fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/smp_lock.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/sunrpc/svcauth_gss.h>
  41 #include <linux/sunrpc/clnt.h>
  42 #include "xdr4.h"
  43 #include "vfs.h"
  44
  45 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  46
  47 /* Globals */
  48 time_t nfsd4_lease = 90;     /* default lease time */
  49 time_t nfsd4_grace = 90;
  50 static time_t boot_time;
  51 static u32 current_ownerid = 1;
  52 static u32 current_fileid = 1;
  53 static u32 current_delegid = 1;
  54 static stateid_t zerostateid;             /* bits all 0 */
  55 static stateid_t onestateid;              /* bits all 1 */
  56 static u64 current_sessionid = 1;
  57
  58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
  59 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
  60
  61 /* forward declarations */
  62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
  63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
  64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  65 static void nfs4_set_recdir(char *recdir);
  66
  67 /* Locking: */
  68
  69 /* Currently used for almost all code touching nfsv4 state: */
  70 static DEFINE_MUTEX(client_mutex);
  71
  72 /*
  73  * Currently used for the del_recall_lru and file hash table.  In an
  74  * effort to decrease the scope of the client_mutex, this spinlock may
  75  * eventually cover more:
  76  */
  77 static DEFINE_SPINLOCK(recall_lock);
  78
  79 static struct kmem_cache *stateowner_slab = NULL;
  80 static struct kmem_cache *file_slab = NULL;
  81 static struct kmem_cache *stateid_slab = NULL;
  82 static struct kmem_cache *deleg_slab = NULL;
  83
  84 void
  85 nfs4_lock_state(void)
  86 {
  87         mutex_lock(&client_mutex);
  88 }
  89
  90 void
  91 nfs4_unlock_state(void)
  92 {
  93         mutex_unlock(&client_mutex);
  94 }
  95
  96 static inline u32
  97 opaque_hashval(const void *ptr, int nbytes)
  98 {
  99         unsigned char *cptr = (unsigned char *) ptr;
 100
 101         u32 x = 0;
 102         while (nbytes--) {
 103                 x *= 37;
 104                 x += *cptr++;
 105         }
 106         return x;
 107 }
 108
 109 static struct list_head del_recall_lru;
 110
 111 static inline void
 112 put_nfs4_file(struct nfs4_file *fi)
 113 {
 114         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 115                 list_del(&fi->fi_hash);
 116                 spin_unlock(&recall_lock);
 117                 iput(fi->fi_inode);
 118                 kmem_cache_free(file_slab, fi);
 119         }
 120 }
 121
 122 static inline void
 123 get_nfs4_file(struct nfs4_file *fi)
 124 {
 125         atomic_inc(&fi->fi_ref);
 126 }
 127
 128 static int num_delegations;
 129 unsigned int max_delegations;
 130
 131 /*
 132  * Open owner state (share locks)
 133  */
 134
 135 /* hash tables for nfs4_stateowner */
 136 #define OWNER_HASH_BITS              8
 137 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 138 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 139
 140 #define ownerid_hashval(id) \
 141         ((id) & OWNER_HASH_MASK)
 142 #define ownerstr_hashval(clientid, ownername) \
 143         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
 144
 145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
 146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 147
 148 /* hash table for nfs4_file */
 149 #define FILE_HASH_BITS                   8
 150 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 151 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
 152 /* hash table for (open)nfs4_stateid */
 153 #define STATEID_HASH_BITS              10
 154 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
 155 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
 156
 157 #define file_hashval(x) \
 158         hash_ptr(x, FILE_HASH_BITS)
 159 #define stateid_hashval(owner_id, file_id)  \
 160         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
 161
 162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
 164
 165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 166 {
 167         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 168         atomic_inc(&fp->fi_access[oflag]);
 169 }
 170
 171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 172 {
 173         if (oflag == O_RDWR) {
 174                 __nfs4_file_get_access(fp, O_RDONLY);
 175                 __nfs4_file_get_access(fp, O_WRONLY);
 176         } else
 177                 __nfs4_file_get_access(fp, oflag);
 178 }
 179
 180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 181 {
 182         if (fp->fi_fds[oflag]) {
 183                 fput(fp->fi_fds[oflag]);
 184                 fp->fi_fds[oflag] = NULL;
 185         }
 186 }
 187
 188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 189 {
 190         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 191                 nfs4_file_put_fd(fp, O_RDWR);
 192                 nfs4_file_put_fd(fp, oflag);
 193         }
 194 }
 195
 196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 197 {
 198         if (oflag == O_RDWR) {
 199                 __nfs4_file_put_access(fp, O_RDONLY);
 200                 __nfs4_file_put_access(fp, O_WRONLY);
 201         } else
 202                 __nfs4_file_put_access(fp, oflag);
 203 }
 204
 205 static struct nfs4_delegation *
 206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
 207 {
 208         struct nfs4_delegation *dp;
 209         struct nfs4_file *fp = stp->st_file;
 210
 211         dprintk("NFSD alloc_init_deleg\n");
 212         /*
 213          * Major work on the lease subsystem (for example, to support
 214          * calbacks on stat) will be required before we can support
 215          * write delegations properly.
 216          */
 217         if (type != NFS4_OPEN_DELEGATE_READ)
 218                 return NULL;
 219         if (fp->fi_had_conflict)
 220                 return NULL;
 221         if (num_delegations > max_delegations)
 222                 return NULL;
 223         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
 224         if (dp == NULL)
 225                 return dp;
 226         num_delegations++;
 227         INIT_LIST_HEAD(&dp->dl_perfile);
 228         INIT_LIST_HEAD(&dp->dl_perclnt);
 229         INIT_LIST_HEAD(&dp->dl_recall_lru);
 230         dp->dl_client = clp;
 231         get_nfs4_file(fp);
 232         dp->dl_file = fp;
 233         nfs4_file_get_access(fp, O_RDONLY);
 234         dp->dl_flock = NULL;
 235         dp->dl_type = type;
 236         dp->dl_stateid.si_boot = boot_time;
 237         dp->dl_stateid.si_stateownerid = current_delegid++;
 238         dp->dl_stateid.si_fileid = 0;
 239         dp->dl_stateid.si_generation = 0;
 240         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 241         dp->dl_time = 0;
 242         atomic_set(&dp->dl_count, 1);
 243         list_add(&dp->dl_perfile, &fp->fi_delegations);
 244         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 245         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
 246         return dp;
 247 }
 248
 249 void
 250 nfs4_put_delegation(struct nfs4_delegation *dp)
 251 {
 252         if (atomic_dec_and_test(&dp->dl_count)) {
 253                 dprintk("NFSD: freeing dp %p\n",dp);
 254                 put_nfs4_file(dp->dl_file);
 255                 kmem_cache_free(deleg_slab, dp);
 256                 num_delegations--;
 257         }
 258 }
 259
 260 /* Remove the associated file_lock first, then remove the delegation.
 261  * lease_modify() is called to remove the FS_LEASE file_lock from
 262  * the i_flock list, eventually calling nfsd's lock_manager
 263  * fl_release_callback.
 264  */
 265 static void
 266 nfs4_close_delegation(struct nfs4_delegation *dp)
 267 {
 268         struct file *filp = find_readable_file(dp->dl_file);
 269
 270         dprintk("NFSD: close_delegation dp %p\n",dp);
 271         if (dp->dl_flock)
 272                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
 273         nfs4_file_put_access(dp->dl_file, O_RDONLY);
 274 }
 275
 276 /* Called under the state lock. */
 277 static void
 278 unhash_delegation(struct nfs4_delegation *dp)
 279 {
 280         list_del_init(&dp->dl_perfile);
 281         list_del_init(&dp->dl_perclnt);
 282         spin_lock(&recall_lock);
 283         list_del_init(&dp->dl_recall_lru);
 284         spin_unlock(&recall_lock);
 285         nfs4_close_delegation(dp);
 286         nfs4_put_delegation(dp);
 287 }
 288
 289 /*
 290  * SETCLIENTID state
 291  */
 292
 293 /* client_lock protects the client lru list and session hash table */
 294 static DEFINE_SPINLOCK(client_lock);
 295
 296 /* Hash tables for nfs4_clientid state */
 297 #define CLIENT_HASH_BITS                 4
 298 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 299 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 300
 301 #define clientid_hashval(id) \
 302         ((id) & CLIENT_HASH_MASK)
 303 #define clientstr_hashval(name) \
 304         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
 305 /*
 306  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 307  * used in reboot/reset lease grace period processing
 308  *
 309  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 310  * setclientid_confirmed info.
 311  *
 312  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 313  * setclientid info.
 314  *
 315  * client_lru holds client queue ordered by nfs4_client.cl_time
 316  * for lease renewal.
 317  *
 318  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 319  * for last close replay.
 320  */
 321 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 322 static int reclaim_str_hashtbl_size = 0;
 323 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 324 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 325 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 326 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 327 static struct list_head client_lru;
 328 static struct list_head close_lru;
 329
 330 static void unhash_generic_stateid(struct nfs4_stateid *stp)
 331 {
 332         list_del(&stp->st_hash);
 333         list_del(&stp->st_perfile);
 334         list_del(&stp->st_perstateowner);
 335 }
 336
 337 static void free_generic_stateid(struct nfs4_stateid *stp)
 338 {
 339         put_nfs4_file(stp->st_file);
 340         kmem_cache_free(stateid_slab, stp);
 341 }
 342
 343 static void release_lock_stateid(struct nfs4_stateid *stp)
 344 {
 345         struct file *file;
 346
 347         unhash_generic_stateid(stp);
 348         file = find_any_file(stp->st_file);
 349         if (file)
 350                 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
 351         free_generic_stateid(stp);
 352 }
 353
 354 static void unhash_lockowner(struct nfs4_stateowner *sop)
 355 {
 356         struct nfs4_stateid *stp;
 357
 358         list_del(&sop->so_idhash);
 359         list_del(&sop->so_strhash);
 360         list_del(&sop->so_perstateid);
 361         while (!list_empty(&sop->so_stateids)) {
 362                 stp = list_first_entry(&sop->so_stateids,
 363                                 struct nfs4_stateid, st_perstateowner);
 364                 release_lock_stateid(stp);
 365         }
 366 }
 367
 368 static void release_lockowner(struct nfs4_stateowner *sop)
 369 {
 370         unhash_lockowner(sop);
 371         nfs4_put_stateowner(sop);
 372 }
 373
 374 static void
 375 release_stateid_lockowners(struct nfs4_stateid *open_stp)
 376 {
 377         struct nfs4_stateowner *lock_sop;
 378
 379         while (!list_empty(&open_stp->st_lockowners)) {
 380                 lock_sop = list_entry(open_stp->st_lockowners.next,
 381                                 struct nfs4_stateowner, so_perstateid);
 382                 /* list_del(&open_stp->st_lockowners);  */
 383                 BUG_ON(lock_sop->so_is_open_owner);
 384                 release_lockowner(lock_sop);
 385         }
 386 }
 387
 388 /*
 389  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 390  * st_{access,deny}_bmap field of the stateid, in order to track not
 391  * only what share bits are currently in force, but also what
 392  * combinations of share bits previous opens have used.  This allows us
 393  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 394  * return an error if the client attempt to downgrade to a combination
 395  * of share bits not explicable by closing some of its previous opens.
 396  *
 397  * XXX: This enforcement is actually incomplete, since we don't keep
 398  * track of access/deny bit combinations; so, e.g., we allow:
 399  *
 400  *      OPEN allow read, deny write
 401  *      OPEN allow both, deny none
 402  *      DOWNGRADE allow read, deny none
 403  *
 404  * which we should reject.
 405  */
 406 static void
 407 set_access(unsigned int *access, unsigned long bmap) {
 408         int i;
 409
 410         *access = 0;
 411         for (i = 1; i < 4; i++) {
 412                 if (test_bit(i, &bmap))
 413                         *access |= i;
 414         }
 415 }
 416
 417 static void
 418 set_deny(unsigned int *deny, unsigned long bmap) {
 419         int i;
 420
 421         *deny = 0;
 422         for (i = 0; i < 4; i++) {
 423                 if (test_bit(i, &bmap))
 424                         *deny |= i ;
 425         }
 426 }
 427
 428 static int
 429 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
 430         unsigned int access, deny;
 431
 432         set_access(&access, stp->st_access_bmap);
 433         set_deny(&deny, stp->st_deny_bmap);
 434         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 435                 return 0;
 436         return 1;
 437 }
 438
 439 static int nfs4_access_to_omode(u32 access)
 440 {
 441         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 442         case NFS4_SHARE_ACCESS_READ:
 443                 return O_RDONLY;
 444         case NFS4_SHARE_ACCESS_WRITE:
 445                 return O_WRONLY;
 446         case NFS4_SHARE_ACCESS_BOTH:
 447                 return O_RDWR;
 448         }
 449         BUG();
 450 }
 451
 452 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
 453 {
 454         unsigned int access;
 455
 456         set_access(&access, stp->st_access_bmap);
 457         return nfs4_access_to_omode(access);
 458 }
 459
 460 static void release_open_stateid(struct nfs4_stateid *stp)
 461 {
 462         int oflag = nfs4_access_bmap_to_omode(stp);
 463
 464         unhash_generic_stateid(stp);
 465         release_stateid_lockowners(stp);
 466         nfs4_file_put_access(stp->st_file, oflag);
 467         free_generic_stateid(stp);
 468 }
 469
 470 static void unhash_openowner(struct nfs4_stateowner *sop)
 471 {
 472         struct nfs4_stateid *stp;
 473
 474         list_del(&sop->so_idhash);
 475         list_del(&sop->so_strhash);
 476         list_del(&sop->so_perclient);
 477         list_del(&sop->so_perstateid); /* XXX: necessary? */
 478         while (!list_empty(&sop->so_stateids)) {
 479                 stp = list_first_entry(&sop->so_stateids,
 480                                 struct nfs4_stateid, st_perstateowner);
 481                 release_open_stateid(stp);
 482         }
 483 }
 484
 485 static void release_openowner(struct nfs4_stateowner *sop)
 486 {
 487         unhash_openowner(sop);
 488         list_del(&sop->so_close_lru);
 489         nfs4_put_stateowner(sop);
 490 }
 491
 492 #define SESSION_HASH_SIZE       512
 493 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
 494
 495 static inline int
 496 hash_sessionid(struct nfs4_sessionid *sessionid)
 497 {
 498         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 499
 500         return sid->sequence % SESSION_HASH_SIZE;
 501 }
 502
 503 static inline void
 504 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 505 {
 506         u32 *ptr = (u32 *)(&sessionid->data[0]);
 507         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 508 }
 509
 510 static void
 511 gen_sessionid(struct nfsd4_session *ses)
 512 {
 513         struct nfs4_client *clp = ses->se_client;
 514         struct nfsd4_sessionid *sid;
 515
 516         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 517         sid->clientid = clp->cl_clientid;
 518         sid->sequence = current_sessionid++;
 519         sid->reserved = 0;
 520 }
 521
 522 /*
 523  * The protocol defines ca_maxresponssize_cached to include the size of
 524  * the rpc header, but all we need to cache is the data starting after
 525  * the end of the initial SEQUENCE operation--the rest we regenerate
 526  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 527  * value that is the number of bytes in our cache plus a few additional
 528  * bytes.  In order to stay on the safe side, and not promise more than
 529  * we can cache, those additional bytes must be the minimum possible: 24
 530  * bytes of rpc header (xid through accept state, with AUTH_NULL
 531  * verifier), 12 for the compound header (with zero-length tag), and 44
 532  * for the SEQUENCE op response:
 533  */
 534 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 535
 536 static void
 537 free_session_slots(struct nfsd4_session *ses)
 538 {
 539         int i;
 540
 541         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 542                 kfree(ses->se_slots[i]);
 543 }
 544
 545 /*
 546  * We don't actually need to cache the rpc and session headers, so we
 547  * can allocate a little less for each slot:
 548  */
 549 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
 550 {
 551         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 552 }
 553
 554 static int nfsd4_sanitize_slot_size(u32 size)
 555 {
 556         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
 557         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
 558
 559         return size;
 560 }
 561
 562 /*
 563  * XXX: If we run out of reserved DRC memory we could (up to a point)
 564  * re-negotiate active sessions and reduce their slot usage to make
 565  * rooom for new connections. For now we just fail the create session.
