SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/nfsd/nfs4state.c

   1 /*
   2 *  linux/fs/nfsd/nfs4state.c
   3 *
   4 *  Copyright (c) 2001 The Regents of the University of Michigan.
   5 *  All rights reserved.
   6 *
   7 *  Kendrick Smith <kmsmith@umich.edu>
   8 *  Andy Adamson <kandros@umich.edu>
   9 *
  10 *  Redistribution and use in source and binary forms, with or without
  11 *  modification, are permitted provided that the following conditions
  12 *  are met:
  13 *
  14 *  1. Redistributions of source code must retain the above copyright
  15 *     notice, this list of conditions and the following disclaimer.
  16 *  2. Redistributions in binary form must reproduce the above copyright
  17 *     notice, this list of conditions and the following disclaimer in the
  18 *     documentation and/or other materials provided with the distribution.
  19 *  3. Neither the name of the University nor the names of its
  20 *     contributors may be used to endorse or promote products derived
  21 *     from this software without specific prior written permission.
  22 *
  23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  34 *
  35 */
  36
  37 #include <linux/param.h>
  38 #include <linux/major.h>
  39 #include <linux/slab.h>
  40
  41 #include <linux/sunrpc/svc.h>
  42 #include <linux/nfsd/nfsd.h>
  43 #include <linux/nfsd/cache.h>
  44 #include <linux/file.h>
  45 #include <linux/mount.h>
  46 #include <linux/workqueue.h>
  47 #include <linux/smp_lock.h>
  48 #include <linux/kthread.h>
  49 #include <linux/nfs4.h>
  50 #include <linux/nfsd/state.h>
  51 #include <linux/nfsd/xdr4.h>
  52 #include <linux/namei.h>
  53 #include <linux/swap.h>
  54 #include <linux/mutex.h>
  55 #include <linux/lockd/bind.h>
  56 #include <linux/module.h>
  57 #include <linux/sunrpc/svcauth_gss.h>
  58
  59 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  60
  61 /* Globals */
  62 static time_t lease_time = 90;     /* default lease time */
  63 static time_t user_lease_time = 90;
  64 static time_t boot_time;
  65 static u32 current_ownerid = 1;
  66 static u32 current_fileid = 1;
  67 static u32 current_delegid = 1;
  68 static u32 nfs4_init;
  69 static stateid_t zerostateid;             /* bits all 0 */
  70 static stateid_t onestateid;              /* bits all 1 */
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
  74
  75 /* forward declarations */
  76 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
  77 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
  78 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  79 static void nfs4_set_recdir(char *recdir);
  80
  81 /* Locking:
  82  *
  83  * client_mutex:
  84  *      protects clientid_hashtbl[], clientstr_hashtbl[],
  85  *      unconfstr_hashtbl[], uncofid_hashtbl[].
  86  */
  87 static DEFINE_MUTEX(client_mutex);
  88
  89 static struct kmem_cache *stateowner_slab = NULL;
  90 static struct kmem_cache *file_slab = NULL;
  91 static struct kmem_cache *stateid_slab = NULL;
  92 static struct kmem_cache *deleg_slab = NULL;
  93
  94 void
  95 nfs4_lock_state(void)
  96 {
  97         mutex_lock(&client_mutex);
  98 }
  99
 100 void
 101 nfs4_unlock_state(void)
 102 {
 103         mutex_unlock(&client_mutex);
 104 }
 105
 106 static inline u32
 107 opaque_hashval(const void *ptr, int nbytes)
 108 {
 109         unsigned char *cptr = (unsigned char *) ptr;
 110
 111         u32 x = 0;
 112         while (nbytes--) {
 113                 x *= 37;
 114                 x += *cptr++;
 115         }
 116         return x;
 117 }
 118
 119 /*
 120  * Delegation state
 121  */
 122
 123 /* recall_lock protects the del_recall_lru */
 124 static DEFINE_SPINLOCK(recall_lock);
 125 static struct list_head del_recall_lru;
 126
 127 static void
 128 free_nfs4_file(struct kref *kref)
 129 {
 130         struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
 131         list_del(&fp->fi_hash);
 132         iput(fp->fi_inode);
 133         kmem_cache_free(file_slab, fp);
 134 }
 135
 136 static inline void
 137 put_nfs4_file(struct nfs4_file *fi)
 138 {
 139         kref_put(&fi->fi_ref, free_nfs4_file);
 140 }
 141
 142 static inline void
 143 get_nfs4_file(struct nfs4_file *fi)
 144 {
 145         kref_get(&fi->fi_ref);
 146 }
 147
 148 static int num_delegations;
 149 unsigned int max_delegations;
 150
 151 /*
 152  * Open owner state (share locks)
 153  */
 154
 155 /* hash tables for nfs4_stateowner */
 156 #define OWNER_HASH_BITS              8
 157 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 158 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 159
 160 #define ownerid_hashval(id) \
 161         ((id) & OWNER_HASH_MASK)
 162 #define ownerstr_hashval(clientid, ownername) \
 163         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
 164
 165 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
 166 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 167
 168 /* hash table for nfs4_file */
 169 #define FILE_HASH_BITS                   8
 170 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 171 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
 172 /* hash table for (open)nfs4_stateid */
 173 #define STATEID_HASH_BITS              10
 174 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
 175 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
 176
 177 #define file_hashval(x) \
 178         hash_ptr(x, FILE_HASH_BITS)
 179 #define stateid_hashval(owner_id, file_id)  \
 180         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
 181
 182 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 183 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 184
 185 static struct nfs4_delegation *
 186 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 187 {
 188         struct nfs4_delegation *dp;
 189         struct nfs4_file *fp = stp->st_file;
 190         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
 191
 192         dprintk("NFSD alloc_init_deleg\n");
 193         if (fp->fi_had_conflict)
 194                 return NULL;
 195         if (num_delegations > max_delegations)
 196                 return NULL;
 197         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
 198         if (dp == NULL)
 199                 return dp;
 200         num_delegations++;
 201         INIT_LIST_HEAD(&dp->dl_perfile);
 202         INIT_LIST_HEAD(&dp->dl_perclnt);
 203         INIT_LIST_HEAD(&dp->dl_recall_lru);
 204         dp->dl_client = clp;
 205         get_nfs4_file(fp);
 206         dp->dl_file = fp;
 207         dp->dl_flock = NULL;
 208         get_file(stp->st_vfs_file);
 209         dp->dl_vfs_file = stp->st_vfs_file;
 210         dp->dl_type = type;
 211         dp->dl_recall.cbr_dp = NULL;
 212         dp->dl_recall.cbr_ident = cb->cb_ident;
 213         dp->dl_recall.cbr_trunc = 0;
 214         dp->dl_stateid.si_boot = boot_time;
 215         dp->dl_stateid.si_stateownerid = current_delegid++;
 216         dp->dl_stateid.si_fileid = 0;
 217         dp->dl_stateid.si_generation = 0;
 218         dp->dl_fhlen = current_fh->fh_handle.fh_size;
 219         memcpy(dp->dl_fhval, &current_fh->fh_handle.fh_base,
 220                         current_fh->fh_handle.fh_size);
 221         dp->dl_time = 0;
 222         atomic_set(&dp->dl_count, 1);
 223         list_add(&dp->dl_perfile, &fp->fi_delegations);
 224         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 225         return dp;
 226 }
 227
 228 void
 229 nfs4_put_delegation(struct nfs4_delegation *dp)
 230 {
 231         if (atomic_dec_and_test(&dp->dl_count)) {
 232                 dprintk("NFSD: freeing dp %p\n",dp);
 233                 put_nfs4_file(dp->dl_file);
 234                 kmem_cache_free(deleg_slab, dp);
 235                 num_delegations--;
 236         }
 237 }
 238
 239 /* Remove the associated file_lock first, then remove the delegation.
 240  * lease_modify() is called to remove the FS_LEASE file_lock from
 241  * the i_flock list, eventually calling nfsd's lock_manager
 242  * fl_release_callback.
 243  */
 244 static void
 245 nfs4_close_delegation(struct nfs4_delegation *dp)
 246 {
 247         struct file *filp = dp->dl_vfs_file;
 248
 249         dprintk("NFSD: close_delegation dp %p\n",dp);
 250         dp->dl_vfs_file = NULL;
 251         /* The following nfsd_close may not actually close the file,
 252          * but we want to remove the lease in any case. */
 253         if (dp->dl_flock)
 254                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
 255         nfsd_close(filp);
 256 }
 257
 258 /* Called under the state lock. */
 259 static void
 260 unhash_delegation(struct nfs4_delegation *dp)
 261 {
 262         list_del_init(&dp->dl_perfile);
 263         list_del_init(&dp->dl_perclnt);
 264         spin_lock(&recall_lock);
 265         list_del_init(&dp->dl_recall_lru);
 266         spin_unlock(&recall_lock);
 267         nfs4_close_delegation(dp);
 268         nfs4_put_delegation(dp);
 269 }
 270
 271 /*
 272  * SETCLIENTID state
 273  */
 274
 275 /* Hash tables for nfs4_clientid state */
 276 #define CLIENT_HASH_BITS                 4
 277 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 278 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 279
 280 #define clientid_hashval(id) \
 281         ((id) & CLIENT_HASH_MASK)
 282 #define clientstr_hashval(name) \
 283         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
 284 /*
 285  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 286  * used in reboot/reset lease grace period processing
 287  *
 288  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 289  * setclientid_confirmed info.
 290  *
 291  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 292  * setclientid info.
 293  *
 294  * client_lru holds client queue ordered by nfs4_client.cl_time
 295  * for lease renewal.
 296  *
 297  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 298  * for last close replay.
 299  */
 300 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 301 static int reclaim_str_hashtbl_size = 0;
 302 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 303 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 304 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 305 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 306 static struct list_head client_lru;
 307 static struct list_head close_lru;
 308
 309 static void unhash_generic_stateid(struct nfs4_stateid *stp)
 310 {
 311         list_del(&stp->st_hash);
 312         list_del(&stp->st_perfile);
 313         list_del(&stp->st_perstateowner);
 314 }
 315
 316 static void free_generic_stateid(struct nfs4_stateid *stp)
 317 {
 318         put_nfs4_file(stp->st_file);
 319         kmem_cache_free(stateid_slab, stp);
 320 }
 321
 322 static void release_lock_stateid(struct nfs4_stateid *stp)
 323 {
 324         unhash_generic_stateid(stp);
 325         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
 326         free_generic_stateid(stp);
 327 }
 328
 329 static void unhash_lockowner(struct nfs4_stateowner *sop)
 330 {
 331         struct nfs4_stateid *stp;
 332
 333         list_del(&sop->so_idhash);
 334         list_del(&sop->so_strhash);
 335         list_del(&sop->so_perstateid);
 336         while (!list_empty(&sop->so_stateids)) {
 337                 stp = list_first_entry(&sop->so_stateids,
 338                                 struct nfs4_stateid, st_perstateowner);
 339                 release_lock_stateid(stp);
 340         }
 341 }
 342
 343 static void release_lockowner(struct nfs4_stateowner *sop)
 344 {
 345         unhash_lockowner(sop);
 346         nfs4_put_stateowner(sop);
 347 }
 348
 349 static void
 350 release_stateid_lockowners(struct nfs4_stateid *open_stp)
 351 {
 352         struct nfs4_stateowner *lock_sop;
 353
 354         while (!list_empty(&open_stp->st_lockowners)) {
 355                 lock_sop = list_entry(open_stp->st_lockowners.next,
 356                                 struct nfs4_stateowner, so_perstateid);
 357                 /* list_del(&open_stp->st_lockowners);  */
 358                 BUG_ON(lock_sop->so_is_open_owner);
 359                 release_lockowner(lock_sop);
 360         }
 361 }
 362
 363 static void release_open_stateid(struct nfs4_stateid *stp)
 364 {
 365         unhash_generic_stateid(stp);
 366         release_stateid_lockowners(stp);
 367         nfsd_close(stp->st_vfs_file);
 368         free_generic_stateid(stp);
 369 }
 370
 371 static void unhash_openowner(struct nfs4_stateowner *sop)
 372 {
 373         struct nfs4_stateid *stp;
 374
 375         list_del(&sop->so_idhash);
 376         list_del(&sop->so_strhash);
 377         list_del(&sop->so_perclient);
 378         list_del(&sop->so_perstateid); /* XXX: necessary? */
 379         while (!list_empty(&sop->so_stateids)) {
 380                 stp = list_first_entry(&sop->so_stateids,
 381                                 struct nfs4_stateid, st_perstateowner);
 382                 release_open_stateid(stp);
 383         }
 384 }
 385
 386 static void release_openowner(struct nfs4_stateowner *sop)
 387 {
 388         unhash_openowner(sop);
 389         list_del(&sop->so_close_lru);
 390         nfs4_put_stateowner(sop);
 391 }
 392
 393 static inline void
 394 renew_client(struct nfs4_client *clp)
 395 {
 396         /*
 397         * Move client to the end to the LRU list.
