fs/nfsd/nfs4state.c

   1 /*
   2 *  linux/fs/nfsd/nfs4state.c
   3 *
   4 *  Copyright (c) 2001 The Regents of the University of Michigan.
   5 *  All rights reserved.
   6 *
   7 *  Kendrick Smith <kmsmith@umich.edu>
   8 *  Andy Adamson <kandros@umich.edu>
   9 *
  10 *  Redistribution and use in source and binary forms, with or without
  11 *  modification, are permitted provided that the following conditions
  12 *  are met:
  13 *
  14 *  1. Redistributions of source code must retain the above copyright
  15 *     notice, this list of conditions and the following disclaimer.
  16 *  2. Redistributions in binary form must reproduce the above copyright
  17 *     notice, this list of conditions and the following disclaimer in the
  18 *     documentation and/or other materials provided with the distribution.
  19 *  3. Neither the name of the University nor the names of its
  20 *     contributors may be used to endorse or promote products derived
  21 *     from this software without specific prior written permission.
  22 *
  23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  34 *
  35 */
  36
  37 #include <linux/param.h>
  38 #include <linux/major.h>
  39 #include <linux/slab.h>
  40
  41 #include <linux/sunrpc/svc.h>
  42 #include <linux/nfsd/nfsd.h>
  43 #include <linux/nfsd/cache.h>
  44 #include <linux/file.h>
  45 #include <linux/mount.h>
  46 #include <linux/workqueue.h>
  47 #include <linux/smp_lock.h>
  48 #include <linux/kthread.h>
  49 #include <linux/nfs4.h>
  50 #include <linux/nfsd/state.h>
  51 #include <linux/nfsd/xdr4.h>
  52 #include <linux/namei.h>
  53 #include <linux/swap.h>
  54 #include <linux/mutex.h>
  55 #include <linux/lockd/bind.h>
  56 #include <linux/module.h>
  57 #include <linux/sunrpc/svcauth_gss.h>
  58
  59 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  60
  61 /* Globals */
  62 static time_t lease_time = 90;     /* default lease time */
  63 static time_t user_lease_time = 90;
  64 static time_t boot_time;
  65 static u32 current_ownerid = 1;
  66 static u32 current_fileid = 1;
  67 static u32 current_delegid = 1;
  68 static u32 nfs4_init;
  69 static stateid_t zerostateid;             /* bits all 0 */
  70 static stateid_t onestateid;              /* bits all 1 */
  71 static u64 current_sessionid = 1;
  72
  73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
  74 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
  75
  76 /* forward declarations */
  77 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
  78 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
  79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  80 static void nfs4_set_recdir(char *recdir);
  81
  82 /* Locking: */
  83
  84 /* Currently used for almost all code touching nfsv4 state: */
  85 static DEFINE_MUTEX(client_mutex);
  86
  87 /*
  88  * Currently used for the del_recall_lru and file hash table.  In an
  89  * effort to decrease the scope of the client_mutex, this spinlock may
  90  * eventually cover more:
  91  */
  92 static DEFINE_SPINLOCK(recall_lock);
  93
  94 static struct kmem_cache *stateowner_slab = NULL;
  95 static struct kmem_cache *file_slab = NULL;
  96 static struct kmem_cache *stateid_slab = NULL;
  97 static struct kmem_cache *deleg_slab = NULL;
  98
  99 void
 100 nfs4_lock_state(void)
 101 {
 102         mutex_lock(&client_mutex);
 103 }
 104
 105 void
 106 nfs4_unlock_state(void)
 107 {
 108         mutex_unlock(&client_mutex);
 109 }
 110
 111 static inline u32
 112 opaque_hashval(const void *ptr, int nbytes)
 113 {
 114         unsigned char *cptr = (unsigned char *) ptr;
 115
 116         u32 x = 0;
 117         while (nbytes--) {
 118                 x *= 37;
 119                 x += *cptr++;
 120         }
 121         return x;
 122 }
 123
 124 static struct list_head del_recall_lru;
 125
 126 static inline void
 127 put_nfs4_file(struct nfs4_file *fi)
 128 {
 129         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 130                 list_del(&fi->fi_hash);
 131                 spin_unlock(&recall_lock);
 132                 iput(fi->fi_inode);
 133                 kmem_cache_free(file_slab, fi);
 134         }
 135 }
 136
 137 static inline void
 138 get_nfs4_file(struct nfs4_file *fi)
 139 {
 140         atomic_inc(&fi->fi_ref);
 141 }
 142
 143 static int num_delegations;
 144 unsigned int max_delegations;
 145
 146 /*
 147  * Open owner state (share locks)
 148  */
 149
 150 /* hash tables for nfs4_stateowner */
 151 #define OWNER_HASH_BITS              8
 152 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 153 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 154
 155 #define ownerid_hashval(id) \
 156         ((id) & OWNER_HASH_MASK)
 157 #define ownerstr_hashval(clientid, ownername) \
 158         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
 159
 160 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
 161 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 162
 163 /* hash table for nfs4_file */
 164 #define FILE_HASH_BITS                   8
 165 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 166 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
 167 /* hash table for (open)nfs4_stateid */
 168 #define STATEID_HASH_BITS              10
 169 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
 170 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
 171
 172 #define file_hashval(x) \
 173         hash_ptr(x, FILE_HASH_BITS)
 174 #define stateid_hashval(owner_id, file_id)  \
 175         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
 176
 177 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 178 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 179
 180 static struct nfs4_delegation *
 181 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 182 {
 183         struct nfs4_delegation *dp;
 184         struct nfs4_file *fp = stp->st_file;
 185         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
 186
 187         dprintk("NFSD alloc_init_deleg\n");
 188         if (fp->fi_had_conflict)
 189                 return NULL;
 190         if (num_delegations > max_delegations)
 191                 return NULL;
 192         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
 193         if (dp == NULL)
 194                 return dp;
 195         num_delegations++;
 196         INIT_LIST_HEAD(&dp->dl_perfile);
 197         INIT_LIST_HEAD(&dp->dl_perclnt);
 198         INIT_LIST_HEAD(&dp->dl_recall_lru);
 199         dp->dl_client = clp;
 200         get_nfs4_file(fp);
 201         dp->dl_file = fp;
 202         dp->dl_flock = NULL;
 203         get_file(stp->st_vfs_file);
 204         dp->dl_vfs_file = stp->st_vfs_file;
 205         dp->dl_type = type;
 206         dp->dl_recall.cbr_dp = NULL;
 207         dp->dl_recall.cbr_ident = cb->cb_ident;
 208         dp->dl_recall.cbr_trunc = 0;
 209         dp->dl_stateid.si_boot = boot_time;
 210         dp->dl_stateid.si_stateownerid = current_delegid++;
 211         dp->dl_stateid.si_fileid = 0;
 212         dp->dl_stateid.si_generation = 0;
 213         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 214         dp->dl_time = 0;
 215         atomic_set(&dp->dl_count, 1);
 216         list_add(&dp->dl_perfile, &fp->fi_delegations);
 217         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 218         return dp;
 219 }
 220
 221 void
 222 nfs4_put_delegation(struct nfs4_delegation *dp)
 223 {
 224         if (atomic_dec_and_test(&dp->dl_count)) {
 225                 dprintk("NFSD: freeing dp %p\n",dp);
 226                 put_nfs4_file(dp->dl_file);
 227                 kmem_cache_free(deleg_slab, dp);
 228                 num_delegations--;
 229         }
 230 }
 231
 232 /* Remove the associated file_lock first, then remove the delegation.
 233  * lease_modify() is called to remove the FS_LEASE file_lock from
 234  * the i_flock list, eventually calling nfsd's lock_manager
 235  * fl_release_callback.
 236  */
 237 static void
 238 nfs4_close_delegation(struct nfs4_delegation *dp)
 239 {
 240         struct file *filp = dp->dl_vfs_file;
 241
 242         dprintk("NFSD: close_delegation dp %p\n",dp);
 243         dp->dl_vfs_file = NULL;
 244         /* The following nfsd_close may not actually close the file,
 245          * but we want to remove the lease in any case. */
 246         if (dp->dl_flock)
 247                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
 248         nfsd_close(filp);
 249 }
 250
 251 /* Called under the state lock. */
 252 static void
 253 unhash_delegation(struct nfs4_delegation *dp)
 254 {
 255         list_del_init(&dp->dl_perfile);
 256         list_del_init(&dp->dl_perclnt);
 257         spin_lock(&recall_lock);
 258         list_del_init(&dp->dl_recall_lru);
 259         spin_unlock(&recall_lock);
 260         nfs4_close_delegation(dp);
 261         nfs4_put_delegation(dp);
 262 }
 263
 264 /*
 265  * SETCLIENTID state
 266  */
 267
 268 /* Hash tables for nfs4_clientid state */
 269 #define CLIENT_HASH_BITS                 4
 270 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 271 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 272
 273 #define clientid_hashval(id) \
 274         ((id) & CLIENT_HASH_MASK)
 275 #define clientstr_hashval(name) \
 276         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
 277 /*
 278  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 279  * used in reboot/reset lease grace period processing
 280  *
 281  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 282  * setclientid_confirmed info.
 283  *
 284  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 285  * setclientid info.
 286  *
 287  * client_lru holds client queue ordered by nfs4_client.cl_time
 288  * for lease renewal.
 289  *
 290  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 291  * for last close replay.
 292  */
 293 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 294 static int reclaim_str_hashtbl_size = 0;
 295 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 296 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 297 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 298 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 299 static struct list_head client_lru;
 300 static struct list_head close_lru;
 301
 302 static void unhash_generic_stateid(struct nfs4_stateid *stp)
 303 {
 304         list_del(&stp->st_hash);
 305         list_del(&stp->st_perfile);
 306         list_del(&stp->st_perstateowner);
 307 }
 308
 309 static void free_generic_stateid(struct nfs4_stateid *stp)
 310 {
 311         put_nfs4_file(stp->st_file);
 312         kmem_cache_free(stateid_slab, stp);
 313 }
 314
 315 static void release_lock_stateid(struct nfs4_stateid *stp)
 316 {
 317         unhash_generic_stateid(stp);
 318         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
 319         free_generic_stateid(stp);
 320 }
 321
 322 static void unhash_lockowner(struct nfs4_stateowner *sop)
 323 {
 324         struct nfs4_stateid *stp;
 325
 326         list_del(&sop->so_idhash);
 327         list_del(&sop->so_strhash);
 328         list_del(&sop->so_perstateid);
 329         while (!list_empty(&sop->so_stateids)) {
 330                 stp = list_first_entry(&sop->so_stateids,
 331                                 struct nfs4_stateid, st_perstateowner);
 332                 release_lock_stateid(stp);
 333         }
 334 }
 335
 336 static void release_lockowner(struct nfs4_stateowner *sop)
 337 {
 338         unhash_lockowner(sop);
 339         nfs4_put_stateowner(sop);
 340 }
 341
 342 static void
 343 release_stateid_lockowners(struct nfs4_stateid *open_stp)
 344 {
 345         struct nfs4_stateowner *lock_sop;
 346
 347         while (!list_empty(&open_stp->st_lockowners)) {
 348                 lock_sop = list_entry(open_stp->st_lockowners.next,
 349                                 struct nfs4_stateowner, so_perstateid);
 350                 /* list_del(&open_stp->st_lockowners);  */
 351                 BUG_ON(lock_sop->so_is_open_owner);
 352                 release_lockowner(lock_sop);
 353         }
 354 }
 355
 356 static void release_open_stateid(struct nfs4_stateid *stp)
 357 {
 358         unhash_generic_stateid(stp);
 359         release_stateid_lockowners(stp);
 360         nfsd_close(stp->st_vfs_file);
 361         free_generic_stateid(stp);
 362 }
 363
 364 static void unhash_openowner(struct nfs4_stateowner *sop)
 365 {
 366         struct nfs4_stateid *stp;
 367
 368         list_del(&sop->so_idhash);
 369         list_del(&sop->so_strhash);
 370         list_del(&sop->so_perclient);
 371         list_del(&sop->so_perstateid); /* XXX: necessary? */
 372         while (!list_empty(&sop->so_stateids)) {
 373                 stp = list_first_entry(&sop->so_stateids,
 374                                 struct nfs4_stateid, st_perstateowner);
 375                 release_open_stateid(stp);
 376         }
 377 }
 378
 379 static void release_openowner(struct nfs4_stateowner *sop)
 380 {
 381         unhash_openowner(sop);
 382         list_del(&sop->so_close_lru);
 383         nfs4_put_stateowner(sop);
 384 }
 385
 386 static DEFINE_SPINLOCK(sessionid_lock);
 387 #define SESSION_HASH_SIZE       512
 388 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
 389
 390 static inline int
 391 hash_sessionid(struct nfs4_sessionid *sessionid)
 392 {
 393         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 394
 395         return sid->sequence % SESSION_HASH_SIZE;
 396 }
 397
 398 static inline void
 399 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 400 {
 401         u32 *ptr = (u32 *)(&sessionid->data[0]);
 402         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 403 }
 404
 405 static void
 406 gen_sessionid(struct nfsd4_session *ses)
 407 {
 408         struct nfs4_client *clp = ses->se_client;
 409         struct nfsd4_sessionid *sid;
 410
 411         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 412         sid->clientid = clp->cl_clientid;
 413         sid->sequence = current_sessionid++;
 414         sid->reserved = 0;
 415 }
 416
 417 /*
 418  * Give the client the number of slots it requests bound by
 419  * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
 420  *
 421  * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
 422  * should (up to a point) re-negotiate active sessions and reduce their
 423  * slot usage to make rooom for new connections. For now we just fail the
 424  * create session.
 425  */
 426 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
 427 {
 428         int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
 429
 430         spin_lock(&nfsd_serv->sv_lock);
 431         if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
 432                 np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
 433         nfsd_serv->sv_drc_pages_used += np;
 434         spin_unlock(&nfsd_serv->sv_lock);
 435
 436         if (np <= 0) {
 437                 status = nfserr_resource;
 438                 fchan->maxreqs = 0;
 439         } else
 440                 fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
 441
 442         return status;
 443 }
 444
 445 /*
 446  * fchan holds the client values on input, and the server values on output
 447  */
 448 static int init_forechannel_attrs(struct svc_rqst *rqstp,
 449                                     struct nfsd4_session *session,
 450                                     struct nfsd4_channel_attrs *fchan)
 451 {
 452         int status = 0;
 453         __u32   maxcount = svc_max_payload(rqstp);
 454
 455         /* headerpadsz set to zero in encode routine */
 456
 457         /* Use the client's max request and max response size if possible */
 458         if (fchan->maxreq_sz > maxcount)
 459                 fchan->maxreq_sz = maxcount;
 460         session->se_fmaxreq_sz = fchan->maxreq_sz;
 461
 462         if (fchan->maxresp_sz > maxcount)
 463                 fchan->maxresp_sz = maxcount;
 464         session->se_fmaxresp_sz = fchan->maxresp_sz;
 465
 466         /* Set the max response cached size our default which is
 467          * a multiple of PAGE_SIZE and small */
 468         session->se_fmaxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
 469         fchan->maxresp_cached = session->se_fmaxresp_cached;
 470
 471         /* Use the client's maxops if possible */
 472         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
 473                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
 474         session->se_fmaxops = fchan->maxops;
 475
 476         /* try to use the client requested number of slots */
 477         if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
 478                 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
 479
 480         /* FIXME: Error means no more DRC pages so the server should
 481          * recover pages from existing sessions. For now fail session
 482          * creation.