 566  */
 567 static int nfsd4_get_drc_mem(int slotsize, u32 num)
 568 {
 569         int avail;
 570
 571         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
 572
 573         spin_lock(&nfsd_drc_lock);
 574         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
 575                         nfsd_drc_max_mem - nfsd_drc_mem_used);
 576         num = min_t(int, num, avail / slotsize);
 577         nfsd_drc_mem_used += num * slotsize;
 578         spin_unlock(&nfsd_drc_lock);
 579
 580         return num;
 581 }
 582
 583 static void nfsd4_put_drc_mem(int slotsize, int num)
 584 {
 585         spin_lock(&nfsd_drc_lock);
 586         nfsd_drc_mem_used -= slotsize * num;
 587         spin_unlock(&nfsd_drc_lock);
 588 }
 589
 590 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
 591 {
 592         struct nfsd4_session *new;
 593         int mem, i;
 594
 595         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 596                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 597         mem = numslots * sizeof(struct nfsd4_slot *);
 598
 599         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 600         if (!new)
 601                 return NULL;
 602         /* allocate each struct nfsd4_slot and data cache in one piece */
 603         for (i = 0; i < numslots; i++) {
 604                 mem = sizeof(struct nfsd4_slot) + slotsize;
 605                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
 606                 if (!new->se_slots[i])
 607                         goto out_free;
 608         }
 609         return new;
 610 out_free:
 611         while (i--)
 612                 kfree(new->se_slots[i]);
 613         kfree(new);
 614         return NULL;
 615 }
 616
 617 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
 618 {
 619         u32 maxrpc = nfsd_serv->sv_max_mesg;
 620
 621         new->maxreqs = numslots;
 622         new->maxresp_cached = slotsize + NFSD_MIN_HDR_SEQ_SZ;
 623         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
 624         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
 625         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
 626 }
 627
 628 static void free_conn(struct nfsd4_conn *c)
 629 {
 630         svc_xprt_put(c->cn_xprt);
 631         kfree(c);
 632 }
 633
 634 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 635 {
 636         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 637         struct nfs4_client *clp = c->cn_session->se_client;
 638
 639         spin_lock(&clp->cl_lock);
 640         if (!list_empty(&c->cn_persession)) {
 641                 list_del(&c->cn_persession);
 642                 free_conn(c);
 643         }
 644         spin_unlock(&clp->cl_lock);
 645 }
 646
 647 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp)
 648 {
 649         struct nfsd4_conn *conn;
 650
 651         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 652         if (!conn)
 653                 return NULL;
 654         svc_xprt_get(rqstp->rq_xprt);
 655         conn->cn_xprt = rqstp->rq_xprt;
 656         conn->cn_flags = NFS4_CDFC4_FORE;
 657         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 658         return conn;
 659 }
 660
 661 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 662 {
 663         conn->cn_session = ses;
 664         list_add(&conn->cn_persession, &ses->se_conns);
 665 }
 666
 667 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 668 {
 669         struct nfs4_client *clp = ses->se_client;
 670
 671         spin_lock(&clp->cl_lock);
 672         __nfsd4_hash_conn(conn, ses);
 673         spin_unlock(&clp->cl_lock);
 674 }
 675
 676 static void nfsd4_register_conn(struct nfsd4_conn *conn)
 677 {
 678         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 679         register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 680 }
 681
 682 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses)
 683 {
 684         struct nfsd4_conn *conn;
 685
 686         conn = alloc_conn(rqstp);
 687         if (!conn)
 688                 return nfserr_jukebox;
 689         nfsd4_hash_conn(conn, ses);
 690         nfsd4_register_conn(conn);
 691         return nfs_ok;
 692 }
 693
 694 static void nfsd4_del_conns(struct nfsd4_session *s)
 695 {
 696         struct nfs4_client *clp = s->se_client;
 697         struct nfsd4_conn *c;
 698
 699         spin_lock(&clp->cl_lock);
 700         while (!list_empty(&s->se_conns)) {
 701                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 702                 list_del_init(&c->cn_persession);
 703                 spin_unlock(&clp->cl_lock);
 704
 705                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 706                 free_conn(c);
 707
 708                 spin_lock(&clp->cl_lock);
 709         }
 710         spin_unlock(&clp->cl_lock);
 711 }
 712
 713 void free_session(struct kref *kref)
 714 {
 715         struct nfsd4_session *ses;
 716         int mem;
 717
 718         ses = container_of(kref, struct nfsd4_session, se_ref);
 719         nfsd4_del_conns(ses);
 720         spin_lock(&nfsd_drc_lock);
 721         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
 722         nfsd_drc_mem_used -= mem;
 723         spin_unlock(&nfsd_drc_lock);
 724         free_session_slots(ses);
 725         kfree(ses);
 726 }
 727
 728
 729 static __be32 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 730 {
 731         struct nfsd4_session *new;
 732         struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
 733         int numslots, slotsize;
 734         int status;
 735         int idx;
 736
 737         /*
 738          * Note decreasing slot size below client's request may
 739          * make it difficult for client to function correctly, whereas
 740          * decreasing the number of slots will (just?) affect
 741          * performance.  When short on memory we therefore prefer to
 742          * decrease number of slots instead of their size.
 743          */
 744         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
 745         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
 746
 747         new = alloc_session(slotsize, numslots);
 748         if (!new) {
 749                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
 750                 return nfserr_jukebox;
 751         }
 752         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
 753
 754         new->se_client = clp;
 755         gen_sessionid(new);
 756         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
 757                NFS4_MAX_SESSIONID_LEN);
 758
 759         INIT_LIST_HEAD(&new->se_conns);
 760
 761         new->se_flags = cses->flags;
 762         kref_init(&new->se_ref);
 763         idx = hash_sessionid(&new->se_sessionid);
 764         spin_lock(&client_lock);
 765         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
 766         list_add(&new->se_perclnt, &clp->cl_sessions);
 767         spin_unlock(&client_lock);
 768
 769         status = nfsd4_new_conn(rqstp, new);
 770         if (status) {
 771                 free_session(&new->se_ref);
 772                 return nfserr_jukebox;
 773         }
 774         return nfs_ok;
 775 }
 776
 777 /* caller must hold client_lock */
 778 static struct nfsd4_session *
 779 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
 780 {
 781         struct nfsd4_session *elem;
 782         int idx;
 783
 784         dump_sessionid(__func__, sessionid);
 785         idx = hash_sessionid(sessionid);
 786         /* Search in the appropriate list */
 787         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
 788                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
 789                             NFS4_MAX_SESSIONID_LEN)) {
 790                         return elem;
 791                 }
 792         }
 793
 794         dprintk("%s: session not found\n", __func__);
 795         return NULL;
 796 }
 797
 798 /* caller must hold client_lock */
 799 static void
 800 unhash_session(struct nfsd4_session *ses)
 801 {
 802         list_del(&ses->se_hash);
 803         list_del(&ses->se_perclnt);
 804 }
 805
 806 /* must be called under the client_lock */
 807 static inline void
 808 renew_client_locked(struct nfs4_client *clp)
 809 {
 810         if (is_client_expired(clp)) {
 811                 dprintk("%s: client (clientid %08x/%08x) already expired\n",
 812                         __func__,
 813                         clp->cl_clientid.cl_boot,
 814                         clp->cl_clientid.cl_id);
 815                 return;
 816         }
 817
 818         /*
 819         * Move client to the end to the LRU list.
 820         */
 821         dprintk("renewing client (clientid %08x/%08x)\n",
 822                         clp->cl_clientid.cl_boot,
 823                         clp->cl_clientid.cl_id);
 824         list_move_tail(&clp->cl_lru, &client_lru);
 825         clp->cl_time = get_seconds();
 826 }
 827
 828 static inline void
 829 renew_client(struct nfs4_client *clp)
 830 {
 831         spin_lock(&client_lock);
 832         renew_client_locked(clp);
 833         spin_unlock(&client_lock);
 834 }
 835
 836 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
 837 static int
 838 STALE_CLIENTID(clientid_t *clid)
 839 {
 840         if (clid->cl_boot == boot_time)
 841                 return 0;
 842         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
 843                 clid->cl_boot, clid->cl_id, boot_time);
 844         return 1;
 845 }
 846
 847 /*
 848  * XXX Should we use a slab cache ?
 849  * This type of memory management is somewhat inefficient, but we use it
 850  * anyway since SETCLIENTID is not a common operation.
 851  */
 852 static struct nfs4_client *alloc_client(struct xdr_netobj name)
 853 {
 854         struct nfs4_client *clp;
 855
 856         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
 857         if (clp == NULL)
 858                 return NULL;
 859         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
 860         if (clp->cl_name.data == NULL) {
 861                 kfree(clp);
 862                 return NULL;
 863         }
 864         memcpy(clp->cl_name.data, name.data, name.len);
 865         clp->cl_name.len = name.len;
 866         return clp;
 867 }
 868
 869 static inline void
 870 free_client(struct nfs4_client *clp)
 871 {
 872         if (clp->cl_cred.cr_group_info)
 873                 put_group_info(clp->cl_cred.cr_group_info);
 874         kfree(clp->cl_principal);
 875         kfree(clp->cl_name.data);
 876         kfree(clp);
 877 }
 878
 879 void
 880 release_session_client(struct nfsd4_session *session)
 881 {
 882         struct nfs4_client *clp = session->se_client;
 883
 884         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
 885                 return;
 886         if (is_client_expired(clp)) {
 887                 free_client(clp);
 888                 session->se_client = NULL;
 889         } else
 890                 renew_client_locked(clp);
 891         spin_unlock(&client_lock);
 892 }
 893
 894 /* must be called under the client_lock */
 895 static inline void
 896 unhash_client_locked(struct nfs4_client *clp)
 897 {
 898         mark_client_expired(clp);
 899         list_del(&clp->cl_lru);
 900         while (!list_empty(&clp->cl_sessions)) {
 901                 struct nfsd4_session  *ses;
 902                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
 903                                  se_perclnt);
 904                 unhash_session(ses);
 905                 nfsd4_put_session(ses);
 906         }
 907 }
 908
 909 static void
 910 expire_client(struct nfs4_client *clp)
 911 {
 912         struct nfs4_stateowner *sop;
 913         struct nfs4_delegation *dp;
 914         struct list_head reaplist;
 915
 916         INIT_LIST_HEAD(&reaplist);
 917         spin_lock(&recall_lock);
 918         while (!list_empty(&clp->cl_delegations)) {
 919                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
 920                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
 921                                 dp->dl_flock);
 922                 list_del_init(&dp->dl_perclnt);
 923                 list_move(&dp->dl_recall_lru, &reaplist);
 924         }
 925         spin_unlock(&recall_lock);
 926         while (!list_empty(&reaplist)) {
 927                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
 928                 list_del_init(&dp->dl_recall_lru);
 929                 unhash_delegation(dp);
 930         }
 931         while (!list_empty(&clp->cl_openowners)) {
 932                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
 933                 release_openowner(sop);
 934         }
 935         nfsd4_shutdown_callback(clp);
 936         if (clp->cl_cb_conn.cb_xprt)
 937                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
 938         list_del(&clp->cl_idhash);
 939         list_del(&clp->cl_strhash);
 940         spin_lock(&client_lock);
 941         unhash_client_locked(clp);
 942         if (atomic_read(&clp->cl_refcount) == 0)
 943                 free_client(clp);
 944         spin_unlock(&client_lock);
 945 }
 946
 947 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
 948 {
 949         memcpy(target->cl_verifier.data, source->data,
 950                         sizeof(target->cl_verifier.data));
 951 }
 952
 953 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
 954 {
 955         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
 956         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
 957 }
 958
 959 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
 960 {
 961         target->cr_uid = source->cr_uid;
 962         target->cr_gid = source->cr_gid;
 963         target->cr_group_info = source->cr_group_info;
 964         get_group_info(target->cr_group_info);
 965 }
 966
 967 static int same_name(const char *n1, const char *n2)
 968 {
 969         return 0 == memcmp(n1, n2, HEXDIR_LEN);
 970 }
 971
 972 static int
 973 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
 974 {
 975         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
 976 }
 977
 978 static int
 979 same_clid(clientid_t *cl1, clientid_t *cl2)
 980 {
 981         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
 982 }
 983
 984 /* XXX what about NGROUP */
 985 static int
 986 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
 987 {
 988         return cr1->cr_uid == cr2->cr_uid;
 989 }
 990
 991 static void gen_clid(struct nfs4_client *clp)
 992 {
 993         static u32 current_clientid = 1;
 994
 995         clp->cl_clientid.cl_boot = boot_time;
 996         clp->cl_clientid.cl_id = current_clientid++;
 997 }
 998
 999 static void gen_confirm(struct nfs4_client *clp)
1000 {
1001         static u32 i;
1002         u32 *p;
1003
1004         p = (u32 *)clp->cl_confirm.data;
1005         *p++ = get_seconds();
1006         *p++ = i++;
1007 }
1008
1009 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1010                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1011 {
1012         struct nfs4_client *clp;
1013         struct sockaddr *sa = svc_addr(rqstp);
1014         char *princ;
1015
1016         clp = alloc_client(name);
1017         if (clp == NULL)
1018                 return NULL;
1019
1020         princ = svc_gss_principal(rqstp);
1021         if (princ) {
1022                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1023                 if (clp->cl_principal == NULL) {
1024                         free_client(clp);
1025                         return NULL;
1026                 }
1027         }
1028
1029         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1030         atomic_set(&clp->cl_refcount, 0);
1031         atomic_set(&clp->cl_cb_set, 0);
1032         INIT_LIST_HEAD(&clp->cl_idhash);
1033         INIT_LIST_HEAD(&clp->cl_strhash);
1034         INIT_LIST_HEAD(&clp->cl_openowners);
1035         INIT_LIST_HEAD(&clp->cl_delegations);
1036         INIT_LIST_HEAD(&clp->cl_sessions);
1037         INIT_LIST_HEAD(&clp->cl_lru);
1038         spin_lock_init(&clp->cl_lock);
1039         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1040         clp->cl_time = get_seconds();
1041         clear_bit(0, &clp->cl_cb_slot_busy);
1042         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1043         copy_verf(clp, verf);
1044         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1045         clp->cl_flavor = rqstp->rq_flavor;
1046         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1047         gen_confirm(clp);
1048
1049         return clp;
1050 }
1051
1052 static int check_name(struct xdr_netobj name)
1053 {
1054         if (name.len == 0)
1055                 return 0;
1056         if (name.len > NFS4_OPAQUE_LIMIT) {
1057                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1058                 return 0;
1059         }
1060         return 1;
1061 }
1062
1063 static void
1064 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1065 {
1066         unsigned int idhashval;
1067
1068         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1069         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1070         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1071         renew_client(clp);
1072 }
1073
1074 static void
1075 move_to_confirmed(struct nfs4_client *clp)
1076 {
1077         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1078         unsigned int strhashval;
1079
1080         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1081         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1082         strhashval = clientstr_hashval(clp->cl_recdir);
1083         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1084         renew_client(clp);
1085 }
1086
1087 static struct nfs4_client *
1088 find_confirmed_client(clientid_t *clid)
1089 {
1090         struct nfs4_client *clp;
1091         unsigned int idhashval = clientid_hashval(clid->cl_id);
1092
1093         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1094                 if (same_clid(&clp->cl_clientid, clid))
1095                         return clp;
1096         }
1097         return NULL;
1098 }
1099
1100 static struct nfs4_client *
1101 find_unconfirmed_client(clientid_t *clid)
1102 {
1103         struct nfs4_client *clp;
1104         unsigned int idhashval = clientid_hashval(clid->cl_id);
1105
1106         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1107                 if (same_clid(&clp->cl_clientid, clid))
1108                         return clp;
1109         }
1110         return NULL;
1111 }
1112
1113 /*
1114  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
1115  * parameter. Matching is based on the fact the at least one of the
1116  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
1117  *
1118  * FIXME: we need to unify the clientid namespaces for nfsv4.x
1119  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
1120  * and SET_CLIENTID{,_CONFIRM}
1121  */
1122 static inline int
1123 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
1124 {
1125         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
1126         return use_exchange_id == has_exchange_flags;
1127 }
1128
1129 static struct nfs4_client *
1130 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
1131                              bool use_exchange_id)
1132 {
1133         struct nfs4_client *clp;
1134
1135         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1136                 if (same_name(clp->cl_recdir, dname) &&
1137                     match_clientid_establishment(clp, use_exchange_id))
1138                         return clp;
1139         }
1140         return NULL;
1141 }
1142
1143 static struct nfs4_client *
1144 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
1145                                bool use_exchange_id)
1146 {
1147         struct nfs4_client *clp;
1148
1149         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1150                 if (same_name(clp->cl_recdir, dname) &&
1151                     match_clientid_establishment(clp, use_exchange_id))
1152                         return clp;
1153         }
1154         return NULL;
1155 }
1156
1157 static void
1158 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1159 {
1160         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1161         unsigned short expected_family;
1162
1163         /* Currently, we only support tcp and tcp6 for the callback channel */
1164         if (se->se_callback_netid_len == 3 &&
1165             !memcmp(se->se_callback_netid_val, "tcp", 3))
1166                 expected_family = AF_INET;
1167         else if (se->se_callback_netid_len == 4 &&
1168                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1169                 expected_family = AF_INET6;
1170         else
1171                 goto out_err;
1172
1173         conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1174                                             se->se_callback_addr_len,
1175                                             (struct sockaddr *)&conn->cb_addr,
1176                                             sizeof(conn->cb_addr));
1177
1178         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1179                 goto out_err;
1180
1181         if (conn->cb_addr.ss_family == AF_INET6)
1182                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1183
1184         conn->cb_minorversion = 0;
1185         conn->cb_prog = se->se_callback_prog;
1186         conn->cb_ident = se->se_callback_ident;
1187         return;
1188 out_err:
1189         conn->cb_addr.ss_family = AF_UNSPEC;
1190         conn->cb_addrlen = 0;
1191         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1192                 "will not receive delegations\n",
1193                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1194
1195         return;
1196 }
1197
1198 /*
1199  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1200  */
1201 void
1202 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1203 {
1204         struct nfsd4_slot *slot = resp->cstate.slot;
1205         unsigned int base;
1206
1207         dprintk("--> %s slot %p\n", __func__, slot);
1208
1209         slot->sl_opcnt = resp->opcnt;
1210         slot->sl_status = resp->cstate.status;
1211
1212         if (nfsd4_not_cached(resp)) {
1213                 slot->sl_datalen = 0;
1214                 return;
1215         }
1216         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1217         base = (char *)resp->cstate.datap -
1218                                         (char *)resp->xbuf->head[0].iov_base;
1219         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1220                                     slot->sl_datalen))
1221                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1222         return;
1223 }
1224
1225 /*
1226  * Encode the replay sequence operation from the slot values.
