte_offset(u64 start, u64 len)
4632 {
4633         u64 end;
4634
4635         WARN_ON_ONCE(!len);
4636         end = start + len;
4637         return end > start ? end - 1: NFS4_MAX_UINT64;
4638 }
4639
4640 /*
4641  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4642  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4643  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4644  * locking, this prevents us from being completely protocol-compliant.  The
4645  * real solution to this problem is to start using unsigned file offsets in
4646  * the VFS, but this is a very deep change!
4647  */
4648 static inline void
4649 nfs4_transform_lock_offset(struct file_lock *lock)
4650 {
4651         if (lock->fl_start < 0)
4652                 lock->fl_start = OFFSET_MAX;
4653         if (lock->fl_end < 0)
4654                 lock->fl_end = OFFSET_MAX;
4655 }
4656
4657 /* Hack!: For now, we're defining this just so we can use a pointer to it
4658  * as a unique cookie to identify our (NFSv4's) posix locks. */
4659 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4660 };
4661
4662 static inline void
4663 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4664 {
4665         struct nfs4_lockowner *lo;
4666
4667         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4668                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4669                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4670                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4671                 if (!deny->ld_owner.data)
4672                         /* We just don't care that much */
4673                         goto nevermind;
4674                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4675                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4676         } else {
4677 nevermind:
4678                 deny->ld_owner.len = 0;
4679                 deny->ld_owner.data = NULL;
4680                 deny->ld_clientid.cl_boot = 0;
4681                 deny->ld_clientid.cl_id = 0;
4682         }
4683         deny->ld_start = fl->fl_start;
4684         deny->ld_length = NFS4_MAX_UINT64;
4685         if (fl->fl_end != NFS4_MAX_UINT64)
4686                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4687         deny->ld_type = NFS4_READ_LT;
4688         if (fl->fl_type != F_RDLCK)
4689                 deny->ld_type = NFS4_WRITE_LT;
4690 }
4691
4692 static struct nfs4_lockowner *
4693 find_lockowner_str(clientid_t *clid, struct xdr_netobj *owner,
4694                 struct nfsd_net *nn)
4695 {
4696         unsigned int strhashval = ownerstr_hashval(clid->cl_id, owner);
4697         struct nfs4_stateowner *so;
4698
4699         list_for_each_entry(so, &nn->ownerstr_hashtbl[strhashval], so_strhash) {
4700                 if (so->so_is_open_owner)
4701                         continue;
4702                 if (!same_owner_str(so, owner, clid))
4703                         continue;
4704                 return lockowner(so);
4705         }
4706         return NULL;
4707 }
4708
4709 /*
4710  * Alloc a lock owner structure.
4711  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4712  * occurred.
4713  *
4714  * strhashval = ownerstr_hashval
4715  */
4716 static struct nfs4_lockowner *
4717 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4718         struct nfs4_lockowner *lo;
4719         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4720
4721         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4722         if (!lo)
4723                 return NULL;
4724         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4725         lo->lo_owner.so_is_open_owner = 0;
4726         /* It is the openowner seqid that will be incremented in encode in the
4727          * case of new lockowners; so increment the lock seqid manually: */
4728         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4729         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4730         return lo;
4731 }
4732
4733 static void
4734 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
4735                   struct nfs4_file *fp, struct inode *inode,
4736                   struct nfs4_ol_stateid *open_stp)
4737 {
4738         struct nfs4_client *clp = lo->lo_owner.so_client;
4739
4740         lockdep_assert_held(&clp->cl_lock);
4741
4742         atomic_inc(&stp->st_stid.sc_count);
4743         stp->st_stid.sc_type = NFS4_LOCK_STID;
4744         stp->st_stateowner = &lo->lo_owner;
4745         get_nfs4_file(fp);
4746         stp->st_stid.sc_file = fp;
4747         stp->st_stid.sc_free = nfs4_free_lock_stateid;
4748         stp->st_access_bmap = 0;
4749         stp->st_deny_bmap = open_stp->st_deny_bmap;
4750         stp->st_openstp = open_stp;
4751         list_add(&stp->st_locks, &open_stp->st_locks);
4752         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4753         spin_lock(&fp->fi_lock);
4754         list_add(&stp->st_perfile, &fp->fi_stateids);
4755         spin_unlock(&fp->fi_lock);
4756 }
4757
4758 static struct nfs4_ol_stateid *
4759 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
4760 {
4761         struct nfs4_ol_stateid *lst;
4762         struct nfs4_client *clp = lo->lo_owner.so_client;
4763
4764         lockdep_assert_held(&clp->cl_lock);
4765
4766         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
4767                 if (lst->st_stid.sc_file == fp) {
4768                         atomic_inc(&lst->st_stid.sc_count);
4769                         return lst;
4770                 }
4771         }
4772         return NULL;
4773 }
4774
4775 static struct nfs4_ol_stateid *
