git.oblomov.eu Git - 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
  58 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  59
  60 /* Globals */
  61 static time_t lease_time = 90;     /* default lease time */
  62 static time_t user_lease_time = 90;
  63 static time_t boot_time;
  64 static int in_grace = 1;
  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 void release_stateid_lockowners(struct nfs4_stateid *open_stp);
  79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  80 static void nfs4_set_recdir(char *recdir);
  81
  82 /* Locking:
  83  *
  84  * client_mutex:
  85  *      protects clientid_hashtbl[], clientstr_hashtbl[],
  86  *      unconfstr_hashtbl[], uncofid_hashtbl[].
  87  */
  88 static DEFINE_MUTEX(client_mutex);
  89
  90 static struct kmem_cache *stateowner_slab = NULL;
  91 static struct kmem_cache *file_slab = NULL;
  92 static struct kmem_cache *stateid_slab = NULL;
  93 static struct kmem_cache *deleg_slab = NULL;
  94
  95 void
  96 nfs4_lock_state(void)
  97 {
  98         mutex_lock(&client_mutex);
  99 }
 100
 101 void
 102 nfs4_unlock_state(void)
 103 {
 104         mutex_unlock(&client_mutex);
 105 }
 106
 107 static inline u32
 108 opaque_hashval(const void *ptr, int nbytes)
 109 {
 110         unsigned char *cptr = (unsigned char *) ptr;
 111
 112         u32 x = 0;
 113         while (nbytes--) {
 114                 x *= 37;
 115                 x += *cptr++;
 116         }
 117         return x;
 118 }
 119
 120 /* forward declarations */
 121 static void release_stateowner(struct nfs4_stateowner *sop);
 122 static void release_stateid(struct nfs4_stateid *stp, int flags);
 123
 124 /*
 125  * Delegation state
 126  */
 127
 128 /* recall_lock protects the del_recall_lru */
 129 static DEFINE_SPINLOCK(recall_lock);
 130 static struct list_head del_recall_lru;
 131
 132 static void
 133 free_nfs4_file(struct kref *kref)
 134 {
 135         struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
 136         list_del(&fp->fi_hash);
 137         iput(fp->fi_inode);
 138         kmem_cache_free(file_slab, fp);
 139 }
 140
 141 static inline void
 142 put_nfs4_file(struct nfs4_file *fi)
 143 {
 144         kref_put(&fi->fi_ref, free_nfs4_file);
 145 }
 146
 147 static inline void
 148 get_nfs4_file(struct nfs4_file *fi)
 149 {
 150         kref_get(&fi->fi_ref);
 151 }
 152
 153 static int num_delegations;
 154 unsigned int max_delegations;
 155
 156 /*
 157  * Open owner state (share locks)
 158  */
 159
 160 /* hash tables for nfs4_stateowner */
 161 #define OWNER_HASH_BITS              8
 162 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 163 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 164
 165 #define ownerid_hashval(id) \
 166         ((id) & OWNER_HASH_MASK)
 167 #define ownerstr_hashval(clientid, ownername) \
 168         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
 169
 170 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
 171 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 172
 173 /* hash table for nfs4_file */
 174 #define FILE_HASH_BITS                   8
 175 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 176 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
 177 /* hash table for (open)nfs4_stateid */
 178 #define STATEID_HASH_BITS              10
 179 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
 180 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
 181
 182 #define file_hashval(x) \
 183         hash_ptr(x, FILE_HASH_BITS)
 184 #define stateid_hashval(owner_id, file_id)  \
 185         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
 186
 187 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 188 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 189
 190 static struct nfs4_delegation *
 191 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 192 {
 193         struct nfs4_delegation *dp;
 194         struct nfs4_file *fp = stp->st_file;
 195         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
 196
 197         dprintk("NFSD alloc_init_deleg\n");
 198         if (fp->fi_had_conflict)
 199                 return NULL;
 200         if (num_delegations > max_delegations)
 201                 return NULL;
 202         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
 203         if (dp == NULL)
 204                 return dp;
 205         num_delegations++;
 206         INIT_LIST_HEAD(&dp->dl_perfile);
 207         INIT_LIST_HEAD(&dp->dl_perclnt);
 208         INIT_LIST_HEAD(&dp->dl_recall_lru);
 209         dp->dl_client = clp;
 210         get_nfs4_file(fp);
 211         dp->dl_file = fp;
 212         dp->dl_flock = NULL;
 213         get_file(stp->st_vfs_file);
 214         dp->dl_vfs_file = stp->st_vfs_file;
 215         dp->dl_type = type;
 216         dp->dl_recall.cbr_dp = NULL;
 217         dp->dl_recall.cbr_ident = cb->cb_ident;
 218         dp->dl_recall.cbr_trunc = 0;
 219         dp->dl_stateid.si_boot = boot_time;
 220         dp->dl_stateid.si_stateownerid = current_delegid++;
 221         dp->dl_stateid.si_fileid = 0;
 222         dp->dl_stateid.si_generation = 0;
 223         dp->dl_fhlen = current_fh->fh_handle.fh_size;
 224         memcpy(dp->dl_fhval, &current_fh->fh_handle.fh_base,
 225                         current_fh->fh_handle.fh_size);
 226         dp->dl_time = 0;
 227         atomic_set(&dp->dl_count, 1);
 228         list_add(&dp->dl_perfile, &fp->fi_delegations);
 229         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 230         return dp;
 231 }
 232
 233 void
 234 nfs4_put_delegation(struct nfs4_delegation *dp)
 235 {
 236         if (atomic_dec_and_test(&dp->dl_count)) {
 237                 dprintk("NFSD: freeing dp %p\n",dp);
 238                 put_nfs4_file(dp->dl_file);
 239                 kmem_cache_free(deleg_slab, dp);
 240                 num_delegations--;
 241         }
 242 }
 243
 244 /* Remove the associated file_lock first, then remove the delegation.
 245  * lease_modify() is called to remove the FS_LEASE file_lock from
 246  * the i_flock list, eventually calling nfsd's lock_manager
 247  * fl_release_callback.
 248  */
 249 static void
 250 nfs4_close_delegation(struct nfs4_delegation *dp)
 251 {
 252         struct file *filp = dp->dl_vfs_file;
 253
 254         dprintk("NFSD: close_delegation dp %p\n",dp);
 255         dp->dl_vfs_file = NULL;
 256         /* The following nfsd_close may not actually close the file,
 257          * but we want to remove the lease in any case. */
 258         if (dp->dl_flock)
 259                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
 260         nfsd_close(filp);
 261 }
 262
 263 /* Called under the state lock. */
 264 static void
 265 unhash_delegation(struct nfs4_delegation *dp)
 266 {
 267         list_del_init(&dp->dl_perfile);
 268         list_del_init(&dp->dl_perclnt);
 269         spin_lock(&recall_lock);
 270         list_del_init(&dp->dl_recall_lru);
 271         spin_unlock(&recall_lock);
 272         nfs4_close_delegation(dp);
 273         nfs4_put_delegation(dp);
 274 }
 275
 276 /*
 277  * SETCLIENTID state
 278  */
 279
 280 /* Hash tables for nfs4_clientid state */
 281 #define CLIENT_HASH_BITS                 4
 282 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 283 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 284
 285 #define clientid_hashval(id) \
 286         ((id) & CLIENT_HASH_MASK)
 287 #define clientstr_hashval(name) \
 288         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
 289 /*
 290  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 291  * used in reboot/reset lease grace period processing
 292  *
 293  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 294  * setclientid_confirmed info.
 295  *
 296  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 297  * setclientid info.
 298  *
 299  * client_lru holds client queue ordered by nfs4_client.cl_time
 300  * for lease renewal.
 301  *
 302  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 303  * for last close replay.
 304  */
 305 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 306 static int reclaim_str_hashtbl_size = 0;
 307 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 308 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 309 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 310 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 311 static struct list_head client_lru;
 312 static struct list_head close_lru;
 313
 314 static inline void
 315 renew_client(struct nfs4_client *clp)
 316 {
 317         /*
 318         * Move client to the end to the LRU list.
 319         */
 320         dprintk("renewing client (clientid %08x/%08x)\n",
 321                         clp->cl_clientid.cl_boot,
 322                         clp->cl_clientid.cl_id);
 323         list_move_tail(&clp->cl_lru, &client_lru);
 324         clp->cl_time = get_seconds();
 325 }
 326
 327 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 328 static int
 329 STALE_CLIENTID(clientid_t *clid)
 330 {
 331         if (clid->cl_boot == boot_time)
 332                 return 0;
 333         dprintk("NFSD stale clientid (%08x/%08x)\n",
 334                         clid->cl_boot, clid->cl_id);
 335         return 1;
 336 }
 337
 338 /*
 339  * XXX Should we use a slab cache ?
 340  * This type of memory management is somewhat inefficient, but we use it
 341  * anyway since SETCLIENTID is not a common operation.
 342  */
 343 static struct nfs4_client *alloc_client(struct xdr_netobj name)
 344 {
 345         struct nfs4_client *clp;
 346
 347         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
 348         if (clp == NULL)
 349                 return NULL;
 350         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
 351         if (clp->cl_name.data == NULL) {
 352                 kfree(clp);
 353                 return NULL;
 354         }
 355         memcpy(clp->cl_name.data, name.data, name.len);
 356         clp->cl_name.len = name.len;
 357         return clp;
 358 }
 359
 360 static void
 361 shutdown_callback_client(struct nfs4_client *clp)
 362 {
 363         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
 364
 365         if (clnt) {
 366                 /*
 367                  * Callback threads take a reference on the client, so there
 368                  * should be no outstanding callbacks at this point.