1227  * If cachethis is FALSE encode the uncached rep error on the next
1228  * operation which sets resp->p and increments resp->opcnt for
1229  * nfs4svc_encode_compoundres.
1230  *
1231  */
1232 static __be32
1233 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1234                           struct nfsd4_compoundres *resp)
1235 {
1236         struct nfsd4_op *op;
1237         struct nfsd4_slot *slot = resp->cstate.slot;
1238
1239         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1240                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1241
1242         /* Encode the replayed sequence operation */
1243         op = &args->ops[resp->opcnt - 1];
1244         nfsd4_encode_operation(resp, op);
1245
1246         /* Return nfserr_retry_uncached_rep in next operation. */
1247         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1248                 op = &args->ops[resp->opcnt++];
1249                 op->status = nfserr_retry_uncached_rep;
1250                 nfsd4_encode_operation(resp, op);
1251         }
1252         return op->status;
1253 }
1254
1255 /*
1256  * The sequence operation is not cached because we can use the slot and
1257  * session values.
1258  */
1259 __be32
1260 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1261                          struct nfsd4_sequence *seq)
1262 {
1263         struct nfsd4_slot *slot = resp->cstate.slot;
1264         __be32 status;
1265
1266         dprintk("--> %s slot %p\n", __func__, slot);
1267
1268         /* Either returns 0 or nfserr_retry_uncached */
1269         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1270         if (status == nfserr_retry_uncached_rep)
1271                 return status;
1272
1273         /* The sequence operation has been encoded, cstate->datap set. */
1274         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1275
1276         resp->opcnt = slot->sl_opcnt;
1277         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1278         status = slot->sl_status;
1279
1280         return status;
1281 }
1282
1283 /*
1284  * Set the exchange_id flags returned by the server.
1285  */
1286 static void
1287 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1288 {
1289         /* pNFS is not supported */
1290         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1291
1292         /* Referrals are supported, Migration is not. */
1293         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1294
1295         /* set the wire flags to return to client. */
1296         clid->flags = new->cl_exchange_flags;
1297 }
1298
1299 __be32
1300 nfsd4_exchange_id(struct svc_rqst *rqstp,
1301                   struct nfsd4_compound_state *cstate,
1302                   struct nfsd4_exchange_id *exid)
1303 {
1304         struct nfs4_client *unconf, *conf, *new;
1305         int status;
1306         unsigned int            strhashval;
1307         char                    dname[HEXDIR_LEN];
1308         char                    addr_str[INET6_ADDRSTRLEN];
1309         nfs4_verifier           verf = exid->verifier;
1310         struct sockaddr         *sa = svc_addr(rqstp);
1311
1312         rpc_ntop(sa, addr_str, sizeof(addr_str));
1313         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1314                 "ip_addr=%s flags %x, spa_how %d\n",
1315                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1316                 addr_str, exid->flags, exid->spa_how);
1317
1318         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1319                 return nfserr_inval;
1320
1321         /* Currently only support SP4_NONE */
1322         switch (exid->spa_how) {
1323         case SP4_NONE:
1324                 break;
1325         case SP4_SSV:
1326                 return nfserr_encr_alg_unsupp;
1327         default:
1328                 BUG();                          /* checked by xdr code */
1329         case SP4_MACH_CRED:
1330                 return nfserr_serverfault;      /* no excuse :-/ */
1331         }
1332
1333         status = nfs4_make_rec_clidname(dname, &exid->clname);
1334
1335         if (status)
1336                 goto error;
1337
1338         strhashval = clientstr_hashval(dname);
1339
1340         nfs4_lock_state();
1341         status = nfs_ok;
1342
1343         conf = find_confirmed_client_by_str(dname, strhashval, true);
1344         if (conf) {
1345                 if (!same_verf(&verf, &conf->cl_verifier)) {
1346                         /* 18.35.4 case 8 */
1347                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1348                                 status = nfserr_not_same;
1349                                 goto out;
1350                         }
1351                         /* Client reboot: destroy old state */
1352                         expire_client(conf);
1353                         goto out_new;
1354                 }
1355                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1356                         /* 18.35.4 case 9 */
1357                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1358                                 status = nfserr_perm;
1359                                 goto out;
1360                         }
1361                         expire_client(conf);
1362                         goto out_new;
1363                 }
1364                 /*
1365                  * Set bit when the owner id and verifier map to an already
1366                  * confirmed client id (18.35.3).
1367                  */
1368                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1369
1370                 /*
1371                  * Falling into 18.35.4 case 2, possible router replay.
1372                  * Leave confirmed record intact and return same result.
1373                  */
1374                 copy_verf(conf, &verf);
1375                 new = conf;
1376                 goto out_copy;
1377         }
1378
1379         /* 18.35.4 case 7 */
1380         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1381                 status = nfserr_noent;
1382                 goto out;
1383         }
1384
1385         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1386         if (unconf) {
1387                 /*
1388                  * Possible retry or client restart.  Per 18.35.4 case 4,
1389                  * a new unconfirmed record should be generated regardless
1390                  * of whether any properties have changed.
1391                  */
1392                 expire_client(unconf);
1393         }
1394
1395 out_new:
1396         /* Normal case */
1397         new = create_client(exid->clname, dname, rqstp, &verf);
1398         if (new == NULL) {
1399                 status = nfserr_jukebox;
1400                 goto out;
1401         }
1402
1403         gen_clid(new);
1404         add_to_unconfirmed(new, strhashval);
1405 out_copy:
1406         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1407         exid->clientid.cl_id = new->cl_clientid.cl_id;
1408
1409         exid->seqid = 1;
1410         nfsd4_set_ex_flags(new, exid);
1411
1412         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1413                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1414         status = nfs_ok;
1415
1416 out:
1417         nfs4_unlock_state();
1418 error:
1419         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1420         return status;
1421 }
1422
1423 static int
1424 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1425 {
1426         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1427                 slot_seqid);
1428
1429         /* The slot is in use, and no response has been sent. */
1430         if (slot_inuse) {
1431                 if (seqid == slot_seqid)
1432                         return nfserr_jukebox;
1433                 else
1434                         return nfserr_seq_misordered;
1435         }
1436         /* Normal */
1437         if (likely(seqid == slot_seqid + 1))
1438                 return nfs_ok;
1439         /* Replay */
1440         if (seqid == slot_seqid)
1441                 return nfserr_replay_cache;
1442         /* Wraparound */
1443         if (seqid == 1 && (slot_seqid + 1) == 0)
1444                 return nfs_ok;
1445         /* Misordered replay or misordered new request */
1446         return nfserr_seq_misordered;
1447 }
1448
1449 /*
1450  * Cache the create session result into the create session single DRC
1451  * slot cache by saving the xdr structure. sl_seqid has been set.
1452  * Do this for solo or embedded create session operations.
1453  */
1454 static void
1455 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1456                            struct nfsd4_clid_slot *slot, int nfserr)
1457 {
1458         slot->sl_status = nfserr;
1459         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1460 }
1461
1462 static __be32
1463 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1464                             struct nfsd4_clid_slot *slot)
1465 {
1466         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1467         return slot->sl_status;
1468 }
1469
1470 __be32
1471 nfsd4_create_session(struct svc_rqst *rqstp,
1472                      struct nfsd4_compound_state *cstate,
1473                      struct nfsd4_create_session *cr_ses)
1474 {
1475         struct sockaddr *sa = svc_addr(rqstp);
1476         struct nfs4_client *conf, *unconf;
1477         struct nfsd4_clid_slot *cs_slot = NULL;
1478         int status = 0;
1479
1480         nfs4_lock_state();
1481         unconf = find_unconfirmed_client(&cr_ses->clientid);
1482         conf = find_confirmed_client(&cr_ses->clientid);
1483
1484         if (conf) {
1485                 cs_slot = &conf->cl_cs_slot;
1486                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1487                 if (status == nfserr_replay_cache) {
1488                         dprintk("Got a create_session replay! seqid= %d\n",
1489                                 cs_slot->sl_seqid);
1490                         /* Return the cached reply status */
1491                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1492                         goto out;
1493                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1494                         status = nfserr_seq_misordered;
1495                         dprintk("Sequence misordered!\n");
1496                         dprintk("Expected seqid= %d but got seqid= %d\n",
1497                                 cs_slot->sl_seqid, cr_ses->seqid);
1498                         goto out;
1499                 }
1500                 cs_slot->sl_seqid++;
1501         } else if (unconf) {
1502                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1503                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1504                         status = nfserr_clid_inuse;
1505                         goto out;
1506                 }
1507
1508                 cs_slot = &unconf->cl_cs_slot;
1509                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1510                 if (status) {
1511                         /* an unconfirmed replay returns misordered */
1512                         status = nfserr_seq_misordered;
1513                         goto out;
1514                 }
1515
1516                 cs_slot->sl_seqid++; /* from 0 to 1 */
1517                 move_to_confirmed(unconf);
1518
1519                 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1520                         unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1521                         svc_xprt_get(rqstp->rq_xprt);
1522                         rpc_copy_addr(
1523                                 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1524                                 sa);
1525                         unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1526                         unconf->cl_cb_conn.cb_minorversion =
1527                                 cstate->minorversion;
1528                         unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1529                         unconf->cl_cb_seq_nr = 1;
1530                         nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1531                 }
1532                 conf = unconf;
1533         } else {
1534                 status = nfserr_stale_clientid;
1535                 goto out;
1536         }
1537
1538         /*
1539          * We do not support RDMA or persistent sessions
1540          */
1541         cr_ses->flags &= ~SESSION4_PERSIST;
1542         cr_ses->flags &= ~SESSION4_RDMA;
1543
1544         status = alloc_init_session(rqstp, conf, cr_ses);
1545         if (status)
1546                 goto out;
1547
1548         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1549                NFS4_MAX_SESSIONID_LEN);
1550         cr_ses->seqid = cs_slot->sl_seqid;
1551
1552         /* cache solo and embedded create sessions under the state lock */
1553         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1554 out:
1555         nfs4_unlock_state();
1556         dprintk("%s returns %d\n", __func__, ntohl(status));
1557         return status;
1558 }
1559
1560 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1561 {
1562         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1563         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1564
1565         return argp->opcnt == resp->opcnt;
1566 }
1567
1568 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1569 {
1570         if (!session)
1571                 return 0;
1572         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1573 }
1574
1575 __be32
1576 nfsd4_destroy_session(struct svc_rqst *r,
1577                       struct nfsd4_compound_state *cstate,
1578                       struct nfsd4_destroy_session *sessionid)
1579 {
1580         struct nfsd4_session *ses;
1581         u32 status = nfserr_badsession;
1582
1583         /* Notes:
1584          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1585          * - Should we return nfserr_back_chan_busy if waiting for
1586          *   callbacks on to-be-destroyed session?
1587          * - Do we need to clear any callback info from previous session?
1588          */
1589
1590         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1591                 if (!nfsd4_last_compound_op(r))
1592                         return nfserr_not_only_op;
1593         }
1594         dump_sessionid(__func__, &sessionid->sessionid);
1595         spin_lock(&client_lock);
1596         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1597         if (!ses) {
1598                 spin_unlock(&client_lock);
1599                 goto out;
1600         }
1601
1602         unhash_session(ses);
1603         spin_unlock(&client_lock);
1604
1605         nfs4_lock_state();
1606         /* wait for callbacks */
1607         nfsd4_shutdown_callback(ses->se_client);
1608         nfs4_unlock_state();
1609
1610         nfsd4_del_conns(ses);
1611
1612         nfsd4_put_session(ses);
1613         status = nfs_ok;
1614 out:
1615         dprintk("%s returns %d\n", __func__, ntohl(status));
1616         return status;
1617 }
1618
1619 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_rqst *r, struct nfsd4_session *s)
1620 {
1621         struct nfsd4_conn *c;
1622
1623         list_for_each_entry(c, &s->se_conns, cn_persession) {
1624                 if (c->cn_xprt == r->rq_xprt) {
1625                         return c;
1626                 }
1627         }
1628         return NULL;
1629 }
1630
1631 static void nfsd4_sequence_check_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses)
1632 {
1633         struct nfs4_client *clp = ses->se_client;
1634         struct nfsd4_conn *c, *new = NULL;
1635
1636         spin_lock(&clp->cl_lock);
1637         c = __nfsd4_find_conn(rqstp, ses);
1638         spin_unlock(&clp->cl_lock);
1639         if (c)
1640                 return;
1641
1642         new = alloc_conn(rqstp);
1643
1644         spin_lock(&clp->cl_lock);
1645         c = __nfsd4_find_conn(rqstp, ses);
1646         if (c) {
1647                 spin_unlock(&clp->cl_lock);
1648                 free_conn(new);
1649                 return;
1650         }
1651         __nfsd4_hash_conn(new, ses);
1652         spin_unlock(&clp->cl_lock);
1653         nfsd4_register_conn(new);
1654         return;
1655 }
1656
1657 __be32
1658 nfsd4_sequence(struct svc_rqst *rqstp,
1659                struct nfsd4_compound_state *cstate,
1660                struct nfsd4_sequence *seq)
1661 {
1662         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1663         struct nfsd4_session *session;
1664         struct nfsd4_slot *slot;
1665         int status;
1666
1667         if (resp->opcnt != 1)
1668                 return nfserr_sequence_pos;
1669
1670         spin_lock(&client_lock);
1671         status = nfserr_badsession;
1672         session = find_in_sessionid_hashtbl(&seq->sessionid);
1673         if (!session)
1674                 goto out;
1675
1676         status = nfserr_badslot;
1677         if (seq->slotid >= session->se_fchannel.maxreqs)
1678                 goto out;
1679
1680         slot = session->se_slots[seq->slotid];
1681         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1682
1683         /* We do not negotiate the number of slots yet, so set the
1684          * maxslots to the session maxreqs which is used to encode
1685          * sr_highest_slotid and the sr_target_slot id to maxslots */
1686         seq->maxslots = session->se_fchannel.maxreqs;
1687
1688         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1689         if (status == nfserr_replay_cache) {
1690                 cstate->slot = slot;
1691                 cstate->session = session;
1692                 /* Return the cached reply status and set cstate->status
1693                  * for nfsd4_proc_compound processing */
1694                 status = nfsd4_replay_cache_entry(resp, seq);
1695                 cstate->status = nfserr_replay_cache;
1696                 goto out;
1697         }
1698         if (status)
1699                 goto out;
1700
1701         nfsd4_sequence_check_conn(rqstp, session);
1702
1703         /* Success! bump slot seqid */
1704         slot->sl_inuse = true;
1705         slot->sl_seqid = seq->seqid;
1706         slot->sl_cachethis = seq->cachethis;
1707
1708         cstate->slot = slot;
1709         cstate->session = session;
1710
1711 out:
1712         /* Hold a session reference until done processing the compound. */
1713         if (cstate->session) {
1714                 nfsd4_get_session(cstate->session);
1715                 atomic_inc(&session->se_client->cl_refcount);
1716         }
1717         spin_unlock(&client_lock);
1718         dprintk("%s: return %d\n", __func__, ntohl(status));
1719         return status;
1720 }
1721
1722 __be32
1723 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1724 {
1725         if (rc->rca_one_fs) {
1726                 if (!cstate->current_fh.fh_dentry)
1727                         return nfserr_nofilehandle;
1728                 /*
1729                  * We don't take advantage of the rca_one_fs case.