 398         */
 399         dprintk("renewing client (clientid %08x/%08x)\n",
 400                         clp->cl_clientid.cl_boot,
 401                         clp->cl_clientid.cl_id);
 402         list_move_tail(&clp->cl_lru, &client_lru);
 403         clp->cl_time = get_seconds();
 404 }
 405
 406 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 407 static int
 408 STALE_CLIENTID(clientid_t *clid)
 409 {
 410         if (clid->cl_boot == boot_time)
 411                 return 0;
 412         dprintk("NFSD stale clientid (%08x/%08x)\n",
 413                         clid->cl_boot, clid->cl_id);
 414         return 1;
 415 }
 416
 417 /*
 418  * XXX Should we use a slab cache ?
 419  * This type of memory management is somewhat inefficient, but we use it
 420  * anyway since SETCLIENTID is not a common operation.
 421  */
 422 static struct nfs4_client *alloc_client(struct xdr_netobj name)
 423 {
 424         struct nfs4_client *clp;
 425
 426         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
 427         if (clp == NULL)
 428                 return NULL;
 429         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
 430         if (clp->cl_name.data == NULL) {
 431                 kfree(clp);
 432                 return NULL;
 433         }
 434         memcpy(clp->cl_name.data, name.data, name.len);
 435         clp->cl_name.len = name.len;
 436         return clp;
 437 }
 438
 439 static void
 440 shutdown_callback_client(struct nfs4_client *clp)
 441 {
 442         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
 443
 444         if (clnt) {
 445                 /*
 446                  * Callback threads take a reference on the client, so there
 447                  * should be no outstanding callbacks at this point.
 448                  */
 449                 clp->cl_callback.cb_client = NULL;
 450                 rpc_shutdown_client(clnt);
 451         }
 452 }
 453
 454 static inline void
 455 free_client(struct nfs4_client *clp)
 456 {
 457         shutdown_callback_client(clp);
 458         if (clp->cl_cred.cr_group_info)
 459                 put_group_info(clp->cl_cred.cr_group_info);
 460         kfree(clp->cl_principal);
 461         kfree(clp->cl_name.data);
 462         kfree(clp);
 463 }
 464
 465 void
 466 put_nfs4_client(struct nfs4_client *clp)
 467 {
 468         if (atomic_dec_and_test(&clp->cl_count))
 469                 free_client(clp);
 470 }
 471
 472 static void
 473 expire_client(struct nfs4_client *clp)
 474 {
 475         struct nfs4_stateowner *sop;
 476         struct nfs4_delegation *dp;
 477         struct list_head reaplist;
 478
 479         dprintk("NFSD: expire_client cl_count %d\n",
 480                             atomic_read(&clp->cl_count));
 481
 482         INIT_LIST_HEAD(&reaplist);
 483         spin_lock(&recall_lock);
 484         while (!list_empty(&clp->cl_delegations)) {
 485                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
 486                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
 487                                 dp->dl_flock);
 488                 list_del_init(&dp->dl_perclnt);
 489                 list_move(&dp->dl_recall_lru, &reaplist);
 490         }
 491         spin_unlock(&recall_lock);
 492         while (!list_empty(&reaplist)) {
 493                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
 494                 list_del_init(&dp->dl_recall_lru);
 495                 unhash_delegation(dp);
 496         }
 497         list_del(&clp->cl_idhash);
 498         list_del(&clp->cl_strhash);
 499         list_del(&clp->cl_lru);
 500         while (!list_empty(&clp->cl_openowners)) {
 501                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
 502                 release_openowner(sop);
 503         }
 504         put_nfs4_client(clp);
 505 }
 506
 507 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
 508 {
 509         struct nfs4_client *clp;
 510
 511         clp = alloc_client(name);
 512         if (clp == NULL)
 513                 return NULL;
 514         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
 515         atomic_set(&clp->cl_count, 1);
 516         atomic_set(&clp->cl_callback.cb_set, 0);
 517         INIT_LIST_HEAD(&clp->cl_idhash);
 518         INIT_LIST_HEAD(&clp->cl_strhash);
 519         INIT_LIST_HEAD(&clp->cl_openowners);
 520         INIT_LIST_HEAD(&clp->cl_delegations);
 521         INIT_LIST_HEAD(&clp->cl_lru);
 522         return clp;
 523 }
 524
 525 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
 526 {
 527         memcpy(target->cl_verifier.data, source->data,
 528                         sizeof(target->cl_verifier.data));
 529 }
 530
 531 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
 532 {
 533         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
 534         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
 535 }
 536
 537 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
 538 {
 539         target->cr_uid = source->cr_uid;
 540         target->cr_gid = source->cr_gid;
 541         target->cr_group_info = source->cr_group_info;
 542         get_group_info(target->cr_group_info);
 543 }
 544
 545 static int same_name(const char *n1, const char *n2)
 546 {
 547         return 0 == memcmp(n1, n2, HEXDIR_LEN);
 548 }
 549
 550 static int
 551 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
 552 {
 553         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
 554 }
 555
 556 static int
 557 same_clid(clientid_t *cl1, clientid_t *cl2)
 558 {
 559         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
 560 }
 561
 562 /* XXX what about NGROUP */
 563 static int
 564 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
 565 {
 566         return cr1->cr_uid == cr2->cr_uid;
 567 }
 568
 569 static void gen_clid(struct nfs4_client *clp)
 570 {
 571         static u32 current_clientid = 1;
 572
 573         clp->cl_clientid.cl_boot = boot_time;
 574         clp->cl_clientid.cl_id = current_clientid++;
 575 }
 576
 577 static void gen_confirm(struct nfs4_client *clp)
 578 {
 579         static u32 i;
 580         u32 *p;
 581
 582         p = (u32 *)clp->cl_confirm.data;
 583         *p++ = get_seconds();
 584         *p++ = i++;
 585 }
 586
 587 static int check_name(struct xdr_netobj name)
 588 {
 589         if (name.len == 0)
 590                 return 0;
 591         if (name.len > NFS4_OPAQUE_LIMIT) {
 592                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
 593                 return 0;
 594         }
 595         return 1;
 596 }
 597
 598 static void
 599 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
 600 {
 601         unsigned int idhashval;
 602
 603         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
 604         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 605         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
 606         list_add_tail(&clp->cl_lru, &client_lru);
 607         clp->cl_time = get_seconds();
 608 }
 609
 610 static void
 611 move_to_confirmed(struct nfs4_client *clp)
 612 {
 613         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 614         unsigned int strhashval;
 615
 616         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
 617         list_del_init(&clp->cl_strhash);
 618         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
 619         strhashval = clientstr_hashval(clp->cl_recdir);
 620         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
 621         renew_client(clp);
 622 }
 623
 624 static struct nfs4_client *
 625 find_confirmed_client(clientid_t *clid)
 626 {
 627         struct nfs4_client *clp;
 628         unsigned int idhashval = clientid_hashval(clid->cl_id);
 629
 630         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
 631                 if (same_clid(&clp->cl_clientid, clid))
 632                         return clp;
 633         }
 634         return NULL;
 635 }
 636
 637 static struct nfs4_client *
 638 find_unconfirmed_client(clientid_t *clid)
 639 {
 640         struct nfs4_client *clp;
 641         unsigned int idhashval = clientid_hashval(clid->cl_id);
 642
 643         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
 644                 if (same_clid(&clp->cl_clientid, clid))
 645                         return clp;
 646         }
 647         return NULL;
 648 }
 649
 650 static struct nfs4_client *
 651 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
 652 {
 653         struct nfs4_client *clp;
 654
 655         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
 656                 if (same_name(clp->cl_recdir, dname))
 657                         return clp;
 658         }
 659         return NULL;
 660 }
 661
 662 static struct nfs4_client *
 663 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
 664 {
 665         struct nfs4_client *clp;
 666
 667         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
 668                 if (same_name(clp->cl_recdir, dname))
 669                         return clp;
 670         }
 671         return NULL;
 672 }
 673
 674 /* a helper function for parse_callback */
 675 static int
 676 parse_octet(unsigned int *lenp, char **addrp)
 677 {
 678         unsigned int len = *lenp;
 679         char *p = *addrp;
 680         int n = -1;
 681         char c;
 682
 683         for (;;) {
 684                 if (!len)
 685                         break;
 686                 len--;
 687                 c = *p++;
 688                 if (c == '.')
 689                         break;
 690                 if ((c < '0') || (c > '9')) {
 691                         n = -1;
 692                         break;
 693                 }
 694                 if (n < 0)
 695                         n = 0;
 696                 n = (n * 10) + (c - '0');
 697                 if (n > 255) {
 698                         n = -1;
 699                         break;
 700                 }
 701         }
 702         *lenp = len;
 703         *addrp = p;
 704         return n;
 705 }
 706
 707 /* parse and set the setclientid ipv4 callback address */
 708 static int
 709 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
 710 {
 711         int temp = 0;
 712         u32 cbaddr = 0;
 713         u16 cbport = 0;
 714         u32 addrlen = addr_len;
 715         char *addr = addr_val;
 716         int i, shift;
 717
 718         /* ipaddress */
 719         shift = 24;
 720         for(i = 4; i > 0  ; i--) {
 721                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 722                         return 0;
 723                 }
 724                 cbaddr |= (temp << shift);
 725                 if (shift > 0)
 726                 shift -= 8;
 727         }
 728         *cbaddrp = cbaddr;
 729
 730         /* port */
 731         shift = 8;
 732         for(i = 2; i > 0  ; i--) {
 733                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 734                         return 0;
 735                 }
 736                 cbport |= (temp << shift);
 737                 if (shift > 0)
 738                         shift -= 8;
 739         }
 740         *cbportp = cbport;
 741         return 1;
 742 }
 743
 744 static void
 745 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
 746 {
 747         struct nfs4_callback *cb = &clp->cl_callback;
 748
 749         /* Currently, we only support tcp for the callback channel */
 750         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
 751                 goto out_err;
 752
 753         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
 754                          &cb->cb_addr, &cb->cb_port)))
 755                 goto out_err;
 756         cb->cb_prog = se->se_callback_prog;
 757         cb->cb_ident = se->se_callback_ident;
 758         return;
 759 out_err:
 760         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
 761                 "will not receive delegations\n",
 762                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
 763
 764         return;
 765 }
 766
 767 __be32
 768 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 769                   struct nfsd4_setclientid *setclid)
 770 {
 771         struct sockaddr_in      *sin = svc_addr_in(rqstp);
 772         struct xdr_netobj       clname = {
 773                 .len = setclid->se_namelen,
 774                 .data = setclid->se_name,
 775         };
 776         nfs4_verifier           clverifier = setclid->se_verf;
 777         unsigned int            strhashval;
 778         struct nfs4_client      *conf, *unconf, *new;
 779         __be32                  status;
 780         char                    *princ;
 781         char                    dname[HEXDIR_LEN];
 782
 783         if (!check_name(clname))
 784                 return nfserr_inval;
 785
 786         status = nfs4_make_rec_clidname(dname, &clname);
 787         if (status)
 788                 return status;
 789
 790         /*
 791          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 792          * We get here on a DRC miss.
 793          */
 794
 795         strhashval = clientstr_hashval(dname);
 796
 797         nfs4_lock_state();
 798         conf = find_confirmed_client_by_str(dname, strhashval);
 799         if (conf) {
 800                 /* RFC 3530 14.2.33 CASE 0: */
 801                 status = nfserr_clid_inuse;
 802                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
 803                                 || conf->cl_addr != sin->sin_addr.s_addr) {
 804                         dprintk("NFSD: setclientid: string in use by clientat %pI4\n",
 805                                 &conf->cl_addr);
 806                         goto out;
 807                 }
 808         }
 809         /*
 810          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
 811          * has a description of SETCLIENTID request processing consisting
 812          * of 5 bullet points, labeled as CASE0 - CASE4 below.
 813          */
 814         unconf = find_unconfirmed_client_by_str(dname, strhashval);
 815         status = nfserr_resource;
 816         if (!conf) {
 817                 /*
 818                  * RFC 3530 14.2.33 CASE 4:
 819                  * placed first, because it is the normal case
 820                  */
 821                 if (unconf)
 822                         expire_client(unconf);
 823                 new = create_client(clname, dname);
 824                 if (new == NULL)
 825                         goto out;
 826                 gen_clid(new);
 827         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
 828                 /*
 829                  * RFC 3530 14.2.33 CASE 1:
 830                  * probable callback update
 831                  */
 832                 if (unconf) {
 833                         /* Note this is removing unconfirmed {*x***},
 834                          * which is stronger than RFC recommended {vxc**}.
