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 static bool is_session_dead(struct nfsd4_session *ses)
 101 {
 102         return ses->se_flags & NFS4_SESSION_DEAD;
 103 }
 104
 105 void nfsd4_put_session(struct nfsd4_session *ses)
 106 {
 107         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 108                 free_session(ses);
 109 }
 110
 111 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 112 {
 113         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 114                 return nfserr_jukebox;
 115         ses->se_flags |= NFS4_SESSION_DEAD;
 116         return nfs_ok;
 117 }
 118
 119 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 120 {
 121         if (is_session_dead(ses))
 122                 return nfserr_badsession;
 123         atomic_inc(&ses->se_ref);
 124         return nfs_ok;
 125 }
 126
 127 void
 128 nfs4_unlock_state(void)
 129 {
 130         mutex_unlock(&client_mutex);
 131 }
 132
 133 static bool is_client_expired(struct nfs4_client *clp)
 134 {
 135         return clp->cl_time == 0;
 136 }
 137
 138 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
 139 {
 140         if (atomic_read(&clp->cl_refcount))
 141                 return nfserr_jukebox;
 142         clp->cl_time = 0;
 143         return nfs_ok;
 144 }
 145
 146 static __be32 mark_client_expired(struct nfs4_client *clp)
 147 {
 148         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 149         __be32 ret;
 150
 151         spin_lock(&nn->client_lock);
 152         ret = mark_client_expired_locked(clp);
 153         spin_unlock(&nn->client_lock);
 154         return ret;
 155 }
 156
 157 static __be32 get_client_locked(struct nfs4_client *clp)
 158 {
 159         if (is_client_expired(clp))
 160                 return nfserr_expired;
 161         atomic_inc(&clp->cl_refcount);
 162         return nfs_ok;
 163 }
 164
 165 /* must be called under the client_lock */
 166 static inline void
 167 renew_client_locked(struct nfs4_client *clp)
 168 {
 169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 170
 171         if (is_client_expired(clp)) {
 172                 WARN_ON(1);
 173                 printk("%s: client (clientid %08x/%08x) already expired\n",
 174                         __func__,
 175                         clp->cl_clientid.cl_boot,
 176                         clp->cl_clientid.cl_id);
 177                 return;
 178         }
 179
 180         dprintk("renewing client (clientid %08x/%08x)\n",
 181                         clp->cl_clientid.cl_boot,
 182                         clp->cl_clientid.cl_id);
 183         list_move_tail(&clp->cl_lru, &nn->client_lru);
 184         clp->cl_time = get_seconds();
 185 }
 186
 187 static inline void
 188 renew_client(struct nfs4_client *clp)
 189 {
 190         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 191
 192         spin_lock(&nn->client_lock);
 193         renew_client_locked(clp);
 194         spin_unlock(&nn->client_lock);
 195 }
 196
 197 static void put_client_renew_locked(struct nfs4_client *clp)
 198 {
 199         if (!atomic_dec_and_test(&clp->cl_refcount))
 200                 return;
 201         if (!is_client_expired(clp))
 202                 renew_client_locked(clp);
 203 }
 204
 205 void put_client_renew(struct nfs4_client *clp)
 206 {
 207         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 208
 209         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 210                 return;
 211         if (!is_client_expired(clp))
 212                 renew_client_locked(clp);
 213         spin_unlock(&nn->client_lock);
 214 }
 215
 216
 217 static inline u32
 218 opaque_hashval(const void *ptr, int nbytes)
 219 {
 220         unsigned char *cptr = (unsigned char *) ptr;
 221
 222         u32 x = 0;
 223         while (nbytes--) {
 224                 x *= 37;
 225                 x += *cptr++;
 226         }
 227         return x;
 228 }
 229
 230 static void nfsd4_free_file(struct nfs4_file *f)
 231 {
 232         kmem_cache_free(file_slab, f);
 233 }
 234
 235 static inline void
 236 put_nfs4_file(struct nfs4_file *fi)
 237 {
 238         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 239                 hlist_del(&fi->fi_hash);
 240                 spin_unlock(&recall_lock);
 241                 iput(fi->fi_inode);
 242                 nfsd4_free_file(fi);
 243         }
 244 }
 245
 246 static inline void
 247 get_nfs4_file(struct nfs4_file *fi)
 248 {
 249         atomic_inc(&fi->fi_ref);
 250 }
 251
 252 static int num_delegations;
 253 unsigned long max_delegations;
 254
 255 /*
 256  * Open owner state (share locks)
 257  */
 258
 259 /* hash tables for lock and open owners */
 260 #define OWNER_HASH_BITS              8
 261 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 262 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 263
 264 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 265 {
 266         unsigned int ret;
 267
 268         ret = opaque_hashval(ownername->data, ownername->len);
 269         ret += clientid;
 270         return ret & OWNER_HASH_MASK;
 271 }
 272
 273 /* hash table for nfs4_file */
 274 #define FILE_HASH_BITS                   8
 275 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 276
 277 static unsigned int file_hashval(struct inode *ino)
 278 {
 279         /* XXX: why are we hashing on inode pointer, anyway? */
 280         return hash_ptr(ino, FILE_HASH_BITS);
 281 }
 282
 283 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 284
 285 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 286 {
 287         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 288         atomic_inc(&fp->fi_access[oflag]);
 289 }
 290
 291 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 292 {
 293         if (oflag == O_RDWR) {
 294                 __nfs4_file_get_access(fp, O_RDONLY);
 295                 __nfs4_file_get_access(fp, O_WRONLY);
 296         } else
 297                 __nfs4_file_get_access(fp, oflag);
 298 }
 299
 300 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 301 {
 302         if (fp->fi_fds[oflag]) {
 303                 fput(fp->fi_fds[oflag]);
 304                 fp->fi_fds[oflag] = NULL;
 305         }
 306 }
 307
 308 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 309 {
 310         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 311                 nfs4_file_put_fd(fp, oflag);
 312                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 313                         nfs4_file_put_fd(fp, O_RDWR);
 314         }
 315 }
 316
 317 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 318 {
 319         if (oflag == O_RDWR) {
 320                 __nfs4_file_put_access(fp, O_RDONLY);
 321                 __nfs4_file_put_access(fp, O_WRONLY);
 322         } else
 323                 __nfs4_file_put_access(fp, oflag);
 324 }
 325
 326 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct
 327 kmem_cache *slab)
 328 {
 329         struct idr *stateids = &cl->cl_stateids;
 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_cyclic(stateids, stid, 0, 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         return stid;
 357 out_free:
 358         kmem_cache_free(slab, stid);
 359         return NULL;
 360 }
 361
 362 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 363 {
 364         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 365 }
 366
 367 static struct nfs4_delegation *
 368 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh)
 369 {
 370         struct nfs4_delegation *dp;
 371
 372         dprintk("NFSD alloc_init_deleg\n");
 373         if (num_delegations > max_delegations)
 374                 return NULL;
 375         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 376         if (dp == NULL)
 377                 return dp;
 378         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 379         /*
 380          * delegation seqid's are never incremented.  The 4.1 special
 381          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 382          * 0 anyway just for consistency and use 1:
 383          */
 384         dp->dl_stid.sc_stateid.si_generation = 1;
 385         num_delegations++;
 386         INIT_LIST_HEAD(&dp->dl_perfile);
 387         INIT_LIST_HEAD(&dp->dl_perclnt);
 388         INIT_LIST_HEAD(&dp->dl_recall_lru);
 389         dp->dl_file = NULL;
 390         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 391         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 392         dp->dl_time = 0;
 393         atomic_set(&dp->dl_count, 1);
 394         nfsd4_init_callback(&dp->dl_recall);
 395         return dp;
 396 }
 397
 398 static void remove_stid(struct nfs4_stid *s)
 399 {
 400         struct idr *stateids = &s->sc_client->cl_stateids;
 401
 402         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 403 }
 404
 405 void
 406 nfs4_put_delegation(struct nfs4_delegation *dp)
 407 {
 408         if (atomic_dec_and_test(&dp->dl_count)) {
 409                 kmem_cache_free(deleg_slab, dp);
 410                 num_delegations--;
 411         }
 412 }
 413
 414 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 415 {
 416         if (atomic_dec_and_test(&fp->fi_delegees)) {
 417                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 418                 fp->fi_lease = NULL;
 419                 fput(fp->fi_deleg_file);
 420                 fp->fi_deleg_file = NULL;
 421         }
 422 }
 423
 424 static void unhash_stid(struct nfs4_stid *s)
 425 {
 426         s->sc_type = 0;
 427 }
 428
 429 /* Called under the state lock. */
 430 static void
 431 unhash_delegation(struct nfs4_delegation *dp)
 432 {
 433         list_del_init(&dp->dl_perclnt);
 434         spin_lock(&recall_lock);
 435         list_del_init(&dp->dl_perfile);
 436         list_del_init(&dp->dl_recall_lru);
 437         spin_unlock(&recall_lock);
 438         nfs4_put_deleg_lease(dp->dl_file);
 439         put_nfs4_file(dp->dl_file);
 440         dp->dl_file = NULL;
 441 }
 442
 443
 444
 445 static void destroy_revoked_delegation(struct nfs4_delegation *dp)
 446 {
 447         list_del_init(&dp->dl_recall_lru);
 448         remove_stid(&dp->dl_stid);
 449         nfs4_put_delegation(dp);
 450 }
 451
 452 static void destroy_delegation(struct nfs4_delegation *dp)
 453 {
 454         unhash_delegation(dp);
 455         remove_stid(&dp->dl_stid);
 456         nfs4_put_delegation(dp);
 457 }
 458
 459 static void revoke_delegation(struct nfs4_delegation *dp)
 460 {
 461         struct nfs4_client *clp = dp->dl_stid.sc_client;
 462
 463         if (clp->cl_minorversion == 0)
 464                 destroy_delegation(dp);
 465         else {
 466                 unhash_delegation(dp);
 467                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 468                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 469         }
 470 }
 471
 472 /*
 473  * SETCLIENTID state
 474  */
 475
 476 static unsigned int clientid_hashval(u32 id)
 477 {
 478         return id & CLIENT_HASH_MASK;
 479 }
 480
 481 static unsigned int clientstr_hashval(const char *name)
 482 {
 483         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 484 }
 485
 486 /*
 487  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 488  * st_{access,deny}_bmap field of the stateid, in order to track not
 489  * only what share bits are currently in force, but also what
 490  * combinations of share bits previous opens have used.  This allows us
 491  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 492  * return an error if the client attempt to downgrade to a combination
 493  * of share bits not explicable by closing some of its previous opens.
 494  *
 495  * XXX: This enforcement is actually incomplete, since we don't keep
 496  * track of access/deny bit combinations; so, e.g., we allow:
 497  *
 498  *      OPEN allow read, deny write
 499  *      OPEN allow both, deny none
 500  *      DOWNGRADE allow read, deny none
 501  *
 502  * which we should reject.
 503  */
 504 static unsigned int
 505 bmap_to_share_mode(unsigned long bmap) {
 506         int i;
 507         unsigned int access = 0;
 508
 509         for (i = 1; i < 4; i++) {
 510                 if (test_bit(i, &bmap))
 511                         access |= i;
 512         }
 513         return access;
 514 }
 515
 516 static bool
 517 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 518         unsigned int access, deny;
 519
 520         access = bmap_to_share_mode(stp->st_access_bmap);
 521         deny = bmap_to_share_mode(stp->st_deny_bmap);
 522         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 523                 return false;
 524         return true;
 525 }
 526
 527 /* set share access for a given stateid */
 528 static inline void
 529 set_access(u32 access, struct nfs4_ol_stateid *stp)
 530 {
 531         __set_bit(access, &stp->st_access_bmap);
 532 }
 533
 534 /* clear share access for a given stateid */
 535 static inline void
 536 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 537 {
 538         __clear_bit(access, &stp->st_access_bmap);
 539 }
 540
 541 /* test whether a given stateid has access */
 542 static inline bool
 543 test_access(u32 access, struct nfs4_ol_stateid *stp)
 544 {
 545         return test_bit(access, &stp->st_access_bmap);
 546 }
 547
 548 /* set share deny for a given stateid */
 549 static inline void
 550 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 551 {
 552         __set_bit(access, &stp->st_deny_bmap);
 553 }
 554
 555 /* clear share deny for a given stateid */
 556 static inline void
 557 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 558 {
 559         __clear_bit(access, &stp->st_deny_bmap);
 560 }
 561
 562 /* test whether a given stateid is denying specific access */
 563 static inline bool
 564 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 565 {
 566         return test_bit(access, &stp->st_deny_bmap);
 567 }
 568
 569 static int nfs4_access_to_omode(u32 access)
 570 {
 571         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 572         case NFS4_SHARE_ACCESS_READ:
 573                 return O_RDONLY;
 574         case NFS4_SHARE_ACCESS_WRITE:
 575                 return O_WRONLY;
 576         case NFS4_SHARE_ACCESS_BOTH:
 577                 return O_RDWR;
 578         }
 579         WARN_ON_ONCE(1);
 580         return O_RDONLY;
 581 }
 582
 583 /* release all access and file references for a given stateid */
 584 static void
 585 release_all_access(struct nfs4_ol_stateid *stp)
 586 {
 587         int i;
 588
 589         for (i = 1; i < 4; i++) {
 590                 if (test_access(i, stp))
 591                         nfs4_file_put_access(stp->st_file,
 592                                              nfs4_access_to_omode(i));
 593                 clear_access(i, stp);
 594         }
 595 }
 596
 597 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 598 {
 599         list_del(&stp->st_perfile);
 600         list_del(&stp->st_perstateowner);
 601 }
 602
 603 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 604 {
 605         release_all_access(stp);
 606         put_nfs4_file(stp->st_file);
 607         stp->st_file = NULL;
 608 }
 609
 610 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 611 {
 612         remove_stid(&stp->st_stid);
 613         kmem_cache_free(stateid_slab, stp);
 614 }
 615
 616 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 617 {
 618         struct file *file;
 619
 620         unhash_generic_stateid(stp);
 621         unhash_stid(&stp->st_stid);
 622         file = find_any_file(stp->st_file);
 623         if (file)
 624                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 625         close_generic_stateid(stp);
 626         free_generic_stateid(stp);
 627 }
 628
 629 static void unhash_lockowner(struct nfs4_lockowner *lo)
 630 {
 631         struct nfs4_ol_stateid *stp;
 632
 633         list_del(&lo->lo_owner.so_strhash);
 634         list_del(&lo->lo_perstateid);
 635         list_del(&lo->lo_owner_ino_hash);
 636         while (!list_empty(&lo->lo_owner.so_stateids)) {
 637                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 638                                 struct nfs4_ol_stateid, st_perstateowner);
 639                 release_lock_stateid(stp);
 640         }
 641 }
 642
 643 static void release_lockowner(struct nfs4_lockowner *lo)
 644 {
 645         unhash_lockowner(lo);
 646         nfs4_free_lockowner(lo);
 647 }
 648
 649 static void
 650 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 651 {
 652         struct nfs4_lockowner *lo;
 653
 654         while (!list_empty(&open_stp->st_lockowners)) {
 655                 lo = list_entry(open_stp->st_lockowners.next,
 656                                 struct nfs4_lockowner, lo_perstateid);
 657                 release_lockowner(lo);
 658         }
 659 }
 660
 661 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 662 {
 663         unhash_generic_stateid(stp);
 664         release_stateid_lockowners(stp);
 665         close_generic_stateid(stp);
 666 }
 667
 668 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 669 {
 670         unhash_open_stateid(stp);
 671         unhash_stid(&stp->st_stid);
 672         free_generic_stateid(stp);
 673 }
 674
 675 static void unhash_openowner(struct nfs4_openowner *oo)
 676 {
 677         struct nfs4_ol_stateid *stp;
 678
 679         list_del(&oo->oo_owner.so_strhash);
 680         list_del(&oo->oo_perclient);
 681         while (!list_empty(&oo->oo_owner.so_stateids)) {
 682                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 683                                 struct nfs4_ol_stateid, st_perstateowner);
 684                 release_open_stateid(stp);
 685         }
 686 }
 687
 688 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 689 {
 690         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 691
 692         if (s) {
 693                 unhash_stid(&s->st_stid);
 694                 free_generic_stateid(s);
 695                 oo->oo_last_closed_stid = NULL;
 696         }
 697 }
 698
 699 static void release_openowner(struct nfs4_openowner *oo)
 700 {
 701         unhash_openowner(oo);
 702         list_del(&oo->oo_close_lru);
 703         release_last_closed_stateid(oo);
 704         nfs4_free_openowner(oo);
 705 }
 706
 707 static inline int
 708 hash_sessionid(struct nfs4_sessionid *sessionid)
 709 {
 710         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 711
 712         return sid->sequence % SESSION_HASH_SIZE;
 713 }
 714
 715 #ifdef NFSD_DEBUG
 716 static inline void
 717 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 718 {
 719         u32 *ptr = (u32 *)(&sessionid->data[0]);
 720         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 721 }
 722 #else
 723 static inline void
 724 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 725 {
 726 }
 727 #endif
 728
 729 /*
 730  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
 731  * won't be used for replay.
 732  */
 733 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
 734 {
 735         struct nfs4_stateowner *so = cstate->replay_owner;
 736
 737         if (nfserr == nfserr_replay_me)
 738                 return;
 739
 740         if (!seqid_mutating_err(ntohl(nfserr))) {
 741                 cstate->replay_owner = NULL;
 742                 return;
 743         }
 744         if (!so)
 745                 return;
 746         if (so->so_is_open_owner)
 747                 release_last_closed_stateid(openowner(so));
 748         so->so_seqid++;
 749         return;
 750 }
 751
 752 static void
 753 gen_sessionid(struct nfsd4_session *ses)
 754 {
 755         struct nfs4_client *clp = ses->se_client;
 756         struct nfsd4_sessionid *sid;
 757
 758         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 759         sid->clientid = clp->cl_clientid;
 760         sid->sequence = current_sessionid++;
 761         sid->reserved = 0;
 762 }
 763
 764 /*
 765  * The protocol defines ca_maxresponssize_cached to include the size of
 766  * the rpc header, but all we need to cache is the data starting after
 767  * the end of the initial SEQUENCE operation--the rest we regenerate
 768  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 769  * value that is the number of bytes in our cache plus a few additional
 770  * bytes.  In order to stay on the safe side, and not promise more than
 771  * we can cache, those additional bytes must be the minimum possible: 24
 772  * bytes of rpc header (xid through accept state, with AUTH_NULL
 773  * verifier), 12 for the compound header (with zero-length tag), and 44
 774  * for the SEQUENCE op response:
 775  */
 776 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 777
 778 static void
 779 free_session_slots(struct nfsd4_session *ses)
 780 {
 781         int i;
 782
 783         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 784                 kfree(ses->se_slots[i]);
 785 }
 786
 787 /*
 788  * We don't actually need to cache the rpc and session headers, so we
 789  * can allocate a little less for each slot:
 790  */
 791 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
 792 {
 793         u32 size;
 794
 795         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
 796                 size = 0;
 797         else
 798                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 799         return size + sizeof(struct nfsd4_slot);
 800 }
 801
 802 /*
 803  * XXX: If we run out of reserved DRC memory we could (up to a point)
 804  * re-negotiate active sessions and reduce their slot usage to make
 805  * room for new connections. For now we just fail the create session.
