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 /*
  79  * Currently used for the del_recall_lru and file hash table.  In an
  80  * effort to decrease the scope of the client_mutex, this spinlock may
  81  * eventually cover more:
  82  */
  83 static DEFINE_SPINLOCK(state_lock);
  84
  85 /*
  86  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  87  * the refcount on the open stateid to drop.
  88  */
  89 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  90
  91 static struct kmem_cache *openowner_slab;
  92 static struct kmem_cache *lockowner_slab;
  93 static struct kmem_cache *file_slab;
  94 static struct kmem_cache *stateid_slab;
  95 static struct kmem_cache *deleg_slab;
  96
  97 static void free_session(struct nfsd4_session *);
  98
  99 static bool is_session_dead(struct nfsd4_session *ses)
 100 {
 101         return ses->se_flags & NFS4_SESSION_DEAD;
 102 }
 103
 104 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 105 {
 106         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 107                 return nfserr_jukebox;
 108         ses->se_flags |= NFS4_SESSION_DEAD;
 109         return nfs_ok;
 110 }
 111
 112 static bool is_client_expired(struct nfs4_client *clp)
 113 {
 114         return clp->cl_time == 0;
 115 }
 116
 117 static __be32 get_client_locked(struct nfs4_client *clp)
 118 {
 119         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 120
 121         lockdep_assert_held(&nn->client_lock);
 122
 123         if (is_client_expired(clp))
 124                 return nfserr_expired;
 125         atomic_inc(&clp->cl_refcount);
 126         return nfs_ok;
 127 }
 128
 129 /* must be called under the client_lock */
 130 static inline void
 131 renew_client_locked(struct nfs4_client *clp)
 132 {
 133         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 134
 135         if (is_client_expired(clp)) {
 136                 WARN_ON(1);
 137                 printk("%s: client (clientid %08x/%08x) already expired\n",
 138                         __func__,
 139                         clp->cl_clientid.cl_boot,
 140                         clp->cl_clientid.cl_id);
 141                 return;
 142         }
 143
 144         dprintk("renewing client (clientid %08x/%08x)\n",
 145                         clp->cl_clientid.cl_boot,
 146                         clp->cl_clientid.cl_id);
 147         list_move_tail(&clp->cl_lru, &nn->client_lru);
 148         clp->cl_time = get_seconds();
 149 }
 150
 151 static inline void
 152 renew_client(struct nfs4_client *clp)
 153 {
 154         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 155
 156         spin_lock(&nn->client_lock);
 157         renew_client_locked(clp);
 158         spin_unlock(&nn->client_lock);
 159 }
 160
 161 static void put_client_renew_locked(struct nfs4_client *clp)
 162 {
 163         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 164
 165         lockdep_assert_held(&nn->client_lock);
 166
 167         if (!atomic_dec_and_test(&clp->cl_refcount))
 168                 return;
 169         if (!is_client_expired(clp))
 170                 renew_client_locked(clp);
 171 }
 172
 173 static void put_client_renew(struct nfs4_client *clp)
 174 {
 175         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 176
 177         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 178                 return;
 179         if (!is_client_expired(clp))
 180                 renew_client_locked(clp);
 181         spin_unlock(&nn->client_lock);
 182 }
 183
 184 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 185 {
 186         __be32 status;
 187
 188         if (is_session_dead(ses))
 189                 return nfserr_badsession;
 190         status = get_client_locked(ses->se_client);
 191         if (status)
 192                 return status;
 193         atomic_inc(&ses->se_ref);
 194         return nfs_ok;
 195 }
 196
 197 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 198 {
 199         struct nfs4_client *clp = ses->se_client;
 200         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 201
 202         lockdep_assert_held(&nn->client_lock);
 203
 204         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 205                 free_session(ses);
 206         put_client_renew_locked(clp);
 207 }
 208
 209 static void nfsd4_put_session(struct nfsd4_session *ses)
 210 {
 211         struct nfs4_client *clp = ses->se_client;
 212         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 213
 214         spin_lock(&nn->client_lock);
 215         nfsd4_put_session_locked(ses);
 216         spin_unlock(&nn->client_lock);
 217 }
 218
 219 static inline struct nfs4_stateowner *
 220 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 221 {
 222         atomic_inc(&sop->so_count);
 223         return sop;
 224 }
 225
 226 static int
 227 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 228 {
 229         return (sop->so_owner.len == owner->len) &&
 230                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 231 }
 232
 233 static struct nfs4_openowner *
 234 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 235                         struct nfs4_client *clp)
 236 {
 237         struct nfs4_stateowner *so;
 238
 239         lockdep_assert_held(&clp->cl_lock);
 240
 241         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 242                             so_strhash) {
 243                 if (!so->so_is_open_owner)
 244                         continue;
 245                 if (same_owner_str(so, &open->op_owner))
 246                         return openowner(nfs4_get_stateowner(so));
 247         }
 248         return NULL;
 249 }
 250
 251 static struct nfs4_openowner *
 252 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 253                         struct nfs4_client *clp)
 254 {
 255         struct nfs4_openowner *oo;
 256
 257         spin_lock(&clp->cl_lock);
 258         oo = find_openstateowner_str_locked(hashval, open, clp);
 259         spin_unlock(&clp->cl_lock);
 260         return oo;
 261 }
 262
 263 static inline u32
 264 opaque_hashval(const void *ptr, int nbytes)
 265 {
 266         unsigned char *cptr = (unsigned char *) ptr;
 267
 268         u32 x = 0;
 269         while (nbytes--) {
 270                 x *= 37;
 271                 x += *cptr++;
 272         }
 273         return x;
 274 }
 275
 276 static void nfsd4_free_file(struct nfs4_file *f)
 277 {
 278         kmem_cache_free(file_slab, f);
 279 }
 280
 281 static inline void
 282 put_nfs4_file(struct nfs4_file *fi)
 283 {
 284         might_lock(&state_lock);
 285
 286         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 287                 hlist_del(&fi->fi_hash);
 288                 spin_unlock(&state_lock);
 289                 nfsd4_free_file(fi);
 290         }
 291 }
 292
 293 static inline void
 294 get_nfs4_file(struct nfs4_file *fi)
 295 {
 296         atomic_inc(&fi->fi_ref);
 297 }
 298
 299 static struct file *
 300 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 301 {
 302         if (f->fi_fds[oflag])
 303                 return get_file(f->fi_fds[oflag]);
 304         return NULL;
 305 }
 306
 307 static struct file *
 308 find_writeable_file_locked(struct nfs4_file *f)
 309 {
 310         struct file *ret;
 311
 312         lockdep_assert_held(&f->fi_lock);
 313
 314         ret = __nfs4_get_fd(f, O_WRONLY);
 315         if (!ret)
 316                 ret = __nfs4_get_fd(f, O_RDWR);
 317         return ret;
 318 }
 319
 320 static struct file *
 321 find_writeable_file(struct nfs4_file *f)
 322 {
 323         struct file *ret;
 324
 325         spin_lock(&f->fi_lock);
 326         ret = find_writeable_file_locked(f);
 327         spin_unlock(&f->fi_lock);
 328
 329         return ret;
 330 }
 331
 332 static struct file *find_readable_file_locked(struct nfs4_file *f)
 333 {
 334         struct file *ret;
 335
 336         lockdep_assert_held(&f->fi_lock);
 337
 338         ret = __nfs4_get_fd(f, O_RDONLY);
 339         if (!ret)
 340                 ret = __nfs4_get_fd(f, O_RDWR);
 341         return ret;
 342 }
 343
 344 static struct file *
 345 find_readable_file(struct nfs4_file *f)
 346 {
 347         struct file *ret;
 348
 349         spin_lock(&f->fi_lock);
 350         ret = find_readable_file_locked(f);
 351         spin_unlock(&f->fi_lock);
 352
 353         return ret;
 354 }
 355
 356 static struct file *
 357 find_any_file(struct nfs4_file *f)
 358 {
 359         struct file *ret;
 360
 361         spin_lock(&f->fi_lock);
 362         ret = __nfs4_get_fd(f, O_RDWR);
 363         if (!ret) {
 364                 ret = __nfs4_get_fd(f, O_WRONLY);
 365                 if (!ret)
 366                         ret = __nfs4_get_fd(f, O_RDONLY);
 367         }
 368         spin_unlock(&f->fi_lock);
 369         return ret;
 370 }
 371
 372 static atomic_long_t num_delegations;
 373 unsigned long max_delegations;
 374
 375 /*
 376  * Open owner state (share locks)
 377  */
 378
 379 /* hash tables for lock and open owners */
 380 #define OWNER_HASH_BITS              8
 381 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 382 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 383
 384 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 385 {
 386         unsigned int ret;
 387
 388         ret = opaque_hashval(ownername->data, ownername->len);
 389         return ret & OWNER_HASH_MASK;
 390 }
 391
 392 /* hash table for nfs4_file */
 393 #define FILE_HASH_BITS                   8
 394 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 395
 396 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 397 {
 398         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 399 }
 400
 401 static unsigned int file_hashval(struct knfsd_fh *fh)
 402 {
 403         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 404 }
 405
 406 static bool nfsd_fh_match(struct knfsd_fh *fh1, struct knfsd_fh *fh2)
 407 {
 408         return fh1->fh_size == fh2->fh_size &&
 409                 !memcmp(fh1->fh_base.fh_pad,
 410                                 fh2->fh_base.fh_pad,
 411                                 fh1->fh_size);
 412 }
 413
 414 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 415
 416 static void
 417 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 418 {
 419         lockdep_assert_held(&fp->fi_lock);
 420
 421         if (access & NFS4_SHARE_ACCESS_WRITE)
 422                 atomic_inc(&fp->fi_access[O_WRONLY]);
 423         if (access & NFS4_SHARE_ACCESS_READ)
 424                 atomic_inc(&fp->fi_access[O_RDONLY]);
 425 }
 426
 427 static __be32
 428 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 429 {
 430         lockdep_assert_held(&fp->fi_lock);
 431
 432         /* Does this access mode make sense? */
 433         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 434                 return nfserr_inval;
 435
 436         /* Does it conflict with a deny mode already set? */
 437         if ((access & fp->fi_share_deny) != 0)
 438                 return nfserr_share_denied;
 439
 440         __nfs4_file_get_access(fp, access);
 441         return nfs_ok;
 442 }
 443
 444 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 445 {
 446         /* Common case is that there is no deny mode. */
 447         if (deny) {
 448                 /* Does this deny mode make sense? */
 449                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 450                         return nfserr_inval;
 451
 452                 if ((deny & NFS4_SHARE_DENY_READ) &&
 453                     atomic_read(&fp->fi_access[O_RDONLY]))
 454                         return nfserr_share_denied;
 455
 456                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 457                     atomic_read(&fp->fi_access[O_WRONLY]))
 458                         return nfserr_share_denied;
 459         }
 460         return nfs_ok;
 461 }
 462
 463 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 464 {
 465         might_lock(&fp->fi_lock);
 466
 467         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 468                 struct file *f1 = NULL;
 469                 struct file *f2 = NULL;
 470
 471                 swap(f1, fp->fi_fds[oflag]);
 472                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 473                         swap(f2, fp->fi_fds[O_RDWR]);
 474                 spin_unlock(&fp->fi_lock);
 475                 if (f1)
 476                         fput(f1);
 477                 if (f2)
 478                         fput(f2);
 479         }
 480 }
 481
 482 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 483 {
 484         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 485
 486         if (access & NFS4_SHARE_ACCESS_WRITE)
 487                 __nfs4_file_put_access(fp, O_WRONLY);
 488         if (access & NFS4_SHARE_ACCESS_READ)
 489                 __nfs4_file_put_access(fp, O_RDONLY);
 490 }
 491
 492 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
 493                                          struct kmem_cache *slab)
 494 {
 495         struct nfs4_stid *stid;
 496         int new_id;
 497
 498         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 499         if (!stid)
 500                 return NULL;
 501
 502         idr_preload(GFP_KERNEL);
 503         spin_lock(&cl->cl_lock);
 504         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 505         spin_unlock(&cl->cl_lock);
 506         idr_preload_end();
 507         if (new_id < 0)
 508                 goto out_free;
 509         stid->sc_client = cl;
 510         stid->sc_stateid.si_opaque.so_id = new_id;
 511         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 512         /* Will be incremented before return to client: */
 513         atomic_set(&stid->sc_count, 1);
 514
 515         /*
 516          * It shouldn't be a problem to reuse an opaque stateid value.
 517          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 518          * example, a stray write retransmission could be accepted by
 519          * the server when it should have been rejected.  Therefore,
 520          * adopt a trick from the sctp code to attempt to maximize the
 521          * amount of time until an id is reused, by ensuring they always
 522          * "increase" (mod INT_MAX):
 523          */
 524         return stid;
 525 out_free:
 526         kmem_cache_free(slab, stid);
 527         return NULL;
 528 }
 529
 530 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 531 {
 532         struct nfs4_stid *stid;
 533         struct nfs4_ol_stateid *stp;
 534
 535         stid = nfs4_alloc_stid(clp, stateid_slab);
 536         if (!stid)
 537                 return NULL;
 538
 539         stp = openlockstateid(stid);
 540         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 541         return stp;
 542 }
 543
 544 static void nfs4_free_deleg(struct nfs4_stid *stid)
 545 {
 546         kmem_cache_free(deleg_slab, stid);
 547         atomic_long_dec(&num_delegations);
 548 }
 549
 550 /*
 551  * When we recall a delegation, we should be careful not to hand it
 552  * out again straight away.
 553  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 554  * in which the filehandles of recalled delegations are "stored".
 555  * If a filehandle appear in either filter, a delegation is blocked.
 556  * When a delegation is recalled, the filehandle is stored in the "new"
 557  * filter.
 558  * Every 30 seconds we swap the filters and clear the "new" one,
 559  * unless both are empty of course.
 560  *
 561  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 562  * low 3 bytes as hash-table indices.
 563  *
 564  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 565  * is used to manage concurrent access.  Testing does not need the lock
 566  * except when swapping the two filters.
 567  */
 568 static DEFINE_SPINLOCK(blocked_delegations_lock);
 569 static struct bloom_pair {
 570         int     entries, old_entries;
 571         time_t  swap_time;
 572         int     new; /* index into 'set' */
 573         DECLARE_BITMAP(set[2], 256);
 574 } blocked_delegations;
 575
 576 static int delegation_blocked(struct knfsd_fh *fh)
 577 {
 578         u32 hash;
 579         struct bloom_pair *bd = &blocked_delegations;
 580
 581         if (bd->entries == 0)
 582                 return 0;
 583         if (seconds_since_boot() - bd->swap_time > 30) {
 584                 spin_lock(&blocked_delegations_lock);
 585                 if (seconds_since_boot() - bd->swap_time > 30) {
 586                         bd->entries -= bd->old_entries;
 587                         bd->old_entries = bd->entries;
 588                         memset(bd->set[bd->new], 0,
 589                                sizeof(bd->set[0]));
 590                         bd->new = 1-bd->new;
 591                         bd->swap_time = seconds_since_boot();
 592                 }
 593                 spin_unlock(&blocked_delegations_lock);
 594         }
 595         hash = arch_fast_hash(&fh->fh_base, fh->fh_size, 0);
 596         if (test_bit(hash&255, bd->set[0]) &&
 597             test_bit((hash>>8)&255, bd->set[0]) &&
 598             test_bit((hash>>16)&255, bd->set[0]))
 599                 return 1;
 600
 601         if (test_bit(hash&255, bd->set[1]) &&
 602             test_bit((hash>>8)&255, bd->set[1]) &&
 603             test_bit((hash>>16)&255, bd->set[1]))
 604                 return 1;
 605
 606         return 0;
 607 }
 608
 609 static void block_delegations(struct knfsd_fh *fh)
 610 {
 611         u32 hash;
 612         struct bloom_pair *bd = &blocked_delegations;
 613
 614         hash = arch_fast_hash(&fh->fh_base, fh->fh_size, 0);
 615
 616         spin_lock(&blocked_delegations_lock);
 617         __set_bit(hash&255, bd->set[bd->new]);
 618         __set_bit((hash>>8)&255, bd->set[bd->new]);
 619         __set_bit((hash>>16)&255, bd->set[bd->new]);
 620         if (bd->entries == 0)
 621                 bd->swap_time = seconds_since_boot();
 622         bd->entries += 1;
 623         spin_unlock(&blocked_delegations_lock);
 624 }
 625
 626 static struct nfs4_delegation *
 627 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh)
 628 {
 629         struct nfs4_delegation *dp;
 630         long n;
 631
 632         dprintk("NFSD alloc_init_deleg\n");
 633         n = atomic_long_inc_return(&num_delegations);
 634         if (n < 0 || n > max_delegations)
 635                 goto out_dec;
 636         if (delegation_blocked(&current_fh->fh_handle))
 637                 goto out_dec;
 638         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 639         if (dp == NULL)
 640                 goto out_dec;
 641
 642         dp->dl_stid.sc_free = nfs4_free_deleg;
 643         /*
 644          * delegation seqid's are never incremented.  The 4.1 special
 645          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 646          * 0 anyway just for consistency and use 1:
 647          */
 648         dp->dl_stid.sc_stateid.si_generation = 1;
 649         INIT_LIST_HEAD(&dp->dl_perfile);
 650         INIT_LIST_HEAD(&dp->dl_perclnt);
 651         INIT_LIST_HEAD(&dp->dl_recall_lru);
 652         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 653         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_run_cb_recall);
 654         return dp;
 655 out_dec:
 656         atomic_long_dec(&num_delegations);
 657         return NULL;
 658 }
 659
 660 void
 661 nfs4_put_stid(struct nfs4_stid *s)
 662 {
 663         struct nfs4_file *fp = s->sc_file;
 664         struct nfs4_client *clp = s->sc_client;
 665
 666         might_lock(&clp->cl_lock);
 667
 668         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 669                 wake_up_all(&close_wq);
 670                 return;
 671         }
 672         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 673         spin_unlock(&clp->cl_lock);
 674         s->sc_free(s);
 675         if (fp)
 676                 put_nfs4_file(fp);
 677 }
 678
 679 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 680 {
 681         struct file *filp = NULL;
 682         struct file_lock *fl;
 683
 684         spin_lock(&fp->fi_lock);
 685         if (fp->fi_lease && atomic_dec_and_test(&fp->fi_delegees)) {
 686                 swap(filp, fp->fi_deleg_file);
 687                 fl = fp->fi_lease;
 688                 fp->fi_lease = NULL;
 689         }
 690         spin_unlock(&fp->fi_lock);
 691
 692         if (filp) {
 693                 vfs_setlease(filp, F_UNLCK, &fl);
 694                 fput(filp);
 695         }
 696 }
 697
 698 static void unhash_stid(struct nfs4_stid *s)
 699 {
 700         s->sc_type = 0;
 701 }
 702
 703 static void
 704 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 705 {
 706         lockdep_assert_held(&state_lock);
 707         lockdep_assert_held(&fp->fi_lock);
 708
 709         atomic_inc(&dp->dl_stid.sc_count);
 710         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 711         list_add(&dp->dl_perfile, &fp->fi_delegations);
 712         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
 713 }
 714
 715 static void
 716 unhash_delegation_locked(struct nfs4_delegation *dp)
 717 {
 718         struct nfs4_file *fp = dp->dl_stid.sc_file;
 719
 720         lockdep_assert_held(&state_lock);
 721
 722         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 723         /* Ensure that deleg break won't try to requeue it */
 724         ++dp->dl_time;
 725         spin_lock(&fp->fi_lock);
 726         list_del_init(&dp->dl_perclnt);
 727         list_del_init(&dp->dl_recall_lru);
 728         list_del_init(&dp->dl_perfile);
 729         spin_unlock(&fp->fi_lock);
 730 }
 731
 732 static void destroy_delegation(struct nfs4_delegation *dp)
 733 {
 734         spin_lock(&state_lock);
 735         unhash_delegation_locked(dp);
 736         spin_unlock(&state_lock);
 737         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 738         nfs4_put_stid(&dp->dl_stid);
 739 }
 740
 741 static void revoke_delegation(struct nfs4_delegation *dp)
 742 {
 743         struct nfs4_client *clp = dp->dl_stid.sc_client;
 744
 745         WARN_ON(!list_empty(&dp->dl_recall_lru));
 746
 747         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 748
 749         if (clp->cl_minorversion == 0)
 750                 nfs4_put_stid(&dp->dl_stid);
 751         else {
 752                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 753                 spin_lock(&clp->cl_lock);
 754                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 755                 spin_unlock(&clp->cl_lock);
 756         }
 757 }
 758
 759 /*
 760  * SETCLIENTID state
 761  */
 762
 763 static unsigned int clientid_hashval(u32 id)
 764 {
 765         return id & CLIENT_HASH_MASK;
 766 }
 767
 768 static unsigned int clientstr_hashval(const char *name)
 769 {
 770         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 771 }
 772
 773 /*
 774  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 775  * st_{access,deny}_bmap field of the stateid, in order to track not
 776  * only what share bits are currently in force, but also what
 777  * combinations of share bits previous opens have used.  This allows us
 778  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 779  * return an error if the client attempt to downgrade to a combination
 780  * of share bits not explicable by closing some of its previous opens.
 781  *
 782  * XXX: This enforcement is actually incomplete, since we don't keep
 783  * track of access/deny bit combinations; so, e.g., we allow:
 784  *
 785  *      OPEN allow read, deny write
 786  *      OPEN allow both, deny none
 787  *      DOWNGRADE allow read, deny none
 788  *
 789  * which we should reject.
 790  */
 791 static unsigned int
 792 bmap_to_share_mode(unsigned long bmap) {
 793         int i;
 794         unsigned int access = 0;
 795
 796         for (i = 1; i < 4; i++) {
 797                 if (test_bit(i, &bmap))
 798                         access |= i;
 799         }
 800         return access;
 801 }
 802
 803 /* set share access for a given stateid */
 804 static inline void
 805 set_access(u32 access, struct nfs4_ol_stateid *stp)
 806 {
 807         unsigned char mask = 1 << access;
 808
 809         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 810         stp->st_access_bmap |= mask;
 811 }
 812
 813 /* clear share access for a given stateid */
 814 static inline void
 815 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 816 {
 817         unsigned char mask = 1 << access;
 818
 819         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
 820         stp->st_access_bmap &= ~mask;
 821 }
 822
 823 /* test whether a given stateid has access */
 824 static inline bool
 825 test_access(u32 access, struct nfs4_ol_stateid *stp)
 826 {
 827         unsigned char mask = 1 << access;
 828
 829         return (bool)(stp->st_access_bmap & mask);
 830 }
 831
 832 /* set share deny for a given stateid */
 833 static inline void
 834 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
 835 {
 836         unsigned char mask = 1 << deny;
 837
 838         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 839         stp->st_deny_bmap |= mask;
 840 }
 841
 842 /* clear share deny for a given stateid */
 843 static inline void
 844 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
 845 {
 846         unsigned char mask = 1 << deny;
 847
 848         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
 849         stp->st_deny_bmap &= ~mask;
 850 }
 851
 852 /* test whether a given stateid is denying specific access */
 853 static inline bool
 854 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
 855 {
 856         unsigned char mask = 1 << deny;
 857
 858         return (bool)(stp->st_deny_bmap & mask);
 859 }
 860
 861 static int nfs4_access_to_omode(u32 access)
 862 {
 863         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 864         case NFS4_SHARE_ACCESS_READ:
 865                 return O_RDONLY;
 866         case NFS4_SHARE_ACCESS_WRITE:
 867                 return O_WRONLY;
 868         case NFS4_SHARE_ACCESS_BOTH:
 869                 return O_RDWR;
 870         }
 871         WARN_ON_ONCE(1);
 872         return O_RDONLY;
 873 }
 874
 875 /*
 876  * A stateid that had a deny mode associated with it is being released
 877  * or downgraded. Recalculate the deny mode on the file.