 835                          * This has the advantage that there is at most
 836                          * one {*x***} in either list at any time.
 837                          */
 838                         expire_client(unconf);
 839                 }
 840                 new = create_client(clname, dname);
 841                 if (new == NULL)
 842                         goto out;
 843                 copy_clid(new, conf);
 844         } else if (!unconf) {
 845                 /*
 846                  * RFC 3530 14.2.33 CASE 2:
 847                  * probable client reboot; state will be removed if
 848                  * confirmed.
 849                  */
 850                 new = create_client(clname, dname);
 851                 if (new == NULL)
 852                         goto out;
 853                 gen_clid(new);
 854         } else {
 855                 /*
 856                  * RFC 3530 14.2.33 CASE 3:
 857                  * probable client reboot; state will be removed if
 858                  * confirmed.
 859                  */
 860                 expire_client(unconf);
 861                 new = create_client(clname, dname);
 862                 if (new == NULL)
 863                         goto out;
 864                 gen_clid(new);
 865         }
 866         copy_verf(new, &clverifier);
 867         new->cl_addr = sin->sin_addr.s_addr;
 868         new->cl_flavor = rqstp->rq_flavor;
 869         princ = svc_gss_principal(rqstp);
 870         if (princ) {
 871                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
 872                 if (new->cl_principal == NULL) {
 873                         free_client(new);
 874                         goto out;
 875                 }
 876         }
 877         copy_cred(&new->cl_cred, &rqstp->rq_cred);
 878         gen_confirm(new);
 879         gen_callback(new, setclid);
 880         add_to_unconfirmed(new, strhashval);
 881         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
 882         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
 883         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
 884         status = nfs_ok;
 885 out:
 886         nfs4_unlock_state();
 887         return status;
 888 }
 889
 890
 891 /*
 892  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
 893  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
 894  * bullets, labeled as CASE1 - CASE4 below.
 895  */
 896 __be32
 897 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
 898                          struct nfsd4_compound_state *cstate,
 899                          struct nfsd4_setclientid_confirm *setclientid_confirm)
 900 {
 901         struct sockaddr_in *sin = svc_addr_in(rqstp);
 902         struct nfs4_client *conf, *unconf;
 903         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
 904         clientid_t * clid = &setclientid_confirm->sc_clientid;
 905         __be32 status;
 906
 907         if (STALE_CLIENTID(clid))
 908                 return nfserr_stale_clientid;
 909         /*
 910          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 911          * We get here on a DRC miss.
 912          */
 913
 914         nfs4_lock_state();
 915
 916         conf = find_confirmed_client(clid);
 917         unconf = find_unconfirmed_client(clid);
 918
 919         status = nfserr_clid_inuse;
 920         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
 921                 goto out;
 922         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
 923                 goto out;
 924
 925         /*
 926          * section 14.2.34 of RFC 3530 has a description of
 927          * SETCLIENTID_CONFIRM request processing consisting
 928          * of 4 bullet points, labeled as CASE1 - CASE4 below.
 929          */
 930         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
 931                 /*
 932                  * RFC 3530 14.2.34 CASE 1:
 933                  * callback update
 934                  */
 935                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
 936                         status = nfserr_clid_inuse;
 937                 else {
 938                         /* XXX: We just turn off callbacks until we can handle
 939                           * change request correctly. */
 940                         atomic_set(&conf->cl_callback.cb_set, 0);
 941                         gen_confirm(conf);
 942                         nfsd4_remove_clid_dir(unconf);
 943                         expire_client(unconf);
 944                         status = nfs_ok;
 945
 946                 }
 947         } else if (conf && !unconf) {
 948                 /*
 949                  * RFC 3530 14.2.34 CASE 2:
 950                  * probable retransmitted request; play it safe and
 951                  * do nothing.
 952                  */
 953                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
 954                         status = nfserr_clid_inuse;
 955                 else
 956                         status = nfs_ok;
 957         } else if (!conf && unconf
 958                         && same_verf(&unconf->cl_confirm, &confirm)) {
 959                 /*
 960                  * RFC 3530 14.2.34 CASE 3:
 961                  * Normal case; new or rebooted client:
 962                  */
 963                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
 964                         status = nfserr_clid_inuse;
 965                 } else {
 966                         unsigned int hash =
 967                                 clientstr_hashval(unconf->cl_recdir);
 968                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
 969                                                                         hash);
 970                         if (conf) {
 971                                 nfsd4_remove_clid_dir(conf);
 972                                 expire_client(conf);
 973                         }
 974                         move_to_confirmed(unconf);
 975                         conf = unconf;
 976                         nfsd4_probe_callback(conf);
 977                         status = nfs_ok;
 978                 }
 979         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
 980             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
 981                                                                 &confirm)))) {
 982                 /*
 983                  * RFC 3530 14.2.34 CASE 4:
 984                  * Client probably hasn't noticed that we rebooted yet.
 985                  */
 986                 status = nfserr_stale_clientid;
 987         } else {
 988                 /* check that we have hit one of the cases...*/
 989                 status = nfserr_clid_inuse;
 990         }
 991 out:
 992         nfs4_unlock_state();
 993         return status;
 994 }
 995
 996 /* OPEN Share state helper functions */
 997 static inline struct nfs4_file *
 998 alloc_init_file(struct inode *ino)
 999 {
1000         struct nfs4_file *fp;
1001         unsigned int hashval = file_hashval(ino);
1002
1003         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1004         if (fp) {
1005                 kref_init(&fp->fi_ref);
1006                 INIT_LIST_HEAD(&fp->fi_hash);
1007                 INIT_LIST_HEAD(&fp->fi_stateids);
1008                 INIT_LIST_HEAD(&fp->fi_delegations);
1009                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1010                 fp->fi_inode = igrab(ino);
1011                 fp->fi_id = current_fileid++;
1012                 fp->fi_had_conflict = false;
1013                 return fp;
1014         }
1015         return NULL;
1016 }
1017
1018 static void
1019 nfsd4_free_slab(struct kmem_cache **slab)
1020 {
1021         if (*slab == NULL)
1022                 return;
1023         kmem_cache_destroy(*slab);
1024         *slab = NULL;
1025 }
1026
1027 void
1028 nfsd4_free_slabs(void)
1029 {
1030         nfsd4_free_slab(&stateowner_slab);
1031         nfsd4_free_slab(&file_slab);
1032         nfsd4_free_slab(&stateid_slab);
1033         nfsd4_free_slab(&deleg_slab);
1034 }
1035
1036 static int
1037 nfsd4_init_slabs(void)
1038 {
1039         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1040                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1041         if (stateowner_slab == NULL)
1042                 goto out_nomem;
1043         file_slab = kmem_cache_create("nfsd4_files",
1044                         sizeof(struct nfs4_file), 0, 0, NULL);
1045         if (file_slab == NULL)
1046                 goto out_nomem;
1047         stateid_slab = kmem_cache_create("nfsd4_stateids",
1048                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1049         if (stateid_slab == NULL)
1050                 goto out_nomem;
1051         deleg_slab = kmem_cache_create("nfsd4_delegations",
1052                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1053         if (deleg_slab == NULL)
1054                 goto out_nomem;
1055         return 0;
1056 out_nomem:
1057         nfsd4_free_slabs();
1058         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1059         return -ENOMEM;
1060 }
1061
1062 void
1063 nfs4_free_stateowner(struct kref *kref)
1064 {
1065         struct nfs4_stateowner *sop =
1066                 container_of(kref, struct nfs4_stateowner, so_ref);
1067         kfree(sop->so_owner.data);
1068         kmem_cache_free(stateowner_slab, sop);
1069 }
1070
1071 static inline struct nfs4_stateowner *
1072 alloc_stateowner(struct xdr_netobj *owner)
1073 {
1074         struct nfs4_stateowner *sop;
1075
1076         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1077                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1078                         memcpy(sop->so_owner.data, owner->data, owner->len);
1079                         sop->so_owner.len = owner->len;
1080                         kref_init(&sop->so_ref);
1081                         return sop;
1082                 }
1083                 kmem_cache_free(stateowner_slab, sop);
1084         }
1085         return NULL;
1086 }
1087
1088 static struct nfs4_stateowner *
1089 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1090         struct nfs4_stateowner *sop;
1091         struct nfs4_replay *rp;
1092         unsigned int idhashval;
1093
1094         if (!(sop = alloc_stateowner(&open->op_owner)))
1095                 return NULL;
1096         idhashval = ownerid_hashval(current_ownerid);
1097         INIT_LIST_HEAD(&sop->so_idhash);
1098         INIT_LIST_HEAD(&sop->so_strhash);
1099         INIT_LIST_HEAD(&sop->so_perclient);
1100         INIT_LIST_HEAD(&sop->so_stateids);
1101         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1102         INIT_LIST_HEAD(&sop->so_close_lru);
1103         sop->so_time = 0;
1104         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1105         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1106         list_add(&sop->so_perclient, &clp->cl_openowners);
1107         sop->so_is_open_owner = 1;
1108         sop->so_id = current_ownerid++;
1109         sop->so_client = clp;
1110         sop->so_seqid = open->op_seqid;
1111         sop->so_confirmed = 0;
1112         rp = &sop->so_replay;
1113         rp->rp_status = nfserr_serverfault;
1114         rp->rp_buflen = 0;
1115         rp->rp_buf = rp->rp_ibuf;
1116         return sop;
1117 }
1118
1119 static inline void
1120 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1121         struct nfs4_stateowner *sop = open->op_stateowner;
1122         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1123
1124         INIT_LIST_HEAD(&stp->st_hash);
1125         INIT_LIST_HEAD(&stp->st_perstateowner);
1126         INIT_LIST_HEAD(&stp->st_lockowners);
1127         INIT_LIST_HEAD(&stp->st_perfile);
1128         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1129         list_add(&stp->st_perstateowner, &sop->so_stateids);
1130         list_add(&stp->st_perfile, &fp->fi_stateids);
1131         stp->st_stateowner = sop;
1132         get_nfs4_file(fp);
1133         stp->st_file = fp;
1134         stp->st_stateid.si_boot = boot_time;
1135         stp->st_stateid.si_stateownerid = sop->so_id;
1136         stp->st_stateid.si_fileid = fp->fi_id;
1137         stp->st_stateid.si_generation = 0;
1138         stp->st_access_bmap = 0;
1139         stp->st_deny_bmap = 0;
1140         __set_bit(open->op_share_access, &stp->st_access_bmap);
1141         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1142         stp->st_openstp = NULL;
1143 }
1144
1145 static void
1146 move_to_close_lru(struct nfs4_stateowner *sop)
1147 {
1148         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1149
1150         list_move_tail(&sop->so_close_lru, &close_lru);
1151         sop->so_time = get_seconds();
1152 }
1153
1154 static int
1155 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1156                                                         clientid_t *clid)
1157 {
1158         return (sop->so_owner.len == owner->len) &&
1159                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1160                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1161 }
1162
1163 static struct nfs4_stateowner *
1164 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1165 {
1166         struct nfs4_stateowner *so = NULL;
1167
1168         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1169                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1170                         return so;
1171         }
1172         return NULL;
1173 }
1174
1175 /* search file_hashtbl[] for file */
1176 static struct nfs4_file *
1177 find_file(struct inode *ino)
1178 {
1179         unsigned int hashval = file_hashval(ino);
1180         struct nfs4_file *fp;
1181
1182         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1183                 if (fp->fi_inode == ino) {
1184                         get_nfs4_file(fp);
1185                         return fp;
1186                 }
1187         }
1188         return NULL;
1189 }
1190
1191 static inline int access_valid(u32 x)
1192 {
1193         if (x < NFS4_SHARE_ACCESS_READ)
1194                 return 0;
1195         if (x > NFS4_SHARE_ACCESS_BOTH)
1196                 return 0;
1197         return 1;
1198 }
1199
1200 static inline int deny_valid(u32 x)
1201 {
1202         /* Note: unlike access bits, deny bits may be zero. */
1203         return x <= NFS4_SHARE_DENY_BOTH;
1204 }
1205
1206 /*
1207  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1208  * st_{access,deny}_bmap field of the stateid, in order to track not
1209  * only what share bits are currently in force, but also what
1210  * combinations of share bits previous opens have used.  This allows us
1211  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1212  * return an error if the client attempt to downgrade to a combination
1213  * of share bits not explicable by closing some of its previous opens.
1214  *
1215  * XXX: This enforcement is actually incomplete, since we don't keep
1216  * track of access/deny bit combinations; so, e.g., we allow:
1217  *
1218  *      OPEN allow read, deny write
1219  *      OPEN allow both, deny none
1220  *      DOWNGRADE allow read, deny none
1221  *
1222  * which we should reject.
