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 /*
1177  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1178  * st_{access,deny}_bmap field of the stateid, in order to track not
1179  * only what share bits are currently in force, but also what
1180  * combinations of share bits previous opens have used.  This allows us
1181  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1182  * return an error if the client attempt to downgrade to a combination
1183  * of share bits not explicable by closing some of its previous opens.
1184  *
1185  * XXX: This enforcement is actually incomplete, since we don't keep
1186  * track of access/deny bit combinations; so, e.g., we allow:
1187  *
1188  *      OPEN allow read, deny write
1189  *      OPEN allow both, deny none
1190  *      DOWNGRADE allow read, deny none
1191  *
1192  * which we should reject.
1193  */
1194 static void
1195 set_access(unsigned int *access, unsigned long bmap) {
1196         int i;
1197
1198         *access = 0;
1199         for (i = 1; i < 4; i++) {
1200                 if (test_bit(i, &bmap))
1201                         *access |= i;
1202         }
1203 }
1204
1205 static void
1206 set_deny(unsigned int *deny, unsigned long bmap) {
1207         int i;
1208
1209         *deny = 0;
1210         for (i = 0; i < 4; i++) {
1211                 if (test_bit(i, &bmap))
1212                         *deny |= i ;
1213         }
1214 }
1215
1216 static int
1217 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1218         unsigned int access, deny;
1219
1220         set_access(&access, stp->st_access_bmap);
1221         set_deny(&deny, stp->st_deny_bmap);
1222         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1223                 return 0;
1224         return 1;
1225 }
1226
1227 /*
1228  * Called to check deny when READ with all zero stateid or
1229  * WRITE with all zero or all one stateid
1230  */
1231 static __be32
1232 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1233 {
1234         struct inode *ino = current_fh->fh_dentry->d_inode;
1235         struct nfs4_file *fp;
1236         struct nfs4_stateid *stp;
1237         __be32 ret;
1238
1239         dprintk("NFSD: nfs4_share_conflict\n");
1240
1241         fp = find_file(ino);
1242         if (!fp)
1243                 return nfs_ok;
1244         ret = nfserr_locked;
1245         /* Search for conflicting share reservations */
1246         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1247                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1248                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1249                         goto out;
1250         }
1251         ret = nfs_ok;
1252 out:
1253         put_nfs4_file(fp);
1254         return ret;
1255 }
1256
1257 static inline void
1258 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1259 {
1260         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1261                 drop_file_write_access(filp);
1262                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1263         }
1264 }
1265
1266 /*
1267  * Recall a delegation
1268  */
1269 static int
1270 do_recall(void *__dp)
1271 {
1272         struct nfs4_delegation *dp = __dp;
1273
1274         dp->dl_file->fi_had_conflict = true;
1275         nfsd4_cb_recall(dp);
1276         return 0;
1277 }
1278
1279 /*
1280  * Spawn a thread to perform a recall on the delegation represented
1281  * by the lease (file_lock)
1282  *
1283  * Called from break_lease() with lock_kernel() held.
1284  * Note: we assume break_lease will only call this *once* for any given
1285  * lease.
1286  */
1287 static
1288 void nfsd_break_deleg_cb(struct file_lock *fl)
1289 {
1290         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1291         struct task_struct *t;
1292
1293         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1294         if (!dp)
1295                 return;
1296
1297         /* We're assuming the state code never drops its reference
1298          * without first removing the lease.  Since we're in this lease
1299          * callback (and since the lease code is serialized by the kernel
1300          * lock) we know the server hasn't removed the lease yet, we know
1301          * it's safe to take a reference: */
1302         atomic_inc(&dp->dl_count);
1303         atomic_inc(&dp->dl_client->cl_count);
1304
1305         spin_lock(&recall_lock);
1306         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1307         spin_unlock(&recall_lock);
1308
1309         /* only place dl_time is set. protected by lock_kernel*/
1310         dp->dl_time = get_seconds();
1311
1312         /*
1313          * We don't want the locks code to timeout the lease for us;
1314          * we'll remove it ourself if the delegation isn't returned
1315          * in time.
1316          */
1317         fl->fl_break_time = 0;
1318
1319         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1320         if (IS_ERR(t)) {
1321                 struct nfs4_client *clp = dp->dl_client;
1322
1323                 printk(KERN_INFO "NFSD: Callback thread failed for "
1324                         "for client (clientid %08x/%08x)\n",
1325                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1326                 put_nfs4_client(dp->dl_client);
1327                 nfs4_put_delegation(dp);
1328         }
1329 }
1330
1331 /*
1332  * The file_lock is being reapd.
1333  *
1334  * Called by locks_free_lock() with lock_kernel() held.
1335  */
1336 static
1337 void nfsd_release_deleg_cb(struct file_lock *fl)
1338 {
1339         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1340
1341         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1342
1343         if (!(fl->fl_flags & FL_LEASE) || !dp)
1344                 return;
1345         dp->dl_flock = NULL;
1346 }
1347
1348 /*
1349  * Set the delegation file_lock back pointer.
1350  *
1351  * Called from setlease() with lock_kernel() held.
1352  */
1353 static
1354 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1355 {
1356         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1357
1358         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1359         if (!dp)
1360                 return;
1361         dp->dl_flock = new;
1362 }
1363
1364 /*
1365  * Called from setlease() with lock_kernel() held
1366  */
1367 static
1368 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1369 {
1370         struct nfs4_delegation *onlistd =
1371                 (struct nfs4_delegation *)onlist->fl_owner;
1372         struct nfs4_delegation *tryd =
1373                 (struct nfs4_delegation *)try->fl_owner;
1374
1375         if (onlist->fl_lmops != try->fl_lmops)
1376                 return 0;
1377
1378         return onlistd->dl_client == tryd->dl_client;
1379 }
1380
1381
1382 static
1383 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1384 {
1385         if (arg & F_UNLCK)
1386                 return lease_modify(onlist, arg);
1387         else
1388                 return -EAGAIN;
1389 }
1390
1391 static struct lock_manager_operations nfsd_lease_mng_ops = {
1392         .fl_break = nfsd_break_deleg_cb,
1393         .fl_release_private = nfsd_release_deleg_cb,
1394         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1395         .fl_mylease = nfsd_same_client_deleg_cb,
1396         .fl_change = nfsd_change_deleg_cb,
1397 };
1398
1399
1400 __be32
1401 nfsd4_process_open1(struct nfsd4_open *open)
1402 {
1403         clientid_t *clientid = &open->op_clientid;
1404         struct nfs4_client *clp = NULL;
1405         unsigned int strhashval;
1406         struct nfs4_stateowner *sop = NULL;
1407
1408         if (!check_name(open->op_owner))
1409                 return nfserr_inval;
1410
1411         if (STALE_CLIENTID(&open->op_clientid))
1412                 return nfserr_stale_clientid;
1413
1414         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1415         sop = find_openstateowner_str(strhashval, open);
1416         open->op_stateowner = sop;
1417         if (!sop) {
1418                 /* Make sure the client's lease hasn't expired. */
1419                 clp = find_confirmed_client(clientid);
1420                 if (clp == NULL)
1421                         return nfserr_expired;
1422                 goto renew;
1423         }
1424         if (!sop->so_confirmed) {
1425                 /* Replace unconfirmed owners without checking for replay. */
1426                 clp = sop->so_client;
1427                 release_stateowner(sop);
1428                 open->op_stateowner = NULL;
1429                 goto renew;
1430         }
1431         if (open->op_seqid == sop->so_seqid - 1) {
1432                 if (sop->so_replay.rp_buflen)
1433                         return nfserr_replay_me;
1434                 /* The original OPEN failed so spectacularly
1435                  * that we don't even have replay data saved!
1436                  * Therefore, we have no choice but to continue
1437                  * processing this OPEN; presumably, we'll
1438                  * fail again for the same reason.