4776 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
4777                             struct inode *inode, struct nfs4_ol_stateid *ost,
4778                             bool *new)
4779 {
4780         struct nfs4_stid *ns = NULL;
4781         struct nfs4_ol_stateid *lst;
4782         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4783         struct nfs4_client *clp = oo->oo_owner.so_client;
4784
4785         spin_lock(&clp->cl_lock);
4786         lst = find_lock_stateid(lo, fi);
4787         if (lst == NULL) {
4788                 spin_unlock(&clp->cl_lock);
4789                 ns = nfs4_alloc_stid(clp, stateid_slab);
4790                 if (ns == NULL)
4791                         return NULL;
4792
4793                 spin_lock(&clp->cl_lock);
4794                 lst = find_lock_stateid(lo, fi);
4795                 if (likely(!lst)) {
4796                         lst = openlockstateid(ns);
4797                         init_lock_stateid(lst, lo, fi, inode, ost);
4798                         ns = NULL;
4799                         *new = true;
4800                 }
4801         }
4802         spin_unlock(&clp->cl_lock);
4803         if (ns)
4804                 nfs4_put_stid(ns);
4805         return lst;
4806 }
4807
4808 static int
4809 check_lock_length(u64 offset, u64 length)
4810 {
4811         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4812              LOFF_OVERFLOW(offset, length)));
4813 }
4814
4815 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4816 {
4817         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
4818
4819         lockdep_assert_held(&fp->fi_lock);
4820
4821         if (test_access(access, lock_stp))
4822                 return;
4823         __nfs4_file_get_access(fp, access);
4824         set_access(access, lock_stp);
4825 }
4826
4827 static __be32
4828 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
4829                             struct nfs4_ol_stateid *ost,
4830                             struct nfsd4_lock *lock,
4831                             struct nfs4_ol_stateid **lst, bool *new)
4832 {
4833         struct nfs4_file *fi = ost->st_stid.sc_file;
4834         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4835         struct nfs4_client *cl = oo->oo_owner.so_client;
4836         struct inode *inode = cstate->current_fh.fh_dentry->d_inode;
4837         struct nfs4_lockowner *lo;
4838         unsigned int strhashval;
4839         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4840
4841         lo = find_lockowner_str(&cl->cl_clientid, &lock->v.new.owner, nn);
4842         if (!lo) {
4843                 strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4844                                 &lock->v.new.owner);
4845                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4846                 if (lo == NULL)
4847                         return nfserr_jukebox;
4848         } else {
4849                 /* with an existing lockowner, seqids must be the same */
4850                 if (!cstate->minorversion &&
4851                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
4852                         return nfserr_bad_seqid;
4853         }
4854
4855         *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
4856         if (*lst == NULL) {
4857                 release_lockowner_if_empty(lo);
4858                 return nfserr_jukebox;
4859         }
4860         return nfs_ok;
4861 }
4862
4863 /*
4864  *  LOCK operation
4865  */
4866 __be32
4867 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4868            struct nfsd4_lock *lock)
4869 {
4870         struct nfs4_openowner *open_sop = NULL;
4871         struct nfs4_lockowner *lock_sop = NULL;
4872         struct nfs4_ol_stateid *lock_stp = NULL;
4873         struct nfs4_file *fp;
4874         struct file *filp = NULL;
4875         struct file_lock *file_lock = NULL;
4876         struct file_lock *conflock = NULL;
4877         __be32 status = 0;
4878         bool new_state = false;
4879         int lkflg;
4880         int err;
4881         struct net *net = SVC_NET(rqstp);
4882         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4883
4884         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4885                 (long long) lock->lk_offset,
4886                 (long long) lock->lk_length);
4887
4888         if (check_lock_length(lock->lk_offset, lock->lk_length))
4889                  return nfserr_inval;
4890
4891         if ((status = fh_verify(rqstp, &cstate->current_fh,
4892                                 S_IFREG, NFSD_MAY_LOCK))) {
4893                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4894                 return status;
4895         }
4896
4897         nfs4_lock_state();
4898
4899         if (lock->lk_is_new) {
4900                 struct nfs4_ol_stateid *open_stp = NULL;
4901
4902                 if (nfsd4_has_session(cstate))
4903                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4904                         memcpy(&lock->v.new.clientid,
4905                                 &cstate->session->se_client->cl_clientid,
4906                                 sizeof(clientid_t));
4907
4908                 status = nfserr_stale_clientid;
4909                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4910                         goto out;
4911
4912                 /* validate and update open stateid and open seqid */
4913                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4914                                         lock->lk_new_open_seqid,
4915                                         &lock->lk_new_open_stateid,
