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 #include <linux/sunrpc/svcauth_gss.h>
  58
  59 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  60
  61 /* Globals */
  62 static time_t lease_time = 90;     /* default lease time */
  63 static time_t user_lease_time = 90;
  64 static time_t boot_time;
  65 static u32 current_ownerid = 1;
  66 static u32 current_fileid = 1;
  67 static u32 current_delegid = 1;
  68 static u32 nfs4_init;
  69 static stateid_t zerostateid;             /* bits all 0 */
  70 static stateid_t onestateid;              /* bits all 1 */
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
  74
  75 /* forward declarations */
  76 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
  77 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
  78 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  79 static void nfs4_set_recdir(char *recdir);
  80
  81 /* Locking:
  82  *
  83  * client_mutex:
  84  *      protects clientid_hashtbl[], clientstr_hashtbl[],
  85  *      unconfstr_hashtbl[], uncofid_hashtbl[].
  86  */
  87 static DEFINE_MUTEX(client_mutex);
  88
  89 static struct kmem_cache *stateowner_slab = NULL;
  90 static struct kmem_cache *file_slab = NULL;
  91 static struct kmem_cache *stateid_slab = NULL;
  92 static struct kmem_cache *deleg_slab = NULL;
  93
  94 void
  95 nfs4_lock_state(void)
  96 {
  97         mutex_lock(&client_mutex);
  98 }
  99
 100 void
 101 nfs4_unlock_state(void)
 102 {
 103         mutex_unlock(&client_mutex);
 104 }
 105
 106 static inline u32
 107 opaque_hashval(const void *ptr, int nbytes)
 108 {
 109         unsigned char *cptr = (unsigned char *) ptr;
 110
 111         u32 x = 0;
 112         while (nbytes--) {
 113                 x *= 37;
 114                 x += *cptr++;
 115         }
 116         return x;
 117 }
 118
 119 /*
 120  * Delegation state
 121  */
 122
 123 /* recall_lock protects the del_recall_lru */
 124 static DEFINE_SPINLOCK(recall_lock);
 125 static struct list_head del_recall_lru;
 126
 127 static void
 128 free_nfs4_file(struct kref *kref)
 129 {
 130         struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
 131         list_del(&fp->fi_hash);
 132         iput(fp->fi_inode);
 133         kmem_cache_free(file_slab, fp);
 134 }
 135
 136 static inline void
 137 put_nfs4_file(struct nfs4_file *fi)
 138 {
 139         kref_put(&fi->fi_ref, free_nfs4_file);
 140 }
 141
 142 static inline void
 143 get_nfs4_file(struct nfs4_file *fi)
 144 {
 145         kref_get(&fi->fi_ref);
 146 }
 147
 148 static int num_delegations;
 149 unsigned int max_delegations;
 150
 151 /*
 152  * Open owner state (share locks)
 153  */
 154
 155 /* hash tables for nfs4_stateowner */
 156 #define OWNER_HASH_BITS              8
 157 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 158 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 159
 160 #define ownerid_hashval(id) \
 161         ((id) & OWNER_HASH_MASK)
 162 #define ownerstr_hashval(clientid, ownername) \
 163         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
 164
 165 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
 166 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 167
 168 /* hash table for nfs4_file */
 169 #define FILE_HASH_BITS                   8
 170 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 171 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
 172 /* hash table for (open)nfs4_stateid */
 173 #define STATEID_HASH_BITS              10
 174 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
 175 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
 176
 177 #define file_hashval(x) \
 178         hash_ptr(x, FILE_HASH_BITS)
 179 #define stateid_hashval(owner_id, file_id)  \
 180         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
 181
 182 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 183 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 184
 185 static struct nfs4_delegation *
 186 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 187 {
 188         struct nfs4_delegation *dp;
 189         struct nfs4_file *fp = stp->st_file;
 190         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
 191
 192         dprintk("NFSD alloc_init_deleg\n");
 193         if (fp->fi_had_conflict)
 194                 return NULL;
 195         if (num_delegations > max_delegations)
 196                 return NULL;
 197         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
 198         if (dp == NULL)
 199                 return dp;
 200         num_delegations++;
 201         INIT_LIST_HEAD(&dp->dl_perfile);
 202         INIT_LIST_HEAD(&dp->dl_perclnt);
 203         INIT_LIST_HEAD(&dp->dl_recall_lru);
 204         dp->dl_client = clp;
 205         get_nfs4_file(fp);
 206         dp->dl_file = fp;
 207         dp->dl_flock = NULL;
 208         get_file(stp->st_vfs_file);
 209         dp->dl_vfs_file = stp->st_vfs_file;
 210         dp->dl_type = type;
 211         dp->dl_recall.cbr_dp = NULL;
 212         dp->dl_recall.cbr_ident = cb->cb_ident;
 213         dp->dl_recall.cbr_trunc = 0;
 214         dp->dl_stateid.si_boot = boot_time;
 215         dp->dl_stateid.si_stateownerid = current_delegid++;
 216         dp->dl_stateid.si_fileid = 0;
 217         dp->dl_stateid.si_generation = 0;
 218         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 219         dp->dl_time = 0;
 220         atomic_set(&dp->dl_count, 1);
 221         list_add(&dp->dl_perfile, &fp->fi_delegations);
 222         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 223         return dp;
 224 }
 225
 226 void
 227 nfs4_put_delegation(struct nfs4_delegation *dp)
 228 {
 229         if (atomic_dec_and_test(&dp->dl_count)) {
 230                 dprintk("NFSD: freeing dp %p\n",dp);
 231                 put_nfs4_file(dp->dl_file);
 232                 kmem_cache_free(deleg_slab, dp);
 233                 num_delegations--;
 234         }
 235 }
 236
 237 /* Remove the associated file_lock first, then remove the delegation.
 238  * lease_modify() is called to remove the FS_LEASE file_lock from
 239  * the i_flock list, eventually calling nfsd's lock_manager
 240  * fl_release_callback.
 241  */
 242 static void
 243 nfs4_close_delegation(struct nfs4_delegation *dp)
 244 {
 245         struct file *filp = dp->dl_vfs_file;
 246
 247         dprintk("NFSD: close_delegation dp %p\n",dp);
 248         dp->dl_vfs_file = NULL;
 249         /* The following nfsd_close may not actually close the file,
 250          * but we want to remove the lease in any case. */
 251         if (dp->dl_flock)
 252                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
 253         nfsd_close(filp);
 254 }
 255
 256 /* Called under the state lock. */
 257 static void
 258 unhash_delegation(struct nfs4_delegation *dp)
 259 {
 260         list_del_init(&dp->dl_perfile);
 261         list_del_init(&dp->dl_perclnt);
 262         spin_lock(&recall_lock);
 263         list_del_init(&dp->dl_recall_lru);
 264         spin_unlock(&recall_lock);
 265         nfs4_close_delegation(dp);
 266         nfs4_put_delegation(dp);
 267 }
 268
 269 /*
 270  * SETCLIENTID state
 271  */
 272
 273 /* Hash tables for nfs4_clientid state */
 274 #define CLIENT_HASH_BITS                 4
 275 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 276 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 277
 278 #define clientid_hashval(id) \
 279         ((id) & CLIENT_HASH_MASK)
 280 #define clientstr_hashval(name) \
 281         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
 282 /*
 283  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 284  * used in reboot/reset lease grace period processing
 285  *
 286  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 287  * setclientid_confirmed info.
 288  *
 289  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 290  * setclientid info.
 291  *
 292  * client_lru holds client queue ordered by nfs4_client.cl_time
 293  * for lease renewal.
 294  *
 295  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 296  * for last close replay.
 297  */
 298 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 299 static int reclaim_str_hashtbl_size = 0;
 300 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 301 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 302 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 303 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 304 static struct list_head client_lru;
 305 static struct list_head close_lru;
 306
 307 static void unhash_generic_stateid(struct nfs4_stateid *stp)
 308 {
 309         list_del(&stp->st_hash);
 310         list_del(&stp->st_perfile);
 311         list_del(&stp->st_perstateowner);
 312 }
 313
 314 static void free_generic_stateid(struct nfs4_stateid *stp)
 315 {
 316         put_nfs4_file(stp->st_file);
 317         kmem_cache_free(stateid_slab, stp);
 318 }
 319
 320 static void release_lock_stateid(struct nfs4_stateid *stp)
 321 {
 322         unhash_generic_stateid(stp);
 323         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
 324         free_generic_stateid(stp);
 325 }
 326
 327 static void unhash_lockowner(struct nfs4_stateowner *sop)
 328 {
 329         struct nfs4_stateid *stp;
 330
 331         list_del(&sop->so_idhash);
 332         list_del(&sop->so_strhash);
 333         list_del(&sop->so_perstateid);
 334         while (!list_empty(&sop->so_stateids)) {
 335                 stp = list_first_entry(&sop->so_stateids,
 336                                 struct nfs4_stateid, st_perstateowner);
 337                 release_lock_stateid(stp);
 338         }
 339 }
 340
 341 static void release_lockowner(struct nfs4_stateowner *sop)
 342 {
 343         unhash_lockowner(sop);
 344         nfs4_put_stateowner(sop);
 345 }
 346
 347 static void
 348 release_stateid_lockowners(struct nfs4_stateid *open_stp)
 349 {
 350         struct nfs4_stateowner *lock_sop;
 351
 352         while (!list_empty(&open_stp->st_lockowners)) {
 353                 lock_sop = list_entry(open_stp->st_lockowners.next,
 354                                 struct nfs4_stateowner, so_perstateid);
 355                 /* list_del(&open_stp->st_lockowners);  */
 356                 BUG_ON(lock_sop->so_is_open_owner);
 357                 release_lockowner(lock_sop);
 358         }
 359 }
 360
 361 static void release_open_stateid(struct nfs4_stateid *stp)
 362 {
 363         unhash_generic_stateid(stp);
 364         release_stateid_lockowners(stp);
 365         nfsd_close(stp->st_vfs_file);
 366         free_generic_stateid(stp);
 367 }
 368
 369 static void unhash_openowner(struct nfs4_stateowner *sop)
 370 {
 371         struct nfs4_stateid *stp;
 372
 373         list_del(&sop->so_idhash);
 374         list_del(&sop->so_strhash);
 375         list_del(&sop->so_perclient);
 376         list_del(&sop->so_perstateid); /* XXX: necessary? */
 377         while (!list_empty(&sop->so_stateids)) {
 378                 stp = list_first_entry(&sop->so_stateids,
 379                                 struct nfs4_stateid, st_perstateowner);
 380                 release_open_stateid(stp);
 381         }
 382 }
 383
 384 static void release_openowner(struct nfs4_stateowner *sop)
 385 {
 386         unhash_openowner(sop);
 387         list_del(&sop->so_close_lru);
 388         nfs4_put_stateowner(sop);
 389 }
 390
 391 static inline void
 392 renew_client(struct nfs4_client *clp)
 393 {
 394         /*
 395         * Move client to the end to the LRU list.
 396         */
 397         dprintk("renewing client (clientid %08x/%08x)\n",
 398                         clp->cl_clientid.cl_boot,
 399                         clp->cl_clientid.cl_id);
 400         list_move_tail(&clp->cl_lru, &client_lru);
 401         clp->cl_time = get_seconds();
 402 }
 403
 404 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 405 static int
 406 STALE_CLIENTID(clientid_t *clid)
 407 {
 408         if (clid->cl_boot == boot_time)
 409                 return 0;
 410         dprintk("NFSD stale clientid (%08x/%08x)\n",
 411                         clid->cl_boot, clid->cl_id);
 412         return 1;
 413 }
 414
 415 /*
 416  * XXX Should we use a slab cache ?
 417  * This type of memory management is somewhat inefficient, but we use it
 418  * anyway since SETCLIENTID is not a common operation.
 419  */
 420 static struct nfs4_client *alloc_client(struct xdr_netobj name)
 421 {
 422         struct nfs4_client *clp;
 423
 424         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
 425         if (clp == NULL)
 426                 return NULL;
 427         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
 428         if (clp->cl_name.data == NULL) {
 429                 kfree(clp);
 430                 return NULL;
 431         }
 432         memcpy(clp->cl_name.data, name.data, name.len);
 433         clp->cl_name.len = name.len;
 434         return clp;
 435 }
 436
 437 static void
 438 shutdown_callback_client(struct nfs4_client *clp)
 439 {
 440         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
 441
 442         if (clnt) {
 443                 /*
 444                  * Callback threads take a reference on the client, so there
 445                  * should be no outstanding callbacks at this point.