1730                  * That's OK, it's optional, we can safely ignore it.
1731                  */
1732                  return nfs_ok;
1733         }
1734         nfs4_lock_state();
1735         if (is_client_expired(cstate->session->se_client)) {
1736                 nfs4_unlock_state();
1737                 /*
1738                  * The following error isn't really legal.
1739                  * But we only get here if the client just explicitly
1740                  * destroyed the client.  Surely it no longer cares what
1741                  * error it gets back on an operation for the dead
1742                  * client.
1743                  */
1744                 return nfserr_stale_clientid;
1745         }
1746         nfsd4_create_clid_dir(cstate->session->se_client);
1747         nfs4_unlock_state();
1748         return nfs_ok;
1749 }
1750
1751 __be32
1752 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1753                   struct nfsd4_setclientid *setclid)
1754 {
1755         struct sockaddr         *sa = svc_addr(rqstp);
1756         struct xdr_netobj       clname = {
1757                 .len = setclid->se_namelen,
1758                 .data = setclid->se_name,
1759         };
1760         nfs4_verifier           clverifier = setclid->se_verf;
1761         unsigned int            strhashval;
1762         struct nfs4_client      *conf, *unconf, *new;
1763         __be32                  status;
1764         char                    dname[HEXDIR_LEN];
1765
1766         if (!check_name(clname))
1767                 return nfserr_inval;
1768
1769         status = nfs4_make_rec_clidname(dname, &clname);
1770         if (status)
1771                 return status;
1772
1773         /*
1774          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1775          * We get here on a DRC miss.
1776          */
1777
1778         strhashval = clientstr_hashval(dname);
1779
1780         nfs4_lock_state();
1781         conf = find_confirmed_client_by_str(dname, strhashval, false);
1782         if (conf) {
1783                 /* RFC 3530 14.2.33 CASE 0: */
1784                 status = nfserr_clid_inuse;
1785                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1786                         char addr_str[INET6_ADDRSTRLEN];
1787                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1788                                  sizeof(addr_str));
1789                         dprintk("NFSD: setclientid: string in use by client "
1790                                 "at %s\n", addr_str);
1791                         goto out;
1792                 }
1793         }
1794         /*
1795          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1796          * has a description of SETCLIENTID request processing consisting
1797          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1798          */
1799         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1800         status = nfserr_resource;
1801         if (!conf) {
1802                 /*
1803                  * RFC 3530 14.2.33 CASE 4:
1804                  * placed first, because it is the normal case
1805                  */
1806                 if (unconf)
1807                         expire_client(unconf);
1808                 new = create_client(clname, dname, rqstp, &clverifier);
1809                 if (new == NULL)
1810                         goto out;
1811                 gen_clid(new);
1812         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1813                 /*
1814                  * RFC 3530 14.2.33 CASE 1:
1815                  * probable callback update
1816                  */
1817                 if (unconf) {
1818                         /* Note this is removing unconfirmed {*x***},
1819                          * which is stronger than RFC recommended {vxc**}.
1820                          * This has the advantage that there is at most
1821                          * one {*x***} in either list at any time.
1822                          */
1823                         expire_client(unconf);
1824                 }
1825                 new = create_client(clname, dname, rqstp, &clverifier);
1826                 if (new == NULL)
1827                         goto out;
1828                 copy_clid(new, conf);
1829         } else if (!unconf) {
1830                 /*
1831                  * RFC 3530 14.2.33 CASE 2:
1832                  * probable client reboot; state will be removed if
1833                  * confirmed.
1834                  */
1835                 new = create_client(clname, dname, rqstp, &clverifier);
1836                 if (new == NULL)
1837                         goto out;
1838                 gen_clid(new);
1839         } else {
1840                 /*
1841                  * RFC 3530 14.2.33 CASE 3:
1842                  * probable client reboot; state will be removed if
1843                  * confirmed.
1844                  */
1845                 expire_client(unconf);
1846                 new = create_client(clname, dname, rqstp, &clverifier);
1847                 if (new == NULL)
1848                         goto out;
1849                 gen_clid(new);
1850         }
1851         gen_callback(new, setclid, rpc_get_scope_id(sa));
1852         add_to_unconfirmed(new, strhashval);
1853         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1854         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1855         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1856         status = nfs_ok;
1857 out:
1858         nfs4_unlock_state();
1859         return status;
1860 }
1861
1862
1863 /*
1864  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1865  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1866  * bullets, labeled as CASE1 - CASE4 below.
1867  */
1868 __be32
1869 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1870                          struct nfsd4_compound_state *cstate,
1871                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1872 {
1873         struct sockaddr *sa = svc_addr(rqstp);
1874         struct nfs4_client *conf, *unconf;
1875         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
1876         clientid_t * clid = &setclientid_confirm->sc_clientid;
1877         __be32 status;
1878
1879         if (STALE_CLIENTID(clid))
1880                 return nfserr_stale_clientid;
1881         /*
1882          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1883          * We get here on a DRC miss.
1884          */
1885
1886         nfs4_lock_state();
1887
1888         conf = find_confirmed_client(clid);
1889         unconf = find_unconfirmed_client(clid);
1890
1891         status = nfserr_clid_inuse;
1892         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1893                 goto out;
1894         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1895                 goto out;
1896
1897         /*
1898          * section 14.2.34 of RFC 3530 has a description of
1899          * SETCLIENTID_CONFIRM request processing consisting
1900          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1901          */
1902         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1903                 /*
1904                  * RFC 3530 14.2.34 CASE 1:
1905                  * callback update
1906                  */
1907                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1908                         status = nfserr_clid_inuse;
1909                 else {
1910                         atomic_set(&conf->cl_cb_set, 0);
1911                         nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1912                         expire_client(unconf);
1913                         status = nfs_ok;
1914
1915                 }
1916         } else if (conf && !unconf) {
1917                 /*
1918                  * RFC 3530 14.2.34 CASE 2:
1919                  * probable retransmitted request; play it safe and
1920                  * do nothing.
1921                  */
1922                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1923                         status = nfserr_clid_inuse;
1924                 else
1925                         status = nfs_ok;
1926         } else if (!conf && unconf
1927                         && same_verf(&unconf->cl_confirm, &confirm)) {
1928                 /*
1929                  * RFC 3530 14.2.34 CASE 3:
1930                  * Normal case; new or rebooted client:
1931                  */
1932                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1933                         status = nfserr_clid_inuse;
1934                 } else {
1935                         unsigned int hash =
1936                                 clientstr_hashval(unconf->cl_recdir);
1937                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1938                                                             hash, false);
1939                         if (conf) {
1940                                 nfsd4_remove_clid_dir(conf);
1941                                 expire_client(conf);
1942                         }
1943                         move_to_confirmed(unconf);
1944                         conf = unconf;
1945                         nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1946                         status = nfs_ok;
1947                 }
1948         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1949             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1950                                                                 &confirm)))) {
1951                 /*
1952                  * RFC 3530 14.2.34 CASE 4:
1953                  * Client probably hasn't noticed that we rebooted yet.
1954                  */
1955                 status = nfserr_stale_clientid;
1956         } else {
1957                 /* check that we have hit one of the cases...*/
1958                 status = nfserr_clid_inuse;
1959         }
1960 out:
1961         nfs4_unlock_state();
1962         return status;
1963 }
1964
1965 /* OPEN Share state helper functions */
1966 static inline struct nfs4_file *
1967 alloc_init_file(struct inode *ino)
1968 {
1969         struct nfs4_file *fp;
1970         unsigned int hashval = file_hashval(ino);
1971
1972         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1973         if (fp) {
1974                 atomic_set(&fp->fi_ref, 1);
1975                 INIT_LIST_HEAD(&fp->fi_hash);
1976                 INIT_LIST_HEAD(&fp->fi_stateids);
1977                 INIT_LIST_HEAD(&fp->fi_delegations);
1978                 fp->fi_inode = igrab(ino);
1979                 fp->fi_id = current_fileid++;
1980                 fp->fi_had_conflict = false;
1981                 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
1982                 memset(fp->fi_access, 0, sizeof(fp->fi_access));
1983                 spin_lock(&recall_lock);
1984                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1985                 spin_unlock(&recall_lock);
1986                 return fp;
1987         }
1988         return NULL;
1989 }
1990
1991 static void
1992 nfsd4_free_slab(struct kmem_cache **slab)
1993 {
1994         if (*slab == NULL)
1995                 return;
1996         kmem_cache_destroy(*slab);
1997         *slab = NULL;
1998 }
1999
2000 void
2001 nfsd4_free_slabs(void)
2002 {
2003         nfsd4_free_slab(&stateowner_slab);
2004         nfsd4_free_slab(&file_slab);
2005         nfsd4_free_slab(&stateid_slab);
2006         nfsd4_free_slab(&deleg_slab);
2007 }
2008
2009 static int
2010 nfsd4_init_slabs(void)
2011 {
2012         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2013                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
2014         if (stateowner_slab == NULL)
2015                 goto out_nomem;
2016         file_slab = kmem_cache_create("nfsd4_files",
2017                         sizeof(struct nfs4_file), 0, 0, NULL);
2018         if (file_slab == NULL)
2019                 goto out_nomem;
2020         stateid_slab = kmem_cache_create("nfsd4_stateids",
2021                         sizeof(struct nfs4_stateid), 0, 0, NULL);
2022         if (stateid_slab == NULL)
2023                 goto out_nomem;
2024         deleg_slab = kmem_cache_create("nfsd4_delegations",
2025                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2026         if (deleg_slab == NULL)
2027                 goto out_nomem;
2028         return 0;
2029 out_nomem:
2030         nfsd4_free_slabs();
2031         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2032         return -ENOMEM;
2033 }
2034
2035 void
2036 nfs4_free_stateowner(struct kref *kref)
2037 {
2038         struct nfs4_stateowner *sop =
2039                 container_of(kref, struct nfs4_stateowner, so_ref);
2040         kfree(sop->so_owner.data);
2041         kmem_cache_free(stateowner_slab, sop);
2042 }
2043
2044 static inline struct nfs4_stateowner *
2045 alloc_stateowner(struct xdr_netobj *owner)
2046 {
2047         struct nfs4_stateowner *sop;
2048
2049         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2050                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2051                         memcpy(sop->so_owner.data, owner->data, owner->len);
2052                         sop->so_owner.len = owner->len;
2053                         kref_init(&sop->so_ref);
2054                         return sop;
2055                 }
2056                 kmem_cache_free(stateowner_slab, sop);
2057         }
2058         return NULL;
2059 }
2060
2061 static struct nfs4_stateowner *
2062 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2063         struct nfs4_stateowner *sop;
2064         struct nfs4_replay *rp;
2065         unsigned int idhashval;
2066
2067         if (!(sop = alloc_stateowner(&open->op_owner)))
2068                 return NULL;
2069         idhashval = ownerid_hashval(current_ownerid);
2070         INIT_LIST_HEAD(&sop->so_idhash);
2071         INIT_LIST_HEAD(&sop->so_strhash);
2072         INIT_LIST_HEAD(&sop->so_perclient);
2073         INIT_LIST_HEAD(&sop->so_stateids);
2074         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
2075         INIT_LIST_HEAD(&sop->so_close_lru);
2076         sop->so_time = 0;
2077         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2078         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2079         list_add(&sop->so_perclient, &clp->cl_openowners);
2080         sop->so_is_open_owner = 1;
2081         sop->so_id = current_ownerid++;
2082         sop->so_client = clp;
2083         sop->so_seqid = open->op_seqid;
2084         sop->so_confirmed = 0;
2085         rp = &sop->so_replay;
2086         rp->rp_status = nfserr_serverfault;
2087         rp->rp_buflen = 0;
2088         rp->rp_buf = rp->rp_ibuf;
2089         return sop;
2090 }
2091
2092 static inline void
2093 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2094         struct nfs4_stateowner *sop = open->op_stateowner;
2095         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2096
2097         INIT_LIST_HEAD(&stp->st_hash);
2098         INIT_LIST_HEAD(&stp->st_perstateowner);
2099         INIT_LIST_HEAD(&stp->st_lockowners);
2100         INIT_LIST_HEAD(&stp->st_perfile);
2101         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2102         list_add(&stp->st_perstateowner, &sop->so_stateids);
2103         list_add(&stp->st_perfile, &fp->fi_stateids);
2104         stp->st_stateowner = sop;
2105         get_nfs4_file(fp);
2106         stp->st_file = fp;
2107         stp->st_stateid.si_boot = boot_time;
2108         stp->st_stateid.si_stateownerid = sop->so_id;
2109         stp->st_stateid.si_fileid = fp->fi_id;
2110         stp->st_stateid.si_generation = 0;
2111         stp->st_access_bmap = 0;
2112         stp->st_deny_bmap = 0;
2113         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2114                   &stp->st_access_bmap);
2115         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2116         stp->st_openstp = NULL;
2117 }
2118
2119 static void
2120 move_to_close_lru(struct nfs4_stateowner *sop)
2121 {
2122         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2123
2124         list_move_tail(&sop->so_close_lru, &close_lru);
2125         sop->so_time = get_seconds();
2126 }
2127
2128 static int
2129 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2130                                                         clientid_t *clid)
2131 {
2132         return (sop->so_owner.len == owner->len) &&
2133                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2134                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2135 }
2136
2137 static struct nfs4_stateowner *
2138 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2139 {
2140         struct nfs4_stateowner *so = NULL;
2141
2142         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2143                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2144                         return so;
2145         }
2146         return NULL;
2147 }
2148
2149 /* search file_hashtbl[] for file */
2150 static struct nfs4_file *
2151 find_file(struct inode *ino)
2152 {
2153         unsigned int hashval = file_hashval(ino);
2154         struct nfs4_file *fp;
2155
2156         spin_lock(&recall_lock);
2157         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2158                 if (fp->fi_inode == ino) {
2159                         get_nfs4_file(fp);
2160                         spin_unlock(&recall_lock);
2161                         return fp;
2162                 }
2163         }
2164         spin_unlock(&recall_lock);
2165         return NULL;
2166 }
2167
2168 static inline int access_valid(u32 x, u32 minorversion)
2169 {
2170         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2171                 return 0;
2172         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2173                 return 0;
2174         x &= ~NFS4_SHARE_ACCESS_MASK;
2175         if (minorversion && x) {
2176                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2177                         return 0;
2178                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2179                         return 0;
2180                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2181         }
2182         if (x)
2183                 return 0;
2184         return 1;
2185 }
2186
2187 static inline int deny_valid(u32 x)
2188 {
2189         /* Note: unlike access bits, deny bits may be zero. */
2190         return x <= NFS4_SHARE_DENY_BOTH;
2191 }
2192
2193 /*
2194  * Called to check deny when READ with all zero stateid or
2195  * WRITE with all zero or all one stateid
2196  */
2197 static __be32
2198 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2199 {
2200         struct inode *ino = current_fh->fh_dentry->d_inode;
2201         struct nfs4_file *fp;
2202         struct nfs4_stateid *stp;
2203         __be32 ret;
2204
2205         dprintk("NFSD: nfs4_share_conflict\n");
2206
2207         fp = find_file(ino);
2208         if (!fp)
2209                 return nfs_ok;
2210         ret = nfserr_locked;
2211         /* Search for conflicting share reservations */
2212         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2213                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2214                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2215                         goto out;
2216         }
2217         ret = nfs_ok;
2218 out:
2219         put_nfs4_file(fp);
2220         return ret;
2221 }
2222
2223 static inline void
2224 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2225 {
2226         if (share_access & NFS4_SHARE_ACCESS_WRITE)
2227                 nfs4_file_put_access(fp, O_WRONLY);
2228         if (share_access & NFS4_SHARE_ACCESS_READ)
2229                 nfs4_file_put_access(fp, O_RDONLY);
2230 }
2231
2232 /*
2233  * Spawn a thread to perform a recall on the delegation represented
2234  * by the lease (file_lock)
2235  *
2236  * Called from break_lease() with lock_kernel() held.
