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/mount.h>
  45 #include <linux/workqueue.h>
  46 #include <linux/smp_lock.h>
  47 #include <linux/kthread.h>
  48 #include <linux/nfs4.h>
  49 #include <linux/nfsd/state.h>
  50 #include <linux/nfsd/xdr4.h>
  51 #include <linux/namei.h>
  52 #include <linux/swap.h>
  53 #include <linux/mutex.h>
  54 #include <linux/lockd/bind.h>
  55 #include <linux/module.h>
  56
  57 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  58
  59 /* Globals */
  60 static time_t lease_time = 90;     /* default lease time */
  61 static time_t user_lease_time = 90;
  62 static time_t boot_time;
  63 static int in_grace = 1;
  64 static u32 current_clientid = 1;
  65 static u32 current_ownerid = 1;
  66 static u32 current_fileid = 1;
  67 static u32 current_delegid = 1;
  68 static u32 nfs4_init;
  69 static stateid_t zerostateid;             /* bits all 0 */
  70 static stateid_t onestateid;              /* bits all 1 */
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
  74
  75 /* forward declarations */
  76 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
  77 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
  78 static void release_stateid_lockowners(struct nfs4_stateid *open_stp);
  79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  80 static void nfs4_set_recdir(char *recdir);
  81
  82 /* Locking:
  83  *
  84  * client_mutex:
  85  *      protects clientid_hashtbl[], clientstr_hashtbl[],
  86  *      unconfstr_hashtbl[], uncofid_hashtbl[].
  87  */
  88 static DEFINE_MUTEX(client_mutex);
  89
  90 static struct kmem_cache *stateowner_slab = NULL;
  91 static struct kmem_cache *file_slab = NULL;
  92 static struct kmem_cache *stateid_slab = NULL;
  93 static struct kmem_cache *deleg_slab = NULL;
  94
  95 void
  96 nfs4_lock_state(void)
  97 {
  98         mutex_lock(&client_mutex);
  99 }
 100
 101 void
 102 nfs4_unlock_state(void)
 103 {
 104         mutex_unlock(&client_mutex);
 105 }
 106
 107 static inline u32
 108 opaque_hashval(const void *ptr, int nbytes)
 109 {
 110         unsigned char *cptr = (unsigned char *) ptr;
 111
 112         u32 x = 0;
 113         while (nbytes--) {
 114                 x *= 37;
 115                 x += *cptr++;
 116         }
 117         return x;
 118 }
 119
 120 /* forward declarations */
 121 static void release_stateowner(struct nfs4_stateowner *sop);
 122 static void release_stateid(struct nfs4_stateid *stp, int flags);
 123
 124 /*
 125  * Delegation state
 126  */
 127
 128 /* recall_lock protects the del_recall_lru */
 129 static DEFINE_SPINLOCK(recall_lock);
 130 static struct list_head del_recall_lru;
 131
 132 static void
 133 free_nfs4_file(struct kref *kref)
 134 {
 135         struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
 136         list_del(&fp->fi_hash);
 137         iput(fp->fi_inode);
 138         kmem_cache_free(file_slab, fp);
 139 }
 140
 141 static inline void
 142 put_nfs4_file(struct nfs4_file *fi)
 143 {
 144         kref_put(&fi->fi_ref, free_nfs4_file);
 145 }
 146
 147 static inline void
 148 get_nfs4_file(struct nfs4_file *fi)
 149 {
 150         kref_get(&fi->fi_ref);
 151 }
 152
 153 static int num_delegations;
 154 unsigned int max_delegations;
 155
 156 /*
 157  * Open owner state (share locks)
 158  */
 159
 160 /* hash tables for nfs4_stateowner */
 161 #define OWNER_HASH_BITS              8
 162 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 163 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 164
 165 #define ownerid_hashval(id) \
 166         ((id) & OWNER_HASH_MASK)
 167 #define ownerstr_hashval(clientid, ownername) \
 168         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
 169
 170 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
 171 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 172
 173 /* hash table for nfs4_file */
 174 #define FILE_HASH_BITS                   8
 175 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 176 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
 177 /* hash table for (open)nfs4_stateid */
 178 #define STATEID_HASH_BITS              10
 179 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
 180 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
 181
 182 #define file_hashval(x) \
 183         hash_ptr(x, FILE_HASH_BITS)
 184 #define stateid_hashval(owner_id, file_id)  \
 185         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
 186
 187 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 188 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 189
 190 static struct nfs4_delegation *
 191 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 192 {
 193         struct nfs4_delegation *dp;
 194         struct nfs4_file *fp = stp->st_file;
 195         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
 196
 197         dprintk("NFSD alloc_init_deleg\n");
 198         if (fp->fi_had_conflict)
 199                 return NULL;
 200         if (num_delegations > max_delegations)
 201                 return NULL;
 202         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
 203         if (dp == NULL)
 204                 return dp;
 205         num_delegations++;
 206         INIT_LIST_HEAD(&dp->dl_perfile);
 207         INIT_LIST_HEAD(&dp->dl_perclnt);
 208         INIT_LIST_HEAD(&dp->dl_recall_lru);
 209         dp->dl_client = clp;
 210         get_nfs4_file(fp);
 211         dp->dl_file = fp;
 212         dp->dl_flock = NULL;
 213         get_file(stp->st_vfs_file);
 214         dp->dl_vfs_file = stp->st_vfs_file;
 215         dp->dl_type = type;
 216         dp->dl_recall.cbr_dp = NULL;
 217         dp->dl_recall.cbr_ident = cb->cb_ident;
 218         dp->dl_recall.cbr_trunc = 0;
 219         dp->dl_stateid.si_boot = boot_time;
 220         dp->dl_stateid.si_stateownerid = current_delegid++;
 221         dp->dl_stateid.si_fileid = 0;
 222         dp->dl_stateid.si_generation = 0;
 223         dp->dl_fhlen = current_fh->fh_handle.fh_size;
 224         memcpy(dp->dl_fhval, &current_fh->fh_handle.fh_base,
 225                         current_fh->fh_handle.fh_size);
 226         dp->dl_time = 0;
 227         atomic_set(&dp->dl_count, 1);
 228         list_add(&dp->dl_perfile, &fp->fi_delegations);
 229         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 230         return dp;
 231 }
 232
 233 void
 234 nfs4_put_delegation(struct nfs4_delegation *dp)
 235 {
 236         if (atomic_dec_and_test(&dp->dl_count)) {
 237                 dprintk("NFSD: freeing dp %p\n",dp);
 238                 put_nfs4_file(dp->dl_file);
 239                 kmem_cache_free(deleg_slab, dp);
 240                 num_delegations--;
 241         }
 242 }
 243
 244 /* Remove the associated file_lock first, then remove the delegation.
 245  * lease_modify() is called to remove the FS_LEASE file_lock from
 246  * the i_flock list, eventually calling nfsd's lock_manager
 247  * fl_release_callback.
 248  */
 249 static void
 250 nfs4_close_delegation(struct nfs4_delegation *dp)
 251 {
 252         struct file *filp = dp->dl_vfs_file;
 253
 254         dprintk("NFSD: close_delegation dp %p\n",dp);
 255         dp->dl_vfs_file = NULL;
 256         /* The following nfsd_close may not actually close the file,
 257          * but we want to remove the lease in any case. */
 258         if (dp->dl_flock)
 259                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
 260         nfsd_close(filp);
 261 }
 262
 263 /* Called under the state lock. */
 264 static void
 265 unhash_delegation(struct nfs4_delegation *dp)
 266 {
 267         list_del_init(&dp->dl_perfile);
 268         list_del_init(&dp->dl_perclnt);
 269         spin_lock(&recall_lock);
 270         list_del_init(&dp->dl_recall_lru);
 271         spin_unlock(&recall_lock);
 272         nfs4_close_delegation(dp);
 273         nfs4_put_delegation(dp);
 274 }
 275
 276 /*
 277  * SETCLIENTID state
 278  */
 279
 280 /* Hash tables for nfs4_clientid state */
 281 #define CLIENT_HASH_BITS                 4
 282 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 283 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 284
 285 #define clientid_hashval(id) \
 286         ((id) & CLIENT_HASH_MASK)
 287 #define clientstr_hashval(name) \
 288         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
 289 /*
 290  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 291  * used in reboot/reset lease grace period processing
 292  *
 293  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 294  * setclientid_confirmed info.
 295  *
 296  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 297  * setclientid info.
 298  *
 299  * client_lru holds client queue ordered by nfs4_client.cl_time
 300  * for lease renewal.
 301  *
 302  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 303  * for last close replay.
 304  */
 305 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 306 static int reclaim_str_hashtbl_size = 0;
 307 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 308 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 309 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 310 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 311 static struct list_head client_lru;
 312 static struct list_head close_lru;
 313
 314 static inline void
 315 renew_client(struct nfs4_client *clp)
 316 {
 317         /*
 318         * Move client to the end to the LRU list.
 319         */
 320         dprintk("renewing client (clientid %08x/%08x)\n",
 321                         clp->cl_clientid.cl_boot,
 322                         clp->cl_clientid.cl_id);
 323         list_move_tail(&clp->cl_lru, &client_lru);
 324         clp->cl_time = get_seconds();
 325 }
 326
 327 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 328 static int
 329 STALE_CLIENTID(clientid_t *clid)
 330 {
 331         if (clid->cl_boot == boot_time)
 332                 return 0;
 333         dprintk("NFSD stale clientid (%08x/%08x)\n",
 334                         clid->cl_boot, clid->cl_id);
 335         return 1;
 336 }
 337
 338 /*
 339  * XXX Should we use a slab cache ?
 340  * This type of memory management is somewhat inefficient, but we use it
 341  * anyway since SETCLIENTID is not a common operation.