 483          */
 484         status = set_forechannel_maxreqs(fchan);
 485
 486         session->se_fnumslots = fchan->maxreqs;
 487         return status;
 488 }
 489
 490 static int
 491 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
 492                    struct nfsd4_create_session *cses)
 493 {
 494         struct nfsd4_session *new, tmp;
 495         int idx, status = nfserr_resource, slotsize;
 496
 497         memset(&tmp, 0, sizeof(tmp));
 498
 499         /* FIXME: For now, we just accept the client back channel attributes. */
 500         status = init_forechannel_attrs(rqstp, &tmp, &cses->fore_channel);
 501         if (status)
 502                 goto out;
 503
 504         /* allocate struct nfsd4_session and slot table in one piece */
 505         slotsize = tmp.se_fnumslots * sizeof(struct nfsd4_slot);
 506         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
 507         if (!new)
 508                 goto out;
 509
 510         memcpy(new, &tmp, sizeof(*new));
 511
 512         new->se_client = clp;
 513         gen_sessionid(new);
 514         idx = hash_sessionid(&new->se_sessionid);
 515         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
 516                NFS4_MAX_SESSIONID_LEN);
 517
 518         new->se_flags = cses->flags;
 519         kref_init(&new->se_ref);
 520         spin_lock(&sessionid_lock);
 521         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
 522         list_add(&new->se_perclnt, &clp->cl_sessions);
 523         spin_unlock(&sessionid_lock);
 524
 525         status = nfs_ok;
 526 out:
 527         return status;
 528 }
 529
 530 /* caller must hold sessionid_lock */
 531 static struct nfsd4_session *
 532 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
 533 {
 534         struct nfsd4_session *elem;
 535         int idx;
 536
 537         dump_sessionid(__func__, sessionid);
 538         idx = hash_sessionid(sessionid);
 539         dprintk("%s: idx is %d\n", __func__, idx);
 540         /* Search in the appropriate list */
 541         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
 542                 dump_sessionid("list traversal", &elem->se_sessionid);
 543                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
 544                             NFS4_MAX_SESSIONID_LEN)) {
 545                         return elem;
 546                 }
 547         }
 548
 549         dprintk("%s: session not found\n", __func__);
 550         return NULL;
 551 }
 552
 553 /* caller must hold sessionid_lock */
 554 static void
 555 unhash_session(struct nfsd4_session *ses)
 556 {
 557         list_del(&ses->se_hash);
 558         list_del(&ses->se_perclnt);
 559 }
 560
 561 static void
 562 release_session(struct nfsd4_session *ses)
 563 {
 564         spin_lock(&sessionid_lock);
 565         unhash_session(ses);
 566         spin_unlock(&sessionid_lock);
 567         nfsd4_put_session(ses);
 568 }
 569
 570 static void nfsd4_release_respages(struct page **respages, short resused);
 571
 572 void
 573 free_session(struct kref *kref)
 574 {
 575         struct nfsd4_session *ses;
 576         int i;
 577
 578         ses = container_of(kref, struct nfsd4_session, se_ref);
 579         for (i = 0; i < ses->se_fnumslots; i++) {
 580                 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
 581                 nfsd4_release_respages(e->ce_respages, e->ce_resused);
 582         }
 583         kfree(ses->se_slots);
 584         kfree(ses);
 585 }
 586
 587 static inline void
 588 renew_client(struct nfs4_client *clp)
 589 {
 590         /*
 591         * Move client to the end to the LRU list.
 592         */
 593         dprintk("renewing client (clientid %08x/%08x)\n",
 594                         clp->cl_clientid.cl_boot,
 595                         clp->cl_clientid.cl_id);
 596         list_move_tail(&clp->cl_lru, &client_lru);
 597         clp->cl_time = get_seconds();
 598 }
 599
 600 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 601 static int
 602 STALE_CLIENTID(clientid_t *clid)
 603 {
 604         if (clid->cl_boot == boot_time)
 605                 return 0;
 606         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
 607                 clid->cl_boot, clid->cl_id, boot_time);
 608         return 1;
 609 }
 610
 611 /*
 612  * XXX Should we use a slab cache ?
 613  * This type of memory management is somewhat inefficient, but we use it
 614  * anyway since SETCLIENTID is not a common operation.
 615  */
 616 static struct nfs4_client *alloc_client(struct xdr_netobj name)
 617 {
 618         struct nfs4_client *clp;
 619
 620         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
 621         if (clp == NULL)
 622                 return NULL;
 623         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
 624         if (clp->cl_name.data == NULL) {
 625                 kfree(clp);
 626                 return NULL;
 627         }
 628         memcpy(clp->cl_name.data, name.data, name.len);
 629         clp->cl_name.len = name.len;
 630         return clp;
 631 }
 632
 633 static void
 634 shutdown_callback_client(struct nfs4_client *clp)
 635 {
 636         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
 637
 638         if (clnt) {
 639                 /*
 640                  * Callback threads take a reference on the client, so there
 641                  * should be no outstanding callbacks at this point.
 642                  */
 643                 clp->cl_callback.cb_client = NULL;
 644                 rpc_shutdown_client(clnt);
 645         }
 646 }
 647
 648 static inline void
 649 free_client(struct nfs4_client *clp)
 650 {
 651         shutdown_callback_client(clp);
 652         nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
 653                              clp->cl_slot.sl_cache_entry.ce_resused);
 654         if (clp->cl_cred.cr_group_info)
 655                 put_group_info(clp->cl_cred.cr_group_info);
 656         kfree(clp->cl_principal);
 657         kfree(clp->cl_name.data);
 658         kfree(clp);
 659 }
 660
 661 void
 662 put_nfs4_client(struct nfs4_client *clp)
 663 {
 664         if (atomic_dec_and_test(&clp->cl_count))
 665                 free_client(clp);
 666 }
 667
 668 static void
 669 expire_client(struct nfs4_client *clp)
 670 {
 671         struct nfs4_stateowner *sop;
 672         struct nfs4_delegation *dp;
 673         struct list_head reaplist;
 674
 675         dprintk("NFSD: expire_client cl_count %d\n",
 676                             atomic_read(&clp->cl_count));
 677
 678         INIT_LIST_HEAD(&reaplist);
 679         spin_lock(&recall_lock);
 680         while (!list_empty(&clp->cl_delegations)) {
 681                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
 682                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
 683                                 dp->dl_flock);
 684                 list_del_init(&dp->dl_perclnt);
 685                 list_move(&dp->dl_recall_lru, &reaplist);
 686         }
 687         spin_unlock(&recall_lock);
 688         while (!list_empty(&reaplist)) {
 689                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
 690                 list_del_init(&dp->dl_recall_lru);
 691                 unhash_delegation(dp);
 692         }
 693         list_del(&clp->cl_idhash);
 694         list_del(&clp->cl_strhash);
 695         list_del(&clp->cl_lru);
 696         while (!list_empty(&clp->cl_openowners)) {
 697                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
 698                 release_openowner(sop);
 699         }
 700         while (!list_empty(&clp->cl_sessions)) {
 701                 struct nfsd4_session  *ses;
 702                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
 703                                  se_perclnt);
 704                 release_session(ses);
 705         }
 706         put_nfs4_client(clp);
 707 }
 708
 709 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
 710 {
 711         struct nfs4_client *clp;
 712
 713         clp = alloc_client(name);
 714         if (clp == NULL)
 715                 return NULL;
 716         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
 717         atomic_set(&clp->cl_count, 1);
 718         atomic_set(&clp->cl_callback.cb_set, 0);
 719         INIT_LIST_HEAD(&clp->cl_idhash);
 720         INIT_LIST_HEAD(&clp->cl_strhash);
 721         INIT_LIST_HEAD(&clp->cl_openowners);
 722         INIT_LIST_HEAD(&clp->cl_delegations);
 723         INIT_LIST_HEAD(&clp->cl_sessions);
 724         INIT_LIST_HEAD(&clp->cl_lru);
 725         return clp;
 726 }
 727
 728 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
 729 {
 730         memcpy(target->cl_verifier.data, source->data,
 731                         sizeof(target->cl_verifier.data));
 732 }
 733
 734 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
 735 {
 736         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
 737         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
 738 }
 739
 740 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
 741 {
 742         target->cr_uid = source->cr_uid;
 743         target->cr_gid = source->cr_gid;
 744         target->cr_group_info = source->cr_group_info;
 745         get_group_info(target->cr_group_info);
 746 }
 747
 748 static int same_name(const char *n1, const char *n2)
 749 {
 750         return 0 == memcmp(n1, n2, HEXDIR_LEN);
 751 }
 752
 753 static int
 754 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
 755 {
 756         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
 757 }
 758
 759 static int
 760 same_clid(clientid_t *cl1, clientid_t *cl2)
 761 {
 762         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
 763 }
 764
 765 /* XXX what about NGROUP */
 766 static int
 767 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
 768 {
 769         return cr1->cr_uid == cr2->cr_uid;
 770 }
 771
 772 static void gen_clid(struct nfs4_client *clp)
 773 {
 774         static u32 current_clientid = 1;
 775
 776         clp->cl_clientid.cl_boot = boot_time;
 777         clp->cl_clientid.cl_id = current_clientid++;
 778 }
 779
 780 static void gen_confirm(struct nfs4_client *clp)
 781 {
 782         static u32 i;
 783         u32 *p;
 784
 785         p = (u32 *)clp->cl_confirm.data;
 786         *p++ = get_seconds();
 787         *p++ = i++;
 788 }
 789
 790 static int check_name(struct xdr_netobj name)
 791 {
 792         if (name.len == 0)
 793                 return 0;
 794         if (name.len > NFS4_OPAQUE_LIMIT) {
 795                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
 796                 return 0;
 797         }
 798         return 1;
 799 }
 800
 801 static void
 802 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
 803 {
 804         unsigned int idhashval;
 805
 806         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
 807         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 808         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
 809         list_add_tail(&clp->cl_lru, &client_lru);
 810         clp->cl_time = get_seconds();
 811 }
 812
 813 static void
 814 move_to_confirmed(struct nfs4_client *clp)
 815 {
 816         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
 817         unsigned int strhashval;
 818
 819         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
 820         list_del_init(&clp->cl_strhash);
 821         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
 822         strhashval = clientstr_hashval(clp->cl_recdir);
 823         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
 824         renew_client(clp);
 825 }
 826
 827 static struct nfs4_client *
 828 find_confirmed_client(clientid_t *clid)
 829 {
 830         struct nfs4_client *clp;
 831         unsigned int idhashval = clientid_hashval(clid->cl_id);
 832
 833         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
 834                 if (same_clid(&clp->cl_clientid, clid))
 835                         return clp;
 836         }
 837         return NULL;
 838 }
 839
 840 static struct nfs4_client *
 841 find_unconfirmed_client(clientid_t *clid)
 842 {
 843         struct nfs4_client *clp;
 844         unsigned int idhashval = clientid_hashval(clid->cl_id);
 845
 846         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
 847                 if (same_clid(&clp->cl_clientid, clid))
 848                         return clp;
 849         }
 850         return NULL;
 851 }
 852
 853 /*
 854  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
 855  * parameter. Matching is based on the fact the at least one of the
 856  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
 857  *
 858  * FIXME: we need to unify the clientid namespaces for nfsv4.x
 859  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
 860  * and SET_CLIENTID{,_CONFIRM}
 861  */
 862 static inline int
 863 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
 864 {
 865         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
 866         return use_exchange_id == has_exchange_flags;
 867 }
 868
 869 static struct nfs4_client *
 870 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
 871                              bool use_exchange_id)
 872 {
 873         struct nfs4_client *clp;
 874
 875         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
 876                 if (same_name(clp->cl_recdir, dname) &&
 877                     match_clientid_establishment(clp, use_exchange_id))
 878                         return clp;
 879         }
 880         return NULL;
 881 }
 882
 883 static struct nfs4_client *
 884 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
 885                                bool use_exchange_id)
 886 {
 887         struct nfs4_client *clp;
 888
 889         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
 890                 if (same_name(clp->cl_recdir, dname) &&
 891                     match_clientid_establishment(clp, use_exchange_id))
 892                         return clp;
 893         }
 894         return NULL;
 895 }
 896
 897 /* a helper function for parse_callback */
 898 static int
 899 parse_octet(unsigned int *lenp, char **addrp)
 900 {
 901         unsigned int len = *lenp;
 902         char *p = *addrp;
 903         int n = -1;
 904         char c;
 905
 906         for (;;) {
 907                 if (!len)
 908                         break;
 909                 len--;
 910                 c = *p++;
 911                 if (c == '.')
 912                         break;
 913                 if ((c < '0') || (c > '9')) {
 914                         n = -1;
 915                         break;
 916                 }
 917                 if (n < 0)
 918                         n = 0;
 919                 n = (n * 10) + (c - '0');
 920                 if (n > 255) {
 921                         n = -1;
 922                         break;
 923                 }
 924         }
 925         *lenp = len;
 926         *addrp = p;
 927         return n;
 928 }
 929
 930 /* parse and set the setclientid ipv4 callback address */
 931 static int
 932 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
 933 {
 934         int temp = 0;
 935         u32 cbaddr = 0;
 936         u16 cbport = 0;
 937         u32 addrlen = addr_len;
 938         char *addr = addr_val;
 939         int i, shift;
 940
 941         /* ipaddress */
 942         shift = 24;
 943         for(i = 4; i > 0  ; i--) {
 944                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 945                         return 0;
 946                 }
 947                 cbaddr |= (temp << shift);
 948                 if (shift > 0)
 949                 shift -= 8;
 950         }
 951         *cbaddrp = cbaddr;
 952
 953         /* port */
 954         shift = 8;
 955         for(i = 2; i > 0  ; i--) {
 956                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
 957                         return 0;
 958                 }
 959                 cbport |= (temp << shift);
 960                 if (shift > 0)
 961                         shift -= 8;
 962         }
 963         *cbportp = cbport;
 964         return 1;
 965 }
 966
 967 static void
 968 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
 969 {
 970         struct nfs4_callback *cb = &clp->cl_callback;
 971
 972         /* Currently, we only support tcp for the callback channel */
 973         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
 974                 goto out_err;
 975
 976         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
 977                          &cb->cb_addr, &cb->cb_port)))
 978                 goto out_err;
 979         cb->cb_prog = se->se_callback_prog;
 980         cb->cb_ident = se->se_callback_ident;
 981         return;
 982 out_err:
 983         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
 984                 "will not receive delegations\n",
 985                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
 986
 987         return;
 988 }
 989
 990 void
 991 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
 992 {
 993         struct nfsd4_compoundres *resp = rqstp->rq_resp;
 994
 995         resp->cstate.statp = statp;
 996 }
 997
 998 /*
 999  * Dereference the result pages.
1000  */
1001 static void
1002 nfsd4_release_respages(struct page **respages, short resused)
1003 {
1004         int i;
1005
1006         dprintk("--> %s\n", __func__);
1007         for (i = 0; i < resused; i++) {
1008                 if (!respages[i])
1009                         continue;
1010                 put_page(respages[i]);
1011                 respages[i] = NULL;
1012         }
1013 }
1014
1015 static void
1016 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1017 {
1018         int i;
1019
1020         for (i = 0; i < count; i++) {
1021                 topages[i] = frompages[i];
1022                 if (!topages[i])
1023                         continue;
1024                 get_page(topages[i]);
1025         }
1026 }
1027
1028 /*
1029  * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1030  * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1031  * length of the XDR response is less than se_fmaxresp_cached
1032  * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1033  * of the reply (e.g. readdir).
1034  *
1035  * Store the base and length of the rq_req.head[0] page
1036  * of the NFSv4.1 data, just past the rpc header.
1037  */
1038 void
1039 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1040 {
1041         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1042         struct svc_rqst *rqstp = resp->rqstp;
1043         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1044         struct nfsd4_op *op = &args->ops[resp->opcnt];
1045         struct kvec *resv = &rqstp->rq_res.head[0];
1046
1047         dprintk("--> %s entry %p\n", __func__, entry);
1048
1049         /* Don't cache a failed OP_SEQUENCE. */
1050         if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1051                 return;
1052
1053         nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1054         entry->ce_opcnt = resp->opcnt;
1055         entry->ce_status = resp->cstate.status;
1056
1057         /*
1058          * Don't need a page to cache just the sequence operation - the slot
1059          * does this for us!
