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 static void nfs4_free_stid(struct kmem_cache *slab, struct nfs4_stid *s)
 406 {
 407         kmem_cache_free(slab, s);
 408 }
 409
 410 void
 411 nfs4_put_delegation(struct nfs4_delegation *dp)
 412 {
 413         if (atomic_dec_and_test(&dp->dl_count)) {
 414                 nfs4_free_stid(deleg_slab, &dp->dl_stid);
 415                 num_delegations--;
 416         }
 417 }
 418
 419 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 420 {
 421         if (atomic_dec_and_test(&fp->fi_delegees)) {
 422                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 423                 fp->fi_lease = NULL;
 424                 fput(fp->fi_deleg_file);
 425                 fp->fi_deleg_file = NULL;
 426         }
 427 }
 428
 429 static void unhash_stid(struct nfs4_stid *s)
 430 {
 431         s->sc_type = 0;
 432 }
 433
 434 /* Called under the state lock. */
 435 static void
 436 unhash_delegation(struct nfs4_delegation *dp)
 437 {
 438         list_del_init(&dp->dl_perclnt);
 439         spin_lock(&recall_lock);
 440         list_del_init(&dp->dl_perfile);
 441         list_del_init(&dp->dl_recall_lru);
 442         spin_unlock(&recall_lock);
 443         nfs4_put_deleg_lease(dp->dl_file);
 444         put_nfs4_file(dp->dl_file);
 445         dp->dl_file = NULL;
 446 }
 447
 448
 449
 450 static void destroy_revoked_delegation(struct nfs4_delegation *dp)
 451 {
 452         list_del_init(&dp->dl_recall_lru);
 453         remove_stid(&dp->dl_stid);
 454         nfs4_put_delegation(dp);
 455 }
 456
 457 static void destroy_delegation(struct nfs4_delegation *dp)
 458 {
 459         unhash_delegation(dp);
 460         remove_stid(&dp->dl_stid);
 461         nfs4_put_delegation(dp);
 462 }
 463
 464 static void revoke_delegation(struct nfs4_delegation *dp)
 465 {
 466         struct nfs4_client *clp = dp->dl_stid.sc_client;
 467
 468         if (clp->cl_minorversion == 0)
 469                 destroy_delegation(dp);
 470         else {
 471                 unhash_delegation(dp);
 472                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 473                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 474         }
 475 }
 476
 477 /*
 478  * SETCLIENTID state
 479  */
 480
 481 static unsigned int clientid_hashval(u32 id)
 482 {
 483         return id & CLIENT_HASH_MASK;
 484 }
 485
 486 static unsigned int clientstr_hashval(const char *name)
 487 {
 488         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 489 }
 490
 491 /*
 492  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 493  * st_{access,deny}_bmap field of the stateid, in order to track not
 494  * only what share bits are currently in force, but also what
 495  * combinations of share bits previous opens have used.  This allows us
 496  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 497  * return an error if the client attempt to downgrade to a combination
 498  * of share bits not explicable by closing some of its previous opens.
 499  *
 500  * XXX: This enforcement is actually incomplete, since we don't keep
 501  * track of access/deny bit combinations; so, e.g., we allow:
 502  *
 503  *      OPEN allow read, deny write
 504  *      OPEN allow both, deny none
 505  *      DOWNGRADE allow read, deny none
 506  *
 507  * which we should reject.