 342  */
 343 static inline struct nfs4_client *
 344 alloc_client(struct xdr_netobj name)
 345 {
 346         struct nfs4_client *clp;
 347
 348         if ((clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL))!= NULL) {
 349                 if ((clp->cl_name.data = kmalloc(name.len, GFP_KERNEL)) != NULL) {
 350                         memcpy(clp->cl_name.data, name.data, name.len);
 351                         clp->cl_name.len = name.len;
 352                 }
 353                 else {
 354                         kfree(clp);
 355                         clp = NULL;
 356                 }
 357         }
 358         return clp;
 359 }
 360
 361 static inline void
 362 free_client(struct nfs4_client *clp)
 363 {
 364         if (clp->cl_cred.cr_group_info)
 365                 put_group_info(clp->cl_cred.cr_group_info);
 366         kfree(clp->cl_name.data);
 367         kfree(clp);
 368 }
 369
 370 void
 371 put_nfs4_client(struct nfs4_client *clp)
 372 {
 373         if (atomic_dec_and_test(&clp->cl_count))
 374                 free_client(clp);
 375 }
 376
 377 static void
 378 shutdown_callback_client(struct nfs4_client *clp)
 379 {
 380         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
 381
 382         /* shutdown rpc client, ending any outstanding recall rpcs */
 383         if (clnt) {
 384                 clp->cl_callback.cb_client = NULL;
 385                 rpc_shutdown_client(clnt);
 386         }
 387 }
 388
 389 static void
 390 expire_client(struct nfs4_client *clp)
 391 {
 392         struct nfs4_stateowner *sop;
 393         struct nfs4_delegation *dp;
 394         struct list_head reaplist;
 395
 396         dprintk("NFSD: expire_client cl_count %d\n",
 397                             atomic_read(&clp->cl_count));
 398
 399         shutdown_callback_client(clp);
 400
 401         INIT_LIST_HEAD(&reaplist);
 402         spin_lock(&recall_lock);
 403         while (!list_empty(&clp->cl_delegations)) {
 404                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
 405                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
 406                                 dp->dl_flock);
 407                 list_del_init(&dp->dl_perclnt);
 408                 list_move(&dp->dl_recall_lru, &reaplist);
 409         }
 410         spin_unlock(&recall_lock);
 411         while (!list_empty(&reaplist)) {
 412                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
 413                 list_del_init(&dp->dl_recall_lru);
 414                 unhash_delegation(dp);
 415         }
 416         list_del(&clp->cl_idhash);
 417         list_del(&clp->cl_strhash);
 418         list_del(&clp->cl_lru);
 419         while (!list_empty(&clp->cl_openowners)) {
 420                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
 421                 release_stateowner(sop);
 422         }
 423         put_nfs4_client(clp);
 424 }
 425
 426 static struct nfs4_client *
 427 create_client(struct xdr_netobj name, char *recdir) {
 428         struct nfs4_client *clp;
 429
 430         if (!(clp = alloc_client(name)))
 431                 goto out;
 432         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
 433         atomic_set(&clp->cl_count, 1);
 434         atomic_set(&clp->cl_callback.cb_set, 0);
 435         INIT_LIST_HEAD(&clp->cl_idhash);
 436         INIT_LIST_HEAD(&clp->cl_strhash);
 437         INIT_LIST_HEAD(&clp->cl_openowners);
 438         INIT_LIST_HEAD(&clp->cl_delegations);
 439         INIT_LIST_HEAD(&clp->cl_lru);
 440 out:
 441         return clp;
 442 }
 443
 444 static void
 445 copy_verf(struct nfs4_client *target, nfs4_verifier *source) {
 446         memcpy(target->cl_verifier.data, source->data, sizeof(target->cl_verifier.data));
 447 }
 448
 449 static void
 450 copy_clid(struct nfs4_client *target, struct nfs4_client *source) {
 451         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
 452         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
 453 }
 454
 455 static void
 456 copy_cred(struct svc_cred *target, struct svc_cred *source) {
 457
 458         target->cr_uid = source->cr_uid;
 459         target->cr_gid = source->cr_gid;
 460         target->cr_group_info = source->cr_group_info;
 461         get_group_info(target->cr_group_info);
 462 }
 463
 464 static inline int
 465 same_name(const char *n1, const char *n2)
 466 {
 467         return 0 == memcmp(n1, n2, HEXDIR_LEN);
 468 }
 469
 470 static int
 471 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
 472 {
 473         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
 474 }
 475
 476 static int
 477 same_clid(clientid_t *cl1, clientid_t *cl2)
 478 {
 479         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
 480 }
 481
 482 /* XXX what about NGROUP */
 483 static int
 484 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
 485 {
 486         return cr1->cr_uid == cr2->cr_uid;
 487 }
 488
 489 static void
 490 gen_clid(struct nfs4_client *clp) {
 491         clp->cl_clientid.cl_boot = boot_time;
 492         clp->cl_clientid.cl_id = current_clientid++;
 493 }
 494
 495 static void
 496 gen_confirm(struct nfs4_client *clp) {
 497         struct timespec         tv;
 498         u32 *                   p;
 499
 500         tv = CURRENT_TIME;
 501         p = (u32 *)clp->cl_confirm.data;
 502         *p++ = tv.tv_sec;
 503         *p++ = tv.tv_nsec;
 504 }
 505
 506 static int
 507 check_name(struct xdr_netobj name) {
 508
 509         if (name.len == 0)
 510                 return 0;
 511         if (name.len > NFS4_OPAQUE_LIMIT) {
 512                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
 513                 return 0;
 514         }
 515         return 1;
 516 }
 517
 518 static void
 519 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
 520 {
 521         unsigned int idhashval;
 522
 523         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
 524         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 525         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
 526         list_add_tail(&clp->cl_lru, &client_lru);
 527         clp->cl_time = get_seconds();
 528 }
 529
 530 static void
 531 move_to_confirmed(struct nfs4_client *clp)
 532 {
 533         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 534         unsigned int strhashval;
 535
 536         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
 537         list_del_init(&clp->cl_strhash);
 538         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
 539         strhashval = clientstr_hashval(clp->cl_recdir);
 540         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
 541         renew_client(clp);
 542 }
 543
 544 static struct nfs4_client *
 545 find_confirmed_client(clientid_t *clid)
 546 {
 547         struct nfs4_client *clp;
 548         unsigned int idhashval = clientid_hashval(clid->cl_id);
 549
 550         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
 551                 if (same_clid(&clp->cl_clientid, clid))
 552                         return clp;
 553         }
 554         return NULL;
 555 }
 556
 557 static struct nfs4_client *
 558 find_unconfirmed_client(clientid_t *clid)
 559 {
 560         struct nfs4_client *clp;
 561         unsigned int idhashval = clientid_hashval(clid->cl_id);
 562
 563         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
 564                 if (same_clid(&clp->cl_clientid, clid))
 565                         return clp;
 566         }
 567         return NULL;
 568 }
 569
 570 static struct nfs4_client *
 571 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
 572 {
 573         struct nfs4_client *clp;
 574
 575         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
 576                 if (same_name(clp->cl_recdir, dname))
 577                         return clp;
 578         }
 579         return NULL;
 580 }
 581
 582 static struct nfs4_client *
 583 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
 584 {
 585         struct nfs4_client *clp;
 586
 587         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
 588                 if (same_name(clp->cl_recdir, dname))
 589                         return clp;
 590         }
 591         return NULL;
 592 }
 593
 594 /* a helper function for parse_callback */
 595 static int
 596 parse_octet(unsigned int *lenp, char **addrp)
 597 {
 598         unsigned int len = *lenp;
 599         char *p = *addrp;
 600         int n = -1;
 601         char c;
 602
 603         for (;;) {
 604                 if (!len)
 605                         break;
 606                 len--;
 607                 c = *p++;
 608                 if (c == '.')
 609                         break;
 610                 if ((c < '0') || (c > '9')) {
 611                         n = -1;
 612                         break;
 613                 }
 614                 if (n < 0)
 615                         n = 0;
 616                 n = (n * 10) + (c - '0');
 617                 if (n > 255) {
 618                         n = -1;
 619                         break;
 620                 }
 621         }
 622         *lenp = len;
 623         *addrp = p;
 624         return n;
 625 }
 626
 627 /* parse and set the setclientid ipv4 callback address */
 628 static int
 629 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
 630 {
 631         int temp = 0;
 632         u32 cbaddr = 0;
 633         u16 cbport = 0;
 634         u32 addrlen = addr_len;
 635         char *addr = addr_val;
 636         int i, shift;
 637
 638         /* ipaddress */
 639         shift = 24;
 640         for(i = 4; i > 0  ; i--) {
 641                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 642                         return 0;
 643                 }
 644                 cbaddr |= (temp << shift);
 645                 if (shift > 0)
 646                 shift -= 8;
 647         }
 648         *cbaddrp = cbaddr;
 649
 650         /* port */
 651         shift = 8;
 652         for(i = 2; i > 0  ; i--) {
 653                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 654                         return 0;
 655                 }
 656                 cbport |= (temp << shift);
 657                 if (shift > 0)
 658                         shift -= 8;
 659         }
 660         *cbportp = cbport;
 661         return 1;
 662 }
 663
 664 static void
 665 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
 666 {
 667         struct nfs4_callback *cb = &clp->cl_callback;
 668
 669         /* Currently, we only support tcp for the callback channel */
 670         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
 671                 goto out_err;
 672
 673         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
 674                          &cb->cb_addr, &cb->cb_port)))
 675                 goto out_err;
 676         cb->cb_prog = se->se_callback_prog;
 677         cb->cb_ident = se->se_callback_ident;
 678         return;
 679 out_err:
 680         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
 681                 "will not receive delegations\n",
 682                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
 683
 684         return;
 685 }
 686
 687 /*
 688  * RFC 3010 has a complex implmentation description of processing a
 689  * SETCLIENTID request consisting of 5 bullets, labeled as
 690  * CASE0 - CASE4 below.
 691  *
 692  * NOTES:
 693  *      callback information will be processed in a future patch
 694  *
 695  *      an unconfirmed record is added when:
 696  *      NORMAL (part of CASE 4): there is no confirmed nor unconfirmed record.
 697  *      CASE 1: confirmed record found with matching name, principal,
 698  *              verifier, and clientid.
 699  *      CASE 2: confirmed record found with matching name, principal,
 700  *              and there is no unconfirmed record with matching
 701  *              name and principal
 702  *
 703  *      an unconfirmed record is replaced when:
 704  *      CASE 3: confirmed record found with matching name, principal,
 705  *              and an unconfirmed record is found with matching
 706  *              name, principal, and with clientid and
 707  *              confirm that does not match the confirmed record.
 708  *      CASE 4: there is no confirmed record with matching name and
 709  *              principal. there is an unconfirmed record with
 710  *              matching name, principal.
 711  *
 712  *      an unconfirmed record is deleted when:
 713  *      CASE 1: an unconfirmed record that matches input name, verifier,
 714  *              and confirmed clientid.
 715  *      CASE 4: any unconfirmed records with matching name and principal
 716  *              that exist after an unconfirmed record has been replaced
 717  *              as described above.
 718  *
 719  */
 720 __be32
 721 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 722                   struct nfsd4_setclientid *setclid)
 723 {
 724         struct sockaddr_in      *sin = svc_addr_in(rqstp);
 725         struct xdr_netobj       clname = {
 726                 .len = setclid->se_namelen,
 727                 .data = setclid->se_name,
 728         };
 729         nfs4_verifier           clverifier = setclid->se_verf;
 730         unsigned int            strhashval;
 731         struct nfs4_client      *conf, *unconf, *new;
 732         __be32                  status;
 733         char                    dname[HEXDIR_LEN];
 734
 735         if (!check_name(clname))
 736                 return nfserr_inval;
 737
 738         status = nfs4_make_rec_clidname(dname, &clname);
 739         if (status)
 740                 return status;
 741
 742         /*
 743          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 744          * We get here on a DRC miss.
 745          */
 746
 747         strhashval = clientstr_hashval(dname);
 748
 749         nfs4_lock_state();
 750         conf = find_confirmed_client_by_str(dname, strhashval);
 751         if (conf) {
 752                 /*
 753                  * CASE 0:
 754                  * clname match, confirmed, different principal
 755                  * or different ip_address
 756                  */
 757                 status = nfserr_clid_inuse;
 758                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
 759                                 || conf->cl_addr != sin->sin_addr.s_addr) {
 760                         dprintk("NFSD: setclientid: string in use by client"
 761                                 "at %u.%u.%u.%u\n", NIPQUAD(conf->cl_addr));
 762                         goto out;
 763                 }
 764         }
 765         unconf = find_unconfirmed_client_by_str(dname, strhashval);
 766         status = nfserr_resource;
 767         if (!conf) {
 768                 /*
 769                  * CASE 4:
 770                  * placed first, because it is the normal case.
 771                  */
 772                 if (unconf)
 773                         expire_client(unconf);
 774                 new = create_client(clname, dname);
 775                 if (new == NULL)
 776                         goto out;
 777                 gen_clid(new);
 778         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
 779                 /*
 780                  * CASE 1:
 781                  * cl_name match, confirmed, principal match
 782                  * verifier match: probable callback update
 783                  *
 784                  * remove any unconfirmed nfs4_client with
 785                  * matching cl_name, cl_verifier, and cl_clientid
 786                  *
 787                  * create and insert an unconfirmed nfs4_client with same
 788                  * cl_name, cl_verifier, and cl_clientid as existing
 789                  * nfs4_client,  but with the new callback info and a
 790                  * new cl_confirm
 791                  */
 792                 if (unconf) {
 793                         /* Note this is removing unconfirmed {*x***},
 794                          * which is stronger than RFC recommended {vxc**}.
 795                          * This has the advantage that there is at most
 796                          * one {*x***} in either list at any time.
 797                          */
 798                         expire_client(unconf);
 799                 }
 800                 new = create_client(clname, dname);
 801                 if (new == NULL)
 802                         goto out;
 803                 copy_clid(new, conf);
 804         } else if (!unconf) {
 805                 /*
 806                  * CASE 2:
 807                  * clname match, confirmed, principal match
 808                  * verfier does not match
 809                  * no unconfirmed. create a new unconfirmed nfs4_client
 810                  * using input clverifier, clname, and callback info
 811                  * and generate a new cl_clientid and cl_confirm.
 812                  */
 813                 new = create_client(clname, dname);
 814                 if (new == NULL)
 815                         goto out;
 816                 gen_clid(new);
 817         } else if (!same_verf(&conf->cl_confirm, &unconf->cl_confirm)) {
 818                 /*
 819                  * CASE3:
 820                  * confirmed found (name, principal match)
 821                  * confirmed verifier does not match input clverifier
 822                  *
 823                  * unconfirmed found (name match)
 824                  * confirmed->cl_confirm != unconfirmed->cl_confirm
 825                  *
 826                  * remove unconfirmed.
 827                  *
 828                  * create an unconfirmed nfs4_client
 829                  * with same cl_name as existing confirmed nfs4_client,
 830                  * but with new callback info, new cl_clientid,
 831                  * new cl_verifier and a new cl_confirm
 832                  */
 833                 expire_client(unconf);
 834                 new = create_client(clname, dname);
 835                 if (new == NULL)
 836                         goto out;
 837                 gen_clid(new);
 838         } else {
 839                 /* No cases hit !!! */
 840                 status = nfserr_inval;
 841                 goto out;
 842
 843         }
 844         copy_verf(new, &clverifier);
 845         new->cl_addr = sin->sin_addr.s_addr;
 846         copy_cred(&new->cl_cred, &rqstp->rq_cred);
 847         gen_confirm(new);
 848         gen_callback(new, setclid);
 849         add_to_unconfirmed(new, strhashval);
 850         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
 851         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
 852         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
 853         status = nfs_ok;
 854 out:
 855         nfs4_unlock_state();
 856         return status;
 857 }
 858
 859
 860 /*
 861  * RFC 3010 has a complex implmentation description of processing a
 862  * SETCLIENTID_CONFIRM request consisting of 4 bullets describing
 863  * processing on a DRC miss, labeled as CASE1 - CASE4 below.