1439                  */
1440                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1441                 goto renew;
1442         }
1443         if (open->op_seqid != sop->so_seqid)
1444                 return nfserr_bad_seqid;
1445 renew:
1446         if (open->op_stateowner == NULL) {
1447                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1448                 if (sop == NULL)
1449                         return nfserr_resource;
1450                 open->op_stateowner = sop;
1451         }
1452         list_del_init(&sop->so_close_lru);
1453         renew_client(sop->so_client);
1454         return nfs_ok;
1455 }
1456
1457 static inline __be32
1458 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1459 {
1460         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1461                 return nfserr_openmode;
1462         else
1463                 return nfs_ok;
1464 }
1465
1466 static struct nfs4_delegation *
1467 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1468 {
1469         struct nfs4_delegation *dp;
1470
1471         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1472                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1473                         return dp;
1474         }
1475         return NULL;
1476 }
1477
1478 static __be32
1479 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1480                 struct nfs4_delegation **dp)
1481 {
1482         int flags;
1483         __be32 status = nfserr_bad_stateid;
1484
1485         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1486         if (*dp == NULL)
1487                 goto out;
1488         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1489                                                 RD_STATE : WR_STATE;
1490         status = nfs4_check_delegmode(*dp, flags);
1491         if (status)
1492                 *dp = NULL;
1493 out:
1494         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1495                 return nfs_ok;
1496         if (status)
1497                 return status;
1498         open->op_stateowner->so_confirmed = 1;
1499         return nfs_ok;
1500 }
1501
1502 static __be32
1503 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1504 {
1505         struct nfs4_stateid *local;
1506         __be32 status = nfserr_share_denied;
1507         struct nfs4_stateowner *sop = open->op_stateowner;
1508
1509         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1510                 /* ignore lock owners */
1511                 if (local->st_stateowner->so_is_open_owner == 0)
1512                         continue;
1513                 /* remember if we have seen this open owner */
1514                 if (local->st_stateowner == sop)
1515                         *stpp = local;
1516                 /* check for conflicting share reservations */
1517                 if (!test_share(local, open))
1518                         goto out;
1519         }
1520         status = 0;
1521 out:
1522         return status;
1523 }
1524
1525 static inline struct nfs4_stateid *
1526 nfs4_alloc_stateid(void)
1527 {
1528         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1529 }
1530
1531 static __be32
1532 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1533                 struct nfs4_delegation *dp,
1534                 struct svc_fh *cur_fh, int flags)
1535 {
1536         struct nfs4_stateid *stp;
1537
1538         stp = nfs4_alloc_stateid();
1539         if (stp == NULL)
1540                 return nfserr_resource;
1541
1542         if (dp) {
1543                 get_file(dp->dl_vfs_file);
1544                 stp->st_vfs_file = dp->dl_vfs_file;
1545         } else {
1546                 __be32 status;
1547                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1548                                 &stp->st_vfs_file);
1549                 if (status) {
1550                         if (status == nfserr_dropit)
1551                                 status = nfserr_jukebox;
1552                         kmem_cache_free(stateid_slab, stp);
1553                         return status;
1554                 }
1555         }
1556         *stpp = stp;
1557         return 0;
1558 }
1559
1560 static inline __be32
1561 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1562                 struct nfsd4_open *open)
1563 {
1564         struct iattr iattr = {
1565                 .ia_valid = ATTR_SIZE,
1566                 .ia_size = 0,
1567         };
1568         if (!open->op_truncate)
1569                 return 0;
1570         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1571                 return nfserr_inval;
1572         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1573 }
1574
1575 static __be32
1576 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1577 {
1578         struct file *filp = stp->st_vfs_file;
1579         struct inode *inode = filp->f_path.dentry->d_inode;
1580         unsigned int share_access, new_writer;
1581         __be32 status;
1582
1583         set_access(&share_access, stp->st_access_bmap);
1584         new_writer = (~share_access) & open->op_share_access
1585                         & NFS4_SHARE_ACCESS_WRITE;
1586
1587         if (new_writer) {
1588                 int err = get_write_access(inode);
1589                 if (err)
1590                         return nfserrno(err);
1591                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
1592                 if (err)
1593                         return nfserrno(err);
1594                 file_take_write(filp);
1595         }
1596         status = nfsd4_truncate(rqstp, cur_fh, open);
1597         if (status) {
1598                 if (new_writer)
1599                         put_write_access(inode);
1600                 return status;
1601         }
1602         /* remember the open */
1603         filp->f_mode |= open->op_share_access;
1604         __set_bit(open->op_share_access, &stp->st_access_bmap);
1605         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1606
1607         return nfs_ok;
1608 }
1609
1610
1611 static void
1612 nfs4_set_claim_prev(struct nfsd4_open *open)
1613 {
1614         open->op_stateowner->so_confirmed = 1;
1615         open->op_stateowner->so_client->cl_firststate = 1;
1616 }
1617
1618 /*
1619  * Attempt to hand out a delegation.
1620  */
1621 static void
1622 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1623 {
1624         struct nfs4_delegation *dp;
1625         struct nfs4_stateowner *sop = stp->st_stateowner;
1626         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1627         struct file_lock fl, *flp = &fl;
1628         int status, flag = 0;
1629
1630         flag = NFS4_OPEN_DELEGATE_NONE;
1631         open->op_recall = 0;
1632         switch (open->op_claim_type) {
1633                 case NFS4_OPEN_CLAIM_PREVIOUS:
1634                         if (!atomic_read(&cb->cb_set))
1635                                 open->op_recall = 1;
1636                         flag = open->op_delegate_type;
1637                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1638                                 goto out;
1639                         break;
1640                 case NFS4_OPEN_CLAIM_NULL:
1641                         /* Let's not give out any delegations till everyone's
1642                          * had the chance to reclaim theirs.... */
1643                         if (nfs4_in_grace())
1644                                 goto out;
1645                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1646                                 goto out;
1647                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1648                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1649                         else
1650                                 flag = NFS4_OPEN_DELEGATE_READ;
1651                         break;
1652                 default:
1653                         goto out;
1654         }
1655
1656         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1657         if (dp == NULL) {
1658                 flag = NFS4_OPEN_DELEGATE_NONE;
1659                 goto out;
1660         }
1661         locks_init_lock(&fl);
1662         fl.fl_lmops = &nfsd_lease_mng_ops;
1663         fl.fl_flags = FL_LEASE;
1664         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
1665         fl.fl_end = OFFSET_MAX;
1666         fl.fl_owner =  (fl_owner_t)dp;
1667         fl.fl_file = stp->st_vfs_file;
1668         fl.fl_pid = current->tgid;
1669
1670         /* vfs_setlease checks to see if delegation should be handed out.
1671          * the lock_manager callbacks fl_mylease and fl_change are used
1672          */
1673         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
1674                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1675                 unhash_delegation(dp);
1676                 flag = NFS4_OPEN_DELEGATE_NONE;
1677                 goto out;
1678         }
1679
1680         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1681
1682         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1683                      dp->dl_stateid.si_boot,
1684                      dp->dl_stateid.si_stateownerid,
1685                      dp->dl_stateid.si_fileid,
1686                      dp->dl_stateid.si_generation);
1687 out:
1688         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1689                         && flag == NFS4_OPEN_DELEGATE_NONE
1690                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1691                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1692         open->op_delegate_type = flag;
1693 }
1694
1695 /*
1696  * called with nfs4_lock_state() held.
1697  */
1698 __be32
1699 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1700 {
1701         struct nfs4_file *fp = NULL;
1702         struct inode *ino = current_fh->fh_dentry->d_inode;
1703         struct nfs4_stateid *stp = NULL;
1704         struct nfs4_delegation *dp = NULL;
1705         __be32 status;
1706
1707         status = nfserr_inval;
1708         if (!access_valid(open->op_share_access)
1709                         || !deny_valid(open->op_share_deny))
1710                 goto out;
1711         /*
1712          * Lookup file; if found, lookup stateid and check open request,
1713          * and check for delegations in the process of being recalled.
1714          * If not found, create the nfs4_file struct
1715          */
1716         fp = find_file(ino);
1717         if (fp) {
1718                 if ((status = nfs4_check_open(fp, open, &stp)))
1719                         goto out;
1720                 status = nfs4_check_deleg(fp, open, &dp);
1721                 if (status)
1722                         goto out;
1723         } else {
1724                 status = nfserr_bad_stateid;
1725                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1726                         goto out;
1727                 status = nfserr_resource;
1728                 fp = alloc_init_file(ino);
1729                 if (fp == NULL)
1730                         goto out;
1731         }
1732
1733         /*
1734          * OPEN the file, or upgrade an existing OPEN.