 508  */
 509 static unsigned int
 510 bmap_to_share_mode(unsigned long bmap) {
 511         int i;
 512         unsigned int access = 0;
 513
 514         for (i = 1; i < 4; i++) {
 515                 if (test_bit(i, &bmap))
 516                         access |= i;
 517         }
 518         return access;
 519 }
 520
 521 static bool
 522 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 523         unsigned int access, deny;
 524
 525         access = bmap_to_share_mode(stp->st_access_bmap);
 526         deny = bmap_to_share_mode(stp->st_deny_bmap);
 527         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 528                 return false;
 529         return true;
 530 }
 531
 532 /* set share access for a given stateid */
 533 static inline void
 534 set_access(u32 access, struct nfs4_ol_stateid *stp)
 535 {
 536         __set_bit(access, &stp->st_access_bmap);
 537 }
 538
 539 /* clear share access for a given stateid */
 540 static inline void
 541 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 542 {
 543         __clear_bit(access, &stp->st_access_bmap);
 544 }
 545
 546 /* test whether a given stateid has access */
 547 static inline bool
 548 test_access(u32 access, struct nfs4_ol_stateid *stp)
 549 {
 550         return test_bit(access, &stp->st_access_bmap);
 551 }
 552
 553 /* set share deny for a given stateid */
 554 static inline void
 555 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 556 {
 557         __set_bit(access, &stp->st_deny_bmap);
 558 }
 559
 560 /* clear share deny for a given stateid */
 561 static inline void
 562 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 563 {
 564         __clear_bit(access, &stp->st_deny_bmap);
 565 }
 566
 567 /* test whether a given stateid is denying specific access */
 568 static inline bool
 569 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 570 {
 571         return test_bit(access, &stp->st_deny_bmap);
 572 }
 573
 574 static int nfs4_access_to_omode(u32 access)
 575 {
 576         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 577         case NFS4_SHARE_ACCESS_READ:
 578                 return O_RDONLY;
 579         case NFS4_SHARE_ACCESS_WRITE:
 580                 return O_WRONLY;
 581         case NFS4_SHARE_ACCESS_BOTH:
 582                 return O_RDWR;
 583         }
 584         WARN_ON_ONCE(1);
 585         return O_RDONLY;
 586 }
 587
 588 /* release all access and file references for a given stateid */
 589 static void
 590 release_all_access(struct nfs4_ol_stateid *stp)
 591 {
 592         int i;
 593
 594         for (i = 1; i < 4; i++) {
 595                 if (test_access(i, stp))
 596                         nfs4_file_put_access(stp->st_file,
 597                                              nfs4_access_to_omode(i));
 598                 clear_access(i, stp);
 599         }
 600 }
 601
 602 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 603 {
 604         list_del(&stp->st_perfile);
 605         list_del(&stp->st_perstateowner);
 606 }
 607
 608 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 609 {
 610         release_all_access(stp);
 611         put_nfs4_file(stp->st_file);
 612         stp->st_file = NULL;
 613 }
 614
 615 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 616 {
 617         remove_stid(&stp->st_stid);
 618         nfs4_free_stid(stateid_slab, &stp->st_stid);
 619 }
 620
 621 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 622 {
 623         struct file *file;
 624
 625         unhash_generic_stateid(stp);
 626         unhash_stid(&stp->st_stid);
 627         file = find_any_file(stp->st_file);
 628         if (file)
 629                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 630         close_generic_stateid(stp);
 631         free_generic_stateid(stp);
 632 }
 633
 634 static void unhash_lockowner(struct nfs4_lockowner *lo)
 635 {
 636         struct nfs4_ol_stateid *stp;
 637
 638         list_del(&lo->lo_owner.so_strhash);
 639         list_del(&lo->lo_perstateid);
 640         list_del(&lo->lo_owner_ino_hash);
 641         while (!list_empty(&lo->lo_owner.so_stateids)) {
 642                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 643                                 struct nfs4_ol_stateid, st_perstateowner);
 644                 release_lock_stateid(stp);
 645         }
 646 }
 647
 648 static void release_lockowner(struct nfs4_lockowner *lo)
 649 {
 650         unhash_lockowner(lo);
 651         nfs4_free_lockowner(lo);
 652 }
 653
 654 static void
 655 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 656 {
 657         struct nfs4_lockowner *lo;
 658
 659         while (!list_empty(&open_stp->st_lockowners)) {
 660                 lo = list_entry(open_stp->st_lockowners.next,
 661                                 struct nfs4_lockowner, lo_perstateid);