 878  */
 879 static void
 880 recalculate_deny_mode(struct nfs4_file *fp)
 881 {
 882         struct nfs4_ol_stateid *stp;
 883
 884         spin_lock(&fp->fi_lock);
 885         fp->fi_share_deny = 0;
 886         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
 887                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
 888         spin_unlock(&fp->fi_lock);
 889 }
 890
 891 static void
 892 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
 893 {
 894         int i;
 895         bool change = false;
 896
 897         for (i = 1; i < 4; i++) {
 898                 if ((i & deny) != i) {
 899                         change = true;
 900                         clear_deny(i, stp);
 901                 }
 902         }
 903
 904         /* Recalculate per-file deny mode if there was a change */
 905         if (change)
 906                 recalculate_deny_mode(stp->st_stid.sc_file);
 907 }
 908
 909 /* release all access and file references for a given stateid */
 910 static void
 911 release_all_access(struct nfs4_ol_stateid *stp)
 912 {
 913         int i;
 914         struct nfs4_file *fp = stp->st_stid.sc_file;
 915
 916         if (fp && stp->st_deny_bmap != 0)
 917                 recalculate_deny_mode(fp);
 918
 919         for (i = 1; i < 4; i++) {
 920                 if (test_access(i, stp))
 921                         nfs4_file_put_access(stp->st_stid.sc_file, i);
 922                 clear_access(i, stp);
 923         }
 924 }
 925
 926 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
 927 {
 928         struct nfs4_client *clp = sop->so_client;
 929
 930         might_lock(&clp->cl_lock);
 931
 932         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
 933                 return;
 934         sop->so_ops->so_unhash(sop);
 935         spin_unlock(&clp->cl_lock);
 936         kfree(sop->so_owner.data);
 937         sop->so_ops->so_free(sop);
 938 }
 939
 940 static void unhash_ol_stateid(struct nfs4_ol_stateid *stp)
 941 {
 942         struct nfs4_file *fp = stp->st_stid.sc_file;
 943
 944         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
 945
 946         spin_lock(&fp->fi_lock);
 947         list_del(&stp->st_perfile);
 948         spin_unlock(&fp->fi_lock);
 949         list_del(&stp->st_perstateowner);
 950 }
 951
 952 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
 953 {
 954         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 955
 956         release_all_access(stp);
 957         if (stp->st_stateowner)
 958                 nfs4_put_stateowner(stp->st_stateowner);
 959         kmem_cache_free(stateid_slab, stid);
 960 }
 961
 962 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
 963 {
 964         struct nfs4_ol_stateid *stp = openlockstateid(stid);
 965         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
 966         struct file *file;
 967
 968         file = find_any_file(stp->st_stid.sc_file);
 969         if (file)
 970                 filp_close(file, (fl_owner_t)lo);
 971         nfs4_free_ol_stateid(stid);
 972 }
 973
 974 /*
 975  * Put the persistent reference to an already unhashed generic stateid, while
 976  * holding the cl_lock. If it's the last reference, then put it onto the
 977  * reaplist for later destruction.
 978  */
 979 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
 980                                        struct list_head *reaplist)
 981 {
 982         struct nfs4_stid *s = &stp->st_stid;
 983         struct nfs4_client *clp = s->sc_client;
 984
 985         lockdep_assert_held(&clp->cl_lock);
 986
 987         WARN_ON_ONCE(!list_empty(&stp->st_locks));
 988
 989         if (!atomic_dec_and_test(&s->sc_count)) {
 990                 wake_up_all(&close_wq);
 991                 return;
 992         }
 993
 994         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 995         list_add(&stp->st_locks, reaplist);
 996 }
 997
 998 static void unhash_lock_stateid(struct nfs4_ol_stateid *stp)
 999 {
1000         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1001
1002         lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1003
1004         list_del_init(&stp->st_locks);
1005         unhash_ol_stateid(stp);
1006         unhash_stid(&stp->st_stid);
1007 }
1008
1009 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1010 {
1011         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1012
1013         spin_lock(&oo->oo_owner.so_client->cl_lock);
1014         unhash_lock_stateid(stp);
1015         spin_unlock(&oo->oo_owner.so_client->cl_lock);
1016         nfs4_put_stid(&stp->st_stid);
1017 }
1018
1019 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1020 {
1021         struct nfs4_client *clp = lo->lo_owner.so_client;
1022
1023         lockdep_assert_held(&clp->cl_lock);
1024
1025         list_del_init(&lo->lo_owner.so_strhash);
1026 }
1027
1028 /*
1029  * Free a list of generic stateids that were collected earlier after being
1030  * fully unhashed.
1031  */
1032 static void
1033 free_ol_stateid_reaplist(struct list_head *reaplist)
1034 {
1035         struct nfs4_ol_stateid *stp;
1036         struct nfs4_file *fp;
1037
1038         might_sleep();
1039
1040         while (!list_empty(reaplist)) {
1041                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1042                                        st_locks);
1043                 list_del(&stp->st_locks);
1044                 fp = stp->st_stid.sc_file;
1045                 stp->st_stid.sc_free(&stp->st_stid);
1046                 if (fp)
1047                         put_nfs4_file(fp);
1048         }
1049 }
1050
1051 static void release_lockowner(struct nfs4_lockowner *lo)
1052 {
1053         struct nfs4_client *clp = lo->lo_owner.so_client;
1054         struct nfs4_ol_stateid *stp;
1055         struct list_head reaplist;
1056
1057         INIT_LIST_HEAD(&reaplist);
1058
1059         spin_lock(&clp->cl_lock);
1060         unhash_lockowner_locked(lo);
1061         while (!list_empty(&lo->lo_owner.so_stateids)) {
1062                 stp = list_first_entry(&lo->lo_owner.so_stateids,
1063                                 struct nfs4_ol_stateid, st_perstateowner);
1064                 unhash_lock_stateid(stp);
1065                 put_ol_stateid_locked(stp, &reaplist);
1066         }
1067         spin_unlock(&clp->cl_lock);
1068         free_ol_stateid_reaplist(&reaplist);
1069         nfs4_put_stateowner(&lo->lo_owner);
1070 }
1071
1072 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1073                                        struct list_head *reaplist)
1074 {
1075         struct nfs4_ol_stateid *stp;
1076
1077         while (!list_empty(&open_stp->st_locks)) {
1078                 stp = list_entry(open_stp->st_locks.next,
1079                                 struct nfs4_ol_stateid, st_locks);
1080                 unhash_lock_stateid(stp);
1081                 put_ol_stateid_locked(stp, reaplist);
1082         }
1083 }
1084
1085 static void unhash_open_stateid(struct nfs4_ol_stateid *stp,
1086                                 struct list_head *reaplist)
1087 {
1088         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1089
1090         unhash_ol_stateid(stp);
1091         release_open_stateid_locks(stp, reaplist);
1092 }
1093
1094 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1095 {
1096         LIST_HEAD(reaplist);
1097
1098         spin_lock(&stp->st_stid.sc_client->cl_lock);
1099         unhash_open_stateid(stp, &reaplist);
1100         put_ol_stateid_locked(stp, &reaplist);
1101         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1102         free_ol_stateid_reaplist(&reaplist);
1103 }
1104
1105 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1106 {
1107         struct nfs4_client *clp = oo->oo_owner.so_client;
1108
1109         lockdep_assert_held(&clp->cl_lock);
1110
1111         list_del_init(&oo->oo_owner.so_strhash);
1112         list_del_init(&oo->oo_perclient);
1113 }
1114
1115 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1116 {
1117         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1118                                           nfsd_net_id);
1119         struct nfs4_ol_stateid *s;
1120
1121         spin_lock(&nn->client_lock);
1122         s = oo->oo_last_closed_stid;
1123         if (s) {
1124                 list_del_init(&oo->oo_close_lru);
1125                 oo->oo_last_closed_stid = NULL;
1126         }
1127         spin_unlock(&nn->client_lock);
1128         if (s)
1129                 nfs4_put_stid(&s->st_stid);
1130 }
1131
1132 static void release_openowner(struct nfs4_openowner *oo)
1133 {
1134         struct nfs4_ol_stateid *stp;
1135         struct nfs4_client *clp = oo->oo_owner.so_client;
1136         struct list_head reaplist;
1137
1138         INIT_LIST_HEAD(&reaplist);
1139
1140         spin_lock(&clp->cl_lock);
1141         unhash_openowner_locked(oo);
1142         while (!list_empty(&oo->oo_owner.so_stateids)) {
1143                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1144                                 struct nfs4_ol_stateid, st_perstateowner);
1145                 unhash_open_stateid(stp, &reaplist);
1146                 put_ol_stateid_locked(stp, &reaplist);
1147         }
1148         spin_unlock(&clp->cl_lock);
1149         free_ol_stateid_reaplist(&reaplist);
1150         release_last_closed_stateid(oo);
1151         nfs4_put_stateowner(&oo->oo_owner);
1152 }
1153
1154 static inline int
1155 hash_sessionid(struct nfs4_sessionid *sessionid)
1156 {
1157         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1158
1159         return sid->sequence % SESSION_HASH_SIZE;
1160 }
1161
1162 #ifdef NFSD_DEBUG
1163 static inline void
1164 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1165 {
1166         u32 *ptr = (u32 *)(&sessionid->data[0]);
1167         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1168 }
1169 #else
1170 static inline void
1171 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1172 {
1173 }
1174 #endif
1175
1176 /*
1177  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1178  * won't be used for replay.
1179  */
1180 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1181 {
1182         struct nfs4_stateowner *so = cstate->replay_owner;
1183
1184         if (nfserr == nfserr_replay_me)
1185                 return;
1186
1187         if (!seqid_mutating_err(ntohl(nfserr))) {
1188                 nfsd4_cstate_clear_replay(cstate);
1189                 return;
1190         }
1191         if (!so)
1192                 return;
1193         if (so->so_is_open_owner)
1194                 release_last_closed_stateid(openowner(so));
1195         so->so_seqid++;
1196         return;
1197 }
1198
1199 static void
1200 gen_sessionid(struct nfsd4_session *ses)
1201 {
1202         struct nfs4_client *clp = ses->se_client;
1203         struct nfsd4_sessionid *sid;
1204
1205         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1206         sid->clientid = clp->cl_clientid;
1207         sid->sequence = current_sessionid++;
1208         sid->reserved = 0;
1209 }
1210
1211 /*
1212  * The protocol defines ca_maxresponssize_cached to include the size of
1213  * the rpc header, but all we need to cache is the data starting after
1214  * the end of the initial SEQUENCE operation--the rest we regenerate
1215  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1216  * value that is the number of bytes in our cache plus a few additional
1217  * bytes.  In order to stay on the safe side, and not promise more than
1218  * we can cache, those additional bytes must be the minimum possible: 24
1219  * bytes of rpc header (xid through accept state, with AUTH_NULL
1220  * verifier), 12 for the compound header (with zero-length tag), and 44
1221  * for the SEQUENCE op response:
1222  */
1223 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1224
1225 static void
1226 free_session_slots(struct nfsd4_session *ses)
1227 {
1228         int i;
1229
1230         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1231                 kfree(ses->se_slots[i]);
1232 }
1233
1234 /*
1235  * We don't actually need to cache the rpc and session headers, so we
1236  * can allocate a little less for each slot:
1237  */
1238 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1239 {
1240         u32 size;
1241
1242         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1243                 size = 0;
1244         else
1245                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1246         return size + sizeof(struct nfsd4_slot);
1247 }
1248
1249 /*
1250  * XXX: If we run out of reserved DRC memory we could (up to a point)
1251  * re-negotiate active sessions and reduce their slot usage to make
1252  * room for new connections. For now we just fail the create session.
1253  */
1254 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1255 {
1256         u32 slotsize = slot_bytes(ca);
1257         u32 num = ca->maxreqs;
1258         int avail;
1259
1260         spin_lock(&nfsd_drc_lock);
1261         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1262                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1263         num = min_t(int, num, avail / slotsize);
1264         nfsd_drc_mem_used += num * slotsize;
1265         spin_unlock(&nfsd_drc_lock);
1266
1267         return num;
1268 }
1269
1270 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1271 {
1272         int slotsize = slot_bytes(ca);
1273
1274         spin_lock(&nfsd_drc_lock);
1275         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1276         spin_unlock(&nfsd_drc_lock);
1277 }
1278
1279 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1280                                            struct nfsd4_channel_attrs *battrs)
1281 {
1282         int numslots = fattrs->maxreqs;
1283         int slotsize = slot_bytes(fattrs);
1284         struct nfsd4_session *new;
1285         int mem, i;
1286
1287         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1288                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1289         mem = numslots * sizeof(struct nfsd4_slot *);
1290
1291         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1292         if (!new)
1293                 return NULL;
1294         /* allocate each struct nfsd4_slot and data cache in one piece */
1295         for (i = 0; i < numslots; i++) {
1296                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1297                 if (!new->se_slots[i])
1298                         goto out_free;
1299         }
1300
1301         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1302         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1303
1304         return new;
1305 out_free:
1306         while (i--)
1307                 kfree(new->se_slots[i]);
1308         kfree(new);
1309         return NULL;
1310 }
1311
1312 static void free_conn(struct nfsd4_conn *c)
1313 {
1314         svc_xprt_put(c->cn_xprt);
1315         kfree(c);
1316 }
1317
1318 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1319 {
1320         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1321         struct nfs4_client *clp = c->cn_session->se_client;
1322
1323         spin_lock(&clp->cl_lock);
1324         if (!list_empty(&c->cn_persession)) {
1325                 list_del(&c->cn_persession);
1326                 free_conn(c);
1327         }
1328         nfsd4_probe_callback(clp);
1329         spin_unlock(&clp->cl_lock);
1330 }
1331
1332 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1333 {
1334         struct nfsd4_conn *conn;
1335
1336         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1337         if (!conn)
1338                 return NULL;
1339         svc_xprt_get(rqstp->rq_xprt);
1340         conn->cn_xprt = rqstp->rq_xprt;
1341         conn->cn_flags = flags;
1342         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1343         return conn;
1344 }
1345
1346 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1347 {
1348         conn->cn_session = ses;
1349         list_add(&conn->cn_persession, &ses->se_conns);
1350 }
1351
1352 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1353 {
1354         struct nfs4_client *clp = ses->se_client;
1355
1356         spin_lock(&clp->cl_lock);
1357         __nfsd4_hash_conn(conn, ses);
1358         spin_unlock(&clp->cl_lock);
1359 }
1360
1361 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1362 {
1363         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1364         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1365 }
1366
1367 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1368 {
1369         int ret;
1370
1371         nfsd4_hash_conn(conn, ses);
1372         ret = nfsd4_register_conn(conn);
1373         if (ret)
1374                 /* oops; xprt is already down: */
1375                 nfsd4_conn_lost(&conn->cn_xpt_user);
1376         /* We may have gained or lost a callback channel: */
1377         nfsd4_probe_callback_sync(ses->se_client);
1378 }
1379
1380 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1381 {
1382         u32 dir = NFS4_CDFC4_FORE;
1383
1384         if (cses->flags & SESSION4_BACK_CHAN)
1385                 dir |= NFS4_CDFC4_BACK;
1386         return alloc_conn(rqstp, dir);
1387 }
1388
1389 /* must be called under client_lock */
1390 static void nfsd4_del_conns(struct nfsd4_session *s)
1391 {
1392         struct nfs4_client *clp = s->se_client;
1393         struct nfsd4_conn *c;
1394
1395         spin_lock(&clp->cl_lock);
1396         while (!list_empty(&s->se_conns)) {
1397                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1398                 list_del_init(&c->cn_persession);
1399                 spin_unlock(&clp->cl_lock);
1400
1401                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1402                 free_conn(c);
1403
1404                 spin_lock(&clp->cl_lock);
1405         }
1406         spin_unlock(&clp->cl_lock);
1407 }
1408
1409 static void __free_session(struct nfsd4_session *ses)
1410 {
1411         free_session_slots(ses);
1412         kfree(ses);
1413 }
1414
1415 static void free_session(struct nfsd4_session *ses)
1416 {
1417         nfsd4_del_conns(ses);
1418         nfsd4_put_drc_mem(&ses->se_fchannel);
1419         __free_session(ses);
1420 }
1421
1422 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1423 {
1424         int idx;
1425         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1426
1427         new->se_client = clp;
1428         gen_sessionid(new);
1429
1430         INIT_LIST_HEAD(&new->se_conns);
1431
1432         new->se_cb_seq_nr = 1;
1433         new->se_flags = cses->flags;
1434         new->se_cb_prog = cses->callback_prog;
1435         new->se_cb_sec = cses->cb_sec;
1436         atomic_set(&new->se_ref, 0);
1437         idx = hash_sessionid(&new->se_sessionid);
1438         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1439         spin_lock(&clp->cl_lock);
1440         list_add(&new->se_perclnt, &clp->cl_sessions);
1441         spin_unlock(&clp->cl_lock);
1442
1443         if (cses->flags & SESSION4_BACK_CHAN) {
1444                 struct sockaddr *sa = svc_addr(rqstp);
1445                 /*
1446                  * This is a little silly; with sessions there's no real
1447                  * use for the callback address.  Use the peer address
1448                  * as a reasonable default for now, but consider fixing
1449                  * the rpc client not to require an address in the
1450                  * future:
1451                  */
1452                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1453                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1454         }
1455 }
1456
1457 /* caller must hold client_lock */
1458 static struct nfsd4_session *
1459 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1460 {
1461         struct nfsd4_session *elem;
1462         int idx;
1463         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1464
1465         lockdep_assert_held(&nn->client_lock);
1466
1467         dump_sessionid(__func__, sessionid);
1468         idx = hash_sessionid(sessionid);
1469         /* Search in the appropriate list */
1470         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1471                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1472                             NFS4_MAX_SESSIONID_LEN)) {
1473                         return elem;
1474                 }
1475         }
1476
1477         dprintk("%s: session not found\n", __func__);
1478         return NULL;
1479 }
1480
1481 static struct nfsd4_session *
1482 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1483                 __be32 *ret)
1484 {
1485         struct nfsd4_session *session;
1486         __be32 status = nfserr_badsession;
1487
1488         session = __find_in_sessionid_hashtbl(sessionid, net);
1489         if (!session)
1490                 goto out;
1491         status = nfsd4_get_session_locked(session);
1492         if (status)
1493                 session = NULL;
1494 out:
1495         *ret = status;
1496         return session;
1497 }
1498
1499 /* caller must hold client_lock */
1500 static void
1501 unhash_session(struct nfsd4_session *ses)
1502 {
1503         struct nfs4_client *clp = ses->se_client;
1504         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1505
1506         lockdep_assert_held(&nn->client_lock);
1507
1508         list_del(&ses->se_hash);
1509         spin_lock(&ses->se_client->cl_lock);
1510         list_del(&ses->se_perclnt);
1511         spin_unlock(&ses->se_client->cl_lock);
1512 }
1513
1514 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1515 static int
1516 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1517 {
1518         if (clid->cl_boot == nn->boot_time)
1519                 return 0;
1520         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1521                 clid->cl_boot, clid->cl_id, nn->boot_time);
1522         return 1;
1523 }
1524
1525 /*
1526  * XXX Should we use a slab cache ?
1527  * This type of memory management is somewhat inefficient, but we use it
1528  * anyway since SETCLIENTID is not a common operation.
1529  */
1530 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1531 {
1532         struct nfs4_client *clp;
1533         int i;
1534
1535         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1536         if (clp == NULL)
1537                 return NULL;
1538         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1539         if (clp->cl_name.data == NULL)
1540                 goto err_no_name;
1541         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1542                         OWNER_HASH_SIZE, GFP_KERNEL);
1543         if (!clp->cl_ownerstr_hashtbl)
1544                 goto err_no_hashtbl;
1545         for (i = 0; i < OWNER_HASH_SIZE; i++)
1546                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1547         clp->cl_name.len = name.len;
1548         INIT_LIST_HEAD(&clp->cl_sessions);
1549         idr_init(&clp->cl_stateids);
1550         atomic_set(&clp->cl_refcount, 0);
1551         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1552         INIT_LIST_HEAD(&clp->cl_idhash);
1553         INIT_LIST_HEAD(&clp->cl_openowners);
1554         INIT_LIST_HEAD(&clp->cl_delegations);
1555         INIT_LIST_HEAD(&clp->cl_lru);
1556         INIT_LIST_HEAD(&clp->cl_callbacks);
1557         INIT_LIST_HEAD(&clp->cl_revoked);
1558         spin_lock_init(&clp->cl_lock);
1559         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1560         return clp;
1561 err_no_hashtbl:
1562         kfree(clp->cl_name.data);
1563 err_no_name:
1564         kfree(clp);
1565         return NULL;
1566 }
1567
1568 static void
1569 free_client(struct nfs4_client *clp)
1570 {
1571         while (!list_empty(&clp->cl_sessions)) {
1572                 struct nfsd4_session *ses;
1573                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1574                                 se_perclnt);
1575                 list_del(&ses->se_perclnt);
1576                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1577                 free_session(ses);
1578         }
1579         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1580         free_svc_cred(&clp->cl_cred);
1581         kfree(clp->cl_ownerstr_hashtbl);
1582         kfree(clp->cl_name.data);
1583         idr_destroy(&clp->cl_stateids);
1584         kfree(clp);
1585 }
1586
1587 /* must be called under the client_lock */
1588 static void
1589 unhash_client_locked(struct nfs4_client *clp)
1590 {
1591         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1592         struct nfsd4_session *ses;
1593
1594         lockdep_assert_held(&nn->client_lock);
1595
1596         /* Mark the client as expired! */
1597         clp->cl_time = 0;
1598         /* Make it invisible */
1599         if (!list_empty(&clp->cl_idhash)) {
1600                 list_del_init(&clp->cl_idhash);
1601                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1602                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1603                 else
1604                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1605         }
1606         list_del_init(&clp->cl_lru);
1607         spin_lock(&clp->cl_lock);
1608         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1609                 list_del_init(&ses->se_hash);
1610         spin_unlock(&clp->cl_lock);
1611 }
1612
1613 static void
1614 unhash_client(struct nfs4_client *clp)
1615 {
1616         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1617
1618         spin_lock(&nn->client_lock);
1619         unhash_client_locked(clp);
1620         spin_unlock(&nn->client_lock);
1621 }
1622
1623 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1624 {
1625         if (atomic_read(&clp->cl_refcount))
1626                 return nfserr_jukebox;
1627         unhash_client_locked(clp);
1628         return nfs_ok;
1629 }
1630
1631 static void
1632 __destroy_client(struct nfs4_client *clp)
1633 {
1634         struct nfs4_openowner *oo;
1635         struct nfs4_delegation *dp;
1636         struct list_head reaplist;
1637
1638         INIT_LIST_HEAD(&reaplist);
1639         spin_lock(&state_lock);
1640         while (!list_empty(&clp->cl_delegations)) {
1641                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1642                 unhash_delegation_locked(dp);
1643                 list_add(&dp->dl_recall_lru, &reaplist);
1644         }
1645         spin_unlock(&state_lock);
1646         while (!list_empty(&reaplist)) {
1647                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1648                 list_del_init(&dp->dl_recall_lru);
1649                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1650                 nfs4_put_stid(&dp->dl_stid);
1651         }
1652         while (!list_empty(&clp->cl_revoked)) {
1653                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1654                 list_del_init(&dp->dl_recall_lru);
1655                 nfs4_put_stid(&dp->dl_stid);
1656         }
1657         while (!list_empty(&clp->cl_openowners)) {
1658                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1659                 nfs4_get_stateowner(&oo->oo_owner);
1660                 release_openowner(oo);
1661         }
1662         nfsd4_shutdown_callback(clp);
1663         if (clp->cl_cb_conn.cb_xprt)
1664                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1665         free_client(clp);
1666 }
1667
1668 static void
1669 destroy_client(struct nfs4_client *clp)
1670 {
1671         unhash_client(clp);
1672         __destroy_client(clp);
1673 }
1674
1675 static void expire_client(struct nfs4_client *clp)
1676 {
1677         unhash_client(clp);
1678         nfsd4_client_record_remove(clp);
1679         __destroy_client(clp);
1680 }
1681
1682 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1683 {
1684         memcpy(target->cl_verifier.data, source->data,
1685                         sizeof(target->cl_verifier.data));
1686 }
1687
1688 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1689 {
1690         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1691         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1692 }
1693
1694 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1695 {
1696         if (source->cr_principal) {
1697                 target->cr_principal =
1698                                 kstrdup(source->cr_principal, GFP_KERNEL);
1699                 if (target->cr_principal == NULL)
1700                         return -ENOMEM;
1701         } else
1702                 target->cr_principal = NULL;
1703         target->cr_flavor = source->cr_flavor;
1704         target->cr_uid = source->cr_uid;
1705         target->cr_gid = source->cr_gid;
1706         target->cr_group_info = source->cr_group_info;
1707         get_group_info(target->cr_group_info);
1708         target->cr_gss_mech = source->cr_gss_mech;
1709         if (source->cr_gss_mech)
1710                 gss_mech_get(source->cr_gss_mech);
1711         return 0;
1712 }
1713
1714 static long long
1715 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1716 {
1717         long long res;
1718
1719         res = o1->len - o2->len;
1720         if (res)
1721                 return res;
1722         return (long long)memcmp(o1->data, o2->data, o1->len);
1723 }
1724
1725 static int same_name(const char *n1, const char *n2)
1726 {
1727         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1728 }
1729
1730 static int
1731 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1732 {
1733         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1734 }
1735
1736 static int
1737 same_clid(clientid_t *cl1, clientid_t *cl2)
1738 {
1739         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1740 }
1741
1742 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1743 {
1744         int i;
1745
1746         if (g1->ngroups != g2->ngroups)
1747                 return false;
1748         for (i=0; i<g1->ngroups; i++)
1749                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1750                         return false;
1751         return true;
1752 }
1753
1754 /*
1755  * RFC 3530 language requires clid_inuse be returned when the
1756  * "principal" associated with a requests differs from that previously
1757  * used.  We use uid, gid's, and gss principal string as our best
1758  * approximation.  We also don't want to allow non-gss use of a client
1759  * established using gss: in theory cr_principal should catch that
1760  * change, but in practice cr_principal can be null even in the gss case
1761  * since gssd doesn't always pass down a principal string.
