fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include <linux/jhash.h>
  45 #include "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51
  52 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  53
  54 #define all_ones {{~0,~0},~0}
  55 static const stateid_t one_stateid = {
  56         .si_generation = ~0,
  57         .si_opaque = all_ones,
  58 };
  59 static const stateid_t zero_stateid = {
  60         /* all fields zero */
  61 };
  62 static const stateid_t currentstateid = {
  63         .si_generation = 1,
  64 };
  65
  66 static u64 current_sessionid = 1;
  67
  68 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  69 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  70 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  71
  72 /* forward declarations */
  73 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  74 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  75
  76 /* Locking: */
  77
  78 /*
  79  * Currently used for the del_recall_lru and file hash table.  In an
  80  * effort to decrease the scope of the client_mutex, this spinlock may
  81  * eventually cover more:
  82  */
  83 static DEFINE_SPINLOCK(state_lock);
  84
  85 /*
  86  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  87  * the refcount on the open stateid to drop.
  88  */
  89 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
  90
  91 static struct kmem_cache *openowner_slab;
  92 static struct kmem_cache *lockowner_slab;
  93 static struct kmem_cache *file_slab;
  94 static struct kmem_cache *stateid_slab;
  95 static struct kmem_cache *deleg_slab;
  96
  97 static void free_session(struct nfsd4_session *);
  98
  99 static struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 100
 101 static bool is_session_dead(struct nfsd4_session *ses)
 102 {
 103         return ses->se_flags & NFS4_SESSION_DEAD;
 104 }
 105
 106 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 107 {
 108         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 109                 return nfserr_jukebox;
 110         ses->se_flags |= NFS4_SESSION_DEAD;
 111         return nfs_ok;
 112 }
 113
 114 static bool is_client_expired(struct nfs4_client *clp)
 115 {
 116         return clp->cl_time == 0;
 117 }
 118
 119 static __be32 get_client_locked(struct nfs4_client *clp)
 120 {
 121         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 122
 123         lockdep_assert_held(&nn->client_lock);
 124
 125         if (is_client_expired(clp))
 126                 return nfserr_expired;
 127         atomic_inc(&clp->cl_refcount);
 128         return nfs_ok;
 129 }
 130
 131 /* must be called under the client_lock */
 132 static inline void
 133 renew_client_locked(struct nfs4_client *clp)
 134 {
 135         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 136
 137         if (is_client_expired(clp)) {
 138                 WARN_ON(1);
 139                 printk("%s: client (clientid %08x/%08x) already expired\n",
 140                         __func__,
 141                         clp->cl_clientid.cl_boot,
 142                         clp->cl_clientid.cl_id);
 143                 return;
 144         }
 145
 146         dprintk("renewing client (clientid %08x/%08x)\n",
 147                         clp->cl_clientid.cl_boot,
 148                         clp->cl_clientid.cl_id);
 149         list_move_tail(&clp->cl_lru, &nn->client_lru);
 150         clp->cl_time = get_seconds();
 151 }
 152
 153 static inline void
 154 renew_client(struct nfs4_client *clp)
 155 {
 156         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 157
 158         spin_lock(&nn->client_lock);
 159         renew_client_locked(clp);
 160         spin_unlock(&nn->client_lock);
 161 }
 162
 163 static void put_client_renew_locked(struct nfs4_client *clp)
 164 {
 165         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 166
 167         lockdep_assert_held(&nn->client_lock);
 168
 169         if (!atomic_dec_and_test(&clp->cl_refcount))
 170                 return;
 171         if (!is_client_expired(clp))
 172                 renew_client_locked(clp);
 173 }
 174
 175 static void put_client_renew(struct nfs4_client *clp)
 176 {
 177         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 178
 179         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 180                 return;
 181         if (!is_client_expired(clp))
 182                 renew_client_locked(clp);
 183         spin_unlock(&nn->client_lock);
 184 }
 185
 186 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 187 {
 188         __be32 status;
 189
 190         if (is_session_dead(ses))
 191                 return nfserr_badsession;
 192         status = get_client_locked(ses->se_client);
 193         if (status)
 194                 return status;
 195         atomic_inc(&ses->se_ref);
 196         return nfs_ok;
 197 }
 198
 199 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 200 {
 201         struct nfs4_client *clp = ses->se_client;
 202         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 203
 204         lockdep_assert_held(&nn->client_lock);
 205
 206         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 207                 free_session(ses);
 208         put_client_renew_locked(clp);
 209 }
 210
 211 static void nfsd4_put_session(struct nfsd4_session *ses)
 212 {
 213         struct nfs4_client *clp = ses->se_client;
 214         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 215
 216         spin_lock(&nn->client_lock);
 217         nfsd4_put_session_locked(ses);
 218         spin_unlock(&nn->client_lock);
 219 }
 220
 221 static inline struct nfs4_stateowner *
 222 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 223 {
 224         atomic_inc(&sop->so_count);
 225         return sop;
 226 }
 227
 228 static int
 229 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 230 {
 231         return (sop->so_owner.len == owner->len) &&
 232                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 233 }
 234
 235 static struct nfs4_openowner *
 236 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 237                         struct nfs4_client *clp)
 238 {
 239         struct nfs4_stateowner *so;
 240
 241         lockdep_assert_held(&clp->cl_lock);
 242
 243         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 244                             so_strhash) {
 245                 if (!so->so_is_open_owner)
 246                         continue;
 247                 if (same_owner_str(so, &open->op_owner))
 248                         return openowner(nfs4_get_stateowner(so));
 249         }
 250         return NULL;
 251 }
 252
 253 static struct nfs4_openowner *
 254 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 255                         struct nfs4_client *clp)
 256 {
 257         struct nfs4_openowner *oo;
 258
 259         spin_lock(&clp->cl_lock);
 260         oo = find_openstateowner_str_locked(hashval, open, clp);
 261         spin_unlock(&clp->cl_lock);
 262         return oo;
 263 }
 264
 265 static inline u32
 266 opaque_hashval(const void *ptr, int nbytes)
 267 {
 268         unsigned char *cptr = (unsigned char *) ptr;
 269
 270         u32 x = 0;
 271         while (nbytes--) {
 272                 x *= 37;
 273                 x += *cptr++;
 274         }
 275         return x;
 276 }
 277
 278 static void nfsd4_free_file(struct nfs4_file *f)
 279 {
 280         kmem_cache_free(file_slab, f);
 281 }
 282
 283 static inline void
 284 put_nfs4_file(struct nfs4_file *fi)
 285 {
 286         might_lock(&state_lock);
 287
 288         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
 289                 hlist_del(&fi->fi_hash);
 290                 spin_unlock(&state_lock);
 291                 nfsd4_free_file(fi);
 292         }
 293 }
 294
 295 static inline void
 296 get_nfs4_file(struct nfs4_file *fi)
 297 {
 298         atomic_inc(&fi->fi_ref);
 299 }
 300
 301 static struct file *
 302 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 303 {
 304         if (f->fi_fds[oflag])
 305                 return get_file(f->fi_fds[oflag]);
 306         return NULL;
 307 }
 308
 309 static struct file *
 310 find_writeable_file_locked(struct nfs4_file *f)
 311 {
 312         struct file *ret;
 313
 314         lockdep_assert_held(&f->fi_lock);
 315
 316         ret = __nfs4_get_fd(f, O_WRONLY);
 317         if (!ret)
 318                 ret = __nfs4_get_fd(f, O_RDWR);
 319         return ret;
 320 }
 321
 322 static struct file *
 323 find_writeable_file(struct nfs4_file *f)
 324 {
 325         struct file *ret;
 326
 327         spin_lock(&f->fi_lock);
 328         ret = find_writeable_file_locked(f);
 329         spin_unlock(&f->fi_lock);
 330
 331         return ret;
 332 }
 333
 334 static struct file *find_readable_file_locked(struct nfs4_file *f)
 335 {
 336         struct file *ret;
 337
 338         lockdep_assert_held(&f->fi_lock);
 339
 340         ret = __nfs4_get_fd(f, O_RDONLY);
 341         if (!ret)
 342                 ret = __nfs4_get_fd(f, O_RDWR);
 343         return ret;
 344 }
 345
 346 static struct file *
 347 find_readable_file(struct nfs4_file *f)
 348 {
 349         struct file *ret;
 350
 351         spin_lock(&f->fi_lock);
 352         ret = find_readable_file_locked(f);
 353         spin_unlock(&f->fi_lock);
 354
 355         return ret;
 356 }
 357
 358 static struct file *
 359 find_any_file(struct nfs4_file *f)
 360 {
 361         struct file *ret;
 362
 363         spin_lock(&f->fi_lock);
 364         ret = __nfs4_get_fd(f, O_RDWR);
 365         if (!ret) {
 366                 ret = __nfs4_get_fd(f, O_WRONLY);
 367                 if (!ret)
 368                         ret = __nfs4_get_fd(f, O_RDONLY);
 369         }
 370         spin_unlock(&f->fi_lock);
 371         return ret;
 372 }
 373
 374 static atomic_long_t num_delegations;
 375 unsigned long max_delegations;
 376
 377 /*
 378  * Open owner state (share locks)
 379  */
 380
 381 /* hash tables for lock and open owners */
 382 #define OWNER_HASH_BITS              8
 383 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 384 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 385
 386 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 387 {
 388         unsigned int ret;
 389
 390         ret = opaque_hashval(ownername->data, ownername->len);
 391         return ret & OWNER_HASH_MASK;
 392 }
 393
 394 /* hash table for nfs4_file */
 395 #define FILE_HASH_BITS                   8
 396 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 397
 398 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 399 {
 400         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 401 }
 402
 403 static unsigned int file_hashval(struct knfsd_fh *fh)
 404 {
 405         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 406 }
 407
 408 static bool nfsd_fh_match(struct knfsd_fh *fh1, struct knfsd_fh *fh2)
 409 {
 410         return fh1->fh_size == fh2->fh_size &&
 411                 !memcmp(fh1->fh_base.fh_pad,
 412                                 fh2->fh_base.fh_pad,
 413                                 fh1->fh_size);
 414 }
 415
 416 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 417
 418 static void
 419 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 420 {
 421         lockdep_assert_held(&fp->fi_lock);
 422
 423         if (access & NFS4_SHARE_ACCESS_WRITE)
 424                 atomic_inc(&fp->fi_access[O_WRONLY]);
 425         if (access & NFS4_SHARE_ACCESS_READ)
 426                 atomic_inc(&fp->fi_access[O_RDONLY]);
 427 }
 428
 429 static __be32
 430 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 431 {
 432         lockdep_assert_held(&fp->fi_lock);
 433
 434         /* Does this access mode make sense? */
 435         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 436                 return nfserr_inval;
 437
 438         /* Does it conflict with a deny mode already set? */
 439         if ((access & fp->fi_share_deny) != 0)
 440                 return nfserr_share_denied;
 441
 442         __nfs4_file_get_access(fp, access);
 443         return nfs_ok;
 444 }
 445
 446 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 447 {
 448         /* Common case is that there is no deny mode. */
 449         if (deny) {
 450                 /* Does this deny mode make sense? */
 451                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 452                         return nfserr_inval;
 453
 454                 if ((deny & NFS4_SHARE_DENY_READ) &&
 455                     atomic_read(&fp->fi_access[O_RDONLY]))
 456                         return nfserr_share_denied;
 457
 458                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 459                     atomic_read(&fp->fi_access[O_WRONLY]))
 460                         return nfserr_share_denied;
 461         }
 462         return nfs_ok;
 463 }
 464
 465 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 466 {
 467         might_lock(&fp->fi_lock);
 468
 469         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 470                 struct file *f1 = NULL;
 471                 struct file *f2 = NULL;
 472
 473                 swap(f1, fp->fi_fds[oflag]);
 474                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 475                         swap(f2, fp->fi_fds[O_RDWR]);
 476                 spin_unlock(&fp->fi_lock);
 477                 if (f1)
 478                         fput(f1);
 479                 if (f2)
 480                         fput(f2);
 481         }
 482 }
 483
 484 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 485 {
 486         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 487
 488         if (access & NFS4_SHARE_ACCESS_WRITE)
 489                 __nfs4_file_put_access(fp, O_WRONLY);
 490         if (access & NFS4_SHARE_ACCESS_READ)
 491                 __nfs4_file_put_access(fp, O_RDONLY);
 492 }
 493
 494 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
 495                                          struct kmem_cache *slab)
 496 {
 497         struct nfs4_stid *stid;
 498         int new_id;
 499
 500         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 501         if (!stid)
 502                 return NULL;
 503
 504         idr_preload(GFP_KERNEL);
 505         spin_lock(&cl->cl_lock);
 506         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 507         spin_unlock(&cl->cl_lock);
 508         idr_preload_end();
 509         if (new_id < 0)
 510                 goto out_free;
 511         stid->sc_client = cl;
 512         stid->sc_stateid.si_opaque.so_id = new_id;
 513         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 514         /* Will be incremented before return to client: */
 515         atomic_set(&stid->sc_count, 1);
 516
 517         /*
 518          * It shouldn't be a problem to reuse an opaque stateid value.
 519          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 520          * example, a stray write retransmission could be accepted by
 521          * the server when it should have been rejected.  Therefore,
 522          * adopt a trick from the sctp code to attempt to maximize the
 523          * amount of time until an id is reused, by ensuring they always
 524          * "increase" (mod INT_MAX):
 525          */
 526         return stid;
 527 out_free:
 528         kmem_cache_free(slab, stid);
 529         return NULL;
 530 }
 531
 532 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 533 {
 534         struct nfs4_stid *stid;
 535         struct nfs4_ol_stateid *stp;
 536
 537         stid = nfs4_alloc_stid(clp, stateid_slab);
 538         if (!stid)
 539                 return NULL;
 540
 541         stp = openlockstateid(stid);
 542         stp->st_stid.sc_free = nfs4_free_ol_stateid;
 543         return stp;
 544 }
 545
 546 static void nfs4_free_deleg(struct nfs4_stid *stid)
 547 {
 548         kmem_cache_free(deleg_slab, stid);
 549         atomic_long_dec(&num_delegations);
 550 }
 551
 552 /*
 553  * When we recall a delegation, we should be careful not to hand it
 554  * out again straight away.
 555  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 556  * in which the filehandles of recalled delegations are "stored".