 806  */
 807 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
 808 {
 809         u32 slotsize = slot_bytes(ca);
 810         u32 num = ca->maxreqs;
 811         int avail;
 812
 813         spin_lock(&nfsd_drc_lock);
 814         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
 815                     nfsd_drc_max_mem - nfsd_drc_mem_used);
 816         num = min_t(int, num, avail / slotsize);
 817         nfsd_drc_mem_used += num * slotsize;
 818         spin_unlock(&nfsd_drc_lock);
 819
 820         return num;
 821 }
 822
 823 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
 824 {
 825         int slotsize = slot_bytes(ca);
 826
 827         spin_lock(&nfsd_drc_lock);
 828         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
 829         spin_unlock(&nfsd_drc_lock);
 830 }
 831
 832 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *attrs)
 833 {
 834         int numslots = attrs->maxreqs;
 835         int slotsize = slot_bytes(attrs);
 836         struct nfsd4_session *new;
 837         int mem, i;
 838
 839         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 840                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 841         mem = numslots * sizeof(struct nfsd4_slot *);
 842
 843         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 844         if (!new)
 845                 return NULL;
 846         /* allocate each struct nfsd4_slot and data cache in one piece */
 847         for (i = 0; i < numslots; i++) {
 848                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
 849                 if (!new->se_slots[i])
 850                         goto out_free;
 851         }
 852         return new;
 853 out_free:
 854         while (i--)
 855                 kfree(new->se_slots[i]);
 856         kfree(new);
 857         return NULL;
 858 }
 859
 860 static void free_conn(struct nfsd4_conn *c)
 861 {
 862         svc_xprt_put(c->cn_xprt);
 863         kfree(c);
 864 }
 865
 866 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 867 {
 868         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 869         struct nfs4_client *clp = c->cn_session->se_client;
 870
 871         spin_lock(&clp->cl_lock);
 872         if (!list_empty(&c->cn_persession)) {
 873                 list_del(&c->cn_persession);
 874                 free_conn(c);
 875         }
 876         nfsd4_probe_callback(clp);
 877         spin_unlock(&clp->cl_lock);
 878 }
 879
 880 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 881 {
 882         struct nfsd4_conn *conn;
 883
 884         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 885         if (!conn)
 886                 return NULL;
 887         svc_xprt_get(rqstp->rq_xprt);
 888         conn->cn_xprt = rqstp->rq_xprt;
 889         conn->cn_flags = flags;
 890         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 891         return conn;
 892 }
 893
 894 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 895 {
 896         conn->cn_session = ses;
 897         list_add(&conn->cn_persession, &ses->se_conns);
 898 }
 899
 900 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 901 {
 902         struct nfs4_client *clp = ses->se_client;
 903
 904         spin_lock(&clp->cl_lock);
 905         __nfsd4_hash_conn(conn, ses);
 906         spin_unlock(&clp->cl_lock);
 907 }
 908
 909 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 910 {
 911         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 912         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 913 }
 914
 915 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 916 {
 917         int ret;
 918
 919         nfsd4_hash_conn(conn, ses);
 920         ret = nfsd4_register_conn(conn);
 921         if (ret)
 922                 /* oops; xprt is already down: */
 923                 nfsd4_conn_lost(&conn->cn_xpt_user);
 924         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 925                 /* callback channel may be back up */
 926                 nfsd4_probe_callback(ses->se_client);
 927         }
 928 }
 929
 930 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 931 {
 932         u32 dir = NFS4_CDFC4_FORE;
 933
 934         if (cses->flags & SESSION4_BACK_CHAN)
 935                 dir |= NFS4_CDFC4_BACK;
 936         return alloc_conn(rqstp, dir);
 937 }
 938
 939 /* must be called under client_lock */
 940 static void nfsd4_del_conns(struct nfsd4_session *s)
 941 {
 942         struct nfs4_client *clp = s->se_client;
 943         struct nfsd4_conn *c;
 944
 945         spin_lock(&clp->cl_lock);
 946         while (!list_empty(&s->se_conns)) {
 947                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 948                 list_del_init(&c->cn_persession);
 949                 spin_unlock(&clp->cl_lock);
 950
 951                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 952                 free_conn(c);
 953
 954                 spin_lock(&clp->cl_lock);
 955         }
 956         spin_unlock(&clp->cl_lock);
 957 }
 958
 959 static void __free_session(struct nfsd4_session *ses)
 960 {
 961         free_session_slots(ses);
 962         kfree(ses);
 963 }
 964
 965 static void free_session(struct nfsd4_session *ses)
 966 {
 967         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
 968
 969         lockdep_assert_held(&nn->client_lock);
 970         nfsd4_del_conns(ses);
 971         nfsd4_put_drc_mem(&ses->se_fchannel);
 972         __free_session(ses);
 973 }
 974
 975 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 976 {
 977         int idx;
 978         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 979
 980         new->se_client = clp;
 981         gen_sessionid(new);
 982
 983         INIT_LIST_HEAD(&new->se_conns);
 984
 985         new->se_cb_seq_nr = 1;
 986         new->se_flags = cses->flags;
 987         new->se_cb_prog = cses->callback_prog;
 988         new->se_cb_sec = cses->cb_sec;
 989         atomic_set(&new->se_ref, 0);
 990         idx = hash_sessionid(&new->se_sessionid);
 991         spin_lock(&nn->client_lock);
 992         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
 993         spin_lock(&clp->cl_lock);
 994         list_add(&new->se_perclnt, &clp->cl_sessions);
 995         spin_unlock(&clp->cl_lock);
 996         spin_unlock(&nn->client_lock);
 997         memcpy(&new->se_fchannel, &cses->fore_channel,
 998                         sizeof(struct nfsd4_channel_attrs));
 999         if (cses->flags & SESSION4_BACK_CHAN) {
1000                 struct sockaddr *sa = svc_addr(rqstp);
1001                 /*
1002                  * This is a little silly; with sessions there's no real
1003                  * use for the callback address.  Use the peer address
1004                  * as a reasonable default for now, but consider fixing
1005                  * the rpc client not to require an address in the
1006                  * future:
1007                  */
1008                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1009                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1010         }
1011 }
1012
1013 /* caller must hold client_lock */
1014 static struct nfsd4_session *
1015 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1016 {
1017         struct nfsd4_session *elem;
1018         int idx;
1019         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1020
1021         dump_sessionid(__func__, sessionid);
1022         idx = hash_sessionid(sessionid);
1023         /* Search in the appropriate list */
1024         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1025                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1026                             NFS4_MAX_SESSIONID_LEN)) {
1027                         return elem;
1028                 }
1029         }
1030
1031         dprintk("%s: session not found\n", __func__);
1032         return NULL;
1033 }
1034
1035 /* caller must hold client_lock */
1036 static void
1037 unhash_session(struct nfsd4_session *ses)
1038 {
1039         list_del(&ses->se_hash);
1040         spin_lock(&ses->se_client->cl_lock);
1041         list_del(&ses->se_perclnt);
1042         spin_unlock(&ses->se_client->cl_lock);
1043 }
1044
1045 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1046 static int
1047 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1048 {
1049         if (clid->cl_boot == nn->boot_time)
1050                 return 0;
1051         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1052                 clid->cl_boot, clid->cl_id, nn->boot_time);
1053         return 1;
1054 }
1055
1056 /*
1057  * XXX Should we use a slab cache ?
1058  * This type of memory management is somewhat inefficient, but we use it
1059  * anyway since SETCLIENTID is not a common operation.
1060  */
1061 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1062 {
1063         struct nfs4_client *clp;
1064
1065         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1066         if (clp == NULL)
1067                 return NULL;
1068         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1069         if (clp->cl_name.data == NULL) {
1070                 kfree(clp);
1071                 return NULL;
1072         }
1073         clp->cl_name.len = name.len;
1074         return clp;
1075 }
1076
1077 static inline void
1078 free_client(struct nfs4_client *clp)
1079 {
1080         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1081
1082         lockdep_assert_held(&nn->client_lock);
1083         while (!list_empty(&clp->cl_sessions)) {
1084                 struct nfsd4_session *ses;
1085                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1086                                 se_perclnt);
1087                 list_del(&ses->se_perclnt);
1088                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1089                 free_session(ses);
1090         }
1091         free_svc_cred(&clp->cl_cred);
1092         kfree(clp->cl_name.data);
1093         idr_destroy(&clp->cl_stateids);
1094         kfree(clp);
1095 }
1096
1097 /* must be called under the client_lock */
1098 static inline void
1099 unhash_client_locked(struct nfs4_client *clp)
1100 {
1101         struct nfsd4_session *ses;
1102
1103         list_del(&clp->cl_lru);
1104         spin_lock(&clp->cl_lock);
1105         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1106                 list_del_init(&ses->se_hash);
1107         spin_unlock(&clp->cl_lock);
1108 }
1109
1110 static void
1111 destroy_client(struct nfs4_client *clp)
1112 {
1113         struct nfs4_openowner *oo;
1114         struct nfs4_delegation *dp;
1115         struct list_head reaplist;
1116         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1117
1118         INIT_LIST_HEAD(&reaplist);
1119         spin_lock(&recall_lock);
1120         while (!list_empty(&clp->cl_delegations)) {
1121                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1122                 list_del_init(&dp->dl_perclnt);
1123                 list_move(&dp->dl_recall_lru, &reaplist);
1124         }
1125         spin_unlock(&recall_lock);
1126         while (!list_empty(&reaplist)) {
1127                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1128                 destroy_delegation(dp);
1129         }
1130         while (!list_empty(&clp->cl_openowners)) {
1131                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1132                 release_openowner(oo);
1133         }
1134         nfsd4_shutdown_callback(clp);
1135         if (clp->cl_cb_conn.cb_xprt)
1136                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1137         list_del(&clp->cl_idhash);
1138         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1139                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1140         else
1141                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1142         spin_lock(&nn->client_lock);
1143         unhash_client_locked(clp);
1144         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1145         free_client(clp);
1146         spin_unlock(&nn->client_lock);
1147 }
1148
1149 static void expire_client(struct nfs4_client *clp)
1150 {
1151         nfsd4_client_record_remove(clp);
1152         destroy_client(clp);
1153 }
1154
1155 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1156 {
1157         memcpy(target->cl_verifier.data, source->data,
1158                         sizeof(target->cl_verifier.data));
1159 }
1160
1161 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1162 {
1163         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1164         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1165 }
1166
1167 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1168 {
1169         if (source->cr_principal) {
1170                 target->cr_principal =
1171                                 kstrdup(source->cr_principal, GFP_KERNEL);
1172                 if (target->cr_principal == NULL)
1173                         return -ENOMEM;
1174         } else
1175                 target->cr_principal = NULL;
1176         target->cr_flavor = source->cr_flavor;
1177         target->cr_uid = source->cr_uid;
1178         target->cr_gid = source->cr_gid;
1179         target->cr_group_info = source->cr_group_info;
1180         get_group_info(target->cr_group_info);
1181         target->cr_gss_mech = source->cr_gss_mech;
1182         if (source->cr_gss_mech)
1183                 gss_mech_get(source->cr_gss_mech);
1184         return 0;
1185 }
1186
1187 static long long
1188 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1189 {
1190         long long res;
1191
1192         res = o1->len - o2->len;
1193         if (res)
1194                 return res;
1195         return (long long)memcmp(o1->data, o2->data, o1->len);
1196 }
1197
1198 static int same_name(const char *n1, const char *n2)
1199 {
1200         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1201 }
1202
1203 static int
1204 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1205 {
1206         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1207 }
1208
1209 static int
1210 same_clid(clientid_t *cl1, clientid_t *cl2)
1211 {
1212         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1213 }
1214
1215 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1216 {
1217         int i;
1218
1219         if (g1->ngroups != g2->ngroups)
1220                 return false;
1221         for (i=0; i<g1->ngroups; i++)
1222                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1223                         return false;
1224         return true;
1225 }
1226
1227 /*
1228  * RFC 3530 language requires clid_inuse be returned when the
1229  * "principal" associated with a requests differs from that previously
1230  * used.  We use uid, gid's, and gss principal string as our best
1231  * approximation.  We also don't want to allow non-gss use of a client
1232  * established using gss: in theory cr_principal should catch that
1233  * change, but in practice cr_principal can be null even in the gss case
1234  * since gssd doesn't always pass down a principal string.
1235  */
1236 static bool is_gss_cred(struct svc_cred *cr)
1237 {
1238         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1239         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1240 }
1241
1242
1243 static bool
1244 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1245 {
1246         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1247                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1248                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1249                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1250                 return false;
1251         if (cr1->cr_principal == cr2->cr_principal)
1252                 return true;
1253         if (!cr1->cr_principal || !cr2->cr_principal)
1254                 return false;
1255         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1256 }
1257
1258 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1259 {
1260         struct svc_cred *cr = &rqstp->rq_cred;
1261         u32 service;
1262
1263         if (!cr->cr_gss_mech)
1264                 return false;
1265         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1266         return service == RPC_GSS_SVC_INTEGRITY ||
1267                service == RPC_GSS_SVC_PRIVACY;
1268 }
1269
1270 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1271 {
1272         struct svc_cred *cr = &rqstp->rq_cred;
1273
1274         if (!cl->cl_mach_cred)
1275                 return true;
1276         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1277                 return false;
1278         if (!svc_rqst_integrity_protected(rqstp))
1279                 return false;
1280         if (!cr->cr_principal)
1281                 return false;
1282         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1283 }
1284
1285 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1286 {
1287         static u32 current_clientid = 1;
1288
1289         clp->cl_clientid.cl_boot = nn->boot_time;
1290         clp->cl_clientid.cl_id = current_clientid++;
1291 }
1292
1293 static void gen_confirm(struct nfs4_client *clp)
1294 {
1295         __be32 verf[2];
1296         static u32 i;
1297
1298         verf[0] = (__be32)get_seconds();
1299         verf[1] = (__be32)i++;
1300         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1301 }
1302
1303 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1304 {
1305         struct nfs4_stid *ret;
1306
1307         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1308         if (!ret || !ret->sc_type)
1309                 return NULL;
1310         return ret;
1311 }
1312
1313 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1314 {
1315         struct nfs4_stid *s;
1316
1317         s = find_stateid(cl, t);
1318         if (!s)
1319                 return NULL;
1320         if (typemask & s->sc_type)
1321                 return s;
1322         return NULL;
1323 }
1324
1325 static struct nfs4_client *create_client(struct xdr_netobj name,
1326                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1327 {
1328         struct nfs4_client *clp;
1329         struct sockaddr *sa = svc_addr(rqstp);
1330         int ret;
1331         struct net *net = SVC_NET(rqstp);
1332         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1333
1334         clp = alloc_client(name);
1335         if (clp == NULL)
1336                 return NULL;
1337
1338         INIT_LIST_HEAD(&clp->cl_sessions);
1339         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1340         if (ret) {
1341                 spin_lock(&nn->client_lock);
1342                 free_client(clp);
1343                 spin_unlock(&nn->client_lock);
1344                 return NULL;
1345         }
1346         idr_init(&clp->cl_stateids);
1347         atomic_set(&clp->cl_refcount, 0);
1348         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1349         INIT_LIST_HEAD(&clp->cl_idhash);
1350         INIT_LIST_HEAD(&clp->cl_openowners);
1351         INIT_LIST_HEAD(&clp->cl_delegations);
1352         INIT_LIST_HEAD(&clp->cl_lru);
1353         INIT_LIST_HEAD(&clp->cl_callbacks);
1354         INIT_LIST_HEAD(&clp->cl_revoked);
1355         spin_lock_init(&clp->cl_lock);
1356         nfsd4_init_callback(&clp->cl_cb_null);
1357         clp->cl_time = get_seconds();
1358         clear_bit(0, &clp->cl_cb_slot_busy);
1359         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1360         copy_verf(clp, verf);
1361         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1362         gen_confirm(clp);
1363         clp->cl_cb_session = NULL;
1364         clp->net = net;
1365         return clp;
1366 }
1367
1368 static void
1369 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1370 {
1371         struct rb_node **new = &(root->rb_node), *parent = NULL;
1372         struct nfs4_client *clp;
1373
1374         while (*new) {
1375                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1376                 parent = *new;
1377
1378                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1379                         new = &((*new)->rb_left);
1380                 else
1381                         new = &((*new)->rb_right);
1382         }
1383
1384         rb_link_node(&new_clp->cl_namenode, parent, new);
1385         rb_insert_color(&new_clp->cl_namenode, root);
1386 }
1387
1388 static struct nfs4_client *
1389 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1390 {
1391         long long cmp;
1392         struct rb_node *node = root->rb_node;
1393         struct nfs4_client *clp;
1394
1395         while (node) {
1396                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1397                 cmp = compare_blob(&clp->cl_name, name);
1398                 if (cmp > 0)
1399                         node = node->rb_left;
1400                 else if (cmp < 0)
1401                         node = node->rb_right;
1402                 else
1403                         return clp;
1404         }
1405         return NULL;
1406 }
1407
1408 static void
1409 add_to_unconfirmed(struct nfs4_client *clp)
1410 {
1411         unsigned int idhashval;
1412         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1413
1414         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1415         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1416         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1417         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1418         renew_client(clp);
1419 }
1420
1421 static void
1422 move_to_confirmed(struct nfs4_client *clp)
1423 {
1424         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1425         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1426
1427         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1428         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1429         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1430         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1431         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1432         renew_client(clp);
1433 }
1434
1435 static struct nfs4_client *
1436 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1437 {
1438         struct nfs4_client *clp;
1439         unsigned int idhashval = clientid_hashval(clid->cl_id);
1440
1441         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1442                 if (same_clid(&clp->cl_clientid, clid)) {
1443                         if ((bool)clp->cl_minorversion != sessions)
1444                                 return NULL;
1445                         renew_client(clp);
1446                         return clp;
1447                 }
1448         }
1449         return NULL;
1450 }
1451
1452 static struct nfs4_client *
1453 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1454 {
1455         struct list_head *tbl = nn->conf_id_hashtbl;
1456
1457         return find_client_in_id_table(tbl, clid, sessions);
1458 }
1459
1460 static struct nfs4_client *
1461 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1462 {
1463         struct list_head *tbl = nn->unconf_id_hashtbl;
1464
1465         return find_client_in_id_table(tbl, clid, sessions);
1466 }
1467
1468 static bool clp_used_exchangeid(struct nfs4_client *clp)
1469 {
1470         return clp->cl_exchange_flags != 0;
1471 }
1472
1473 static struct nfs4_client *
1474 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1475 {
1476         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1477 }
1478
1479 static struct nfs4_client *
1480 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1481 {
1482         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1483 }
1484
1485 static void
1486 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1487 {
1488         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1489         struct sockaddr *sa = svc_addr(rqstp);
1490         u32 scopeid = rpc_get_scope_id(sa);
1491         unsigned short expected_family;
1492
1493         /* Currently, we only support tcp and tcp6 for the callback channel */
1494         if (se->se_callback_netid_len == 3 &&
1495             !memcmp(se->se_callback_netid_val, "tcp", 3))
1496                 expected_family = AF_INET;
1497         else if (se->se_callback_netid_len == 4 &&
1498                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1499                 expected_family = AF_INET6;
1500         else
1501                 goto out_err;
1502
1503         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1504                                             se->se_callback_addr_len,
1505                                             (struct sockaddr *)&conn->cb_addr,
1506                                             sizeof(conn->cb_addr));
1507
1508         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1509                 goto out_err;
1510
1511         if (conn->cb_addr.ss_family == AF_INET6)
1512                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1513
1514         conn->cb_prog = se->se_callback_prog;
1515         conn->cb_ident = se->se_callback_ident;
1516         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1517         return;
1518 out_err:
1519         conn->cb_addr.ss_family = AF_UNSPEC;
1520         conn->cb_addrlen = 0;
1521         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1522                 "will not receive delegations\n",
1523                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1524
1525         return;
1526 }
1527
1528 /*
1529  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1530  */
1531 void
1532 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1533 {
1534         struct nfsd4_slot *slot = resp->cstate.slot;
1535         unsigned int base;
1536
1537         dprintk("--> %s slot %p\n", __func__, slot);
1538
1539         slot->sl_opcnt = resp->opcnt;
1540         slot->sl_status = resp->cstate.status;
1541
1542         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1543         if (nfsd4_not_cached(resp)) {
1544                 slot->sl_datalen = 0;
1545                 return;
1546         }
1547         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1548         base = (char *)resp->cstate.datap -
1549                                         (char *)resp->xbuf->head[0].iov_base;
1550         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1551                                     slot->sl_datalen))
1552                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1553         return;
1554 }
1555
1556 /*
1557  * Encode the replay sequence operation from the slot values.
