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/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include "xdr4.h"
  45 #include "xdr4cb.h"
  46 #include "vfs.h"
  47 #include "current_stateid.h"
  48
  49 #include "netns.h"
  50
  51 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  52
  53 #define all_ones {{~0,~0},~0}
  54 static const stateid_t one_stateid = {
  55         .si_generation = ~0,
  56         .si_opaque = all_ones,
  57 };
  58 static const stateid_t zero_stateid = {
  59         /* all fields zero */
  60 };
  61 static const stateid_t currentstateid = {
  62         .si_generation = 1,
  63 };
  64
  65 static u64 current_sessionid = 1;
  66
  67 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  68 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  69 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  70
  71 /* forward declarations */
  72 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
  73
  74 /* Locking: */
  75
  76 /* Currently used for almost all code touching nfsv4 state: */
  77 static DEFINE_MUTEX(client_mutex);
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(recall_lock);
  85
  86 static struct kmem_cache *openowner_slab = NULL;
  87 static struct kmem_cache *lockowner_slab = NULL;
  88 static struct kmem_cache *file_slab = NULL;
  89 static struct kmem_cache *stateid_slab = NULL;
  90 static struct kmem_cache *deleg_slab = NULL;
  91
  92 void
  93 nfs4_lock_state(void)
  94 {
  95         mutex_lock(&client_mutex);
  96 }
  97
  98 static void free_session(struct nfsd4_session *);
  99
 100 void nfsd4_put_session(struct nfsd4_session *ses)
 101 {
 102         atomic_dec(&ses->se_ref);
 103 }
 104
 105 static bool is_session_dead(struct nfsd4_session *ses)
 106 {
 107         return ses->se_flags & NFS4_SESSION_DEAD;
 108 }
 109
 110 static __be32 mark_session_dead_locked(struct nfsd4_session *ses)
 111 {
 112         if (atomic_read(&ses->se_ref))
 113                 return nfserr_jukebox;
 114         ses->se_flags |= NFS4_SESSION_DEAD;
 115         return nfs_ok;
 116 }
 117
 118 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 119 {
 120         if (is_session_dead(ses))
 121                 return nfserr_badsession;
 122         atomic_inc(&ses->se_ref);
 123         return nfs_ok;
 124 }
 125
 126 void
 127 nfs4_unlock_state(void)
 128 {
 129         mutex_unlock(&client_mutex);
 130 }
 131
 132 static bool is_client_expired(struct nfs4_client *clp)
 133 {
 134         return clp->cl_time == 0;
 135 }
 136
 137 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
 138 {
 139         if (atomic_read(&clp->cl_refcount))
 140                 return nfserr_jukebox;
 141         clp->cl_time = 0;
 142         return nfs_ok;
 143 }
 144
 145 static __be32 mark_client_expired(struct nfs4_client *clp)
 146 {
 147         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 148         __be32 ret;
 149
 150         spin_lock(&nn->client_lock);
 151         ret = mark_client_expired_locked(clp);
 152         spin_unlock(&nn->client_lock);
 153         return ret;
 154 }
 155
 156 static __be32 get_client_locked(struct nfs4_client *clp)
 157 {
 158         if (is_client_expired(clp))
 159                 return nfserr_expired;
 160         atomic_inc(&clp->cl_refcount);
 161         return nfs_ok;
 162 }
 163
 164 /* must be called under the client_lock */
 165 static inline void
 166 renew_client_locked(struct nfs4_client *clp)
 167 {
 168         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 169
 170         if (is_client_expired(clp)) {
 171                 WARN_ON(1);
 172                 printk("%s: client (clientid %08x/%08x) already expired\n",
 173                         __func__,
 174                         clp->cl_clientid.cl_boot,
 175                         clp->cl_clientid.cl_id);
 176                 return;
 177         }
 178
 179         dprintk("renewing client (clientid %08x/%08x)\n",
 180                         clp->cl_clientid.cl_boot,
 181                         clp->cl_clientid.cl_id);
 182         list_move_tail(&clp->cl_lru, &nn->client_lru);
 183         clp->cl_time = get_seconds();
 184 }
 185
 186 static inline void
 187 renew_client(struct nfs4_client *clp)
 188 {
 189         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 190
 191         spin_lock(&nn->client_lock);
 192         renew_client_locked(clp);
 193         spin_unlock(&nn->client_lock);
 194 }
 195
 196 void put_client_renew_locked(struct nfs4_client *clp)
 197 {
 198         if (!atomic_dec_and_test(&clp->cl_refcount))
 199                 return;
 200         if (!is_client_expired(clp))
 201                 renew_client_locked(clp);
 202 }
 203
 204 void put_client_renew(struct nfs4_client *clp)
 205 {
 206         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 207
 208         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 209                 return;
 210         if (!is_client_expired(clp))
 211                 renew_client_locked(clp);
 212         spin_unlock(&nn->client_lock);
 213 }
 214
 215
 216 static inline u32
 217 opaque_hashval(const void *ptr, int nbytes)
 218 {
 219         unsigned char *cptr = (unsigned char *) ptr;
 220
 221         u32 x = 0;
 222         while (nbytes--) {
 223                 x *= 37;
 224                 x += *cptr++;
 225         }
 226         return x;
 227 }
 228
 229 static void nfsd4_free_file(struct nfs4_file *f)
 230 {
 231         kmem_cache_free(file_slab, f);
 232 }
 233
 234 static inline void
 235 put_nfs4_file(struct nfs4_file *fi)
 236 {
 237         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 238                 hlist_del(&fi->fi_hash);
 239                 spin_unlock(&recall_lock);
 240                 iput(fi->fi_inode);
 241                 nfsd4_free_file(fi);
 242         }
 243 }
 244
 245 static inline void
 246 get_nfs4_file(struct nfs4_file *fi)
 247 {
 248         atomic_inc(&fi->fi_ref);
 249 }
 250
 251 static int num_delegations;
 252 unsigned long max_delegations;
 253
 254 /*
 255  * Open owner state (share locks)
 256  */
 257
 258 /* hash tables for lock and open owners */
 259 #define OWNER_HASH_BITS              8
 260 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 261 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 262
 263 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 264 {
 265         unsigned int ret;
 266
 267         ret = opaque_hashval(ownername->data, ownername->len);
 268         ret += clientid;
 269         return ret & OWNER_HASH_MASK;
 270 }
 271
 272 /* hash table for nfs4_file */
 273 #define FILE_HASH_BITS                   8
 274 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 275
 276 static unsigned int file_hashval(struct inode *ino)
 277 {
 278         /* XXX: why are we hashing on inode pointer, anyway? */
 279         return hash_ptr(ino, FILE_HASH_BITS);
 280 }
 281
 282 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 283
 284 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 285 {
 286         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 287         atomic_inc(&fp->fi_access[oflag]);
 288 }
 289
 290 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 291 {
 292         if (oflag == O_RDWR) {
 293                 __nfs4_file_get_access(fp, O_RDONLY);
 294                 __nfs4_file_get_access(fp, O_WRONLY);
 295         } else
 296                 __nfs4_file_get_access(fp, oflag);
 297 }
 298
 299 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 300 {
 301         if (fp->fi_fds[oflag]) {
 302                 fput(fp->fi_fds[oflag]);
 303                 fp->fi_fds[oflag] = NULL;
 304         }
 305 }
 306
 307 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 308 {
 309         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 310                 nfs4_file_put_fd(fp, oflag);
 311                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 312                         nfs4_file_put_fd(fp, O_RDWR);
 313         }
 314 }
 315
 316 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 317 {
 318         if (oflag == O_RDWR) {
 319                 __nfs4_file_put_access(fp, O_RDONLY);
 320                 __nfs4_file_put_access(fp, O_WRONLY);
 321         } else
 322                 __nfs4_file_put_access(fp, oflag);
 323 }
 324
 325 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct
 326 kmem_cache *slab)
 327 {
 328         struct idr *stateids = &cl->cl_stateids;
 329         static int min_stateid = 0;
 330         struct nfs4_stid *stid;
 331         int new_id;
 332
 333         stid = kmem_cache_alloc(slab, GFP_KERNEL);
 334         if (!stid)
 335                 return NULL;
 336
 337         new_id = idr_alloc(stateids, stid, min_stateid, 0, GFP_KERNEL);
 338         if (new_id < 0)
 339                 goto out_free;
 340         stid->sc_client = cl;
 341         stid->sc_type = 0;
 342         stid->sc_stateid.si_opaque.so_id = new_id;
 343         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 344         /* Will be incremented before return to client: */
 345         stid->sc_stateid.si_generation = 0;
 346
 347         /*
 348          * It shouldn't be a problem to reuse an opaque stateid value.
 349          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 350          * example, a stray write retransmission could be accepted by
 351          * the server when it should have been rejected.  Therefore,
 352          * adopt a trick from the sctp code to attempt to maximize the
 353          * amount of time until an id is reused, by ensuring they always
 354          * "increase" (mod INT_MAX):
 355          */
 356
 357         min_stateid = new_id+1;
 358         if (min_stateid == INT_MAX)
 359                 min_stateid = 0;
 360         return stid;
 361 out_free:
 362         kmem_cache_free(slab, stid);
 363         return NULL;
 364 }
 365
 366 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 367 {
 368         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 369 }
 370
 371 static struct nfs4_delegation *
 372 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
 373 {
 374         struct nfs4_delegation *dp;
 375         struct nfs4_file *fp = stp->st_file;
 376
 377         dprintk("NFSD alloc_init_deleg\n");
 378         /*
 379          * Major work on the lease subsystem (for example, to support
 380          * calbacks on stat) will be required before we can support
 381          * write delegations properly.
 382          */
 383         if (type != NFS4_OPEN_DELEGATE_READ)
 384                 return NULL;
 385         if (fp->fi_had_conflict)
 386                 return NULL;
 387         if (num_delegations > max_delegations)
 388                 return NULL;
 389         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 390         if (dp == NULL)
 391                 return dp;
 392         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 393         /*
 394          * delegation seqid's are never incremented.  The 4.1 special
 395          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 396          * 0 anyway just for consistency and use 1:
 397          */
 398         dp->dl_stid.sc_stateid.si_generation = 1;
 399         num_delegations++;
 400         INIT_LIST_HEAD(&dp->dl_perfile);
 401         INIT_LIST_HEAD(&dp->dl_perclnt);
 402         INIT_LIST_HEAD(&dp->dl_recall_lru);
 403         get_nfs4_file(fp);
 404         dp->dl_file = fp;
 405         dp->dl_type = type;
 406         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 407         dp->dl_time = 0;
 408         atomic_set(&dp->dl_count, 1);
 409         nfsd4_init_callback(&dp->dl_recall);
 410         return dp;
 411 }
 412
 413 static void remove_stid(struct nfs4_stid *s)
 414 {
 415         struct idr *stateids = &s->sc_client->cl_stateids;
 416
 417         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 418 }
 419
 420 void
 421 nfs4_put_delegation(struct nfs4_delegation *dp)
 422 {
 423         if (atomic_dec_and_test(&dp->dl_count)) {
 424                 kmem_cache_free(deleg_slab, dp);
 425                 num_delegations--;
 426         }
 427 }
 428
 429 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 430 {
 431         if (atomic_dec_and_test(&fp->fi_delegees)) {
 432                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 433                 fp->fi_lease = NULL;
 434                 fput(fp->fi_deleg_file);
 435                 fp->fi_deleg_file = NULL;
 436         }
 437 }
 438
 439 static void unhash_stid(struct nfs4_stid *s)
 440 {
 441         s->sc_type = 0;
 442 }
 443
 444 /* Called under the state lock. */
 445 static void
 446 unhash_delegation(struct nfs4_delegation *dp)
 447 {
 448         unhash_stid(&dp->dl_stid);
 449         list_del_init(&dp->dl_perclnt);
 450         spin_lock(&recall_lock);
 451         list_del_init(&dp->dl_perfile);
 452         list_del_init(&dp->dl_recall_lru);
 453         spin_unlock(&recall_lock);
 454         nfs4_put_deleg_lease(dp->dl_file);
 455         put_nfs4_file(dp->dl_file);
 456         dp->dl_file = NULL;
 457         remove_stid(&dp->dl_stid);
 458         nfs4_put_delegation(dp);
 459 }
 460
 461 /*
 462  * SETCLIENTID state
 463  */
 464
 465 static unsigned int clientid_hashval(u32 id)
 466 {
 467         return id & CLIENT_HASH_MASK;
 468 }
 469
 470 static unsigned int clientstr_hashval(const char *name)
 471 {
 472         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 473 }
 474
 475 /*
 476  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 477  * st_{access,deny}_bmap field of the stateid, in order to track not
 478  * only what share bits are currently in force, but also what
 479  * combinations of share bits previous opens have used.  This allows us
 480  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 481  * return an error if the client attempt to downgrade to a combination
 482  * of share bits not explicable by closing some of its previous opens.
 483  *
 484  * XXX: This enforcement is actually incomplete, since we don't keep
 485  * track of access/deny bit combinations; so, e.g., we allow:
 486  *
 487  *      OPEN allow read, deny write
 488  *      OPEN allow both, deny none
 489  *      DOWNGRADE allow read, deny none
 490  *
 491  * which we should reject.
 492  */
 493 static unsigned int
 494 bmap_to_share_mode(unsigned long bmap) {
 495         int i;
 496         unsigned int access = 0;
 497
 498         for (i = 1; i < 4; i++) {
 499                 if (test_bit(i, &bmap))
 500                         access |= i;
 501         }
 502         return access;
 503 }
 504
 505 static bool
 506 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 507         unsigned int access, deny;
 508
 509         access = bmap_to_share_mode(stp->st_access_bmap);
 510         deny = bmap_to_share_mode(stp->st_deny_bmap);
 511         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 512                 return false;
 513         return true;
 514 }
 515
 516 /* set share access for a given stateid */
 517 static inline void
 518 set_access(u32 access, struct nfs4_ol_stateid *stp)
 519 {
 520         __set_bit(access, &stp->st_access_bmap);
 521 }
 522
 523 /* clear share access for a given stateid */
 524 static inline void
 525 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 526 {
 527         __clear_bit(access, &stp->st_access_bmap);
 528 }
 529
 530 /* test whether a given stateid has access */
 531 static inline bool
 532 test_access(u32 access, struct nfs4_ol_stateid *stp)
 533 {
 534         return test_bit(access, &stp->st_access_bmap);
 535 }
 536
 537 /* set share deny for a given stateid */
 538 static inline void
 539 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 540 {
 541         __set_bit(access, &stp->st_deny_bmap);
 542 }
 543
 544 /* clear share deny for a given stateid */
 545 static inline void
 546 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 547 {
 548         __clear_bit(access, &stp->st_deny_bmap);
 549 }
 550
 551 /* test whether a given stateid is denying specific access */
 552 static inline bool
 553 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 554 {
 555         return test_bit(access, &stp->st_deny_bmap);
 556 }
 557
 558 static int nfs4_access_to_omode(u32 access)
 559 {
 560         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 561         case NFS4_SHARE_ACCESS_READ:
 562                 return O_RDONLY;
 563         case NFS4_SHARE_ACCESS_WRITE:
 564                 return O_WRONLY;
 565         case NFS4_SHARE_ACCESS_BOTH:
 566                 return O_RDWR;
 567         }
 568         WARN_ON_ONCE(1);
 569         return O_RDONLY;
 570 }
 571
 572 /* release all access and file references for a given stateid */
 573 static void
 574 release_all_access(struct nfs4_ol_stateid *stp)
 575 {
 576         int i;
 577
 578         for (i = 1; i < 4; i++) {
 579                 if (test_access(i, stp))
 580                         nfs4_file_put_access(stp->st_file,
 581                                              nfs4_access_to_omode(i));
 582                 clear_access(i, stp);
 583         }
 584 }
 585
 586 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 587 {
 588         list_del(&stp->st_perfile);
 589         list_del(&stp->st_perstateowner);
 590 }
 591
 592 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 593 {
 594         release_all_access(stp);
 595         put_nfs4_file(stp->st_file);
 596         stp->st_file = NULL;
 597 }
 598
 599 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 600 {
 601         remove_stid(&stp->st_stid);
 602         kmem_cache_free(stateid_slab, stp);
 603 }
 604
 605 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 606 {
 607         struct file *file;
 608
 609         unhash_generic_stateid(stp);
 610         unhash_stid(&stp->st_stid);
 611         file = find_any_file(stp->st_file);
 612         if (file)
 613                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 614         close_generic_stateid(stp);
 615         free_generic_stateid(stp);
 616 }
 617
 618 static void unhash_lockowner(struct nfs4_lockowner *lo)
 619 {
 620         struct nfs4_ol_stateid *stp;
 621
 622         list_del(&lo->lo_owner.so_strhash);
 623         list_del(&lo->lo_perstateid);
 624         list_del(&lo->lo_owner_ino_hash);
 625         while (!list_empty(&lo->lo_owner.so_stateids)) {
 626                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 627                                 struct nfs4_ol_stateid, st_perstateowner);
 628                 release_lock_stateid(stp);
 629         }
 630 }
 631
 632 static void release_lockowner(struct nfs4_lockowner *lo)
 633 {
 634         unhash_lockowner(lo);
 635         nfs4_free_lockowner(lo);
 636 }
 637
 638 static void
 639 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 640 {
 641         struct nfs4_lockowner *lo;
 642
 643         while (!list_empty(&open_stp->st_lockowners)) {
 644                 lo = list_entry(open_stp->st_lockowners.next,
 645                                 struct nfs4_lockowner, lo_perstateid);
 646                 release_lockowner(lo);
 647         }
 648 }
 649
 650 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 651 {
 652         unhash_generic_stateid(stp);
 653         release_stateid_lockowners(stp);
 654         close_generic_stateid(stp);
 655 }
 656
 657 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 658 {
 659         unhash_open_stateid(stp);
 660         unhash_stid(&stp->st_stid);
 661         free_generic_stateid(stp);
 662 }
 663
 664 static void unhash_openowner(struct nfs4_openowner *oo)
 665 {
 666         struct nfs4_ol_stateid *stp;
 667
 668         list_del(&oo->oo_owner.so_strhash);
 669         list_del(&oo->oo_perclient);
 670         while (!list_empty(&oo->oo_owner.so_stateids)) {
 671                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 672                                 struct nfs4_ol_stateid, st_perstateowner);
 673                 release_open_stateid(stp);
 674         }
 675 }
 676
 677 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 678 {
 679         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 680
 681         if (s) {
 682                 unhash_stid(&s->st_stid);
 683                 free_generic_stateid(s);
 684                 oo->oo_last_closed_stid = NULL;
 685         }
 686 }
 687
 688 static void release_openowner(struct nfs4_openowner *oo)
 689 {
 690         unhash_openowner(oo);
 691         list_del(&oo->oo_close_lru);
 692         release_last_closed_stateid(oo);
 693         nfs4_free_openowner(oo);
 694 }
 695
 696 static inline int
 697 hash_sessionid(struct nfs4_sessionid *sessionid)
 698 {
 699         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 700
 701         return sid->sequence % SESSION_HASH_SIZE;
 702 }
 703
 704 #ifdef NFSD_DEBUG
 705 static inline void
 706 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 707 {
 708         u32 *ptr = (u32 *)(&sessionid->data[0]);
 709         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 710 }
 711 #else
 712 static inline void
 713 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 714 {
 715 }
 716 #endif
 717
 718 /*
 719  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
 720  * won't be used for replay.