4916                                         &open_stp, nn);
4917                 if (status)
4918                         goto out;
4919                 open_sop = openowner(open_stp->st_stateowner);
4920                 status = nfserr_bad_stateid;
4921                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4922                                                 &lock->v.new.clientid))
4923                         goto out;
4924                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4925                                                         &lock_stp, &new_state);
4926         } else {
4927                 status = nfs4_preprocess_seqid_op(cstate,
4928                                        lock->lk_old_lock_seqid,
4929                                        &lock->lk_old_lock_stateid,
4930                                        NFS4_LOCK_STID, &lock_stp, nn);
4931                 /* FIXME: move into nfs4_preprocess_seqid_op */
4932                 if (!status)
4933                         atomic_inc(&lock_stp->st_stid.sc_count);
4934         }
4935         if (status)
4936                 goto out;
4937         lock_sop = lockowner(lock_stp->st_stateowner);
4938
4939         lkflg = setlkflg(lock->lk_type);
4940         status = nfs4_check_openmode(lock_stp, lkflg);
4941         if (status)
4942                 goto out;
4943
4944         status = nfserr_grace;
4945         if (locks_in_grace(net) && !lock->lk_reclaim)
4946                 goto out;
4947         status = nfserr_no_grace;
4948         if (!locks_in_grace(net) && lock->lk_reclaim)
4949                 goto out;
4950
4951         file_lock = locks_alloc_lock();
4952         if (!file_lock) {
4953                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4954                 status = nfserr_jukebox;
4955                 goto out;
4956         }
4957
4958         fp = lock_stp->st_stid.sc_file;
4959         locks_init_lock(file_lock);
4960         switch (lock->lk_type) {
4961                 case NFS4_READ_LT:
4962                 case NFS4_READW_LT:
4963                         spin_lock(&fp->fi_lock);
4964                         filp = find_readable_file_locked(fp);
4965                         if (filp)
4966                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4967                         spin_unlock(&fp->fi_lock);
4968                         file_lock->fl_type = F_RDLCK;
4969                         break;
4970                 case NFS4_WRITE_LT:
4971                 case NFS4_WRITEW_LT:
4972                         spin_lock(&fp->fi_lock);
4973                         filp = find_writeable_file_locked(fp);
4974                         if (filp)
4975                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4976                         spin_unlock(&fp->fi_lock);
4977                         file_lock->fl_type = F_WRLCK;
4978                         break;
4979                 default:
4980                         status = nfserr_inval;
4981                 goto out;
4982         }
4983         if (!filp) {
4984                 status = nfserr_openmode;
4985                 goto out;
4986         }
4987         file_lock->fl_owner = (fl_owner_t)lock_sop;
4988         file_lock->fl_pid = current->tgid;
4989         file_lock->fl_file = filp;
4990         file_lock->fl_flags = FL_POSIX;
4991         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4992         file_lock->fl_start = lock->lk_offset;
4993         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4994         nfs4_transform_lock_offset(file_lock);
4995
4996         conflock = locks_alloc_lock();
4997         if (!conflock) {
4998                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4999                 status = nfserr_jukebox;
5000                 goto out;
5001         }
5002
5003         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5004         switch (-err) {
5005         case 0: /* success! */
5006                 update_stateid(&lock_stp->st_stid.sc_stateid);
5007                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
5008                                 sizeof(stateid_t));
5009                 status = 0;
5010                 break;
5011         case (EAGAIN):          /* conflock holds conflicting lock */
5012                 status = nfserr_denied;
5013                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5014                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5015                 break;
5016         case (EDEADLK):
5017                 status = nfserr_deadlock;
5018                 break;
5019         default:
5020                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5021                 status = nfserrno(err);
5022                 break;
5023         }
5024 out:
5025         if (filp)
5026                 fput(filp);
5027         if (lock_stp)
5028                 nfs4_put_stid(&lock_stp->st_stid);
5029         if (status && new_state)
5030                 release_lock_stateid(lock_stp);
5031         nfsd4_bump_seqid(cstate, status);
5032         if (!cstate->replay_owner)
5033                 nfs4_unlock_state();
5034         if (file_lock)
5035                 locks_free_lock(file_lock);
5036         if (conflock)
5037                 locks_free_lock(conflock);
5038         return status;
5039 }
5040
5041 /*
5042  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5043  * so we do a temporary open here just to get an open file to pass to
5044  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5045  * inode operation.)