1762  */
1763 static bool is_gss_cred(struct svc_cred *cr)
1764 {
1765         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1766         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1767 }
1768
1769
1770 static bool
1771 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1772 {
1773         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1774                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1775                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1776                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1777                 return false;
1778         if (cr1->cr_principal == cr2->cr_principal)
1779                 return true;
1780         if (!cr1->cr_principal || !cr2->cr_principal)
1781                 return false;
1782         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1783 }
1784
1785 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1786 {
1787         struct svc_cred *cr = &rqstp->rq_cred;
1788         u32 service;
1789
1790         if (!cr->cr_gss_mech)
1791                 return false;
1792         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1793         return service == RPC_GSS_SVC_INTEGRITY ||
1794                service == RPC_GSS_SVC_PRIVACY;
1795 }
1796
1797 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1798 {
1799         struct svc_cred *cr = &rqstp->rq_cred;
1800
1801         if (!cl->cl_mach_cred)
1802                 return true;
1803         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1804                 return false;
1805         if (!svc_rqst_integrity_protected(rqstp))
1806                 return false;
1807         if (!cr->cr_principal)
1808                 return false;
1809         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1810 }
1811
1812 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1813 {
1814         __be32 verf[2];
1815
1816         /*
1817          * This is opaque to client, so no need to byte-swap. Use
1818          * __force to keep sparse happy
1819          */
1820         verf[0] = (__force __be32)get_seconds();
1821         verf[1] = (__force __be32)nn->clientid_counter;
1822         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1823 }
1824
1825 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1826 {
1827         clp->cl_clientid.cl_boot = nn->boot_time;
1828         clp->cl_clientid.cl_id = nn->clientid_counter++;
1829         gen_confirm(clp, nn);
1830 }
1831
1832 static struct nfs4_stid *
1833 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
1834 {
1835         struct nfs4_stid *ret;
1836
1837         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1838         if (!ret || !ret->sc_type)
1839                 return NULL;
1840         return ret;
1841 }
1842
1843 static struct nfs4_stid *
1844 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1845 {
1846         struct nfs4_stid *s;
1847
1848         spin_lock(&cl->cl_lock);
1849         s = find_stateid_locked(cl, t);
1850         if (s != NULL) {
1851                 if (typemask & s->sc_type)
1852                         atomic_inc(&s->sc_count);
1853                 else
1854                         s = NULL;
1855         }
1856         spin_unlock(&cl->cl_lock);
1857         return s;
1858 }
1859
1860 static struct nfs4_client *create_client(struct xdr_netobj name,
1861                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1862 {
1863         struct nfs4_client *clp;
1864         struct sockaddr *sa = svc_addr(rqstp);
1865         int ret;
1866         struct net *net = SVC_NET(rqstp);
1867
1868         clp = alloc_client(name);
1869         if (clp == NULL)
1870                 return NULL;
1871
1872         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1873         if (ret) {
1874                 free_client(clp);
1875                 return NULL;
1876         }
1877         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_run_cb_null);
1878         clp->cl_time = get_seconds();
1879         clear_bit(0, &clp->cl_cb_slot_busy);
1880         copy_verf(clp, verf);
1881         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1882         clp->cl_cb_session = NULL;
1883         clp->net = net;
1884         return clp;
1885 }
1886
1887 static void
1888 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1889 {
1890         struct rb_node **new = &(root->rb_node), *parent = NULL;
1891         struct nfs4_client *clp;
1892
1893         while (*new) {
1894                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1895                 parent = *new;
1896
1897                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1898                         new = &((*new)->rb_left);
1899                 else
1900                         new = &((*new)->rb_right);
1901         }
1902
1903         rb_link_node(&new_clp->cl_namenode, parent, new);
1904         rb_insert_color(&new_clp->cl_namenode, root);
1905 }
1906
1907 static struct nfs4_client *
1908 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1909 {
1910         long long cmp;
1911         struct rb_node *node = root->rb_node;
1912         struct nfs4_client *clp;
1913
1914         while (node) {
1915                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1916                 cmp = compare_blob(&clp->cl_name, name);
1917                 if (cmp > 0)
1918                         node = node->rb_left;
1919                 else if (cmp < 0)
1920                         node = node->rb_right;
1921                 else
1922                         return clp;
1923         }
1924         return NULL;
1925 }
1926
1927 static void
1928 add_to_unconfirmed(struct nfs4_client *clp)
1929 {
1930         unsigned int idhashval;
1931         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1932
1933         lockdep_assert_held(&nn->client_lock);
1934
1935         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1936         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1937         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1938         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1939         renew_client_locked(clp);
1940 }
1941
1942 static void
1943 move_to_confirmed(struct nfs4_client *clp)
1944 {
1945         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1946         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1947
1948         lockdep_assert_held(&nn->client_lock);
1949
1950         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1951         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1952         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1953         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1954         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1955         renew_client_locked(clp);
1956 }
1957
1958 static struct nfs4_client *
1959 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1960 {
1961         struct nfs4_client *clp;
1962         unsigned int idhashval = clientid_hashval(clid->cl_id);
1963
1964         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1965                 if (same_clid(&clp->cl_clientid, clid)) {
1966                         if ((bool)clp->cl_minorversion != sessions)
1967                                 return NULL;
1968                         renew_client_locked(clp);
1969                         return clp;
1970                 }
1971         }
1972         return NULL;
1973 }
1974
1975 static struct nfs4_client *
1976 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1977 {
1978         struct list_head *tbl = nn->conf_id_hashtbl;
1979
1980         lockdep_assert_held(&nn->client_lock);
1981         return find_client_in_id_table(tbl, clid, sessions);
1982 }
1983
1984 static struct nfs4_client *
1985 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1986 {
1987         struct list_head *tbl = nn->unconf_id_hashtbl;
1988
1989         lockdep_assert_held(&nn->client_lock);
1990         return find_client_in_id_table(tbl, clid, sessions);
1991 }
1992
1993 static bool clp_used_exchangeid(struct nfs4_client *clp)
1994 {
1995         return clp->cl_exchange_flags != 0;
1996 }
1997
1998 static struct nfs4_client *
1999 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2000 {
2001         lockdep_assert_held(&nn->client_lock);
2002         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2003 }
2004
2005 static struct nfs4_client *
2006 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2007 {
2008         lockdep_assert_held(&nn->client_lock);
2009         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2010 }
2011
2012 static void
2013 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2014 {
2015         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2016         struct sockaddr *sa = svc_addr(rqstp);
2017         u32 scopeid = rpc_get_scope_id(sa);
2018         unsigned short expected_family;
2019
2020         /* Currently, we only support tcp and tcp6 for the callback channel */
2021         if (se->se_callback_netid_len == 3 &&
2022             !memcmp(se->se_callback_netid_val, "tcp", 3))
2023                 expected_family = AF_INET;
2024         else if (se->se_callback_netid_len == 4 &&
2025                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2026                 expected_family = AF_INET6;
2027         else
2028                 goto out_err;
2029
2030         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2031                                             se->se_callback_addr_len,
2032                                             (struct sockaddr *)&conn->cb_addr,
2033                                             sizeof(conn->cb_addr));
2034
2035         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2036                 goto out_err;
2037
2038         if (conn->cb_addr.ss_family == AF_INET6)
2039                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2040
2041         conn->cb_prog = se->se_callback_prog;
2042         conn->cb_ident = se->se_callback_ident;
2043         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2044         return;
2045 out_err:
2046         conn->cb_addr.ss_family = AF_UNSPEC;
2047         conn->cb_addrlen = 0;
2048         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2049                 "will not receive delegations\n",
2050                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2051
2052         return;
2053 }
2054
2055 /*
2056  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2057  */
2058 static void
2059 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2060 {
2061         struct xdr_buf *buf = resp->xdr.buf;
2062         struct nfsd4_slot *slot = resp->cstate.slot;
2063         unsigned int base;
2064
2065         dprintk("--> %s slot %p\n", __func__, slot);
2066
2067         slot->sl_opcnt = resp->opcnt;
2068         slot->sl_status = resp->cstate.status;
2069
2070         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2071         if (nfsd4_not_cached(resp)) {
2072                 slot->sl_datalen = 0;
2073                 return;
2074         }
2075         base = resp->cstate.data_offset;
2076         slot->sl_datalen = buf->len - base;
2077         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2078                 WARN("%s: sessions DRC could not cache compound\n", __func__);
2079         return;
2080 }
2081
2082 /*
2083  * Encode the replay sequence operation from the slot values.
2084  * If cachethis is FALSE encode the uncached rep error on the next
2085  * operation which sets resp->p and increments resp->opcnt for
2086  * nfs4svc_encode_compoundres.
2087  *
2088  */
2089 static __be32
2090 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2091                           struct nfsd4_compoundres *resp)
2092 {
2093         struct nfsd4_op *op;
2094         struct nfsd4_slot *slot = resp->cstate.slot;
2095
2096         /* Encode the replayed sequence operation */
2097         op = &args->ops[resp->opcnt - 1];
2098         nfsd4_encode_operation(resp, op);
2099
2100         /* Return nfserr_retry_uncached_rep in next operation. */
2101         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2102                 op = &args->ops[resp->opcnt++];
2103                 op->status = nfserr_retry_uncached_rep;
2104                 nfsd4_encode_operation(resp, op);
2105         }
2106         return op->status;
2107 }
2108
2109 /*
2110  * The sequence operation is not cached because we can use the slot and
2111  * session values.
2112  */
2113 static __be32
2114 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2115                          struct nfsd4_sequence *seq)
2116 {
2117         struct nfsd4_slot *slot = resp->cstate.slot;
2118         struct xdr_stream *xdr = &resp->xdr;
2119         __be32 *p;
2120         __be32 status;
2121
2122         dprintk("--> %s slot %p\n", __func__, slot);
2123
2124         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2125         if (status)
2126                 return status;
2127
2128         p = xdr_reserve_space(xdr, slot->sl_datalen);
2129         if (!p) {
2130                 WARN_ON_ONCE(1);
2131                 return nfserr_serverfault;
2132         }
2133         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2134         xdr_commit_encode(xdr);
2135
2136         resp->opcnt = slot->sl_opcnt;
2137         return slot->sl_status;
2138 }
2139
2140 /*
2141  * Set the exchange_id flags returned by the server.
2142  */
2143 static void
2144 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2145 {
2146         /* pNFS is not supported */
2147         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2148
2149         /* Referrals are supported, Migration is not. */
2150         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2151
2152         /* set the wire flags to return to client. */
2153         clid->flags = new->cl_exchange_flags;
2154 }
2155
2156 static bool client_has_state(struct nfs4_client *clp)
2157 {
2158         /*
2159          * Note clp->cl_openowners check isn't quite right: there's no
2160          * need to count owners without stateid's.
2161          *
2162          * Also note we should probably be using this in 4.0 case too.
2163          */
2164         return !list_empty(&clp->cl_openowners)
2165                 || !list_empty(&clp->cl_delegations)
2166                 || !list_empty(&clp->cl_sessions);
2167 }
2168
2169 __be32
2170 nfsd4_exchange_id(struct svc_rqst *rqstp,
2171                   struct nfsd4_compound_state *cstate,
2172                   struct nfsd4_exchange_id *exid)
2173 {
2174         struct nfs4_client *conf, *new;
2175         struct nfs4_client *unconf = NULL;
2176         __be32 status;
2177         char                    addr_str[INET6_ADDRSTRLEN];
2178         nfs4_verifier           verf = exid->verifier;
2179         struct sockaddr         *sa = svc_addr(rqstp);
2180         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2181         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2182
2183         rpc_ntop(sa, addr_str, sizeof(addr_str));
2184         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2185                 "ip_addr=%s flags %x, spa_how %d\n",
2186                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2187                 addr_str, exid->flags, exid->spa_how);
2188
2189         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2190                 return nfserr_inval;
2191
2192         switch (exid->spa_how) {
2193         case SP4_MACH_CRED:
2194                 if (!svc_rqst_integrity_protected(rqstp))
2195                         return nfserr_inval;
2196         case SP4_NONE:
2197                 break;
2198         default:                                /* checked by xdr code */
2199                 WARN_ON_ONCE(1);
2200         case SP4_SSV:
2201                 return nfserr_encr_alg_unsupp;
2202         }
2203
2204         new = create_client(exid->clname, rqstp, &verf);
2205         if (new == NULL)
2206                 return nfserr_jukebox;
2207
2208         /* Cases below refer to rfc 5661 section 18.35.4: */
2209         spin_lock(&nn->client_lock);
2210         conf = find_confirmed_client_by_name(&exid->clname, nn);
2211         if (conf) {
2212                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2213                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2214
2215                 if (update) {
2216                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2217                                 status = nfserr_inval;
2218                                 goto out;
2219                         }
2220                         if (!mach_creds_match(conf, rqstp)) {
2221                                 status = nfserr_wrong_cred;
2222                                 goto out;
2223                         }
2224                         if (!creds_match) { /* case 9 */
2225                                 status = nfserr_perm;
2226                                 goto out;
2227                         }
2228                         if (!verfs_match) { /* case 8 */
2229                                 status = nfserr_not_same;
2230                                 goto out;
2231                         }
2232                         /* case 6 */
2233                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2234                         goto out_copy;
2235                 }
2236                 if (!creds_match) { /* case 3 */
2237                         if (client_has_state(conf)) {
2238                                 status = nfserr_clid_inuse;
2239                                 goto out;
2240                         }
2241                         goto out_new;
2242                 }
2243                 if (verfs_match) { /* case 2 */
2244                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2245                         goto out_copy;
2246                 }
2247                 /* case 5, client reboot */
2248                 conf = NULL;
2249                 goto out_new;
2250         }
2251
2252         if (update) { /* case 7 */
2253                 status = nfserr_noent;
2254                 goto out;
2255         }
2256
2257         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2258         if (unconf) /* case 4, possible retry or client restart */
2259                 unhash_client_locked(unconf);
2260
2261         /* case 1 (normal case) */
2262 out_new:
2263         if (conf) {
2264                 status = mark_client_expired_locked(conf);
2265                 if (status)
2266                         goto out;
2267         }
2268         new->cl_minorversion = cstate->minorversion;
2269         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
2270
2271         gen_clid(new, nn);
2272         add_to_unconfirmed(new);
2273         swap(new, conf);
2274 out_copy:
2275         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2276         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2277
2278         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2279         nfsd4_set_ex_flags(conf, exid);
2280
2281         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2282                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2283         status = nfs_ok;
2284
2285 out:
2286         spin_unlock(&nn->client_lock);
2287         if (new)
2288                 expire_client(new);
2289         if (unconf)
2290                 expire_client(unconf);
2291         return status;
2292 }
2293
2294 static __be32
2295 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2296 {
2297         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2298                 slot_seqid);
2299
2300         /* The slot is in use, and no response has been sent. */
2301         if (slot_inuse) {
2302                 if (seqid == slot_seqid)
2303                         return nfserr_jukebox;
2304                 else
2305                         return nfserr_seq_misordered;
2306         }
2307         /* Note unsigned 32-bit arithmetic handles wraparound: */
2308         if (likely(seqid == slot_seqid + 1))
2309                 return nfs_ok;
2310         if (seqid == slot_seqid)
2311                 return nfserr_replay_cache;
2312         return nfserr_seq_misordered;
2313 }
2314
2315 /*
2316  * Cache the create session result into the create session single DRC
2317  * slot cache by saving the xdr structure. sl_seqid has been set.
2318  * Do this for solo or embedded create session operations.
2319  */
2320 static void
2321 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2322                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2323 {
2324         slot->sl_status = nfserr;
2325         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2326 }
2327
2328 static __be32
2329 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2330                             struct nfsd4_clid_slot *slot)
2331 {
2332         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2333         return slot->sl_status;
2334 }
2335
2336 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2337                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2338                         1 +     /* MIN tag is length with zero, only length */ \
2339                         3 +     /* version, opcount, opcode */ \
2340                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2341                                 /* seqid, slotID, slotID, cache */ \
2342                         4 ) * sizeof(__be32))
2343
2344 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2345                         2 +     /* verifier: AUTH_NULL, length 0 */\
2346                         1 +     /* status */ \
2347                         1 +     /* MIN tag is length with zero, only length */ \
2348                         3 +     /* opcount, opcode, opstatus*/ \
2349                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2350                                 /* seqid, slotID, slotID, slotID, status */ \
2351                         5 ) * sizeof(__be32))
2352
2353 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2354 {
2355         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2356
2357         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2358                 return nfserr_toosmall;
2359         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2360                 return nfserr_toosmall;
2361         ca->headerpadsz = 0;
2362         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2363         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2364         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2365         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2366                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2367         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2368         /*
2369          * Note decreasing slot size below client's request may make it
2370          * difficult for client to function correctly, whereas
2371          * decreasing the number of slots will (just?) affect
2372          * performance.  When short on memory we therefore prefer to
2373          * decrease number of slots instead of their size.  Clients that
2374          * request larger slots than they need will get poor results:
2375          */
2376         ca->maxreqs = nfsd4_get_drc_mem(ca);
2377         if (!ca->maxreqs)
2378                 return nfserr_jukebox;
2379
2380         return nfs_ok;
2381 }
2382
2383 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2384                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2385 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2386                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2387
2388 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2389 {
2390         ca->headerpadsz = 0;
2391
2392         /*
2393          * These RPC_MAX_HEADER macros are overkill, especially since we
2394          * don't even do gss on the backchannel yet.  But this is still
2395          * less than 1k.  Tighten up this estimate in the unlikely event
2396          * it turns out to be a problem for some client:
2397          */
2398         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2399                 return nfserr_toosmall;
2400         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2401                 return nfserr_toosmall;
2402         ca->maxresp_cached = 0;
2403         if (ca->maxops < 2)
2404                 return nfserr_toosmall;
2405
2406         return nfs_ok;
2407 }
2408
2409 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2410 {
2411         switch (cbs->flavor) {
2412         case RPC_AUTH_NULL:
2413         case RPC_AUTH_UNIX:
2414                 return nfs_ok;
2415         default:
2416                 /*
2417                  * GSS case: the spec doesn't allow us to return this
2418                  * error.  But it also doesn't allow us not to support
2419                  * GSS.