 721  */
 722 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
 723 {
 724         struct nfs4_stateowner *so = cstate->replay_owner;
 725
 726         if (nfserr == nfserr_replay_me)
 727                 return;
 728
 729         if (!seqid_mutating_err(ntohl(nfserr))) {
 730                 cstate->replay_owner = NULL;
 731                 return;
 732         }
 733         if (!so)
 734                 return;
 735         if (so->so_is_open_owner)
 736                 release_last_closed_stateid(openowner(so));
 737         so->so_seqid++;
 738         return;
 739 }
 740
 741 static void
 742 gen_sessionid(struct nfsd4_session *ses)
 743 {
 744         struct nfs4_client *clp = ses->se_client;
 745         struct nfsd4_sessionid *sid;
 746
 747         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 748         sid->clientid = clp->cl_clientid;
 749         sid->sequence = current_sessionid++;
 750         sid->reserved = 0;
 751 }
 752
 753 /*
 754  * The protocol defines ca_maxresponssize_cached to include the size of
 755  * the rpc header, but all we need to cache is the data starting after
 756  * the end of the initial SEQUENCE operation--the rest we regenerate
 757  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 758  * value that is the number of bytes in our cache plus a few additional
 759  * bytes.  In order to stay on the safe side, and not promise more than
 760  * we can cache, those additional bytes must be the minimum possible: 24
 761  * bytes of rpc header (xid through accept state, with AUTH_NULL
 762  * verifier), 12 for the compound header (with zero-length tag), and 44
 763  * for the SEQUENCE op response:
 764  */
 765 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 766
 767 static void
 768 free_session_slots(struct nfsd4_session *ses)
 769 {
 770         int i;
 771
 772         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 773                 kfree(ses->se_slots[i]);
 774 }
 775
 776 /*
 777  * We don't actually need to cache the rpc and session headers, so we
 778  * can allocate a little less for each slot:
 779  */
 780 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
 781 {
 782         u32 size;
 783
 784         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
 785                 size = 0;
 786         else
 787                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 788         return size + sizeof(struct nfsd4_slot);
 789 }
 790
 791 /*
 792  * XXX: If we run out of reserved DRC memory we could (up to a point)
 793  * re-negotiate active sessions and reduce their slot usage to make
 794  * room for new connections. For now we just fail the create session.
 795  */
 796 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
 797 {
 798         u32 slotsize = slot_bytes(ca);
 799         u32 num = ca->maxreqs;
 800         int avail;
 801
 802         spin_lock(&nfsd_drc_lock);
 803         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
 804                     nfsd_drc_max_mem - nfsd_drc_mem_used);
 805         num = min_t(int, num, avail / slotsize);
 806         nfsd_drc_mem_used += num * slotsize;
 807         spin_unlock(&nfsd_drc_lock);
 808
 809         return num;
 810 }
 811
 812 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
 813 {
 814         int slotsize = slot_bytes(ca);
 815
 816         spin_lock(&nfsd_drc_lock);
 817         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
 818         spin_unlock(&nfsd_drc_lock);
 819 }
 820
 821 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *attrs)
 822 {
 823         int numslots = attrs->maxreqs;
 824         int slotsize = slot_bytes(attrs);
 825         struct nfsd4_session *new;
 826         int mem, i;
 827
 828         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 829                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 830         mem = numslots * sizeof(struct nfsd4_slot *);
 831
 832         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 833         if (!new)
 834                 return NULL;
 835         /* allocate each struct nfsd4_slot and data cache in one piece */
 836         for (i = 0; i < numslots; i++) {
 837                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
 838                 if (!new->se_slots[i])
 839                         goto out_free;
 840         }
 841         return new;
 842 out_free:
 843         while (i--)
 844                 kfree(new->se_slots[i]);
 845         kfree(new);
 846         return NULL;
 847 }
 848
 849 static void free_conn(struct nfsd4_conn *c)
 850 {
 851         svc_xprt_put(c->cn_xprt);
 852         kfree(c);
 853 }
 854
 855 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 856 {
 857         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 858         struct nfs4_client *clp = c->cn_session->se_client;
 859
 860         spin_lock(&clp->cl_lock);
 861         if (!list_empty(&c->cn_persession)) {
 862                 list_del(&c->cn_persession);
 863                 free_conn(c);
 864         }
 865         nfsd4_probe_callback(clp);
 866         spin_unlock(&clp->cl_lock);
 867 }
 868
 869 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 870 {
 871         struct nfsd4_conn *conn;
 872
 873         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 874         if (!conn)
 875                 return NULL;
 876         svc_xprt_get(rqstp->rq_xprt);
 877         conn->cn_xprt = rqstp->rq_xprt;
 878         conn->cn_flags = flags;
 879         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 880         return conn;
 881 }
 882
 883 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 884 {
 885         conn->cn_session = ses;
 886         list_add(&conn->cn_persession, &ses->se_conns);
 887 }
 888
 889 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 890 {
 891         struct nfs4_client *clp = ses->se_client;
 892
 893         spin_lock(&clp->cl_lock);
 894         __nfsd4_hash_conn(conn, ses);
 895         spin_unlock(&clp->cl_lock);
 896 }
 897
 898 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 899 {
 900         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 901         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 902 }
 903
 904 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 905 {
 906         int ret;
 907
 908         nfsd4_hash_conn(conn, ses);
 909         ret = nfsd4_register_conn(conn);
 910         if (ret)
 911                 /* oops; xprt is already down: */
 912                 nfsd4_conn_lost(&conn->cn_xpt_user);
 913         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 914                 /* callback channel may be back up */
 915                 nfsd4_probe_callback(ses->se_client);
 916         }
 917 }
 918
 919 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 920 {
 921         u32 dir = NFS4_CDFC4_FORE;
 922
 923         if (cses->flags & SESSION4_BACK_CHAN)
 924                 dir |= NFS4_CDFC4_BACK;
 925         return alloc_conn(rqstp, dir);
 926 }
 927
 928 /* must be called under client_lock */
 929 static void nfsd4_del_conns(struct nfsd4_session *s)
 930 {
 931         struct nfs4_client *clp = s->se_client;
 932         struct nfsd4_conn *c;
 933
 934         spin_lock(&clp->cl_lock);
 935         while (!list_empty(&s->se_conns)) {
 936                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 937                 list_del_init(&c->cn_persession);
 938                 spin_unlock(&clp->cl_lock);
 939
 940                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 941                 free_conn(c);
 942
 943                 spin_lock(&clp->cl_lock);
 944         }
 945         spin_unlock(&clp->cl_lock);
 946 }
 947
 948 static void __free_session(struct nfsd4_session *ses)
 949 {
 950         free_session_slots(ses);
 951         kfree(ses);
 952 }
 953
 954 static void free_session(struct nfsd4_session *ses)
 955 {
 956         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
 957
 958         lockdep_assert_held(&nn->client_lock);
 959         nfsd4_del_conns(ses);
 960         nfsd4_put_drc_mem(&ses->se_fchannel);
 961         __free_session(ses);
 962 }
 963
 964 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 965 {
 966         int idx;
 967         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 968
 969         new->se_client = clp;
 970         gen_sessionid(new);
 971
 972         INIT_LIST_HEAD(&new->se_conns);
 973
 974         new->se_cb_seq_nr = 1;
 975         new->se_flags = cses->flags;
 976         new->se_cb_prog = cses->callback_prog;
 977         new->se_cb_sec = cses->cb_sec;
 978         atomic_set(&new->se_ref, 0);
 979         idx = hash_sessionid(&new->se_sessionid);
 980         spin_lock(&nn->client_lock);
 981         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
 982         spin_lock(&clp->cl_lock);
 983         list_add(&new->se_perclnt, &clp->cl_sessions);
 984         spin_unlock(&clp->cl_lock);
 985         spin_unlock(&nn->client_lock);
 986         memcpy(&new->se_fchannel, &cses->fore_channel,
 987                         sizeof(struct nfsd4_channel_attrs));
 988         if (cses->flags & SESSION4_BACK_CHAN) {
 989                 struct sockaddr *sa = svc_addr(rqstp);
 990                 /*
 991                  * This is a little silly; with sessions there's no real
 992                  * use for the callback address.  Use the peer address
 993                  * as a reasonable default for now, but consider fixing
 994                  * the rpc client not to require an address in the
 995                  * future:
 996                  */
 997                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
 998                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
 999         }
1000 }
1001
1002 /* caller must hold client_lock */
1003 static struct nfsd4_session *
1004 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1005 {
1006         struct nfsd4_session *elem;
1007         int idx;
1008         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1009
1010         dump_sessionid(__func__, sessionid);
1011         idx = hash_sessionid(sessionid);
1012         /* Search in the appropriate list */
1013         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1014                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1015                             NFS4_MAX_SESSIONID_LEN)) {
1016                         return elem;
1017                 }
1018         }
1019
1020         dprintk("%s: session not found\n", __func__);
1021         return NULL;
1022 }
1023
1024 /* caller must hold client_lock */
1025 static void
1026 unhash_session(struct nfsd4_session *ses)
1027 {
1028         list_del(&ses->se_hash);
1029         spin_lock(&ses->se_client->cl_lock);
1030         list_del(&ses->se_perclnt);
1031         spin_unlock(&ses->se_client->cl_lock);
1032 }
1033
1034 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1035 static int
1036 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1037 {
1038         if (clid->cl_boot == nn->boot_time)
1039                 return 0;
1040         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1041                 clid->cl_boot, clid->cl_id, nn->boot_time);
1042         return 1;
1043 }
1044
1045 /*
1046  * XXX Should we use a slab cache ?
1047  * This type of memory management is somewhat inefficient, but we use it
1048  * anyway since SETCLIENTID is not a common operation.
1049  */
1050 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1051 {
1052         struct nfs4_client *clp;
1053
1054         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1055         if (clp == NULL)
1056                 return NULL;
1057         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1058         if (clp->cl_name.data == NULL) {
1059                 kfree(clp);
1060                 return NULL;
1061         }
1062         clp->cl_name.len = name.len;
1063         return clp;
1064 }
1065
1066 static inline void
1067 free_client(struct nfs4_client *clp)
1068 {
1069         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1070
1071         lockdep_assert_held(&nn->client_lock);
1072         while (!list_empty(&clp->cl_sessions)) {
1073                 struct nfsd4_session *ses;
1074                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1075                                 se_perclnt);
1076                 list_del(&ses->se_perclnt);
1077                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1078                 free_session(ses);
1079         }
1080         free_svc_cred(&clp->cl_cred);
1081         kfree(clp->cl_name.data);
1082         idr_destroy(&clp->cl_stateids);
1083         kfree(clp);
1084 }
1085
1086 /* must be called under the client_lock */
1087 static inline void
1088 unhash_client_locked(struct nfs4_client *clp)
1089 {
1090         struct nfsd4_session *ses;
1091
1092         list_del(&clp->cl_lru);
1093         spin_lock(&clp->cl_lock);
1094         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1095                 list_del_init(&ses->se_hash);
1096         spin_unlock(&clp->cl_lock);
1097 }
1098
1099 static void
1100 destroy_client(struct nfs4_client *clp)
1101 {
1102         struct nfs4_openowner *oo;
1103         struct nfs4_delegation *dp;
1104         struct list_head reaplist;
1105         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1106
1107         INIT_LIST_HEAD(&reaplist);
1108         spin_lock(&recall_lock);
1109         while (!list_empty(&clp->cl_delegations)) {
1110                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1111                 list_del_init(&dp->dl_perclnt);
1112                 list_move(&dp->dl_recall_lru, &reaplist);
1113         }
1114         spin_unlock(&recall_lock);
1115         while (!list_empty(&reaplist)) {
1116                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1117                 unhash_delegation(dp);
1118         }
1119         while (!list_empty(&clp->cl_openowners)) {
1120                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1121                 release_openowner(oo);
1122         }
1123         nfsd4_shutdown_callback(clp);
1124         if (clp->cl_cb_conn.cb_xprt)
1125                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1126         list_del(&clp->cl_idhash);
1127         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1128                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1129         else
1130                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1131         spin_lock(&nn->client_lock);
1132         unhash_client_locked(clp);
1133         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1134         free_client(clp);
1135         spin_unlock(&nn->client_lock);
1136 }
1137
1138 static void expire_client(struct nfs4_client *clp)
1139 {
1140         nfsd4_client_record_remove(clp);
1141         destroy_client(clp);
1142 }
1143
1144 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1145 {
1146         memcpy(target->cl_verifier.data, source->data,
1147                         sizeof(target->cl_verifier.data));
1148 }
1149
1150 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1151 {
1152         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1153         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1154 }
1155
1156 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1157 {
1158         if (source->cr_principal) {
1159                 target->cr_principal =
1160                                 kstrdup(source->cr_principal, GFP_KERNEL);
1161                 if (target->cr_principal == NULL)
1162                         return -ENOMEM;
1163         } else
1164                 target->cr_principal = NULL;
1165         target->cr_flavor = source->cr_flavor;
1166         target->cr_uid = source->cr_uid;
1167         target->cr_gid = source->cr_gid;
1168         target->cr_group_info = source->cr_group_info;
1169         get_group_info(target->cr_group_info);
1170         return 0;
1171 }
1172
1173 static long long
1174 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1175 {
1176         long long res;
1177
1178         res = o1->len - o2->len;
1179         if (res)
1180                 return res;
1181         return (long long)memcmp(o1->data, o2->data, o1->len);
1182 }
1183
1184 static int same_name(const char *n1, const char *n2)
1185 {
1186         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1187 }
1188
1189 static int
1190 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1191 {
1192         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1193 }
1194
1195 static int
1196 same_clid(clientid_t *cl1, clientid_t *cl2)
1197 {
1198         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1199 }
1200
1201 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1202 {
1203         int i;
1204
1205         if (g1->ngroups != g2->ngroups)
1206                 return false;
1207         for (i=0; i<g1->ngroups; i++)
1208                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1209                         return false;
1210         return true;
1211 }
1212
1213 /*
1214  * RFC 3530 language requires clid_inuse be returned when the
1215  * "principal" associated with a requests differs from that previously
1216  * used.  We use uid, gid's, and gss principal string as our best
1217  * approximation.  We also don't want to allow non-gss use of a client
1218  * established using gss: in theory cr_principal should catch that
1219  * change, but in practice cr_principal can be null even in the gss case
1220  * since gssd doesn't always pass down a principal string.