 369                  */
 370                 clp->cl_callback.cb_client = NULL;
 371                 rpc_shutdown_client(clnt);
 372         }
 373 }
 374
 375 static inline void
 376 free_client(struct nfs4_client *clp)
 377 {
 378         shutdown_callback_client(clp);
 379         if (clp->cl_cred.cr_group_info)
 380                 put_group_info(clp->cl_cred.cr_group_info);
 381         kfree(clp->cl_name.data);
 382         kfree(clp);
 383 }
 384
 385 void
 386 put_nfs4_client(struct nfs4_client *clp)
 387 {
 388         if (atomic_dec_and_test(&clp->cl_count))
 389                 free_client(clp);
 390 }
 391
 392 static void
 393 expire_client(struct nfs4_client *clp)
 394 {
 395         struct nfs4_stateowner *sop;
 396         struct nfs4_delegation *dp;
 397         struct list_head reaplist;
 398
 399         dprintk("NFSD: expire_client cl_count %d\n",
 400                             atomic_read(&clp->cl_count));
 401
 402         INIT_LIST_HEAD(&reaplist);
 403         spin_lock(&recall_lock);
 404         while (!list_empty(&clp->cl_delegations)) {
 405                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
 406                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
 407                                 dp->dl_flock);
 408                 list_del_init(&dp->dl_perclnt);
 409                 list_move(&dp->dl_recall_lru, &reaplist);
 410         }
 411         spin_unlock(&recall_lock);
 412         while (!list_empty(&reaplist)) {
 413                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
 414                 list_del_init(&dp->dl_recall_lru);
 415                 unhash_delegation(dp);
 416         }
 417         list_del(&clp->cl_idhash);
 418         list_del(&clp->cl_strhash);
 419         list_del(&clp->cl_lru);
 420         while (!list_empty(&clp->cl_openowners)) {
 421                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
 422                 release_stateowner(sop);
 423         }
 424         put_nfs4_client(clp);
 425 }
 426
 427 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
 428 {
 429         struct nfs4_client *clp;
 430
 431         clp = alloc_client(name);
 432         if (clp == NULL)
 433                 return NULL;
 434         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
 435         atomic_set(&clp->cl_count, 1);
 436         atomic_set(&clp->cl_callback.cb_set, 0);
 437         INIT_LIST_HEAD(&clp->cl_idhash);
 438         INIT_LIST_HEAD(&clp->cl_strhash);
 439         INIT_LIST_HEAD(&clp->cl_openowners);
 440         INIT_LIST_HEAD(&clp->cl_delegations);
 441         INIT_LIST_HEAD(&clp->cl_lru);
 442         return clp;
 443 }
 444
 445 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
 446 {
 447         memcpy(target->cl_verifier.data, source->data,
 448                         sizeof(target->cl_verifier.data));
 449 }
 450
 451 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
 452 {
 453         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
 454         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
 455 }
 456
 457 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
 458 {
 459         target->cr_uid = source->cr_uid;
 460         target->cr_gid = source->cr_gid;
 461         target->cr_group_info = source->cr_group_info;
 462         get_group_info(target->cr_group_info);
 463 }
 464
 465 static int same_name(const char *n1, const char *n2)
 466 {
 467         return 0 == memcmp(n1, n2, HEXDIR_LEN);
 468 }
 469
 470 static int
 471 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
 472 {
 473         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
 474 }
 475
 476 static int
 477 same_clid(clientid_t *cl1, clientid_t *cl2)
 478 {
 479         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
 480 }
 481
 482 /* XXX what about NGROUP */
 483 static int
 484 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
 485 {
 486         return cr1->cr_uid == cr2->cr_uid;
 487 }
 488
 489 static void gen_clid(struct nfs4_client *clp)
 490 {
 491         static u32 current_clientid = 1;
 492
 493         clp->cl_clientid.cl_boot = boot_time;
 494         clp->cl_clientid.cl_id = current_clientid++;
 495 }
 496
 497 static void gen_confirm(struct nfs4_client *clp)
 498 {
 499         static u32 i;
 500         u32 *p;
 501
 502         p = (u32 *)clp->cl_confirm.data;
 503         *p++ = get_seconds();
 504         *p++ = i++;
 505 }
 506
 507 static int check_name(struct xdr_netobj name)
 508 {
 509         if (name.len == 0)
 510                 return 0;
 511         if (name.len > NFS4_OPAQUE_LIMIT) {
 512                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
 513                 return 0;
 514         }
 515         return 1;
 516 }
 517
 518 static void
 519 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
 520 {
 521         unsigned int idhashval;
 522
 523         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
 524         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 525         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
 526         list_add_tail(&clp->cl_lru, &client_lru);
 527         clp->cl_time = get_seconds();
 528 }
 529
 530 static void
 531 move_to_confirmed(struct nfs4_client *clp)
 532 {
 533         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 534         unsigned int strhashval;
 535
 536         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
 537         list_del_init(&clp->cl_strhash);
 538         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
 539         strhashval = clientstr_hashval(clp->cl_recdir);
 540         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
 541         renew_client(clp);
 542 }
 543
 544 static struct nfs4_client *
 545 find_confirmed_client(clientid_t *clid)
 546 {
 547         struct nfs4_client *clp;
 548         unsigned int idhashval = clientid_hashval(clid->cl_id);
 549
 550         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
 551                 if (same_clid(&clp->cl_clientid, clid))
 552                         return clp;
 553         }
 554         return NULL;
 555 }
 556
 557 static struct nfs4_client *
 558 find_unconfirmed_client(clientid_t *clid)
 559 {
 560         struct nfs4_client *clp;
 561         unsigned int idhashval = clientid_hashval(clid->cl_id);
 562
 563         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
 564                 if (same_clid(&clp->cl_clientid, clid))
 565                         return clp;
 566         }
 567         return NULL;
 568 }
 569
 570 static struct nfs4_client *
 571 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
 572 {
 573         struct nfs4_client *clp;
 574
 575         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
 576                 if (same_name(clp->cl_recdir, dname))
 577                         return clp;
 578         }
 579         return NULL;
 580 }
 581
 582 static struct nfs4_client *
 583 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
 584 {
 585         struct nfs4_client *clp;
 586
 587         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
 588                 if (same_name(clp->cl_recdir, dname))
 589                         return clp;
 590         }
 591         return NULL;
 592 }
 593
 594 /* a helper function for parse_callback */
 595 static int
 596 parse_octet(unsigned int *lenp, char **addrp)
 597 {
 598         unsigned int len = *lenp;
 599         char *p = *addrp;
 600         int n = -1;
 601         char c;
 602
 603         for (;;) {
 604                 if (!len)
 605                         break;
 606                 len--;
 607                 c = *p++;
 608                 if (c == '.')
 609                         break;
 610                 if ((c < '0') || (c > '9')) {
 611                         n = -1;
 612                         break;
 613                 }
 614                 if (n < 0)
 615                         n = 0;
 616                 n = (n * 10) + (c - '0');
 617                 if (n > 255) {
 618                         n = -1;
 619                         break;
 620                 }
 621         }
 622         *lenp = len;
 623         *addrp = p;
 624         return n;
 625 }
 626
 627 /* parse and set the setclientid ipv4 callback address */
 628 static int
 629 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
 630 {
 631         int temp = 0;
 632         u32 cbaddr = 0;
 633         u16 cbport = 0;
 634         u32 addrlen = addr_len;
 635         char *addr = addr_val;
 636         int i, shift;
 637
 638         /* ipaddress */
 639         shift = 24;
 640         for(i = 4; i > 0  ; i--) {
 641                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 642                         return 0;
 643                 }
 644                 cbaddr |= (temp << shift);
 645                 if (shift > 0)
 646                 shift -= 8;
 647         }
 648         *cbaddrp = cbaddr;
 649
 650         /* port */
 651         shift = 8;
 652         for(i = 2; i > 0  ; i--) {
 653                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 654                         return 0;
 655                 }
 656                 cbport |= (temp << shift);
 657                 if (shift > 0)
 658                         shift -= 8;
 659         }
 660         *cbportp = cbport;
 661         return 1;
 662 }
 663
 664 static void
 665 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
 666 {
 667         struct nfs4_callback *cb = &clp->cl_callback;
 668
 669         /* Currently, we only support tcp for the callback channel */
 670         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
 671                 goto out_err;
 672
 673         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
 674                          &cb->cb_addr, &cb->cb_port)))
 675                 goto out_err;
 676         cb->cb_prog = se->se_callback_prog;
 677         cb->cb_ident = se->se_callback_ident;
 678         return;
 679 out_err:
 680         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
 681                 "will not receive delegations\n",
 682                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
 683
 684         return;
 685 }
 686
 687 __be32
 688 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 689                   struct nfsd4_setclientid *setclid)
 690 {
 691         struct sockaddr_in      *sin = svc_addr_in(rqstp);
 692         struct xdr_netobj       clname = {
 693                 .len = setclid->se_namelen,
 694                 .data = setclid->se_name,
 695         };
 696         nfs4_verifier           clverifier = setclid->se_verf;
 697         unsigned int            strhashval;
 698         struct nfs4_client      *conf, *unconf, *new;
 699         __be32                  status;
 700         char                    dname[HEXDIR_LEN];
 701
 702         if (!check_name(clname))
 703                 return nfserr_inval;
 704
 705         status = nfs4_make_rec_clidname(dname, &clname);
 706         if (status)
 707                 return status;
 708
 709         /*
 710          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 711          * We get here on a DRC miss.
 712          */
 713
 714         strhashval = clientstr_hashval(dname);
 715
 716         nfs4_lock_state();
 717         conf = find_confirmed_client_by_str(dname, strhashval);
 718         if (conf) {
 719                 /* RFC 3530 14.2.33 CASE 0: */
 720                 status = nfserr_clid_inuse;
 721                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
 722                                 || conf->cl_addr != sin->sin_addr.s_addr) {
 723                         dprintk("NFSD: setclientid: string in use by client"
 724                                 "at %u.%u.%u.%u\n", NIPQUAD(conf->cl_addr));
 725                         goto out;
 726                 }
 727         }
 728         /*
 729          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
 730          * has a description of SETCLIENTID request processing consisting
 731          * of 5 bullet points, labeled as CASE0 - CASE4 below.
 732          */
 733         unconf = find_unconfirmed_client_by_str(dname, strhashval);
 734         status = nfserr_resource;
 735         if (!conf) {
 736                 /*
 737                  * RFC 3530 14.2.33 CASE 4:
 738                  * placed first, because it is the normal case
 739                  */
 740                 if (unconf)
 741                         expire_client(unconf);
 742                 new = create_client(clname, dname);
 743                 if (new == NULL)
 744                         goto out;
 745                 gen_clid(new);
 746         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
 747                 /*
 748                  * RFC 3530 14.2.33 CASE 1:
 749                  * probable callback update
 750                  */
 751                 if (unconf) {
 752                         /* Note this is removing unconfirmed {*x***},
 753                          * which is stronger than RFC recommended {vxc**}.
 754                          * This has the advantage that there is at most
 755                          * one {*x***} in either list at any time.
 756                          */
 757                         expire_client(unconf);
 758                 }
 759                 new = create_client(clname, dname);
 760                 if (new == NULL)
 761                         goto out;
 762                 copy_clid(new, conf);
 763         } else if (!unconf) {
 764                 /*
 765                  * RFC 3530 14.2.33 CASE 2:
 766                  * probable client reboot; state will be removed if
 767                  * confirmed.
 768                  */
 769                 new = create_client(clname, dname);
 770                 if (new == NULL)
 771                         goto out;
 772                 gen_clid(new);
 773         } else {
 774                 /*
 775                  * RFC 3530 14.2.33 CASE 3:
 776                  * probable client reboot; state will be removed if
 777                  * confirmed.
 778                  */
 779                 expire_client(unconf);
 780                 new = create_client(clname, dname);
 781                 if (new == NULL)
 782                         goto out;
 783                 gen_clid(new);
 784         }
 785         copy_verf(new, &clverifier);
 786         new->cl_addr = sin->sin_addr.s_addr;
 787         copy_cred(&new->cl_cred, &rqstp->rq_cred);
 788         gen_confirm(new);
 789         gen_callback(new, setclid);
 790         add_to_unconfirmed(new, strhashval);
 791         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
 792         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
 793         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
 794         status = nfs_ok;
 795 out:
 796         nfs4_unlock_state();
 797         return status;
 798 }
 799
 800
 801 /*
 802  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
 803  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
 804  * bullets, labeled as CASE1 - CASE4 below.
 805  */
 806 __be32
 807 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
 808                          struct nfsd4_compound_state *cstate,
 809                          struct nfsd4_setclientid_confirm *setclientid_confirm)
 810 {
 811         struct sockaddr_in *sin = svc_addr_in(rqstp);
 812         struct nfs4_client *conf, *unconf;
 813         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
 814         clientid_t * clid = &setclientid_confirm->sc_clientid;
 815         __be32 status;
 816
 817         if (STALE_CLIENTID(clid))
 818                 return nfserr_stale_clientid;
 819         /*
 820          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 821          * We get here on a DRC miss.
 822          */
 823
 824         nfs4_lock_state();
 825
 826         conf = find_confirmed_client(clid);
 827         unconf = find_unconfirmed_client(clid);
 828
 829         status = nfserr_clid_inuse;
 830         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
 831                 goto out;
 832         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
 833                 goto out;
 834
 835         /*
 836          * section 14.2.34 of RFC 3530 has a description of
 837          * SETCLIENTID_CONFIRM request processing consisting
 838          * of 4 bullet points, labeled as CASE1 - CASE4 below.
 839          */
 840         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
 841                 /*
 842                  * RFC 3530 14.2.34 CASE 1:
 843                  * callback update
 844                  */
 845                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
 846                         status = nfserr_clid_inuse;
 847                 else {
 848                         /* XXX: We just turn off callbacks until we can handle
 849                           * change request correctly. */
 850                         atomic_set(&conf->cl_callback.cb_set, 0);
 851                         gen_confirm(conf);
 852                         nfsd4_remove_clid_dir(unconf);
 853                         expire_client(unconf);
 854                         status = nfs_ok;
 855
 856                 }
 857         } else if (conf && !unconf) {
 858                 /*
 859                  * RFC 3530 14.2.34 CASE 2:
 860                  * probable retransmitted request; play it safe and
 861                  * do nothing.
 862                  */
 863                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
 864                         status = nfserr_clid_inuse;
 865                 else
 866                         status = nfs_ok;
 867         } else if (!conf && unconf
 868                         && same_verf(&unconf->cl_confirm, &confirm)) {
 869                 /*
 870                  * RFC 3530 14.2.34 CASE 3:
 871                  * Normal case; new or rebooted client:
 872                  */
 873                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
 874                         status = nfserr_clid_inuse;
 875                 } else {
 876                         unsigned int hash =
 877                                 clientstr_hashval(unconf->cl_recdir);
 878                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
 879                                                                         hash);
 880                         if (conf) {
 881                                 nfsd4_remove_clid_dir(conf);
 882                                 expire_client(conf);
 883                         }
 884                         move_to_confirmed(unconf);
 885                         conf = unconf;
 886                         nfsd4_probe_callback(conf);
 887                         status = nfs_ok;
 888                 }
 889         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
 890             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
 891                                                                 &confirm)))) {
 892                 /*
 893                  * RFC 3530 14.2.34 CASE 4:
 894                  * Client probably hasn't noticed that we rebooted yet.