 446                  */
 447                 clp->cl_callback.cb_client = NULL;
 448                 rpc_shutdown_client(clnt);
 449         }
 450 }
 451
 452 static inline void
 453 free_client(struct nfs4_client *clp)
 454 {
 455         shutdown_callback_client(clp);
 456         if (clp->cl_cred.cr_group_info)
 457                 put_group_info(clp->cl_cred.cr_group_info);
 458         kfree(clp->cl_principal);
 459         kfree(clp->cl_name.data);
 460         kfree(clp);
 461 }
 462
 463 void
 464 put_nfs4_client(struct nfs4_client *clp)
 465 {
 466         if (atomic_dec_and_test(&clp->cl_count))
 467                 free_client(clp);
 468 }
 469
 470 static void
 471 expire_client(struct nfs4_client *clp)
 472 {
 473         struct nfs4_stateowner *sop;
 474         struct nfs4_delegation *dp;
 475         struct list_head reaplist;
 476
 477         dprintk("NFSD: expire_client cl_count %d\n",
 478                             atomic_read(&clp->cl_count));
 479
 480         INIT_LIST_HEAD(&reaplist);
 481         spin_lock(&recall_lock);
 482         while (!list_empty(&clp->cl_delegations)) {
 483                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
 484                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
 485                                 dp->dl_flock);
 486                 list_del_init(&dp->dl_perclnt);
 487                 list_move(&dp->dl_recall_lru, &reaplist);
 488         }
 489         spin_unlock(&recall_lock);
 490         while (!list_empty(&reaplist)) {
 491                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
 492                 list_del_init(&dp->dl_recall_lru);
 493                 unhash_delegation(dp);
 494         }
 495         list_del(&clp->cl_idhash);
 496         list_del(&clp->cl_strhash);
 497         list_del(&clp->cl_lru);
 498         while (!list_empty(&clp->cl_openowners)) {
 499                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
 500                 release_openowner(sop);
 501         }
 502         put_nfs4_client(clp);
 503 }
 504
 505 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
 506 {
 507         struct nfs4_client *clp;
 508
 509         clp = alloc_client(name);
 510         if (clp == NULL)
 511                 return NULL;
 512         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
 513         atomic_set(&clp->cl_count, 1);
 514         atomic_set(&clp->cl_callback.cb_set, 0);
 515         INIT_LIST_HEAD(&clp->cl_idhash);
 516         INIT_LIST_HEAD(&clp->cl_strhash);
 517         INIT_LIST_HEAD(&clp->cl_openowners);
 518         INIT_LIST_HEAD(&clp->cl_delegations);
 519         INIT_LIST_HEAD(&clp->cl_lru);
 520         return clp;
 521 }
 522
 523 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
 524 {
 525         memcpy(target->cl_verifier.data, source->data,
 526                         sizeof(target->cl_verifier.data));
 527 }
 528
 529 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
 530 {
 531         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
 532         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
 533 }
 534
 535 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
 536 {
 537         target->cr_uid = source->cr_uid;
 538         target->cr_gid = source->cr_gid;
 539         target->cr_group_info = source->cr_group_info;
 540         get_group_info(target->cr_group_info);
 541 }
 542
 543 static int same_name(const char *n1, const char *n2)
 544 {
 545         return 0 == memcmp(n1, n2, HEXDIR_LEN);
 546 }
 547
 548 static int
 549 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
 550 {
 551         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
 552 }
 553
 554 static int
 555 same_clid(clientid_t *cl1, clientid_t *cl2)
 556 {
 557         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
 558 }
 559
 560 /* XXX what about NGROUP */
 561 static int
 562 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
 563 {
 564         return cr1->cr_uid == cr2->cr_uid;
 565 }
 566
 567 static void gen_clid(struct nfs4_client *clp)
 568 {
 569         static u32 current_clientid = 1;
 570
 571         clp->cl_clientid.cl_boot = boot_time;
 572         clp->cl_clientid.cl_id = current_clientid++;
 573 }
 574
 575 static void gen_confirm(struct nfs4_client *clp)
 576 {
 577         static u32 i;
 578         u32 *p;
 579
 580         p = (u32 *)clp->cl_confirm.data;
 581         *p++ = get_seconds();
 582         *p++ = i++;
 583 }
 584
 585 static int check_name(struct xdr_netobj name)
 586 {
 587         if (name.len == 0)
 588                 return 0;
 589         if (name.len > NFS4_OPAQUE_LIMIT) {
 590                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
 591                 return 0;
 592         }
 593         return 1;
 594 }
 595
 596 static void
 597 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
 598 {
 599         unsigned int idhashval;
 600
 601         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
 602         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 603         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
 604         list_add_tail(&clp->cl_lru, &client_lru);
 605         clp->cl_time = get_seconds();
 606 }
 607
 608 static void
 609 move_to_confirmed(struct nfs4_client *clp)
 610 {
 611         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 612         unsigned int strhashval;
 613
 614         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
 615         list_del_init(&clp->cl_strhash);
 616         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
 617         strhashval = clientstr_hashval(clp->cl_recdir);
 618         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
 619         renew_client(clp);
 620 }
 621
 622 static struct nfs4_client *
 623 find_confirmed_client(clientid_t *clid)
 624 {
 625         struct nfs4_client *clp;
 626         unsigned int idhashval = clientid_hashval(clid->cl_id);
 627
 628         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
 629                 if (same_clid(&clp->cl_clientid, clid))
 630                         return clp;
 631         }
 632         return NULL;
 633 }
 634
 635 static struct nfs4_client *
 636 find_unconfirmed_client(clientid_t *clid)
 637 {
 638         struct nfs4_client *clp;
 639         unsigned int idhashval = clientid_hashval(clid->cl_id);
 640
 641         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
 642                 if (same_clid(&clp->cl_clientid, clid))
 643                         return clp;
 644         }
 645         return NULL;
 646 }
 647
 648 static struct nfs4_client *
 649 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
 650 {
 651         struct nfs4_client *clp;
 652
 653         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
 654                 if (same_name(clp->cl_recdir, dname))
 655                         return clp;
 656         }
 657         return NULL;
 658 }
 659
 660 static struct nfs4_client *
 661 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
 662 {
 663         struct nfs4_client *clp;
 664
 665         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
 666                 if (same_name(clp->cl_recdir, dname))
 667                         return clp;
 668         }
 669         return NULL;
 670 }
 671
 672 /* a helper function for parse_callback */
 673 static int
 674 parse_octet(unsigned int *lenp, char **addrp)
 675 {
 676         unsigned int len = *lenp;
 677         char *p = *addrp;
 678         int n = -1;
 679         char c;
 680
 681         for (;;) {
 682                 if (!len)
 683                         break;
 684                 len--;
 685                 c = *p++;
 686                 if (c == '.')
 687                         break;
 688                 if ((c < '0') || (c > '9')) {
 689                         n = -1;
 690                         break;
 691                 }
 692                 if (n < 0)
 693                         n = 0;
 694                 n = (n * 10) + (c - '0');
 695                 if (n > 255) {
 696                         n = -1;
 697                         break;
 698                 }
 699         }
 700         *lenp = len;
 701         *addrp = p;
 702         return n;
 703 }
 704
 705 /* parse and set the setclientid ipv4 callback address */
 706 static int
 707 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
 708 {
 709         int temp = 0;
 710         u32 cbaddr = 0;
 711         u16 cbport = 0;
 712         u32 addrlen = addr_len;
 713         char *addr = addr_val;
 714         int i, shift;
 715
 716         /* ipaddress */
 717         shift = 24;
 718         for(i = 4; i > 0  ; i--) {
 719                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 720                         return 0;
 721                 }
 722                 cbaddr |= (temp << shift);
 723                 if (shift > 0)
 724                 shift -= 8;
 725         }
 726         *cbaddrp = cbaddr;
 727
 728         /* port */
 729         shift = 8;
 730         for(i = 2; i > 0  ; i--) {
 731                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 732                         return 0;
 733                 }
 734                 cbport |= (temp << shift);
 735                 if (shift > 0)
 736                         shift -= 8;
 737         }
 738         *cbportp = cbport;
 739         return 1;
 740 }
 741
 742 static void
 743 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
 744 {
 745         struct nfs4_callback *cb = &clp->cl_callback;
 746
 747         /* Currently, we only support tcp for the callback channel */
 748         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
 749                 goto out_err;
 750
 751         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
 752                          &cb->cb_addr, &cb->cb_port)))
 753                 goto out_err;
 754         cb->cb_prog = se->se_callback_prog;
 755         cb->cb_ident = se->se_callback_ident;
 756         return;
 757 out_err:
 758         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
 759                 "will not receive delegations\n",
 760                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
 761
 762         return;
 763 }
 764
 765 __be32
 766 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 767                   struct nfsd4_setclientid *setclid)
 768 {
 769         struct sockaddr_in      *sin = svc_addr_in(rqstp);
 770         struct xdr_netobj       clname = {
 771                 .len = setclid->se_namelen,
 772                 .data = setclid->se_name,
 773         };
 774         nfs4_verifier           clverifier = setclid->se_verf;
 775         unsigned int            strhashval;
 776         struct nfs4_client      *conf, *unconf, *new;
 777         __be32                  status;
 778         char                    *princ;
 779         char                    dname[HEXDIR_LEN];
 780
 781         if (!check_name(clname))
 782                 return nfserr_inval;
 783
 784         status = nfs4_make_rec_clidname(dname, &clname);
 785         if (status)
 786                 return status;
 787
 788         /*
 789          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 790          * We get here on a DRC miss.
 791          */
 792
 793         strhashval = clientstr_hashval(dname);
 794
 795         nfs4_lock_state();
 796         conf = find_confirmed_client_by_str(dname, strhashval);
 797         if (conf) {
 798                 /* RFC 3530 14.2.33 CASE 0: */
 799                 status = nfserr_clid_inuse;
 800                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
 801                                 || conf->cl_addr != sin->sin_addr.s_addr) {
 802                         dprintk("NFSD: setclientid: string in use by clientat %pI4\n",
 803                                 &conf->cl_addr);
 804                         goto out;
 805                 }
 806         }
 807         /*
 808          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
 809          * has a description of SETCLIENTID request processing consisting
 810          * of 5 bullet points, labeled as CASE0 - CASE4 below.
 811          */
 812         unconf = find_unconfirmed_client_by_str(dname, strhashval);
 813         status = nfserr_resource;
 814         if (!conf) {
 815                 /*
 816                  * RFC 3530 14.2.33 CASE 4:
 817                  * placed first, because it is the normal case
 818                  */
 819                 if (unconf)
 820                         expire_client(unconf);
 821                 new = create_client(clname, dname);
 822                 if (new == NULL)
 823                         goto out;
 824                 gen_clid(new);
 825         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
 826                 /*
 827                  * RFC 3530 14.2.33 CASE 1:
 828                  * probable callback update
 829                  */
 830                 if (unconf) {
 831                         /* Note this is removing unconfirmed {*x***},
 832                          * which is stronger than RFC recommended {vxc**}.
 833                          * This has the advantage that there is at most
 834                          * one {*x***} in either list at any time.
 835                          */
 836                         expire_client(unconf);
 837                 }
 838                 new = create_client(clname, dname);
 839                 if (new == NULL)
 840                         goto out;
 841                 copy_clid(new, conf);
 842         } else if (!unconf) {
 843                 /*
 844                  * RFC 3530 14.2.33 CASE 2:
 845                  * probable client reboot; state will be removed if
 846                  * confirmed.
 847                  */
 848                 new = create_client(clname, dname);
 849                 if (new == NULL)
 850                         goto out;
 851                 gen_clid(new);
 852         } else {
 853                 /*
 854                  * RFC 3530 14.2.33 CASE 3:
 855                  * probable client reboot; state will be removed if
 856                  * confirmed.
 857                  */
 858                 expire_client(unconf);
 859                 new = create_client(clname, dname);
 860                 if (new == NULL)
 861                         goto out;
 862                 gen_clid(new);
 863         }
 864         copy_verf(new, &clverifier);
 865         new->cl_addr = sin->sin_addr.s_addr;
 866         new->cl_flavor = rqstp->rq_flavor;
 867         princ = svc_gss_principal(rqstp);
 868         if (princ) {
 869                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
 870                 if (new->cl_principal == NULL) {
 871                         free_client(new);
 872                         goto out;
 873                 }
 874         }
 875         copy_cred(&new->cl_cred, &rqstp->rq_cred);
 876         gen_confirm(new);
 877         gen_callback(new, setclid);
 878         add_to_unconfirmed(new, strhashval);
 879         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
 880         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
 881         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
 882         status = nfs_ok;
 883 out:
 884         nfs4_unlock_state();
 885         return status;
 886 }
 887
 888
 889 /*
 890  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
 891  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
 892  * bullets, labeled as CASE1 - CASE4 below.