 557  * If a filehandle appear in either filter, a delegation is blocked.
 558  * When a delegation is recalled, the filehandle is stored in the "new"
 559  * filter.
 560  * Every 30 seconds we swap the filters and clear the "new" one,
 561  * unless both are empty of course.
 562  *
 563  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 564  * low 3 bytes as hash-table indices.
 565  *
 566  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 567  * is used to manage concurrent access.  Testing does not need the lock
 568  * except when swapping the two filters.
 569  */
 570 static DEFINE_SPINLOCK(blocked_delegations_lock);
 571 static struct bloom_pair {
 572         int     entries, old_entries;
 573         time_t  swap_time;
 574         int     new; /* index into 'set' */
 575         DECLARE_BITMAP(set[2], 256);
 576 } blocked_delegations;
 577
 578 static int delegation_blocked(struct knfsd_fh *fh)
 579 {
 580         u32 hash;
 581         struct bloom_pair *bd = &blocked_delegations;
 582
 583         if (bd->entries == 0)
 584                 return 0;
 585         if (seconds_since_boot() - bd->swap_time > 30) {
 586                 spin_lock(&blocked_delegations_lock);
 587                 if (seconds_since_boot() - bd->swap_time > 30) {
 588                         bd->entries -= bd->old_entries;
 589                         bd->old_entries = bd->entries;
 590                         memset(bd->set[bd->new], 0,
 591                                sizeof(bd->set[0]));
 592                         bd->new = 1-bd->new;
 593                         bd->swap_time = seconds_since_boot();
 594                 }
 595                 spin_unlock(&blocked_delegations_lock);
 596         }
 597         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 598         if (test_bit(hash&255, bd->set[0]) &&
 599             test_bit((hash>>8)&255, bd->set[0]) &&
 600             test_bit((hash>>16)&255, bd->set[0]))
 601                 return 1;
 602
 603         if (test_bit(hash&255, bd->set[1]) &&
 604             test_bit((hash>>8)&255, bd->set[1]) &&
 605             test_bit((hash>>16)&255, bd->set[1]))
 606                 return 1;
 607
 608         return 0;
 609 }
 610
 611 static void block_delegations(struct knfsd_fh *fh)
 612 {
 613         u32 hash;
 614         struct bloom_pair *bd = &blocked_delegations;
 615
 616         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 617
 618         spin_lock(&blocked_delegations_lock);
 619         __set_bit(hash&255, bd->set[bd->new]);
 620         __set_bit((hash>>8)&255, bd->set[bd->new]);
 621         __set_bit((hash>>16)&255, bd->set[bd->new]);
 622         if (bd->entries == 0)
 623                 bd->swap_time = seconds_since_boot();
 624         bd->entries += 1;
 625         spin_unlock(&blocked_delegations_lock);
 626 }
 627
 628 static struct nfs4_delegation *
 629 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh)
 630 {
 631         struct nfs4_delegation *dp;
 632         long n;
 633
 634         dprintk("NFSD alloc_init_deleg\n");
 635         n = atomic_long_inc_return(&num_delegations);
 636         if (n < 0 || n > max_delegations)
 637                 goto out_dec;
 638         if (delegation_blocked(&current_fh->fh_handle))
 639                 goto out_dec;
 640         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 641         if (dp == NULL)
 642                 goto out_dec;
 643
 644         dp->dl_stid.sc_free = nfs4_free_deleg;
 645         /*
 646          * delegation seqid's are never incremented.  The 4.1 special
 647          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 648          * 0 anyway just for consistency and use 1:
 649          */
 650         dp->dl_stid.sc_stateid.si_generation = 1;
 651         INIT_LIST_HEAD(&dp->dl_perfile);
 652         INIT_LIST_HEAD(&dp->dl_perclnt);
 653         INIT_LIST_HEAD(&dp->dl_recall_lru);
 654         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 655         dp->dl_retries = 1;
 656         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 657                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 658         return dp;
 659 out_dec:
 660         atomic_long_dec(&num_delegations);
 661         return NULL;
 662 }
 663
 664 void
 665 nfs4_put_stid(struct nfs4_stid *s)
 666 {
 667         struct nfs4_file *fp = s->sc_file;
 668         struct nfs4_client *clp = s->sc_client;
 669
 670         might_lock(&clp->cl_lock);
 671
 672         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 673                 wake_up_all(&close_wq);
 674                 return;
 675         }
 676         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 677         spin_unlock(&clp->cl_lock);
 678         s->sc_free(s);
 679         if (fp)
 680                 put_nfs4_file(fp);
 681 }
 682
 683 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 684 {
 685         struct file *filp = NULL;
 686
 687         spin_lock(&fp->fi_lock);
 688         if (fp->fi_deleg_file && atomic_dec_and_test(&fp->fi_delegees))
 689                 swap(filp, fp->fi_deleg_file);
 690         spin_unlock(&fp->fi_lock);
 691
 692         if (filp) {
 693                 vfs_setlease(filp, F_UNLCK, NULL, NULL);
 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         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_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_deleg_file = NULL;
3072         fp->fi_had_conflict = false;
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 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3363 {
3364         struct nfs4_delegation *dp = cb_to_delegation(cb);
3365         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3366                                           nfsd_net_id);
3367
3368         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3369
3370         /*
3371          * We can't do this in nfsd_break_deleg_cb because it is
3372          * already holding inode->i_lock.
3373          *
3374          * If the dl_time != 0, then we know that it has already been
3375          * queued for a lease break. Don't queue it again.
3376          */
3377         spin_lock(&state_lock);
3378         if (dp->dl_time == 0) {
3379                 dp->dl_time = get_seconds();
3380                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3381         }
3382         spin_unlock(&state_lock);
3383 }
3384
3385 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3386                 struct rpc_task *task)
3387 {
3388         struct nfs4_delegation *dp = cb_to_delegation(cb);
3389
3390         switch (task->tk_status) {
3391         case 0:
3392                 return 1;
3393         case -EBADHANDLE:
3394         case -NFS4ERR_BAD_STATEID:
3395                 /*
3396                  * Race: client probably got cb_recall before open reply
3397                  * granting delegation.
3398                  */
3399                 if (dp->dl_retries--) {
3400                         rpc_delay(task, 2 * HZ);
3401                         return 0;
3402                 }
3403                 /*FALLTHRU*/
3404         default:
3405                 return -1;
3406         }
3407 }
3408
3409 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3410 {
3411         struct nfs4_delegation *dp = cb_to_delegation(cb);
3412
3413         nfs4_put_stid(&dp->dl_stid);
3414 }
3415
3416 static struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3417         .prepare        = nfsd4_cb_recall_prepare,
3418         .done           = nfsd4_cb_recall_done,
3419         .release        = nfsd4_cb_recall_release,
3420 };
3421
3422 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3423 {
3424         /*
3425          * We're assuming the state code never drops its reference
3426          * without first removing the lease.  Since we're in this lease
3427          * callback (and since the lease code is serialized by the kernel
3428          * lock) we know the server hasn't removed the lease yet, we know
3429          * it's safe to take a reference.
3430          */
3431         atomic_inc(&dp->dl_stid.sc_count);
3432         nfsd4_run_cb(&dp->dl_recall);
3433 }
3434
3435 /* Called from break_lease() with i_lock held. */
3436 static bool
3437 nfsd_break_deleg_cb(struct file_lock *fl)
3438 {
3439         bool ret = false;
3440         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3441         struct nfs4_delegation *dp;
3442
3443         if (!fp) {
3444                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3445                 return ret;
3446         }
3447         if (fp->fi_had_conflict) {
3448                 WARN(1, "duplicate break on %p\n", fp);
3449                 return ret;
3450         }
3451         /*
3452          * We don't want the locks code to timeout the lease for us;
3453          * we'll remove it ourself if a delegation isn't returned
3454          * in time:
3455          */
3456         fl->fl_break_time = 0;
3457
3458         spin_lock(&fp->fi_lock);
3459         fp->fi_had_conflict = true;
3460         /*
3461          * If there are no delegations on the list, then return true
3462          * so that the lease code will go ahead and delete it.
3463          */
3464         if (list_empty(&fp->fi_delegations))
3465                 ret = true;
3466         else
3467                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3468                         nfsd_break_one_deleg(dp);
3469         spin_unlock(&fp->fi_lock);
3470         return ret;
3471 }
3472
3473 static int
3474 nfsd_change_deleg_cb(struct file_lock **onlist, int arg, struct list_head *dispose)
3475 {
3476         if (arg & F_UNLCK)
3477                 return lease_modify(onlist, arg, dispose);
3478         else
3479                 return -EAGAIN;
3480 }
3481
3482 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3483         .lm_break = nfsd_break_deleg_cb,
3484         .lm_change = nfsd_change_deleg_cb,
3485 };
3486
3487 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3488 {
3489         if (nfsd4_has_session(cstate))
3490                 return nfs_ok;
3491         if (seqid == so->so_seqid - 1)
3492                 return nfserr_replay_me;
3493         if (seqid == so->so_seqid)
3494                 return nfs_ok;
3495         return nfserr_bad_seqid;
3496 }
3497
3498 static __be32 lookup_clientid(clientid_t *clid,
3499                 struct nfsd4_compound_state *cstate,
3500                 struct nfsd_net *nn)
3501 {
3502         struct nfs4_client *found;
3503
3504         if (cstate->clp) {
3505                 found = cstate->clp;
3506                 if (!same_clid(&found->cl_clientid, clid))
3507                         return nfserr_stale_clientid;
3508                 return nfs_ok;
3509         }
3510
3511         if (STALE_CLIENTID(clid, nn))
3512                 return nfserr_stale_clientid;
3513
3514         /*
3515          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3516          * cached already then we know this is for is for v4.0 and "sessions"
3517          * will be false.
3518          */
3519         WARN_ON_ONCE(cstate->session);
3520         spin_lock(&nn->client_lock);
3521         found = find_confirmed_client(clid, false, nn);
3522         if (!found) {
3523                 spin_unlock(&nn->client_lock);
3524                 return nfserr_expired;
3525         }
3526         atomic_inc(&found->cl_refcount);
3527         spin_unlock(&nn->client_lock);
3528
3529         /* Cache the nfs4_client in cstate! */
3530         cstate->clp = found;
3531         return nfs_ok;
3532 }
3533
3534 __be32
3535 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3536                     struct nfsd4_open *open, struct nfsd_net *nn)
3537 {
3538         clientid_t *clientid = &open->op_clientid;
3539         struct nfs4_client *clp = NULL;
3540         unsigned int strhashval;
3541         struct nfs4_openowner *oo = NULL;
3542         __be32 status;
3543
3544         if (STALE_CLIENTID(&open->op_clientid, nn))
3545                 return nfserr_stale_clientid;
3546         /*
3547          * In case we need it later, after we've already created the
3548          * file and don't want to risk a further failure:
3549          */
3550         open->op_file = nfsd4_alloc_file();
3551         if (open->op_file == NULL)
3552                 return nfserr_jukebox;
3553
3554         status = lookup_clientid(clientid, cstate, nn);
3555         if (status)
3556                 return status;
3557         clp = cstate->clp;
3558
3559         strhashval = ownerstr_hashval(&open->op_owner);
3560         oo = find_openstateowner_str(strhashval, open, clp);
3561         open->op_openowner = oo;
3562         if (!oo) {
3563                 goto new_owner;
3564         }
3565         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3566                 /* Replace unconfirmed owners without checking for replay. */
3567                 release_openowner(oo);
3568                 open->op_openowner = NULL;
3569                 goto new_owner;
3570         }
3571         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3572         if (status)
3573                 return status;
3574         goto alloc_stateid;
3575 new_owner:
3576         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3577         if (oo == NULL)
3578                 return nfserr_jukebox;
3579         open->op_openowner = oo;
3580 alloc_stateid:
3581         open->op_stp = nfs4_alloc_open_stateid(clp);
3582         if (!open->op_stp)
3583                 return nfserr_jukebox;
3584         return nfs_ok;
3585 }
3586
3587 static inline __be32
3588 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3589 {
3590         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3591                 return nfserr_openmode;
3592         else
3593                 return nfs_ok;
3594 }
3595
3596 static int share_access_to_flags(u32 share_access)
3597 {
3598         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3599 }
3600
3601 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3602 {
3603         struct nfs4_stid *ret;
3604
3605         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3606         if (!ret)
3607                 return NULL;
3608         return delegstateid(ret);
3609 }
3610
3611 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3612 {
3613         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3614                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3615 }
3616
3617 static __be32
3618 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3619                 struct nfs4_delegation **dp)
3620 {
3621         int flags;
3622         __be32 status = nfserr_bad_stateid;
3623         struct nfs4_delegation *deleg;
3624
3625         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3626         if (deleg == NULL)
3627                 goto out;
3628         flags = share_access_to_flags(open->op_share_access);
3629         status = nfs4_check_delegmode(deleg, flags);
3630         if (status) {
3631                 nfs4_put_stid(&deleg->dl_stid);
3632                 goto out;
3633         }
3634         *dp = deleg;
3635 out:
3636         if (!nfsd4_is_deleg_cur(open))
3637                 return nfs_ok;
3638         if (status)
3639                 return status;
3640         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3641         return nfs_ok;
3642 }
3643
3644 static struct nfs4_ol_stateid *
3645 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3646 {
3647         struct nfs4_ol_stateid *local, *ret = NULL;
3648         struct nfs4_openowner *oo = open->op_openowner;
3649
3650         spin_lock(&fp->fi_lock);
3651         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3652                 /* ignore lock owners */
3653                 if (local->st_stateowner->so_is_open_owner == 0)
3654                         continue;
3655                 if (local->st_stateowner == &oo->oo_owner) {
3656                         ret = local;
3657                         atomic_inc(&ret->st_stid.sc_count);
3658                         break;
3659                 }
3660         }
3661         spin_unlock(&fp->fi_lock);
3662         return ret;
3663 }
3664
3665 static inline int nfs4_access_to_access(u32 nfs4_access)
3666 {
3667         int flags = 0;
3668
3669         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3670                 flags |= NFSD_MAY_READ;
3671         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3672                 flags |= NFSD_MAY_WRITE;
3673         return flags;
3674 }
3675
3676 static inline __be32
3677 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3678                 struct nfsd4_open *open)
3679 {
3680         struct iattr iattr = {
3681                 .ia_valid = ATTR_SIZE,
3682                 .ia_size = 0,
3683         };
3684         if (!open->op_truncate)
3685                 return 0;
3686         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3687                 return nfserr_inval;
3688         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3689 }
3690
3691 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3692                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3693                 struct nfsd4_open *open)
3694 {
3695         struct file *filp = NULL;
3696         __be32 status;
3697         int oflag = nfs4_access_to_omode(open->op_share_access);
3698         int access = nfs4_access_to_access(open->op_share_access);
3699         unsigned char old_access_bmap, old_deny_bmap;
3700
3701         spin_lock(&fp->fi_lock);
3702
3703         /*
3704          * Are we trying to set a deny mode that would conflict with
3705          * current access?