 895                  */
 896                 status = nfserr_stale_clientid;
 897         } else {
 898                 /* check that we have hit one of the cases...*/
 899                 status = nfserr_clid_inuse;
 900         }
 901 out:
 902         nfs4_unlock_state();
 903         return status;
 904 }
 905
 906 /* OPEN Share state helper functions */
 907 static inline struct nfs4_file *
 908 alloc_init_file(struct inode *ino)
 909 {
 910         struct nfs4_file *fp;
 911         unsigned int hashval = file_hashval(ino);
 912
 913         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
 914         if (fp) {
 915                 kref_init(&fp->fi_ref);
 916                 INIT_LIST_HEAD(&fp->fi_hash);
 917                 INIT_LIST_HEAD(&fp->fi_stateids);
 918                 INIT_LIST_HEAD(&fp->fi_delegations);
 919                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
 920                 fp->fi_inode = igrab(ino);
 921                 fp->fi_id = current_fileid++;
 922                 fp->fi_had_conflict = false;
 923                 return fp;
 924         }
 925         return NULL;
 926 }
 927
 928 static void
 929 nfsd4_free_slab(struct kmem_cache **slab)
 930 {
 931         if (*slab == NULL)
 932                 return;
 933         kmem_cache_destroy(*slab);
 934         *slab = NULL;
 935 }
 936
 937 void
 938 nfsd4_free_slabs(void)
 939 {
 940         nfsd4_free_slab(&stateowner_slab);
 941         nfsd4_free_slab(&file_slab);
 942         nfsd4_free_slab(&stateid_slab);
 943         nfsd4_free_slab(&deleg_slab);
 944 }
 945
 946 static int
 947 nfsd4_init_slabs(void)
 948 {
 949         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
 950                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
 951         if (stateowner_slab == NULL)
 952                 goto out_nomem;
 953         file_slab = kmem_cache_create("nfsd4_files",
 954                         sizeof(struct nfs4_file), 0, 0, NULL);
 955         if (file_slab == NULL)
 956                 goto out_nomem;
 957         stateid_slab = kmem_cache_create("nfsd4_stateids",
 958                         sizeof(struct nfs4_stateid), 0, 0, NULL);
 959         if (stateid_slab == NULL)
 960                 goto out_nomem;
 961         deleg_slab = kmem_cache_create("nfsd4_delegations",
 962                         sizeof(struct nfs4_delegation), 0, 0, NULL);
 963         if (deleg_slab == NULL)
 964                 goto out_nomem;
 965         return 0;
 966 out_nomem:
 967         nfsd4_free_slabs();
 968         dprintk("nfsd4: out of memory while initializing nfsv4\n");
 969         return -ENOMEM;
 970 }
 971
 972 void
 973 nfs4_free_stateowner(struct kref *kref)
 974 {
 975         struct nfs4_stateowner *sop =
 976                 container_of(kref, struct nfs4_stateowner, so_ref);
 977         kfree(sop->so_owner.data);
 978         kmem_cache_free(stateowner_slab, sop);
 979 }
 980
 981 static inline struct nfs4_stateowner *
 982 alloc_stateowner(struct xdr_netobj *owner)
 983 {
 984         struct nfs4_stateowner *sop;
 985
 986         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
 987                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
 988                         memcpy(sop->so_owner.data, owner->data, owner->len);
 989                         sop->so_owner.len = owner->len;
 990                         kref_init(&sop->so_ref);
 991                         return sop;
 992                 }
 993                 kmem_cache_free(stateowner_slab, sop);
 994         }
 995         return NULL;
 996 }
 997
 998 static struct nfs4_stateowner *
 999 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1000         struct nfs4_stateowner *sop;
1001         struct nfs4_replay *rp;
1002         unsigned int idhashval;
1003
1004         if (!(sop = alloc_stateowner(&open->op_owner)))
1005                 return NULL;
1006         idhashval = ownerid_hashval(current_ownerid);
1007         INIT_LIST_HEAD(&sop->so_idhash);
1008         INIT_LIST_HEAD(&sop->so_strhash);
1009         INIT_LIST_HEAD(&sop->so_perclient);
1010         INIT_LIST_HEAD(&sop->so_stateids);
1011         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1012         INIT_LIST_HEAD(&sop->so_close_lru);
1013         sop->so_time = 0;
1014         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1015         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1016         list_add(&sop->so_perclient, &clp->cl_openowners);
1017         sop->so_is_open_owner = 1;
1018         sop->so_id = current_ownerid++;
1019         sop->so_client = clp;
1020         sop->so_seqid = open->op_seqid;
1021         sop->so_confirmed = 0;
1022         rp = &sop->so_replay;
1023         rp->rp_status = nfserr_serverfault;
1024         rp->rp_buflen = 0;
1025         rp->rp_buf = rp->rp_ibuf;
1026         return sop;
1027 }
1028
1029 static void
1030 release_stateid_lockowners(struct nfs4_stateid *open_stp)
1031 {
1032         struct nfs4_stateowner *lock_sop;
1033
1034         while (!list_empty(&open_stp->st_lockowners)) {
1035                 lock_sop = list_entry(open_stp->st_lockowners.next,
1036                                 struct nfs4_stateowner, so_perstateid);
1037                 /* list_del(&open_stp->st_lockowners);  */
1038                 BUG_ON(lock_sop->so_is_open_owner);
1039                 release_stateowner(lock_sop);
1040         }
1041 }
1042
1043 static void
1044 unhash_stateowner(struct nfs4_stateowner *sop)
1045 {
1046         struct nfs4_stateid *stp;
1047
1048         list_del(&sop->so_idhash);
1049         list_del(&sop->so_strhash);
1050         if (sop->so_is_open_owner)
1051                 list_del(&sop->so_perclient);
1052         list_del(&sop->so_perstateid);
1053         while (!list_empty(&sop->so_stateids)) {
1054                 stp = list_entry(sop->so_stateids.next,
1055                         struct nfs4_stateid, st_perstateowner);
1056                 if (sop->so_is_open_owner)
1057                         release_stateid(stp, OPEN_STATE);
1058                 else
1059                         release_stateid(stp, LOCK_STATE);
1060         }
1061 }
1062
1063 static void
1064 release_stateowner(struct nfs4_stateowner *sop)
1065 {
1066         unhash_stateowner(sop);
1067         list_del(&sop->so_close_lru);
1068         nfs4_put_stateowner(sop);
1069 }
1070
1071 static inline void
1072 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1073         struct nfs4_stateowner *sop = open->op_stateowner;
1074         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1075
1076         INIT_LIST_HEAD(&stp->st_hash);
1077         INIT_LIST_HEAD(&stp->st_perstateowner);
1078         INIT_LIST_HEAD(&stp->st_lockowners);
1079         INIT_LIST_HEAD(&stp->st_perfile);
1080         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1081         list_add(&stp->st_perstateowner, &sop->so_stateids);
1082         list_add(&stp->st_perfile, &fp->fi_stateids);
1083         stp->st_stateowner = sop;
1084         get_nfs4_file(fp);
1085         stp->st_file = fp;
1086         stp->st_stateid.si_boot = boot_time;
1087         stp->st_stateid.si_stateownerid = sop->so_id;
1088         stp->st_stateid.si_fileid = fp->fi_id;
1089         stp->st_stateid.si_generation = 0;
1090         stp->st_access_bmap = 0;
1091         stp->st_deny_bmap = 0;
1092         __set_bit(open->op_share_access, &stp->st_access_bmap);
1093         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1094         stp->st_openstp = NULL;
1095 }
1096
1097 static void
1098 release_stateid(struct nfs4_stateid *stp, int flags)
1099 {
1100         struct file *filp = stp->st_vfs_file;
1101
1102         list_del(&stp->st_hash);
1103         list_del(&stp->st_perfile);
1104         list_del(&stp->st_perstateowner);
1105         if (flags & OPEN_STATE) {
1106                 release_stateid_lockowners(stp);
1107                 stp->st_vfs_file = NULL;
1108                 nfsd_close(filp);
1109         } else if (flags & LOCK_STATE)
1110                 locks_remove_posix(filp, (fl_owner_t) stp->st_stateowner);
1111         put_nfs4_file(stp->st_file);
1112         kmem_cache_free(stateid_slab, stp);
1113 }
1114
1115 static void
1116 move_to_close_lru(struct nfs4_stateowner *sop)
1117 {
1118         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1119
1120         list_move_tail(&sop->so_close_lru, &close_lru);
1121         sop->so_time = get_seconds();
1122 }
1123
1124 static int
1125 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1126                                                         clientid_t *clid)
1127 {
1128         return (sop->so_owner.len == owner->len) &&
1129                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1130                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1131 }
1132
1133 static struct nfs4_stateowner *
1134 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1135 {
1136         struct nfs4_stateowner *so = NULL;
1137
1138         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1139                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1140                         return so;
1141         }
1142         return NULL;
1143 }
1144
1145 /* search file_hashtbl[] for file */
1146 static struct nfs4_file *
1147 find_file(struct inode *ino)
1148 {
1149         unsigned int hashval = file_hashval(ino);
1150         struct nfs4_file *fp;
1151
1152         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1153                 if (fp->fi_inode == ino) {
1154                         get_nfs4_file(fp);
1155                         return fp;
1156                 }
1157         }
1158         return NULL;
1159 }
1160
1161 static inline int access_valid(u32 x)
1162 {
1163         if (x < NFS4_SHARE_ACCESS_READ)
1164                 return 0;
1165         if (x > NFS4_SHARE_ACCESS_BOTH)
1166                 return 0;
1167         return 1;
1168 }
1169
1170 static inline int deny_valid(u32 x)
1171 {
1172         /* Note: unlike access bits, deny bits may be zero. */
1173         return x <= NFS4_SHARE_DENY_BOTH;
1174 }
1175
1176 static void
1177 set_access(unsigned int *access, unsigned long bmap) {
1178         int i;
1179
1180         *access = 0;
1181         for (i = 1; i < 4; i++) {
1182                 if (test_bit(i, &bmap))
1183                         *access |= i;
1184         }
1185 }
1186
1187 static void
1188 set_deny(unsigned int *deny, unsigned long bmap) {
1189         int i;
1190
1191         *deny = 0;
1192         for (i = 0; i < 4; i++) {
1193                 if (test_bit(i, &bmap))
1194                         *deny |= i ;
1195         }
1196 }
1197
1198 static int
1199 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1200         unsigned int access, deny;
1201
1202         set_access(&access, stp->st_access_bmap);
1203         set_deny(&deny, stp->st_deny_bmap);
1204         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1205                 return 0;
1206         return 1;
1207 }
1208
1209 /*
1210  * Called to check deny when READ with all zero stateid or
1211  * WRITE with all zero or all one stateid
1212  */
1213 static __be32
1214 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1215 {
1216         struct inode *ino = current_fh->fh_dentry->d_inode;
1217         struct nfs4_file *fp;
1218         struct nfs4_stateid *stp;
1219         __be32 ret;
1220
1221         dprintk("NFSD: nfs4_share_conflict\n");
1222
1223         fp = find_file(ino);
1224         if (!fp)
1225                 return nfs_ok;
1226         ret = nfserr_locked;
1227         /* Search for conflicting share reservations */
1228         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1229                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1230                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1231                         goto out;
1232         }
1233         ret = nfs_ok;
1234 out:
1235         put_nfs4_file(fp);
1236         return ret;
1237 }
1238
1239 static inline void
1240 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1241 {
1242         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1243                 drop_file_write_access(filp);
1244                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1245         }
1246 }
1247
1248 /*
1249  * Recall a delegation
1250  */
1251 static int
1252 do_recall(void *__dp)
1253 {
1254         struct nfs4_delegation *dp = __dp;
1255
1256         dp->dl_file->fi_had_conflict = true;
1257         nfsd4_cb_recall(dp);
1258         return 0;
1259 }
1260
1261 /*
1262  * Spawn a thread to perform a recall on the delegation represented
1263  * by the lease (file_lock)
1264  *
1265  * Called from break_lease() with lock_kernel() held.