2237  * Note: we assume break_lease will only call this *once* for any given
2238  * lease.
2239  */
2240 static
2241 void nfsd_break_deleg_cb(struct file_lock *fl)
2242 {
2243         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2244
2245         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2246         if (!dp)
2247                 return;
2248
2249         /* We're assuming the state code never drops its reference
2250          * without first removing the lease.  Since we're in this lease
2251          * callback (and since the lease code is serialized by the kernel
2252          * lock) we know the server hasn't removed the lease yet, we know
2253          * it's safe to take a reference: */
2254         atomic_inc(&dp->dl_count);
2255
2256         spin_lock(&recall_lock);
2257         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2258         spin_unlock(&recall_lock);
2259
2260         /* only place dl_time is set. protected by lock_kernel*/
2261         dp->dl_time = get_seconds();
2262
2263         /*
2264          * We don't want the locks code to timeout the lease for us;
2265          * we'll remove it ourself if the delegation isn't returned
2266          * in time.
2267          */
2268         fl->fl_break_time = 0;
2269
2270         dp->dl_file->fi_had_conflict = true;
2271         nfsd4_cb_recall(dp);
2272 }
2273
2274 /*
2275  * The file_lock is being reapd.
2276  *
2277  * Called by locks_free_lock() with lock_kernel() held.
2278  */
2279 static
2280 void nfsd_release_deleg_cb(struct file_lock *fl)
2281 {
2282         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2283
2284         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2285
2286         if (!(fl->fl_flags & FL_LEASE) || !dp)
2287                 return;
2288         dp->dl_flock = NULL;
2289 }
2290
2291 /*
2292  * Set the delegation file_lock back pointer.
2293  *
2294  * Called from setlease() with lock_kernel() held.
2295  */
2296 static
2297 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2298 {
2299         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2300
2301         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2302         if (!dp)
2303                 return;
2304         dp->dl_flock = new;
2305 }
2306
2307 /*
2308  * Called from setlease() with lock_kernel() held
2309  */
2310 static
2311 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2312 {
2313         struct nfs4_delegation *onlistd =
2314                 (struct nfs4_delegation *)onlist->fl_owner;
2315         struct nfs4_delegation *tryd =
2316                 (struct nfs4_delegation *)try->fl_owner;
2317
2318         if (onlist->fl_lmops != try->fl_lmops)
2319                 return 0;
2320
2321         return onlistd->dl_client == tryd->dl_client;
2322 }
2323
2324
2325 static
2326 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2327 {
2328         if (arg & F_UNLCK)
2329                 return lease_modify(onlist, arg);
2330         else
2331                 return -EAGAIN;
2332 }
2333
2334 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2335         .fl_break = nfsd_break_deleg_cb,
2336         .fl_release_private = nfsd_release_deleg_cb,
2337         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2338         .fl_mylease = nfsd_same_client_deleg_cb,
2339         .fl_change = nfsd_change_deleg_cb,
2340 };
2341
2342
2343 __be32
2344 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2345                     struct nfsd4_open *open)
2346 {
2347         clientid_t *clientid = &open->op_clientid;
2348         struct nfs4_client *clp = NULL;
2349         unsigned int strhashval;
2350         struct nfs4_stateowner *sop = NULL;
2351
2352         if (!check_name(open->op_owner))
2353                 return nfserr_inval;
2354
2355         if (STALE_CLIENTID(&open->op_clientid))
2356                 return nfserr_stale_clientid;
2357
2358         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2359         sop = find_openstateowner_str(strhashval, open);
2360         open->op_stateowner = sop;
2361         if (!sop) {
2362                 /* Make sure the client's lease hasn't expired. */
2363                 clp = find_confirmed_client(clientid);
2364                 if (clp == NULL)
2365                         return nfserr_expired;
2366                 goto renew;
2367         }
2368         /* When sessions are used, skip open sequenceid processing */
2369         if (nfsd4_has_session(cstate))
2370                 goto renew;
2371         if (!sop->so_confirmed) {
2372                 /* Replace unconfirmed owners without checking for replay. */
2373                 clp = sop->so_client;
2374                 release_openowner(sop);
2375                 open->op_stateowner = NULL;
2376                 goto renew;
2377         }
2378         if (open->op_seqid == sop->so_seqid - 1) {
2379                 if (sop->so_replay.rp_buflen)
2380                         return nfserr_replay_me;
2381                 /* The original OPEN failed so spectacularly
2382                  * that we don't even have replay data saved!
2383                  * Therefore, we have no choice but to continue
2384                  * processing this OPEN; presumably, we'll
2385                  * fail again for the same reason.
2386                  */
2387                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2388                 goto renew;
2389         }
2390         if (open->op_seqid != sop->so_seqid)
2391                 return nfserr_bad_seqid;
2392 renew:
2393         if (open->op_stateowner == NULL) {
2394                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2395                 if (sop == NULL)
2396                         return nfserr_resource;
2397                 open->op_stateowner = sop;
2398         }
2399         list_del_init(&sop->so_close_lru);
2400         renew_client(sop->so_client);
2401         return nfs_ok;
2402 }
2403
2404 static inline __be32
2405 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2406 {
2407         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2408                 return nfserr_openmode;
2409         else
2410                 return nfs_ok;
2411 }
2412
2413 static struct nfs4_delegation *
2414 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2415 {
2416         struct nfs4_delegation *dp;
2417
2418         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2419                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2420                         return dp;
2421         }
2422         return NULL;
2423 }
2424
2425 int share_access_to_flags(u32 share_access)
2426 {
2427         share_access &= ~NFS4_SHARE_WANT_MASK;
2428
2429         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2430 }
2431
2432 static __be32
2433 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2434                 struct nfs4_delegation **dp)
2435 {
2436         int flags;
2437         __be32 status = nfserr_bad_stateid;
2438
2439         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2440         if (*dp == NULL)
2441                 goto out;
2442         flags = share_access_to_flags(open->op_share_access);
2443         status = nfs4_check_delegmode(*dp, flags);
2444         if (status)
2445                 *dp = NULL;
2446 out:
2447         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2448                 return nfs_ok;
2449         if (status)
2450                 return status;
2451         open->op_stateowner->so_confirmed = 1;
2452         return nfs_ok;
2453 }
2454
2455 static __be32
2456 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2457 {
2458         struct nfs4_stateid *local;
2459         __be32 status = nfserr_share_denied;
2460         struct nfs4_stateowner *sop = open->op_stateowner;
2461
2462         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2463                 /* ignore lock owners */
2464                 if (local->st_stateowner->so_is_open_owner == 0)
2465                         continue;
2466                 /* remember if we have seen this open owner */
2467                 if (local->st_stateowner == sop)
2468                         *stpp = local;
2469                 /* check for conflicting share reservations */
2470                 if (!test_share(local, open))
2471                         goto out;
2472         }
2473         status = 0;
2474 out:
2475         return status;
2476 }
2477
2478 static inline struct nfs4_stateid *
2479 nfs4_alloc_stateid(void)
2480 {
2481         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2482 }
2483
2484 static inline int nfs4_access_to_access(u32 nfs4_access)
2485 {
2486         int flags = 0;
2487
2488         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2489                 flags |= NFSD_MAY_READ;
2490         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2491                 flags |= NFSD_MAY_WRITE;
2492         return flags;
2493 }
2494
2495 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2496 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2497 {
2498         __be32 status;
2499         int oflag = nfs4_access_to_omode(nfs4_access);
2500         int access = nfs4_access_to_access(nfs4_access);
2501
2502         if (!fp->fi_fds[oflag]) {
2503                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2504                         &fp->fi_fds[oflag]);
2505                 if (status == nfserr_dropit)
2506                         status = nfserr_jukebox;
2507                 if (status)
2508                         return status;
2509         }
2510         nfs4_file_get_access(fp, oflag);
2511
2512         return nfs_ok;
2513 }
2514
2515 static __be32
2516 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2517                 struct nfs4_file *fp, struct svc_fh *cur_fh,
2518                 struct nfsd4_open *open)
2519 {
2520         struct nfs4_stateid *stp;
2521         __be32 status;
2522
2523         stp = nfs4_alloc_stateid();
2524         if (stp == NULL)
2525                 return nfserr_resource;
2526
2527         status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2528         if (status) {
2529                 kmem_cache_free(stateid_slab, stp);
2530                 return status;
2531         }
2532         *stpp = stp;
2533         return 0;
2534 }
2535
2536 static inline __be32
2537 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2538                 struct nfsd4_open *open)
2539 {
2540         struct iattr iattr = {
2541                 .ia_valid = ATTR_SIZE,
2542                 .ia_size = 0,
2543         };
2544         if (!open->op_truncate)
2545                 return 0;
2546         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2547                 return nfserr_inval;
2548         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2549 }
2550
2551 static __be32
2552 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2553 {
2554         u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2555         bool new_access;
2556         __be32 status;
2557
2558         new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2559         if (new_access) {
2560                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2561                 if (status)
2562                         return status;
2563         }
2564         status = nfsd4_truncate(rqstp, cur_fh, open);
2565         if (status) {
2566                 if (new_access) {
2567                         int oflag = nfs4_access_to_omode(new_access);
2568                         nfs4_file_put_access(fp, oflag);
2569                 }
2570                 return status;
2571         }
2572         /* remember the open */
2573         __set_bit(op_share_access, &stp->st_access_bmap);
2574         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2575
2576         return nfs_ok;
2577 }
2578
2579
2580 static void
2581 nfs4_set_claim_prev(struct nfsd4_open *open)
2582 {
2583         open->op_stateowner->so_confirmed = 1;
2584         open->op_stateowner->so_client->cl_firststate = 1;
2585 }
2586
2587 /*
2588  * Attempt to hand out a delegation.
2589  */
2590 static void
2591 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2592 {
2593         struct nfs4_delegation *dp;
2594         struct nfs4_stateowner *sop = stp->st_stateowner;
2595         int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2596         struct file_lock fl, *flp = &fl;
2597         int status, flag = 0;
2598
2599         flag = NFS4_OPEN_DELEGATE_NONE;
2600         open->op_recall = 0;
2601         switch (open->op_claim_type) {
2602                 case NFS4_OPEN_CLAIM_PREVIOUS:
2603                         if (!cb_up)
2604                                 open->op_recall = 1;
2605                         flag = open->op_delegate_type;
2606                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2607                                 goto out;
2608                         break;
2609                 case NFS4_OPEN_CLAIM_NULL:
2610                         /* Let's not give out any delegations till everyone's
2611                          * had the chance to reclaim theirs.... */
2612                         if (locks_in_grace())
2613                                 goto out;
2614                         if (!cb_up || !sop->so_confirmed)
2615                                 goto out;
2616                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2617                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2618                         else
2619                                 flag = NFS4_OPEN_DELEGATE_READ;
2620                         break;
2621                 default:
2622                         goto out;
2623         }
2624
2625         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2626         if (dp == NULL) {
2627                 flag = NFS4_OPEN_DELEGATE_NONE;
2628                 goto out;
2629         }
2630         locks_init_lock(&fl);
2631         fl.fl_lmops = &nfsd_lease_mng_ops;
2632         fl.fl_flags = FL_LEASE;
2633         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2634         fl.fl_end = OFFSET_MAX;
2635         fl.fl_owner =  (fl_owner_t)dp;
2636         fl.fl_file = find_readable_file(stp->st_file);
2637         BUG_ON(!fl.fl_file);
2638         fl.fl_pid = current->tgid;
2639
2640         /* vfs_setlease checks to see if delegation should be handed out.
2641          * the lock_manager callbacks fl_mylease and fl_change are used
2642          */
2643         if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
2644                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2645                 unhash_delegation(dp);
2646                 flag = NFS4_OPEN_DELEGATE_NONE;
2647                 goto out;
2648         }
2649
2650         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2651
2652         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2653                 STATEID_VAL(&dp->dl_stateid));
2654 out:
2655         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2656                         && flag == NFS4_OPEN_DELEGATE_NONE
2657                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2658                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2659         open->op_delegate_type = flag;
2660 }
2661
2662 /*
2663  * called with nfs4_lock_state() held.
2664  */
2665 __be32
2666 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2667 {
2668         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2669         struct nfs4_file *fp = NULL;
2670         struct inode *ino = current_fh->fh_dentry->d_inode;
2671         struct nfs4_stateid *stp = NULL;
2672         struct nfs4_delegation *dp = NULL;
2673         __be32 status;
2674
2675         status = nfserr_inval;
2676         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2677                         || !deny_valid(open->op_share_deny))
2678                 goto out;
2679         /*
2680          * Lookup file; if found, lookup stateid and check open request,
2681          * and check for delegations in the process of being recalled.
2682          * If not found, create the nfs4_file struct
2683          */
2684         fp = find_file(ino);
2685         if (fp) {
2686                 if ((status = nfs4_check_open(fp, open, &stp)))
2687                         goto out;
2688                 status = nfs4_check_deleg(fp, open, &dp);
2689                 if (status)
2690                         goto out;
2691         } else {
2692                 status = nfserr_bad_stateid;
2693                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2694                         goto out;
2695                 status = nfserr_resource;
2696                 fp = alloc_init_file(ino);
2697                 if (fp == NULL)
2698                         goto out;
2699         }
2700
2701         /*
2702          * OPEN the file, or upgrade an existing OPEN.
2703          * If truncate fails, the OPEN fails.
2704          */
2705         if (stp) {
2706                 /* Stateid was found, this is an OPEN upgrade */
2707                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2708                 if (status)
2709                         goto out;
2710                 update_stateid(&stp->st_stateid);
2711         } else {
2712                 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2713                 if (status)
2714                         goto out;
2715                 init_stateid(stp, fp, open);
2716                 status = nfsd4_truncate(rqstp, current_fh, open);
2717                 if (status) {
2718                         release_open_stateid(stp);
2719                         goto out;
2720                 }
2721                 if (nfsd4_has_session(&resp->cstate))
2722                         update_stateid(&stp->st_stateid);
2723         }
2724         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2725
2726         if (nfsd4_has_session(&resp->cstate))
2727                 open->op_stateowner->so_confirmed = 1;
2728
2729         /*
2730         * Attempt to hand out a delegation. No error return, because the
2731         * OPEN succeeds even if we fail.
2732         */
2733         nfs4_open_delegation(current_fh, open, stp);
2734
2735         status = nfs_ok;
2736
2737         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2738                 STATEID_VAL(&stp->st_stateid));
2739 out:
2740         if (fp)
2741                 put_nfs4_file(fp);
2742         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2743                 nfs4_set_claim_prev(open);
2744         /*
2745         * To finish the open response, we just need to set the rflags.