3706          */
3707         status = nfs4_file_check_deny(fp, open->op_share_deny);
3708         if (status != nfs_ok) {
3709                 spin_unlock(&fp->fi_lock);
3710                 goto out;
3711         }
3712
3713         /* set access to the file */
3714         status = nfs4_file_get_access(fp, open->op_share_access);
3715         if (status != nfs_ok) {
3716                 spin_unlock(&fp->fi_lock);
3717                 goto out;
3718         }
3719
3720         /* Set access bits in stateid */
3721         old_access_bmap = stp->st_access_bmap;
3722         set_access(open->op_share_access, stp);
3723
3724         /* Set new deny mask */
3725         old_deny_bmap = stp->st_deny_bmap;
3726         set_deny(open->op_share_deny, stp);
3727         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3728
3729         if (!fp->fi_fds[oflag]) {
3730                 spin_unlock(&fp->fi_lock);
3731                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3732                 if (status)
3733                         goto out_put_access;
3734                 spin_lock(&fp->fi_lock);
3735                 if (!fp->fi_fds[oflag]) {
3736                         fp->fi_fds[oflag] = filp;
3737                         filp = NULL;
3738                 }
3739         }
3740         spin_unlock(&fp->fi_lock);
3741         if (filp)
3742                 fput(filp);
3743
3744         status = nfsd4_truncate(rqstp, cur_fh, open);
3745         if (status)
3746                 goto out_put_access;
3747 out:
3748         return status;
3749 out_put_access:
3750         stp->st_access_bmap = old_access_bmap;
3751         nfs4_file_put_access(fp, open->op_share_access);
3752         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3753         goto out;
3754 }
3755
3756 static __be32
3757 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
3758 {
3759         __be32 status;
3760         unsigned char old_deny_bmap;
3761
3762         if (!test_access(open->op_share_access, stp))
3763                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3764
3765         /* test and set deny mode */
3766         spin_lock(&fp->fi_lock);
3767         status = nfs4_file_check_deny(fp, open->op_share_deny);
3768         if (status == nfs_ok) {
3769                 old_deny_bmap = stp->st_deny_bmap;
3770                 set_deny(open->op_share_deny, stp);
3771                 fp->fi_share_deny |=
3772                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3773         }
3774         spin_unlock(&fp->fi_lock);
3775
3776         if (status != nfs_ok)
3777                 return status;
3778
3779         status = nfsd4_truncate(rqstp, cur_fh, open);
3780         if (status != nfs_ok)
3781                 reset_union_bmap_deny(old_deny_bmap, stp);
3782         return status;
3783 }
3784
3785 static void
3786 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3787 {
3788         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3789 }
3790
3791 /* Should we give out recallable state?: */
3792 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3793 {
3794         if (clp->cl_cb_state == NFSD4_CB_UP)
3795                 return true;
3796         /*
3797          * In the sessions case, since we don't have to establish a
3798          * separate connection for callbacks, we assume it's OK
3799          * until we hear otherwise:
3800          */
3801         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3802 }
3803
3804 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
3805 {
3806         struct file_lock *fl;
3807
3808         fl = locks_alloc_lock();
3809         if (!fl)
3810                 return NULL;
3811         fl->fl_lmops = &nfsd_lease_mng_ops;
3812         fl->fl_flags = FL_DELEG;
3813         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3814         fl->fl_end = OFFSET_MAX;
3815         fl->fl_owner = (fl_owner_t)fp;
3816         fl->fl_pid = current->tgid;
3817         return fl;
3818 }
3819
3820 static int nfs4_setlease(struct nfs4_delegation *dp)
3821 {
3822         struct nfs4_file *fp = dp->dl_stid.sc_file;
3823         struct file_lock *fl, *ret;
3824         struct file *filp;
3825         int status = 0;
3826
3827         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
3828         if (!fl)
3829                 return -ENOMEM;
3830         filp = find_readable_file(fp);
3831         if (!filp) {
3832                 /* We should always have a readable file here */
3833                 WARN_ON_ONCE(1);
3834                 return -EBADF;
3835         }
3836         fl->fl_file = filp;
3837         ret = fl;
3838         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
3839         if (fl)
3840                 locks_free_lock(fl);
3841         if (status)
3842                 goto out_fput;
3843         spin_lock(&state_lock);
3844         spin_lock(&fp->fi_lock);
3845         /* Did the lease get broken before we took the lock? */
3846         status = -EAGAIN;
3847         if (fp->fi_had_conflict)
3848                 goto out_unlock;
3849         /* Race breaker */
3850         if (fp->fi_deleg_file) {
3851                 status = 0;
3852                 atomic_inc(&fp->fi_delegees);
3853                 hash_delegation_locked(dp, fp);
3854                 goto out_unlock;
3855         }
3856         fp->fi_deleg_file = filp;
3857         atomic_set(&fp->fi_delegees, 1);
3858         hash_delegation_locked(dp, fp);
3859         spin_unlock(&fp->fi_lock);
3860         spin_unlock(&state_lock);
3861         return 0;
3862 out_unlock:
3863         spin_unlock(&fp->fi_lock);
3864         spin_unlock(&state_lock);
3865 out_fput:
3866         fput(filp);
3867         return status;
3868 }
3869
3870 static struct nfs4_delegation *
3871 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
3872                     struct nfs4_file *fp)
3873 {
3874         int status;
3875         struct nfs4_delegation *dp;
3876
3877         if (fp->fi_had_conflict)
3878                 return ERR_PTR(-EAGAIN);
3879
3880         dp = alloc_init_deleg(clp, fh);
3881         if (!dp)
3882                 return ERR_PTR(-ENOMEM);
3883
3884         get_nfs4_file(fp);
3885         spin_lock(&state_lock);
3886         spin_lock(&fp->fi_lock);
3887         dp->dl_stid.sc_file = fp;
3888         if (!fp->fi_deleg_file) {
3889                 spin_unlock(&fp->fi_lock);
3890                 spin_unlock(&state_lock);
3891                 status = nfs4_setlease(dp);
3892                 goto out;
3893         }
3894         atomic_inc(&fp->fi_delegees);
3895         if (fp->fi_had_conflict) {
3896                 status = -EAGAIN;
3897                 goto out_unlock;
3898         }
3899         hash_delegation_locked(dp, fp);
3900         status = 0;
3901 out_unlock:
3902         spin_unlock(&fp->fi_lock);
3903         spin_unlock(&state_lock);
3904 out:
3905         if (status) {
3906                 nfs4_put_stid(&dp->dl_stid);
3907                 return ERR_PTR(status);
3908         }
3909         return dp;
3910 }
3911
3912 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3913 {
3914         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3915         if (status == -EAGAIN)
3916                 open->op_why_no_deleg = WND4_CONTENTION;
3917         else {
3918                 open->op_why_no_deleg = WND4_RESOURCE;
3919                 switch (open->op_deleg_want) {
3920                 case NFS4_SHARE_WANT_READ_DELEG:
3921                 case NFS4_SHARE_WANT_WRITE_DELEG:
3922                 case NFS4_SHARE_WANT_ANY_DELEG:
3923                         break;
3924                 case NFS4_SHARE_WANT_CANCEL:
3925                         open->op_why_no_deleg = WND4_CANCELLED;
3926                         break;
3927                 case NFS4_SHARE_WANT_NO_DELEG:
3928                         WARN_ON_ONCE(1);
3929                 }
3930         }
3931 }
3932
3933 /*
3934  * Attempt to hand out a delegation.
3935  *
3936  * Note we don't support write delegations, and won't until the vfs has
3937  * proper support for them.
3938  */
3939 static void
3940 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
3941                         struct nfs4_ol_stateid *stp)
3942 {
3943         struct nfs4_delegation *dp;
3944         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
3945         struct nfs4_client *clp = stp->st_stid.sc_client;
3946         int cb_up;
3947         int status = 0;
3948
3949         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3950         open->op_recall = 0;
3951         switch (open->op_claim_type) {
3952                 case NFS4_OPEN_CLAIM_PREVIOUS:
3953                         if (!cb_up)
3954                                 open->op_recall = 1;
3955                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3956                                 goto out_no_deleg;
3957                         break;
3958                 case NFS4_OPEN_CLAIM_NULL:
3959                 case NFS4_OPEN_CLAIM_FH:
3960                         /*
3961                          * Let's not give out any delegations till everyone's
3962                          * had the chance to reclaim theirs....
3963                          */
3964                         if (locks_in_grace(clp->net))
3965                                 goto out_no_deleg;
3966                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3967                                 goto out_no_deleg;
3968                         /*
3969                          * Also, if the file was opened for write or
3970                          * create, there's a good chance the client's
3971                          * about to write to it, resulting in an
3972                          * immediate recall (since we don't support
3973                          * write delegations):
3974                          */
3975                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3976                                 goto out_no_deleg;
3977                         if (open->op_create == NFS4_OPEN_CREATE)
3978                                 goto out_no_deleg;
3979                         break;
3980                 default:
3981                         goto out_no_deleg;
3982         }
3983         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file);
3984         if (IS_ERR(dp))
3985                 goto out_no_deleg;
3986
3987         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3988
3989         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3990                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3991         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3992         nfs4_put_stid(&dp->dl_stid);
3993         return;
3994 out_no_deleg:
3995         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3996         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3997             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3998                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3999                 open->op_recall = 1;
4000         }
4001
4002         /* 4.1 client asking for a delegation? */
4003         if (open->op_deleg_want)
4004                 nfsd4_open_deleg_none_ext(open, status);
4005         return;
4006 }
4007
4008 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4009                                         struct nfs4_delegation *dp)
4010 {
4011         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4012             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4013                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4014                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4015         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4016                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4017                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4018                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4019         }
4020         /* Otherwise the client must be confused wanting a delegation
4021          * it already has, therefore we don't return
4022          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4023          */
4024 }
4025
4026 __be32
4027 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4028 {
4029         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4030         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4031         struct nfs4_file *fp = NULL;
4032         struct nfs4_ol_stateid *stp = NULL;
4033         struct nfs4_delegation *dp = NULL;
4034         __be32 status;
4035
4036         /*
4037          * Lookup file; if found, lookup stateid and check open request,
4038          * and check for delegations in the process of being recalled.
4039          * If not found, create the nfs4_file struct
4040          */
4041         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4042         if (fp != open->op_file) {
4043                 status = nfs4_check_deleg(cl, open, &dp);
4044                 if (status)
4045                         goto out;
4046                 stp = nfsd4_find_existing_open(fp, open);
4047         } else {
4048                 open->op_file = NULL;
4049                 status = nfserr_bad_stateid;
4050                 if (nfsd4_is_deleg_cur(open))
4051                         goto out;
4052                 status = nfserr_jukebox;
4053         }
4054
4055         /*
4056          * OPEN the file, or upgrade an existing OPEN.
4057          * If truncate fails, the OPEN fails.
4058          */
4059         if (stp) {
4060                 /* Stateid was found, this is an OPEN upgrade */
4061                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4062                 if (status)
4063                         goto out;
4064         } else {
4065                 stp = open->op_stp;
4066                 open->op_stp = NULL;
4067                 init_open_stateid(stp, fp, open);
4068                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4069                 if (status) {
4070                         release_open_stateid(stp);
4071                         goto out;
4072                 }
4073         }
4074         update_stateid(&stp->st_stid.sc_stateid);
4075         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4076
4077         if (nfsd4_has_session(&resp->cstate)) {
4078                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4079                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4080                         open->op_why_no_deleg = WND4_NOT_WANTED;
4081                         goto nodeleg;
4082                 }
4083         }
4084
4085         /*
4086         * Attempt to hand out a delegation. No error return, because the
4087         * OPEN succeeds even if we fail.
4088         */
4089         nfs4_open_delegation(current_fh, open, stp);
4090 nodeleg:
4091         status = nfs_ok;
4092
4093         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4094                 STATEID_VAL(&stp->st_stid.sc_stateid));
4095 out:
4096         /* 4.1 client trying to upgrade/downgrade delegation? */
4097         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4098             open->op_deleg_want)
4099                 nfsd4_deleg_xgrade_none_ext(open, dp);
4100
4101         if (fp)
4102                 put_nfs4_file(fp);
4103         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4104                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
4105         /*
4106         * To finish the open response, we just need to set the rflags.
4107         */
4108         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4109         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4110             !nfsd4_has_session(&resp->cstate))
4111                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4112         if (dp)
4113                 nfs4_put_stid(&dp->dl_stid);
4114         if (stp)
4115                 nfs4_put_stid(&stp->st_stid);
4116
4117         return status;
4118 }
4119
4120 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4121                               struct nfsd4_open *open, __be32 status)
4122 {
4123         if (open->op_openowner) {
4124                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4125
4126                 nfsd4_cstate_assign_replay(cstate, so);
4127                 nfs4_put_stateowner(so);
4128         }
4129         if (open->op_file)
4130                 nfsd4_free_file(open->op_file);
4131         if (open->op_stp)
4132                 nfs4_put_stid(&open->op_stp->st_stid);
4133 }
4134
4135 __be32
4136 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4137             clientid_t *clid)
4138 {
4139         struct nfs4_client *clp;
4140         __be32 status;
4141         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4142
4143         dprintk("process_renew(%08x/%08x): starting\n",
4144                         clid->cl_boot, clid->cl_id);
4145         status = lookup_clientid(clid, cstate, nn);
4146         if (status)
4147                 goto out;
4148         clp = cstate->clp;
4149         status = nfserr_cb_path_down;
4150         if (!list_empty(&clp->cl_delegations)
4151                         && clp->cl_cb_state != NFSD4_CB_UP)
4152                 goto out;
4153         status = nfs_ok;
4154 out:
4155         return status;
4156 }
4157
4158 void
4159 nfsd4_end_grace(struct nfsd_net *nn)
4160 {
4161         /* do nothing if grace period already ended */
4162         if (nn->grace_ended)
4163                 return;
4164
4165         dprintk("NFSD: end of grace period\n");
4166         nn->grace_ended = true;
4167         /*
4168          * If the server goes down again right now, an NFSv4
4169          * client will still be allowed to reclaim after it comes back up,
4170          * even if it hasn't yet had a chance to reclaim state this time.