 662                 release_lockowner(lo);
 663         }
 664 }
 665
 666 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 667 {
 668         unhash_generic_stateid(stp);
 669         release_stateid_lockowners(stp);
 670         close_generic_stateid(stp);
 671 }
 672
 673 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 674 {
 675         unhash_open_stateid(stp);
 676         free_generic_stateid(stp);
 677 }
 678
 679 static void unhash_openowner(struct nfs4_openowner *oo)
 680 {
 681         struct nfs4_ol_stateid *stp;
 682
 683         list_del(&oo->oo_owner.so_strhash);
 684         list_del(&oo->oo_perclient);
 685         while (!list_empty(&oo->oo_owner.so_stateids)) {
 686                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 687                                 struct nfs4_ol_stateid, st_perstateowner);
 688                 release_open_stateid(stp);
 689         }
 690 }
 691
 692 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 693 {
 694         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 695
 696         if (s) {
 697                 free_generic_stateid(s);
 698                 oo->oo_last_closed_stid = NULL;
 699         }
 700 }
 701
 702 static void release_openowner(struct nfs4_openowner *oo)
 703 {
 704         unhash_openowner(oo);
 705         list_del(&oo->oo_close_lru);
 706         release_last_closed_stateid(oo);
 707         nfs4_free_openowner(oo);
 708 }
 709
 710 static inline int
 711 hash_sessionid(struct nfs4_sessionid *sessionid)
 712 {
 713         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 714
 715         return sid->sequence % SESSION_HASH_SIZE;
 716 }
 717
 718 #ifdef NFSD_DEBUG
 719 static inline void
 720 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 721 {
 722         u32 *ptr = (u32 *)(&sessionid->data[0]);
 723         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 724 }
 725 #else
 726 static inline void
 727 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 728 {
 729 }
 730 #endif
 731
 732 /*
 733  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
 734  * won't be used for replay.
 735  */
 736 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
 737 {
 738         struct nfs4_stateowner *so = cstate->replay_owner;
 739
 740         if (nfserr == nfserr_replay_me)
 741                 return;
 742
 743         if (!seqid_mutating_err(ntohl(nfserr))) {
 744                 cstate->replay_owner = NULL;
 745                 return;
 746         }
 747         if (!so)
 748                 return;
 749         if (so->so_is_open_owner)
 750                 release_last_closed_stateid(openowner(so));
 751         so->so_seqid++;
 752         return;
 753 }
 754
 755 static void
 756 gen_sessionid(struct nfsd4_session *ses)
 757 {
 758         struct nfs4_client *clp = ses->se_client;
 759         struct nfsd4_sessionid *sid;
 760
 761         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 762         sid->clientid = clp->cl_clientid;
 763         sid->sequence = current_sessionid++;
 764         sid->reserved = 0;
 765 }
 766
 767 /*
 768  * The protocol defines ca_maxresponssize_cached to include the size of
 769  * the rpc header, but all we need to cache is the data starting after
 770  * the end of the initial SEQUENCE operation--the rest we regenerate
 771  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 772  * value that is the number of bytes in our cache plus a few additional
 773  * bytes.  In order to stay on the safe side, and not promise more than
 774  * we can cache, those additional bytes must be the minimum possible: 24
 775  * bytes of rpc header (xid through accept state, with AUTH_NULL
 776  * verifier), 12 for the compound header (with zero-length tag), and 44
 777  * for the SEQUENCE op response:
 778  */
 779 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 780
 781 static void
 782 free_session_slots(struct nfsd4_session *ses)
 783 {
 784         int i;
 785
 786         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 787                 kfree(ses->se_slots[i]);
 788 }
 789
 790 /*
 791  * We don't actually need to cache the rpc and session headers, so we
 792  * can allocate a little less for each slot:
 793  */
 794 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
 795 {
 796         u32 size;
 797
 798         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
 799                 size = 0;
 800         else
 801                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 802         return size + sizeof(struct nfsd4_slot);
 803 }
 804
 805 /*
 806  * XXX: If we run out of reserved DRC memory we could (up to a point)
 807  * re-negotiate active sessions and reduce their slot usage to make
 808  * room for new connections. For now we just fail the create session.