2746         */
2747         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2748         if (!open->op_stateowner->so_confirmed &&
2749             !nfsd4_has_session(&resp->cstate))
2750                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2751
2752         return status;
2753 }
2754
2755 __be32
2756 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2757             clientid_t *clid)
2758 {
2759         struct nfs4_client *clp;
2760         __be32 status;
2761
2762         nfs4_lock_state();
2763         dprintk("process_renew(%08x/%08x): starting\n",
2764                         clid->cl_boot, clid->cl_id);
2765         status = nfserr_stale_clientid;
2766         if (STALE_CLIENTID(clid))
2767                 goto out;
2768         clp = find_confirmed_client(clid);
2769         status = nfserr_expired;
2770         if (clp == NULL) {
2771                 /* We assume the client took too long to RENEW. */
2772                 dprintk("nfsd4_renew: clientid not found!\n");
2773                 goto out;
2774         }
2775         renew_client(clp);
2776         status = nfserr_cb_path_down;
2777         if (!list_empty(&clp->cl_delegations)
2778                         && !atomic_read(&clp->cl_cb_set))
2779                 goto out;
2780         status = nfs_ok;
2781 out:
2782         nfs4_unlock_state();
2783         return status;
2784 }
2785
2786 struct lock_manager nfsd4_manager = {
2787 };
2788
2789 static void
2790 nfsd4_end_grace(void)
2791 {
2792         dprintk("NFSD: end of grace period\n");
2793         nfsd4_recdir_purge_old();
2794         locks_end_grace(&nfsd4_manager);
2795         /*
2796          * Now that every NFSv4 client has had the chance to recover and
2797          * to see the (possibly new, possibly shorter) lease time, we
2798          * can safely set the next grace time to the current lease time:
2799          */
2800         nfsd4_grace = nfsd4_lease;
2801 }
2802
2803 static time_t
2804 nfs4_laundromat(void)
2805 {
2806         struct nfs4_client *clp;
2807         struct nfs4_stateowner *sop;
2808         struct nfs4_delegation *dp;
2809         struct list_head *pos, *next, reaplist;
2810         time_t cutoff = get_seconds() - nfsd4_lease;
2811         time_t t, clientid_val = nfsd4_lease;
2812         time_t u, test_val = nfsd4_lease;
2813
2814         nfs4_lock_state();
2815
2816         dprintk("NFSD: laundromat service - starting\n");
2817         if (locks_in_grace())
2818                 nfsd4_end_grace();
2819         INIT_LIST_HEAD(&reaplist);
2820         spin_lock(&client_lock);
2821         list_for_each_safe(pos, next, &client_lru) {
2822                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2823                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2824                         t = clp->cl_time - cutoff;
2825                         if (clientid_val > t)
2826                                 clientid_val = t;
2827                         break;
2828                 }
2829                 if (atomic_read(&clp->cl_refcount)) {
2830                         dprintk("NFSD: client in use (clientid %08x)\n",
2831                                 clp->cl_clientid.cl_id);
2832                         continue;
2833                 }
2834                 unhash_client_locked(clp);
2835                 list_add(&clp->cl_lru, &reaplist);
2836         }
2837         spin_unlock(&client_lock);
2838         list_for_each_safe(pos, next, &reaplist) {
2839                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2840                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2841                         clp->cl_clientid.cl_id);
2842                 nfsd4_remove_clid_dir(clp);
2843                 expire_client(clp);
2844         }
2845         spin_lock(&recall_lock);
2846         list_for_each_safe(pos, next, &del_recall_lru) {
2847                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2848                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2849                         u = dp->dl_time - cutoff;
2850                         if (test_val > u)
2851                                 test_val = u;
2852                         break;
2853                 }
2854                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2855                                     dp, dp->dl_flock);
2856                 list_move(&dp->dl_recall_lru, &reaplist);
2857         }
2858         spin_unlock(&recall_lock);
2859         list_for_each_safe(pos, next, &reaplist) {
2860                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2861                 list_del_init(&dp->dl_recall_lru);
2862                 unhash_delegation(dp);
2863         }
2864         test_val = nfsd4_lease;
2865         list_for_each_safe(pos, next, &close_lru) {
2866                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2867                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2868                         u = sop->so_time - cutoff;
2869                         if (test_val > u)
2870                                 test_val = u;
2871                         break;
2872                 }
2873                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2874                         sop->so_id);
2875                 release_openowner(sop);
2876         }
2877         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2878                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2879         nfs4_unlock_state();
2880         return clientid_val;
2881 }
2882
2883 static struct workqueue_struct *laundry_wq;
2884 static void laundromat_main(struct work_struct *);
2885 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2886
2887 static void
2888 laundromat_main(struct work_struct *not_used)
2889 {
2890         time_t t;
2891
2892         t = nfs4_laundromat();
2893         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2894         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2895 }
2896
2897 static struct nfs4_stateowner *
2898 search_close_lru(u32 st_id, int flags)
2899 {
2900         struct nfs4_stateowner *local = NULL;
2901
2902         if (flags & CLOSE_STATE) {
2903                 list_for_each_entry(local, &close_lru, so_close_lru) {
2904                         if (local->so_id == st_id)
2905                                 return local;
2906                 }
2907         }
2908         return NULL;
2909 }
2910
2911 static inline int
2912 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2913 {
2914         return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2915 }
2916
2917 static int
2918 STALE_STATEID(stateid_t *stateid)
2919 {
2920         if (stateid->si_boot == boot_time)
2921                 return 0;
2922         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2923                 STATEID_VAL(stateid));
2924         return 1;
2925 }
2926
2927 static inline int
2928 access_permit_read(unsigned long access_bmap)
2929 {
2930         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2931                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2932                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2933 }
2934
2935 static inline int
2936 access_permit_write(unsigned long access_bmap)
2937 {
2938         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2939                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2940 }
2941
2942 static
2943 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2944 {
2945         __be32 status = nfserr_openmode;
2946
2947         /* For lock stateid's, we test the parent open, not the lock: */
2948         if (stp->st_openstp)
2949                 stp = stp->st_openstp;
2950         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2951                 goto out;
2952         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2953                 goto out;
2954         status = nfs_ok;
2955 out:
2956         return status;
2957 }
2958
2959 static inline __be32
2960 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2961 {
2962         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2963                 return nfs_ok;
2964         else if (locks_in_grace()) {
2965                 /* Answer in remaining cases depends on existance of
2966                  * conflicting state; so we must wait out the grace period. */
2967                 return nfserr_grace;
2968         } else if (flags & WR_STATE)
2969                 return nfs4_share_conflict(current_fh,
2970                                 NFS4_SHARE_DENY_WRITE);
2971         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2972                 return nfs4_share_conflict(current_fh,
2973                                 NFS4_SHARE_DENY_READ);
2974 }
2975
2976 /*
2977  * Allow READ/WRITE during grace period on recovered state only for files
2978  * that are not able to provide mandatory locking.
2979  */
2980 static inline int
2981 grace_disallows_io(struct inode *inode)
2982 {
2983         return locks_in_grace() && mandatory_lock(inode);
2984 }
2985
2986 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2987 {
2988         /*
2989          * When sessions are used the stateid generation number is ignored
2990          * when it is zero.
2991          */
2992         if ((flags & HAS_SESSION) && in->si_generation == 0)
2993                 goto out;
2994
2995         /* If the client sends us a stateid from the future, it's buggy: */
2996         if (in->si_generation > ref->si_generation)
2997                 return nfserr_bad_stateid;
2998         /*
2999          * The following, however, can happen.  For example, if the
3000          * client sends an open and some IO at the same time, the open
3001          * may bump si_generation while the IO is still in flight.
3002          * Thanks to hard links and renames, the client never knows what
3003          * file an open will affect.  So it could avoid that situation
3004          * only by serializing all opens and IO from the same open
3005          * owner.  To recover from the old_stateid error, the client
3006          * will just have to retry the IO:
3007          */
3008         if (in->si_generation < ref->si_generation)
3009                 return nfserr_old_stateid;
3010 out:
3011         return nfs_ok;
3012 }
3013
3014 static int is_delegation_stateid(stateid_t *stateid)
3015 {
3016         return stateid->si_fileid == 0;
3017 }
3018
3019 /*
3020 * Checks for stateid operations
3021 */
3022 __be32
3023 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3024                            stateid_t *stateid, int flags, struct file **filpp)
3025 {
3026         struct nfs4_stateid *stp = NULL;
3027         struct nfs4_delegation *dp = NULL;
3028         struct svc_fh *current_fh = &cstate->current_fh;
3029         struct inode *ino = current_fh->fh_dentry->d_inode;
3030         __be32 status;
3031
3032         if (filpp)
3033                 *filpp = NULL;
3034
3035         if (grace_disallows_io(ino))
3036                 return nfserr_grace;
3037
3038         if (nfsd4_has_session(cstate))
3039                 flags |= HAS_SESSION;
3040
3041         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3042                 return check_special_stateids(current_fh, stateid, flags);
3043
3044         status = nfserr_stale_stateid;
3045         if (STALE_STATEID(stateid))
3046                 goto out;
3047
3048         /*
3049          * We assume that any stateid that has the current boot time,
3050          * but that we can't find, is expired:
3051          */
3052         status = nfserr_expired;
3053         if (is_delegation_stateid(stateid)) {
3054                 dp = find_delegation_stateid(ino, stateid);
3055                 if (!dp)
3056                         goto out;
3057                 status = check_stateid_generation(stateid, &dp->dl_stateid,
3058                                                   flags);
3059                 if (status)
3060                         goto out;
3061                 status = nfs4_check_delegmode(dp, flags);
3062                 if (status)
3063                         goto out;
3064                 renew_client(dp->dl_client);
3065                 if (filpp)
3066                         *filpp = find_readable_file(dp->dl_file);
3067                 BUG_ON(!*filpp);
3068         } else { /* open or lock stateid */
3069                 stp = find_stateid(stateid, flags);
3070                 if (!stp)
3071                         goto out;
3072                 status = nfserr_bad_stateid;
3073                 if (nfs4_check_fh(current_fh, stp))
3074                         goto out;
3075                 if (!stp->st_stateowner->so_confirmed)
3076                         goto out;
3077                 status = check_stateid_generation(stateid, &stp->st_stateid,
3078                                                   flags);
3079                 if (status)
3080                         goto out;
3081                 status = nfs4_check_openmode(stp, flags);
3082                 if (status)
3083                         goto out;
3084                 renew_client(stp->st_stateowner->so_client);
3085                 if (filpp) {
3086                         if (flags & RD_STATE)
3087                                 *filpp = find_readable_file(stp->st_file);
3088                         else
3089                                 *filpp = find_writeable_file(stp->st_file);
3090                 }
3091         }
3092         status = nfs_ok;
3093 out:
3094         return status;
3095 }
3096
3097 static inline int
3098 setlkflg (int type)
3099 {
3100         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3101                 RD_STATE : WR_STATE;
3102 }
3103
3104 /*
3105  * Checks for sequence id mutating operations.
3106  */
3107 static __be32
3108 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3109                          stateid_t *stateid, int flags,
3110                          struct nfs4_stateowner **sopp,
3111                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3112 {
3113         struct nfs4_stateid *stp;
3114         struct nfs4_stateowner *sop;
3115         struct svc_fh *current_fh = &cstate->current_fh;
3116         __be32 status;
3117
3118         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3119                 seqid, STATEID_VAL(stateid));
3120
3121         *stpp = NULL;
3122         *sopp = NULL;
3123
3124         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3125                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3126                 return nfserr_bad_stateid;
3127         }
3128
3129         if (STALE_STATEID(stateid))
3130                 return nfserr_stale_stateid;
3131
3132         if (nfsd4_has_session(cstate))
3133                 flags |= HAS_SESSION;
3134
3135         /*
3136         * We return BAD_STATEID if filehandle doesn't match stateid,
3137         * the confirmed flag is incorrecly set, or the generation
3138         * number is incorrect.
3139         */
3140         stp = find_stateid(stateid, flags);
3141         if (stp == NULL) {
3142                 /*
3143                  * Also, we should make sure this isn't just the result of
3144                  * a replayed close:
3145                  */
3146                 sop = search_close_lru(stateid->si_stateownerid, flags);
3147                 /* It's not stale; let's assume it's expired: */
3148                 if (sop == NULL)
3149                         return nfserr_expired;
3150                 *sopp = sop;
3151                 goto check_replay;
3152         }
3153
3154         *stpp = stp;
3155         *sopp = sop = stp->st_stateowner;
3156
3157         if (lock) {
3158                 clientid_t *lockclid = &lock->v.new.clientid;
3159                 struct nfs4_client *clp = sop->so_client;
3160                 int lkflg = 0;
3161                 __be32 status;
3162
3163                 lkflg = setlkflg(lock->lk_type);
3164
3165                 if (lock->lk_is_new) {
3166                         if (!sop->so_is_open_owner)
3167                                 return nfserr_bad_stateid;
3168                         if (!(flags & HAS_SESSION) &&
3169                             !same_clid(&clp->cl_clientid, lockclid))
3170                                 return nfserr_bad_stateid;
3171                         /* stp is the open stateid */
3172                         status = nfs4_check_openmode(stp, lkflg);
3173                         if (status)
3174                                 return status;
3175                 } else {
3176                         /* stp is the lock stateid */
3177                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
3178                         if (status)
3179                                 return status;
3180                }
3181         }
3182
3183         if (nfs4_check_fh(current_fh, stp)) {
3184                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3185                 return nfserr_bad_stateid;
3186         }
3187
3188         /*
3189         *  We now validate the seqid and stateid generation numbers.
3190         *  For the moment, we ignore the possibility of
3191         *  generation number wraparound.