1266  * Note: we assume break_lease will only call this *once* for any given
1267  * lease.
1268  */
1269 static
1270 void nfsd_break_deleg_cb(struct file_lock *fl)
1271 {
1272         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1273         struct task_struct *t;
1274
1275         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1276         if (!dp)
1277                 return;
1278
1279         /* We're assuming the state code never drops its reference
1280          * without first removing the lease.  Since we're in this lease
1281          * callback (and since the lease code is serialized by the kernel
1282          * lock) we know the server hasn't removed the lease yet, we know
1283          * it's safe to take a reference: */
1284         atomic_inc(&dp->dl_count);
1285         atomic_inc(&dp->dl_client->cl_count);
1286
1287         spin_lock(&recall_lock);
1288         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1289         spin_unlock(&recall_lock);
1290
1291         /* only place dl_time is set. protected by lock_kernel*/
1292         dp->dl_time = get_seconds();
1293
1294         /*
1295          * We don't want the locks code to timeout the lease for us;
1296          * we'll remove it ourself if the delegation isn't returned
1297          * in time.
1298          */
1299         fl->fl_break_time = 0;
1300
1301         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1302         if (IS_ERR(t)) {
1303                 struct nfs4_client *clp = dp->dl_client;
1304
1305                 printk(KERN_INFO "NFSD: Callback thread failed for "
1306                         "for client (clientid %08x/%08x)\n",
1307                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1308                 put_nfs4_client(dp->dl_client);
1309                 nfs4_put_delegation(dp);
1310         }
1311 }
1312
1313 /*
1314  * The file_lock is being reapd.
1315  *
1316  * Called by locks_free_lock() with lock_kernel() held.
1317  */
1318 static
1319 void nfsd_release_deleg_cb(struct file_lock *fl)
1320 {
1321         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1322
1323         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1324
1325         if (!(fl->fl_flags & FL_LEASE) || !dp)
1326                 return;
1327         dp->dl_flock = NULL;
1328 }
1329
1330 /*
1331  * Set the delegation file_lock back pointer.
1332  *
1333  * Called from setlease() with lock_kernel() held.
1334  */
1335 static
1336 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1337 {
1338         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1339
1340         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1341         if (!dp)
1342                 return;
1343         dp->dl_flock = new;
1344 }
1345
1346 /*
1347  * Called from setlease() with lock_kernel() held
1348  */
1349 static
1350 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1351 {
1352         struct nfs4_delegation *onlistd =
1353                 (struct nfs4_delegation *)onlist->fl_owner;
1354         struct nfs4_delegation *tryd =
1355                 (struct nfs4_delegation *)try->fl_owner;
1356
1357         if (onlist->fl_lmops != try->fl_lmops)
1358                 return 0;
1359
1360         return onlistd->dl_client == tryd->dl_client;
1361 }
1362
1363
1364 static
1365 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1366 {
1367         if (arg & F_UNLCK)
1368                 return lease_modify(onlist, arg);
1369         else
1370                 return -EAGAIN;
1371 }
1372
1373 static struct lock_manager_operations nfsd_lease_mng_ops = {
1374         .fl_break = nfsd_break_deleg_cb,
1375         .fl_release_private = nfsd_release_deleg_cb,
1376         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1377         .fl_mylease = nfsd_same_client_deleg_cb,
1378         .fl_change = nfsd_change_deleg_cb,
1379 };
1380
1381
1382 __be32
1383 nfsd4_process_open1(struct nfsd4_open *open)
1384 {
1385         clientid_t *clientid = &open->op_clientid;
1386         struct nfs4_client *clp = NULL;
1387         unsigned int strhashval;
1388         struct nfs4_stateowner *sop = NULL;
1389
1390         if (!check_name(open->op_owner))
1391                 return nfserr_inval;
1392
1393         if (STALE_CLIENTID(&open->op_clientid))
1394                 return nfserr_stale_clientid;
1395
1396         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1397         sop = find_openstateowner_str(strhashval, open);
1398         open->op_stateowner = sop;
1399         if (!sop) {
1400                 /* Make sure the client's lease hasn't expired. */
1401                 clp = find_confirmed_client(clientid);
1402                 if (clp == NULL)
1403                         return nfserr_expired;
1404                 goto renew;
1405         }
1406         if (!sop->so_confirmed) {
1407                 /* Replace unconfirmed owners without checking for replay. */
1408                 clp = sop->so_client;
1409                 release_stateowner(sop);
1410                 open->op_stateowner = NULL;
1411                 goto renew;
1412         }
1413         if (open->op_seqid == sop->so_seqid - 1) {
1414                 if (sop->so_replay.rp_buflen)
1415                         return nfserr_replay_me;
1416                 /* The original OPEN failed so spectacularly
1417                  * that we don't even have replay data saved!
1418                  * Therefore, we have no choice but to continue
1419                  * processing this OPEN; presumably, we'll
1420                  * fail again for the same reason.
1421                  */
1422                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1423                 goto renew;
1424         }
1425         if (open->op_seqid != sop->so_seqid)
1426                 return nfserr_bad_seqid;
1427 renew:
1428         if (open->op_stateowner == NULL) {
1429                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1430                 if (sop == NULL)
1431                         return nfserr_resource;
1432                 open->op_stateowner = sop;
1433         }
1434         list_del_init(&sop->so_close_lru);
1435         renew_client(sop->so_client);
1436         return nfs_ok;
1437 }
1438
1439 static inline __be32
1440 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1441 {
1442         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1443                 return nfserr_openmode;
1444         else
1445                 return nfs_ok;
1446 }
1447
1448 static struct nfs4_delegation *
1449 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1450 {
1451         struct nfs4_delegation *dp;
1452
1453         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1454                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1455                         return dp;
1456         }
1457         return NULL;
1458 }
1459
1460 static __be32
1461 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1462                 struct nfs4_delegation **dp)
1463 {
1464         int flags;
1465         __be32 status = nfserr_bad_stateid;
1466
1467         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1468         if (*dp == NULL)
1469                 goto out;
1470         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1471                                                 RD_STATE : WR_STATE;
1472         status = nfs4_check_delegmode(*dp, flags);
1473         if (status)
1474                 *dp = NULL;
1475 out:
1476         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1477                 return nfs_ok;
1478         if (status)
1479                 return status;
1480         open->op_stateowner->so_confirmed = 1;
1481         return nfs_ok;
1482 }
1483
1484 static __be32
1485 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1486 {
1487         struct nfs4_stateid *local;
1488         __be32 status = nfserr_share_denied;
1489         struct nfs4_stateowner *sop = open->op_stateowner;
1490
1491         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1492                 /* ignore lock owners */
1493                 if (local->st_stateowner->so_is_open_owner == 0)
1494                         continue;
1495                 /* remember if we have seen this open owner */
1496                 if (local->st_stateowner == sop)
1497                         *stpp = local;
1498                 /* check for conflicting share reservations */
1499                 if (!test_share(local, open))
1500                         goto out;
1501         }
1502         status = 0;
1503 out:
1504         return status;
1505 }
1506
1507 static inline struct nfs4_stateid *
1508 nfs4_alloc_stateid(void)
1509 {
1510         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1511 }
1512
1513 static __be32
1514 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1515                 struct nfs4_delegation *dp,
1516                 struct svc_fh *cur_fh, int flags)
1517 {
1518         struct nfs4_stateid *stp;
1519
1520         stp = nfs4_alloc_stateid();
1521         if (stp == NULL)
1522                 return nfserr_resource;
1523
1524         if (dp) {
1525                 get_file(dp->dl_vfs_file);
1526                 stp->st_vfs_file = dp->dl_vfs_file;
1527         } else {
1528                 __be32 status;
1529                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1530                                 &stp->st_vfs_file);
1531                 if (status) {
1532                         if (status == nfserr_dropit)
1533                                 status = nfserr_jukebox;
1534                         kmem_cache_free(stateid_slab, stp);
1535                         return status;
1536                 }
1537         }
1538         *stpp = stp;
1539         return 0;
1540 }
1541
1542 static inline __be32
1543 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1544                 struct nfsd4_open *open)
1545 {
1546         struct iattr iattr = {
1547                 .ia_valid = ATTR_SIZE,
1548                 .ia_size = 0,
1549         };
1550         if (!open->op_truncate)
1551                 return 0;
1552         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1553                 return nfserr_inval;
1554         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1555 }
1556
1557 static __be32
1558 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1559 {
1560         struct file *filp = stp->st_vfs_file;
1561         struct inode *inode = filp->f_path.dentry->d_inode;
1562         unsigned int share_access, new_writer;
1563         __be32 status;
1564
1565         set_access(&share_access, stp->st_access_bmap);
1566         new_writer = (~share_access) & open->op_share_access
1567                         & NFS4_SHARE_ACCESS_WRITE;
1568
1569         if (new_writer) {
1570                 int err = get_write_access(inode);
1571                 if (err)
1572                         return nfserrno(err);
1573         }
1574         status = nfsd4_truncate(rqstp, cur_fh, open);
1575         if (status) {
1576                 if (new_writer)
1577                         put_write_access(inode);
1578                 return status;
1579         }
1580         /* remember the open */
1581         filp->f_mode |= open->op_share_access;
1582         set_bit(open->op_share_access, &stp->st_access_bmap);
1583         set_bit(open->op_share_deny, &stp->st_deny_bmap);
1584
1585         return nfs_ok;
1586 }
1587
1588
1589 static void
1590 nfs4_set_claim_prev(struct nfsd4_open *open)
1591 {
1592         open->op_stateowner->so_confirmed = 1;
1593         open->op_stateowner->so_client->cl_firststate = 1;
1594 }
1595
1596 /*
1597  * Attempt to hand out a delegation.
1598  */
1599 static void
1600 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1601 {
1602         struct nfs4_delegation *dp;
1603         struct nfs4_stateowner *sop = stp->st_stateowner;
1604         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1605         struct file_lock fl, *flp = &fl;
1606         int status, flag = 0;
1607
1608         flag = NFS4_OPEN_DELEGATE_NONE;
1609         open->op_recall = 0;
1610         switch (open->op_claim_type) {
1611                 case NFS4_OPEN_CLAIM_PREVIOUS:
1612                         if (!atomic_read(&cb->cb_set))
1613                                 open->op_recall = 1;
1614                         flag = open->op_delegate_type;
1615                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1616                                 goto out;
1617                         break;
1618                 case NFS4_OPEN_CLAIM_NULL:
1619                         /* Let's not give out any delegations till everyone's
1620                          * had the chance to reclaim theirs.... */
1621                         if (nfs4_in_grace())
1622                                 goto out;
1623                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1624                                 goto out;
1625                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1626                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1627                         else
1628                                 flag = NFS4_OPEN_DELEGATE_READ;
1629                         break;
1630                 default:
1631                         goto out;
1632         }
1633
1634         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1635         if (dp == NULL) {
1636                 flag = NFS4_OPEN_DELEGATE_NONE;
1637                 goto out;
1638         }
1639         locks_init_lock(&fl);
1640         fl.fl_lmops = &nfsd_lease_mng_ops;
1641         fl.fl_flags = FL_LEASE;
1642         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
1643         fl.fl_end = OFFSET_MAX;
1644         fl.fl_owner =  (fl_owner_t)dp;
1645         fl.fl_file = stp->st_vfs_file;
1646         fl.fl_pid = current->tgid;
1647
1648         /* vfs_setlease checks to see if delegation should be handed out.
1649          * the lock_manager callbacks fl_mylease and fl_change are used
1650          */
1651         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
1652                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1653                 unhash_delegation(dp);
1654                 flag = NFS4_OPEN_DELEGATE_NONE;
1655                 goto out;
1656         }
1657
1658         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1659
1660         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1661                      dp->dl_stateid.si_boot,
1662                      dp->dl_stateid.si_stateownerid,
1663                      dp->dl_stateid.si_fileid,
1664                      dp->dl_stateid.si_generation);
1665 out:
1666         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1667                         && flag == NFS4_OPEN_DELEGATE_NONE
1668                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1669                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1670         open->op_delegate_type = flag;
1671 }
1672
1673 /*
1674  * called with nfs4_lock_state() held.
