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
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 /* Globals */
  56 static time_t lease_time = 90;     /* default lease time */
  57 static time_t user_lease_time = 90;
  58 static time_t boot_time;
  59 static int in_grace = 1;
  60 static u32 current_clientid = 1;
  61 static u32 current_ownerid = 1;
  62 static u32 current_fileid = 1;
  63 static u32 current_delegid = 1;
  64 static u32 nfs4_init;
  65 static stateid_t zerostateid;             /* bits all 0 */
  66 static stateid_t onestateid;              /* bits all 1 */
  67
  68 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
  69 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
  70
  71 /* forward declarations */
  72 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
  73 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
  74 static void release_stateid_lockowners(struct nfs4_stateid *open_stp);
  75 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  76 static void nfs4_set_recdir(char *recdir);
  77
  78 /* Locking:
  79  *
  80  * client_sema:
  81  *      protects clientid_hashtbl[], clientstr_hashtbl[],
  82  *      unconfstr_hashtbl[], uncofid_hashtbl[].
  83  */
  84 static DECLARE_MUTEX(client_sema);
  85
  86 static kmem_cache_t *stateowner_slab = NULL;
  87 static kmem_cache_t *file_slab = NULL;
  88 static kmem_cache_t *stateid_slab = NULL;
  89 static kmem_cache_t *deleg_slab = NULL;
  90
  91 void
  92 nfs4_lock_state(void)
  93 {
  94         down(&client_sema);
  95 }
  96
  97 void
  98 nfs4_unlock_state(void)
  99 {
 100         up(&client_sema);
 101 }
 102
 103 static inline u32
 104 opaque_hashval(const void *ptr, int nbytes)
 105 {
 106         unsigned char *cptr = (unsigned char *) ptr;
 107
 108         u32 x = 0;
 109         while (nbytes--) {
 110                 x *= 37;
 111                 x += *cptr++;
 112         }
 113         return x;
 114 }
 115
 116 /* forward declarations */
 117 static void release_stateowner(struct nfs4_stateowner *sop);
 118 static void release_stateid(struct nfs4_stateid *stp, int flags);
 119
 120 /*
 121  * Delegation state
 122  */
 123
 124 /* recall_lock protects the del_recall_lru */
 125 static spinlock_t recall_lock = SPIN_LOCK_UNLOCKED;
 126 static struct list_head del_recall_lru;
 127
 128 static void
 129 free_nfs4_file(struct kref *kref)
 130 {
 131         struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
 132         list_del(&fp->fi_hash);
 133         iput(fp->fi_inode);
 134         kmem_cache_free(file_slab, fp);
 135 }
 136
 137 static inline void
 138 put_nfs4_file(struct nfs4_file *fi)
 139 {
 140         kref_put(&fi->fi_ref, free_nfs4_file);
 141 }
 142
 143 static inline void
 144 get_nfs4_file(struct nfs4_file *fi)
 145 {
 146         kref_get(&fi->fi_ref);
 147 }
 148
 149 static struct nfs4_delegation *
 150 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 151 {
 152         struct nfs4_delegation *dp;
 153         struct nfs4_file *fp = stp->st_file;
 154         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
 155
 156         dprintk("NFSD alloc_init_deleg\n");
 157         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
 158         if (dp == NULL)
 159                 return dp;
 160         INIT_LIST_HEAD(&dp->dl_perfile);
 161         INIT_LIST_HEAD(&dp->dl_perclnt);
 162         INIT_LIST_HEAD(&dp->dl_recall_lru);
 163         dp->dl_client = clp;
 164         get_nfs4_file(fp);
 165         dp->dl_file = fp;
 166         dp->dl_flock = NULL;
 167         get_file(stp->st_vfs_file);
 168         dp->dl_vfs_file = stp->st_vfs_file;
 169         dp->dl_type = type;
 170         dp->dl_recall.cbr_dp = NULL;
 171         dp->dl_recall.cbr_ident = cb->cb_ident;
 172         dp->dl_recall.cbr_trunc = 0;
 173         dp->dl_stateid.si_boot = boot_time;
 174         dp->dl_stateid.si_stateownerid = current_delegid++;
 175         dp->dl_stateid.si_fileid = 0;
 176         dp->dl_stateid.si_generation = 0;
 177         dp->dl_fhlen = current_fh->fh_handle.fh_size;
 178         memcpy(dp->dl_fhval, &current_fh->fh_handle.fh_base,
 179                         current_fh->fh_handle.fh_size);
 180         dp->dl_time = 0;
 181         atomic_set(&dp->dl_count, 1);
 182         list_add(&dp->dl_perfile, &fp->fi_delegations);
 183         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 184         return dp;
 185 }
 186
 187 void
 188 nfs4_put_delegation(struct nfs4_delegation *dp)
 189 {
 190         if (atomic_dec_and_test(&dp->dl_count)) {
 191                 dprintk("NFSD: freeing dp %p\n",dp);
 192                 put_nfs4_file(dp->dl_file);
 193                 kmem_cache_free(deleg_slab, dp);
 194         }
 195 }
 196
 197 /* Remove the associated file_lock first, then remove the delegation.
 198  * lease_modify() is called to remove the FS_LEASE file_lock from
 199  * the i_flock list, eventually calling nfsd's lock_manager
 200  * fl_release_callback.
 201  */
 202 static void
 203 nfs4_close_delegation(struct nfs4_delegation *dp)
 204 {
 205         struct file *filp = dp->dl_vfs_file;
 206
 207         dprintk("NFSD: close_delegation dp %p\n",dp);
 208         dp->dl_vfs_file = NULL;
 209         /* The following nfsd_close may not actually close the file,
 210          * but we want to remove the lease in any case. */
 211         if (dp->dl_flock)
 212                 setlease(filp, F_UNLCK, &dp->dl_flock);
 213         nfsd_close(filp);
 214 }
 215
 216 /* Called under the state lock. */
 217 static void
 218 unhash_delegation(struct nfs4_delegation *dp)
 219 {
 220         list_del_init(&dp->dl_perfile);
 221         list_del_init(&dp->dl_perclnt);
 222         spin_lock(&recall_lock);
 223         list_del_init(&dp->dl_recall_lru);
 224         spin_unlock(&recall_lock);
 225         nfs4_close_delegation(dp);
 226         nfs4_put_delegation(dp);
 227 }
 228
 229 /*
 230  * SETCLIENTID state
 231  */
 232
 233 /* Hash tables for nfs4_clientid state */
 234 #define CLIENT_HASH_BITS                 4
 235 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 236 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 237
 238 #define clientid_hashval(id) \
 239         ((id) & CLIENT_HASH_MASK)
 240 #define clientstr_hashval(name) \
 241         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
 242 /*
 243  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 244  * used in reboot/reset lease grace period processing
 245  *
 246  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 247  * setclientid_confirmed info.
 248  *
 249  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 250  * setclientid info.
 251  *
 252  * client_lru holds client queue ordered by nfs4_client.cl_time
 253  * for lease renewal.
 254  *
 255  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 256  * for last close replay.
 257  */
 258 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 259 static int reclaim_str_hashtbl_size = 0;
 260 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 261 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 262 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 263 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 264 static struct list_head client_lru;
 265 static struct list_head close_lru;
 266
 267 static inline void
 268 renew_client(struct nfs4_client *clp)
 269 {
 270         /*
 271         * Move client to the end to the LRU list.
 272         */
 273         dprintk("renewing client (clientid %08x/%08x)\n",
 274                         clp->cl_clientid.cl_boot,
 275                         clp->cl_clientid.cl_id);
 276         list_move_tail(&clp->cl_lru, &client_lru);
 277         clp->cl_time = get_seconds();
 278 }
 279
 280 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 281 static int
 282 STALE_CLIENTID(clientid_t *clid)
 283 {
 284         if (clid->cl_boot == boot_time)
 285                 return 0;
 286         dprintk("NFSD stale clientid (%08x/%08x)\n",
 287                         clid->cl_boot, clid->cl_id);
 288         return 1;
 289 }
 290
 291 /*
 292  * XXX Should we use a slab cache ?
 293  * This type of memory management is somewhat inefficient, but we use it
 294  * anyway since SETCLIENTID is not a common operation.
 295  */
 296 static inline struct nfs4_client *
 297 alloc_client(struct xdr_netobj name)
 298 {
 299         struct nfs4_client *clp;
 300
 301         if ((clp = kmalloc(sizeof(struct nfs4_client), GFP_KERNEL))!= NULL) {
 302                 memset(clp, 0, sizeof(*clp));
 303                 if ((clp->cl_name.data = kmalloc(name.len, GFP_KERNEL)) != NULL) {
 304                         memcpy(clp->cl_name.data, name.data, name.len);
 305                         clp->cl_name.len = name.len;
 306                 }
 307                 else {
 308                         kfree(clp);
 309                         clp = NULL;
 310                 }
 311         }
 312         return clp;
 313 }
 314
 315 static inline void
 316 free_client(struct nfs4_client *clp)
 317 {
 318         if (clp->cl_cred.cr_group_info)
 319                 put_group_info(clp->cl_cred.cr_group_info);
 320         kfree(clp->cl_name.data);
 321         kfree(clp);
 322 }
 323
 324 void
 325 put_nfs4_client(struct nfs4_client *clp)
 326 {
 327         if (atomic_dec_and_test(&clp->cl_count))
 328                 free_client(clp);
 329 }
 330
 331 static void
 332 expire_client(struct nfs4_client *clp)
 333 {
 334         struct nfs4_stateowner *sop;
 335         struct nfs4_delegation *dp;
 336         struct nfs4_callback *cb = &clp->cl_callback;
 337         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
 338         struct list_head reaplist;
 339
 340         dprintk("NFSD: expire_client cl_count %d\n",
 341                             atomic_read(&clp->cl_count));
 342
 343         /* shutdown rpc client, ending any outstanding recall rpcs */
 344         if (atomic_read(&cb->cb_set) == 1 && clnt) {
 345                 rpc_shutdown_client(clnt);
 346                 clnt = clp->cl_callback.cb_client = NULL;
 347         }
 348
 349         INIT_LIST_HEAD(&reaplist);
 350         spin_lock(&recall_lock);
 351         while (!list_empty(&clp->cl_delegations)) {
 352                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
 353                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
 354                                 dp->dl_flock);
 355                 list_del_init(&dp->dl_perclnt);
 356                 list_move(&dp->dl_recall_lru, &reaplist);
 357         }
 358         spin_unlock(&recall_lock);
 359         while (!list_empty(&reaplist)) {
 360                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
 361                 list_del_init(&dp->dl_recall_lru);
 362                 unhash_delegation(dp);
 363         }
 364         list_del(&clp->cl_idhash);
 365         list_del(&clp->cl_strhash);
 366         list_del(&clp->cl_lru);
 367         while (!list_empty(&clp->cl_openowners)) {
 368                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
 369                 release_stateowner(sop);
 370         }
 371         put_nfs4_client(clp);
 372 }
 373
 374 static struct nfs4_client *
 375 create_client(struct xdr_netobj name, char *recdir) {
 376         struct nfs4_client *clp;
 377
 378         if (!(clp = alloc_client(name)))
 379                 goto out;
 380         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
 381         atomic_set(&clp->cl_count, 1);
 382         atomic_set(&clp->cl_callback.cb_set, 0);
 383         INIT_LIST_HEAD(&clp->cl_idhash);
 384         INIT_LIST_HEAD(&clp->cl_strhash);
 385         INIT_LIST_HEAD(&clp->cl_openowners);
 386         INIT_LIST_HEAD(&clp->cl_delegations);
 387         INIT_LIST_HEAD(&clp->cl_lru);
 388 out:
 389         return clp;
 390 }
 391
 392 static void
 393 copy_verf(struct nfs4_client *target, nfs4_verifier *source) {
 394         memcpy(target->cl_verifier.data, source->data, sizeof(target->cl_verifier.data));
 395 }
 396
 397 static void
 398 copy_clid(struct nfs4_client *target, struct nfs4_client *source) {
 399         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
 400         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
 401 }
 402
 403 static void
 404 copy_cred(struct svc_cred *target, struct svc_cred *source) {
 405
 406         target->cr_uid = source->cr_uid;
 407         target->cr_gid = source->cr_gid;
 408         target->cr_group_info = source->cr_group_info;
 409         get_group_info(target->cr_group_info);
 410 }
 411
 412 static inline int
 413 same_name(const char *n1, const char *n2) {
 414         return 0 == memcmp(n1, n2, HEXDIR_LEN);
 415 }
 416
 417 static int
 418 cmp_verf(nfs4_verifier *v1, nfs4_verifier *v2) {
 419         return(!memcmp(v1->data,v2->data,sizeof(v1->data)));
 420 }
 421
 422 static int
 423 cmp_clid(clientid_t * cl1, clientid_t * cl2) {
 424         return((cl1->cl_boot == cl2->cl_boot) &&
 425                 (cl1->cl_id == cl2->cl_id));
 426 }
 427
 428 /* XXX what about NGROUP */
 429 static int
 430 cmp_creds(struct svc_cred *cr1, struct svc_cred *cr2){
 431         return(cr1->cr_uid == cr2->cr_uid);
 432
 433 }
 434
 435 static void
 436 gen_clid(struct nfs4_client *clp) {
 437         clp->cl_clientid.cl_boot = boot_time;
 438         clp->cl_clientid.cl_id = current_clientid++;
 439 }
 440
 441 static void
 442 gen_confirm(struct nfs4_client *clp) {
 443         struct timespec         tv;
 444         u32 *                   p;
 445
 446         tv = CURRENT_TIME;
 447         p = (u32 *)clp->cl_confirm.data;
 448         *p++ = tv.tv_sec;
 449         *p++ = tv.tv_nsec;
 450 }
 451
 452 static int
 453 check_name(struct xdr_netobj name) {
 454
 455         if (name.len == 0)
 456                 return 0;
 457         if (name.len > NFS4_OPAQUE_LIMIT) {
 458                 printk("NFSD: check_name: name too long(%d)!\n", name.len);
 459                 return 0;
 460         }
 461         return 1;
 462 }
 463
 464 static void
 465 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
 466 {
 467         unsigned int idhashval;
 468
 469         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
 470         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 471         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
 472         list_add_tail(&clp->cl_lru, &client_lru);
 473         clp->cl_time = get_seconds();
 474 }
 475
 476 static void
 477 move_to_confirmed(struct nfs4_client *clp)
 478 {
 479         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 480         unsigned int strhashval;
 481
 482         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
 483         list_del_init(&clp->cl_strhash);
 484         list_del_init(&clp->cl_idhash);
 485         list_add(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
 486         strhashval = clientstr_hashval(clp->cl_recdir);
 487         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
 488         renew_client(clp);
 489 }
 490
 491 static struct nfs4_client *
 492 find_confirmed_client(clientid_t *clid)
 493 {
 494         struct nfs4_client *clp;
 495         unsigned int idhashval = clientid_hashval(clid->cl_id);
 496
 497         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
 498                 if (cmp_clid(&clp->cl_clientid, clid))
 499                         return clp;
 500         }
 501         return NULL;
 502 }
 503
 504 static struct nfs4_client *
 505 find_unconfirmed_client(clientid_t *clid)
 506 {
 507         struct nfs4_client *clp;
 508         unsigned int idhashval = clientid_hashval(clid->cl_id);
 509
 510         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
 511                 if (cmp_clid(&clp->cl_clientid, clid))
 512                         return clp;
 513         }
 514         return NULL;
 515 }
 516
 517 static struct nfs4_client *
 518 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
 519 {
 520         struct nfs4_client *clp;
 521
 522         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
 523                 if (same_name(clp->cl_recdir, dname))
 524                         return clp;
 525         }
 526         return NULL;
 527 }
 528
 529 static struct nfs4_client *
 530 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
 531 {
 532         struct nfs4_client *clp;
 533
 534         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
 535                 if (same_name(clp->cl_recdir, dname))
 536                         return clp;
 537         }
 538         return NULL;
 539 }
 540
 541 /* a helper function for parse_callback */
 542 static int
 543 parse_octet(unsigned int *lenp, char **addrp)
 544 {
 545         unsigned int len = *lenp;
 546         char *p = *addrp;
 547         int n = -1;
 548         char c;
 549
 550         for (;;) {
 551                 if (!len)
 552                         break;
 553                 len--;
 554                 c = *p++;
 555                 if (c == '.')