1060          */
1061
1062         if (nfsd4_not_cached(resp)) {
1063                 entry->ce_resused = 0;
1064                 entry->ce_rpchdrlen = 0;
1065                 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1066                         resp->cstate.slot->sl_cache_entry.ce_cachethis);
1067                 return;
1068         }
1069         entry->ce_resused = rqstp->rq_resused;
1070         if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1071                 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1072         nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1073                          entry->ce_resused);
1074         entry->ce_datav.iov_base = resp->cstate.statp;
1075         entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1076                                 (char *)page_address(rqstp->rq_respages[0]));
1077         /* Current request rpc header length*/
1078         entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1079                                 (char *)page_address(rqstp->rq_respages[0]);
1080 }
1081
1082 /*
1083  * We keep the rpc header, but take the nfs reply from the replycache.
1084  */
1085 static int
1086 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1087                         struct nfsd4_cache_entry *entry)
1088 {
1089         struct svc_rqst *rqstp = resp->rqstp;
1090         struct kvec *resv = &resp->rqstp->rq_res.head[0];
1091         int len;
1092
1093         /* Current request rpc header length*/
1094         len = (char *)resp->cstate.statp -
1095                         (char *)page_address(rqstp->rq_respages[0]);
1096         if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1097                 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1098                         entry->ce_datav.iov_len);
1099                 return 0;
1100         }
1101         /* copy the cached reply nfsd data past the current rpc header */
1102         memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1103                 entry->ce_datav.iov_len);
1104         resv->iov_len = len + entry->ce_datav.iov_len;
1105         return 1;
1106 }
1107
1108 /*
1109  * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1110  * cached page.  Replace any futher replay pages from the cache.
1111  */
1112 __be32
1113 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1114                          struct nfsd4_sequence *seq)
1115 {
1116         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1117         __be32 status;
1118
1119         dprintk("--> %s entry %p\n", __func__, entry);
1120
1121         /*
1122          * If this is just the sequence operation, we did not keep
1123          * a page in the cache entry because we can just use the
1124          * slot info stored in struct nfsd4_sequence that was checked
1125          * against the slot in nfsd4_sequence().
1126          *
1127          * This occurs when seq->cachethis is FALSE, or when the client
1128          * session inactivity timer fires and a solo sequence operation
1129          * is sent (lease renewal).
1130          */
1131         if (seq && nfsd4_not_cached(resp)) {
1132                 seq->maxslots = resp->cstate.session->se_fnumslots;
1133                 return nfs_ok;
1134         }
1135
1136         if (!nfsd41_copy_replay_data(resp, entry)) {
1137                 /*
1138                  * Not enough room to use the replay rpc header, send the
1139                  * cached header. Release all the allocated result pages.
1140                  */
1141                 svc_free_res_pages(resp->rqstp);
1142                 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1143                         entry->ce_resused);
1144         } else {
1145                 /* Release all but the first allocated result page */
1146
1147                 resp->rqstp->rq_resused--;
1148                 svc_free_res_pages(resp->rqstp);
1149
1150                 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1151                                  &entry->ce_respages[1],
1152                                  entry->ce_resused - 1);
1153         }
1154
1155         resp->rqstp->rq_resused = entry->ce_resused;
1156         resp->opcnt = entry->ce_opcnt;
1157         resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1158         status = entry->ce_status;
1159
1160         return status;
1161 }
1162
1163 /*
1164  * Set the exchange_id flags returned by the server.
1165  */
1166 static void
1167 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1168 {
1169         /* pNFS is not supported */
1170         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1171
1172         /* Referrals are supported, Migration is not. */
1173         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1174
1175         /* set the wire flags to return to client. */
1176         clid->flags = new->cl_exchange_flags;
1177 }
1178
1179 __be32
1180 nfsd4_exchange_id(struct svc_rqst *rqstp,
1181                   struct nfsd4_compound_state *cstate,
1182                   struct nfsd4_exchange_id *exid)
1183 {
1184         struct nfs4_client *unconf, *conf, *new;
1185         int status;
1186         unsigned int            strhashval;
1187         char                    dname[HEXDIR_LEN];
1188         nfs4_verifier           verf = exid->verifier;
1189         u32                     ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1190
1191         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1192                 " ip_addr=%u flags %x, spa_how %d\n",
1193                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1194                 ip_addr, exid->flags, exid->spa_how);
1195
1196         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1197                 return nfserr_inval;
1198
1199         /* Currently only support SP4_NONE */
1200         switch (exid->spa_how) {
1201         case SP4_NONE:
1202                 break;
1203         case SP4_SSV:
1204                 return nfserr_encr_alg_unsupp;
1205         default:
1206                 BUG();                          /* checked by xdr code */
1207         case SP4_MACH_CRED:
1208                 return nfserr_serverfault;      /* no excuse :-/ */
1209         }
1210
1211         status = nfs4_make_rec_clidname(dname, &exid->clname);
1212
1213         if (status)
1214                 goto error;
1215
1216         strhashval = clientstr_hashval(dname);
1217
1218         nfs4_lock_state();
1219         status = nfs_ok;
1220
1221         conf = find_confirmed_client_by_str(dname, strhashval, true);
1222         if (conf) {
1223                 if (!same_verf(&verf, &conf->cl_verifier)) {
1224                         /* 18.35.4 case 8 */
1225                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1226                                 status = nfserr_not_same;
1227                                 goto out;
1228                         }
1229                         /* Client reboot: destroy old state */
1230                         expire_client(conf);
1231                         goto out_new;
1232                 }
1233                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1234                         /* 18.35.4 case 9 */
1235                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1236                                 status = nfserr_perm;
1237                                 goto out;
1238                         }
1239                         expire_client(conf);
1240                         goto out_new;
1241                 }
1242                 if (ip_addr != conf->cl_addr &&
1243                     !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
1244                         /* Client collision. 18.35.4 case 3 */
1245                         status = nfserr_clid_inuse;
1246                         goto out;
1247                 }
1248                 /*
1249                  * Set bit when the owner id and verifier map to an already
1250                  * confirmed client id (18.35.3).
1251                  */
1252                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1253
1254                 /*
1255                  * Falling into 18.35.4 case 2, possible router replay.
1256                  * Leave confirmed record intact and return same result.
1257                  */
1258                 copy_verf(conf, &verf);
1259                 new = conf;
1260                 goto out_copy;
1261         } else {
1262                 /* 18.35.4 case 7 */
1263                 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1264                         status = nfserr_noent;
1265                         goto out;
1266                 }
1267         }
1268
1269         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1270         if (unconf) {
1271                 /*
1272                  * Possible retry or client restart.  Per 18.35.4 case 4,
1273                  * a new unconfirmed record should be generated regardless
1274                  * of whether any properties have changed.
1275                  */
1276                 expire_client(unconf);
1277         }
1278
1279 out_new:
1280         /* Normal case */
1281         new = create_client(exid->clname, dname);
1282         if (new == NULL) {
1283                 status = nfserr_resource;
1284                 goto out;
1285         }
1286
1287         copy_verf(new, &verf);
1288         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1289         new->cl_addr = ip_addr;
1290         gen_clid(new);
1291         gen_confirm(new);
1292         add_to_unconfirmed(new, strhashval);
1293 out_copy:
1294         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1295         exid->clientid.cl_id = new->cl_clientid.cl_id;
1296
1297         new->cl_slot.sl_seqid = 0;
1298         exid->seqid = 1;
1299         nfsd4_set_ex_flags(new, exid);
1300
1301         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1302                 new->cl_slot.sl_seqid, new->cl_exchange_flags);
1303         status = nfs_ok;
1304
1305 out:
1306         nfs4_unlock_state();
1307 error:
1308         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1309         return status;
1310 }
1311
1312 static int
1313 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1314 {
1315         dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1316                 slot->sl_seqid);
1317
1318         /* The slot is in use, and no response has been sent. */
1319         if (slot->sl_inuse) {
1320                 if (seqid == slot->sl_seqid)
1321                         return nfserr_jukebox;
1322                 else
1323                         return nfserr_seq_misordered;
1324         }
1325         /* Normal */
1326         if (likely(seqid == slot->sl_seqid + 1))
1327                 return nfs_ok;
1328         /* Replay */
1329         if (seqid == slot->sl_seqid)
1330                 return nfserr_replay_cache;
1331         /* Wraparound */
1332         if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1333                 return nfs_ok;
1334         /* Misordered replay or misordered new request */
1335         return nfserr_seq_misordered;
1336 }
1337
1338 __be32
1339 nfsd4_create_session(struct svc_rqst *rqstp,
1340                      struct nfsd4_compound_state *cstate,
1341                      struct nfsd4_create_session *cr_ses)
1342 {
1343         u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1344         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1345         struct nfs4_client *conf, *unconf;
1346         struct nfsd4_slot *slot = NULL;
1347         int status = 0;
1348
1349         nfs4_lock_state();
1350         unconf = find_unconfirmed_client(&cr_ses->clientid);
1351         conf = find_confirmed_client(&cr_ses->clientid);
1352
1353         if (conf) {
1354                 slot = &conf->cl_slot;
1355                 status = check_slot_seqid(cr_ses->seqid, slot);
1356                 if (status == nfserr_replay_cache) {
1357                         dprintk("Got a create_session replay! seqid= %d\n",
1358                                 slot->sl_seqid);
1359                         cstate->slot = slot;
1360                         cstate->status = status;
1361                         /* Return the cached reply status */
1362                         status = nfsd4_replay_cache_entry(resp, NULL);
1363                         goto out;
1364                 } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
1365                         status = nfserr_seq_misordered;
1366                         dprintk("Sequence misordered!\n");
1367                         dprintk("Expected seqid= %d but got seqid= %d\n",
1368                                 slot->sl_seqid, cr_ses->seqid);
1369                         goto out;
1370                 }
1371                 conf->cl_slot.sl_seqid++;
1372         } else if (unconf) {
1373                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1374                     (ip_addr != unconf->cl_addr)) {
1375                         status = nfserr_clid_inuse;
1376                         goto out;
1377                 }
1378
1379                 slot = &unconf->cl_slot;
1380                 status = check_slot_seqid(cr_ses->seqid, slot);
1381                 if (status) {
1382                         /* an unconfirmed replay returns misordered */
1383                         status = nfserr_seq_misordered;
1384                         goto out;
1385                 }
1386
1387                 slot->sl_seqid++; /* from 0 to 1 */
1388                 move_to_confirmed(unconf);
1389
1390                 /*
1391                  * We do not support RDMA or persistent sessions
1392                  */
1393                 cr_ses->flags &= ~SESSION4_PERSIST;
1394                 cr_ses->flags &= ~SESSION4_RDMA;
1395
1396                 conf = unconf;
1397         } else {
1398                 status = nfserr_stale_clientid;
1399                 goto out;
1400         }
1401
1402         status = alloc_init_session(rqstp, conf, cr_ses);
1403         if (status)
1404                 goto out;
1405
1406         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1407                NFS4_MAX_SESSIONID_LEN);
1408         cr_ses->seqid = slot->sl_seqid;
1409
1410         slot->sl_inuse = true;
1411         cstate->slot = slot;
1412         /* Ensure a page is used for the cache */
1413         slot->sl_cache_entry.ce_cachethis = 1;
1414 out:
1415         nfs4_unlock_state();
1416         dprintk("%s returns %d\n", __func__, ntohl(status));
1417         return status;
1418 }
1419
1420 __be32
1421 nfsd4_destroy_session(struct svc_rqst *r,
1422                       struct nfsd4_compound_state *cstate,
1423                       struct nfsd4_destroy_session *sessionid)
1424 {
1425         struct nfsd4_session *ses;
1426         u32 status = nfserr_badsession;
1427
1428         /* Notes:
1429          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1430          * - Should we return nfserr_back_chan_busy if waiting for
1431          *   callbacks on to-be-destroyed session?
1432          * - Do we need to clear any callback info from previous session?
1433          */
1434
1435         dump_sessionid(__func__, &sessionid->sessionid);
1436         spin_lock(&sessionid_lock);
1437         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1438         if (!ses) {
1439                 spin_unlock(&sessionid_lock);
1440                 goto out;
1441         }
1442
1443         unhash_session(ses);
1444         spin_unlock(&sessionid_lock);
1445
1446         /* wait for callbacks */
1447         shutdown_callback_client(ses->se_client);
1448         nfsd4_put_session(ses);
1449         status = nfs_ok;
1450 out:
1451         dprintk("%s returns %d\n", __func__, ntohl(status));
1452         return status;
1453 }
1454
1455 __be32
1456 nfsd4_sequence(struct svc_rqst *rqstp,
1457                struct nfsd4_compound_state *cstate,
1458                struct nfsd4_sequence *seq)
1459 {
1460         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1461         struct nfsd4_session *session;
1462         struct nfsd4_slot *slot;
1463         int status;
1464
1465         if (resp->opcnt != 1)
1466                 return nfserr_sequence_pos;
1467
1468         spin_lock(&sessionid_lock);
1469         status = nfserr_badsession;
1470         session = find_in_sessionid_hashtbl(&seq->sessionid);
1471         if (!session)
1472                 goto out;
1473
1474         status = nfserr_badslot;
1475         if (seq->slotid >= session->se_fnumslots)
1476                 goto out;
1477
1478         slot = &session->se_slots[seq->slotid];
1479         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1480
1481         status = check_slot_seqid(seq->seqid, slot);
1482         if (status == nfserr_replay_cache) {
1483                 cstate->slot = slot;
1484                 cstate->session = session;
1485                 /* Return the cached reply status and set cstate->status
1486                  * for nfsd4_svc_encode_compoundres processing */
1487                 status = nfsd4_replay_cache_entry(resp, seq);
1488                 cstate->status = nfserr_replay_cache;
1489                 goto replay_cache;
1490         }
1491         if (status)
1492                 goto out;
1493
1494         /* Success! bump slot seqid */
1495         slot->sl_inuse = true;
1496         slot->sl_seqid = seq->seqid;
1497         slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1498         /* Always set the cache entry cachethis for solo sequence */
1499         if (nfsd4_is_solo_sequence(resp))
1500                 slot->sl_cache_entry.ce_cachethis = 1;
1501
1502         cstate->slot = slot;
1503         cstate->session = session;
1504
1505 replay_cache:
1506         /* Renew the clientid on success and on replay.
1507          * Hold a session reference until done processing the compound:
1508          * nfsd4_put_session called only if the cstate slot is set.
1509          */
1510         renew_client(session->se_client);
1511         nfsd4_get_session(session);
1512 out:
1513         spin_unlock(&sessionid_lock);
1514         dprintk("%s: return %d\n", __func__, ntohl(status));
1515         return status;
1516 }
1517
1518 __be32
1519 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1520                   struct nfsd4_setclientid *setclid)
1521 {
1522         struct sockaddr_in      *sin = svc_addr_in(rqstp);
1523         struct xdr_netobj       clname = {
1524                 .len = setclid->se_namelen,
1525                 .data = setclid->se_name,
1526         };
1527         nfs4_verifier           clverifier = setclid->se_verf;
1528         unsigned int            strhashval;
1529         struct nfs4_client      *conf, *unconf, *new;
1530         __be32                  status;
1531         char                    *princ;
1532         char                    dname[HEXDIR_LEN];
1533
1534         if (!check_name(clname))
1535                 return nfserr_inval;
1536
1537         status = nfs4_make_rec_clidname(dname, &clname);
1538         if (status)
1539                 return status;
1540
1541         /*
1542          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1543          * We get here on a DRC miss.