1735          * If truncate fails, the OPEN fails.
1736          */
1737         if (stp) {
1738                 /* Stateid was found, this is an OPEN upgrade */
1739                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1740                 if (status)
1741                         goto out;
1742                 update_stateid(&stp->st_stateid);
1743         } else {
1744                 /* Stateid was not found, this is a new OPEN */
1745                 int flags = 0;
1746                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1747                         flags |= NFSD_MAY_READ;
1748                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1749                         flags |= NFSD_MAY_WRITE;
1750                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1751                 if (status)
1752                         goto out;
1753                 init_stateid(stp, fp, open);
1754                 status = nfsd4_truncate(rqstp, current_fh, open);
1755                 if (status) {
1756                         release_stateid(stp, OPEN_STATE);
1757                         goto out;
1758                 }
1759         }
1760         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1761
1762         /*
1763         * Attempt to hand out a delegation. No error return, because the
1764         * OPEN succeeds even if we fail.
1765         */
1766         nfs4_open_delegation(current_fh, open, stp);
1767
1768         status = nfs_ok;
1769
1770         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1771                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1772                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1773 out:
1774         if (fp)
1775                 put_nfs4_file(fp);
1776         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1777                 nfs4_set_claim_prev(open);
1778         /*
1779         * To finish the open response, we just need to set the rflags.
1780         */
1781         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1782         if (!open->op_stateowner->so_confirmed)
1783                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1784
1785         return status;
1786 }
1787
1788 __be32
1789 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1790             clientid_t *clid)
1791 {
1792         struct nfs4_client *clp;
1793         __be32 status;
1794
1795         nfs4_lock_state();
1796         dprintk("process_renew(%08x/%08x): starting\n",
1797                         clid->cl_boot, clid->cl_id);
1798         status = nfserr_stale_clientid;
1799         if (STALE_CLIENTID(clid))
1800                 goto out;
1801         clp = find_confirmed_client(clid);
1802         status = nfserr_expired;
1803         if (clp == NULL) {
1804                 /* We assume the client took too long to RENEW. */
1805                 dprintk("nfsd4_renew: clientid not found!\n");
1806                 goto out;
1807         }
1808         renew_client(clp);
1809         status = nfserr_cb_path_down;
1810         if (!list_empty(&clp->cl_delegations)
1811                         && !atomic_read(&clp->cl_callback.cb_set))
1812                 goto out;
1813         status = nfs_ok;
1814 out:
1815         nfs4_unlock_state();
1816         return status;
1817 }
1818
1819 static void
1820 end_grace(void)
1821 {
1822         dprintk("NFSD: end of grace period\n");
1823         nfsd4_recdir_purge_old();
1824         in_grace = 0;
1825 }
1826
1827 static time_t
1828 nfs4_laundromat(void)
1829 {
1830         struct nfs4_client *clp;
1831         struct nfs4_stateowner *sop;
1832         struct nfs4_delegation *dp;
1833         struct list_head *pos, *next, reaplist;
1834         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1835         time_t t, clientid_val = NFSD_LEASE_TIME;
1836         time_t u, test_val = NFSD_LEASE_TIME;
1837
1838         nfs4_lock_state();
1839
1840         dprintk("NFSD: laundromat service - starting\n");
1841         if (in_grace)
1842                 end_grace();
1843         list_for_each_safe(pos, next, &client_lru) {
1844                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1845                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1846                         t = clp->cl_time - cutoff;
1847                         if (clientid_val > t)
1848                                 clientid_val = t;
1849                         break;
1850                 }
1851                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1852                         clp->cl_clientid.cl_id);
1853                 nfsd4_remove_clid_dir(clp);
1854                 expire_client(clp);
1855         }
1856         INIT_LIST_HEAD(&reaplist);
1857         spin_lock(&recall_lock);
1858         list_for_each_safe(pos, next, &del_recall_lru) {
1859                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1860                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1861                         u = dp->dl_time - cutoff;
1862                         if (test_val > u)
1863                                 test_val = u;
1864                         break;
1865                 }
1866                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1867                                     dp, dp->dl_flock);
1868                 list_move(&dp->dl_recall_lru, &reaplist);
1869         }
1870         spin_unlock(&recall_lock);
1871         list_for_each_safe(pos, next, &reaplist) {
1872                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1873                 list_del_init(&dp->dl_recall_lru);
1874                 unhash_delegation(dp);
1875         }
1876         test_val = NFSD_LEASE_TIME;
1877         list_for_each_safe(pos, next, &close_lru) {
1878                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1879                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1880                         u = sop->so_time - cutoff;
1881                         if (test_val > u)
1882                                 test_val = u;
1883                         break;
1884                 }
1885                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1886                         sop->so_id);
1887                 release_stateowner(sop);
1888         }
1889         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1890                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1891         nfs4_unlock_state();
1892         return clientid_val;
1893 }
1894
1895 static struct workqueue_struct *laundry_wq;
1896 static void laundromat_main(struct work_struct *);
1897 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1898
1899 static void
1900 laundromat_main(struct work_struct *not_used)
1901 {
1902         time_t t;
1903
1904         t = nfs4_laundromat();
1905         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1906         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1907 }
1908
1909 static struct nfs4_stateowner *
1910 search_close_lru(u32 st_id, int flags)
1911 {
1912         struct nfs4_stateowner *local = NULL;
1913
1914         if (flags & CLOSE_STATE) {
1915                 list_for_each_entry(local, &close_lru, so_close_lru) {
1916                         if (local->so_id == st_id)
1917                                 return local;
1918                 }
1919         }
1920         return NULL;
1921 }
1922
1923 static inline int
1924 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1925 {
1926         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1927 }
1928
1929 static int
1930 STALE_STATEID(stateid_t *stateid)
1931 {
1932         if (stateid->si_boot == boot_time)
1933                 return 0;
1934         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1935                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1936                 stateid->si_generation);
1937         return 1;
1938 }
1939
1940 static inline int
1941 access_permit_read(unsigned long access_bmap)
1942 {
1943         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1944                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1945                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1946 }
1947
1948 static inline int
1949 access_permit_write(unsigned long access_bmap)
1950 {
1951         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1952                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1953 }
1954
1955 static
1956 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1957 {
1958         __be32 status = nfserr_openmode;
1959
1960         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1961                 goto out;
1962         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1963                 goto out;
1964         status = nfs_ok;
1965 out:
1966         return status;
1967 }
1968
1969 static inline __be32
1970 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
1971 {
1972         /* Trying to call delegreturn with a special stateid? Yuch: */
1973         if (!(flags & (RD_STATE | WR_STATE)))
1974                 return nfserr_bad_stateid;
1975         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
1976                 return nfs_ok;
1977         else if (nfs4_in_grace()) {
1978                 /* Answer in remaining cases depends on existance of
1979                  * conflicting state; so we must wait out the grace period. */
1980                 return nfserr_grace;
1981         } else if (flags & WR_STATE)
1982                 return nfs4_share_conflict(current_fh,
1983                                 NFS4_SHARE_DENY_WRITE);
1984         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
1985                 return nfs4_share_conflict(current_fh,
1986                                 NFS4_SHARE_DENY_READ);
1987 }
1988
1989 /*
1990  * Allow READ/WRITE during grace period on recovered state only for files
1991  * that are not able to provide mandatory locking.
1992  */
1993 static inline int
1994 io_during_grace_disallowed(struct inode *inode, int flags)
1995 {
1996         return nfs4_in_grace() && (flags & (RD_STATE | WR_STATE))
1997                 && mandatory_lock(inode);
1998 }
1999
2000 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
2001 {
2002         /* If the client sends us a stateid from the future, it's buggy: */
2003         if (in->si_generation > ref->si_generation)
2004                 return nfserr_bad_stateid;
2005         /*
2006          * The following, however, can happen.  For example, if the
2007          * client sends an open and some IO at the same time, the open
2008          * may bump si_generation while the IO is still in flight.