5046  */
5047 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5048 {
5049         struct file *file;
5050         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5051         if (!err) {
5052                 err = nfserrno(vfs_test_lock(file, lock));
5053                 nfsd_close(file);
5054         }
5055         return err;
5056 }
5057
5058 /*
5059  * LOCKT operation
5060  */
5061 __be32
5062 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5063             struct nfsd4_lockt *lockt)
5064 {
5065         struct file_lock *file_lock = NULL;
5066         struct nfs4_lockowner *lo;
5067         __be32 status;
5068         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5069
5070         if (locks_in_grace(SVC_NET(rqstp)))
5071                 return nfserr_grace;
5072
5073         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5074                  return nfserr_inval;
5075
5076         nfs4_lock_state();
5077
5078         if (!nfsd4_has_session(cstate)) {
5079                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5080                 if (status)
5081                         goto out;
5082         }
5083
5084         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5085                 goto out;
5086
5087         file_lock = locks_alloc_lock();
5088         if (!file_lock) {
5089                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5090                 status = nfserr_jukebox;
5091                 goto out;
5092         }
5093         locks_init_lock(file_lock);
5094         switch (lockt->lt_type) {
5095                 case NFS4_READ_LT:
5096                 case NFS4_READW_LT:
5097                         file_lock->fl_type = F_RDLCK;
5098                 break;
5099                 case NFS4_WRITE_LT:
5100                 case NFS4_WRITEW_LT:
5101                         file_lock->fl_type = F_WRLCK;
5102                 break;
5103                 default:
5104                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5105                         status = nfserr_inval;
5106                 goto out;
5107         }
5108
5109         lo = find_lockowner_str(&lockt->lt_clientid, &lockt->lt_owner, nn);
5110         if (lo)
5111                 file_lock->fl_owner = (fl_owner_t)lo;
5112         file_lock->fl_pid = current->tgid;
5113         file_lock->fl_flags = FL_POSIX;
5114
5115         file_lock->fl_start = lockt->lt_offset;
5116         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5117
5118         nfs4_transform_lock_offset(file_lock);
5119
5120         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5121         if (status)
5122                 goto out;
5123
5124         if (file_lock->fl_type != F_UNLCK) {
5125                 status = nfserr_denied;
5126                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5127         }
5128 out:
5129         nfs4_unlock_state();
5130         if (file_lock)
5131                 locks_free_lock(file_lock);
5132         return status;
5133 }
5134
5135 __be32
5136 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5137             struct nfsd4_locku *locku)
5138 {
5139         struct nfs4_ol_stateid *stp;
5140         struct file *filp = NULL;
5141         struct file_lock *file_lock = NULL;
5142         __be32 status;
5143         int err;
5144         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5145
5146         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5147                 (long long) locku->lu_offset,
5148                 (long long) locku->lu_length);
5149
5150         if (check_lock_length(locku->lu_offset, locku->lu_length))
5151                  return nfserr_inval;
5152
5153         nfs4_lock_state();
5154
5155         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5156                                         &locku->lu_stateid, NFS4_LOCK_STID,
5157                                         &stp, nn);
5158         if (status)
5159                 goto out;
5160         /* FIXME: move into nfs4_preprocess_seqid_op */
5161         atomic_inc(&stp->st_stid.sc_count);
5162         filp = find_any_file(stp->st_stid.sc_file);
5163         if (!filp) {
5164                 status = nfserr_lock_range;
5165                 goto put_stateid;
5166         }
5167         file_lock = locks_alloc_lock();
5168         if (!file_lock) {
5169                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5170                 status = nfserr_jukebox;
5171                 goto fput;
5172         }
5173         locks_init_lock(file_lock);
5174         file_lock->fl_type = F_UNLCK;
5175         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
5176         file_lock->fl_pid = current->tgid;
5177         file_lock->fl_file = filp;
5178         file_lock->fl_flags = FL_POSIX;
5179         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5180         file_lock->fl_start = locku->lu_offset;
5181
5182         file_lock->fl_end = last_byte_offset(locku->lu_offset,
5183                                                 locku->lu_length);
5184         nfs4_transform_lock_offset(file_lock);
5185
5186         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5187         if (err) {
5188                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5189                 goto out_nfserr;
5190         }
5191         update_stateid(&stp->st_stid.sc_stateid);
5192         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
5193 fput:
5194         fput(filp);
5195 put_stateid:
5196         nfs4_put_stid(&stp->st_stid);
5197 out:
5198         nfsd4_bump_seqid(cstate, status);
5199         if (!cstate->replay_owner)
5200                 nfs4_unlock_state();
5201         if (file_lock)
5202                 locks_free_lock(file_lock);
5203         return status;
5204