1544          */
1545
1546         strhashval = clientstr_hashval(dname);
1547
1548         nfs4_lock_state();
1549         conf = find_confirmed_client_by_str(dname, strhashval, false);
1550         if (conf) {
1551                 /* RFC 3530 14.2.33 CASE 0: */
1552                 status = nfserr_clid_inuse;
1553                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1554                         dprintk("NFSD: setclientid: string in use by client"
1555                                 " at %pI4\n", &conf->cl_addr);
1556                         goto out;
1557                 }
1558         }
1559         /*
1560          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1561          * has a description of SETCLIENTID request processing consisting
1562          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1563          */
1564         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1565         status = nfserr_resource;
1566         if (!conf) {
1567                 /*
1568                  * RFC 3530 14.2.33 CASE 4:
1569                  * placed first, because it is the normal case
1570                  */
1571                 if (unconf)
1572                         expire_client(unconf);
1573                 new = create_client(clname, dname);
1574                 if (new == NULL)
1575                         goto out;
1576                 gen_clid(new);
1577         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1578                 /*
1579                  * RFC 3530 14.2.33 CASE 1:
1580                  * probable callback update
1581                  */
1582                 if (unconf) {
1583                         /* Note this is removing unconfirmed {*x***},
1584                          * which is stronger than RFC recommended {vxc**}.
1585                          * This has the advantage that there is at most
1586                          * one {*x***} in either list at any time.
1587                          */
1588                         expire_client(unconf);
1589                 }
1590                 new = create_client(clname, dname);
1591                 if (new == NULL)
1592                         goto out;
1593                 copy_clid(new, conf);
1594         } else if (!unconf) {
1595                 /*
1596                  * RFC 3530 14.2.33 CASE 2:
1597                  * probable client reboot; state will be removed if
1598                  * confirmed.
1599                  */
1600                 new = create_client(clname, dname);
1601                 if (new == NULL)
1602                         goto out;
1603                 gen_clid(new);
1604         } else {
1605                 /*
1606                  * RFC 3530 14.2.33 CASE 3:
1607                  * probable client reboot; state will be removed if
1608                  * confirmed.
1609                  */
1610                 expire_client(unconf);
1611                 new = create_client(clname, dname);
1612                 if (new == NULL)
1613                         goto out;
1614                 gen_clid(new);
1615         }
1616         copy_verf(new, &clverifier);
1617         new->cl_addr = sin->sin_addr.s_addr;
1618         new->cl_flavor = rqstp->rq_flavor;
1619         princ = svc_gss_principal(rqstp);
1620         if (princ) {
1621                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1622                 if (new->cl_principal == NULL) {
1623                         free_client(new);
1624                         goto out;
1625                 }
1626         }
1627         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1628         gen_confirm(new);
1629         gen_callback(new, setclid);
1630         add_to_unconfirmed(new, strhashval);
1631         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1632         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1633         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1634         status = nfs_ok;
1635 out:
1636         nfs4_unlock_state();
1637         return status;
1638 }
1639
1640
1641 /*
1642  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1643  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1644  * bullets, labeled as CASE1 - CASE4 below.
1645  */
1646 __be32
1647 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1648                          struct nfsd4_compound_state *cstate,
1649                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1650 {
1651         struct sockaddr_in *sin = svc_addr_in(rqstp);
1652         struct nfs4_client *conf, *unconf;
1653         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1654         clientid_t * clid = &setclientid_confirm->sc_clientid;
1655         __be32 status;
1656
1657         if (STALE_CLIENTID(clid))
1658                 return nfserr_stale_clientid;
1659         /*
1660          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1661          * We get here on a DRC miss.
1662          */
1663
1664         nfs4_lock_state();
1665
1666         conf = find_confirmed_client(clid);
1667         unconf = find_unconfirmed_client(clid);
1668
1669         status = nfserr_clid_inuse;
1670         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1671                 goto out;
1672         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1673                 goto out;
1674
1675         /*
1676          * section 14.2.34 of RFC 3530 has a description of
1677          * SETCLIENTID_CONFIRM request processing consisting
1678          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1679          */
1680         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1681                 /*
1682                  * RFC 3530 14.2.34 CASE 1:
1683                  * callback update
1684                  */
1685                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1686                         status = nfserr_clid_inuse;
1687                 else {
1688                         /* XXX: We just turn off callbacks until we can handle
1689                           * change request correctly. */
1690                         atomic_set(&conf->cl_callback.cb_set, 0);
1691                         gen_confirm(conf);
1692                         nfsd4_remove_clid_dir(unconf);
1693                         expire_client(unconf);
1694                         status = nfs_ok;
1695
1696                 }
1697         } else if (conf && !unconf) {
1698                 /*
1699                  * RFC 3530 14.2.34 CASE 2:
1700                  * probable retransmitted request; play it safe and
1701                  * do nothing.
1702                  */
1703                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1704                         status = nfserr_clid_inuse;
1705                 else
1706                         status = nfs_ok;
1707         } else if (!conf && unconf
1708                         && same_verf(&unconf->cl_confirm, &confirm)) {
1709                 /*
1710                  * RFC 3530 14.2.34 CASE 3:
1711                  * Normal case; new or rebooted client:
1712                  */
1713                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1714                         status = nfserr_clid_inuse;
1715                 } else {
1716                         unsigned int hash =
1717                                 clientstr_hashval(unconf->cl_recdir);
1718                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1719                                                             hash, false);
1720                         if (conf) {
1721                                 nfsd4_remove_clid_dir(conf);
1722                                 expire_client(conf);
1723                         }
1724                         move_to_confirmed(unconf);
1725                         conf = unconf;
1726                         nfsd4_probe_callback(conf);
1727                         status = nfs_ok;
1728                 }
1729         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1730             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1731                                                                 &confirm)))) {
1732                 /*
1733                  * RFC 3530 14.2.34 CASE 4:
1734                  * Client probably hasn't noticed that we rebooted yet.
1735                  */
1736                 status = nfserr_stale_clientid;
1737         } else {
1738                 /* check that we have hit one of the cases...*/
1739                 status = nfserr_clid_inuse;
1740         }
1741 out:
1742         nfs4_unlock_state();
1743         return status;
1744 }
1745
1746 /* OPEN Share state helper functions */
1747 static inline struct nfs4_file *
1748 alloc_init_file(struct inode *ino)
1749 {
1750         struct nfs4_file *fp;
1751         unsigned int hashval = file_hashval(ino);
1752
1753         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1754         if (fp) {
1755                 atomic_set(&fp->fi_ref, 1);
1756                 INIT_LIST_HEAD(&fp->fi_hash);
1757                 INIT_LIST_HEAD(&fp->fi_stateids);
1758                 INIT_LIST_HEAD(&fp->fi_delegations);
1759                 spin_lock(&recall_lock);
1760                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1761                 spin_unlock(&recall_lock);
1762                 fp->fi_inode = igrab(ino);
1763                 fp->fi_id = current_fileid++;
1764                 fp->fi_had_conflict = false;
1765                 return fp;
1766         }
1767         return NULL;
1768 }
1769
1770 static void
1771 nfsd4_free_slab(struct kmem_cache **slab)
1772 {
1773         if (*slab == NULL)
1774                 return;
1775         kmem_cache_destroy(*slab);
1776         *slab = NULL;
1777 }
1778
1779 void
1780 nfsd4_free_slabs(void)
1781 {
1782         nfsd4_free_slab(&stateowner_slab);
1783         nfsd4_free_slab(&file_slab);
1784         nfsd4_free_slab(&stateid_slab);
1785         nfsd4_free_slab(&deleg_slab);
1786 }
1787
1788 static int
1789 nfsd4_init_slabs(void)
1790 {
1791         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1792                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1793         if (stateowner_slab == NULL)
1794                 goto out_nomem;
1795         file_slab = kmem_cache_create("nfsd4_files",
1796                         sizeof(struct nfs4_file), 0, 0, NULL);
1797         if (file_slab == NULL)
1798                 goto out_nomem;
1799         stateid_slab = kmem_cache_create("nfsd4_stateids",
1800                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1801         if (stateid_slab == NULL)
1802                 goto out_nomem;
1803         deleg_slab = kmem_cache_create("nfsd4_delegations",
1804                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1805         if (deleg_slab == NULL)
1806                 goto out_nomem;
1807         return 0;
1808 out_nomem:
1809         nfsd4_free_slabs();
1810         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1811         return -ENOMEM;
1812 }
1813
1814 void
1815 nfs4_free_stateowner(struct kref *kref)
1816 {
1817         struct nfs4_stateowner *sop =
1818                 container_of(kref, struct nfs4_stateowner, so_ref);
1819         kfree(sop->so_owner.data);
1820         kmem_cache_free(stateowner_slab, sop);
1821 }
1822
1823 static inline struct nfs4_stateowner *
1824 alloc_stateowner(struct xdr_netobj *owner)
1825 {
1826         struct nfs4_stateowner *sop;
1827
1828         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1829                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1830                         memcpy(sop->so_owner.data, owner->data, owner->len);
1831                         sop->so_owner.len = owner->len;
1832                         kref_init(&sop->so_ref);
1833                         return sop;
1834                 }
1835                 kmem_cache_free(stateowner_slab, sop);
1836         }
1837         return NULL;
1838 }
1839
1840 static struct nfs4_stateowner *
1841 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1842         struct nfs4_stateowner *sop;
1843         struct nfs4_replay *rp;
1844         unsigned int idhashval;
1845
1846         if (!(sop = alloc_stateowner(&open->op_owner)))
1847                 return NULL;
1848         idhashval = ownerid_hashval(current_ownerid);
1849         INIT_LIST_HEAD(&sop->so_idhash);
1850         INIT_LIST_HEAD(&sop->so_strhash);
1851         INIT_LIST_HEAD(&sop->so_perclient);
1852         INIT_LIST_HEAD(&sop->so_stateids);
1853         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1854         INIT_LIST_HEAD(&sop->so_close_lru);
1855         sop->so_time = 0;
1856         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1857         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1858         list_add(&sop->so_perclient, &clp->cl_openowners);
1859         sop->so_is_open_owner = 1;
1860         sop->so_id = current_ownerid++;
1861         sop->so_client = clp;
1862         sop->so_seqid = open->op_seqid;
1863         sop->so_confirmed = 0;
1864         rp = &sop->so_replay;
1865         rp->rp_status = nfserr_serverfault;
1866         rp->rp_buflen = 0;
1867         rp->rp_buf = rp->rp_ibuf;
1868         return sop;
1869 }
1870
1871 static inline void
1872 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1873         struct nfs4_stateowner *sop = open->op_stateowner;
1874         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1875
1876         INIT_LIST_HEAD(&stp->st_hash);
1877         INIT_LIST_HEAD(&stp->st_perstateowner);
1878         INIT_LIST_HEAD(&stp->st_lockowners);
1879         INIT_LIST_HEAD(&stp->st_perfile);
1880         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1881         list_add(&stp->st_perstateowner, &sop->so_stateids);
1882         list_add(&stp->st_perfile, &fp->fi_stateids);
1883         stp->st_stateowner = sop;
1884         get_nfs4_file(fp);
1885         stp->st_file = fp;
1886         stp->st_stateid.si_boot = boot_time;
1887         stp->st_stateid.si_stateownerid = sop->so_id;
1888         stp->st_stateid.si_fileid = fp->fi_id;
1889         stp->st_stateid.si_generation = 0;
1890         stp->st_access_bmap = 0;
1891         stp->st_deny_bmap = 0;
1892         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1893                   &stp->st_access_bmap);
1894         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1895         stp->st_openstp = NULL;
1896 }
1897
1898 static void
1899 move_to_close_lru(struct nfs4_stateowner *sop)
1900 {
1901         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1902
1903         list_move_tail(&sop->so_close_lru, &close_lru);
1904         sop->so_time = get_seconds();
1905 }
1906
1907 static int
1908 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1909                                                         clientid_t *clid)
1910 {
1911         return (sop->so_owner.len == owner->len) &&
1912                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1913                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1914 }
1915
1916 static struct nfs4_stateowner *
1917 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1918 {
1919         struct nfs4_stateowner *so = NULL;
1920
1921         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1922                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1923                         return so;
1924         }
1925         return NULL;
1926 }
1927
1928 /* search file_hashtbl[] for file */
1929 static struct nfs4_file *
1930 find_file(struct inode *ino)
1931 {
1932         unsigned int hashval = file_hashval(ino);
1933         struct nfs4_file *fp;
1934
1935         spin_lock(&recall_lock);
1936         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1937                 if (fp->fi_inode == ino) {
1938                         get_nfs4_file(fp);
1939                         spin_unlock(&recall_lock);
1940                         return fp;
1941                 }
1942         }
1943         spin_unlock(&recall_lock);
1944         return NULL;
1945 }
1946
1947 static inline int access_valid(u32 x, u32 minorversion)
1948 {
1949         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1950                 return 0;
1951         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1952                 return 0;
1953         x &= ~NFS4_SHARE_ACCESS_MASK;
1954         if (minorversion && x) {
1955                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1956                         return 0;
1957                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1958                         return 0;
1959                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1960         }
1961         if (x)
1962                 return 0;
1963         return 1;
1964 }
1965
1966 static inline int deny_valid(u32 x)
1967 {
1968         /* Note: unlike access bits, deny bits may be zero. */
1969         return x <= NFS4_SHARE_DENY_BOTH;
1970 }
1971
1972 /*
1973  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1974  * st_{access,deny}_bmap field of the stateid, in order to track not
1975  * only what share bits are currently in force, but also what
1976  * combinations of share bits previous opens have used.  This allows us
1977  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1978  * return an error if the client attempt to downgrade to a combination
1979  * of share bits not explicable by closing some of its previous opens.
1980  *
1981  * XXX: This enforcement is actually incomplete, since we don't keep
1982  * track of access/deny bit combinations; so, e.g., we allow:
1983  *
1984  *      OPEN allow read, deny write
1985  *      OPEN allow both, deny none
1986  *      DOWNGRADE allow read, deny none
1987  *
1988  * which we should reject.
1989  */
1990 static void
1991 set_access(unsigned int *access, unsigned long bmap) {
1992         int i;
1993
1994         *access = 0;
1995         for (i = 1; i < 4; i++) {
1996                 if (test_bit(i, &bmap))
1997                         *access |= i;
1998         }
1999 }
2000
2001 static void
2002 set_deny(unsigned int *deny, unsigned long bmap) {
2003         int i;
2004
2005         *deny = 0;
2006         for (i = 0; i < 4; i++) {
2007                 if (test_bit(i, &bmap))
2008                         *deny |= i ;
2009         }
2010 }
2011
2012 static int
2013 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2014         unsigned int access, deny;
2015
2016         set_access(&access, stp->st_access_bmap);
2017         set_deny(&deny, stp->st_deny_bmap);
2018         if ((access & open->op_share_deny) || (deny & open->op_share_access))
2019                 return 0;
2020         return 1;
2021 }
2022
2023 /*
2024  * Called to check deny when READ with all zero stateid or
2025  * WRITE with all zero or all one stateid
2026  */
2027 static __be32
2028 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2029 {
2030         struct inode *ino = current_fh->fh_dentry->d_inode;
2031         struct nfs4_file *fp;
2032         struct nfs4_stateid *stp;
2033         __be32 ret;
2034
2035         dprintk("NFSD: nfs4_share_conflict\n");
2036
2037         fp = find_file(ino);
2038         if (!fp)
2039                 return nfs_ok;
2040         ret = nfserr_locked;
2041         /* Search for conflicting share reservations */
2042         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2043                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2044                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2045                         goto out;
2046         }
2047         ret = nfs_ok;
2048 out:
2049         put_nfs4_file(fp);
2050         return ret;
2051 }
2052
2053 static inline void
2054 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2055 {
2056         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2057                 drop_file_write_access(filp);
2058                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2059         }
2060 }
2061
2062 /*
2063  * Recall a delegation
2064  */
2065 static int
2066 do_recall(void *__dp)
2067 {
2068         struct nfs4_delegation *dp = __dp;
2069
2070         dp->dl_file->fi_had_conflict = true;
2071         nfsd4_cb_recall(dp);
2072         return 0;
2073 }
2074
2075 /*
2076  * Spawn a thread to perform a recall on the delegation represented
2077  * by the lease (file_lock)
2078  *
2079  * Called from break_lease() with lock_kernel() held.