2009          * Thanks to hard links and renames, the client never knows what
2010          * file an open will affect.  So it could avoid that situation
2011          * only by serializing all opens and IO from the same open
2012          * owner.  To recover from the old_stateid error, the client
2013          * will just have to retry the IO:
2014          */
2015         if (in->si_generation < ref->si_generation)
2016                 return nfserr_old_stateid;
2017         return nfs_ok;
2018 }
2019
2020 /*
2021 * Checks for stateid operations
2022 */
2023 __be32
2024 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2025 {
2026         struct nfs4_stateid *stp = NULL;
2027         struct nfs4_delegation *dp = NULL;
2028         stateid_t *stidp;
2029         struct inode *ino = current_fh->fh_dentry->d_inode;
2030         __be32 status;
2031
2032         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2033                 stateid->si_boot, stateid->si_stateownerid,
2034                 stateid->si_fileid, stateid->si_generation);
2035         if (filpp)
2036                 *filpp = NULL;
2037
2038         if (io_during_grace_disallowed(ino, flags))
2039                 return nfserr_grace;
2040
2041         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2042                 return check_special_stateids(current_fh, stateid, flags);
2043
2044         /* STALE STATEID */
2045         status = nfserr_stale_stateid;
2046         if (STALE_STATEID(stateid))
2047                 goto out;
2048
2049         /* BAD STATEID */
2050         status = nfserr_bad_stateid;
2051         if (!stateid->si_fileid) { /* delegation stateid */
2052                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2053                         dprintk("NFSD: delegation stateid not found\n");
2054                         goto out;
2055                 }
2056                 stidp = &dp->dl_stateid;
2057         } else { /* open or lock stateid */
2058                 if (!(stp = find_stateid(stateid, flags))) {
2059                         dprintk("NFSD: open or lock stateid not found\n");
2060                         goto out;
2061                 }
2062                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2063                         goto out;
2064                 if (!stp->st_stateowner->so_confirmed)
2065                         goto out;
2066                 stidp = &stp->st_stateid;
2067         }
2068         status = check_stateid_generation(stateid, stidp);
2069         if (status)
2070                 goto out;
2071         if (stp) {
2072                 if ((status = nfs4_check_openmode(stp,flags)))
2073                         goto out;
2074                 renew_client(stp->st_stateowner->so_client);
2075                 if (filpp)
2076                         *filpp = stp->st_vfs_file;
2077         } else {
2078                 if ((status = nfs4_check_delegmode(dp, flags)))
2079                         goto out;
2080                 renew_client(dp->dl_client);
2081                 if (flags & DELEG_RET)
2082                         unhash_delegation(dp);
2083                 if (filpp)
2084                         *filpp = dp->dl_vfs_file;
2085         }
2086         status = nfs_ok;
2087 out:
2088         return status;
2089 }
2090
2091 static inline int
2092 setlkflg (int type)
2093 {
2094         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2095                 RD_STATE : WR_STATE;
2096 }
2097
2098 /*
2099  * Checks for sequence id mutating operations.
2100  */
2101 static __be32
2102 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)
2103 {
2104         struct nfs4_stateid *stp;
2105         struct nfs4_stateowner *sop;
2106         __be32 status;
2107
2108         dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2109                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2110                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2111                 stateid->si_generation);
2112
2113         *stpp = NULL;
2114         *sopp = NULL;
2115
2116         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2117                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2118                 return nfserr_bad_stateid;
2119         }
2120
2121         if (STALE_STATEID(stateid))
2122                 return nfserr_stale_stateid;
2123         /*
2124         * We return BAD_STATEID if filehandle doesn't match stateid,
2125         * the confirmed flag is incorrecly set, or the generation
2126         * number is incorrect.
2127         */
2128         stp = find_stateid(stateid, flags);
2129         if (stp == NULL) {
2130                 /*
2131                  * Also, we should make sure this isn't just the result of
2132                  * a replayed close:
2133                  */
2134                 sop = search_close_lru(stateid->si_stateownerid, flags);
2135                 if (sop == NULL)
2136                         return nfserr_bad_stateid;
2137                 *sopp = sop;
2138                 goto check_replay;
2139         }
2140
2141         *stpp = stp;
2142         *sopp = sop = stp->st_stateowner;
2143
2144         if (lock) {
2145                 clientid_t *lockclid = &lock->v.new.clientid;
2146                 struct nfs4_client *clp = sop->so_client;
2147                 int lkflg = 0;
2148                 __be32 status;
2149
2150                 lkflg = setlkflg(lock->lk_type);
2151
2152                 if (lock->lk_is_new) {
2153                         if (!sop->so_is_open_owner)
2154                                 return nfserr_bad_stateid;
2155                         if (!same_clid(&clp->cl_clientid, lockclid))
2156                                return nfserr_bad_stateid;
2157                         /* stp is the open stateid */
2158                         status = nfs4_check_openmode(stp, lkflg);
2159                         if (status)
2160                                 return status;
2161                 } else {
2162                         /* stp is the lock stateid */
2163                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2164                         if (status)
2165                                 return status;
2166                }
2167         }
2168
2169         if (nfs4_check_fh(current_fh, stp)) {
2170                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2171                 return nfserr_bad_stateid;
2172         }
2173
2174         /*
2175         *  We now validate the seqid and stateid generation numbers.
2176         *  For the moment, we ignore the possibility of
2177         *  generation number wraparound.
2178         */
2179         if (seqid != sop->so_seqid)
2180                 goto check_replay;
2181
2182         if (sop->so_confirmed && flags & CONFIRM) {
2183                 dprintk("NFSD: preprocess_seqid_op: expected"
2184                                 " unconfirmed stateowner!\n");
2185                 return nfserr_bad_stateid;
2186         }
2187         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2188                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2189                                 " confirmed yet!\n");
2190                 return nfserr_bad_stateid;
2191         }
2192         status = check_stateid_generation(stateid, &stp->st_stateid);
2193         if (status)
2194                 return status;
2195         renew_client(sop->so_client);
2196         return nfs_ok;
2197
2198 check_replay:
2199         if (seqid == sop->so_seqid - 1) {
2200                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2201                 /* indicate replay to calling function */
2202                 return nfserr_replay_me;
2203         }
2204         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2205                         sop->so_seqid, seqid);
2206         *sopp = NULL;
2207         return nfserr_bad_seqid;
2208 }
2209
2210 __be32
2211 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2212                    struct nfsd4_open_confirm *oc)
2213 {
2214         __be32 status;
2215         struct nfs4_stateowner *sop;
2216         struct nfs4_stateid *stp;
2217
2218         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2219                         (int)cstate->current_fh.fh_dentry->d_name.len,
2220                         cstate->current_fh.fh_dentry->d_name.name);
2221
2222         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2223         if (status)
2224                 return status;
2225
2226         nfs4_lock_state();
2227
2228         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2229                                         oc->oc_seqid, &oc->oc_req_stateid,
2230                                         CONFIRM | OPEN_STATE,
2231                                         &oc->oc_stateowner, &stp, NULL)))
2232                 goto out;
2233
2234         sop = oc->oc_stateowner;
2235         sop->so_confirmed = 1;
2236         update_stateid(&stp->st_stateid);
2237         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2238         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2239                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2240                          stp->st_stateid.si_boot,
2241                          stp->st_stateid.si_stateownerid,
2242                          stp->st_stateid.si_fileid,
2243                          stp->st_stateid.si_generation);
2244
2245         nfsd4_create_clid_dir(sop->so_client);
2246 out:
2247         if (oc->oc_stateowner) {
2248                 nfs4_get_stateowner(oc->oc_stateowner);
2249                 cstate->replay_owner = oc->oc_stateowner;
2250         }
2251         nfs4_unlock_state();
2252         return status;
2253 }
2254
2255
2256 /*
2257  * unset all bits in union bitmap (bmap) that
2258  * do not exist in share (from successful OPEN_DOWNGRADE)
2259  */
2260 static void
2261 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2262 {
2263         int i;
2264         for (i = 1; i < 4; i++) {
2265                 if ((i & access) != i)
2266                         __clear_bit(i, bmap);
2267         }
2268 }