 556                         break;
 557                 if ((c < '0') || (c > '9')) {
 558                         n = -1;
 559                         break;
 560                 }
 561                 if (n < 0)
 562                         n = 0;
 563                 n = (n * 10) + (c - '0');
 564                 if (n > 255) {
 565                         n = -1;
 566                         break;
 567                 }
 568         }
 569         *lenp = len;
 570         *addrp = p;
 571         return n;
 572 }
 573
 574 /* parse and set the setclientid ipv4 callback address */
 575 static int
 576 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
 577 {
 578         int temp = 0;
 579         u32 cbaddr = 0;
 580         u16 cbport = 0;
 581         u32 addrlen = addr_len;
 582         char *addr = addr_val;
 583         int i, shift;
 584
 585         /* ipaddress */
 586         shift = 24;
 587         for(i = 4; i > 0  ; i--) {
 588                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 589                         return 0;
 590                 }
 591                 cbaddr |= (temp << shift);
 592                 if (shift > 0)
 593                 shift -= 8;
 594         }
 595         *cbaddrp = cbaddr;
 596
 597         /* port */
 598         shift = 8;
 599         for(i = 2; i > 0  ; i--) {
 600                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 601                         return 0;
 602                 }
 603                 cbport |= (temp << shift);
 604                 if (shift > 0)
 605                         shift -= 8;
 606         }
 607         *cbportp = cbport;
 608         return 1;
 609 }
 610
 611 static void
 612 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
 613 {
 614         struct nfs4_callback *cb = &clp->cl_callback;
 615
 616         /* Currently, we only support tcp for the callback channel */
 617         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
 618                 goto out_err;
 619
 620         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
 621                          &cb->cb_addr, &cb->cb_port)))
 622                 goto out_err;
 623         cb->cb_prog = se->se_callback_prog;
 624         cb->cb_ident = se->se_callback_ident;
 625         return;
 626 out_err:
 627         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
 628                 "will not receive delegations\n",
 629                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
 630
 631         return;
 632 }
 633
 634 /*
 635  * RFC 3010 has a complex implmentation description of processing a
 636  * SETCLIENTID request consisting of 5 bullets, labeled as
 637  * CASE0 - CASE4 below.
 638  *
 639  * NOTES:
 640  *      callback information will be processed in a future patch
 641  *
 642  *      an unconfirmed record is added when:
 643  *      NORMAL (part of CASE 4): there is no confirmed nor unconfirmed record.
 644  *      CASE 1: confirmed record found with matching name, principal,
 645  *              verifier, and clientid.
 646  *      CASE 2: confirmed record found with matching name, principal,
 647  *              and there is no unconfirmed record with matching
 648  *              name and principal
 649  *
 650  *      an unconfirmed record is replaced when:
 651  *      CASE 3: confirmed record found with matching name, principal,
 652  *              and an unconfirmed record is found with matching
 653  *              name, principal, and with clientid and
 654  *              confirm that does not match the confirmed record.
 655  *      CASE 4: there is no confirmed record with matching name and
 656  *              principal. there is an unconfirmed record with
 657  *              matching name, principal.
 658  *
 659  *      an unconfirmed record is deleted when:
 660  *      CASE 1: an unconfirmed record that matches input name, verifier,
 661  *              and confirmed clientid.
 662  *      CASE 4: any unconfirmed records with matching name and principal
 663  *              that exist after an unconfirmed record has been replaced
 664  *              as described above.
 665  *
 666  */
 667 int
 668 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_setclientid *setclid)
 669 {
 670         u32                     ip_addr = rqstp->rq_addr.sin_addr.s_addr;
 671         struct xdr_netobj       clname = {
 672                 .len = setclid->se_namelen,
 673                 .data = setclid->se_name,
 674         };
 675         nfs4_verifier           clverifier = setclid->se_verf;
 676         unsigned int            strhashval;
 677         struct nfs4_client      *conf, *unconf, *new;
 678         int                     status;
 679         char                    dname[HEXDIR_LEN];
 680
 681         if (!check_name(clname))
 682                 return nfserr_inval;
 683
 684         status = nfs4_make_rec_clidname(dname, &clname);
 685         if (status)
 686                 return status;
 687
 688         /*
 689          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 690          * We get here on a DRC miss.
 691          */
 692
 693         strhashval = clientstr_hashval(dname);
 694
 695         nfs4_lock_state();
 696         conf = find_confirmed_client_by_str(dname, strhashval);
 697         if (conf) {
 698                 /*
 699                  * CASE 0:
 700                  * clname match, confirmed, different principal
 701                  * or different ip_address
 702                  */
 703                 status = nfserr_clid_inuse;
 704                 if (!cmp_creds(&conf->cl_cred, &rqstp->rq_cred)
 705                                 || conf->cl_addr != ip_addr) {
 706                         printk("NFSD: setclientid: string in use by client"
 707                         "(clientid %08x/%08x)\n",
 708                         conf->cl_clientid.cl_boot, conf->cl_clientid.cl_id);
 709                         goto out;
 710                 }
 711         }
 712         unconf = find_unconfirmed_client_by_str(dname, strhashval);
 713         status = nfserr_resource;
 714         if (!conf) {
 715                 /*
 716                  * CASE 4:
 717                  * placed first, because it is the normal case.
 718                  */
 719                 if (unconf)
 720                         expire_client(unconf);
 721                 new = create_client(clname, dname);
 722                 if (new == NULL)
 723                         goto out;
 724                 copy_verf(new, &clverifier);
 725                 new->cl_addr = ip_addr;
 726                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
 727                 gen_clid(new);
 728                 gen_confirm(new);
 729                 gen_callback(new, setclid);
 730                 add_to_unconfirmed(new, strhashval);
 731         } else if (cmp_verf(&conf->cl_verifier, &clverifier)) {
 732                 /*
 733                  * CASE 1:
 734                  * cl_name match, confirmed, principal match
 735                  * verifier match: probable callback update
 736                  *
 737                  * remove any unconfirmed nfs4_client with
 738                  * matching cl_name, cl_verifier, and cl_clientid
 739                  *
 740                  * create and insert an unconfirmed nfs4_client with same
 741                  * cl_name, cl_verifier, and cl_clientid as existing
 742                  * nfs4_client,  but with the new callback info and a
 743                  * new cl_confirm
 744                  */
 745                 if (unconf) {
 746                         /* Note this is removing unconfirmed {*x***},
 747                          * which is stronger than RFC recommended {vxc**}.
 748                          * This has the advantage that there is at most
 749                          * one {*x***} in either list at any time.
 750                          */
 751                         expire_client(unconf);
 752                 }
 753                 new = create_client(clname, dname);
 754                 if (new == NULL)
 755                         goto out;
 756                 copy_verf(new,&conf->cl_verifier);
 757                 new->cl_addr = ip_addr;
 758                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
 759                 copy_clid(new, conf);
 760                 gen_confirm(new);
 761                 gen_callback(new, setclid);
 762                 add_to_unconfirmed(new,strhashval);
 763         } else if (!unconf) {
 764                 /*
 765                  * CASE 2:
 766                  * clname match, confirmed, principal match
 767                  * verfier does not match
 768                  * no unconfirmed. create a new unconfirmed nfs4_client
 769                  * using input clverifier, clname, and callback info
 770                  * and generate a new cl_clientid and cl_confirm.
 771                  */
 772                 new = create_client(clname, dname);
 773                 if (new == NULL)
 774                         goto out;
 775                 copy_verf(new,&clverifier);
 776                 new->cl_addr = ip_addr;
 777                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
 778                 gen_clid(new);
 779                 gen_confirm(new);
 780                 gen_callback(new, setclid);
 781                 add_to_unconfirmed(new, strhashval);
 782         } else if (!cmp_verf(&conf->cl_confirm, &unconf->cl_confirm)) {
 783                 /*
 784                  * CASE3:
 785                  * confirmed found (name, principal match)
 786                  * confirmed verifier does not match input clverifier
 787                  *
 788                  * unconfirmed found (name match)
 789                  * confirmed->cl_confirm != unconfirmed->cl_confirm
 790                  *
 791                  * remove unconfirmed.
 792                  *
 793                  * create an unconfirmed nfs4_client
 794                  * with same cl_name as existing confirmed nfs4_client,
 795                  * but with new callback info, new cl_clientid,
 796                  * new cl_verifier and a new cl_confirm
 797                  */
 798                 expire_client(unconf);
 799                 new = create_client(clname, dname);
 800                 if (new == NULL)
 801                         goto out;
 802                 copy_verf(new,&clverifier);
 803                 new->cl_addr = ip_addr;
 804                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
 805                 gen_clid(new);
 806                 gen_confirm(new);
 807                 gen_callback(new, setclid);
 808                 add_to_unconfirmed(new, strhashval);
 809         } else {
 810                 /* No cases hit !!! */
 811                 status = nfserr_inval;
 812                 goto out;
 813
 814         }
 815         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
 816         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
 817         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
 818         status = nfs_ok;
 819 out:
 820         nfs4_unlock_state();
 821         return status;
 822 }
 823
 824
 825 /*
 826  * RFC 3010 has a complex implmentation description of processing a
 827  * SETCLIENTID_CONFIRM request consisting of 4 bullets describing
 828  * processing on a DRC miss, labeled as CASE1 - CASE4 below.
 829  *
 830  * NOTE: callback information will be processed here in a future patch
 831  */
 832 int
 833 nfsd4_setclientid_confirm(struct svc_rqst *rqstp, struct nfsd4_setclientid_confirm *setclientid_confirm)
 834 {
 835         u32 ip_addr = rqstp->rq_addr.sin_addr.s_addr;
 836         struct nfs4_client *conf, *unconf;
 837         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
 838         clientid_t * clid = &setclientid_confirm->sc_clientid;
 839         int status;
 840
 841         if (STALE_CLIENTID(clid))
 842                 return nfserr_stale_clientid;
 843         /*
 844          * XXX The Duplicate Request Cache (DRC) has been checked (??)
 845          * We get here on a DRC miss.
 846          */
 847
 848         nfs4_lock_state();
 849
 850         conf = find_confirmed_client(clid);
 851         unconf = find_unconfirmed_client(clid);
 852
 853         status = nfserr_clid_inuse;
 854         if (conf && conf->cl_addr != ip_addr)
 855                 goto out;
 856         if (unconf && unconf->cl_addr != ip_addr)
 857                 goto out;
 858
 859         if ((conf && unconf) &&
 860             (cmp_verf(&unconf->cl_confirm, &confirm)) &&
 861             (cmp_verf(&conf->cl_verifier, &unconf->cl_verifier)) &&
 862             (same_name(conf->cl_recdir,unconf->cl_recdir))  &&
 863             (!cmp_verf(&conf->cl_confirm, &unconf->cl_confirm))) {
 864                 /* CASE 1:
 865                 * unconf record that matches input clientid and input confirm.