 864  *
 865  * NOTE: callback information will be processed here in a future patch
 866  */
 867 __be32
 868 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
 869                          struct nfsd4_compound_state *cstate,
 870                          struct nfsd4_setclientid_confirm *setclientid_confirm)
 871 {
 872         struct sockaddr_in *sin = svc_addr_in(rqstp);
 873         struct nfs4_client *conf, *unconf;
 874         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
 875         clientid_t * clid = &setclientid_confirm->sc_clientid;
 876         __be32 status;
 877
 878         if (STALE_CLIENTID(clid))
 879                 return nfserr_stale_clientid;
 880         /*
 881          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 882          * We get here on a DRC miss.
 883          */
 884
 885         nfs4_lock_state();
 886
 887         conf = find_confirmed_client(clid);
 888         unconf = find_unconfirmed_client(clid);
 889
 890         status = nfserr_clid_inuse;
 891         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
 892                 goto out;
 893         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
 894                 goto out;
 895
 896         if ((conf && unconf) &&
 897             (same_verf(&unconf->cl_confirm, &confirm)) &&
 898             (same_verf(&conf->cl_verifier, &unconf->cl_verifier)) &&
 899             (same_name(conf->cl_recdir,unconf->cl_recdir))  &&
 900             (!same_verf(&conf->cl_confirm, &unconf->cl_confirm))) {
 901                 /* CASE 1:
 902                 * unconf record that matches input clientid and input confirm.
 903                 * conf record that matches input clientid.
 904                 * conf and unconf records match names, verifiers
 905                 */
 906                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
 907                         status = nfserr_clid_inuse;
 908                 else {
 909                         /* XXX: We just turn off callbacks until we can handle
 910                           * change request correctly. */
 911                         atomic_set(&conf->cl_callback.cb_set, 0);
 912                         gen_confirm(conf);
 913                         nfsd4_remove_clid_dir(unconf);
 914                         expire_client(unconf);
 915                         status = nfs_ok;
 916
 917                 }
 918         } else if ((conf && !unconf) ||
 919             ((conf && unconf) &&
 920              (!same_verf(&conf->cl_verifier, &unconf->cl_verifier) ||
 921               !same_name(conf->cl_recdir, unconf->cl_recdir)))) {
 922                 /* CASE 2:
 923                  * conf record that matches input clientid.
 924                  * if unconf record matches input clientid, then
 925                  * unconf->cl_name or unconf->cl_verifier don't match the
 926                  * conf record.
 927                  */
 928                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
 929                         status = nfserr_clid_inuse;
 930                 else
 931                         status = nfs_ok;
 932         } else if (!conf && unconf
 933                         && same_verf(&unconf->cl_confirm, &confirm)) {
 934                 /* CASE 3:
 935                  * conf record not found.
 936                  * unconf record found.
 937                  * unconf->cl_confirm matches input confirm
 938                  */
 939                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
 940                         status = nfserr_clid_inuse;
 941                 } else {
 942                         unsigned int hash =
 943                                 clientstr_hashval(unconf->cl_recdir);
 944                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
 945                                                                         hash);
 946                         if (conf) {
 947                                 nfsd4_remove_clid_dir(conf);
 948                                 expire_client(conf);
 949                         }
 950                         move_to_confirmed(unconf);
 951                         conf = unconf;
 952                         status = nfs_ok;
 953                 }
 954         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
 955             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
 956                                                                 &confirm)))) {
 957                 /* CASE 4:
 958                  * conf record not found, or if conf, conf->cl_confirm does not
 959                  * match input confirm.
 960                  * unconf record not found, or if unconf, unconf->cl_confirm
 961                  * does not match input confirm.
 962                  */
 963                 status = nfserr_stale_clientid;
 964         } else {
 965                 /* check that we have hit one of the cases...*/
 966                 status = nfserr_clid_inuse;
 967         }
 968 out:
 969         if (!status)
 970                 nfsd4_probe_callback(conf);
 971         nfs4_unlock_state();
 972         return status;
 973 }
 974
 975 /* OPEN Share state helper functions */
 976 static inline struct nfs4_file *
 977 alloc_init_file(struct inode *ino)
 978 {
 979         struct nfs4_file *fp;
 980         unsigned int hashval = file_hashval(ino);
 981
 982         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
 983         if (fp) {
 984                 kref_init(&fp->fi_ref);
 985                 INIT_LIST_HEAD(&fp->fi_hash);
 986                 INIT_LIST_HEAD(&fp->fi_stateids);
 987                 INIT_LIST_HEAD(&fp->fi_delegations);
 988                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
 989                 fp->fi_inode = igrab(ino);
 990                 fp->fi_id = current_fileid++;
 991                 fp->fi_had_conflict = false;
 992                 return fp;
 993         }
 994         return NULL;
 995 }
 996
 997 static void
 998 nfsd4_free_slab(struct kmem_cache **slab)
 999 {
1000         if (*slab == NULL)
1001                 return;
1002         kmem_cache_destroy(*slab);
1003         *slab = NULL;
1004 }
1005
1006 void
1007 nfsd4_free_slabs(void)
1008 {
1009         nfsd4_free_slab(&stateowner_slab);
1010         nfsd4_free_slab(&file_slab);
1011         nfsd4_free_slab(&stateid_slab);
1012         nfsd4_free_slab(&deleg_slab);
1013 }
1014
1015 static int
1016 nfsd4_init_slabs(void)
1017 {
1018         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1019                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1020         if (stateowner_slab == NULL)
1021                 goto out_nomem;
1022         file_slab = kmem_cache_create("nfsd4_files",
1023                         sizeof(struct nfs4_file), 0, 0, NULL);
1024         if (file_slab == NULL)
1025                 goto out_nomem;
1026         stateid_slab = kmem_cache_create("nfsd4_stateids",
1027                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1028         if (stateid_slab == NULL)
1029                 goto out_nomem;
1030         deleg_slab = kmem_cache_create("nfsd4_delegations",
1031                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1032         if (deleg_slab == NULL)
1033                 goto out_nomem;
1034         return 0;
1035 out_nomem:
1036         nfsd4_free_slabs();
1037         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1038         return -ENOMEM;
1039 }
1040
1041 void
1042 nfs4_free_stateowner(struct kref *kref)
1043 {
1044         struct nfs4_stateowner *sop =
1045                 container_of(kref, struct nfs4_stateowner, so_ref);
1046         kfree(sop->so_owner.data);
1047         kmem_cache_free(stateowner_slab, sop);
1048 }
1049
1050 static inline struct nfs4_stateowner *
1051 alloc_stateowner(struct xdr_netobj *owner)
1052 {
1053         struct nfs4_stateowner *sop;
1054
1055         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1056                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1057                         memcpy(sop->so_owner.data, owner->data, owner->len);
1058                         sop->so_owner.len = owner->len;
1059                         kref_init(&sop->so_ref);
1060                         return sop;
1061                 }
1062                 kmem_cache_free(stateowner_slab, sop);
1063         }
1064         return NULL;
1065 }
1066
1067 static struct nfs4_stateowner *
1068 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1069         struct nfs4_stateowner *sop;
1070         struct nfs4_replay *rp;
1071         unsigned int idhashval;
1072
1073         if (!(sop = alloc_stateowner(&open->op_owner)))
1074                 return NULL;
1075         idhashval = ownerid_hashval(current_ownerid);
1076         INIT_LIST_HEAD(&sop->so_idhash);
1077         INIT_LIST_HEAD(&sop->so_strhash);
1078         INIT_LIST_HEAD(&sop->so_perclient);
1079         INIT_LIST_HEAD(&sop->so_stateids);
1080         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1081         INIT_LIST_HEAD(&sop->so_close_lru);
1082         sop->so_time = 0;
1083         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1084         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1085         list_add(&sop->so_perclient, &clp->cl_openowners);
1086         sop->so_is_open_owner = 1;
1087         sop->so_id = current_ownerid++;
1088         sop->so_client = clp;
1089         sop->so_seqid = open->op_seqid;
1090         sop->so_confirmed = 0;
1091         rp = &sop->so_replay;
1092         rp->rp_status = nfserr_serverfault;
1093         rp->rp_buflen = 0;
1094         rp->rp_buf = rp->rp_ibuf;
1095         return sop;
1096 }
1097
1098 static void
1099 release_stateid_lockowners(struct nfs4_stateid *open_stp)
1100 {
1101         struct nfs4_stateowner *lock_sop;
1102
1103         while (!list_empty(&open_stp->st_lockowners)) {
1104                 lock_sop = list_entry(open_stp->st_lockowners.next,
1105                                 struct nfs4_stateowner, so_perstateid);
1106                 /* list_del(&open_stp->st_lockowners);  */
1107                 BUG_ON(lock_sop->so_is_open_owner);
1108                 release_stateowner(lock_sop);
1109         }
1110 }
1111
1112 static void
1113 unhash_stateowner(struct nfs4_stateowner *sop)
1114 {
1115         struct nfs4_stateid *stp;
1116
1117         list_del(&sop->so_idhash);
1118         list_del(&sop->so_strhash);
1119         if (sop->so_is_open_owner)
1120                 list_del(&sop->so_perclient);
1121         list_del(&sop->so_perstateid);
1122         while (!list_empty(&sop->so_stateids)) {
1123                 stp = list_entry(sop->so_stateids.next,
1124                         struct nfs4_stateid, st_perstateowner);
1125                 if (sop->so_is_open_owner)
1126                         release_stateid(stp, OPEN_STATE);
1127                 else
1128                         release_stateid(stp, LOCK_STATE);
1129         }
1130 }
1131
1132 static void
1133 release_stateowner(struct nfs4_stateowner *sop)
1134 {
1135         unhash_stateowner(sop);
1136         list_del(&sop->so_close_lru);
1137         nfs4_put_stateowner(sop);
1138 }
1139
1140 static inline void
1141 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1142         struct nfs4_stateowner *sop = open->op_stateowner;
1143         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1144
1145         INIT_LIST_HEAD(&stp->st_hash);
1146         INIT_LIST_HEAD(&stp->st_perstateowner);
1147         INIT_LIST_HEAD(&stp->st_lockowners);
1148         INIT_LIST_HEAD(&stp->st_perfile);
1149         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1150         list_add(&stp->st_perstateowner, &sop->so_stateids);
1151         list_add(&stp->st_perfile, &fp->fi_stateids);
1152         stp->st_stateowner = sop;
1153         get_nfs4_file(fp);
1154         stp->st_file = fp;
1155         stp->st_stateid.si_boot = boot_time;
1156         stp->st_stateid.si_stateownerid = sop->so_id;
1157         stp->st_stateid.si_fileid = fp->fi_id;
1158         stp->st_stateid.si_generation = 0;
1159         stp->st_access_bmap = 0;
1160         stp->st_deny_bmap = 0;
1161         __set_bit(open->op_share_access, &stp->st_access_bmap);
1162         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1163         stp->st_openstp = NULL;
1164 }
1165
1166 static void
1167 release_stateid(struct nfs4_stateid *stp, int flags)
1168 {
1169         struct file *filp = stp->st_vfs_file;
1170
1171         list_del(&stp->st_hash);
1172         list_del(&stp->st_perfile);
1173         list_del(&stp->st_perstateowner);
1174         if (flags & OPEN_STATE) {
1175                 release_stateid_lockowners(stp);
1176                 stp->st_vfs_file = NULL;
1177                 nfsd_close(filp);
1178         } else if (flags & LOCK_STATE)
1179                 locks_remove_posix(filp, (fl_owner_t) stp->st_stateowner);
1180         put_nfs4_file(stp->st_file);
1181         kmem_cache_free(stateid_slab, stp);
1182 }
1183
1184 static void
1185 move_to_close_lru(struct nfs4_stateowner *sop)
1186 {
1187         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1188
1189         list_move_tail(&sop->so_close_lru, &close_lru);
1190         sop->so_time = get_seconds();
1191 }
1192
1193 static int
1194 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1195                                                         clientid_t *clid)
1196 {
1197         return (sop->so_owner.len == owner->len) &&
1198                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1199                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1200 }
1201
1202 static struct nfs4_stateowner *
1203 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1204 {
1205         struct nfs4_stateowner *so = NULL;
1206
1207         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1208                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1209                         return so;
1210         }
1211         return NULL;
1212 }
1213
1214 /* search file_hashtbl[] for file */
1215 static struct nfs4_file *
1216 find_file(struct inode *ino)
1217 {
1218         unsigned int hashval = file_hashval(ino);
1219         struct nfs4_file *fp;
1220
1221         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1222                 if (fp->fi_inode == ino) {
1223                         get_nfs4_file(fp);
1224                         return fp;
1225                 }
1226         }
1227         return NULL;
1228 }
1229
1230 static int access_valid(u32 x)
1231 {
1232         return (x > 0 && x < 4);
1233 }
1234
1235 static int deny_valid(u32 x)
1236 {
1237         return (x >= 0 && x < 5);
1238 }
1239
1240 static void
1241 set_access(unsigned int *access, unsigned long bmap) {
1242         int i;
1243
1244         *access = 0;
1245         for (i = 1; i < 4; i++) {
1246                 if (test_bit(i, &bmap))
1247                         *access |= i;
1248         }
1249 }
1250
1251 static void
1252 set_deny(unsigned int *deny, unsigned long bmap) {
1253         int i;
1254
1255         *deny = 0;
1256         for (i = 0; i < 4; i++) {
1257                 if (test_bit(i, &bmap))
1258                         *deny |= i ;
1259         }
1260 }
1261
1262 static int
1263 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1264         unsigned int access, deny;
1265
1266         set_access(&access, stp->st_access_bmap);
1267         set_deny(&deny, stp->st_deny_bmap);
1268         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1269                 return 0;
1270         return 1;
1271 }
1272
1273 /*
1274  * Called to check deny when READ with all zero stateid or
1275  * WRITE with all zero or all one stateid
1276  */
1277 static __be32
1278 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1279 {
1280         struct inode *ino = current_fh->fh_dentry->d_inode;
1281         struct nfs4_file *fp;
1282         struct nfs4_stateid *stp;
1283         __be32 ret;
1284
1285         dprintk("NFSD: nfs4_share_conflict\n");
1286
1287         fp = find_file(ino);
1288         if (!fp)
1289                 return nfs_ok;
1290         ret = nfserr_locked;
1291         /* Search for conflicting share reservations */
1292         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1293                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1294                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1295                         goto out;
1296         }
1297         ret = nfs_ok;
1298 out:
1299         put_nfs4_file(fp);
1300         return ret;
1301 }
1302
1303 static inline void
1304 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1305 {
1306         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1307                 put_write_access(filp->f_path.dentry->d_inode);
1308                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1309         }
1310 }
1311
1312 /*
1313  * Recall a delegation
1314  */
1315 static int
1316 do_recall(void *__dp)
1317 {
1318         struct nfs4_delegation *dp = __dp;
1319
1320         dp->dl_file->fi_had_conflict = true;
1321         nfsd4_cb_recall(dp);
1322         return 0;
1323 }
1324
1325 /*
1326  * Spawn a thread to perform a recall on the delegation represented
1327  * by the lease (file_lock)
1328  *
1329  * Called from break_lease() with lock_kernel() held.