1558  * If cachethis is FALSE encode the uncached rep error on the next
1559  * operation which sets resp->p and increments resp->opcnt for
1560  * nfs4svc_encode_compoundres.
1561  *
1562  */
1563 static __be32
1564 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1565                           struct nfsd4_compoundres *resp)
1566 {
1567         struct nfsd4_op *op;
1568         struct nfsd4_slot *slot = resp->cstate.slot;
1569
1570         /* Encode the replayed sequence operation */
1571         op = &args->ops[resp->opcnt - 1];
1572         nfsd4_encode_operation(resp, op);
1573
1574         /* Return nfserr_retry_uncached_rep in next operation. */
1575         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1576                 op = &args->ops[resp->opcnt++];
1577                 op->status = nfserr_retry_uncached_rep;
1578                 nfsd4_encode_operation(resp, op);
1579         }
1580         return op->status;
1581 }
1582
1583 /*
1584  * The sequence operation is not cached because we can use the slot and
1585  * session values.
1586  */
1587 __be32
1588 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1589                          struct nfsd4_sequence *seq)
1590 {
1591         struct nfsd4_slot *slot = resp->cstate.slot;
1592         __be32 status;
1593
1594         dprintk("--> %s slot %p\n", __func__, slot);
1595
1596         /* Either returns 0 or nfserr_retry_uncached */
1597         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1598         if (status == nfserr_retry_uncached_rep)
1599                 return status;
1600
1601         /* The sequence operation has been encoded, cstate->datap set. */
1602         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1603
1604         resp->opcnt = slot->sl_opcnt;
1605         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1606         status = slot->sl_status;
1607
1608         return status;
1609 }
1610
1611 /*
1612  * Set the exchange_id flags returned by the server.
1613  */
1614 static void
1615 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1616 {
1617         /* pNFS is not supported */
1618         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1619
1620         /* Referrals are supported, Migration is not. */
1621         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1622
1623         /* set the wire flags to return to client. */
1624         clid->flags = new->cl_exchange_flags;
1625 }
1626
1627 static bool client_has_state(struct nfs4_client *clp)
1628 {
1629         /*
1630          * Note clp->cl_openowners check isn't quite right: there's no
1631          * need to count owners without stateid's.
1632          *
1633          * Also note we should probably be using this in 4.0 case too.
1634          */
1635         return !list_empty(&clp->cl_openowners)
1636                 || !list_empty(&clp->cl_delegations)
1637                 || !list_empty(&clp->cl_sessions);
1638 }
1639
1640 __be32
1641 nfsd4_exchange_id(struct svc_rqst *rqstp,
1642                   struct nfsd4_compound_state *cstate,
1643                   struct nfsd4_exchange_id *exid)
1644 {
1645         struct nfs4_client *unconf, *conf, *new;
1646         __be32 status;
1647         char                    addr_str[INET6_ADDRSTRLEN];
1648         nfs4_verifier           verf = exid->verifier;
1649         struct sockaddr         *sa = svc_addr(rqstp);
1650         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1651         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1652
1653         rpc_ntop(sa, addr_str, sizeof(addr_str));
1654         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1655                 "ip_addr=%s flags %x, spa_how %d\n",
1656                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1657                 addr_str, exid->flags, exid->spa_how);
1658
1659         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1660                 return nfserr_inval;
1661
1662         switch (exid->spa_how) {
1663         case SP4_MACH_CRED:
1664                 if (!svc_rqst_integrity_protected(rqstp))
1665                         return nfserr_inval;
1666         case SP4_NONE:
1667                 break;
1668         default:                                /* checked by xdr code */
1669                 WARN_ON_ONCE(1);
1670         case SP4_SSV:
1671                 return nfserr_encr_alg_unsupp;
1672         }
1673
1674         /* Cases below refer to rfc 5661 section 18.35.4: */
1675         nfs4_lock_state();
1676         conf = find_confirmed_client_by_name(&exid->clname, nn);
1677         if (conf) {
1678                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1679                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1680
1681                 if (update) {
1682                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1683                                 status = nfserr_inval;
1684                                 goto out;
1685                         }
1686                         if (!mach_creds_match(conf, rqstp)) {
1687                                 status = nfserr_wrong_cred;
1688                                 goto out;
1689                         }
1690                         if (!creds_match) { /* case 9 */
1691                                 status = nfserr_perm;
1692                                 goto out;
1693                         }
1694                         if (!verfs_match) { /* case 8 */
1695                                 status = nfserr_not_same;
1696                                 goto out;
1697                         }
1698                         /* case 6 */
1699                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1700                         new = conf;
1701                         goto out_copy;
1702                 }
1703                 if (!creds_match) { /* case 3 */
1704                         if (client_has_state(conf)) {
1705                                 status = nfserr_clid_inuse;
1706                                 goto out;
1707                         }
1708                         expire_client(conf);
1709                         goto out_new;
1710                 }
1711                 if (verfs_match) { /* case 2 */
1712                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1713                         new = conf;
1714                         goto out_copy;
1715                 }
1716                 /* case 5, client reboot */
1717                 goto out_new;
1718         }
1719
1720         if (update) { /* case 7 */
1721                 status = nfserr_noent;
1722                 goto out;
1723         }
1724
1725         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1726         if (unconf) /* case 4, possible retry or client restart */
1727                 expire_client(unconf);
1728
1729         /* case 1 (normal case) */
1730 out_new:
1731         new = create_client(exid->clname, rqstp, &verf);
1732         if (new == NULL) {
1733                 status = nfserr_jukebox;
1734                 goto out;
1735         }
1736         new->cl_minorversion = cstate->minorversion;
1737         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
1738
1739         gen_clid(new, nn);
1740         add_to_unconfirmed(new);
1741 out_copy:
1742         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1743         exid->clientid.cl_id = new->cl_clientid.cl_id;
1744
1745         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1746         nfsd4_set_ex_flags(new, exid);
1747
1748         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1749                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1750         status = nfs_ok;
1751
1752 out:
1753         nfs4_unlock_state();
1754         return status;
1755 }
1756
1757 static __be32
1758 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1759 {
1760         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1761                 slot_seqid);
1762
1763         /* The slot is in use, and no response has been sent. */
1764         if (slot_inuse) {
1765                 if (seqid == slot_seqid)
1766                         return nfserr_jukebox;
1767                 else
1768                         return nfserr_seq_misordered;
1769         }
1770         /* Note unsigned 32-bit arithmetic handles wraparound: */
1771         if (likely(seqid == slot_seqid + 1))
1772                 return nfs_ok;
1773         if (seqid == slot_seqid)
1774                 return nfserr_replay_cache;
1775         return nfserr_seq_misordered;
1776 }
1777
1778 /*
1779  * Cache the create session result into the create session single DRC
1780  * slot cache by saving the xdr structure. sl_seqid has been set.
1781  * Do this for solo or embedded create session operations.
1782  */
1783 static void
1784 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1785                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1786 {
1787         slot->sl_status = nfserr;
1788         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1789 }
1790
1791 static __be32
1792 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1793                             struct nfsd4_clid_slot *slot)
1794 {
1795         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1796         return slot->sl_status;
1797 }
1798
1799 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1800                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1801                         1 +     /* MIN tag is length with zero, only length */ \
1802                         3 +     /* version, opcount, opcode */ \
1803                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1804                                 /* seqid, slotID, slotID, cache */ \
1805                         4 ) * sizeof(__be32))
1806
1807 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1808                         2 +     /* verifier: AUTH_NULL, length 0 */\
1809                         1 +     /* status */ \
1810                         1 +     /* MIN tag is length with zero, only length */ \
1811                         3 +     /* opcount, opcode, opstatus*/ \
1812                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1813                                 /* seqid, slotID, slotID, slotID, status */ \
1814                         5 ) * sizeof(__be32))
1815
1816 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1817 {
1818         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
1819
1820         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
1821                 return nfserr_toosmall;
1822         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
1823                 return nfserr_toosmall;
1824         ca->headerpadsz = 0;
1825         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
1826         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
1827         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
1828         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
1829                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
1830         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
1831         /*
1832          * Note decreasing slot size below client's request may make it
1833          * difficult for client to function correctly, whereas
1834          * decreasing the number of slots will (just?) affect
1835          * performance.  When short on memory we therefore prefer to
1836          * decrease number of slots instead of their size.  Clients that
1837          * request larger slots than they need will get poor results:
1838          */
1839         ca->maxreqs = nfsd4_get_drc_mem(ca);
1840         if (!ca->maxreqs)
1841                 return nfserr_jukebox;
1842
1843         return nfs_ok;
1844 }
1845
1846 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
1847 {
1848         ca->headerpadsz = 0;
1849
1850         /*
1851          * These RPC_MAX_HEADER macros are overkill, especially since we
1852          * don't even do gss on the backchannel yet.  But this is still
1853          * less than 1k.  Tighten up this estimate in the unlikely event
1854          * it turns out to be a problem for some client:
1855          */
1856         if (ca->maxreq_sz < NFS4_enc_cb_recall_sz + RPC_MAX_HEADER_WITH_AUTH)
1857                 return nfserr_toosmall;
1858         if (ca->maxresp_sz < NFS4_dec_cb_recall_sz + RPC_MAX_REPHEADER_WITH_AUTH)
1859                 return nfserr_toosmall;
1860         ca->maxresp_cached = 0;
1861         if (ca->maxops < 2)
1862                 return nfserr_toosmall;
1863
1864         return nfs_ok;
1865 }
1866
1867 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
1868 {
1869         switch (cbs->flavor) {
1870         case RPC_AUTH_NULL:
1871         case RPC_AUTH_UNIX:
1872                 return nfs_ok;
1873         default:
1874                 /*
1875                  * GSS case: the spec doesn't allow us to return this
1876                  * error.  But it also doesn't allow us not to support
1877                  * GSS.
1878                  * I'd rather this fail hard than return some error the
1879                  * client might think it can already handle:
1880                  */
1881                 return nfserr_encr_alg_unsupp;
1882         }
1883 }
1884
1885 __be32
1886 nfsd4_create_session(struct svc_rqst *rqstp,
1887                      struct nfsd4_compound_state *cstate,
1888                      struct nfsd4_create_session *cr_ses)
1889 {
1890         struct sockaddr *sa = svc_addr(rqstp);
1891         struct nfs4_client *conf, *unconf;
1892         struct nfsd4_session *new;
1893         struct nfsd4_conn *conn;
1894         struct nfsd4_clid_slot *cs_slot = NULL;
1895         __be32 status = 0;
1896         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1897
1898         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1899                 return nfserr_inval;
1900         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
1901         if (status)
1902                 return status;
1903         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
1904         if (status)
1905                 return status;
1906         status = check_backchannel_attrs(&cr_ses->back_channel);
1907         if (status)
1908                 return status;
1909         status = nfserr_jukebox;
1910         new = alloc_session(&cr_ses->fore_channel);
1911         if (!new)
1912                 goto out_release_drc_mem;
1913         conn = alloc_conn_from_crses(rqstp, cr_ses);
1914         if (!conn)
1915                 goto out_free_session;
1916
1917         nfs4_lock_state();
1918         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1919         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1920         WARN_ON_ONCE(conf && unconf);
1921
1922         if (conf) {
1923                 status = nfserr_wrong_cred;
1924                 if (!mach_creds_match(conf, rqstp))
1925                         goto out_free_conn;
1926                 cs_slot = &conf->cl_cs_slot;
1927                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1928                 if (status == nfserr_replay_cache) {
1929                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1930                         goto out_free_conn;
1931                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1932                         status = nfserr_seq_misordered;
1933                         goto out_free_conn;
1934                 }
1935         } else if (unconf) {
1936                 struct nfs4_client *old;
1937                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1938                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1939                         status = nfserr_clid_inuse;
1940                         goto out_free_conn;
1941                 }
1942                 status = nfserr_wrong_cred;
1943                 if (!mach_creds_match(unconf, rqstp))
1944                         goto out_free_conn;
1945                 cs_slot = &unconf->cl_cs_slot;
1946                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1947                 if (status) {
1948                         /* an unconfirmed replay returns misordered */
1949                         status = nfserr_seq_misordered;
1950                         goto out_free_conn;
1951                 }
1952                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1953                 if (old) {
1954                         status = mark_client_expired(old);
1955                         if (status)
1956                                 goto out_free_conn;
1957                         expire_client(old);
1958                 }
1959                 move_to_confirmed(unconf);
1960                 conf = unconf;
1961         } else {
1962                 status = nfserr_stale_clientid;
1963                 goto out_free_conn;
1964         }
1965         status = nfs_ok;
1966         /*
1967          * We do not support RDMA or persistent sessions
1968          */
1969         cr_ses->flags &= ~SESSION4_PERSIST;
1970         cr_ses->flags &= ~SESSION4_RDMA;
1971
1972         init_session(rqstp, new, conf, cr_ses);
1973         nfsd4_init_conn(rqstp, conn, new);
1974
1975         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1976                NFS4_MAX_SESSIONID_LEN);
1977         cs_slot->sl_seqid++;
1978         cr_ses->seqid = cs_slot->sl_seqid;
1979
1980         /* cache solo and embedded create sessions under the state lock */
1981         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1982         nfs4_unlock_state();
1983         return status;
1984 out_free_conn:
1985         nfs4_unlock_state();
1986         free_conn(conn);
1987 out_free_session:
1988         __free_session(new);
1989 out_release_drc_mem:
1990         nfsd4_put_drc_mem(&cr_ses->fore_channel);
1991         return status;
1992 }
1993
1994 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1995 {
1996         switch (*dir) {
1997         case NFS4_CDFC4_FORE:
1998         case NFS4_CDFC4_BACK:
1999                 return nfs_ok;
2000         case NFS4_CDFC4_FORE_OR_BOTH:
2001         case NFS4_CDFC4_BACK_OR_BOTH:
2002                 *dir = NFS4_CDFC4_BOTH;
2003                 return nfs_ok;
2004         };
2005         return nfserr_inval;
2006 }
2007
2008 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2009 {
2010         struct nfsd4_session *session = cstate->session;
2011         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2012         __be32 status;
2013
2014         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2015         if (status)
2016                 return status;
2017         spin_lock(&nn->client_lock);
2018         session->se_cb_prog = bc->bc_cb_program;
2019         session->se_cb_sec = bc->bc_cb_sec;
2020         spin_unlock(&nn->client_lock);
2021
2022         nfsd4_probe_callback(session->se_client);
2023
2024         return nfs_ok;
2025 }
2026
2027 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2028                      struct nfsd4_compound_state *cstate,
2029                      struct nfsd4_bind_conn_to_session *bcts)
2030 {
2031         __be32 status;
2032         struct nfsd4_conn *conn;
2033         struct nfsd4_session *session;
2034         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2035
2036         if (!nfsd4_last_compound_op(rqstp))
2037                 return nfserr_not_only_op;
2038         nfs4_lock_state();
2039         spin_lock(&nn->client_lock);
2040         session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
2041         spin_unlock(&nn->client_lock);
2042         status = nfserr_badsession;
2043         if (!session)
2044                 goto out;
2045         status = nfserr_wrong_cred;
2046         if (!mach_creds_match(session->se_client, rqstp))
2047                 goto out;
2048         status = nfsd4_map_bcts_dir(&bcts->dir);
2049         if (status)
2050                 goto out;
2051         conn = alloc_conn(rqstp, bcts->dir);
2052         status = nfserr_jukebox;
2053         if (!conn)
2054                 goto out;
2055         nfsd4_init_conn(rqstp, conn, session);
2056         status = nfs_ok;
2057 out:
2058         nfs4_unlock_state();
2059         return status;
2060 }
2061
2062 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2063 {
2064         if (!session)
2065                 return 0;
2066         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2067 }
2068
2069 __be32
2070 nfsd4_destroy_session(struct svc_rqst *r,
2071                       struct nfsd4_compound_state *cstate,
2072                       struct nfsd4_destroy_session *sessionid)
2073 {
2074         struct nfsd4_session *ses;
2075         __be32 status;
2076         int ref_held_by_me = 0;
2077         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
2078
2079         nfs4_lock_state();
2080         status = nfserr_not_only_op;
2081         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2082                 if (!nfsd4_last_compound_op(r))
2083                         goto out;
2084                 ref_held_by_me++;
2085         }
2086         dump_sessionid(__func__, &sessionid->sessionid);
2087         spin_lock(&nn->client_lock);
2088         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
2089         status = nfserr_badsession;
2090         if (!ses)
2091                 goto out_client_lock;
2092         status = nfserr_wrong_cred;
2093         if (!mach_creds_match(ses->se_client, r))
2094                 goto out_client_lock;
2095         nfsd4_get_session_locked(ses);
2096         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2097         if (status)
2098                 goto out_put_session;
2099         unhash_session(ses);
2100         spin_unlock(&nn->client_lock);
2101
2102         nfsd4_probe_callback_sync(ses->se_client);
2103
2104         spin_lock(&nn->client_lock);
2105         status = nfs_ok;
2106 out_put_session:
2107         nfsd4_put_session(ses);
2108 out_client_lock:
2109         spin_unlock(&nn->client_lock);
2110 out:
2111         nfs4_unlock_state();
2112         return status;
2113 }
2114
2115 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2116 {
2117         struct nfsd4_conn *c;
2118
2119         list_for_each_entry(c, &s->se_conns, cn_persession) {
2120                 if (c->cn_xprt == xpt) {
2121                         return c;
2122                 }
2123         }
2124         return NULL;
2125 }
2126
2127 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2128 {
2129         struct nfs4_client *clp = ses->se_client;
2130         struct nfsd4_conn *c;
2131         __be32 status = nfs_ok;
2132         int ret;
2133
2134         spin_lock(&clp->cl_lock);
2135         c = __nfsd4_find_conn(new->cn_xprt, ses);
2136         if (c)
2137                 goto out_free;
2138         status = nfserr_conn_not_bound_to_session;
2139         if (clp->cl_mach_cred)
2140                 goto out_free;
2141         __nfsd4_hash_conn(new, ses);
2142         spin_unlock(&clp->cl_lock);
2143         ret = nfsd4_register_conn(new);
2144         if (ret)
2145                 /* oops; xprt is already down: */
2146                 nfsd4_conn_lost(&new->cn_xpt_user);
2147         return nfs_ok;
2148 out_free:
2149         spin_unlock(&clp->cl_lock);
2150         free_conn(new);
2151         return status;
2152 }
2153
2154 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2155 {
2156         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2157
2158         return args->opcnt > session->se_fchannel.maxops;
2159 }
2160
2161 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2162                                   struct nfsd4_session *session)
2163 {
2164         struct xdr_buf *xb = &rqstp->rq_arg;
2165
2166         return xb->len > session->se_fchannel.maxreq_sz;
2167 }
2168
2169 __be32
2170 nfsd4_sequence(struct svc_rqst *rqstp,
2171                struct nfsd4_compound_state *cstate,
2172                struct nfsd4_sequence *seq)
2173 {
2174         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2175         struct nfsd4_session *session;
2176         struct nfs4_client *clp;
2177         struct nfsd4_slot *slot;
2178         struct nfsd4_conn *conn;
2179         __be32 status;
2180         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2181
2182         if (resp->opcnt != 1)
2183                 return nfserr_sequence_pos;
2184
2185         /*
2186          * Will be either used or freed by nfsd4_sequence_check_conn
2187          * below.