1223  */
1224 static void
1225 set_access(unsigned int *access, unsigned long bmap) {
1226         int i;
1227
1228         *access = 0;
1229         for (i = 1; i < 4; i++) {
1230                 if (test_bit(i, &bmap))
1231                         *access |= i;
1232         }
1233 }
1234
1235 static void
1236 set_deny(unsigned int *deny, unsigned long bmap) {
1237         int i;
1238
1239         *deny = 0;
1240         for (i = 0; i < 4; i++) {
1241                 if (test_bit(i, &bmap))
1242                         *deny |= i ;
1243         }
1244 }
1245
1246 static int
1247 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1248         unsigned int access, deny;
1249
1250         set_access(&access, stp->st_access_bmap);
1251         set_deny(&deny, stp->st_deny_bmap);
1252         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1253                 return 0;
1254         return 1;
1255 }
1256
1257 /*
1258  * Called to check deny when READ with all zero stateid or
1259  * WRITE with all zero or all one stateid
1260  */
1261 static __be32
1262 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1263 {
1264         struct inode *ino = current_fh->fh_dentry->d_inode;
1265         struct nfs4_file *fp;
1266         struct nfs4_stateid *stp;
1267         __be32 ret;
1268
1269         dprintk("NFSD: nfs4_share_conflict\n");
1270
1271         fp = find_file(ino);
1272         if (!fp)
1273                 return nfs_ok;
1274         ret = nfserr_locked;
1275         /* Search for conflicting share reservations */
1276         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1277                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1278                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1279                         goto out;
1280         }
1281         ret = nfs_ok;
1282 out:
1283         put_nfs4_file(fp);
1284         return ret;
1285 }
1286
1287 static inline void
1288 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1289 {
1290         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1291                 drop_file_write_access(filp);
1292                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1293         }
1294 }
1295
1296 /*
1297  * Recall a delegation
1298  */
1299 static int
1300 do_recall(void *__dp)
1301 {
1302         struct nfs4_delegation *dp = __dp;
1303
1304         dp->dl_file->fi_had_conflict = true;
1305         nfsd4_cb_recall(dp);
1306         return 0;
1307 }
1308
1309 /*
1310  * Spawn a thread to perform a recall on the delegation represented
1311  * by the lease (file_lock)
1312  *
1313  * Called from break_lease() with lock_kernel() held.
1314  * Note: we assume break_lease will only call this *once* for any given
1315  * lease.
1316  */
1317 static
1318 void nfsd_break_deleg_cb(struct file_lock *fl)
1319 {
1320         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1321         struct task_struct *t;
1322
1323         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1324         if (!dp)
1325                 return;
1326
1327         /* We're assuming the state code never drops its reference
1328          * without first removing the lease.  Since we're in this lease
1329          * callback (and since the lease code is serialized by the kernel
1330          * lock) we know the server hasn't removed the lease yet, we know
1331          * it's safe to take a reference: */
1332         atomic_inc(&dp->dl_count);
1333         atomic_inc(&dp->dl_client->cl_count);
1334
1335         spin_lock(&recall_lock);
1336         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1337         spin_unlock(&recall_lock);
1338
1339         /* only place dl_time is set. protected by lock_kernel*/
1340         dp->dl_time = get_seconds();
1341
1342         /*
1343          * We don't want the locks code to timeout the lease for us;
1344          * we'll remove it ourself if the delegation isn't returned
1345          * in time.
1346          */
1347         fl->fl_break_time = 0;
1348
1349         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1350         if (IS_ERR(t)) {
1351                 struct nfs4_client *clp = dp->dl_client;
1352
1353                 printk(KERN_INFO "NFSD: Callback thread failed for "
1354                         "for client (clientid %08x/%08x)\n",
1355                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1356                 put_nfs4_client(dp->dl_client);
1357                 nfs4_put_delegation(dp);
1358         }
1359 }
1360
1361 /*
1362  * The file_lock is being reapd.
1363  *
1364  * Called by locks_free_lock() with lock_kernel() held.
1365  */
1366 static
1367 void nfsd_release_deleg_cb(struct file_lock *fl)
1368 {
1369         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1370
1371         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1372
1373         if (!(fl->fl_flags & FL_LEASE) || !dp)
1374                 return;
1375         dp->dl_flock = NULL;
1376 }
1377
1378 /*
1379  * Set the delegation file_lock back pointer.
1380  *
1381  * Called from setlease() with lock_kernel() held.
1382  */
1383 static
1384 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1385 {
1386         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1387
1388         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1389         if (!dp)
1390                 return;
1391         dp->dl_flock = new;
1392 }
1393
1394 /*
1395  * Called from setlease() with lock_kernel() held
1396  */
1397 static
1398 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1399 {
1400         struct nfs4_delegation *onlistd =
1401                 (struct nfs4_delegation *)onlist->fl_owner;
1402         struct nfs4_delegation *tryd =
1403                 (struct nfs4_delegation *)try->fl_owner;
1404
1405         if (onlist->fl_lmops != try->fl_lmops)
1406                 return 0;
1407
1408         return onlistd->dl_client == tryd->dl_client;
1409 }
1410
1411
1412 static
1413 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1414 {
1415         if (arg & F_UNLCK)
1416                 return lease_modify(onlist, arg);
1417         else
1418                 return -EAGAIN;
1419 }
1420
1421 static struct lock_manager_operations nfsd_lease_mng_ops = {
1422         .fl_break = nfsd_break_deleg_cb,
1423         .fl_release_private = nfsd_release_deleg_cb,
1424         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1425         .fl_mylease = nfsd_same_client_deleg_cb,
1426         .fl_change = nfsd_change_deleg_cb,
1427 };
1428
1429
1430 __be32
1431 nfsd4_process_open1(struct nfsd4_open *open)
1432 {
1433         clientid_t *clientid = &open->op_clientid;
1434         struct nfs4_client *clp = NULL;
1435         unsigned int strhashval;
1436         struct nfs4_stateowner *sop = NULL;
1437
1438         if (!check_name(open->op_owner))
1439                 return nfserr_inval;
1440
1441         if (STALE_CLIENTID(&open->op_clientid))
1442                 return nfserr_stale_clientid;
1443
1444         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1445         sop = find_openstateowner_str(strhashval, open);
1446         open->op_stateowner = sop;
1447         if (!sop) {
1448                 /* Make sure the client's lease hasn't expired. */
1449                 clp = find_confirmed_client(clientid);
1450                 if (clp == NULL)
1451                         return nfserr_expired;
1452                 goto renew;
1453         }
1454         if (!sop->so_confirmed) {
1455                 /* Replace unconfirmed owners without checking for replay. */
1456                 clp = sop->so_client;
1457                 release_openowner(sop);
1458                 open->op_stateowner = NULL;
1459                 goto renew;
1460         }
1461         if (open->op_seqid == sop->so_seqid - 1) {
1462                 if (sop->so_replay.rp_buflen)
1463                         return nfserr_replay_me;
1464                 /* The original OPEN failed so spectacularly
1465                  * that we don't even have replay data saved!
1466                  * Therefore, we have no choice but to continue
1467                  * processing this OPEN; presumably, we'll
1468                  * fail again for the same reason.
1469                  */
1470                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1471                 goto renew;
1472         }
1473         if (open->op_seqid != sop->so_seqid)
1474                 return nfserr_bad_seqid;
1475 renew:
1476         if (open->op_stateowner == NULL) {
1477                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1478                 if (sop == NULL)
1479                         return nfserr_resource;
1480                 open->op_stateowner = sop;
1481         }
1482         list_del_init(&sop->so_close_lru);
1483         renew_client(sop->so_client);
1484         return nfs_ok;
1485 }
1486
1487 static inline __be32
1488 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1489 {
1490         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1491                 return nfserr_openmode;
1492         else
1493                 return nfs_ok;
1494 }
1495
1496 static struct nfs4_delegation *
1497 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1498 {
1499         struct nfs4_delegation *dp;
1500
1501         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1502                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1503                         return dp;
1504         }
1505         return NULL;
1506 }
1507
1508 static __be32
1509 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1510                 struct nfs4_delegation **dp)
1511 {
1512         int flags;
1513         __be32 status = nfserr_bad_stateid;
1514
1515         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1516         if (*dp == NULL)
1517                 goto out;
1518         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1519                                                 RD_STATE : WR_STATE;
1520         status = nfs4_check_delegmode(*dp, flags);
1521         if (status)
1522                 *dp = NULL;
1523 out:
1524         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1525                 return nfs_ok;
1526         if (status)
1527                 return status;
1528         open->op_stateowner->so_confirmed = 1;
1529         return nfs_ok;
1530 }
1531
1532 static __be32
1533 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1534 {
1535         struct nfs4_stateid *local;
1536         __be32 status = nfserr_share_denied;
1537         struct nfs4_stateowner *sop = open->op_stateowner;
1538
1539         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1540                 /* ignore lock owners */
1541                 if (local->st_stateowner->so_is_open_owner == 0)
1542                         continue;
1543                 /* remember if we have seen this open owner */
1544                 if (local->st_stateowner == sop)
1545                         *stpp = local;
1546                 /* check for conflicting share reservations */
1547                 if (!test_share(local, open))
1548                         goto out;
1549         }
1550         status = 0;
1551 out:
1552         return status;
1553 }
1554
1555 static inline struct nfs4_stateid *
1556 nfs4_alloc_stateid(void)
1557 {
1558         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1559 }
1560
1561 static __be32
1562 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1563                 struct nfs4_delegation *dp,
1564                 struct svc_fh *cur_fh, int flags)
1565 {
1566         struct nfs4_stateid *stp;
1567
1568         stp = nfs4_alloc_stateid();
1569         if (stp == NULL)
1570                 return nfserr_resource;
1571
1572         if (dp) {
1573                 get_file(dp->dl_vfs_file);
1574                 stp->st_vfs_file = dp->dl_vfs_file;
1575         } else {
1576                 __be32 status;
1577                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1578                                 &stp->st_vfs_file);
1579                 if (status) {
1580                         if (status == nfserr_dropit)
1581                                 status = nfserr_jukebox;
1582                         kmem_cache_free(stateid_slab, stp);
1583                         return status;
1584                 }
1585         }
1586         *stpp = stp;
1587         return 0;
1588 }
1589
1590 static inline __be32
1591 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1592                 struct nfsd4_open *open)
1593 {
1594         struct iattr iattr = {
1595                 .ia_valid = ATTR_SIZE,
1596                 .ia_size = 0,
1597         };
1598         if (!open->op_truncate)
1599                 return 0;
1600         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1601                 return nfserr_inval;
1602         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1603 }
1604
1605 static __be32
1606 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1607 {
1608         struct file *filp = stp->st_vfs_file;
1609         struct inode *inode = filp->f_path.dentry->d_inode;
1610         unsigned int share_access, new_writer;
1611         __be32 status;
1612
1613         set_access(&share_access, stp->st_access_bmap);
1614         new_writer = (~share_access) & open->op_share_access
1615                         & NFS4_SHARE_ACCESS_WRITE;
1616
1617         if (new_writer) {
1618                 int err = get_write_access(inode);
1619                 if (err)
1620                         return nfserrno(err);
1621                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
1622                 if (err)
1623                         return nfserrno(err);
1624                 file_take_write(filp);
1625         }
1626         status = nfsd4_truncate(rqstp, cur_fh, open);
1627         if (status) {
1628                 if (new_writer)
1629                         put_write_access(inode);
1630                 return status;
1631         }
1632         /* remember the open */
1633         filp->f_mode |= open->op_share_access;
1634         __set_bit(open->op_share_access, &stp->st_access_bmap);
1635         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1636
1637         return nfs_ok;
1638 }
1639
1640
1641 static void
1642 nfs4_set_claim_prev(struct nfsd4_open *open)
1643 {
1644         open->op_stateowner->so_confirmed = 1;
1645         open->op_stateowner->so_client->cl_firststate = 1;
1646 }
1647
1648 /*
1649  * Attempt to hand out a delegation.
1650  */
1651 static void
1652 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1653 {
1654         struct nfs4_delegation *dp;
1655         struct nfs4_stateowner *sop = stp->st_stateowner;
1656         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1657         struct file_lock fl, *flp = &fl;
1658         int status, flag = 0;
1659
1660         flag = NFS4_OPEN_DELEGATE_NONE;
1661         open->op_recall = 0;
1662         switch (open->op_claim_type) {
1663                 case NFS4_OPEN_CLAIM_PREVIOUS:
1664                         if (!atomic_read(&cb->cb_set))
1665                                 open->op_recall = 1;
1666                         flag = open->op_delegate_type;
1667                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1668                                 goto out;
1669                         break;
1670                 case NFS4_OPEN_CLAIM_NULL:
1671                         /* Let's not give out any delegations till everyone's
1672                          * had the chance to reclaim theirs.... */
1673                         if (locks_in_grace())
1674                                 goto out;
1675                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1676                                 goto out;
1677                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1678                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1679                         else
1680                                 flag = NFS4_OPEN_DELEGATE_READ;
1681                         break;
1682                 default:
1683                         goto out;
1684         }
1685
1686         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1687         if (dp == NULL) {
1688                 flag = NFS4_OPEN_DELEGATE_NONE;
1689                 goto out;
1690         }
1691         locks_init_lock(&fl);
1692         fl.fl_lmops = &nfsd_lease_mng_ops;
1693         fl.fl_flags = FL_LEASE;
1694         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
1695         fl.fl_end = OFFSET_MAX;
1696         fl.fl_owner =  (fl_owner_t)dp;
1697         fl.fl_file = stp->st_vfs_file;
1698         fl.fl_pid = current->tgid;
1699
1700         /* vfs_setlease checks to see if delegation should be handed out.