4171          *
4172          */
4173         nfsd4_record_grace_done(nn);
4174         /*
4175          * At this point, NFSv4 clients can still reclaim.  But if the
4176          * server crashes, any that have not yet reclaimed will be out
4177          * of luck on the next boot.
4178          *
4179          * (NFSv4.1+ clients are considered to have reclaimed once they
4180          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4181          * have reclaimed after their first OPEN.)
4182          */
4183         locks_end_grace(&nn->nfsd4_manager);
4184         /*
4185          * At this point, and once lockd and/or any other containers
4186          * exit their grace period, further reclaims will fail and
4187          * regular locking can resume.
4188          */
4189 }
4190
4191 static time_t
4192 nfs4_laundromat(struct nfsd_net *nn)
4193 {
4194         struct nfs4_client *clp;
4195         struct nfs4_openowner *oo;
4196         struct nfs4_delegation *dp;
4197         struct nfs4_ol_stateid *stp;
4198         struct list_head *pos, *next, reaplist;
4199         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4200         time_t t, new_timeo = nn->nfsd4_lease;
4201
4202         dprintk("NFSD: laundromat service - starting\n");
4203         nfsd4_end_grace(nn);
4204         INIT_LIST_HEAD(&reaplist);
4205         spin_lock(&nn->client_lock);
4206         list_for_each_safe(pos, next, &nn->client_lru) {
4207                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4208                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4209                         t = clp->cl_time - cutoff;
4210                         new_timeo = min(new_timeo, t);
4211                         break;
4212                 }
4213                 if (mark_client_expired_locked(clp)) {
4214                         dprintk("NFSD: client in use (clientid %08x)\n",
4215                                 clp->cl_clientid.cl_id);
4216                         continue;
4217                 }
4218                 list_add(&clp->cl_lru, &reaplist);
4219         }
4220         spin_unlock(&nn->client_lock);
4221         list_for_each_safe(pos, next, &reaplist) {
4222                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4223                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4224                         clp->cl_clientid.cl_id);
4225                 list_del_init(&clp->cl_lru);
4226                 expire_client(clp);
4227         }
4228         spin_lock(&state_lock);
4229         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4230                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4231                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
4232                         continue;
4233                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4234                         t = dp->dl_time - cutoff;
4235                         new_timeo = min(new_timeo, t);
4236                         break;
4237                 }
4238                 unhash_delegation_locked(dp);
4239                 list_add(&dp->dl_recall_lru, &reaplist);
4240         }
4241         spin_unlock(&state_lock);
4242         while (!list_empty(&reaplist)) {
4243                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4244                                         dl_recall_lru);
4245                 list_del_init(&dp->dl_recall_lru);
4246                 revoke_delegation(dp);
4247         }
4248
4249         spin_lock(&nn->client_lock);
4250         while (!list_empty(&nn->close_lru)) {
4251                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4252                                         oo_close_lru);
4253                 if (time_after((unsigned long)oo->oo_time,
4254                                (unsigned long)cutoff)) {
4255                         t = oo->oo_time - cutoff;
4256                         new_timeo = min(new_timeo, t);
4257                         break;
4258                 }
4259                 list_del_init(&oo->oo_close_lru);
4260                 stp = oo->oo_last_closed_stid;
4261                 oo->oo_last_closed_stid = NULL;
4262                 spin_unlock(&nn->client_lock);
4263                 nfs4_put_stid(&stp->st_stid);
4264                 spin_lock(&nn->client_lock);
4265         }
4266         spin_unlock(&nn->client_lock);
4267
4268         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4269         return new_timeo;
4270 }
4271
4272 static struct workqueue_struct *laundry_wq;
4273 static void laundromat_main(struct work_struct *);
4274
4275 static void
4276 laundromat_main(struct work_struct *laundry)
4277 {
4278         time_t t;
4279         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
4280                                                   work);
4281         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4282                                            laundromat_work);
4283
4284         t = nfs4_laundromat(nn);
4285         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4286         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4287 }
4288
4289 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
4290 {
4291         if (!nfsd_fh_match(&fhp->fh_handle, &stp->st_stid.sc_file->fi_fhandle))
4292                 return nfserr_bad_stateid;
4293         return nfs_ok;
4294 }
4295
4296 static inline int
4297 access_permit_read(struct nfs4_ol_stateid *stp)
4298 {
4299         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4300                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4301                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4302 }
4303
4304 static inline int
4305 access_permit_write(struct nfs4_ol_stateid *stp)
4306 {
4307         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4308                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4309 }
4310
4311 static
4312 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4313 {
4314         __be32 status = nfserr_openmode;
4315
4316         /* For lock stateid's, we test the parent open, not the lock: */
4317         if (stp->st_openstp)
4318                 stp = stp->st_openstp;
4319         if ((flags & WR_STATE) && !access_permit_write(stp))
4320                 goto out;
4321         if ((flags & RD_STATE) && !access_permit_read(stp))
4322                 goto out;
4323         status = nfs_ok;
4324 out:
4325         return status;
4326 }
4327
4328 static inline __be32
4329 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4330 {
4331         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4332                 return nfs_ok;
4333         else if (locks_in_grace(net)) {
4334                 /* Answer in remaining cases depends on existence of
4335                  * conflicting state; so we must wait out the grace period. */
4336                 return nfserr_grace;
4337         } else if (flags & WR_STATE)
4338                 return nfs4_share_conflict(current_fh,
4339                                 NFS4_SHARE_DENY_WRITE);
4340         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4341                 return nfs4_share_conflict(current_fh,
4342                                 NFS4_SHARE_DENY_READ);
4343 }
4344
4345 /*
4346  * Allow READ/WRITE during grace period on recovered state only for files
4347  * that are not able to provide mandatory locking.
4348  */
4349 static inline int
4350 grace_disallows_io(struct net *net, struct inode *inode)
4351 {
4352         return locks_in_grace(net) && mandatory_lock(inode);
4353 }
4354
4355 /* Returns true iff a is later than b: */
4356 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4357 {
4358         return (s32)(a->si_generation - b->si_generation) > 0;
4359 }
4360
4361 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4362 {
4363         /*
4364          * When sessions are used the stateid generation number is ignored
4365          * when it is zero.
4366          */
4367         if (has_session && in->si_generation == 0)
4368                 return nfs_ok;
4369
4370         if (in->si_generation == ref->si_generation)
4371                 return nfs_ok;
4372
4373         /* If the client sends us a stateid from the future, it's buggy: */
4374         if (stateid_generation_after(in, ref))
4375                 return nfserr_bad_stateid;
4376         /*
4377          * However, we could see a stateid from the past, even from a
4378          * non-buggy client.  For example, if the client sends a lock
4379          * while some IO is outstanding, the lock may bump si_generation
4380          * while the IO is still in flight.  The client could avoid that
4381          * situation by waiting for responses on all the IO requests,
4382          * but better performance may result in retrying IO that
4383          * receives an old_stateid error if requests are rarely
4384          * reordered in flight:
4385          */
4386         return nfserr_old_stateid;
4387 }
4388
4389 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4390 {
4391         struct nfs4_stid *s;
4392         struct nfs4_ol_stateid *ols;
4393         __be32 status = nfserr_bad_stateid;
4394
4395         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4396                 return status;
4397         /* Client debugging aid. */
4398         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4399                 char addr_str[INET6_ADDRSTRLEN];
4400                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4401                                  sizeof(addr_str));
4402                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4403                                         "with incorrect client ID\n", addr_str);
4404                 return status;
4405         }
4406         spin_lock(&cl->cl_lock);
4407         s = find_stateid_locked(cl, stateid);
4408         if (!s)
4409                 goto out_unlock;
4410         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4411         if (status)
4412                 goto out_unlock;
4413         switch (s->sc_type) {
4414         case NFS4_DELEG_STID:
4415                 status = nfs_ok;
4416                 break;
4417         case NFS4_REVOKED_DELEG_STID:
4418                 status = nfserr_deleg_revoked;
4419                 break;
4420         case NFS4_OPEN_STID:
4421         case NFS4_LOCK_STID:
4422                 ols = openlockstateid(s);
4423                 if (ols->st_stateowner->so_is_open_owner
4424                                 && !(openowner(ols->st_stateowner)->oo_flags
4425                                                 & NFS4_OO_CONFIRMED))
4426                         status = nfserr_bad_stateid;
4427                 else
4428                         status = nfs_ok;
4429                 break;
4430         default:
4431                 printk("unknown stateid type %x\n", s->sc_type);
4432                 /* Fallthrough */
4433         case NFS4_CLOSED_STID:
4434         case NFS4_CLOSED_DELEG_STID:
4435                 status = nfserr_bad_stateid;
4436         }
4437 out_unlock:
4438         spin_unlock(&cl->cl_lock);
4439         return status;
4440 }
4441
4442 static __be32
4443 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4444                      stateid_t *stateid, unsigned char typemask,
4445                      struct nfs4_stid **s, struct nfsd_net *nn)
4446 {
4447         __be32 status;
4448
4449         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4450                 return nfserr_bad_stateid;
4451         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4452         if (status == nfserr_stale_clientid) {
4453                 if (cstate->session)
4454                         return nfserr_bad_stateid;
4455                 return nfserr_stale_stateid;
4456         }
4457         if (status)
4458                 return status;
4459         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4460         if (!*s)
4461                 return nfserr_bad_stateid;
4462         return nfs_ok;
4463 }
4464
4465 /*
4466 * Checks for stateid operations
4467 */
4468 __be32
4469 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
4470                            stateid_t *stateid, int flags, struct file **filpp)
4471 {
4472         struct nfs4_stid *s;
4473         struct nfs4_ol_stateid *stp = NULL;
4474         struct nfs4_delegation *dp = NULL;
4475         struct svc_fh *current_fh = &cstate->current_fh;
4476         struct inode *ino = current_fh->fh_dentry->d_inode;
4477         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4478         struct file *file = NULL;
4479         __be32 status;
4480
4481         if (filpp)
4482                 *filpp = NULL;
4483
4484         if (grace_disallows_io(net, ino))
4485                 return nfserr_grace;
4486
4487         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4488                 return check_special_stateids(net, current_fh, stateid, flags);
4489
4490         status = nfsd4_lookup_stateid(cstate, stateid,
4491                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4492                                 &s, nn);
4493         if (status)
4494                 return status;
4495         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
4496         if (status)
4497                 goto out;
4498         switch (s->sc_type) {
4499         case NFS4_DELEG_STID:
4500                 dp = delegstateid(s);
4501                 status = nfs4_check_delegmode(dp, flags);
4502                 if (status)
4503                         goto out;
4504                 if (filpp) {
4505                         file = dp->dl_stid.sc_file->fi_deleg_file;
4506                         if (!file) {
4507                                 WARN_ON_ONCE(1);
4508                                 status = nfserr_serverfault;
4509                                 goto out;
4510                         }
4511                         get_file(file);
4512                 }
4513                 break;
4514         case NFS4_OPEN_STID:
4515         case NFS4_LOCK_STID:
4516                 stp = openlockstateid(s);
4517                 status = nfs4_check_fh(current_fh, stp);
4518                 if (status)
4519                         goto out;
4520                 if (stp->st_stateowner->so_is_open_owner
4521                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4522                         goto out;
4523                 status = nfs4_check_openmode(stp, flags);
4524                 if (status)
4525                         goto out;
4526                 if (filpp) {
4527                         struct nfs4_file *fp = stp->st_stid.sc_file;
4528
4529                         if (flags & RD_STATE)
4530                                 file = find_readable_file(fp);
4531                         else
4532                                 file = find_writeable_file(fp);
4533                 }
4534                 break;
4535         default:
4536                 status = nfserr_bad_stateid;
4537                 goto out;
4538         }
4539         status = nfs_ok;
4540         if (file)
4541                 *filpp = file;
4542 out:
4543         nfs4_put_stid(s);
4544         return status;
4545 }
4546
4547 /*
4548  * Test if the stateid is valid
4549  */
4550 __be32
4551 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4552                    struct nfsd4_test_stateid *test_stateid)
4553 {
4554         struct nfsd4_test_stateid_id *stateid;
4555         struct nfs4_client *cl = cstate->session->se_client;
4556
4557         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4558                 stateid->ts_id_status =
4559                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4560
4561         return nfs_ok;
4562 }
4563
4564 __be32
4565 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4566                    struct nfsd4_free_stateid *free_stateid)
4567 {
4568         stateid_t *stateid = &free_stateid->fr_stateid;
4569         struct nfs4_stid *s;
4570         struct nfs4_delegation *dp;
4571         struct nfs4_ol_stateid *stp;
4572         struct nfs4_client *cl = cstate->session->se_client;
4573         __be32 ret = nfserr_bad_stateid;
4574
4575         spin_lock(&cl->cl_lock);
4576         s = find_stateid_locked(cl, stateid);
4577         if (!s)
4578                 goto out_unlock;
4579         switch (s->sc_type) {
4580         case NFS4_DELEG_STID:
4581                 ret = nfserr_locks_held;
4582                 break;
4583         case NFS4_OPEN_STID:
4584                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4585                 if (ret)
4586                         break;
4587                 ret = nfserr_locks_held;
4588                 break;
4589         case NFS4_LOCK_STID:
4590                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4591                 if (ret)
4592                         break;
4593                 stp = openlockstateid(s);
4594                 ret = nfserr_locks_held;
4595                 if (check_for_locks(stp->st_stid.sc_file,
4596                                     lockowner(stp->st_stateowner)))
4597                         break;
4598                 unhash_lock_stateid(stp);
4599                 spin_unlock(&cl->cl_lock);
4600                 nfs4_put_stid(s);
4601                 ret = nfs_ok;
4602                 goto out;
4603         case NFS4_REVOKED_DELEG_STID:
4604                 dp = delegstateid(s);
4605                 list_del_init(&dp->dl_recall_lru);
4606                 spin_unlock(&cl->cl_lock);
4607                 nfs4_put_stid(s);
4608                 ret = nfs_ok;
4609                 goto out;
4610         /* Default falls through and returns nfserr_bad_stateid */
4611         }
4612 out_unlock:
4613         spin_unlock(&cl->cl_lock);
4614 out:
4615         return ret;
4616 }
4617
4618 static inline int
4619 setlkflg (int type)
4620 {
4621         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4622                 RD_STATE : WR_STATE;
4623 }
4624
4625 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4626 {
4627         struct svc_fh *current_fh = &cstate->current_fh;
4628         struct nfs4_stateowner *sop = stp->st_stateowner;
4629         __be32 status;
4630
4631         status = nfsd4_check_seqid(cstate, sop, seqid);
4632         if (status)
4633                 return status;
4634         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4635                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4636                 /*
4637                  * "Closed" stateid's exist *only* to return
4638                  * nfserr_replay_me from the previous step, and
4639                  * revoked delegations are kept only for free_stateid.