1221  */
1222 static bool is_gss_cred(struct svc_cred *cr)
1223 {
1224         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1225         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1226 }
1227
1228
1229 static bool
1230 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1231 {
1232         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1233                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1234                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1235                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1236                 return false;
1237         if (cr1->cr_principal == cr2->cr_principal)
1238                 return true;
1239         if (!cr1->cr_principal || !cr2->cr_principal)
1240                 return false;
1241         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1242 }
1243
1244 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1245 {
1246         static u32 current_clientid = 1;
1247
1248         clp->cl_clientid.cl_boot = nn->boot_time;
1249         clp->cl_clientid.cl_id = current_clientid++;
1250 }
1251
1252 static void gen_confirm(struct nfs4_client *clp)
1253 {
1254         __be32 verf[2];
1255         static u32 i;
1256
1257         verf[0] = (__be32)get_seconds();
1258         verf[1] = (__be32)i++;
1259         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1260 }
1261
1262 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1263 {
1264         struct nfs4_stid *ret;
1265
1266         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1267         if (!ret || !ret->sc_type)
1268                 return NULL;
1269         return ret;
1270 }
1271
1272 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1273 {
1274         struct nfs4_stid *s;
1275
1276         s = find_stateid(cl, t);
1277         if (!s)
1278                 return NULL;
1279         if (typemask & s->sc_type)
1280                 return s;
1281         return NULL;
1282 }
1283
1284 static struct nfs4_client *create_client(struct xdr_netobj name,
1285                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1286 {
1287         struct nfs4_client *clp;
1288         struct sockaddr *sa = svc_addr(rqstp);
1289         int ret;
1290         struct net *net = SVC_NET(rqstp);
1291         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1292
1293         clp = alloc_client(name);
1294         if (clp == NULL)
1295                 return NULL;
1296
1297         INIT_LIST_HEAD(&clp->cl_sessions);
1298         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1299         if (ret) {
1300                 spin_lock(&nn->client_lock);
1301                 free_client(clp);
1302                 spin_unlock(&nn->client_lock);
1303                 return NULL;
1304         }
1305         idr_init(&clp->cl_stateids);
1306         atomic_set(&clp->cl_refcount, 0);
1307         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1308         INIT_LIST_HEAD(&clp->cl_idhash);
1309         INIT_LIST_HEAD(&clp->cl_openowners);
1310         INIT_LIST_HEAD(&clp->cl_delegations);
1311         INIT_LIST_HEAD(&clp->cl_lru);
1312         INIT_LIST_HEAD(&clp->cl_callbacks);
1313         spin_lock_init(&clp->cl_lock);
1314         nfsd4_init_callback(&clp->cl_cb_null);
1315         clp->cl_time = get_seconds();
1316         clear_bit(0, &clp->cl_cb_slot_busy);
1317         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1318         copy_verf(clp, verf);
1319         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1320         gen_confirm(clp);
1321         clp->cl_cb_session = NULL;
1322         clp->net = net;
1323         return clp;
1324 }
1325
1326 static void
1327 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1328 {
1329         struct rb_node **new = &(root->rb_node), *parent = NULL;
1330         struct nfs4_client *clp;
1331
1332         while (*new) {
1333                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1334                 parent = *new;
1335
1336                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1337                         new = &((*new)->rb_left);
1338                 else
1339                         new = &((*new)->rb_right);
1340         }
1341
1342         rb_link_node(&new_clp->cl_namenode, parent, new);
1343         rb_insert_color(&new_clp->cl_namenode, root);
1344 }
1345
1346 static struct nfs4_client *
1347 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1348 {
1349         long long cmp;
1350         struct rb_node *node = root->rb_node;
1351         struct nfs4_client *clp;
1352
1353         while (node) {
1354                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1355                 cmp = compare_blob(&clp->cl_name, name);
1356                 if (cmp > 0)
1357                         node = node->rb_left;
1358                 else if (cmp < 0)
1359                         node = node->rb_right;
1360                 else
1361                         return clp;
1362         }
1363         return NULL;
1364 }
1365
1366 static void
1367 add_to_unconfirmed(struct nfs4_client *clp)
1368 {
1369         unsigned int idhashval;
1370         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1371
1372         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1373         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1374         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1375         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1376         renew_client(clp);
1377 }
1378
1379 static void
1380 move_to_confirmed(struct nfs4_client *clp)
1381 {
1382         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1383         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1384
1385         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1386         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1387         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1388         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1389         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1390         renew_client(clp);
1391 }
1392
1393 static struct nfs4_client *
1394 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1395 {
1396         struct nfs4_client *clp;
1397         unsigned int idhashval = clientid_hashval(clid->cl_id);
1398
1399         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1400                 if (same_clid(&clp->cl_clientid, clid)) {
1401                         if ((bool)clp->cl_minorversion != sessions)
1402                                 return NULL;
1403                         renew_client(clp);
1404                         return clp;
1405                 }
1406         }
1407         return NULL;
1408 }
1409
1410 static struct nfs4_client *
1411 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1412 {
1413         struct list_head *tbl = nn->conf_id_hashtbl;
1414
1415         return find_client_in_id_table(tbl, clid, sessions);
1416 }
1417
1418 static struct nfs4_client *
1419 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1420 {
1421         struct list_head *tbl = nn->unconf_id_hashtbl;
1422
1423         return find_client_in_id_table(tbl, clid, sessions);
1424 }
1425
1426 static bool clp_used_exchangeid(struct nfs4_client *clp)
1427 {
1428         return clp->cl_exchange_flags != 0;
1429 }
1430
1431 static struct nfs4_client *
1432 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1433 {
1434         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1435 }
1436
1437 static struct nfs4_client *
1438 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1439 {
1440         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1441 }
1442
1443 static void
1444 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1445 {
1446         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1447         struct sockaddr *sa = svc_addr(rqstp);
1448         u32 scopeid = rpc_get_scope_id(sa);
1449         unsigned short expected_family;
1450
1451         /* Currently, we only support tcp and tcp6 for the callback channel */
1452         if (se->se_callback_netid_len == 3 &&
1453             !memcmp(se->se_callback_netid_val, "tcp", 3))
1454                 expected_family = AF_INET;
1455         else if (se->se_callback_netid_len == 4 &&
1456                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1457                 expected_family = AF_INET6;
1458         else
1459                 goto out_err;
1460
1461         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1462                                             se->se_callback_addr_len,
1463                                             (struct sockaddr *)&conn->cb_addr,
1464                                             sizeof(conn->cb_addr));
1465
1466         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1467                 goto out_err;
1468
1469         if (conn->cb_addr.ss_family == AF_INET6)
1470                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1471
1472         conn->cb_prog = se->se_callback_prog;
1473         conn->cb_ident = se->se_callback_ident;
1474         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1475         return;
1476 out_err:
1477         conn->cb_addr.ss_family = AF_UNSPEC;
1478         conn->cb_addrlen = 0;
1479         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1480                 "will not receive delegations\n",
1481                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1482
1483         return;
1484 }
1485
1486 /*
1487  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1488  */
1489 void
1490 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1491 {
1492         struct nfsd4_slot *slot = resp->cstate.slot;
1493         unsigned int base;
1494
1495         dprintk("--> %s slot %p\n", __func__, slot);
1496
1497         slot->sl_opcnt = resp->opcnt;
1498         slot->sl_status = resp->cstate.status;
1499
1500         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1501         if (nfsd4_not_cached(resp)) {
1502                 slot->sl_datalen = 0;
1503                 return;
1504         }
1505         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1506         base = (char *)resp->cstate.datap -
1507                                         (char *)resp->xbuf->head[0].iov_base;
1508         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1509                                     slot->sl_datalen))
1510                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1511         return;
1512 }
1513
1514 /*
1515  * Encode the replay sequence operation from the slot values.
1516  * If cachethis is FALSE encode the uncached rep error on the next
1517  * operation which sets resp->p and increments resp->opcnt for
1518  * nfs4svc_encode_compoundres.
1519  *
1520  */
1521 static __be32
1522 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1523                           struct nfsd4_compoundres *resp)
1524 {
1525         struct nfsd4_op *op;
1526         struct nfsd4_slot *slot = resp->cstate.slot;
1527
1528         /* Encode the replayed sequence operation */
1529         op = &args->ops[resp->opcnt - 1];
1530         nfsd4_encode_operation(resp, op);
1531
1532         /* Return nfserr_retry_uncached_rep in next operation. */
1533         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1534                 op = &args->ops[resp->opcnt++];
1535                 op->status = nfserr_retry_uncached_rep;
1536                 nfsd4_encode_operation(resp, op);
1537         }
1538         return op->status;
1539 }
1540
1541 /*
1542  * The sequence operation is not cached because we can use the slot and
1543  * session values.
1544  */
1545 __be32
1546 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1547                          struct nfsd4_sequence *seq)
1548 {
1549         struct nfsd4_slot *slot = resp->cstate.slot;
1550         __be32 status;
1551
1552         dprintk("--> %s slot %p\n", __func__, slot);
1553
1554         /* Either returns 0 or nfserr_retry_uncached */
1555         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1556         if (status == nfserr_retry_uncached_rep)
1557                 return status;
1558
1559         /* The sequence operation has been encoded, cstate->datap set. */
1560         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1561
1562         resp->opcnt = slot->sl_opcnt;
1563         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1564         status = slot->sl_status;
1565
1566         return status;
1567 }
1568
1569 /*
1570  * Set the exchange_id flags returned by the server.
1571  */
1572 static void
1573 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1574 {
1575         /* pNFS is not supported */
1576         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1577
1578         /* Referrals are supported, Migration is not. */
1579         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1580
1581         /* set the wire flags to return to client. */
1582         clid->flags = new->cl_exchange_flags;
1583 }
1584
1585 static bool client_has_state(struct nfs4_client *clp)
1586 {
1587         /*
1588          * Note clp->cl_openowners check isn't quite right: there's no
1589          * need to count owners without stateid's.
1590          *
1591          * Also note we should probably be using this in 4.0 case too.
1592          */
1593         return !list_empty(&clp->cl_openowners)
1594                 || !list_empty(&clp->cl_delegations)
1595                 || !list_empty(&clp->cl_sessions);
1596 }
1597
1598 __be32
1599 nfsd4_exchange_id(struct svc_rqst *rqstp,
1600                   struct nfsd4_compound_state *cstate,
1601                   struct nfsd4_exchange_id *exid)
1602 {
1603         struct nfs4_client *unconf, *conf, *new;
1604         __be32 status;
1605         char                    addr_str[INET6_ADDRSTRLEN];
1606         nfs4_verifier           verf = exid->verifier;
1607         struct sockaddr         *sa = svc_addr(rqstp);
1608         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1609         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1610
1611         rpc_ntop(sa, addr_str, sizeof(addr_str));
1612         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1613                 "ip_addr=%s flags %x, spa_how %d\n",
1614                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1615                 addr_str, exid->flags, exid->spa_how);
1616
1617         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1618                 return nfserr_inval;
1619
1620         /* Currently only support SP4_NONE */
1621         switch (exid->spa_how) {
1622         case SP4_NONE:
1623                 break;
1624         default:                                /* checked by xdr code */
1625                 WARN_ON_ONCE(1);
1626         case SP4_SSV:
1627         case SP4_MACH_CRED:
1628                 return nfserr_serverfault;      /* no excuse :-/ */
1629         }
1630
1631         /* Cases below refer to rfc 5661 section 18.35.4: */
1632         nfs4_lock_state();
1633         conf = find_confirmed_client_by_name(&exid->clname, nn);
1634         if (conf) {
1635                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1636                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1637
1638                 if (update) {
1639                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1640                                 status = nfserr_inval;
1641                                 goto out;
1642                         }
1643                         if (!creds_match) { /* case 9 */
1644                                 status = nfserr_perm;
1645                                 goto out;
1646                         }
1647                         if (!verfs_match) { /* case 8 */
1648                                 status = nfserr_not_same;
1649                                 goto out;
1650                         }
1651                         /* case 6 */
1652                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1653                         new = conf;
1654                         goto out_copy;
1655                 }
1656                 if (!creds_match) { /* case 3 */
1657                         if (client_has_state(conf)) {
1658                                 status = nfserr_clid_inuse;
1659                                 goto out;
1660                         }
1661                         expire_client(conf);
1662                         goto out_new;
1663                 }
1664                 if (verfs_match) { /* case 2 */
1665                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1666                         new = conf;
1667                         goto out_copy;
1668                 }
1669                 /* case 5, client reboot */
1670                 goto out_new;
1671         }
1672
1673         if (update) { /* case 7 */
1674                 status = nfserr_noent;
1675                 goto out;
1676         }
1677
1678         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1679         if (unconf) /* case 4, possible retry or client restart */
1680                 expire_client(unconf);
1681
1682         /* case 1 (normal case) */
1683 out_new:
1684         new = create_client(exid->clname, rqstp, &verf);
1685         if (new == NULL) {
1686                 status = nfserr_jukebox;
1687                 goto out;
1688         }
1689         new->cl_minorversion = 1;
1690
1691         gen_clid(new, nn);
1692         add_to_unconfirmed(new);
1693 out_copy:
1694         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1695         exid->clientid.cl_id = new->cl_clientid.cl_id;
1696
1697         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1698         nfsd4_set_ex_flags(new, exid);
1699
1700         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1701                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1702         status = nfs_ok;
1703
1704 out:
1705         nfs4_unlock_state();
1706         return status;
1707 }
1708
1709 static __be32
1710 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1711 {
1712         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1713                 slot_seqid);
1714
1715         /* The slot is in use, and no response has been sent. */
1716         if (slot_inuse) {
1717                 if (seqid == slot_seqid)
1718                         return nfserr_jukebox;
1719                 else
1720                         return nfserr_seq_misordered;
1721         }
1722         /* Note unsigned 32-bit arithmetic handles wraparound: */
1723         if (likely(seqid == slot_seqid + 1))
1724                 return nfs_ok;
1725         if (seqid == slot_seqid)
1726                 return nfserr_replay_cache;
1727         return nfserr_seq_misordered;
1728 }
1729
1730 /*
1731  * Cache the create session result into the create session single DRC
1732  * slot cache by saving the xdr structure. sl_seqid has been set.
1733  * Do this for solo or embedded create session operations.
1734  */
1735 static void
1736 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1737                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1738 {
1739         slot->sl_status = nfserr;
1740         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1741 }
1742
1743 static __be32
1744 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1745                             struct nfsd4_clid_slot *slot)
1746 {
1747         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1748         return slot->sl_status;
1749 }
1750
1751 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1752                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1753                         1 +     /* MIN tag is length with zero, only length */ \
1754                         3 +     /* version, opcount, opcode */ \
1755                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1756                                 /* seqid, slotID, slotID, cache */ \
1757                         4 ) * sizeof(__be32))
1758
1759 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1760                         2 +     /* verifier: AUTH_NULL, length 0 */\
1761                         1 +     /* status */ \
1762                         1 +     /* MIN tag is length with zero, only length */ \
1763                         3 +     /* opcount, opcode, opstatus*/ \
1764                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1765                                 /* seqid, slotID, slotID, slotID, status */ \
1766                         5 ) * sizeof(__be32))
1767
1768 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1769 {
1770         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
1771
1772         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
1773                 return nfserr_toosmall;
1774         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
1775                 return nfserr_toosmall;
1776         ca->headerpadsz = 0;
1777         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
1778         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
1779         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
1780         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
1781                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
1782         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
1783         /*
1784          * Note decreasing slot size below client's request may make it
1785          * difficult for client to function correctly, whereas
1786          * decreasing the number of slots will (just?) affect
1787          * performance.  When short on memory we therefore prefer to
1788          * decrease number of slots instead of their size.  Clients that
1789          * request larger slots than they need will get poor results:
1790          */
1791         ca->maxreqs = nfsd4_get_drc_mem(ca);
1792         if (!ca->maxreqs)
1793                 return nfserr_jukebox;
1794
1795         return nfs_ok;
1796 }
1797
1798 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
1799 {
1800         ca->headerpadsz = 0;
1801
1802         /*
1803          * These RPC_MAX_HEADER macros are overkill, especially since we
1804          * don't even do gss on the backchannel yet.  But this is still
1805          * less than 1k.  Tighten up this estimate in the unlikely event
1806          * it turns out to be a problem for some client:
1807          */
1808         if (ca->maxreq_sz < NFS4_enc_cb_recall_sz + RPC_MAX_HEADER_WITH_AUTH)
1809                 return nfserr_toosmall;
1810         if (ca->maxresp_sz < NFS4_dec_cb_recall_sz + RPC_MAX_REPHEADER_WITH_AUTH)
1811                 return nfserr_toosmall;
1812         ca->maxresp_cached = 0;
1813         if (ca->maxops < 2)
1814                 return nfserr_toosmall;
1815
1816         return nfs_ok;
1817 }
1818
1819 __be32
1820 nfsd4_create_session(struct svc_rqst *rqstp,
1821                      struct nfsd4_compound_state *cstate,
1822                      struct nfsd4_create_session *cr_ses)
1823 {
1824         struct sockaddr *sa = svc_addr(rqstp);
1825         struct nfs4_client *conf, *unconf;
1826         struct nfsd4_session *new;
1827         struct nfsd4_conn *conn;
1828         struct nfsd4_clid_slot *cs_slot = NULL;
1829         __be32 status = 0;
1830         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1831
1832         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1833                 return nfserr_inval;
1834         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
1835         if (status)
1836                 return status;
1837         status = check_backchannel_attrs(&cr_ses->back_channel);
1838         if (status)
1839                 return status;
1840         status = nfserr_jukebox;
1841         new = alloc_session(&cr_ses->fore_channel);
1842         if (!new)
1843                 goto out_release_drc_mem;
1844         conn = alloc_conn_from_crses(rqstp, cr_ses);
1845         if (!conn)
1846                 goto out_free_session;
1847
1848         nfs4_lock_state();
1849         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1850         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1851         WARN_ON_ONCE(conf && unconf);
1852
1853         if (conf) {
1854                 cs_slot = &conf->cl_cs_slot;
1855                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1856                 if (status == nfserr_replay_cache) {
1857                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1858                         goto out_free_conn;
1859                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1860                         status = nfserr_seq_misordered;
1861                         goto out_free_conn;
1862                 }
1863         } else if (unconf) {
1864                 struct nfs4_client *old;
1865                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1866                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1867                         status = nfserr_clid_inuse;
1868                         goto out_free_conn;
1869                 }
1870                 cs_slot = &unconf->cl_cs_slot;
1871                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1872                 if (status) {
1873                         /* an unconfirmed replay returns misordered */
1874                         status = nfserr_seq_misordered;
1875                         goto out_free_conn;
1876                 }
1877                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1878                 if (old) {
1879                         status = mark_client_expired(old);
1880                         if (status)
1881                                 goto out_free_conn;
1882                         expire_client(old);
1883                 }
1884                 move_to_confirmed(unconf);
1885                 conf = unconf;
1886         } else {
1887                 status = nfserr_stale_clientid;
1888                 goto out_free_conn;
1889         }
1890         status = nfs_ok;
1891         /*
1892          * We do not support RDMA or persistent sessions
1893          */
1894         cr_ses->flags &= ~SESSION4_PERSIST;
1895         cr_ses->flags &= ~SESSION4_RDMA;
1896
1897         init_session(rqstp, new, conf, cr_ses);
1898         nfsd4_init_conn(rqstp, conn, new);
1899
1900         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1901                NFS4_MAX_SESSIONID_LEN);
1902         cs_slot->sl_seqid++;
1903         cr_ses->seqid = cs_slot->sl_seqid;
1904
1905         /* cache solo and embedded create sessions under the state lock */
1906         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1907         nfs4_unlock_state();
1908         return status;
1909 out_free_conn:
1910         nfs4_unlock_state();
1911         free_conn(conn);
1912 out_free_session:
1913         __free_session(new);
1914 out_release_drc_mem:
1915         nfsd4_put_drc_mem(&cr_ses->fore_channel);
1916         return status;
1917 }
1918
1919 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1920 {
1921         switch (*dir) {
1922         case NFS4_CDFC4_FORE:
1923         case NFS4_CDFC4_BACK:
1924                 return nfs_ok;
1925         case NFS4_CDFC4_FORE_OR_BOTH:
1926         case NFS4_CDFC4_BACK_OR_BOTH:
1927                 *dir = NFS4_CDFC4_BOTH;
1928                 return nfs_ok;
1929         };
1930         return nfserr_inval;
1931 }
1932
1933 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1934 {
1935         struct nfsd4_session *session = cstate->session;
1936         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1937
1938         spin_lock(&nn->client_lock);
1939         session->se_cb_prog = bc->bc_cb_program;
1940         session->se_cb_sec = bc->bc_cb_sec;
1941         spin_unlock(&nn->client_lock);
1942
1943         nfsd4_probe_callback(session->se_client);
1944
1945         return nfs_ok;
1946 }
1947
1948 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1949                      struct nfsd4_compound_state *cstate,
1950                      struct nfsd4_bind_conn_to_session *bcts)
1951 {
1952         __be32 status;
1953         struct nfsd4_conn *conn;
1954         struct nfsd4_session *session;
1955         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1956
1957         if (!nfsd4_last_compound_op(rqstp))
1958                 return nfserr_not_only_op;
1959         nfs4_lock_state();
1960         spin_lock(&nn->client_lock);
1961         session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
1962         spin_unlock(&nn->client_lock);
1963         status = nfserr_badsession;
1964         if (!session)
1965                 goto out;
1966         status = nfsd4_map_bcts_dir(&bcts->dir);
1967         if (status)
1968                 goto out;
1969         conn = alloc_conn(rqstp, bcts->dir);
1970         status = nfserr_jukebox;
1971         if (!conn)
1972                 goto out;
1973         nfsd4_init_conn(rqstp, conn, session);
1974         status = nfs_ok;
1975 out:
1976         nfs4_unlock_state();
1977         return status;
1978 }
1979
1980 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1981 {
1982         if (!session)
1983                 return 0;
1984         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1985 }
1986
1987 __be32
1988 nfsd4_destroy_session(struct svc_rqst *r,
1989                       struct nfsd4_compound_state *cstate,
1990                       struct nfsd4_destroy_session *sessionid)
1991 {
1992         struct nfsd4_session *ses;
1993         __be32 status;
1994         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
1995
1996         nfs4_lock_state();
1997         status = nfserr_not_only_op;
1998         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1999                 if (!nfsd4_last_compound_op(r))
2000                         goto out;
2001         }
2002         dump_sessionid(__func__, &sessionid->sessionid);
2003         spin_lock(&nn->client_lock);
2004         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
2005         status = nfserr_badsession;
2006         if (!ses)
2007                 goto out_client_lock;
2008         status = mark_session_dead_locked(ses);
2009         if (status)
2010                 goto out_client_lock;
2011         unhash_session(ses);
2012         spin_unlock(&nn->client_lock);
2013
2014         nfsd4_probe_callback_sync(ses->se_client);
2015
2016         spin_lock(&nn->client_lock);
2017         free_session(ses);
2018         status = nfs_ok;
2019 out_client_lock:
2020         spin_unlock(&nn->client_lock);
2021 out:
2022         nfs4_unlock_state();
2023         return status;
2024 }
2025
2026 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2027 {
2028         struct nfsd4_conn *c;
2029
2030         list_for_each_entry(c, &s->se_conns, cn_persession) {
2031                 if (c->cn_xprt == xpt) {
2032                         return c;
2033                 }
2034         }
2035         return NULL;
2036 }
2037
2038 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2039 {
2040         struct nfs4_client *clp = ses->se_client;
2041         struct nfsd4_conn *c;
2042         int ret;
2043
2044         spin_lock(&clp->cl_lock);
2045         c = __nfsd4_find_conn(new->cn_xprt, ses);
2046         if (c) {
2047                 spin_unlock(&clp->cl_lock);
2048                 free_conn(new);
2049                 return;
2050         }
2051         __nfsd4_hash_conn(new, ses);
2052         spin_unlock(&clp->cl_lock);
2053         ret = nfsd4_register_conn(new);
2054         if (ret)
2055                 /* oops; xprt is already down: */
2056                 nfsd4_conn_lost(&new->cn_xpt_user);
2057         return;
2058 }
2059
2060 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2061 {
2062         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2063
2064         return args->opcnt > session->se_fchannel.maxops;
2065 }
2066
2067 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2068                                   struct nfsd4_session *session)
2069 {
2070         struct xdr_buf *xb = &rqstp->rq_arg;
2071
2072         return xb->len > session->se_fchannel.maxreq_sz;
2073 }
2074
2075 __be32
2076 nfsd4_sequence(struct svc_rqst *rqstp,
2077                struct nfsd4_compound_state *cstate,
2078                struct nfsd4_sequence *seq)
2079 {
2080         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2081         struct nfsd4_session *session;
2082         struct nfs4_client *clp;
2083         struct nfsd4_slot *slot;
2084         struct nfsd4_conn *conn;
2085         __be32 status;
2086         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2087
2088         if (resp->opcnt != 1)
2089                 return nfserr_sequence_pos;
2090
2091         /*
2092          * Will be either used or freed by nfsd4_sequence_check_conn
2093          * below.