2080  * Note: we assume break_lease will only call this *once* for any given
2081  * lease.
2082  */
2083 static
2084 void nfsd_break_deleg_cb(struct file_lock *fl)
2085 {
2086         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
2087         struct task_struct *t;
2088
2089         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2090         if (!dp)
2091                 return;
2092
2093         /* We're assuming the state code never drops its reference
2094          * without first removing the lease.  Since we're in this lease
2095          * callback (and since the lease code is serialized by the kernel
2096          * lock) we know the server hasn't removed the lease yet, we know
2097          * it's safe to take a reference: */
2098         atomic_inc(&dp->dl_count);
2099         atomic_inc(&dp->dl_client->cl_count);
2100
2101         spin_lock(&recall_lock);
2102         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2103         spin_unlock(&recall_lock);
2104
2105         /* only place dl_time is set. protected by lock_kernel*/
2106         dp->dl_time = get_seconds();
2107
2108         /*
2109          * We don't want the locks code to timeout the lease for us;
2110          * we'll remove it ourself if the delegation isn't returned
2111          * in time.
2112          */
2113         fl->fl_break_time = 0;
2114
2115         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
2116         if (IS_ERR(t)) {
2117                 struct nfs4_client *clp = dp->dl_client;
2118
2119                 printk(KERN_INFO "NFSD: Callback thread failed for "
2120                         "for client (clientid %08x/%08x)\n",
2121                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2122                 put_nfs4_client(dp->dl_client);
2123                 nfs4_put_delegation(dp);
2124         }
2125 }
2126
2127 /*
2128  * The file_lock is being reapd.
2129  *
2130  * Called by locks_free_lock() with lock_kernel() held.
2131  */
2132 static
2133 void nfsd_release_deleg_cb(struct file_lock *fl)
2134 {
2135         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2136
2137         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2138
2139         if (!(fl->fl_flags & FL_LEASE) || !dp)
2140                 return;
2141         dp->dl_flock = NULL;
2142 }
2143
2144 /*
2145  * Set the delegation file_lock back pointer.
2146  *
2147  * Called from setlease() with lock_kernel() held.
2148  */
2149 static
2150 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2151 {
2152         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2153
2154         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2155         if (!dp)
2156                 return;
2157         dp->dl_flock = new;
2158 }
2159
2160 /*
2161  * Called from setlease() with lock_kernel() held
2162  */
2163 static
2164 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2165 {
2166         struct nfs4_delegation *onlistd =
2167                 (struct nfs4_delegation *)onlist->fl_owner;
2168         struct nfs4_delegation *tryd =
2169                 (struct nfs4_delegation *)try->fl_owner;
2170
2171         if (onlist->fl_lmops != try->fl_lmops)
2172                 return 0;
2173
2174         return onlistd->dl_client == tryd->dl_client;
2175 }
2176
2177
2178 static
2179 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2180 {
2181         if (arg & F_UNLCK)
2182                 return lease_modify(onlist, arg);
2183         else
2184                 return -EAGAIN;
2185 }
2186
2187 static struct lock_manager_operations nfsd_lease_mng_ops = {
2188         .fl_break = nfsd_break_deleg_cb,
2189         .fl_release_private = nfsd_release_deleg_cb,
2190         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2191         .fl_mylease = nfsd_same_client_deleg_cb,
2192         .fl_change = nfsd_change_deleg_cb,
2193 };
2194
2195
2196 __be32
2197 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2198                     struct nfsd4_open *open)
2199 {
2200         clientid_t *clientid = &open->op_clientid;
2201         struct nfs4_client *clp = NULL;
2202         unsigned int strhashval;
2203         struct nfs4_stateowner *sop = NULL;
2204
2205         if (!check_name(open->op_owner))
2206                 return nfserr_inval;
2207
2208         if (STALE_CLIENTID(&open->op_clientid))
2209                 return nfserr_stale_clientid;
2210
2211         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2212         sop = find_openstateowner_str(strhashval, open);
2213         open->op_stateowner = sop;
2214         if (!sop) {
2215                 /* Make sure the client's lease hasn't expired. */
2216                 clp = find_confirmed_client(clientid);
2217                 if (clp == NULL)
2218                         return nfserr_expired;
2219                 goto renew;
2220         }
2221         /* When sessions are used, skip open sequenceid processing */
2222         if (nfsd4_has_session(cstate))
2223                 goto renew;
2224         if (!sop->so_confirmed) {
2225                 /* Replace unconfirmed owners without checking for replay. */
2226                 clp = sop->so_client;
2227                 release_openowner(sop);
2228                 open->op_stateowner = NULL;
2229                 goto renew;
2230         }
2231         if (open->op_seqid == sop->so_seqid - 1) {
2232                 if (sop->so_replay.rp_buflen)
2233                         return nfserr_replay_me;
2234                 /* The original OPEN failed so spectacularly
2235                  * that we don't even have replay data saved!
2236                  * Therefore, we have no choice but to continue
2237                  * processing this OPEN; presumably, we'll
2238                  * fail again for the same reason.
2239                  */
2240                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2241                 goto renew;
2242         }
2243         if (open->op_seqid != sop->so_seqid)
2244                 return nfserr_bad_seqid;
2245 renew:
2246         if (open->op_stateowner == NULL) {
2247                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2248                 if (sop == NULL)
2249                         return nfserr_resource;
2250                 open->op_stateowner = sop;
2251         }
2252         list_del_init(&sop->so_close_lru);
2253         renew_client(sop->so_client);
2254         return nfs_ok;
2255 }
2256
2257 static inline __be32
2258 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2259 {
2260         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2261                 return nfserr_openmode;
2262         else
2263                 return nfs_ok;
2264 }
2265
2266 static struct nfs4_delegation *
2267 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2268 {
2269         struct nfs4_delegation *dp;
2270
2271         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2272                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2273                         return dp;
2274         }
2275         return NULL;
2276 }
2277
2278 static __be32
2279 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2280                 struct nfs4_delegation **dp)
2281 {
2282         int flags;
2283         __be32 status = nfserr_bad_stateid;
2284
2285         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2286         if (*dp == NULL)
2287                 goto out;
2288         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2289                                                 RD_STATE : WR_STATE;
2290         status = nfs4_check_delegmode(*dp, flags);
2291         if (status)
2292                 *dp = NULL;
2293 out:
2294         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2295                 return nfs_ok;
2296         if (status)
2297                 return status;
2298         open->op_stateowner->so_confirmed = 1;
2299         return nfs_ok;
2300 }
2301
2302 static __be32
2303 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2304 {
2305         struct nfs4_stateid *local;
2306         __be32 status = nfserr_share_denied;
2307         struct nfs4_stateowner *sop = open->op_stateowner;
2308
2309         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2310                 /* ignore lock owners */
2311                 if (local->st_stateowner->so_is_open_owner == 0)
2312                         continue;
2313                 /* remember if we have seen this open owner */
2314                 if (local->st_stateowner == sop)
2315                         *stpp = local;
2316                 /* check for conflicting share reservations */
2317                 if (!test_share(local, open))
2318                         goto out;
2319         }
2320         status = 0;
2321 out:
2322         return status;
2323 }
2324
2325 static inline struct nfs4_stateid *
2326 nfs4_alloc_stateid(void)
2327 {
2328         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2329 }
2330
2331 static __be32
2332 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2333                 struct nfs4_delegation *dp,
2334                 struct svc_fh *cur_fh, int flags)
2335 {
2336         struct nfs4_stateid *stp;
2337
2338         stp = nfs4_alloc_stateid();
2339         if (stp == NULL)
2340                 return nfserr_resource;
2341
2342         if (dp) {
2343                 get_file(dp->dl_vfs_file);
2344                 stp->st_vfs_file = dp->dl_vfs_file;
2345         } else {
2346                 __be32 status;
2347                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2348                                 &stp->st_vfs_file);
2349                 if (status) {
2350                         if (status == nfserr_dropit)
2351                                 status = nfserr_jukebox;
2352                         kmem_cache_free(stateid_slab, stp);
2353                         return status;
2354                 }
2355         }
2356         *stpp = stp;
2357         return 0;
2358 }
2359
2360 static inline __be32
2361 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2362                 struct nfsd4_open *open)
2363 {
2364         struct iattr iattr = {
2365                 .ia_valid = ATTR_SIZE,
2366                 .ia_size = 0,
2367         };
2368         if (!open->op_truncate)
2369                 return 0;
2370         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2371                 return nfserr_inval;
2372         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2373 }
2374
2375 static __be32
2376 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2377 {
2378         struct file *filp = stp->st_vfs_file;
2379         struct inode *inode = filp->f_path.dentry->d_inode;
2380         unsigned int share_access, new_writer;
2381         __be32 status;
2382
2383         set_access(&share_access, stp->st_access_bmap);
2384         new_writer = (~share_access) & open->op_share_access
2385                         & NFS4_SHARE_ACCESS_WRITE;
2386
2387         if (new_writer) {
2388                 int err = get_write_access(inode);
2389                 if (err)
2390                         return nfserrno(err);
2391                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2392                 if (err)
2393                         return nfserrno(err);
2394                 file_take_write(filp);
2395         }
2396         status = nfsd4_truncate(rqstp, cur_fh, open);
2397         if (status) {
2398                 if (new_writer)
2399                         put_write_access(inode);
2400                 return status;
2401         }
2402         /* remember the open */
2403         filp->f_mode |= open->op_share_access;
2404         __set_bit(open->op_share_access, &stp->st_access_bmap);
2405         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2406
2407         return nfs_ok;
2408 }
2409
2410
2411 static void
2412 nfs4_set_claim_prev(struct nfsd4_open *open)
2413 {
2414         open->op_stateowner->so_confirmed = 1;
2415         open->op_stateowner->so_client->cl_firststate = 1;
2416 }
2417
2418 /*
2419  * Attempt to hand out a delegation.
2420  */
2421 static void
2422 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2423 {
2424         struct nfs4_delegation *dp;
2425         struct nfs4_stateowner *sop = stp->st_stateowner;
2426         struct nfs4_callback *cb = &sop->so_client->cl_callback;
2427         struct file_lock fl, *flp = &fl;
2428         int status, flag = 0;
2429
2430         flag = NFS4_OPEN_DELEGATE_NONE;
2431         open->op_recall = 0;
2432         switch (open->op_claim_type) {
2433                 case NFS4_OPEN_CLAIM_PREVIOUS:
2434                         if (!atomic_read(&cb->cb_set))
2435                                 open->op_recall = 1;
2436                         flag = open->op_delegate_type;
2437                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2438                                 goto out;
2439                         break;
2440                 case NFS4_OPEN_CLAIM_NULL:
2441                         /* Let's not give out any delegations till everyone's
2442                          * had the chance to reclaim theirs.... */
2443                         if (locks_in_grace())
2444                                 goto out;
2445                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2446                                 goto out;
2447                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2448                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2449                         else
2450                                 flag = NFS4_OPEN_DELEGATE_READ;
2451                         break;
2452                 default:
2453                         goto out;
2454         }
2455
2456         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2457         if (dp == NULL) {
2458                 flag = NFS4_OPEN_DELEGATE_NONE;
2459                 goto out;
2460         }
2461         locks_init_lock(&fl);
2462         fl.fl_lmops = &nfsd_lease_mng_ops;
2463         fl.fl_flags = FL_LEASE;
2464         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2465         fl.fl_end = OFFSET_MAX;
2466         fl.fl_owner =  (fl_owner_t)dp;
2467         fl.fl_file = stp->st_vfs_file;
2468         fl.fl_pid = current->tgid;
2469
2470         /* vfs_setlease checks to see if delegation should be handed out.
2471          * the lock_manager callbacks fl_mylease and fl_change are used
2472          */
2473         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2474                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2475                 unhash_delegation(dp);
2476                 flag = NFS4_OPEN_DELEGATE_NONE;
2477                 goto out;
2478         }
2479
2480         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2481
2482         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2483                      dp->dl_stateid.si_boot,
2484                      dp->dl_stateid.si_stateownerid,
2485                      dp->dl_stateid.si_fileid,
2486                      dp->dl_stateid.si_generation);
2487 out:
2488         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2489                         && flag == NFS4_OPEN_DELEGATE_NONE
2490                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2491                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2492         open->op_delegate_type = flag;
2493 }
2494
2495 /*
2496  * called with nfs4_lock_state() held.
2497  */
2498 __be32
2499 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2500 {
2501         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2502         struct nfs4_file *fp = NULL;
2503         struct inode *ino = current_fh->fh_dentry->d_inode;
2504         struct nfs4_stateid *stp = NULL;
2505         struct nfs4_delegation *dp = NULL;
2506         __be32 status;
2507
2508         status = nfserr_inval;
2509         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2510                         || !deny_valid(open->op_share_deny))
2511                 goto out;
2512         /*
2513          * Lookup file; if found, lookup stateid and check open request,
2514          * and check for delegations in the process of being recalled.
2515          * If not found, create the nfs4_file struct
2516          */
2517         fp = find_file(ino);
2518         if (fp) {
2519                 if ((status = nfs4_check_open(fp, open, &stp)))
2520                         goto out;
2521                 status = nfs4_check_deleg(fp, open, &dp);
2522                 if (status)
2523                         goto out;
2524         } else {
2525                 status = nfserr_bad_stateid;
2526                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2527                         goto out;
2528                 status = nfserr_resource;
2529                 fp = alloc_init_file(ino);
2530                 if (fp == NULL)
2531                         goto out;
2532         }
2533
2534         /*
2535          * OPEN the file, or upgrade an existing OPEN.
2536          * If truncate fails, the OPEN fails.
2537          */
2538         if (stp) {
2539                 /* Stateid was found, this is an OPEN upgrade */
2540                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2541                 if (status)
2542                         goto out;
2543                 update_stateid(&stp->st_stateid);
2544         } else {
2545                 /* Stateid was not found, this is a new OPEN */
2546                 int flags = 0;
2547                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2548                         flags |= NFSD_MAY_READ;
2549                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2550                         flags |= NFSD_MAY_WRITE;
2551                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2552                 if (status)
2553                         goto out;
2554                 init_stateid(stp, fp, open);
2555                 status = nfsd4_truncate(rqstp, current_fh, open);
2556                 if (status) {
2557                         release_open_stateid(stp);
2558                         goto out;
2559                 }
2560                 if (nfsd4_has_session(&resp->cstate))
2561                         update_stateid(&stp->st_stateid);
2562         }
2563         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2564
2565         if (nfsd4_has_session(&resp->cstate))
2566                 open->op_stateowner->so_confirmed = 1;
2567
2568         /*
2569         * Attempt to hand out a delegation. No error return, because the
2570         * OPEN succeeds even if we fail.
2571         */
2572         nfs4_open_delegation(current_fh, open, stp);
2573
2574         status = nfs_ok;
2575
2576         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2577                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2578                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2579 out:
2580         if (fp)
2581                 put_nfs4_file(fp);
2582         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2583                 nfs4_set_claim_prev(open);
2584         /*
2585         * To finish the open response, we just need to set the rflags.