 809  */
 810 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
 811 {
 812         u32 slotsize = slot_bytes(ca);
 813         u32 num = ca->maxreqs;
 814         int avail;
 815
 816         spin_lock(&nfsd_drc_lock);
 817         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
 818                     nfsd_drc_max_mem - nfsd_drc_mem_used);
 819         num = min_t(int, num, avail / slotsize);
 820         nfsd_drc_mem_used += num * slotsize;
 821         spin_unlock(&nfsd_drc_lock);
 822
 823         return num;
 824 }
 825
 826 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
 827 {
 828         int slotsize = slot_bytes(ca);
 829
 830         spin_lock(&nfsd_drc_lock);
 831         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
 832         spin_unlock(&nfsd_drc_lock);
 833 }
 834
 835 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
 836                                            struct nfsd4_channel_attrs *battrs)
 837 {
 838         int numslots = fattrs->maxreqs;
 839         int slotsize = slot_bytes(fattrs);
 840         struct nfsd4_session *new;
 841         int mem, i;
 842
 843         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 844                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 845         mem = numslots * sizeof(struct nfsd4_slot *);
 846
 847         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 848         if (!new)
 849                 return NULL;
 850         /* allocate each struct nfsd4_slot and data cache in one piece */
 851         for (i = 0; i < numslots; i++) {
 852                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
 853                 if (!new->se_slots[i])
 854                         goto out_free;
 855         }
 856
 857         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
 858         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
 859
 860         return new;
 861 out_free:
 862         while (i--)
 863                 kfree(new->se_slots[i]);
 864         kfree(new);
 865         return NULL;
 866 }
 867
 868 static void free_conn(struct nfsd4_conn *c)
 869 {
 870         svc_xprt_put(c->cn_xprt);
 871         kfree(c);
 872 }
 873
 874 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 875 {
 876         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 877         struct nfs4_client *clp = c->cn_session->se_client;
 878
 879         spin_lock(&clp->cl_lock);
 880         if (!list_empty(&c->cn_persession)) {
 881                 list_del(&c->cn_persession);
 882                 free_conn(c);
 883         }
 884         nfsd4_probe_callback(clp);
 885         spin_unlock(&clp->cl_lock);
 886 }
 887
 888 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 889 {
 890         struct nfsd4_conn *conn;
 891
 892         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 893         if (!conn)
 894                 return NULL;
 895         svc_xprt_get(rqstp->rq_xprt);
 896         conn->cn_xprt = rqstp->rq_xprt;
 897         conn->cn_flags = flags;
 898         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 899         return conn;
 900 }
 901
 902 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 903 {
 904         conn->cn_session = ses;
 905         list_add(&conn->cn_persession, &ses->se_conns);
 906 }
 907
 908 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 909 {
 910         struct nfs4_client *clp = ses->se_client;
 911
 912         spin_lock(&clp->cl_lock);
 913         __nfsd4_hash_conn(conn, ses);
 914         spin_unlock(&clp->cl_lock);
 915 }
 916
 917 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 918 {
 919         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 920         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 921 }
 922
 923 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 924 {
 925         int ret;
 926
 927         nfsd4_hash_conn(conn, ses);
 928         ret = nfsd4_register_conn(conn);
 929         if (ret)
 930                 /* oops; xprt is already down: */
 931                 nfsd4_conn_lost(&conn->cn_xpt_user);
 932         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 933                 /* callback channel may be back up */
 934                 nfsd4_probe_callback(ses->se_client);
 935         }
 936 }
 937
 938 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 939 {
 940         u32 dir = NFS4_CDFC4_FORE;
 941
 942         if (cses->flags & SESSION4_BACK_CHAN)
 943                 dir |= NFS4_CDFC4_BACK;
 944         return alloc_conn(rqstp, dir);
 945 }
 946
 947 /* must be called under client_lock */
 948 static void nfsd4_del_conns(struct nfsd4_session *s)
 949 {
 950         struct nfs4_client *clp = s->se_client;
 951         struct nfsd4_conn *c;
 952
 953         spin_lock(&clp->cl_lock);
 954         while (!list_empty(&s->se_conns)) {
 955                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 956                 list_del_init(&c->cn_persession);
 957                 spin_unlock(&clp->cl_lock);
 958
 959                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 960                 free_conn(c);
 961
 962                 spin_lock(&clp->cl_lock);