2094          */
2095         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2096         if (!conn)
2097                 return nfserr_jukebox;
2098
2099         spin_lock(&nn->client_lock);
2100         status = nfserr_badsession;
2101         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2102         if (!session)
2103                 goto out_no_session;
2104         clp = session->se_client;
2105         status = get_client_locked(clp);
2106         if (status)
2107                 goto out_no_session;
2108         status = nfsd4_get_session_locked(session);
2109         if (status)
2110                 goto out_put_client;
2111
2112         status = nfserr_too_many_ops;
2113         if (nfsd4_session_too_many_ops(rqstp, session))
2114                 goto out_put_session;
2115
2116         status = nfserr_req_too_big;
2117         if (nfsd4_request_too_big(rqstp, session))
2118                 goto out_put_session;
2119
2120         status = nfserr_badslot;
2121         if (seq->slotid >= session->se_fchannel.maxreqs)
2122                 goto out_put_session;
2123
2124         slot = session->se_slots[seq->slotid];
2125         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2126
2127         /* We do not negotiate the number of slots yet, so set the
2128          * maxslots to the session maxreqs which is used to encode
2129          * sr_highest_slotid and the sr_target_slot id to maxslots */
2130         seq->maxslots = session->se_fchannel.maxreqs;
2131
2132         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2133                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2134         if (status == nfserr_replay_cache) {
2135                 status = nfserr_seq_misordered;
2136                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2137                         goto out_put_session;
2138                 cstate->slot = slot;
2139                 cstate->session = session;
2140                 /* Return the cached reply status and set cstate->status
2141                  * for nfsd4_proc_compound processing */
2142                 status = nfsd4_replay_cache_entry(resp, seq);
2143                 cstate->status = nfserr_replay_cache;
2144                 goto out;
2145         }
2146         if (status)
2147                 goto out_put_session;
2148
2149         nfsd4_sequence_check_conn(conn, session);
2150         conn = NULL;
2151
2152         /* Success! bump slot seqid */
2153         slot->sl_seqid = seq->seqid;
2154         slot->sl_flags |= NFSD4_SLOT_INUSE;
2155         if (seq->cachethis)
2156                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2157         else
2158                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2159
2160         cstate->slot = slot;
2161         cstate->session = session;
2162
2163 out:
2164         switch (clp->cl_cb_state) {
2165         case NFSD4_CB_DOWN:
2166                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2167                 break;
2168         case NFSD4_CB_FAULT:
2169                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2170                 break;
2171         default:
2172                 seq->status_flags = 0;
2173         }
2174 out_no_session:
2175         kfree(conn);
2176         spin_unlock(&nn->client_lock);
2177         return status;
2178 out_put_session:
2179         nfsd4_put_session(session);
2180 out_put_client:
2181         put_client_renew_locked(clp);
2182         goto out_no_session;
2183 }
2184
2185 __be32
2186 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2187 {
2188         struct nfs4_client *conf, *unconf, *clp;
2189         __be32 status = 0;
2190         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2191
2192         nfs4_lock_state();
2193         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2194         conf = find_confirmed_client(&dc->clientid, true, nn);
2195         WARN_ON_ONCE(conf && unconf);
2196
2197         if (conf) {
2198                 clp = conf;
2199
2200                 if (client_has_state(conf)) {
2201                         status = nfserr_clientid_busy;
2202                         goto out;
2203                 }
2204         } else if (unconf)
2205                 clp = unconf;
2206         else {
2207                 status = nfserr_stale_clientid;
2208                 goto out;
2209         }
2210
2211         expire_client(clp);
2212 out:
2213         nfs4_unlock_state();
2214         return status;
2215 }
2216
2217 __be32
2218 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2219 {
2220         __be32 status = 0;
2221
2222         if (rc->rca_one_fs) {
2223                 if (!cstate->current_fh.fh_dentry)
2224                         return nfserr_nofilehandle;
2225                 /*
2226                  * We don't take advantage of the rca_one_fs case.
2227                  * That's OK, it's optional, we can safely ignore it.
2228                  */
2229                  return nfs_ok;
2230         }
2231
2232         nfs4_lock_state();
2233         status = nfserr_complete_already;
2234         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2235                              &cstate->session->se_client->cl_flags))
2236                 goto out;
2237
2238         status = nfserr_stale_clientid;
2239         if (is_client_expired(cstate->session->se_client))
2240                 /*
2241                  * The following error isn't really legal.
2242                  * But we only get here if the client just explicitly
2243                  * destroyed the client.  Surely it no longer cares what
2244                  * error it gets back on an operation for the dead
2245                  * client.
2246                  */
2247                 goto out;
2248
2249         status = nfs_ok;
2250         nfsd4_client_record_create(cstate->session->se_client);
2251 out:
2252         nfs4_unlock_state();
2253         return status;
2254 }
2255
2256 __be32
2257 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2258                   struct nfsd4_setclientid *setclid)
2259 {
2260         struct xdr_netobj       clname = setclid->se_name;
2261         nfs4_verifier           clverifier = setclid->se_verf;
2262         struct nfs4_client      *conf, *unconf, *new;
2263         __be32                  status;
2264         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2265
2266         /* Cases below refer to rfc 3530 section 14.2.33: */
2267         nfs4_lock_state();
2268         conf = find_confirmed_client_by_name(&clname, nn);
2269         if (conf) {
2270                 /* case 0: */
2271                 status = nfserr_clid_inuse;
2272                 if (clp_used_exchangeid(conf))
2273                         goto out;
2274                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2275                         char addr_str[INET6_ADDRSTRLEN];
2276                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2277                                  sizeof(addr_str));
2278                         dprintk("NFSD: setclientid: string in use by client "
2279                                 "at %s\n", addr_str);
2280                         goto out;
2281                 }
2282         }
2283         unconf = find_unconfirmed_client_by_name(&clname, nn);
2284         if (unconf)
2285                 expire_client(unconf);
2286         status = nfserr_jukebox;
2287         new = create_client(clname, rqstp, &clverifier);
2288         if (new == NULL)
2289                 goto out;
2290         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2291                 /* case 1: probable callback update */
2292                 copy_clid(new, conf);
2293         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2294                 gen_clid(new, nn);
2295         new->cl_minorversion = 0;
2296         gen_callback(new, setclid, rqstp);
2297         add_to_unconfirmed(new);
2298         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2299         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2300         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2301         status = nfs_ok;
2302 out:
2303         nfs4_unlock_state();
2304         return status;
2305 }
2306
2307
2308 __be32
2309 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2310                          struct nfsd4_compound_state *cstate,
2311                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2312 {
2313         struct nfs4_client *conf, *unconf;
2314         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2315         clientid_t * clid = &setclientid_confirm->sc_clientid;
2316         __be32 status;
2317         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2318
2319         if (STALE_CLIENTID(clid, nn))
2320                 return nfserr_stale_clientid;
2321         nfs4_lock_state();
2322
2323         conf = find_confirmed_client(clid, false, nn);
2324         unconf = find_unconfirmed_client(clid, false, nn);
2325         /*
2326          * We try hard to give out unique clientid's, so if we get an
2327          * attempt to confirm the same clientid with a different cred,
2328          * there's a bug somewhere.  Let's charitably assume it's our
2329          * bug.
2330          */
2331         status = nfserr_serverfault;
2332         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2333                 goto out;
2334         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2335                 goto out;
2336         /* cases below refer to rfc 3530 section 14.2.34: */
2337         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2338                 if (conf && !unconf) /* case 2: probable retransmit */
2339                         status = nfs_ok;
2340                 else /* case 4: client hasn't noticed we rebooted yet? */
2341                         status = nfserr_stale_clientid;
2342                 goto out;
2343         }
2344         status = nfs_ok;
2345         if (conf) { /* case 1: callback update */
2346                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2347                 nfsd4_probe_callback(conf);
2348                 expire_client(unconf);
2349         } else { /* case 3: normal case; new or rebooted client */
2350                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2351                 if (conf) {
2352                         status = mark_client_expired(conf);
2353                         if (status)
2354                                 goto out;
2355                         expire_client(conf);
2356                 }
2357                 move_to_confirmed(unconf);
2358                 nfsd4_probe_callback(unconf);
2359         }
2360 out:
2361         nfs4_unlock_state();
2362         return status;
2363 }
2364
2365 static struct nfs4_file *nfsd4_alloc_file(void)
2366 {
2367         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2368 }
2369
2370 /* OPEN Share state helper functions */
2371 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2372 {
2373         unsigned int hashval = file_hashval(ino);
2374
2375         atomic_set(&fp->fi_ref, 1);
2376         INIT_LIST_HEAD(&fp->fi_stateids);
2377         INIT_LIST_HEAD(&fp->fi_delegations);
2378         fp->fi_inode = igrab(ino);
2379         fp->fi_had_conflict = false;
2380         fp->fi_lease = NULL;
2381         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2382         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2383         spin_lock(&recall_lock);
2384         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2385         spin_unlock(&recall_lock);
2386 }
2387
2388 static void
2389 nfsd4_free_slab(struct kmem_cache **slab)
2390 {
2391         if (*slab == NULL)
2392                 return;
2393         kmem_cache_destroy(*slab);
2394         *slab = NULL;
2395 }
2396
2397 void
2398 nfsd4_free_slabs(void)
2399 {
2400         nfsd4_free_slab(&openowner_slab);
2401         nfsd4_free_slab(&lockowner_slab);
2402         nfsd4_free_slab(&file_slab);
2403         nfsd4_free_slab(&stateid_slab);
2404         nfsd4_free_slab(&deleg_slab);
2405 }
2406
2407 int
2408 nfsd4_init_slabs(void)
2409 {
2410         openowner_slab = kmem_cache_create("nfsd4_openowners",
2411                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2412         if (openowner_slab == NULL)
2413                 goto out_nomem;
2414         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2415                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2416         if (lockowner_slab == NULL)
2417                 goto out_nomem;
2418         file_slab = kmem_cache_create("nfsd4_files",
2419                         sizeof(struct nfs4_file), 0, 0, NULL);
2420         if (file_slab == NULL)
2421                 goto out_nomem;
2422         stateid_slab = kmem_cache_create("nfsd4_stateids",
2423                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2424         if (stateid_slab == NULL)
2425                 goto out_nomem;
2426         deleg_slab = kmem_cache_create("nfsd4_delegations",
2427                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2428         if (deleg_slab == NULL)
2429                 goto out_nomem;
2430         return 0;
2431 out_nomem:
2432         nfsd4_free_slabs();
2433         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2434         return -ENOMEM;
2435 }
2436
2437 void nfs4_free_openowner(struct nfs4_openowner *oo)
2438 {
2439         kfree(oo->oo_owner.so_owner.data);
2440         kmem_cache_free(openowner_slab, oo);
2441 }
2442
2443 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2444 {
2445         kfree(lo->lo_owner.so_owner.data);
2446         kmem_cache_free(lockowner_slab, lo);
2447 }
2448
2449 static void init_nfs4_replay(struct nfs4_replay *rp)
2450 {
2451         rp->rp_status = nfserr_serverfault;
2452         rp->rp_buflen = 0;
2453         rp->rp_buf = rp->rp_ibuf;
2454 }
2455
2456 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2457 {
2458         struct nfs4_stateowner *sop;
2459
2460         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2461         if (!sop)
2462                 return NULL;
2463
2464         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2465         if (!sop->so_owner.data) {
2466                 kmem_cache_free(slab, sop);
2467                 return NULL;
2468         }
2469         sop->so_owner.len = owner->len;
2470
2471         INIT_LIST_HEAD(&sop->so_stateids);
2472         sop->so_client = clp;
2473         init_nfs4_replay(&sop->so_replay);
2474         return sop;
2475 }
2476
2477 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2478 {
2479         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2480
2481         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2482         list_add(&oo->oo_perclient, &clp->cl_openowners);
2483 }
2484
2485 static struct nfs4_openowner *
2486 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2487         struct nfs4_openowner *oo;
2488
2489         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2490         if (!oo)
2491                 return NULL;
2492         oo->oo_owner.so_is_open_owner = 1;
2493         oo->oo_owner.so_seqid = open->op_seqid;
2494         oo->oo_flags = NFS4_OO_NEW;
2495         oo->oo_time = 0;
2496         oo->oo_last_closed_stid = NULL;
2497         INIT_LIST_HEAD(&oo->oo_close_lru);
2498         hash_openowner(oo, clp, strhashval);
2499         return oo;
2500 }
2501
2502 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2503         struct nfs4_openowner *oo = open->op_openowner;
2504
2505         stp->st_stid.sc_type = NFS4_OPEN_STID;
2506         INIT_LIST_HEAD(&stp->st_lockowners);
2507         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2508         list_add(&stp->st_perfile, &fp->fi_stateids);
2509         stp->st_stateowner = &oo->oo_owner;
2510         get_nfs4_file(fp);
2511         stp->st_file = fp;
2512         stp->st_access_bmap = 0;
2513         stp->st_deny_bmap = 0;
2514         set_access(open->op_share_access, stp);
2515         set_deny(open->op_share_deny, stp);
2516         stp->st_openstp = NULL;
2517 }
2518
2519 static void
2520 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2521 {
2522         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2523
2524         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2525
2526         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2527         oo->oo_time = get_seconds();
2528 }
2529
2530 static int
2531 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2532                                                         clientid_t *clid)
2533 {
2534         return (sop->so_owner.len == owner->len) &&
2535                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2536                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2537 }
2538
2539 static struct nfs4_openowner *
2540 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2541                         bool sessions, struct nfsd_net *nn)
2542 {
2543         struct nfs4_stateowner *so;
2544         struct nfs4_openowner *oo;
2545         struct nfs4_client *clp;
2546
2547         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2548                 if (!so->so_is_open_owner)
2549                         continue;
2550                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2551                         oo = openowner(so);
2552                         clp = oo->oo_owner.so_client;
2553                         if ((bool)clp->cl_minorversion != sessions)
2554                                 return NULL;
2555                         renew_client(oo->oo_owner.so_client);
2556                         return oo;
2557                 }
2558         }
2559         return NULL;
2560 }
2561
2562 /* search file_hashtbl[] for file */
2563 static struct nfs4_file *
2564 find_file(struct inode *ino)
2565 {
2566         unsigned int hashval = file_hashval(ino);
2567         struct nfs4_file *fp;
2568
2569         spin_lock(&recall_lock);
2570         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2571                 if (fp->fi_inode == ino) {
2572                         get_nfs4_file(fp);
2573                         spin_unlock(&recall_lock);
2574                         return fp;
2575                 }
2576         }
2577         spin_unlock(&recall_lock);
2578         return NULL;
2579 }
2580
2581 /*
2582  * Called to check deny when READ with all zero stateid or
2583  * WRITE with all zero or all one stateid
2584  */
2585 static __be32
2586 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2587 {
2588         struct inode *ino = current_fh->fh_dentry->d_inode;
2589         struct nfs4_file *fp;
2590         struct nfs4_ol_stateid *stp;
2591         __be32 ret;
2592
2593         fp = find_file(ino);
2594         if (!fp)
2595                 return nfs_ok;
2596         ret = nfserr_locked;
2597         /* Search for conflicting share reservations */
2598         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2599                 if (test_deny(deny_type, stp) ||
2600                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2601                         goto out;
2602         }
2603         ret = nfs_ok;
2604 out:
2605         put_nfs4_file(fp);
2606         return ret;
2607 }
2608
2609 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2610 {
2611         struct nfs4_client *clp = dp->dl_stid.sc_client;
2612         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2613
2614         /* We're assuming the state code never drops its reference
2615          * without first removing the lease.  Since we're in this lease
2616          * callback (and since the lease code is serialized by the kernel
2617          * lock) we know the server hasn't removed the lease yet, we know
2618          * it's safe to take a reference: */
2619         atomic_inc(&dp->dl_count);
2620
2621         list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
2622
2623         /* only place dl_time is set. protected by lock_flocks*/
2624         dp->dl_time = get_seconds();
2625