 866                 * conf record that matches input clientid.
 867                 * conf and unconf records match names, verifiers
 868                 */
 869                 if (!cmp_creds(&conf->cl_cred, &unconf->cl_cred))
 870                         status = nfserr_clid_inuse;
 871                 else {
 872                         /* XXX: We just turn off callbacks until we can handle
 873                           * change request correctly. */
 874                         atomic_set(&conf->cl_callback.cb_set, 0);
 875                         gen_confirm(conf);
 876                         nfsd4_remove_clid_dir(unconf);
 877                         expire_client(unconf);
 878                         status = nfs_ok;
 879
 880                 }
 881         } else if ((conf && !unconf) ||
 882             ((conf && unconf) &&
 883              (!cmp_verf(&conf->cl_verifier, &unconf->cl_verifier) ||
 884               !same_name(conf->cl_recdir, unconf->cl_recdir)))) {
 885                 /* CASE 2:
 886                  * conf record that matches input clientid.
 887                  * if unconf record matches input clientid, then
 888                  * unconf->cl_name or unconf->cl_verifier don't match the
 889                  * conf record.
 890                  */
 891                 if (!cmp_creds(&conf->cl_cred,&rqstp->rq_cred))
 892                         status = nfserr_clid_inuse;
 893                 else
 894                         status = nfs_ok;
 895         } else if (!conf && unconf
 896                         && cmp_verf(&unconf->cl_confirm, &confirm)) {
 897                 /* CASE 3:
 898                  * conf record not found.
 899                  * unconf record found.
 900                  * unconf->cl_confirm matches input confirm
 901                  */
 902                 if (!cmp_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
 903                         status = nfserr_clid_inuse;
 904                 } else {
 905                         unsigned int hash =
 906                                 clientstr_hashval(unconf->cl_recdir);
 907                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
 908                                                                         hash);
 909                         if (conf) {
 910                                 nfsd4_remove_clid_dir(conf);
 911                                 expire_client(conf);
 912                         }
 913                         move_to_confirmed(unconf);
 914                         conf = unconf;
 915                         status = nfs_ok;
 916                 }
 917         } else if ((!conf || (conf && !cmp_verf(&conf->cl_confirm, &confirm)))
 918             && (!unconf || (unconf && !cmp_verf(&unconf->cl_confirm,
 919                                                                 &confirm)))) {
 920                 /* CASE 4:
 921                  * conf record not found, or if conf, conf->cl_confirm does not
 922                  * match input confirm.
 923                  * unconf record not found, or if unconf, unconf->cl_confirm
 924                  * does not match input confirm.
 925                  */
 926                 status = nfserr_stale_clientid;
 927         } else {
 928                 /* check that we have hit one of the cases...*/
 929                 status = nfserr_clid_inuse;
 930         }
 931 out:
 932         if (!status)
 933                 nfsd4_probe_callback(conf);
 934         nfs4_unlock_state();
 935         return status;
 936 }
 937
 938 /*
 939  * Open owner state (share locks)
 940  */
 941
 942 /* hash tables for nfs4_stateowner */
 943 #define OWNER_HASH_BITS              8
 944 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 945 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 946
 947 #define ownerid_hashval(id) \
 948         ((id) & OWNER_HASH_MASK)
 949 #define ownerstr_hashval(clientid, ownername) \
 950         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
 951
 952 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
 953 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 954
 955 /* hash table for nfs4_file */
 956 #define FILE_HASH_BITS                   8
 957 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 958 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
 959 /* hash table for (open)nfs4_stateid */
 960 #define STATEID_HASH_BITS              10
 961 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
 962 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
 963
 964 #define file_hashval(x) \
 965         hash_ptr(x, FILE_HASH_BITS)
 966 #define stateid_hashval(owner_id, file_id)  \
 967         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
 968
 969 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 970 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 971
 972 /* OPEN Share state helper functions */
 973 static inline struct nfs4_file *
 974 alloc_init_file(struct inode *ino)
 975 {
 976         struct nfs4_file *fp;
 977         unsigned int hashval = file_hashval(ino);
 978
 979         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
 980         if (fp) {
 981                 kref_init(&fp->fi_ref);
 982                 INIT_LIST_HEAD(&fp->fi_hash);
 983                 INIT_LIST_HEAD(&fp->fi_stateids);
 984                 INIT_LIST_HEAD(&fp->fi_delegations);
 985                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
 986                 fp->fi_inode = igrab(ino);
 987                 fp->fi_id = current_fileid++;
 988                 return fp;
 989         }
 990         return NULL;
 991 }
 992
 993 static void
 994 nfsd4_free_slab(kmem_cache_t **slab)
 995 {
 996         int status;
 997
 998         if (*slab == NULL)
 999                 return;
1000         status = kmem_cache_destroy(*slab);
1001         *slab = NULL;
1002         WARN_ON(status);
1003 }
1004
1005 static void
1006 nfsd4_free_slabs(void)
1007 {
1008         nfsd4_free_slab(&stateowner_slab);
1009         nfsd4_free_slab(&file_slab);
1010         nfsd4_free_slab(&stateid_slab);
1011         nfsd4_free_slab(&deleg_slab);
1012 }
1013
1014 static int
1015 nfsd4_init_slabs(void)
1016 {
1017         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1018                         sizeof(struct nfs4_stateowner), 0, 0, NULL, NULL);
1019         if (stateowner_slab == NULL)
1020                 goto out_nomem;
1021         file_slab = kmem_cache_create("nfsd4_files",
1022                         sizeof(struct nfs4_file), 0, 0, NULL, NULL);
1023         if (file_slab == NULL)
1024                 goto out_nomem;
1025         stateid_slab = kmem_cache_create("nfsd4_stateids",
1026                         sizeof(struct nfs4_stateid), 0, 0, NULL, NULL);
1027         if (stateid_slab == NULL)
1028                 goto out_nomem;
1029         deleg_slab = kmem_cache_create("nfsd4_delegations",
1030                         sizeof(struct nfs4_delegation), 0, 0, NULL, NULL);
1031         if (deleg_slab == NULL)
1032                 goto out_nomem;
1033         return 0;
1034 out_nomem:
1035         nfsd4_free_slabs();
1036         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1037         return -ENOMEM;
1038 }
1039
1040 void
1041 nfs4_free_stateowner(struct kref *kref)
1042 {
1043         struct nfs4_stateowner *sop =
1044                 container_of(kref, struct nfs4_stateowner, so_ref);
1045         kfree(sop->so_owner.data);
1046         kmem_cache_free(stateowner_slab, sop);
1047 }
1048
1049 static inline struct nfs4_stateowner *
1050 alloc_stateowner(struct xdr_netobj *owner)
1051 {
1052         struct nfs4_stateowner *sop;
1053
1054         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1055                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1056                         memcpy(sop->so_owner.data, owner->data, owner->len);
1057                         sop->so_owner.len = owner->len;
1058                         kref_init(&sop->so_ref);
1059                         return sop;
1060                 }
1061                 kmem_cache_free(stateowner_slab, sop);
1062         }
1063         return NULL;
1064 }
1065
1066 static struct nfs4_stateowner *
1067 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1068         struct nfs4_stateowner *sop;
1069         struct nfs4_replay *rp;
1070         unsigned int idhashval;
1071
1072         if (!(sop = alloc_stateowner(&open->op_owner)))
1073                 return NULL;
1074         idhashval = ownerid_hashval(current_ownerid);
1075         INIT_LIST_HEAD(&sop->so_idhash);
1076         INIT_LIST_HEAD(&sop->so_strhash);
1077         INIT_LIST_HEAD(&sop->so_perclient);
1078         INIT_LIST_HEAD(&sop->so_stateids);
1079         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1080         INIT_LIST_HEAD(&sop->so_close_lru);
1081         sop->so_time = 0;
1082         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1083         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1084         list_add(&sop->so_perclient, &clp->cl_openowners);
1085         sop->so_is_open_owner = 1;
1086         sop->so_id = current_ownerid++;
1087         sop->so_client = clp;
1088         sop->so_seqid = open->op_seqid;
1089         sop->so_confirmed = 0;
1090         rp = &sop->so_replay;
1091         rp->rp_status = NFSERR_SERVERFAULT;
1092         rp->rp_buflen = 0;
1093         rp->rp_buf = rp->rp_ibuf;
1094         return sop;
1095 }
1096
1097 static void
1098 release_stateid_lockowners(struct nfs4_stateid *open_stp)
1099 {
1100         struct nfs4_stateowner *lock_sop;
1101
1102         while (!list_empty(&open_stp->st_lockowners)) {
1103                 lock_sop = list_entry(open_stp->st_lockowners.next,
1104                                 struct nfs4_stateowner, so_perstateid);
1105                 /* list_del(&open_stp->st_lockowners);  */
1106                 BUG_ON(lock_sop->so_is_open_owner);
1107                 release_stateowner(lock_sop);
1108         }
1109 }
1110
1111 static void
1112 unhash_stateowner(struct nfs4_stateowner *sop)
1113 {
1114         struct nfs4_stateid *stp;
1115
1116         list_del(&sop->so_idhash);
1117         list_del(&sop->so_strhash);
1118         if (sop->so_is_open_owner)
1119                 list_del(&sop->so_perclient);
1120         list_del(&sop->so_perstateid);
1121         while (!list_empty(&sop->so_stateids)) {
1122                 stp = list_entry(sop->so_stateids.next,
1123                         struct nfs4_stateid, st_perstateowner);
1124                 if (sop->so_is_open_owner)
1125                         release_stateid(stp, OPEN_STATE);
1126                 else
1127                         release_stateid(stp, LOCK_STATE);
1128         }
1129 }
1130
1131 static void
1132 release_stateowner(struct nfs4_stateowner *sop)
1133 {
1134         unhash_stateowner(sop);
1135         list_del(&sop->so_close_lru);
1136         nfs4_put_stateowner(sop);
1137 }
1138
1139 static inline void
1140 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1141         struct nfs4_stateowner *sop = open->op_stateowner;
1142         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1143
1144         INIT_LIST_HEAD(&stp->st_hash);
1145         INIT_LIST_HEAD(&stp->st_perstateowner);
1146         INIT_LIST_HEAD(&stp->st_lockowners);
1147         INIT_LIST_HEAD(&stp->st_perfile);
1148         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1149         list_add(&stp->st_perstateowner, &sop->so_stateids);
1150         list_add(&stp->st_perfile, &fp->fi_stateids);
1151         stp->st_stateowner = sop;
1152         get_nfs4_file(fp);
1153         stp->st_file = fp;
1154         stp->st_stateid.si_boot = boot_time;
1155         stp->st_stateid.si_stateownerid = sop->so_id;
1156         stp->st_stateid.si_fileid = fp->fi_id;
1157         stp->st_stateid.si_generation = 0;
1158         stp->st_access_bmap = 0;
1159         stp->st_deny_bmap = 0;
1160         __set_bit(open->op_share_access, &stp->st_access_bmap);
1161         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1162         stp->st_openstp = NULL;
1163 }
1164
1165 static void
1166 release_stateid(struct nfs4_stateid *stp, int flags)
1167 {
1168         struct file *filp = stp->st_vfs_file;
1169
1170         list_del(&stp->st_hash);
1171         list_del(&stp->st_perfile);
1172         list_del(&stp->st_perstateowner);
1173         if (flags & OPEN_STATE) {
1174                 release_stateid_lockowners(stp);
1175                 stp->st_vfs_file = NULL;
1176                 nfsd_close(filp);
1177         } else if (flags & LOCK_STATE)
1178                 locks_remove_posix(filp, (fl_owner_t) stp->st_stateowner);
1179         put_nfs4_file(stp->st_file);
1180         kmem_cache_free(stateid_slab, stp);
1181         stp = NULL;
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         unhash_stateowner(sop);
1190         list_add_tail(&sop->so_close_lru, &close_lru);
1191         sop->so_time = get_seconds();
1192 }
1193
1194 static void
1195 release_state_owner(struct nfs4_stateid *stp, int flag)
1196 {
1197         struct nfs4_stateowner *sop = stp->st_stateowner;
1198
1199         dprintk("NFSD: release_state_owner\n");
1200         release_stateid(stp, flag);
1201
1202         /* place unused nfs4_stateowners on so_close_lru list to be
1203          * released by the laundromat service after the lease period
1204          * to enable us to handle CLOSE replay
1205          */
1206         if (sop->so_confirmed && list_empty(&sop->so_stateids))
1207                 move_to_close_lru(sop);
1208 }
1209
1210 static int
1211 cmp_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner, clientid_t *clid) {
1212         return ((sop->so_owner.len == owner->len) &&
1213          !memcmp(sop->so_owner.data, owner->data, owner->len) &&
1214           (sop->so_client->cl_clientid.cl_id == clid->cl_id));
1215 }
1216
1217 static struct nfs4_stateowner *
1218 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1219 {
1220         struct nfs4_stateowner *so = NULL;
1221
1222         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1223                 if (cmp_owner_str(so, &open->op_owner, &open->op_clientid))
1224                         return so;
1225         }
1226         return NULL;
1227 }
1228
1229 /* search file_hashtbl[] for file */
1230 static struct nfs4_file *
1231 find_file(struct inode *ino)
1232 {
1233         unsigned int hashval = file_hashval(ino);
1234         struct nfs4_file *fp;
1235
1236         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1237                 if (fp->fi_inode == ino) {
1238                         get_nfs4_file(fp);
1239                         return fp;
1240                 }
1241         }
1242         return NULL;
1243 }
1244
1245 #define TEST_ACCESS(x) ((x > 0 || x < 4)?1:0)
1246 #define TEST_DENY(x) ((x >= 0 || x < 5)?1:0)
1247
1248 static void
1249 set_access(unsigned int *access, unsigned long bmap) {
1250         int i;
1251
1252         *access = 0;
1253         for (i = 1; i < 4; i++) {
1254                 if (test_bit(i, &bmap))
1255                         *access |= i;
1256         }
1257 }
1258
1259 static void
1260 set_deny(unsigned int *deny, unsigned long bmap) {
1261         int i;
1262
1263         *deny = 0;
1264         for (i = 0; i < 4; i++) {
1265                 if (test_bit(i, &bmap))
1266                         *deny |= i ;
1267         }
1268 }
1269
1270 static int
1271 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1272         unsigned int access, deny;
1273
1274         set_access(&access, stp->st_access_bmap);
1275         set_deny(&deny, stp->st_deny_bmap);
1276         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1277                 return 0;
1278         return 1;
1279 }
1280
1281 /*
1282  * Called to check deny when READ with all zero stateid or
1283  * WRITE with all zero or all one stateid
1284  */
1285 static int
1286 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1287 {
1288         struct inode *ino = current_fh->fh_dentry->d_inode;
1289         struct nfs4_file *fp;
1290         struct nfs4_stateid *stp;
1291         int ret;
1292
1293         dprintk("NFSD: nfs4_share_conflict\n");
1294
1295         fp = find_file(ino);
1296         if (!fp)
1297                 return nfs_ok;
1298         ret = nfserr_locked;
1299         /* Search for conflicting share reservations */
1300         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1301                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1302                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1303                         goto out;
1304         }
1305         ret = nfs_ok;
1306 out:
1307         put_nfs4_file(fp);
1308         return ret;
1309 }
1310
1311 static inline void
1312 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1313 {
1314         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1315                 put_write_access(filp->f_dentry->d_inode);
1316                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1317         }
1318 }
1319
1320 /*
1321  * Recall a delegation
1322  */
1323 static int
1324 do_recall(void *__dp)
1325 {
1326         struct nfs4_delegation *dp = __dp;
1327
1328         daemonize("nfsv4-recall");
1329
1330         nfsd4_cb_recall(dp);
1331         return 0;
1332 }
1333
1334 /*
1335  * Spawn a thread to perform a recall on the delegation represented
1336  * by the lease (file_lock)
1337  *
1338  * Called from break_lease() with lock_kernel() held.