4640                  */
4641                 return nfserr_bad_stateid;
4642         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4643         if (status)
4644                 return status;
4645         return nfs4_check_fh(current_fh, stp);
4646 }
4647
4648 /*
4649  * Checks for sequence id mutating operations.
4650  */
4651 static __be32
4652 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4653                          stateid_t *stateid, char typemask,
4654                          struct nfs4_ol_stateid **stpp,
4655                          struct nfsd_net *nn)
4656 {
4657         __be32 status;
4658         struct nfs4_stid *s;
4659         struct nfs4_ol_stateid *stp = NULL;
4660
4661         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4662                 seqid, STATEID_VAL(stateid));
4663
4664         *stpp = NULL;
4665         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4666         if (status)
4667                 return status;
4668         stp = openlockstateid(s);
4669         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
4670
4671         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4672         if (!status)
4673                 *stpp = stp;
4674         else
4675                 nfs4_put_stid(&stp->st_stid);
4676         return status;
4677 }
4678
4679 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4680                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4681 {
4682         __be32 status;
4683         struct nfs4_openowner *oo;
4684         struct nfs4_ol_stateid *stp;
4685
4686         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4687                                                 NFS4_OPEN_STID, &stp, nn);
4688         if (status)
4689                 return status;
4690         oo = openowner(stp->st_stateowner);
4691         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4692                 nfs4_put_stid(&stp->st_stid);
4693                 return nfserr_bad_stateid;
4694         }
4695         *stpp = stp;
4696         return nfs_ok;
4697 }
4698
4699 __be32
4700 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4701                    struct nfsd4_open_confirm *oc)
4702 {
4703         __be32 status;
4704         struct nfs4_openowner *oo;
4705         struct nfs4_ol_stateid *stp;
4706         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4707
4708         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4709                         cstate->current_fh.fh_dentry);
4710
4711         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4712         if (status)
4713                 return status;
4714
4715         status = nfs4_preprocess_seqid_op(cstate,
4716                                         oc->oc_seqid, &oc->oc_req_stateid,
4717                                         NFS4_OPEN_STID, &stp, nn);
4718         if (status)
4719                 goto out;
4720         oo = openowner(stp->st_stateowner);
4721         status = nfserr_bad_stateid;
4722         if (oo->oo_flags & NFS4_OO_CONFIRMED)
4723                 goto put_stateid;
4724         oo->oo_flags |= NFS4_OO_CONFIRMED;
4725         update_stateid(&stp->st_stid.sc_stateid);
4726         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4727         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
4728                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
4729
4730         nfsd4_client_record_create(oo->oo_owner.so_client);
4731         status = nfs_ok;
4732 put_stateid:
4733         nfs4_put_stid(&stp->st_stid);
4734 out:
4735         nfsd4_bump_seqid(cstate, status);
4736         return status;
4737 }
4738
4739 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
4740 {
4741         if (!test_access(access, stp))
4742                 return;
4743         nfs4_file_put_access(stp->st_stid.sc_file, access);
4744         clear_access(access, stp);
4745 }
4746
4747 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
4748 {
4749         switch (to_access) {
4750         case NFS4_SHARE_ACCESS_READ:
4751                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
4752                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4753                 break;
4754         case NFS4_SHARE_ACCESS_WRITE:
4755                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
4756                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4757                 break;
4758         case NFS4_SHARE_ACCESS_BOTH:
4759                 break;
4760         default:
4761                 WARN_ON_ONCE(1);
4762         }
4763 }
4764
4765 __be32
4766 nfsd4_open_downgrade(struct svc_rqst *rqstp,
4767                      struct nfsd4_compound_state *cstate,
4768                      struct nfsd4_open_downgrade *od)
4769 {
4770         __be32 status;
4771         struct nfs4_ol_stateid *stp;
4772         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4773
4774         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
4775                         cstate->current_fh.fh_dentry);
4776
4777         /* We don't yet support WANT bits: */
4778         if (od->od_deleg_want)
4779                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
4780                         od->od_deleg_want);
4781
4782         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
4783                                         &od->od_stateid, &stp, nn);
4784         if (status)
4785                 goto out;
4786         status = nfserr_inval;
4787         if (!test_access(od->od_share_access, stp)) {
4788                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
4789                         stp->st_access_bmap, od->od_share_access);
4790                 goto put_stateid;
4791         }
4792         if (!test_deny(od->od_share_deny, stp)) {
4793                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
4794                         stp->st_deny_bmap, od->od_share_deny);
4795                 goto put_stateid;
4796         }
4797         nfs4_stateid_downgrade(stp, od->od_share_access);
4798
4799         reset_union_bmap_deny(od->od_share_deny, stp);
4800
4801         update_stateid(&stp->st_stid.sc_stateid);
4802         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4803         status = nfs_ok;
4804 put_stateid:
4805         nfs4_put_stid(&stp->st_stid);
4806 out:
4807         nfsd4_bump_seqid(cstate, status);
4808         return status;
4809 }
4810
4811 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
4812 {
4813         struct nfs4_client *clp = s->st_stid.sc_client;
4814         LIST_HEAD(reaplist);
4815
4816         s->st_stid.sc_type = NFS4_CLOSED_STID;
4817         spin_lock(&clp->cl_lock);
4818         unhash_open_stateid(s, &reaplist);
4819
4820         if (clp->cl_minorversion) {
4821                 put_ol_stateid_locked(s, &reaplist);
4822                 spin_unlock(&clp->cl_lock);
4823                 free_ol_stateid_reaplist(&reaplist);
4824         } else {
4825                 spin_unlock(&clp->cl_lock);
4826                 free_ol_stateid_reaplist(&reaplist);
4827                 move_to_close_lru(s, clp->net);
4828         }
4829 }
4830
4831 /*
4832  * nfs4_unlock_state() called after encode
4833  */
4834 __be32
4835 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4836             struct nfsd4_close *close)
4837 {
4838         __be32 status;
4839         struct nfs4_ol_stateid *stp;
4840         struct net *net = SVC_NET(rqstp);
4841         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4842
4843         dprintk("NFSD: nfsd4_close on file %pd\n",
4844                         cstate->current_fh.fh_dentry);
4845
4846         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4847                                         &close->cl_stateid,
4848                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4849                                         &stp, nn);
4850         nfsd4_bump_seqid(cstate, status);
4851         if (status)
4852                 goto out;
4853         update_stateid(&stp->st_stid.sc_stateid);
4854         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4855
4856         nfsd4_close_open_stateid(stp);
4857
4858         /* put reference from nfs4_preprocess_seqid_op */
4859         nfs4_put_stid(&stp->st_stid);
4860 out:
4861         return status;
4862 }
4863
4864 __be32
4865 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4866                   struct nfsd4_delegreturn *dr)
4867 {
4868         struct nfs4_delegation *dp;
4869         stateid_t *stateid = &dr->dr_stateid;
4870         struct nfs4_stid *s;
4871         __be32 status;
4872         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4873
4874         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4875                 return status;
4876
4877         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
4878         if (status)
4879                 goto out;
4880         dp = delegstateid(s);
4881         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4882         if (status)
4883                 goto put_stateid;
4884
4885         destroy_delegation(dp);
4886 put_stateid:
4887         nfs4_put_stid(&dp->dl_stid);
4888 out:
4889         return status;
4890 }
4891
4892
4893 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4894
4895 static inline u64
4896 end_offset(u64 start, u64 len)
4897 {
4898         u64 end;
4899
4900         end = start + len;
4901         return end >= start ? end: NFS4_MAX_UINT64;
4902 }
4903
4904 /* last octet in a range */
4905 static inline u64
4906 last_byte_offset(u64 start, u64 len)
4907 {
4908         u64 end;
4909
4910         WARN_ON_ONCE(!len);
4911         end = start + len;
4912         return end > start ? end - 1: NFS4_MAX_UINT64;
4913 }
4914
4915 /*
4916  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4917  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4918  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4919  * locking, this prevents us from being completely protocol-compliant.  The
4920  * real solution to this problem is to start using unsigned file offsets in
4921  * the VFS, but this is a very deep change!
4922  */
4923 static inline void
4924 nfs4_transform_lock_offset(struct file_lock *lock)
4925 {
4926         if (lock->fl_start < 0)
4927                 lock->fl_start = OFFSET_MAX;
4928         if (lock->fl_end < 0)
4929                 lock->fl_end = OFFSET_MAX;
4930 }
4931
4932 static void nfsd4_fl_get_owner(struct file_lock *dst, struct file_lock *src)
4933 {
4934         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)src->fl_owner;
4935         dst->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lo->lo_owner));
4936 }
4937
4938 static void nfsd4_fl_put_owner(struct file_lock *fl)
4939 {
4940         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
4941
4942         if (lo) {
4943                 nfs4_put_stateowner(&lo->lo_owner);
4944                 fl->fl_owner = NULL;
4945         }
4946 }
4947
4948 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4949         .lm_get_owner = nfsd4_fl_get_owner,
4950         .lm_put_owner = nfsd4_fl_put_owner,
4951 };
4952
4953 static inline void
4954 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4955 {
4956         struct nfs4_lockowner *lo;
4957
4958         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4959                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4960                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4961                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4962                 if (!deny->ld_owner.data)
4963                         /* We just don't care that much */
4964                         goto nevermind;
4965                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4966                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4967         } else {
4968 nevermind:
4969                 deny->ld_owner.len = 0;
4970                 deny->ld_owner.data = NULL;
4971                 deny->ld_clientid.cl_boot = 0;
4972                 deny->ld_clientid.cl_id = 0;
4973         }
4974         deny->ld_start = fl->fl_start;
4975         deny->ld_length = NFS4_MAX_UINT64;
4976         if (fl->fl_end != NFS4_MAX_UINT64)
4977                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4978         deny->ld_type = NFS4_READ_LT;
4979         if (fl->fl_type != F_RDLCK)
4980                 deny->ld_type = NFS4_WRITE_LT;
4981 }
4982
4983 static struct nfs4_lockowner *
4984 find_lockowner_str_locked(clientid_t *clid, struct xdr_netobj *owner,
4985                 struct nfs4_client *clp)
4986 {
4987         unsigned int strhashval = ownerstr_hashval(owner);
4988         struct nfs4_stateowner *so;
4989
4990         lockdep_assert_held(&clp->cl_lock);
4991
4992         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
4993                             so_strhash) {
4994                 if (so->so_is_open_owner)
4995                         continue;
4996                 if (same_owner_str(so, owner))
4997                         return lockowner(nfs4_get_stateowner(so));
4998         }
4999         return NULL;
5000 }
5001
5002 static struct nfs4_lockowner *
5003 find_lockowner_str(clientid_t *clid, struct xdr_netobj *owner,
5004                 struct nfs4_client *clp)
5005 {
5006         struct nfs4_lockowner *lo;
5007
5008         spin_lock(&clp->cl_lock);
5009         lo = find_lockowner_str_locked(clid, owner, clp);
5010         spin_unlock(&clp->cl_lock);
5011         return lo;
5012 }
5013
5014 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5015 {
5016         unhash_lockowner_locked(lockowner(sop));
5017 }
5018
5019 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5020 {
5021         struct nfs4_lockowner *lo = lockowner(sop);
5022
5023         kmem_cache_free(lockowner_slab, lo);
5024 }
5025
5026 static const struct nfs4_stateowner_operations lockowner_ops = {
5027         .so_unhash =    nfs4_unhash_lockowner,
5028         .so_free =      nfs4_free_lockowner,
5029 };
5030
5031 /*
5032  * Alloc a lock owner structure.
5033  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5034  * occurred.