2269
2270 static void
2271 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2272 {
2273         int i;
2274         for (i = 0; i < 4; i++) {
2275                 if ((i & deny) != i)
2276                         __clear_bit(i, bmap);
2277         }
2278 }
2279
2280 __be32
2281 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2282                      struct nfsd4_compound_state *cstate,
2283                      struct nfsd4_open_downgrade *od)
2284 {
2285         __be32 status;
2286         struct nfs4_stateid *stp;
2287         unsigned int share_access;
2288
2289         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2290                         (int)cstate->current_fh.fh_dentry->d_name.len,
2291                         cstate->current_fh.fh_dentry->d_name.name);
2292
2293         if (!access_valid(od->od_share_access)
2294                         || !deny_valid(od->od_share_deny))
2295                 return nfserr_inval;
2296
2297         nfs4_lock_state();
2298         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2299                                         od->od_seqid,
2300                                         &od->od_stateid,
2301                                         OPEN_STATE,
2302                                         &od->od_stateowner, &stp, NULL)))
2303                 goto out;
2304
2305         status = nfserr_inval;
2306         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2307                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2308                         stp->st_access_bmap, od->od_share_access);
2309                 goto out;
2310         }
2311         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2312                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2313                         stp->st_deny_bmap, od->od_share_deny);
2314                 goto out;
2315         }
2316         set_access(&share_access, stp->st_access_bmap);
2317         nfs4_file_downgrade(stp->st_vfs_file,
2318                             share_access & ~od->od_share_access);
2319
2320         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2321         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2322
2323         update_stateid(&stp->st_stateid);
2324         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2325         status = nfs_ok;
2326 out:
2327         if (od->od_stateowner) {
2328                 nfs4_get_stateowner(od->od_stateowner);
2329                 cstate->replay_owner = od->od_stateowner;
2330         }
2331         nfs4_unlock_state();
2332         return status;
2333 }
2334
2335 /*
2336  * nfs4_unlock_state() called after encode
2337  */
2338 __be32
2339 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2340             struct nfsd4_close *close)
2341 {
2342         __be32 status;
2343         struct nfs4_stateid *stp;
2344
2345         dprintk("NFSD: nfsd4_close on file %.*s\n",
2346                         (int)cstate->current_fh.fh_dentry->d_name.len,
2347                         cstate->current_fh.fh_dentry->d_name.name);
2348
2349         nfs4_lock_state();
2350         /* check close_lru for replay */
2351         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2352                                         close->cl_seqid,
2353                                         &close->cl_stateid,
2354                                         OPEN_STATE | CLOSE_STATE,
2355                                         &close->cl_stateowner, &stp, NULL)))
2356                 goto out;
2357         status = nfs_ok;
2358         update_stateid(&stp->st_stateid);
2359         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2360
2361         /* release_stateid() calls nfsd_close() if needed */
2362         release_stateid(stp, OPEN_STATE);
2363
2364         /* place unused nfs4_stateowners on so_close_lru list to be
2365          * released by the laundromat service after the lease period
2366          * to enable us to handle CLOSE replay
2367          */
2368         if (list_empty(&close->cl_stateowner->so_stateids))
2369                 move_to_close_lru(close->cl_stateowner);
2370 out:
2371         if (close->cl_stateowner) {
2372                 nfs4_get_stateowner(close->cl_stateowner);
2373                 cstate->replay_owner = close->cl_stateowner;
2374         }
2375         nfs4_unlock_state();
2376         return status;
2377 }
2378
2379 __be32
2380 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2381                   struct nfsd4_delegreturn *dr)
2382 {
2383         __be32 status;
2384
2385         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2386                 goto out;
2387
2388         nfs4_lock_state();
2389         status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2390                                             &dr->dr_stateid, DELEG_RET, NULL);
2391         nfs4_unlock_state();
2392 out:
2393         return status;
2394 }
2395
2396
2397 /*
2398  * Lock owner state (byte-range locks)
2399  */
2400 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2401 #define LOCK_HASH_BITS              8
2402 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2403 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2404
2405 #define lockownerid_hashval(id) \
2406         ((id) & LOCK_HASH_MASK)
2407
2408 static inline unsigned int
2409 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2410                 struct xdr_netobj *ownername)
2411 {
2412         return (file_hashval(inode) + cl_id
2413                         + opaque_hashval(ownername->data, ownername->len))
2414                 & LOCK_HASH_MASK;
2415 }
2416
2417 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2418 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2419 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2420
2421 static struct nfs4_stateid *
2422 find_stateid(stateid_t *stid, int flags)
2423 {
2424         struct nfs4_stateid *local = NULL;
2425         u32 st_id = stid->si_stateownerid;
2426         u32 f_id = stid->si_fileid;
2427         unsigned int hashval;
2428
2429         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2430         if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2431                 hashval = stateid_hashval(st_id, f_id);
2432                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2433                         if ((local->st_stateid.si_stateownerid == st_id) &&
2434                             (local->st_stateid.si_fileid == f_id))
2435                                 return local;
2436                 }
2437         }
2438         if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2439                 hashval = stateid_hashval(st_id, f_id);
2440                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2441                         if ((local->st_stateid.si_stateownerid == st_id) &&
2442                             (local->st_stateid.si_fileid == f_id))
2443                                 return local;
2444                 }
2445         }
2446         return NULL;
2447 }
2448
2449 static struct nfs4_delegation *
2450 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2451 {
2452         struct nfs4_file *fp;
2453         struct nfs4_delegation *dl;
2454
2455         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2456                     stid->si_boot, stid->si_stateownerid,
2457                     stid->si_fileid, stid->si_generation);
2458
2459         fp = find_file(ino);
2460         if (!fp)
2461                 return NULL;
2462         dl = find_delegation_file(fp, stid);
2463         put_nfs4_file(fp);
2464         return dl;
2465 }
2466
2467 /*
2468  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2469  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2470  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2471  * locking, this prevents us from being completely protocol-compliant.  The
2472  * real solution to this problem is to start using unsigned file offsets in
2473  * the VFS, but this is a very deep change!
2474  */
2475 static inline void
2476 nfs4_transform_lock_offset(struct file_lock *lock)
2477 {
2478         if (lock->fl_start < 0)
2479                 lock->fl_start = OFFSET_MAX;
2480         if (lock->fl_end < 0)
2481                 lock->fl_end = OFFSET_MAX;
2482 }
2483
2484 /* Hack!: For now, we're defining this just so we can use a pointer to it
2485  * as a unique cookie to identify our (NFSv4's) posix locks. */
2486 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2487 };
2488
2489 static inline void
2490 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2491 {
2492         struct nfs4_stateowner *sop;
2493         unsigned int hval;
2494
2495         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2496                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2497                 hval = lockownerid_hashval(sop->so_id);
2498                 kref_get(&sop->so_ref);
2499                 deny->ld_sop = sop;
2500                 deny->ld_clientid = sop->so_client->cl_clientid;
2501         } else {
2502                 deny->ld_sop = NULL;
2503                 deny->ld_clientid.cl_boot = 0;
2504                 deny->ld_clientid.cl_id = 0;
2505         }
2506         deny->ld_start = fl->fl_start;
2507         deny->ld_length = ~(u64)0;
2508         if (fl->fl_end != ~(u64)0)
2509                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2510         deny->ld_type = NFS4_READ_LT;
2511         if (fl->fl_type != F_RDLCK)
2512                 deny->ld_type = NFS4_WRITE_LT;
2513 }
2514
2515 static struct nfs4_stateowner *
2516 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2517                 struct xdr_netobj *owner)
2518 {
2519         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2520         struct nfs4_stateowner *op;
2521
2522         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2523                 if (same_owner_str(op, owner, clid))
2524                         return op;
2525         }
2526         return NULL;
2527 }
2528
2529 /*
2530  * Alloc a lock owner structure.
2531  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2532  * occured.