2420                  * I'd rather this fail hard than return some error the
2421                  * client might think it can already handle:
2422                  */
2423                 return nfserr_encr_alg_unsupp;
2424         }
2425 }
2426
2427 __be32
2428 nfsd4_create_session(struct svc_rqst *rqstp,
2429                      struct nfsd4_compound_state *cstate,
2430                      struct nfsd4_create_session *cr_ses)
2431 {
2432         struct sockaddr *sa = svc_addr(rqstp);
2433         struct nfs4_client *conf, *unconf;
2434         struct nfs4_client *old = NULL;
2435         struct nfsd4_session *new;
2436         struct nfsd4_conn *conn;
2437         struct nfsd4_clid_slot *cs_slot = NULL;
2438         __be32 status = 0;
2439         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2440
2441         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2442                 return nfserr_inval;
2443         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2444         if (status)
2445                 return status;
2446         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2447         if (status)
2448                 return status;
2449         status = check_backchannel_attrs(&cr_ses->back_channel);
2450         if (status)
2451                 goto out_release_drc_mem;
2452         status = nfserr_jukebox;
2453         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2454         if (!new)
2455                 goto out_release_drc_mem;
2456         conn = alloc_conn_from_crses(rqstp, cr_ses);
2457         if (!conn)
2458                 goto out_free_session;
2459
2460         spin_lock(&nn->client_lock);
2461         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2462         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2463         WARN_ON_ONCE(conf && unconf);
2464
2465         if (conf) {
2466                 status = nfserr_wrong_cred;
2467                 if (!mach_creds_match(conf, rqstp))
2468                         goto out_free_conn;
2469                 cs_slot = &conf->cl_cs_slot;
2470                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2471                 if (status == nfserr_replay_cache) {
2472                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
2473                         goto out_free_conn;
2474                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
2475                         status = nfserr_seq_misordered;
2476                         goto out_free_conn;
2477                 }
2478         } else if (unconf) {
2479                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2480                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2481                         status = nfserr_clid_inuse;
2482                         goto out_free_conn;
2483                 }
2484                 status = nfserr_wrong_cred;
2485                 if (!mach_creds_match(unconf, rqstp))
2486                         goto out_free_conn;
2487                 cs_slot = &unconf->cl_cs_slot;
2488                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2489                 if (status) {
2490                         /* an unconfirmed replay returns misordered */
2491                         status = nfserr_seq_misordered;
2492                         goto out_free_conn;
2493                 }
2494                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2495                 if (old) {
2496                         status = mark_client_expired_locked(old);
2497                         if (status) {
2498                                 old = NULL;
2499                                 goto out_free_conn;
2500                         }
2501                 }
2502                 move_to_confirmed(unconf);
2503                 conf = unconf;
2504         } else {
2505                 status = nfserr_stale_clientid;
2506                 goto out_free_conn;
2507         }
2508         status = nfs_ok;
2509         /*
2510          * We do not support RDMA or persistent sessions
2511          */
2512         cr_ses->flags &= ~SESSION4_PERSIST;
2513         cr_ses->flags &= ~SESSION4_RDMA;
2514
2515         init_session(rqstp, new, conf, cr_ses);
2516         nfsd4_get_session_locked(new);
2517
2518         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2519                NFS4_MAX_SESSIONID_LEN);
2520         cs_slot->sl_seqid++;
2521         cr_ses->seqid = cs_slot->sl_seqid;
2522
2523         /* cache solo and embedded create sessions under the client_lock */
2524         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2525         spin_unlock(&nn->client_lock);
2526         /* init connection and backchannel */
2527         nfsd4_init_conn(rqstp, conn, new);
2528         nfsd4_put_session(new);
2529         if (old)
2530                 expire_client(old);
2531         return status;
2532 out_free_conn:
2533         spin_unlock(&nn->client_lock);
2534         free_conn(conn);
2535         if (old)
2536                 expire_client(old);
2537 out_free_session:
2538         __free_session(new);
2539 out_release_drc_mem:
2540         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2541         return status;
2542 }
2543
2544 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2545 {
2546         switch (*dir) {
2547         case NFS4_CDFC4_FORE:
2548         case NFS4_CDFC4_BACK:
2549                 return nfs_ok;
2550         case NFS4_CDFC4_FORE_OR_BOTH:
2551         case NFS4_CDFC4_BACK_OR_BOTH:
2552                 *dir = NFS4_CDFC4_BOTH;
2553                 return nfs_ok;
2554         };
2555         return nfserr_inval;
2556 }
2557
2558 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2559 {
2560         struct nfsd4_session *session = cstate->session;
2561         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2562         __be32 status;
2563
2564         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2565         if (status)
2566                 return status;
2567         spin_lock(&nn->client_lock);
2568         session->se_cb_prog = bc->bc_cb_program;
2569         session->se_cb_sec = bc->bc_cb_sec;
2570         spin_unlock(&nn->client_lock);
2571
2572         nfsd4_probe_callback(session->se_client);
2573
2574         return nfs_ok;
2575 }
2576
2577 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2578                      struct nfsd4_compound_state *cstate,
2579                      struct nfsd4_bind_conn_to_session *bcts)
2580 {
2581         __be32 status;
2582         struct nfsd4_conn *conn;
2583         struct nfsd4_session *session;
2584         struct net *net = SVC_NET(rqstp);
2585         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2586
2587         if (!nfsd4_last_compound_op(rqstp))
2588                 return nfserr_not_only_op;
2589         spin_lock(&nn->client_lock);
2590         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2591         spin_unlock(&nn->client_lock);
2592         if (!session)
2593                 goto out_no_session;
2594         status = nfserr_wrong_cred;
2595         if (!mach_creds_match(session->se_client, rqstp))
2596                 goto out;
2597         status = nfsd4_map_bcts_dir(&bcts->dir);
2598         if (status)
2599                 goto out;
2600         conn = alloc_conn(rqstp, bcts->dir);
2601         status = nfserr_jukebox;
2602         if (!conn)
2603                 goto out;
2604         nfsd4_init_conn(rqstp, conn, session);
2605         status = nfs_ok;
2606 out:
2607         nfsd4_put_session(session);
2608 out_no_session:
2609         return status;
2610 }
2611
2612 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2613 {
2614         if (!session)
2615                 return 0;
2616         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2617 }
2618
2619 __be32
2620 nfsd4_destroy_session(struct svc_rqst *r,
2621                       struct nfsd4_compound_state *cstate,
2622                       struct nfsd4_destroy_session *sessionid)
2623 {
2624         struct nfsd4_session *ses;
2625         __be32 status;
2626         int ref_held_by_me = 0;
2627         struct net *net = SVC_NET(r);
2628         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2629
2630         status = nfserr_not_only_op;
2631         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2632                 if (!nfsd4_last_compound_op(r))
2633                         goto out;
2634                 ref_held_by_me++;
2635         }
2636         dump_sessionid(__func__, &sessionid->sessionid);
2637         spin_lock(&nn->client_lock);
2638         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2639         if (!ses)
2640                 goto out_client_lock;
2641         status = nfserr_wrong_cred;
2642         if (!mach_creds_match(ses->se_client, r))
2643                 goto out_put_session;
2644         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2645         if (status)
2646                 goto out_put_session;
2647         unhash_session(ses);
2648         spin_unlock(&nn->client_lock);
2649
2650         nfsd4_probe_callback_sync(ses->se_client);
2651
2652         spin_lock(&nn->client_lock);
2653         status = nfs_ok;
2654 out_put_session:
2655         nfsd4_put_session_locked(ses);
2656 out_client_lock:
2657         spin_unlock(&nn->client_lock);
2658 out:
2659         return status;
2660 }
2661
2662 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2663 {
2664         struct nfsd4_conn *c;
2665
2666         list_for_each_entry(c, &s->se_conns, cn_persession) {
2667                 if (c->cn_xprt == xpt) {
2668                         return c;
2669                 }
2670         }
2671         return NULL;
2672 }
2673
2674 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2675 {
2676         struct nfs4_client *clp = ses->se_client;
2677         struct nfsd4_conn *c;
2678         __be32 status = nfs_ok;
2679         int ret;
2680
2681         spin_lock(&clp->cl_lock);
2682         c = __nfsd4_find_conn(new->cn_xprt, ses);
2683         if (c)
2684                 goto out_free;
2685         status = nfserr_conn_not_bound_to_session;
2686         if (clp->cl_mach_cred)
2687                 goto out_free;
2688         __nfsd4_hash_conn(new, ses);
2689         spin_unlock(&clp->cl_lock);
2690         ret = nfsd4_register_conn(new);
2691         if (ret)
2692                 /* oops; xprt is already down: */
2693                 nfsd4_conn_lost(&new->cn_xpt_user);
2694         return nfs_ok;
2695 out_free:
2696         spin_unlock(&clp->cl_lock);
2697         free_conn(new);
2698         return status;
2699 }
2700
2701 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2702 {
2703         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2704
2705         return args->opcnt > session->se_fchannel.maxops;
2706 }
2707
2708 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2709                                   struct nfsd4_session *session)
2710 {
2711         struct xdr_buf *xb = &rqstp->rq_arg;
2712
2713         return xb->len > session->se_fchannel.maxreq_sz;
2714 }
2715
2716 __be32
2717 nfsd4_sequence(struct svc_rqst *rqstp,
2718                struct nfsd4_compound_state *cstate,
2719                struct nfsd4_sequence *seq)
2720 {
2721         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2722         struct xdr_stream *xdr = &resp->xdr;
2723         struct nfsd4_session *session;
2724         struct nfs4_client *clp;
2725         struct nfsd4_slot *slot;
2726         struct nfsd4_conn *conn;
2727         __be32 status;
2728         int buflen;
2729         struct net *net = SVC_NET(rqstp);
2730         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2731
2732         if (resp->opcnt != 1)
2733                 return nfserr_sequence_pos;
2734
2735         /*
2736          * Will be either used or freed by nfsd4_sequence_check_conn
2737          * below.
2738          */
2739         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2740         if (!conn)
2741                 return nfserr_jukebox;
2742
2743         spin_lock(&nn->client_lock);
2744         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2745         if (!session)
2746                 goto out_no_session;
2747         clp = session->se_client;
2748
2749         status = nfserr_too_many_ops;
2750         if (nfsd4_session_too_many_ops(rqstp, session))
2751                 goto out_put_session;
2752
2753         status = nfserr_req_too_big;
2754         if (nfsd4_request_too_big(rqstp, session))
2755                 goto out_put_session;
2756
2757         status = nfserr_badslot;
2758         if (seq->slotid >= session->se_fchannel.maxreqs)
2759                 goto out_put_session;
2760
2761         slot = session->se_slots[seq->slotid];
2762         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2763
2764         /* We do not negotiate the number of slots yet, so set the
2765          * maxslots to the session maxreqs which is used to encode
2766          * sr_highest_slotid and the sr_target_slot id to maxslots */
2767         seq->maxslots = session->se_fchannel.maxreqs;
2768
2769         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2770                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2771         if (status == nfserr_replay_cache) {
2772                 status = nfserr_seq_misordered;
2773                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2774                         goto out_put_session;
2775                 cstate->slot = slot;
2776                 cstate->session = session;
2777                 cstate->clp = clp;
2778                 /* Return the cached reply status and set cstate->status
2779                  * for nfsd4_proc_compound processing */
2780                 status = nfsd4_replay_cache_entry(resp, seq);
2781                 cstate->status = nfserr_replay_cache;
2782                 goto out;
2783         }
2784         if (status)
2785                 goto out_put_session;
2786
2787         status = nfsd4_sequence_check_conn(conn, session);
2788         conn = NULL;
2789         if (status)
2790                 goto out_put_session;
2791
2792         buflen = (seq->cachethis) ?
2793                         session->se_fchannel.maxresp_cached :
2794                         session->se_fchannel.maxresp_sz;
2795         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
2796                                     nfserr_rep_too_big;
2797         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
2798                 goto out_put_session;
2799         svc_reserve(rqstp, buflen);
2800
2801         status = nfs_ok;
2802         /* Success! bump slot seqid */
2803         slot->sl_seqid = seq->seqid;
2804         slot->sl_flags |= NFSD4_SLOT_INUSE;
2805         if (seq->cachethis)
2806                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2807         else
2808                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2809
2810         cstate->slot = slot;
2811         cstate->session = session;
2812         cstate->clp = clp;
2813
2814 out:
2815         switch (clp->cl_cb_state) {
2816         case NFSD4_CB_DOWN:
2817                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2818                 break;
2819         case NFSD4_CB_FAULT:
2820                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2821                 break;
2822         default:
2823                 seq->status_flags = 0;
2824         }
2825         if (!list_empty(&clp->cl_revoked))
2826                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2827 out_no_session:
2828         if (conn)
2829                 free_conn(conn);
2830         spin_unlock(&nn->client_lock);
2831         return status;
2832 out_put_session:
2833         nfsd4_put_session_locked(session);
2834         goto out_no_session;
2835 }
2836
2837 void
2838 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
2839 {
2840         struct nfsd4_compound_state *cs = &resp->cstate;
2841
2842         if (nfsd4_has_session(cs)) {
2843                 if (cs->status != nfserr_replay_cache) {
2844                         nfsd4_store_cache_entry(resp);
2845                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
2846                 }
2847                 /* Drop session reference that was taken in nfsd4_sequence() */
2848                 nfsd4_put_session(cs->session);
2849         } else if (cs->clp)
2850                 put_client_renew(cs->clp);
2851 }
2852
2853 __be32
2854 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2855 {
2856         struct nfs4_client *conf, *unconf;
2857         struct nfs4_client *clp = NULL;
2858         __be32 status = 0;
2859         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2860
2861         spin_lock(&nn->client_lock);
2862         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2863         conf = find_confirmed_client(&dc->clientid, true, nn);
2864         WARN_ON_ONCE(conf && unconf);
2865
2866         if (conf) {
2867                 if (client_has_state(conf)) {
2868                         status = nfserr_clientid_busy;
2869                         goto out;
2870                 }
2871                 status = mark_client_expired_locked(conf);
2872                 if (status)
2873                         goto out;
2874                 clp = conf;
2875         } else if (unconf)
2876                 clp = unconf;
2877         else {
2878                 status = nfserr_stale_clientid;
2879                 goto out;
2880         }
2881         if (!mach_creds_match(clp, rqstp)) {
2882                 clp = NULL;
2883                 status = nfserr_wrong_cred;
2884                 goto out;
2885         }
2886         unhash_client_locked(clp);
2887 out:
2888         spin_unlock(&nn->client_lock);
2889         if (clp)
2890                 expire_client(clp);
2891         return status;
2892 }
2893
2894 __be32
2895 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2896 {
2897         __be32 status = 0;
2898
2899         if (rc->rca_one_fs) {
2900                 if (!cstate->current_fh.fh_dentry)
2901                         return nfserr_nofilehandle;
2902                 /*
2903                  * We don't take advantage of the rca_one_fs case.
2904                  * That's OK, it's optional, we can safely ignore it.
2905                  */
2906                  return nfs_ok;
2907         }
2908
2909         status = nfserr_complete_already;
2910         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2911                              &cstate->session->se_client->cl_flags))
2912                 goto out;
2913
2914         status = nfserr_stale_clientid;
2915         if (is_client_expired(cstate->session->se_client))
2916                 /*
2917                  * The following error isn't really legal.
2918                  * But we only get here if the client just explicitly
2919                  * destroyed the client.  Surely it no longer cares what
2920                  * error it gets back on an operation for the dead
2921                  * client.
2922                  */
2923                 goto out;
2924
2925         status = nfs_ok;
2926         nfsd4_client_record_create(cstate->session->se_client);
2927 out:
2928         return status;
2929 }
2930
2931 __be32
2932 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2933                   struct nfsd4_setclientid *setclid)
2934 {
2935         struct xdr_netobj       clname = setclid->se_name;
2936         nfs4_verifier           clverifier = setclid->se_verf;
2937         struct nfs4_client      *conf, *new;
2938         struct nfs4_client      *unconf = NULL;
2939         __be32                  status;
2940         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2941
2942         new = create_client(clname, rqstp, &clverifier);
2943         if (new == NULL)
2944                 return nfserr_jukebox;
2945         /* Cases below refer to rfc 3530 section 14.2.33: */
2946         spin_lock(&nn->client_lock);
2947         conf = find_confirmed_client_by_name(&clname, nn);
2948         if (conf) {
2949                 /* case 0: */
2950                 status = nfserr_clid_inuse;
2951                 if (clp_used_exchangeid(conf))
2952                         goto out;
2953                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2954                         char addr_str[INET6_ADDRSTRLEN];
2955                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2956                                  sizeof(addr_str));
2957                         dprintk("NFSD: setclientid: string in use by client "
2958                                 "at %s\n", addr_str);
2959                         goto out;
2960                 }
2961         }
2962         unconf = find_unconfirmed_client_by_name(&clname, nn);
2963         if (unconf)
2964                 unhash_client_locked(unconf);
2965         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2966                 /* case 1: probable callback update */
2967                 copy_clid(new, conf);
2968         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2969                 gen_clid(new, nn);
2970         new->cl_minorversion = 0;
2971         gen_callback(new, setclid, rqstp);
2972         add_to_unconfirmed(new);
2973         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2974         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2975         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2976         new = NULL;
2977         status = nfs_ok;
2978 out:
2979         spin_unlock(&nn->client_lock);
2980         if (new)
2981                 free_client(new);
2982         if (unconf)
2983                 expire_client(unconf);
2984         return status;
2985 }
2986
2987
2988 __be32
2989 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2990                          struct nfsd4_compound_state *cstate,
2991                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2992 {
2993         struct nfs4_client *conf, *unconf;
2994         struct nfs4_client *old = NULL;
2995         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2996         clientid_t * clid = &setclientid_confirm->sc_clientid;
2997         __be32 status;
2998         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2999
3000         if (STALE_CLIENTID(clid, nn))
3001                 return nfserr_stale_clientid;
3002
3003         spin_lock(&nn->client_lock);
3004         conf = find_confirmed_client(clid, false, nn);
3005         unconf = find_unconfirmed_client(clid, false, nn);
3006         /*
3007          * We try hard to give out unique clientid's, so if we get an
3008          * attempt to confirm the same clientid with a different cred,
3009          * there's a bug somewhere.  Let's charitably assume it's our
3010          * bug.
3011          */
3012         status = nfserr_serverfault;
3013         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3014                 goto out;
3015         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3016                 goto out;
3017         /* cases below refer to rfc 3530 section 14.2.34: */
3018         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3019                 if (conf && !unconf) /* case 2: probable retransmit */
3020                         status = nfs_ok;
3021                 else /* case 4: client hasn't noticed we rebooted yet? */
3022                         status = nfserr_stale_clientid;
3023                 goto out;
3024         }
3025         status = nfs_ok;
3026         if (conf) { /* case 1: callback update */
3027                 old = unconf;
3028                 unhash_client_locked(old);
3029                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3030         } else { /* case 3: normal case; new or rebooted client */
3031                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3032                 if (old) {
3033                         status = mark_client_expired_locked(old);
3034                         if (status) {
3035                                 old = NULL;
3036                                 goto out;
3037                         }
3038                 }
3039                 move_to_confirmed(unconf);
3040                 conf = unconf;
3041         }
3042         get_client_locked(conf);
3043         spin_unlock(&nn->client_lock);
3044         nfsd4_probe_callback(conf);
3045         spin_lock(&nn->client_lock);
3046         put_client_renew_locked(conf);
3047 out:
3048         spin_unlock(&nn->client_lock);
3049         if (old)
3050                 expire_client(old);
3051         return status;
3052 }
3053
3054 static struct nfs4_file *nfsd4_alloc_file(void)
3055 {
3056         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3057 }
3058
3059 /* OPEN Share state helper functions */
3060 static void nfsd4_init_file(struct nfs4_file *fp, struct knfsd_fh *fh)
3061 {
3062         unsigned int hashval = file_hashval(fh);
3063
3064         lockdep_assert_held(&state_lock);
3065
3066         atomic_set(&fp->fi_ref, 1);
3067         spin_lock_init(&fp->fi_lock);
3068         INIT_LIST_HEAD(&fp->fi_stateids);
3069         INIT_LIST_HEAD(&fp->fi_delegations);
3070         fh_copy_shallow(&fp->fi_fhandle, fh);
3071         fp->fi_had_conflict = false;
3072         fp->fi_lease = NULL;
3073         fp->fi_share_deny = 0;
3074         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3075         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3076         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
3077 }
3078
3079 void
3080 nfsd4_free_slabs(void)
3081 {
3082         kmem_cache_destroy(openowner_slab);
3083         kmem_cache_destroy(lockowner_slab);
3084         kmem_cache_destroy(file_slab);
3085         kmem_cache_destroy(stateid_slab);
3086         kmem_cache_destroy(deleg_slab);
3087 }
3088
3089 int
3090 nfsd4_init_slabs(void)
3091 {
3092         openowner_slab = kmem_cache_create("nfsd4_openowners",
3093                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3094         if (openowner_slab == NULL)
3095                 goto out;
3096         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3097                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3098         if (lockowner_slab == NULL)
3099                 goto out_free_openowner_slab;
3100         file_slab = kmem_cache_create("nfsd4_files",
3101                         sizeof(struct nfs4_file), 0, 0, NULL);
3102         if (file_slab == NULL)
3103                 goto out_free_lockowner_slab;
3104         stateid_slab = kmem_cache_create("nfsd4_stateids",
3105                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3106         if (stateid_slab == NULL)
3107                 goto out_free_file_slab;
3108         deleg_slab = kmem_cache_create("nfsd4_delegations",
3109                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3110         if (deleg_slab == NULL)
3111                 goto out_free_stateid_slab;
3112         return 0;
3113
3114 out_free_stateid_slab:
3115         kmem_cache_destroy(stateid_slab);
3116 out_free_file_slab:
3117         kmem_cache_destroy(file_slab);
3118 out_free_lockowner_slab:
3119         kmem_cache_destroy(lockowner_slab);
3120 out_free_openowner_slab:
3121         kmem_cache_destroy(openowner_slab);
3122 out:
3123         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3124         return -ENOMEM;
3125 }
3126
3127 static void init_nfs4_replay(struct nfs4_replay *rp)
3128 {
3129         rp->rp_status = nfserr_serverfault;
3130         rp->rp_buflen = 0;
3131         rp->rp_buf = rp->rp_ibuf;
3132         mutex_init(&rp->rp_mutex);
3133 }
3134
3135 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3136                 struct nfs4_stateowner *so)
3137 {
3138         if (!nfsd4_has_session(cstate)) {
3139                 mutex_lock(&so->so_replay.rp_mutex);
3140                 cstate->replay_owner = nfs4_get_stateowner(so);
3141         }
3142 }
3143
3144 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3145 {
3146         struct nfs4_stateowner *so = cstate->replay_owner;
3147
3148         if (so != NULL) {
3149                 cstate->replay_owner = NULL;
3150                 mutex_unlock(&so->so_replay.rp_mutex);
3151                 nfs4_put_stateowner(so);
3152         }
3153 }
3154
3155 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3156 {
3157         struct nfs4_stateowner *sop;
3158
3159         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3160         if (!sop)
3161                 return NULL;
3162
3163         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3164         if (!sop->so_owner.data) {
3165                 kmem_cache_free(slab, sop);
3166                 return NULL;
3167         }
3168         sop->so_owner.len = owner->len;
3169
3170         INIT_LIST_HEAD(&sop->so_stateids);
3171         sop->so_client = clp;
3172         init_nfs4_replay(&sop->so_replay);
3173         atomic_set(&sop->so_count, 1);
3174         return sop;
3175 }
3176
3177 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3178 {
3179         lockdep_assert_held(&clp->cl_lock);
3180
3181         list_add(&oo->oo_owner.so_strhash,
3182                  &clp->cl_ownerstr_hashtbl[strhashval]);
3183         list_add(&oo->oo_perclient, &clp->cl_openowners);
3184 }
3185
3186 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3187 {
3188         unhash_openowner_locked(openowner(so));
3189 }
3190
3191 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3192 {
3193         struct nfs4_openowner *oo = openowner(so);
3194
3195         kmem_cache_free(openowner_slab, oo);
3196 }
3197
3198 static const struct nfs4_stateowner_operations openowner_ops = {
3199         .so_unhash =    nfs4_unhash_openowner,
3200         .so_free =      nfs4_free_openowner,
3201 };
3202
3203 static struct nfs4_openowner *
3204 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3205                            struct nfsd4_compound_state *cstate)
3206 {
3207         struct nfs4_client *clp = cstate->clp;
3208         struct nfs4_openowner *oo, *ret;
3209
3210         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3211         if (!oo)
3212                 return NULL;
3213         oo->oo_owner.so_ops = &openowner_ops;
3214         oo->oo_owner.so_is_open_owner = 1;
3215         oo->oo_owner.so_seqid = open->op_seqid;
3216         oo->oo_flags = 0;
3217         if (nfsd4_has_session(cstate))
3218                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3219         oo->oo_time = 0;
3220         oo->oo_last_closed_stid = NULL;
3221         INIT_LIST_HEAD(&oo->oo_close_lru);
3222         spin_lock(&clp->cl_lock);
3223         ret = find_openstateowner_str_locked(strhashval, open, clp);
3224         if (ret == NULL) {
3225                 hash_openowner(oo, clp, strhashval);
3226                 ret = oo;
3227         } else
3228                 nfs4_free_openowner(&oo->oo_owner);
3229         spin_unlock(&clp->cl_lock);
3230         return oo;
3231 }
3232
3233 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
3234         struct nfs4_openowner *oo = open->op_openowner;
3235
3236         atomic_inc(&stp->st_stid.sc_count);
3237         stp->st_stid.sc_type = NFS4_OPEN_STID;
3238         INIT_LIST_HEAD(&stp->st_locks);
3239         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3240         get_nfs4_file(fp);
3241         stp->st_stid.sc_file = fp;
3242         stp->st_access_bmap = 0;
3243         stp->st_deny_bmap = 0;
3244         stp->st_openstp = NULL;
3245         spin_lock(&oo->oo_owner.so_client->cl_lock);
3246         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3247         spin_lock(&fp->fi_lock);
3248         list_add(&stp->st_perfile, &fp->fi_stateids);
3249         spin_unlock(&fp->fi_lock);
3250         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3251 }
3252
3253 /*
3254  * In the 4.0 case we need to keep the owners around a little while to handle
3255  * CLOSE replay. We still do need to release any file access that is held by
3256  * them before returning however.
3257  */
3258 static void
3259 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3260 {
3261         struct nfs4_ol_stateid *last;
3262         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3263         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3264                                                 nfsd_net_id);
3265
3266         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3267
3268         /*
3269          * We know that we hold one reference via nfsd4_close, and another
3270          * "persistent" reference for the client. If the refcount is higher
3271          * than 2, then there are still calls in progress that are using this
3272          * stateid. We can't put the sc_file reference until they are finished.
3273          * Wait for the refcount to drop to 2. Since it has been unhashed,
3274          * there should be no danger of the refcount going back up again at
3275          * this point.