1339  * Note: we assume break_lease will only call this *once* for any given
1340  * lease.
1341  */
1342 static
1343 void nfsd_break_deleg_cb(struct file_lock *fl)
1344 {
1345         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1346         struct task_struct *t;
1347
1348         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1349         if (!dp)
1350                 return;
1351
1352         /* We're assuming the state code never drops its reference
1353          * without first removing the lease.  Since we're in this lease
1354          * callback (and since the lease code is serialized by the kernel
1355          * lock) we know the server hasn't removed the lease yet, we know
1356          * it's safe to take a reference: */
1357         atomic_inc(&dp->dl_count);
1358
1359         spin_lock(&recall_lock);
1360         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1361         spin_unlock(&recall_lock);
1362
1363         /* only place dl_time is set. protected by lock_kernel*/
1364         dp->dl_time = get_seconds();
1365
1366         /* XXX need to merge NFSD_LEASE_TIME with fs/locks.c:lease_break_time */
1367         fl->fl_break_time = jiffies + NFSD_LEASE_TIME * HZ;
1368
1369         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1370         if (IS_ERR(t)) {
1371                 struct nfs4_client *clp = dp->dl_client;
1372
1373                 printk(KERN_INFO "NFSD: Callback thread failed for "
1374                         "for client (clientid %08x/%08x)\n",
1375                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1376                 nfs4_put_delegation(dp);
1377         }
1378 }
1379
1380 /*
1381  * The file_lock is being reapd.
1382  *
1383  * Called by locks_free_lock() with lock_kernel() held.
1384  */
1385 static
1386 void nfsd_release_deleg_cb(struct file_lock *fl)
1387 {
1388         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1389
1390         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1391
1392         if (!(fl->fl_flags & FL_LEASE) || !dp)
1393                 return;
1394         dp->dl_flock = NULL;
1395 }
1396
1397 /*
1398  * Set the delegation file_lock back pointer.
1399  *
1400  * Called from __setlease() with lock_kernel() held.
1401  */
1402 static
1403 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1404 {
1405         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1406
1407         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1408         if (!dp)
1409                 return;
1410         dp->dl_flock = new;
1411 }
1412
1413 /*
1414  * Called from __setlease() with lock_kernel() held
1415  */
1416 static
1417 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1418 {
1419         struct nfs4_delegation *onlistd =
1420                 (struct nfs4_delegation *)onlist->fl_owner;
1421         struct nfs4_delegation *tryd =
1422                 (struct nfs4_delegation *)try->fl_owner;
1423
1424         if (onlist->fl_lmops != try->fl_lmops)
1425                 return 0;
1426
1427         return onlistd->dl_client == tryd->dl_client;
1428 }
1429
1430
1431 static
1432 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1433 {
1434         if (arg & F_UNLCK)
1435                 return lease_modify(onlist, arg);
1436         else
1437                 return -EAGAIN;
1438 }
1439
1440 static struct lock_manager_operations nfsd_lease_mng_ops = {
1441         .fl_break = nfsd_break_deleg_cb,
1442         .fl_release_private = nfsd_release_deleg_cb,
1443         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1444         .fl_mylease = nfsd_same_client_deleg_cb,
1445         .fl_change = nfsd_change_deleg_cb,
1446 };
1447
1448
1449 /*
1450  * nfsd4_process_open1()
1451  *      lookup stateowner.
1452  *              found:
1453  *                      check confirmed
1454  *                              confirmed:
1455  *                                      check seqid
1456  *                              not confirmed:
1457  *                                      delete owner
1458  *                                      create new owner
1459  *              notfound:
1460  *                      verify clientid
1461  *                      create new owner
1462  *
1463  * called with nfs4_lock_state() held.
1464  */
1465 int
1466 nfsd4_process_open1(struct nfsd4_open *open)
1467 {
1468         int status;
1469         clientid_t *clientid = &open->op_clientid;
1470         struct nfs4_client *clp = NULL;
1471         unsigned int strhashval;
1472         struct nfs4_stateowner *sop = NULL;
1473
1474         status = nfserr_inval;
1475         if (!check_name(open->op_owner))
1476                 goto out;
1477
1478         if (STALE_CLIENTID(&open->op_clientid))
1479                 return nfserr_stale_clientid;
1480
1481         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1482         sop = find_openstateowner_str(strhashval, open);
1483         if (sop) {
1484                 open->op_stateowner = sop;
1485                 /* check for replay */
1486                 if (open->op_seqid == sop->so_seqid - 1){
1487                         if (sop->so_replay.rp_buflen)
1488                                 return NFSERR_REPLAY_ME;
1489                         else {
1490                                 /* The original OPEN failed so spectacularly
1491                                  * that we don't even have replay data saved!
1492                                  * Therefore, we have no choice but to continue
1493                                  * processing this OPEN; presumably, we'll
1494                                  * fail again for the same reason.
1495                                  */
1496                                 dprintk("nfsd4_process_open1:"
1497                                         " replay with no replay cache\n");
1498                                 goto renew;
1499                         }
1500                 } else if (sop->so_confirmed) {
1501                         if (open->op_seqid == sop->so_seqid)
1502                                 goto renew;
1503                         status = nfserr_bad_seqid;
1504                         goto out;
1505                 } else {
1506                         /* If we get here, we received an OPEN for an
1507                          * unconfirmed nfs4_stateowner. Since the seqid's are
1508                          * different, purge the existing nfs4_stateowner, and
1509                          * instantiate a new one.
1510                          */
1511                         clp = sop->so_client;
1512                         release_stateowner(sop);
1513                 }
1514         } else {
1515                 /* nfs4_stateowner not found.
1516                  * Verify clientid and instantiate new nfs4_stateowner.
1517                  * If verify fails this is presumably the result of the
1518                  * client's lease expiring.
1519                  */
1520                 status = nfserr_expired;
1521                 clp = find_confirmed_client(clientid);
1522                 if (clp == NULL)
1523                         goto out;
1524         }
1525         status = nfserr_resource;
1526         sop = alloc_init_open_stateowner(strhashval, clp, open);
1527         if (sop == NULL)
1528                 goto out;
1529         open->op_stateowner = sop;
1530 renew:
1531         status = nfs_ok;
1532         renew_client(sop->so_client);
1533 out:
1534         return status;
1535 }
1536
1537 static inline int
1538 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1539 {
1540         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1541                 return nfserr_openmode;
1542         else
1543                 return nfs_ok;
1544 }
1545
1546 static struct nfs4_delegation *
1547 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1548 {
1549         struct nfs4_delegation *dp;
1550
1551         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1552                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1553                         return dp;
1554         }
1555         return NULL;
1556 }
1557
1558 static int
1559 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1560                 struct nfs4_delegation **dp)
1561 {
1562         int flags;
1563         int status = nfserr_bad_stateid;
1564
1565         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1566         if (*dp == NULL)
1567                 goto out;
1568         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1569                                                 RD_STATE : WR_STATE;
1570         status = nfs4_check_delegmode(*dp, flags);
1571         if (status)
1572                 *dp = NULL;
1573 out:
1574         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1575                 return nfs_ok;
1576         if (status)
1577                 return status;
1578         open->op_stateowner->so_confirmed = 1;
1579         return nfs_ok;
1580 }
1581
1582 static int
1583 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1584 {
1585         struct nfs4_stateid *local;
1586         int status = nfserr_share_denied;
1587         struct nfs4_stateowner *sop = open->op_stateowner;
1588
1589         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1590                 /* ignore lock owners */
1591                 if (local->st_stateowner->so_is_open_owner == 0)
1592                         continue;
1593                 /* remember if we have seen this open owner */
1594                 if (local->st_stateowner == sop)
1595                         *stpp = local;
1596                 /* check for conflicting share reservations */
1597                 if (!test_share(local, open))
1598                         goto out;
1599         }
1600         status = 0;
1601 out:
1602         return status;
1603 }
1604
1605 static inline struct nfs4_stateid *
1606 nfs4_alloc_stateid(void)
1607 {
1608         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1609 }
1610
1611 static int
1612 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1613                 struct nfs4_delegation *dp,
1614                 struct svc_fh *cur_fh, int flags)
1615 {
1616         struct nfs4_stateid *stp;
1617
1618         stp = nfs4_alloc_stateid();
1619         if (stp == NULL)
1620                 return nfserr_resource;
1621
1622         if (dp) {
1623                 get_file(dp->dl_vfs_file);
1624                 stp->st_vfs_file = dp->dl_vfs_file;
1625         } else {
1626                 int status;
1627                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1628                                 &stp->st_vfs_file);
1629                 if (status) {
1630                         if (status == nfserr_dropit)
1631                                 status = nfserr_jukebox;
1632                         kmem_cache_free(stateid_slab, stp);
1633                         return status;
1634                 }
1635         }
1636         *stpp = stp;
1637         return 0;
1638 }
1639
1640 static inline int
1641 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1642                 struct nfsd4_open *open)
1643 {
1644         struct iattr iattr = {
1645                 .ia_valid = ATTR_SIZE,
1646                 .ia_size = 0,
1647         };
1648         if (!open->op_truncate)
1649                 return 0;
1650         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1651                 return -EINVAL;
1652         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1653 }
1654
1655 static int
1656 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1657 {
1658         struct file *filp = stp->st_vfs_file;
1659         struct inode *inode = filp->f_dentry->d_inode;
1660         unsigned int share_access;
1661         int status;
1662
1663         set_access(&share_access, stp->st_access_bmap);
1664         share_access = ~share_access;
1665         share_access &= open->op_share_access;
1666
1667         if (!(share_access & NFS4_SHARE_ACCESS_WRITE))
1668                 return nfsd4_truncate(rqstp, cur_fh, open);
1669
1670         status = get_write_access(inode);
1671         if (status)
1672                 return nfserrno(status);
1673         status = nfsd4_truncate(rqstp, cur_fh, open);
1674         if (status) {
1675                 put_write_access(inode);
1676                 return status;
1677         }
1678         /* remember the open */
1679         filp->f_mode = (filp->f_mode | FMODE_WRITE) & ~FMODE_READ;
1680         set_bit(open->op_share_access, &stp->st_access_bmap);
1681         set_bit(open->op_share_deny, &stp->st_deny_bmap);
1682
1683         return nfs_ok;
1684 }
1685
1686
1687 static void
1688 nfs4_set_claim_prev(struct nfsd4_open *open)
1689 {
1690         open->op_stateowner->so_confirmed = 1;
1691         open->op_stateowner->so_client->cl_firststate = 1;
1692 }
1693
1694 /*
1695  * Attempt to hand out a delegation.