2533  *
2534  * strhashval = lock_ownerstr_hashval
2535  */
2536
2537 static struct nfs4_stateowner *
2538 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2539         struct nfs4_stateowner *sop;
2540         struct nfs4_replay *rp;
2541         unsigned int idhashval;
2542
2543         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2544                 return NULL;
2545         idhashval = lockownerid_hashval(current_ownerid);
2546         INIT_LIST_HEAD(&sop->so_idhash);
2547         INIT_LIST_HEAD(&sop->so_strhash);
2548         INIT_LIST_HEAD(&sop->so_perclient);
2549         INIT_LIST_HEAD(&sop->so_stateids);
2550         INIT_LIST_HEAD(&sop->so_perstateid);
2551         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2552         sop->so_time = 0;
2553         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2554         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2555         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2556         sop->so_is_open_owner = 0;
2557         sop->so_id = current_ownerid++;
2558         sop->so_client = clp;
2559         /* It is the openowner seqid that will be incremented in encode in the
2560          * case of new lockowners; so increment the lock seqid manually: */
2561         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2562         sop->so_confirmed = 1;
2563         rp = &sop->so_replay;
2564         rp->rp_status = nfserr_serverfault;
2565         rp->rp_buflen = 0;
2566         rp->rp_buf = rp->rp_ibuf;
2567         return sop;
2568 }
2569
2570 static struct nfs4_stateid *
2571 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2572 {
2573         struct nfs4_stateid *stp;
2574         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2575
2576         stp = nfs4_alloc_stateid();
2577         if (stp == NULL)
2578                 goto out;
2579         INIT_LIST_HEAD(&stp->st_hash);
2580         INIT_LIST_HEAD(&stp->st_perfile);
2581         INIT_LIST_HEAD(&stp->st_perstateowner);
2582         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2583         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2584         list_add(&stp->st_perfile, &fp->fi_stateids);
2585         list_add(&stp->st_perstateowner, &sop->so_stateids);
2586         stp->st_stateowner = sop;
2587         get_nfs4_file(fp);
2588         stp->st_file = fp;
2589         stp->st_stateid.si_boot = boot_time;
2590         stp->st_stateid.si_stateownerid = sop->so_id;
2591         stp->st_stateid.si_fileid = fp->fi_id;
2592         stp->st_stateid.si_generation = 0;
2593         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2594         stp->st_access_bmap = open_stp->st_access_bmap;
2595         stp->st_deny_bmap = open_stp->st_deny_bmap;
2596         stp->st_openstp = open_stp;
2597
2598 out:
2599         return stp;
2600 }
2601
2602 static int
2603 check_lock_length(u64 offset, u64 length)
2604 {
2605         return ((length == 0)  || ((length != ~(u64)0) &&
2606              LOFF_OVERFLOW(offset, length)));
2607 }
2608
2609 /*
2610  *  LOCK operation
2611  */
2612 __be32
2613 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2614            struct nfsd4_lock *lock)
2615 {
2616         struct nfs4_stateowner *open_sop = NULL;
2617         struct nfs4_stateowner *lock_sop = NULL;
2618         struct nfs4_stateid *lock_stp;
2619         struct file *filp;
2620         struct file_lock file_lock;
2621         struct file_lock conflock;
2622         __be32 status = 0;
2623         unsigned int strhashval;
2624         unsigned int cmd;
2625         int err;
2626
2627         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2628                 (long long) lock->lk_offset,
2629                 (long long) lock->lk_length);
2630
2631         if (check_lock_length(lock->lk_offset, lock->lk_length))
2632                  return nfserr_inval;
2633
2634         if ((status = fh_verify(rqstp, &cstate->current_fh,
2635                                 S_IFREG, NFSD_MAY_LOCK))) {
2636                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2637                 return status;
2638         }
2639
2640         nfs4_lock_state();
2641
2642         if (lock->lk_is_new) {
2643                 /*
2644                  * Client indicates that this is a new lockowner.
2645                  * Use open owner and open stateid to create lock owner and
2646                  * lock stateid.
2647                  */
2648                 struct nfs4_stateid *open_stp = NULL;
2649                 struct nfs4_file *fp;
2650
2651                 status = nfserr_stale_clientid;
2652                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2653                         goto out;
2654
2655                 /* validate and update open stateid and open seqid */
2656                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2657                                         lock->lk_new_open_seqid,
2658                                         &lock->lk_new_open_stateid,
2659                                         OPEN_STATE,
2660                                         &lock->lk_replay_owner, &open_stp,
2661                                         lock);
2662                 if (status)
2663                         goto out;
2664                 open_sop = lock->lk_replay_owner;
2665                 /* create lockowner and lock stateid */
2666                 fp = open_stp->st_file;
2667                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
2668                                 open_sop->so_client->cl_clientid.cl_id,
2669                                 &lock->v.new.owner);
2670                 /* XXX: Do we need to check for duplicate stateowners on
2671                  * the same file, or should they just be allowed (and
2672                  * create new stateids)? */
2673                 status = nfserr_resource;
2674                 lock_sop = alloc_init_lock_stateowner(strhashval,
2675                                 open_sop->so_client, open_stp, lock);
2676                 if (lock_sop == NULL)
2677                         goto out;
2678                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2679                 if (lock_stp == NULL)
2680                         goto out;
2681         } else {
2682                 /* lock (lock owner + lock stateid) already exists */
2683                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2684                                        lock->lk_old_lock_seqid,
2685                                        &lock->lk_old_lock_stateid,
2686                                        LOCK_STATE,
2687                                        &lock->lk_replay_owner, &lock_stp, lock);
2688                 if (status)
2689                         goto out;
2690                 lock_sop = lock->lk_replay_owner;
2691         }
2692         /* lock->lk_replay_owner and lock_stp have been created or found */
2693         filp = lock_stp->st_vfs_file;
2694
2695         status = nfserr_grace;
2696         if (nfs4_in_grace() && !lock->lk_reclaim)
2697                 goto out;
2698         status = nfserr_no_grace;
2699         if (!nfs4_in_grace() && lock->lk_reclaim)
2700                 goto out;
2701
2702         locks_init_lock(&file_lock);
2703         switch (lock->lk_type) {
2704                 case NFS4_READ_LT:
2705                 case NFS4_READW_LT:
2706                         file_lock.fl_type = F_RDLCK;
2707                         cmd = F_SETLK;
2708                 break;
2709                 case NFS4_WRITE_LT:
2710                 case NFS4_WRITEW_LT:
2711                         file_lock.fl_type = F_WRLCK;
2712                         cmd = F_SETLK;
2713                 break;
2714                 default:
2715                         status = nfserr_inval;
2716                 goto out;
2717         }
2718         file_lock.fl_owner = (fl_owner_t)lock_sop;
2719         file_lock.fl_pid = current->tgid;
2720         file_lock.fl_file = filp;
2721         file_lock.fl_flags = FL_POSIX;
2722         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2723
2724         file_lock.fl_start = lock->lk_offset;
2725         if ((lock->lk_length == ~(u64)0) ||
2726                         LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2727                 file_lock.fl_end = ~(u64)0;
2728         else
2729                 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2730         nfs4_transform_lock_offset(&file_lock);
2731
2732         /*
2733         * Try to lock the file in the VFS.
2734         * Note: locks.c uses the BKL to protect the inode's lock list.