5205 out_nfserr:
5206         status = nfserrno(err);
5207         goto fput;
5208 }
5209
5210 /*
5211  * returns
5212  *      true:  locks held by lockowner
5213  *      false: no locks held by lockowner
5214  */
5215 static bool
5216 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5217 {
5218         struct file_lock **flpp;
5219         int status = false;
5220         struct file *filp = find_any_file(fp);
5221         struct inode *inode;
5222
5223         if (!filp) {
5224                 /* Any valid lock stateid should have some sort of access */
5225                 WARN_ON_ONCE(1);
5226                 return status;
5227         }
5228
5229         inode = file_inode(filp);
5230
5231         spin_lock(&inode->i_lock);
5232         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
5233                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
5234                         status = true;
5235                         break;
5236                 }
5237         }
5238         spin_unlock(&inode->i_lock);
5239         fput(filp);
5240         return status;
5241 }
5242
5243 __be32
5244 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5245                         struct nfsd4_compound_state *cstate,
5246                         struct nfsd4_release_lockowner *rlockowner)
5247 {
5248         clientid_t *clid = &rlockowner->rl_clientid;
5249         struct nfs4_stateowner *sop = NULL, *tmp;
5250         struct nfs4_lockowner *lo;
5251         struct nfs4_ol_stateid *stp;
5252         struct xdr_netobj *owner = &rlockowner->rl_owner;
5253         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
5254         __be32 status;
5255         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5256
5257         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5258                 clid->cl_boot, clid->cl_id);
5259
5260         nfs4_lock_state();
5261
5262         status = lookup_clientid(clid, cstate, nn);
5263         if (status)
5264                 goto out;
5265
5266         status = nfserr_locks_held;
5267
5268         /* Find the matching lock stateowner */
5269         list_for_each_entry(tmp, &nn->ownerstr_hashtbl[hashval], so_strhash) {
5270                 if (tmp->so_is_open_owner)
5271                         continue;
5272                 if (same_owner_str(tmp, owner, clid)) {
5273                         sop = tmp;
5274                         break;
5275                 }
5276         }
5277
5278         /* No matching owner found, maybe a replay? Just declare victory... */
5279         if (!sop) {
5280                 status = nfs_ok;
5281                 goto out;
5282         }
5283
5284         lo = lockowner(sop);
5285         /* see if there are still any locks associated with it */
5286         list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5287                 if (check_for_locks(stp->st_stid.sc_file, lo))
5288                         goto out;
5289         }
5290
5291         status = nfs_ok;
5292         release_lockowner(lo);
5293 out:
5294         nfs4_unlock_state();
5295         return status;
5296 }
5297
5298 static inline struct nfs4_client_reclaim *
5299 alloc_reclaim(void)
5300 {
5301         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5302 }
5303
5304 bool
5305 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5306 {
5307         struct nfs4_client_reclaim *crp;
5308
5309         crp = nfsd4_find_reclaim_client(name, nn);
5310         return (crp && crp->cr_clp);
5311 }
5312
5313 /*
5314  * failure => all reset bets are off, nfserr_no_grace...
5315  */
5316 struct nfs4_client_reclaim *
5317 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5318 {
5319         unsigned int strhashval;
5320         struct nfs4_client_reclaim *crp;
5321
5322         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5323         crp = alloc_reclaim();
5324         if (crp) {
5325                 strhashval = clientstr_hashval(name);
5326                 INIT_LIST_HEAD(&crp->cr_strhash);
5327                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5328                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5329                 crp->cr_clp = NULL;
5330                 nn->reclaim_str_hashtbl_size++;
5331         }
5332         return crp;
5333 }
5334
5335 void
5336 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5337 {
5338         list_del(&crp->cr_strhash);
5339         kfree(crp);
5340         nn->reclaim_str_hashtbl_size--;
5341 }
5342
5343 void
5344 nfs4_release_reclaim(struct nfsd_net *nn)
5345 {
5346         struct nfs4_client_reclaim *crp = NULL;
5347         int i;
5348
5349         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5350                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5351                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5352                                         struct nfs4_client_reclaim, cr_strhash);
5353                         nfs4_remove_reclaim_record(crp, nn);
5354                 }
5355         }
5356         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5357 }
5358
5359 /*
5360  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5361 struct nfs4_client_reclaim *
5362 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5363 {
5364         unsigned int strhashval;
5365         struct nfs4_client_reclaim *crp = NULL;
5366
5367         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5368
5369         strhashval = clientstr_hashval(recdir);
5370         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5371                 if (same_name(crp->cr_recdir, recdir)) {
5372                         return crp;
5373                 }
5374         }
5375         return NULL;
5376 }
5377
5378 /*
5379 * Called from OPEN. Look for clientid in reclaim list.