 893  */
 894 __be32
 895 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
 896                          struct nfsd4_compound_state *cstate,
 897                          struct nfsd4_setclientid_confirm *setclientid_confirm)
 898 {
 899         struct sockaddr_in *sin = svc_addr_in(rqstp);
 900         struct nfs4_client *conf, *unconf;
 901         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
 902         clientid_t * clid = &setclientid_confirm->sc_clientid;
 903         __be32 status;
 904
 905         if (STALE_CLIENTID(clid))
 906                 return nfserr_stale_clientid;
 907         /*
 908          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 909          * We get here on a DRC miss.
 910          */
 911
 912         nfs4_lock_state();
 913
 914         conf = find_confirmed_client(clid);
 915         unconf = find_unconfirmed_client(clid);
 916
 917         status = nfserr_clid_inuse;
 918         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
 919                 goto out;
 920         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
 921                 goto out;
 922
 923         /*
 924          * section 14.2.34 of RFC 3530 has a description of
 925          * SETCLIENTID_CONFIRM request processing consisting
 926          * of 4 bullet points, labeled as CASE1 - CASE4 below.
 927          */
 928         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
 929                 /*
 930                  * RFC 3530 14.2.34 CASE 1:
 931                  * callback update
 932                  */
 933                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
 934                         status = nfserr_clid_inuse;
 935                 else {
 936                         /* XXX: We just turn off callbacks until we can handle
 937                           * change request correctly. */
 938                         atomic_set(&conf->cl_callback.cb_set, 0);
 939                         gen_confirm(conf);
 940                         nfsd4_remove_clid_dir(unconf);
 941                         expire_client(unconf);
 942                         status = nfs_ok;
 943
 944                 }
 945         } else if (conf && !unconf) {
 946                 /*
 947                  * RFC 3530 14.2.34 CASE 2:
 948                  * probable retransmitted request; play it safe and
 949                  * do nothing.
 950                  */
 951                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
 952                         status = nfserr_clid_inuse;
 953                 else
 954                         status = nfs_ok;
 955         } else if (!conf && unconf
 956                         && same_verf(&unconf->cl_confirm, &confirm)) {
 957                 /*
 958                  * RFC 3530 14.2.34 CASE 3:
 959                  * Normal case; new or rebooted client:
 960                  */
 961                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
 962                         status = nfserr_clid_inuse;
 963                 } else {
 964                         unsigned int hash =
 965                                 clientstr_hashval(unconf->cl_recdir);
 966                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
 967                                                                         hash);
 968                         if (conf) {
 969                                 nfsd4_remove_clid_dir(conf);
 970                                 expire_client(conf);
 971                         }
 972                         move_to_confirmed(unconf);
 973                         conf = unconf;
 974                         nfsd4_probe_callback(conf);
 975                         status = nfs_ok;
 976                 }
 977         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
 978             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
 979                                                                 &confirm)))) {
 980                 /*
 981                  * RFC 3530 14.2.34 CASE 4:
 982                  * Client probably hasn't noticed that we rebooted yet.
 983                  */
 984                 status = nfserr_stale_clientid;
 985         } else {
 986                 /* check that we have hit one of the cases...*/
 987                 status = nfserr_clid_inuse;
 988         }
 989 out:
 990         nfs4_unlock_state();
 991         return status;
 992 }
 993
 994 /* OPEN Share state helper functions */
 995 static inline struct nfs4_file *
 996 alloc_init_file(struct inode *ino)
 997 {
 998         struct nfs4_file *fp;
 999         unsigned int hashval = file_hashval(ino);
1000
1001         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1002         if (fp) {
1003                 kref_init(&fp->fi_ref);
1004                 INIT_LIST_HEAD(&fp->fi_hash);
1005                 INIT_LIST_HEAD(&fp->fi_stateids);
1006                 INIT_LIST_HEAD(&fp->fi_delegations);
1007                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1008                 fp->fi_inode = igrab(ino);
1009                 fp->fi_id = current_fileid++;
1010                 fp->fi_had_conflict = false;
1011                 return fp;
1012         }
1013         return NULL;
1014 }
1015
1016 static void
1017 nfsd4_free_slab(struct kmem_cache **slab)
1018 {
1019         if (*slab == NULL)
1020                 return;
1021         kmem_cache_destroy(*slab);
1022         *slab = NULL;
1023 }
1024
1025 void
1026 nfsd4_free_slabs(void)
1027 {
1028         nfsd4_free_slab(&stateowner_slab);
1029         nfsd4_free_slab(&file_slab);
1030         nfsd4_free_slab(&stateid_slab);
1031         nfsd4_free_slab(&deleg_slab);
1032 }
1033
1034 static int
1035 nfsd4_init_slabs(void)
1036 {
1037         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1038                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1039         if (stateowner_slab == NULL)
1040                 goto out_nomem;
1041         file_slab = kmem_cache_create("nfsd4_files",
1042                         sizeof(struct nfs4_file), 0, 0, NULL);
1043         if (file_slab == NULL)
1044                 goto out_nomem;
1045         stateid_slab = kmem_cache_create("nfsd4_stateids",
1046                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1047         if (stateid_slab == NULL)
1048                 goto out_nomem;
1049         deleg_slab = kmem_cache_create("nfsd4_delegations",
1050                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1051         if (deleg_slab == NULL)
1052                 goto out_nomem;
1053         return 0;
1054 out_nomem:
1055         nfsd4_free_slabs();
1056         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1057         return -ENOMEM;
1058 }
1059
1060 void
1061 nfs4_free_stateowner(struct kref *kref)
1062 {
1063         struct nfs4_stateowner *sop =
1064                 container_of(kref, struct nfs4_stateowner, so_ref);
1065         kfree(sop->so_owner.data);
1066         kmem_cache_free(stateowner_slab, sop);
1067 }
1068
1069 static inline struct nfs4_stateowner *
1070 alloc_stateowner(struct xdr_netobj *owner)
1071 {
1072         struct nfs4_stateowner *sop;
1073
1074         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1075                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1076                         memcpy(sop->so_owner.data, owner->data, owner->len);
1077                         sop->so_owner.len = owner->len;
1078                         kref_init(&sop->so_ref);
1079                         return sop;
1080                 }
1081                 kmem_cache_free(stateowner_slab, sop);
1082         }
1083         return NULL;
1084 }
1085
1086 static struct nfs4_stateowner *
1087 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1088         struct nfs4_stateowner *sop;
1089         struct nfs4_replay *rp;
1090         unsigned int idhashval;
1091
1092         if (!(sop = alloc_stateowner(&open->op_owner)))
1093                 return NULL;
1094         idhashval = ownerid_hashval(current_ownerid);
1095         INIT_LIST_HEAD(&sop->so_idhash);
1096         INIT_LIST_HEAD(&sop->so_strhash);
1097         INIT_LIST_HEAD(&sop->so_perclient);
1098         INIT_LIST_HEAD(&sop->so_stateids);
1099         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1100         INIT_LIST_HEAD(&sop->so_close_lru);
1101         sop->so_time = 0;
1102         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1103         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1104         list_add(&sop->so_perclient, &clp->cl_openowners);
1105         sop->so_is_open_owner = 1;
1106         sop->so_id = current_ownerid++;
1107         sop->so_client = clp;
1108         sop->so_seqid = open->op_seqid;
1109         sop->so_confirmed = 0;
1110         rp = &sop->so_replay;
1111         rp->rp_status = nfserr_serverfault;
1112         rp->rp_buflen = 0;
1113         rp->rp_buf = rp->rp_ibuf;
1114         return sop;
1115 }
1116
1117 static inline void
1118 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1119         struct nfs4_stateowner *sop = open->op_stateowner;
1120         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1121
1122         INIT_LIST_HEAD(&stp->st_hash);
1123         INIT_LIST_HEAD(&stp->st_perstateowner);
1124         INIT_LIST_HEAD(&stp->st_lockowners);
1125         INIT_LIST_HEAD(&stp->st_perfile);
1126         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1127         list_add(&stp->st_perstateowner, &sop->so_stateids);
1128         list_add(&stp->st_perfile, &fp->fi_stateids);
1129         stp->st_stateowner = sop;
1130         get_nfs4_file(fp);
1131         stp->st_file = fp;
1132         stp->st_stateid.si_boot = boot_time;
1133         stp->st_stateid.si_stateownerid = sop->so_id;
1134         stp->st_stateid.si_fileid = fp->fi_id;
1135         stp->st_stateid.si_generation = 0;
1136         stp->st_access_bmap = 0;
1137         stp->st_deny_bmap = 0;
1138         __set_bit(open->op_share_access, &stp->st_access_bmap);
1139         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1140         stp->st_openstp = NULL;
1141 }
1142
1143 static void
1144 move_to_close_lru(struct nfs4_stateowner *sop)
1145 {
1146         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1147
1148         list_move_tail(&sop->so_close_lru, &close_lru);
1149         sop->so_time = get_seconds();
1150 }
1151
1152 static int
1153 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1154                                                         clientid_t *clid)
1155 {
1156         return (sop->so_owner.len == owner->len) &&
1157                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1158                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1159 }
1160
1161 static struct nfs4_stateowner *
1162 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1163 {
1164         struct nfs4_stateowner *so = NULL;
1165
1166         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1167                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1168                         return so;
1169         }
1170         return NULL;
1171 }
1172
1173 /* search file_hashtbl[] for file */
1174 static struct nfs4_file *
1175 find_file(struct inode *ino)
1176 {
1177         unsigned int hashval = file_hashval(ino);
1178         struct nfs4_file *fp;
1179
1180         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1181                 if (fp->fi_inode == ino) {
1182                         get_nfs4_file(fp);
1183                         return fp;
1184                 }
1185         }
1186         return NULL;
1187 }
1188
1189 static inline int access_valid(u32 x)
1190 {
1191         if (x < NFS4_SHARE_ACCESS_READ)
1192                 return 0;
1193         if (x > NFS4_SHARE_ACCESS_BOTH)
1194                 return 0;
1195         return 1;
1196 }
1197
1198 static inline int deny_valid(u32 x)
1199 {
1200         /* Note: unlike access bits, deny bits may be zero. */
1201         return x <= NFS4_SHARE_DENY_BOTH;
1202 }
1203
1204 /*
1205  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1206  * st_{access,deny}_bmap field of the stateid, in order to track not
1207  * only what share bits are currently in force, but also what
1208  * combinations of share bits previous opens have used.  This allows us
1209  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1210  * return an error if the client attempt to downgrade to a combination
1211  * of share bits not explicable by closing some of its previous opens.
1212  *
1213  * XXX: This enforcement is actually incomplete, since we don't keep
1214  * track of access/deny bit combinations; so, e.g., we allow:
1215  *
1216  *      OPEN allow read, deny write
1217  *      OPEN allow both, deny none
1218  *      DOWNGRADE allow read, deny none
1219  *
1220  * which we should reject.
1221  */
1222 static void
1223 set_access(unsigned int *access, unsigned long bmap) {
1224         int i;
1225
1226         *access = 0;
1227         for (i = 1; i < 4; i++) {
1228                 if (test_bit(i, &bmap))
1229                         *access |= i;
1230         }
1231 }
1232
1233 static void
1234 set_deny(unsigned int *deny, unsigned long bmap) {
1235         int i;
1236
1237         *deny = 0;
1238         for (i = 0; i < 4; i++) {
1239                 if (test_bit(i, &bmap))
1240                         *deny |= i ;
1241         }
1242 }
1243
1244 static int
1245 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1246         unsigned int access, deny;
1247
1248         set_access(&access, stp->st_access_bmap);
1249         set_deny(&deny, stp->st_deny_bmap);
1250         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1251                 return 0;
1252         return 1;
1253 }
1254
1255 /*
1256  * Called to check deny when READ with all zero stateid or
1257  * WRITE with all zero or all one stateid
1258  */
1259 static __be32
1260 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1261 {
1262         struct inode *ino = current_fh->fh_dentry->d_inode;
1263         struct nfs4_file *fp;
1264         struct nfs4_stateid *stp;
1265         __be32 ret;
1266
1267         dprintk("NFSD: nfs4_share_conflict\n");
1268
1269         fp = find_file(ino);
1270         if (!fp)
1271                 return nfs_ok;
1272         ret = nfserr_locked;
1273         /* Search for conflicting share reservations */
1274         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1275                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1276                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1277                         goto out;
1278         }
1279         ret = nfs_ok;
1280 out:
1281         put_nfs4_file(fp);
1282         return ret;
1283 }
1284
1285 static inline void
1286 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1287 {
1288         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1289                 drop_file_write_access(filp);
1290                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1291         }
1292 }
1293
1294 /*
1295  * Recall a delegation
1296  */
1297 static int
1298 do_recall(void *__dp)
1299 {
1300         struct nfs4_delegation *dp = __dp;
1301
1302         dp->dl_file->fi_had_conflict = true;
1303         nfsd4_cb_recall(dp);
1304         return 0;
1305 }
1306
1307 /*
1308  * Spawn a thread to perform a recall on the delegation represented
1309  * by the lease (file_lock)
1310  *
1311  * Called from break_lease() with lock_kernel() held.