3192         */
3193         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3194                 goto check_replay;
3195
3196         if (sop->so_confirmed && flags & CONFIRM) {
3197                 dprintk("NFSD: preprocess_seqid_op: expected"
3198                                 " unconfirmed stateowner!\n");
3199                 return nfserr_bad_stateid;
3200         }
3201         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3202                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3203                                 " confirmed yet!\n");
3204                 return nfserr_bad_stateid;
3205         }
3206         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3207         if (status)
3208                 return status;
3209         renew_client(sop->so_client);
3210         return nfs_ok;
3211
3212 check_replay:
3213         if (seqid == sop->so_seqid - 1) {
3214                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3215                 /* indicate replay to calling function */
3216                 return nfserr_replay_me;
3217         }
3218         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3219                         sop->so_seqid, seqid);
3220         *sopp = NULL;
3221         return nfserr_bad_seqid;
3222 }
3223
3224 __be32
3225 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3226                    struct nfsd4_open_confirm *oc)
3227 {
3228         __be32 status;
3229         struct nfs4_stateowner *sop;
3230         struct nfs4_stateid *stp;
3231
3232         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3233                         (int)cstate->current_fh.fh_dentry->d_name.len,
3234                         cstate->current_fh.fh_dentry->d_name.name);
3235
3236         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3237         if (status)
3238                 return status;
3239
3240         nfs4_lock_state();
3241
3242         if ((status = nfs4_preprocess_seqid_op(cstate,
3243                                         oc->oc_seqid, &oc->oc_req_stateid,
3244                                         CONFIRM | OPEN_STATE,
3245                                         &oc->oc_stateowner, &stp, NULL)))
3246                 goto out;
3247
3248         sop = oc->oc_stateowner;
3249         sop->so_confirmed = 1;
3250         update_stateid(&stp->st_stateid);
3251         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3252         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3253                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3254
3255         nfsd4_create_clid_dir(sop->so_client);
3256 out:
3257         if (oc->oc_stateowner) {
3258                 nfs4_get_stateowner(oc->oc_stateowner);
3259                 cstate->replay_owner = oc->oc_stateowner;
3260         }
3261         nfs4_unlock_state();
3262         return status;
3263 }
3264
3265
3266 /*
3267  * unset all bits in union bitmap (bmap) that
3268  * do not exist in share (from successful OPEN_DOWNGRADE)
3269  */
3270 static void
3271 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3272 {
3273         int i;
3274         for (i = 1; i < 4; i++) {
3275                 if ((i & access) != i)
3276                         __clear_bit(i, bmap);
3277         }
3278 }
3279
3280 static void
3281 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3282 {
3283         int i;
3284         for (i = 0; i < 4; i++) {
3285                 if ((i & deny) != i)
3286                         __clear_bit(i, bmap);
3287         }
3288 }
3289
3290 __be32
3291 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3292                      struct nfsd4_compound_state *cstate,
3293                      struct nfsd4_open_downgrade *od)
3294 {
3295         __be32 status;
3296         struct nfs4_stateid *stp;
3297         unsigned int share_access;
3298
3299         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3300                         (int)cstate->current_fh.fh_dentry->d_name.len,
3301                         cstate->current_fh.fh_dentry->d_name.name);
3302
3303         if (!access_valid(od->od_share_access, cstate->minorversion)
3304                         || !deny_valid(od->od_share_deny))
3305                 return nfserr_inval;
3306
3307         nfs4_lock_state();
3308         if ((status = nfs4_preprocess_seqid_op(cstate,
3309                                         od->od_seqid,
3310                                         &od->od_stateid,
3311                                         OPEN_STATE,
3312                                         &od->od_stateowner, &stp, NULL)))
3313                 goto out;
3314
3315         status = nfserr_inval;
3316         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3317                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3318                         stp->st_access_bmap, od->od_share_access);
3319                 goto out;
3320         }
3321         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3322                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3323                         stp->st_deny_bmap, od->od_share_deny);
3324                 goto out;
3325         }
3326         set_access(&share_access, stp->st_access_bmap);
3327         nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3328
3329         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3330         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3331
3332         update_stateid(&stp->st_stateid);
3333         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3334         status = nfs_ok;
3335 out:
3336         if (od->od_stateowner) {
3337                 nfs4_get_stateowner(od->od_stateowner);
3338                 cstate->replay_owner = od->od_stateowner;
3339         }
3340         nfs4_unlock_state();
3341         return status;
3342 }
3343
3344 /*
3345  * nfs4_unlock_state() called after encode
3346  */
3347 __be32
3348 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3349             struct nfsd4_close *close)
3350 {
3351         __be32 status;
3352         struct nfs4_stateid *stp;
3353
3354         dprintk("NFSD: nfsd4_close on file %.*s\n",
3355                         (int)cstate->current_fh.fh_dentry->d_name.len,
3356                         cstate->current_fh.fh_dentry->d_name.name);
3357
3358         nfs4_lock_state();
3359         /* check close_lru for replay */
3360         if ((status = nfs4_preprocess_seqid_op(cstate,
3361                                         close->cl_seqid,
3362                                         &close->cl_stateid,
3363                                         OPEN_STATE | CLOSE_STATE,
3364                                         &close->cl_stateowner, &stp, NULL)))
3365                 goto out;
3366         status = nfs_ok;
3367         update_stateid(&stp->st_stateid);
3368         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3369
3370         /* release_stateid() calls nfsd_close() if needed */
3371         release_open_stateid(stp);
3372
3373         /* place unused nfs4_stateowners on so_close_lru list to be
3374          * released by the laundromat service after the lease period
3375          * to enable us to handle CLOSE replay
3376          */
3377         if (list_empty(&close->cl_stateowner->so_stateids))
3378                 move_to_close_lru(close->cl_stateowner);
3379 out:
3380         if (close->cl_stateowner) {
3381                 nfs4_get_stateowner(close->cl_stateowner);
3382                 cstate->replay_owner = close->cl_stateowner;
3383         }
3384         nfs4_unlock_state();
3385         return status;
3386 }
3387
3388 __be32
3389 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3390                   struct nfsd4_delegreturn *dr)
3391 {
3392         struct nfs4_delegation *dp;
3393         stateid_t *stateid = &dr->dr_stateid;
3394         struct inode *inode;
3395         __be32 status;
3396         int flags = 0;
3397
3398         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3399                 return status;
3400         inode = cstate->current_fh.fh_dentry->d_inode;
3401
3402         if (nfsd4_has_session(cstate))
3403                 flags |= HAS_SESSION;
3404         nfs4_lock_state();
3405         status = nfserr_bad_stateid;
3406         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3407                 goto out;
3408         status = nfserr_stale_stateid;
3409         if (STALE_STATEID(stateid))
3410                 goto out;
3411         status = nfserr_bad_stateid;
3412         if (!is_delegation_stateid(stateid))
3413                 goto out;
3414         status = nfserr_expired;
3415         dp = find_delegation_stateid(inode, stateid);
3416         if (!dp)
3417                 goto out;
3418         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3419         if (status)
3420                 goto out;
3421         renew_client(dp->dl_client);
3422
3423         unhash_delegation(dp);
3424 out:
3425         nfs4_unlock_state();
3426
3427         return status;
3428 }
3429
3430
3431 /*
3432  * Lock owner state (byte-range locks)
3433  */
3434 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3435 #define LOCK_HASH_BITS              8
3436 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3437 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3438
3439 static inline u64
3440 end_offset(u64 start, u64 len)
3441 {
3442         u64 end;
3443
3444         end = start + len;
3445         return end >= start ? end: NFS4_MAX_UINT64;
3446 }
3447
3448 /* last octet in a range */
3449 static inline u64
3450 last_byte_offset(u64 start, u64 len)
3451 {
3452         u64 end;
3453
3454         BUG_ON(!len);
3455         end = start + len;
3456         return end > start ? end - 1: NFS4_MAX_UINT64;
3457 }
3458
3459 #define lockownerid_hashval(id) \
3460         ((id) & LOCK_HASH_MASK)
3461
3462 static inline unsigned int
3463 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3464                 struct xdr_netobj *ownername)
3465 {
3466         return (file_hashval(inode) + cl_id
3467                         + opaque_hashval(ownername->data, ownername->len))
3468                 & LOCK_HASH_MASK;
3469 }
3470
3471 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3472 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3473 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3474
3475 static struct nfs4_stateid *
3476 find_stateid(stateid_t *stid, int flags)
3477 {
3478         struct nfs4_stateid *local;
3479         u32 st_id = stid->si_stateownerid;
3480         u32 f_id = stid->si_fileid;
3481         unsigned int hashval;
3482
3483         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3484         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3485                 hashval = stateid_hashval(st_id, f_id);
3486                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3487                         if ((local->st_stateid.si_stateownerid == st_id) &&
3488                             (local->st_stateid.si_fileid == f_id))
3489                                 return local;
3490                 }
3491         }
3492
3493         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3494                 hashval = stateid_hashval(st_id, f_id);
3495                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3496                         if ((local->st_stateid.si_stateownerid == st_id) &&
3497                             (local->st_stateid.si_fileid == f_id))
3498                                 return local;
3499                 }
3500         }
3501         return NULL;
3502 }
3503
3504 static struct nfs4_delegation *
3505 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3506 {
3507         struct nfs4_file *fp;
3508         struct nfs4_delegation *dl;
3509
3510         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3511                 STATEID_VAL(stid));
3512
3513         fp = find_file(ino);
3514         if (!fp)
3515                 return NULL;
3516         dl = find_delegation_file(fp, stid);
3517         put_nfs4_file(fp);
3518         return dl;
3519 }
3520
3521 /*
3522  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3523  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3524  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3525  * locking, this prevents us from being completely protocol-compliant.  The
3526  * real solution to this problem is to start using unsigned file offsets in
3527  * the VFS, but this is a very deep change!
3528  */
3529 static inline void
3530 nfs4_transform_lock_offset(struct file_lock *lock)
3531 {
3532         if (lock->fl_start < 0)
3533                 lock->fl_start = OFFSET_MAX;
3534         if (lock->fl_end < 0)
3535                 lock->fl_end = OFFSET_MAX;
3536 }
3537
3538 /* Hack!: For now, we're defining this just so we can use a pointer to it
3539  * as a unique cookie to identify our (NFSv4's) posix locks. */
3540 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3541 };
3542
3543 static inline void
3544 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3545 {
3546         struct nfs4_stateowner *sop;
3547
3548         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3549                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3550                 kref_get(&sop->so_ref);
3551                 deny->ld_sop = sop;
3552                 deny->ld_clientid = sop->so_client->cl_clientid;
3553         } else {
3554                 deny->ld_sop = NULL;
3555                 deny->ld_clientid.cl_boot = 0;
3556                 deny->ld_clientid.cl_id = 0;
3557         }
3558         deny->ld_start = fl->fl_start;
3559         deny->ld_length = NFS4_MAX_UINT64;
3560         if (fl->fl_end != NFS4_MAX_UINT64)
3561                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3562         deny->ld_type = NFS4_READ_LT;
3563         if (fl->fl_type != F_RDLCK)
3564                 deny->ld_type = NFS4_WRITE_LT;
3565 }
3566
3567 static struct nfs4_stateowner *
3568 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3569                 struct xdr_netobj *owner)
3570 {
3571         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3572         struct nfs4_stateowner *op;
3573
3574         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3575                 if (same_owner_str(op, owner, clid))
3576                         return op;
3577         }
3578         return NULL;
3579 }
3580
3581 /*
3582  * Alloc a lock owner structure.
3583  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3584  * occured.
3585  *
3586  * strhashval = lock_ownerstr_hashval
3587  */
3588
3589 static struct nfs4_stateowner *
3590 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3591         struct nfs4_stateowner *sop;
3592         struct nfs4_replay *rp;
3593         unsigned int idhashval;
3594
3595         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3596                 return NULL;
3597         idhashval = lockownerid_hashval(current_ownerid);
3598         INIT_LIST_HEAD(&sop->so_idhash);
3599         INIT_LIST_HEAD(&sop->so_strhash);
3600         INIT_LIST_HEAD(&sop->so_perclient);
3601         INIT_LIST_HEAD(&sop->so_stateids);
3602         INIT_LIST_HEAD(&sop->so_perstateid);
3603         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3604         sop->so_time = 0;
3605         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3606         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3607         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3608         sop->so_is_open_owner = 0;
3609         sop->so_id = current_ownerid++;
3610         sop->so_client = clp;
3611         /* It is the openowner seqid that will be incremented in encode in the
3612          * case of new lockowners; so increment the lock seqid manually: */
3613         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3614         sop->so_confirmed = 1;
3615         rp = &sop->so_replay;
3616         rp->rp_status = nfserr_serverfault;
3617         rp->rp_buflen = 0;
3618         rp->rp_buf = rp->rp_ibuf;
3619         return sop;
3620 }
3621
3622 static struct nfs4_stateid *
3623 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3624 {
3625         struct nfs4_stateid *stp;
3626         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3627
3628         stp = nfs4_alloc_stateid();
3629         if (stp == NULL)
3630                 goto out;
3631         INIT_LIST_HEAD(&stp->st_hash);
3632         INIT_LIST_HEAD(&stp->st_perfile);
3633         INIT_LIST_HEAD(&stp->st_perstateowner);
3634         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3635         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3636         list_add(&stp->st_perfile, &fp->fi_stateids);
3637         list_add(&stp->st_perstateowner, &sop->so_stateids);
3638         stp->st_stateowner = sop;
3639         get_nfs4_file(fp);
3640         stp->st_file = fp;
3641         stp->st_stateid.si_boot = boot_time;
3642         stp->st_stateid.si_stateownerid = sop->so_id;
3643         stp->st_stateid.si_fileid = fp->fi_id;
3644         stp->st_stateid.si_generation = 0;
3645         stp->st_deny_bmap = open_stp->st_deny_bmap;
3646         stp->st_openstp = open_stp;
3647
3648 out:
3649         return stp;
3650 }
3651
3652 static int
3653 check_lock_length(u64 offset, u64 length)
3654 {
3655         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3656              LOFF_OVERFLOW(offset, length)));
3657 }
3658
3659 /*
3660  *  LOCK operation
3661  */
3662 __be32
3663 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3664            struct nfsd4_lock *lock)
3665 {
3666         struct nfs4_stateowner *open_sop = NULL;
3667         struct nfs4_stateowner *lock_sop = NULL;
3668         struct nfs4_stateid *lock_stp;
3669         struct nfs4_file *fp;
3670         struct file *filp = NULL;
3671         struct file_lock file_lock;
3672         struct file_lock conflock;
3673         __be32 status = 0;
3674         unsigned int strhashval;
3675         unsigned int cmd;
3676         int err;
3677
3678         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3679                 (long long) lock->lk_offset,
3680                 (long long) lock->lk_length);
3681
3682         if (check_lock_length(lock->lk_offset, lock->lk_length))
3683                  return nfserr_inval;
3684
3685         if ((status = fh_verify(rqstp, &cstate->current_fh,
3686                                 S_IFREG, NFSD_MAY_LOCK))) {
3687                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3688                 return status;
3689         }
3690
3691         nfs4_lock_state();
3692
3693         if (lock->lk_is_new) {
3694                 /*
3695                  * Client indicates that this is a new lockowner.
3696                  * Use open owner and open stateid to create lock owner and
3697                  * lock stateid.
3698                  */
3699                 struct nfs4_stateid *open_stp = NULL;
3700
3701                 status = nfserr_stale_clientid;
3702                 if (!nfsd4_has_session(cstate) &&
3703                     STALE_CLIENTID(&lock->lk_new_clientid))
3704                         goto out;
3705
3706                 /* validate and update open stateid and open seqid */
3707                 status = nfs4_preprocess_seqid_op(cstate,
3708                                         lock->lk_new_open_seqid,
3709                                         &lock->lk_new_open_stateid,
3710                                         OPEN_STATE,
3711                                         &lock->lk_replay_owner, &open_stp,
3712                                         lock);
3713                 if (status)
3714                         goto out;
3715                 open_sop = lock->lk_replay_owner;
3716                 /* create lockowner and lock stateid */
3717                 fp = open_stp->st_file;
3718                 strhashval = lock_ownerstr_hashval(fp->fi_inode,
3719                                 open_sop->so_client->cl_clientid.cl_id,
3720                                 &lock->v.new.owner);
3721                 /* XXX: Do we need to check for duplicate stateowners on
3722                  * the same file, or should they just be allowed (and
3723                  * create new stateids)? */
3724                 status = nfserr_resource;
3725                 lock_sop = alloc_init_lock_stateowner(strhashval,
3726                                 open_sop->so_client, open_stp, lock);
3727                 if (lock_sop == NULL)
3728                         goto out;
3729                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3730                 if (lock_stp == NULL)
3731                         goto out;
3732         } else {
3733                 /* lock (lock owner + lock stateid) already exists */
3734                 status = nfs4_preprocess_seqid_op(cstate,
3735                                        lock->lk_old_lock_seqid,
3736                                        &lock->lk_old_lock_stateid,
3737                                        LOCK_STATE,
3738                                        &lock->lk_replay_owner, &lock_stp, lock);
3739                 if (status)
3740                         goto out;
3741                 lock_sop = lock->lk_replay_owner;
3742                 fp = lock_stp->st_file;
3743         }
3744         /* lock->lk_replay_owner and lock_stp have been created or found */
3745
3746         status = nfserr_grace;
3747         if (locks_in_grace() && !lock->lk_reclaim)
3748                 goto out;
3749         status = nfserr_no_grace;
3750         if (!locks_in_grace() && lock->lk_reclaim)
3751                 goto out;
3752
3753         locks_init_lock(&file_lock);
3754         switch (lock->lk_type) {
3755                 case NFS4_READ_LT:
3756                 case NFS4_READW_LT:
3757                         if (find_readable_file(lock_stp->st_file)) {
3758                                 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3759                                 filp = find_readable_file(lock_stp->st_file);
3760                         }
3761                         file_lock.fl_type = F_RDLCK;
3762                         cmd = F_SETLK;
3763                 break;
3764                 case NFS4_WRITE_LT:
3765                 case NFS4_WRITEW_LT:
3766                         if (find_writeable_file(lock_stp->st_file)) {
3767                                 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3768                                 filp = find_writeable_file(lock_stp->st_file);
3769                         }
3770                         file_lock.fl_type = F_WRLCK;
3771                         cmd = F_SETLK;
3772                 break;
3773                 default:
3774                         status = nfserr_inval;
3775                 goto out;
3776         }
3777         if (!filp) {
3778                 status = nfserr_openmode;
3779                 goto out;
3780         }
3781         file_lock.fl_owner = (fl_owner_t)lock_sop;
3782         file_lock.fl_pid = current->tgid;
3783         file_lock.fl_file = filp;
3784         file_lock.fl_flags = FL_POSIX;
3785         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3786
3787         file_lock.fl_start = lock->lk_offset;
3788         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3789         nfs4_transform_lock_offset(&file_lock);
3790
3791         /*
3792         * Try to lock the file in the VFS.