1696  */
1697 static void
1698 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1699 {
1700         struct nfs4_delegation *dp;
1701         struct nfs4_stateowner *sop = stp->st_stateowner;
1702         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1703         struct file_lock fl, *flp = &fl;
1704         int status, flag = 0;
1705
1706         flag = NFS4_OPEN_DELEGATE_NONE;
1707         open->op_recall = 0;
1708         switch (open->op_claim_type) {
1709                 case NFS4_OPEN_CLAIM_PREVIOUS:
1710                         if (!atomic_read(&cb->cb_set))
1711                                 open->op_recall = 1;
1712                         flag = open->op_delegate_type;
1713                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1714                                 goto out;
1715                         break;
1716                 case NFS4_OPEN_CLAIM_NULL:
1717                         /* Let's not give out any delegations till everyone's
1718                          * had the chance to reclaim theirs.... */
1719                         if (nfs4_in_grace())
1720                                 goto out;
1721                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1722                                 goto out;
1723                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1724                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1725                         else
1726                                 flag = NFS4_OPEN_DELEGATE_READ;
1727                         break;
1728                 default:
1729                         goto out;
1730         }
1731
1732         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1733         if (dp == NULL) {
1734                 flag = NFS4_OPEN_DELEGATE_NONE;
1735                 goto out;
1736         }
1737         locks_init_lock(&fl);
1738         fl.fl_lmops = &nfsd_lease_mng_ops;
1739         fl.fl_flags = FL_LEASE;
1740         fl.fl_end = OFFSET_MAX;
1741         fl.fl_owner =  (fl_owner_t)dp;
1742         fl.fl_file = stp->st_vfs_file;
1743         fl.fl_pid = current->tgid;
1744
1745         /* setlease checks to see if delegation should be handed out.
1746          * the lock_manager callbacks fl_mylease and fl_change are used
1747          */
1748         if ((status = setlease(stp->st_vfs_file,
1749                 flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK, &flp))) {
1750                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1751                 unhash_delegation(dp);
1752                 flag = NFS4_OPEN_DELEGATE_NONE;
1753                 goto out;
1754         }
1755
1756         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1757
1758         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1759                      dp->dl_stateid.si_boot,
1760                      dp->dl_stateid.si_stateownerid,
1761                      dp->dl_stateid.si_fileid,
1762                      dp->dl_stateid.si_generation);
1763 out:
1764         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1765                         && flag == NFS4_OPEN_DELEGATE_NONE
1766                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1767                 printk("NFSD: WARNING: refusing delegation reclaim\n");
1768         open->op_delegate_type = flag;
1769 }
1770
1771 /*
1772  * called with nfs4_lock_state() held.
1773  */
1774 int
1775 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1776 {
1777         struct nfs4_file *fp = NULL;
1778         struct inode *ino = current_fh->fh_dentry->d_inode;
1779         struct nfs4_stateid *stp = NULL;
1780         struct nfs4_delegation *dp = NULL;
1781         int status;
1782
1783         if (nfs4_in_grace() && open->op_claim_type != NFS4_OPEN_CLAIM_PREVIOUS)
1784                 return nfserr_grace;
1785
1786         if (!nfs4_in_grace() && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1787                 return nfserr_no_grace;
1788
1789         status = nfserr_inval;
1790         if (!TEST_ACCESS(open->op_share_access) || !TEST_DENY(open->op_share_deny))
1791                 goto out;
1792         /*
1793          * Lookup file; if found, lookup stateid and check open request,
1794          * and check for delegations in the process of being recalled.
1795          * If not found, create the nfs4_file struct
1796          */
1797         fp = find_file(ino);
1798         if (fp) {
1799                 if ((status = nfs4_check_open(fp, open, &stp)))
1800                         goto out;
1801                 status = nfs4_check_deleg(fp, open, &dp);
1802                 if (status)
1803                         goto out;
1804         } else {
1805                 status = nfserr_bad_stateid;
1806                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1807                         goto out;
1808                 status = nfserr_resource;
1809                 fp = alloc_init_file(ino);
1810                 if (fp == NULL)
1811                         goto out;
1812         }
1813
1814         /*
1815          * OPEN the file, or upgrade an existing OPEN.
1816          * If truncate fails, the OPEN fails.
1817          */
1818         if (stp) {
1819                 /* Stateid was found, this is an OPEN upgrade */
1820                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1821                 if (status)
1822                         goto out;
1823                 update_stateid(&stp->st_stateid);
1824         } else {
1825                 /* Stateid was not found, this is a new OPEN */
1826                 int flags = 0;
1827                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1828                         flags = MAY_WRITE;
1829                 else
1830                         flags = MAY_READ;
1831                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1832                 if (status)
1833                         goto out;
1834                 init_stateid(stp, fp, open);
1835                 status = nfsd4_truncate(rqstp, current_fh, open);
1836                 if (status) {
1837                         release_stateid(stp, OPEN_STATE);
1838                         goto out;
1839                 }
1840         }
1841         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1842
1843         /*
1844         * Attempt to hand out a delegation. No error return, because the
1845         * OPEN succeeds even if we fail.
1846         */
1847         nfs4_open_delegation(current_fh, open, stp);
1848
1849         status = nfs_ok;
1850
1851         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1852                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1853                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1854 out:
1855         if (fp)
1856                 put_nfs4_file(fp);
1857         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1858                 nfs4_set_claim_prev(open);
1859         /*
1860         * To finish the open response, we just need to set the rflags.
1861         */
1862         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1863         if (!open->op_stateowner->so_confirmed)
1864                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1865
1866         return status;
1867 }
1868
1869 static struct workqueue_struct *laundry_wq;
1870 static struct work_struct laundromat_work;
1871 static void laundromat_main(void *);
1872 static DECLARE_WORK(laundromat_work, laundromat_main, NULL);
1873
1874 int
1875 nfsd4_renew(clientid_t *clid)
1876 {
1877         struct nfs4_client *clp;
1878         int status;
1879
1880         nfs4_lock_state();
1881         dprintk("process_renew(%08x/%08x): starting\n",
1882                         clid->cl_boot, clid->cl_id);
1883         status = nfserr_stale_clientid;
1884         if (STALE_CLIENTID(clid))
1885                 goto out;
1886         clp = find_confirmed_client(clid);
1887         status = nfserr_expired;
1888         if (clp == NULL) {
1889                 /* We assume the client took too long to RENEW. */
1890                 dprintk("nfsd4_renew: clientid not found!\n");
1891                 goto out;
1892         }
1893         renew_client(clp);
1894         status = nfserr_cb_path_down;
1895         if (!list_empty(&clp->cl_delegations)
1896                         && !atomic_read(&clp->cl_callback.cb_set))
1897                 goto out;
1898         status = nfs_ok;
1899 out:
1900         nfs4_unlock_state();
1901         return status;
1902 }
1903
1904 static void
1905 end_grace(void)
1906 {
1907         dprintk("NFSD: end of grace period\n");
1908         nfsd4_recdir_purge_old();
1909         in_grace = 0;
1910 }
1911
1912 static time_t
1913 nfs4_laundromat(void)
1914 {
1915         struct nfs4_client *clp;
1916         struct nfs4_stateowner *sop;
1917         struct nfs4_delegation *dp;
1918         struct list_head *pos, *next, reaplist;
1919         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1920         time_t t, clientid_val = NFSD_LEASE_TIME;
1921         time_t u, test_val = NFSD_LEASE_TIME;
1922
1923         nfs4_lock_state();
1924
1925         dprintk("NFSD: laundromat service - starting\n");
1926         if (in_grace)
1927                 end_grace();
1928         list_for_each_safe(pos, next, &client_lru) {
1929                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1930                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1931                         t = clp->cl_time - cutoff;
1932                         if (clientid_val > t)
1933                                 clientid_val = t;
1934                         break;
1935                 }
1936                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1937                         clp->cl_clientid.cl_id);
1938                 nfsd4_remove_clid_dir(clp);
1939                 expire_client(clp);
1940         }
1941         INIT_LIST_HEAD(&reaplist);
1942         spin_lock(&recall_lock);
1943         list_for_each_safe(pos, next, &del_recall_lru) {
1944                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1945                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1946                         u = dp->dl_time - cutoff;
1947                         if (test_val > u)
1948                                 test_val = u;
1949                         break;
1950                 }
1951                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1952                                     dp, dp->dl_flock);
1953                 list_move(&dp->dl_recall_lru, &reaplist);
1954         }
1955         spin_unlock(&recall_lock);
1956         list_for_each_safe(pos, next, &reaplist) {
1957                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1958                 list_del_init(&dp->dl_recall_lru);
1959                 unhash_delegation(dp);
1960         }
1961         test_val = NFSD_LEASE_TIME;
1962         list_for_each_safe(pos, next, &close_lru) {
1963                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1964                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1965                         u = sop->so_time - cutoff;
1966                         if (test_val > u)
1967                                 test_val = u;
1968                         break;
1969                 }
1970                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1971                         sop->so_id);
1972                 list_del(&sop->so_close_lru);
1973                 nfs4_put_stateowner(sop);
1974         }
1975         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1976                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1977         nfs4_unlock_state();
1978         return clientid_val;
1979 }
1980
1981 void
1982 laundromat_main(void *not_used)
1983 {
1984         time_t t;
1985
1986         t = nfs4_laundromat();
1987         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1988         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1989 }
1990
1991 static struct nfs4_stateowner *
1992 search_close_lru(u32 st_id, int flags)
1993 {
1994         struct nfs4_stateowner *local = NULL;
1995
1996         if (flags & CLOSE_STATE) {
1997                 list_for_each_entry(local, &close_lru, so_close_lru) {
1998                         if (local->so_id == st_id)
1999                                 return local;
2000                 }
2001         }
2002         return NULL;
2003 }
2004
2005 static inline int
2006 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2007 {
2008         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_dentry->d_inode;
2009 }
2010
2011 static int
2012 STALE_STATEID(stateid_t *stateid)
2013 {
2014         if (stateid->si_boot == boot_time)
2015                 return 0;
2016         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2017                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2018                 stateid->si_generation);
2019         return 1;
2020 }
2021
2022 static inline int
2023 access_permit_read(unsigned long access_bmap)
2024 {
2025         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2026                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2027                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2028 }
2029
2030 static inline int
2031 access_permit_write(unsigned long access_bmap)
2032 {
2033         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2034                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2035 }
2036
2037 static
2038 int nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2039 {
2040         int status = nfserr_openmode;
2041
2042         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2043                 goto out;
2044         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2045                 goto out;
2046         status = nfs_ok;
2047 out:
2048         return status;
2049 }
2050
2051 static inline int
2052 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2053 {
2054         /* Trying to call delegreturn with a special stateid? Yuch: */
2055         if (!(flags & (RD_STATE | WR_STATE)))
2056                 return nfserr_bad_stateid;
2057         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
2058                 return nfs_ok;
2059         else if (nfs4_in_grace()) {
2060                 /* Answer in remaining cases depends on existance of
2061                  * conflicting state; so we must wait out the grace period. */
2062                 return nfserr_grace;
2063         } else if (flags & WR_STATE)
2064                 return nfs4_share_conflict(current_fh,
2065                                 NFS4_SHARE_DENY_WRITE);
2066         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2067                 return nfs4_share_conflict(current_fh,
2068                                 NFS4_SHARE_DENY_READ);
2069 }
2070
2071 /*
2072  * Allow READ/WRITE during grace period on recovered state only for files
2073  * that are not able to provide mandatory locking.
2074  */
2075 static inline int
2076 io_during_grace_disallowed(struct inode *inode, int flags)
2077 {
2078         return nfs4_in_grace() && (flags & (RD_STATE | WR_STATE))
2079                 && MANDATORY_LOCK(inode);
2080 }
2081
2082 /*
2083 * Checks for stateid operations
2084 */
2085 int
2086 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2087 {
2088         struct nfs4_stateid *stp = NULL;
2089         struct nfs4_delegation *dp = NULL;
2090         stateid_t *stidp;
2091         struct inode *ino = current_fh->fh_dentry->d_inode;
2092         int status;
2093
2094         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2095                 stateid->si_boot, stateid->si_stateownerid,
2096                 stateid->si_fileid, stateid->si_generation);
2097         if (filpp)
2098                 *filpp = NULL;
2099
2100         if (io_during_grace_disallowed(ino, flags))
2101                 return nfserr_grace;
2102
2103         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2104                 return check_special_stateids(current_fh, stateid, flags);
2105
2106         /* STALE STATEID */
2107         status = nfserr_stale_stateid;
2108         if (STALE_STATEID(stateid))
2109                 goto out;
2110
2111         /* BAD STATEID */
2112         status = nfserr_bad_stateid;
2113         if (!stateid->si_fileid) { /* delegation stateid */
2114                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2115                         dprintk("NFSD: delegation stateid not found\n");
2116                         if (nfs4_in_grace())
2117                                 status = nfserr_grace;
2118                         goto out;
2119                 }
2120                 stidp = &dp->dl_stateid;
2121         } else { /* open or lock stateid */
2122                 if (!(stp = find_stateid(stateid, flags))) {
2123                         dprintk("NFSD: open or lock stateid not found\n");
2124                         if (nfs4_in_grace())
2125                                 status = nfserr_grace;
2126                         goto out;
2127                 }
2128                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2129                         goto out;
2130                 if (!stp->st_stateowner->so_confirmed)
2131                         goto out;
2132                 stidp = &stp->st_stateid;
2133         }
2134         if (stateid->si_generation > stidp->si_generation)
2135                 goto out;
2136
2137         /* OLD STATEID */
2138         status = nfserr_old_stateid;
2139         if (stateid->si_generation < stidp->si_generation)
2140                 goto out;
2141         if (stp) {
2142                 if ((status = nfs4_check_openmode(stp,flags)))
2143                         goto out;
2144                 renew_client(stp->st_stateowner->so_client);
2145                 if (filpp)
2146                         *filpp = stp->st_vfs_file;
2147         } else if (dp) {
2148                 if ((status = nfs4_check_delegmode(dp, flags)))
2149                         goto out;
2150                 renew_client(dp->dl_client);
2151                 if (flags & DELEG_RET)
2152                         unhash_delegation(dp);
2153                 if (filpp)
2154                         *filpp = dp->dl_vfs_file;
2155         }
2156         status = nfs_ok;
2157 out:
2158         return status;
2159 }
2160
2161 static inline int
2162 setlkflg (int type)
2163 {
2164         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2165                 RD_STATE : WR_STATE;
2166 }
2167
2168 /*
2169  * Checks for sequence id mutating operations.