2188          */
2189         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2190         if (!conn)
2191                 return nfserr_jukebox;
2192
2193         spin_lock(&nn->client_lock);
2194         status = nfserr_badsession;
2195         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2196         if (!session)
2197                 goto out_no_session;
2198         clp = session->se_client;
2199         status = get_client_locked(clp);
2200         if (status)
2201                 goto out_no_session;
2202         status = nfsd4_get_session_locked(session);
2203         if (status)
2204                 goto out_put_client;
2205
2206         status = nfserr_too_many_ops;
2207         if (nfsd4_session_too_many_ops(rqstp, session))
2208                 goto out_put_session;
2209
2210         status = nfserr_req_too_big;
2211         if (nfsd4_request_too_big(rqstp, session))
2212                 goto out_put_session;
2213
2214         status = nfserr_badslot;
2215         if (seq->slotid >= session->se_fchannel.maxreqs)
2216                 goto out_put_session;
2217
2218         slot = session->se_slots[seq->slotid];
2219         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2220
2221         /* We do not negotiate the number of slots yet, so set the
2222          * maxslots to the session maxreqs which is used to encode
2223          * sr_highest_slotid and the sr_target_slot id to maxslots */
2224         seq->maxslots = session->se_fchannel.maxreqs;
2225
2226         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2227                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2228         if (status == nfserr_replay_cache) {
2229                 status = nfserr_seq_misordered;
2230                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2231                         goto out_put_session;
2232                 cstate->slot = slot;
2233                 cstate->session = session;
2234                 /* Return the cached reply status and set cstate->status
2235                  * for nfsd4_proc_compound processing */
2236                 status = nfsd4_replay_cache_entry(resp, seq);
2237                 cstate->status = nfserr_replay_cache;
2238                 goto out;
2239         }
2240         if (status)
2241                 goto out_put_session;
2242
2243         status = nfsd4_sequence_check_conn(conn, session);
2244         conn = NULL;
2245         if (status)
2246                 goto out_put_session;
2247
2248         /* Success! bump slot seqid */
2249         slot->sl_seqid = seq->seqid;
2250         slot->sl_flags |= NFSD4_SLOT_INUSE;
2251         if (seq->cachethis)
2252                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2253         else
2254                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2255
2256         cstate->slot = slot;
2257         cstate->session = session;
2258
2259 out:
2260         switch (clp->cl_cb_state) {
2261         case NFSD4_CB_DOWN:
2262                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2263                 break;
2264         case NFSD4_CB_FAULT:
2265                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2266                 break;
2267         default:
2268                 seq->status_flags = 0;
2269         }
2270         if (!list_empty(&clp->cl_revoked))
2271                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2272 out_no_session:
2273         kfree(conn);
2274         spin_unlock(&nn->client_lock);
2275         return status;
2276 out_put_session:
2277         nfsd4_put_session(session);
2278 out_put_client:
2279         put_client_renew_locked(clp);
2280         goto out_no_session;
2281 }
2282
2283 __be32
2284 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2285 {
2286         struct nfs4_client *conf, *unconf, *clp;
2287         __be32 status = 0;
2288         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2289
2290         nfs4_lock_state();
2291         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2292         conf = find_confirmed_client(&dc->clientid, true, nn);
2293         WARN_ON_ONCE(conf && unconf);
2294
2295         if (conf) {
2296                 clp = conf;
2297
2298                 if (client_has_state(conf)) {
2299                         status = nfserr_clientid_busy;
2300                         goto out;
2301                 }
2302         } else if (unconf)
2303                 clp = unconf;
2304         else {
2305                 status = nfserr_stale_clientid;
2306                 goto out;
2307         }
2308         if (!mach_creds_match(clp, rqstp)) {
2309                 status = nfserr_wrong_cred;
2310                 goto out;
2311         }
2312         expire_client(clp);
2313 out:
2314         nfs4_unlock_state();
2315         return status;
2316 }
2317
2318 __be32
2319 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2320 {
2321         __be32 status = 0;
2322
2323         if (rc->rca_one_fs) {
2324                 if (!cstate->current_fh.fh_dentry)
2325                         return nfserr_nofilehandle;
2326                 /*
2327                  * We don't take advantage of the rca_one_fs case.
2328                  * That's OK, it's optional, we can safely ignore it.
2329                  */
2330                  return nfs_ok;
2331         }
2332
2333         nfs4_lock_state();
2334         status = nfserr_complete_already;
2335         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2336                              &cstate->session->se_client->cl_flags))
2337                 goto out;
2338
2339         status = nfserr_stale_clientid;
2340         if (is_client_expired(cstate->session->se_client))
2341                 /*
2342                  * The following error isn't really legal.
2343                  * But we only get here if the client just explicitly
2344                  * destroyed the client.  Surely it no longer cares what
2345                  * error it gets back on an operation for the dead
2346                  * client.
2347                  */
2348                 goto out;
2349
2350         status = nfs_ok;
2351         nfsd4_client_record_create(cstate->session->se_client);
2352 out:
2353         nfs4_unlock_state();
2354         return status;
2355 }
2356
2357 __be32
2358 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2359                   struct nfsd4_setclientid *setclid)
2360 {
2361         struct xdr_netobj       clname = setclid->se_name;
2362         nfs4_verifier           clverifier = setclid->se_verf;
2363         struct nfs4_client      *conf, *unconf, *new;
2364         __be32                  status;
2365         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2366
2367         /* Cases below refer to rfc 3530 section 14.2.33: */
2368         nfs4_lock_state();
2369         conf = find_confirmed_client_by_name(&clname, nn);
2370         if (conf) {
2371                 /* case 0: */
2372                 status = nfserr_clid_inuse;
2373                 if (clp_used_exchangeid(conf))
2374                         goto out;
2375                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2376                         char addr_str[INET6_ADDRSTRLEN];
2377                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2378                                  sizeof(addr_str));
2379                         dprintk("NFSD: setclientid: string in use by client "
2380                                 "at %s\n", addr_str);
2381                         goto out;
2382                 }
2383         }
2384         unconf = find_unconfirmed_client_by_name(&clname, nn);
2385         if (unconf)
2386                 expire_client(unconf);
2387         status = nfserr_jukebox;
2388         new = create_client(clname, rqstp, &clverifier);
2389         if (new == NULL)
2390                 goto out;
2391         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2392                 /* case 1: probable callback update */
2393                 copy_clid(new, conf);
2394         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2395                 gen_clid(new, nn);
2396         new->cl_minorversion = 0;
2397         gen_callback(new, setclid, rqstp);
2398         add_to_unconfirmed(new);
2399         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2400         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2401         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2402         status = nfs_ok;
2403 out:
2404         nfs4_unlock_state();
2405         return status;
2406 }
2407
2408
2409 __be32
2410 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2411                          struct nfsd4_compound_state *cstate,
2412                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2413 {
2414         struct nfs4_client *conf, *unconf;
2415         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2416         clientid_t * clid = &setclientid_confirm->sc_clientid;
2417         __be32 status;
2418         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2419
2420         if (STALE_CLIENTID(clid, nn))
2421                 return nfserr_stale_clientid;
2422         nfs4_lock_state();
2423
2424         conf = find_confirmed_client(clid, false, nn);
2425         unconf = find_unconfirmed_client(clid, false, nn);
2426         /*
2427          * We try hard to give out unique clientid's, so if we get an
2428          * attempt to confirm the same clientid with a different cred,
2429          * there's a bug somewhere.  Let's charitably assume it's our
2430          * bug.
2431          */
2432         status = nfserr_serverfault;
2433         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2434                 goto out;
2435         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2436                 goto out;
2437         /* cases below refer to rfc 3530 section 14.2.34: */
2438         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2439                 if (conf && !unconf) /* case 2: probable retransmit */
2440                         status = nfs_ok;
2441                 else /* case 4: client hasn't noticed we rebooted yet? */
2442                         status = nfserr_stale_clientid;
2443                 goto out;
2444         }
2445         status = nfs_ok;
2446         if (conf) { /* case 1: callback update */
2447                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2448                 nfsd4_probe_callback(conf);
2449                 expire_client(unconf);
2450         } else { /* case 3: normal case; new or rebooted client */
2451                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2452                 if (conf) {
2453                         status = mark_client_expired(conf);
2454                         if (status)
2455                                 goto out;
2456                         expire_client(conf);
2457                 }
2458                 move_to_confirmed(unconf);
2459                 nfsd4_probe_callback(unconf);
2460         }
2461 out:
2462         nfs4_unlock_state();
2463         return status;
2464 }
2465
2466 static struct nfs4_file *nfsd4_alloc_file(void)
2467 {
2468         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2469 }
2470
2471 /* OPEN Share state helper functions */
2472 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2473 {
2474         unsigned int hashval = file_hashval(ino);
2475
2476         atomic_set(&fp->fi_ref, 1);
2477         INIT_LIST_HEAD(&fp->fi_stateids);
2478         INIT_LIST_HEAD(&fp->fi_delegations);
2479         fp->fi_inode = igrab(ino);
2480         fp->fi_had_conflict = false;
2481         fp->fi_lease = NULL;
2482         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2483         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2484         spin_lock(&recall_lock);
2485         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2486         spin_unlock(&recall_lock);
2487 }
2488
2489 static void
2490 nfsd4_free_slab(struct kmem_cache **slab)
2491 {
2492         if (*slab == NULL)
2493                 return;
2494         kmem_cache_destroy(*slab);
2495         *slab = NULL;
2496 }
2497
2498 void
2499 nfsd4_free_slabs(void)
2500 {
2501         nfsd4_free_slab(&openowner_slab);
2502         nfsd4_free_slab(&lockowner_slab);
2503         nfsd4_free_slab(&file_slab);
2504         nfsd4_free_slab(&stateid_slab);
2505         nfsd4_free_slab(&deleg_slab);
2506 }
2507
2508 int
2509 nfsd4_init_slabs(void)
2510 {
2511         openowner_slab = kmem_cache_create("nfsd4_openowners",
2512                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2513         if (openowner_slab == NULL)
2514                 goto out_nomem;
2515         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2516                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2517         if (lockowner_slab == NULL)
2518                 goto out_nomem;
2519         file_slab = kmem_cache_create("nfsd4_files",
2520                         sizeof(struct nfs4_file), 0, 0, NULL);
2521         if (file_slab == NULL)
2522                 goto out_nomem;
2523         stateid_slab = kmem_cache_create("nfsd4_stateids",
2524                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2525         if (stateid_slab == NULL)
2526                 goto out_nomem;
2527         deleg_slab = kmem_cache_create("nfsd4_delegations",
2528                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2529         if (deleg_slab == NULL)
2530                 goto out_nomem;
2531         return 0;
2532 out_nomem:
2533         nfsd4_free_slabs();
2534         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2535         return -ENOMEM;
2536 }
2537
2538 void nfs4_free_openowner(struct nfs4_openowner *oo)
2539 {
2540         kfree(oo->oo_owner.so_owner.data);
2541         kmem_cache_free(openowner_slab, oo);
2542 }
2543
2544 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2545 {
2546         kfree(lo->lo_owner.so_owner.data);
2547         kmem_cache_free(lockowner_slab, lo);
2548 }
2549
2550 static void init_nfs4_replay(struct nfs4_replay *rp)
2551 {
2552         rp->rp_status = nfserr_serverfault;
2553         rp->rp_buflen = 0;
2554         rp->rp_buf = rp->rp_ibuf;
2555 }
2556
2557 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2558 {
2559         struct nfs4_stateowner *sop;
2560
2561         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2562         if (!sop)
2563                 return NULL;
2564
2565         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2566         if (!sop->so_owner.data) {
2567                 kmem_cache_free(slab, sop);
2568                 return NULL;
2569         }
2570         sop->so_owner.len = owner->len;
2571
2572         INIT_LIST_HEAD(&sop->so_stateids);
2573         sop->so_client = clp;
2574         init_nfs4_replay(&sop->so_replay);
2575         return sop;
2576 }
2577
2578 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2579 {
2580         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2581
2582         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2583         list_add(&oo->oo_perclient, &clp->cl_openowners);
2584 }
2585
2586 static struct nfs4_openowner *
2587 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2588         struct nfs4_openowner *oo;
2589
2590         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2591         if (!oo)
2592                 return NULL;
2593         oo->oo_owner.so_is_open_owner = 1;
2594         oo->oo_owner.so_seqid = open->op_seqid;
2595         oo->oo_flags = NFS4_OO_NEW;
2596         oo->oo_time = 0;
2597         oo->oo_last_closed_stid = NULL;
2598         INIT_LIST_HEAD(&oo->oo_close_lru);
2599         hash_openowner(oo, clp, strhashval);
2600         return oo;
2601 }
2602
2603 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2604         struct nfs4_openowner *oo = open->op_openowner;
2605
2606         stp->st_stid.sc_type = NFS4_OPEN_STID;
2607         INIT_LIST_HEAD(&stp->st_lockowners);
2608         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2609         list_add(&stp->st_perfile, &fp->fi_stateids);
2610         stp->st_stateowner = &oo->oo_owner;
2611         get_nfs4_file(fp);
2612         stp->st_file = fp;
2613         stp->st_access_bmap = 0;
2614         stp->st_deny_bmap = 0;
2615         set_access(open->op_share_access, stp);
2616         set_deny(open->op_share_deny, stp);
2617         stp->st_openstp = NULL;
2618 }
2619
2620 static void
2621 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2622 {
2623         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2624
2625         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2626
2627         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2628         oo->oo_time = get_seconds();
2629 }
2630
2631 static int
2632 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2633                                                         clientid_t *clid)
2634 {
2635         return (sop->so_owner.len == owner->len) &&
2636                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2637                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2638 }
2639
2640 static struct nfs4_openowner *
2641 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2642                         bool sessions, struct nfsd_net *nn)
2643 {
2644         struct nfs4_stateowner *so;
2645         struct nfs4_openowner *oo;
2646         struct nfs4_client *clp;
2647
2648         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2649                 if (!so->so_is_open_owner)
2650                         continue;
2651                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2652                         oo = openowner(so);
2653                         clp = oo->oo_owner.so_client;
2654                         if ((bool)clp->cl_minorversion != sessions)
2655                                 return NULL;
2656                         renew_client(oo->oo_owner.so_client);
2657                         return oo;
2658                 }
2659         }
2660         return NULL;
2661 }
2662
2663 /* search file_hashtbl[] for file */
2664 static struct nfs4_file *
2665 find_file(struct inode *ino)
2666 {
2667         unsigned int hashval = file_hashval(ino);
2668         struct nfs4_file *fp;
2669
2670         spin_lock(&recall_lock);
2671         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2672                 if (fp->fi_inode == ino) {
2673                         get_nfs4_file(fp);
2674                         spin_unlock(&recall_lock);
2675                         return fp;
2676                 }
2677         }
2678         spin_unlock(&recall_lock);
2679         return NULL;
2680 }
2681
2682 /*
2683  * Called to check deny when READ with all zero stateid or
2684  * WRITE with all zero or all one stateid
2685  */
2686 static __be32
2687 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2688 {
2689         struct inode *ino = current_fh->fh_dentry->d_inode;
2690         struct nfs4_file *fp;
2691         struct nfs4_ol_stateid *stp;
2692         __be32 ret;
2693
2694         fp = find_file(ino);
2695         if (!fp)
2696                 return nfs_ok;
2697         ret = nfserr_locked;
2698         /* Search for conflicting share reservations */
2699         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2700                 if (test_deny(deny_type, stp) ||
2701                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2702                         goto out;
2703         }
2704         ret = nfs_ok;
2705 out:
2706         put_nfs4_file(fp);
2707         return ret;
2708 }
2709
2710 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2711 {
2712         struct nfs4_client *clp = dp->dl_stid.sc_client;
2713         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2714
2715         /* We're assuming the state code never drops its reference
2716          * without first removing the lease.  Since we're in this lease
2717          * callback (and since the lease code is serialized by the kernel
2718          * lock) we know the server hasn't removed the lease yet, we know
2719          * it's safe to take a reference: */
2720         atomic_inc(&dp->dl_count);
2721
2722         list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
2723
2724         /* Only place dl_time is set; protected by i_lock: */
2725         dp->dl_time = get_seconds();
2726
2727         nfsd4_cb_recall(dp);
2728 }
2729
2730 /* Called from break_lease() with i_lock held. */
2731 static void nfsd_break_deleg_cb(struct file_lock *fl)
2732 {
2733         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2734         struct nfs4_delegation *dp;
2735
2736         if (!fp) {