2626         nfsd4_cb_recall(dp);
2627 }
2628
2629 /* Called from break_lease() with lock_flocks() held. */
2630 static void nfsd_break_deleg_cb(struct file_lock *fl)
2631 {
2632         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2633         struct nfs4_delegation *dp;
2634
2635         if (!fp) {
2636                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2637                 return;
2638         }
2639         if (fp->fi_had_conflict) {
2640                 WARN(1, "duplicate break on %p\n", fp);
2641                 return;
2642         }
2643         /*
2644          * We don't want the locks code to timeout the lease for us;
2645          * we'll remove it ourself if a delegation isn't returned
2646          * in time:
2647          */
2648         fl->fl_break_time = 0;
2649
2650         spin_lock(&recall_lock);
2651         fp->fi_had_conflict = true;
2652         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2653                 nfsd_break_one_deleg(dp);
2654         spin_unlock(&recall_lock);
2655 }
2656
2657 static
2658 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2659 {
2660         if (arg & F_UNLCK)
2661                 return lease_modify(onlist, arg);
2662         else
2663                 return -EAGAIN;
2664 }
2665
2666 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2667         .lm_break = nfsd_break_deleg_cb,
2668         .lm_change = nfsd_change_deleg_cb,
2669 };
2670
2671 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2672 {
2673         if (nfsd4_has_session(cstate))
2674                 return nfs_ok;
2675         if (seqid == so->so_seqid - 1)
2676                 return nfserr_replay_me;
2677         if (seqid == so->so_seqid)
2678                 return nfs_ok;
2679         return nfserr_bad_seqid;
2680 }
2681
2682 __be32
2683 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2684                     struct nfsd4_open *open, struct nfsd_net *nn)
2685 {
2686         clientid_t *clientid = &open->op_clientid;
2687         struct nfs4_client *clp = NULL;
2688         unsigned int strhashval;
2689         struct nfs4_openowner *oo = NULL;
2690         __be32 status;
2691
2692         if (STALE_CLIENTID(&open->op_clientid, nn))
2693                 return nfserr_stale_clientid;
2694         /*
2695          * In case we need it later, after we've already created the
2696          * file and don't want to risk a further failure:
2697          */
2698         open->op_file = nfsd4_alloc_file();
2699         if (open->op_file == NULL)
2700                 return nfserr_jukebox;
2701
2702         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2703         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2704         open->op_openowner = oo;
2705         if (!oo) {
2706                 clp = find_confirmed_client(clientid, cstate->minorversion,
2707                                             nn);
2708                 if (clp == NULL)
2709                         return nfserr_expired;
2710                 goto new_owner;
2711         }
2712         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2713                 /* Replace unconfirmed owners without checking for replay. */
2714                 clp = oo->oo_owner.so_client;
2715                 release_openowner(oo);
2716                 open->op_openowner = NULL;
2717                 goto new_owner;
2718         }
2719         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2720         if (status)
2721                 return status;
2722         clp = oo->oo_owner.so_client;
2723         goto alloc_stateid;
2724 new_owner:
2725         oo = alloc_init_open_stateowner(strhashval, clp, open);
2726         if (oo == NULL)
2727                 return nfserr_jukebox;
2728         open->op_openowner = oo;
2729 alloc_stateid:
2730         open->op_stp = nfs4_alloc_stateid(clp);
2731         if (!open->op_stp)
2732                 return nfserr_jukebox;
2733         return nfs_ok;
2734 }
2735
2736 static inline __be32
2737 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2738 {
2739         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2740                 return nfserr_openmode;
2741         else
2742                 return nfs_ok;
2743 }
2744
2745 static int share_access_to_flags(u32 share_access)
2746 {
2747         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2748 }
2749
2750 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2751 {
2752         struct nfs4_stid *ret;
2753
2754         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2755         if (!ret)
2756                 return NULL;
2757         return delegstateid(ret);
2758 }
2759
2760 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2761 {
2762         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2763                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2764 }
2765
2766 static __be32
2767 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
2768                 struct nfs4_delegation **dp)
2769 {
2770         int flags;
2771         __be32 status = nfserr_bad_stateid;
2772
2773         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2774         if (*dp == NULL)
2775                 goto out;
2776         flags = share_access_to_flags(open->op_share_access);
2777         status = nfs4_check_delegmode(*dp, flags);
2778         if (status)
2779                 *dp = NULL;
2780 out:
2781         if (!nfsd4_is_deleg_cur(open))
2782                 return nfs_ok;
2783         if (status)
2784                 return status;
2785         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2786         return nfs_ok;
2787 }
2788
2789 static __be32
2790 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2791 {
2792         struct nfs4_ol_stateid *local;
2793         struct nfs4_openowner *oo = open->op_openowner;
2794
2795         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2796                 /* ignore lock owners */
2797                 if (local->st_stateowner->so_is_open_owner == 0)
2798                         continue;
2799                 /* remember if we have seen this open owner */
2800                 if (local->st_stateowner == &oo->oo_owner)
2801                         *stpp = local;
2802                 /* check for conflicting share reservations */
2803                 if (!test_share(local, open))
2804                         return nfserr_share_denied;
2805         }
2806         return nfs_ok;
2807 }
2808
2809 static inline int nfs4_access_to_access(u32 nfs4_access)
2810 {
2811         int flags = 0;
2812
2813         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2814                 flags |= NFSD_MAY_READ;
2815         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2816                 flags |= NFSD_MAY_WRITE;
2817         return flags;
2818 }
2819
2820 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2821                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2822 {
2823         __be32 status;
2824         int oflag = nfs4_access_to_omode(open->op_share_access);
2825         int access = nfs4_access_to_access(open->op_share_access);
2826
2827         if (!fp->fi_fds[oflag]) {
2828                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2829                         &fp->fi_fds[oflag]);
2830                 if (status)
2831                         return status;
2832         }
2833         nfs4_file_get_access(fp, oflag);
2834
2835         return nfs_ok;
2836 }
2837
2838 static inline __be32
2839 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2840                 struct nfsd4_open *open)
2841 {
2842         struct iattr iattr = {
2843                 .ia_valid = ATTR_SIZE,
2844                 .ia_size = 0,
2845         };
2846         if (!open->op_truncate)
2847                 return 0;
2848         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2849                 return nfserr_inval;
2850         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2851 }
2852
2853 static __be32
2854 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
2855 {
2856         u32 op_share_access = open->op_share_access;
2857         bool new_access;
2858         __be32 status;
2859
2860         new_access = !test_access(op_share_access, stp);
2861         if (new_access) {
2862                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2863                 if (status)
2864                         return status;
2865         }
2866         status = nfsd4_truncate(rqstp, cur_fh, open);
2867         if (status) {
2868                 if (new_access) {
2869                         int oflag = nfs4_access_to_omode(op_share_access);
2870                         nfs4_file_put_access(fp, oflag);
2871                 }
2872                 return status;
2873         }
2874         /* remember the open */
2875         set_access(op_share_access, stp);
2876         set_deny(open->op_share_deny, stp);
2877
2878         return nfs_ok;
2879 }
2880
2881
2882 static void
2883 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2884 {
2885         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2886 }
2887
2888 /* Should we give out recallable state?: */
2889 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2890 {
2891         if (clp->cl_cb_state == NFSD4_CB_UP)
2892                 return true;
2893         /*
2894          * In the sessions case, since we don't have to establish a
2895          * separate connection for callbacks, we assume it's OK
2896          * until we hear otherwise:
2897          */
2898         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2899 }
2900
2901 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2902 {
2903         struct file_lock *fl;
2904
2905         fl = locks_alloc_lock();
2906         if (!fl)
2907                 return NULL;
2908         locks_init_lock(fl);
2909         fl->fl_lmops = &nfsd_lease_mng_ops;
2910         fl->fl_flags = FL_LEASE;
2911         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2912         fl->fl_end = OFFSET_MAX;
2913         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2914         fl->fl_pid = current->tgid;
2915         return fl;
2916 }
2917
2918 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2919 {
2920         struct nfs4_file *fp = dp->dl_file;
2921         struct file_lock *fl;
2922         int status;
2923
2924         fl = nfs4_alloc_init_lease(dp, flag);
2925         if (!fl)
2926                 return -ENOMEM;
2927         fl->fl_file = find_readable_file(fp);
2928         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2929         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2930         if (status) {
2931                 list_del_init(&dp->dl_perclnt);
2932                 locks_free_lock(fl);
2933                 return -ENOMEM;
2934         }
2935         fp->fi_lease = fl;
2936         fp->fi_deleg_file = get_file(fl->fl_file);
2937         atomic_set(&fp->fi_delegees, 1);
2938         list_add(&dp->dl_perfile, &fp->fi_delegations);
2939         return 0;
2940 }
2941
2942 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2943 {
2944         struct nfs4_file *fp = dp->dl_file;
2945
2946         if (!fp->fi_lease)
2947                 return nfs4_setlease(dp, flag);
2948         spin_lock(&recall_lock);
2949         if (fp->fi_had_conflict) {
2950                 spin_unlock(&recall_lock);
2951                 return -EAGAIN;
2952         }
2953         atomic_inc(&fp->fi_delegees);
2954         list_add(&dp->dl_perfile, &fp->fi_delegations);
2955         spin_unlock(&recall_lock);
2956         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2957         return 0;
2958 }
2959
2960 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2961 {
2962         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2963         if (status == -EAGAIN)
2964                 open->op_why_no_deleg = WND4_CONTENTION;
2965         else {
2966                 open->op_why_no_deleg = WND4_RESOURCE;
2967                 switch (open->op_deleg_want) {
2968                 case NFS4_SHARE_WANT_READ_DELEG:
2969                 case NFS4_SHARE_WANT_WRITE_DELEG:
2970                 case NFS4_SHARE_WANT_ANY_DELEG:
2971                         break;
2972                 case NFS4_SHARE_WANT_CANCEL:
2973                         open->op_why_no_deleg = WND4_CANCELLED;
2974                         break;
2975                 case NFS4_SHARE_WANT_NO_DELEG:
2976                         WARN_ON_ONCE(1);
2977                 }
2978         }
2979 }
2980
2981 /*
2982  * Attempt to hand out a delegation.
2983  */
2984 static void
2985 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2986                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2987 {
2988         struct nfs4_delegation *dp;
2989         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2990         int cb_up;
2991         int status = 0, flag = 0;
2992
2993         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2994         flag = NFS4_OPEN_DELEGATE_NONE;
2995         open->op_recall = 0;
2996         switch (open->op_claim_type) {
2997                 case NFS4_OPEN_CLAIM_PREVIOUS:
2998                         if (!cb_up)
2999                                 open->op_recall = 1;
3000                         flag = open->op_delegate_type;
3001                         if (flag == NFS4_OPEN_DELEGATE_NONE)
3002                                 goto out;
3003                         break;
3004                 case NFS4_OPEN_CLAIM_NULL:
3005                         /* Let's not give out any delegations till everyone's
3006                          * had the chance to reclaim theirs.... */
3007                         if (locks_in_grace(net))
3008                                 goto out;
3009                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3010                                 goto out;
3011                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3012                                 flag = NFS4_OPEN_DELEGATE_WRITE;
3013                         else
3014                                 flag = NFS4_OPEN_DELEGATE_READ;
3015                         break;
3016                 default:
3017                         goto out;
3018         }
3019
3020         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
3021         if (dp == NULL)
3022                 goto out_no_deleg;
3023         status = nfs4_set_delegation(dp, flag);
3024         if (status)
3025                 goto out_free;
3026
3027         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3028
3029         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3030                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3031 out:
3032         open->op_delegate_type = flag;
3033         if (flag == NFS4_OPEN_DELEGATE_NONE) {
3034                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3035                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
3036                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3037
3038                 /* 4.1 client asking for a delegation? */
3039                 if (open->op_deleg_want)
3040                         nfsd4_open_deleg_none_ext(open, status);
3041         }
3042         return;
3043 out_free:
3044         unhash_stid(&dp->dl_stid);
3045         nfs4_put_delegation(dp);
3046 out_no_deleg:
3047         flag = NFS4_OPEN_DELEGATE_NONE;
3048         goto out;
3049 }
3050
3051 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3052                                         struct nfs4_delegation *dp)
3053 {
3054         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3055             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3056                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3057                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3058         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3059                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3060                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3061                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3062         }
3063         /* Otherwise the client must be confused wanting a delegation
3064          * it already has, therefore we don't return
3065          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3066          */
3067 }
3068
3069 /*
3070  * called with nfs4_lock_state() held.
3071  */
3072 __be32
3073 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3074 {
3075         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3076         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3077         struct nfs4_file *fp = NULL;
3078         struct inode *ino = current_fh->fh_dentry->d_inode;
3079         struct nfs4_ol_stateid *stp = NULL;
3080         struct nfs4_delegation *dp = NULL;
3081         __be32 status;
3082
3083         /*
3084          * Lookup file; if found, lookup stateid and check open request,
3085          * and check for delegations in the process of being recalled.
3086          * If not found, create the nfs4_file struct
3087          */
3088         fp = find_file(ino);
3089         if (fp) {
3090                 if ((status = nfs4_check_open(fp, open, &stp)))
3091                         goto out;
3092                 status = nfs4_check_deleg(cl, open, &dp);
3093                 if (status)
3094                         goto out;
3095         } else {
3096                 status = nfserr_bad_stateid;
3097                 if (nfsd4_is_deleg_cur(open))
3098                         goto out;
3099                 status = nfserr_jukebox;
3100                 fp = open->op_file;
3101                 open->op_file = NULL;
3102                 nfsd4_init_file(fp, ino);
3103         }
3104
3105         /*
3106          * OPEN the file, or upgrade an existing OPEN.
3107          * If truncate fails, the OPEN fails.
3108          */
3109         if (stp) {
3110                 /* Stateid was found, this is an OPEN upgrade */
3111                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3112                 if (status)
3113                         goto out;
3114         } else {
3115                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3116                 if (status)
3117                         goto out;
3118                 status = nfsd4_truncate(rqstp, current_fh, open);
3119                 if (status)
3120                         goto out;
3121                 stp = open->op_stp;
3122                 open->op_stp = NULL;
3123                 init_open_stateid(stp, fp, open);
3124         }
3125         update_stateid(&stp->st_stid.sc_stateid);
3126         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3127
3128         if (nfsd4_has_session(&resp->cstate)) {
3129                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3130
3131                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3132                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3133                         open->op_why_no_deleg = WND4_NOT_WANTED;
3134                         goto nodeleg;
3135                 }
3136         }
3137
3138         /*
3139         * Attempt to hand out a delegation. No error return, because the
3140         * OPEN succeeds even if we fail.
3141         */
3142         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3143 nodeleg:
3144         status = nfs_ok;
3145
3146         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3147                 STATEID_VAL(&stp->st_stid.sc_stateid));
3148 out:
3149         /* 4.1 client trying to upgrade/downgrade delegation? */
3150         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3151             open->op_deleg_want)
3152                 nfsd4_deleg_xgrade_none_ext(open, dp);
3153
3154         if (fp)
3155                 put_nfs4_file(fp);
3156         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3157                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3158         /*
3159         * To finish the open response, we just need to set the rflags.