1701          * the lock_manager callbacks fl_mylease and fl_change are used
1702          */
1703         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
1704                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1705                 unhash_delegation(dp);
1706                 flag = NFS4_OPEN_DELEGATE_NONE;
1707                 goto out;
1708         }
1709
1710         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1711
1712         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1713                      dp->dl_stateid.si_boot,
1714                      dp->dl_stateid.si_stateownerid,
1715                      dp->dl_stateid.si_fileid,
1716                      dp->dl_stateid.si_generation);
1717 out:
1718         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1719                         && flag == NFS4_OPEN_DELEGATE_NONE
1720                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1721                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1722         open->op_delegate_type = flag;
1723 }
1724
1725 /*
1726  * called with nfs4_lock_state() held.
1727  */
1728 __be32
1729 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1730 {
1731         struct nfs4_file *fp = NULL;
1732         struct inode *ino = current_fh->fh_dentry->d_inode;
1733         struct nfs4_stateid *stp = NULL;
1734         struct nfs4_delegation *dp = NULL;
1735         __be32 status;
1736
1737         status = nfserr_inval;
1738         if (!access_valid(open->op_share_access)
1739                         || !deny_valid(open->op_share_deny))
1740                 goto out;
1741         /*
1742          * Lookup file; if found, lookup stateid and check open request,
1743          * and check for delegations in the process of being recalled.
1744          * If not found, create the nfs4_file struct
1745          */
1746         fp = find_file(ino);
1747         if (fp) {
1748                 if ((status = nfs4_check_open(fp, open, &stp)))
1749                         goto out;
1750                 status = nfs4_check_deleg(fp, open, &dp);
1751                 if (status)
1752                         goto out;
1753         } else {
1754                 status = nfserr_bad_stateid;
1755                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1756                         goto out;
1757                 status = nfserr_resource;
1758                 fp = alloc_init_file(ino);
1759                 if (fp == NULL)
1760                         goto out;
1761         }
1762
1763         /*
1764          * OPEN the file, or upgrade an existing OPEN.
1765          * If truncate fails, the OPEN fails.
1766          */
1767         if (stp) {
1768                 /* Stateid was found, this is an OPEN upgrade */
1769                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1770                 if (status)
1771                         goto out;
1772                 update_stateid(&stp->st_stateid);
1773         } else {
1774                 /* Stateid was not found, this is a new OPEN */
1775                 int flags = 0;
1776                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1777                         flags |= NFSD_MAY_READ;
1778                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1779                         flags |= NFSD_MAY_WRITE;
1780                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1781                 if (status)
1782                         goto out;
1783                 init_stateid(stp, fp, open);
1784                 status = nfsd4_truncate(rqstp, current_fh, open);
1785                 if (status) {
1786                         release_open_stateid(stp);
1787                         goto out;
1788                 }
1789         }
1790         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1791
1792         /*
1793         * Attempt to hand out a delegation. No error return, because the
1794         * OPEN succeeds even if we fail.
1795         */
1796         nfs4_open_delegation(current_fh, open, stp);
1797
1798         status = nfs_ok;
1799
1800         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1801                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1802                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1803 out:
1804         if (fp)
1805                 put_nfs4_file(fp);
1806         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1807                 nfs4_set_claim_prev(open);
1808         /*
1809         * To finish the open response, we just need to set the rflags.
1810         */
1811         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1812         if (!open->op_stateowner->so_confirmed)
1813                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1814
1815         return status;
1816 }
1817
1818 __be32
1819 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1820             clientid_t *clid)
1821 {
1822         struct nfs4_client *clp;
1823         __be32 status;
1824
1825         nfs4_lock_state();
1826         dprintk("process_renew(%08x/%08x): starting\n",
1827                         clid->cl_boot, clid->cl_id);
1828         status = nfserr_stale_clientid;
1829         if (STALE_CLIENTID(clid))
1830                 goto out;
1831         clp = find_confirmed_client(clid);
1832         status = nfserr_expired;
1833         if (clp == NULL) {
1834                 /* We assume the client took too long to RENEW. */
1835                 dprintk("nfsd4_renew: clientid not found!\n");
1836                 goto out;
1837         }
1838         renew_client(clp);
1839         status = nfserr_cb_path_down;
1840         if (!list_empty(&clp->cl_delegations)
1841                         && !atomic_read(&clp->cl_callback.cb_set))
1842                 goto out;
1843         status = nfs_ok;
1844 out:
1845         nfs4_unlock_state();
1846         return status;
1847 }
1848
1849 struct lock_manager nfsd4_manager = {
1850 };
1851
1852 static void
1853 nfsd4_end_grace(void)
1854 {
1855         dprintk("NFSD: end of grace period\n");
1856         nfsd4_recdir_purge_old();
1857         locks_end_grace(&nfsd4_manager);
1858 }
1859
1860 static time_t
1861 nfs4_laundromat(void)
1862 {
1863         struct nfs4_client *clp;
1864         struct nfs4_stateowner *sop;
1865         struct nfs4_delegation *dp;
1866         struct list_head *pos, *next, reaplist;
1867         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1868         time_t t, clientid_val = NFSD_LEASE_TIME;
1869         time_t u, test_val = NFSD_LEASE_TIME;
1870
1871         nfs4_lock_state();
1872
1873         dprintk("NFSD: laundromat service - starting\n");
1874         if (locks_in_grace())
1875                 nfsd4_end_grace();
1876         list_for_each_safe(pos, next, &client_lru) {
1877                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1878                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1879                         t = clp->cl_time - cutoff;
1880                         if (clientid_val > t)
1881                                 clientid_val = t;
1882                         break;
1883                 }
1884                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1885                         clp->cl_clientid.cl_id);
1886                 nfsd4_remove_clid_dir(clp);
1887                 expire_client(clp);
1888         }
1889         INIT_LIST_HEAD(&reaplist);
1890         spin_lock(&recall_lock);
1891         list_for_each_safe(pos, next, &del_recall_lru) {
1892                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1893                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1894                         u = dp->dl_time - cutoff;
1895                         if (test_val > u)
1896                                 test_val = u;
1897                         break;
1898                 }
1899                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1900                                     dp, dp->dl_flock);
1901                 list_move(&dp->dl_recall_lru, &reaplist);
1902         }
1903         spin_unlock(&recall_lock);
1904         list_for_each_safe(pos, next, &reaplist) {
1905                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1906                 list_del_init(&dp->dl_recall_lru);
1907                 unhash_delegation(dp);
1908         }
1909         test_val = NFSD_LEASE_TIME;
1910         list_for_each_safe(pos, next, &close_lru) {
1911                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1912                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1913                         u = sop->so_time - cutoff;
1914                         if (test_val > u)
1915                                 test_val = u;
1916                         break;
1917                 }
1918                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1919                         sop->so_id);
1920                 release_openowner(sop);
1921         }
1922         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1923                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1924         nfs4_unlock_state();
1925         return clientid_val;
1926 }
1927
1928 static struct workqueue_struct *laundry_wq;
1929 static void laundromat_main(struct work_struct *);
1930 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1931
1932 static void
1933 laundromat_main(struct work_struct *not_used)
1934 {
1935         time_t t;
1936
1937         t = nfs4_laundromat();
1938         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1939         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1940 }
1941
1942 static struct nfs4_stateowner *
1943 search_close_lru(u32 st_id, int flags)
1944 {
1945         struct nfs4_stateowner *local = NULL;
1946
1947         if (flags & CLOSE_STATE) {
1948                 list_for_each_entry(local, &close_lru, so_close_lru) {
1949                         if (local->so_id == st_id)
1950                                 return local;
1951                 }
1952         }
1953         return NULL;
1954 }
1955
1956 static inline int
1957 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1958 {
1959         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1960 }
1961
1962 static int
1963 STALE_STATEID(stateid_t *stateid)
1964 {
1965         if (stateid->si_boot == boot_time)
1966                 return 0;
1967         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1968                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1969                 stateid->si_generation);
1970         return 1;
1971 }
1972
1973 static inline int
1974 access_permit_read(unsigned long access_bmap)
1975 {
1976         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1977                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1978                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1979 }
1980
1981 static inline int
1982 access_permit_write(unsigned long access_bmap)
1983 {
1984         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1985                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1986 }
1987
1988 static
1989 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1990 {
1991         __be32 status = nfserr_openmode;
1992
1993         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1994                 goto out;
1995         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1996                 goto out;
1997         status = nfs_ok;
1998 out:
1999         return status;
2000 }
2001
2002 static inline __be32
2003 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2004 {
2005         /* Trying to call delegreturn with a special stateid? Yuch: */
2006         if (!(flags & (RD_STATE | WR_STATE)))
2007                 return nfserr_bad_stateid;
2008         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
2009                 return nfs_ok;
2010         else if (locks_in_grace()) {
2011                 /* Answer in remaining cases depends on existance of
2012                  * conflicting state; so we must wait out the grace period. */
2013                 return nfserr_grace;
2014         } else if (flags & WR_STATE)
2015                 return nfs4_share_conflict(current_fh,
2016                                 NFS4_SHARE_DENY_WRITE);
2017         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2018                 return nfs4_share_conflict(current_fh,
2019                                 NFS4_SHARE_DENY_READ);
2020 }
2021
2022 /*
2023  * Allow READ/WRITE during grace period on recovered state only for files
2024  * that are not able to provide mandatory locking.
2025  */
2026 static inline int
2027 io_during_grace_disallowed(struct inode *inode, int flags)
2028 {
2029         return locks_in_grace() && (flags & (RD_STATE | WR_STATE))
2030                 && mandatory_lock(inode);
2031 }
2032
2033 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
2034 {
2035         /* If the client sends us a stateid from the future, it's buggy: */
2036         if (in->si_generation > ref->si_generation)
2037                 return nfserr_bad_stateid;
2038         /*
2039          * The following, however, can happen.  For example, if the
2040          * client sends an open and some IO at the same time, the open
2041          * may bump si_generation while the IO is still in flight.
2042          * Thanks to hard links and renames, the client never knows what
2043          * file an open will affect.  So it could avoid that situation
2044          * only by serializing all opens and IO from the same open
2045          * owner.  To recover from the old_stateid error, the client
2046          * will just have to retry the IO:
2047          */
2048         if (in->si_generation < ref->si_generation)
2049                 return nfserr_old_stateid;
2050         return nfs_ok;
2051 }
2052
2053 /*
2054 * Checks for stateid operations
2055 */
2056 __be32
2057 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2058 {
2059         struct nfs4_stateid *stp = NULL;
2060         struct nfs4_delegation *dp = NULL;
2061         stateid_t *stidp;
2062         struct inode *ino = current_fh->fh_dentry->d_inode;
2063         __be32 status;
2064
2065         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2066                 stateid->si_boot, stateid->si_stateownerid,
2067                 stateid->si_fileid, stateid->si_generation);
2068         if (filpp)
2069                 *filpp = NULL;
2070
2071         if (io_during_grace_disallowed(ino, flags))
2072                 return nfserr_grace;
2073
2074         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2075                 return check_special_stateids(current_fh, stateid, flags);
2076
2077         /* STALE STATEID */
2078         status = nfserr_stale_stateid;
2079         if (STALE_STATEID(stateid))
2080                 goto out;
2081
2082         /* BAD STATEID */
2083         status = nfserr_bad_stateid;
2084         if (!stateid->si_fileid) { /* delegation stateid */
2085                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2086                         dprintk("NFSD: delegation stateid not found\n");
2087                         goto out;
2088                 }
2089                 stidp = &dp->dl_stateid;
2090         } else { /* open or lock stateid */
2091                 if (!(stp = find_stateid(stateid, flags))) {
2092                         dprintk("NFSD: open or lock stateid not found\n");
2093                         goto out;
2094                 }
2095                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2096                         goto out;
2097                 if (!stp->st_stateowner->so_confirmed)
2098                         goto out;
2099                 stidp = &stp->st_stateid;
2100         }
2101         status = check_stateid_generation(stateid, stidp);
2102         if (status)
2103                 goto out;
2104         if (stp) {
2105                 if ((status = nfs4_check_openmode(stp,flags)))
2106                         goto out;
2107                 renew_client(stp->st_stateowner->so_client);
2108                 if (filpp)
2109                         *filpp = stp->st_vfs_file;
2110         } else {
2111                 if ((status = nfs4_check_delegmode(dp, flags)))
2112                         goto out;
2113                 renew_client(dp->dl_client);
2114                 if (flags & DELEG_RET)
2115                         unhash_delegation(dp);
2116                 if (filpp)
2117                         *filpp = dp->dl_vfs_file;
2118         }
2119         status = nfs_ok;
2120 out:
2121         return status;
2122 }
2123
2124 static inline int
2125 setlkflg (int type)
2126 {
2127         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2128                 RD_STATE : WR_STATE;
2129 }
2130
2131 /*
2132  * Checks for sequence id mutating operations.