1330  * Note: we assume break_lease will only call this *once* for any given
1331  * lease.
1332  */
1333 static
1334 void nfsd_break_deleg_cb(struct file_lock *fl)
1335 {
1336         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1337         struct task_struct *t;
1338
1339         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1340         if (!dp)
1341                 return;
1342
1343         /* We're assuming the state code never drops its reference
1344          * without first removing the lease.  Since we're in this lease
1345          * callback (and since the lease code is serialized by the kernel
1346          * lock) we know the server hasn't removed the lease yet, we know
1347          * it's safe to take a reference: */
1348         atomic_inc(&dp->dl_count);
1349
1350         spin_lock(&recall_lock);
1351         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1352         spin_unlock(&recall_lock);
1353
1354         /* only place dl_time is set. protected by lock_kernel*/
1355         dp->dl_time = get_seconds();
1356
1357         /* XXX need to merge NFSD_LEASE_TIME with fs/locks.c:lease_break_time */
1358         fl->fl_break_time = jiffies + NFSD_LEASE_TIME * HZ;
1359
1360         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1361         if (IS_ERR(t)) {
1362                 struct nfs4_client *clp = dp->dl_client;
1363
1364                 printk(KERN_INFO "NFSD: Callback thread failed for "
1365                         "for client (clientid %08x/%08x)\n",
1366                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1367                 nfs4_put_delegation(dp);
1368         }
1369 }
1370
1371 /*
1372  * The file_lock is being reapd.
1373  *
1374  * Called by locks_free_lock() with lock_kernel() held.
1375  */
1376 static
1377 void nfsd_release_deleg_cb(struct file_lock *fl)
1378 {
1379         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1380
1381         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1382
1383         if (!(fl->fl_flags & FL_LEASE) || !dp)
1384                 return;
1385         dp->dl_flock = NULL;
1386 }
1387
1388 /*
1389  * Set the delegation file_lock back pointer.
1390  *
1391  * Called from setlease() with lock_kernel() held.
1392  */
1393 static
1394 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1395 {
1396         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1397
1398         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1399         if (!dp)
1400                 return;
1401         dp->dl_flock = new;
1402 }
1403
1404 /*
1405  * Called from setlease() with lock_kernel() held
1406  */
1407 static
1408 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1409 {
1410         struct nfs4_delegation *onlistd =
1411                 (struct nfs4_delegation *)onlist->fl_owner;
1412         struct nfs4_delegation *tryd =
1413                 (struct nfs4_delegation *)try->fl_owner;
1414
1415         if (onlist->fl_lmops != try->fl_lmops)
1416                 return 0;
1417
1418         return onlistd->dl_client == tryd->dl_client;
1419 }
1420
1421
1422 static
1423 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1424 {
1425         if (arg & F_UNLCK)
1426                 return lease_modify(onlist, arg);
1427         else
1428                 return -EAGAIN;
1429 }
1430
1431 static struct lock_manager_operations nfsd_lease_mng_ops = {
1432         .fl_break = nfsd_break_deleg_cb,
1433         .fl_release_private = nfsd_release_deleg_cb,
1434         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1435         .fl_mylease = nfsd_same_client_deleg_cb,
1436         .fl_change = nfsd_change_deleg_cb,
1437 };
1438
1439
1440 __be32
1441 nfsd4_process_open1(struct nfsd4_open *open)
1442 {
1443         clientid_t *clientid = &open->op_clientid;
1444         struct nfs4_client *clp = NULL;
1445         unsigned int strhashval;
1446         struct nfs4_stateowner *sop = NULL;
1447
1448         if (!check_name(open->op_owner))
1449                 return nfserr_inval;
1450
1451         if (STALE_CLIENTID(&open->op_clientid))
1452                 return nfserr_stale_clientid;
1453
1454         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1455         sop = find_openstateowner_str(strhashval, open);
1456         open->op_stateowner = sop;
1457         if (!sop) {
1458                 /* Make sure the client's lease hasn't expired. */
1459                 clp = find_confirmed_client(clientid);
1460                 if (clp == NULL)
1461                         return nfserr_expired;
1462                 goto renew;
1463         }
1464         if (!sop->so_confirmed) {
1465                 /* Replace unconfirmed owners without checking for replay. */
1466                 clp = sop->so_client;
1467                 release_stateowner(sop);
1468                 open->op_stateowner = NULL;
1469                 goto renew;
1470         }
1471         if (open->op_seqid == sop->so_seqid - 1) {
1472                 if (sop->so_replay.rp_buflen)
1473                         return nfserr_replay_me;
1474                 /* The original OPEN failed so spectacularly
1475                  * that we don't even have replay data saved!
1476                  * Therefore, we have no choice but to continue
1477                  * processing this OPEN; presumably, we'll
1478                  * fail again for the same reason.
1479                  */
1480                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1481                 goto renew;
1482         }
1483         if (open->op_seqid != sop->so_seqid)
1484                 return nfserr_bad_seqid;
1485 renew:
1486         if (open->op_stateowner == NULL) {
1487                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1488                 if (sop == NULL)
1489                         return nfserr_resource;
1490                 open->op_stateowner = sop;
1491         }
1492         list_del_init(&sop->so_close_lru);
1493         renew_client(sop->so_client);
1494         return nfs_ok;
1495 }
1496
1497 static inline __be32
1498 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1499 {
1500         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1501                 return nfserr_openmode;
1502         else
1503                 return nfs_ok;
1504 }
1505
1506 static struct nfs4_delegation *
1507 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1508 {
1509         struct nfs4_delegation *dp;
1510
1511         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1512                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1513                         return dp;
1514         }
1515         return NULL;
1516 }
1517
1518 static __be32
1519 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1520                 struct nfs4_delegation **dp)
1521 {
1522         int flags;
1523         __be32 status = nfserr_bad_stateid;
1524
1525         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1526         if (*dp == NULL)
1527                 goto out;
1528         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1529                                                 RD_STATE : WR_STATE;
1530         status = nfs4_check_delegmode(*dp, flags);
1531         if (status)
1532                 *dp = NULL;
1533 out:
1534         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1535                 return nfs_ok;
1536         if (status)
1537                 return status;
1538         open->op_stateowner->so_confirmed = 1;
1539         return nfs_ok;
1540 }
1541
1542 static __be32
1543 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1544 {
1545         struct nfs4_stateid *local;
1546         __be32 status = nfserr_share_denied;
1547         struct nfs4_stateowner *sop = open->op_stateowner;
1548
1549         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1550                 /* ignore lock owners */
1551                 if (local->st_stateowner->so_is_open_owner == 0)
1552                         continue;
1553                 /* remember if we have seen this open owner */
1554                 if (local->st_stateowner == sop)
1555                         *stpp = local;
1556                 /* check for conflicting share reservations */
1557                 if (!test_share(local, open))
1558                         goto out;
1559         }
1560         status = 0;
1561 out:
1562         return status;
1563 }
1564
1565 static inline struct nfs4_stateid *
1566 nfs4_alloc_stateid(void)
1567 {
1568         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1569 }
1570
1571 static __be32
1572 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1573                 struct nfs4_delegation *dp,
1574                 struct svc_fh *cur_fh, int flags)
1575 {
1576         struct nfs4_stateid *stp;
1577
1578         stp = nfs4_alloc_stateid();
1579         if (stp == NULL)
1580                 return nfserr_resource;
1581
1582         if (dp) {
1583                 get_file(dp->dl_vfs_file);
1584                 stp->st_vfs_file = dp->dl_vfs_file;
1585         } else {
1586                 __be32 status;
1587                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1588                                 &stp->st_vfs_file);
1589                 if (status) {
1590                         if (status == nfserr_dropit)
1591                                 status = nfserr_jukebox;
1592                         kmem_cache_free(stateid_slab, stp);
1593                         return status;
1594                 }
1595         }
1596         *stpp = stp;
1597         return 0;
1598 }
1599
1600 static inline __be32
1601 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1602                 struct nfsd4_open *open)
1603 {
1604         struct iattr iattr = {
1605                 .ia_valid = ATTR_SIZE,
1606                 .ia_size = 0,
1607         };
1608         if (!open->op_truncate)
1609                 return 0;
1610         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1611                 return nfserr_inval;
1612         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1613 }
1614
1615 static __be32
1616 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1617 {
1618         struct file *filp = stp->st_vfs_file;
1619         struct inode *inode = filp->f_path.dentry->d_inode;
1620         unsigned int share_access, new_writer;
1621         __be32 status;
1622
1623         set_access(&share_access, stp->st_access_bmap);
1624         new_writer = (~share_access) & open->op_share_access
1625                         & NFS4_SHARE_ACCESS_WRITE;
1626
1627         if (new_writer) {
1628                 int err = get_write_access(inode);
1629                 if (err)
1630                         return nfserrno(err);
1631         }
1632         status = nfsd4_truncate(rqstp, cur_fh, open);
1633         if (status) {
1634                 if (new_writer)
1635                         put_write_access(inode);
1636                 return status;
1637         }
1638         /* remember the open */
1639         filp->f_mode |= open->op_share_access;
1640         set_bit(open->op_share_access, &stp->st_access_bmap);
1641         set_bit(open->op_share_deny, &stp->st_deny_bmap);
1642
1643         return nfs_ok;
1644 }
1645
1646
1647 static void
1648 nfs4_set_claim_prev(struct nfsd4_open *open)
1649 {
1650         open->op_stateowner->so_confirmed = 1;
1651         open->op_stateowner->so_client->cl_firststate = 1;
1652 }
1653
1654 /*
1655  * Attempt to hand out a delegation.