1312  * Note: we assume break_lease will only call this *once* for any given
1313  * lease.
1314  */
1315 static
1316 void nfsd_break_deleg_cb(struct file_lock *fl)
1317 {
1318         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1319         struct task_struct *t;
1320
1321         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1322         if (!dp)
1323                 return;
1324
1325         /* We're assuming the state code never drops its reference
1326          * without first removing the lease.  Since we're in this lease
1327          * callback (and since the lease code is serialized by the kernel
1328          * lock) we know the server hasn't removed the lease yet, we know
1329          * it's safe to take a reference: */
1330         atomic_inc(&dp->dl_count);
1331         atomic_inc(&dp->dl_client->cl_count);
1332
1333         spin_lock(&recall_lock);
1334         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1335         spin_unlock(&recall_lock);
1336
1337         /* only place dl_time is set. protected by lock_kernel*/
1338         dp->dl_time = get_seconds();
1339
1340         /*
1341          * We don't want the locks code to timeout the lease for us;
1342          * we'll remove it ourself if the delegation isn't returned
1343          * in time.
1344          */
1345         fl->fl_break_time = 0;
1346
1347         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1348         if (IS_ERR(t)) {
1349                 struct nfs4_client *clp = dp->dl_client;
1350
1351                 printk(KERN_INFO "NFSD: Callback thread failed for "
1352                         "for client (clientid %08x/%08x)\n",
1353                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1354                 put_nfs4_client(dp->dl_client);
1355                 nfs4_put_delegation(dp);
1356         }
1357 }
1358
1359 /*
1360  * The file_lock is being reapd.
1361  *
1362  * Called by locks_free_lock() with lock_kernel() held.
1363  */
1364 static
1365 void nfsd_release_deleg_cb(struct file_lock *fl)
1366 {
1367         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1368
1369         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1370
1371         if (!(fl->fl_flags & FL_LEASE) || !dp)
1372                 return;
1373         dp->dl_flock = NULL;
1374 }
1375
1376 /*
1377  * Set the delegation file_lock back pointer.
1378  *
1379  * Called from setlease() with lock_kernel() held.
1380  */
1381 static
1382 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1383 {
1384         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1385
1386         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1387         if (!dp)
1388                 return;
1389         dp->dl_flock = new;
1390 }
1391
1392 /*
1393  * Called from setlease() with lock_kernel() held
1394  */
1395 static
1396 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1397 {
1398         struct nfs4_delegation *onlistd =
1399                 (struct nfs4_delegation *)onlist->fl_owner;
1400         struct nfs4_delegation *tryd =
1401                 (struct nfs4_delegation *)try->fl_owner;
1402
1403         if (onlist->fl_lmops != try->fl_lmops)
1404                 return 0;
1405
1406         return onlistd->dl_client == tryd->dl_client;
1407 }
1408
1409
1410 static
1411 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1412 {
1413         if (arg & F_UNLCK)
1414                 return lease_modify(onlist, arg);
1415         else
1416                 return -EAGAIN;
1417 }
1418
1419 static struct lock_manager_operations nfsd_lease_mng_ops = {
1420         .fl_break = nfsd_break_deleg_cb,
1421         .fl_release_private = nfsd_release_deleg_cb,
1422         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1423         .fl_mylease = nfsd_same_client_deleg_cb,
1424         .fl_change = nfsd_change_deleg_cb,
1425 };
1426
1427
1428 __be32
1429 nfsd4_process_open1(struct nfsd4_open *open)
1430 {
1431         clientid_t *clientid = &open->op_clientid;
1432         struct nfs4_client *clp = NULL;
1433         unsigned int strhashval;
1434         struct nfs4_stateowner *sop = NULL;
1435
1436         if (!check_name(open->op_owner))
1437                 return nfserr_inval;
1438
1439         if (STALE_CLIENTID(&open->op_clientid))
1440                 return nfserr_stale_clientid;
1441
1442         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1443         sop = find_openstateowner_str(strhashval, open);
1444         open->op_stateowner = sop;
1445         if (!sop) {
1446                 /* Make sure the client's lease hasn't expired. */
1447                 clp = find_confirmed_client(clientid);
1448                 if (clp == NULL)
1449                         return nfserr_expired;
1450                 goto renew;
1451         }
1452         if (!sop->so_confirmed) {
1453                 /* Replace unconfirmed owners without checking for replay. */
1454                 clp = sop->so_client;
1455                 release_openowner(sop);
1456                 open->op_stateowner = NULL;
1457                 goto renew;
1458         }
1459         if (open->op_seqid == sop->so_seqid - 1) {
1460                 if (sop->so_replay.rp_buflen)
1461                         return nfserr_replay_me;
1462                 /* The original OPEN failed so spectacularly
1463                  * that we don't even have replay data saved!
1464                  * Therefore, we have no choice but to continue
1465                  * processing this OPEN; presumably, we'll
1466                  * fail again for the same reason.
1467                  */
1468                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1469                 goto renew;
1470         }
1471         if (open->op_seqid != sop->so_seqid)
1472                 return nfserr_bad_seqid;
1473 renew:
1474         if (open->op_stateowner == NULL) {
1475                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1476                 if (sop == NULL)
1477                         return nfserr_resource;
1478                 open->op_stateowner = sop;
1479         }
1480         list_del_init(&sop->so_close_lru);
1481         renew_client(sop->so_client);
1482         return nfs_ok;
1483 }
1484
1485 static inline __be32
1486 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1487 {
1488         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1489                 return nfserr_openmode;
1490         else
1491                 return nfs_ok;
1492 }
1493
1494 static struct nfs4_delegation *
1495 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1496 {
1497         struct nfs4_delegation *dp;
1498
1499         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1500                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1501                         return dp;
1502         }
1503         return NULL;
1504 }
1505
1506 static __be32
1507 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1508                 struct nfs4_delegation **dp)
1509 {
1510         int flags;
1511         __be32 status = nfserr_bad_stateid;
1512
1513         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1514         if (*dp == NULL)
1515                 goto out;
1516         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1517                                                 RD_STATE : WR_STATE;
1518         status = nfs4_check_delegmode(*dp, flags);
1519         if (status)
1520                 *dp = NULL;
1521 out:
1522         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1523                 return nfs_ok;
1524         if (status)
1525                 return status;
1526         open->op_stateowner->so_confirmed = 1;
1527         return nfs_ok;
1528 }
1529
1530 static __be32
1531 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1532 {
1533         struct nfs4_stateid *local;
1534         __be32 status = nfserr_share_denied;
1535         struct nfs4_stateowner *sop = open->op_stateowner;
1536
1537         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1538                 /* ignore lock owners */
1539                 if (local->st_stateowner->so_is_open_owner == 0)
1540                         continue;
1541                 /* remember if we have seen this open owner */
1542                 if (local->st_stateowner == sop)
1543                         *stpp = local;
1544                 /* check for conflicting share reservations */
1545                 if (!test_share(local, open))
1546                         goto out;
1547         }
1548         status = 0;
1549 out:
1550         return status;
1551 }
1552
1553 static inline struct nfs4_stateid *
1554 nfs4_alloc_stateid(void)
1555 {
1556         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1557 }
1558
1559 static __be32
1560 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1561                 struct nfs4_delegation *dp,
1562                 struct svc_fh *cur_fh, int flags)
1563 {
1564         struct nfs4_stateid *stp;
1565
1566         stp = nfs4_alloc_stateid();
1567         if (stp == NULL)
1568                 return nfserr_resource;
1569
1570         if (dp) {
1571                 get_file(dp->dl_vfs_file);
1572                 stp->st_vfs_file = dp->dl_vfs_file;
1573         } else {
1574                 __be32 status;
1575                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1576                                 &stp->st_vfs_file);
1577                 if (status) {
1578                         if (status == nfserr_dropit)
1579                                 status = nfserr_jukebox;
1580                         kmem_cache_free(stateid_slab, stp);
1581                         return status;
1582                 }
1583         }
1584         *stpp = stp;
1585         return 0;
1586 }
1587
1588 static inline __be32
1589 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1590                 struct nfsd4_open *open)
1591 {
1592         struct iattr iattr = {
1593                 .ia_valid = ATTR_SIZE,
1594                 .ia_size = 0,
1595         };
1596         if (!open->op_truncate)
1597                 return 0;
1598         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1599                 return nfserr_inval;
1600         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1601 }
1602
1603 static __be32
1604 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1605 {
1606         struct file *filp = stp->st_vfs_file;
1607         struct inode *inode = filp->f_path.dentry->d_inode;
1608         unsigned int share_access, new_writer;
1609         __be32 status;
1610
1611         set_access(&share_access, stp->st_access_bmap);
1612         new_writer = (~share_access) & open->op_share_access
1613                         & NFS4_SHARE_ACCESS_WRITE;
1614
1615         if (new_writer) {
1616                 int err = get_write_access(inode);
1617                 if (err)
1618                         return nfserrno(err);
1619                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
1620                 if (err)
1621                         return nfserrno(err);
1622                 file_take_write(filp);
1623         }
1624         status = nfsd4_truncate(rqstp, cur_fh, open);
1625         if (status) {
1626                 if (new_writer)
1627                         put_write_access(inode);
1628                 return status;
1629         }
1630         /* remember the open */
1631         filp->f_mode |= open->op_share_access;
1632         __set_bit(open->op_share_access, &stp->st_access_bmap);
1633         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1634
1635         return nfs_ok;
1636 }
1637
1638
1639 static void
1640 nfs4_set_claim_prev(struct nfsd4_open *open)
1641 {
1642         open->op_stateowner->so_confirmed = 1;
1643         open->op_stateowner->so_client->cl_firststate = 1;
1644 }
1645
1646 /*
1647  * Attempt to hand out a delegation.
1648  */
1649 static void
1650 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1651 {
1652         struct nfs4_delegation *dp;
1653         struct nfs4_stateowner *sop = stp->st_stateowner;
1654         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1655         struct file_lock fl, *flp = &fl;
1656         int status, flag = 0;
1657
1658         flag = NFS4_OPEN_DELEGATE_NONE;
1659         open->op_recall = 0;
1660         switch (open->op_claim_type) {
1661                 case NFS4_OPEN_CLAIM_PREVIOUS:
1662                         if (!atomic_read(&cb->cb_set))
1663                                 open->op_recall = 1;
1664                         flag = open->op_delegate_type;
1665                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1666                                 goto out;
1667                         break;
1668                 case NFS4_OPEN_CLAIM_NULL:
1669                         /* Let's not give out any delegations till everyone's
1670                          * had the chance to reclaim theirs.... */
1671                         if (locks_in_grace())
1672                                 goto out;
1673                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1674                                 goto out;
1675                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1676                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1677                         else
1678                                 flag = NFS4_OPEN_DELEGATE_READ;
1679                         break;
1680                 default:
1681                         goto out;
1682         }
1683
1684         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1685         if (dp == NULL) {
1686                 flag = NFS4_OPEN_DELEGATE_NONE;
1687                 goto out;
1688         }
1689         locks_init_lock(&fl);
1690         fl.fl_lmops = &nfsd_lease_mng_ops;
1691         fl.fl_flags = FL_LEASE;
1692         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
1693         fl.fl_end = OFFSET_MAX;
1694         fl.fl_owner =  (fl_owner_t)dp;
1695         fl.fl_file = stp->st_vfs_file;
1696         fl.fl_pid = current->tgid;
1697
1698         /* vfs_setlease checks to see if delegation should be handed out.