2586         */
2587         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2588         if (!open->op_stateowner->so_confirmed &&
2589             !nfsd4_has_session(&resp->cstate))
2590                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2591
2592         return status;
2593 }
2594
2595 __be32
2596 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2597             clientid_t *clid)
2598 {
2599         struct nfs4_client *clp;
2600         __be32 status;
2601
2602         nfs4_lock_state();
2603         dprintk("process_renew(%08x/%08x): starting\n",
2604                         clid->cl_boot, clid->cl_id);
2605         status = nfserr_stale_clientid;
2606         if (STALE_CLIENTID(clid))
2607                 goto out;
2608         clp = find_confirmed_client(clid);
2609         status = nfserr_expired;
2610         if (clp == NULL) {
2611                 /* We assume the client took too long to RENEW. */
2612                 dprintk("nfsd4_renew: clientid not found!\n");
2613                 goto out;
2614         }
2615         renew_client(clp);
2616         status = nfserr_cb_path_down;
2617         if (!list_empty(&clp->cl_delegations)
2618                         && !atomic_read(&clp->cl_callback.cb_set))
2619                 goto out;
2620         status = nfs_ok;
2621 out:
2622         nfs4_unlock_state();
2623         return status;
2624 }
2625
2626 struct lock_manager nfsd4_manager = {
2627 };
2628
2629 static void
2630 nfsd4_end_grace(void)
2631 {
2632         dprintk("NFSD: end of grace period\n");
2633         nfsd4_recdir_purge_old();
2634         locks_end_grace(&nfsd4_manager);
2635 }
2636
2637 static time_t
2638 nfs4_laundromat(void)
2639 {
2640         struct nfs4_client *clp;
2641         struct nfs4_stateowner *sop;
2642         struct nfs4_delegation *dp;
2643         struct list_head *pos, *next, reaplist;
2644         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2645         time_t t, clientid_val = NFSD_LEASE_TIME;
2646         time_t u, test_val = NFSD_LEASE_TIME;
2647
2648         nfs4_lock_state();
2649
2650         dprintk("NFSD: laundromat service - starting\n");
2651         if (locks_in_grace())
2652                 nfsd4_end_grace();
2653         list_for_each_safe(pos, next, &client_lru) {
2654                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2655                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2656                         t = clp->cl_time - cutoff;
2657                         if (clientid_val > t)
2658                                 clientid_val = t;
2659                         break;
2660                 }
2661                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2662                         clp->cl_clientid.cl_id);
2663                 nfsd4_remove_clid_dir(clp);
2664                 expire_client(clp);
2665         }
2666         INIT_LIST_HEAD(&reaplist);
2667         spin_lock(&recall_lock);
2668         list_for_each_safe(pos, next, &del_recall_lru) {
2669                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2670                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2671                         u = dp->dl_time - cutoff;
2672                         if (test_val > u)
2673                                 test_val = u;
2674                         break;
2675                 }
2676                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2677                                     dp, dp->dl_flock);
2678                 list_move(&dp->dl_recall_lru, &reaplist);
2679         }
2680         spin_unlock(&recall_lock);
2681         list_for_each_safe(pos, next, &reaplist) {
2682                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2683                 list_del_init(&dp->dl_recall_lru);
2684                 unhash_delegation(dp);
2685         }
2686         test_val = NFSD_LEASE_TIME;
2687         list_for_each_safe(pos, next, &close_lru) {
2688                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2689                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2690                         u = sop->so_time - cutoff;
2691                         if (test_val > u)
2692                                 test_val = u;
2693                         break;
2694                 }
2695                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2696                         sop->so_id);
2697                 release_openowner(sop);
2698         }
2699         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2700                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2701         nfs4_unlock_state();
2702         return clientid_val;
2703 }
2704
2705 static struct workqueue_struct *laundry_wq;
2706 static void laundromat_main(struct work_struct *);
2707 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2708
2709 static void
2710 laundromat_main(struct work_struct *not_used)
2711 {
2712         time_t t;
2713
2714         t = nfs4_laundromat();
2715         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2716         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2717 }
2718
2719 static struct nfs4_stateowner *
2720 search_close_lru(u32 st_id, int flags)
2721 {
2722         struct nfs4_stateowner *local = NULL;
2723
2724         if (flags & CLOSE_STATE) {
2725                 list_for_each_entry(local, &close_lru, so_close_lru) {
2726                         if (local->so_id == st_id)
2727                                 return local;
2728                 }
2729         }
2730         return NULL;
2731 }
2732
2733 static inline int
2734 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2735 {
2736         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2737 }
2738
2739 static int
2740 STALE_STATEID(stateid_t *stateid)
2741 {
2742         if (stateid->si_boot == boot_time)
2743                 return 0;
2744         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2745                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2746                 stateid->si_generation);
2747         return 1;
2748 }
2749
2750 static inline int
2751 access_permit_read(unsigned long access_bmap)
2752 {
2753         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2754                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2755                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2756 }
2757
2758 static inline int
2759 access_permit_write(unsigned long access_bmap)
2760 {
2761         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2762                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2763 }
2764
2765 static
2766 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2767 {
2768         __be32 status = nfserr_openmode;
2769
2770         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2771                 goto out;
2772         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2773                 goto out;
2774         status = nfs_ok;
2775 out:
2776         return status;
2777 }
2778
2779 static inline __be32
2780 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2781 {
2782         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2783                 return nfs_ok;
2784         else if (locks_in_grace()) {
2785                 /* Answer in remaining cases depends on existance of
2786                  * conflicting state; so we must wait out the grace period. */
2787                 return nfserr_grace;
2788         } else if (flags & WR_STATE)
2789                 return nfs4_share_conflict(current_fh,
2790                                 NFS4_SHARE_DENY_WRITE);
2791         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2792                 return nfs4_share_conflict(current_fh,
2793                                 NFS4_SHARE_DENY_READ);
2794 }
2795
2796 /*
2797  * Allow READ/WRITE during grace period on recovered state only for files
2798  * that are not able to provide mandatory locking.
2799  */
2800 static inline int
2801 grace_disallows_io(struct inode *inode)
2802 {
2803         return locks_in_grace() && mandatory_lock(inode);
2804 }
2805
2806 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2807 {
2808         /*
2809          * When sessions are used the stateid generation number is ignored
2810          * when it is zero.
2811          */
2812         if ((flags & HAS_SESSION) && in->si_generation == 0)
2813                 goto out;
2814
2815         /* If the client sends us a stateid from the future, it's buggy: */
2816         if (in->si_generation > ref->si_generation)
2817                 return nfserr_bad_stateid;
2818         /*
2819          * The following, however, can happen.  For example, if the
2820          * client sends an open and some IO at the same time, the open
2821          * may bump si_generation while the IO is still in flight.
2822          * Thanks to hard links and renames, the client never knows what
2823          * file an open will affect.  So it could avoid that situation
2824          * only by serializing all opens and IO from the same open
2825          * owner.  To recover from the old_stateid error, the client
2826          * will just have to retry the IO:
2827          */
2828         if (in->si_generation < ref->si_generation)
2829                 return nfserr_old_stateid;
2830 out:
2831         return nfs_ok;
2832 }
2833
2834 static int is_delegation_stateid(stateid_t *stateid)
2835 {
2836         return stateid->si_fileid == 0;
2837 }
2838
2839 /*
2840 * Checks for stateid operations
2841 */
2842 __be32
2843 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2844                            stateid_t *stateid, int flags, struct file **filpp)
2845 {
2846         struct nfs4_stateid *stp = NULL;
2847         struct nfs4_delegation *dp = NULL;
2848         struct svc_fh *current_fh = &cstate->current_fh;
2849         struct inode *ino = current_fh->fh_dentry->d_inode;
2850         __be32 status;
2851
2852         if (filpp)
2853                 *filpp = NULL;
2854
2855         if (grace_disallows_io(ino))
2856                 return nfserr_grace;
2857
2858         if (nfsd4_has_session(cstate))
2859                 flags |= HAS_SESSION;
2860
2861         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2862                 return check_special_stateids(current_fh, stateid, flags);
2863
2864         status = nfserr_stale_stateid;
2865         if (STALE_STATEID(stateid))
2866                 goto out;
2867
2868         status = nfserr_bad_stateid;
2869         if (is_delegation_stateid(stateid)) {
2870                 dp = find_delegation_stateid(ino, stateid);
2871                 if (!dp)
2872                         goto out;
2873                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2874                                                   flags);
2875                 if (status)
2876                         goto out;
2877                 status = nfs4_check_delegmode(dp, flags);
2878                 if (status)
2879                         goto out;
2880                 renew_client(dp->dl_client);
2881                 if (filpp)
2882                         *filpp = dp->dl_vfs_file;
2883         } else { /* open or lock stateid */
2884                 stp = find_stateid(stateid, flags);
2885                 if (!stp)
2886                         goto out;
2887                 if (nfs4_check_fh(current_fh, stp))
2888                         goto out;
2889                 if (!stp->st_stateowner->so_confirmed)
2890                         goto out;
2891                 status = check_stateid_generation(stateid, &stp->st_stateid,
2892                                                   flags);
2893                 if (status)
2894                         goto out;
2895                 status = nfs4_check_openmode(stp, flags);
2896                 if (status)
2897                         goto out;
2898                 renew_client(stp->st_stateowner->so_client);
2899                 if (filpp)
2900                         *filpp = stp->st_vfs_file;
2901         }
2902         status = nfs_ok;
2903 out:
2904         return status;
2905 }
2906
2907 static inline int
2908 setlkflg (int type)
2909 {
2910         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2911                 RD_STATE : WR_STATE;
2912 }
2913
2914 /*
2915  * Checks for sequence id mutating operations.
2916  */
2917 static __be32
2918 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2919                          stateid_t *stateid, int flags,
2920                          struct nfs4_stateowner **sopp,
2921                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2922 {
2923         struct nfs4_stateid *stp;
2924         struct nfs4_stateowner *sop;
2925         struct svc_fh *current_fh = &cstate->current_fh;
2926         __be32 status;
2927
2928         dprintk("NFSD: preprocess_seqid_op: seqid=%d "
2929                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2930                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2931                 stateid->si_generation);
2932
2933         *stpp = NULL;
2934         *sopp = NULL;
2935
2936         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2937                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2938                 return nfserr_bad_stateid;
2939         }
2940
2941         if (STALE_STATEID(stateid))
2942                 return nfserr_stale_stateid;
2943
2944         if (nfsd4_has_session(cstate))
2945                 flags |= HAS_SESSION;
2946
2947         /*
2948         * We return BAD_STATEID if filehandle doesn't match stateid,
2949         * the confirmed flag is incorrecly set, or the generation
2950         * number is incorrect.
2951         */
2952         stp = find_stateid(stateid, flags);
2953         if (stp == NULL) {
2954                 /*
2955                  * Also, we should make sure this isn't just the result of
2956                  * a replayed close:
2957                  */
2958                 sop = search_close_lru(stateid->si_stateownerid, flags);
2959                 if (sop == NULL)
2960                         return nfserr_bad_stateid;
2961                 *sopp = sop;
2962                 goto check_replay;
2963         }
2964
2965         *stpp = stp;
2966         *sopp = sop = stp->st_stateowner;
2967
2968         if (lock) {
2969                 clientid_t *lockclid = &lock->v.new.clientid;
2970                 struct nfs4_client *clp = sop->so_client;
2971                 int lkflg = 0;
2972                 __be32 status;
2973
2974                 lkflg = setlkflg(lock->lk_type);
2975
2976                 if (lock->lk_is_new) {
2977                         if (!sop->so_is_open_owner)
2978                                 return nfserr_bad_stateid;
2979                         if (!(flags & HAS_SESSION) &&
2980                             !same_clid(&clp->cl_clientid, lockclid))
2981                                 return nfserr_bad_stateid;
2982                         /* stp is the open stateid */
2983                         status = nfs4_check_openmode(stp, lkflg);
2984                         if (status)
2985                                 return status;
2986                 } else {
2987                         /* stp is the lock stateid */
2988                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2989                         if (status)
2990                                 return status;
2991                }
2992         }
2993
2994         if (nfs4_check_fh(current_fh, stp)) {
2995                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2996                 return nfserr_bad_stateid;
2997         }
2998
2999         /*
3000         *  We now validate the seqid and stateid generation numbers.
3001         *  For the moment, we ignore the possibility of
3002         *  generation number wraparound.