2735         */
2736
2737         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2738         switch (-err) {
2739         case 0: /* success! */
2740                 update_stateid(&lock_stp->st_stateid);
2741                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
2742                                 sizeof(stateid_t));
2743                 status = 0;
2744                 break;
2745         case (EAGAIN):          /* conflock holds conflicting lock */
2746                 status = nfserr_denied;
2747                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2748                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2749                 break;
2750         case (EDEADLK):
2751                 status = nfserr_deadlock;
2752                 break;
2753         default:
2754                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2755                 status = nfserr_resource;
2756                 break;
2757         }
2758 out:
2759         if (status && lock->lk_is_new && lock_sop)
2760                 release_stateowner(lock_sop);
2761         if (lock->lk_replay_owner) {
2762                 nfs4_get_stateowner(lock->lk_replay_owner);
2763                 cstate->replay_owner = lock->lk_replay_owner;
2764         }
2765         nfs4_unlock_state();
2766         return status;
2767 }
2768
2769 /*
2770  * LOCKT operation
2771  */
2772 __be32
2773 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2774             struct nfsd4_lockt *lockt)
2775 {
2776         struct inode *inode;
2777         struct file file;
2778         struct file_lock file_lock;
2779         int error;
2780         __be32 status;
2781
2782         if (nfs4_in_grace())
2783                 return nfserr_grace;
2784
2785         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2786                  return nfserr_inval;
2787
2788         lockt->lt_stateowner = NULL;
2789         nfs4_lock_state();
2790
2791         status = nfserr_stale_clientid;
2792         if (STALE_CLIENTID(&lockt->lt_clientid))
2793                 goto out;
2794
2795         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2796                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2797                 if (status == nfserr_symlink)
2798                         status = nfserr_inval;
2799                 goto out;
2800         }
2801
2802         inode = cstate->current_fh.fh_dentry->d_inode;
2803         locks_init_lock(&file_lock);
2804         switch (lockt->lt_type) {
2805                 case NFS4_READ_LT:
2806                 case NFS4_READW_LT:
2807                         file_lock.fl_type = F_RDLCK;
2808                 break;
2809                 case NFS4_WRITE_LT:
2810                 case NFS4_WRITEW_LT:
2811                         file_lock.fl_type = F_WRLCK;
2812                 break;
2813                 default:
2814                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2815                         status = nfserr_inval;
2816                 goto out;
2817         }
2818
2819         lockt->lt_stateowner = find_lockstateowner_str(inode,
2820                         &lockt->lt_clientid, &lockt->lt_owner);
2821         if (lockt->lt_stateowner)
2822                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2823         file_lock.fl_pid = current->tgid;
2824         file_lock.fl_flags = FL_POSIX;
2825         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2826
2827         file_lock.fl_start = lockt->lt_offset;
2828         if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2829                 file_lock.fl_end = ~(u64)0;
2830         else
2831                 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2832
2833         nfs4_transform_lock_offset(&file_lock);
2834
2835         /* vfs_test_lock uses the struct file _only_ to resolve the inode.
2836          * since LOCKT doesn't require an OPEN, and therefore a struct
2837          * file may not exist, pass vfs_test_lock a struct file with
2838          * only the dentry:inode set.
2839          */
2840         memset(&file, 0, sizeof (struct file));
2841         file.f_path.dentry = cstate->current_fh.fh_dentry;
2842
2843         status = nfs_ok;
2844         error = vfs_test_lock(&file, &file_lock);
2845         if (error) {
2846                 status = nfserrno(error);
2847                 goto out;
2848         }
2849         if (file_lock.fl_type != F_UNLCK) {
2850                 status = nfserr_denied;
2851                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2852         }
2853 out:
2854         nfs4_unlock_state();
2855         return status;
2856 }
2857
2858 __be32
2859 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2860             struct nfsd4_locku *locku)
2861 {
2862         struct nfs4_stateid *stp;
2863         struct file *filp = NULL;
2864         struct file_lock file_lock;
2865         __be32 status;
2866         int err;
2867
2868         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2869                 (long long) locku->lu_offset,
2870                 (long long) locku->lu_length);
2871
2872         if (check_lock_length(locku->lu_offset, locku->lu_length))
2873                  return nfserr_inval;
2874
2875         nfs4_lock_state();
2876
2877         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2878                                         locku->lu_seqid,
2879                                         &locku->lu_stateid,
2880                                         LOCK_STATE,
2881                                         &locku->lu_stateowner, &stp, NULL)))
2882                 goto out;
2883
2884         filp = stp->st_vfs_file;
2885         BUG_ON(!filp);
2886         locks_init_lock(&file_lock);
2887         file_lock.fl_type = F_UNLCK;
2888         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2889         file_lock.fl_pid = current->tgid;
2890         file_lock.fl_file = filp;
2891         file_lock.fl_flags = FL_POSIX;
2892         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2893         file_lock.fl_start = locku->lu_offset;
2894
2895         if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2896                 file_lock.fl_end = ~(u64)0;
2897         else
2898                 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2899         nfs4_transform_lock_offset(&file_lock);
2900
2901         /*
2902         *  Try to unlock the file in the VFS.
2903         */
2904         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2905         if (err) {
2906                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2907                 goto out_nfserr;
2908         }
2909         /*
2910         * OK, unlock succeeded; the only thing left to do is update the stateid.
2911         */
2912         update_stateid(&stp->st_stateid);
2913         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2914
2915 out:
2916         if (locku->lu_stateowner) {
2917                 nfs4_get_stateowner(locku->lu_stateowner);
2918                 cstate->replay_owner = locku->lu_stateowner;
2919         }
2920         nfs4_unlock_state();
2921         return status;
2922
2923 out_nfserr:
2924         status = nfserrno(err);
2925         goto out;
2926 }
2927
2928 /*
2929  * returns
2930  *      1: locks held by lockowner
2931  *      0: no locks held by lockowner
2932  */
2933 static int
2934 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2935 {
2936         struct file_lock **flpp;
2937         struct inode *inode = filp->f_path.dentry->d_inode;
2938         int status = 0;
2939
2940         lock_kernel();
2941         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2942                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2943                         status = 1;
2944                         goto out;
2945                 }
2946         }
2947 out:
2948         unlock_kernel();
2949         return status;
2950 }
2951
2952 __be32
2953 nfsd4_release_lockowner(struct svc_rqst *rqstp,
2954                         struct nfsd4_compound_state *cstate,
2955                         struct nfsd4_release_lockowner *rlockowner)
2956 {
2957         clientid_t *clid = &rlockowner->rl_clientid;
2958         struct nfs4_stateowner *sop;
2959         struct nfs4_stateid *stp;
2960         struct xdr_netobj *owner = &rlockowner->rl_owner;
2961         struct list_head matches;
2962         int i;
2963         __be32 status;
2964
2965         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
2966                 clid->cl_boot, clid->cl_id);
2967
2968         /* XXX check for lease expiration */
2969
2970         status = nfserr_stale_clientid;
2971         if (STALE_CLIENTID(clid))
2972                 return status;
2973
2974         nfs4_lock_state();
2975
2976         status = nfserr_locks_held;
2977         /* XXX: we're doing a linear search through all the lockowners.
2978          * Yipes!  For now we'll just hope clients aren't really using
2979          * release_lockowner much, but eventually we have to fix these
2980          * data structures. */
2981         INIT_LIST_HEAD(&matches);
2982         for (i = 0; i < LOCK_HASH_SIZE; i++) {
2983                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
2984                         if (!same_owner_str(sop, owner, clid))
2985                                 continue;
2986                         list_for_each_entry(stp, &sop->so_stateids,
2987                                         st_perstateowner) {
2988                                 if (check_for_locks(stp->st_vfs_file, sop))
2989                                         goto out;
2990                                 /* Note: so_perclient unused for lockowners,
2991                                  * so it's OK to fool with here. */
2992                                 list_add(&sop->so_perclient, &matches);
2993                         }
2994                 }
2995         }
2996         /* Clients probably won't expect us to return with some (but not all)
2997          * of the lockowner state released; so don't release any until all
2998          * have been checked. */
2999         status = nfs_ok;
3000         while (!list_empty(&matches)) {
3001                 sop = list_entry(matches.next, struct nfs4_stateowner,
3002                                                                 so_perclient);
3003                 /* unhash_stateowner deletes so_perclient only
3004                  * for openowners. */
3005                 list_del(&sop->so_perclient);
3006                 release_stateowner(sop);
3007         }
3008 out:
3009         nfs4_unlock_state();
3010         return status;
3011 }
3012
3013 static inline struct nfs4_client_reclaim *
3014 alloc_reclaim(void)
3015 {
3016         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3017 }
3018
3019 int
3020 nfs4_has_reclaimed_state(const char *name)
3021 {
3022         unsigned int strhashval = clientstr_hashval(name);
3023         struct nfs4_client *clp;
3024
3025         clp = find_confirmed_client_by_str(name, strhashval);
3026         return clp ? 1 : 0;
3027 }
3028
3029 /*
3030  * failure => all reset bets are off, nfserr_no_grace...