3276          */
3277         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3278
3279         release_all_access(s);
3280         if (s->st_stid.sc_file) {
3281                 put_nfs4_file(s->st_stid.sc_file);
3282                 s->st_stid.sc_file = NULL;
3283         }
3284
3285         spin_lock(&nn->client_lock);
3286         last = oo->oo_last_closed_stid;
3287         oo->oo_last_closed_stid = s;
3288         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3289         oo->oo_time = get_seconds();
3290         spin_unlock(&nn->client_lock);
3291         if (last)
3292                 nfs4_put_stid(&last->st_stid);
3293 }
3294
3295 /* search file_hashtbl[] for file */
3296 static struct nfs4_file *
3297 find_file_locked(struct knfsd_fh *fh)
3298 {
3299         unsigned int hashval = file_hashval(fh);
3300         struct nfs4_file *fp;
3301
3302         lockdep_assert_held(&state_lock);
3303
3304         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
3305                 if (nfsd_fh_match(&fp->fi_fhandle, fh)) {
3306                         get_nfs4_file(fp);
3307                         return fp;
3308                 }
3309         }
3310         return NULL;
3311 }
3312
3313 static struct nfs4_file *
3314 find_file(struct knfsd_fh *fh)
3315 {
3316         struct nfs4_file *fp;
3317
3318         spin_lock(&state_lock);
3319         fp = find_file_locked(fh);
3320         spin_unlock(&state_lock);
3321         return fp;
3322 }
3323
3324 static struct nfs4_file *
3325 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3326 {
3327         struct nfs4_file *fp;
3328
3329         spin_lock(&state_lock);
3330         fp = find_file_locked(fh);
3331         if (fp == NULL) {
3332                 nfsd4_init_file(new, fh);
3333                 fp = new;
3334         }
3335         spin_unlock(&state_lock);
3336
3337         return fp;
3338 }
3339
3340 /*
3341  * Called to check deny when READ with all zero stateid or
3342  * WRITE with all zero or all one stateid
3343  */
3344 static __be32
3345 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3346 {
3347         struct nfs4_file *fp;
3348         __be32 ret = nfs_ok;
3349
3350         fp = find_file(&current_fh->fh_handle);
3351         if (!fp)
3352                 return ret;
3353         /* Check for conflicting share reservations */
3354         spin_lock(&fp->fi_lock);
3355         if (fp->fi_share_deny & deny_type)
3356                 ret = nfserr_locked;
3357         spin_unlock(&fp->fi_lock);
3358         put_nfs4_file(fp);
3359         return ret;
3360 }
3361
3362 void nfsd4_prepare_cb_recall(struct nfs4_delegation *dp)
3363 {
3364         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3365                                           nfsd_net_id);
3366
3367         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3368
3369         /*
3370          * We can't do this in nfsd_break_deleg_cb because it is
3371          * already holding inode->i_lock.
3372          *
3373          * If the dl_time != 0, then we know that it has already been
3374          * queued for a lease break. Don't queue it again.
3375          */
3376         spin_lock(&state_lock);
3377         if (dp->dl_time == 0) {
3378                 dp->dl_time = get_seconds();
3379                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3380         }
3381         spin_unlock(&state_lock);
3382 }
3383
3384 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3385 {
3386         /*
3387          * We're assuming the state code never drops its reference
3388          * without first removing the lease.  Since we're in this lease
3389          * callback (and since the lease code is serialized by the kernel
3390          * lock) we know the server hasn't removed the lease yet, we know
3391          * it's safe to take a reference.
3392          */
3393         atomic_inc(&dp->dl_stid.sc_count);
3394         nfsd4_cb_recall(dp);
3395 }
3396
3397 /* Called from break_lease() with i_lock held. */
3398 static void nfsd_break_deleg_cb(struct file_lock *fl)
3399 {
3400         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3401         struct nfs4_delegation *dp;
3402
3403         if (!fp) {
3404                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3405                 return;
3406         }
3407         if (fp->fi_had_conflict) {
3408                 WARN(1, "duplicate break on %p\n", fp);
3409                 return;
3410         }
3411         /*
3412          * We don't want the locks code to timeout the lease for us;
3413          * we'll remove it ourself if a delegation isn't returned
3414          * in time:
3415          */
3416         fl->fl_break_time = 0;
3417
3418         spin_lock(&fp->fi_lock);
3419         fp->fi_had_conflict = true;
3420         /*
3421          * If there are no delegations on the list, then we can't count on this
3422          * lease ever being cleaned up. Set the fl_break_time to jiffies so that
3423          * time_out_leases will do it ASAP. The fact that fi_had_conflict is now
3424          * true should keep any new delegations from being hashed.
3425          */
3426         if (list_empty(&fp->fi_delegations))
3427                 fl->fl_break_time = jiffies;
3428         else
3429                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3430                         nfsd_break_one_deleg(dp);
3431         spin_unlock(&fp->fi_lock);
3432 }
3433
3434 static
3435 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
3436 {
3437         if (arg & F_UNLCK)
3438                 return lease_modify(onlist, arg);
3439         else
3440                 return -EAGAIN;
3441 }
3442
3443 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3444         .lm_break = nfsd_break_deleg_cb,
3445         .lm_change = nfsd_change_deleg_cb,
3446 };
3447
3448 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3449 {
3450         if (nfsd4_has_session(cstate))
3451                 return nfs_ok;
3452         if (seqid == so->so_seqid - 1)
3453                 return nfserr_replay_me;
3454         if (seqid == so->so_seqid)
3455                 return nfs_ok;
3456         return nfserr_bad_seqid;
3457 }
3458
3459 static __be32 lookup_clientid(clientid_t *clid,
3460                 struct nfsd4_compound_state *cstate,
3461                 struct nfsd_net *nn)
3462 {
3463         struct nfs4_client *found;
3464
3465         if (cstate->clp) {
3466                 found = cstate->clp;
3467                 if (!same_clid(&found->cl_clientid, clid))
3468                         return nfserr_stale_clientid;
3469                 return nfs_ok;
3470         }
3471
3472         if (STALE_CLIENTID(clid, nn))
3473                 return nfserr_stale_clientid;
3474
3475         /*
3476          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3477          * cached already then we know this is for is for v4.0 and "sessions"
3478          * will be false.
3479          */
3480         WARN_ON_ONCE(cstate->session);
3481         spin_lock(&nn->client_lock);
3482         found = find_confirmed_client(clid, false, nn);
3483         if (!found) {
3484                 spin_unlock(&nn->client_lock);
3485                 return nfserr_expired;
3486         }
3487         atomic_inc(&found->cl_refcount);
3488         spin_unlock(&nn->client_lock);
3489
3490         /* Cache the nfs4_client in cstate! */
3491         cstate->clp = found;
3492         return nfs_ok;
3493 }
3494
3495 __be32
3496 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3497                     struct nfsd4_open *open, struct nfsd_net *nn)
3498 {
3499         clientid_t *clientid = &open->op_clientid;
3500         struct nfs4_client *clp = NULL;
3501         unsigned int strhashval;
3502         struct nfs4_openowner *oo = NULL;
3503         __be32 status;
3504
3505         if (STALE_CLIENTID(&open->op_clientid, nn))
3506                 return nfserr_stale_clientid;
3507         /*
3508          * In case we need it later, after we've already created the
3509          * file and don't want to risk a further failure:
3510          */
3511         open->op_file = nfsd4_alloc_file();
3512         if (open->op_file == NULL)
3513                 return nfserr_jukebox;
3514
3515         status = lookup_clientid(clientid, cstate, nn);
3516         if (status)
3517                 return status;
3518         clp = cstate->clp;
3519
3520         strhashval = ownerstr_hashval(&open->op_owner);
3521         oo = find_openstateowner_str(strhashval, open, clp);
3522         open->op_openowner = oo;
3523         if (!oo) {
3524                 goto new_owner;
3525         }
3526         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3527                 /* Replace unconfirmed owners without checking for replay. */
3528                 release_openowner(oo);
3529                 open->op_openowner = NULL;
3530                 goto new_owner;
3531         }
3532         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3533         if (status)
3534                 return status;
3535         goto alloc_stateid;
3536 new_owner:
3537         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3538         if (oo == NULL)
3539                 return nfserr_jukebox;
3540         open->op_openowner = oo;
3541 alloc_stateid:
3542         open->op_stp = nfs4_alloc_open_stateid(clp);
3543         if (!open->op_stp)
3544                 return nfserr_jukebox;
3545         return nfs_ok;
3546 }
3547
3548 static inline __be32
3549 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3550 {
3551         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3552                 return nfserr_openmode;
3553         else
3554                 return nfs_ok;
3555 }
3556
3557 static int share_access_to_flags(u32 share_access)
3558 {
3559         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3560 }
3561
3562 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3563 {
3564         struct nfs4_stid *ret;
3565
3566         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3567         if (!ret)
3568                 return NULL;
3569         return delegstateid(ret);
3570 }
3571
3572 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3573 {
3574         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3575                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3576 }
3577
3578 static __be32
3579 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3580                 struct nfs4_delegation **dp)
3581 {
3582         int flags;
3583         __be32 status = nfserr_bad_stateid;
3584         struct nfs4_delegation *deleg;
3585
3586         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3587         if (deleg == NULL)
3588                 goto out;
3589         flags = share_access_to_flags(open->op_share_access);
3590         status = nfs4_check_delegmode(deleg, flags);
3591         if (status) {
3592                 nfs4_put_stid(&deleg->dl_stid);
3593                 goto out;
3594         }
3595         *dp = deleg;
3596 out:
3597         if (!nfsd4_is_deleg_cur(open))
3598                 return nfs_ok;
3599         if (status)
3600                 return status;
3601         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3602         return nfs_ok;
3603 }
3604
3605 static struct nfs4_ol_stateid *
3606 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3607 {
3608         struct nfs4_ol_stateid *local, *ret = NULL;
3609         struct nfs4_openowner *oo = open->op_openowner;
3610
3611         spin_lock(&fp->fi_lock);
3612         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3613                 /* ignore lock owners */
3614                 if (local->st_stateowner->so_is_open_owner == 0)
3615                         continue;
3616                 if (local->st_stateowner == &oo->oo_owner) {
3617                         ret = local;
3618                         atomic_inc(&ret->st_stid.sc_count);
3619                         break;
3620                 }
3621         }
3622         spin_unlock(&fp->fi_lock);
3623         return ret;
3624 }
3625
3626 static inline int nfs4_access_to_access(u32 nfs4_access)
3627 {
3628         int flags = 0;
3629
3630         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3631                 flags |= NFSD_MAY_READ;
3632         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3633                 flags |= NFSD_MAY_WRITE;
3634         return flags;
3635 }
3636
3637 static inline __be32
3638 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3639                 struct nfsd4_open *open)
3640 {
3641         struct iattr iattr = {
3642                 .ia_valid = ATTR_SIZE,
3643                 .ia_size = 0,
3644         };
3645         if (!open->op_truncate)
3646                 return 0;
3647         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3648                 return nfserr_inval;
3649         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3650 }
3651
3652 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3653                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3654                 struct nfsd4_open *open)
3655 {
3656         struct file *filp = NULL;
3657         __be32 status;
3658         int oflag = nfs4_access_to_omode(open->op_share_access);
3659         int access = nfs4_access_to_access(open->op_share_access);
3660         unsigned char old_access_bmap, old_deny_bmap;
3661
3662         spin_lock(&fp->fi_lock);
3663
3664         /*
3665          * Are we trying to set a deny mode that would conflict with
3666          * current access?
3667          */
3668         status = nfs4_file_check_deny(fp, open->op_share_deny);
3669         if (status != nfs_ok) {
3670                 spin_unlock(&fp->fi_lock);
3671                 goto out;
3672         }
3673
3674         /* set access to the file */
3675         status = nfs4_file_get_access(fp, open->op_share_access);
3676         if (status != nfs_ok) {
3677                 spin_unlock(&fp->fi_lock);
3678                 goto out;
3679         }
3680
3681         /* Set access bits in stateid */
3682         old_access_bmap = stp->st_access_bmap;
3683         set_access(open->op_share_access, stp);
3684
3685         /* Set new deny mask */
3686         old_deny_bmap = stp->st_deny_bmap;
3687         set_deny(open->op_share_deny, stp);
3688         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3689
3690         if (!fp->fi_fds[oflag]) {
3691                 spin_unlock(&fp->fi_lock);
3692                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3693                 if (status)
3694                         goto out_put_access;
3695                 spin_lock(&fp->fi_lock);
3696                 if (!fp->fi_fds[oflag]) {
3697                         fp->fi_fds[oflag] = filp;
3698                         filp = NULL;
3699                 }
3700         }
3701         spin_unlock(&fp->fi_lock);
3702         if (filp)
3703                 fput(filp);
3704
3705         status = nfsd4_truncate(rqstp, cur_fh, open);
3706         if (status)
3707                 goto out_put_access;
3708 out:
3709         return status;
3710 out_put_access:
3711         stp->st_access_bmap = old_access_bmap;
3712         nfs4_file_put_access(fp, open->op_share_access);
3713         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3714         goto out;
3715 }
3716
3717 static __be32
3718 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)
3719 {
3720         __be32 status;
3721         unsigned char old_deny_bmap;
3722
3723         if (!test_access(open->op_share_access, stp))
3724                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3725
3726         /* test and set deny mode */
3727         spin_lock(&fp->fi_lock);
3728         status = nfs4_file_check_deny(fp, open->op_share_deny);
3729         if (status == nfs_ok) {
3730                 old_deny_bmap = stp->st_deny_bmap;
3731                 set_deny(open->op_share_deny, stp);
3732                 fp->fi_share_deny |=
3733                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3734         }
3735         spin_unlock(&fp->fi_lock);
3736
3737         if (status != nfs_ok)
3738                 return status;
3739
3740         status = nfsd4_truncate(rqstp, cur_fh, open);
3741         if (status != nfs_ok)
3742                 reset_union_bmap_deny(old_deny_bmap, stp);
3743         return status;
3744 }
3745
3746 static void
3747 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3748 {
3749         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3750 }
3751
3752 /* Should we give out recallable state?: */
3753 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3754 {
3755         if (clp->cl_cb_state == NFSD4_CB_UP)
3756                 return true;
3757         /*
3758          * In the sessions case, since we don't have to establish a
3759          * separate connection for callbacks, we assume it's OK
3760          * until we hear otherwise:
3761          */
3762         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3763 }
3764
3765 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
3766 {
3767         struct file_lock *fl;
3768
3769         fl = locks_alloc_lock();
3770         if (!fl)
3771                 return NULL;
3772         fl->fl_lmops = &nfsd_lease_mng_ops;
3773         fl->fl_flags = FL_DELEG;
3774         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3775         fl->fl_end = OFFSET_MAX;
3776         fl->fl_owner = (fl_owner_t)fp;
3777         fl->fl_pid = current->tgid;
3778         return fl;
3779 }
3780
3781 static int nfs4_setlease(struct nfs4_delegation *dp)
3782 {
3783         struct nfs4_file *fp = dp->dl_stid.sc_file;
3784         struct file_lock *fl;
3785         struct file *filp;
3786         int status = 0;
3787
3788         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
3789         if (!fl)
3790                 return -ENOMEM;
3791         filp = find_readable_file(fp);
3792         if (!filp) {
3793                 /* We should always have a readable file here */
3794                 WARN_ON_ONCE(1);
3795                 return -EBADF;
3796         }
3797         fl->fl_file = filp;
3798         status = vfs_setlease(filp, fl->fl_type, &fl);
3799         if (status) {
3800                 locks_free_lock(fl);
3801                 goto out_fput;
3802         }
3803         spin_lock(&state_lock);
3804         spin_lock(&fp->fi_lock);
3805         /* Did the lease get broken before we took the lock? */
3806         status = -EAGAIN;
3807         if (fp->fi_had_conflict)
3808                 goto out_unlock;
3809         /* Race breaker */
3810         if (fp->fi_lease) {
3811                 status = 0;
3812                 atomic_inc(&fp->fi_delegees);
3813                 hash_delegation_locked(dp, fp);
3814                 goto out_unlock;
3815         }
3816         fp->fi_lease = fl;
3817         fp->fi_deleg_file = filp;
3818         atomic_set(&fp->fi_delegees, 1);
3819         hash_delegation_locked(dp, fp);
3820         spin_unlock(&fp->fi_lock);
3821         spin_unlock(&state_lock);
3822         return 0;
3823 out_unlock:
3824         spin_unlock(&fp->fi_lock);
3825         spin_unlock(&state_lock);
3826 out_fput:
3827         fput(filp);
3828         return status;
3829 }
3830
3831 static struct nfs4_delegation *
3832 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
3833                     struct nfs4_file *fp)
3834 {
3835         int status;
3836         struct nfs4_delegation *dp;
3837
3838         if (fp->fi_had_conflict)
3839                 return ERR_PTR(-EAGAIN);
3840
3841         dp = alloc_init_deleg(clp, fh);
3842         if (!dp)
3843                 return ERR_PTR(-ENOMEM);
3844
3845         get_nfs4_file(fp);
3846         spin_lock(&state_lock);
3847         spin_lock(&fp->fi_lock);
3848         dp->dl_stid.sc_file = fp;
3849         if (!fp->fi_lease) {
3850                 spin_unlock(&fp->fi_lock);
3851                 spin_unlock(&state_lock);
3852                 status = nfs4_setlease(dp);
3853                 goto out;
3854         }
3855         atomic_inc(&fp->fi_delegees);
3856         if (fp->fi_had_conflict) {
3857                 status = -EAGAIN;
3858                 goto out_unlock;
3859         }
3860         hash_delegation_locked(dp, fp);
3861         status = 0;
3862 out_unlock:
3863         spin_unlock(&fp->fi_lock);
3864         spin_unlock(&state_lock);
3865 out:
3866         if (status) {
3867                 nfs4_put_stid(&dp->dl_stid);
3868                 return ERR_PTR(status);
3869         }
3870         return dp;
3871 }
3872
3873 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3874 {
3875         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3876         if (status == -EAGAIN)
3877                 open->op_why_no_deleg = WND4_CONTENTION;
3878         else {
3879                 open->op_why_no_deleg = WND4_RESOURCE;
3880                 switch (open->op_deleg_want) {
3881                 case NFS4_SHARE_WANT_READ_DELEG:
3882                 case NFS4_SHARE_WANT_WRITE_DELEG:
3883                 case NFS4_SHARE_WANT_ANY_DELEG:
3884                         break;
3885                 case NFS4_SHARE_WANT_CANCEL:
3886                         open->op_why_no_deleg = WND4_CANCELLED;
3887                         break;
3888                 case NFS4_SHARE_WANT_NO_DELEG:
3889                         WARN_ON_ONCE(1);
3890                 }
3891         }
3892 }
3893
3894 /*
3895  * Attempt to hand out a delegation.
3896  *
3897  * Note we don't support write delegations, and won't until the vfs has
3898  * proper support for them.
3899  */
3900 static void
3901 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
3902                         struct nfs4_ol_stateid *stp)
3903 {
3904         struct nfs4_delegation *dp;
3905         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
3906         struct nfs4_client *clp = stp->st_stid.sc_client;
3907         int cb_up;
3908         int status = 0;
3909
3910         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3911         open->op_recall = 0;
3912         switch (open->op_claim_type) {
3913                 case NFS4_OPEN_CLAIM_PREVIOUS:
3914                         if (!cb_up)
3915                                 open->op_recall = 1;
3916                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3917                                 goto out_no_deleg;
3918                         break;
3919                 case NFS4_OPEN_CLAIM_NULL:
3920                 case NFS4_OPEN_CLAIM_FH:
3921                         /*
3922                          * Let's not give out any delegations till everyone's
3923                          * had the chance to reclaim theirs....
3924                          */
3925                         if (locks_in_grace(clp->net))
3926                                 goto out_no_deleg;
3927                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3928                                 goto out_no_deleg;
3929                         /*
3930                          * Also, if the file was opened for write or
3931                          * create, there's a good chance the client's
3932                          * about to write to it, resulting in an
3933                          * immediate recall (since we don't support
3934                          * write delegations):
3935                          */
3936                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3937                                 goto out_no_deleg;
3938                         if (open->op_create == NFS4_OPEN_CREATE)
3939                                 goto out_no_deleg;
3940                         break;
3941                 default:
3942                         goto out_no_deleg;
3943         }
3944         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file);
3945         if (IS_ERR(dp))
3946                 goto out_no_deleg;
3947
3948         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3949
3950         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3951                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3952         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3953         nfs4_put_stid(&dp->dl_stid);
3954         return;
3955 out_no_deleg:
3956         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3957         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3958             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3959                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3960                 open->op_recall = 1;
3961         }
3962
3963         /* 4.1 client asking for a delegation? */
3964         if (open->op_deleg_want)
3965                 nfsd4_open_deleg_none_ext(open, status);
3966         return;
3967 }
3968
3969 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3970                                         struct nfs4_delegation *dp)
3971 {
3972         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3973             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3974                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3975                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3976         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3977                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3978                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3979                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3980         }
3981         /* Otherwise the client must be confused wanting a delegation
3982          * it already has, therefore we don't return
3983          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3984          */
3985 }
3986
3987 __be32
3988 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3989 {
3990         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3991         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3992         struct nfs4_file *fp = NULL;
3993         struct nfs4_ol_stateid *stp = NULL;
3994         struct nfs4_delegation *dp = NULL;
3995         __be32 status;
3996
3997         /*
3998          * Lookup file; if found, lookup stateid and check open request,
3999          * and check for delegations in the process of being recalled.
4000          * If not found, create the nfs4_file struct
4001          */
4002         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4003         if (fp != open->op_file) {
4004                 status = nfs4_check_deleg(cl, open, &dp);
4005                 if (status)
4006                         goto out;
4007                 stp = nfsd4_find_existing_open(fp, open);
4008         } else {
4009                 open->op_file = NULL;
4010                 status = nfserr_bad_stateid;
4011                 if (nfsd4_is_deleg_cur(open))
4012                         goto out;
4013                 status = nfserr_jukebox;
4014         }
4015
4016         /*
4017          * OPEN the file, or upgrade an existing OPEN.
4018          * If truncate fails, the OPEN fails.
4019          */
4020         if (stp) {
4021                 /* Stateid was found, this is an OPEN upgrade */
4022                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4023                 if (status)
4024                         goto out;
4025         } else {
4026                 stp = open->op_stp;
4027                 open->op_stp = NULL;
4028                 init_open_stateid(stp, fp, open);
4029                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4030                 if (status) {
4031                         release_open_stateid(stp);
4032                         goto out;
4033                 }
4034         }
4035         update_stateid(&stp->st_stid.sc_stateid);
4036         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4037
4038         if (nfsd4_has_session(&resp->cstate)) {
4039                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4040                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4041                         open->op_why_no_deleg = WND4_NOT_WANTED;
4042                         goto nodeleg;
4043                 }
4044         }
4045
4046         /*
4047         * Attempt to hand out a delegation. No error return, because the
4048         * OPEN succeeds even if we fail.
4049         */
4050         nfs4_open_delegation(current_fh, open, stp);
4051 nodeleg:
4052         status = nfs_ok;
4053
4054         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4055                 STATEID_VAL(&stp->st_stid.sc_stateid));
4056 out:
4057         /* 4.1 client trying to upgrade/downgrade delegation? */
4058         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4059             open->op_deleg_want)
4060                 nfsd4_deleg_xgrade_none_ext(open, dp);
4061
4062         if (fp)
4063                 put_nfs4_file(fp);
4064         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4065                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
4066         /*
4067         * To finish the open response, we just need to set the rflags.