5380 */
5381 __be32
5382 nfs4_check_open_reclaim(clientid_t *clid,
5383                 struct nfsd4_compound_state *cstate,
5384                 struct nfsd_net *nn)
5385 {
5386         __be32 status;
5387
5388         /* find clientid in conf_id_hashtbl */
5389         status = lookup_clientid(clid, cstate, nn);
5390         if (status)
5391                 return nfserr_reclaim_bad;
5392
5393         if (nfsd4_client_record_check(cstate->clp))
5394                 return nfserr_reclaim_bad;
5395
5396         return nfs_ok;
5397 }
5398
5399 #ifdef CONFIG_NFSD_FAULT_INJECTION
5400
5401 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
5402 {
5403         if (mark_client_expired(clp))
5404                 return 0;
5405         expire_client(clp);
5406         return 1;
5407 }
5408
5409 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
5410 {
5411         char buf[INET6_ADDRSTRLEN];
5412         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5413         printk(KERN_INFO "NFS Client: %s\n", buf);
5414         return 1;
5415 }
5416
5417 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
5418                              const char *type)
5419 {
5420         char buf[INET6_ADDRSTRLEN];
5421         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5422         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
5423 }
5424
5425 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
5426                                     void (*func)(struct nfs4_ol_stateid *))
5427 {
5428         struct nfs4_openowner *oop;
5429         struct nfs4_ol_stateid *stp, *st_next;
5430         struct nfs4_ol_stateid *lst, *lst_next;
5431         u64 count = 0;
5432
5433         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
5434                 list_for_each_entry_safe(stp, st_next,
5435                                 &oop->oo_owner.so_stateids, st_perstateowner) {
5436                         list_for_each_entry_safe(lst, lst_next,
5437                                         &stp->st_locks, st_locks) {
5438                                 if (func)
5439                                         func(lst);
5440                                 if (++count == max)
5441                                         return count;
5442                         }
5443                 }
5444         }
5445
5446         return count;
5447 }
5448
5449 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
5450 {
5451         return nfsd_foreach_client_lock(clp, max, release_lock_stateid);
5452 }
5453
5454 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
5455 {
5456         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
5457         nfsd_print_count(clp, count, "locked files");
5458         return count;
5459 }
5460
5461 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
5462 {
5463         struct nfs4_openowner *oop, *next;
5464         u64 count = 0;
5465
5466         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
5467                 if (func)
5468                         func(oop);
5469                 if (++count == max)
5470                         break;
5471         }
5472
5473         return count;
5474 }
5475
5476 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
5477 {
5478         return nfsd_foreach_client_open(clp, max, release_openowner);
5479 }
5480
5481 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
5482 {
5483         u64 count = nfsd_foreach_client_open(clp, max, NULL);
5484         nfsd_print_count(clp, count, "open files");
5485         return count;
5486 }
5487
5488 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
5489                                      struct list_head *victims)
5490 {
5491         struct nfs4_delegation *dp, *next;
5492         u64 count = 0;
5493
5494         lockdep_assert_held(&state_lock);
5495         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
5496                 if (victims) {
5497                         /*
5498                          * It's not safe to mess with delegations that have a
5499                          * non-zero dl_time. They might have already been broken
5500                          * and could be processed by the laundromat outside of
5501                          * the state_lock. Just leave them be.