1656  */
1657 static void
1658 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1659 {
1660         struct nfs4_delegation *dp;
1661         struct nfs4_stateowner *sop = stp->st_stateowner;
1662         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1663         struct file_lock fl, *flp = &fl;
1664         int status, flag = 0;
1665
1666         flag = NFS4_OPEN_DELEGATE_NONE;
1667         open->op_recall = 0;
1668         switch (open->op_claim_type) {
1669                 case NFS4_OPEN_CLAIM_PREVIOUS:
1670                         if (!atomic_read(&cb->cb_set))
1671                                 open->op_recall = 1;
1672                         flag = open->op_delegate_type;
1673                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1674                                 goto out;
1675                         break;
1676                 case NFS4_OPEN_CLAIM_NULL:
1677                         /* Let's not give out any delegations till everyone's
1678                          * had the chance to reclaim theirs.... */
1679                         if (nfs4_in_grace())
1680                                 goto out;
1681                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1682                                 goto out;
1683                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1684                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1685                         else
1686                                 flag = NFS4_OPEN_DELEGATE_READ;
1687                         break;
1688                 default:
1689                         goto out;
1690         }
1691
1692         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1693         if (dp == NULL) {
1694                 flag = NFS4_OPEN_DELEGATE_NONE;
1695                 goto out;
1696         }
1697         locks_init_lock(&fl);
1698         fl.fl_lmops = &nfsd_lease_mng_ops;
1699         fl.fl_flags = FL_LEASE;
1700         fl.fl_end = OFFSET_MAX;
1701         fl.fl_owner =  (fl_owner_t)dp;
1702         fl.fl_file = stp->st_vfs_file;
1703         fl.fl_pid = current->tgid;
1704
1705         /* vfs_setlease checks to see if delegation should be handed out.
1706          * the lock_manager callbacks fl_mylease and fl_change are used
1707          */
1708         if ((status = vfs_setlease(stp->st_vfs_file,
1709                 flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK, &flp))) {
1710                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1711                 unhash_delegation(dp);
1712                 flag = NFS4_OPEN_DELEGATE_NONE;
1713                 goto out;
1714         }
1715
1716         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1717
1718         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1719                      dp->dl_stateid.si_boot,
1720                      dp->dl_stateid.si_stateownerid,
1721                      dp->dl_stateid.si_fileid,
1722                      dp->dl_stateid.si_generation);
1723 out:
1724         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1725                         && flag == NFS4_OPEN_DELEGATE_NONE
1726                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1727                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1728         open->op_delegate_type = flag;
1729 }
1730
1731 /*
1732  * called with nfs4_lock_state() held.
1733  */
1734 __be32
1735 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1736 {
1737         struct nfs4_file *fp = NULL;
1738         struct inode *ino = current_fh->fh_dentry->d_inode;
1739         struct nfs4_stateid *stp = NULL;
1740         struct nfs4_delegation *dp = NULL;
1741         __be32 status;
1742
1743         status = nfserr_inval;
1744         if (!access_valid(open->op_share_access)
1745                         || !deny_valid(open->op_share_deny))
1746                 goto out;
1747         /*
1748          * Lookup file; if found, lookup stateid and check open request,
1749          * and check for delegations in the process of being recalled.
1750          * If not found, create the nfs4_file struct
1751          */
1752         fp = find_file(ino);
1753         if (fp) {
1754                 if ((status = nfs4_check_open(fp, open, &stp)))
1755                         goto out;
1756                 status = nfs4_check_deleg(fp, open, &dp);
1757                 if (status)
1758                         goto out;
1759         } else {
1760                 status = nfserr_bad_stateid;
1761                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1762                         goto out;
1763                 status = nfserr_resource;
1764                 fp = alloc_init_file(ino);
1765                 if (fp == NULL)
1766                         goto out;
1767         }
1768
1769         /*
1770          * OPEN the file, or upgrade an existing OPEN.
1771          * If truncate fails, the OPEN fails.
1772          */
1773         if (stp) {
1774                 /* Stateid was found, this is an OPEN upgrade */
1775                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1776                 if (status)
1777                         goto out;
1778                 update_stateid(&stp->st_stateid);
1779         } else {
1780                 /* Stateid was not found, this is a new OPEN */
1781                 int flags = 0;
1782                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1783                         flags |= MAY_READ;
1784                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1785                         flags |= MAY_WRITE;
1786                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1787                 if (status)
1788                         goto out;
1789                 init_stateid(stp, fp, open);
1790                 status = nfsd4_truncate(rqstp, current_fh, open);
1791                 if (status) {
1792                         release_stateid(stp, OPEN_STATE);
1793                         goto out;
1794                 }
1795         }
1796         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1797
1798         /*
1799         * Attempt to hand out a delegation. No error return, because the
1800         * OPEN succeeds even if we fail.
1801         */
1802         nfs4_open_delegation(current_fh, open, stp);
1803
1804         status = nfs_ok;
1805
1806         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1807                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1808                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1809 out:
1810         if (fp)
1811                 put_nfs4_file(fp);
1812         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1813                 nfs4_set_claim_prev(open);
1814         /*
1815         * To finish the open response, we just need to set the rflags.
1816         */
1817         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1818         if (!open->op_stateowner->so_confirmed)
1819                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1820
1821         return status;
1822 }
1823
1824 static struct workqueue_struct *laundry_wq;
1825 static void laundromat_main(struct work_struct *);
1826 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1827
1828 __be32
1829 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1830             clientid_t *clid)
1831 {
1832         struct nfs4_client *clp;
1833         __be32 status;
1834
1835         nfs4_lock_state();
1836         dprintk("process_renew(%08x/%08x): starting\n",
1837                         clid->cl_boot, clid->cl_id);
1838         status = nfserr_stale_clientid;
1839         if (STALE_CLIENTID(clid))
1840                 goto out;
1841         clp = find_confirmed_client(clid);
1842         status = nfserr_expired;
1843         if (clp == NULL) {
1844                 /* We assume the client took too long to RENEW. */
1845                 dprintk("nfsd4_renew: clientid not found!\n");
1846                 goto out;
1847         }
1848         renew_client(clp);
1849         status = nfserr_cb_path_down;
1850         if (!list_empty(&clp->cl_delegations)
1851                         && !atomic_read(&clp->cl_callback.cb_set))
1852                 goto out;
1853         status = nfs_ok;
1854 out:
1855         nfs4_unlock_state();
1856         return status;
1857 }
1858
1859 static void
1860 end_grace(void)
1861 {
1862         dprintk("NFSD: end of grace period\n");
1863         nfsd4_recdir_purge_old();
1864         in_grace = 0;
1865 }
1866
1867 static time_t
1868 nfs4_laundromat(void)
1869 {
1870         struct nfs4_client *clp;
1871         struct nfs4_stateowner *sop;
1872         struct nfs4_delegation *dp;
1873         struct list_head *pos, *next, reaplist;
1874         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1875         time_t t, clientid_val = NFSD_LEASE_TIME;
1876         time_t u, test_val = NFSD_LEASE_TIME;
1877
1878         nfs4_lock_state();
1879
1880         dprintk("NFSD: laundromat service - starting\n");
1881         if (in_grace)
1882                 end_grace();
1883         list_for_each_safe(pos, next, &client_lru) {
1884                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1885                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1886                         t = clp->cl_time - cutoff;
1887                         if (clientid_val > t)
1888                                 clientid_val = t;
1889                         break;
1890                 }
1891                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1892                         clp->cl_clientid.cl_id);
1893                 nfsd4_remove_clid_dir(clp);
1894                 expire_client(clp);
1895         }
1896         INIT_LIST_HEAD(&reaplist);
1897         spin_lock(&recall_lock);
1898         list_for_each_safe(pos, next, &del_recall_lru) {
1899                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1900                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1901                         u = dp->dl_time - cutoff;
1902                         if (test_val > u)
1903                                 test_val = u;
1904                         break;
1905                 }
1906                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1907                                     dp, dp->dl_flock);
1908                 list_move(&dp->dl_recall_lru, &reaplist);
1909         }
1910         spin_unlock(&recall_lock);
1911         list_for_each_safe(pos, next, &reaplist) {
1912                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1913                 list_del_init(&dp->dl_recall_lru);
1914                 unhash_delegation(dp);
1915         }
1916         test_val = NFSD_LEASE_TIME;
1917         list_for_each_safe(pos, next, &close_lru) {
1918                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1919                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1920                         u = sop->so_time - cutoff;
1921                         if (test_val > u)
1922                                 test_val = u;
1923                         break;
1924                 }
1925                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1926                         sop->so_id);
1927                 release_stateowner(sop);
1928         }
1929         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1930                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1931         nfs4_unlock_state();
1932         return clientid_val;
1933 }
1934
1935 void
1936 laundromat_main(struct work_struct *not_used)
1937 {
1938         time_t t;
1939
1940         t = nfs4_laundromat();
1941         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1942         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1943 }
1944
1945 static struct nfs4_stateowner *
1946 search_close_lru(u32 st_id, int flags)
1947 {
1948         struct nfs4_stateowner *local = NULL;
1949
1950         if (flags & CLOSE_STATE) {
1951                 list_for_each_entry(local, &close_lru, so_close_lru) {
1952                         if (local->so_id == st_id)
1953                                 return local;
1954                 }
1955         }
1956         return NULL;
1957 }
1958
1959 static inline int
1960 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1961 {
1962         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1963 }
1964
1965 static int
1966 STALE_STATEID(stateid_t *stateid)
1967 {
1968         if (stateid->si_boot == boot_time)
1969                 return 0;
1970         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1971                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1972                 stateid->si_generation);
1973         return 1;
1974 }
1975
1976 static inline int
1977 access_permit_read(unsigned long access_bmap)
1978 {
1979         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1980                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1981                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1982 }
1983
1984 static inline int
1985 access_permit_write(unsigned long access_bmap)
1986 {
1987         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1988                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1989 }
1990
1991 static
1992 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1993 {
1994         __be32 status = nfserr_openmode;
1995
1996         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1997                 goto out;
1998         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1999                 goto out;
2000         status = nfs_ok;
2001 out:
2002         return status;
2003 }
2004
2005 static inline __be32
2006 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2007 {
2008         /* Trying to call delegreturn with a special stateid? Yuch: */
2009         if (!(flags & (RD_STATE | WR_STATE)))
2010                 return nfserr_bad_stateid;
2011         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
2012                 return nfs_ok;
2013         else if (nfs4_in_grace()) {
2014                 /* Answer in remaining cases depends on existance of
2015                  * conflicting state; so we must wait out the grace period. */
2016                 return nfserr_grace;
2017         } else if (flags & WR_STATE)
2018                 return nfs4_share_conflict(current_fh,
2019                                 NFS4_SHARE_DENY_WRITE);
2020         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2021                 return nfs4_share_conflict(current_fh,
2022                                 NFS4_SHARE_DENY_READ);
2023 }
2024
2025 /*
2026  * Allow READ/WRITE during grace period on recovered state only for files
2027  * that are not able to provide mandatory locking.
2028  */
2029 static inline int
2030 io_during_grace_disallowed(struct inode *inode, int flags)
2031 {
2032         return nfs4_in_grace() && (flags & (RD_STATE | WR_STATE))
2033                 && MANDATORY_LOCK(inode);
2034 }
2035
2036 /*
2037 * Checks for stateid operations
2038 */
2039 __be32
2040 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2041 {
2042         struct nfs4_stateid *stp = NULL;
2043         struct nfs4_delegation *dp = NULL;
2044         stateid_t *stidp;
2045         struct inode *ino = current_fh->fh_dentry->d_inode;
2046         __be32 status;
2047
2048         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2049                 stateid->si_boot, stateid->si_stateownerid,
2050                 stateid->si_fileid, stateid->si_generation);
2051         if (filpp)
2052                 *filpp = NULL;
2053
2054         if (io_during_grace_disallowed(ino, flags))
2055                 return nfserr_grace;
2056
2057         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2058                 return check_special_stateids(current_fh, stateid, flags);
2059
2060         /* STALE STATEID */
2061         status = nfserr_stale_stateid;
2062         if (STALE_STATEID(stateid))
2063                 goto out;
2064
2065         /* BAD STATEID */
2066         status = nfserr_bad_stateid;
2067         if (!stateid->si_fileid) { /* delegation stateid */
2068                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2069                         dprintk("NFSD: delegation stateid not found\n");
2070                         goto out;
2071                 }
2072                 stidp = &dp->dl_stateid;
2073         } else { /* open or lock stateid */
2074                 if (!(stp = find_stateid(stateid, flags))) {
2075                         dprintk("NFSD: open or lock stateid not found\n");
2076                         goto out;
2077                 }
2078                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2079                         goto out;
2080                 if (!stp->st_stateowner->so_confirmed)
2081                         goto out;
2082                 stidp = &stp->st_stateid;
2083         }
2084         if (stateid->si_generation > stidp->si_generation)
2085                 goto out;
2086
2087         /* OLD STATEID */
2088         status = nfserr_old_stateid;
2089         if (stateid->si_generation < stidp->si_generation)
2090                 goto out;
2091         if (stp) {
2092                 if ((status = nfs4_check_openmode(stp,flags)))
2093                         goto out;
2094                 renew_client(stp->st_stateowner->so_client);
2095                 if (filpp)
2096                         *filpp = stp->st_vfs_file;
2097         } else if (dp) {
2098                 if ((status = nfs4_check_delegmode(dp, flags)))
2099                         goto out;
2100                 renew_client(dp->dl_client);
2101                 if (flags & DELEG_RET)
2102                         unhash_delegation(dp);
2103                 if (filpp)
2104                         *filpp = dp->dl_vfs_file;
2105         }
2106         status = nfs_ok;
2107 out:
2108         return status;
2109 }
2110
2111 static inline int
2112 setlkflg (int type)
2113 {
2114         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2115                 RD_STATE : WR_STATE;
2116 }
2117
2118 /*
2119  * Checks for sequence id mutating operations.