2133  */
2134 static __be32
2135 nfs4_preprocess_seqid_op(struct svc_fh *current_fh, u32 seqid, stateid_t *stateid, int flags, struct nfs4_stateowner **sopp, struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2136 {
2137         struct nfs4_stateid *stp;
2138         struct nfs4_stateowner *sop;
2139         __be32 status;
2140
2141         dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2142                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2143                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2144                 stateid->si_generation);
2145
2146         *stpp = NULL;
2147         *sopp = NULL;
2148
2149         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2150                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2151                 return nfserr_bad_stateid;
2152         }
2153
2154         if (STALE_STATEID(stateid))
2155                 return nfserr_stale_stateid;
2156         /*
2157         * We return BAD_STATEID if filehandle doesn't match stateid,
2158         * the confirmed flag is incorrecly set, or the generation
2159         * number is incorrect.
2160         */
2161         stp = find_stateid(stateid, flags);
2162         if (stp == NULL) {
2163                 /*
2164                  * Also, we should make sure this isn't just the result of
2165                  * a replayed close:
2166                  */
2167                 sop = search_close_lru(stateid->si_stateownerid, flags);
2168                 if (sop == NULL)
2169                         return nfserr_bad_stateid;
2170                 *sopp = sop;
2171                 goto check_replay;
2172         }
2173
2174         *stpp = stp;
2175         *sopp = sop = stp->st_stateowner;
2176
2177         if (lock) {
2178                 clientid_t *lockclid = &lock->v.new.clientid;
2179                 struct nfs4_client *clp = sop->so_client;
2180                 int lkflg = 0;
2181                 __be32 status;
2182
2183                 lkflg = setlkflg(lock->lk_type);
2184
2185                 if (lock->lk_is_new) {
2186                         if (!sop->so_is_open_owner)
2187                                 return nfserr_bad_stateid;
2188                         if (!same_clid(&clp->cl_clientid, lockclid))
2189                                return nfserr_bad_stateid;
2190                         /* stp is the open stateid */
2191                         status = nfs4_check_openmode(stp, lkflg);
2192                         if (status)
2193                                 return status;
2194                 } else {
2195                         /* stp is the lock stateid */
2196                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2197                         if (status)
2198                                 return status;
2199                }
2200         }
2201
2202         if (nfs4_check_fh(current_fh, stp)) {
2203                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2204                 return nfserr_bad_stateid;
2205         }
2206
2207         /*
2208         *  We now validate the seqid and stateid generation numbers.
2209         *  For the moment, we ignore the possibility of
2210         *  generation number wraparound.
2211         */
2212         if (seqid != sop->so_seqid)
2213                 goto check_replay;
2214
2215         if (sop->so_confirmed && flags & CONFIRM) {
2216                 dprintk("NFSD: preprocess_seqid_op: expected"
2217                                 " unconfirmed stateowner!\n");
2218                 return nfserr_bad_stateid;
2219         }
2220         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2221                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2222                                 " confirmed yet!\n");
2223                 return nfserr_bad_stateid;
2224         }
2225         status = check_stateid_generation(stateid, &stp->st_stateid);
2226         if (status)
2227                 return status;
2228         renew_client(sop->so_client);
2229         return nfs_ok;
2230
2231 check_replay:
2232         if (seqid == sop->so_seqid - 1) {
2233                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2234                 /* indicate replay to calling function */
2235                 return nfserr_replay_me;
2236         }
2237         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2238                         sop->so_seqid, seqid);
2239         *sopp = NULL;
2240         return nfserr_bad_seqid;
2241 }
2242
2243 __be32
2244 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2245                    struct nfsd4_open_confirm *oc)
2246 {
2247         __be32 status;
2248         struct nfs4_stateowner *sop;
2249         struct nfs4_stateid *stp;
2250
2251         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2252                         (int)cstate->current_fh.fh_dentry->d_name.len,
2253                         cstate->current_fh.fh_dentry->d_name.name);
2254
2255         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2256         if (status)
2257                 return status;
2258
2259         nfs4_lock_state();
2260
2261         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2262                                         oc->oc_seqid, &oc->oc_req_stateid,
2263                                         CONFIRM | OPEN_STATE,
2264                                         &oc->oc_stateowner, &stp, NULL)))
2265                 goto out;
2266
2267         sop = oc->oc_stateowner;
2268         sop->so_confirmed = 1;
2269         update_stateid(&stp->st_stateid);
2270         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2271         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2272                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2273                          stp->st_stateid.si_boot,
2274                          stp->st_stateid.si_stateownerid,
2275                          stp->st_stateid.si_fileid,
2276                          stp->st_stateid.si_generation);
2277
2278         nfsd4_create_clid_dir(sop->so_client);
2279 out:
2280         if (oc->oc_stateowner) {
2281                 nfs4_get_stateowner(oc->oc_stateowner);
2282                 cstate->replay_owner = oc->oc_stateowner;
2283         }
2284         nfs4_unlock_state();
2285         return status;
2286 }
2287
2288
2289 /*
2290  * unset all bits in union bitmap (bmap) that
2291  * do not exist in share (from successful OPEN_DOWNGRADE)
2292  */
2293 static void
2294 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2295 {
2296         int i;
2297         for (i = 1; i < 4; i++) {
2298                 if ((i & access) != i)
2299                         __clear_bit(i, bmap);
2300         }
2301 }
2302
2303 static void
2304 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2305 {
2306         int i;
2307         for (i = 0; i < 4; i++) {
2308                 if ((i & deny) != i)
2309                         __clear_bit(i, bmap);
2310         }
2311 }
2312
2313 __be32
2314 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2315                      struct nfsd4_compound_state *cstate,
2316                      struct nfsd4_open_downgrade *od)
2317 {
2318         __be32 status;
2319         struct nfs4_stateid *stp;
2320         unsigned int share_access;
2321
2322         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2323                         (int)cstate->current_fh.fh_dentry->d_name.len,
2324                         cstate->current_fh.fh_dentry->d_name.name);
2325
2326         if (!access_valid(od->od_share_access)
2327                         || !deny_valid(od->od_share_deny))
2328                 return nfserr_inval;
2329
2330         nfs4_lock_state();
2331         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2332                                         od->od_seqid,
2333                                         &od->od_stateid,
2334                                         OPEN_STATE,
2335                                         &od->od_stateowner, &stp, NULL)))
2336                 goto out;
2337
2338         status = nfserr_inval;
2339         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2340                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2341                         stp->st_access_bmap, od->od_share_access);
2342                 goto out;
2343         }
2344         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2345                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2346                         stp->st_deny_bmap, od->od_share_deny);
2347                 goto out;
2348         }
2349         set_access(&share_access, stp->st_access_bmap);
2350         nfs4_file_downgrade(stp->st_vfs_file,
2351                             share_access & ~od->od_share_access);
2352
2353         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2354         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2355
2356         update_stateid(&stp->st_stateid);
2357         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2358         status = nfs_ok;
2359 out:
2360         if (od->od_stateowner) {
2361                 nfs4_get_stateowner(od->od_stateowner);
2362                 cstate->replay_owner = od->od_stateowner;
2363         }
2364         nfs4_unlock_state();
2365         return status;
2366 }
2367
2368 /*
2369  * nfs4_unlock_state() called after encode
2370  */
2371 __be32
2372 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2373             struct nfsd4_close *close)
2374 {
2375         __be32 status;
2376         struct nfs4_stateid *stp;
2377
2378         dprintk("NFSD: nfsd4_close on file %.*s\n",
2379                         (int)cstate->current_fh.fh_dentry->d_name.len,
2380                         cstate->current_fh.fh_dentry->d_name.name);
2381
2382         nfs4_lock_state();
2383         /* check close_lru for replay */
2384         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2385                                         close->cl_seqid,
2386                                         &close->cl_stateid,
2387                                         OPEN_STATE | CLOSE_STATE,
2388                                         &close->cl_stateowner, &stp, NULL)))
2389                 goto out;
2390         status = nfs_ok;
2391         update_stateid(&stp->st_stateid);
2392         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2393
2394         /* release_stateid() calls nfsd_close() if needed */
2395         release_open_stateid(stp);
2396
2397         /* place unused nfs4_stateowners on so_close_lru list to be
2398          * released by the laundromat service after the lease period
2399          * to enable us to handle CLOSE replay
2400          */
2401         if (list_empty(&close->cl_stateowner->so_stateids))
2402                 move_to_close_lru(close->cl_stateowner);
2403 out:
2404         if (close->cl_stateowner) {
2405                 nfs4_get_stateowner(close->cl_stateowner);
2406                 cstate->replay_owner = close->cl_stateowner;
2407         }
2408         nfs4_unlock_state();
2409         return status;
2410 }
2411
2412 __be32
2413 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2414                   struct nfsd4_delegreturn *dr)
2415 {
2416         __be32 status;
2417
2418         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2419                 goto out;
2420
2421         nfs4_lock_state();
2422         status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2423                                             &dr->dr_stateid, DELEG_RET, NULL);
2424         nfs4_unlock_state();
2425 out:
2426         return status;
2427 }
2428
2429
2430 /*
2431  * Lock owner state (byte-range locks)
2432  */
2433 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2434 #define LOCK_HASH_BITS              8
2435 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2436 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2437
2438 static inline u64
2439 end_offset(u64 start, u64 len)
2440 {
2441         u64 end;
2442
2443         end = start + len;
2444         return end >= start ? end: NFS4_MAX_UINT64;
2445 }
2446
2447 /* last octet in a range */
2448 static inline u64
2449 last_byte_offset(u64 start, u64 len)
2450 {
2451         u64 end;
2452
2453         BUG_ON(!len);
2454         end = start + len;
2455         return end > start ? end - 1: NFS4_MAX_UINT64;
2456 }
2457
2458 #define lockownerid_hashval(id) \
2459         ((id) & LOCK_HASH_MASK)
2460
2461 static inline unsigned int
2462 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2463                 struct xdr_netobj *ownername)
2464 {
2465         return (file_hashval(inode) + cl_id
2466                         + opaque_hashval(ownername->data, ownername->len))
2467                 & LOCK_HASH_MASK;
2468 }
2469
2470 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2471 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2472 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2473
2474 static struct nfs4_stateid *
2475 find_stateid(stateid_t *stid, int flags)
2476 {
2477         struct nfs4_stateid *local;
2478         u32 st_id = stid->si_stateownerid;
2479         u32 f_id = stid->si_fileid;
2480         unsigned int hashval;
2481
2482         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2483         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
2484                 hashval = stateid_hashval(st_id, f_id);
2485                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2486                         if ((local->st_stateid.si_stateownerid == st_id) &&
2487                             (local->st_stateid.si_fileid == f_id))
2488                                 return local;
2489                 }
2490         }
2491
2492         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
2493                 hashval = stateid_hashval(st_id, f_id);
2494                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2495                         if ((local->st_stateid.si_stateownerid == st_id) &&
2496                             (local->st_stateid.si_fileid == f_id))
2497                                 return local;
2498                 }
2499         }
2500         return NULL;
2501 }
2502
2503 static struct nfs4_delegation *
2504 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2505 {
2506         struct nfs4_file *fp;
2507         struct nfs4_delegation *dl;
2508
2509         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2510                     stid->si_boot, stid->si_stateownerid,
2511                     stid->si_fileid, stid->si_generation);
2512
2513         fp = find_file(ino);
2514         if (!fp)
2515                 return NULL;
2516         dl = find_delegation_file(fp, stid);
2517         put_nfs4_file(fp);
2518         return dl;
2519 }
2520
2521 /*
2522  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2523  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2524  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2525  * locking, this prevents us from being completely protocol-compliant.  The
2526  * real solution to this problem is to start using unsigned file offsets in
2527  * the VFS, but this is a very deep change!