2170  */
2171 static int
2172 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)
2173 {
2174         struct nfs4_stateid *stp;
2175         struct nfs4_stateowner *sop;
2176
2177         dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2178                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2179                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2180                 stateid->si_generation);
2181
2182         *stpp = NULL;
2183         *sopp = NULL;
2184
2185         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2186                 printk("NFSD: preprocess_seqid_op: magic stateid!\n");
2187                 return nfserr_bad_stateid;
2188         }
2189
2190         if (STALE_STATEID(stateid))
2191                 return nfserr_stale_stateid;
2192         /*
2193         * We return BAD_STATEID if filehandle doesn't match stateid,
2194         * the confirmed flag is incorrecly set, or the generation
2195         * number is incorrect.
2196         */
2197         stp = find_stateid(stateid, flags);
2198         if (stp == NULL) {
2199                 /*
2200                  * Also, we should make sure this isn't just the result of
2201                  * a replayed close:
2202                  */
2203                 sop = search_close_lru(stateid->si_stateownerid, flags);
2204                 if (sop == NULL)
2205                         return nfserr_bad_stateid;
2206                 *sopp = sop;
2207                 goto check_replay;
2208         }
2209
2210         if (lock) {
2211                 struct nfs4_stateowner *sop = stp->st_stateowner;
2212                 clientid_t *lockclid = &lock->v.new.clientid;
2213                 struct nfs4_client *clp = sop->so_client;
2214                 int lkflg = 0;
2215                 int status;
2216
2217                 lkflg = setlkflg(lock->lk_type);
2218
2219                 if (lock->lk_is_new) {
2220                        if (!sop->so_is_open_owner)
2221                                return nfserr_bad_stateid;
2222                        if (!cmp_clid(&clp->cl_clientid, lockclid))
2223                                return nfserr_bad_stateid;
2224                        /* stp is the open stateid */
2225                        status = nfs4_check_openmode(stp, lkflg);
2226                        if (status)
2227                                return status;
2228                } else {
2229                        /* stp is the lock stateid */
2230                        status = nfs4_check_openmode(stp->st_openstp, lkflg);
2231                        if (status)
2232                                return status;
2233                }
2234
2235         }
2236
2237         if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp)) {
2238                 printk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2239                 return nfserr_bad_stateid;
2240         }
2241
2242         *stpp = stp;
2243         *sopp = sop = stp->st_stateowner;
2244
2245         /*
2246         *  We now validate the seqid and stateid generation numbers.
2247         *  For the moment, we ignore the possibility of
2248         *  generation number wraparound.
2249         */
2250         if (seqid != sop->so_seqid)
2251                 goto check_replay;
2252
2253         if (sop->so_confirmed && flags & CONFIRM) {
2254                 printk("NFSD: preprocess_seqid_op: expected"
2255                                 " unconfirmed stateowner!\n");
2256                 return nfserr_bad_stateid;
2257         }
2258         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2259                 printk("NFSD: preprocess_seqid_op: stateowner not"
2260                                 " confirmed yet!\n");
2261                 return nfserr_bad_stateid;
2262         }
2263         if (stateid->si_generation > stp->st_stateid.si_generation) {
2264                 printk("NFSD: preprocess_seqid_op: future stateid?!\n");
2265                 return nfserr_bad_stateid;
2266         }
2267
2268         if (stateid->si_generation < stp->st_stateid.si_generation) {
2269                 printk("NFSD: preprocess_seqid_op: old stateid!\n");
2270                 return nfserr_old_stateid;
2271         }
2272         renew_client(sop->so_client);
2273         return nfs_ok;
2274
2275 check_replay:
2276         if (seqid == sop->so_seqid - 1) {
2277                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2278                 /* indicate replay to calling function */
2279                 return NFSERR_REPLAY_ME;
2280         }
2281         printk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2282                         sop->so_seqid, seqid);
2283         *sopp = NULL;
2284         return nfserr_bad_seqid;
2285 }
2286
2287 int
2288 nfsd4_open_confirm(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open_confirm *oc, struct nfs4_stateowner **replay_owner)
2289 {
2290         int status;
2291         struct nfs4_stateowner *sop;
2292         struct nfs4_stateid *stp;
2293
2294         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2295                         (int)current_fh->fh_dentry->d_name.len,
2296                         current_fh->fh_dentry->d_name.name);
2297
2298         if ((status = fh_verify(rqstp, current_fh, S_IFREG, 0)))
2299                 goto out;
2300
2301         nfs4_lock_state();
2302
2303         if ((status = nfs4_preprocess_seqid_op(current_fh, oc->oc_seqid,
2304                                         &oc->oc_req_stateid,
2305                                         CHECK_FH | CONFIRM | OPEN_STATE,
2306                                         &oc->oc_stateowner, &stp, NULL)))
2307                 goto out;
2308
2309         sop = oc->oc_stateowner;
2310         sop->so_confirmed = 1;
2311         update_stateid(&stp->st_stateid);
2312         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2313         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
2314                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2315                          stp->st_stateid.si_boot,
2316                          stp->st_stateid.si_stateownerid,
2317                          stp->st_stateid.si_fileid,
2318                          stp->st_stateid.si_generation);
2319
2320         nfsd4_create_clid_dir(sop->so_client);
2321 out:
2322         if (oc->oc_stateowner) {
2323                 nfs4_get_stateowner(oc->oc_stateowner);
2324                 *replay_owner = oc->oc_stateowner;
2325         }
2326         nfs4_unlock_state();
2327         return status;
2328 }
2329
2330
2331 /*
2332  * unset all bits in union bitmap (bmap) that
2333  * do not exist in share (from successful OPEN_DOWNGRADE)
2334  */
2335 static void
2336 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2337 {
2338         int i;
2339         for (i = 1; i < 4; i++) {
2340                 if ((i & access) != i)
2341                         __clear_bit(i, bmap);
2342         }
2343 }
2344
2345 static void
2346 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2347 {
2348         int i;
2349         for (i = 0; i < 4; i++) {
2350                 if ((i & deny) != i)
2351                         __clear_bit(i, bmap);
2352         }
2353 }
2354
2355 int
2356 nfsd4_open_downgrade(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open_downgrade *od, struct nfs4_stateowner **replay_owner)
2357 {
2358         int status;
2359         struct nfs4_stateid *stp;
2360         unsigned int share_access;
2361
2362         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
2363                         (int)current_fh->fh_dentry->d_name.len,
2364                         current_fh->fh_dentry->d_name.name);
2365
2366         if (!TEST_ACCESS(od->od_share_access) || !TEST_DENY(od->od_share_deny))
2367                 return nfserr_inval;
2368
2369         nfs4_lock_state();
2370         if ((status = nfs4_preprocess_seqid_op(current_fh, od->od_seqid,
2371                                         &od->od_stateid,
2372                                         CHECK_FH | OPEN_STATE,
2373                                         &od->od_stateowner, &stp, NULL)))
2374                 goto out;
2375
2376         status = nfserr_inval;
2377         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2378                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2379                         stp->st_access_bmap, od->od_share_access);
2380                 goto out;
2381         }
2382         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2383                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2384                         stp->st_deny_bmap, od->od_share_deny);
2385                 goto out;
2386         }
2387         set_access(&share_access, stp->st_access_bmap);
2388         nfs4_file_downgrade(stp->st_vfs_file,
2389                             share_access & ~od->od_share_access);
2390
2391         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2392         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2393
2394         update_stateid(&stp->st_stateid);
2395         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2396         status = nfs_ok;
2397 out:
2398         if (od->od_stateowner) {
2399                 nfs4_get_stateowner(od->od_stateowner);
2400                 *replay_owner = od->od_stateowner;
2401         }
2402         nfs4_unlock_state();
2403         return status;
2404 }
2405
2406 /*
2407  * nfs4_unlock_state() called after encode
2408  */
2409 int
2410 nfsd4_close(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_close *close, struct nfs4_stateowner **replay_owner)
2411 {
2412         int status;
2413         struct nfs4_stateid *stp;
2414
2415         dprintk("NFSD: nfsd4_close on file %.*s\n",
2416                         (int)current_fh->fh_dentry->d_name.len,
2417                         current_fh->fh_dentry->d_name.name);
2418
2419         nfs4_lock_state();
2420         /* check close_lru for replay */
2421         if ((status = nfs4_preprocess_seqid_op(current_fh, close->cl_seqid,
2422                                         &close->cl_stateid,
2423                                         CHECK_FH | OPEN_STATE | CLOSE_STATE,
2424                                         &close->cl_stateowner, &stp, NULL)))
2425                 goto out;
2426         /*
2427         *  Return success, but first update the stateid.
2428         */
2429         status = nfs_ok;
2430         update_stateid(&stp->st_stateid);
2431         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2432
2433         /* release_state_owner() calls nfsd_close() if needed */
2434         release_state_owner(stp, OPEN_STATE);
2435 out:
2436         if (close->cl_stateowner) {
2437                 nfs4_get_stateowner(close->cl_stateowner);
2438                 *replay_owner = close->cl_stateowner;
2439         }
2440         nfs4_unlock_state();
2441         return status;
2442 }
2443
2444 int
2445 nfsd4_delegreturn(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_delegreturn *dr)
2446 {
2447         int status;
2448
2449         if ((status = fh_verify(rqstp, current_fh, S_IFREG, 0)))
2450                 goto out;
2451
2452         nfs4_lock_state();
2453         status = nfs4_preprocess_stateid_op(current_fh, &dr->dr_stateid, DELEG_RET, NULL);
2454         nfs4_unlock_state();
2455 out:
2456         return status;
2457 }
2458
2459
2460 /*
2461  * Lock owner state (byte-range locks)
2462  */
2463 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2464 #define LOCK_HASH_BITS              8
2465 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2466 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2467
2468 #define lockownerid_hashval(id) \
2469         ((id) & LOCK_HASH_MASK)
2470
2471 static inline unsigned int
2472 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2473                 struct xdr_netobj *ownername)
2474 {
2475         return (file_hashval(inode) + cl_id
2476                         + opaque_hashval(ownername->data, ownername->len))
2477                 & LOCK_HASH_MASK;
2478 }
2479
2480 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2481 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2482 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2483
2484 static struct nfs4_stateid *
2485 find_stateid(stateid_t *stid, int flags)
2486 {
2487         struct nfs4_stateid *local = NULL;
2488         u32 st_id = stid->si_stateownerid;
2489         u32 f_id = stid->si_fileid;
2490         unsigned int hashval;
2491
2492         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2493         if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2494                 hashval = stateid_hashval(st_id, f_id);
2495                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2496                         if ((local->st_stateid.si_stateownerid == st_id) &&
2497                             (local->st_stateid.si_fileid == f_id))
2498                                 return local;
2499                 }
2500         }
2501         if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2502                 hashval = stateid_hashval(st_id, f_id);
2503                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2504                         if ((local->st_stateid.si_stateownerid == st_id) &&
2505                             (local->st_stateid.si_fileid == f_id))
2506                                 return local;
2507                 }
2508         }
2509         return NULL;
2510 }
2511
2512 static struct nfs4_delegation *
2513 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2514 {
2515         struct nfs4_file *fp;
2516         struct nfs4_delegation *dl;
2517
2518         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2519                     stid->si_boot, stid->si_stateownerid,
2520                     stid->si_fileid, stid->si_generation);
2521
2522         fp = find_file(ino);
2523         if (!fp)
2524                 return NULL;
2525         dl = find_delegation_file(fp, stid);
2526         put_nfs4_file(fp);
2527         return dl;
2528 }
2529
2530 /*
2531  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2532  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2533  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2534  * locking, this prevents us from being completely protocol-compliant.  The
2535  * real solution to this problem is to start using unsigned file offsets in
2536  * the VFS, but this is a very deep change!