5035  *
5036  * strhashval = ownerstr_hashval
5037  */
5038 static struct nfs4_lockowner *
5039 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5040                            struct nfs4_ol_stateid *open_stp,
5041                            struct nfsd4_lock *lock)
5042 {
5043         struct nfs4_lockowner *lo, *ret;
5044
5045         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5046         if (!lo)
5047                 return NULL;
5048         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5049         lo->lo_owner.so_is_open_owner = 0;
5050         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5051         lo->lo_owner.so_ops = &lockowner_ops;
5052         spin_lock(&clp->cl_lock);
5053         ret = find_lockowner_str_locked(&clp->cl_clientid,
5054                         &lock->lk_new_owner, clp);
5055         if (ret == NULL) {
5056                 list_add(&lo->lo_owner.so_strhash,
5057                          &clp->cl_ownerstr_hashtbl[strhashval]);
5058                 ret = lo;
5059         } else
5060                 nfs4_free_lockowner(&lo->lo_owner);
5061         spin_unlock(&clp->cl_lock);
5062         return lo;
5063 }
5064
5065 static void
5066 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5067                   struct nfs4_file *fp, struct inode *inode,
5068                   struct nfs4_ol_stateid *open_stp)
5069 {
5070         struct nfs4_client *clp = lo->lo_owner.so_client;
5071
5072         lockdep_assert_held(&clp->cl_lock);
5073
5074         atomic_inc(&stp->st_stid.sc_count);
5075         stp->st_stid.sc_type = NFS4_LOCK_STID;
5076         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5077         get_nfs4_file(fp);
5078         stp->st_stid.sc_file = fp;
5079         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5080         stp->st_access_bmap = 0;
5081         stp->st_deny_bmap = open_stp->st_deny_bmap;
5082         stp->st_openstp = open_stp;
5083         list_add(&stp->st_locks, &open_stp->st_locks);
5084         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5085         spin_lock(&fp->fi_lock);
5086         list_add(&stp->st_perfile, &fp->fi_stateids);
5087         spin_unlock(&fp->fi_lock);
5088 }
5089
5090 static struct nfs4_ol_stateid *
5091 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5092 {
5093         struct nfs4_ol_stateid *lst;
5094         struct nfs4_client *clp = lo->lo_owner.so_client;
5095
5096         lockdep_assert_held(&clp->cl_lock);
5097
5098         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5099                 if (lst->st_stid.sc_file == fp) {
5100                         atomic_inc(&lst->st_stid.sc_count);
5101                         return lst;
5102                 }
5103         }
5104         return NULL;
5105 }
5106
5107 static struct nfs4_ol_stateid *
5108 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5109                             struct inode *inode, struct nfs4_ol_stateid *ost,
5110                             bool *new)
5111 {
5112         struct nfs4_stid *ns = NULL;
5113         struct nfs4_ol_stateid *lst;
5114         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5115         struct nfs4_client *clp = oo->oo_owner.so_client;
5116
5117         spin_lock(&clp->cl_lock);
5118         lst = find_lock_stateid(lo, fi);
5119         if (lst == NULL) {
5120                 spin_unlock(&clp->cl_lock);
5121                 ns = nfs4_alloc_stid(clp, stateid_slab);
5122                 if (ns == NULL)
5123                         return NULL;
5124
5125                 spin_lock(&clp->cl_lock);
5126                 lst = find_lock_stateid(lo, fi);
5127                 if (likely(!lst)) {
5128                         lst = openlockstateid(ns);
5129                         init_lock_stateid(lst, lo, fi, inode, ost);
5130                         ns = NULL;
5131                         *new = true;
5132                 }
5133         }
5134         spin_unlock(&clp->cl_lock);
5135         if (ns)
5136                 nfs4_put_stid(ns);
5137         return lst;
5138 }
5139
5140 static int
5141 check_lock_length(u64 offset, u64 length)
5142 {
5143         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
5144              LOFF_OVERFLOW(offset, length)));
5145 }
5146
5147 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5148 {
5149         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5150
5151         lockdep_assert_held(&fp->fi_lock);
5152
5153         if (test_access(access, lock_stp))
5154                 return;
5155         __nfs4_file_get_access(fp, access);
5156         set_access(access, lock_stp);
5157 }
5158
5159 static __be32
5160 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5161                             struct nfs4_ol_stateid *ost,
5162                             struct nfsd4_lock *lock,
5163                             struct nfs4_ol_stateid **lst, bool *new)
5164 {
5165         __be32 status;
5166         struct nfs4_file *fi = ost->st_stid.sc_file;
5167         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5168         struct nfs4_client *cl = oo->oo_owner.so_client;
5169         struct inode *inode = cstate->current_fh.fh_dentry->d_inode;
5170         struct nfs4_lockowner *lo;
5171         unsigned int strhashval;
5172
5173         lo = find_lockowner_str(&cl->cl_clientid, &lock->v.new.owner, cl);
5174         if (!lo) {
5175                 strhashval = ownerstr_hashval(&lock->v.new.owner);
5176                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5177                 if (lo == NULL)
5178                         return nfserr_jukebox;
5179         } else {
5180                 /* with an existing lockowner, seqids must be the same */
5181                 status = nfserr_bad_seqid;
5182                 if (!cstate->minorversion &&
5183                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5184                         goto out;
5185         }
5186
5187         *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5188         if (*lst == NULL) {
5189                 status = nfserr_jukebox;
5190                 goto out;
5191         }
5192         status = nfs_ok;
5193 out:
5194         nfs4_put_stateowner(&lo->lo_owner);
5195         return status;
5196 }
5197
5198 /*
5199  *  LOCK operation
5200  */
5201 __be32
5202 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5203            struct nfsd4_lock *lock)
5204 {
5205         struct nfs4_openowner *open_sop = NULL;
5206         struct nfs4_lockowner *lock_sop = NULL;
5207         struct nfs4_ol_stateid *lock_stp = NULL;
5208         struct nfs4_ol_stateid *open_stp = NULL;
5209         struct nfs4_file *fp;
5210         struct file *filp = NULL;
5211         struct file_lock *file_lock = NULL;
5212         struct file_lock *conflock = NULL;
5213         __be32 status = 0;
5214         int lkflg;
5215         int err;
5216         bool new = false;
5217         struct net *net = SVC_NET(rqstp);
5218         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5219
5220         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5221                 (long long) lock->lk_offset,
5222                 (long long) lock->lk_length);
5223
5224         if (check_lock_length(lock->lk_offset, lock->lk_length))
5225                  return nfserr_inval;
5226
5227         if ((status = fh_verify(rqstp, &cstate->current_fh,
5228                                 S_IFREG, NFSD_MAY_LOCK))) {
5229                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5230                 return status;
5231         }
5232
5233         if (lock->lk_is_new) {
5234                 if (nfsd4_has_session(cstate))
5235                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5236                         memcpy(&lock->v.new.clientid,
5237                                 &cstate->session->se_client->cl_clientid,
5238                                 sizeof(clientid_t));
5239
5240                 status = nfserr_stale_clientid;
5241                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5242                         goto out;
5243
5244                 /* validate and update open stateid and open seqid */
5245                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5246                                         lock->lk_new_open_seqid,
5247                                         &lock->lk_new_open_stateid,
5248                                         &open_stp, nn);
5249                 if (status)
5250                         goto out;
5251                 open_sop = openowner(open_stp->st_stateowner);
5252                 status = nfserr_bad_stateid;
5253                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5254                                                 &lock->v.new.clientid))
5255                         goto out;
5256                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5257                                                         &lock_stp, &new);
5258         } else {
5259                 status = nfs4_preprocess_seqid_op(cstate,
5260                                        lock->lk_old_lock_seqid,
5261                                        &lock->lk_old_lock_stateid,
5262                                        NFS4_LOCK_STID, &lock_stp, nn);
5263         }
5264         if (status)
5265                 goto out;
5266         lock_sop = lockowner(lock_stp->st_stateowner);
5267
5268         lkflg = setlkflg(lock->lk_type);
5269         status = nfs4_check_openmode(lock_stp, lkflg);
5270         if (status)
5271                 goto out;
5272
5273         status = nfserr_grace;
5274         if (locks_in_grace(net) && !lock->lk_reclaim)
5275                 goto out;
5276         status = nfserr_no_grace;
5277         if (!locks_in_grace(net) && lock->lk_reclaim)
5278                 goto out;
5279
5280         file_lock = locks_alloc_lock();
5281         if (!file_lock) {
5282                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5283                 status = nfserr_jukebox;
5284                 goto out;
5285         }
5286
5287         fp = lock_stp->st_stid.sc_file;
5288         switch (lock->lk_type) {
5289                 case NFS4_READ_LT:
5290                 case NFS4_READW_LT:
5291                         spin_lock(&fp->fi_lock);
5292                         filp = find_readable_file_locked(fp);
5293                         if (filp)
5294                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5295                         spin_unlock(&fp->fi_lock);
5296                         file_lock->fl_type = F_RDLCK;
5297                         break;
5298                 case NFS4_WRITE_LT:
5299                 case NFS4_WRITEW_LT:
5300                         spin_lock(&fp->fi_lock);
5301                         filp = find_writeable_file_locked(fp);
5302                         if (filp)
5303                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5304                         spin_unlock(&fp->fi_lock);
5305                         file_lock->fl_type = F_WRLCK;
5306                         break;
5307                 default:
5308                         status = nfserr_inval;
5309                 goto out;
5310         }
5311         if (!filp) {
5312                 status = nfserr_openmode;
5313                 goto out;
5314         }
5315
5316         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5317         file_lock->fl_pid = current->tgid;
5318         file_lock->fl_file = filp;
5319         file_lock->fl_flags = FL_POSIX;
5320         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5321         file_lock->fl_start = lock->lk_offset;
5322         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5323         nfs4_transform_lock_offset(file_lock);
5324
5325         conflock = locks_alloc_lock();
5326         if (!conflock) {
5327                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5328                 status = nfserr_jukebox;
5329                 goto out;
5330         }
5331
5332         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5333         switch (-err) {
5334         case 0: /* success! */
5335                 update_stateid(&lock_stp->st_stid.sc_stateid);
5336                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
5337                                 sizeof(stateid_t));
5338                 status = 0;
5339                 break;
5340         case (EAGAIN):          /* conflock holds conflicting lock */
5341                 status = nfserr_denied;
5342                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5343                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5344                 break;
5345         case (EDEADLK):
5346                 status = nfserr_deadlock;
5347                 break;
5348         default:
5349                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5350                 status = nfserrno(err);
5351                 break;
5352         }
5353 out:
5354         if (filp)
5355                 fput(filp);
5356         if (lock_stp) {
5357                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5358                 if (cstate->replay_owner &&
5359                     cstate->replay_owner != &lock_sop->lo_owner &&
5360                     seqid_mutating_err(ntohl(status)))
5361                         lock_sop->lo_owner.so_seqid++;
5362
5363                 /*
5364                  * If this is a new, never-before-used stateid, and we are
5365                  * returning an error, then just go ahead and release it.
5366                  */
5367                 if (status && new)
5368                         release_lock_stateid(lock_stp);
5369
5370                 nfs4_put_stid(&lock_stp->st_stid);
5371         }
5372         if (open_stp)
5373                 nfs4_put_stid(&open_stp->st_stid);
5374         nfsd4_bump_seqid(cstate, status);
5375         if (file_lock)
5376                 locks_free_lock(file_lock);
5377         if (conflock)
5378                 locks_free_lock(conflock);
5379         return status;
5380 }
5381
5382 /*
5383  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5384  * so we do a temporary open here just to get an open file to pass to
5385  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5386  * inode operation.)
5387  */
5388 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5389 {
5390         struct file *file;
5391         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5392         if (!err) {
5393                 err = nfserrno(vfs_test_lock(file, lock));
5394                 nfsd_close(file);
5395         }
5396         return err;
5397 }
5398
5399 /*
5400  * LOCKT operation
5401  */
5402 __be32
5403 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5404             struct nfsd4_lockt *lockt)
5405 {
5406         struct file_lock *file_lock = NULL;
5407         struct nfs4_lockowner *lo = NULL;
5408         __be32 status;
5409         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5410
5411         if (locks_in_grace(SVC_NET(rqstp)))
5412                 return nfserr_grace;
5413
5414         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5415                  return nfserr_inval;
5416
5417         if (!nfsd4_has_session(cstate)) {
5418                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5419                 if (status)
5420                         goto out;
5421         }
5422
5423         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5424                 goto out;
5425
5426         file_lock = locks_alloc_lock();
5427         if (!file_lock) {
5428                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5429                 status = nfserr_jukebox;
5430                 goto out;
5431         }
5432
5433         switch (lockt->lt_type) {
5434                 case NFS4_READ_LT:
5435                 case NFS4_READW_LT:
5436                         file_lock->fl_type = F_RDLCK;
5437                 break;
5438                 case NFS4_WRITE_LT:
5439                 case NFS4_WRITEW_LT:
5440                         file_lock->fl_type = F_WRLCK;
5441                 break;
5442                 default:
5443                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5444                         status = nfserr_inval;
5445                 goto out;
5446         }
5447
5448         lo = find_lockowner_str(&lockt->lt_clientid, &lockt->lt_owner,
5449                                 cstate->clp);
5450         if (lo)
5451                 file_lock->fl_owner = (fl_owner_t)lo;
5452         file_lock->fl_pid = current->tgid;
5453         file_lock->fl_flags = FL_POSIX;
5454
5455         file_lock->fl_start = lockt->lt_offset;
5456         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5457
5458         nfs4_transform_lock_offset(file_lock);
5459
5460         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5461         if (status)
5462                 goto out;
5463
5464         if (file_lock->fl_type != F_UNLCK) {
5465                 status = nfserr_denied;
5466                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5467         }
5468 out:
5469         if (lo)
5470                 nfs4_put_stateowner(&lo->lo_owner);
5471         if (file_lock)
5472                 locks_free_lock(file_lock);
5473         return status;
5474 }
5475
5476 __be32
5477 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5478             struct nfsd4_locku *locku)
5479 {
5480         struct nfs4_ol_stateid *stp;
5481         struct file *filp = NULL;
5482         struct file_lock *file_lock = NULL;
5483         __be32 status;
5484         int err;
5485         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5486
5487         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5488                 (long long) locku->lu_offset,
5489                 (long long) locku->lu_length);
5490
5491         if (check_lock_length(locku->lu_offset, locku->lu_length))
5492                  return nfserr_inval;
5493
5494         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5495                                         &locku->lu_stateid, NFS4_LOCK_STID,
5496                                         &stp, nn);
5497         if (status)
5498                 goto out;
5499         filp = find_any_file(stp->st_stid.sc_file);
5500         if (!filp) {
5501                 status = nfserr_lock_range;
5502                 goto put_stateid;
5503         }
5504         file_lock = locks_alloc_lock();
5505         if (!file_lock) {
5506                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5507                 status = nfserr_jukebox;
5508                 goto fput;
5509         }
5510
5511         file_lock->fl_type = F_UNLCK;
5512         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5513         file_lock->fl_pid = current->tgid;
5514         file_lock->fl_file = filp;
5515         file_lock->fl_flags = FL_POSIX;
5516         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5517         file_lock->fl_start = locku->lu_offset;
5518
5519         file_lock->fl_end = last_byte_offset(locku->lu_offset,
5520                                                 locku->lu_length);
5521         nfs4_transform_lock_offset(file_lock);
5522
5523         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5524         if (err) {
5525                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5526                 goto out_nfserr;
5527         }
5528         update_stateid(&stp->st_stid.sc_stateid);
5529         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
5530 fput:
5531         fput(filp);
5532 put_stateid:
5533         nfs4_put_stid(&stp->st_stid);
5534 out:
5535         nfsd4_bump_seqid(cstate, status);
5536         if (file_lock)
5537                 locks_free_lock(file_lock);
5538         return status;
5539
5540 out_nfserr:
5541         status = nfserrno(err);
5542         goto fput;
5543 }
5544
5545 /*
5546  * returns
5547  *      true:  locks held by lockowner
5548  *      false: no locks held by lockowner
5549  */
5550 static bool
5551 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5552 {
5553         struct file_lock **flpp;
5554         int status = false;
5555         struct file *filp = find_any_file(fp);
5556         struct inode *inode;
5557
5558         if (!filp) {
5559                 /* Any valid lock stateid should have some sort of access */
5560                 WARN_ON_ONCE(1);
5561                 return status;
5562         }
5563
5564         inode = file_inode(filp);
5565
5566         spin_lock(&inode->i_lock);
5567         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
5568                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
5569                         status = true;
5570                         break;
5571                 }
5572         }
5573         spin_unlock(&inode->i_lock);
5574         fput(filp);
5575         return status;
5576 }
5577
5578 __be32
5579 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5580                         struct nfsd4_compound_state *cstate,
5581                         struct nfsd4_release_lockowner *rlockowner)
5582 {
5583         clientid_t *clid = &rlockowner->rl_clientid;
5584         struct nfs4_stateowner *sop;
5585         struct nfs4_lockowner *lo = NULL;
5586         struct nfs4_ol_stateid *stp;
5587         struct xdr_netobj *owner = &rlockowner->rl_owner;
5588         unsigned int hashval = ownerstr_hashval(owner);
5589         __be32 status;
5590         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5591         struct nfs4_client *clp;
5592
5593         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5594                 clid->cl_boot, clid->cl_id);
5595
5596         status = lookup_clientid(clid, cstate, nn);
5597         if (status)
5598                 return status;
5599
5600         clp = cstate->clp;
5601         /* Find the matching lock stateowner */
5602         spin_lock(&clp->cl_lock);
5603         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5604                             so_strhash) {
5605
5606                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5607                         continue;
5608
5609                 /* see if there are still any locks associated with it */
5610                 lo = lockowner(sop);
5611                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5612                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
5613                                 status = nfserr_locks_held;
5614                                 spin_unlock(&clp->cl_lock);
5615                                 return status;
5616                         }
5617                 }
5618
5619                 nfs4_get_stateowner(sop);
5620                 break;
5621         }
5622         spin_unlock(&clp->cl_lock);
5623         if (lo)
5624                 release_lockowner(lo);
5625         return status;
5626 }
5627
5628 static inline struct nfs4_client_reclaim *
5629 alloc_reclaim(void)
5630 {
5631         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5632 }
5633
5634 bool
5635 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5636 {
5637         struct nfs4_client_reclaim *crp;
5638
5639         crp = nfsd4_find_reclaim_client(name, nn);
5640         return (crp && crp->cr_clp);
5641 }
5642
5643 /*
5644  * failure => all reset bets are off, nfserr_no_grace...