3031  */
3032 int
3033 nfs4_client_to_reclaim(const char *name)
3034 {
3035         unsigned int strhashval;
3036         struct nfs4_client_reclaim *crp = NULL;
3037
3038         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3039         crp = alloc_reclaim();
3040         if (!crp)
3041                 return 0;
3042         strhashval = clientstr_hashval(name);
3043         INIT_LIST_HEAD(&crp->cr_strhash);
3044         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3045         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3046         reclaim_str_hashtbl_size++;
3047         return 1;
3048 }
3049
3050 static void
3051 nfs4_release_reclaim(void)
3052 {
3053         struct nfs4_client_reclaim *crp = NULL;
3054         int i;
3055
3056         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3057                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3058                         crp = list_entry(reclaim_str_hashtbl[i].next,
3059                                         struct nfs4_client_reclaim, cr_strhash);
3060                         list_del(&crp->cr_strhash);
3061                         kfree(crp);
3062                         reclaim_str_hashtbl_size--;
3063                 }
3064         }
3065         BUG_ON(reclaim_str_hashtbl_size);
3066 }
3067
3068 /*
3069  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3070 static struct nfs4_client_reclaim *
3071 nfs4_find_reclaim_client(clientid_t *clid)
3072 {
3073         unsigned int strhashval;
3074         struct nfs4_client *clp;
3075         struct nfs4_client_reclaim *crp = NULL;
3076
3077
3078         /* find clientid in conf_id_hashtbl */
3079         clp = find_confirmed_client(clid);
3080         if (clp == NULL)
3081                 return NULL;
3082
3083         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3084                             clp->cl_name.len, clp->cl_name.data,
3085                             clp->cl_recdir);
3086
3087         /* find clp->cl_name in reclaim_str_hashtbl */
3088         strhashval = clientstr_hashval(clp->cl_recdir);
3089         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3090                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3091                         return crp;
3092                 }
3093         }
3094         return NULL;
3095 }
3096
3097 /*
3098 * Called from OPEN. Look for clientid in reclaim list.
3099 */
3100 __be32
3101 nfs4_check_open_reclaim(clientid_t *clid)
3102 {
3103         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3104 }
3105
3106 /* initialization to perform at module load time: */
3107
3108 int
3109 nfs4_state_init(void)
3110 {
3111         int i, status;
3112
3113         status = nfsd4_init_slabs();
3114         if (status)
3115                 return status;
3116         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3117                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3118                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3119                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3120                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3121         }
3122         for (i = 0; i < FILE_HASH_SIZE; i++) {
3123                 INIT_LIST_HEAD(&file_hashtbl[i]);
3124         }
3125         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3126                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3127                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3128         }
3129         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3130                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3131                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3132         }
3133         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3134                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3135                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3136         }
3137         memset(&onestateid, ~0, sizeof(stateid_t));
3138         INIT_LIST_HEAD(&close_lru);
3139         INIT_LIST_HEAD(&client_lru);
3140         INIT_LIST_HEAD(&del_recall_lru);
3141         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3142                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3143         reclaim_str_hashtbl_size = 0;
3144         return 0;
3145 }
3146
3147 static void
3148 nfsd4_load_reboot_recovery_data(void)
3149 {
3150         int status;
3151
3152         nfs4_lock_state();
3153         nfsd4_init_recdir(user_recovery_dirname);
3154         status = nfsd4_recdir_load();
3155         nfs4_unlock_state();
3156         if (status)
3157                 printk("NFSD: Failure reading reboot recovery data\n");
3158 }
3159
3160 unsigned long
3161 get_nfs4_grace_period(void)
3162 {
3163         return max(user_lease_time, lease_time) * HZ;
3164 }
3165
3166 /*
3167  * Since the lifetime of a delegation isn't limited to that of an open, a
3168  * client may quite reasonably hang on to a delegation as long as it has
3169  * the inode cached.  This becomes an obvious problem the first time a
3170  * client's inode cache approaches the size of the server's total memory.
3171  *
3172  * For now we avoid this problem by imposing a hard limit on the number
3173  * of delegations, which varies according to the server's memory size.
3174  */
3175 static void
3176 set_max_delegations(void)
3177 {
3178         /*
3179          * Allow at most 4 delegations per megabyte of RAM.  Quick
3180          * estimates suggest that in the worst case (where every delegation
3181          * is for a different inode), a delegation could take about 1.5K,
3182          * giving a worst case usage of about 6% of memory.
3183          */
3184         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3185 }
3186
3187 /* initialization to perform when the nfsd service is started: */
3188
3189 static void
3190 __nfs4_state_start(void)
3191 {
3192         unsigned long grace_time;
3193
3194         boot_time = get_seconds();
3195         grace_time = get_nfs_grace_period();
3196         lease_time = user_lease_time;
3197         in_grace = 1;
3198         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3199                grace_time/HZ);
3200         laundry_wq = create_singlethread_workqueue("nfsd4");
3201         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3202         set_max_delegations();
3203 }
3204
3205 void
3206 nfs4_state_start(void)
3207 {
3208         if (nfs4_init)
3209                 return;
3210         nfsd4_load_reboot_recovery_data();
3211         __nfs4_state_start();
3212         nfs4_init = 1;
3213         return;
3214 }
3215
3216 int
3217 nfs4_in_grace(void)
3218 {
3219         return in_grace;
3220 }
3221
3222 time_t
3223 nfs4_lease_time(void)
3224 {
3225         return lease_time;
3226 }
3227
3228 static void
3229 __nfs4_state_shutdown(void)
3230 {
3231         int i;
3232         struct nfs4_client *clp = NULL;
3233         struct nfs4_delegation *dp = NULL;
3234         struct list_head *pos, *next, reaplist;
3235
3236         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3237                 while (!list_empty(&conf_id_hashtbl[i])) {
3238                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3239                         expire_client(clp);
3240                 }
3241                 while (!list_empty(&unconf_str_hashtbl[i])) {
3242                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3243                         expire_client(clp);
3244                 }
3245         }
3246         INIT_LIST_HEAD(&reaplist);
3247         spin_lock(&recall_lock);
3248         list_for_each_safe(pos, next, &del_recall_lru) {
3249                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3250                 list_move(&dp->dl_recall_lru, &reaplist);
3251         }
3252         spin_unlock(&recall_lock);
3253         list_for_each_safe(pos, next, &reaplist) {
3254                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3255                 list_del_init(&dp->dl_recall_lru);
3256                 unhash_delegation(dp);
3257         }
3258
3259         nfsd4_shutdown_recdir();
3260         nfs4_init = 0;
3261 }
3262
3263 void
3264 nfs4_state_shutdown(void)
3265 {
3266         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3267         destroy_workqueue(laundry_wq);
3268         nfs4_lock_state();
3269         nfs4_release_reclaim();
3270         __nfs4_state_shutdown();
3271         nfs4_unlock_state();
3272 }
3273
3274 /*
3275  * user_recovery_dirname is protected by the nfsd_mutex since it's only
3276  * accessed when nfsd is starting.
3277  */
3278 static void
3279 nfs4_set_recdir(char *recdir)
3280 {
3281         strcpy(user_recovery_dirname, recdir);
3282 }
3283
3284 /*
3285  * Change the NFSv4 recovery directory to recdir.
3286  */
3287 int
3288 nfs4_reset_recoverydir(char *recdir)
3289 {
3290         int status;
3291         struct nameidata nd;
3292
3293         status = path_lookup(recdir, LOOKUP_FOLLOW, &nd);
3294         if (status)
3295                 return status;
3296         status = -ENOTDIR;
3297         if (S_ISDIR(nd.path.dentry->d_inode->i_mode)) {
3298                 nfs4_set_recdir(recdir);
3299                 status = 0;
3300         }
3301         path_put(&nd.path);
3302         return status;
3303 }
3304
3305 char *
3306 nfs4_recoverydir(void)
3307 {
3308         return user_recovery_dirname;
3309 }
3310
3311 /*
3312  * Called when leasetime is changed.
3313  *
3314  * The only way the protocol gives us to handle on-the-fly lease changes is to
3315  * simulate a reboot.  Instead of doing that, we just wait till the next time
3316  * we start to register any changes in lease time.  If the administrator
3317  * really wants to change the lease time *now*, they can go ahead and bring
3318  * nfsd down and then back up again after changing the lease time.
3319  *
3320  * user_lease_time is protected by nfsd_mutex since it's only really accessed
3321  * when nfsd is starting
3322  */
3323 void
3324 nfs4_reset_lease(time_t leasetime)
3325 {
3326         user_lease_time = leasetime;
3327 }