2120  */
2121 static __be32
2122 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)
2123 {
2124         struct nfs4_stateid *stp;
2125         struct nfs4_stateowner *sop;
2126
2127         dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2128                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2129                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2130                 stateid->si_generation);
2131
2132         *stpp = NULL;
2133         *sopp = NULL;
2134
2135         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2136                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2137                 return nfserr_bad_stateid;
2138         }
2139
2140         if (STALE_STATEID(stateid))
2141                 return nfserr_stale_stateid;
2142         /*
2143         * We return BAD_STATEID if filehandle doesn't match stateid,
2144         * the confirmed flag is incorrecly set, or the generation
2145         * number is incorrect.
2146         */
2147         stp = find_stateid(stateid, flags);
2148         if (stp == NULL) {
2149                 /*
2150                  * Also, we should make sure this isn't just the result of
2151                  * a replayed close:
2152                  */
2153                 sop = search_close_lru(stateid->si_stateownerid, flags);
2154                 if (sop == NULL)
2155                         return nfserr_bad_stateid;
2156                 *sopp = sop;
2157                 goto check_replay;
2158         }
2159
2160         if (lock) {
2161                 struct nfs4_stateowner *sop = stp->st_stateowner;
2162                 clientid_t *lockclid = &lock->v.new.clientid;
2163                 struct nfs4_client *clp = sop->so_client;
2164                 int lkflg = 0;
2165                 __be32 status;
2166
2167                 lkflg = setlkflg(lock->lk_type);
2168
2169                 if (lock->lk_is_new) {
2170                         if (!sop->so_is_open_owner)
2171                                 return nfserr_bad_stateid;
2172                         if (!same_clid(&clp->cl_clientid, lockclid))
2173                                return nfserr_bad_stateid;
2174                         /* stp is the open stateid */
2175                         status = nfs4_check_openmode(stp, lkflg);
2176                         if (status)
2177                                 return status;
2178                 } else {
2179                         /* stp is the lock stateid */
2180                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2181                         if (status)
2182                                 return status;
2183                }
2184         }
2185
2186         if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp)) {
2187                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2188                 return nfserr_bad_stateid;
2189         }
2190
2191         *stpp = stp;
2192         *sopp = sop = stp->st_stateowner;
2193
2194         /*
2195         *  We now validate the seqid and stateid generation numbers.
2196         *  For the moment, we ignore the possibility of
2197         *  generation number wraparound.
2198         */
2199         if (seqid != sop->so_seqid)
2200                 goto check_replay;
2201
2202         if (sop->so_confirmed && flags & CONFIRM) {
2203                 dprintk("NFSD: preprocess_seqid_op: expected"
2204                                 " unconfirmed stateowner!\n");
2205                 return nfserr_bad_stateid;
2206         }
2207         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2208                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2209                                 " confirmed yet!\n");
2210                 return nfserr_bad_stateid;
2211         }
2212         if (stateid->si_generation > stp->st_stateid.si_generation) {
2213                 dprintk("NFSD: preprocess_seqid_op: future stateid?!\n");
2214                 return nfserr_bad_stateid;
2215         }
2216
2217         if (stateid->si_generation < stp->st_stateid.si_generation) {
2218                 dprintk("NFSD: preprocess_seqid_op: old stateid!\n");
2219                 return nfserr_old_stateid;
2220         }
2221         renew_client(sop->so_client);
2222         return nfs_ok;
2223
2224 check_replay:
2225         if (seqid == sop->so_seqid - 1) {
2226                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2227                 /* indicate replay to calling function */
2228                 return nfserr_replay_me;
2229         }
2230         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2231                         sop->so_seqid, seqid);
2232         *sopp = NULL;
2233         return nfserr_bad_seqid;
2234 }
2235
2236 __be32
2237 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2238                    struct nfsd4_open_confirm *oc)
2239 {
2240         __be32 status;
2241         struct nfs4_stateowner *sop;
2242         struct nfs4_stateid *stp;
2243
2244         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2245                         (int)cstate->current_fh.fh_dentry->d_name.len,
2246                         cstate->current_fh.fh_dentry->d_name.name);
2247
2248         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2249         if (status)
2250                 return status;
2251
2252         nfs4_lock_state();
2253
2254         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2255                                         oc->oc_seqid, &oc->oc_req_stateid,
2256                                         CHECK_FH | CONFIRM | OPEN_STATE,
2257                                         &oc->oc_stateowner, &stp, NULL)))
2258                 goto out;
2259
2260         sop = oc->oc_stateowner;
2261         sop->so_confirmed = 1;
2262         update_stateid(&stp->st_stateid);
2263         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2264         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2265                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2266                          stp->st_stateid.si_boot,
2267                          stp->st_stateid.si_stateownerid,
2268                          stp->st_stateid.si_fileid,
2269                          stp->st_stateid.si_generation);
2270
2271         nfsd4_create_clid_dir(sop->so_client);
2272 out:
2273         if (oc->oc_stateowner) {
2274                 nfs4_get_stateowner(oc->oc_stateowner);
2275                 cstate->replay_owner = oc->oc_stateowner;
2276         }
2277         nfs4_unlock_state();
2278         return status;
2279 }
2280
2281
2282 /*
2283  * unset all bits in union bitmap (bmap) that
2284  * do not exist in share (from successful OPEN_DOWNGRADE)
2285  */
2286 static void
2287 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2288 {
2289         int i;
2290         for (i = 1; i < 4; i++) {
2291                 if ((i & access) != i)
2292                         __clear_bit(i, bmap);
2293         }
2294 }
2295
2296 static void
2297 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2298 {
2299         int i;
2300         for (i = 0; i < 4; i++) {
2301                 if ((i & deny) != i)
2302                         __clear_bit(i, bmap);
2303         }
2304 }
2305
2306 __be32
2307 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2308                      struct nfsd4_compound_state *cstate,
2309                      struct nfsd4_open_downgrade *od)
2310 {
2311         __be32 status;
2312         struct nfs4_stateid *stp;
2313         unsigned int share_access;
2314
2315         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2316                         (int)cstate->current_fh.fh_dentry->d_name.len,
2317                         cstate->current_fh.fh_dentry->d_name.name);
2318
2319         if (!access_valid(od->od_share_access)
2320                         || !deny_valid(od->od_share_deny))
2321                 return nfserr_inval;
2322
2323         nfs4_lock_state();
2324         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2325                                         od->od_seqid,
2326                                         &od->od_stateid,
2327                                         CHECK_FH | OPEN_STATE,
2328                                         &od->od_stateowner, &stp, NULL)))
2329                 goto out;
2330
2331         status = nfserr_inval;
2332         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2333                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2334                         stp->st_access_bmap, od->od_share_access);
2335                 goto out;
2336         }
2337         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2338                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2339                         stp->st_deny_bmap, od->od_share_deny);
2340                 goto out;
2341         }
2342         set_access(&share_access, stp->st_access_bmap);
2343         nfs4_file_downgrade(stp->st_vfs_file,
2344                             share_access & ~od->od_share_access);
2345
2346         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2347         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2348
2349         update_stateid(&stp->st_stateid);
2350         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2351         status = nfs_ok;
2352 out:
2353         if (od->od_stateowner) {
2354                 nfs4_get_stateowner(od->od_stateowner);
2355                 cstate->replay_owner = od->od_stateowner;
2356         }
2357         nfs4_unlock_state();
2358         return status;
2359 }
2360
2361 /*
2362  * nfs4_unlock_state() called after encode
2363  */
2364 __be32
2365 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2366             struct nfsd4_close *close)
2367 {
2368         __be32 status;
2369         struct nfs4_stateid *stp;
2370
2371         dprintk("NFSD: nfsd4_close on file %.*s\n",
2372                         (int)cstate->current_fh.fh_dentry->d_name.len,
2373                         cstate->current_fh.fh_dentry->d_name.name);
2374
2375         nfs4_lock_state();
2376         /* check close_lru for replay */
2377         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2378                                         close->cl_seqid,
2379                                         &close->cl_stateid,
2380                                         CHECK_FH | OPEN_STATE | CLOSE_STATE,
2381                                         &close->cl_stateowner, &stp, NULL)))
2382                 goto out;
2383         status = nfs_ok;
2384         update_stateid(&stp->st_stateid);
2385         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2386
2387         /* release_stateid() calls nfsd_close() if needed */
2388         release_stateid(stp, OPEN_STATE);
2389
2390         /* place unused nfs4_stateowners on so_close_lru list to be
2391          * released by the laundromat service after the lease period
2392          * to enable us to handle CLOSE replay
2393          */
2394         if (list_empty(&close->cl_stateowner->so_stateids))
2395                 move_to_close_lru(close->cl_stateowner);
2396 out:
2397         if (close->cl_stateowner) {
2398                 nfs4_get_stateowner(close->cl_stateowner);
2399                 cstate->replay_owner = close->cl_stateowner;
2400         }
2401         nfs4_unlock_state();
2402         return status;
2403 }
2404
2405 __be32
2406 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2407                   struct nfsd4_delegreturn *dr)
2408 {
2409         __be32 status;
2410
2411         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2412                 goto out;
2413
2414         nfs4_lock_state();
2415         status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2416                                             &dr->dr_stateid, DELEG_RET, NULL);
2417         nfs4_unlock_state();
2418 out:
2419         return status;
2420 }
2421
2422
2423 /*
2424  * Lock owner state (byte-range locks)
2425  */
2426 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2427 #define LOCK_HASH_BITS              8
2428 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2429 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2430
2431 #define lockownerid_hashval(id) \
2432         ((id) & LOCK_HASH_MASK)
2433
2434 static inline unsigned int
2435 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2436                 struct xdr_netobj *ownername)
2437 {
2438         return (file_hashval(inode) + cl_id
2439                         + opaque_hashval(ownername->data, ownername->len))
2440                 & LOCK_HASH_MASK;
2441 }
2442
2443 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2444 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2445 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2446
2447 static struct nfs4_stateid *
2448 find_stateid(stateid_t *stid, int flags)
2449 {
2450         struct nfs4_stateid *local = NULL;
2451         u32 st_id = stid->si_stateownerid;
2452         u32 f_id = stid->si_fileid;
2453         unsigned int hashval;
2454
2455         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2456         if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2457                 hashval = stateid_hashval(st_id, f_id);
2458                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2459                         if ((local->st_stateid.si_stateownerid == st_id) &&
2460                             (local->st_stateid.si_fileid == f_id))
2461                                 return local;
2462                 }
2463         }
2464         if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2465                 hashval = stateid_hashval(st_id, f_id);
2466                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2467                         if ((local->st_stateid.si_stateownerid == st_id) &&
2468                             (local->st_stateid.si_fileid == f_id))
2469                                 return local;
2470                 }
2471         }
2472         return NULL;
2473 }
2474
2475 static struct nfs4_delegation *
2476 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2477 {
2478         struct nfs4_file *fp;
2479         struct nfs4_delegation *dl;
2480
2481         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2482                     stid->si_boot, stid->si_stateownerid,
2483                     stid->si_fileid, stid->si_generation);
2484
2485         fp = find_file(ino);
2486         if (!fp)
2487                 return NULL;
2488         dl = find_delegation_file(fp, stid);
2489         put_nfs4_file(fp);
2490         return dl;
2491 }
2492
2493 /*
2494  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2495  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2496  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2497  * locking, this prevents us from being completely protocol-compliant.  The
2498  * real solution to this problem is to start using unsigned file offsets in
2499  * the VFS, but this is a very deep change!