1699          * the lock_manager callbacks fl_mylease and fl_change are used
1700          */
1701         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
1702                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1703                 unhash_delegation(dp);
1704                 flag = NFS4_OPEN_DELEGATE_NONE;
1705                 goto out;
1706         }
1707
1708         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1709
1710         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1711                      dp->dl_stateid.si_boot,
1712                      dp->dl_stateid.si_stateownerid,
1713                      dp->dl_stateid.si_fileid,
1714                      dp->dl_stateid.si_generation);
1715 out:
1716         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1717                         && flag == NFS4_OPEN_DELEGATE_NONE
1718                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1719                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1720         open->op_delegate_type = flag;
1721 }
1722
1723 /*
1724  * called with nfs4_lock_state() held.
1725  */
1726 __be32
1727 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1728 {
1729         struct nfs4_file *fp = NULL;
1730         struct inode *ino = current_fh->fh_dentry->d_inode;
1731         struct nfs4_stateid *stp = NULL;
1732         struct nfs4_delegation *dp = NULL;
1733         __be32 status;
1734
1735         status = nfserr_inval;
1736         if (!access_valid(open->op_share_access)
1737                         || !deny_valid(open->op_share_deny))
1738                 goto out;
1739         /*
1740          * Lookup file; if found, lookup stateid and check open request,
1741          * and check for delegations in the process of being recalled.
1742          * If not found, create the nfs4_file struct
1743          */
1744         fp = find_file(ino);
1745         if (fp) {
1746                 if ((status = nfs4_check_open(fp, open, &stp)))
1747                         goto out;
1748                 status = nfs4_check_deleg(fp, open, &dp);
1749                 if (status)
1750                         goto out;
1751         } else {
1752                 status = nfserr_bad_stateid;
1753                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1754                         goto out;
1755                 status = nfserr_resource;
1756                 fp = alloc_init_file(ino);
1757                 if (fp == NULL)
1758                         goto out;
1759         }
1760
1761         /*
1762          * OPEN the file, or upgrade an existing OPEN.
1763          * If truncate fails, the OPEN fails.
1764          */
1765         if (stp) {
1766                 /* Stateid was found, this is an OPEN upgrade */
1767                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1768                 if (status)
1769                         goto out;
1770                 update_stateid(&stp->st_stateid);
1771         } else {
1772                 /* Stateid was not found, this is a new OPEN */
1773                 int flags = 0;
1774                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1775                         flags |= NFSD_MAY_READ;
1776                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1777                         flags |= NFSD_MAY_WRITE;
1778                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1779                 if (status)
1780                         goto out;
1781                 init_stateid(stp, fp, open);
1782                 status = nfsd4_truncate(rqstp, current_fh, open);
1783                 if (status) {
1784                         release_open_stateid(stp);
1785                         goto out;
1786                 }
1787         }
1788         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1789
1790         /*
1791         * Attempt to hand out a delegation. No error return, because the
1792         * OPEN succeeds even if we fail.
1793         */
1794         nfs4_open_delegation(current_fh, open, stp);
1795
1796         status = nfs_ok;
1797
1798         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1799                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1800                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1801 out:
1802         if (fp)
1803                 put_nfs4_file(fp);
1804         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1805                 nfs4_set_claim_prev(open);
1806         /*
1807         * To finish the open response, we just need to set the rflags.
1808         */
1809         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1810         if (!open->op_stateowner->so_confirmed)
1811                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1812
1813         return status;
1814 }
1815
1816 __be32
1817 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1818             clientid_t *clid)
1819 {
1820         struct nfs4_client *clp;
1821         __be32 status;
1822
1823         nfs4_lock_state();
1824         dprintk("process_renew(%08x/%08x): starting\n",
1825                         clid->cl_boot, clid->cl_id);
1826         status = nfserr_stale_clientid;
1827         if (STALE_CLIENTID(clid))
1828                 goto out;
1829         clp = find_confirmed_client(clid);
1830         status = nfserr_expired;
1831         if (clp == NULL) {
1832                 /* We assume the client took too long to RENEW. */
1833                 dprintk("nfsd4_renew: clientid not found!\n");
1834                 goto out;
1835         }
1836         renew_client(clp);
1837         status = nfserr_cb_path_down;
1838         if (!list_empty(&clp->cl_delegations)
1839                         && !atomic_read(&clp->cl_callback.cb_set))
1840                 goto out;
1841         status = nfs_ok;
1842 out:
1843         nfs4_unlock_state();
1844         return status;
1845 }
1846
1847 struct lock_manager nfsd4_manager = {
1848 };
1849
1850 static void
1851 nfsd4_end_grace(void)
1852 {
1853         dprintk("NFSD: end of grace period\n");
1854         nfsd4_recdir_purge_old();
1855         locks_end_grace(&nfsd4_manager);
1856 }
1857
1858 static time_t
1859 nfs4_laundromat(void)
1860 {
1861         struct nfs4_client *clp;
1862         struct nfs4_stateowner *sop;
1863         struct nfs4_delegation *dp;
1864         struct list_head *pos, *next, reaplist;
1865         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1866         time_t t, clientid_val = NFSD_LEASE_TIME;
1867         time_t u, test_val = NFSD_LEASE_TIME;
1868
1869         nfs4_lock_state();
1870
1871         dprintk("NFSD: laundromat service - starting\n");
1872         if (locks_in_grace())
1873                 nfsd4_end_grace();
1874         list_for_each_safe(pos, next, &client_lru) {
1875                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1876                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1877                         t = clp->cl_time - cutoff;
1878                         if (clientid_val > t)
1879                                 clientid_val = t;
1880                         break;
1881                 }
1882                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1883                         clp->cl_clientid.cl_id);
1884                 nfsd4_remove_clid_dir(clp);
1885                 expire_client(clp);
1886         }
1887         INIT_LIST_HEAD(&reaplist);
1888         spin_lock(&recall_lock);
1889         list_for_each_safe(pos, next, &del_recall_lru) {
1890                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1891                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1892                         u = dp->dl_time - cutoff;
1893                         if (test_val > u)
1894                                 test_val = u;
1895                         break;
1896                 }
1897                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1898                                     dp, dp->dl_flock);
1899                 list_move(&dp->dl_recall_lru, &reaplist);
1900         }
1901         spin_unlock(&recall_lock);
1902         list_for_each_safe(pos, next, &reaplist) {
1903                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1904                 list_del_init(&dp->dl_recall_lru);
1905                 unhash_delegation(dp);
1906         }
1907         test_val = NFSD_LEASE_TIME;
1908         list_for_each_safe(pos, next, &close_lru) {
1909                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1910                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1911                         u = sop->so_time - cutoff;
1912                         if (test_val > u)
1913                                 test_val = u;
1914                         break;
1915                 }
1916                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1917                         sop->so_id);
1918                 release_openowner(sop);
1919         }
1920         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1921                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1922         nfs4_unlock_state();
1923         return clientid_val;
1924 }
1925
1926 static struct workqueue_struct *laundry_wq;
1927 static void laundromat_main(struct work_struct *);
1928 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1929
1930 static void
1931 laundromat_main(struct work_struct *not_used)
1932 {
1933         time_t t;
1934
1935         t = nfs4_laundromat();
1936         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1937         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1938 }
1939
1940 static struct nfs4_stateowner *
1941 search_close_lru(u32 st_id, int flags)
1942 {
1943         struct nfs4_stateowner *local = NULL;
1944
1945         if (flags & CLOSE_STATE) {
1946                 list_for_each_entry(local, &close_lru, so_close_lru) {
1947                         if (local->so_id == st_id)
1948                                 return local;
1949                 }
1950         }
1951         return NULL;
1952 }
1953
1954 static inline int
1955 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1956 {
1957         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1958 }
1959
1960 static int
1961 STALE_STATEID(stateid_t *stateid)
1962 {
1963         if (stateid->si_boot == boot_time)
1964                 return 0;
1965         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1966                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1967                 stateid->si_generation);
1968         return 1;
1969 }
1970
1971 static inline int
1972 access_permit_read(unsigned long access_bmap)
1973 {
1974         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1975                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1976                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1977 }
1978
1979 static inline int
1980 access_permit_write(unsigned long access_bmap)
1981 {
1982         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1983                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1984 }
1985
1986 static
1987 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1988 {
1989         __be32 status = nfserr_openmode;
1990
1991         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1992                 goto out;
1993         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1994                 goto out;
1995         status = nfs_ok;
1996 out:
1997         return status;
1998 }
1999
2000 static inline __be32
2001 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2002 {
2003         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2004                 return nfs_ok;
2005         else if (locks_in_grace()) {
2006                 /* Answer in remaining cases depends on existance of
2007                  * conflicting state; so we must wait out the grace period. */
2008                 return nfserr_grace;
2009         } else if (flags & WR_STATE)
2010                 return nfs4_share_conflict(current_fh,
2011                                 NFS4_SHARE_DENY_WRITE);
2012         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2013                 return nfs4_share_conflict(current_fh,
2014                                 NFS4_SHARE_DENY_READ);
2015 }
2016
2017 /*
2018  * Allow READ/WRITE during grace period on recovered state only for files
2019  * that are not able to provide mandatory locking.
2020  */
2021 static inline int
2022 io_during_grace_disallowed(struct inode *inode, int flags)
2023 {
2024         return locks_in_grace() && mandatory_lock(inode);
2025 }
2026
2027 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
2028 {
2029         /* If the client sends us a stateid from the future, it's buggy: */
2030         if (in->si_generation > ref->si_generation)
2031                 return nfserr_bad_stateid;
2032         /*
2033          * The following, however, can happen.  For example, if the
2034          * client sends an open and some IO at the same time, the open
2035          * may bump si_generation while the IO is still in flight.
2036          * Thanks to hard links and renames, the client never knows what
2037          * file an open will affect.  So it could avoid that situation
2038          * only by serializing all opens and IO from the same open
2039          * owner.  To recover from the old_stateid error, the client
2040          * will just have to retry the IO:
2041          */
2042         if (in->si_generation < ref->si_generation)
2043                 return nfserr_old_stateid;
2044         return nfs_ok;
2045 }
2046
2047 static int is_delegation_stateid(stateid_t *stateid)
2048 {
2049         return stateid->si_fileid == 0;
2050 }
2051
2052 /*
2053 * Checks for stateid operations
2054 */
2055 __be32
2056 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2057 {
2058         struct nfs4_stateid *stp = NULL;
2059         struct nfs4_delegation *dp = NULL;
2060         struct inode *ino = current_fh->fh_dentry->d_inode;
2061         __be32 status;
2062
2063         if (filpp)
2064                 *filpp = NULL;
2065
2066         if (io_during_grace_disallowed(ino, flags))
2067                 return nfserr_grace;
2068
2069         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2070                 return check_special_stateids(current_fh, stateid, flags);
2071
2072         status = nfserr_stale_stateid;
2073         if (STALE_STATEID(stateid))
2074                 goto out;
2075
2076         status = nfserr_bad_stateid;
2077         if (is_delegation_stateid(stateid)) {
2078                 dp = find_delegation_stateid(ino, stateid);
2079                 if (!dp)
2080                         goto out;
2081                 status = check_stateid_generation(stateid, &dp->dl_stateid);
2082                 if (status)
2083                         goto out;
2084                 status = nfs4_check_delegmode(dp, flags);
2085                 if (status)
2086                         goto out;
2087                 renew_client(dp->dl_client);
2088                 if (filpp)
2089                         *filpp = dp->dl_vfs_file;
2090         } else { /* open or lock stateid */
2091                 stp = find_stateid(stateid, flags);
2092                 if (!stp)
2093                         goto out;
2094                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2095                         goto out;
2096                 if (!stp->st_stateowner->so_confirmed)
2097                         goto out;
2098                 status = check_stateid_generation(stateid, &stp->st_stateid);
2099                 if (status)
2100                         goto out;
2101                 status = nfs4_check_openmode(stp, flags);
2102                 if (status)
2103                         goto out;
2104                 renew_client(stp->st_stateowner->so_client);
2105                 if (filpp)
2106                         *filpp = stp->st_vfs_file;
2107         }
2108         status = nfs_ok;
2109 out:
2110         return status;
2111 }
2112
2113 static inline int
2114 setlkflg (int type)
2115 {
2116         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2117                 RD_STATE : WR_STATE;
2118 }
2119
2120 /*
2121  * Checks for sequence id mutating operations.
2122  */
2123 static __be32
2124 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)
2125 {
2126         struct nfs4_stateid *stp;
2127         struct nfs4_stateowner *sop;
2128         __be32 status;
2129
2130         dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2131                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2132                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2133                 stateid->si_generation);
2134
2135         *stpp = NULL;
2136         *sopp = NULL;
2137
2138         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2139                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2140                 return nfserr_bad_stateid;
2141         }
2142
2143         if (STALE_STATEID(stateid))
2144                 return nfserr_stale_stateid;
2145         /*
2146         * We return BAD_STATEID if filehandle doesn't match stateid,
2147         * the confirmed flag is incorrecly set, or the generation
2148         * number is incorrect.