 963         }
 964         spin_unlock(&clp->cl_lock);
 965 }
 966
 967 static void __free_session(struct nfsd4_session *ses)
 968 {
 969         free_session_slots(ses);
 970         kfree(ses);
 971 }
 972
 973 static void free_session(struct nfsd4_session *ses)
 974 {
 975         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
 976
 977         lockdep_assert_held(&nn->client_lock);
 978         nfsd4_del_conns(ses);
 979         nfsd4_put_drc_mem(&ses->se_fchannel);
 980         __free_session(ses);
 981 }
 982
 983 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 984 {
 985         int idx;
 986         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 987
 988         new->se_client = clp;
 989         gen_sessionid(new);
 990
 991         INIT_LIST_HEAD(&new->se_conns);
 992
 993         new->se_cb_seq_nr = 1;
 994         new->se_flags = cses->flags;
 995         new->se_cb_prog = cses->callback_prog;
 996         new->se_cb_sec = cses->cb_sec;
 997         atomic_set(&new->se_ref, 0);
 998         idx = hash_sessionid(&new->se_sessionid);
 999         spin_lock(&nn->client_lock);
1000         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1001         spin_lock(&clp->cl_lock);
1002         list_add(&new->se_perclnt, &clp->cl_sessions);
1003         spin_unlock(&clp->cl_lock);
1004         spin_unlock(&nn->client_lock);
1005
1006         if (cses->flags & SESSION4_BACK_CHAN) {
1007                 struct sockaddr *sa = svc_addr(rqstp);
1008                 /*
1009                  * This is a little silly; with sessions there's no real
1010                  * use for the callback address.  Use the peer address
1011                  * as a reasonable default for now, but consider fixing
1012                  * the rpc client not to require an address in the
1013                  * future:
1014                  */
1015                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1016                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1017         }
1018 }
1019
1020 /* caller must hold client_lock */
1021 static struct nfsd4_session *
1022 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1023 {
1024         struct nfsd4_session *elem;
1025         int idx;
1026         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1027
1028         dump_sessionid(__func__, sessionid);
1029         idx = hash_sessionid(sessionid);
1030         /* Search in the appropriate list */
1031         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1032                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1033                             NFS4_MAX_SESSIONID_LEN)) {
1034                         return elem;
1035                 }
1036         }
1037
1038         dprintk("%s: session not found\n", __func__);
1039         return NULL;
1040 }
1041
1042 /* caller must hold client_lock */
1043 static void
1044 unhash_session(struct nfsd4_session *ses)
1045 {
1046         list_del(&ses->se_hash);
1047         spin_lock(&ses->se_client->cl_lock);
1048         list_del(&ses->se_perclnt);
1049         spin_unlock(&ses->se_client->cl_lock);
1050 }
1051
1052 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1053 static int
1054 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1055 {
1056         if (clid->cl_boot == nn->boot_time)
1057                 return 0;
1058         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1059                 clid->cl_boot, clid->cl_id, nn->boot_time);
1060         return 1;
1061 }
1062
1063 /*
1064  * XXX Should we use a slab cache ?
1065  * This type of memory management is somewhat inefficient, but we use it
1066  * anyway since SETCLIENTID is not a common operation.
1067  */
1068 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1069 {
1070         struct nfs4_client *clp;
1071
1072         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1073         if (clp == NULL)
1074                 return NULL;
1075         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1076         if (clp->cl_name.data == NULL) {
1077                 kfree(clp);
1078                 return NULL;
1079         }
1080         clp->cl_name.len = name.len;
1081         return clp;
1082 }
1083
1084 static inline void
1085 free_client(struct nfs4_client *clp)
1086 {
1087         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1088
1089         lockdep_assert_held(&nn->client_lock);
1090         while (!list_empty(&clp->cl_sessions)) {
1091                 struct nfsd4_session *ses;
1092                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1093                                 se_perclnt);
1094                 list_del(&ses->se_perclnt);
1095                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1096                 free_session(ses);
1097         }
1098         free_svc_cred(&clp->cl_cred);
1099         kfree(clp->cl_name.data);
1100         idr_destroy(&clp->cl_stateids);
1101         kfree(clp);
1102 }
1103
1104 /* must be called under the client_lock */
1105 static inline void
1106 unhash_client_locked(struct nfs4_client *clp)
1107 {
1108         struct nfsd4_session *ses;