2500  */
2501 static inline void
2502 nfs4_transform_lock_offset(struct file_lock *lock)
2503 {
2504         if (lock->fl_start < 0)
2505                 lock->fl_start = OFFSET_MAX;
2506         if (lock->fl_end < 0)
2507                 lock->fl_end = OFFSET_MAX;
2508 }
2509
2510 /* Hack!: For now, we're defining this just so we can use a pointer to it
2511  * as a unique cookie to identify our (NFSv4's) posix locks. */
2512 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2513 };
2514
2515 static inline void
2516 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2517 {
2518         struct nfs4_stateowner *sop;
2519         unsigned int hval;
2520
2521         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2522                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2523                 hval = lockownerid_hashval(sop->so_id);
2524                 kref_get(&sop->so_ref);
2525                 deny->ld_sop = sop;
2526                 deny->ld_clientid = sop->so_client->cl_clientid;
2527         } else {
2528                 deny->ld_sop = NULL;
2529                 deny->ld_clientid.cl_boot = 0;
2530                 deny->ld_clientid.cl_id = 0;
2531         }
2532         deny->ld_start = fl->fl_start;
2533         deny->ld_length = ~(u64)0;
2534         if (fl->fl_end != ~(u64)0)
2535                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2536         deny->ld_type = NFS4_READ_LT;
2537         if (fl->fl_type != F_RDLCK)
2538                 deny->ld_type = NFS4_WRITE_LT;
2539 }
2540
2541 static struct nfs4_stateowner *
2542 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2543                 struct xdr_netobj *owner)
2544 {
2545         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2546         struct nfs4_stateowner *op;
2547
2548         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2549                 if (same_owner_str(op, owner, clid))
2550                         return op;
2551         }
2552         return NULL;
2553 }
2554
2555 /*
2556  * Alloc a lock owner structure.
2557  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2558  * occured.
2559  *
2560  * strhashval = lock_ownerstr_hashval
2561  */
2562
2563 static struct nfs4_stateowner *
2564 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2565         struct nfs4_stateowner *sop;
2566         struct nfs4_replay *rp;
2567         unsigned int idhashval;
2568
2569         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2570                 return NULL;
2571         idhashval = lockownerid_hashval(current_ownerid);
2572         INIT_LIST_HEAD(&sop->so_idhash);
2573         INIT_LIST_HEAD(&sop->so_strhash);
2574         INIT_LIST_HEAD(&sop->so_perclient);
2575         INIT_LIST_HEAD(&sop->so_stateids);
2576         INIT_LIST_HEAD(&sop->so_perstateid);
2577         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2578         sop->so_time = 0;
2579         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2580         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2581         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2582         sop->so_is_open_owner = 0;
2583         sop->so_id = current_ownerid++;
2584         sop->so_client = clp;
2585         /* It is the openowner seqid that will be incremented in encode in the
2586          * case of new lockowners; so increment the lock seqid manually: */
2587         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2588         sop->so_confirmed = 1;
2589         rp = &sop->so_replay;
2590         rp->rp_status = nfserr_serverfault;
2591         rp->rp_buflen = 0;
2592         rp->rp_buf = rp->rp_ibuf;
2593         return sop;
2594 }
2595
2596 static struct nfs4_stateid *
2597 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2598 {
2599         struct nfs4_stateid *stp;
2600         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2601
2602         stp = nfs4_alloc_stateid();
2603         if (stp == NULL)
2604                 goto out;
2605         INIT_LIST_HEAD(&stp->st_hash);
2606         INIT_LIST_HEAD(&stp->st_perfile);
2607         INIT_LIST_HEAD(&stp->st_perstateowner);
2608         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2609         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2610         list_add(&stp->st_perfile, &fp->fi_stateids);
2611         list_add(&stp->st_perstateowner, &sop->so_stateids);
2612         stp->st_stateowner = sop;
2613         get_nfs4_file(fp);
2614         stp->st_file = fp;
2615         stp->st_stateid.si_boot = boot_time;
2616         stp->st_stateid.si_stateownerid = sop->so_id;
2617         stp->st_stateid.si_fileid = fp->fi_id;
2618         stp->st_stateid.si_generation = 0;
2619         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2620         stp->st_access_bmap = open_stp->st_access_bmap;
2621         stp->st_deny_bmap = open_stp->st_deny_bmap;
2622         stp->st_openstp = open_stp;
2623
2624 out:
2625         return stp;
2626 }
2627
2628 static int
2629 check_lock_length(u64 offset, u64 length)
2630 {
2631         return ((length == 0)  || ((length != ~(u64)0) &&
2632              LOFF_OVERFLOW(offset, length)));
2633 }
2634
2635 /*
2636  *  LOCK operation
2637  */
2638 __be32
2639 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2640            struct nfsd4_lock *lock)
2641 {
2642         struct nfs4_stateowner *open_sop = NULL;
2643         struct nfs4_stateowner *lock_sop = NULL;
2644         struct nfs4_stateid *lock_stp;
2645         struct file *filp;
2646         struct file_lock file_lock;
2647         struct file_lock conflock;
2648         __be32 status = 0;
2649         unsigned int strhashval;
2650         unsigned int cmd;
2651         int err;
2652
2653         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2654                 (long long) lock->lk_offset,
2655                 (long long) lock->lk_length);
2656
2657         if (check_lock_length(lock->lk_offset, lock->lk_length))
2658                  return nfserr_inval;
2659
2660         if ((status = fh_verify(rqstp, &cstate->current_fh,
2661                                 S_IFREG, MAY_LOCK))) {
2662                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2663                 return status;
2664         }
2665
2666         nfs4_lock_state();
2667
2668         if (lock->lk_is_new) {
2669                 /*
2670                  * Client indicates that this is a new lockowner.
2671                  * Use open owner and open stateid to create lock owner and
2672                  * lock stateid.
2673                  */
2674                 struct nfs4_stateid *open_stp = NULL;
2675                 struct nfs4_file *fp;
2676
2677                 status = nfserr_stale_clientid;
2678                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2679                         goto out;
2680
2681                 /* validate and update open stateid and open seqid */
2682                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2683                                         lock->lk_new_open_seqid,
2684                                         &lock->lk_new_open_stateid,
2685                                         CHECK_FH | OPEN_STATE,
2686                                         &lock->lk_replay_owner, &open_stp,
2687                                         lock);
2688                 if (status)
2689                         goto out;
2690                 open_sop = lock->lk_replay_owner;
2691                 /* create lockowner and lock stateid */
2692                 fp = open_stp->st_file;
2693                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
2694                                 open_sop->so_client->cl_clientid.cl_id,
2695                                 &lock->v.new.owner);
2696                 /* XXX: Do we need to check for duplicate stateowners on
2697                  * the same file, or should they just be allowed (and
2698                  * create new stateids)? */
2699                 status = nfserr_resource;
2700                 lock_sop = alloc_init_lock_stateowner(strhashval,
2701                                 open_sop->so_client, open_stp, lock);
2702                 if (lock_sop == NULL)
2703                         goto out;
2704                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2705                 if (lock_stp == NULL)
2706                         goto out;
2707         } else {
2708                 /* lock (lock owner + lock stateid) already exists */
2709                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2710                                        lock->lk_old_lock_seqid,
2711                                        &lock->lk_old_lock_stateid,
2712                                        CHECK_FH | LOCK_STATE,
2713                                        &lock->lk_replay_owner, &lock_stp, lock);
2714                 if (status)
2715                         goto out;
2716                 lock_sop = lock->lk_replay_owner;
2717         }
2718         /* lock->lk_replay_owner and lock_stp have been created or found */
2719         filp = lock_stp->st_vfs_file;
2720
2721         status = nfserr_grace;
2722         if (nfs4_in_grace() && !lock->lk_reclaim)
2723                 goto out;
2724         status = nfserr_no_grace;
2725         if (!nfs4_in_grace() && lock->lk_reclaim)
2726                 goto out;
2727
2728         locks_init_lock(&file_lock);
2729         switch (lock->lk_type) {
2730                 case NFS4_READ_LT:
2731                 case NFS4_READW_LT:
2732                         file_lock.fl_type = F_RDLCK;
2733                         cmd = F_SETLK;
2734                 break;
2735                 case NFS4_WRITE_LT:
2736                 case NFS4_WRITEW_LT:
2737                         file_lock.fl_type = F_WRLCK;
2738                         cmd = F_SETLK;
2739                 break;
2740                 default:
2741                         status = nfserr_inval;
2742                 goto out;
2743         }
2744         file_lock.fl_owner = (fl_owner_t)lock_sop;
2745         file_lock.fl_pid = current->tgid;
2746         file_lock.fl_file = filp;
2747         file_lock.fl_flags = FL_POSIX;
2748         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2749
2750         file_lock.fl_start = lock->lk_offset;
2751         if ((lock->lk_length == ~(u64)0) ||
2752                         LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2753                 file_lock.fl_end = ~(u64)0;
2754         else
2755                 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2756         nfs4_transform_lock_offset(&file_lock);
2757
2758         /*
2759         * Try to lock the file in the VFS.
2760         * Note: locks.c uses the BKL to protect the inode's lock list.
2761         */
2762
2763         /* XXX?: Just to divert the locks_release_private at the start of
2764          * locks_copy_lock: */
2765         locks_init_lock(&conflock);
2766         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2767         switch (-err) {
2768         case 0: /* success! */
2769                 update_stateid(&lock_stp->st_stateid);
2770                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
2771                                 sizeof(stateid_t));
2772                 status = 0;
2773                 break;
2774         case (EAGAIN):          /* conflock holds conflicting lock */
2775                 status = nfserr_denied;
2776                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2777                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2778                 break;
2779         case (EDEADLK):
2780                 status = nfserr_deadlock;
2781                 break;
2782         default:
2783                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2784                 status = nfserr_resource;
2785                 break;
2786         }
2787 out:
2788         if (status && lock->lk_is_new && lock_sop)
2789                 release_stateowner(lock_sop);
2790         if (lock->lk_replay_owner) {
2791                 nfs4_get_stateowner(lock->lk_replay_owner);
2792                 cstate->replay_owner = lock->lk_replay_owner;
2793         }
2794         nfs4_unlock_state();
2795         return status;
2796 }
2797
2798 /*
2799  * LOCKT operation
2800  */
2801 __be32
2802 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2803             struct nfsd4_lockt *lockt)
2804 {
2805         struct inode *inode;
2806         struct file file;
2807         struct file_lock file_lock;
2808         int error;
2809         __be32 status;
2810
2811         if (nfs4_in_grace())
2812                 return nfserr_grace;
2813
2814         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2815                  return nfserr_inval;
2816
2817         lockt->lt_stateowner = NULL;
2818         nfs4_lock_state();
2819
2820         status = nfserr_stale_clientid;
2821         if (STALE_CLIENTID(&lockt->lt_clientid))
2822                 goto out;
2823
2824         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2825                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2826                 if (status == nfserr_symlink)
2827                         status = nfserr_inval;
2828                 goto out;
2829         }
2830
2831         inode = cstate->current_fh.fh_dentry->d_inode;
2832         locks_init_lock(&file_lock);
2833         switch (lockt->lt_type) {
2834                 case NFS4_READ_LT:
2835                 case NFS4_READW_LT:
2836                         file_lock.fl_type = F_RDLCK;
2837                 break;
2838                 case NFS4_WRITE_LT:
2839                 case NFS4_WRITEW_LT:
2840                         file_lock.fl_type = F_WRLCK;
2841                 break;
2842                 default:
2843                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2844                         status = nfserr_inval;
2845                 goto out;
2846         }
2847
2848         lockt->lt_stateowner = find_lockstateowner_str(inode,
2849                         &lockt->lt_clientid, &lockt->lt_owner);
2850         if (lockt->lt_stateowner)
2851                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2852         file_lock.fl_pid = current->tgid;
2853         file_lock.fl_flags = FL_POSIX;
2854         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2855
2856         file_lock.fl_start = lockt->lt_offset;
2857         if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2858                 file_lock.fl_end = ~(u64)0;
2859         else
2860                 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2861
2862         nfs4_transform_lock_offset(&file_lock);
2863
2864         /* vfs_test_lock uses the struct file _only_ to resolve the inode.
2865          * since LOCKT doesn't require an OPEN, and therefore a struct
2866          * file may not exist, pass vfs_test_lock a struct file with
2867          * only the dentry:inode set.
2868          */
2869         memset(&file, 0, sizeof (struct file));
2870         file.f_path.dentry = cstate->current_fh.fh_dentry;
2871
2872         status = nfs_ok;
2873         error = vfs_test_lock(&file, &file_lock);
2874         if (error) {
2875                 status = nfserrno(error);
2876                 goto out;
2877         }
2878         if (file_lock.fl_type != F_UNLCK) {
2879                 status = nfserr_denied;
2880                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2881         }
2882 out:
2883         nfs4_unlock_state();
2884         return status;
2885 }
2886
2887 __be32
2888 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2889             struct nfsd4_locku *locku)
2890 {
2891         struct nfs4_stateid *stp;
2892         struct file *filp = NULL;
2893         struct file_lock file_lock;
2894         __be32 status;
2895         int err;
2896
2897         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2898                 (long long) locku->lu_offset,
2899                 (long long) locku->lu_length);
2900
2901         if (check_lock_length(locku->lu_offset, locku->lu_length))
2902                  return nfserr_inval;
2903
2904         nfs4_lock_state();
2905
2906         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2907                                         locku->lu_seqid,
2908                                         &locku->lu_stateid,
2909                                         CHECK_FH | LOCK_STATE,
2910                                         &locku->lu_stateowner, &stp, NULL)))
2911                 goto out;
2912
2913         filp = stp->st_vfs_file;
2914         BUG_ON(!filp);
2915         locks_init_lock(&file_lock);
2916         file_lock.fl_type = F_UNLCK;
2917         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2918         file_lock.fl_pid = current->tgid;
2919         file_lock.fl_file = filp;
2920         file_lock.fl_flags = FL_POSIX;
2921         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2922         file_lock.fl_start = locku->lu_offset;
2923
2924         if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2925                 file_lock.fl_end = ~(u64)0;
2926         else
2927                 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2928         nfs4_transform_lock_offset(&file_lock);
2929
2930         /*
2931         *  Try to unlock the file in the VFS.