2537  */
2538 static inline void
2539 nfs4_transform_lock_offset(struct file_lock *lock)
2540 {
2541         if (lock->fl_start < 0)
2542                 lock->fl_start = OFFSET_MAX;
2543         if (lock->fl_end < 0)
2544                 lock->fl_end = OFFSET_MAX;
2545 }
2546
2547 static int
2548 nfs4_verify_lock_stateowner(struct nfs4_stateowner *sop, unsigned int hashval)
2549 {
2550         struct nfs4_stateowner *local = NULL;
2551         int status = 0;
2552
2553         if (hashval >= LOCK_HASH_SIZE)
2554                 goto out;
2555         list_for_each_entry(local, &lock_ownerid_hashtbl[hashval], so_idhash) {
2556                 if (local == sop) {
2557                         status = 1;
2558                         goto out;
2559                 }
2560         }
2561 out:
2562         return status;
2563 }
2564
2565
2566 static inline void
2567 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2568 {
2569         struct nfs4_stateowner *sop = (struct nfs4_stateowner *) fl->fl_owner;
2570         unsigned int hval = lockownerid_hashval(sop->so_id);
2571
2572         deny->ld_sop = NULL;
2573         if (nfs4_verify_lock_stateowner(sop, hval)) {
2574                 kref_get(&sop->so_ref);
2575                 deny->ld_sop = sop;
2576                 deny->ld_clientid = sop->so_client->cl_clientid;
2577         }
2578         deny->ld_start = fl->fl_start;
2579         deny->ld_length = ~(u64)0;
2580         if (fl->fl_end != ~(u64)0)
2581                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
2582         deny->ld_type = NFS4_READ_LT;
2583         if (fl->fl_type != F_RDLCK)
2584                 deny->ld_type = NFS4_WRITE_LT;
2585 }
2586
2587 static struct nfs4_stateowner *
2588 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2589                 struct xdr_netobj *owner)
2590 {
2591         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2592         struct nfs4_stateowner *op;
2593
2594         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2595                 if (cmp_owner_str(op, owner, clid))
2596                         return op;
2597         }
2598         return NULL;
2599 }
2600
2601 /*
2602  * Alloc a lock owner structure.
2603  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
2604  * occured.
2605  *
2606  * strhashval = lock_ownerstr_hashval
2607  */
2608
2609 static struct nfs4_stateowner *
2610 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2611         struct nfs4_stateowner *sop;
2612         struct nfs4_replay *rp;
2613         unsigned int idhashval;
2614
2615         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2616                 return NULL;
2617         idhashval = lockownerid_hashval(current_ownerid);
2618         INIT_LIST_HEAD(&sop->so_idhash);
2619         INIT_LIST_HEAD(&sop->so_strhash);
2620         INIT_LIST_HEAD(&sop->so_perclient);
2621         INIT_LIST_HEAD(&sop->so_stateids);
2622         INIT_LIST_HEAD(&sop->so_perstateid);
2623         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2624         sop->so_time = 0;
2625         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2626         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2627         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2628         sop->so_is_open_owner = 0;
2629         sop->so_id = current_ownerid++;
2630         sop->so_client = clp;
2631         /* It is the openowner seqid that will be incremented in encode in the
2632          * case of new lockowners; so increment the lock seqid manually: */
2633         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2634         sop->so_confirmed = 1;
2635         rp = &sop->so_replay;
2636         rp->rp_status = NFSERR_SERVERFAULT;
2637         rp->rp_buflen = 0;
2638         rp->rp_buf = rp->rp_ibuf;
2639         return sop;
2640 }
2641
2642 static struct nfs4_stateid *
2643 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2644 {
2645         struct nfs4_stateid *stp;
2646         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2647
2648         stp = nfs4_alloc_stateid();
2649         if (stp == NULL)
2650                 goto out;
2651         INIT_LIST_HEAD(&stp->st_hash);
2652         INIT_LIST_HEAD(&stp->st_perfile);
2653         INIT_LIST_HEAD(&stp->st_perstateowner);
2654         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2655         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2656         list_add(&stp->st_perfile, &fp->fi_stateids);
2657         list_add(&stp->st_perstateowner, &sop->so_stateids);
2658         stp->st_stateowner = sop;
2659         get_nfs4_file(fp);
2660         stp->st_file = fp;
2661         stp->st_stateid.si_boot = boot_time;
2662         stp->st_stateid.si_stateownerid = sop->so_id;
2663         stp->st_stateid.si_fileid = fp->fi_id;
2664         stp->st_stateid.si_generation = 0;
2665         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2666         stp->st_access_bmap = open_stp->st_access_bmap;
2667         stp->st_deny_bmap = open_stp->st_deny_bmap;
2668         stp->st_openstp = open_stp;
2669
2670 out:
2671         return stp;
2672 }
2673
2674 static int
2675 check_lock_length(u64 offset, u64 length)
2676 {
2677         return ((length == 0)  || ((length != ~(u64)0) &&
2678              LOFF_OVERFLOW(offset, length)));
2679 }
2680
2681 /*
2682  *  LOCK operation
2683  */
2684 int
2685 nfsd4_lock(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_lock *lock, struct nfs4_stateowner **replay_owner)
2686 {
2687         struct nfs4_stateowner *open_sop = NULL;
2688         struct nfs4_stateowner *lock_sop = NULL;
2689         struct nfs4_stateid *lock_stp;
2690         struct file *filp;
2691         struct file_lock file_lock;
2692         struct file_lock *conflock;
2693         int status = 0;
2694         unsigned int strhashval;
2695
2696         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2697                 (long long) lock->lk_offset,
2698                 (long long) lock->lk_length);
2699
2700         if (check_lock_length(lock->lk_offset, lock->lk_length))
2701                  return nfserr_inval;
2702
2703         nfs4_lock_state();
2704
2705         if (lock->lk_is_new) {
2706                 /*
2707                  * Client indicates that this is a new lockowner.
2708                  * Use open owner and open stateid to create lock owner and
2709                  * lock stateid.
2710                  */
2711                 struct nfs4_stateid *open_stp = NULL;
2712                 struct nfs4_file *fp;
2713
2714                 status = nfserr_stale_clientid;
2715                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2716                         goto out;
2717
2718                 /* validate and update open stateid and open seqid */
2719                 status = nfs4_preprocess_seqid_op(current_fh,
2720                                         lock->lk_new_open_seqid,
2721                                         &lock->lk_new_open_stateid,
2722                                         CHECK_FH | OPEN_STATE,
2723                                         &lock->lk_stateowner, &open_stp,
2724                                         lock);
2725                 if (status)
2726                         goto out;
2727                 open_sop = lock->lk_stateowner;
2728                 /* create lockowner and lock stateid */
2729                 fp = open_stp->st_file;
2730                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
2731                                 open_sop->so_client->cl_clientid.cl_id,
2732                                 &lock->v.new.owner);
2733                 /* XXX: Do we need to check for duplicate stateowners on
2734                  * the same file, or should they just be allowed (and
2735                  * create new stateids)? */
2736                 status = nfserr_resource;
2737                 lock_sop = alloc_init_lock_stateowner(strhashval,
2738                                 open_sop->so_client, open_stp, lock);
2739                 if (lock_sop == NULL)
2740                         goto out;
2741                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2742                 if (lock_stp == NULL) {
2743                         release_stateowner(lock_sop);
2744                         goto out;
2745                 }
2746         } else {
2747                 /* lock (lock owner + lock stateid) already exists */
2748                 status = nfs4_preprocess_seqid_op(current_fh,
2749                                        lock->lk_old_lock_seqid,
2750                                        &lock->lk_old_lock_stateid,
2751                                        CHECK_FH | LOCK_STATE,
2752                                        &lock->lk_stateowner, &lock_stp, lock);
2753                 if (status)
2754                         goto out;
2755                 lock_sop = lock->lk_stateowner;
2756         }
2757         /* lock->lk_stateowner and lock_stp have been created or found */
2758         filp = lock_stp->st_vfs_file;
2759
2760         if ((status = fh_verify(rqstp, current_fh, S_IFREG, MAY_LOCK))) {
2761                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2762                 goto out;
2763         }
2764
2765         status = nfserr_grace;
2766         if (nfs4_in_grace() && !lock->lk_reclaim)
2767                 goto out;
2768         status = nfserr_no_grace;
2769         if (!nfs4_in_grace() && lock->lk_reclaim)
2770                 goto out;
2771
2772         locks_init_lock(&file_lock);
2773         switch (lock->lk_type) {
2774                 case NFS4_READ_LT:
2775                 case NFS4_READW_LT:
2776                         file_lock.fl_type = F_RDLCK;
2777                 break;
2778                 case NFS4_WRITE_LT:
2779                 case NFS4_WRITEW_LT:
2780                         file_lock.fl_type = F_WRLCK;
2781                 break;
2782                 default:
2783                         status = nfserr_inval;
2784                 goto out;
2785         }
2786         file_lock.fl_owner = (fl_owner_t)lock_sop;
2787         file_lock.fl_pid = current->tgid;
2788         file_lock.fl_file = filp;
2789         file_lock.fl_flags = FL_POSIX;
2790
2791         file_lock.fl_start = lock->lk_offset;
2792         if ((lock->lk_length == ~(u64)0) ||
2793                         LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2794                 file_lock.fl_end = ~(u64)0;
2795         else
2796                 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2797         nfs4_transform_lock_offset(&file_lock);
2798
2799         /*
2800         * Try to lock the file in the VFS.
2801         * Note: locks.c uses the BKL to protect the inode's lock list.
2802         */
2803
2804         status = posix_lock_file(filp, &file_lock);
2805         if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
2806                 file_lock.fl_ops->fl_release_private(&file_lock);
2807         dprintk("NFSD: nfsd4_lock: posix_lock_file status %d\n",status);
2808         switch (-status) {
2809         case 0: /* success! */
2810                 update_stateid(&lock_stp->st_stateid);
2811                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
2812                                 sizeof(stateid_t));
2813                 goto out;
2814         case (EAGAIN):
2815                 goto conflicting_lock;
2816         case (EDEADLK):
2817                 status = nfserr_deadlock;
2818         default:
2819                 dprintk("NFSD: nfsd4_lock: posix_lock_file() failed! status %d\n",status);
2820                 goto out_destroy_new_stateid;
2821         }
2822
2823 conflicting_lock:
2824         dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2825         status = nfserr_denied;
2826         /* XXX There is a race here. Future patch needed to provide
2827          * an atomic posix_lock_and_test_file
2828          */
2829         if (!(conflock = posix_test_lock(filp, &file_lock))) {
2830                 status = nfserr_serverfault;
2831                 goto out;
2832         }
2833         nfs4_set_lock_denied(conflock, &lock->lk_denied);
2834
2835 out_destroy_new_stateid:
2836         if (lock->lk_is_new) {
2837                 dprintk("NFSD: nfsd4_lock: destroy new stateid!\n");
2838                 /*
2839                  * An error encountered after instantiation of the new
2840                  * stateid has forced us to destroy it.
2841                  */
2842                 release_state_owner(lock_stp, LOCK_STATE);
2843         }
2844 out:
2845         if (lock->lk_stateowner) {
2846                 nfs4_get_stateowner(lock->lk_stateowner);
2847                 *replay_owner = lock->lk_stateowner;
2848         }
2849         nfs4_unlock_state();
2850         return status;
2851 }
2852
2853 /*
2854  * LOCKT operation
2855  */
2856 int
2857 nfsd4_lockt(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_lockt *lockt)
2858 {
2859         struct inode *inode;
2860         struct file file;
2861         struct file_lock file_lock;
2862         struct file_lock *conflicting_lock;
2863         int status;
2864
2865         if (nfs4_in_grace())
2866                 return nfserr_grace;
2867
2868         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2869                  return nfserr_inval;
2870
2871         lockt->lt_stateowner = NULL;
2872         nfs4_lock_state();
2873
2874         status = nfserr_stale_clientid;
2875         if (STALE_CLIENTID(&lockt->lt_clientid))
2876                 goto out;
2877
2878         if ((status = fh_verify(rqstp, current_fh, S_IFREG, 0))) {
2879                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2880                 if (status == nfserr_symlink)
2881                         status = nfserr_inval;
2882                 goto out;
2883         }
2884
2885         inode = current_fh->fh_dentry->d_inode;
2886         locks_init_lock(&file_lock);
2887         switch (lockt->lt_type) {
2888                 case NFS4_READ_LT:
2889                 case NFS4_READW_LT:
2890                         file_lock.fl_type = F_RDLCK;
2891                 break;
2892                 case NFS4_WRITE_LT:
2893                 case NFS4_WRITEW_LT:
2894                         file_lock.fl_type = F_WRLCK;
2895                 break;
2896                 default:
2897                         printk("NFSD: nfs4_lockt: bad lock type!\n");
2898                         status = nfserr_inval;
2899                 goto out;
2900         }
2901
2902         lockt->lt_stateowner = find_lockstateowner_str(inode,
2903                         &lockt->lt_clientid, &lockt->lt_owner);
2904         if (lockt->lt_stateowner)
2905                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2906         file_lock.fl_pid = current->tgid;
2907         file_lock.fl_flags = FL_POSIX;
2908
2909         file_lock.fl_start = lockt->lt_offset;
2910         if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2911                 file_lock.fl_end = ~(u64)0;
2912         else
2913                 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2914
2915         nfs4_transform_lock_offset(&file_lock);
2916
2917         /* posix_test_lock uses the struct file _only_ to resolve the inode.
2918          * since LOCKT doesn't require an OPEN, and therefore a struct
2919          * file may not exist, pass posix_test_lock a struct file with
2920          * only the dentry:inode set.