2528  */
2529 static inline void
2530 nfs4_transform_lock_offset(struct file_lock *lock)
2531 {
2532         if (lock->fl_start < 0)
2533                 lock->fl_start = OFFSET_MAX;
2534         if (lock->fl_end < 0)
2535                 lock->fl_end = OFFSET_MAX;
2536 }
2537
2538 /* Hack!: For now, we're defining this just so we can use a pointer to it
2539  * as a unique cookie to identify our (NFSv4's) posix locks. */
2540 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2541 };
2542
2543 static inline void
2544 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2545 {
2546         struct nfs4_stateowner *sop;
2547         unsigned int hval;
2548
2549         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2550                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2551                 hval = lockownerid_hashval(sop->so_id);
2552                 kref_get(&sop->so_ref);
2553                 deny->ld_sop = sop;
2554                 deny->ld_clientid = sop->so_client->cl_clientid;
2555         } else {
2556                 deny->ld_sop = NULL;
2557                 deny->ld_clientid.cl_boot = 0;
2558                 deny->ld_clientid.cl_id = 0;
2559         }
2560         deny->ld_start = fl->fl_start;
2561         deny->ld_length = NFS4_MAX_UINT64;
2562         if (fl->fl_end != NFS4_MAX_UINT64)
2563                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2564         deny->ld_type = NFS4_READ_LT;
2565         if (fl->fl_type != F_RDLCK)
2566                 deny->ld_type = NFS4_WRITE_LT;
2567 }
2568
2569 static struct nfs4_stateowner *
2570 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2571                 struct xdr_netobj *owner)
2572 {
2573         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2574         struct nfs4_stateowner *op;
2575
2576         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2577                 if (same_owner_str(op, owner, clid))
2578                         return op;
2579         }
2580         return NULL;
2581 }
2582
2583 /*
2584  * Alloc a lock owner structure.
2585  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2586  * occured.
2587  *
2588  * strhashval = lock_ownerstr_hashval
2589  */
2590
2591 static struct nfs4_stateowner *
2592 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2593         struct nfs4_stateowner *sop;
2594         struct nfs4_replay *rp;
2595         unsigned int idhashval;
2596
2597         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2598                 return NULL;
2599         idhashval = lockownerid_hashval(current_ownerid);
2600         INIT_LIST_HEAD(&sop->so_idhash);
2601         INIT_LIST_HEAD(&sop->so_strhash);
2602         INIT_LIST_HEAD(&sop->so_perclient);
2603         INIT_LIST_HEAD(&sop->so_stateids);
2604         INIT_LIST_HEAD(&sop->so_perstateid);
2605         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2606         sop->so_time = 0;
2607         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2608         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2609         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2610         sop->so_is_open_owner = 0;
2611         sop->so_id = current_ownerid++;
2612         sop->so_client = clp;
2613         /* It is the openowner seqid that will be incremented in encode in the
2614          * case of new lockowners; so increment the lock seqid manually: */
2615         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2616         sop->so_confirmed = 1;
2617         rp = &sop->so_replay;
2618         rp->rp_status = nfserr_serverfault;
2619         rp->rp_buflen = 0;
2620         rp->rp_buf = rp->rp_ibuf;
2621         return sop;
2622 }
2623
2624 static struct nfs4_stateid *
2625 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2626 {
2627         struct nfs4_stateid *stp;
2628         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2629
2630         stp = nfs4_alloc_stateid();
2631         if (stp == NULL)
2632                 goto out;
2633         INIT_LIST_HEAD(&stp->st_hash);
2634         INIT_LIST_HEAD(&stp->st_perfile);
2635         INIT_LIST_HEAD(&stp->st_perstateowner);
2636         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2637         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2638         list_add(&stp->st_perfile, &fp->fi_stateids);
2639         list_add(&stp->st_perstateowner, &sop->so_stateids);
2640         stp->st_stateowner = sop;
2641         get_nfs4_file(fp);
2642         stp->st_file = fp;
2643         stp->st_stateid.si_boot = boot_time;
2644         stp->st_stateid.si_stateownerid = sop->so_id;
2645         stp->st_stateid.si_fileid = fp->fi_id;
2646         stp->st_stateid.si_generation = 0;
2647         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2648         stp->st_access_bmap = open_stp->st_access_bmap;
2649         stp->st_deny_bmap = open_stp->st_deny_bmap;
2650         stp->st_openstp = open_stp;
2651
2652 out:
2653         return stp;
2654 }
2655
2656 static int
2657 check_lock_length(u64 offset, u64 length)
2658 {
2659         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
2660              LOFF_OVERFLOW(offset, length)));
2661 }
2662
2663 /*
2664  *  LOCK operation
2665  */
2666 __be32
2667 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2668            struct nfsd4_lock *lock)
2669 {
2670         struct nfs4_stateowner *open_sop = NULL;
2671         struct nfs4_stateowner *lock_sop = NULL;
2672         struct nfs4_stateid *lock_stp;
2673         struct file *filp;
2674         struct file_lock file_lock;
2675         struct file_lock conflock;
2676         __be32 status = 0;
2677         unsigned int strhashval;
2678         unsigned int cmd;
2679         int err;
2680
2681         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2682                 (long long) lock->lk_offset,
2683                 (long long) lock->lk_length);
2684
2685         if (check_lock_length(lock->lk_offset, lock->lk_length))
2686                  return nfserr_inval;
2687
2688         if ((status = fh_verify(rqstp, &cstate->current_fh,
2689                                 S_IFREG, NFSD_MAY_LOCK))) {
2690                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2691                 return status;
2692         }
2693
2694         nfs4_lock_state();
2695
2696         if (lock->lk_is_new) {
2697                 /*
2698                  * Client indicates that this is a new lockowner.
2699                  * Use open owner and open stateid to create lock owner and
2700                  * lock stateid.
2701                  */
2702                 struct nfs4_stateid *open_stp = NULL;
2703                 struct nfs4_file *fp;
2704
2705                 status = nfserr_stale_clientid;
2706                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2707                         goto out;
2708
2709                 /* validate and update open stateid and open seqid */
2710                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2711                                         lock->lk_new_open_seqid,
2712                                         &lock->lk_new_open_stateid,
2713                                         OPEN_STATE,
2714                                         &lock->lk_replay_owner, &open_stp,
2715                                         lock);
2716                 if (status)
2717                         goto out;
2718                 open_sop = lock->lk_replay_owner;
2719                 /* create lockowner and lock stateid */
2720                 fp = open_stp->st_file;
2721                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
2722                                 open_sop->so_client->cl_clientid.cl_id,
2723                                 &lock->v.new.owner);
2724                 /* XXX: Do we need to check for duplicate stateowners on
2725                  * the same file, or should they just be allowed (and
2726                  * create new stateids)? */
2727                 status = nfserr_resource;
2728                 lock_sop = alloc_init_lock_stateowner(strhashval,
2729                                 open_sop->so_client, open_stp, lock);
2730                 if (lock_sop == NULL)
2731                         goto out;
2732                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2733                 if (lock_stp == NULL)
2734                         goto out;
2735         } else {
2736                 /* lock (lock owner + lock stateid) already exists */
2737                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2738                                        lock->lk_old_lock_seqid,
2739                                        &lock->lk_old_lock_stateid,
2740                                        LOCK_STATE,
2741                                        &lock->lk_replay_owner, &lock_stp, lock);
2742                 if (status)
2743                         goto out;
2744                 lock_sop = lock->lk_replay_owner;
2745         }
2746         /* lock->lk_replay_owner and lock_stp have been created or found */
2747         filp = lock_stp->st_vfs_file;
2748
2749         status = nfserr_grace;
2750         if (locks_in_grace() && !lock->lk_reclaim)
2751                 goto out;
2752         status = nfserr_no_grace;
2753         if (!locks_in_grace() && lock->lk_reclaim)
2754                 goto out;
2755
2756         locks_init_lock(&file_lock);
2757         switch (lock->lk_type) {
2758                 case NFS4_READ_LT:
2759                 case NFS4_READW_LT:
2760                         file_lock.fl_type = F_RDLCK;
2761                         cmd = F_SETLK;
2762                 break;
2763                 case NFS4_WRITE_LT:
2764                 case NFS4_WRITEW_LT:
2765                         file_lock.fl_type = F_WRLCK;
2766                         cmd = F_SETLK;
2767                 break;
2768                 default:
2769                         status = nfserr_inval;
2770                 goto out;
2771         }
2772         file_lock.fl_owner = (fl_owner_t)lock_sop;
2773         file_lock.fl_pid = current->tgid;
2774         file_lock.fl_file = filp;
2775         file_lock.fl_flags = FL_POSIX;
2776         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2777
2778         file_lock.fl_start = lock->lk_offset;
2779         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
2780         nfs4_transform_lock_offset(&file_lock);
2781
2782         /*
2783         * Try to lock the file in the VFS.
2784         * Note: locks.c uses the BKL to protect the inode's lock list.
2785         */
2786
2787         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2788         switch (-err) {
2789         case 0: /* success! */
2790                 update_stateid(&lock_stp->st_stateid);
2791                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
2792                                 sizeof(stateid_t));
2793                 status = 0;
2794                 break;
2795         case (EAGAIN):          /* conflock holds conflicting lock */
2796                 status = nfserr_denied;
2797                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2798                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2799                 break;
2800         case (EDEADLK):
2801                 status = nfserr_deadlock;
2802                 break;
2803         default:
2804                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2805                 status = nfserr_resource;
2806                 break;
2807         }
2808 out:
2809         if (status && lock->lk_is_new && lock_sop)
2810                 release_lockowner(lock_sop);
2811         if (lock->lk_replay_owner) {
2812                 nfs4_get_stateowner(lock->lk_replay_owner);
2813                 cstate->replay_owner = lock->lk_replay_owner;
2814         }
2815         nfs4_unlock_state();
2816         return status;
2817 }
2818
2819 /*
2820  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
2821  * so we do a temporary open here just to get an open file to pass to
2822  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
2823  * inode operation.)
2824  */
2825 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
2826 {
2827         struct file *file;
2828         int err;
2829
2830         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
2831         if (err)
2832                 return err;
2833         err = vfs_test_lock(file, lock);
2834         nfsd_close(file);
2835         return err;
2836 }
2837
2838 /*
2839  * LOCKT operation
2840  */
2841 __be32
2842 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2843             struct nfsd4_lockt *lockt)
2844 {
2845         struct inode *inode;
2846         struct file_lock file_lock;
2847         int error;
2848         __be32 status;
2849
2850         if (locks_in_grace())
2851                 return nfserr_grace;
2852
2853         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2854                  return nfserr_inval;
2855
2856         lockt->lt_stateowner = NULL;
2857         nfs4_lock_state();
2858
2859         status = nfserr_stale_clientid;
2860         if (STALE_CLIENTID(&lockt->lt_clientid))
2861                 goto out;
2862
2863         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2864                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2865                 if (status == nfserr_symlink)
2866                         status = nfserr_inval;
2867                 goto out;
2868         }
2869
2870         inode = cstate->current_fh.fh_dentry->d_inode;
2871         locks_init_lock(&file_lock);
2872         switch (lockt->lt_type) {
2873                 case NFS4_READ_LT:
2874                 case NFS4_READW_LT:
2875                         file_lock.fl_type = F_RDLCK;
2876                 break;
2877                 case NFS4_WRITE_LT:
2878                 case NFS4_WRITEW_LT:
2879                         file_lock.fl_type = F_WRLCK;
2880                 break;
2881                 default:
2882                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2883                         status = nfserr_inval;
2884                 goto out;
2885         }
2886
2887         lockt->lt_stateowner = find_lockstateowner_str(inode,
2888                         &lockt->lt_clientid, &lockt->lt_owner);
2889         if (lockt->lt_stateowner)
2890                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2891         file_lock.fl_pid = current->tgid;
2892         file_lock.fl_flags = FL_POSIX;
2893
2894         file_lock.fl_start = lockt->lt_offset;
2895         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
2896
2897         nfs4_transform_lock_offset(&file_lock);
2898
2899         status = nfs_ok;
2900         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
2901         if (error) {
2902                 status = nfserrno(error);
2903                 goto out;
2904         }
2905         if (file_lock.fl_type != F_UNLCK) {
2906                 status = nfserr_denied;
2907                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2908         }
2909 out:
2910         nfs4_unlock_state();
2911         return status;
2912 }
2913
2914 __be32
2915 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2916             struct nfsd4_locku *locku)
2917 {
2918         struct nfs4_stateid *stp;
2919         struct file *filp = NULL;
2920         struct file_lock file_lock;
2921         __be32 status;
2922         int err;
2923
2924         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2925                 (long long) locku->lu_offset,
2926                 (long long) locku->lu_length);
2927
2928         if (check_lock_length(locku->lu_offset, locku->lu_length))
2929                  return nfserr_inval;
2930
2931         nfs4_lock_state();
2932
2933         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2934                                         locku->lu_seqid,
2935                                         &locku->lu_stateid,
2936                                         LOCK_STATE,
2937                                         &locku->lu_stateowner, &stp, NULL)))
2938                 goto out;
2939
2940         filp = stp->st_vfs_file;
2941         BUG_ON(!filp);
2942         locks_init_lock(&file_lock);
2943         file_lock.fl_type = F_UNLCK;
2944         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2945         file_lock.fl_pid = current->tgid;
2946         file_lock.fl_file = filp;
2947         file_lock.fl_flags = FL_POSIX;
2948         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2949         file_lock.fl_start = locku->lu_offset;
2950
2951         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
2952         nfs4_transform_lock_offset(&file_lock);
2953
2954         /*
2955         *  Try to unlock the file in the VFS.