2737                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2738                 return;
2739         }
2740         if (fp->fi_had_conflict) {
2741                 WARN(1, "duplicate break on %p\n", fp);
2742                 return;
2743         }
2744         /*
2745          * We don't want the locks code to timeout the lease for us;
2746          * we'll remove it ourself if a delegation isn't returned
2747          * in time:
2748          */
2749         fl->fl_break_time = 0;
2750
2751         spin_lock(&recall_lock);
2752         fp->fi_had_conflict = true;
2753         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2754                 nfsd_break_one_deleg(dp);
2755         spin_unlock(&recall_lock);
2756 }
2757
2758 static
2759 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2760 {
2761         if (arg & F_UNLCK)
2762                 return lease_modify(onlist, arg);
2763         else
2764                 return -EAGAIN;
2765 }
2766
2767 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2768         .lm_break = nfsd_break_deleg_cb,
2769         .lm_change = nfsd_change_deleg_cb,
2770 };
2771
2772 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2773 {
2774         if (nfsd4_has_session(cstate))
2775                 return nfs_ok;
2776         if (seqid == so->so_seqid - 1)
2777                 return nfserr_replay_me;
2778         if (seqid == so->so_seqid)
2779                 return nfs_ok;
2780         return nfserr_bad_seqid;
2781 }
2782
2783 __be32
2784 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2785                     struct nfsd4_open *open, struct nfsd_net *nn)
2786 {
2787         clientid_t *clientid = &open->op_clientid;
2788         struct nfs4_client *clp = NULL;
2789         unsigned int strhashval;
2790         struct nfs4_openowner *oo = NULL;
2791         __be32 status;
2792
2793         if (STALE_CLIENTID(&open->op_clientid, nn))
2794                 return nfserr_stale_clientid;
2795         /*
2796          * In case we need it later, after we've already created the
2797          * file and don't want to risk a further failure:
2798          */
2799         open->op_file = nfsd4_alloc_file();
2800         if (open->op_file == NULL)
2801                 return nfserr_jukebox;
2802
2803         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2804         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2805         open->op_openowner = oo;
2806         if (!oo) {
2807                 clp = find_confirmed_client(clientid, cstate->minorversion,
2808                                             nn);
2809                 if (clp == NULL)
2810                         return nfserr_expired;
2811                 goto new_owner;
2812         }
2813         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2814                 /* Replace unconfirmed owners without checking for replay. */
2815                 clp = oo->oo_owner.so_client;
2816                 release_openowner(oo);
2817                 open->op_openowner = NULL;
2818                 goto new_owner;
2819         }
2820         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2821         if (status)
2822                 return status;
2823         clp = oo->oo_owner.so_client;
2824         goto alloc_stateid;
2825 new_owner:
2826         oo = alloc_init_open_stateowner(strhashval, clp, open);
2827         if (oo == NULL)
2828                 return nfserr_jukebox;
2829         open->op_openowner = oo;
2830 alloc_stateid:
2831         open->op_stp = nfs4_alloc_stateid(clp);
2832         if (!open->op_stp)
2833                 return nfserr_jukebox;
2834         return nfs_ok;
2835 }
2836
2837 static inline __be32
2838 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2839 {
2840         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2841                 return nfserr_openmode;
2842         else
2843                 return nfs_ok;
2844 }
2845
2846 static int share_access_to_flags(u32 share_access)
2847 {
2848         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2849 }
2850
2851 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2852 {
2853         struct nfs4_stid *ret;
2854
2855         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2856         if (!ret)
2857                 return NULL;
2858         return delegstateid(ret);
2859 }
2860
2861 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2862 {
2863         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2864                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2865 }
2866
2867 static __be32
2868 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
2869                 struct nfs4_delegation **dp)
2870 {
2871         int flags;
2872         __be32 status = nfserr_bad_stateid;
2873
2874         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2875         if (*dp == NULL)
2876                 goto out;
2877         flags = share_access_to_flags(open->op_share_access);
2878         status = nfs4_check_delegmode(*dp, flags);
2879         if (status)
2880                 *dp = NULL;
2881 out:
2882         if (!nfsd4_is_deleg_cur(open))
2883                 return nfs_ok;
2884         if (status)
2885                 return status;
2886         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2887         return nfs_ok;
2888 }
2889
2890 static __be32
2891 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2892 {
2893         struct nfs4_ol_stateid *local;
2894         struct nfs4_openowner *oo = open->op_openowner;
2895
2896         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2897                 /* ignore lock owners */
2898                 if (local->st_stateowner->so_is_open_owner == 0)
2899                         continue;
2900                 /* remember if we have seen this open owner */
2901                 if (local->st_stateowner == &oo->oo_owner)
2902                         *stpp = local;
2903                 /* check for conflicting share reservations */
2904                 if (!test_share(local, open))
2905                         return nfserr_share_denied;
2906         }
2907         return nfs_ok;
2908 }
2909
2910 static inline int nfs4_access_to_access(u32 nfs4_access)
2911 {
2912         int flags = 0;
2913
2914         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2915                 flags |= NFSD_MAY_READ;
2916         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2917                 flags |= NFSD_MAY_WRITE;
2918         return flags;
2919 }
2920
2921 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2922                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2923 {
2924         __be32 status;
2925         int oflag = nfs4_access_to_omode(open->op_share_access);
2926         int access = nfs4_access_to_access(open->op_share_access);
2927
2928         if (!fp->fi_fds[oflag]) {
2929                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2930                         &fp->fi_fds[oflag]);
2931                 if (status)
2932                         return status;
2933         }
2934         nfs4_file_get_access(fp, oflag);
2935
2936         return nfs_ok;
2937 }
2938
2939 static inline __be32
2940 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2941                 struct nfsd4_open *open)
2942 {
2943         struct iattr iattr = {
2944                 .ia_valid = ATTR_SIZE,
2945                 .ia_size = 0,
2946         };
2947         if (!open->op_truncate)
2948                 return 0;
2949         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2950                 return nfserr_inval;
2951         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2952 }
2953
2954 static __be32
2955 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)
2956 {
2957         u32 op_share_access = open->op_share_access;
2958         bool new_access;
2959         __be32 status;
2960
2961         new_access = !test_access(op_share_access, stp);
2962         if (new_access) {
2963                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2964                 if (status)
2965                         return status;
2966         }
2967         status = nfsd4_truncate(rqstp, cur_fh, open);
2968         if (status) {
2969                 if (new_access) {
2970                         int oflag = nfs4_access_to_omode(op_share_access);
2971                         nfs4_file_put_access(fp, oflag);
2972                 }
2973                 return status;
2974         }
2975         /* remember the open */
2976         set_access(op_share_access, stp);
2977         set_deny(open->op_share_deny, stp);
2978
2979         return nfs_ok;
2980 }
2981
2982
2983 static void
2984 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2985 {
2986         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2987 }
2988
2989 /* Should we give out recallable state?: */
2990 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2991 {
2992         if (clp->cl_cb_state == NFSD4_CB_UP)
2993                 return true;
2994         /*
2995          * In the sessions case, since we don't have to establish a
2996          * separate connection for callbacks, we assume it's OK
2997          * until we hear otherwise:
2998          */
2999         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3000 }
3001
3002 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
3003 {
3004         struct file_lock *fl;
3005
3006         fl = locks_alloc_lock();
3007         if (!fl)
3008                 return NULL;
3009         locks_init_lock(fl);
3010         fl->fl_lmops = &nfsd_lease_mng_ops;
3011         fl->fl_flags = FL_DELEG;
3012         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3013         fl->fl_end = OFFSET_MAX;
3014         fl->fl_owner = (fl_owner_t)(dp->dl_file);
3015         fl->fl_pid = current->tgid;
3016         return fl;
3017 }
3018
3019 static int nfs4_setlease(struct nfs4_delegation *dp)
3020 {
3021         struct nfs4_file *fp = dp->dl_file;
3022         struct file_lock *fl;
3023         int status;
3024
3025         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
3026         if (!fl)
3027                 return -ENOMEM;
3028         fl->fl_file = find_readable_file(fp);
3029         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3030         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
3031         if (status) {
3032                 list_del_init(&dp->dl_perclnt);
3033                 locks_free_lock(fl);
3034                 return status;
3035         }
3036         fp->fi_lease = fl;
3037         fp->fi_deleg_file = get_file(fl->fl_file);
3038         atomic_set(&fp->fi_delegees, 1);
3039         list_add(&dp->dl_perfile, &fp->fi_delegations);
3040         return 0;
3041 }
3042
3043 static int nfs4_set_delegation(struct nfs4_delegation *dp, struct nfs4_file *fp)
3044 {
3045         int status;
3046
3047         if (fp->fi_had_conflict)
3048                 return -EAGAIN;
3049         get_nfs4_file(fp);
3050         dp->dl_file = fp;
3051         if (!fp->fi_lease) {
3052                 status = nfs4_setlease(dp);
3053                 if (status)
3054                         goto out_free;
3055                 return 0;
3056         }
3057         spin_lock(&recall_lock);
3058         if (fp->fi_had_conflict) {
3059                 spin_unlock(&recall_lock);
3060                 status = -EAGAIN;
3061                 goto out_free;
3062         }
3063         atomic_inc(&fp->fi_delegees);
3064         list_add(&dp->dl_perfile, &fp->fi_delegations);
3065         spin_unlock(&recall_lock);
3066         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3067         return 0;
3068 out_free:
3069         put_nfs4_file(fp);
3070         dp->dl_file = fp;
3071         return status;
3072 }
3073
3074 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3075 {
3076         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3077         if (status == -EAGAIN)
3078                 open->op_why_no_deleg = WND4_CONTENTION;
3079         else {
3080                 open->op_why_no_deleg = WND4_RESOURCE;
3081                 switch (open->op_deleg_want) {
3082                 case NFS4_SHARE_WANT_READ_DELEG:
3083                 case NFS4_SHARE_WANT_WRITE_DELEG:
3084                 case NFS4_SHARE_WANT_ANY_DELEG:
3085                         break;
3086                 case NFS4_SHARE_WANT_CANCEL:
3087                         open->op_why_no_deleg = WND4_CANCELLED;
3088                         break;
3089                 case NFS4_SHARE_WANT_NO_DELEG:
3090                         WARN_ON_ONCE(1);
3091                 }
3092         }
3093 }
3094
3095 /*
3096  * Attempt to hand out a delegation.
3097  *
3098  * Note we don't support write delegations, and won't until the vfs has
3099  * proper support for them.
3100  */
3101 static void
3102 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
3103                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
3104 {
3105         struct nfs4_delegation *dp;
3106         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
3107         int cb_up;
3108         int status = 0;
3109
3110         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3111         open->op_recall = 0;
3112         switch (open->op_claim_type) {
3113                 case NFS4_OPEN_CLAIM_PREVIOUS:
3114                         if (!cb_up)
3115                                 open->op_recall = 1;
3116                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3117                                 goto out_no_deleg;
3118                         break;
3119                 case NFS4_OPEN_CLAIM_NULL:
3120                         /*
3121                          * Let's not give out any delegations till everyone's
3122                          * had the chance to reclaim theirs....
3123                          */
3124                         if (locks_in_grace(net))
3125                                 goto out_no_deleg;
3126                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3127                                 goto out_no_deleg;
3128                         /*
3129                          * Also, if the file was opened for write or
3130                          * create, there's a good chance the client's
3131                          * about to write to it, resulting in an
3132                          * immediate recall (since we don't support
3133                          * write delegations):
3134                          */
3135                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3136                                 goto out_no_deleg;
3137                         if (open->op_create == NFS4_OPEN_CREATE)
3138                                 goto out_no_deleg;
3139                         break;
3140                 default:
3141                         goto out_no_deleg;
3142         }
3143         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh);
3144         if (dp == NULL)
3145                 goto out_no_deleg;
3146         status = nfs4_set_delegation(dp, stp->st_file);
3147         if (status)
3148                 goto out_free;
3149
3150         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3151
3152         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3153                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3154         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3155         return;
3156 out_free:
3157         unhash_stid(&dp->dl_stid);
3158         nfs4_put_delegation(dp);
3159 out_no_deleg:
3160         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3161         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3162             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3163                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3164                 open->op_recall = 1;
3165         }
3166
3167         /* 4.1 client asking for a delegation? */
3168         if (open->op_deleg_want)
3169                 nfsd4_open_deleg_none_ext(open, status);
3170         return;
3171 }
3172
3173 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3174                                         struct nfs4_delegation *dp)
3175 {
3176         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3177             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3178                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3179                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3180         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3181                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3182                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3183                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3184         }
3185         /* Otherwise the client must be confused wanting a delegation
3186          * it already has, therefore we don't return
3187          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3188          */
3189 }
3190
3191 /*
3192  * called with nfs4_lock_state() held.
3193  */
3194 __be32
3195 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3196 {
3197         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3198         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3199         struct nfs4_file *fp = NULL;
3200         struct inode *ino = current_fh->fh_dentry->d_inode;
3201         struct nfs4_ol_stateid *stp = NULL;
3202         struct nfs4_delegation *dp = NULL;
3203         __be32 status;
3204
3205         /*
3206          * Lookup file; if found, lookup stateid and check open request,
3207          * and check for delegations in the process of being recalled.
3208          * If not found, create the nfs4_file struct
3209          */
3210         fp = find_file(ino);
3211         if (fp) {
3212                 if ((status = nfs4_check_open(fp, open, &stp)))
3213                         goto out;
3214                 status = nfs4_check_deleg(cl, open, &dp);
3215                 if (status)
3216                         goto out;
3217         } else {
3218                 status = nfserr_bad_stateid;
3219                 if (nfsd4_is_deleg_cur(open))
3220                         goto out;
3221                 status = nfserr_jukebox;
3222                 fp = open->op_file;
3223                 open->op_file = NULL;
3224                 nfsd4_init_file(fp, ino);
3225         }
3226
3227         /*
3228          * OPEN the file, or upgrade an existing OPEN.
3229          * If truncate fails, the OPEN fails.
3230          */
3231         if (stp) {
3232                 /* Stateid was found, this is an OPEN upgrade */
3233                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3234                 if (status)
3235                         goto out;
3236         } else {
3237                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3238                 if (status)
3239                         goto out;
3240                 status = nfsd4_truncate(rqstp, current_fh, open);
3241                 if (status)
3242                         goto out;
3243                 stp = open->op_stp;
3244                 open->op_stp = NULL;
3245                 init_open_stateid(stp, fp, open);
3246         }
3247         update_stateid(&stp->st_stid.sc_stateid);
3248         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3249
3250         if (nfsd4_has_session(&resp->cstate)) {
3251                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3252
3253                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3254                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3255                         open->op_why_no_deleg = WND4_NOT_WANTED;
3256                         goto nodeleg;
3257                 }
3258         }
3259
3260         /*
3261         * Attempt to hand out a delegation. No error return, because the
3262         * OPEN succeeds even if we fail.
3263         */
3264         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3265 nodeleg:
3266         status = nfs_ok;
3267
3268         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3269                 STATEID_VAL(&stp->st_stid.sc_stateid));
3270 out:
3271         /* 4.1 client trying to upgrade/downgrade delegation? */
3272         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3273             open->op_deleg_want)
3274                 nfsd4_deleg_xgrade_none_ext(open, dp);
3275
3276         if (fp)
3277                 put_nfs4_file(fp);
3278         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3279                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3280         /*
3281         * To finish the open response, we just need to set the rflags.