3160         */
3161         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3162         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3163             !nfsd4_has_session(&resp->cstate))
3164                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3165
3166         return status;
3167 }
3168
3169 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3170 {
3171         if (open->op_openowner) {
3172                 struct nfs4_openowner *oo = open->op_openowner;
3173
3174                 if (!list_empty(&oo->oo_owner.so_stateids))
3175                         list_del_init(&oo->oo_close_lru);
3176                 if (oo->oo_flags & NFS4_OO_NEW) {
3177                         if (status) {
3178                                 release_openowner(oo);
3179                                 open->op_openowner = NULL;
3180                         } else
3181                                 oo->oo_flags &= ~NFS4_OO_NEW;
3182                 }
3183         }
3184         if (open->op_file)
3185                 nfsd4_free_file(open->op_file);
3186         if (open->op_stp)
3187                 free_generic_stateid(open->op_stp);
3188 }
3189
3190 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3191 {
3192         struct nfs4_client *found;
3193
3194         if (STALE_CLIENTID(clid, nn))
3195                 return nfserr_stale_clientid;
3196         found = find_confirmed_client(clid, session, nn);
3197         if (clp)
3198                 *clp = found;
3199         return found ? nfs_ok : nfserr_expired;
3200 }
3201
3202 __be32
3203 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3204             clientid_t *clid)
3205 {
3206         struct nfs4_client *clp;
3207         __be32 status;
3208         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3209
3210         nfs4_lock_state();
3211         dprintk("process_renew(%08x/%08x): starting\n",
3212                         clid->cl_boot, clid->cl_id);
3213         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3214         if (status)
3215                 goto out;
3216         status = nfserr_cb_path_down;
3217         if (!list_empty(&clp->cl_delegations)
3218                         && clp->cl_cb_state != NFSD4_CB_UP)
3219                 goto out;
3220         status = nfs_ok;
3221 out:
3222         nfs4_unlock_state();
3223         return status;
3224 }
3225
3226 static void
3227 nfsd4_end_grace(struct nfsd_net *nn)
3228 {
3229         /* do nothing if grace period already ended */
3230         if (nn->grace_ended)
3231                 return;
3232
3233         dprintk("NFSD: end of grace period\n");
3234         nn->grace_ended = true;
3235         nfsd4_record_grace_done(nn, nn->boot_time);
3236         locks_end_grace(&nn->nfsd4_manager);
3237         /*
3238          * Now that every NFSv4 client has had the chance to recover and
3239          * to see the (possibly new, possibly shorter) lease time, we
3240          * can safely set the next grace time to the current lease time:
3241          */
3242         nn->nfsd4_grace = nn->nfsd4_lease;
3243 }
3244
3245 static time_t
3246 nfs4_laundromat(struct nfsd_net *nn)
3247 {
3248         struct nfs4_client *clp;
3249         struct nfs4_openowner *oo;
3250         struct nfs4_delegation *dp;
3251         struct list_head *pos, *next, reaplist;
3252         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3253         time_t t, clientid_val = nn->nfsd4_lease;
3254         time_t u, test_val = nn->nfsd4_lease;
3255
3256         nfs4_lock_state();
3257
3258         dprintk("NFSD: laundromat service - starting\n");
3259         nfsd4_end_grace(nn);
3260         INIT_LIST_HEAD(&reaplist);
3261         spin_lock(&nn->client_lock);
3262         list_for_each_safe(pos, next, &nn->client_lru) {
3263                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3264                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3265                         t = clp->cl_time - cutoff;
3266                         if (clientid_val > t)
3267                                 clientid_val = t;
3268                         break;
3269                 }
3270                 if (mark_client_expired_locked(clp)) {
3271                         dprintk("NFSD: client in use (clientid %08x)\n",
3272                                 clp->cl_clientid.cl_id);
3273                         continue;
3274                 }
3275                 list_move(&clp->cl_lru, &reaplist);
3276         }
3277         spin_unlock(&nn->client_lock);
3278         list_for_each_safe(pos, next, &reaplist) {
3279                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3280                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3281                         clp->cl_clientid.cl_id);
3282                 expire_client(clp);
3283         }
3284         spin_lock(&recall_lock);
3285         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3286                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3287                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3288                         continue;
3289                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3290                         u = dp->dl_time - cutoff;
3291                         if (test_val > u)
3292                                 test_val = u;
3293                         break;
3294                 }
3295                 list_move(&dp->dl_recall_lru, &reaplist);
3296         }
3297         spin_unlock(&recall_lock);
3298         list_for_each_safe(pos, next, &reaplist) {
3299                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3300                 unhash_delegation(dp);
3301         }
3302         test_val = nn->nfsd4_lease;
3303         list_for_each_safe(pos, next, &nn->close_lru) {
3304                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3305                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3306                         u = oo->oo_time - cutoff;
3307                         if (test_val > u)
3308                                 test_val = u;
3309                         break;
3310                 }
3311                 release_openowner(oo);
3312         }
3313         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3314                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3315         nfs4_unlock_state();
3316         return clientid_val;
3317 }
3318
3319 static struct workqueue_struct *laundry_wq;
3320 static void laundromat_main(struct work_struct *);
3321
3322 static void
3323 laundromat_main(struct work_struct *laundry)
3324 {
3325         time_t t;
3326         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3327                                                   work);
3328         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3329                                            laundromat_work);
3330
3331         t = nfs4_laundromat(nn);
3332         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3333         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3334 }
3335
3336 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3337 {
3338         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3339                 return nfserr_bad_stateid;
3340         return nfs_ok;
3341 }
3342
3343 static inline int
3344 access_permit_read(struct nfs4_ol_stateid *stp)
3345 {
3346         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3347                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3348                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3349 }
3350
3351 static inline int
3352 access_permit_write(struct nfs4_ol_stateid *stp)
3353 {
3354         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3355                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3356 }
3357
3358 static
3359 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3360 {
3361         __be32 status = nfserr_openmode;
3362
3363         /* For lock stateid's, we test the parent open, not the lock: */
3364         if (stp->st_openstp)
3365                 stp = stp->st_openstp;
3366         if ((flags & WR_STATE) && !access_permit_write(stp))
3367                 goto out;
3368         if ((flags & RD_STATE) && !access_permit_read(stp))
3369                 goto out;
3370         status = nfs_ok;
3371 out:
3372         return status;
3373 }
3374
3375 static inline __be32
3376 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3377 {
3378         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3379                 return nfs_ok;
3380         else if (locks_in_grace(net)) {
3381                 /* Answer in remaining cases depends on existence of
3382                  * conflicting state; so we must wait out the grace period. */
3383                 return nfserr_grace;
3384         } else if (flags & WR_STATE)
3385                 return nfs4_share_conflict(current_fh,
3386                                 NFS4_SHARE_DENY_WRITE);
3387         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3388                 return nfs4_share_conflict(current_fh,
3389                                 NFS4_SHARE_DENY_READ);
3390 }
3391
3392 /*
3393  * Allow READ/WRITE during grace period on recovered state only for files
3394  * that are not able to provide mandatory locking.
3395  */
3396 static inline int
3397 grace_disallows_io(struct net *net, struct inode *inode)
3398 {
3399         return locks_in_grace(net) && mandatory_lock(inode);
3400 }
3401
3402 /* Returns true iff a is later than b: */
3403 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3404 {
3405         return (s32)a->si_generation - (s32)b->si_generation > 0;
3406 }
3407
3408 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3409 {
3410         /*
3411          * When sessions are used the stateid generation number is ignored
3412          * when it is zero.
3413          */
3414         if (has_session && in->si_generation == 0)
3415                 return nfs_ok;
3416
3417         if (in->si_generation == ref->si_generation)
3418                 return nfs_ok;
3419
3420         /* If the client sends us a stateid from the future, it's buggy: */
3421         if (stateid_generation_after(in, ref))
3422                 return nfserr_bad_stateid;
3423         /*
3424          * However, we could see a stateid from the past, even from a
3425          * non-buggy client.  For example, if the client sends a lock
3426          * while some IO is outstanding, the lock may bump si_generation
3427          * while the IO is still in flight.  The client could avoid that
3428          * situation by waiting for responses on all the IO requests,
3429          * but better performance may result in retrying IO that
3430          * receives an old_stateid error if requests are rarely
3431          * reordered in flight:
3432          */
3433         return nfserr_old_stateid;
3434 }
3435
3436 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3437 {
3438         struct nfs4_stid *s;
3439         struct nfs4_ol_stateid *ols;
3440         __be32 status;
3441
3442         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3443                 return nfserr_bad_stateid;
3444         /* Client debugging aid. */
3445         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3446                 char addr_str[INET6_ADDRSTRLEN];
3447                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3448                                  sizeof(addr_str));
3449                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3450                                         "with incorrect client ID\n", addr_str);
3451                 return nfserr_bad_stateid;
3452         }
3453         s = find_stateid(cl, stateid);
3454         if (!s)
3455                 return nfserr_bad_stateid;
3456         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3457         if (status)
3458                 return status;
3459         switch (s->sc_type) {
3460         case NFS4_DELEG_STID:
3461                 return nfs_ok;
3462         case NFS4_OPEN_STID:
3463         case NFS4_LOCK_STID:
3464                 ols = openlockstateid(s);
3465                 if (ols->st_stateowner->so_is_open_owner
3466                                 && !(openowner(ols->st_stateowner)->oo_flags
3467                                                 & NFS4_OO_CONFIRMED))
3468                         return nfserr_bad_stateid;
3469                 return nfs_ok;
3470         default:
3471                 printk("unknown stateid type %x\n", s->sc_type);
3472         case NFS4_CLOSED_STID:
3473                 return nfserr_bad_stateid;
3474         }
3475 }
3476
3477 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3478                                    struct nfs4_stid **s, bool sessions,
3479                                    struct nfsd_net *nn)
3480 {
3481         struct nfs4_client *cl;
3482         __be32 status;
3483
3484         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3485                 return nfserr_bad_stateid;
3486         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3487                                                         nn, &cl);
3488         if (status == nfserr_stale_clientid)
3489                 return nfserr_stale_stateid;
3490         if (status)
3491                 return status;
3492         *s = find_stateid_by_type(cl, stateid, typemask);
3493         if (!*s)
3494                 return nfserr_bad_stateid;
3495         return nfs_ok;
3496 }
3497
3498 /*
3499 * Checks for stateid operations
3500 */
3501 __be32
3502 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3503                            stateid_t *stateid, int flags, struct file **filpp)
3504 {
3505         struct nfs4_stid *s;
3506         struct nfs4_ol_stateid *stp = NULL;
3507         struct nfs4_delegation *dp = NULL;
3508         struct svc_fh *current_fh = &cstate->current_fh;
3509         struct inode *ino = current_fh->fh_dentry->d_inode;
3510         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3511         __be32 status;
3512
3513         if (filpp)
3514                 *filpp = NULL;
3515
3516         if (grace_disallows_io(net, ino))
3517                 return nfserr_grace;
3518
3519         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3520                 return check_special_stateids(net, current_fh, stateid, flags);
3521
3522         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3523                                       &s, cstate->minorversion, nn);
3524         if (status)
3525                 return status;
3526         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3527         if (status)
3528                 goto out;
3529         switch (s->sc_type) {
3530         case NFS4_DELEG_STID:
3531                 dp = delegstateid(s);
3532                 status = nfs4_check_delegmode(dp, flags);
3533                 if (status)
3534                         goto out;
3535                 if (filpp) {
3536                         *filpp = dp->dl_file->fi_deleg_file;
3537                         if (!*filpp) {
3538                                 WARN_ON_ONCE(1);
3539                                 status = nfserr_serverfault;
3540                                 goto out;
3541                         }
3542                 }
3543                 break;
3544         case NFS4_OPEN_STID:
3545         case NFS4_LOCK_STID:
3546                 stp = openlockstateid(s);
3547                 status = nfs4_check_fh(current_fh, stp);
3548                 if (status)
3549                         goto out;
3550                 if (stp->st_stateowner->so_is_open_owner
3551                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3552                         goto out;
3553                 status = nfs4_check_openmode(stp, flags);
3554                 if (status)
3555                         goto out;
3556                 if (filpp) {
3557                         if (flags & RD_STATE)
3558                                 *filpp = find_readable_file(stp->st_file);
3559                         else
3560                                 *filpp = find_writeable_file(stp->st_file);
3561                 }
3562                 break;
3563         default:
3564                 return nfserr_bad_stateid;
3565         }
3566         status = nfs_ok;
3567 out:
3568         return status;
3569 }
3570
3571 static __be32
3572 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3573 {
3574         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3575                 return nfserr_locks_held;
3576         release_lock_stateid(stp);
3577         return nfs_ok;
3578 }
3579
3580 /*
3581  * Test if the stateid is valid
3582  */
3583 __be32
3584 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3585                    struct nfsd4_test_stateid *test_stateid)
3586 {
3587         struct nfsd4_test_stateid_id *stateid;
3588         struct nfs4_client *cl = cstate->session->se_client;
3589
3590         nfs4_lock_state();
3591         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3592                 stateid->ts_id_status =
3593                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3594         nfs4_unlock_state();
3595
3596         return nfs_ok;
3597 }
3598
3599 __be32
3600 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3601                    struct nfsd4_free_stateid *free_stateid)
3602 {
3603         stateid_t *stateid = &free_stateid->fr_stateid;
3604         struct nfs4_stid *s;
3605         struct nfs4_client *cl = cstate->session->se_client;
3606         __be32 ret = nfserr_bad_stateid;
3607
3608         nfs4_lock_state();
3609         s = find_stateid(cl, stateid);
3610         if (!s)
3611                 goto out;
3612         switch (s->sc_type) {
3613         case NFS4_DELEG_STID:
3614                 ret = nfserr_locks_held;
3615                 goto out;
3616         case NFS4_OPEN_STID:
3617         case NFS4_LOCK_STID:
3618                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3619                 if (ret)
3620                         goto out;
3621                 if (s->sc_type == NFS4_LOCK_STID)
3622                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3623                 else
3624                         ret = nfserr_locks_held;
3625                 break;
3626         default:
3627                 ret = nfserr_bad_stateid;
3628         }
3629 out:
3630         nfs4_unlock_state();
3631         return ret;
3632 }
3633
3634 static inline int
3635 setlkflg (int type)
3636 {
3637         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3638                 RD_STATE : WR_STATE;
3639 }
3640
3641 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3642 {
3643         struct svc_fh *current_fh = &cstate->current_fh;
3644         struct nfs4_stateowner *sop = stp->st_stateowner;
3645         __be32 status;
3646
3647         status = nfsd4_check_seqid(cstate, sop, seqid);
3648         if (status)
3649                 return status;
3650         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3651                 /*
3652                  * "Closed" stateid's exist *only* to return
3653                  * nfserr_replay_me from the previous step.
3654                  */
3655                 return nfserr_bad_stateid;
3656         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3657         if (status)
3658                 return status;
3659         return nfs4_check_fh(current_fh, stp);
3660 }
3661
3662 /*
3663  * Checks for sequence id mutating operations.
3664  */
3665 static __be32
3666 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3667                          stateid_t *stateid, char typemask,
3668                          struct nfs4_ol_stateid **stpp,
3669                          struct nfsd_net *nn)
3670 {
3671         __be32 status;
3672         struct nfs4_stid *s;
3673
3674         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3675                 seqid, STATEID_VAL(stateid));
3676
3677         *stpp = NULL;
3678         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3679                                       cstate->minorversion, nn);
3680         if (status)
3681                 return status;
3682         *stpp = openlockstateid(s);
3683         if (!nfsd4_has_session(cstate))
3684                 cstate->replay_owner = (*stpp)->st_stateowner;
3685
3686         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3687 }
3688
3689 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3690                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3691 {
3692         __be32 status;
3693         struct nfs4_openowner *oo;
3694
3695         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3696                                                 NFS4_OPEN_STID, stpp, nn);
3697         if (status)
3698                 return status;
3699         oo = openowner((*stpp)->st_stateowner);
3700         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3701                 return nfserr_bad_stateid;
3702         return nfs_ok;
3703 }
3704
3705 __be32
3706 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3707                    struct nfsd4_open_confirm *oc)
3708 {
3709         __be32 status;
3710         struct nfs4_openowner *oo;
3711         struct nfs4_ol_stateid *stp;
3712         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3713
3714         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3715                         (int)cstate->current_fh.fh_dentry->d_name.len,
3716                         cstate->current_fh.fh_dentry->d_name.name);
3717
3718         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3719         if (status)
3720                 return status;
3721
3722         nfs4_lock_state();
3723
3724         status = nfs4_preprocess_seqid_op(cstate,
3725                                         oc->oc_seqid, &oc->oc_req_stateid,
3726                                         NFS4_OPEN_STID, &stp, nn);
3727         if (status)
3728                 goto out;
3729         oo = openowner(stp->st_stateowner);
3730         status = nfserr_bad_stateid;
3731         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3732                 goto out;
3733         oo->oo_flags |= NFS4_OO_CONFIRMED;
3734         update_stateid(&stp->st_stid.sc_stateid);
3735         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3736         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3737                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3738
3739         nfsd4_client_record_create(oo->oo_owner.so_client);
3740         status = nfs_ok;
3741 out:
3742         nfsd4_bump_seqid(cstate, status);
3743         if (!cstate->replay_owner)
3744                 nfs4_unlock_state();
3745         return status;
3746 }
3747
3748 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3749 {
3750         if (!test_access(access, stp))
3751                 return;
3752         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3753         clear_access(access, stp);
3754 }
3755
3756 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3757 {
3758         switch (to_access) {
3759         case NFS4_SHARE_ACCESS_READ:
3760                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3761                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3762                 break;
3763         case NFS4_SHARE_ACCESS_WRITE:
3764                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3765                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3766                 break;
3767         case NFS4_SHARE_ACCESS_BOTH:
3768                 break;
3769         default:
3770                 WARN_ON_ONCE(1);
3771         }
3772 }
3773
3774 static void
3775 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3776 {
3777         int i;
3778         for (i = 0; i < 4; i++) {
3779                 if ((i & deny) != i)
3780                         clear_deny(i, stp);
3781         }
3782 }
3783
3784 __be32
3785 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3786                      struct nfsd4_compound_state *cstate,
3787                      struct nfsd4_open_downgrade *od)
3788 {
3789         __be32 status;
3790         struct nfs4_ol_stateid *stp;
3791         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3792
3793         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3794                         (int)cstate->current_fh.fh_dentry->d_name.len,
3795                         cstate->current_fh.fh_dentry->d_name.name);
3796
3797         /* We don't yet support WANT bits: */
3798         if (od->od_deleg_want)
3799                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3800                         od->od_deleg_want);
3801
3802         nfs4_lock_state();
3803         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3804                                         &od->od_stateid, &stp, nn);
3805         if (status)
3806                 goto out;
3807         status = nfserr_inval;
3808         if (!test_access(od->od_share_access, stp)) {
3809                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3810                         stp->st_access_bmap, od->od_share_access);
3811                 goto out;
3812         }
3813         if (!test_deny(od->od_share_deny, stp)) {
3814                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3815                         stp->st_deny_bmap, od->od_share_deny);
3816                 goto out;
3817         }
3818         nfs4_stateid_downgrade(stp, od->od_share_access);
3819
3820         reset_union_bmap_deny(od->od_share_deny, stp);
3821
3822         update_stateid(&stp->st_stid.sc_stateid);
3823         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3824         status = nfs_ok;
3825 out:
3826         nfsd4_bump_seqid(cstate, status);
3827         if (!cstate->replay_owner)
3828                 nfs4_unlock_state();
3829         return status;
3830 }
3831
3832 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3833 {
3834         unhash_open_stateid(s);
3835         s->st_stid.sc_type = NFS4_CLOSED_STID;
3836 }
3837
3838 /*
3839  * nfs4_unlock_state() called after encode
3840  */
3841 __be32
3842 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3843             struct nfsd4_close *close)
3844 {
3845         __be32 status;
3846         struct nfs4_openowner *oo;
3847         struct nfs4_ol_stateid *stp;
3848         struct net *net = SVC_NET(rqstp);
3849         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3850
3851         dprintk("NFSD: nfsd4_close on file %.*s\n",
3852                         (int)cstate->current_fh.fh_dentry->d_name.len,
3853                         cstate->current_fh.fh_dentry->d_name.name);
3854
3855         nfs4_lock_state();
3856         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3857                                         &close->cl_stateid,
3858                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3859                                         &stp, nn);
3860         nfsd4_bump_seqid(cstate, status);
3861         if (status)
3862                 goto out;
3863         oo = openowner(stp->st_stateowner);
3864         update_stateid(&stp->st_stid.sc_stateid);
3865         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3866
3867         nfsd4_close_open_stateid(stp);
3868
3869         if (cstate->minorversion) {
3870                 unhash_stid(&stp->st_stid);
3871                 free_generic_stateid(stp);
3872         } else
3873                 oo->oo_last_closed_stid = stp;
3874
3875         if (list_empty(&oo->oo_owner.so_stateids)) {
3876                 if (cstate->minorversion)
3877                         release_openowner(oo);
3878                 else {
3879                         /*
3880                          * In the 4.0 case we need to keep the owners around a
3881                          * little while to handle CLOSE replay.