2149         */
2150         stp = find_stateid(stateid, flags);
2151         if (stp == NULL) {
2152                 /*
2153                  * Also, we should make sure this isn't just the result of
2154                  * a replayed close:
2155                  */
2156                 sop = search_close_lru(stateid->si_stateownerid, flags);
2157                 if (sop == NULL)
2158                         return nfserr_bad_stateid;
2159                 *sopp = sop;
2160                 goto check_replay;
2161         }
2162
2163         *stpp = stp;
2164         *sopp = sop = stp->st_stateowner;
2165
2166         if (lock) {
2167                 clientid_t *lockclid = &lock->v.new.clientid;
2168                 struct nfs4_client *clp = sop->so_client;
2169                 int lkflg = 0;
2170                 __be32 status;
2171
2172                 lkflg = setlkflg(lock->lk_type);
2173
2174                 if (lock->lk_is_new) {
2175                         if (!sop->so_is_open_owner)
2176                                 return nfserr_bad_stateid;
2177                         if (!same_clid(&clp->cl_clientid, lockclid))
2178                                return nfserr_bad_stateid;
2179                         /* stp is the open stateid */
2180                         status = nfs4_check_openmode(stp, lkflg);
2181                         if (status)
2182                                 return status;
2183                 } else {
2184                         /* stp is the lock stateid */
2185                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2186                         if (status)
2187                                 return status;
2188                }
2189         }
2190
2191         if (nfs4_check_fh(current_fh, stp)) {
2192                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2193                 return nfserr_bad_stateid;
2194         }
2195
2196         /*
2197         *  We now validate the seqid and stateid generation numbers.
2198         *  For the moment, we ignore the possibility of
2199         *  generation number wraparound.
2200         */
2201         if (seqid != sop->so_seqid)
2202                 goto check_replay;
2203
2204         if (sop->so_confirmed && flags & CONFIRM) {
2205                 dprintk("NFSD: preprocess_seqid_op: expected"
2206                                 " unconfirmed stateowner!\n");
2207                 return nfserr_bad_stateid;
2208         }
2209         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2210                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2211                                 " confirmed yet!\n");
2212                 return nfserr_bad_stateid;
2213         }
2214         status = check_stateid_generation(stateid, &stp->st_stateid);
2215         if (status)
2216                 return status;
2217         renew_client(sop->so_client);
2218         return nfs_ok;
2219
2220 check_replay:
2221         if (seqid == sop->so_seqid - 1) {
2222                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2223                 /* indicate replay to calling function */
2224                 return nfserr_replay_me;
2225         }
2226         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2227                         sop->so_seqid, seqid);
2228         *sopp = NULL;
2229         return nfserr_bad_seqid;
2230 }
2231
2232 __be32
2233 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2234                    struct nfsd4_open_confirm *oc)
2235 {
2236         __be32 status;
2237         struct nfs4_stateowner *sop;
2238         struct nfs4_stateid *stp;
2239
2240         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2241                         (int)cstate->current_fh.fh_dentry->d_name.len,
2242                         cstate->current_fh.fh_dentry->d_name.name);
2243
2244         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2245         if (status)
2246                 return status;
2247
2248         nfs4_lock_state();
2249
2250         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2251                                         oc->oc_seqid, &oc->oc_req_stateid,
2252                                         CONFIRM | OPEN_STATE,
2253                                         &oc->oc_stateowner, &stp, NULL)))
2254                 goto out;
2255
2256         sop = oc->oc_stateowner;
2257         sop->so_confirmed = 1;
2258         update_stateid(&stp->st_stateid);
2259         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2260         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2261                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2262                          stp->st_stateid.si_boot,
2263                          stp->st_stateid.si_stateownerid,
2264                          stp->st_stateid.si_fileid,
2265                          stp->st_stateid.si_generation);
2266
2267         nfsd4_create_clid_dir(sop->so_client);
2268 out:
2269         if (oc->oc_stateowner) {
2270                 nfs4_get_stateowner(oc->oc_stateowner);
2271                 cstate->replay_owner = oc->oc_stateowner;
2272         }
2273         nfs4_unlock_state();
2274         return status;
2275 }
2276
2277
2278 /*
2279  * unset all bits in union bitmap (bmap) that
2280  * do not exist in share (from successful OPEN_DOWNGRADE)
2281  */
2282 static void
2283 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2284 {
2285         int i;
2286         for (i = 1; i < 4; i++) {
2287                 if ((i & access) != i)
2288                         __clear_bit(i, bmap);
2289         }
2290 }
2291
2292 static void
2293 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2294 {
2295         int i;
2296         for (i = 0; i < 4; i++) {
2297                 if ((i & deny) != i)
2298                         __clear_bit(i, bmap);
2299         }
2300 }
2301
2302 __be32
2303 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2304                      struct nfsd4_compound_state *cstate,
2305                      struct nfsd4_open_downgrade *od)
2306 {
2307         __be32 status;
2308         struct nfs4_stateid *stp;
2309         unsigned int share_access;
2310
2311         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2312                         (int)cstate->current_fh.fh_dentry->d_name.len,
2313                         cstate->current_fh.fh_dentry->d_name.name);
2314
2315         if (!access_valid(od->od_share_access)
2316                         || !deny_valid(od->od_share_deny))
2317                 return nfserr_inval;
2318
2319         nfs4_lock_state();
2320         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2321                                         od->od_seqid,
2322                                         &od->od_stateid,
2323                                         OPEN_STATE,
2324                                         &od->od_stateowner, &stp, NULL)))
2325                 goto out;
2326
2327         status = nfserr_inval;
2328         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2329                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2330                         stp->st_access_bmap, od->od_share_access);
2331                 goto out;
2332         }
2333         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2334                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2335                         stp->st_deny_bmap, od->od_share_deny);
2336                 goto out;
2337         }
2338         set_access(&share_access, stp->st_access_bmap);
2339         nfs4_file_downgrade(stp->st_vfs_file,
2340                             share_access & ~od->od_share_access);
2341
2342         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2343         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2344
2345         update_stateid(&stp->st_stateid);
2346         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2347         status = nfs_ok;
2348 out:
2349         if (od->od_stateowner) {
2350                 nfs4_get_stateowner(od->od_stateowner);
2351                 cstate->replay_owner = od->od_stateowner;
2352         }
2353         nfs4_unlock_state();
2354         return status;
2355 }
2356
2357 /*
2358  * nfs4_unlock_state() called after encode
2359  */
2360 __be32
2361 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2362             struct nfsd4_close *close)
2363 {
2364         __be32 status;
2365         struct nfs4_stateid *stp;
2366
2367         dprintk("NFSD: nfsd4_close on file %.*s\n",
2368                         (int)cstate->current_fh.fh_dentry->d_name.len,
2369                         cstate->current_fh.fh_dentry->d_name.name);
2370
2371         nfs4_lock_state();
2372         /* check close_lru for replay */
2373         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2374                                         close->cl_seqid,
2375                                         &close->cl_stateid,
2376                                         OPEN_STATE | CLOSE_STATE,
2377                                         &close->cl_stateowner, &stp, NULL)))
2378                 goto out;
2379         status = nfs_ok;
2380         update_stateid(&stp->st_stateid);
2381         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2382
2383         /* release_stateid() calls nfsd_close() if needed */
2384         release_open_stateid(stp);
2385
2386         /* place unused nfs4_stateowners on so_close_lru list to be
2387          * released by the laundromat service after the lease period
2388          * to enable us to handle CLOSE replay
2389          */
2390         if (list_empty(&close->cl_stateowner->so_stateids))
2391                 move_to_close_lru(close->cl_stateowner);
2392 out:
2393         if (close->cl_stateowner) {
2394                 nfs4_get_stateowner(close->cl_stateowner);
2395                 cstate->replay_owner = close->cl_stateowner;
2396         }
2397         nfs4_unlock_state();
2398         return status;
2399 }
2400
2401 __be32
2402 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2403                   struct nfsd4_delegreturn *dr)
2404 {
2405         struct nfs4_delegation *dp;
2406         stateid_t *stateid = &dr->dr_stateid;
2407         struct inode *inode;
2408         __be32 status;
2409
2410         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2411                 return status;
2412         inode = cstate->current_fh.fh_dentry->d_inode;
2413
2414         nfs4_lock_state();
2415         status = nfserr_bad_stateid;
2416         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2417                 goto out;
2418         status = nfserr_stale_stateid;
2419         if (STALE_STATEID(stateid))
2420                 goto out;
2421         status = nfs_ok;
2422         if (!is_delegation_stateid(stateid))
2423                 goto out;
2424         status = nfserr_bad_stateid;
2425         dp = find_delegation_stateid(inode, stateid);
2426         if (!dp)
2427                 goto out;
2428         status = check_stateid_generation(stateid, &dp->dl_stateid);
2429         if (status)
2430                 goto out;
2431         renew_client(dp->dl_client);
2432
2433         unhash_delegation(dp);
2434 out:
2435         nfs4_unlock_state();
2436
2437         return status;
2438 }
2439
2440
2441 /*
2442  * Lock owner state (byte-range locks)
2443  */
2444 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2445 #define LOCK_HASH_BITS              8
2446 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2447 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2448
2449 static inline u64
2450 end_offset(u64 start, u64 len)
2451 {
2452         u64 end;
2453
2454         end = start + len;
2455         return end >= start ? end: NFS4_MAX_UINT64;
2456 }
2457
2458 /* last octet in a range */
2459 static inline u64
2460 last_byte_offset(u64 start, u64 len)
2461 {
2462         u64 end;
2463
2464         BUG_ON(!len);
2465         end = start + len;
2466         return end > start ? end - 1: NFS4_MAX_UINT64;
2467 }
2468
2469 #define lockownerid_hashval(id) \
2470         ((id) & LOCK_HASH_MASK)
2471
2472 static inline unsigned int
2473 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2474                 struct xdr_netobj *ownername)
2475 {
2476         return (file_hashval(inode) + cl_id
2477                         + opaque_hashval(ownername->data, ownername->len))
2478                 & LOCK_HASH_MASK;
2479 }
2480
2481 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2482 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2483 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2484
2485 static struct nfs4_stateid *
2486 find_stateid(stateid_t *stid, int flags)
2487 {
2488         struct nfs4_stateid *local;
2489         u32 st_id = stid->si_stateownerid;
2490         u32 f_id = stid->si_fileid;
2491         unsigned int hashval;
2492
2493         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2494         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
2495                 hashval = stateid_hashval(st_id, f_id);
2496                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2497                         if ((local->st_stateid.si_stateownerid == st_id) &&
2498                             (local->st_stateid.si_fileid == f_id))
2499                                 return local;
2500                 }
2501         }
2502
2503         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
2504                 hashval = stateid_hashval(st_id, f_id);
2505                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2506                         if ((local->st_stateid.si_stateownerid == st_id) &&
2507                             (local->st_stateid.si_fileid == f_id))
2508                                 return local;
2509                 }
2510         }
2511         return NULL;
2512 }
2513
2514 static struct nfs4_delegation *
2515 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2516 {
2517         struct nfs4_file *fp;
2518         struct nfs4_delegation *dl;
2519
2520         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2521                     stid->si_boot, stid->si_stateownerid,
2522                     stid->si_fileid, stid->si_generation);
2523
2524         fp = find_file(ino);
2525         if (!fp)
2526                 return NULL;
2527         dl = find_delegation_file(fp, stid);
2528         put_nfs4_file(fp);
2529         return dl;
2530 }
2531
2532 /*
2533  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2534  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2535  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2536  * locking, this prevents us from being completely protocol-compliant.  The
2537  * real solution to this problem is to start using unsigned file offsets in
2538  * the VFS, but this is a very deep change!