3282         */
3283         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3284         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3285             !nfsd4_has_session(&resp->cstate))
3286                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3287
3288         return status;
3289 }
3290
3291 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3292 {
3293         if (open->op_openowner) {
3294                 struct nfs4_openowner *oo = open->op_openowner;
3295
3296                 if (!list_empty(&oo->oo_owner.so_stateids))
3297                         list_del_init(&oo->oo_close_lru);
3298                 if (oo->oo_flags & NFS4_OO_NEW) {
3299                         if (status) {
3300                                 release_openowner(oo);
3301                                 open->op_openowner = NULL;
3302                         } else
3303                                 oo->oo_flags &= ~NFS4_OO_NEW;
3304                 }
3305         }
3306         if (open->op_file)
3307                 nfsd4_free_file(open->op_file);
3308         if (open->op_stp)
3309                 free_generic_stateid(open->op_stp);
3310 }
3311
3312 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3313 {
3314         struct nfs4_client *found;
3315
3316         if (STALE_CLIENTID(clid, nn))
3317                 return nfserr_stale_clientid;
3318         found = find_confirmed_client(clid, session, nn);
3319         if (clp)
3320                 *clp = found;
3321         return found ? nfs_ok : nfserr_expired;
3322 }
3323
3324 __be32
3325 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3326             clientid_t *clid)
3327 {
3328         struct nfs4_client *clp;
3329         __be32 status;
3330         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3331
3332         nfs4_lock_state();
3333         dprintk("process_renew(%08x/%08x): starting\n",
3334                         clid->cl_boot, clid->cl_id);
3335         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3336         if (status)
3337                 goto out;
3338         status = nfserr_cb_path_down;
3339         if (!list_empty(&clp->cl_delegations)
3340                         && clp->cl_cb_state != NFSD4_CB_UP)
3341                 goto out;
3342         status = nfs_ok;
3343 out:
3344         nfs4_unlock_state();
3345         return status;
3346 }
3347
3348 static void
3349 nfsd4_end_grace(struct nfsd_net *nn)
3350 {
3351         /* do nothing if grace period already ended */
3352         if (nn->grace_ended)
3353                 return;
3354
3355         dprintk("NFSD: end of grace period\n");
3356         nn->grace_ended = true;
3357         nfsd4_record_grace_done(nn, nn->boot_time);
3358         locks_end_grace(&nn->nfsd4_manager);
3359         /*
3360          * Now that every NFSv4 client has had the chance to recover and
3361          * to see the (possibly new, possibly shorter) lease time, we
3362          * can safely set the next grace time to the current lease time:
3363          */
3364         nn->nfsd4_grace = nn->nfsd4_lease;
3365 }
3366
3367 static time_t
3368 nfs4_laundromat(struct nfsd_net *nn)
3369 {
3370         struct nfs4_client *clp;
3371         struct nfs4_openowner *oo;
3372         struct nfs4_delegation *dp;
3373         struct list_head *pos, *next, reaplist;
3374         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3375         time_t t, clientid_val = nn->nfsd4_lease;
3376         time_t u, test_val = nn->nfsd4_lease;
3377
3378         nfs4_lock_state();
3379
3380         dprintk("NFSD: laundromat service - starting\n");
3381         nfsd4_end_grace(nn);
3382         INIT_LIST_HEAD(&reaplist);
3383         spin_lock(&nn->client_lock);
3384         list_for_each_safe(pos, next, &nn->client_lru) {
3385                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3386                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3387                         t = clp->cl_time - cutoff;
3388                         if (clientid_val > t)
3389                                 clientid_val = t;
3390                         break;
3391                 }
3392                 if (mark_client_expired_locked(clp)) {
3393                         dprintk("NFSD: client in use (clientid %08x)\n",
3394                                 clp->cl_clientid.cl_id);
3395                         continue;
3396                 }
3397                 list_move(&clp->cl_lru, &reaplist);
3398         }
3399         spin_unlock(&nn->client_lock);
3400         list_for_each_safe(pos, next, &reaplist) {
3401                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3402                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3403                         clp->cl_clientid.cl_id);
3404                 expire_client(clp);
3405         }
3406         spin_lock(&recall_lock);
3407         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3408                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3409                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3410                         continue;
3411                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3412                         u = dp->dl_time - cutoff;
3413                         if (test_val > u)
3414                                 test_val = u;
3415                         break;
3416                 }
3417                 list_move(&dp->dl_recall_lru, &reaplist);
3418         }
3419         spin_unlock(&recall_lock);
3420         list_for_each_safe(pos, next, &reaplist) {
3421                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3422                 revoke_delegation(dp);
3423         }
3424         test_val = nn->nfsd4_lease;
3425         list_for_each_safe(pos, next, &nn->close_lru) {
3426                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3427                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3428                         u = oo->oo_time - cutoff;
3429                         if (test_val > u)
3430                                 test_val = u;
3431                         break;
3432                 }
3433                 release_openowner(oo);
3434         }
3435         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3436                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3437         nfs4_unlock_state();
3438         return clientid_val;
3439 }
3440
3441 static struct workqueue_struct *laundry_wq;
3442 static void laundromat_main(struct work_struct *);
3443
3444 static void
3445 laundromat_main(struct work_struct *laundry)
3446 {
3447         time_t t;
3448         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3449                                                   work);
3450         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3451                                            laundromat_work);
3452
3453         t = nfs4_laundromat(nn);
3454         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3455         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3456 }
3457
3458 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3459 {
3460         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3461                 return nfserr_bad_stateid;
3462         return nfs_ok;
3463 }
3464
3465 static inline int
3466 access_permit_read(struct nfs4_ol_stateid *stp)
3467 {
3468         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3469                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3470                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3471 }
3472
3473 static inline int
3474 access_permit_write(struct nfs4_ol_stateid *stp)
3475 {
3476         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3477                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3478 }
3479
3480 static
3481 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3482 {
3483         __be32 status = nfserr_openmode;
3484
3485         /* For lock stateid's, we test the parent open, not the lock: */
3486         if (stp->st_openstp)
3487                 stp = stp->st_openstp;
3488         if ((flags & WR_STATE) && !access_permit_write(stp))
3489                 goto out;
3490         if ((flags & RD_STATE) && !access_permit_read(stp))
3491                 goto out;
3492         status = nfs_ok;
3493 out:
3494         return status;
3495 }
3496
3497 static inline __be32
3498 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3499 {
3500         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3501                 return nfs_ok;
3502         else if (locks_in_grace(net)) {
3503                 /* Answer in remaining cases depends on existence of
3504                  * conflicting state; so we must wait out the grace period. */
3505                 return nfserr_grace;
3506         } else if (flags & WR_STATE)
3507                 return nfs4_share_conflict(current_fh,
3508                                 NFS4_SHARE_DENY_WRITE);
3509         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3510                 return nfs4_share_conflict(current_fh,
3511                                 NFS4_SHARE_DENY_READ);
3512 }
3513
3514 /*
3515  * Allow READ/WRITE during grace period on recovered state only for files
3516  * that are not able to provide mandatory locking.
3517  */
3518 static inline int
3519 grace_disallows_io(struct net *net, struct inode *inode)
3520 {
3521         return locks_in_grace(net) && mandatory_lock(inode);
3522 }
3523
3524 /* Returns true iff a is later than b: */
3525 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3526 {
3527         return (s32)(a->si_generation - b->si_generation) > 0;
3528 }
3529
3530 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3531 {
3532         /*
3533          * When sessions are used the stateid generation number is ignored
3534          * when it is zero.
3535          */
3536         if (has_session && in->si_generation == 0)
3537                 return nfs_ok;
3538
3539         if (in->si_generation == ref->si_generation)
3540                 return nfs_ok;
3541
3542         /* If the client sends us a stateid from the future, it's buggy: */
3543         if (stateid_generation_after(in, ref))
3544                 return nfserr_bad_stateid;
3545         /*
3546          * However, we could see a stateid from the past, even from a
3547          * non-buggy client.  For example, if the client sends a lock
3548          * while some IO is outstanding, the lock may bump si_generation
3549          * while the IO is still in flight.  The client could avoid that
3550          * situation by waiting for responses on all the IO requests,
3551          * but better performance may result in retrying IO that
3552          * receives an old_stateid error if requests are rarely
3553          * reordered in flight:
3554          */
3555         return nfserr_old_stateid;
3556 }
3557
3558 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3559 {
3560         struct nfs4_stid *s;
3561         struct nfs4_ol_stateid *ols;
3562         __be32 status;
3563
3564         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3565                 return nfserr_bad_stateid;
3566         /* Client debugging aid. */
3567         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3568                 char addr_str[INET6_ADDRSTRLEN];
3569                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3570                                  sizeof(addr_str));
3571                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3572                                         "with incorrect client ID\n", addr_str);
3573                 return nfserr_bad_stateid;
3574         }
3575         s = find_stateid(cl, stateid);
3576         if (!s)
3577                 return nfserr_bad_stateid;
3578         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3579         if (status)
3580                 return status;
3581         switch (s->sc_type) {
3582         case NFS4_DELEG_STID:
3583                 return nfs_ok;
3584         case NFS4_REVOKED_DELEG_STID:
3585                 return nfserr_deleg_revoked;
3586         case NFS4_OPEN_STID:
3587         case NFS4_LOCK_STID:
3588                 ols = openlockstateid(s);
3589                 if (ols->st_stateowner->so_is_open_owner
3590                                 && !(openowner(ols->st_stateowner)->oo_flags
3591                                                 & NFS4_OO_CONFIRMED))
3592                         return nfserr_bad_stateid;
3593                 return nfs_ok;
3594         default:
3595                 printk("unknown stateid type %x\n", s->sc_type);
3596         case NFS4_CLOSED_STID:
3597                 return nfserr_bad_stateid;
3598         }
3599 }
3600
3601 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3602                                    struct nfs4_stid **s, bool sessions,
3603                                    struct nfsd_net *nn)
3604 {
3605         struct nfs4_client *cl;
3606         __be32 status;
3607
3608         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3609                 return nfserr_bad_stateid;
3610         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3611                                                         nn, &cl);
3612         if (status == nfserr_stale_clientid)
3613                 return nfserr_stale_stateid;
3614         if (status)
3615                 return status;
3616         *s = find_stateid_by_type(cl, stateid, typemask);
3617         if (!*s)
3618                 return nfserr_bad_stateid;
3619         return nfs_ok;
3620 }
3621
3622 /*
3623 * Checks for stateid operations
3624 */
3625 __be32
3626 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3627                            stateid_t *stateid, int flags, struct file **filpp)
3628 {
3629         struct nfs4_stid *s;
3630         struct nfs4_ol_stateid *stp = NULL;
3631         struct nfs4_delegation *dp = NULL;
3632         struct svc_fh *current_fh = &cstate->current_fh;
3633         struct inode *ino = current_fh->fh_dentry->d_inode;
3634         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3635         __be32 status;
3636
3637         if (filpp)
3638                 *filpp = NULL;
3639
3640         if (grace_disallows_io(net, ino))
3641                 return nfserr_grace;
3642
3643         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3644                 return check_special_stateids(net, current_fh, stateid, flags);
3645
3646         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3647                                       &s, cstate->minorversion, nn);
3648         if (status)
3649                 return status;
3650         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3651         if (status)
3652                 goto out;
3653         switch (s->sc_type) {
3654         case NFS4_DELEG_STID:
3655                 dp = delegstateid(s);
3656                 status = nfs4_check_delegmode(dp, flags);
3657                 if (status)
3658                         goto out;
3659                 if (filpp) {
3660                         *filpp = dp->dl_file->fi_deleg_file;
3661                         if (!*filpp) {
3662                                 WARN_ON_ONCE(1);
3663                                 status = nfserr_serverfault;
3664                                 goto out;
3665                         }
3666                 }
3667                 break;
3668         case NFS4_OPEN_STID:
3669         case NFS4_LOCK_STID:
3670                 stp = openlockstateid(s);
3671                 status = nfs4_check_fh(current_fh, stp);
3672                 if (status)
3673                         goto out;
3674                 if (stp->st_stateowner->so_is_open_owner
3675                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3676                         goto out;
3677                 status = nfs4_check_openmode(stp, flags);
3678                 if (status)
3679                         goto out;
3680                 if (filpp) {
3681                         if (flags & RD_STATE)
3682                                 *filpp = find_readable_file(stp->st_file);
3683                         else
3684                                 *filpp = find_writeable_file(stp->st_file);
3685                 }
3686                 break;
3687         default:
3688                 return nfserr_bad_stateid;
3689         }
3690         status = nfs_ok;
3691 out:
3692         return status;
3693 }
3694
3695 static __be32
3696 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3697 {
3698         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3699                 return nfserr_locks_held;
3700         release_lock_stateid(stp);
3701         return nfs_ok;
3702 }
3703
3704 /*
3705  * Test if the stateid is valid
3706  */
3707 __be32
3708 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3709                    struct nfsd4_test_stateid *test_stateid)
3710 {
3711         struct nfsd4_test_stateid_id *stateid;
3712         struct nfs4_client *cl = cstate->session->se_client;
3713
3714         nfs4_lock_state();
3715         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3716                 stateid->ts_id_status =
3717                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3718         nfs4_unlock_state();
3719
3720         return nfs_ok;
3721 }
3722
3723 __be32
3724 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3725                    struct nfsd4_free_stateid *free_stateid)
3726 {
3727         stateid_t *stateid = &free_stateid->fr_stateid;
3728         struct nfs4_stid *s;
3729         struct nfs4_delegation *dp;
3730         struct nfs4_client *cl = cstate->session->se_client;
3731         __be32 ret = nfserr_bad_stateid;
3732
3733         nfs4_lock_state();
3734         s = find_stateid(cl, stateid);
3735         if (!s)
3736                 goto out;
3737         switch (s->sc_type) {
3738         case NFS4_DELEG_STID:
3739                 ret = nfserr_locks_held;
3740                 goto out;
3741         case NFS4_OPEN_STID:
3742         case NFS4_LOCK_STID:
3743                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3744                 if (ret)
3745                         goto out;
3746                 if (s->sc_type == NFS4_LOCK_STID)
3747                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3748                 else
3749                         ret = nfserr_locks_held;
3750                 break;
3751         case NFS4_REVOKED_DELEG_STID:
3752                 dp = delegstateid(s);
3753                 destroy_revoked_delegation(dp);
3754                 ret = nfs_ok;
3755                 break;
3756         default:
3757                 ret = nfserr_bad_stateid;
3758         }
3759 out:
3760         nfs4_unlock_state();
3761         return ret;
3762 }
3763
3764 static inline int
3765 setlkflg (int type)
3766 {
3767         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3768                 RD_STATE : WR_STATE;
3769 }
3770
3771 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3772 {
3773         struct svc_fh *current_fh = &cstate->current_fh;
3774         struct nfs4_stateowner *sop = stp->st_stateowner;
3775         __be32 status;
3776
3777         status = nfsd4_check_seqid(cstate, sop, seqid);
3778         if (status)
3779                 return status;
3780         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
3781                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
3782                 /*
3783                  * "Closed" stateid's exist *only* to return
3784                  * nfserr_replay_me from the previous step, and
3785                  * revoked delegations are kept only for free_stateid.
3786                  */
3787                 return nfserr_bad_stateid;
3788         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3789         if (status)
3790                 return status;
3791         return nfs4_check_fh(current_fh, stp);
3792 }
3793
3794 /*
3795  * Checks for sequence id mutating operations.