3882                          */
3883                         if (list_empty(&oo->oo_owner.so_stateids))
3884                                 move_to_close_lru(oo, SVC_NET(rqstp));
3885                 }
3886         }
3887 out:
3888         if (!cstate->replay_owner)
3889                 nfs4_unlock_state();
3890         return status;
3891 }
3892
3893 __be32
3894 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3895                   struct nfsd4_delegreturn *dr)
3896 {
3897         struct nfs4_delegation *dp;
3898         stateid_t *stateid = &dr->dr_stateid;
3899         struct nfs4_stid *s;
3900         __be32 status;
3901         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3902
3903         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3904                 return status;
3905
3906         nfs4_lock_state();
3907         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
3908                                       cstate->minorversion, nn);
3909         if (status)
3910                 goto out;
3911         dp = delegstateid(s);
3912         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3913         if (status)
3914                 goto out;
3915
3916         unhash_delegation(dp);
3917 out:
3918         nfs4_unlock_state();
3919
3920         return status;
3921 }
3922
3923
3924 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3925
3926 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3927
3928 static inline u64
3929 end_offset(u64 start, u64 len)
3930 {
3931         u64 end;
3932
3933         end = start + len;
3934         return end >= start ? end: NFS4_MAX_UINT64;
3935 }
3936
3937 /* last octet in a range */
3938 static inline u64
3939 last_byte_offset(u64 start, u64 len)
3940 {
3941         u64 end;
3942
3943         WARN_ON_ONCE(!len);
3944         end = start + len;
3945         return end > start ? end - 1: NFS4_MAX_UINT64;
3946 }
3947
3948 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3949 {
3950         return (file_hashval(inode) + cl_id
3951                         + opaque_hashval(ownername->data, ownername->len))
3952                 & LOCKOWNER_INO_HASH_MASK;
3953 }
3954
3955 /*
3956  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3957  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3958  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3959  * locking, this prevents us from being completely protocol-compliant.  The
3960  * real solution to this problem is to start using unsigned file offsets in
3961  * the VFS, but this is a very deep change!
3962  */
3963 static inline void
3964 nfs4_transform_lock_offset(struct file_lock *lock)
3965 {
3966         if (lock->fl_start < 0)
3967                 lock->fl_start = OFFSET_MAX;
3968         if (lock->fl_end < 0)
3969                 lock->fl_end = OFFSET_MAX;
3970 }
3971
3972 /* Hack!: For now, we're defining this just so we can use a pointer to it
3973  * as a unique cookie to identify our (NFSv4's) posix locks. */
3974 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3975 };
3976
3977 static inline void
3978 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3979 {
3980         struct nfs4_lockowner *lo;
3981
3982         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3983                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3984                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3985                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3986                 if (!deny->ld_owner.data)
3987                         /* We just don't care that much */
3988                         goto nevermind;
3989                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3990                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3991         } else {
3992 nevermind:
3993                 deny->ld_owner.len = 0;
3994                 deny->ld_owner.data = NULL;
3995                 deny->ld_clientid.cl_boot = 0;
3996                 deny->ld_clientid.cl_id = 0;
3997         }
3998         deny->ld_start = fl->fl_start;
3999         deny->ld_length = NFS4_MAX_UINT64;
4000         if (fl->fl_end != NFS4_MAX_UINT64)
4001                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4002         deny->ld_type = NFS4_READ_LT;
4003         if (fl->fl_type != F_RDLCK)
4004                 deny->ld_type = NFS4_WRITE_LT;
4005 }
4006
4007 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
4008 {
4009         struct nfs4_ol_stateid *lst;
4010
4011         if (!same_owner_str(&lo->lo_owner, owner, clid))
4012                 return false;
4013         lst = list_first_entry(&lo->lo_owner.so_stateids,
4014                                struct nfs4_ol_stateid, st_perstateowner);
4015         return lst->st_file->fi_inode == inode;
4016 }
4017
4018 static struct nfs4_lockowner *
4019 find_lockowner_str(struct inode *inode, clientid_t *clid,
4020                    struct xdr_netobj *owner, struct nfsd_net *nn)
4021 {
4022         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4023         struct nfs4_lockowner *lo;
4024
4025         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4026                 if (same_lockowner_ino(lo, inode, clid, owner))
4027                         return lo;
4028         }
4029         return NULL;
4030 }
4031
4032 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4033 {
4034         struct inode *inode = open_stp->st_file->fi_inode;
4035         unsigned int inohash = lockowner_ino_hashval(inode,
4036                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4037         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4038
4039         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4040         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4041         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4042 }
4043
4044 /*
4045  * Alloc a lock owner structure.
4046  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4047  * occurred.
4048  *
4049  * strhashval = ownerstr_hashval
4050  */
4051
4052 static struct nfs4_lockowner *
4053 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4054         struct nfs4_lockowner *lo;
4055
4056         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4057         if (!lo)
4058                 return NULL;
4059         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4060         lo->lo_owner.so_is_open_owner = 0;
4061         /* It is the openowner seqid that will be incremented in encode in the
4062          * case of new lockowners; so increment the lock seqid manually: */
4063         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4064         hash_lockowner(lo, strhashval, clp, open_stp);
4065         return lo;
4066 }
4067
4068 static struct nfs4_ol_stateid *
4069 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4070 {
4071         struct nfs4_ol_stateid *stp;
4072         struct nfs4_client *clp = lo->lo_owner.so_client;
4073
4074         stp = nfs4_alloc_stateid(clp);
4075         if (stp == NULL)
4076                 return NULL;
4077         stp->st_stid.sc_type = NFS4_LOCK_STID;
4078         list_add(&stp->st_perfile, &fp->fi_stateids);
4079         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4080         stp->st_stateowner = &lo->lo_owner;
4081         get_nfs4_file(fp);
4082         stp->st_file = fp;
4083         stp->st_access_bmap = 0;
4084         stp->st_deny_bmap = open_stp->st_deny_bmap;
4085         stp->st_openstp = open_stp;
4086         return stp;
4087 }
4088
4089 static int
4090 check_lock_length(u64 offset, u64 length)
4091 {
4092         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4093              LOFF_OVERFLOW(offset, length)));
4094 }
4095
4096 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4097 {
4098         struct nfs4_file *fp = lock_stp->st_file;
4099         int oflag = nfs4_access_to_omode(access);
4100
4101         if (test_access(access, lock_stp))
4102                 return;
4103         nfs4_file_get_access(fp, oflag);
4104         set_access(access, lock_stp);
4105 }
4106
4107 static __be32 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate, struct nfs4_ol_stateid *ost, struct nfsd4_lock *lock, struct nfs4_ol_stateid **lst, bool *new)
4108 {
4109         struct nfs4_file *fi = ost->st_file;
4110         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4111         struct nfs4_client *cl = oo->oo_owner.so_client;
4112         struct nfs4_lockowner *lo;
4113         unsigned int strhashval;
4114         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4115
4116         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4117                                 &lock->v.new.owner, nn);
4118         if (lo) {
4119                 if (!cstate->minorversion)
4120                         return nfserr_bad_seqid;
4121                 /* XXX: a lockowner always has exactly one stateid: */
4122                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4123                                 struct nfs4_ol_stateid, st_perstateowner);
4124                 return nfs_ok;
4125         }
4126         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4127                         &lock->v.new.owner);
4128         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4129         if (lo == NULL)
4130                 return nfserr_jukebox;
4131         *lst = alloc_init_lock_stateid(lo, fi, ost);
4132         if (*lst == NULL) {
4133                 release_lockowner(lo);
4134                 return nfserr_jukebox;
4135         }
4136         *new = true;
4137         return nfs_ok;
4138 }
4139
4140 /*
4141  *  LOCK operation
4142  */
4143 __be32
4144 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4145            struct nfsd4_lock *lock)
4146 {
4147         struct nfs4_openowner *open_sop = NULL;
4148         struct nfs4_lockowner *lock_sop = NULL;
4149         struct nfs4_ol_stateid *lock_stp;
4150         struct file *filp = NULL;
4151         struct file_lock *file_lock = NULL;
4152         struct file_lock *conflock = NULL;
4153         __be32 status = 0;
4154         bool new_state = false;
4155         int lkflg;
4156         int err;
4157         struct net *net = SVC_NET(rqstp);
4158         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4159
4160         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4161                 (long long) lock->lk_offset,
4162                 (long long) lock->lk_length);
4163
4164         if (check_lock_length(lock->lk_offset, lock->lk_length))
4165                  return nfserr_inval;
4166
4167         if ((status = fh_verify(rqstp, &cstate->current_fh,
4168                                 S_IFREG, NFSD_MAY_LOCK))) {
4169                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4170                 return status;
4171         }
4172
4173         nfs4_lock_state();
4174
4175         if (lock->lk_is_new) {
4176                 struct nfs4_ol_stateid *open_stp = NULL;
4177
4178                 if (nfsd4_has_session(cstate))
4179                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4180                         memcpy(&lock->v.new.clientid,
4181                                 &cstate->session->se_client->cl_clientid,
4182                                 sizeof(clientid_t));
4183
4184                 status = nfserr_stale_clientid;
4185                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4186                         goto out;
4187
4188                 /* validate and update open stateid and open seqid */
4189                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4190                                         lock->lk_new_open_seqid,
4191                                         &lock->lk_new_open_stateid,
4192                                         &open_stp, nn);
4193                 if (status)
4194                         goto out;
4195                 open_sop = openowner(open_stp->st_stateowner);
4196                 status = nfserr_bad_stateid;
4197                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4198                                                 &lock->v.new.clientid))
4199                         goto out;
4200                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4201                                                         &lock_stp, &new_state);
4202         } else
4203                 status = nfs4_preprocess_seqid_op(cstate,
4204                                        lock->lk_old_lock_seqid,
4205                                        &lock->lk_old_lock_stateid,
4206                                        NFS4_LOCK_STID, &lock_stp, nn);
4207         if (status)
4208                 goto out;
4209         lock_sop = lockowner(lock_stp->st_stateowner);
4210
4211         lkflg = setlkflg(lock->lk_type);
4212         status = nfs4_check_openmode(lock_stp, lkflg);
4213         if (status)
4214                 goto out;
4215
4216         status = nfserr_grace;
4217         if (locks_in_grace(net) && !lock->lk_reclaim)
4218                 goto out;
4219         status = nfserr_no_grace;
4220         if (!locks_in_grace(net) && lock->lk_reclaim)
4221                 goto out;
4222
4223         file_lock = locks_alloc_lock();
4224         if (!file_lock) {
4225                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4226                 status = nfserr_jukebox;
4227                 goto out;
4228         }
4229
4230         locks_init_lock(file_lock);
4231         switch (lock->lk_type) {
4232                 case NFS4_READ_LT:
4233                 case NFS4_READW_LT:
4234                         filp = find_readable_file(lock_stp->st_file);
4235                         if (filp)
4236                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4237                         file_lock->fl_type = F_RDLCK;
4238                         break;
4239                 case NFS4_WRITE_LT:
4240                 case NFS4_WRITEW_LT:
4241                         filp = find_writeable_file(lock_stp->st_file);
4242                         if (filp)
4243                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4244                         file_lock->fl_type = F_WRLCK;
4245                         break;
4246                 default:
4247                         status = nfserr_inval;
4248                 goto out;
4249         }
4250         if (!filp) {
4251                 status = nfserr_openmode;
4252                 goto out;
4253         }
4254         file_lock->fl_owner = (fl_owner_t)lock_sop;
4255         file_lock->fl_pid = current->tgid;
4256         file_lock->fl_file = filp;
4257         file_lock->fl_flags = FL_POSIX;
4258         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4259         file_lock->fl_start = lock->lk_offset;
4260         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4261         nfs4_transform_lock_offset(file_lock);
4262
4263         conflock = locks_alloc_lock();
4264         if (!conflock) {
4265                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4266                 status = nfserr_jukebox;
4267                 goto out;
4268         }
4269
4270         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4271         switch (-err) {
4272         case 0: /* success! */
4273                 update_stateid(&lock_stp->st_stid.sc_stateid);
4274                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4275                                 sizeof(stateid_t));
4276                 status = 0;
4277                 break;
4278         case (EAGAIN):          /* conflock holds conflicting lock */
4279                 status = nfserr_denied;
4280                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4281                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4282                 break;
4283         case (EDEADLK):
4284                 status = nfserr_deadlock;
4285                 break;
4286         default:
4287                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4288                 status = nfserrno(err);
4289                 break;
4290         }
4291 out:
4292         if (status && new_state)
4293                 release_lockowner(lock_sop);
4294         nfsd4_bump_seqid(cstate, status);
4295         if (!cstate->replay_owner)
4296                 nfs4_unlock_state();
4297         if (file_lock)
4298                 locks_free_lock(file_lock);
4299         if (conflock)
4300                 locks_free_lock(conflock);
4301         return status;
4302 }
4303
4304 /*
4305  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4306  * so we do a temporary open here just to get an open file to pass to
4307  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4308  * inode operation.)