3003         */
3004         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3005                 goto check_replay;
3006
3007         if (sop->so_confirmed && flags & CONFIRM) {
3008                 dprintk("NFSD: preprocess_seqid_op: expected"
3009                                 " unconfirmed stateowner!\n");
3010                 return nfserr_bad_stateid;
3011         }
3012         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3013                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3014                                 " confirmed yet!\n");
3015                 return nfserr_bad_stateid;
3016         }
3017         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3018         if (status)
3019                 return status;
3020         renew_client(sop->so_client);
3021         return nfs_ok;
3022
3023 check_replay:
3024         if (seqid == sop->so_seqid - 1) {
3025                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3026                 /* indicate replay to calling function */
3027                 return nfserr_replay_me;
3028         }
3029         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3030                         sop->so_seqid, seqid);
3031         *sopp = NULL;
3032         return nfserr_bad_seqid;
3033 }
3034
3035 __be32
3036 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3037                    struct nfsd4_open_confirm *oc)
3038 {
3039         __be32 status;
3040         struct nfs4_stateowner *sop;
3041         struct nfs4_stateid *stp;
3042
3043         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3044                         (int)cstate->current_fh.fh_dentry->d_name.len,
3045                         cstate->current_fh.fh_dentry->d_name.name);
3046
3047         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3048         if (status)
3049                 return status;
3050
3051         nfs4_lock_state();
3052
3053         if ((status = nfs4_preprocess_seqid_op(cstate,
3054                                         oc->oc_seqid, &oc->oc_req_stateid,
3055                                         CONFIRM | OPEN_STATE,
3056                                         &oc->oc_stateowner, &stp, NULL)))
3057                 goto out;
3058
3059         sop = oc->oc_stateowner;
3060         sop->so_confirmed = 1;
3061         update_stateid(&stp->st_stateid);
3062         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3063         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d "
3064                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3065                          stp->st_stateid.si_boot,
3066                          stp->st_stateid.si_stateownerid,
3067                          stp->st_stateid.si_fileid,
3068                          stp->st_stateid.si_generation);
3069
3070         nfsd4_create_clid_dir(sop->so_client);
3071 out:
3072         if (oc->oc_stateowner) {
3073                 nfs4_get_stateowner(oc->oc_stateowner);
3074                 cstate->replay_owner = oc->oc_stateowner;
3075         }
3076         nfs4_unlock_state();
3077         return status;
3078 }
3079
3080
3081 /*
3082  * unset all bits in union bitmap (bmap) that
3083  * do not exist in share (from successful OPEN_DOWNGRADE)
3084  */
3085 static void
3086 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3087 {
3088         int i;
3089         for (i = 1; i < 4; i++) {
3090                 if ((i & access) != i)
3091                         __clear_bit(i, bmap);
3092         }
3093 }
3094
3095 static void
3096 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3097 {
3098         int i;
3099         for (i = 0; i < 4; i++) {
3100                 if ((i & deny) != i)
3101                         __clear_bit(i, bmap);
3102         }
3103 }
3104
3105 __be32
3106 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3107                      struct nfsd4_compound_state *cstate,
3108                      struct nfsd4_open_downgrade *od)
3109 {
3110         __be32 status;
3111         struct nfs4_stateid *stp;
3112         unsigned int share_access;
3113
3114         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3115                         (int)cstate->current_fh.fh_dentry->d_name.len,
3116                         cstate->current_fh.fh_dentry->d_name.name);
3117
3118         if (!access_valid(od->od_share_access, cstate->minorversion)
3119                         || !deny_valid(od->od_share_deny))
3120                 return nfserr_inval;
3121
3122         nfs4_lock_state();
3123         if ((status = nfs4_preprocess_seqid_op(cstate,
3124                                         od->od_seqid,
3125                                         &od->od_stateid,
3126                                         OPEN_STATE,
3127                                         &od->od_stateowner, &stp, NULL)))
3128                 goto out;
3129
3130         status = nfserr_inval;
3131         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3132                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3133                         stp->st_access_bmap, od->od_share_access);
3134                 goto out;
3135         }
3136         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3137                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3138                         stp->st_deny_bmap, od->od_share_deny);
3139                 goto out;
3140         }
3141         set_access(&share_access, stp->st_access_bmap);
3142         nfs4_file_downgrade(stp->st_vfs_file,
3143                             share_access & ~od->od_share_access);
3144
3145         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3146         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3147
3148         update_stateid(&stp->st_stateid);
3149         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3150         status = nfs_ok;
3151 out:
3152         if (od->od_stateowner) {
3153                 nfs4_get_stateowner(od->od_stateowner);
3154                 cstate->replay_owner = od->od_stateowner;
3155         }
3156         nfs4_unlock_state();
3157         return status;
3158 }
3159
3160 /*
3161  * nfs4_unlock_state() called after encode
3162  */
3163 __be32
3164 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3165             struct nfsd4_close *close)
3166 {
3167         __be32 status;
3168         struct nfs4_stateid *stp;
3169
3170         dprintk("NFSD: nfsd4_close on file %.*s\n",
3171                         (int)cstate->current_fh.fh_dentry->d_name.len,
3172                         cstate->current_fh.fh_dentry->d_name.name);
3173
3174         nfs4_lock_state();
3175         /* check close_lru for replay */
3176         if ((status = nfs4_preprocess_seqid_op(cstate,
3177                                         close->cl_seqid,
3178                                         &close->cl_stateid,
3179                                         OPEN_STATE | CLOSE_STATE,
3180                                         &close->cl_stateowner, &stp, NULL)))
3181                 goto out;
3182         status = nfs_ok;
3183         update_stateid(&stp->st_stateid);
3184         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3185
3186         /* release_stateid() calls nfsd_close() if needed */
3187         release_open_stateid(stp);
3188
3189         /* place unused nfs4_stateowners on so_close_lru list to be
3190          * released by the laundromat service after the lease period
3191          * to enable us to handle CLOSE replay
3192          */
3193         if (list_empty(&close->cl_stateowner->so_stateids))
3194                 move_to_close_lru(close->cl_stateowner);
3195 out:
3196         if (close->cl_stateowner) {
3197                 nfs4_get_stateowner(close->cl_stateowner);
3198                 cstate->replay_owner = close->cl_stateowner;
3199         }
3200         nfs4_unlock_state();
3201         return status;
3202 }
3203
3204 __be32
3205 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3206                   struct nfsd4_delegreturn *dr)
3207 {
3208         struct nfs4_delegation *dp;
3209         stateid_t *stateid = &dr->dr_stateid;
3210         struct inode *inode;
3211         __be32 status;
3212         int flags = 0;
3213
3214         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3215                 return status;
3216         inode = cstate->current_fh.fh_dentry->d_inode;
3217
3218         if (nfsd4_has_session(cstate))
3219                 flags |= HAS_SESSION;
3220         nfs4_lock_state();
3221         status = nfserr_bad_stateid;
3222         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3223                 goto out;
3224         status = nfserr_stale_stateid;
3225         if (STALE_STATEID(stateid))
3226                 goto out;
3227         status = nfserr_bad_stateid;
3228         if (!is_delegation_stateid(stateid))
3229                 goto out;
3230         dp = find_delegation_stateid(inode, stateid);
3231         if (!dp)
3232                 goto out;
3233         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3234         if (status)
3235                 goto out;
3236         renew_client(dp->dl_client);
3237
3238         unhash_delegation(dp);
3239 out:
3240         nfs4_unlock_state();
3241
3242         return status;
3243 }
3244
3245
3246 /*
3247  * Lock owner state (byte-range locks)
3248  */
3249 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3250 #define LOCK_HASH_BITS              8
3251 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3252 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3253
3254 static inline u64
3255 end_offset(u64 start, u64 len)
3256 {
3257         u64 end;
3258
3259         end = start + len;
3260         return end >= start ? end: NFS4_MAX_UINT64;
3261 }
3262
3263 /* last octet in a range */
3264 static inline u64
3265 last_byte_offset(u64 start, u64 len)
3266 {
3267         u64 end;
3268
3269         BUG_ON(!len);
3270         end = start + len;
3271         return end > start ? end - 1: NFS4_MAX_UINT64;
3272 }
3273
3274 #define lockownerid_hashval(id) \
3275         ((id) & LOCK_HASH_MASK)
3276
3277 static inline unsigned int
3278 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3279                 struct xdr_netobj *ownername)
3280 {
3281         return (file_hashval(inode) + cl_id
3282                         + opaque_hashval(ownername->data, ownername->len))
3283                 & LOCK_HASH_MASK;
3284 }
3285
3286 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3287 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3288 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3289
3290 static struct nfs4_stateid *
3291 find_stateid(stateid_t *stid, int flags)
3292 {
3293         struct nfs4_stateid *local;
3294         u32 st_id = stid->si_stateownerid;
3295         u32 f_id = stid->si_fileid;
3296         unsigned int hashval;
3297
3298         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3299         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3300                 hashval = stateid_hashval(st_id, f_id);
3301                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3302                         if ((local->st_stateid.si_stateownerid == st_id) &&
3303                             (local->st_stateid.si_fileid == f_id))
3304                                 return local;
3305                 }
3306         }
3307
3308         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3309                 hashval = stateid_hashval(st_id, f_id);
3310                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3311                         if ((local->st_stateid.si_stateownerid == st_id) &&
3312                             (local->st_stateid.si_fileid == f_id))
3313                                 return local;
3314                 }
3315         }
3316         return NULL;
3317 }
3318
3319 static struct nfs4_delegation *
3320 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3321 {
3322         struct nfs4_file *fp;
3323         struct nfs4_delegation *dl;
3324
3325         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3326                     stid->si_boot, stid->si_stateownerid,
3327                     stid->si_fileid, stid->si_generation);
3328
3329         fp = find_file(ino);
3330         if (!fp)
3331                 return NULL;
3332         dl = find_delegation_file(fp, stid);
3333         put_nfs4_file(fp);
3334         return dl;
3335 }
3336
3337 /*
3338  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3339  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3340  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3341  * locking, this prevents us from being completely protocol-compliant.  The
3342  * real solution to this problem is to start using unsigned file offsets in
3343  * the VFS, but this is a very deep change!
3344  */
3345 static inline void
3346 nfs4_transform_lock_offset(struct file_lock *lock)
3347 {
3348         if (lock->fl_start < 0)
3349                 lock->fl_start = OFFSET_MAX;
3350         if (lock->fl_end < 0)
3351                 lock->fl_end = OFFSET_MAX;
3352 }
3353
3354 /* Hack!: For now, we're defining this just so we can use a pointer to it
3355  * as a unique cookie to identify our (NFSv4's) posix locks. */
3356 static struct lock_manager_operations nfsd_posix_mng_ops  = {
3357 };
3358
3359 static inline void
3360 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3361 {
3362         struct nfs4_stateowner *sop;
3363         unsigned int hval;
3364
3365         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3366                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3367                 hval = lockownerid_hashval(sop->so_id);
3368                 kref_get(&sop->so_ref);
3369                 deny->ld_sop = sop;
3370                 deny->ld_clientid = sop->so_client->cl_clientid;
3371         } else {
3372                 deny->ld_sop = NULL;
3373                 deny->ld_clientid.cl_boot = 0;
3374                 deny->ld_clientid.cl_id = 0;
3375         }
3376         deny->ld_start = fl->fl_start;
3377         deny->ld_length = NFS4_MAX_UINT64;
3378         if (fl->fl_end != NFS4_MAX_UINT64)
3379                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3380         deny->ld_type = NFS4_READ_LT;
3381         if (fl->fl_type != F_RDLCK)
3382                 deny->ld_type = NFS4_WRITE_LT;
3383 }
3384
3385 static struct nfs4_stateowner *
3386 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3387                 struct xdr_netobj *owner)
3388 {
3389         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3390         struct nfs4_stateowner *op;
3391
3392         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3393                 if (same_owner_str(op, owner, clid))
3394                         return op;
3395         }
3396         return NULL;
3397 }
3398
3399 /*
3400  * Alloc a lock owner structure.
3401  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3402  * occured.
3403  *
3404  * strhashval = lock_ownerstr_hashval
3405  */
3406
3407 static struct nfs4_stateowner *
3408 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3409         struct nfs4_stateowner *sop;
3410         struct nfs4_replay *rp;
3411         unsigned int idhashval;
3412
3413         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3414                 return NULL;
3415         idhashval = lockownerid_hashval(current_ownerid);
3416         INIT_LIST_HEAD(&sop->so_idhash);
3417         INIT_LIST_HEAD(&sop->so_strhash);
3418         INIT_LIST_HEAD(&sop->so_perclient);
3419         INIT_LIST_HEAD(&sop->so_stateids);
3420         INIT_LIST_HEAD(&sop->so_perstateid);
3421         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3422         sop->so_time = 0;
3423         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3424         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3425         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3426         sop->so_is_open_owner = 0;
3427         sop->so_id = current_ownerid++;
3428         sop->so_client = clp;
3429         /* It is the openowner seqid that will be incremented in encode in the
3430          * case of new lockowners; so increment the lock seqid manually: */
3431         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3432         sop->so_confirmed = 1;
3433         rp = &sop->so_replay;
3434         rp->rp_status = nfserr_serverfault;
3435         rp->rp_buflen = 0;
3436         rp->rp_buf = rp->rp_ibuf;
3437         return sop;
3438 }
3439
3440 static struct nfs4_stateid *
3441 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3442 {
3443         struct nfs4_stateid *stp;
3444         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3445
3446         stp = nfs4_alloc_stateid();
3447         if (stp == NULL)
3448                 goto out;
3449         INIT_LIST_HEAD(&stp->st_hash);
3450         INIT_LIST_HEAD(&stp->st_perfile);
3451         INIT_LIST_HEAD(&stp->st_perstateowner);
3452         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3453         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3454         list_add(&stp->st_perfile, &fp->fi_stateids);
3455         list_add(&stp->st_perstateowner, &sop->so_stateids);
3456         stp->st_stateowner = sop;
3457         get_nfs4_file(fp);
3458         stp->st_file = fp;
3459         stp->st_stateid.si_boot = boot_time;
3460         stp->st_stateid.si_stateownerid = sop->so_id;
3461         stp->st_stateid.si_fileid = fp->fi_id;
3462         stp->st_stateid.si_generation = 0;
3463         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3464         stp->st_access_bmap = open_stp->st_access_bmap;
3465         stp->st_deny_bmap = open_stp->st_deny_bmap;
3466         stp->st_openstp = open_stp;
3467
3468 out:
3469         return stp;
3470 }
3471
3472 static int
3473 check_lock_length(u64 offset, u64 length)
3474 {
3475         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3476              LOFF_OVERFLOW(offset, length)));
3477 }
3478
3479 /*
3480  *  LOCK operation
3481  */
3482 __be32
3483 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3484            struct nfsd4_lock *lock)
3485 {
3486         struct nfs4_stateowner *open_sop = NULL;
3487         struct nfs4_stateowner *lock_sop = NULL;
3488         struct nfs4_stateid *lock_stp;
3489         struct file *filp;
3490         struct file_lock file_lock;
3491         struct file_lock conflock;
3492         __be32 status = 0;
3493         unsigned int strhashval;
3494         unsigned int cmd;
3495         int err;
3496
3497         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3498                 (long long) lock->lk_offset,
3499                 (long long) lock->lk_length);
3500
3501         if (check_lock_length(lock->lk_offset, lock->lk_length))
3502                  return nfserr_inval;
3503
3504         if ((status = fh_verify(rqstp, &cstate->current_fh,
3505                                 S_IFREG, NFSD_MAY_LOCK))) {
3506                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3507                 return status;
3508         }
3509
3510         nfs4_lock_state();
3511
3512         if (lock->lk_is_new) {
3513                 /*
3514                  * Client indicates that this is a new lockowner.
3515                  * Use open owner and open stateid to create lock owner and
3516                  * lock stateid.
3517                  */
3518                 struct nfs4_stateid *open_stp = NULL;
3519                 struct nfs4_file *fp;
3520
3521                 status = nfserr_stale_clientid;
3522                 if (!nfsd4_has_session(cstate) &&
3523                     STALE_CLIENTID(&lock->lk_new_clientid))
3524                         goto out;
3525
3526                 /* validate and update open stateid and open seqid */
3527                 status = nfs4_preprocess_seqid_op(cstate,
3528                                         lock->lk_new_open_seqid,
3529                                         &lock->lk_new_open_stateid,
3530                                         OPEN_STATE,
3531                                         &lock->lk_replay_owner, &open_stp,
3532                                         lock);
3533                 if (status)
3534                         goto out;
3535                 open_sop = lock->lk_replay_owner;
3536                 /* create lockowner and lock stateid */
3537                 fp = open_stp->st_file;
3538                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3539                                 open_sop->so_client->cl_clientid.cl_id,
3540                                 &lock->v.new.owner);
3541                 /* XXX: Do we need to check for duplicate stateowners on
3542                  * the same file, or should they just be allowed (and
3543                  * create new stateids)? */
3544                 status = nfserr_resource;
3545                 lock_sop = alloc_init_lock_stateowner(strhashval,
3546                                 open_sop->so_client, open_stp, lock);
3547                 if (lock_sop == NULL)
3548                         goto out;
3549                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3550                 if (lock_stp == NULL)
3551                         goto out;
3552         } else {
3553                 /* lock (lock owner + lock stateid) already exists */
3554                 status = nfs4_preprocess_seqid_op(cstate,
3555                                        lock->lk_old_lock_seqid,
3556                                        &lock->lk_old_lock_stateid,
3557                                        LOCK_STATE,
3558                                        &lock->lk_replay_owner, &lock_stp, lock);
3559                 if (status)
3560                         goto out;
3561                 lock_sop = lock->lk_replay_owner;
3562         }
3563         /* lock->lk_replay_owner and lock_stp have been created or found */
3564         filp = lock_stp->st_vfs_file;
3565
3566         status = nfserr_grace;
3567         if (locks_in_grace() && !lock->lk_reclaim)
3568                 goto out;
3569         status = nfserr_no_grace;
3570         if (!locks_in_grace() && lock->lk_reclaim)
3571                 goto out;
3572
3573         locks_init_lock(&file_lock);
3574         switch (lock->lk_type) {
3575                 case NFS4_READ_LT:
3576                 case NFS4_READW_LT:
3577                         file_lock.fl_type = F_RDLCK;
3578                         cmd = F_SETLK;
3579                 break;
3580                 case NFS4_WRITE_LT:
3581                 case NFS4_WRITEW_LT:
3582                         file_lock.fl_type = F_WRLCK;
3583                         cmd = F_SETLK;
3584                 break;
3585                 default:
3586                         status = nfserr_inval;
3587                 goto out;
3588         }
3589         file_lock.fl_owner = (fl_owner_t)lock_sop;
3590         file_lock.fl_pid = current->tgid;
3591         file_lock.fl_file = filp;
3592         file_lock.fl_flags = FL_POSIX;
3593         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3594
3595         file_lock.fl_start = lock->lk_offset;
3596         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3597         nfs4_transform_lock_offset(&file_lock);
3598
3599         /*
3600         * Try to lock the file in the VFS.
3601         * Note: locks.c uses the BKL to protect the inode's lock list.
3602         */
3603
3604         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3605         switch (-err) {
3606         case 0: /* success! */
3607                 update_stateid(&lock_stp->st_stateid);
3608                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3609                                 sizeof(stateid_t));
3610                 status = 0;
3611                 break;
3612         case (EAGAIN):          /* conflock holds conflicting lock */
3613                 status = nfserr_denied;
3614                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3615                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3616                 break;
3617         case (EDEADLK):
3618                 status = nfserr_deadlock;
3619                 break;
3620         default:
3621                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3622                 status = nfserr_resource;
3623                 break;
3624         }
3625 out:
3626         if (status && lock->lk_is_new && lock_sop)
3627                 release_lockowner(lock_sop);
3628         if (lock->lk_replay_owner) {
3629                 nfs4_get_stateowner(lock->lk_replay_owner);
3630                 cstate->replay_owner = lock->lk_replay_owner;
3631         }
3632         nfs4_unlock_state();
3633         return status;
3634 }
3635
3636 /*
3637  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3638  * so we do a temporary open here just to get an open file to pass to
3639  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3640  * inode operation.)