2539  */
2540 static inline void
2541 nfs4_transform_lock_offset(struct file_lock *lock)
2542 {
2543         if (lock->fl_start < 0)
2544                 lock->fl_start = OFFSET_MAX;
2545         if (lock->fl_end < 0)
2546                 lock->fl_end = OFFSET_MAX;
2547 }
2548
2549 /* Hack!: For now, we're defining this just so we can use a pointer to it
2550  * as a unique cookie to identify our (NFSv4's) posix locks. */
2551 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2552 };
2553
2554 static inline void
2555 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2556 {
2557         struct nfs4_stateowner *sop;
2558         unsigned int hval;
2559
2560         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2561                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2562                 hval = lockownerid_hashval(sop->so_id);
2563                 kref_get(&sop->so_ref);
2564                 deny->ld_sop = sop;
2565                 deny->ld_clientid = sop->so_client->cl_clientid;
2566         } else {
2567                 deny->ld_sop = NULL;
2568                 deny->ld_clientid.cl_boot = 0;
2569                 deny->ld_clientid.cl_id = 0;
2570         }
2571         deny->ld_start = fl->fl_start;
2572         deny->ld_length = NFS4_MAX_UINT64;
2573         if (fl->fl_end != NFS4_MAX_UINT64)
2574                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2575         deny->ld_type = NFS4_READ_LT;
2576         if (fl->fl_type != F_RDLCK)
2577                 deny->ld_type = NFS4_WRITE_LT;
2578 }
2579
2580 static struct nfs4_stateowner *
2581 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2582                 struct xdr_netobj *owner)
2583 {
2584         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2585         struct nfs4_stateowner *op;
2586
2587         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2588                 if (same_owner_str(op, owner, clid))
2589                         return op;
2590         }
2591         return NULL;
2592 }
2593
2594 /*
2595  * Alloc a lock owner structure.
2596  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2597  * occured.
2598  *
2599  * strhashval = lock_ownerstr_hashval
2600  */
2601
2602 static struct nfs4_stateowner *
2603 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2604         struct nfs4_stateowner *sop;
2605         struct nfs4_replay *rp;
2606         unsigned int idhashval;
2607
2608         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2609                 return NULL;
2610         idhashval = lockownerid_hashval(current_ownerid);
2611         INIT_LIST_HEAD(&sop->so_idhash);
2612         INIT_LIST_HEAD(&sop->so_strhash);
2613         INIT_LIST_HEAD(&sop->so_perclient);
2614         INIT_LIST_HEAD(&sop->so_stateids);
2615         INIT_LIST_HEAD(&sop->so_perstateid);
2616         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2617         sop->so_time = 0;
2618         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2619         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2620         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2621         sop->so_is_open_owner = 0;
2622         sop->so_id = current_ownerid++;
2623         sop->so_client = clp;
2624         /* It is the openowner seqid that will be incremented in encode in the
2625          * case of new lockowners; so increment the lock seqid manually: */
2626         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2627         sop->so_confirmed = 1;
2628         rp = &sop->so_replay;
2629         rp->rp_status = nfserr_serverfault;
2630         rp->rp_buflen = 0;
2631         rp->rp_buf = rp->rp_ibuf;
2632         return sop;
2633 }
2634
2635 static struct nfs4_stateid *
2636 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2637 {
2638         struct nfs4_stateid *stp;
2639         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2640
2641         stp = nfs4_alloc_stateid();
2642         if (stp == NULL)
2643                 goto out;
2644         INIT_LIST_HEAD(&stp->st_hash);
2645         INIT_LIST_HEAD(&stp->st_perfile);
2646         INIT_LIST_HEAD(&stp->st_perstateowner);
2647         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2648         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2649         list_add(&stp->st_perfile, &fp->fi_stateids);
2650         list_add(&stp->st_perstateowner, &sop->so_stateids);
2651         stp->st_stateowner = sop;
2652         get_nfs4_file(fp);
2653         stp->st_file = fp;
2654         stp->st_stateid.si_boot = boot_time;
2655         stp->st_stateid.si_stateownerid = sop->so_id;
2656         stp->st_stateid.si_fileid = fp->fi_id;
2657         stp->st_stateid.si_generation = 0;
2658         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2659         stp->st_access_bmap = open_stp->st_access_bmap;
2660         stp->st_deny_bmap = open_stp->st_deny_bmap;
2661         stp->st_openstp = open_stp;
2662
2663 out:
2664         return stp;
2665 }
2666
2667 static int
2668 check_lock_length(u64 offset, u64 length)
2669 {
2670         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
2671              LOFF_OVERFLOW(offset, length)));
2672 }
2673
2674 /*
2675  *  LOCK operation
2676  */
2677 __be32
2678 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2679            struct nfsd4_lock *lock)
2680 {
2681         struct nfs4_stateowner *open_sop = NULL;
2682         struct nfs4_stateowner *lock_sop = NULL;
2683         struct nfs4_stateid *lock_stp;
2684         struct file *filp;
2685         struct file_lock file_lock;
2686         struct file_lock conflock;
2687         __be32 status = 0;
2688         unsigned int strhashval;
2689         unsigned int cmd;
2690         int err;
2691
2692         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2693                 (long long) lock->lk_offset,
2694                 (long long) lock->lk_length);
2695
2696         if (check_lock_length(lock->lk_offset, lock->lk_length))
2697                  return nfserr_inval;
2698
2699         if ((status = fh_verify(rqstp, &cstate->current_fh,
2700                                 S_IFREG, NFSD_MAY_LOCK))) {
2701                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2702                 return status;
2703         }
2704
2705         nfs4_lock_state();
2706
2707         if (lock->lk_is_new) {
2708                 /*
2709                  * Client indicates that this is a new lockowner.
2710                  * Use open owner and open stateid to create lock owner and
2711                  * lock stateid.
2712                  */
2713                 struct nfs4_stateid *open_stp = NULL;
2714                 struct nfs4_file *fp;
2715
2716                 status = nfserr_stale_clientid;
2717                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2718                         goto out;
2719
2720                 /* validate and update open stateid and open seqid */
2721                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2722                                         lock->lk_new_open_seqid,
2723                                         &lock->lk_new_open_stateid,
2724                                         OPEN_STATE,
2725                                         &lock->lk_replay_owner, &open_stp,
2726                                         lock);
2727                 if (status)
2728                         goto out;
2729                 open_sop = lock->lk_replay_owner;
2730                 /* create lockowner and lock stateid */
2731                 fp = open_stp->st_file;
2732                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
2733                                 open_sop->so_client->cl_clientid.cl_id,
2734                                 &lock->v.new.owner);
2735                 /* XXX: Do we need to check for duplicate stateowners on
2736                  * the same file, or should they just be allowed (and
2737                  * create new stateids)? */
2738                 status = nfserr_resource;
2739                 lock_sop = alloc_init_lock_stateowner(strhashval,
2740                                 open_sop->so_client, open_stp, lock);
2741                 if (lock_sop == NULL)
2742                         goto out;
2743                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2744                 if (lock_stp == NULL)
2745                         goto out;
2746         } else {
2747                 /* lock (lock owner + lock stateid) already exists */
2748                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2749                                        lock->lk_old_lock_seqid,
2750                                        &lock->lk_old_lock_stateid,
2751                                        LOCK_STATE,
2752                                        &lock->lk_replay_owner, &lock_stp, lock);
2753                 if (status)
2754                         goto out;
2755                 lock_sop = lock->lk_replay_owner;
2756         }
2757         /* lock->lk_replay_owner and lock_stp have been created or found */
2758         filp = lock_stp->st_vfs_file;
2759
2760         status = nfserr_grace;
2761         if (locks_in_grace() && !lock->lk_reclaim)
2762                 goto out;
2763         status = nfserr_no_grace;
2764         if (!locks_in_grace() && lock->lk_reclaim)
2765                 goto out;
2766
2767         locks_init_lock(&file_lock);
2768         switch (lock->lk_type) {
2769                 case NFS4_READ_LT:
2770                 case NFS4_READW_LT:
2771                         file_lock.fl_type = F_RDLCK;
2772                         cmd = F_SETLK;
2773                 break;
2774                 case NFS4_WRITE_LT:
2775                 case NFS4_WRITEW_LT:
2776                         file_lock.fl_type = F_WRLCK;
2777                         cmd = F_SETLK;
2778                 break;
2779                 default:
2780                         status = nfserr_inval;
2781                 goto out;
2782         }
2783         file_lock.fl_owner = (fl_owner_t)lock_sop;
2784         file_lock.fl_pid = current->tgid;
2785         file_lock.fl_file = filp;
2786         file_lock.fl_flags = FL_POSIX;
2787         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2788
2789         file_lock.fl_start = lock->lk_offset;
2790         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
2791         nfs4_transform_lock_offset(&file_lock);
2792
2793         /*
2794         * Try to lock the file in the VFS.
2795         * Note: locks.c uses the BKL to protect the inode's lock list.
2796         */
2797
2798         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2799         switch (-err) {
2800         case 0: /* success! */
2801                 update_stateid(&lock_stp->st_stateid);
2802                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
2803                                 sizeof(stateid_t));
2804                 status = 0;
2805                 break;
2806         case (EAGAIN):          /* conflock holds conflicting lock */
2807                 status = nfserr_denied;
2808                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2809                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2810                 break;
2811         case (EDEADLK):
2812                 status = nfserr_deadlock;
2813                 break;
2814         default:
2815                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2816                 status = nfserr_resource;
2817                 break;
2818         }
2819 out:
2820         if (status && lock->lk_is_new && lock_sop)
2821                 release_lockowner(lock_sop);
2822         if (lock->lk_replay_owner) {
2823                 nfs4_get_stateowner(lock->lk_replay_owner);
2824                 cstate->replay_owner = lock->lk_replay_owner;
2825         }
2826         nfs4_unlock_state();
2827         return status;
2828 }
2829
2830 /*
2831  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
2832  * so we do a temporary open here just to get an open file to pass to
2833  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
2834  * inode operation.)
2835  */
2836 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
2837 {
2838         struct file *file;
2839         int err;
2840
2841         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
2842         if (err)
2843                 return err;
2844         err = vfs_test_lock(file, lock);
2845         nfsd_close(file);
2846         return err;
2847 }
2848
2849 /*
2850  * LOCKT operation
2851  */
2852 __be32
2853 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2854             struct nfsd4_lockt *lockt)
2855 {
2856         struct inode *inode;
2857         struct file_lock file_lock;
2858         int error;
2859         __be32 status;
2860
2861         if (locks_in_grace())
2862                 return nfserr_grace;
2863
2864         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2865                  return nfserr_inval;
2866
2867         lockt->lt_stateowner = NULL;
2868         nfs4_lock_state();
2869
2870         status = nfserr_stale_clientid;
2871         if (STALE_CLIENTID(&lockt->lt_clientid))
2872                 goto out;
2873
2874         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2875                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2876                 if (status == nfserr_symlink)
2877                         status = nfserr_inval;
2878                 goto out;
2879         }
2880
2881         inode = cstate->current_fh.fh_dentry->d_inode;
2882         locks_init_lock(&file_lock);
2883         switch (lockt->lt_type) {
2884                 case NFS4_READ_LT:
2885                 case NFS4_READW_LT:
2886                         file_lock.fl_type = F_RDLCK;
2887                 break;
2888                 case NFS4_WRITE_LT:
2889                 case NFS4_WRITEW_LT:
2890                         file_lock.fl_type = F_WRLCK;
2891                 break;
2892                 default:
2893                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2894                         status = nfserr_inval;
2895                 goto out;
2896         }
2897
2898         lockt->lt_stateowner = find_lockstateowner_str(inode,
2899                         &lockt->lt_clientid, &lockt->lt_owner);
2900         if (lockt->lt_stateowner)
2901                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2902         file_lock.fl_pid = current->tgid;
2903         file_lock.fl_flags = FL_POSIX;
2904
2905         file_lock.fl_start = lockt->lt_offset;
2906         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
2907
2908         nfs4_transform_lock_offset(&file_lock);
2909
2910         status = nfs_ok;
2911         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
2912         if (error) {
2913                 status = nfserrno(error);
2914                 goto out;
2915         }
2916         if (file_lock.fl_type != F_UNLCK) {
2917                 status = nfserr_denied;
2918                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2919         }
2920 out:
2921         nfs4_unlock_state();
2922         return status;
2923 }
2924
2925 __be32
2926 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2927             struct nfsd4_locku *locku)
2928 {
2929         struct nfs4_stateid *stp;
2930         struct file *filp = NULL;
2931         struct file_lock file_lock;
2932         __be32 status;
2933         int err;
2934
2935         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2936                 (long long) locku->lu_offset,
2937                 (long long) locku->lu_length);
2938
2939         if (check_lock_length(locku->lu_offset, locku->lu_length))
2940                  return nfserr_inval;
2941
2942         nfs4_lock_state();
2943
2944         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2945                                         locku->lu_seqid,
2946                                         &locku->lu_stateid,
2947                                         LOCK_STATE,
2948                                         &locku->lu_stateowner, &stp, NULL)))
2949                 goto out;
2950
2951         filp = stp->st_vfs_file;
2952         BUG_ON(!filp);
2953         locks_init_lock(&file_lock);
2954         file_lock.fl_type = F_UNLCK;
2955         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2956         file_lock.fl_pid = current->tgid;
2957         file_lock.fl_file = filp;
2958         file_lock.fl_flags = FL_POSIX;
2959         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2960         file_lock.fl_start = locku->lu_offset;
2961
2962         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
2963         nfs4_transform_lock_offset(&file_lock);
2964
2965         /*
2966         *  Try to unlock the file in the VFS.