4309  */
4310 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4311 {
4312         struct file *file;
4313         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4314         if (!err) {
4315                 err = nfserrno(vfs_test_lock(file, lock));
4316                 nfsd_close(file);
4317         }
4318         return err;
4319 }
4320
4321 /*
4322  * LOCKT operation
4323  */
4324 __be32
4325 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4326             struct nfsd4_lockt *lockt)
4327 {
4328         struct inode *inode;
4329         struct file_lock *file_lock = NULL;
4330         struct nfs4_lockowner *lo;
4331         __be32 status;
4332         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4333
4334         if (locks_in_grace(SVC_NET(rqstp)))
4335                 return nfserr_grace;
4336
4337         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4338                  return nfserr_inval;
4339
4340         nfs4_lock_state();
4341
4342         if (!nfsd4_has_session(cstate)) {
4343                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4344                 if (status)
4345                         goto out;
4346         }
4347
4348         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4349                 goto out;
4350
4351         inode = cstate->current_fh.fh_dentry->d_inode;
4352         file_lock = locks_alloc_lock();
4353         if (!file_lock) {
4354                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4355                 status = nfserr_jukebox;
4356                 goto out;
4357         }
4358         locks_init_lock(file_lock);
4359         switch (lockt->lt_type) {
4360                 case NFS4_READ_LT:
4361                 case NFS4_READW_LT:
4362                         file_lock->fl_type = F_RDLCK;
4363                 break;
4364                 case NFS4_WRITE_LT:
4365                 case NFS4_WRITEW_LT:
4366                         file_lock->fl_type = F_WRLCK;
4367                 break;
4368                 default:
4369                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4370                         status = nfserr_inval;
4371                 goto out;
4372         }
4373
4374         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4375         if (lo)
4376                 file_lock->fl_owner = (fl_owner_t)lo;
4377         file_lock->fl_pid = current->tgid;
4378         file_lock->fl_flags = FL_POSIX;
4379
4380         file_lock->fl_start = lockt->lt_offset;
4381         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4382
4383         nfs4_transform_lock_offset(file_lock);
4384
4385         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4386         if (status)
4387                 goto out;
4388
4389         if (file_lock->fl_type != F_UNLCK) {
4390                 status = nfserr_denied;
4391                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4392         }
4393 out:
4394         nfs4_unlock_state();
4395         if (file_lock)
4396                 locks_free_lock(file_lock);
4397         return status;
4398 }
4399
4400 __be32
4401 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4402             struct nfsd4_locku *locku)
4403 {
4404         struct nfs4_lockowner *lo;
4405         struct nfs4_ol_stateid *stp;
4406         struct file *filp = NULL;
4407         struct file_lock *file_lock = NULL;
4408         __be32 status;
4409         int err;
4410         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4411
4412         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4413                 (long long) locku->lu_offset,
4414                 (long long) locku->lu_length);
4415
4416         if (check_lock_length(locku->lu_offset, locku->lu_length))
4417                  return nfserr_inval;
4418
4419         nfs4_lock_state();
4420
4421         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4422                                         &locku->lu_stateid, NFS4_LOCK_STID,
4423                                         &stp, nn);
4424         if (status)
4425                 goto out;
4426         filp = find_any_file(stp->st_file);
4427         if (!filp) {
4428                 status = nfserr_lock_range;
4429                 goto out;
4430         }
4431         file_lock = locks_alloc_lock();
4432         if (!file_lock) {
4433                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4434                 status = nfserr_jukebox;
4435                 goto out;
4436         }
4437         lo = lockowner(stp->st_stateowner);
4438         locks_init_lock(file_lock);
4439         file_lock->fl_type = F_UNLCK;
4440         file_lock->fl_owner = (fl_owner_t)lo;
4441         file_lock->fl_pid = current->tgid;
4442         file_lock->fl_file = filp;
4443         file_lock->fl_flags = FL_POSIX;
4444         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4445         file_lock->fl_start = locku->lu_offset;
4446
4447         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4448                                                 locku->lu_length);
4449         nfs4_transform_lock_offset(file_lock);
4450
4451         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4452         if (err) {
4453                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4454                 goto out_nfserr;
4455         }
4456         update_stateid(&stp->st_stid.sc_stateid);
4457         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4458
4459         if (nfsd4_has_session(cstate) && !check_for_locks(stp->st_file, lo)) {
4460                 WARN_ON_ONCE(cstate->replay_owner);
4461                 release_lockowner(lo);
4462         }
4463
4464 out:
4465         nfsd4_bump_seqid(cstate, status);
4466         if (!cstate->replay_owner)
4467                 nfs4_unlock_state();
4468         if (file_lock)
4469                 locks_free_lock(file_lock);
4470         return status;
4471
4472 out_nfserr:
4473         status = nfserrno(err);
4474         goto out;
4475 }
4476
4477 /*
4478  * returns
4479  *      1: locks held by lockowner
4480  *      0: no locks held by lockowner
4481  */
4482 static int
4483 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4484 {
4485         struct file_lock **flpp;
4486         struct inode *inode = filp->fi_inode;
4487         int status = 0;
4488
4489         lock_flocks();
4490         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4491                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4492                         status = 1;
4493                         goto out;
4494                 }
4495         }
4496 out:
4497         unlock_flocks();
4498         return status;
4499 }
4500
4501 __be32
4502 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4503                         struct nfsd4_compound_state *cstate,
4504                         struct nfsd4_release_lockowner *rlockowner)
4505 {
4506         clientid_t *clid = &rlockowner->rl_clientid;
4507         struct nfs4_stateowner *sop;
4508         struct nfs4_lockowner *lo;
4509         struct nfs4_ol_stateid *stp;
4510         struct xdr_netobj *owner = &rlockowner->rl_owner;
4511         struct list_head matches;
4512         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4513         __be32 status;
4514         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4515
4516         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4517                 clid->cl_boot, clid->cl_id);
4518
4519         nfs4_lock_state();
4520
4521         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4522         if (status)
4523                 goto out;
4524
4525         status = nfserr_locks_held;
4526         INIT_LIST_HEAD(&matches);
4527
4528         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4529                 if (sop->so_is_open_owner)
4530                         continue;
4531                 if (!same_owner_str(sop, owner, clid))
4532                         continue;
4533                 list_for_each_entry(stp, &sop->so_stateids,
4534                                 st_perstateowner) {
4535                         lo = lockowner(sop);
4536                         if (check_for_locks(stp->st_file, lo))
4537                                 goto out;
4538                         list_add(&lo->lo_list, &matches);
4539                 }
4540         }
4541         /* Clients probably won't expect us to return with some (but not all)
4542          * of the lockowner state released; so don't release any until all
4543          * have been checked. */
4544         status = nfs_ok;
4545         while (!list_empty(&matches)) {
4546                 lo = list_entry(matches.next, struct nfs4_lockowner,
4547                                                                 lo_list);
4548                 /* unhash_stateowner deletes so_perclient only
4549                  * for openowners. */
4550                 list_del(&lo->lo_list);
4551                 release_lockowner(lo);
4552         }
4553 out:
4554         nfs4_unlock_state();
4555         return status;
4556 }
4557
4558 static inline struct nfs4_client_reclaim *
4559 alloc_reclaim(void)
4560 {
4561         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4562 }
4563
4564 bool
4565 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4566 {
4567         struct nfs4_client_reclaim *crp;
4568
4569         crp = nfsd4_find_reclaim_client(name, nn);
4570         return (crp && crp->cr_clp);
4571 }
4572
4573 /*
4574  * failure => all reset bets are off, nfserr_no_grace...
4575  */
4576 struct nfs4_client_reclaim *
4577 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4578 {
4579         unsigned int strhashval;
4580         struct nfs4_client_reclaim *crp;
4581
4582         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4583         crp = alloc_reclaim();
4584         if (crp) {
4585                 strhashval = clientstr_hashval(name);
4586                 INIT_LIST_HEAD(&crp->cr_strhash);
4587                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4588                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4589                 crp->cr_clp = NULL;
4590                 nn->reclaim_str_hashtbl_size++;
4591         }
4592         return crp;
4593 }
4594
4595 void
4596 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4597 {
4598         list_del(&crp->cr_strhash);
4599         kfree(crp);
4600         nn->reclaim_str_hashtbl_size--;
4601 }
4602
4603 void
4604 nfs4_release_reclaim(struct nfsd_net *nn)
4605 {
4606         struct nfs4_client_reclaim *crp = NULL;
4607         int i;
4608
4609         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4610                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4611                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4612                                         struct nfs4_client_reclaim, cr_strhash);
4613                         nfs4_remove_reclaim_record(crp, nn);
4614                 }
4615         }
4616         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4617 }
4618
4619 /*
4620  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4621 struct nfs4_client_reclaim *
4622 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4623 {
4624         unsigned int strhashval;
4625         struct nfs4_client_reclaim *crp = NULL;
4626
4627         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4628
4629         strhashval = clientstr_hashval(recdir);
4630         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4631                 if (same_name(crp->cr_recdir, recdir)) {
4632                         return crp;
4633                 }
4634         }
4635         return NULL;
4636 }
4637
4638 /*
4639 * Called from OPEN. Look for clientid in reclaim list.
4640 */
4641 __be32
4642 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4643 {
4644         struct nfs4_client *clp;
4645
4646         /* find clientid in conf_id_hashtbl */
4647         clp = find_confirmed_client(clid, sessions, nn);
4648         if (clp == NULL)
4649                 return nfserr_reclaim_bad;
4650
4651         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4652 }
4653
4654 #ifdef CONFIG_NFSD_FAULT_INJECTION
4655
4656 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4657 {
4658         if (mark_client_expired(clp))
4659                 return 0;
4660         expire_client(clp);
4661         return 1;
4662 }
4663
4664 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4665 {
4666         char buf[INET6_ADDRSTRLEN];
4667         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4668         printk(KERN_INFO "NFS Client: %s\n", buf);
4669         return 1;
4670 }
4671
4672 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4673                              const char *type)
4674 {
4675         char buf[INET6_ADDRSTRLEN];
4676         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4677         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4678 }
4679
4680 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4681 {
4682         struct nfs4_openowner *oop;
4683         struct nfs4_lockowner *lop, *lo_next;
4684         struct nfs4_ol_stateid *stp, *st_next;
4685         u64 count = 0;
4686
4687         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4688                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4689                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4690                                 if (func)
4691                                         func(lop);
4692                                 if (++count == max)
4693                                         return count;
4694                         }
4695                 }
4696         }
4697
4698         return count;
4699 }
4700
4701 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4702 {
4703         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4704 }
4705
4706 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4707 {
4708         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4709         nfsd_print_count(clp, count, "locked files");
4710         return count;
4711 }
4712
4713 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4714 {
4715         struct nfs4_openowner *oop, *next;
4716         u64 count = 0;
4717
4718         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4719                 if (func)
4720                         func(oop);
4721                 if (++count == max)
4722                         break;
4723         }
4724
4725         return count;
4726 }
4727
4728 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4729 {
4730         return nfsd_foreach_client_open(clp, max, release_openowner);
4731 }
4732
4733 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4734 {
4735         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4736         nfsd_print_count(clp, count, "open files");
4737         return count;
4738 }
4739
4740 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4741                                      struct list_head *victims)
4742 {
4743         struct nfs4_delegation *dp, *next;
4744         u64 count = 0;
4745
4746         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4747                 if (victims)
4748                         list_move(&dp->dl_recall_lru, victims);
4749                 if (++count == max)
4750                         break;
4751         }
4752         return count;
4753 }
4754
4755 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4756 {
4757         struct nfs4_delegation *dp, *next;
4758         LIST_HEAD(victims);
4759         u64 count;
4760
4761         spin_lock(&recall_lock);
4762         count = nfsd_find_all_delegations(clp, max, &victims);
4763         spin_unlock(&recall_lock);
4764
4765         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4766                 unhash_delegation(dp);
4767
4768         return count;
4769 }
4770
4771 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4772 {
4773         struct nfs4_delegation *dp, *next;
4774         LIST_HEAD(victims);
4775         u64 count;
4776
4777         spin_lock(&recall_lock);
4778         count = nfsd_find_all_delegations(clp, max, &victims);
4779         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4780                 nfsd_break_one_deleg(dp);
4781         spin_unlock(&recall_lock);
4782
4783         return count;
4784 }
4785
4786 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4787 {
4788         u64 count = 0;
4789
4790         spin_lock(&recall_lock);
4791         count = nfsd_find_all_delegations(clp, max, NULL);
4792         spin_unlock(&recall_lock);
4793
4794         nfsd_print_count(clp, count, "delegations");
4795         return count;
4796 }
4797
4798 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4799 {
4800         struct nfs4_client *clp, *next;
4801         u64 count = 0;
4802         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4803
4804         if (!nfsd_netns_ready(nn))
4805                 return 0;
4806
4807         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4808                 count += func(clp, max - count);
4809                 if ((max != 0) && (count >= max))
4810                         break;
4811         }
4812
4813         return count;
4814 }
4815
4816 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4817 {
4818         struct nfs4_client *clp;
4819         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4820
4821         if (!nfsd_netns_ready(nn))
4822                 return NULL;
4823
4824         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4825                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4826                         return clp;
4827         }
4828         return NULL;
4829 }
4830
4831 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4832
4833 /* initialization to perform at module load time: */
4834
4835 void
4836 nfs4_state_init(void)
4837 {
4838 }
4839
4840 /*
4841  * Since the lifetime of a delegation isn't limited to that of an open, a
4842  * client may quite reasonably hang on to a delegation as long as it has
4843  * the inode cached.  This becomes an obvious problem the first time a
4844  * client's inode cache approaches the size of the server's total memory.
4845  *
4846  * For now we avoid this problem by imposing a hard limit on the number
4847  * of delegations, which varies according to the server's memory size.
4848  */
4849 static void
4850 set_max_delegations(void)
4851 {
4852         /*
4853          * Allow at most 4 delegations per megabyte of RAM.  Quick
4854          * estimates suggest that in the worst case (where every delegation
4855          * is for a different inode), a delegation could take about 1.5K,
4856          * giving a worst case usage of about 6% of memory.
4857          */
4858         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4859 }
4860
4861 static int nfs4_state_create_net(struct net *net)
4862 {
4863         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4864         int i;
4865
4866         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4867                         CLIENT_HASH_SIZE, GFP_KERNEL);
4868         if (!nn->conf_id_hashtbl)
4869                 goto err;
4870         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4871                         CLIENT_HASH_SIZE, GFP_KERNEL);
4872         if (!nn->unconf_id_hashtbl)
4873                 goto err_unconf_id;
4874         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4875                         OWNER_HASH_SIZE, GFP_KERNEL);
4876         if (!nn->ownerstr_hashtbl)
4877                 goto err_ownerstr;
4878         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
4879                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
4880         if (!nn->lockowner_ino_hashtbl)
4881                 goto err_lockowner_ino;
4882         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
4883                         SESSION_HASH_SIZE, GFP_KERNEL);
4884         if (!nn->sessionid_hashtbl)
4885                 goto err_sessionid;
4886
4887         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4888                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
4889                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
4890         }
4891         for (i = 0; i < OWNER_HASH_SIZE; i++)
4892                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
4893         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4894                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
4895         for (i = 0; i < SESSION_HASH_SIZE; i++)
4896                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
4897         nn->conf_name_tree = RB_ROOT;
4898         nn->unconf_name_tree = RB_ROOT;
4899         INIT_LIST_HEAD(&nn->client_lru);
4900         INIT_LIST_HEAD(&nn->close_lru);
4901         INIT_LIST_HEAD(&nn->del_recall_lru);
4902         spin_lock_init(&nn->client_lock);
4903
4904         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
4905         get_net(net);
4906
4907         return 0;
4908
4909 err_sessionid:
4910         kfree(nn->lockowner_ino_hashtbl);
4911 err_lockowner_ino:
4912         kfree(nn->ownerstr_hashtbl);
4913 err_ownerstr:
4914         kfree(nn->unconf_id_hashtbl);
4915 err_unconf_id:
4916         kfree(nn->conf_id_hashtbl);
4917 err:
4918         return -ENOMEM;
4919 }
4920
4921 static void
4922 nfs4_state_destroy_net(struct net *net)
4923 {
4924         int i;
4925         struct nfs4_client *clp = NULL;
4926         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4927         struct rb_node *node, *tmp;
4928
4929         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4930                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
4931                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4932                         destroy_client(clp);
4933                 }
4934         }
4935
4936         node = rb_first(&nn->unconf_name_tree);
4937         while (node != NULL) {
4938                 tmp = node;
4939                 node = rb_next(tmp);
4940                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
4941                 rb_erase(tmp, &nn->unconf_name_tree);
4942                 destroy_client(clp);
4943         }
4944
4945         kfree(nn->sessionid_hashtbl);
4946         kfree(nn->lockowner_ino_hashtbl);
4947         kfree(nn->ownerstr_hashtbl);
4948         kfree(nn->unconf_id_hashtbl);
4949         kfree(nn->conf_id_hashtbl);
4950         put_net(net);
4951 }
4952
4953 int
4954 nfs4_state_start_net(struct net *net)
4955 {
4956         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4957         int ret;
4958
4959         ret = nfs4_state_create_net(net);
4960         if (ret)
4961                 return ret;
4962         nfsd4_client_tracking_init(net);
4963         nn->boot_time = get_seconds();
4964         locks_start_grace(net, &nn->nfsd4_manager);
4965         nn->grace_ended = false;
4966         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
4967                nn->nfsd4_grace, net);
4968         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
4969         return 0;
4970 }
4971
4972 /* initialization to perform when the nfsd service is started: */
4973
4974 int
4975 nfs4_state_start(void)
4976 {
4977         int ret;
4978
4979         ret = set_callback_cred();
4980         if (ret)
4981                 return -ENOMEM;
4982         laundry_wq = create_singlethread_workqueue("nfsd4");
4983         if (laundry_wq == NULL) {
4984                 ret = -ENOMEM;
4985                 goto out_recovery;
4986         }
4987         ret = nfsd4_create_callback_queue();
4988         if (ret)
4989                 goto out_free_laundry;
4990
4991         set_max_delegations();
4992
4993         return 0;
4994
4995 out_free_laundry:
4996         destroy_workqueue(laundry_wq);
4997 out_recovery:
4998         return ret;
4999 }
5000
5001 /* should be called with the state lock held */
5002 void
5003 nfs4_state_shutdown_net(struct net *net)
5004 {
5005         struct nfs4_delegation *dp = NULL;
5006         struct list_head *pos, *next, reaplist;
5007         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5008
5009         cancel_delayed_work_sync(&nn->laundromat_work);
5010         locks_end_grace(&nn->nfsd4_manager);
5011
5012         INIT_LIST_HEAD(&reaplist);
5013         spin_lock(&recall_lock);
5014         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5015                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5016                 list_move(&dp->dl_recall_lru, &reaplist);
5017         }
5018         spin_unlock(&recall_lock);
5019         list_for_each_safe(pos, next, &reaplist) {
5020                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5021                 unhash_delegation(dp);
5022         }
5023
5024         nfsd4_client_tracking_exit(net);
5025         nfs4_state_destroy_net(net);
5026 }
5027
5028 void
5029 nfs4_state_shutdown(void)
5030 {
5031         destroy_workqueue(laundry_wq);
5032         nfsd4_destroy_callback_queue();
5033 }
5034
5035 static void
5036 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5037 {
5038         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5039                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5040 }
5041
5042 static void
5043 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5044 {
5045         if (cstate->minorversion) {
5046                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5047                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5048         }
5049 }
5050
5051 void
5052 clear_current_stateid(struct nfsd4_compound_state *cstate)
5053 {
5054         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5055 }
5056
5057 /*
5058  * functions to set current state id
5059  */
5060 void
5061 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5062 {
5063         put_stateid(cstate, &odp->od_stateid);
5064 }
5065
5066 void
5067 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5068 {
5069         put_stateid(cstate, &open->op_stateid);
5070 }
5071
5072 void
5073 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5074 {
5075         put_stateid(cstate, &close->cl_stateid);
5076 }
5077
5078 void
5079 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5080 {
5081         put_stateid(cstate, &lock->lk_resp_stateid);
5082 }
5083
5084 /*
5085  * functions to consume current state id
5086  */
5087
5088 void
5089 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5090 {
5091         get_stateid(cstate, &odp->od_stateid);
5092 }
5093
5094 void
5095 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5096 {
5097         get_stateid(cstate, &drp->dr_stateid);
5098 }
5099
5100 void
5101 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5102 {
5103         get_stateid(cstate, &fsp->fr_stateid);
5104 }
5105
5106 void
5107 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5108 {
5109         get_stateid(cstate, &setattr->sa_stateid);
5110 }
5111
5112 void
5113 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5114 {
5115         get_stateid(cstate, &close->cl_stateid);
5116 }
5117
5118 void
5119 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5120 {
5121         get_stateid(cstate, &locku->lu_stateid);
5122 }
5123
5124 void
5125 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5126 {
5127         get_stateid(cstate, &read->rd_stateid);
5128 }
5129
5130 void
5131 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5132 {
5133         get_stateid(cstate, &write->wr_stateid);
5134 }