4068         */
4069         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4070         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4071             !nfsd4_has_session(&resp->cstate))
4072                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4073         if (dp)
4074                 nfs4_put_stid(&dp->dl_stid);
4075         if (stp)
4076                 nfs4_put_stid(&stp->st_stid);
4077
4078         return status;
4079 }
4080
4081 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4082                               struct nfsd4_open *open, __be32 status)
4083 {
4084         if (open->op_openowner) {
4085                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4086
4087                 nfsd4_cstate_assign_replay(cstate, so);
4088                 nfs4_put_stateowner(so);
4089         }
4090         if (open->op_file)
4091                 nfsd4_free_file(open->op_file);
4092         if (open->op_stp)
4093                 nfs4_put_stid(&open->op_stp->st_stid);
4094 }
4095
4096 __be32
4097 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4098             clientid_t *clid)
4099 {
4100         struct nfs4_client *clp;
4101         __be32 status;
4102         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4103
4104         dprintk("process_renew(%08x/%08x): starting\n",
4105                         clid->cl_boot, clid->cl_id);
4106         status = lookup_clientid(clid, cstate, nn);
4107         if (status)
4108                 goto out;
4109         clp = cstate->clp;
4110         status = nfserr_cb_path_down;
4111         if (!list_empty(&clp->cl_delegations)
4112                         && clp->cl_cb_state != NFSD4_CB_UP)
4113                 goto out;
4114         status = nfs_ok;
4115 out:
4116         return status;
4117 }
4118
4119 static void
4120 nfsd4_end_grace(struct nfsd_net *nn)
4121 {
4122         /* do nothing if grace period already ended */
4123         if (nn->grace_ended)
4124                 return;
4125
4126         dprintk("NFSD: end of grace period\n");
4127         nn->grace_ended = true;
4128         nfsd4_record_grace_done(nn, nn->boot_time);
4129         locks_end_grace(&nn->nfsd4_manager);
4130         /*
4131          * Now that every NFSv4 client has had the chance to recover and
4132          * to see the (possibly new, possibly shorter) lease time, we
4133          * can safely set the next grace time to the current lease time:
4134          */
4135         nn->nfsd4_grace = nn->nfsd4_lease;
4136 }
4137
4138 static time_t
4139 nfs4_laundromat(struct nfsd_net *nn)
4140 {
4141         struct nfs4_client *clp;
4142         struct nfs4_openowner *oo;
4143         struct nfs4_delegation *dp;
4144         struct nfs4_ol_stateid *stp;
4145         struct list_head *pos, *next, reaplist;
4146         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4147         time_t t, new_timeo = nn->nfsd4_lease;
4148
4149         dprintk("NFSD: laundromat service - starting\n");
4150         nfsd4_end_grace(nn);
4151         INIT_LIST_HEAD(&reaplist);
4152         spin_lock(&nn->client_lock);
4153         list_for_each_safe(pos, next, &nn->client_lru) {
4154                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4155                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4156                         t = clp->cl_time - cutoff;
4157                         new_timeo = min(new_timeo, t);
4158                         break;
4159                 }
4160                 if (mark_client_expired_locked(clp)) {
4161                         dprintk("NFSD: client in use (clientid %08x)\n",
4162                                 clp->cl_clientid.cl_id);
4163                         continue;
4164                 }
4165                 list_add(&clp->cl_lru, &reaplist);
4166         }
4167         spin_unlock(&nn->client_lock);
4168         list_for_each_safe(pos, next, &reaplist) {
4169                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4170                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4171                         clp->cl_clientid.cl_id);
4172                 list_del_init(&clp->cl_lru);
4173                 expire_client(clp);
4174         }
4175         spin_lock(&state_lock);
4176         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4177                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4178                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
4179                         continue;
4180                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4181                         t = dp->dl_time - cutoff;
4182                         new_timeo = min(new_timeo, t);
4183                         break;
4184                 }
4185                 unhash_delegation_locked(dp);
4186                 list_add(&dp->dl_recall_lru, &reaplist);
4187         }
4188         spin_unlock(&state_lock);
4189         while (!list_empty(&reaplist)) {
4190                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4191                                         dl_recall_lru);
4192                 list_del_init(&dp->dl_recall_lru);
4193                 revoke_delegation(dp);
4194         }
4195
4196         spin_lock(&nn->client_lock);
4197         while (!list_empty(&nn->close_lru)) {
4198                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4199                                         oo_close_lru);
4200                 if (time_after((unsigned long)oo->oo_time,
4201                                (unsigned long)cutoff)) {
4202                         t = oo->oo_time - cutoff;
4203                         new_timeo = min(new_timeo, t);
4204                         break;
4205                 }
4206                 list_del_init(&oo->oo_close_lru);
4207                 stp = oo->oo_last_closed_stid;
4208                 oo->oo_last_closed_stid = NULL;
4209                 spin_unlock(&nn->client_lock);
4210                 nfs4_put_stid(&stp->st_stid);
4211                 spin_lock(&nn->client_lock);
4212         }
4213         spin_unlock(&nn->client_lock);
4214
4215         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4216         return new_timeo;
4217 }
4218
4219 static struct workqueue_struct *laundry_wq;
4220 static void laundromat_main(struct work_struct *);
4221
4222 static void
4223 laundromat_main(struct work_struct *laundry)
4224 {
4225         time_t t;
4226         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
4227                                                   work);
4228         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4229                                            laundromat_work);
4230
4231         t = nfs4_laundromat(nn);
4232         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4233         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4234 }
4235
4236 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
4237 {
4238         if (!nfsd_fh_match(&fhp->fh_handle, &stp->st_stid.sc_file->fi_fhandle))
4239                 return nfserr_bad_stateid;
4240         return nfs_ok;
4241 }
4242
4243 static inline int
4244 access_permit_read(struct nfs4_ol_stateid *stp)
4245 {
4246         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4247                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4248                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4249 }
4250
4251 static inline int
4252 access_permit_write(struct nfs4_ol_stateid *stp)
4253 {
4254         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4255                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4256 }
4257
4258 static
4259 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4260 {
4261         __be32 status = nfserr_openmode;
4262
4263         /* For lock stateid's, we test the parent open, not the lock: */
4264         if (stp->st_openstp)
4265                 stp = stp->st_openstp;
4266         if ((flags & WR_STATE) && !access_permit_write(stp))
4267                 goto out;
4268         if ((flags & RD_STATE) && !access_permit_read(stp))
4269                 goto out;
4270         status = nfs_ok;
4271 out:
4272         return status;
4273 }
4274
4275 static inline __be32
4276 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4277 {
4278         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4279                 return nfs_ok;
4280         else if (locks_in_grace(net)) {
4281                 /* Answer in remaining cases depends on existence of
4282                  * conflicting state; so we must wait out the grace period. */
4283                 return nfserr_grace;
4284         } else if (flags & WR_STATE)
4285                 return nfs4_share_conflict(current_fh,
4286                                 NFS4_SHARE_DENY_WRITE);
4287         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4288                 return nfs4_share_conflict(current_fh,
4289                                 NFS4_SHARE_DENY_READ);
4290 }
4291
4292 /*
4293  * Allow READ/WRITE during grace period on recovered state only for files
4294  * that are not able to provide mandatory locking.
4295  */
4296 static inline int
4297 grace_disallows_io(struct net *net, struct inode *inode)
4298 {
4299         return locks_in_grace(net) && mandatory_lock(inode);
4300 }
4301
4302 /* Returns true iff a is later than b: */
4303 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4304 {
4305         return (s32)(a->si_generation - b->si_generation) > 0;
4306 }
4307
4308 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4309 {
4310         /*
4311          * When sessions are used the stateid generation number is ignored
4312          * when it is zero.
4313          */
4314         if (has_session && in->si_generation == 0)
4315                 return nfs_ok;
4316
4317         if (in->si_generation == ref->si_generation)
4318                 return nfs_ok;
4319
4320         /* If the client sends us a stateid from the future, it's buggy: */
4321         if (stateid_generation_after(in, ref))
4322                 return nfserr_bad_stateid;
4323         /*
4324          * However, we could see a stateid from the past, even from a
4325          * non-buggy client.  For example, if the client sends a lock
4326          * while some IO is outstanding, the lock may bump si_generation
4327          * while the IO is still in flight.  The client could avoid that
4328          * situation by waiting for responses on all the IO requests,
4329          * but better performance may result in retrying IO that
4330          * receives an old_stateid error if requests are rarely
4331          * reordered in flight:
4332          */
4333         return nfserr_old_stateid;
4334 }
4335
4336 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4337 {
4338         struct nfs4_stid *s;
4339         struct nfs4_ol_stateid *ols;
4340         __be32 status = nfserr_bad_stateid;
4341
4342         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4343                 return status;
4344         /* Client debugging aid. */
4345         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4346                 char addr_str[INET6_ADDRSTRLEN];
4347                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4348                                  sizeof(addr_str));
4349                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4350                                         "with incorrect client ID\n", addr_str);
4351                 return status;
4352         }
4353         spin_lock(&cl->cl_lock);
4354         s = find_stateid_locked(cl, stateid);
4355         if (!s)
4356                 goto out_unlock;
4357         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4358         if (status)
4359                 goto out_unlock;
4360         switch (s->sc_type) {
4361         case NFS4_DELEG_STID:
4362                 status = nfs_ok;
4363                 break;
4364         case NFS4_REVOKED_DELEG_STID:
4365                 status = nfserr_deleg_revoked;
4366                 break;
4367         case NFS4_OPEN_STID:
4368         case NFS4_LOCK_STID:
4369                 ols = openlockstateid(s);
4370                 if (ols->st_stateowner->so_is_open_owner
4371                                 && !(openowner(ols->st_stateowner)->oo_flags
4372                                                 & NFS4_OO_CONFIRMED))
4373                         status = nfserr_bad_stateid;
4374                 else
4375                         status = nfs_ok;
4376                 break;
4377         default:
4378                 printk("unknown stateid type %x\n", s->sc_type);
4379                 /* Fallthrough */
4380         case NFS4_CLOSED_STID:
4381         case NFS4_CLOSED_DELEG_STID:
4382                 status = nfserr_bad_stateid;
4383         }
4384 out_unlock:
4385         spin_unlock(&cl->cl_lock);
4386         return status;
4387 }
4388
4389 static __be32
4390 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4391                      stateid_t *stateid, unsigned char typemask,
4392                      struct nfs4_stid **s, struct nfsd_net *nn)
4393 {
4394         __be32 status;
4395
4396         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4397                 return nfserr_bad_stateid;
4398         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4399         if (status == nfserr_stale_clientid) {
4400                 if (cstate->session)
4401                         return nfserr_bad_stateid;
4402                 return nfserr_stale_stateid;
4403         }
4404         if (status)
4405                 return status;
4406         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4407         if (!*s)
4408                 return nfserr_bad_stateid;
4409         return nfs_ok;
4410 }
4411
4412 /*
4413 * Checks for stateid operations
4414 */
4415 __be32
4416 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
4417                            stateid_t *stateid, int flags, struct file **filpp)
4418 {
4419         struct nfs4_stid *s;
4420         struct nfs4_ol_stateid *stp = NULL;
4421         struct nfs4_delegation *dp = NULL;
4422         struct svc_fh *current_fh = &cstate->current_fh;
4423         struct inode *ino = current_fh->fh_dentry->d_inode;
4424         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4425         struct file *file = NULL;
4426         __be32 status;
4427
4428         if (filpp)
4429                 *filpp = NULL;
4430
4431         if (grace_disallows_io(net, ino))
4432                 return nfserr_grace;
4433
4434         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4435                 return check_special_stateids(net, current_fh, stateid, flags);
4436
4437         status = nfsd4_lookup_stateid(cstate, stateid,
4438                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4439                                 &s, nn);
4440         if (status)
4441                 return status;
4442         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
4443         if (status)
4444                 goto out;
4445         switch (s->sc_type) {
4446         case NFS4_DELEG_STID:
4447                 dp = delegstateid(s);
4448                 status = nfs4_check_delegmode(dp, flags);
4449                 if (status)
4450                         goto out;
4451                 if (filpp) {
4452                         file = dp->dl_stid.sc_file->fi_deleg_file;
4453                         if (!file) {
4454                                 WARN_ON_ONCE(1);
4455                                 status = nfserr_serverfault;
4456                                 goto out;
4457                         }
4458                         get_file(file);
4459                 }
4460                 break;
4461         case NFS4_OPEN_STID:
4462         case NFS4_LOCK_STID:
4463                 stp = openlockstateid(s);
4464                 status = nfs4_check_fh(current_fh, stp);
4465                 if (status)
4466                         goto out;
4467                 if (stp->st_stateowner->so_is_open_owner
4468                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4469                         goto out;
4470                 status = nfs4_check_openmode(stp, flags);
4471                 if (status)
4472                         goto out;
4473                 if (filpp) {
4474                         struct nfs4_file *fp = stp->st_stid.sc_file;
4475
4476                         if (flags & RD_STATE)
4477                                 file = find_readable_file(fp);
4478                         else
4479                                 file = find_writeable_file(fp);
4480                 }
4481                 break;
4482         default:
4483                 status = nfserr_bad_stateid;
4484                 goto out;
4485         }
4486         status = nfs_ok;
4487         if (file)
4488                 *filpp = file;
4489 out:
4490         nfs4_put_stid(s);
4491         return status;
4492 }
4493
4494 /*
4495  * Test if the stateid is valid
4496  */
4497 __be32
4498 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4499                    struct nfsd4_test_stateid *test_stateid)
4500 {
4501         struct nfsd4_test_stateid_id *stateid;
4502         struct nfs4_client *cl = cstate->session->se_client;
4503
4504         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4505                 stateid->ts_id_status =
4506                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4507
4508         return nfs_ok;
4509 }
4510
4511 __be32
4512 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4513                    struct nfsd4_free_stateid *free_stateid)
4514 {
4515         stateid_t *stateid = &free_stateid->fr_stateid;
4516         struct nfs4_stid *s;
4517         struct nfs4_delegation *dp;
4518         struct nfs4_ol_stateid *stp;
4519         struct nfs4_client *cl = cstate->session->se_client;
4520         __be32 ret = nfserr_bad_stateid;
4521
4522         spin_lock(&cl->cl_lock);
4523         s = find_stateid_locked(cl, stateid);
4524         if (!s)
4525                 goto out_unlock;
4526         switch (s->sc_type) {
4527         case NFS4_DELEG_STID:
4528                 ret = nfserr_locks_held;
4529                 break;
4530         case NFS4_OPEN_STID:
4531                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4532                 if (ret)
4533                         break;
4534                 ret = nfserr_locks_held;
4535                 break;
4536         case NFS4_LOCK_STID:
4537                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4538                 if (ret)
4539                         break;
4540                 stp = openlockstateid(s);
4541                 ret = nfserr_locks_held;
4542                 if (check_for_locks(stp->st_stid.sc_file,
4543                                     lockowner(stp->st_stateowner)))
4544                         break;
4545                 unhash_lock_stateid(stp);
4546                 spin_unlock(&cl->cl_lock);
4547                 nfs4_put_stid(s);
4548                 ret = nfs_ok;
4549                 goto out;
4550         case NFS4_REVOKED_DELEG_STID:
4551                 dp = delegstateid(s);
4552                 list_del_init(&dp->dl_recall_lru);
4553                 spin_unlock(&cl->cl_lock);
4554                 nfs4_put_stid(s);
4555                 ret = nfs_ok;
4556                 goto out;
4557         /* Default falls through and returns nfserr_bad_stateid */
4558         }
4559 out_unlock:
4560         spin_unlock(&cl->cl_lock);
4561 out:
4562         return ret;
4563 }
4564
4565 static inline int
4566 setlkflg (int type)
4567 {
4568         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4569                 RD_STATE : WR_STATE;
4570 }
4571
4572 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4573 {
4574         struct svc_fh *current_fh = &cstate->current_fh;
4575         struct nfs4_stateowner *sop = stp->st_stateowner;
4576         __be32 status;
4577
4578         status = nfsd4_check_seqid(cstate, sop, seqid);
4579         if (status)
4580                 return status;
4581         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4582                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4583                 /*
4584                  * "Closed" stateid's exist *only* to return
4585                  * nfserr_replay_me from the previous step, and
4586                  * revoked delegations are kept only for free_stateid.
4587                  */
4588                 return nfserr_bad_stateid;
4589         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4590         if (status)
4591                 return status;
4592         return nfs4_check_fh(current_fh, stp);
4593 }
4594
4595 /*
4596  * Checks for sequence id mutating operations.
4597  */
4598 static __be32
4599 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4600                          stateid_t *stateid, char typemask,
4601                          struct nfs4_ol_stateid **stpp,
4602                          struct nfsd_net *nn)
4603 {
4604         __be32 status;
4605         struct nfs4_stid *s;
4606         struct nfs4_ol_stateid *stp = NULL;
4607
4608         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4609                 seqid, STATEID_VAL(stateid));
4610
4611         *stpp = NULL;
4612         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4613         if (status)
4614                 return status;
4615         stp = openlockstateid(s);
4616         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
4617
4618         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4619         if (!status)
4620                 *stpp = stp;
4621         else
4622                 nfs4_put_stid(&stp->st_stid);
4623         return status;
4624 }
4625
4626 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4627                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4628 {
4629         __be32 status;
4630         struct nfs4_openowner *oo;
4631         struct nfs4_ol_stateid *stp;
4632
4633         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4634                                                 NFS4_OPEN_STID, &stp, nn);
4635         if (status)
4636                 return status;
4637         oo = openowner(stp->st_stateowner);
4638         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4639                 nfs4_put_stid(&stp->st_stid);
4640                 return nfserr_bad_stateid;
4641         }
4642         *stpp = stp;
4643         return nfs_ok;
4644 }
4645
4646 __be32
4647 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4648                    struct nfsd4_open_confirm *oc)
4649 {
4650         __be32 status;
4651         struct nfs4_openowner *oo;
4652         struct nfs4_ol_stateid *stp;
4653         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4654
4655         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4656                         cstate->current_fh.fh_dentry);
4657
4658         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4659         if (status)
4660                 return status;
4661
4662         status = nfs4_preprocess_seqid_op(cstate,
4663                                         oc->oc_seqid, &oc->oc_req_stateid,
4664                                         NFS4_OPEN_STID, &stp, nn);
4665         if (status)
4666                 goto out;
4667         oo = openowner(stp->st_stateowner);
4668         status = nfserr_bad_stateid;
4669         if (oo->oo_flags & NFS4_OO_CONFIRMED)
4670                 goto put_stateid;
4671         oo->oo_flags |= NFS4_OO_CONFIRMED;
4672         update_stateid(&stp->st_stid.sc_stateid);
4673         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4674         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
4675                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
4676
4677         nfsd4_client_record_create(oo->oo_owner.so_client);
4678         status = nfs_ok;
4679 put_stateid:
4680         nfs4_put_stid(&stp->st_stid);
4681 out:
4682         nfsd4_bump_seqid(cstate, status);
4683         return status;
4684 }
4685
4686 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
4687 {
4688         if (!test_access(access, stp))
4689                 return;
4690         nfs4_file_put_access(stp->st_stid.sc_file, access);
4691         clear_access(access, stp);
4692 }
4693
4694 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
4695 {
4696         switch (to_access) {
4697         case NFS4_SHARE_ACCESS_READ:
4698                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
4699                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4700                 break;
4701         case NFS4_SHARE_ACCESS_WRITE:
4702                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
4703                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4704                 break;
4705         case NFS4_SHARE_ACCESS_BOTH:
4706                 break;
4707         default:
4708                 WARN_ON_ONCE(1);
4709         }
4710 }
4711
4712 __be32
4713 nfsd4_open_downgrade(struct svc_rqst *rqstp,
4714                      struct nfsd4_compound_state *cstate,
4715                      struct nfsd4_open_downgrade *od)
4716 {
4717         __be32 status;
4718         struct nfs4_ol_stateid *stp;
4719         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4720
4721         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
4722                         cstate->current_fh.fh_dentry);
4723
4724         /* We don't yet support WANT bits: */
4725         if (od->od_deleg_want)
4726                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
4727                         od->od_deleg_want);
4728
4729         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
4730                                         &od->od_stateid, &stp, nn);
4731         if (status)
4732                 goto out;
4733         status = nfserr_inval;
4734         if (!test_access(od->od_share_access, stp)) {
4735                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
4736                         stp->st_access_bmap, od->od_share_access);
4737                 goto put_stateid;
4738         }
4739         if (!test_deny(od->od_share_deny, stp)) {
4740                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
4741                         stp->st_deny_bmap, od->od_share_deny);
4742                 goto put_stateid;
4743         }
4744         nfs4_stateid_downgrade(stp, od->od_share_access);
4745
4746         reset_union_bmap_deny(od->od_share_deny, stp);
4747
4748         update_stateid(&stp->st_stid.sc_stateid);
4749         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4750         status = nfs_ok;
4751 put_stateid:
4752         nfs4_put_stid(&stp->st_stid);
4753 out:
4754         nfsd4_bump_seqid(cstate, status);
4755         return status;
4756 }
4757
4758 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
4759 {
4760         struct nfs4_client *clp = s->st_stid.sc_client;
4761         LIST_HEAD(reaplist);
4762
4763         s->st_stid.sc_type = NFS4_CLOSED_STID;
4764         spin_lock(&clp->cl_lock);
4765         unhash_open_stateid(s, &reaplist);
4766
4767         if (clp->cl_minorversion) {
4768                 put_ol_stateid_locked(s, &reaplist);
4769                 spin_unlock(&clp->cl_lock);
4770                 free_ol_stateid_reaplist(&reaplist);
4771         } else {
4772                 spin_unlock(&clp->cl_lock);
4773                 free_ol_stateid_reaplist(&reaplist);
4774                 move_to_close_lru(s, clp->net);
4775         }
4776 }
4777
4778 /*
4779  * nfs4_unlock_state() called after encode
4780  */
4781 __be32
4782 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4783             struct nfsd4_close *close)
4784 {
4785         __be32 status;
4786         struct nfs4_ol_stateid *stp;
4787         struct net *net = SVC_NET(rqstp);
4788         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4789
4790         dprintk("NFSD: nfsd4_close on file %pd\n",
4791                         cstate->current_fh.fh_dentry);
4792
4793         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4794                                         &close->cl_stateid,
4795                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4796                                         &stp, nn);
4797         nfsd4_bump_seqid(cstate, status);
4798         if (status)
4799                 goto out;
4800         update_stateid(&stp->st_stid.sc_stateid);
4801         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4802
4803         nfsd4_close_open_stateid(stp);
4804
4805         /* put reference from nfs4_preprocess_seqid_op */
4806         nfs4_put_stid(&stp->st_stid);
4807 out:
4808         return status;
4809 }
4810
4811 __be32
4812 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4813                   struct nfsd4_delegreturn *dr)
4814 {
4815         struct nfs4_delegation *dp;
4816         stateid_t *stateid = &dr->dr_stateid;
4817         struct nfs4_stid *s;
4818         __be32 status;
4819         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4820
4821         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4822                 return status;
4823
4824         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
4825         if (status)
4826                 goto out;
4827         dp = delegstateid(s);
4828         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4829         if (status)
4830                 goto put_stateid;
4831
4832         destroy_delegation(dp);
4833 put_stateid:
4834         nfs4_put_stid(&dp->dl_stid);
4835 out:
4836         return status;
4837 }
4838
4839
4840 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4841
4842 static inline u64
4843 end_offset(u64 start, u64 len)
4844 {
4845         u64 end;
4846
4847         end = start + len;
4848         return end >= start ? end: NFS4_MAX_UINT64;
4849 }
4850
4851 /* last octet in a range */
4852 static inline u64
4853 last_byte_offset(u64 start, u64 len)
4854 {
4855         u64 end;
4856
4857         WARN_ON_ONCE(!len);
4858         end = start + len;
4859         return end > start ? end - 1: NFS4_MAX_UINT64;
4860 }
4861
4862 /*
4863  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4864  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4865  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4866  * locking, this prevents us from being completely protocol-compliant.  The
4867  * real solution to this problem is to start using unsigned file offsets in
4868  * the VFS, but this is a very deep change!