5502                          */
5503                         if (dp->dl_time != 0)
5504                                 continue;
5505
5506                         unhash_delegation_locked(dp);
5507                         list_add(&dp->dl_recall_lru, victims);
5508                 }
5509                 if (++count == max)
5510                         break;
5511         }
5512         return count;
5513 }
5514
5515 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
5516 {
5517         struct nfs4_delegation *dp, *next;
5518         LIST_HEAD(victims);
5519         u64 count;
5520
5521         spin_lock(&state_lock);
5522         count = nfsd_find_all_delegations(clp, max, &victims);
5523         spin_unlock(&state_lock);
5524
5525         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru) {
5526                 list_del_init(&dp->dl_recall_lru);
5527                 revoke_delegation(dp);
5528         }
5529
5530         return count;
5531 }
5532
5533 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
5534 {
5535         struct nfs4_delegation *dp;
5536         LIST_HEAD(victims);
5537         u64 count;
5538
5539         spin_lock(&state_lock);
5540         count = nfsd_find_all_delegations(clp, max, &victims);
5541         while (!list_empty(&victims)) {
5542                 dp = list_first_entry(&victims, struct nfs4_delegation,
5543                                         dl_recall_lru);
5544                 list_del_init(&dp->dl_recall_lru);
5545                 dp->dl_time = 0;
5546                 nfsd_break_one_deleg(dp);
5547         }
5548         spin_unlock(&state_lock);
5549
5550         return count;
5551 }
5552
5553 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
5554 {
5555         u64 count = 0;
5556
5557         spin_lock(&state_lock);
5558         count = nfsd_find_all_delegations(clp, max, NULL);
5559         spin_unlock(&state_lock);
5560
5561         nfsd_print_count(clp, count, "delegations");
5562         return count;
5563 }
5564
5565 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
5566 {
5567         struct nfs4_client *clp, *next;
5568         u64 count = 0;
5569         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
5570
5571         if (!nfsd_netns_ready(nn))
5572                 return 0;
5573
5574         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
5575                 count += func(clp, max - count);
5576                 if ((max != 0) && (count >= max))
5577                         break;
5578         }
5579
5580         return count;
5581 }
5582
5583 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
5584 {
5585         struct nfs4_client *clp;
5586         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
5587
5588         if (!nfsd_netns_ready(nn))
5589                 return NULL;
5590
5591         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5592                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
5593                         return clp;
5594         }
5595         return NULL;
5596 }
5597
5598 #endif /* CONFIG_NFSD_FAULT_INJECTION */
5599
5600 /*
5601  * Since the lifetime of a delegation isn't limited to that of an open, a
5602  * client may quite reasonably hang on to a delegation as long as it has
5603  * the inode cached.  This becomes an obvious problem the first time a
5604  * client's inode cache approaches the size of the server's total memory.
5605  *
5606  * For now we avoid this problem by imposing a hard limit on the number
5607  * of delegations, which varies according to the server's memory size.
5608  */
5609 static void
5610 set_max_delegations(void)
5611 {
5612         /*
5613          * Allow at most 4 delegations per megabyte of RAM.  Quick
5614          * estimates suggest that in the worst case (where every delegation
5615          * is for a different inode), a delegation could take about 1.5K,
5616          * giving a worst case usage of about 6% of memory.
5617          */
5618         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
5619 }
5620
5621 static int nfs4_state_create_net(struct net *net)
5622 {
5623         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5624         int i;
5625
5626         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
5627                         CLIENT_HASH_SIZE, GFP_KERNEL);
5628         if (!nn->conf_id_hashtbl)
5629                 goto err;
5630         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
5631                         CLIENT_HASH_SIZE, GFP_KERNEL);
5632         if (!nn->unconf_id_hashtbl)
5633                 goto err_unconf_id;
5634         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
5635                         OWNER_HASH_SIZE, GFP_KERNEL);
5636         if (!nn->ownerstr_hashtbl)
5637                 goto err_ownerstr;
5638         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
5639                         SESSION_HASH_SIZE, GFP_KERNEL);
5640         if (!nn->sessionid_hashtbl)
5641                 goto err_sessionid;
5642
5643         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5644                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
5645                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
5646         }
5647         for (i = 0; i < OWNER_HASH_SIZE; i++)
5648                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
5649         for (i = 0; i < SESSION_HASH_SIZE; i++)
5650                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
5651         nn->conf_name_tree = RB_ROOT;
5652         nn->unconf_name_tree = RB_ROOT;
5653         INIT_LIST_HEAD(&nn->client_lru);
5654         INIT_LIST_HEAD(&nn->close_lru);
5655         INIT_LIST_HEAD(&nn->del_recall_lru);
5656         spin_lock_init(&nn->client_lock);
5657
5658         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
5659         get_net(net);
5660
5661         return 0;
5662
5663 err_sessionid:
5664         kfree(nn->ownerstr_hashtbl);
5665 err_ownerstr:
5666         kfree(nn->unconf_id_hashtbl);
5667 err_unconf_id:
5668         kfree(nn->conf_id_hashtbl);
5669 err:
5670         return -ENOMEM;
5671 }
5672
5673 static void