1109
1110         list_del(&clp->cl_lru);
1111         spin_lock(&clp->cl_lock);
1112         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1113                 list_del_init(&ses->se_hash);
1114         spin_unlock(&clp->cl_lock);
1115 }
1116
1117 static void
1118 destroy_client(struct nfs4_client *clp)
1119 {
1120         struct nfs4_openowner *oo;
1121         struct nfs4_delegation *dp;
1122         struct list_head reaplist;
1123         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1124
1125         INIT_LIST_HEAD(&reaplist);
1126         spin_lock(&recall_lock);
1127         while (!list_empty(&clp->cl_delegations)) {
1128                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1129                 list_del_init(&dp->dl_perclnt);
1130                 list_move(&dp->dl_recall_lru, &reaplist);
1131         }
1132         spin_unlock(&recall_lock);
1133         while (!list_empty(&reaplist)) {
1134                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1135                 destroy_delegation(dp);
1136         }
1137         list_splice_init(&clp->cl_revoked, &reaplist);
1138         while (!list_empty(&reaplist)) {
1139                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1140                 destroy_revoked_delegation(dp);
1141         }
1142         while (!list_empty(&clp->cl_openowners)) {
1143                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1144                 release_openowner(oo);
1145         }
1146         nfsd4_shutdown_callback(clp);
1147         if (clp->cl_cb_conn.cb_xprt)
1148                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1149         list_del(&clp->cl_idhash);
1150         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1151                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1152         else
1153                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1154         spin_lock(&nn->client_lock);
1155         unhash_client_locked(clp);
1156         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1157         free_client(clp);
1158         spin_unlock(&nn->client_lock);
1159 }
1160
1161 static void expire_client(struct nfs4_client *clp)
1162 {
1163         nfsd4_client_record_remove(clp);
1164         destroy_client(clp);
1165 }
1166
1167 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1168 {
1169         memcpy(target->cl_verifier.data, source->data,
1170                         sizeof(target->cl_verifier.data));
1171 }
1172
1173 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1174 {
1175         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1176         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1177 }
1178
1179 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1180 {
1181         if (source->cr_principal) {
1182                 target->cr_principal =
1183                                 kstrdup(source->cr_principal, GFP_KERNEL);
1184                 if (target->cr_principal == NULL)
1185                         return -ENOMEM;
1186         } else
1187                 target->cr_principal = NULL;
1188         target->cr_flavor = source->cr_flavor;
1189         target->cr_uid = source->cr_uid;
1190         target->cr_gid = source->cr_gid;
1191         target->cr_group_info = source->cr_group_info;
1192         get_group_info(target->cr_group_info);
1193         target->cr_gss_mech = source->cr_gss_mech;
1194         if (source->cr_gss_mech)
1195                 gss_mech_get(source->cr_gss_mech);
1196         return 0;
1197 }
1198
1199 static long long
1200 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1201 {
1202         long long res;
1203
1204         res = o1->len - o2->len;
1205         if (res)
1206                 return res;
1207         return (long long)memcmp(o1->data, o2->data, o1->len);
1208 }
1209
1210 static int same_name(const char *n1, const char *n2)
1211 {
1212         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1213 }
1214
1215 static int
1216 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1217 {
1218         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1219 }
1220
1221 static int
1222 same_clid(clientid_t *cl1, clientid_t *cl2)
1223 {
1224         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1225 }
1226
1227 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1228 {
1229         int i;
1230
1231         if (g1->ngroups != g2->ngroups)
1232                 return false;
1233         for (i=0; i<g1->ngroups; i++)
1234                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1235                         return false;
1236         return true;
1237 }
1238
1239 /*
1240  * RFC 3530 language requires clid_inuse be returned when the
1241  * "principal" associated with a requests differs from that previously
1242  * used.  We use uid, gid's, and gss principal string as our best
1243  * approximation.  We also don't want to allow non-gss use of a client
1244  * established using gss: in theory cr_principal should catch that
1245  * change, but in practice cr_principal can be null even in the gss case
1246  * since gssd doesn't always pass down a principal string.