4869  */
4870 static inline void
4871 nfs4_transform_lock_offset(struct file_lock *lock)
4872 {
4873         if (lock->fl_start < 0)
4874                 lock->fl_start = OFFSET_MAX;
4875         if (lock->fl_end < 0)
4876                 lock->fl_end = OFFSET_MAX;
4877 }
4878
4879 static void nfsd4_fl_get_owner(struct file_lock *dst, struct file_lock *src)
4880 {
4881         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)src->fl_owner;
4882         dst->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lo->lo_owner));
4883 }
4884
4885 static void nfsd4_fl_put_owner(struct file_lock *fl)
4886 {
4887         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
4888
4889         if (lo) {
4890                 nfs4_put_stateowner(&lo->lo_owner);
4891                 fl->fl_owner = NULL;
4892         }
4893 }
4894
4895 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4896         .lm_get_owner = nfsd4_fl_get_owner,
4897         .lm_put_owner = nfsd4_fl_put_owner,
4898 };
4899
4900 static inline void
4901 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4902 {
4903         struct nfs4_lockowner *lo;
4904
4905         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4906                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4907                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4908                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4909                 if (!deny->ld_owner.data)
4910                         /* We just don't care that much */
4911                         goto nevermind;
4912                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4913                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4914         } else {
4915 nevermind:
4916                 deny->ld_owner.len = 0;
4917                 deny->ld_owner.data = NULL;
4918                 deny->ld_clientid.cl_boot = 0;
4919                 deny->ld_clientid.cl_id = 0;
4920         }
4921         deny->ld_start = fl->fl_start;
4922         deny->ld_length = NFS4_MAX_UINT64;
4923         if (fl->fl_end != NFS4_MAX_UINT64)
4924                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4925         deny->ld_type = NFS4_READ_LT;
4926         if (fl->fl_type != F_RDLCK)
4927                 deny->ld_type = NFS4_WRITE_LT;
4928 }
4929
4930 static struct nfs4_lockowner *
4931 find_lockowner_str_locked(clientid_t *clid, struct xdr_netobj *owner,
4932                 struct nfs4_client *clp)
4933 {
4934         unsigned int strhashval = ownerstr_hashval(owner);
4935         struct nfs4_stateowner *so;
4936
4937         lockdep_assert_held(&clp->cl_lock);
4938
4939         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
4940                             so_strhash) {
4941                 if (so->so_is_open_owner)
4942                         continue;
4943                 if (same_owner_str(so, owner))
4944                         return lockowner(nfs4_get_stateowner(so));
4945         }
4946         return NULL;
4947 }
4948
4949 static struct nfs4_lockowner *
4950 find_lockowner_str(clientid_t *clid, struct xdr_netobj *owner,
4951                 struct nfs4_client *clp)
4952 {
4953         struct nfs4_lockowner *lo;
4954
4955         spin_lock(&clp->cl_lock);
4956         lo = find_lockowner_str_locked(clid, owner, clp);
4957         spin_unlock(&clp->cl_lock);
4958         return lo;
4959 }
4960
4961 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
4962 {
4963         unhash_lockowner_locked(lockowner(sop));
4964 }
4965
4966 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
4967 {
4968         struct nfs4_lockowner *lo = lockowner(sop);
4969
4970         kmem_cache_free(lockowner_slab, lo);
4971 }
4972
4973 static const struct nfs4_stateowner_operations lockowner_ops = {
4974         .so_unhash =    nfs4_unhash_lockowner,
4975         .so_free =      nfs4_free_lockowner,
4976 };
4977
4978 /*
4979  * Alloc a lock owner structure.
4980  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4981  * occurred.
4982  *
4983  * strhashval = ownerstr_hashval
4984  */
4985 static struct nfs4_lockowner *
4986 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
4987                            struct nfs4_ol_stateid *open_stp,
4988                            struct nfsd4_lock *lock)
4989 {
4990         struct nfs4_lockowner *lo, *ret;
4991
4992         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4993         if (!lo)
4994                 return NULL;
4995         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4996         lo->lo_owner.so_is_open_owner = 0;
4997         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
4998         lo->lo_owner.so_ops = &lockowner_ops;
4999         spin_lock(&clp->cl_lock);
5000         ret = find_lockowner_str_locked(&clp->cl_clientid,
5001                         &lock->lk_new_owner, clp);
5002         if (ret == NULL) {
5003                 list_add(&lo->lo_owner.so_strhash,
5004                          &clp->cl_ownerstr_hashtbl[strhashval]);
5005                 ret = lo;
5006         } else
5007                 nfs4_free_lockowner(&lo->lo_owner);
5008         spin_unlock(&clp->cl_lock);
5009         return lo;
5010 }
5011
5012 static void
5013 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5014                   struct nfs4_file *fp, struct inode *inode,
5015                   struct nfs4_ol_stateid *open_stp)
5016 {
5017         struct nfs4_client *clp = lo->lo_owner.so_client;
5018
5019         lockdep_assert_held(&clp->cl_lock);
5020
5021         atomic_inc(&stp->st_stid.sc_count);
5022         stp->st_stid.sc_type = NFS4_LOCK_STID;
5023         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5024         get_nfs4_file(fp);
5025         stp->st_stid.sc_file = fp;
5026         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5027         stp->st_access_bmap = 0;
5028         stp->st_deny_bmap = open_stp->st_deny_bmap;
5029         stp->st_openstp = open_stp;
5030         list_add(&stp->st_locks, &open_stp->st_locks);
5031         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5032         spin_lock(&fp->fi_lock);
5033         list_add(&stp->st_perfile, &fp->fi_stateids);
5034         spin_unlock(&fp->fi_lock);
5035 }
5036
5037 static struct nfs4_ol_stateid *
5038 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5039 {
5040         struct nfs4_ol_stateid *lst;
5041         struct nfs4_client *clp = lo->lo_owner.so_client;
5042
5043         lockdep_assert_held(&clp->cl_lock);
5044
5045         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5046                 if (lst->st_stid.sc_file == fp) {
5047                         atomic_inc(&lst->st_stid.sc_count);
5048                         return lst;
5049                 }
5050         }
5051         return NULL;
5052 }
5053
5054 static struct nfs4_ol_stateid *
5055 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5056                             struct inode *inode, struct nfs4_ol_stateid *ost,
5057                             bool *new)
5058 {
5059         struct nfs4_stid *ns = NULL;
5060         struct nfs4_ol_stateid *lst;
5061         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5062         struct nfs4_client *clp = oo->oo_owner.so_client;
5063
5064         spin_lock(&clp->cl_lock);
5065         lst = find_lock_stateid(lo, fi);
5066         if (lst == NULL) {
5067                 spin_unlock(&clp->cl_lock);
5068                 ns = nfs4_alloc_stid(clp, stateid_slab);
5069                 if (ns == NULL)
5070                         return NULL;
5071
5072                 spin_lock(&clp->cl_lock);
5073                 lst = find_lock_stateid(lo, fi);
5074                 if (likely(!lst)) {
5075                         lst = openlockstateid(ns);
5076                         init_lock_stateid(lst, lo, fi, inode, ost);
5077                         ns = NULL;
5078                         *new = true;
5079                 }
5080         }
5081         spin_unlock(&clp->cl_lock);
5082         if (ns)
5083                 nfs4_put_stid(ns);
5084         return lst;
5085 }
5086
5087 static int
5088 check_lock_length(u64 offset, u64 length)
5089 {
5090         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
5091              LOFF_OVERFLOW(offset, length)));
5092 }
5093
5094 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5095 {
5096         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5097
5098         lockdep_assert_held(&fp->fi_lock);
5099
5100         if (test_access(access, lock_stp))
5101                 return;
5102         __nfs4_file_get_access(fp, access);
5103         set_access(access, lock_stp);
5104 }
5105
5106 static __be32
5107 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5108                             struct nfs4_ol_stateid *ost,
5109                             struct nfsd4_lock *lock,
5110                             struct nfs4_ol_stateid **lst, bool *new)
5111 {
5112         __be32 status;
5113         struct nfs4_file *fi = ost->st_stid.sc_file;
5114         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5115         struct nfs4_client *cl = oo->oo_owner.so_client;
5116         struct inode *inode = cstate->current_fh.fh_dentry->d_inode;
5117         struct nfs4_lockowner *lo;
5118         unsigned int strhashval;
5119
5120         lo = find_lockowner_str(&cl->cl_clientid, &lock->v.new.owner, cl);
5121         if (!lo) {
5122                 strhashval = ownerstr_hashval(&lock->v.new.owner);
5123                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5124                 if (lo == NULL)
5125                         return nfserr_jukebox;
5126         } else {
5127                 /* with an existing lockowner, seqids must be the same */
5128                 status = nfserr_bad_seqid;
5129                 if (!cstate->minorversion &&
5130                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5131                         goto out;
5132         }
5133
5134         *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5135         if (*lst == NULL) {
5136                 status = nfserr_jukebox;
5137                 goto out;
5138         }
5139         status = nfs_ok;
5140 out:
5141         nfs4_put_stateowner(&lo->lo_owner);
5142         return status;
5143 }
5144
5145 /*
5146  *  LOCK operation
5147  */
5148 __be32
5149 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5150            struct nfsd4_lock *lock)
5151 {
5152         struct nfs4_openowner *open_sop = NULL;
5153         struct nfs4_lockowner *lock_sop = NULL;
5154         struct nfs4_ol_stateid *lock_stp = NULL;
5155         struct nfs4_ol_stateid *open_stp = NULL;
5156         struct nfs4_file *fp;
5157         struct file *filp = NULL;
5158         struct file_lock *file_lock = NULL;
5159         struct file_lock *conflock = NULL;
5160         __be32 status = 0;
5161         int lkflg;
5162         int err;
5163         bool new = false;
5164         struct net *net = SVC_NET(rqstp);
5165         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5166
5167         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5168                 (long long) lock->lk_offset,
5169                 (long long) lock->lk_length);
5170
5171         if (check_lock_length(lock->lk_offset, lock->lk_length))
5172                  return nfserr_inval;
5173
5174         if ((status = fh_verify(rqstp, &cstate->current_fh,
5175                                 S_IFREG, NFSD_MAY_LOCK))) {
5176                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5177                 return status;
5178         }
5179
5180         if (lock->lk_is_new) {
5181                 if (nfsd4_has_session(cstate))
5182                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5183                         memcpy(&lock->v.new.clientid,
5184                                 &cstate->session->se_client->cl_clientid,
5185                                 sizeof(clientid_t));
5186
5187                 status = nfserr_stale_clientid;
5188                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5189                         goto out;
5190
5191                 /* validate and update open stateid and open seqid */
5192                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5193                                         lock->lk_new_open_seqid,
5194                                         &lock->lk_new_open_stateid,
5195                                         &open_stp, nn);
5196                 if (status)
5197                         goto out;
5198                 open_sop = openowner(open_stp->st_stateowner);
5199                 status = nfserr_bad_stateid;
5200                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5201                                                 &lock->v.new.clientid))
5202                         goto out;
5203                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5204                                                         &lock_stp, &new);
5205         } else {
5206                 status = nfs4_preprocess_seqid_op(cstate,
5207                                        lock->lk_old_lock_seqid,
5208                                        &lock->lk_old_lock_stateid,
5209                                        NFS4_LOCK_STID, &lock_stp, nn);
5210         }
5211         if (status)
5212                 goto out;
5213         lock_sop = lockowner(lock_stp->st_stateowner);
5214
5215         lkflg = setlkflg(lock->lk_type);
5216         status = nfs4_check_openmode(lock_stp, lkflg);
5217         if (status)
5218                 goto out;
5219
5220         status = nfserr_grace;
5221         if (locks_in_grace(net) && !lock->lk_reclaim)
5222                 goto out;
5223         status = nfserr_no_grace;
5224         if (!locks_in_grace(net) && lock->lk_reclaim)
5225                 goto out;
5226
5227         file_lock = locks_alloc_lock();
5228         if (!file_lock) {
5229                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5230                 status = nfserr_jukebox;
5231                 goto out;
5232         }
5233
5234         fp = lock_stp->st_stid.sc_file;
5235         switch (lock->lk_type) {
5236                 case NFS4_READ_LT:
5237                 case NFS4_READW_LT:
5238                         spin_lock(&fp->fi_lock);
5239                         filp = find_readable_file_locked(fp);
5240                         if (filp)
5241                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5242                         spin_unlock(&fp->fi_lock);
5243                         file_lock->fl_type = F_RDLCK;
5244                         break;
5245                 case NFS4_WRITE_LT:
5246                 case NFS4_WRITEW_LT:
5247                         spin_lock(&fp->fi_lock);
5248                         filp = find_writeable_file_locked(fp);
5249                         if (filp)
5250                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5251                         spin_unlock(&fp->fi_lock);
5252                         file_lock->fl_type = F_WRLCK;
5253                         break;
5254                 default:
5255                         status = nfserr_inval;
5256                 goto out;
5257         }
5258         if (!filp) {
5259                 status = nfserr_openmode;
5260                 goto out;
5261         }
5262
5263         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5264         file_lock->fl_pid = current->tgid;
5265         file_lock->fl_file = filp;
5266         file_lock->fl_flags = FL_POSIX;
5267         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5268         file_lock->fl_start = lock->lk_offset;
5269         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5270         nfs4_transform_lock_offset(file_lock);
5271
5272         conflock = locks_alloc_lock();
5273         if (!conflock) {
5274                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5275                 status = nfserr_jukebox;
5276                 goto out;
5277         }
5278
5279         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5280         switch (-err) {
5281         case 0: /* success! */
5282                 update_stateid(&lock_stp->st_stid.sc_stateid);
5283                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
5284                                 sizeof(stateid_t));
5285                 status = 0;
5286                 break;
5287         case (EAGAIN):          /* conflock holds conflicting lock */
5288                 status = nfserr_denied;
5289                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5290                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5291                 break;
5292         case (EDEADLK):
5293                 status = nfserr_deadlock;
5294                 break;
5295         default:
5296                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5297                 status = nfserrno(err);
5298                 break;
5299         }
5300 out:
5301         if (filp)
5302                 fput(filp);
5303         if (lock_stp) {
5304                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5305                 if (cstate->replay_owner &&
5306                     cstate->replay_owner != &lock_sop->lo_owner &&
5307                     seqid_mutating_err(ntohl(status)))
5308                         lock_sop->lo_owner.so_seqid++;
5309
5310                 /*
5311                  * If this is a new, never-before-used stateid, and we are
5312                  * returning an error, then just go ahead and release it.
5313                  */
5314                 if (status && new)
5315                         release_lock_stateid(lock_stp);
5316
5317                 nfs4_put_stid(&lock_stp->st_stid);
5318         }
5319         if (open_stp)
5320                 nfs4_put_stid(&open_stp->st_stid);
5321         nfsd4_bump_seqid(cstate, status);
5322         if (file_lock)
5323                 locks_free_lock(file_lock);
5324         if (conflock)
5325                 locks_free_lock(conflock);
5326         return status;
5327 }
5328
5329 /*
5330  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5331  * so we do a temporary open here just to get an open file to pass to
5332  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5333  * inode operation.)
5334  */
5335 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5336 {
5337         struct file *file;
5338         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5339         if (!err) {
5340                 err = nfserrno(vfs_test_lock(file, lock));
5341                 nfsd_close(file);
5342         }
5343         return err;
5344 }
5345
5346 /*
5347  * LOCKT operation
5348  */
5349 __be32
5350 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5351             struct nfsd4_lockt *lockt)
5352 {
5353         struct file_lock *file_lock = NULL;
5354         struct nfs4_lockowner *lo = NULL;
5355         __be32 status;
5356         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5357
5358         if (locks_in_grace(SVC_NET(rqstp)))
5359                 return nfserr_grace;
5360
5361         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5362                  return nfserr_inval;
5363
5364         if (!nfsd4_has_session(cstate)) {
5365                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5366                 if (status)
5367                         goto out;
5368         }
5369
5370         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5371                 goto out;
5372
5373         file_lock = locks_alloc_lock();
5374         if (!file_lock) {
5375                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5376                 status = nfserr_jukebox;
5377                 goto out;
5378         }
5379
5380         switch (lockt->lt_type) {
5381                 case NFS4_READ_LT:
5382                 case NFS4_READW_LT:
5383                         file_lock->fl_type = F_RDLCK;
5384                 break;
5385                 case NFS4_WRITE_LT:
5386                 case NFS4_WRITEW_LT:
5387                         file_lock->fl_type = F_WRLCK;
5388                 break;
5389                 default:
5390                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5391                         status = nfserr_inval;
5392                 goto out;
5393         }
5394
5395         lo = find_lockowner_str(&lockt->lt_clientid, &lockt->lt_owner,
5396                                 cstate->clp);
5397         if (lo)
5398                 file_lock->fl_owner = (fl_owner_t)lo;
5399         file_lock->fl_pid = current->tgid;
5400         file_lock->fl_flags = FL_POSIX;
5401
5402         file_lock->fl_start = lockt->lt_offset;
5403         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5404
5405         nfs4_transform_lock_offset(file_lock);
5406
5407         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5408         if (status)
5409                 goto out;
5410
5411         if (file_lock->fl_type != F_UNLCK) {
5412                 status = nfserr_denied;
5413                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5414         }
5415 out:
5416         if (lo)
5417                 nfs4_put_stateowner(&lo->lo_owner);
5418         if (file_lock)
5419                 locks_free_lock(file_lock);
5420         return status;
5421 }
5422
5423 __be32
5424 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5425             struct nfsd4_locku *locku)
5426 {
5427         struct nfs4_ol_stateid *stp;
5428         struct file *filp = NULL;
5429         struct file_lock *file_lock = NULL;
5430         __be32 status;
5431         int err;
5432         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5433
5434         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5435                 (long long) locku->lu_offset,
5436                 (long long) locku->lu_length);
5437
5438         if (check_lock_length(locku->lu_offset, locku->lu_length))
5439                  return nfserr_inval;
5440
5441         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5442                                         &locku->lu_stateid, NFS4_LOCK_STID,
5443                                         &stp, nn);
5444         if (status)
5445                 goto out;
5446         filp = find_any_file(stp->st_stid.sc_file);
5447         if (!filp) {
5448                 status = nfserr_lock_range;
5449                 goto put_stateid;
5450         }
5451         file_lock = locks_alloc_lock();
5452         if (!file_lock) {
5453                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5454                 status = nfserr_jukebox;
5455                 goto fput;
5456         }
5457
5458         file_lock->fl_type = F_UNLCK;
5459         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5460         file_lock->fl_pid = current->tgid;
5461         file_lock->fl_file = filp;
5462         file_lock->fl_flags = FL_POSIX;
5463         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5464         file_lock->fl_start = locku->lu_offset;
5465
5466         file_lock->fl_end = last_byte_offset(locku->lu_offset,
5467                                                 locku->lu_length);
5468         nfs4_transform_lock_offset(file_lock);
5469
5470         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5471         if (err) {
5472                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5473                 goto out_nfserr;
5474         }
5475         update_stateid(&stp->st_stid.sc_stateid);
5476         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
5477 fput:
5478         fput(filp);
5479 put_stateid:
5480         nfs4_put_stid(&stp->st_stid);
5481 out:
5482         nfsd4_bump_seqid(cstate, status);
5483         if (file_lock)
5484                 locks_free_lock(file_lock);
5485         return status;
5486
5487 out_nfserr:
5488         status = nfserrno(err);
5489         goto fput;
5490 }
5491
5492 /*
5493  * returns
5494  *      true:  locks held by lockowner
5495  *      false: no locks held by lockowner
5496  */
5497 static bool
5498 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5499 {
5500         struct file_lock **flpp;
5501         int status = false;
5502         struct file *filp = find_any_file(fp);
5503         struct inode *inode;
5504
5505         if (!filp) {
5506                 /* Any valid lock stateid should have some sort of access */
5507                 WARN_ON_ONCE(1);
5508                 return status;
5509         }
5510
5511         inode = file_inode(filp);
5512
5513         spin_lock(&inode->i_lock);
5514         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
5515                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
5516                         status = true;
5517                         break;
5518                 }
5519         }
5520         spin_unlock(&inode->i_lock);
5521         fput(filp);
5522         return status;
5523 }
5524
5525 __be32
5526 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5527                         struct nfsd4_compound_state *cstate,
5528                         struct nfsd4_release_lockowner *rlockowner)
5529 {
5530         clientid_t *clid = &rlockowner->rl_clientid;
5531         struct nfs4_stateowner *sop;
5532         struct nfs4_lockowner *lo = NULL;
5533         struct nfs4_ol_stateid *stp;
5534         struct xdr_netobj *owner = &rlockowner->rl_owner;
5535         unsigned int hashval = ownerstr_hashval(owner);
5536         __be32 status;
5537         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5538         struct nfs4_client *clp;
5539
5540         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5541                 clid->cl_boot, clid->cl_id);
5542
5543         status = lookup_clientid(clid, cstate, nn);
5544         if (status)
5545                 return status;
5546
5547         clp = cstate->clp;
5548         /* Find the matching lock stateowner */
5549         spin_lock(&clp->cl_lock);
5550         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5551                             so_strhash) {
5552
5553                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5554                         continue;
5555
5556                 /* see if there are still any locks associated with it */
5557                 lo = lockowner(sop);
5558                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5559                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
5560                                 status = nfserr_locks_held;
5561                                 spin_unlock(&clp->cl_lock);
5562                                 return status;
5563                         }
5564                 }
5565
5566                 nfs4_get_stateowner(sop);
5567                 break;
5568         }
5569         spin_unlock(&clp->cl_lock);
5570         if (lo)
5571                 release_lockowner(lo);
5572         return status;
5573 }
5574
5575 static inline struct nfs4_client_reclaim *
5576 alloc_reclaim(void)
5577 {
5578         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5579 }
5580
5581 bool
5582 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5583 {
5584         struct nfs4_client_reclaim *crp;
5585
5586         crp = nfsd4_find_reclaim_client(name, nn);
5587         return (crp && crp->cr_clp);
5588 }
5589
5590 /*
5591  * failure => all reset bets are off, nfserr_no_grace...
5592  */
5593 struct nfs4_client_reclaim *
5594 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5595 {
5596         unsigned int strhashval;
5597         struct nfs4_client_reclaim *crp;
5598
5599         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5600         crp = alloc_reclaim();
5601         if (crp) {
5602                 strhashval = clientstr_hashval(name);
5603                 INIT_LIST_HEAD(&crp->cr_strhash);
5604                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5605                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5606                 crp->cr_clp = NULL;
5607                 nn->reclaim_str_hashtbl_size++;
5608         }
5609         return crp;
5610 }
5611
5612 void
5613 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5614 {
5615         list_del(&crp->cr_strhash);
5616         kfree(crp);
5617         nn->reclaim_str_hashtbl_size--;
5618 }
5619
5620 void
5621 nfs4_release_reclaim(struct nfsd_net *nn)
5622 {
5623         struct nfs4_client_reclaim *crp = NULL;
5624         int i;
5625
5626         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5627                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5628                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5629                                         struct nfs4_client_reclaim, cr_strhash);
5630                         nfs4_remove_reclaim_record(crp, nn);
5631                 }
5632         }
5633         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5634 }
5635
5636 /*
5637  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5638 struct nfs4_client_reclaim *
5639 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5640 {
5641         unsigned int strhashval;
5642         struct nfs4_client_reclaim *crp = NULL;
5643
5644         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5645
5646         strhashval = clientstr_hashval(recdir);
5647         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5648                 if (same_name(crp->cr_recdir, recdir)) {
5649                         return crp;
5650                 }
5651         }
5652         return NULL;
5653 }
5654
5655 /*
5656 * Called from OPEN. Look for clientid in reclaim list.