2956         */
2957         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2958         if (err) {
2959                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2960                 goto out_nfserr;
2961         }
2962         /*
2963         * OK, unlock succeeded; the only thing left to do is update the stateid.
2964         */
2965         update_stateid(&stp->st_stateid);
2966         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2967
2968 out:
2969         if (locku->lu_stateowner) {
2970                 nfs4_get_stateowner(locku->lu_stateowner);
2971                 cstate->replay_owner = locku->lu_stateowner;
2972         }
2973         nfs4_unlock_state();
2974         return status;
2975
2976 out_nfserr:
2977         status = nfserrno(err);
2978         goto out;
2979 }
2980
2981 /*
2982  * returns
2983  *      1: locks held by lockowner
2984  *      0: no locks held by lockowner
2985  */
2986 static int
2987 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2988 {
2989         struct file_lock **flpp;
2990         struct inode *inode = filp->f_path.dentry->d_inode;
2991         int status = 0;
2992
2993         lock_kernel();
2994         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2995                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2996                         status = 1;
2997                         goto out;
2998                 }
2999         }
3000 out:
3001         unlock_kernel();
3002         return status;
3003 }
3004
3005 __be32
3006 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3007                         struct nfsd4_compound_state *cstate,
3008                         struct nfsd4_release_lockowner *rlockowner)
3009 {
3010         clientid_t *clid = &rlockowner->rl_clientid;
3011         struct nfs4_stateowner *sop;
3012         struct nfs4_stateid *stp;
3013         struct xdr_netobj *owner = &rlockowner->rl_owner;
3014         struct list_head matches;
3015         int i;
3016         __be32 status;
3017
3018         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3019                 clid->cl_boot, clid->cl_id);
3020
3021         /* XXX check for lease expiration */
3022
3023         status = nfserr_stale_clientid;
3024         if (STALE_CLIENTID(clid))
3025                 return status;
3026
3027         nfs4_lock_state();
3028
3029         status = nfserr_locks_held;
3030         /* XXX: we're doing a linear search through all the lockowners.
3031          * Yipes!  For now we'll just hope clients aren't really using
3032          * release_lockowner much, but eventually we have to fix these
3033          * data structures. */
3034         INIT_LIST_HEAD(&matches);
3035         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3036                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3037                         if (!same_owner_str(sop, owner, clid))
3038                                 continue;
3039                         list_for_each_entry(stp, &sop->so_stateids,
3040                                         st_perstateowner) {
3041                                 if (check_for_locks(stp->st_vfs_file, sop))
3042                                         goto out;
3043                                 /* Note: so_perclient unused for lockowners,
3044                                  * so it's OK to fool with here. */
3045                                 list_add(&sop->so_perclient, &matches);
3046                         }
3047                 }
3048         }
3049         /* Clients probably won't expect us to return with some (but not all)
3050          * of the lockowner state released; so don't release any until all
3051          * have been checked. */
3052         status = nfs_ok;
3053         while (!list_empty(&matches)) {
3054                 sop = list_entry(matches.next, struct nfs4_stateowner,
3055                                                                 so_perclient);
3056                 /* unhash_stateowner deletes so_perclient only
3057                  * for openowners. */
3058                 list_del(&sop->so_perclient);
3059                 release_lockowner(sop);
3060         }
3061 out:
3062         nfs4_unlock_state();
3063         return status;
3064 }
3065
3066 static inline struct nfs4_client_reclaim *
3067 alloc_reclaim(void)
3068 {
3069         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3070 }
3071
3072 int
3073 nfs4_has_reclaimed_state(const char *name)
3074 {
3075         unsigned int strhashval = clientstr_hashval(name);
3076         struct nfs4_client *clp;
3077
3078         clp = find_confirmed_client_by_str(name, strhashval);
3079         return clp ? 1 : 0;
3080 }
3081
3082 /*
3083  * failure => all reset bets are off, nfserr_no_grace...
3084  */
3085 int
3086 nfs4_client_to_reclaim(const char *name)
3087 {
3088         unsigned int strhashval;
3089         struct nfs4_client_reclaim *crp = NULL;
3090
3091         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3092         crp = alloc_reclaim();
3093         if (!crp)
3094                 return 0;
3095         strhashval = clientstr_hashval(name);
3096         INIT_LIST_HEAD(&crp->cr_strhash);
3097         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3098         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3099         reclaim_str_hashtbl_size++;
3100         return 1;
3101 }
3102
3103 static void
3104 nfs4_release_reclaim(void)
3105 {
3106         struct nfs4_client_reclaim *crp = NULL;
3107         int i;
3108
3109         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3110                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3111                         crp = list_entry(reclaim_str_hashtbl[i].next,
3112                                         struct nfs4_client_reclaim, cr_strhash);
3113                         list_del(&crp->cr_strhash);
3114                         kfree(crp);
3115                         reclaim_str_hashtbl_size--;
3116                 }
3117         }
3118         BUG_ON(reclaim_str_hashtbl_size);
3119 }
3120
3121 /*
3122  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3123 static struct nfs4_client_reclaim *
3124 nfs4_find_reclaim_client(clientid_t *clid)
3125 {
3126         unsigned int strhashval;
3127         struct nfs4_client *clp;
3128         struct nfs4_client_reclaim *crp = NULL;
3129
3130
3131         /* find clientid in conf_id_hashtbl */
3132         clp = find_confirmed_client(clid);
3133         if (clp == NULL)
3134                 return NULL;
3135
3136         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3137                             clp->cl_name.len, clp->cl_name.data,
3138                             clp->cl_recdir);
3139
3140         /* find clp->cl_name in reclaim_str_hashtbl */
3141         strhashval = clientstr_hashval(clp->cl_recdir);
3142         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3143                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3144                         return crp;
3145                 }
3146         }
3147         return NULL;
3148 }
3149
3150 /*
3151 * Called from OPEN. Look for clientid in reclaim list.
3152 */
3153 __be32
3154 nfs4_check_open_reclaim(clientid_t *clid)
3155 {
3156         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3157 }
3158
3159 /* initialization to perform at module load time: */
3160
3161 int
3162 nfs4_state_init(void)
3163 {
3164         int i, status;
3165
3166         status = nfsd4_init_slabs();
3167         if (status)
3168                 return status;
3169         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3170                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3171                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3172                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3173                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3174         }
3175         for (i = 0; i < FILE_HASH_SIZE; i++) {
3176                 INIT_LIST_HEAD(&file_hashtbl[i]);
3177         }
3178         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3179                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3180                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3181         }
3182         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3183                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3184                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3185         }
3186         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3187                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3188                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3189         }
3190         memset(&onestateid, ~0, sizeof(stateid_t));
3191         INIT_LIST_HEAD(&close_lru);
3192         INIT_LIST_HEAD(&client_lru);
3193         INIT_LIST_HEAD(&del_recall_lru);
3194         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3195                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3196         reclaim_str_hashtbl_size = 0;
3197         return 0;
3198 }
3199
3200 static void
3201 nfsd4_load_reboot_recovery_data(void)
3202 {
3203         int status;
3204
3205         nfs4_lock_state();
3206         nfsd4_init_recdir(user_recovery_dirname);
3207         status = nfsd4_recdir_load();
3208         nfs4_unlock_state();
3209         if (status)
3210                 printk("NFSD: Failure reading reboot recovery data\n");
3211 }
3212
3213 unsigned long
3214 get_nfs4_grace_period(void)
3215 {
3216         return max(user_lease_time, lease_time) * HZ;
3217 }
3218
3219 /*
3220  * Since the lifetime of a delegation isn't limited to that of an open, a
3221  * client may quite reasonably hang on to a delegation as long as it has
3222  * the inode cached.  This becomes an obvious problem the first time a
3223  * client's inode cache approaches the size of the server's total memory.
3224  *
3225  * For now we avoid this problem by imposing a hard limit on the number
3226  * of delegations, which varies according to the server's memory size.
3227  */
3228 static void
3229 set_max_delegations(void)
3230 {
3231         /*
3232          * Allow at most 4 delegations per megabyte of RAM.  Quick
3233          * estimates suggest that in the worst case (where every delegation
3234          * is for a different inode), a delegation could take about 1.5K,
3235          * giving a worst case usage of about 6% of memory.
3236          */
3237         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3238 }
3239
3240 /* initialization to perform when the nfsd service is started: */
3241
3242 static void
3243 __nfs4_state_start(void)
3244 {
3245         unsigned long grace_time;
3246
3247         boot_time = get_seconds();
3248         grace_time = get_nfs4_grace_period();
3249         lease_time = user_lease_time;
3250         locks_start_grace(&nfsd4_manager);
3251         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3252                grace_time/HZ);
3253         laundry_wq = create_singlethread_workqueue("nfsd4");
3254         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3255         set_max_delegations();
3256 }
3257
3258 void
3259 nfs4_state_start(void)
3260 {
3261         if (nfs4_init)
3262                 return;
3263         nfsd4_load_reboot_recovery_data();
3264         __nfs4_state_start();
3265         nfs4_init = 1;
3266         return;
3267 }
3268
3269 time_t
3270 nfs4_lease_time(void)
3271 {
3272         return lease_time;
3273 }
3274
3275 static void
3276 __nfs4_state_shutdown(void)
3277 {
3278         int i;
3279         struct nfs4_client *clp = NULL;
3280         struct nfs4_delegation *dp = NULL;
3281         struct list_head *pos, *next, reaplist;
3282
3283         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3284                 while (!list_empty(&conf_id_hashtbl[i])) {
3285                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3286                         expire_client(clp);
3287                 }
3288                 while (!list_empty(&unconf_str_hashtbl[i])) {
3289                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3290                         expire_client(clp);
3291                 }
3292         }
3293         INIT_LIST_HEAD(&reaplist);
3294         spin_lock(&recall_lock);
3295         list_for_each_safe(pos, next, &del_recall_lru) {
3296                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3297                 list_move(&dp->dl_recall_lru, &reaplist);
3298         }
3299         spin_unlock(&recall_lock);
3300         list_for_each_safe(pos, next, &reaplist) {
3301                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3302                 list_del_init(&dp->dl_recall_lru);
3303                 unhash_delegation(dp);
3304         }
3305
3306         nfsd4_shutdown_recdir();
3307         nfs4_init = 0;
3308 }
3309
3310 void
3311 nfs4_state_shutdown(void)
3312 {
3313         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3314         destroy_workqueue(laundry_wq);
3315         locks_end_grace(&nfsd4_manager);
3316         nfs4_lock_state();
3317         nfs4_release_reclaim();
3318         __nfs4_state_shutdown();
3319         nfs4_unlock_state();
3320 }
3321
3322 /*
3323  * user_recovery_dirname is protected by the nfsd_mutex since it's only
3324  * accessed when nfsd is starting.
3325  */
3326 static void
3327 nfs4_set_recdir(char *recdir)
3328 {
3329         strcpy(user_recovery_dirname, recdir);
3330 }
3331
3332 /*
3333  * Change the NFSv4 recovery directory to recdir.
3334  */
3335 int
3336 nfs4_reset_recoverydir(char *recdir)
3337 {
3338         int status;
3339         struct path path;
3340
3341         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
3342         if (status)
3343                 return status;
3344         status = -ENOTDIR;
3345         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
3346                 nfs4_set_recdir(recdir);
3347                 status = 0;
3348         }
3349         path_put(&path);
3350         return status;
3351 }
3352
3353 char *
3354 nfs4_recoverydir(void)
3355 {
3356         return user_recovery_dirname;
3357 }
3358
3359 /*
3360  * Called when leasetime is changed.
3361  *
3362  * The only way the protocol gives us to handle on-the-fly lease changes is to
3363  * simulate a reboot.  Instead of doing that, we just wait till the next time
3364  * we start to register any changes in lease time.  If the administrator
3365  * really wants to change the lease time *now*, they can go ahead and bring
3366  * nfsd down and then back up again after changing the lease time.
3367  *
3368  * user_lease_time is protected by nfsd_mutex since it's only really accessed
3369  * when nfsd is starting
3370  */
3371 void
3372 nfs4_reset_lease(time_t leasetime)
3373 {
3374         user_lease_time = leasetime;
3375 }