1247  */
1248 static bool is_gss_cred(struct svc_cred *cr)
1249 {
1250         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1251         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1252 }
1253
1254
1255 static bool
1256 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1257 {
1258         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1259                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1260                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1261                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1262                 return false;
1263         if (cr1->cr_principal == cr2->cr_principal)
1264                 return true;
1265         if (!cr1->cr_principal || !cr2->cr_principal)
1266                 return false;
1267         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1268 }
1269
1270 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1271 {
1272         struct svc_cred *cr = &rqstp->rq_cred;
1273         u32 service;
1274
1275         if (!cr->cr_gss_mech)
1276                 return false;
1277         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1278         return service == RPC_GSS_SVC_INTEGRITY ||
1279                service == RPC_GSS_SVC_PRIVACY;
1280 }
1281
1282 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1283 {
1284         struct svc_cred *cr = &rqstp->rq_cred;
1285
1286         if (!cl->cl_mach_cred)
1287                 return true;
1288         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1289                 return false;
1290         if (!svc_rqst_integrity_protected(rqstp))
1291                 return false;
1292         if (!cr->cr_principal)
1293                 return false;
1294         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1295 }
1296
1297 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1298 {
1299         static u32 current_clientid = 1;
1300
1301         clp->cl_clientid.cl_boot = nn->boot_time;
1302         clp->cl_clientid.cl_id = current_clientid++;
1303 }
1304
1305 static void gen_confirm(struct nfs4_client *clp)
1306 {
1307         __be32 verf[2];
1308         static u32 i;
1309
1310         verf[0] = (__be32)get_seconds();
1311         verf[1] = (__be32)i++;
1312         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1313 }
1314
1315 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1316 {
1317         struct nfs4_stid *ret;
1318
1319         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1320         if (!ret || !ret->sc_type)
1321                 return NULL;
1322         return ret;
1323 }
1324
1325 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1326 {
1327         struct nfs4_stid *s;
1328
1329         s = find_stateid(cl, t);
1330         if (!s)
1331                 return NULL;
1332         if (typemask & s->sc_type)
1333                 return s;
1334         return NULL;
1335 }
1336
1337 static struct nfs4_client *create_client(struct xdr_netobj name,
1338                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1339 {
1340         struct nfs4_client *clp;
1341         struct sockaddr *sa = svc_addr(rqstp);
1342         int ret;
1343         struct net *net = SVC_NET(rqstp);
1344         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1345
1346         clp = alloc_client(name);
1347         if (clp == NULL)
1348                 return NULL;
1349
1350         INIT_LIST_HEAD(&clp->cl_sessions);
1351         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1352         if (ret) {
1353                 spin_lock(&nn->client_lock);
1354                 free_client(clp);
1355                 spin_unlock(&nn->client_lock);
1356                 return NULL;
1357         }
1358         idr_init(&clp->cl_stateids);
1359         atomic_set(&clp->cl_refcount, 0);
1360         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1361         INIT_LIST_HEAD(&clp->cl_idhash);
1362         INIT_LIST_HEAD(&clp->cl_openowners);
1363         INIT_LIST_HEAD(&clp->cl_delegations);
1364         INIT_LIST_HEAD(&clp->cl_lru);
1365         INIT_LIST_HEAD(&clp->cl_callbacks);
1366         INIT_LIST_HEAD(&clp->cl_revoked);
1367         spin_lock_init(&clp->cl_lock);
1368         nfsd4_init_callback(&clp->cl_cb_null);
1369         clp->cl_time = get_seconds();
1370         clear_bit(0, &clp->cl_cb_slot_busy);
1371         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1372         copy_verf(clp, verf);
1373         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1374         gen_confirm(clp);
1375         clp->cl_cb_session = NULL;
1376         clp->net = net;
1377         return clp;
1378 }
1379
1380 static void
1381 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1382 {
1383         struct rb_node **new = &(root->rb_node), *parent = NULL;
1384         struct nfs4_client *clp;
1385
1386         while (*new) {
1387                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1388                 parent = *new;
1389
1390                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1391                         new = &((*new)->rb_left);
1392                 else
1393                         new = &((*new)->rb_right);
1394         }
1395
1396         rb_link_node(&new_clp->cl_namenode, parent, new);
1397         rb_insert_color(&new_clp->cl_namenode, root);
1398 }
1399
1400 static struct nfs4_client *
1401 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1402 {
1403         long long cmp;
1404         struct rb_node *node = root->rb_node;
1405         struct nfs4_client *clp;
1406
1407         while (node) {
1408                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1409                 cmp = compare_blob(&clp->cl_name, name);
1410                 if (cmp > 0)
1411                         node = node->rb_left;