1675  */
1676 __be32
1677 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1678 {
1679         struct nfs4_file *fp = NULL;
1680         struct inode *ino = current_fh->fh_dentry->d_inode;
1681         struct nfs4_stateid *stp = NULL;
1682         struct nfs4_delegation *dp = NULL;
1683         __be32 status;
1684
1685         status = nfserr_inval;
1686         if (!access_valid(open->op_share_access)
1687                         || !deny_valid(open->op_share_deny))
1688                 goto out;
1689         /*
1690          * Lookup file; if found, lookup stateid and check open request,
1691          * and check for delegations in the process of being recalled.
1692          * If not found, create the nfs4_file struct
1693          */
1694         fp = find_file(ino);
1695         if (fp) {
1696                 if ((status = nfs4_check_open(fp, open, &stp)))
1697                         goto out;
1698                 status = nfs4_check_deleg(fp, open, &dp);
1699                 if (status)
1700                         goto out;
1701         } else {
1702                 status = nfserr_bad_stateid;
1703                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1704                         goto out;
1705                 status = nfserr_resource;
1706                 fp = alloc_init_file(ino);
1707                 if (fp == NULL)
1708                         goto out;
1709         }
1710
1711         /*
1712          * OPEN the file, or upgrade an existing OPEN.
1713          * If truncate fails, the OPEN fails.
1714          */
1715         if (stp) {
1716                 /* Stateid was found, this is an OPEN upgrade */
1717                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1718                 if (status)
1719                         goto out;
1720                 update_stateid(&stp->st_stateid);
1721         } else {
1722                 /* Stateid was not found, this is a new OPEN */
1723                 int flags = 0;
1724                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1725                         flags |= MAY_READ;
1726                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1727                         flags |= MAY_WRITE;
1728                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1729                 if (status)
1730                         goto out;
1731                 init_stateid(stp, fp, open);
1732                 status = nfsd4_truncate(rqstp, current_fh, open);
1733                 if (status) {
1734                         release_stateid(stp, OPEN_STATE);
1735                         goto out;
1736                 }
1737         }
1738         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1739
1740         /*
1741         * Attempt to hand out a delegation. No error return, because the
1742         * OPEN succeeds even if we fail.
1743         */
1744         nfs4_open_delegation(current_fh, open, stp);
1745
1746         status = nfs_ok;
1747
1748         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1749                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1750                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1751 out:
1752         if (fp)
1753                 put_nfs4_file(fp);
1754         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1755                 nfs4_set_claim_prev(open);
1756         /*
1757         * To finish the open response, we just need to set the rflags.
1758         */
1759         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1760         if (!open->op_stateowner->so_confirmed)
1761                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1762
1763         return status;
1764 }
1765
1766 __be32
1767 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1768             clientid_t *clid)
1769 {
1770         struct nfs4_client *clp;
1771         __be32 status;
1772
1773         nfs4_lock_state();
1774         dprintk("process_renew(%08x/%08x): starting\n",
1775                         clid->cl_boot, clid->cl_id);
1776         status = nfserr_stale_clientid;
1777         if (STALE_CLIENTID(clid))
1778                 goto out;
1779         clp = find_confirmed_client(clid);
1780         status = nfserr_expired;
1781         if (clp == NULL) {
1782                 /* We assume the client took too long to RENEW. */
1783                 dprintk("nfsd4_renew: clientid not found!\n");
1784                 goto out;
1785         }
1786         renew_client(clp);
1787         status = nfserr_cb_path_down;
1788         if (!list_empty(&clp->cl_delegations)
1789                         && !atomic_read(&clp->cl_callback.cb_set))
1790                 goto out;
1791         status = nfs_ok;
1792 out:
1793         nfs4_unlock_state();
1794         return status;
1795 }
1796
1797 static void
1798 end_grace(void)
1799 {
1800         dprintk("NFSD: end of grace period\n");
1801         nfsd4_recdir_purge_old();
1802         in_grace = 0;
1803 }
1804
1805 static time_t
1806 nfs4_laundromat(void)
1807 {
1808         struct nfs4_client *clp;
1809         struct nfs4_stateowner *sop;
1810         struct nfs4_delegation *dp;
1811         struct list_head *pos, *next, reaplist;
1812         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1813         time_t t, clientid_val = NFSD_LEASE_TIME;
1814         time_t u, test_val = NFSD_LEASE_TIME;
1815
1816         nfs4_lock_state();
1817
1818         dprintk("NFSD: laundromat service - starting\n");
1819         if (in_grace)
1820                 end_grace();
1821         list_for_each_safe(pos, next, &client_lru) {
1822                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1823                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1824                         t = clp->cl_time - cutoff;
1825                         if (clientid_val > t)
1826                                 clientid_val = t;
1827                         break;
1828                 }
1829                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1830                         clp->cl_clientid.cl_id);
1831                 nfsd4_remove_clid_dir(clp);
1832                 expire_client(clp);
1833         }
1834         INIT_LIST_HEAD(&reaplist);
1835         spin_lock(&recall_lock);
1836         list_for_each_safe(pos, next, &del_recall_lru) {
1837                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1838                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1839                         u = dp->dl_time - cutoff;
1840                         if (test_val > u)
1841                                 test_val = u;
1842                         break;
1843                 }
1844                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1845                                     dp, dp->dl_flock);
1846                 list_move(&dp->dl_recall_lru, &reaplist);
1847         }
1848         spin_unlock(&recall_lock);
1849         list_for_each_safe(pos, next, &reaplist) {
1850                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1851                 list_del_init(&dp->dl_recall_lru);
1852                 unhash_delegation(dp);
1853         }
1854         test_val = NFSD_LEASE_TIME;
1855         list_for_each_safe(pos, next, &close_lru) {
1856                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1857                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1858                         u = sop->so_time - cutoff;
1859                         if (test_val > u)
1860                                 test_val = u;
1861                         break;
1862                 }
1863                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1864                         sop->so_id);
1865                 release_stateowner(sop);
1866         }
1867         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1868                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1869         nfs4_unlock_state();
1870         return clientid_val;
1871 }
1872
1873 static struct workqueue_struct *laundry_wq;
1874 static void laundromat_main(struct work_struct *);
1875 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1876
1877 static void
1878 laundromat_main(struct work_struct *not_used)
1879 {
1880         time_t t;
1881
1882         t = nfs4_laundromat();
1883         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1884         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1885 }
1886
1887 static struct nfs4_stateowner *
1888 search_close_lru(u32 st_id, int flags)
1889 {
1890         struct nfs4_stateowner *local = NULL;
1891
1892         if (flags & CLOSE_STATE) {
1893                 list_for_each_entry(local, &close_lru, so_close_lru) {
1894                         if (local->so_id == st_id)
1895                                 return local;
1896                 }
1897         }
1898         return NULL;
1899 }
1900
1901 static inline int
1902 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1903 {
1904         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1905 }
1906
1907 static int
1908 STALE_STATEID(stateid_t *stateid)
1909 {
1910         if (stateid->si_boot == boot_time)
1911                 return 0;
1912         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1913                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1914                 stateid->si_generation);
1915         return 1;
1916 }
1917
1918 static inline int
1919 access_permit_read(unsigned long access_bmap)
1920 {
1921         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1922                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1923                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1924 }
1925
1926 static inline int
1927 access_permit_write(unsigned long access_bmap)
1928 {
1929         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1930                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1931 }
1932
1933 static
1934 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1935 {
1936         __be32 status = nfserr_openmode;
1937
1938         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1939                 goto out;
1940         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1941                 goto out;
1942         status = nfs_ok;
1943 out:
1944         return status;
1945 }
1946
1947 static inline __be32
1948 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
1949 {
1950         /* Trying to call delegreturn with a special stateid? Yuch: */
1951         if (!(flags & (RD_STATE | WR_STATE)))
1952                 return nfserr_bad_stateid;
1953         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
1954                 return nfs_ok;
1955         else if (nfs4_in_grace()) {
1956                 /* Answer in remaining cases depends on existance of
1957                  * conflicting state; so we must wait out the grace period. */
1958                 return nfserr_grace;
1959         } else if (flags & WR_STATE)
1960                 return nfs4_share_conflict(current_fh,
1961                                 NFS4_SHARE_DENY_WRITE);
1962         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
1963                 return nfs4_share_conflict(current_fh,
1964                                 NFS4_SHARE_DENY_READ);
1965 }
1966
1967 /*
1968  * Allow READ/WRITE during grace period on recovered state only for files
1969  * that are not able to provide mandatory locking.
1970  */
1971 static inline int
1972 io_during_grace_disallowed(struct inode *inode, int flags)
1973 {
1974         return nfs4_in_grace() && (flags & (RD_STATE | WR_STATE))
1975                 && mandatory_lock(inode);
1976 }
1977
1978 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
1979 {
1980         /* If the client sends us a stateid from the future, it's buggy: */
1981         if (in->si_generation > ref->si_generation)
1982                 return nfserr_bad_stateid;
1983         /*
1984          * The following, however, can happen.  For example, if the
1985          * client sends an open and some IO at the same time, the open
1986          * may bump si_generation while the IO is still in flight.
1987          * Thanks to hard links and renames, the client never knows what
1988          * file an open will affect.  So it could avoid that situation
1989          * only by serializing all opens and IO from the same open
1990          * owner.  To recover from the old_stateid error, the client
1991          * will just have to retry the IO:
1992          */
1993         if (in->si_generation < ref->si_generation)
1994                 return nfserr_old_stateid;
1995         return nfs_ok;
1996 }
1997
1998 /*
1999 * Checks for stateid operations
2000 */
2001 __be32
2002 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2003 {
2004         struct nfs4_stateid *stp = NULL;
2005         struct nfs4_delegation *dp = NULL;
2006         stateid_t *stidp;
2007         struct inode *ino = current_fh->fh_dentry->d_inode;
2008         __be32 status;
2009
2010         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2011                 stateid->si_boot, stateid->si_stateownerid,
2012                 stateid->si_fileid, stateid->si_generation);
2013         if (filpp)
2014                 *filpp = NULL;
2015
2016         if (io_during_grace_disallowed(ino, flags))
2017                 return nfserr_grace;
2018
2019         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2020                 return check_special_stateids(current_fh, stateid, flags);
2021
2022         /* STALE STATEID */
2023         status = nfserr_stale_stateid;
2024         if (STALE_STATEID(stateid))
2025                 goto out;
2026
2027         /* BAD STATEID */
2028         status = nfserr_bad_stateid;
2029         if (!stateid->si_fileid) { /* delegation stateid */
2030                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2031                         dprintk("NFSD: delegation stateid not found\n");
2032                         goto out;
2033                 }
2034                 stidp = &dp->dl_stateid;
2035         } else { /* open or lock stateid */
2036                 if (!(stp = find_stateid(stateid, flags))) {
2037                         dprintk("NFSD: open or lock stateid not found\n");
2038                         goto out;
2039                 }
2040                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2041                         goto out;
2042                 if (!stp->st_stateowner->so_confirmed)
2043                         goto out;
2044                 stidp = &stp->st_stateid;
2045         }
2046         status = check_stateid_generation(stateid, stidp);
2047         if (status)
2048                 goto out;
2049         if (stp) {
2050                 if ((status = nfs4_check_openmode(stp,flags)))
2051                         goto out;
2052                 renew_client(stp->st_stateowner->so_client);
2053                 if (filpp)
2054                         *filpp = stp->st_vfs_file;
2055         } else {
2056                 if ((status = nfs4_check_delegmode(dp, flags)))
2057                         goto out;
2058                 renew_client(dp->dl_client);
2059                 if (flags & DELEG_RET)
2060                         unhash_delegation(dp);
2061                 if (filpp)
2062                         *filpp = dp->dl_vfs_file;
2063         }
2064         status = nfs_ok;
2065 out:
2066         return status;
2067 }
2068
2069 static inline int
2070 setlkflg (int type)
2071 {
2072         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2073                 RD_STATE : WR_STATE;
2074 }
2075
2076 /*
2077  * Checks for sequence id mutating operations.
2078  */
2079 static __be32
2080 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)
2081 {
2082         struct nfs4_stateid *stp;
2083         struct nfs4_stateowner *sop;
2084         __be32 status;
2085
2086         dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2087                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2088                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2089                 stateid->si_generation);
2090
2091         *stpp = NULL;
2092         *sopp = NULL;
2093
2094         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2095                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2096                 return nfserr_bad_stateid;
2097         }
2098
2099         if (STALE_STATEID(stateid))
2100                 return nfserr_stale_stateid;
2101         /*
2102         * We return BAD_STATEID if filehandle doesn't match stateid,
2103         * the confirmed flag is incorrecly set, or the generation
2104         * number is incorrect.