3796  */
3797 static __be32
3798 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3799                          stateid_t *stateid, char typemask,
3800                          struct nfs4_ol_stateid **stpp,
3801                          struct nfsd_net *nn)
3802 {
3803         __be32 status;
3804         struct nfs4_stid *s;
3805
3806         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3807                 seqid, STATEID_VAL(stateid));
3808
3809         *stpp = NULL;
3810         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3811                                       cstate->minorversion, nn);
3812         if (status)
3813                 return status;
3814         *stpp = openlockstateid(s);
3815         if (!nfsd4_has_session(cstate))
3816                 cstate->replay_owner = (*stpp)->st_stateowner;
3817
3818         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3819 }
3820
3821 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3822                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3823 {
3824         __be32 status;
3825         struct nfs4_openowner *oo;
3826
3827         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3828                                                 NFS4_OPEN_STID, stpp, nn);
3829         if (status)
3830                 return status;
3831         oo = openowner((*stpp)->st_stateowner);
3832         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3833                 return nfserr_bad_stateid;
3834         return nfs_ok;
3835 }
3836
3837 __be32
3838 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3839                    struct nfsd4_open_confirm *oc)
3840 {
3841         __be32 status;
3842         struct nfs4_openowner *oo;
3843         struct nfs4_ol_stateid *stp;
3844         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3845
3846         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
3847                         cstate->current_fh.fh_dentry);
3848
3849         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3850         if (status)
3851                 return status;
3852
3853         nfs4_lock_state();
3854
3855         status = nfs4_preprocess_seqid_op(cstate,
3856                                         oc->oc_seqid, &oc->oc_req_stateid,
3857                                         NFS4_OPEN_STID, &stp, nn);
3858         if (status)
3859                 goto out;
3860         oo = openowner(stp->st_stateowner);
3861         status = nfserr_bad_stateid;
3862         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3863                 goto out;
3864         oo->oo_flags |= NFS4_OO_CONFIRMED;
3865         update_stateid(&stp->st_stid.sc_stateid);
3866         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3867         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3868                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3869
3870         nfsd4_client_record_create(oo->oo_owner.so_client);
3871         status = nfs_ok;
3872 out:
3873         nfsd4_bump_seqid(cstate, status);
3874         if (!cstate->replay_owner)
3875                 nfs4_unlock_state();
3876         return status;
3877 }
3878
3879 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3880 {
3881         if (!test_access(access, stp))
3882                 return;
3883         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3884         clear_access(access, stp);
3885 }
3886
3887 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3888 {
3889         switch (to_access) {
3890         case NFS4_SHARE_ACCESS_READ:
3891                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3892                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3893                 break;
3894         case NFS4_SHARE_ACCESS_WRITE:
3895                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3896                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3897                 break;
3898         case NFS4_SHARE_ACCESS_BOTH:
3899                 break;
3900         default:
3901                 WARN_ON_ONCE(1);
3902         }
3903 }
3904
3905 static void
3906 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3907 {
3908         int i;
3909         for (i = 0; i < 4; i++) {
3910                 if ((i & deny) != i)
3911                         clear_deny(i, stp);
3912         }
3913 }
3914
3915 __be32
3916 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3917                      struct nfsd4_compound_state *cstate,
3918                      struct nfsd4_open_downgrade *od)
3919 {
3920         __be32 status;
3921         struct nfs4_ol_stateid *stp;
3922         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3923
3924         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
3925                         cstate->current_fh.fh_dentry);
3926
3927         /* We don't yet support WANT bits: */
3928         if (od->od_deleg_want)
3929                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3930                         od->od_deleg_want);
3931
3932         nfs4_lock_state();
3933         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3934                                         &od->od_stateid, &stp, nn);
3935         if (status)
3936                 goto out;
3937         status = nfserr_inval;
3938         if (!test_access(od->od_share_access, stp)) {
3939                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3940                         stp->st_access_bmap, od->od_share_access);
3941                 goto out;
3942         }
3943         if (!test_deny(od->od_share_deny, stp)) {
3944                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3945                         stp->st_deny_bmap, od->od_share_deny);
3946                 goto out;
3947         }
3948         nfs4_stateid_downgrade(stp, od->od_share_access);
3949
3950         reset_union_bmap_deny(od->od_share_deny, stp);
3951
3952         update_stateid(&stp->st_stid.sc_stateid);
3953         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3954         status = nfs_ok;
3955 out:
3956         nfsd4_bump_seqid(cstate, status);
3957         if (!cstate->replay_owner)
3958                 nfs4_unlock_state();
3959         return status;
3960 }
3961
3962 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3963 {
3964         unhash_open_stateid(s);
3965         s->st_stid.sc_type = NFS4_CLOSED_STID;
3966 }
3967
3968 /*
3969  * nfs4_unlock_state() called after encode
3970  */
3971 __be32
3972 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3973             struct nfsd4_close *close)
3974 {
3975         __be32 status;
3976         struct nfs4_openowner *oo;
3977         struct nfs4_ol_stateid *stp;
3978         struct net *net = SVC_NET(rqstp);
3979         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3980
3981         dprintk("NFSD: nfsd4_close on file %pd\n",
3982                         cstate->current_fh.fh_dentry);
3983
3984         nfs4_lock_state();
3985         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3986                                         &close->cl_stateid,
3987                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3988                                         &stp, nn);
3989         nfsd4_bump_seqid(cstate, status);
3990         if (status)
3991                 goto out;
3992         oo = openowner(stp->st_stateowner);
3993         update_stateid(&stp->st_stid.sc_stateid);
3994         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3995
3996         nfsd4_close_open_stateid(stp);
3997
3998         if (cstate->minorversion) {
3999                 unhash_stid(&stp->st_stid);
4000                 free_generic_stateid(stp);
4001         } else
4002                 oo->oo_last_closed_stid = stp;
4003
4004         if (list_empty(&oo->oo_owner.so_stateids)) {
4005                 if (cstate->minorversion)
4006                         release_openowner(oo);
4007                 else {
4008                         /*
4009                          * In the 4.0 case we need to keep the owners around a
4010                          * little while to handle CLOSE replay.
4011                          */
4012                         move_to_close_lru(oo, SVC_NET(rqstp));
4013                 }
4014         }
4015 out:
4016         if (!cstate->replay_owner)
4017                 nfs4_unlock_state();
4018         return status;
4019 }
4020
4021 __be32
4022 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4023                   struct nfsd4_delegreturn *dr)
4024 {
4025         struct nfs4_delegation *dp;
4026         stateid_t *stateid = &dr->dr_stateid;
4027         struct nfs4_stid *s;
4028         __be32 status;
4029         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4030
4031         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4032                 return status;
4033
4034         nfs4_lock_state();
4035         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
4036                                       cstate->minorversion, nn);
4037         if (status)
4038                 goto out;
4039         dp = delegstateid(s);
4040         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4041         if (status)
4042                 goto out;
4043
4044         destroy_delegation(dp);
4045 out:
4046         nfs4_unlock_state();
4047
4048         return status;
4049 }
4050
4051
4052 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4053
4054 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
4055
4056 static inline u64
4057 end_offset(u64 start, u64 len)
4058 {
4059         u64 end;
4060
4061         end = start + len;
4062         return end >= start ? end: NFS4_MAX_UINT64;
4063 }
4064
4065 /* last octet in a range */
4066 static inline u64
4067 last_byte_offset(u64 start, u64 len)
4068 {
4069         u64 end;
4070
4071         WARN_ON_ONCE(!len);
4072         end = start + len;
4073         return end > start ? end - 1: NFS4_MAX_UINT64;
4074 }
4075
4076 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
4077 {
4078         return (file_hashval(inode) + cl_id
4079                         + opaque_hashval(ownername->data, ownername->len))
4080                 & LOCKOWNER_INO_HASH_MASK;
4081 }
4082
4083 /*
4084  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4085  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4086  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4087  * locking, this prevents us from being completely protocol-compliant.  The
4088  * real solution to this problem is to start using unsigned file offsets in
4089  * the VFS, but this is a very deep change!
4090  */
4091 static inline void
4092 nfs4_transform_lock_offset(struct file_lock *lock)
4093 {
4094         if (lock->fl_start < 0)
4095                 lock->fl_start = OFFSET_MAX;
4096         if (lock->fl_end < 0)
4097                 lock->fl_end = OFFSET_MAX;
4098 }
4099
4100 /* Hack!: For now, we're defining this just so we can use a pointer to it
4101  * as a unique cookie to identify our (NFSv4's) posix locks. */
4102 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4103 };
4104
4105 static inline void
4106 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4107 {
4108         struct nfs4_lockowner *lo;
4109
4110         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4111                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4112                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4113                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4114                 if (!deny->ld_owner.data)
4115                         /* We just don't care that much */
4116                         goto nevermind;
4117                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4118                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4119         } else {
4120 nevermind:
4121                 deny->ld_owner.len = 0;
4122                 deny->ld_owner.data = NULL;
4123                 deny->ld_clientid.cl_boot = 0;
4124                 deny->ld_clientid.cl_id = 0;
4125         }
4126         deny->ld_start = fl->fl_start;
4127         deny->ld_length = NFS4_MAX_UINT64;
4128         if (fl->fl_end != NFS4_MAX_UINT64)
4129                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4130         deny->ld_type = NFS4_READ_LT;
4131         if (fl->fl_type != F_RDLCK)
4132                 deny->ld_type = NFS4_WRITE_LT;
4133 }
4134
4135 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
4136 {
4137         struct nfs4_ol_stateid *lst;
4138
4139         if (!same_owner_str(&lo->lo_owner, owner, clid))
4140                 return false;
4141         lst = list_first_entry(&lo->lo_owner.so_stateids,
4142                                struct nfs4_ol_stateid, st_perstateowner);
4143         return lst->st_file->fi_inode == inode;
4144 }
4145
4146 static struct nfs4_lockowner *
4147 find_lockowner_str(struct inode *inode, clientid_t *clid,
4148                    struct xdr_netobj *owner, struct nfsd_net *nn)
4149 {
4150         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4151         struct nfs4_lockowner *lo;
4152
4153         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4154                 if (same_lockowner_ino(lo, inode, clid, owner))
4155                         return lo;
4156         }
4157         return NULL;
4158 }
4159
4160 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4161 {
4162         struct inode *inode = open_stp->st_file->fi_inode;
4163         unsigned int inohash = lockowner_ino_hashval(inode,
4164                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4165         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4166
4167         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4168         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4169         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4170 }
4171
4172 /*
4173  * Alloc a lock owner structure.
4174  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4175  * occurred.
4176  *
4177  * strhashval = ownerstr_hashval
4178  */
4179
4180 static struct nfs4_lockowner *
4181 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4182         struct nfs4_lockowner *lo;
4183
4184         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4185         if (!lo)
4186                 return NULL;
4187         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4188         lo->lo_owner.so_is_open_owner = 0;
4189         /* It is the openowner seqid that will be incremented in encode in the
4190          * case of new lockowners; so increment the lock seqid manually: */
4191         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4192         hash_lockowner(lo, strhashval, clp, open_stp);
4193         return lo;
4194 }
4195
4196 static struct nfs4_ol_stateid *
4197 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4198 {
4199         struct nfs4_ol_stateid *stp;
4200         struct nfs4_client *clp = lo->lo_owner.so_client;
4201
4202         stp = nfs4_alloc_stateid(clp);
4203         if (stp == NULL)
4204                 return NULL;
4205         stp->st_stid.sc_type = NFS4_LOCK_STID;
4206         list_add(&stp->st_perfile, &fp->fi_stateids);
4207         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4208         stp->st_stateowner = &lo->lo_owner;
4209         get_nfs4_file(fp);
4210         stp->st_file = fp;
4211         stp->st_access_bmap = 0;
4212         stp->st_deny_bmap = open_stp->st_deny_bmap;
4213         stp->st_openstp = open_stp;
4214         return stp;
4215 }
4216
4217 static int
4218 check_lock_length(u64 offset, u64 length)
4219 {
4220         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4221              LOFF_OVERFLOW(offset, length)));
4222 }
4223
4224 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4225 {
4226         struct nfs4_file *fp = lock_stp->st_file;
4227         int oflag = nfs4_access_to_omode(access);
4228
4229         if (test_access(access, lock_stp))
4230                 return;
4231         nfs4_file_get_access(fp, oflag);
4232         set_access(access, lock_stp);
4233 }
4234
4235 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)
4236 {
4237         struct nfs4_file *fi = ost->st_file;
4238         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4239         struct nfs4_client *cl = oo->oo_owner.so_client;
4240         struct nfs4_lockowner *lo;
4241         unsigned int strhashval;
4242         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4243
4244         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4245                                 &lock->v.new.owner, nn);
4246         if (lo) {
4247                 if (!cstate->minorversion)
4248                         return nfserr_bad_seqid;
4249                 /* XXX: a lockowner always has exactly one stateid: */
4250                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4251                                 struct nfs4_ol_stateid, st_perstateowner);
4252                 return nfs_ok;
4253         }
4254         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4255                         &lock->v.new.owner);
4256         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4257         if (lo == NULL)
4258                 return nfserr_jukebox;
4259         *lst = alloc_init_lock_stateid(lo, fi, ost);
4260         if (*lst == NULL) {
4261                 release_lockowner(lo);
4262                 return nfserr_jukebox;
4263         }
4264         *new = true;
4265         return nfs_ok;
4266 }
4267
4268 /*
4269  *  LOCK operation
4270  */
4271 __be32
4272 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4273            struct nfsd4_lock *lock)
4274 {
4275         struct nfs4_openowner *open_sop = NULL;
4276         struct nfs4_lockowner *lock_sop = NULL;
4277         struct nfs4_ol_stateid *lock_stp;
4278         struct file *filp = NULL;
4279         struct file_lock *file_lock = NULL;
4280         struct file_lock *conflock = NULL;
4281         __be32 status = 0;
4282         bool new_state = false;
4283         int lkflg;
4284         int err;
4285         struct net *net = SVC_NET(rqstp);
4286         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4287
4288         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4289                 (long long) lock->lk_offset,
4290                 (long long) lock->lk_length);
4291
4292         if (check_lock_length(lock->lk_offset, lock->lk_length))
4293                  return nfserr_inval;
4294
4295         if ((status = fh_verify(rqstp, &cstate->current_fh,
4296                                 S_IFREG, NFSD_MAY_LOCK))) {
4297                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4298                 return status;
4299         }
4300
4301         nfs4_lock_state();
4302
4303         if (lock->lk_is_new) {
4304                 struct nfs4_ol_stateid *open_stp = NULL;
4305
4306                 if (nfsd4_has_session(cstate))
4307                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4308                         memcpy(&lock->v.new.clientid,
4309                                 &cstate->session->se_client->cl_clientid,
4310                                 sizeof(clientid_t));
4311
4312                 status = nfserr_stale_clientid;
4313                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4314                         goto out;
4315
4316                 /* validate and update open stateid and open seqid */
4317                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4318                                         lock->lk_new_open_seqid,
4319                                         &lock->lk_new_open_stateid,
4320                                         &open_stp, nn);
4321                 if (status)
4322                         goto out;
4323                 open_sop = openowner(open_stp->st_stateowner);
4324                 status = nfserr_bad_stateid;
4325                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4326                                                 &lock->v.new.clientid))
4327                         goto out;
4328                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4329                                                         &lock_stp, &new_state);
4330         } else
4331                 status = nfs4_preprocess_seqid_op(cstate,
4332                                        lock->lk_old_lock_seqid,
4333                                        &lock->lk_old_lock_stateid,
4334                                        NFS4_LOCK_STID, &lock_stp, nn);
4335         if (status)
4336                 goto out;
4337         lock_sop = lockowner(lock_stp->st_stateowner);
4338
4339         lkflg = setlkflg(lock->lk_type);
4340         status = nfs4_check_openmode(lock_stp, lkflg);
4341         if (status)
4342                 goto out;
4343
4344         status = nfserr_grace;
4345         if (locks_in_grace(net) && !lock->lk_reclaim)
4346                 goto out;
4347         status = nfserr_no_grace;
4348         if (!locks_in_grace(net) && lock->lk_reclaim)
4349                 goto out;
4350
4351         file_lock = locks_alloc_lock();
4352         if (!file_lock) {
4353                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4354                 status = nfserr_jukebox;
4355                 goto out;
4356         }
4357
4358         locks_init_lock(file_lock);
4359         switch (lock->lk_type) {
4360                 case NFS4_READ_LT:
4361                 case NFS4_READW_LT:
4362                         filp = find_readable_file(lock_stp->st_file);
4363                         if (filp)
4364                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4365                         file_lock->fl_type = F_RDLCK;
4366                         break;
4367                 case NFS4_WRITE_LT:
4368                 case NFS4_WRITEW_LT:
4369                         filp = find_writeable_file(lock_stp->st_file);
4370                         if (filp)
4371                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4372                         file_lock->fl_type = F_WRLCK;
4373                         break;
4374                 default:
4375                         status = nfserr_inval;
4376                 goto out;
4377         }
4378         if (!filp) {
4379                 status = nfserr_openmode;
4380                 goto out;
4381         }
4382         file_lock->fl_owner = (fl_owner_t)lock_sop;
4383         file_lock->fl_pid = current->tgid;
4384         file_lock->fl_file = filp;
4385         file_lock->fl_flags = FL_POSIX;
4386         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4387         file_lock->fl_start = lock->lk_offset;
4388         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4389         nfs4_transform_lock_offset(file_lock);
4390
4391         conflock = locks_alloc_lock();
4392         if (!conflock) {
4393                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4394                 status = nfserr_jukebox;
4395                 goto out;
4396         }
4397
4398         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4399         switch (-err) {
4400         case 0: /* success! */
4401                 update_stateid(&lock_stp->st_stid.sc_stateid);
4402                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4403                                 sizeof(stateid_t));
4404                 status = 0;
4405                 break;
4406         case (EAGAIN):          /* conflock holds conflicting lock */
4407                 status = nfserr_denied;
4408                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4409                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4410                 break;
4411         case (EDEADLK):
4412                 status = nfserr_deadlock;
4413                 break;
4414         default:
4415                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4416                 status = nfserrno(err);
4417                 break;
4418         }
4419 out:
4420         if (status && new_state)
4421                 release_lockowner(lock_sop);
4422         nfsd4_bump_seqid(cstate, status);
4423         if (!cstate->replay_owner)
4424                 nfs4_unlock_state();
4425         if (file_lock)
4426                 locks_free_lock(file_lock);
4427         if (conflock)
4428                 locks_free_lock(conflock);
4429         return status;
4430 }
4431
4432 /*
4433  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4434  * so we do a temporary open here just to get an open file to pass to
4435  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4436  * inode operation.)