5674 nfs4_state_destroy_net(struct net *net)
5675 {
5676         int i;
5677         struct nfs4_client *clp = NULL;
5678         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5679
5680         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5681                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
5682                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
5683                         destroy_client(clp);
5684                 }
5685         }
5686
5687         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5688                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
5689                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
5690                         destroy_client(clp);
5691                 }
5692         }
5693
5694         kfree(nn->sessionid_hashtbl);
5695         kfree(nn->ownerstr_hashtbl);
5696         kfree(nn->unconf_id_hashtbl);
5697         kfree(nn->conf_id_hashtbl);
5698         put_net(net);
5699 }
5700
5701 int
5702 nfs4_state_start_net(struct net *net)
5703 {
5704         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5705         int ret;
5706
5707         ret = nfs4_state_create_net(net);
5708         if (ret)
5709                 return ret;
5710         nfsd4_client_tracking_init(net);
5711         nn->boot_time = get_seconds();
5712         locks_start_grace(net, &nn->nfsd4_manager);
5713         nn->grace_ended = false;
5714         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
5715                nn->nfsd4_grace, net);
5716         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
5717         return 0;
5718 }
5719
5720 /* initialization to perform when the nfsd service is started: */
5721
5722 int
5723 nfs4_state_start(void)
5724 {
5725         int ret;
5726
5727         ret = set_callback_cred();
5728         if (ret)
5729                 return -ENOMEM;
5730         laundry_wq = create_singlethread_workqueue("nfsd4");
5731         if (laundry_wq == NULL) {
5732                 ret = -ENOMEM;
5733                 goto out_recovery;
5734         }
5735         ret = nfsd4_create_callback_queue();
5736         if (ret)
5737                 goto out_free_laundry;
5738
5739         set_max_delegations();
5740
5741         return 0;
5742
5743 out_free_laundry:
5744         destroy_workqueue(laundry_wq);
5745 out_recovery:
5746         return ret;
5747 }
5748
5749 void
5750 nfs4_state_shutdown_net(struct net *net)
5751 {
5752         struct nfs4_delegation *dp = NULL;
5753         struct list_head *pos, *next, reaplist;
5754         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5755
5756         cancel_delayed_work_sync(&nn->laundromat_work);
5757         locks_end_grace(&nn->nfsd4_manager);
5758
5759         nfs4_lock_state();
5760         INIT_LIST_HEAD(&reaplist);
5761         spin_lock(&state_lock);
5762         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5763                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5764                 unhash_delegation_locked(dp);
5765                 list_add(&dp->dl_recall_lru, &reaplist);
5766         }
5767         spin_unlock(&state_lock);
5768         list_for_each_safe(pos, next, &reaplist) {
5769                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5770                 list_del_init(&dp->dl_recall_lru);
5771                 nfs4_put_stid(&dp->dl_stid);
5772         }
5773
5774         nfsd4_client_tracking_exit(net);
5775         nfs4_state_destroy_net(net);
5776         nfs4_unlock_state();
5777 }
5778
5779 void
5780 nfs4_state_shutdown(void)
5781 {
5782         destroy_workqueue(laundry_wq);
5783         nfsd4_destroy_callback_queue();
5784 }
5785
5786 static void
5787 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5788 {
5789         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5790                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5791 }
5792
5793 static void
5794 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5795 {
5796         if (cstate->minorversion) {
5797                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5798                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5799         }
5800 }
5801
5802 void
5803 clear_current_stateid(struct nfsd4_compound_state *cstate)
5804 {
5805         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5806 }
5807
5808 /*
5809  * functions to set current state id
5810  */
5811 void
5812 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5813 {
5814         put_stateid(cstate, &odp->od_stateid);
5815 }
5816
5817 void
5818 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5819 {
5820         put_stateid(cstate, &open->op_stateid);
5821 }
5822
5823 void
5824 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5825 {
5826         put_stateid(cstate, &close->cl_stateid);
5827 }
5828
5829 void
5830 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5831 {
5832         put_stateid(cstate, &lock->lk_resp_stateid);
5833 }
5834
5835 /*
5836  * functions to consume current state id
5837  */
5838
5839 void
5840 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5841 {
5842         get_stateid(cstate, &odp->od_stateid);
5843 }
5844
5845 void
5846 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5847 {
5848         get_stateid(cstate, &drp->dr_stateid);
5849 }
5850
5851 void
5852 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5853 {
5854         get_stateid(cstate, &fsp->fr_stateid);
5855 }
5856
5857 void
5858 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5859 {
5860         get_stateid(cstate, &setattr->sa_stateid);
5861 }
5862
5863 void
5864 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5865 {
5866         get_stateid(cstate, &close->cl_stateid);
5867 }
5868
5869 void
5870 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5871 {
5872         get_stateid(cstate, &locku->lu_stateid);
5873 }
5874
5875 void
5876 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5877 {
5878         get_stateid(cstate, &read->rd_stateid);
5879 }
5880
5881 void
5882 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5883 {
5884         get_stateid(cstate, &write->wr_stateid);
5885 }