2967         */
2968         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2969         if (err) {
2970                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2971                 goto out_nfserr;
2972         }
2973         /*
2974         * OK, unlock succeeded; the only thing left to do is update the stateid.
2975         */
2976         update_stateid(&stp->st_stateid);
2977         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2978
2979 out:
2980         if (locku->lu_stateowner) {
2981                 nfs4_get_stateowner(locku->lu_stateowner);
2982                 cstate->replay_owner = locku->lu_stateowner;
2983         }
2984         nfs4_unlock_state();
2985         return status;
2986
2987 out_nfserr:
2988         status = nfserrno(err);
2989         goto out;
2990 }
2991
2992 /*
2993  * returns
2994  *      1: locks held by lockowner
2995  *      0: no locks held by lockowner
2996  */
2997 static int
2998 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2999 {
3000         struct file_lock **flpp;
3001         struct inode *inode = filp->f_path.dentry->d_inode;
3002         int status = 0;
3003
3004         lock_kernel();
3005         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3006                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3007                         status = 1;
3008                         goto out;
3009                 }
3010         }
3011 out:
3012         unlock_kernel();
3013         return status;
3014 }
3015
3016 __be32
3017 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3018                         struct nfsd4_compound_state *cstate,
3019                         struct nfsd4_release_lockowner *rlockowner)
3020 {
3021         clientid_t *clid = &rlockowner->rl_clientid;
3022         struct nfs4_stateowner *sop;
3023         struct nfs4_stateid *stp;
3024         struct xdr_netobj *owner = &rlockowner->rl_owner;
3025         struct list_head matches;
3026         int i;
3027         __be32 status;
3028
3029         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3030                 clid->cl_boot, clid->cl_id);
3031
3032         /* XXX check for lease expiration */
3033
3034         status = nfserr_stale_clientid;
3035         if (STALE_CLIENTID(clid))
3036                 return status;
3037
3038         nfs4_lock_state();
3039
3040         status = nfserr_locks_held;
3041         /* XXX: we're doing a linear search through all the lockowners.
3042          * Yipes!  For now we'll just hope clients aren't really using
3043          * release_lockowner much, but eventually we have to fix these
3044          * data structures. */
3045         INIT_LIST_HEAD(&matches);
3046         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3047                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3048                         if (!same_owner_str(sop, owner, clid))
3049                                 continue;
3050                         list_for_each_entry(stp, &sop->so_stateids,
3051                                         st_perstateowner) {
3052                                 if (check_for_locks(stp->st_vfs_file, sop))
3053                                         goto out;
3054                                 /* Note: so_perclient unused for lockowners,
3055                                  * so it's OK to fool with here. */
3056                                 list_add(&sop->so_perclient, &matches);
3057                         }
3058                 }
3059         }
3060         /* Clients probably won't expect us to return with some (but not all)
3061          * of the lockowner state released; so don't release any until all
3062          * have been checked. */
3063         status = nfs_ok;
3064         while (!list_empty(&matches)) {
3065                 sop = list_entry(matches.next, struct nfs4_stateowner,
3066                                                                 so_perclient);
3067                 /* unhash_stateowner deletes so_perclient only
3068                  * for openowners. */
3069                 list_del(&sop->so_perclient);
3070                 release_lockowner(sop);
3071         }
3072 out:
3073         nfs4_unlock_state();
3074         return status;
3075 }
3076
3077 static inline struct nfs4_client_reclaim *
3078 alloc_reclaim(void)
3079 {
3080         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3081 }
3082
3083 int
3084 nfs4_has_reclaimed_state(const char *name)
3085 {
3086         unsigned int strhashval = clientstr_hashval(name);
3087         struct nfs4_client *clp;
3088
3089         clp = find_confirmed_client_by_str(name, strhashval);
3090         return clp ? 1 : 0;
3091 }
3092
3093 /*
3094  * failure => all reset bets are off, nfserr_no_grace...
3095  */
3096 int
3097 nfs4_client_to_reclaim(const char *name)
3098 {
3099         unsigned int strhashval;
3100         struct nfs4_client_reclaim *crp = NULL;
3101
3102         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3103         crp = alloc_reclaim();
3104         if (!crp)
3105                 return 0;
3106         strhashval = clientstr_hashval(name);
3107         INIT_LIST_HEAD(&crp->cr_strhash);
3108         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3109         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3110         reclaim_str_hashtbl_size++;
3111         return 1;
3112 }
3113
3114 static void
3115 nfs4_release_reclaim(void)
3116 {
3117         struct nfs4_client_reclaim *crp = NULL;
3118         int i;
3119
3120         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3121                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3122                         crp = list_entry(reclaim_str_hashtbl[i].next,
3123                                         struct nfs4_client_reclaim, cr_strhash);
3124                         list_del(&crp->cr_strhash);
3125                         kfree(crp);
3126                         reclaim_str_hashtbl_size--;
3127                 }
3128         }
3129         BUG_ON(reclaim_str_hashtbl_size);
3130 }
3131
3132 /*
3133  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3134 static struct nfs4_client_reclaim *
3135 nfs4_find_reclaim_client(clientid_t *clid)
3136 {
3137         unsigned int strhashval;
3138         struct nfs4_client *clp;
3139         struct nfs4_client_reclaim *crp = NULL;
3140
3141
3142         /* find clientid in conf_id_hashtbl */
3143         clp = find_confirmed_client(clid);
3144         if (clp == NULL)
3145                 return NULL;
3146
3147         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3148                             clp->cl_name.len, clp->cl_name.data,
3149                             clp->cl_recdir);
3150
3151         /* find clp->cl_name in reclaim_str_hashtbl */
3152         strhashval = clientstr_hashval(clp->cl_recdir);
3153         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3154                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3155                         return crp;
3156                 }
3157         }
3158         return NULL;
3159 }
3160
3161 /*
3162 * Called from OPEN. Look for clientid in reclaim list.
3163 */
3164 __be32
3165 nfs4_check_open_reclaim(clientid_t *clid)
3166 {
3167         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3168 }
3169
3170 /* initialization to perform at module load time: */
3171
3172 int
3173 nfs4_state_init(void)
3174 {
3175         int i, status;
3176
3177         status = nfsd4_init_slabs();
3178         if (status)
3179                 return status;
3180         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3181                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3182                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3183                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3184                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3185         }
3186         for (i = 0; i < FILE_HASH_SIZE; i++) {
3187                 INIT_LIST_HEAD(&file_hashtbl[i]);
3188         }
3189         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3190                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3191                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3192         }
3193         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3194                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3195                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3196         }
3197         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3198                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3199                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3200         }
3201         memset(&onestateid, ~0, sizeof(stateid_t));
3202         INIT_LIST_HEAD(&close_lru);
3203         INIT_LIST_HEAD(&client_lru);
3204         INIT_LIST_HEAD(&del_recall_lru);
3205         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3206                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3207         reclaim_str_hashtbl_size = 0;
3208         return 0;
3209 }
3210
3211 static void
3212 nfsd4_load_reboot_recovery_data(void)
3213 {
3214         int status;
3215
3216         nfs4_lock_state();
3217         nfsd4_init_recdir(user_recovery_dirname);
3218         status = nfsd4_recdir_load();
3219         nfs4_unlock_state();
3220         if (status)
3221                 printk("NFSD: Failure reading reboot recovery data\n");
3222 }
3223
3224 unsigned long
3225 get_nfs4_grace_period(void)
3226 {
3227         return max(user_lease_time, lease_time) * HZ;
3228 }
3229
3230 /*
3231  * Since the lifetime of a delegation isn't limited to that of an open, a
3232  * client may quite reasonably hang on to a delegation as long as it has
3233  * the inode cached.  This becomes an obvious problem the first time a
3234  * client's inode cache approaches the size of the server's total memory.
3235  *
3236  * For now we avoid this problem by imposing a hard limit on the number
3237  * of delegations, which varies according to the server's memory size.
3238  */
3239 static void
3240 set_max_delegations(void)
3241 {
3242         /*
3243          * Allow at most 4 delegations per megabyte of RAM.  Quick
3244          * estimates suggest that in the worst case (where every delegation
3245          * is for a different inode), a delegation could take about 1.5K,
3246          * giving a worst case usage of about 6% of memory.
3247          */
3248         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3249 }
3250
3251 /* initialization to perform when the nfsd service is started: */
3252
3253 static void
3254 __nfs4_state_start(void)
3255 {
3256         unsigned long grace_time;
3257
3258         boot_time = get_seconds();
3259         grace_time = get_nfs4_grace_period();
3260         lease_time = user_lease_time;
3261         locks_start_grace(&nfsd4_manager);
3262         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3263                grace_time/HZ);
3264         laundry_wq = create_singlethread_workqueue("nfsd4");
3265         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3266         set_max_delegations();
3267 }
3268
3269 void
3270 nfs4_state_start(void)
3271 {
3272         if (nfs4_init)
3273                 return;
3274         nfsd4_load_reboot_recovery_data();
3275         __nfs4_state_start();
3276         nfs4_init = 1;
3277         return;
3278 }
3279
3280 time_t
3281 nfs4_lease_time(void)
3282 {
3283         return lease_time;
3284 }
3285
3286 static void
3287 __nfs4_state_shutdown(void)
3288 {
3289         int i;
3290         struct nfs4_client *clp = NULL;
3291         struct nfs4_delegation *dp = NULL;
3292         struct list_head *pos, *next, reaplist;
3293
3294         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3295                 while (!list_empty(&conf_id_hashtbl[i])) {
3296                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3297                         expire_client(clp);
3298                 }
3299                 while (!list_empty(&unconf_str_hashtbl[i])) {
3300                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3301                         expire_client(clp);
3302                 }
3303         }
3304         INIT_LIST_HEAD(&reaplist);
3305         spin_lock(&recall_lock);
3306         list_for_each_safe(pos, next, &del_recall_lru) {
3307                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3308                 list_move(&dp->dl_recall_lru, &reaplist);
3309         }
3310         spin_unlock(&recall_lock);
3311         list_for_each_safe(pos, next, &reaplist) {
3312                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3313                 list_del_init(&dp->dl_recall_lru);
3314                 unhash_delegation(dp);
3315         }
3316
3317         nfsd4_shutdown_recdir();
3318         nfs4_init = 0;
3319 }
3320
3321 void
3322 nfs4_state_shutdown(void)
3323 {
3324         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3325         destroy_workqueue(laundry_wq);
3326         locks_end_grace(&nfsd4_manager);
3327         nfs4_lock_state();
3328         nfs4_release_reclaim();
3329         __nfs4_state_shutdown();
3330         nfs4_unlock_state();
3331 }
3332
3333 /*
3334  * user_recovery_dirname is protected by the nfsd_mutex since it's only
3335  * accessed when nfsd is starting.
3336  */
3337 static void
3338 nfs4_set_recdir(char *recdir)
3339 {
3340         strcpy(user_recovery_dirname, recdir);
3341 }
3342
3343 /*
3344  * Change the NFSv4 recovery directory to recdir.
3345  */
3346 int
3347 nfs4_reset_recoverydir(char *recdir)
3348 {
3349         int status;
3350         struct path path;
3351
3352         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
3353         if (status)
3354                 return status;
3355         status = -ENOTDIR;
3356         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
3357                 nfs4_set_recdir(recdir);
3358                 status = 0;
3359         }
3360         path_put(&path);
3361         return status;
3362 }
3363
3364 char *
3365 nfs4_recoverydir(void)
3366 {
3367         return user_recovery_dirname;
3368 }
3369
3370 /*
3371  * Called when leasetime is changed.
3372  *
3373  * The only way the protocol gives us to handle on-the-fly lease changes is to
3374  * simulate a reboot.  Instead of doing that, we just wait till the next time
3375  * we start to register any changes in lease time.  If the administrator
3376  * really wants to change the lease time *now*, they can go ahead and bring
3377  * nfsd down and then back up again after changing the lease time.
3378  *
3379  * user_lease_time is protected by nfsd_mutex since it's only really accessed
3380  * when nfsd is starting
3381  */
3382 void
3383 nfs4_reset_lease(time_t leasetime)
3384 {
3385         user_lease_time = leasetime;
3386 }