2105         */
2106         stp = find_stateid(stateid, flags);
2107         if (stp == NULL) {
2108                 /*
2109                  * Also, we should make sure this isn't just the result of
2110                  * a replayed close:
2111                  */
2112                 sop = search_close_lru(stateid->si_stateownerid, flags);
2113                 if (sop == NULL)
2114                         return nfserr_bad_stateid;
2115                 *sopp = sop;
2116                 goto check_replay;
2117         }
2118
2119         *stpp = stp;
2120         *sopp = sop = stp->st_stateowner;
2121
2122         if (lock) {
2123                 clientid_t *lockclid = &lock->v.new.clientid;
2124                 struct nfs4_client *clp = sop->so_client;
2125                 int lkflg = 0;
2126                 __be32 status;
2127
2128                 lkflg = setlkflg(lock->lk_type);
2129
2130                 if (lock->lk_is_new) {
2131                         if (!sop->so_is_open_owner)
2132                                 return nfserr_bad_stateid;
2133                         if (!same_clid(&clp->cl_clientid, lockclid))
2134                                return nfserr_bad_stateid;
2135                         /* stp is the open stateid */
2136                         status = nfs4_check_openmode(stp, lkflg);
2137                         if (status)
2138                                 return status;
2139                 } else {
2140                         /* stp is the lock stateid */
2141                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2142                         if (status)
2143                                 return status;
2144                }
2145         }
2146
2147         if (nfs4_check_fh(current_fh, stp)) {
2148                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2149                 return nfserr_bad_stateid;
2150         }
2151
2152         /*
2153         *  We now validate the seqid and stateid generation numbers.
2154         *  For the moment, we ignore the possibility of
2155         *  generation number wraparound.
2156         */
2157         if (seqid != sop->so_seqid)
2158                 goto check_replay;
2159
2160         if (sop->so_confirmed && flags & CONFIRM) {
2161                 dprintk("NFSD: preprocess_seqid_op: expected"
2162                                 " unconfirmed stateowner!\n");
2163                 return nfserr_bad_stateid;
2164         }
2165         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2166                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2167                                 " confirmed yet!\n");
2168                 return nfserr_bad_stateid;
2169         }
2170         status = check_stateid_generation(stateid, &stp->st_stateid);
2171         if (status)
2172                 return status;
2173         renew_client(sop->so_client);
2174         return nfs_ok;
2175
2176 check_replay:
2177         if (seqid == sop->so_seqid - 1) {
2178                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2179                 /* indicate replay to calling function */
2180                 return nfserr_replay_me;
2181         }
2182         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2183                         sop->so_seqid, seqid);
2184         *sopp = NULL;
2185         return nfserr_bad_seqid;
2186 }
2187
2188 __be32
2189 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2190                    struct nfsd4_open_confirm *oc)
2191 {
2192         __be32 status;
2193         struct nfs4_stateowner *sop;
2194         struct nfs4_stateid *stp;
2195
2196         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2197                         (int)cstate->current_fh.fh_dentry->d_name.len,
2198                         cstate->current_fh.fh_dentry->d_name.name);
2199
2200         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2201         if (status)
2202                 return status;
2203
2204         nfs4_lock_state();
2205
2206         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2207                                         oc->oc_seqid, &oc->oc_req_stateid,
2208                                         CONFIRM | OPEN_STATE,
2209                                         &oc->oc_stateowner, &stp, NULL)))
2210                 goto out;
2211
2212         sop = oc->oc_stateowner;
2213         sop->so_confirmed = 1;
2214         update_stateid(&stp->st_stateid);
2215         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2216         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2217                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2218                          stp->st_stateid.si_boot,
2219                          stp->st_stateid.si_stateownerid,
2220                          stp->st_stateid.si_fileid,
2221                          stp->st_stateid.si_generation);
2222
2223         nfsd4_create_clid_dir(sop->so_client);
2224 out:
2225         if (oc->oc_stateowner) {
2226                 nfs4_get_stateowner(oc->oc_stateowner);
2227                 cstate->replay_owner = oc->oc_stateowner;
2228         }
2229         nfs4_unlock_state();
2230         return status;
2231 }
2232
2233
2234 /*
2235  * unset all bits in union bitmap (bmap) that
2236  * do not exist in share (from successful OPEN_DOWNGRADE)
2237  */
2238 static void
2239 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2240 {
2241         int i;
2242         for (i = 1; i < 4; i++) {
2243                 if ((i & access) != i)
2244                         __clear_bit(i, bmap);
2245         }
2246 }
2247
2248 static void
2249 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2250 {
2251         int i;
2252         for (i = 0; i < 4; i++) {
2253                 if ((i & deny) != i)
2254                         __clear_bit(i, bmap);
2255         }
2256 }
2257
2258 __be32
2259 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2260                      struct nfsd4_compound_state *cstate,
2261                      struct nfsd4_open_downgrade *od)
2262 {
2263         __be32 status;
2264         struct nfs4_stateid *stp;
2265         unsigned int share_access;
2266
2267         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2268                         (int)cstate->current_fh.fh_dentry->d_name.len,
2269                         cstate->current_fh.fh_dentry->d_name.name);
2270
2271         if (!access_valid(od->od_share_access)
2272                         || !deny_valid(od->od_share_deny))
2273                 return nfserr_inval;
2274
2275         nfs4_lock_state();
2276         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2277                                         od->od_seqid,
2278                                         &od->od_stateid,
2279                                         OPEN_STATE,
2280                                         &od->od_stateowner, &stp, NULL)))
2281                 goto out;
2282
2283         status = nfserr_inval;
2284         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2285                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2286                         stp->st_access_bmap, od->od_share_access);
2287                 goto out;
2288         }
2289         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2290                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2291                         stp->st_deny_bmap, od->od_share_deny);
2292                 goto out;
2293         }
2294         set_access(&share_access, stp->st_access_bmap);
2295         nfs4_file_downgrade(stp->st_vfs_file,
2296                             share_access & ~od->od_share_access);
2297
2298         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2299         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2300
2301         update_stateid(&stp->st_stateid);
2302         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2303         status = nfs_ok;
2304 out:
2305         if (od->od_stateowner) {
2306                 nfs4_get_stateowner(od->od_stateowner);
2307                 cstate->replay_owner = od->od_stateowner;
2308         }
2309         nfs4_unlock_state();
2310         return status;
2311 }
2312
2313 /*
2314  * nfs4_unlock_state() called after encode
2315  */
2316 __be32
2317 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2318             struct nfsd4_close *close)
2319 {
2320         __be32 status;
2321         struct nfs4_stateid *stp;
2322
2323         dprintk("NFSD: nfsd4_close on file %.*s\n",
2324                         (int)cstate->current_fh.fh_dentry->d_name.len,
2325                         cstate->current_fh.fh_dentry->d_name.name);
2326
2327         nfs4_lock_state();
2328         /* check close_lru for replay */
2329         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2330                                         close->cl_seqid,
2331                                         &close->cl_stateid,
2332                                         OPEN_STATE | CLOSE_STATE,
2333                                         &close->cl_stateowner, &stp, NULL)))
2334                 goto out;
2335         status = nfs_ok;
2336         update_stateid(&stp->st_stateid);
2337         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2338
2339         /* release_stateid() calls nfsd_close() if needed */
2340         release_stateid(stp, OPEN_STATE);
2341
2342         /* place unused nfs4_stateowners on so_close_lru list to be
2343          * released by the laundromat service after the lease period
2344          * to enable us to handle CLOSE replay
2345          */
2346         if (list_empty(&close->cl_stateowner->so_stateids))
2347                 move_to_close_lru(close->cl_stateowner);
2348 out:
2349         if (close->cl_stateowner) {
2350                 nfs4_get_stateowner(close->cl_stateowner);
2351                 cstate->replay_owner = close->cl_stateowner;
2352         }
2353         nfs4_unlock_state();
2354         return status;
2355 }
2356
2357 __be32
2358 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2359                   struct nfsd4_delegreturn *dr)
2360 {
2361         __be32 status;
2362
2363         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2364                 goto out;
2365
2366         nfs4_lock_state();
2367         status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2368                                             &dr->dr_stateid, DELEG_RET, NULL);
2369         nfs4_unlock_state();
2370 out:
2371         return status;
2372 }
2373
2374
2375 /*
2376  * Lock owner state (byte-range locks)
2377  */
2378 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2379 #define LOCK_HASH_BITS              8
2380 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2381 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2382
2383 #define lockownerid_hashval(id) \
2384         ((id) & LOCK_HASH_MASK)
2385
2386 static inline unsigned int
2387 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2388                 struct xdr_netobj *ownername)
2389 {
2390         return (file_hashval(inode) + cl_id
2391                         + opaque_hashval(ownername->data, ownername->len))
2392                 & LOCK_HASH_MASK;
2393 }
2394
2395 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2396 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2397 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2398
2399 static struct nfs4_stateid *
2400 find_stateid(stateid_t *stid, int flags)
2401 {
2402         struct nfs4_stateid *local = NULL;
2403         u32 st_id = stid->si_stateownerid;
2404         u32 f_id = stid->si_fileid;
2405         unsigned int hashval;
2406
2407         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2408         if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2409                 hashval = stateid_hashval(st_id, f_id);
2410                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2411                         if ((local->st_stateid.si_stateownerid == st_id) &&
2412                             (local->st_stateid.si_fileid == f_id))
2413                                 return local;
2414                 }
2415         }
2416         if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2417                 hashval = stateid_hashval(st_id, f_id);
2418                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2419                         if ((local->st_stateid.si_stateownerid == st_id) &&
2420                             (local->st_stateid.si_fileid == f_id))
2421                                 return local;
2422                 }
2423         }
2424         return NULL;
2425 }
2426
2427 static struct nfs4_delegation *
2428 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2429 {
2430         struct nfs4_file *fp;
2431         struct nfs4_delegation *dl;
2432
2433         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2434                     stid->si_boot, stid->si_stateownerid,
2435                     stid->si_fileid, stid->si_generation);
2436
2437         fp = find_file(ino);
2438         if (!fp)
2439                 return NULL;
2440         dl = find_delegation_file(fp, stid);
2441         put_nfs4_file(fp);
2442         return dl;
2443 }
2444
2445 /*
2446  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2447  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2448  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2449  * locking, this prevents us from being completely protocol-compliant.  The
2450  * real solution to this problem is to start using unsigned file offsets in
2451  * the VFS, but this is a very deep change!
2452  */
2453 static inline void
2454 nfs4_transform_lock_offset(struct file_lock *lock)
2455 {
2456         if (lock->fl_start < 0)
2457                 lock->fl_start = OFFSET_MAX;
2458         if (lock->fl_end < 0)
2459                 lock->fl_end = OFFSET_MAX;
2460 }
2461
2462 /* Hack!: For now, we're defining this just so we can use a pointer to it
2463  * as a unique cookie to identify our (NFSv4's) posix locks. */
2464 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2465 };
2466
2467 static inline void
2468 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2469 {
2470         struct nfs4_stateowner *sop;
2471         unsigned int hval;
2472
2473         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2474                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2475                 hval = lockownerid_hashval(sop->so_id);
2476                 kref_get(&sop->so_ref);
2477                 deny->ld_sop = sop;
2478                 deny->ld_clientid = sop->so_client->cl_clientid;
2479         } else {
2480                 deny->ld_sop = NULL;
2481                 deny->ld_clientid.cl_boot = 0;
2482                 deny->ld_clientid.cl_id = 0;
2483         }
2484         deny->ld_start = fl->fl_start;
2485         deny->ld_length = ~(u64)0;
2486         if (fl->fl_end != ~(u64)0)
2487                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2488         deny->ld_type = NFS4_READ_LT;
2489         if (fl->fl_type != F_RDLCK)
2490                 deny->ld_type = NFS4_WRITE_LT;
2491 }
2492
2493 static struct nfs4_stateowner *
2494 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2495                 struct xdr_netobj *owner)
2496 {
2497         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2498         struct nfs4_stateowner *op;
2499
2500         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2501                 if (same_owner_str(op, owner, clid))
2502                         return op;
2503         }
2504         return NULL;
2505 }
2506
2507 /*
2508  * Alloc a lock owner structure.