5657 */
5658 __be32
5659 nfs4_check_open_reclaim(clientid_t *clid,
5660                 struct nfsd4_compound_state *cstate,
5661                 struct nfsd_net *nn)
5662 {
5663         __be32 status;
5664
5665         /* find clientid in conf_id_hashtbl */
5666         status = lookup_clientid(clid, cstate, nn);
5667         if (status)
5668                 return nfserr_reclaim_bad;
5669
5670         if (nfsd4_client_record_check(cstate->clp))
5671                 return nfserr_reclaim_bad;
5672
5673         return nfs_ok;
5674 }
5675
5676 #ifdef CONFIG_NFSD_FAULT_INJECTION
5677 static inline void
5678 put_client(struct nfs4_client *clp)
5679 {
5680         atomic_dec(&clp->cl_refcount);
5681 }
5682
5683 static struct nfs4_client *
5684 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
5685 {
5686         struct nfs4_client *clp;
5687         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5688                                           nfsd_net_id);
5689
5690         if (!nfsd_netns_ready(nn))
5691                 return NULL;
5692
5693         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5694                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
5695                         return clp;
5696         }
5697         return NULL;
5698 }
5699
5700 u64
5701 nfsd_inject_print_clients(void)
5702 {
5703         struct nfs4_client *clp;
5704         u64 count = 0;
5705         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5706                                           nfsd_net_id);
5707         char buf[INET6_ADDRSTRLEN];
5708
5709         if (!nfsd_netns_ready(nn))
5710                 return 0;
5711
5712         spin_lock(&nn->client_lock);
5713         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5714                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5715                 pr_info("NFS Client: %s\n", buf);
5716                 ++count;
5717         }
5718         spin_unlock(&nn->client_lock);
5719
5720         return count;
5721 }
5722
5723 u64
5724 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
5725 {
5726         u64 count = 0;
5727         struct nfs4_client *clp;
5728         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5729                                           nfsd_net_id);
5730
5731         if (!nfsd_netns_ready(nn))
5732                 return count;
5733
5734         spin_lock(&nn->client_lock);
5735         clp = nfsd_find_client(addr, addr_size);
5736         if (clp) {
5737                 if (mark_client_expired_locked(clp) == nfs_ok)
5738                         ++count;
5739                 else
5740                         clp = NULL;
5741         }
5742         spin_unlock(&nn->client_lock);
5743
5744         if (clp)
5745                 expire_client(clp);
5746
5747         return count;
5748 }
5749
5750 u64
5751 nfsd_inject_forget_clients(u64 max)
5752 {
5753         u64 count = 0;
5754         struct nfs4_client *clp, *next;
5755         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5756                                                 nfsd_net_id);
5757         LIST_HEAD(reaplist);
5758
5759         if (!nfsd_netns_ready(nn))
5760                 return count;
5761
5762         spin_lock(&nn->client_lock);
5763         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
5764                 if (mark_client_expired_locked(clp) == nfs_ok) {
5765                         list_add(&clp->cl_lru, &reaplist);
5766                         if (max != 0 && ++count >= max)
5767                                 break;
5768                 }
5769         }
5770         spin_unlock(&nn->client_lock);
5771
5772         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
5773                 expire_client(clp);
5774
5775         return count;
5776 }
5777
5778 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
5779                              const char *type)
5780 {
5781         char buf[INET6_ADDRSTRLEN];
5782         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5783         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
5784 }
5785
5786 static void
5787 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
5788                              struct list_head *collect)
5789 {
5790         struct nfs4_client *clp = lst->st_stid.sc_client;
5791         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5792                                           nfsd_net_id);
5793
5794         if (!collect)
5795                 return;
5796
5797         lockdep_assert_held(&nn->client_lock);
5798         atomic_inc(&clp->cl_refcount);
5799         list_add(&lst->st_locks, collect);
5800 }
5801
5802 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
5803                                     struct list_head *collect,
5804                                     void (*func)(struct nfs4_ol_stateid *))
5805 {
5806         struct nfs4_openowner *oop;
5807         struct nfs4_ol_stateid *stp, *st_next;
5808         struct nfs4_ol_stateid *lst, *lst_next;
5809         u64 count = 0;
5810
5811         spin_lock(&clp->cl_lock);
5812         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
5813                 list_for_each_entry_safe(stp, st_next,
5814                                 &oop->oo_owner.so_stateids, st_perstateowner) {
5815                         list_for_each_entry_safe(lst, lst_next,
5816                                         &stp->st_locks, st_locks) {
5817                                 if (func) {
5818                                         func(lst);
5819                                         nfsd_inject_add_lock_to_list(lst,
5820                                                                 collect);
5821                                 }
5822                                 ++count;
5823                                 /*
5824                                  * Despite the fact that these functions deal
5825                                  * with 64-bit integers for "count", we must
5826                                  * ensure that it doesn't blow up the
5827                                  * clp->cl_refcount. Throw a warning if we
5828                                  * start to approach INT_MAX here.
5829                                  */
5830                                 WARN_ON_ONCE(count == (INT_MAX / 2));
5831                                 if (count == max)
5832                                         goto out;
5833                         }
5834                 }
5835         }
5836 out:
5837         spin_unlock(&clp->cl_lock);
5838
5839         return count;
5840 }
5841
5842 static u64
5843 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
5844                           u64 max)
5845 {
5846         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
5847 }
5848
5849 static u64
5850 nfsd_print_client_locks(struct nfs4_client *clp)
5851 {
5852         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
5853         nfsd_print_count(clp, count, "locked files");
5854         return count;
5855 }
5856
5857 u64
5858 nfsd_inject_print_locks(void)
5859 {
5860         struct nfs4_client *clp;
5861         u64 count = 0;
5862         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5863                                                 nfsd_net_id);
5864
5865         if (!nfsd_netns_ready(nn))
5866                 return 0;
5867
5868         spin_lock(&nn->client_lock);
5869         list_for_each_entry(clp, &nn->client_lru, cl_lru)
5870                 count += nfsd_print_client_locks(clp);
5871         spin_unlock(&nn->client_lock);
5872
5873         return count;
5874 }
5875
5876 static void
5877 nfsd_reap_locks(struct list_head *reaplist)
5878 {
5879         struct nfs4_client *clp;
5880         struct nfs4_ol_stateid *stp, *next;
5881
5882         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
5883                 list_del_init(&stp->st_locks);
5884                 clp = stp->st_stid.sc_client;
5885                 nfs4_put_stid(&stp->st_stid);
5886                 put_client(clp);
5887         }
5888 }
5889
5890 u64
5891 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
5892 {
5893         unsigned int count = 0;
5894         struct nfs4_client *clp;
5895         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5896                                                 nfsd_net_id);
5897         LIST_HEAD(reaplist);
5898
5899         if (!nfsd_netns_ready(nn))
5900                 return count;
5901
5902         spin_lock(&nn->client_lock);
5903         clp = nfsd_find_client(addr, addr_size);
5904         if (clp)
5905                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
5906         spin_unlock(&nn->client_lock);
5907         nfsd_reap_locks(&reaplist);
5908         return count;
5909 }
5910
5911 u64
5912 nfsd_inject_forget_locks(u64 max)
5913 {
5914         u64 count = 0;
5915         struct nfs4_client *clp;
5916         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5917                                                 nfsd_net_id);
5918         LIST_HEAD(reaplist);
5919
5920         if (!nfsd_netns_ready(nn))
5921                 return count;
5922
5923         spin_lock(&nn->client_lock);
5924         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5925                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
5926                 if (max != 0 && count >= max)
5927                         break;
5928         }
5929         spin_unlock(&nn->client_lock);
5930         nfsd_reap_locks(&reaplist);
5931         return count;
5932 }
5933
5934 static u64
5935 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
5936                               struct list_head *collect,
5937                               void (*func)(struct nfs4_openowner *))
5938 {
5939         struct nfs4_openowner *oop, *next;
5940         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5941                                                 nfsd_net_id);
5942         u64 count = 0;
5943
5944         lockdep_assert_held(&nn->client_lock);
5945
5946         spin_lock(&clp->cl_lock);
5947         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
5948                 if (func) {
5949                         func(oop);
5950                         if (collect) {
5951                                 atomic_inc(&clp->cl_refcount);
5952                                 list_add(&oop->oo_perclient, collect);
5953                         }
5954                 }
5955                 ++count;
5956                 /*
5957                  * Despite the fact that these functions deal with
5958                  * 64-bit integers for "count", we must ensure that
5959                  * it doesn't blow up the clp->cl_refcount. Throw a
5960                  * warning if we start to approach INT_MAX here.
5961                  */
5962                 WARN_ON_ONCE(count == (INT_MAX / 2));
5963                 if (count == max)
5964                         break;
5965         }
5966         spin_unlock(&clp->cl_lock);
5967
5968         return count;
5969 }
5970
5971 static u64
5972 nfsd_print_client_openowners(struct nfs4_client *clp)
5973 {
5974         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
5975
5976         nfsd_print_count(clp, count, "openowners");
5977         return count;
5978 }
5979
5980 static u64
5981 nfsd_collect_client_openowners(struct nfs4_client *clp,
5982                                struct list_head *collect, u64 max)
5983 {
5984         return nfsd_foreach_client_openowner(clp, max, collect,
5985                                                 unhash_openowner_locked);
5986 }
5987
5988 u64
5989 nfsd_inject_print_openowners(void)
5990 {
5991         struct nfs4_client *clp;
5992         u64 count = 0;
5993         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5994                                                 nfsd_net_id);
5995
5996         if (!nfsd_netns_ready(nn))
5997                 return 0;
5998
5999         spin_lock(&nn->client_lock);
6000         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6001                 count += nfsd_print_client_openowners(clp);
6002         spin_unlock(&nn->client_lock);
6003
6004         return count;
6005 }
6006
6007 static void
6008 nfsd_reap_openowners(struct list_head *reaplist)
6009 {
6010         struct nfs4_client *clp;
6011         struct nfs4_openowner *oop, *next;
6012
6013         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6014                 list_del_init(&oop->oo_perclient);
6015                 clp = oop->oo_owner.so_client;
6016                 release_openowner(oop);
6017                 put_client(clp);
6018         }
6019 }
6020
6021 u64
6022 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6023                                      size_t addr_size)
6024 {
6025         unsigned int count = 0;
6026         struct nfs4_client *clp;
6027         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6028                                                 nfsd_net_id);
6029         LIST_HEAD(reaplist);
6030
6031         if (!nfsd_netns_ready(nn))
6032                 return count;
6033
6034         spin_lock(&nn->client_lock);
6035         clp = nfsd_find_client(addr, addr_size);
6036         if (clp)
6037                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6038         spin_unlock(&nn->client_lock);
6039         nfsd_reap_openowners(&reaplist);
6040         return count;
6041 }
6042
6043 u64
6044 nfsd_inject_forget_openowners(u64 max)
6045 {
6046         u64 count = 0;
6047         struct nfs4_client *clp;
6048         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6049                                                 nfsd_net_id);
6050         LIST_HEAD(reaplist);
6051
6052         if (!nfsd_netns_ready(nn))
6053                 return count;
6054
6055         spin_lock(&nn->client_lock);
6056         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6057                 count += nfsd_collect_client_openowners(clp, &reaplist,
6058                                                         max - count);
6059                 if (max != 0 && count >= max)
6060                         break;
6061         }
6062         spin_unlock(&nn->client_lock);
6063         nfsd_reap_openowners(&reaplist);
6064         return count;
6065 }
6066
6067 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6068                                      struct list_head *victims)
6069 {
6070         struct nfs4_delegation *dp, *next;
6071         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6072                                                 nfsd_net_id);
6073         u64 count = 0;
6074
6075         lockdep_assert_held(&nn->client_lock);
6076
6077         spin_lock(&state_lock);
6078         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6079                 if (victims) {
6080                         /*
6081                          * It's not safe to mess with delegations that have a
6082                          * non-zero dl_time. They might have already been broken
6083                          * and could be processed by the laundromat outside of
6084                          * the state_lock. Just leave them be.
6085                          */
6086                         if (dp->dl_time != 0)
6087                                 continue;
6088
6089                         atomic_inc(&clp->cl_refcount);
6090                         unhash_delegation_locked(dp);
6091                         list_add(&dp->dl_recall_lru, victims);
6092                 }
6093                 ++count;
6094                 /*
6095                  * Despite the fact that these functions deal with
6096                  * 64-bit integers for "count", we must ensure that
6097                  * it doesn't blow up the clp->cl_refcount. Throw a
6098                  * warning if we start to approach INT_MAX here.
6099                  */
6100                 WARN_ON_ONCE(count == (INT_MAX / 2));
6101                 if (count == max)
6102                         break;
6103         }
6104         spin_unlock(&state_lock);
6105         return count;
6106 }
6107
6108 static u64
6109 nfsd_print_client_delegations(struct nfs4_client *clp)
6110 {
6111         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6112
6113         nfsd_print_count(clp, count, "delegations");
6114         return count;
6115 }
6116
6117 u64
6118 nfsd_inject_print_delegations(void)
6119 {
6120         struct nfs4_client *clp;
6121         u64 count = 0;
6122         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6123                                                 nfsd_net_id);
6124
6125         if (!nfsd_netns_ready(nn))
6126                 return 0;
6127
6128         spin_lock(&nn->client_lock);
6129         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6130                 count += nfsd_print_client_delegations(clp);
6131         spin_unlock(&nn->client_lock);
6132
6133         return count;
6134 }
6135
6136 static void
6137 nfsd_forget_delegations(struct list_head *reaplist)
6138 {
6139         struct nfs4_client *clp;
6140         struct nfs4_delegation *dp, *next;
6141
6142         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6143                 list_del_init(&dp->dl_recall_lru);
6144                 clp = dp->dl_stid.sc_client;
6145                 revoke_delegation(dp);
6146                 put_client(clp);
6147         }
6148 }
6149
6150 u64
6151 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6152                                       size_t addr_size)
6153 {
6154         u64 count = 0;
6155         struct nfs4_client *clp;
6156         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6157                                                 nfsd_net_id);
6158         LIST_HEAD(reaplist);
6159
6160         if (!nfsd_netns_ready(nn))
6161                 return count;
6162
6163         spin_lock(&nn->client_lock);
6164         clp = nfsd_find_client(addr, addr_size);
6165         if (clp)
6166                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6167         spin_unlock(&nn->client_lock);
6168
6169         nfsd_forget_delegations(&reaplist);
6170         return count;
6171 }
6172
6173 u64
6174 nfsd_inject_forget_delegations(u64 max)
6175 {
6176         u64 count = 0;
6177         struct nfs4_client *clp;
6178         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6179                                                 nfsd_net_id);
6180         LIST_HEAD(reaplist);
6181
6182         if (!nfsd_netns_ready(nn))
6183                 return count;
6184
6185         spin_lock(&nn->client_lock);
6186         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6187                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6188                 if (max != 0 && count >= max)
6189                         break;
6190         }
6191         spin_unlock(&nn->client_lock);
6192         nfsd_forget_delegations(&reaplist);
6193         return count;
6194 }
6195
6196 static void
6197 nfsd_recall_delegations(struct list_head *reaplist)
6198 {
6199         struct nfs4_client *clp;
6200         struct nfs4_delegation *dp, *next;
6201
6202         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6203                 list_del_init(&dp->dl_recall_lru);
6204                 clp = dp->dl_stid.sc_client;
6205                 /*
6206                  * We skipped all entries that had a zero dl_time before,
6207                  * so we can now reset the dl_time back to 0. If a delegation
6208                  * break comes in now, then it won't make any difference since
6209                  * we're recalling it either way.
6210                  */
6211                 spin_lock(&state_lock);
6212                 dp->dl_time = 0;
6213                 spin_unlock(&state_lock);
6214                 nfsd_break_one_deleg(dp);
6215                 put_client(clp);
6216         }
6217 }
6218
6219 u64
6220 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6221                                       size_t addr_size)
6222 {
6223         u64 count = 0;
6224         struct nfs4_client *clp;
6225         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6226                                                 nfsd_net_id);
6227         LIST_HEAD(reaplist);
6228
6229         if (!nfsd_netns_ready(nn))
6230                 return count;
6231
6232         spin_lock(&nn->client_lock);
6233         clp = nfsd_find_client(addr, addr_size);
6234         if (clp)
6235                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6236         spin_unlock(&nn->client_lock);
6237
6238         nfsd_recall_delegations(&reaplist);
6239         return count;
6240 }
6241
6242 u64
6243 nfsd_inject_recall_delegations(u64 max)
6244 {
6245         u64 count = 0;
6246         struct nfs4_client *clp, *next;
6247         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6248                                                 nfsd_net_id);
6249         LIST_HEAD(reaplist);
6250
6251         if (!nfsd_netns_ready(nn))
6252                 return count;
6253
6254         spin_lock(&nn->client_lock);
6255         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6256                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6257                 if (max != 0 && ++count >= max)
6258                         break;
6259         }
6260         spin_unlock(&nn->client_lock);
6261         nfsd_recall_delegations(&reaplist);
6262         return count;
6263 }
6264 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6265
6266 /*
6267  * Since the lifetime of a delegation isn't limited to that of an open, a
6268  * client may quite reasonably hang on to a delegation as long as it has
6269  * the inode cached.  This becomes an obvious problem the first time a
6270  * client's inode cache approaches the size of the server's total memory.
6271  *
6272  * For now we avoid this problem by imposing a hard limit on the number
6273  * of delegations, which varies according to the server's memory size.
6274  */
6275 static void
6276 set_max_delegations(void)
6277 {
6278         /*
6279          * Allow at most 4 delegations per megabyte of RAM.  Quick
6280          * estimates suggest that in the worst case (where every delegation
6281          * is for a different inode), a delegation could take about 1.5K,
6282          * giving a worst case usage of about 6% of memory.
6283          */
6284         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6285 }
6286
6287 static int nfs4_state_create_net(struct net *net)
6288 {
6289         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6290         int i;
6291
6292         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6293                         CLIENT_HASH_SIZE, GFP_KERNEL);
6294         if (!nn->conf_id_hashtbl)
6295                 goto err;
6296         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6297                         CLIENT_HASH_SIZE, GFP_KERNEL);
6298         if (!nn->unconf_id_hashtbl)
6299                 goto err_unconf_id;
6300         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6301                         SESSION_HASH_SIZE, GFP_KERNEL);
6302         if (!nn->sessionid_hashtbl)
6303                 goto err_sessionid;
6304
6305         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6306                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6307                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6308         }
6309         for (i = 0; i < SESSION_HASH_SIZE; i++)
6310                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6311         nn->conf_name_tree = RB_ROOT;
6312         nn->unconf_name_tree = RB_ROOT;
6313         INIT_LIST_HEAD(&nn->client_lru);
6314         INIT_LIST_HEAD(&nn->close_lru);
6315         INIT_LIST_HEAD(&nn->del_recall_lru);
6316         spin_lock_init(&nn->client_lock);
6317
6318         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6319         get_net(net);
6320
6321         return 0;
6322
6323 err_sessionid:
6324         kfree(nn->unconf_id_hashtbl);
6325 err_unconf_id:
6326         kfree(nn->conf_id_hashtbl);
6327 err:
6328         return -ENOMEM;
6329 }
6330
6331 static void
6332 nfs4_state_destroy_net(struct net *net)
6333 {
6334         int i;
6335         struct nfs4_client *clp = NULL;
6336         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6337
6338         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6339                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6340                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6341                         destroy_client(clp);
6342                 }
6343         }
6344
6345         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6346                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6347                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6348                         destroy_client(clp);
6349                 }
6350         }
6351
6352         kfree(nn->sessionid_hashtbl);
6353         kfree(nn->unconf_id_hashtbl);
6354         kfree(nn->conf_id_hashtbl);
6355         put_net(net);
6356 }
6357
6358 int
6359 nfs4_state_start_net(struct net *net)
6360 {
6361         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6362         int ret;
6363
6364         ret = nfs4_state_create_net(net);
6365         if (ret)
6366                 return ret;
6367         nfsd4_client_tracking_init(net);
6368         nn->boot_time = get_seconds();
6369         locks_start_grace(net, &nn->nfsd4_manager);
6370         nn->grace_ended = false;
6371         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6372                nn->nfsd4_grace, net);
6373         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6374         return 0;
6375 }
6376
6377 /* initialization to perform when the nfsd service is started: */
6378
6379 int
6380 nfs4_state_start(void)
6381 {
6382         int ret;
6383
6384         ret = set_callback_cred();
6385         if (ret)
6386                 return -ENOMEM;
6387         laundry_wq = create_singlethread_workqueue("nfsd4");
6388         if (laundry_wq == NULL) {
6389                 ret = -ENOMEM;
6390                 goto out_recovery;
6391         }
6392         ret = nfsd4_create_callback_queue();
6393         if (ret)
6394                 goto out_free_laundry;
6395
6396         set_max_delegations();
6397
6398         return 0;
6399
6400 out_free_laundry:
6401         destroy_workqueue(laundry_wq);
6402 out_recovery:
6403         return ret;
6404 }
6405
6406 void
6407 nfs4_state_shutdown_net(struct net *net)
6408 {
6409         struct nfs4_delegation *dp = NULL;
6410         struct list_head *pos, *next, reaplist;
6411         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6412
6413         cancel_delayed_work_sync(&nn->laundromat_work);
6414         locks_end_grace(&nn->nfsd4_manager);
6415
6416         INIT_LIST_HEAD(&reaplist);
6417         spin_lock(&state_lock);
6418         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6419                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6420                 unhash_delegation_locked(dp);
6421                 list_add(&dp->dl_recall_lru, &reaplist);
6422         }
6423         spin_unlock(&state_lock);
6424         list_for_each_safe(pos, next, &reaplist) {
6425                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6426                 list_del_init(&dp->dl_recall_lru);
6427                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6428                 nfs4_put_stid(&dp->dl_stid);
6429         }
6430
6431         nfsd4_client_tracking_exit(net);
6432         nfs4_state_destroy_net(net);
6433 }
6434
6435 void
6436 nfs4_state_shutdown(void)
6437 {
6438         destroy_workqueue(laundry_wq);
6439         nfsd4_destroy_callback_queue();
6440 }
6441
6442 static void
6443 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6444 {
6445         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6446                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6447 }
6448
6449 static void
6450 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6451 {
6452         if (cstate->minorversion) {
6453                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6454                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6455         }
6456 }
6457
6458 void
6459 clear_current_stateid(struct nfsd4_compound_state *cstate)
6460 {
6461         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6462 }
6463
6464 /*
6465  * functions to set current state id
6466  */
6467 void
6468 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6469 {
6470         put_stateid(cstate, &odp->od_stateid);
6471 }
6472
6473 void
6474 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6475 {
6476         put_stateid(cstate, &open->op_stateid);
6477 }
6478
6479 void
6480 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6481 {
6482         put_stateid(cstate, &close->cl_stateid);
6483 }
6484
6485 void
6486 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6487 {
6488         put_stateid(cstate, &lock->lk_resp_stateid);
6489 }
6490
6491 /*
6492  * functions to consume current state id
6493  */
6494
6495 void
6496 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6497 {
6498         get_stateid(cstate, &odp->od_stateid);
6499 }
6500
6501 void
6502 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6503 {
6504         get_stateid(cstate, &drp->dr_stateid);
6505 }
6506
6507 void
6508 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6509 {
6510         get_stateid(cstate, &fsp->fr_stateid);
6511 }
6512
6513 void
6514 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6515 {
6516         get_stateid(cstate, &setattr->sa_stateid);
6517 }
6518
6519 void
6520 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6521 {
6522         get_stateid(cstate, &close->cl_stateid);
6523 }
6524
6525 void
6526 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6527 {
6528         get_stateid(cstate, &locku->lu_stateid);
6529 }
6530
6531 void
6532 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6533 {
6534         get_stateid(cstate, &read->rd_stateid);
6535 }
6536
6537 void
6538 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6539 {
6540         get_stateid(cstate, &write->wr_stateid);
6541 }