2932         */
2933         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2934         if (err) {
2935                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2936                 goto out_nfserr;
2937         }
2938         /*
2939         * OK, unlock succeeded; the only thing left to do is update the stateid.
2940         */
2941         update_stateid(&stp->st_stateid);
2942         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2943
2944 out:
2945         if (locku->lu_stateowner) {
2946                 nfs4_get_stateowner(locku->lu_stateowner);
2947                 cstate->replay_owner = locku->lu_stateowner;
2948         }
2949         nfs4_unlock_state();
2950         return status;
2951
2952 out_nfserr:
2953         status = nfserrno(err);
2954         goto out;
2955 }
2956
2957 /*
2958  * returns
2959  *      1: locks held by lockowner
2960  *      0: no locks held by lockowner
2961  */
2962 static int
2963 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2964 {
2965         struct file_lock **flpp;
2966         struct inode *inode = filp->f_path.dentry->d_inode;
2967         int status = 0;
2968
2969         lock_kernel();
2970         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2971                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2972                         status = 1;
2973                         goto out;
2974                 }
2975         }
2976 out:
2977         unlock_kernel();
2978         return status;
2979 }
2980
2981 __be32
2982 nfsd4_release_lockowner(struct svc_rqst *rqstp,
2983                         struct nfsd4_compound_state *cstate,
2984                         struct nfsd4_release_lockowner *rlockowner)
2985 {
2986         clientid_t *clid = &rlockowner->rl_clientid;
2987         struct nfs4_stateowner *sop;
2988         struct nfs4_stateid *stp;
2989         struct xdr_netobj *owner = &rlockowner->rl_owner;
2990         struct list_head matches;
2991         int i;
2992         __be32 status;
2993
2994         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
2995                 clid->cl_boot, clid->cl_id);
2996
2997         /* XXX check for lease expiration */
2998
2999         status = nfserr_stale_clientid;
3000         if (STALE_CLIENTID(clid))
3001                 return status;
3002
3003         nfs4_lock_state();
3004
3005         status = nfserr_locks_held;
3006         /* XXX: we're doing a linear search through all the lockowners.
3007          * Yipes!  For now we'll just hope clients aren't really using
3008          * release_lockowner much, but eventually we have to fix these
3009          * data structures. */
3010         INIT_LIST_HEAD(&matches);
3011         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3012                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3013                         if (!same_owner_str(sop, owner, clid))
3014                                 continue;
3015                         list_for_each_entry(stp, &sop->so_stateids,
3016                                         st_perstateowner) {
3017                                 if (check_for_locks(stp->st_vfs_file, sop))
3018                                         goto out;
3019                                 /* Note: so_perclient unused for lockowners,
3020                                  * so it's OK to fool with here. */
3021                                 list_add(&sop->so_perclient, &matches);
3022                         }
3023                 }
3024         }
3025         /* Clients probably won't expect us to return with some (but not all)
3026          * of the lockowner state released; so don't release any until all
3027          * have been checked. */
3028         status = nfs_ok;
3029         while (!list_empty(&matches)) {
3030                 sop = list_entry(matches.next, struct nfs4_stateowner,
3031                                                                 so_perclient);
3032                 /* unhash_stateowner deletes so_perclient only
3033                  * for openowners. */
3034                 list_del(&sop->so_perclient);
3035                 release_stateowner(sop);
3036         }
3037 out:
3038         nfs4_unlock_state();
3039         return status;
3040 }
3041
3042 static inline struct nfs4_client_reclaim *
3043 alloc_reclaim(void)
3044 {
3045         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3046 }
3047
3048 int
3049 nfs4_has_reclaimed_state(const char *name)
3050 {
3051         unsigned int strhashval = clientstr_hashval(name);
3052         struct nfs4_client *clp;
3053
3054         clp = find_confirmed_client_by_str(name, strhashval);
3055         return clp ? 1 : 0;
3056 }
3057
3058 /*
3059  * failure => all reset bets are off, nfserr_no_grace...
3060  */
3061 int
3062 nfs4_client_to_reclaim(const char *name)
3063 {
3064         unsigned int strhashval;
3065         struct nfs4_client_reclaim *crp = NULL;
3066
3067         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3068         crp = alloc_reclaim();
3069         if (!crp)
3070                 return 0;
3071         strhashval = clientstr_hashval(name);
3072         INIT_LIST_HEAD(&crp->cr_strhash);
3073         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3074         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3075         reclaim_str_hashtbl_size++;
3076         return 1;
3077 }
3078
3079 static void
3080 nfs4_release_reclaim(void)
3081 {
3082         struct nfs4_client_reclaim *crp = NULL;
3083         int i;
3084
3085         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3086                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3087                         crp = list_entry(reclaim_str_hashtbl[i].next,
3088                                         struct nfs4_client_reclaim, cr_strhash);
3089                         list_del(&crp->cr_strhash);
3090                         kfree(crp);
3091                         reclaim_str_hashtbl_size--;
3092                 }
3093         }
3094         BUG_ON(reclaim_str_hashtbl_size);
3095 }
3096
3097 /*
3098  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3099 static struct nfs4_client_reclaim *
3100 nfs4_find_reclaim_client(clientid_t *clid)
3101 {
3102         unsigned int strhashval;
3103         struct nfs4_client *clp;
3104         struct nfs4_client_reclaim *crp = NULL;
3105
3106
3107         /* find clientid in conf_id_hashtbl */
3108         clp = find_confirmed_client(clid);
3109         if (clp == NULL)
3110                 return NULL;
3111
3112         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3113                             clp->cl_name.len, clp->cl_name.data,
3114                             clp->cl_recdir);
3115
3116         /* find clp->cl_name in reclaim_str_hashtbl */
3117         strhashval = clientstr_hashval(clp->cl_recdir);
3118         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3119                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3120                         return crp;
3121                 }
3122         }
3123         return NULL;
3124 }
3125
3126 /*
3127 * Called from OPEN. Look for clientid in reclaim list.
3128 */
3129 __be32
3130 nfs4_check_open_reclaim(clientid_t *clid)
3131 {
3132         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3133 }
3134
3135 /* initialization to perform at module load time: */
3136
3137 int
3138 nfs4_state_init(void)
3139 {
3140         int i, status;
3141
3142         status = nfsd4_init_slabs();
3143         if (status)
3144                 return status;
3145         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3146                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3147                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3148                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3149                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3150         }
3151         for (i = 0; i < FILE_HASH_SIZE; i++) {
3152                 INIT_LIST_HEAD(&file_hashtbl[i]);
3153         }
3154         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3155                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3156                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3157         }
3158         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3159                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3160                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3161         }
3162         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3163                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3164                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3165         }
3166         memset(&onestateid, ~0, sizeof(stateid_t));
3167         INIT_LIST_HEAD(&close_lru);
3168         INIT_LIST_HEAD(&client_lru);
3169         INIT_LIST_HEAD(&del_recall_lru);
3170         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3171                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3172         reclaim_str_hashtbl_size = 0;
3173         return 0;
3174 }
3175
3176 static void
3177 nfsd4_load_reboot_recovery_data(void)
3178 {
3179         int status;
3180
3181         nfs4_lock_state();
3182         nfsd4_init_recdir(user_recovery_dirname);
3183         status = nfsd4_recdir_load();
3184         nfs4_unlock_state();
3185         if (status)
3186                 printk("NFSD: Failure reading reboot recovery data\n");
3187 }
3188
3189 unsigned long
3190 get_nfs4_grace_period(void)
3191 {
3192         return max(user_lease_time, lease_time) * HZ;
3193 }
3194
3195 /*
3196  * Since the lifetime of a delegation isn't limited to that of an open, a
3197  * client may quite reasonably hang on to a delegation as long as it has
3198  * the inode cached.  This becomes an obvious problem the first time a
3199  * client's inode cache approaches the size of the server's total memory.
3200  *
3201  * For now we avoid this problem by imposing a hard limit on the number
3202  * of delegations, which varies according to the server's memory size.
3203  */
3204 static void
3205 set_max_delegations(void)
3206 {
3207         /*
3208          * Allow at most 4 delegations per megabyte of RAM.  Quick
3209          * estimates suggest that in the worst case (where every delegation
3210          * is for a different inode), a delegation could take about 1.5K,
3211          * giving a worst case usage of about 6% of memory.
3212          */
3213         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3214 }
3215
3216 /* initialization to perform when the nfsd service is started: */
3217
3218 static void
3219 __nfs4_state_start(void)
3220 {
3221         unsigned long grace_time;
3222
3223         boot_time = get_seconds();
3224         grace_time = get_nfs_grace_period();
3225         lease_time = user_lease_time;
3226         in_grace = 1;
3227         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3228                grace_time/HZ);
3229         laundry_wq = create_singlethread_workqueue("nfsd4");
3230         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3231         set_max_delegations();
3232 }
3233
3234 void
3235 nfs4_state_start(void)
3236 {
3237         if (nfs4_init)
3238                 return;
3239         nfsd4_load_reboot_recovery_data();
3240         __nfs4_state_start();
3241         nfs4_init = 1;
3242         return;
3243 }
3244
3245 int
3246 nfs4_in_grace(void)
3247 {
3248         return in_grace;
3249 }
3250
3251 time_t
3252 nfs4_lease_time(void)
3253 {
3254         return lease_time;
3255 }
3256
3257 static void
3258 __nfs4_state_shutdown(void)
3259 {
3260         int i;
3261         struct nfs4_client *clp = NULL;
3262         struct nfs4_delegation *dp = NULL;
3263         struct list_head *pos, *next, reaplist;
3264
3265         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3266                 while (!list_empty(&conf_id_hashtbl[i])) {
3267                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3268                         expire_client(clp);
3269                 }
3270                 while (!list_empty(&unconf_str_hashtbl[i])) {
3271                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3272                         expire_client(clp);
3273                 }
3274         }
3275         INIT_LIST_HEAD(&reaplist);
3276         spin_lock(&recall_lock);
3277         list_for_each_safe(pos, next, &del_recall_lru) {
3278                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3279                 list_move(&dp->dl_recall_lru, &reaplist);
3280         }
3281         spin_unlock(&recall_lock);
3282         list_for_each_safe(pos, next, &reaplist) {
3283                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3284                 list_del_init(&dp->dl_recall_lru);
3285                 unhash_delegation(dp);
3286         }
3287
3288         nfsd4_shutdown_recdir();
3289         nfs4_init = 0;
3290 }
3291
3292 void
3293 nfs4_state_shutdown(void)
3294 {
3295         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3296         destroy_workqueue(laundry_wq);
3297         nfs4_lock_state();
3298         nfs4_release_reclaim();
3299         __nfs4_state_shutdown();
3300         nfs4_unlock_state();
3301 }
3302
3303 static void
3304 nfs4_set_recdir(char *recdir)
3305 {
3306         nfs4_lock_state();
3307         strcpy(user_recovery_dirname, recdir);
3308         nfs4_unlock_state();
3309 }
3310
3311 /*
3312  * Change the NFSv4 recovery directory to recdir.
3313  */
3314 int
3315 nfs4_reset_recoverydir(char *recdir)
3316 {
3317         int status;
3318         struct nameidata nd;
3319
3320         status = path_lookup(recdir, LOOKUP_FOLLOW, &nd);
3321         if (status)
3322                 return status;
3323         status = -ENOTDIR;
3324         if (S_ISDIR(nd.dentry->d_inode->i_mode)) {
3325                 nfs4_set_recdir(recdir);
3326                 status = 0;
3327         }
3328         path_release(&nd);
3329         return status;
3330 }
3331
3332 /*
3333  * Called when leasetime is changed.
3334  *
3335  * The only way the protocol gives us to handle on-the-fly lease changes is to
3336  * simulate a reboot.  Instead of doing that, we just wait till the next time
3337  * we start to register any changes in lease time.  If the administrator
3338  * really wants to change the lease time *now*, they can go ahead and bring
3339  * nfsd down and then back up again after changing the lease time.
3340  */
3341 void
3342 nfs4_reset_lease(time_t leasetime)
3343 {
3344         lock_kernel();
3345         user_lease_time = leasetime;
3346         unlock_kernel();
3347 }