2509  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2510  * occured.
2511  *
2512  * strhashval = lock_ownerstr_hashval
2513  */
2514
2515 static struct nfs4_stateowner *
2516 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2517         struct nfs4_stateowner *sop;
2518         struct nfs4_replay *rp;
2519         unsigned int idhashval;
2520
2521         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2522                 return NULL;
2523         idhashval = lockownerid_hashval(current_ownerid);
2524         INIT_LIST_HEAD(&sop->so_idhash);
2525         INIT_LIST_HEAD(&sop->so_strhash);
2526         INIT_LIST_HEAD(&sop->so_perclient);
2527         INIT_LIST_HEAD(&sop->so_stateids);
2528         INIT_LIST_HEAD(&sop->so_perstateid);
2529         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2530         sop->so_time = 0;
2531         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2532         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2533         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2534         sop->so_is_open_owner = 0;
2535         sop->so_id = current_ownerid++;
2536         sop->so_client = clp;
2537         /* It is the openowner seqid that will be incremented in encode in the
2538          * case of new lockowners; so increment the lock seqid manually: */
2539         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2540         sop->so_confirmed = 1;
2541         rp = &sop->so_replay;
2542         rp->rp_status = nfserr_serverfault;
2543         rp->rp_buflen = 0;
2544         rp->rp_buf = rp->rp_ibuf;
2545         return sop;
2546 }
2547
2548 static struct nfs4_stateid *
2549 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2550 {
2551         struct nfs4_stateid *stp;
2552         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2553
2554         stp = nfs4_alloc_stateid();
2555         if (stp == NULL)
2556                 goto out;
2557         INIT_LIST_HEAD(&stp->st_hash);
2558         INIT_LIST_HEAD(&stp->st_perfile);
2559         INIT_LIST_HEAD(&stp->st_perstateowner);
2560         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2561         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2562         list_add(&stp->st_perfile, &fp->fi_stateids);
2563         list_add(&stp->st_perstateowner, &sop->so_stateids);
2564         stp->st_stateowner = sop;
2565         get_nfs4_file(fp);
2566         stp->st_file = fp;
2567         stp->st_stateid.si_boot = boot_time;
2568         stp->st_stateid.si_stateownerid = sop->so_id;
2569         stp->st_stateid.si_fileid = fp->fi_id;
2570         stp->st_stateid.si_generation = 0;
2571         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2572         stp->st_access_bmap = open_stp->st_access_bmap;
2573         stp->st_deny_bmap = open_stp->st_deny_bmap;
2574         stp->st_openstp = open_stp;
2575
2576 out:
2577         return stp;
2578 }
2579
2580 static int
2581 check_lock_length(u64 offset, u64 length)
2582 {
2583         return ((length == 0)  || ((length != ~(u64)0) &&
2584              LOFF_OVERFLOW(offset, length)));
2585 }
2586
2587 /*
2588  *  LOCK operation
2589  */
2590 __be32
2591 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2592            struct nfsd4_lock *lock)
2593 {
2594         struct nfs4_stateowner *open_sop = NULL;
2595         struct nfs4_stateowner *lock_sop = NULL;
2596         struct nfs4_stateid *lock_stp;
2597         struct file *filp;
2598         struct file_lock file_lock;
2599         struct file_lock conflock;
2600         __be32 status = 0;
2601         unsigned int strhashval;
2602         unsigned int cmd;
2603         int err;
2604
2605         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2606                 (long long) lock->lk_offset,
2607                 (long long) lock->lk_length);
2608
2609         if (check_lock_length(lock->lk_offset, lock->lk_length))
2610                  return nfserr_inval;
2611
2612         if ((status = fh_verify(rqstp, &cstate->current_fh,
2613                                 S_IFREG, MAY_LOCK))) {
2614                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2615                 return status;
2616         }
2617
2618         nfs4_lock_state();
2619
2620         if (lock->lk_is_new) {
2621                 /*
2622                  * Client indicates that this is a new lockowner.
2623                  * Use open owner and open stateid to create lock owner and
2624                  * lock stateid.
2625                  */
2626                 struct nfs4_stateid *open_stp = NULL;
2627                 struct nfs4_file *fp;
2628
2629                 status = nfserr_stale_clientid;
2630                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2631                         goto out;
2632
2633                 /* validate and update open stateid and open seqid */
2634                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2635                                         lock->lk_new_open_seqid,
2636                                         &lock->lk_new_open_stateid,
2637                                         OPEN_STATE,
2638                                         &lock->lk_replay_owner, &open_stp,
2639                                         lock);
2640                 if (status)
2641                         goto out;
2642                 open_sop = lock->lk_replay_owner;
2643                 /* create lockowner and lock stateid */
2644                 fp = open_stp->st_file;
2645                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
2646                                 open_sop->so_client->cl_clientid.cl_id,
2647                                 &lock->v.new.owner);
2648                 /* XXX: Do we need to check for duplicate stateowners on
2649                  * the same file, or should they just be allowed (and
2650                  * create new stateids)? */
2651                 status = nfserr_resource;
2652                 lock_sop = alloc_init_lock_stateowner(strhashval,
2653                                 open_sop->so_client, open_stp, lock);
2654                 if (lock_sop == NULL)
2655                         goto out;
2656                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2657                 if (lock_stp == NULL)
2658                         goto out;
2659         } else {
2660                 /* lock (lock owner + lock stateid) already exists */
2661                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2662                                        lock->lk_old_lock_seqid,
2663                                        &lock->lk_old_lock_stateid,
2664                                        LOCK_STATE,
2665                                        &lock->lk_replay_owner, &lock_stp, lock);
2666                 if (status)
2667                         goto out;
2668                 lock_sop = lock->lk_replay_owner;
2669         }
2670         /* lock->lk_replay_owner and lock_stp have been created or found */
2671         filp = lock_stp->st_vfs_file;
2672
2673         status = nfserr_grace;
2674         if (nfs4_in_grace() && !lock->lk_reclaim)
2675                 goto out;
2676         status = nfserr_no_grace;
2677         if (!nfs4_in_grace() && lock->lk_reclaim)
2678                 goto out;
2679
2680         locks_init_lock(&file_lock);
2681         switch (lock->lk_type) {
2682                 case NFS4_READ_LT:
2683                 case NFS4_READW_LT:
2684                         file_lock.fl_type = F_RDLCK;
2685                         cmd = F_SETLK;
2686                 break;
2687                 case NFS4_WRITE_LT:
2688                 case NFS4_WRITEW_LT:
2689                         file_lock.fl_type = F_WRLCK;
2690                         cmd = F_SETLK;
2691                 break;
2692                 default:
2693                         status = nfserr_inval;
2694                 goto out;
2695         }
2696         file_lock.fl_owner = (fl_owner_t)lock_sop;
2697         file_lock.fl_pid = current->tgid;
2698         file_lock.fl_file = filp;
2699         file_lock.fl_flags = FL_POSIX;
2700         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2701
2702         file_lock.fl_start = lock->lk_offset;
2703         if ((lock->lk_length == ~(u64)0) ||
2704                         LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2705                 file_lock.fl_end = ~(u64)0;
2706         else
2707                 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2708         nfs4_transform_lock_offset(&file_lock);
2709
2710         /*
2711         * Try to lock the file in the VFS.
2712         * Note: locks.c uses the BKL to protect the inode's lock list.
2713         */
2714
2715         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2716         switch (-err) {
2717         case 0: /* success! */
2718                 update_stateid(&lock_stp->st_stateid);
2719                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
2720                                 sizeof(stateid_t));
2721                 status = 0;
2722                 break;
2723         case (EAGAIN):          /* conflock holds conflicting lock */
2724                 status = nfserr_denied;
2725                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2726                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2727                 break;
2728         case (EDEADLK):
2729                 status = nfserr_deadlock;
2730                 break;
2731         default:
2732                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2733                 status = nfserr_resource;
2734                 break;
2735         }
2736 out:
2737         if (status && lock->lk_is_new && lock_sop)
2738                 release_stateowner(lock_sop);
2739         if (lock->lk_replay_owner) {
2740                 nfs4_get_stateowner(lock->lk_replay_owner);
2741                 cstate->replay_owner = lock->lk_replay_owner;
2742         }
2743         nfs4_unlock_state();
2744         return status;
2745 }
2746
2747 /*
2748  * LOCKT operation
2749  */
2750 __be32
2751 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2752             struct nfsd4_lockt *lockt)
2753 {
2754         struct inode *inode;
2755         struct file file;
2756         struct file_lock file_lock;
2757         int error;
2758         __be32 status;
2759
2760         if (nfs4_in_grace())
2761                 return nfserr_grace;
2762
2763         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2764                  return nfserr_inval;
2765
2766         lockt->lt_stateowner = NULL;
2767         nfs4_lock_state();
2768
2769         status = nfserr_stale_clientid;
2770         if (STALE_CLIENTID(&lockt->lt_clientid))
2771                 goto out;
2772
2773         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2774                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2775                 if (status == nfserr_symlink)
2776                         status = nfserr_inval;
2777                 goto out;
2778         }
2779
2780         inode = cstate->current_fh.fh_dentry->d_inode;
2781         locks_init_lock(&file_lock);
2782         switch (lockt->lt_type) {
2783                 case NFS4_READ_LT:
2784                 case NFS4_READW_LT:
2785                         file_lock.fl_type = F_RDLCK;
2786                 break;
2787                 case NFS4_WRITE_LT:
2788                 case NFS4_WRITEW_LT:
2789                         file_lock.fl_type = F_WRLCK;
2790                 break;
2791                 default:
2792                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2793                         status = nfserr_inval;
2794                 goto out;
2795         }
2796
2797         lockt->lt_stateowner = find_lockstateowner_str(inode,
2798                         &lockt->lt_clientid, &lockt->lt_owner);
2799         if (lockt->lt_stateowner)
2800                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2801         file_lock.fl_pid = current->tgid;
2802         file_lock.fl_flags = FL_POSIX;
2803         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2804
2805         file_lock.fl_start = lockt->lt_offset;
2806         if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2807                 file_lock.fl_end = ~(u64)0;
2808         else
2809                 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2810
2811         nfs4_transform_lock_offset(&file_lock);
2812
2813         /* vfs_test_lock uses the struct file _only_ to resolve the inode.
2814          * since LOCKT doesn't require an OPEN, and therefore a struct
2815          * file may not exist, pass vfs_test_lock a struct file with
2816          * only the dentry:inode set.
2817          */
2818         memset(&file, 0, sizeof (struct file));
2819         file.f_path.dentry = cstate->current_fh.fh_dentry;
2820
2821         status = nfs_ok;
2822         error = vfs_test_lock(&file, &file_lock);
2823         if (error) {
2824                 status = nfserrno(error);
2825                 goto out;
2826         }
2827         if (file_lock.fl_type != F_UNLCK) {
2828                 status = nfserr_denied;
2829                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2830         }
2831 out:
2832         nfs4_unlock_state();
2833         return status;
2834 }
2835
2836 __be32
2837 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2838             struct nfsd4_locku *locku)
2839 {
2840         struct nfs4_stateid *stp;
2841         struct file *filp = NULL;
2842         struct file_lock file_lock;
2843         __be32 status;
2844         int err;
2845
2846         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2847                 (long long) locku->lu_offset,
2848                 (long long) locku->lu_length);
2849
2850         if (check_lock_length(locku->lu_offset, locku->lu_length))
2851                  return nfserr_inval;
2852
2853         nfs4_lock_state();
2854
2855         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2856                                         locku->lu_seqid,
2857                                         &locku->lu_stateid,
2858                                         LOCK_STATE,
2859                                         &locku->lu_stateowner, &stp, NULL)))
2860                 goto out;
2861
2862         filp = stp->st_vfs_file;
2863         BUG_ON(!filp);
2864         locks_init_lock(&file_lock);
2865         file_lock.fl_type = F_UNLCK;
2866         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2867         file_lock.fl_pid = current->tgid;
2868         file_lock.fl_file = filp;
2869         file_lock.fl_flags = FL_POSIX;
2870         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2871         file_lock.fl_start = locku->lu_offset;
2872
2873         if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2874                 file_lock.fl_end = ~(u64)0;
2875         else
2876                 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2877         nfs4_transform_lock_offset(&file_lock);
2878
2879         /*
2880         *  Try to unlock the file in the VFS.