5645  */
5646 struct nfs4_client_reclaim *
5647 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5648 {
5649         unsigned int strhashval;
5650         struct nfs4_client_reclaim *crp;
5651
5652         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5653         crp = alloc_reclaim();
5654         if (crp) {
5655                 strhashval = clientstr_hashval(name);
5656                 INIT_LIST_HEAD(&crp->cr_strhash);
5657                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5658                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5659                 crp->cr_clp = NULL;
5660                 nn->reclaim_str_hashtbl_size++;
5661         }
5662         return crp;
5663 }
5664
5665 void
5666 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5667 {
5668         list_del(&crp->cr_strhash);
5669         kfree(crp);
5670         nn->reclaim_str_hashtbl_size--;
5671 }
5672
5673 void
5674 nfs4_release_reclaim(struct nfsd_net *nn)
5675 {
5676         struct nfs4_client_reclaim *crp = NULL;
5677         int i;
5678
5679         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5680                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5681                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5682                                         struct nfs4_client_reclaim, cr_strhash);
5683                         nfs4_remove_reclaim_record(crp, nn);
5684                 }
5685         }
5686         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5687 }
5688
5689 /*
5690  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5691 struct nfs4_client_reclaim *
5692 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5693 {
5694         unsigned int strhashval;
5695         struct nfs4_client_reclaim *crp = NULL;
5696
5697         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5698
5699         strhashval = clientstr_hashval(recdir);
5700         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5701                 if (same_name(crp->cr_recdir, recdir)) {
5702                         return crp;
5703                 }
5704         }
5705         return NULL;
5706 }
5707
5708 /*
5709 * Called from OPEN. Look for clientid in reclaim list.
5710 */
5711 __be32
5712 nfs4_check_open_reclaim(clientid_t *clid,
5713                 struct nfsd4_compound_state *cstate,
5714                 struct nfsd_net *nn)
5715 {
5716         __be32 status;
5717
5718         /* find clientid in conf_id_hashtbl */
5719         status = lookup_clientid(clid, cstate, nn);
5720         if (status)
5721                 return nfserr_reclaim_bad;
5722
5723         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
5724                 return nfserr_no_grace;
5725
5726         if (nfsd4_client_record_check(cstate->clp))
5727                 return nfserr_reclaim_bad;
5728
5729         return nfs_ok;
5730 }
5731
5732 #ifdef CONFIG_NFSD_FAULT_INJECTION
5733 static inline void
5734 put_client(struct nfs4_client *clp)
5735 {
5736         atomic_dec(&clp->cl_refcount);
5737 }
5738
5739 static struct nfs4_client *
5740 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
5741 {
5742         struct nfs4_client *clp;
5743         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5744                                           nfsd_net_id);
5745
5746         if (!nfsd_netns_ready(nn))
5747                 return NULL;
5748
5749         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5750                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
5751                         return clp;
5752         }
5753         return NULL;
5754 }
5755
5756 u64
5757 nfsd_inject_print_clients(void)
5758 {
5759         struct nfs4_client *clp;
5760         u64 count = 0;
5761         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5762                                           nfsd_net_id);
5763         char buf[INET6_ADDRSTRLEN];
5764
5765         if (!nfsd_netns_ready(nn))
5766                 return 0;
5767
5768         spin_lock(&nn->client_lock);
5769         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5770                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5771                 pr_info("NFS Client: %s\n", buf);
5772                 ++count;
5773         }
5774         spin_unlock(&nn->client_lock);
5775
5776         return count;
5777 }
5778
5779 u64
5780 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
5781 {
5782         u64 count = 0;
5783         struct nfs4_client *clp;
5784         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5785                                           nfsd_net_id);
5786
5787         if (!nfsd_netns_ready(nn))
5788                 return count;
5789
5790         spin_lock(&nn->client_lock);
5791         clp = nfsd_find_client(addr, addr_size);
5792         if (clp) {
5793                 if (mark_client_expired_locked(clp) == nfs_ok)
5794                         ++count;
5795                 else
5796                         clp = NULL;
5797         }
5798         spin_unlock(&nn->client_lock);
5799
5800         if (clp)
5801                 expire_client(clp);
5802
5803         return count;
5804 }
5805
5806 u64
5807 nfsd_inject_forget_clients(u64 max)
5808 {
5809         u64 count = 0;
5810         struct nfs4_client *clp, *next;
5811         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5812                                                 nfsd_net_id);
5813         LIST_HEAD(reaplist);
5814
5815         if (!nfsd_netns_ready(nn))
5816                 return count;
5817
5818         spin_lock(&nn->client_lock);
5819         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
5820                 if (mark_client_expired_locked(clp) == nfs_ok) {
5821                         list_add(&clp->cl_lru, &reaplist);
5822                         if (max != 0 && ++count >= max)
5823                                 break;
5824                 }
5825         }
5826         spin_unlock(&nn->client_lock);
5827
5828         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
5829                 expire_client(clp);
5830
5831         return count;
5832 }
5833
5834 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
5835                              const char *type)
5836 {
5837         char buf[INET6_ADDRSTRLEN];
5838         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5839         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
5840 }
5841
5842 static void
5843 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
5844                              struct list_head *collect)
5845 {
5846         struct nfs4_client *clp = lst->st_stid.sc_client;
5847         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5848                                           nfsd_net_id);
5849
5850         if (!collect)
5851                 return;
5852
5853         lockdep_assert_held(&nn->client_lock);
5854         atomic_inc(&clp->cl_refcount);
5855         list_add(&lst->st_locks, collect);
5856 }
5857
5858 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
5859                                     struct list_head *collect,
5860                                     void (*func)(struct nfs4_ol_stateid *))
5861 {
5862         struct nfs4_openowner *oop;
5863         struct nfs4_ol_stateid *stp, *st_next;
5864         struct nfs4_ol_stateid *lst, *lst_next;
5865         u64 count = 0;
5866
5867         spin_lock(&clp->cl_lock);
5868         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
5869                 list_for_each_entry_safe(stp, st_next,
5870                                 &oop->oo_owner.so_stateids, st_perstateowner) {
5871                         list_for_each_entry_safe(lst, lst_next,
5872                                         &stp->st_locks, st_locks) {
5873                                 if (func) {
5874                                         func(lst);
5875                                         nfsd_inject_add_lock_to_list(lst,
5876                                                                 collect);
5877                                 }
5878                                 ++count;
5879                                 /*
5880                                  * Despite the fact that these functions deal
5881                                  * with 64-bit integers for "count", we must
5882                                  * ensure that it doesn't blow up the
5883                                  * clp->cl_refcount. Throw a warning if we
5884                                  * start to approach INT_MAX here.
5885                                  */
5886                                 WARN_ON_ONCE(count == (INT_MAX / 2));
5887                                 if (count == max)
5888                                         goto out;
5889                         }
5890                 }
5891         }
5892 out:
5893         spin_unlock(&clp->cl_lock);
5894
5895         return count;
5896 }
5897
5898 static u64
5899 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
5900                           u64 max)
5901 {
5902         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
5903 }
5904
5905 static u64
5906 nfsd_print_client_locks(struct nfs4_client *clp)
5907 {
5908         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
5909         nfsd_print_count(clp, count, "locked files");
5910         return count;
5911 }
5912
5913 u64
5914 nfsd_inject_print_locks(void)
5915 {
5916         struct nfs4_client *clp;
5917         u64 count = 0;
5918         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5919                                                 nfsd_net_id);
5920
5921         if (!nfsd_netns_ready(nn))
5922                 return 0;
5923
5924         spin_lock(&nn->client_lock);
5925         list_for_each_entry(clp, &nn->client_lru, cl_lru)
5926                 count += nfsd_print_client_locks(clp);
5927         spin_unlock(&nn->client_lock);
5928
5929         return count;
5930 }
5931
5932 static void
5933 nfsd_reap_locks(struct list_head *reaplist)
5934 {
5935         struct nfs4_client *clp;
5936         struct nfs4_ol_stateid *stp, *next;
5937
5938         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
5939                 list_del_init(&stp->st_locks);
5940                 clp = stp->st_stid.sc_client;
5941                 nfs4_put_stid(&stp->st_stid);
5942                 put_client(clp);
5943         }
5944 }
5945
5946 u64
5947 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
5948 {
5949         unsigned int count = 0;
5950         struct nfs4_client *clp;
5951         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5952                                                 nfsd_net_id);
5953         LIST_HEAD(reaplist);
5954
5955         if (!nfsd_netns_ready(nn))
5956                 return count;
5957
5958         spin_lock(&nn->client_lock);
5959         clp = nfsd_find_client(addr, addr_size);
5960         if (clp)
5961                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
5962         spin_unlock(&nn->client_lock);
5963         nfsd_reap_locks(&reaplist);
5964         return count;
5965 }
5966
5967 u64
5968 nfsd_inject_forget_locks(u64 max)
5969 {
5970         u64 count = 0;
5971         struct nfs4_client *clp;
5972         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5973                                                 nfsd_net_id);
5974         LIST_HEAD(reaplist);
5975
5976         if (!nfsd_netns_ready(nn))
5977                 return count;
5978
5979         spin_lock(&nn->client_lock);
5980         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5981                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
5982                 if (max != 0 && count >= max)
5983                         break;
5984         }
5985         spin_unlock(&nn->client_lock);
5986         nfsd_reap_locks(&reaplist);
5987         return count;
5988 }
5989
5990 static u64
5991 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
5992                               struct list_head *collect,
5993                               void (*func)(struct nfs4_openowner *))
5994 {
5995         struct nfs4_openowner *oop, *next;
5996         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5997                                                 nfsd_net_id);
5998         u64 count = 0;
5999
6000         lockdep_assert_held(&nn->client_lock);
6001
6002         spin_lock(&clp->cl_lock);
6003         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6004                 if (func) {
6005                         func(oop);
6006                         if (collect) {
6007                                 atomic_inc(&clp->cl_refcount);
6008                                 list_add(&oop->oo_perclient, collect);
6009                         }
6010                 }
6011                 ++count;
6012                 /*
6013                  * Despite the fact that these functions deal with
6014                  * 64-bit integers for "count", we must ensure that
6015                  * it doesn't blow up the clp->cl_refcount. Throw a
6016                  * warning if we start to approach INT_MAX here.