2921          */
2922         memset(&file, 0, sizeof (struct file));
2923         file.f_dentry = current_fh->fh_dentry;
2924
2925         status = nfs_ok;
2926         conflicting_lock = posix_test_lock(&file, &file_lock);
2927         if (conflicting_lock) {
2928                 status = nfserr_denied;
2929                 nfs4_set_lock_denied(conflicting_lock, &lockt->lt_denied);
2930         }
2931 out:
2932         nfs4_unlock_state();
2933         return status;
2934 }
2935
2936 int
2937 nfsd4_locku(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_locku *locku, struct nfs4_stateowner **replay_owner)
2938 {
2939         struct nfs4_stateid *stp;
2940         struct file *filp = NULL;
2941         struct file_lock file_lock;
2942         int status;
2943
2944         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2945                 (long long) locku->lu_offset,
2946                 (long long) locku->lu_length);
2947
2948         if (check_lock_length(locku->lu_offset, locku->lu_length))
2949                  return nfserr_inval;
2950
2951         nfs4_lock_state();
2952
2953         if ((status = nfs4_preprocess_seqid_op(current_fh,
2954                                         locku->lu_seqid,
2955                                         &locku->lu_stateid,
2956                                         CHECK_FH | LOCK_STATE,
2957                                         &locku->lu_stateowner, &stp, NULL)))
2958                 goto out;
2959
2960         filp = stp->st_vfs_file;
2961         BUG_ON(!filp);
2962         locks_init_lock(&file_lock);
2963         file_lock.fl_type = F_UNLCK;
2964         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2965         file_lock.fl_pid = current->tgid;
2966         file_lock.fl_file = filp;
2967         file_lock.fl_flags = FL_POSIX;
2968         file_lock.fl_start = locku->lu_offset;
2969
2970         if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2971                 file_lock.fl_end = ~(u64)0;
2972         else
2973                 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2974         nfs4_transform_lock_offset(&file_lock);
2975
2976         /*
2977         *  Try to unlock the file in the VFS.
2978         */
2979         status = posix_lock_file(filp, &file_lock);
2980         if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
2981                 file_lock.fl_ops->fl_release_private(&file_lock);
2982         if (status) {
2983                 dprintk("NFSD: nfs4_locku: posix_lock_file failed!\n");
2984                 goto out_nfserr;
2985         }
2986         /*
2987         * OK, unlock succeeded; the only thing left to do is update the stateid.
2988         */
2989         update_stateid(&stp->st_stateid);
2990         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2991
2992 out:
2993         if (locku->lu_stateowner) {
2994                 nfs4_get_stateowner(locku->lu_stateowner);
2995                 *replay_owner = locku->lu_stateowner;
2996         }
2997         nfs4_unlock_state();
2998         return status;
2999
3000 out_nfserr:
3001         status = nfserrno(status);
3002         goto out;
3003 }
3004
3005 /*
3006  * returns
3007  *      1: locks held by lockowner
3008  *      0: no locks held by lockowner
3009  */
3010 static int
3011 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3012 {
3013         struct file_lock **flpp;
3014         struct inode *inode = filp->f_dentry->d_inode;
3015         int status = 0;
3016
3017         lock_kernel();
3018         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3019                 if ((*flpp)->fl_owner == (fl_owner_t)lowner)
3020                         status = 1;
3021                         goto out;
3022         }
3023 out:
3024         unlock_kernel();
3025         return status;
3026 }
3027
3028 int
3029 nfsd4_release_lockowner(struct svc_rqst *rqstp, struct nfsd4_release_lockowner *rlockowner)
3030 {
3031         clientid_t *clid = &rlockowner->rl_clientid;
3032         struct nfs4_stateowner *sop;
3033         struct nfs4_stateid *stp;
3034         struct xdr_netobj *owner = &rlockowner->rl_owner;
3035         struct list_head matches;
3036         int i;
3037         int status;
3038
3039         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3040                 clid->cl_boot, clid->cl_id);
3041
3042         /* XXX check for lease expiration */
3043
3044         status = nfserr_stale_clientid;
3045         if (STALE_CLIENTID(clid))
3046                 return status;
3047
3048         nfs4_lock_state();
3049
3050         status = nfserr_locks_held;
3051         /* XXX: we're doing a linear search through all the lockowners.
3052          * Yipes!  For now we'll just hope clients aren't really using
3053          * release_lockowner much, but eventually we have to fix these
3054          * data structures. */
3055         INIT_LIST_HEAD(&matches);
3056         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3057                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3058                         if (!cmp_owner_str(sop, owner, clid))
3059                                 continue;
3060                         list_for_each_entry(stp, &sop->so_stateids,
3061                                         st_perstateowner) {
3062                                 if (check_for_locks(stp->st_vfs_file, sop))
3063                                         goto out;
3064                                 /* Note: so_perclient unused for lockowners,
3065                                  * so it's OK to fool with here. */
3066                                 list_add(&sop->so_perclient, &matches);
3067                         }
3068                 }
3069         }
3070         /* Clients probably won't expect us to return with some (but not all)
3071          * of the lockowner state released; so don't release any until all
3072          * have been checked. */
3073         status = nfs_ok;
3074         while (!list_empty(&matches)) {
3075                 sop = list_entry(matches.next, struct nfs4_stateowner,
3076                                                                 so_perclient);
3077                 /* unhash_stateowner deletes so_perclient only
3078                  * for openowners. */
3079                 list_del(&sop->so_perclient);
3080                 release_stateowner(sop);
3081         }
3082 out:
3083         nfs4_unlock_state();
3084         return status;
3085 }
3086
3087 static inline struct nfs4_client_reclaim *
3088 alloc_reclaim(void)
3089 {
3090         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3091 }
3092
3093 int
3094 nfs4_has_reclaimed_state(const char *name)
3095 {
3096         unsigned int strhashval = clientstr_hashval(name);
3097         struct nfs4_client *clp;
3098
3099         clp = find_confirmed_client_by_str(name, strhashval);
3100         return clp ? 1 : 0;
3101 }
3102
3103 /*
3104  * failure => all reset bets are off, nfserr_no_grace...
3105  */
3106 int
3107 nfs4_client_to_reclaim(const char *name)
3108 {
3109         unsigned int strhashval;
3110         struct nfs4_client_reclaim *crp = NULL;
3111
3112         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3113         crp = alloc_reclaim();
3114         if (!crp)
3115                 return 0;
3116         strhashval = clientstr_hashval(name);
3117         INIT_LIST_HEAD(&crp->cr_strhash);
3118         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3119         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3120         reclaim_str_hashtbl_size++;
3121         return 1;
3122 }
3123
3124 static void
3125 nfs4_release_reclaim(void)
3126 {
3127         struct nfs4_client_reclaim *crp = NULL;
3128         int i;
3129
3130         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3131                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3132                         crp = list_entry(reclaim_str_hashtbl[i].next,
3133                                         struct nfs4_client_reclaim, cr_strhash);
3134                         list_del(&crp->cr_strhash);
3135                         kfree(crp);
3136                         reclaim_str_hashtbl_size--;
3137                 }
3138         }
3139         BUG_ON(reclaim_str_hashtbl_size);
3140 }
3141
3142 /*
3143  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3144 static struct nfs4_client_reclaim *
3145 nfs4_find_reclaim_client(clientid_t *clid)
3146 {
3147         unsigned int strhashval;
3148         struct nfs4_client *clp;
3149         struct nfs4_client_reclaim *crp = NULL;
3150
3151
3152         /* find clientid in conf_id_hashtbl */
3153         clp = find_confirmed_client(clid);
3154         if (clp == NULL)
3155                 return NULL;
3156
3157         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3158                             clp->cl_name.len, clp->cl_name.data,
3159                             clp->cl_recdir);
3160
3161         /* find clp->cl_name in reclaim_str_hashtbl */
3162         strhashval = clientstr_hashval(clp->cl_recdir);
3163         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3164                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3165                         return crp;
3166                 }
3167         }
3168         return NULL;
3169 }
3170
3171 /*
3172 * Called from OPEN. Look for clientid in reclaim list.
3173 */
3174 int
3175 nfs4_check_open_reclaim(clientid_t *clid)
3176 {
3177         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3178 }
3179
3180 /* initialization to perform at module load time: */
3181
3182 void
3183 nfs4_state_init(void)
3184 {
3185         int i;
3186
3187         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3188                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3189                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3190                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3191                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3192         }
3193         for (i = 0; i < FILE_HASH_SIZE; i++) {
3194                 INIT_LIST_HEAD(&file_hashtbl[i]);
3195         }
3196         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3197                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3198                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3199         }
3200         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3201                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3202                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3203         }
3204         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3205                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3206                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3207         }
3208         memset(&onestateid, ~0, sizeof(stateid_t));
3209         INIT_LIST_HEAD(&close_lru);
3210         INIT_LIST_HEAD(&client_lru);
3211         INIT_LIST_HEAD(&del_recall_lru);
3212         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3213                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3214         reclaim_str_hashtbl_size = 0;
3215 }
3216
3217 static void
3218 nfsd4_load_reboot_recovery_data(void)
3219 {
3220         int status;
3221
3222         nfs4_lock_state();
3223         nfsd4_init_recdir(user_recovery_dirname);
3224         status = nfsd4_recdir_load();
3225         nfs4_unlock_state();
3226         if (status)
3227                 printk("NFSD: Failure reading reboot recovery data\n");
3228 }
3229
3230 /* initialization to perform when the nfsd service is started: */
3231
3232 static void
3233 __nfs4_state_start(void)
3234 {
3235         time_t grace_time;
3236
3237         boot_time = get_seconds();
3238         grace_time = max(user_lease_time, lease_time);
3239         lease_time = user_lease_time;
3240         in_grace = 1;
3241         printk("NFSD: starting %ld-second grace period\n", grace_time);
3242         laundry_wq = create_singlethread_workqueue("nfsd4");
3243         queue_delayed_work(laundry_wq, &laundromat_work, grace_time*HZ);
3244 }
3245
3246 int
3247 nfs4_state_start(void)
3248 {
3249         int status;
3250
3251         if (nfs4_init)
3252                 return 0;
3253         status = nfsd4_init_slabs();
3254         if (status)
3255                 return status;
3256         nfsd4_load_reboot_recovery_data();
3257         __nfs4_state_start();
3258         nfs4_init = 1;
3259         return 0;
3260 }
3261
3262 int
3263 nfs4_in_grace(void)
3264 {
3265         return in_grace;
3266 }
3267
3268 time_t
3269 nfs4_lease_time(void)
3270 {
3271         return lease_time;
3272 }
3273
3274 static void
3275 __nfs4_state_shutdown(void)
3276 {
3277         int i;
3278         struct nfs4_client *clp = NULL;
3279         struct nfs4_delegation *dp = NULL;
3280         struct nfs4_stateowner *sop = NULL;
3281         struct list_head *pos, *next, reaplist;
3282
3283         list_for_each_safe(pos, next, &close_lru) {
3284                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
3285                 list_del(&sop->so_close_lru);
3286                 nfs4_put_stateowner(sop);
3287         }
3288
3289         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3290                 while (!list_empty(&conf_id_hashtbl[i])) {
3291                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3292                         expire_client(clp);
3293                 }
3294                 while (!list_empty(&unconf_str_hashtbl[i])) {
3295                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3296                         expire_client(clp);
3297                 }
3298         }
3299         INIT_LIST_HEAD(&reaplist);
3300         spin_lock(&recall_lock);
3301         list_for_each_safe(pos, next, &del_recall_lru) {
3302                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3303                 list_move(&dp->dl_recall_lru, &reaplist);
3304         }
3305         spin_unlock(&recall_lock);
3306         list_for_each_safe(pos, next, &reaplist) {
3307                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3308                 list_del_init(&dp->dl_recall_lru);
3309                 unhash_delegation(dp);
3310         }
3311
3312         cancel_delayed_work(&laundromat_work);
3313         flush_workqueue(laundry_wq);
3314         destroy_workqueue(laundry_wq);
3315         nfsd4_shutdown_recdir();
3316         nfs4_init = 0;
3317 }
3318
3319 void
3320 nfs4_state_shutdown(void)
3321 {
3322         nfs4_lock_state();
3323         nfs4_release_reclaim();
3324         __nfs4_state_shutdown();
3325         nfsd4_free_slabs();
3326         nfs4_unlock_state();
3327 }
3328
3329 static void
3330 nfs4_set_recdir(char *recdir)
3331 {
3332         nfs4_lock_state();
3333         strcpy(user_recovery_dirname, recdir);
3334         nfs4_unlock_state();
3335 }
3336
3337 /*
3338  * Change the NFSv4 recovery directory to recdir.
3339  */
3340 int
3341 nfs4_reset_recoverydir(char *recdir)
3342 {
3343         int status;
3344         struct nameidata nd;
3345
3346         status = path_lookup(recdir, LOOKUP_FOLLOW, &nd);
3347         if (status)
3348                 return status;
3349         status = -ENOTDIR;
3350         if (S_ISDIR(nd.dentry->d_inode->i_mode)) {
3351                 nfs4_set_recdir(recdir);
3352                 status = 0;
3353         }
3354         path_release(&nd);
3355         return status;
3356 }
3357
3358 /*
3359  * Called when leasetime is changed.
3360  *
3361  * The only way the protocol gives us to handle on-the-fly lease changes is to
3362  * simulate a reboot.  Instead of doing that, we just wait till the next time
3363  * we start to register any changes in lease time.  If the administrator
3364  * really wants to change the lease time *now*, they can go ahead and bring
3365  * nfsd down and then back up again after changing the lease time.
3366  */
3367 void
3368 nfs4_reset_lease(time_t leasetime)
3369 {
3370         lock_kernel();
3371         user_lease_time = leasetime;
3372         unlock_kernel();
3373 }