2881         */
2882         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2883         if (err) {
2884                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2885                 goto out_nfserr;
2886         }
2887         /*
2888         * OK, unlock succeeded; the only thing left to do is update the stateid.
2889         */
2890         update_stateid(&stp->st_stateid);
2891         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2892
2893 out:
2894         if (locku->lu_stateowner) {
2895                 nfs4_get_stateowner(locku->lu_stateowner);
2896                 cstate->replay_owner = locku->lu_stateowner;
2897         }
2898         nfs4_unlock_state();
2899         return status;
2900
2901 out_nfserr:
2902         status = nfserrno(err);
2903         goto out;
2904 }
2905
2906 /*
2907  * returns
2908  *      1: locks held by lockowner
2909  *      0: no locks held by lockowner
2910  */
2911 static int
2912 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2913 {
2914         struct file_lock **flpp;
2915         struct inode *inode = filp->f_path.dentry->d_inode;
2916         int status = 0;
2917
2918         lock_kernel();
2919         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2920                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2921                         status = 1;
2922                         goto out;
2923                 }
2924         }
2925 out:
2926         unlock_kernel();
2927         return status;
2928 }
2929
2930 __be32
2931 nfsd4_release_lockowner(struct svc_rqst *rqstp,
2932                         struct nfsd4_compound_state *cstate,
2933                         struct nfsd4_release_lockowner *rlockowner)
2934 {
2935         clientid_t *clid = &rlockowner->rl_clientid;
2936         struct nfs4_stateowner *sop;
2937         struct nfs4_stateid *stp;
2938         struct xdr_netobj *owner = &rlockowner->rl_owner;
2939         struct list_head matches;
2940         int i;
2941         __be32 status;
2942
2943         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
2944                 clid->cl_boot, clid->cl_id);
2945
2946         /* XXX check for lease expiration */
2947
2948         status = nfserr_stale_clientid;
2949         if (STALE_CLIENTID(clid))
2950                 return status;
2951
2952         nfs4_lock_state();
2953
2954         status = nfserr_locks_held;
2955         /* XXX: we're doing a linear search through all the lockowners.
2956          * Yipes!  For now we'll just hope clients aren't really using
2957          * release_lockowner much, but eventually we have to fix these
2958          * data structures. */
2959         INIT_LIST_HEAD(&matches);
2960         for (i = 0; i < LOCK_HASH_SIZE; i++) {
2961                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
2962                         if (!same_owner_str(sop, owner, clid))
2963                                 continue;
2964                         list_for_each_entry(stp, &sop->so_stateids,
2965                                         st_perstateowner) {
2966                                 if (check_for_locks(stp->st_vfs_file, sop))
2967                                         goto out;
2968                                 /* Note: so_perclient unused for lockowners,
2969                                  * so it's OK to fool with here. */
2970                                 list_add(&sop->so_perclient, &matches);
2971                         }
2972                 }
2973         }
2974         /* Clients probably won't expect us to return with some (but not all)
2975          * of the lockowner state released; so don't release any until all
2976          * have been checked. */
2977         status = nfs_ok;
2978         while (!list_empty(&matches)) {
2979                 sop = list_entry(matches.next, struct nfs4_stateowner,
2980                                                                 so_perclient);
2981                 /* unhash_stateowner deletes so_perclient only
2982                  * for openowners. */
2983                 list_del(&sop->so_perclient);
2984                 release_stateowner(sop);
2985         }
2986 out:
2987         nfs4_unlock_state();
2988         return status;
2989 }
2990
2991 static inline struct nfs4_client_reclaim *
2992 alloc_reclaim(void)
2993 {
2994         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
2995 }
2996
2997 int
2998 nfs4_has_reclaimed_state(const char *name)
2999 {
3000         unsigned int strhashval = clientstr_hashval(name);
3001         struct nfs4_client *clp;
3002
3003         clp = find_confirmed_client_by_str(name, strhashval);
3004         return clp ? 1 : 0;
3005 }
3006
3007 /*
3008  * failure => all reset bets are off, nfserr_no_grace...
3009  */
3010 int
3011 nfs4_client_to_reclaim(const char *name)
3012 {
3013         unsigned int strhashval;
3014         struct nfs4_client_reclaim *crp = NULL;
3015
3016         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3017         crp = alloc_reclaim();
3018         if (!crp)
3019                 return 0;
3020         strhashval = clientstr_hashval(name);
3021         INIT_LIST_HEAD(&crp->cr_strhash);
3022         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3023         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3024         reclaim_str_hashtbl_size++;
3025         return 1;
3026 }
3027
3028 static void
3029 nfs4_release_reclaim(void)
3030 {
3031         struct nfs4_client_reclaim *crp = NULL;
3032         int i;
3033
3034         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3035                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3036                         crp = list_entry(reclaim_str_hashtbl[i].next,
3037                                         struct nfs4_client_reclaim, cr_strhash);
3038                         list_del(&crp->cr_strhash);
3039                         kfree(crp);
3040                         reclaim_str_hashtbl_size--;
3041                 }
3042         }
3043         BUG_ON(reclaim_str_hashtbl_size);
3044 }
3045
3046 /*
3047  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3048 static struct nfs4_client_reclaim *
3049 nfs4_find_reclaim_client(clientid_t *clid)
3050 {
3051         unsigned int strhashval;
3052         struct nfs4_client *clp;
3053         struct nfs4_client_reclaim *crp = NULL;
3054
3055
3056         /* find clientid in conf_id_hashtbl */
3057         clp = find_confirmed_client(clid);
3058         if (clp == NULL)
3059                 return NULL;
3060
3061         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3062                             clp->cl_name.len, clp->cl_name.data,
3063                             clp->cl_recdir);
3064
3065         /* find clp->cl_name in reclaim_str_hashtbl */
3066         strhashval = clientstr_hashval(clp->cl_recdir);
3067         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3068                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3069                         return crp;
3070                 }
3071         }
3072         return NULL;
3073 }
3074
3075 /*
3076 * Called from OPEN. Look for clientid in reclaim list.
3077 */
3078 __be32
3079 nfs4_check_open_reclaim(clientid_t *clid)
3080 {
3081         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3082 }
3083
3084 /* initialization to perform at module load time: */
3085
3086 int
3087 nfs4_state_init(void)
3088 {
3089         int i, status;
3090
3091         status = nfsd4_init_slabs();
3092         if (status)
3093                 return status;
3094         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3095                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3096                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3097                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3098                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3099         }
3100         for (i = 0; i < FILE_HASH_SIZE; i++) {
3101                 INIT_LIST_HEAD(&file_hashtbl[i]);
3102         }
3103         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3104                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3105                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3106         }
3107         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3108                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3109                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3110         }
3111         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3112                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3113                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3114         }
3115         memset(&onestateid, ~0, sizeof(stateid_t));
3116         INIT_LIST_HEAD(&close_lru);
3117         INIT_LIST_HEAD(&client_lru);
3118         INIT_LIST_HEAD(&del_recall_lru);
3119         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3120                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3121         reclaim_str_hashtbl_size = 0;
3122         return 0;
3123 }
3124
3125 static void
3126 nfsd4_load_reboot_recovery_data(void)
3127 {
3128         int status;
3129
3130         nfs4_lock_state();
3131         nfsd4_init_recdir(user_recovery_dirname);
3132         status = nfsd4_recdir_load();
3133         nfs4_unlock_state();
3134         if (status)
3135                 printk("NFSD: Failure reading reboot recovery data\n");
3136 }
3137
3138 unsigned long
3139 get_nfs4_grace_period(void)
3140 {
3141         return max(user_lease_time, lease_time) * HZ;
3142 }
3143
3144 /*
3145  * Since the lifetime of a delegation isn't limited to that of an open, a
3146  * client may quite reasonably hang on to a delegation as long as it has
3147  * the inode cached.  This becomes an obvious problem the first time a
3148  * client's inode cache approaches the size of the server's total memory.
3149  *
3150  * For now we avoid this problem by imposing a hard limit on the number
3151  * of delegations, which varies according to the server's memory size.
3152  */
3153 static void
3154 set_max_delegations(void)
3155 {
3156         /*
3157          * Allow at most 4 delegations per megabyte of RAM.  Quick
3158          * estimates suggest that in the worst case (where every delegation
3159          * is for a different inode), a delegation could take about 1.5K,
3160          * giving a worst case usage of about 6% of memory.
3161          */
3162         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3163 }
3164
3165 /* initialization to perform when the nfsd service is started: */
3166
3167 static void
3168 __nfs4_state_start(void)
3169 {
3170         unsigned long grace_time;
3171
3172         boot_time = get_seconds();
3173         grace_time = get_nfs_grace_period();
3174         lease_time = user_lease_time;
3175         in_grace = 1;
3176         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3177                grace_time/HZ);
3178         laundry_wq = create_singlethread_workqueue("nfsd4");
3179         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3180         set_max_delegations();
3181 }
3182
3183 void
3184 nfs4_state_start(void)
3185 {
3186         if (nfs4_init)
3187                 return;
3188         nfsd4_load_reboot_recovery_data();
3189         __nfs4_state_start();
3190         nfs4_init = 1;
3191         return;
3192 }
3193
3194 int
3195 nfs4_in_grace(void)
3196 {
3197         return in_grace;
3198 }
3199
3200 time_t
3201 nfs4_lease_time(void)
3202 {
3203         return lease_time;
3204 }
3205
3206 static void
3207 __nfs4_state_shutdown(void)
3208 {
3209         int i;
3210         struct nfs4_client *clp = NULL;
3211         struct nfs4_delegation *dp = NULL;
3212         struct list_head *pos, *next, reaplist;
3213
3214         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3215                 while (!list_empty(&conf_id_hashtbl[i])) {
3216                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3217                         expire_client(clp);
3218                 }
3219                 while (!list_empty(&unconf_str_hashtbl[i])) {
3220                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3221                         expire_client(clp);
3222                 }
3223         }
3224         INIT_LIST_HEAD(&reaplist);
3225         spin_lock(&recall_lock);
3226         list_for_each_safe(pos, next, &del_recall_lru) {
3227                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3228                 list_move(&dp->dl_recall_lru, &reaplist);
3229         }
3230         spin_unlock(&recall_lock);
3231         list_for_each_safe(pos, next, &reaplist) {
3232                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3233                 list_del_init(&dp->dl_recall_lru);
3234                 unhash_delegation(dp);
3235         }
3236
3237         nfsd4_shutdown_recdir();
3238         nfs4_init = 0;
3239 }
3240
3241 void
3242 nfs4_state_shutdown(void)
3243 {
3244         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3245         destroy_workqueue(laundry_wq);
3246         nfs4_lock_state();
3247         nfs4_release_reclaim();
3248         __nfs4_state_shutdown();
3249         nfs4_unlock_state();
3250 }
3251
3252 static void
3253 nfs4_set_recdir(char *recdir)
3254 {
3255         nfs4_lock_state();
3256         strcpy(user_recovery_dirname, recdir);
3257         nfs4_unlock_state();
3258 }
3259
3260 /*
3261  * Change the NFSv4 recovery directory to recdir.
3262  */
3263 int
3264 nfs4_reset_recoverydir(char *recdir)
3265 {
3266         int status;
3267         struct nameidata nd;
3268
3269         status = path_lookup(recdir, LOOKUP_FOLLOW, &nd);
3270         if (status)
3271                 return status;
3272         status = -ENOTDIR;
3273         if (S_ISDIR(nd.path.dentry->d_inode->i_mode)) {
3274                 nfs4_set_recdir(recdir);
3275                 status = 0;
3276         }
3277         path_put(&nd.path);
3278         return status;
3279 }
3280
3281 /*
3282  * Called when leasetime is changed.
3283  *
3284  * The only way the protocol gives us to handle on-the-fly lease changes is to
3285  * simulate a reboot.  Instead of doing that, we just wait till the next time
3286  * we start to register any changes in lease time.  If the administrator
3287  * really wants to change the lease time *now*, they can go ahead and bring
3288  * nfsd down and then back up again after changing the lease time.
3289  */
3290 void
3291 nfs4_reset_lease(time_t leasetime)
3292 {
3293         lock_kernel();
3294         user_lease_time = leasetime;
3295         unlock_kernel();
3296 }