3793         * Note: locks.c uses the BKL to protect the inode's lock list.
3794         */
3795
3796         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3797         switch (-err) {
3798         case 0: /* success! */
3799                 update_stateid(&lock_stp->st_stateid);
3800                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
3801                                 sizeof(stateid_t));
3802                 status = 0;
3803                 break;
3804         case (EAGAIN):          /* conflock holds conflicting lock */
3805                 status = nfserr_denied;
3806                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3807                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3808                 break;
3809         case (EDEADLK):
3810                 status = nfserr_deadlock;
3811                 break;
3812         default:
3813                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3814                 status = nfserr_resource;
3815                 break;
3816         }
3817 out:
3818         if (status && lock->lk_is_new && lock_sop)
3819                 release_lockowner(lock_sop);
3820         if (lock->lk_replay_owner) {
3821                 nfs4_get_stateowner(lock->lk_replay_owner);
3822                 cstate->replay_owner = lock->lk_replay_owner;
3823         }
3824         nfs4_unlock_state();
3825         return status;
3826 }
3827
3828 /*
3829  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3830  * so we do a temporary open here just to get an open file to pass to
3831  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3832  * inode operation.)
3833  */
3834 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3835 {
3836         struct file *file;
3837         int err;
3838
3839         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3840         if (err)
3841                 return err;
3842         err = vfs_test_lock(file, lock);
3843         nfsd_close(file);
3844         return err;
3845 }
3846
3847 /*
3848  * LOCKT operation
3849  */
3850 __be32
3851 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3852             struct nfsd4_lockt *lockt)
3853 {
3854         struct inode *inode;
3855         struct file_lock file_lock;
3856         int error;
3857         __be32 status;
3858
3859         if (locks_in_grace())
3860                 return nfserr_grace;
3861
3862         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3863                  return nfserr_inval;
3864
3865         lockt->lt_stateowner = NULL;
3866         nfs4_lock_state();
3867
3868         status = nfserr_stale_clientid;
3869         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3870                 goto out;
3871
3872         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3873                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3874                 if (status == nfserr_symlink)
3875                         status = nfserr_inval;
3876                 goto out;
3877         }
3878
3879         inode = cstate->current_fh.fh_dentry->d_inode;
3880         locks_init_lock(&file_lock);
3881         switch (lockt->lt_type) {
3882                 case NFS4_READ_LT:
3883                 case NFS4_READW_LT:
3884                         file_lock.fl_type = F_RDLCK;
3885                 break;
3886                 case NFS4_WRITE_LT:
3887                 case NFS4_WRITEW_LT:
3888                         file_lock.fl_type = F_WRLCK;
3889                 break;
3890                 default:
3891                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3892                         status = nfserr_inval;
3893                 goto out;
3894         }
3895
3896         lockt->lt_stateowner = find_lockstateowner_str(inode,
3897                         &lockt->lt_clientid, &lockt->lt_owner);
3898         if (lockt->lt_stateowner)
3899                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3900         file_lock.fl_pid = current->tgid;
3901         file_lock.fl_flags = FL_POSIX;
3902
3903         file_lock.fl_start = lockt->lt_offset;
3904         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3905
3906         nfs4_transform_lock_offset(&file_lock);
3907
3908         status = nfs_ok;
3909         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3910         if (error) {
3911                 status = nfserrno(error);
3912                 goto out;
3913         }
3914         if (file_lock.fl_type != F_UNLCK) {
3915                 status = nfserr_denied;
3916                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3917         }
3918 out:
3919         nfs4_unlock_state();
3920         return status;
3921 }
3922
3923 __be32
3924 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3925             struct nfsd4_locku *locku)
3926 {
3927         struct nfs4_stateid *stp;
3928         struct file *filp = NULL;
3929         struct file_lock file_lock;
3930         __be32 status;
3931         int err;
3932
3933         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3934                 (long long) locku->lu_offset,
3935                 (long long) locku->lu_length);
3936
3937         if (check_lock_length(locku->lu_offset, locku->lu_length))
3938                  return nfserr_inval;
3939
3940         nfs4_lock_state();
3941
3942         if ((status = nfs4_preprocess_seqid_op(cstate,
3943                                         locku->lu_seqid,
3944                                         &locku->lu_stateid,
3945                                         LOCK_STATE,
3946                                         &locku->lu_stateowner, &stp, NULL)))
3947                 goto out;
3948
3949         filp = find_any_file(stp->st_file);
3950         if (!filp) {
3951                 status = nfserr_lock_range;
3952                 goto out;
3953         }
3954         BUG_ON(!filp);
3955         locks_init_lock(&file_lock);
3956         file_lock.fl_type = F_UNLCK;
3957         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3958         file_lock.fl_pid = current->tgid;
3959         file_lock.fl_file = filp;
3960         file_lock.fl_flags = FL_POSIX;
3961         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3962         file_lock.fl_start = locku->lu_offset;
3963
3964         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3965         nfs4_transform_lock_offset(&file_lock);
3966
3967         /*
3968         *  Try to unlock the file in the VFS.
3969         */
3970         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3971         if (err) {
3972                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3973                 goto out_nfserr;
3974         }
3975         /*
3976         * OK, unlock succeeded; the only thing left to do is update the stateid.
3977         */
3978         update_stateid(&stp->st_stateid);
3979         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3980
3981 out:
3982         if (locku->lu_stateowner) {
3983                 nfs4_get_stateowner(locku->lu_stateowner);
3984                 cstate->replay_owner = locku->lu_stateowner;
3985         }
3986         nfs4_unlock_state();
3987         return status;
3988
3989 out_nfserr:
3990         status = nfserrno(err);
3991         goto out;
3992 }
3993
3994 /*
3995  * returns
3996  *      1: locks held by lockowner
3997  *      0: no locks held by lockowner
3998  */
3999 static int
4000 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4001 {
4002         struct file_lock **flpp;
4003         struct inode *inode = filp->fi_inode;
4004         int status = 0;
4005
4006         lock_kernel();
4007         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4008                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4009                         status = 1;
4010                         goto out;
4011                 }
4012         }
4013 out:
4014         unlock_kernel();
4015         return status;
4016 }
4017
4018 __be32
4019 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4020                         struct nfsd4_compound_state *cstate,
4021                         struct nfsd4_release_lockowner *rlockowner)
4022 {
4023         clientid_t *clid = &rlockowner->rl_clientid;
4024         struct nfs4_stateowner *sop;
4025         struct nfs4_stateid *stp;
4026         struct xdr_netobj *owner = &rlockowner->rl_owner;
4027         struct list_head matches;
4028         int i;
4029         __be32 status;
4030
4031         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4032                 clid->cl_boot, clid->cl_id);
4033
4034         /* XXX check for lease expiration */
4035
4036         status = nfserr_stale_clientid;
4037         if (STALE_CLIENTID(clid))
4038                 return status;
4039
4040         nfs4_lock_state();
4041
4042         status = nfserr_locks_held;
4043         /* XXX: we're doing a linear search through all the lockowners.
4044          * Yipes!  For now we'll just hope clients aren't really using
4045          * release_lockowner much, but eventually we have to fix these
4046          * data structures. */
4047         INIT_LIST_HEAD(&matches);
4048         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4049                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4050                         if (!same_owner_str(sop, owner, clid))
4051                                 continue;
4052                         list_for_each_entry(stp, &sop->so_stateids,
4053                                         st_perstateowner) {
4054                                 if (check_for_locks(stp->st_file, sop))
4055                                         goto out;
4056                                 /* Note: so_perclient unused for lockowners,
4057                                  * so it's OK to fool with here. */
4058                                 list_add(&sop->so_perclient, &matches);
4059                         }
4060                 }
4061         }
4062         /* Clients probably won't expect us to return with some (but not all)
4063          * of the lockowner state released; so don't release any until all
4064          * have been checked. */
4065         status = nfs_ok;
4066         while (!list_empty(&matches)) {
4067                 sop = list_entry(matches.next, struct nfs4_stateowner,
4068                                                                 so_perclient);
4069                 /* unhash_stateowner deletes so_perclient only
4070                  * for openowners. */
4071                 list_del(&sop->so_perclient);
4072                 release_lockowner(sop);
4073         }
4074 out:
4075         nfs4_unlock_state();
4076         return status;
4077 }
4078
4079 static inline struct nfs4_client_reclaim *
4080 alloc_reclaim(void)
4081 {
4082         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4083 }
4084
4085 int
4086 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4087 {
4088         unsigned int strhashval = clientstr_hashval(name);
4089         struct nfs4_client *clp;
4090
4091         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
4092         return clp ? 1 : 0;
4093 }
4094
4095 /*
4096  * failure => all reset bets are off, nfserr_no_grace...
4097  */
4098 int
4099 nfs4_client_to_reclaim(const char *name)
4100 {
4101         unsigned int strhashval;
4102         struct nfs4_client_reclaim *crp = NULL;
4103
4104         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4105         crp = alloc_reclaim();
4106         if (!crp)
4107                 return 0;
4108         strhashval = clientstr_hashval(name);
4109         INIT_LIST_HEAD(&crp->cr_strhash);
4110         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4111         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4112         reclaim_str_hashtbl_size++;
4113         return 1;
4114 }
4115
4116 static void
4117 nfs4_release_reclaim(void)
4118 {
4119         struct nfs4_client_reclaim *crp = NULL;
4120         int i;
4121
4122         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4123                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4124                         crp = list_entry(reclaim_str_hashtbl[i].next,
4125                                         struct nfs4_client_reclaim, cr_strhash);
4126                         list_del(&crp->cr_strhash);
4127                         kfree(crp);
4128                         reclaim_str_hashtbl_size--;
4129                 }
4130         }
4131         BUG_ON(reclaim_str_hashtbl_size);
4132 }
4133
4134 /*
4135  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4136 static struct nfs4_client_reclaim *
4137 nfs4_find_reclaim_client(clientid_t *clid)
4138 {
4139         unsigned int strhashval;
4140         struct nfs4_client *clp;
4141         struct nfs4_client_reclaim *crp = NULL;
4142
4143
4144         /* find clientid in conf_id_hashtbl */
4145         clp = find_confirmed_client(clid);
4146         if (clp == NULL)
4147                 return NULL;
4148
4149         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4150                             clp->cl_name.len, clp->cl_name.data,
4151                             clp->cl_recdir);
4152
4153         /* find clp->cl_name in reclaim_str_hashtbl */
4154         strhashval = clientstr_hashval(clp->cl_recdir);
4155         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4156                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4157                         return crp;
4158                 }
4159         }
4160         return NULL;
4161 }
4162
4163 /*
4164 * Called from OPEN. Look for clientid in reclaim list.
4165 */
4166 __be32
4167 nfs4_check_open_reclaim(clientid_t *clid)
4168 {
4169         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4170 }
4171
4172 /* initialization to perform at module load time: */
4173
4174 int
4175 nfs4_state_init(void)
4176 {
4177         int i, status;
4178
4179         status = nfsd4_init_slabs();
4180         if (status)
4181                 return status;
4182         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4183                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4184                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4185                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4186                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4187                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4188         }
4189         for (i = 0; i < SESSION_HASH_SIZE; i++)
4190                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4191         for (i = 0; i < FILE_HASH_SIZE; i++) {
4192                 INIT_LIST_HEAD(&file_hashtbl[i]);
4193         }
4194         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4195                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4196                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4197         }
4198         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4199                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4200                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4201         }
4202         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4203                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4204                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4205         }
4206         memset(&onestateid, ~0, sizeof(stateid_t));
4207         INIT_LIST_HEAD(&close_lru);
4208         INIT_LIST_HEAD(&client_lru);
4209         INIT_LIST_HEAD(&del_recall_lru);
4210         reclaim_str_hashtbl_size = 0;
4211         return 0;
4212 }
4213
4214 static void
4215 nfsd4_load_reboot_recovery_data(void)
4216 {
4217         int status;
4218
4219         nfs4_lock_state();
4220         nfsd4_init_recdir(user_recovery_dirname);
4221         status = nfsd4_recdir_load();
4222         nfs4_unlock_state();
4223         if (status)
4224                 printk("NFSD: Failure reading reboot recovery data\n");
4225 }
4226
4227 /*
4228  * Since the lifetime of a delegation isn't limited to that of an open, a
4229  * client may quite reasonably hang on to a delegation as long as it has
4230  * the inode cached.  This becomes an obvious problem the first time a
4231  * client's inode cache approaches the size of the server's total memory.
4232  *
4233  * For now we avoid this problem by imposing a hard limit on the number
4234  * of delegations, which varies according to the server's memory size.
4235  */
4236 static void
4237 set_max_delegations(void)
4238 {
4239         /*
4240          * Allow at most 4 delegations per megabyte of RAM.  Quick
4241          * estimates suggest that in the worst case (where every delegation
4242          * is for a different inode), a delegation could take about 1.5K,
4243          * giving a worst case usage of about 6% of memory.
4244          */
4245         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4246 }
4247
4248 /* initialization to perform when the nfsd service is started: */
4249
4250 static int
4251 __nfs4_state_start(void)
4252 {
4253         int ret;
4254
4255         boot_time = get_seconds();
4256         locks_start_grace(&nfsd4_manager);
4257         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4258                nfsd4_grace);
4259         ret = set_callback_cred();
4260         if (ret)
4261                 return -ENOMEM;
4262         laundry_wq = create_singlethread_workqueue("nfsd4");
4263         if (laundry_wq == NULL)
4264                 return -ENOMEM;
4265         ret = nfsd4_create_callback_queue();
4266         if (ret)
4267                 goto out_free_laundry;
4268         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4269         set_max_delegations();
4270         return 0;
4271 out_free_laundry:
4272         destroy_workqueue(laundry_wq);
4273         return ret;
4274 }
4275
4276 int
4277 nfs4_state_start(void)
4278 {
4279         nfsd4_load_reboot_recovery_data();
4280         return __nfs4_state_start();
4281 }
4282
4283 static void
4284 __nfs4_state_shutdown(void)
4285 {
4286         int i;
4287         struct nfs4_client *clp = NULL;
4288         struct nfs4_delegation *dp = NULL;
4289         struct list_head *pos, *next, reaplist;
4290
4291         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4292                 while (!list_empty(&conf_id_hashtbl[i])) {
4293                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4294                         expire_client(clp);
4295                 }
4296                 while (!list_empty(&unconf_str_hashtbl[i])) {
4297                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4298                         expire_client(clp);
4299                 }
4300         }
4301         INIT_LIST_HEAD(&reaplist);
4302         spin_lock(&recall_lock);
4303         list_for_each_safe(pos, next, &del_recall_lru) {
4304                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4305                 list_move(&dp->dl_recall_lru, &reaplist);
4306         }
4307         spin_unlock(&recall_lock);
4308         list_for_each_safe(pos, next, &reaplist) {
4309                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4310                 list_del_init(&dp->dl_recall_lru);
4311                 unhash_delegation(dp);
4312         }
4313
4314         nfsd4_shutdown_recdir();
4315 }
4316
4317 void
4318 nfs4_state_shutdown(void)
4319 {
4320         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4321         destroy_workqueue(laundry_wq);
4322         locks_end_grace(&nfsd4_manager);
4323         nfs4_lock_state();
4324         nfs4_release_reclaim();
4325         __nfs4_state_shutdown();
4326         nfs4_unlock_state();
4327         nfsd4_destroy_callback_queue();
4328 }
4329
4330 /*
4331  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4332  * accessed when nfsd is starting.
4333  */
4334 static void
4335 nfs4_set_recdir(char *recdir)
4336 {
4337         strcpy(user_recovery_dirname, recdir);
4338 }
4339
4340 /*
4341  * Change the NFSv4 recovery directory to recdir.
4342  */
4343 int
4344 nfs4_reset_recoverydir(char *recdir)
4345 {
4346         int status;
4347         struct path path;
4348
4349         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4350         if (status)
4351                 return status;
4352         status = -ENOTDIR;
4353         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4354                 nfs4_set_recdir(recdir);
4355                 status = 0;
4356         }
4357         path_put(&path);
4358         return status;
4359 }
4360
4361 char *
4362 nfs4_recoverydir(void)
4363 {
4364         return user_recovery_dirname;
4365 }