3641  */
3642 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3643 {
3644         struct file *file;
3645         int err;
3646
3647         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3648         if (err)
3649                 return err;
3650         err = vfs_test_lock(file, lock);
3651         nfsd_close(file);
3652         return err;
3653 }
3654
3655 /*
3656  * LOCKT operation
3657  */
3658 __be32
3659 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3660             struct nfsd4_lockt *lockt)
3661 {
3662         struct inode *inode;
3663         struct file_lock file_lock;
3664         int error;
3665         __be32 status;
3666
3667         if (locks_in_grace())
3668                 return nfserr_grace;
3669
3670         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3671                  return nfserr_inval;
3672
3673         lockt->lt_stateowner = NULL;
3674         nfs4_lock_state();
3675
3676         status = nfserr_stale_clientid;
3677         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3678                 goto out;
3679
3680         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3681                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3682                 if (status == nfserr_symlink)
3683                         status = nfserr_inval;
3684                 goto out;
3685         }
3686
3687         inode = cstate->current_fh.fh_dentry->d_inode;
3688         locks_init_lock(&file_lock);
3689         switch (lockt->lt_type) {
3690                 case NFS4_READ_LT:
3691                 case NFS4_READW_LT:
3692                         file_lock.fl_type = F_RDLCK;
3693                 break;
3694                 case NFS4_WRITE_LT:
3695                 case NFS4_WRITEW_LT:
3696                         file_lock.fl_type = F_WRLCK;
3697                 break;
3698                 default:
3699                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3700                         status = nfserr_inval;
3701                 goto out;
3702         }
3703
3704         lockt->lt_stateowner = find_lockstateowner_str(inode,
3705                         &lockt->lt_clientid, &lockt->lt_owner);
3706         if (lockt->lt_stateowner)
3707                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3708         file_lock.fl_pid = current->tgid;
3709         file_lock.fl_flags = FL_POSIX;
3710
3711         file_lock.fl_start = lockt->lt_offset;
3712         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3713
3714         nfs4_transform_lock_offset(&file_lock);
3715
3716         status = nfs_ok;
3717         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3718         if (error) {
3719                 status = nfserrno(error);
3720                 goto out;
3721         }
3722         if (file_lock.fl_type != F_UNLCK) {
3723                 status = nfserr_denied;
3724                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3725         }
3726 out:
3727         nfs4_unlock_state();
3728         return status;
3729 }
3730
3731 __be32
3732 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3733             struct nfsd4_locku *locku)
3734 {
3735         struct nfs4_stateid *stp;
3736         struct file *filp = NULL;
3737         struct file_lock file_lock;
3738         __be32 status;
3739         int err;
3740
3741         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3742                 (long long) locku->lu_offset,
3743                 (long long) locku->lu_length);
3744
3745         if (check_lock_length(locku->lu_offset, locku->lu_length))
3746                  return nfserr_inval;
3747
3748         nfs4_lock_state();
3749
3750         if ((status = nfs4_preprocess_seqid_op(cstate,
3751                                         locku->lu_seqid,
3752                                         &locku->lu_stateid,
3753                                         LOCK_STATE,
3754                                         &locku->lu_stateowner, &stp, NULL)))
3755                 goto out;
3756
3757         filp = stp->st_vfs_file;
3758         BUG_ON(!filp);
3759         locks_init_lock(&file_lock);
3760         file_lock.fl_type = F_UNLCK;
3761         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3762         file_lock.fl_pid = current->tgid;
3763         file_lock.fl_file = filp;
3764         file_lock.fl_flags = FL_POSIX;
3765         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3766         file_lock.fl_start = locku->lu_offset;
3767
3768         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3769         nfs4_transform_lock_offset(&file_lock);
3770
3771         /*
3772         *  Try to unlock the file in the VFS.
3773         */
3774         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3775         if (err) {
3776                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3777                 goto out_nfserr;
3778         }
3779         /*
3780         * OK, unlock succeeded; the only thing left to do is update the stateid.
3781         */
3782         update_stateid(&stp->st_stateid);
3783         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3784
3785 out:
3786         if (locku->lu_stateowner) {
3787                 nfs4_get_stateowner(locku->lu_stateowner);
3788                 cstate->replay_owner = locku->lu_stateowner;
3789         }
3790         nfs4_unlock_state();
3791         return status;
3792
3793 out_nfserr:
3794         status = nfserrno(err);
3795         goto out;
3796 }
3797
3798 /*
3799  * returns
3800  *      1: locks held by lockowner
3801  *      0: no locks held by lockowner
3802  */
3803 static int
3804 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3805 {
3806         struct file_lock **flpp;
3807         struct inode *inode = filp->f_path.dentry->d_inode;
3808         int status = 0;
3809
3810         lock_kernel();
3811         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3812                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3813                         status = 1;
3814                         goto out;
3815                 }
3816         }
3817 out:
3818         unlock_kernel();
3819         return status;
3820 }
3821
3822 __be32
3823 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3824                         struct nfsd4_compound_state *cstate,
3825                         struct nfsd4_release_lockowner *rlockowner)
3826 {
3827         clientid_t *clid = &rlockowner->rl_clientid;
3828         struct nfs4_stateowner *sop;
3829         struct nfs4_stateid *stp;
3830         struct xdr_netobj *owner = &rlockowner->rl_owner;
3831         struct list_head matches;
3832         int i;
3833         __be32 status;
3834
3835         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3836                 clid->cl_boot, clid->cl_id);
3837
3838         /* XXX check for lease expiration */
3839
3840         status = nfserr_stale_clientid;
3841         if (STALE_CLIENTID(clid))
3842                 return status;
3843
3844         nfs4_lock_state();
3845
3846         status = nfserr_locks_held;
3847         /* XXX: we're doing a linear search through all the lockowners.
3848          * Yipes!  For now we'll just hope clients aren't really using
3849          * release_lockowner much, but eventually we have to fix these
3850          * data structures. */
3851         INIT_LIST_HEAD(&matches);
3852         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3853                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3854                         if (!same_owner_str(sop, owner, clid))
3855                                 continue;
3856                         list_for_each_entry(stp, &sop->so_stateids,
3857                                         st_perstateowner) {
3858                                 if (check_for_locks(stp->st_vfs_file, sop))
3859                                         goto out;
3860                                 /* Note: so_perclient unused for lockowners,
3861                                  * so it's OK to fool with here. */
3862                                 list_add(&sop->so_perclient, &matches);
3863                         }
3864                 }
3865         }
3866         /* Clients probably won't expect us to return with some (but not all)
3867          * of the lockowner state released; so don't release any until all
3868          * have been checked. */
3869         status = nfs_ok;
3870         while (!list_empty(&matches)) {
3871                 sop = list_entry(matches.next, struct nfs4_stateowner,
3872                                                                 so_perclient);
3873                 /* unhash_stateowner deletes so_perclient only
3874                  * for openowners. */
3875                 list_del(&sop->so_perclient);
3876                 release_lockowner(sop);
3877         }
3878 out:
3879         nfs4_unlock_state();
3880         return status;
3881 }
3882
3883 static inline struct nfs4_client_reclaim *
3884 alloc_reclaim(void)
3885 {
3886         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3887 }
3888
3889 int
3890 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3891 {
3892         unsigned int strhashval = clientstr_hashval(name);
3893         struct nfs4_client *clp;
3894
3895         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3896         return clp ? 1 : 0;
3897 }
3898
3899 /*
3900  * failure => all reset bets are off, nfserr_no_grace...
3901  */
3902 int
3903 nfs4_client_to_reclaim(const char *name)
3904 {
3905         unsigned int strhashval;
3906         struct nfs4_client_reclaim *crp = NULL;
3907
3908         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3909         crp = alloc_reclaim();
3910         if (!crp)
3911                 return 0;
3912         strhashval = clientstr_hashval(name);
3913         INIT_LIST_HEAD(&crp->cr_strhash);
3914         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3915         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3916         reclaim_str_hashtbl_size++;
3917         return 1;
3918 }
3919
3920 static void
3921 nfs4_release_reclaim(void)
3922 {
3923         struct nfs4_client_reclaim *crp = NULL;
3924         int i;
3925
3926         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3927                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3928                         crp = list_entry(reclaim_str_hashtbl[i].next,
3929                                         struct nfs4_client_reclaim, cr_strhash);
3930                         list_del(&crp->cr_strhash);
3931                         kfree(crp);
3932                         reclaim_str_hashtbl_size--;
3933                 }
3934         }
3935         BUG_ON(reclaim_str_hashtbl_size);
3936 }
3937
3938 /*
3939  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3940 static struct nfs4_client_reclaim *
3941 nfs4_find_reclaim_client(clientid_t *clid)
3942 {
3943         unsigned int strhashval;
3944         struct nfs4_client *clp;
3945         struct nfs4_client_reclaim *crp = NULL;
3946
3947
3948         /* find clientid in conf_id_hashtbl */
3949         clp = find_confirmed_client(clid);
3950         if (clp == NULL)
3951                 return NULL;
3952
3953         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3954                             clp->cl_name.len, clp->cl_name.data,
3955                             clp->cl_recdir);
3956
3957         /* find clp->cl_name in reclaim_str_hashtbl */
3958         strhashval = clientstr_hashval(clp->cl_recdir);
3959         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3960                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3961                         return crp;
3962                 }
3963         }
3964         return NULL;
3965 }
3966
3967 /*
3968 * Called from OPEN. Look for clientid in reclaim list.
3969 */
3970 __be32
3971 nfs4_check_open_reclaim(clientid_t *clid)
3972 {
3973         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3974 }
3975
3976 /* initialization to perform at module load time: */
3977
3978 int
3979 nfs4_state_init(void)
3980 {
3981         int i, status;
3982
3983         status = nfsd4_init_slabs();
3984         if (status)
3985                 return status;
3986         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3987                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3988                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3989                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3990                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3991         }
3992         for (i = 0; i < SESSION_HASH_SIZE; i++)
3993                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3994         for (i = 0; i < FILE_HASH_SIZE; i++) {
3995                 INIT_LIST_HEAD(&file_hashtbl[i]);
3996         }
3997         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3998                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3999                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4000         }
4001         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4002                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4003                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4004         }
4005         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4006                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4007                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4008         }
4009         memset(&onestateid, ~0, sizeof(stateid_t));
4010         INIT_LIST_HEAD(&close_lru);
4011         INIT_LIST_HEAD(&client_lru);
4012         INIT_LIST_HEAD(&del_recall_lru);
4013         for (i = 0; i < CLIENT_HASH_SIZE; i++)
4014                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4015         reclaim_str_hashtbl_size = 0;
4016         return 0;
4017 }
4018
4019 static void
4020 nfsd4_load_reboot_recovery_data(void)
4021 {
4022         int status;
4023
4024         nfs4_lock_state();
4025         nfsd4_init_recdir(user_recovery_dirname);
4026         status = nfsd4_recdir_load();
4027         nfs4_unlock_state();
4028         if (status)
4029                 printk("NFSD: Failure reading reboot recovery data\n");
4030 }
4031
4032 unsigned long
4033 get_nfs4_grace_period(void)
4034 {
4035         return max(user_lease_time, lease_time) * HZ;
4036 }
4037
4038 /*
4039  * Since the lifetime of a delegation isn't limited to that of an open, a
4040  * client may quite reasonably hang on to a delegation as long as it has
4041  * the inode cached.  This becomes an obvious problem the first time a
4042  * client's inode cache approaches the size of the server's total memory.
4043  *
4044  * For now we avoid this problem by imposing a hard limit on the number
4045  * of delegations, which varies according to the server's memory size.
4046  */
4047 static void
4048 set_max_delegations(void)
4049 {
4050         /*
4051          * Allow at most 4 delegations per megabyte of RAM.  Quick
4052          * estimates suggest that in the worst case (where every delegation
4053          * is for a different inode), a delegation could take about 1.5K,
4054          * giving a worst case usage of about 6% of memory.
4055          */
4056         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4057 }
4058
4059 /* initialization to perform when the nfsd service is started: */
4060
4061 static void
4062 __nfs4_state_start(void)
4063 {
4064         unsigned long grace_time;
4065
4066         boot_time = get_seconds();
4067         grace_time = get_nfs4_grace_period();
4068         lease_time = user_lease_time;
4069         locks_start_grace(&nfsd4_manager);
4070         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4071                grace_time/HZ);
4072         laundry_wq = create_singlethread_workqueue("nfsd4");
4073         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4074         set_max_delegations();
4075 }
4076
4077 void
4078 nfs4_state_start(void)
4079 {
4080         if (nfs4_init)
4081                 return;
4082         nfsd4_load_reboot_recovery_data();
4083         __nfs4_state_start();
4084         nfs4_init = 1;
4085         return;
4086 }
4087
4088 time_t
4089 nfs4_lease_time(void)
4090 {
4091         return lease_time;
4092 }
4093
4094 static void
4095 __nfs4_state_shutdown(void)
4096 {
4097         int i;
4098         struct nfs4_client *clp = NULL;
4099         struct nfs4_delegation *dp = NULL;
4100         struct list_head *pos, *next, reaplist;
4101
4102         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4103                 while (!list_empty(&conf_id_hashtbl[i])) {
4104                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4105                         expire_client(clp);
4106                 }
4107                 while (!list_empty(&unconf_str_hashtbl[i])) {
4108                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4109                         expire_client(clp);
4110                 }
4111         }
4112         INIT_LIST_HEAD(&reaplist);
4113         spin_lock(&recall_lock);
4114         list_for_each_safe(pos, next, &del_recall_lru) {
4115                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4116                 list_move(&dp->dl_recall_lru, &reaplist);
4117         }
4118         spin_unlock(&recall_lock);
4119         list_for_each_safe(pos, next, &reaplist) {
4120                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4121                 list_del_init(&dp->dl_recall_lru);
4122                 unhash_delegation(dp);
4123         }
4124
4125         nfsd4_shutdown_recdir();
4126         nfs4_init = 0;
4127 }
4128
4129 void
4130 nfs4_state_shutdown(void)
4131 {
4132         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4133         destroy_workqueue(laundry_wq);
4134         locks_end_grace(&nfsd4_manager);
4135         nfs4_lock_state();
4136         nfs4_release_reclaim();
4137         __nfs4_state_shutdown();
4138         nfs4_unlock_state();
4139 }
4140
4141 /*
4142  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4143  * accessed when nfsd is starting.
4144  */
4145 static void
4146 nfs4_set_recdir(char *recdir)
4147 {
4148         strcpy(user_recovery_dirname, recdir);
4149 }
4150
4151 /*
4152  * Change the NFSv4 recovery directory to recdir.
4153  */
4154 int
4155 nfs4_reset_recoverydir(char *recdir)
4156 {
4157         int status;
4158         struct path path;
4159
4160         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4161         if (status)
4162                 return status;
4163         status = -ENOTDIR;
4164         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4165                 nfs4_set_recdir(recdir);
4166                 status = 0;
4167         }
4168         path_put(&path);
4169         return status;
4170 }
4171
4172 char *
4173 nfs4_recoverydir(void)
4174 {
4175         return user_recovery_dirname;
4176 }
4177
4178 /*
4179  * Called when leasetime is changed.
4180  *
4181  * The only way the protocol gives us to handle on-the-fly lease changes is to
4182  * simulate a reboot.  Instead of doing that, we just wait till the next time
4183  * we start to register any changes in lease time.  If the administrator
4184  * really wants to change the lease time *now*, they can go ahead and bring
4185  * nfsd down and then back up again after changing the lease time.
4186  *
4187  * user_lease_time is protected by nfsd_mutex since it's only really accessed
4188  * when nfsd is starting
4189  */
4190 void
4191 nfs4_reset_lease(time_t leasetime)
4192 {
4193         user_lease_time = leasetime;
4194 }