4437  */
4438 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4439 {
4440         struct file *file;
4441         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4442         if (!err) {
4443                 err = nfserrno(vfs_test_lock(file, lock));
4444                 nfsd_close(file);
4445         }
4446         return err;
4447 }
4448
4449 /*
4450  * LOCKT operation
4451  */
4452 __be32
4453 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4454             struct nfsd4_lockt *lockt)
4455 {
4456         struct inode *inode;
4457         struct file_lock *file_lock = NULL;
4458         struct nfs4_lockowner *lo;
4459         __be32 status;
4460         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4461
4462         if (locks_in_grace(SVC_NET(rqstp)))
4463                 return nfserr_grace;
4464
4465         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4466                  return nfserr_inval;
4467
4468         nfs4_lock_state();
4469
4470         if (!nfsd4_has_session(cstate)) {
4471                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4472                 if (status)
4473                         goto out;
4474         }
4475
4476         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4477                 goto out;
4478
4479         inode = cstate->current_fh.fh_dentry->d_inode;
4480         file_lock = locks_alloc_lock();
4481         if (!file_lock) {
4482                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4483                 status = nfserr_jukebox;
4484                 goto out;
4485         }
4486         locks_init_lock(file_lock);
4487         switch (lockt->lt_type) {
4488                 case NFS4_READ_LT:
4489                 case NFS4_READW_LT:
4490                         file_lock->fl_type = F_RDLCK;
4491                 break;
4492                 case NFS4_WRITE_LT:
4493                 case NFS4_WRITEW_LT:
4494                         file_lock->fl_type = F_WRLCK;
4495                 break;
4496                 default:
4497                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4498                         status = nfserr_inval;
4499                 goto out;
4500         }
4501
4502         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4503         if (lo)
4504                 file_lock->fl_owner = (fl_owner_t)lo;
4505         file_lock->fl_pid = current->tgid;
4506         file_lock->fl_flags = FL_POSIX;
4507
4508         file_lock->fl_start = lockt->lt_offset;
4509         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4510
4511         nfs4_transform_lock_offset(file_lock);
4512
4513         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4514         if (status)
4515                 goto out;
4516
4517         if (file_lock->fl_type != F_UNLCK) {
4518                 status = nfserr_denied;
4519                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4520         }
4521 out:
4522         nfs4_unlock_state();
4523         if (file_lock)
4524                 locks_free_lock(file_lock);
4525         return status;
4526 }
4527
4528 __be32
4529 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4530             struct nfsd4_locku *locku)
4531 {
4532         struct nfs4_ol_stateid *stp;
4533         struct file *filp = NULL;
4534         struct file_lock *file_lock = NULL;
4535         __be32 status;
4536         int err;
4537         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4538
4539         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4540                 (long long) locku->lu_offset,
4541                 (long long) locku->lu_length);
4542
4543         if (check_lock_length(locku->lu_offset, locku->lu_length))
4544                  return nfserr_inval;
4545
4546         nfs4_lock_state();
4547
4548         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4549                                         &locku->lu_stateid, NFS4_LOCK_STID,
4550                                         &stp, nn);
4551         if (status)
4552                 goto out;
4553         filp = find_any_file(stp->st_file);
4554         if (!filp) {
4555                 status = nfserr_lock_range;
4556                 goto out;
4557         }
4558         file_lock = locks_alloc_lock();
4559         if (!file_lock) {
4560                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4561                 status = nfserr_jukebox;
4562                 goto out;
4563         }
4564         locks_init_lock(file_lock);
4565         file_lock->fl_type = F_UNLCK;
4566         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4567         file_lock->fl_pid = current->tgid;
4568         file_lock->fl_file = filp;
4569         file_lock->fl_flags = FL_POSIX;
4570         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4571         file_lock->fl_start = locku->lu_offset;
4572
4573         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4574                                                 locku->lu_length);
4575         nfs4_transform_lock_offset(file_lock);
4576
4577         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4578         if (err) {
4579                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4580                 goto out_nfserr;
4581         }
4582         update_stateid(&stp->st_stid.sc_stateid);
4583         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4584
4585 out:
4586         nfsd4_bump_seqid(cstate, status);
4587         if (!cstate->replay_owner)
4588                 nfs4_unlock_state();
4589         if (file_lock)
4590                 locks_free_lock(file_lock);
4591         return status;
4592
4593 out_nfserr:
4594         status = nfserrno(err);
4595         goto out;
4596 }
4597
4598 /*
4599  * returns
4600  *      1: locks held by lockowner
4601  *      0: no locks held by lockowner
4602  */
4603 static int
4604 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4605 {
4606         struct file_lock **flpp;
4607         struct inode *inode = filp->fi_inode;
4608         int status = 0;
4609
4610         spin_lock(&inode->i_lock);
4611         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4612                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4613                         status = 1;
4614                         goto out;
4615                 }
4616         }
4617 out:
4618         spin_unlock(&inode->i_lock);
4619         return status;
4620 }
4621
4622 __be32
4623 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4624                         struct nfsd4_compound_state *cstate,
4625                         struct nfsd4_release_lockowner *rlockowner)
4626 {
4627         clientid_t *clid = &rlockowner->rl_clientid;
4628         struct nfs4_stateowner *sop;
4629         struct nfs4_lockowner *lo;
4630         struct nfs4_ol_stateid *stp;
4631         struct xdr_netobj *owner = &rlockowner->rl_owner;
4632         struct list_head matches;
4633         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4634         __be32 status;
4635         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4636
4637         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4638                 clid->cl_boot, clid->cl_id);
4639
4640         nfs4_lock_state();
4641
4642         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4643         if (status)
4644                 goto out;
4645
4646         status = nfserr_locks_held;
4647         INIT_LIST_HEAD(&matches);
4648
4649         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4650                 if (sop->so_is_open_owner)
4651                         continue;
4652                 if (!same_owner_str(sop, owner, clid))
4653                         continue;
4654                 list_for_each_entry(stp, &sop->so_stateids,
4655                                 st_perstateowner) {
4656                         lo = lockowner(sop);
4657                         if (check_for_locks(stp->st_file, lo))
4658                                 goto out;
4659                         list_add(&lo->lo_list, &matches);
4660                 }
4661         }
4662         /* Clients probably won't expect us to return with some (but not all)
4663          * of the lockowner state released; so don't release any until all
4664          * have been checked. */
4665         status = nfs_ok;
4666         while (!list_empty(&matches)) {
4667                 lo = list_entry(matches.next, struct nfs4_lockowner,
4668                                                                 lo_list);
4669                 /* unhash_stateowner deletes so_perclient only
4670                  * for openowners. */
4671                 list_del(&lo->lo_list);
4672                 release_lockowner(lo);
4673         }
4674 out:
4675         nfs4_unlock_state();
4676         return status;
4677 }
4678
4679 static inline struct nfs4_client_reclaim *
4680 alloc_reclaim(void)
4681 {
4682         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4683 }
4684
4685 bool
4686 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4687 {
4688         struct nfs4_client_reclaim *crp;
4689
4690         crp = nfsd4_find_reclaim_client(name, nn);
4691         return (crp && crp->cr_clp);
4692 }
4693
4694 /*
4695  * failure => all reset bets are off, nfserr_no_grace...
4696  */
4697 struct nfs4_client_reclaim *
4698 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4699 {
4700         unsigned int strhashval;
4701         struct nfs4_client_reclaim *crp;
4702
4703         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4704         crp = alloc_reclaim();
4705         if (crp) {
4706                 strhashval = clientstr_hashval(name);
4707                 INIT_LIST_HEAD(&crp->cr_strhash);
4708                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4709                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4710                 crp->cr_clp = NULL;
4711                 nn->reclaim_str_hashtbl_size++;
4712         }
4713         return crp;
4714 }
4715
4716 void
4717 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4718 {
4719         list_del(&crp->cr_strhash);
4720         kfree(crp);
4721         nn->reclaim_str_hashtbl_size--;
4722 }
4723
4724 void
4725 nfs4_release_reclaim(struct nfsd_net *nn)
4726 {
4727         struct nfs4_client_reclaim *crp = NULL;
4728         int i;
4729
4730         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4731                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4732                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4733                                         struct nfs4_client_reclaim, cr_strhash);
4734                         nfs4_remove_reclaim_record(crp, nn);
4735                 }
4736         }
4737         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4738 }
4739
4740 /*
4741  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4742 struct nfs4_client_reclaim *
4743 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4744 {
4745         unsigned int strhashval;
4746         struct nfs4_client_reclaim *crp = NULL;
4747
4748         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4749
4750         strhashval = clientstr_hashval(recdir);
4751         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4752                 if (same_name(crp->cr_recdir, recdir)) {
4753                         return crp;
4754                 }
4755         }
4756         return NULL;
4757 }
4758
4759 /*
4760 * Called from OPEN. Look for clientid in reclaim list.
4761 */
4762 __be32
4763 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4764 {
4765         struct nfs4_client *clp;
4766
4767         /* find clientid in conf_id_hashtbl */
4768         clp = find_confirmed_client(clid, sessions, nn);
4769         if (clp == NULL)
4770                 return nfserr_reclaim_bad;
4771
4772         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4773 }
4774
4775 #ifdef CONFIG_NFSD_FAULT_INJECTION
4776
4777 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4778 {
4779         if (mark_client_expired(clp))
4780                 return 0;
4781         expire_client(clp);
4782         return 1;
4783 }
4784
4785 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4786 {
4787         char buf[INET6_ADDRSTRLEN];
4788         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4789         printk(KERN_INFO "NFS Client: %s\n", buf);
4790         return 1;
4791 }
4792
4793 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4794                              const char *type)
4795 {
4796         char buf[INET6_ADDRSTRLEN];
4797         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4798         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4799 }
4800
4801 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4802 {
4803         struct nfs4_openowner *oop;
4804         struct nfs4_lockowner *lop, *lo_next;
4805         struct nfs4_ol_stateid *stp, *st_next;
4806         u64 count = 0;
4807
4808         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4809                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4810                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4811                                 if (func)
4812                                         func(lop);
4813                                 if (++count == max)
4814                                         return count;
4815                         }
4816                 }
4817         }
4818
4819         return count;
4820 }
4821
4822 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4823 {
4824         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4825 }
4826
4827 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4828 {
4829         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4830         nfsd_print_count(clp, count, "locked files");
4831         return count;
4832 }
4833
4834 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4835 {
4836         struct nfs4_openowner *oop, *next;
4837         u64 count = 0;
4838
4839         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4840                 if (func)
4841                         func(oop);
4842                 if (++count == max)
4843                         break;
4844         }
4845
4846         return count;
4847 }
4848
4849 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4850 {
4851         return nfsd_foreach_client_open(clp, max, release_openowner);
4852 }
4853
4854 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4855 {
4856         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4857         nfsd_print_count(clp, count, "open files");
4858         return count;
4859 }
4860
4861 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4862                                      struct list_head *victims)
4863 {
4864         struct nfs4_delegation *dp, *next;
4865         u64 count = 0;
4866
4867         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4868                 if (victims)
4869                         list_move(&dp->dl_recall_lru, victims);
4870                 if (++count == max)
4871                         break;
4872         }
4873         return count;
4874 }
4875
4876 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4877 {
4878         struct nfs4_delegation *dp, *next;
4879         LIST_HEAD(victims);
4880         u64 count;
4881
4882         spin_lock(&recall_lock);
4883         count = nfsd_find_all_delegations(clp, max, &victims);
4884         spin_unlock(&recall_lock);
4885
4886         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4887                 revoke_delegation(dp);
4888
4889         return count;
4890 }
4891
4892 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4893 {
4894         struct nfs4_delegation *dp, *next;
4895         LIST_HEAD(victims);
4896         u64 count;
4897
4898         spin_lock(&recall_lock);
4899         count = nfsd_find_all_delegations(clp, max, &victims);
4900         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4901                 nfsd_break_one_deleg(dp);
4902         spin_unlock(&recall_lock);
4903
4904         return count;
4905 }
4906
4907 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4908 {
4909         u64 count = 0;
4910
4911         spin_lock(&recall_lock);
4912         count = nfsd_find_all_delegations(clp, max, NULL);
4913         spin_unlock(&recall_lock);
4914
4915         nfsd_print_count(clp, count, "delegations");
4916         return count;
4917 }
4918
4919 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4920 {
4921         struct nfs4_client *clp, *next;
4922         u64 count = 0;
4923         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4924
4925         if (!nfsd_netns_ready(nn))
4926                 return 0;
4927
4928         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4929                 count += func(clp, max - count);
4930                 if ((max != 0) && (count >= max))
4931                         break;
4932         }
4933
4934         return count;
4935 }
4936
4937 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4938 {
4939         struct nfs4_client *clp;
4940         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4941
4942         if (!nfsd_netns_ready(nn))
4943                 return NULL;
4944
4945         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4946                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4947                         return clp;
4948         }
4949         return NULL;
4950 }
4951
4952 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4953
4954 /* initialization to perform at module load time: */
4955
4956 void
4957 nfs4_state_init(void)
4958 {
4959 }
4960
4961 /*
4962  * Since the lifetime of a delegation isn't limited to that of an open, a
4963  * client may quite reasonably hang on to a delegation as long as it has
4964  * the inode cached.  This becomes an obvious problem the first time a
4965  * client's inode cache approaches the size of the server's total memory.
4966  *
4967  * For now we avoid this problem by imposing a hard limit on the number
4968  * of delegations, which varies according to the server's memory size.
4969  */
4970 static void
4971 set_max_delegations(void)
4972 {
4973         /*
4974          * Allow at most 4 delegations per megabyte of RAM.  Quick
4975          * estimates suggest that in the worst case (where every delegation
4976          * is for a different inode), a delegation could take about 1.5K,
4977          * giving a worst case usage of about 6% of memory.
4978          */
4979         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4980 }
4981
4982 static int nfs4_state_create_net(struct net *net)
4983 {
4984         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4985         int i;
4986
4987         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4988                         CLIENT_HASH_SIZE, GFP_KERNEL);
4989         if (!nn->conf_id_hashtbl)
4990                 goto err;
4991         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4992                         CLIENT_HASH_SIZE, GFP_KERNEL);
4993         if (!nn->unconf_id_hashtbl)
4994                 goto err_unconf_id;
4995         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4996                         OWNER_HASH_SIZE, GFP_KERNEL);
4997         if (!nn->ownerstr_hashtbl)
4998                 goto err_ownerstr;
4999         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
5000                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
5001         if (!nn->lockowner_ino_hashtbl)
5002                 goto err_lockowner_ino;
5003         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
5004                         SESSION_HASH_SIZE, GFP_KERNEL);
5005         if (!nn->sessionid_hashtbl)
5006                 goto err_sessionid;
5007
5008         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5009                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
5010                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
5011         }
5012         for (i = 0; i < OWNER_HASH_SIZE; i++)
5013                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
5014         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
5015                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
5016         for (i = 0; i < SESSION_HASH_SIZE; i++)
5017                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
5018         nn->conf_name_tree = RB_ROOT;
5019         nn->unconf_name_tree = RB_ROOT;
5020         INIT_LIST_HEAD(&nn->client_lru);
5021         INIT_LIST_HEAD(&nn->close_lru);
5022         INIT_LIST_HEAD(&nn->del_recall_lru);
5023         spin_lock_init(&nn->client_lock);
5024
5025         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
5026         get_net(net);
5027
5028         return 0;
5029
5030 err_sessionid:
5031         kfree(nn->lockowner_ino_hashtbl);
5032 err_lockowner_ino:
5033         kfree(nn->ownerstr_hashtbl);
5034 err_ownerstr:
5035         kfree(nn->unconf_id_hashtbl);
5036 err_unconf_id:
5037         kfree(nn->conf_id_hashtbl);
5038 err:
5039         return -ENOMEM;
5040 }
5041
5042 static void
5043 nfs4_state_destroy_net(struct net *net)
5044 {
5045         int i;
5046         struct nfs4_client *clp = NULL;
5047         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5048         struct rb_node *node, *tmp;
5049
5050         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5051                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
5052                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
5053                         destroy_client(clp);
5054                 }
5055         }
5056
5057         node = rb_first(&nn->unconf_name_tree);
5058         while (node != NULL) {
5059                 tmp = node;
5060                 node = rb_next(tmp);
5061                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
5062                 rb_erase(tmp, &nn->unconf_name_tree);
5063                 destroy_client(clp);
5064         }
5065
5066         kfree(nn->sessionid_hashtbl);
5067         kfree(nn->lockowner_ino_hashtbl);
5068         kfree(nn->ownerstr_hashtbl);
5069         kfree(nn->unconf_id_hashtbl);
5070         kfree(nn->conf_id_hashtbl);
5071         put_net(net);
5072 }
5073
5074 int
5075 nfs4_state_start_net(struct net *net)
5076 {
5077         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5078         int ret;
5079
5080         ret = nfs4_state_create_net(net);
5081         if (ret)
5082                 return ret;
5083         nfsd4_client_tracking_init(net);
5084         nn->boot_time = get_seconds();
5085         locks_start_grace(net, &nn->nfsd4_manager);
5086         nn->grace_ended = false;
5087         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
5088                nn->nfsd4_grace, net);
5089         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
5090         return 0;
5091 }
5092
5093 /* initialization to perform when the nfsd service is started: */
5094
5095 int
5096 nfs4_state_start(void)
5097 {
5098         int ret;
5099
5100         ret = set_callback_cred();
5101         if (ret)
5102                 return -ENOMEM;
5103         laundry_wq = create_singlethread_workqueue("nfsd4");
5104         if (laundry_wq == NULL) {
5105                 ret = -ENOMEM;
5106                 goto out_recovery;
5107         }
5108         ret = nfsd4_create_callback_queue();
5109         if (ret)
5110                 goto out_free_laundry;
5111
5112         set_max_delegations();
5113
5114         return 0;
5115
5116 out_free_laundry:
5117         destroy_workqueue(laundry_wq);
5118 out_recovery:
5119         return ret;
5120 }
5121
5122 /* should be called with the state lock held */
5123 void
5124 nfs4_state_shutdown_net(struct net *net)
5125 {
5126         struct nfs4_delegation *dp = NULL;
5127         struct list_head *pos, *next, reaplist;
5128         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5129
5130         cancel_delayed_work_sync(&nn->laundromat_work);
5131         locks_end_grace(&nn->nfsd4_manager);
5132
5133         INIT_LIST_HEAD(&reaplist);
5134         spin_lock(&recall_lock);
5135         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5136                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5137                 list_move(&dp->dl_recall_lru, &reaplist);
5138         }
5139         spin_unlock(&recall_lock);
5140         list_for_each_safe(pos, next, &reaplist) {
5141                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5142                 destroy_delegation(dp);
5143         }
5144
5145         nfsd4_client_tracking_exit(net);
5146         nfs4_state_destroy_net(net);
5147 }
5148
5149 void
5150 nfs4_state_shutdown(void)
5151 {
5152         destroy_workqueue(laundry_wq);
5153         nfsd4_destroy_callback_queue();
5154 }
5155
5156 static void
5157 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5158 {
5159         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5160                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5161 }
5162
5163 static void
5164 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5165 {
5166         if (cstate->minorversion) {
5167                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5168                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5169         }
5170 }
5171
5172 void
5173 clear_current_stateid(struct nfsd4_compound_state *cstate)
5174 {
5175         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5176 }
5177
5178 /*
5179  * functions to set current state id
5180  */
5181 void
5182 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5183 {
5184         put_stateid(cstate, &odp->od_stateid);
5185 }
5186
5187 void
5188 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5189 {
5190         put_stateid(cstate, &open->op_stateid);
5191 }
5192
5193 void
5194 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5195 {
5196         put_stateid(cstate, &close->cl_stateid);
5197 }
5198
5199 void
5200 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5201 {
5202         put_stateid(cstate, &lock->lk_resp_stateid);
5203 }
5204
5205 /*
5206  * functions to consume current state id
5207  */
5208
5209 void
5210 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5211 {
5212         get_stateid(cstate, &odp->od_stateid);
5213 }
5214
5215 void
5216 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5217 {
5218         get_stateid(cstate, &drp->dr_stateid);
5219 }
5220
5221 void
5222 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5223 {
5224         get_stateid(cstate, &fsp->fr_stateid);
5225 }
5226
5227 void
5228 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5229 {
5230         get_stateid(cstate, &setattr->sa_stateid);
5231 }
5232
5233 void
5234 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5235 {
5236         get_stateid(cstate, &close->cl_stateid);
5237 }
5238
5239 void
5240 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5241 {
5242         get_stateid(cstate, &locku->lu_stateid);
5243 }
5244
5245 void
5246 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5247 {
5248         get_stateid(cstate, &read->rd_stateid);
5249 }
5250
5251 void
5252 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5253 {
5254         get_stateid(cstate, &write->wr_stateid);
5255 }