6017                  */
6018                 WARN_ON_ONCE(count == (INT_MAX / 2));
6019                 if (count == max)
6020                         break;
6021         }
6022         spin_unlock(&clp->cl_lock);
6023
6024         return count;
6025 }
6026
6027 static u64
6028 nfsd_print_client_openowners(struct nfs4_client *clp)
6029 {
6030         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6031
6032         nfsd_print_count(clp, count, "openowners");
6033         return count;
6034 }
6035
6036 static u64
6037 nfsd_collect_client_openowners(struct nfs4_client *clp,
6038                                struct list_head *collect, u64 max)
6039 {
6040         return nfsd_foreach_client_openowner(clp, max, collect,
6041                                                 unhash_openowner_locked);
6042 }
6043
6044 u64
6045 nfsd_inject_print_openowners(void)
6046 {
6047         struct nfs4_client *clp;
6048         u64 count = 0;
6049         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6050                                                 nfsd_net_id);
6051
6052         if (!nfsd_netns_ready(nn))
6053                 return 0;
6054
6055         spin_lock(&nn->client_lock);
6056         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6057                 count += nfsd_print_client_openowners(clp);
6058         spin_unlock(&nn->client_lock);
6059
6060         return count;
6061 }
6062
6063 static void
6064 nfsd_reap_openowners(struct list_head *reaplist)
6065 {
6066         struct nfs4_client *clp;
6067         struct nfs4_openowner *oop, *next;
6068
6069         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6070                 list_del_init(&oop->oo_perclient);
6071                 clp = oop->oo_owner.so_client;
6072                 release_openowner(oop);
6073                 put_client(clp);
6074         }
6075 }
6076
6077 u64
6078 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6079                                      size_t addr_size)
6080 {
6081         unsigned int count = 0;
6082         struct nfs4_client *clp;
6083         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6084                                                 nfsd_net_id);
6085         LIST_HEAD(reaplist);
6086
6087         if (!nfsd_netns_ready(nn))
6088                 return count;
6089
6090         spin_lock(&nn->client_lock);
6091         clp = nfsd_find_client(addr, addr_size);
6092         if (clp)
6093                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6094         spin_unlock(&nn->client_lock);
6095         nfsd_reap_openowners(&reaplist);
6096         return count;
6097 }
6098
6099 u64
6100 nfsd_inject_forget_openowners(u64 max)
6101 {
6102         u64 count = 0;
6103         struct nfs4_client *clp;
6104         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6105                                                 nfsd_net_id);
6106         LIST_HEAD(reaplist);
6107
6108         if (!nfsd_netns_ready(nn))
6109                 return count;
6110
6111         spin_lock(&nn->client_lock);
6112         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6113                 count += nfsd_collect_client_openowners(clp, &reaplist,
6114                                                         max - count);
6115                 if (max != 0 && count >= max)
6116                         break;
6117         }
6118         spin_unlock(&nn->client_lock);
6119         nfsd_reap_openowners(&reaplist);
6120         return count;
6121 }
6122
6123 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6124                                      struct list_head *victims)
6125 {
6126         struct nfs4_delegation *dp, *next;
6127         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6128                                                 nfsd_net_id);
6129         u64 count = 0;
6130
6131         lockdep_assert_held(&nn->client_lock);
6132
6133         spin_lock(&state_lock);
6134         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6135                 if (victims) {
6136                         /*
6137                          * It's not safe to mess with delegations that have a
6138                          * non-zero dl_time. They might have already been broken
6139                          * and could be processed by the laundromat outside of
6140                          * the state_lock. Just leave them be.
6141                          */
6142                         if (dp->dl_time != 0)
6143                                 continue;
6144
6145                         atomic_inc(&clp->cl_refcount);
6146                         unhash_delegation_locked(dp);
6147                         list_add(&dp->dl_recall_lru, victims);
6148                 }
6149                 ++count;
6150                 /*
6151                  * Despite the fact that these functions deal with
6152                  * 64-bit integers for "count", we must ensure that
6153                  * it doesn't blow up the clp->cl_refcount. Throw a
6154                  * warning if we start to approach INT_MAX here.
6155                  */
6156                 WARN_ON_ONCE(count == (INT_MAX / 2));
6157                 if (count == max)
6158                         break;
6159         }
6160         spin_unlock(&state_lock);
6161         return count;
6162 }
6163
6164 static u64
6165 nfsd_print_client_delegations(struct nfs4_client *clp)
6166 {
6167         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6168
6169         nfsd_print_count(clp, count, "delegations");
6170         return count;
6171 }
6172
6173 u64
6174 nfsd_inject_print_delegations(void)
6175 {
6176         struct nfs4_client *clp;
6177         u64 count = 0;
6178         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6179                                                 nfsd_net_id);
6180
6181         if (!nfsd_netns_ready(nn))
6182                 return 0;
6183
6184         spin_lock(&nn->client_lock);
6185         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6186                 count += nfsd_print_client_delegations(clp);
6187         spin_unlock(&nn->client_lock);
6188
6189         return count;
6190 }
6191
6192 static void
6193 nfsd_forget_delegations(struct list_head *reaplist)
6194 {
6195         struct nfs4_client *clp;
6196         struct nfs4_delegation *dp, *next;
6197
6198         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6199                 list_del_init(&dp->dl_recall_lru);
6200                 clp = dp->dl_stid.sc_client;
6201                 revoke_delegation(dp);
6202                 put_client(clp);
6203         }
6204 }
6205
6206 u64
6207 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6208                                       size_t addr_size)
6209 {
6210         u64 count = 0;
6211         struct nfs4_client *clp;
6212         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6213                                                 nfsd_net_id);
6214         LIST_HEAD(reaplist);
6215
6216         if (!nfsd_netns_ready(nn))
6217                 return count;
6218
6219         spin_lock(&nn->client_lock);
6220         clp = nfsd_find_client(addr, addr_size);
6221         if (clp)
6222                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6223         spin_unlock(&nn->client_lock);
6224
6225         nfsd_forget_delegations(&reaplist);
6226         return count;
6227 }
6228
6229 u64
6230 nfsd_inject_forget_delegations(u64 max)
6231 {
6232         u64 count = 0;
6233         struct nfs4_client *clp;
6234         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6235                                                 nfsd_net_id);
6236         LIST_HEAD(reaplist);
6237
6238         if (!nfsd_netns_ready(nn))
6239                 return count;
6240
6241         spin_lock(&nn->client_lock);
6242         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6243                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6244                 if (max != 0 && count >= max)
6245                         break;
6246         }
6247         spin_unlock(&nn->client_lock);
6248         nfsd_forget_delegations(&reaplist);
6249         return count;
6250 }
6251
6252 static void
6253 nfsd_recall_delegations(struct list_head *reaplist)
6254 {
6255         struct nfs4_client *clp;
6256         struct nfs4_delegation *dp, *next;
6257
6258         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6259                 list_del_init(&dp->dl_recall_lru);
6260                 clp = dp->dl_stid.sc_client;
6261                 /*
6262                  * We skipped all entries that had a zero dl_time before,
6263                  * so we can now reset the dl_time back to 0. If a delegation
6264                  * break comes in now, then it won't make any difference since
6265                  * we're recalling it either way.
6266                  */
6267                 spin_lock(&state_lock);
6268                 dp->dl_time = 0;
6269                 spin_unlock(&state_lock);
6270                 nfsd_break_one_deleg(dp);
6271                 put_client(clp);
6272         }
6273 }
6274
6275 u64
6276 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6277                                       size_t addr_size)
6278 {
6279         u64 count = 0;
6280         struct nfs4_client *clp;
6281         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6282                                                 nfsd_net_id);
6283         LIST_HEAD(reaplist);
6284
6285         if (!nfsd_netns_ready(nn))
6286                 return count;
6287
6288         spin_lock(&nn->client_lock);
6289         clp = nfsd_find_client(addr, addr_size);
6290         if (clp)
6291                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6292         spin_unlock(&nn->client_lock);
6293
6294         nfsd_recall_delegations(&reaplist);
6295         return count;
6296 }
6297
6298 u64
6299 nfsd_inject_recall_delegations(u64 max)
6300 {
6301         u64 count = 0;
6302         struct nfs4_client *clp, *next;
6303         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6304                                                 nfsd_net_id);
6305         LIST_HEAD(reaplist);
6306
6307         if (!nfsd_netns_ready(nn))
6308                 return count;
6309
6310         spin_lock(&nn->client_lock);
6311         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6312                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6313                 if (max != 0 && ++count >= max)
6314                         break;
6315         }
6316         spin_unlock(&nn->client_lock);
6317         nfsd_recall_delegations(&reaplist);
6318         return count;
6319 }
6320 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6321
6322 /*
6323  * Since the lifetime of a delegation isn't limited to that of an open, a
6324  * client may quite reasonably hang on to a delegation as long as it has
6325  * the inode cached.  This becomes an obvious problem the first time a
6326  * client's inode cache approaches the size of the server's total memory.
6327  *
6328  * For now we avoid this problem by imposing a hard limit on the number
6329  * of delegations, which varies according to the server's memory size.
6330  */
6331 static void
6332 set_max_delegations(void)
6333 {
6334         /*
6335          * Allow at most 4 delegations per megabyte of RAM.  Quick
6336          * estimates suggest that in the worst case (where every delegation
6337          * is for a different inode), a delegation could take about 1.5K,
6338          * giving a worst case usage of about 6% of memory.
6339          */
6340         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6341 }
6342
6343 static int nfs4_state_create_net(struct net *net)
6344 {
6345         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6346         int i;
6347
6348         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6349                         CLIENT_HASH_SIZE, GFP_KERNEL);
6350         if (!nn->conf_id_hashtbl)
6351                 goto err;
6352         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6353                         CLIENT_HASH_SIZE, GFP_KERNEL);
6354         if (!nn->unconf_id_hashtbl)
6355                 goto err_unconf_id;
6356         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6357                         SESSION_HASH_SIZE, GFP_KERNEL);
6358         if (!nn->sessionid_hashtbl)
6359                 goto err_sessionid;
6360
6361         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6362                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6363                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6364         }
6365         for (i = 0; i < SESSION_HASH_SIZE; i++)
6366                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6367         nn->conf_name_tree = RB_ROOT;
6368         nn->unconf_name_tree = RB_ROOT;
6369         INIT_LIST_HEAD(&nn->client_lru);
6370         INIT_LIST_HEAD(&nn->close_lru);
6371         INIT_LIST_HEAD(&nn->del_recall_lru);
6372         spin_lock_init(&nn->client_lock);
6373
6374         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6375         get_net(net);
6376
6377         return 0;
6378
6379 err_sessionid:
6380         kfree(nn->unconf_id_hashtbl);
6381 err_unconf_id:
6382         kfree(nn->conf_id_hashtbl);
6383 err:
6384         return -ENOMEM;
6385 }
6386
6387 static void
6388 nfs4_state_destroy_net(struct net *net)
6389 {
6390         int i;
6391         struct nfs4_client *clp = NULL;
6392         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6393
6394         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6395                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6396                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6397                         destroy_client(clp);
6398                 }
6399         }
6400
6401         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6402                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6403                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6404                         destroy_client(clp);
6405                 }
6406         }
6407
6408         kfree(nn->sessionid_hashtbl);
6409         kfree(nn->unconf_id_hashtbl);
6410         kfree(nn->conf_id_hashtbl);
6411         put_net(net);
6412 }
6413
6414 int
6415 nfs4_state_start_net(struct net *net)
6416 {
6417         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6418         int ret;
6419
6420         ret = nfs4_state_create_net(net);
6421         if (ret)
6422                 return ret;
6423         nn->boot_time = get_seconds();
6424         nn->grace_ended = false;
6425         locks_start_grace(net, &nn->nfsd4_manager);
6426         nfsd4_client_tracking_init(net);
6427         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6428                nn->nfsd4_grace, net);
6429         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6430         return 0;
6431 }
6432
6433 /* initialization to perform when the nfsd service is started: */
6434
6435 int
6436 nfs4_state_start(void)
6437 {
6438         int ret;
6439
6440         ret = set_callback_cred();
6441         if (ret)
6442                 return -ENOMEM;
6443         laundry_wq = create_singlethread_workqueue("nfsd4");
6444         if (laundry_wq == NULL) {
6445                 ret = -ENOMEM;
6446                 goto out_recovery;
6447         }
6448         ret = nfsd4_create_callback_queue();
6449         if (ret)
6450                 goto out_free_laundry;
6451
6452         set_max_delegations();
6453
6454         return 0;
6455
6456 out_free_laundry:
6457         destroy_workqueue(laundry_wq);
6458 out_recovery:
6459         return ret;
6460 }
6461
6462 void
6463 nfs4_state_shutdown_net(struct net *net)
6464 {
6465         struct nfs4_delegation *dp = NULL;
6466         struct list_head *pos, *next, reaplist;
6467         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6468
6469         cancel_delayed_work_sync(&nn->laundromat_work);
6470         locks_end_grace(&nn->nfsd4_manager);
6471
6472         INIT_LIST_HEAD(&reaplist);
6473         spin_lock(&state_lock);
6474         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6475                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6476                 unhash_delegation_locked(dp);
6477                 list_add(&dp->dl_recall_lru, &reaplist);
6478         }
6479         spin_unlock(&state_lock);
6480         list_for_each_safe(pos, next, &reaplist) {
6481                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6482                 list_del_init(&dp->dl_recall_lru);
6483                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6484                 nfs4_put_stid(&dp->dl_stid);
6485         }
6486
6487         nfsd4_client_tracking_exit(net);
6488         nfs4_state_destroy_net(net);
6489 }
6490
6491 void
6492 nfs4_state_shutdown(void)
6493 {
6494         destroy_workqueue(laundry_wq);
6495         nfsd4_destroy_callback_queue();
6496 }
6497
6498 static void
6499 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6500 {
6501         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6502                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6503 }
6504
6505 static void
6506 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6507 {
6508         if (cstate->minorversion) {
6509                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6510                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6511         }
6512 }
6513
6514 void
6515 clear_current_stateid(struct nfsd4_compound_state *cstate)
6516 {
6517         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6518 }
6519
6520 /*
6521  * functions to set current state id
6522  */
6523 void
6524 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6525 {
6526         put_stateid(cstate, &odp->od_stateid);
6527 }
6528
6529 void
6530 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6531 {
6532         put_stateid(cstate, &open->op_stateid);
6533 }
6534
6535 void
6536 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6537 {
6538         put_stateid(cstate, &close->cl_stateid);
6539 }
6540
6541 void
6542 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6543 {
6544         put_stateid(cstate, &lock->lk_resp_stateid);
6545 }
6546
6547 /*
6548  * functions to consume current state id
6549  */
6550
6551 void
6552 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6553 {
6554         get_stateid(cstate, &odp->od_stateid);
6555 }
6556
6557 void
6558 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6559 {
6560         get_stateid(cstate, &drp->dr_stateid);
6561 }
6562
6563 void
6564 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6565 {
6566         get_stateid(cstate, &fsp->fr_stateid);
6567 }
6568
6569 void
6570 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6571 {
6572         get_stateid(cstate, &setattr->sa_stateid);
6573 }
6574
6575 void
6576 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6577 {
6578         get_stateid(cstate, &close->cl_stateid);
6579 }
6580
6581 void
6582 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6583 {
6584         get_stateid(cstate, &locku->lu_stateid);
6585 }
6586
6587 void
6588 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6589 {
6590         get_stateid(cstate, &read->rd_stateid);
6591 }
6592
6593 void
6594 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6595 {
6596         get_stateid(cstate, &write->wr_stateid);
6597 }