1412                 else if (cmp < 0)
1413                         node = node->rb_right;
1414                 else
1415                         return clp;
1416         }
1417         return NULL;
1418 }
1419
1420 static void
1421 add_to_unconfirmed(struct nfs4_client *clp)
1422 {
1423         unsigned int idhashval;
1424         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1425
1426         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1427         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1428         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1429         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1430         renew_client(clp);
1431 }
1432
1433 static void
1434 move_to_confirmed(struct nfs4_client *clp)
1435 {
1436         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1437         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1438
1439         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1440         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1441         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1442         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1443         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1444         renew_client(clp);
1445 }
1446
1447 static struct nfs4_client *
1448 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1449 {
1450         struct nfs4_client *clp;
1451         unsigned int idhashval = clientid_hashval(clid->cl_id);
1452
1453         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1454                 if (same_clid(&clp->cl_clientid, clid)) {
1455                         if ((bool)clp->cl_minorversion != sessions)
1456                                 return NULL;
1457                         renew_client(clp);
1458                         return clp;
1459                 }
1460         }
1461         return NULL;
1462 }
1463
1464 static struct nfs4_client *
1465 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1466 {
1467         struct list_head *tbl = nn->conf_id_hashtbl;
1468
1469         return find_client_in_id_table(tbl, clid, sessions);
1470 }
1471
1472 static struct nfs4_client *
1473 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1474 {
1475         struct list_head *tbl = nn->unconf_id_hashtbl;
1476
1477         return find_client_in_id_table(tbl, clid, sessions);
1478 }
1479
1480 static bool clp_used_exchangeid(struct nfs4_client *clp)
1481 {
1482         return clp->cl_exchange_flags != 0;
1483 }
1484
1485 static struct nfs4_client *
1486 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1487 {
1488         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1489 }
1490
1491 static struct nfs4_client *
1492 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1493 {
1494         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1495 }
1496
1497 static void
1498 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1499 {
1500         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1501         struct sockaddr *sa = svc_addr(rqstp);
1502         u32 scopeid = rpc_get_scope_id(sa);
1503         unsigned short expected_family;
1504
1505         /* Currently, we only support tcp and tcp6 for the callback channel */
1506         if (se->se_callback_netid_len == 3 &&
1507             !memcmp(se->se_callback_netid_val, "tcp", 3))
1508                 expected_family = AF_INET;
1509         else if (se->se_callback_netid_len == 4 &&
1510                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1511                 expected_family = AF_INET6;
1512         else
1513                 goto out_err;
1514
1515         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1516                                             se->se_callback_addr_len,
1517                                             (struct sockaddr *)&conn->cb_addr,
1518                                             sizeof(conn->cb_addr));
1519
1520         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1521                 goto out_err;
1522
1523         if (conn->cb_addr.ss_family == AF_INET6)
1524                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1525
1526         conn->cb_prog = se->se_callback_prog;
1527         conn->cb_ident = se->se_callback_ident;
1528         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1529         return;
1530 out_err:
1531         conn->cb_addr.ss_family = AF_UNSPEC;
1532         conn->cb_addrlen = 0;
1533         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1534                 "will not receive delegations\n",
1535                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1536
1537         return;
1538 }
1539
1540 /*
1541  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1542  */
1543 void
1544 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1545 {
1546         struct nfsd4_slot *slot = resp->cstate.slot;
1547         unsigned int base;
1548
1549         dprintk("--> %s slot %p\n", __func__, slot);
1550
1551         slot->sl_opcnt = resp->opcnt;
1552         slot->sl_status = resp->cstate.status;
1553
1554         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1555         if (nfsd4_not_cached(resp)) {
1556                 slot->sl_datalen = 0;
1557                 return;
1558         }
1559         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1560         base = (char *)resp->cstate.datap -
1561                                         (char *)resp->xbuf->head[0].iov_base;
1562         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1563                                     slot->sl_datalen))
1564                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1565         return;
1566 }
1567
1568 /*
1569  * Encode the replay sequence operation from the slot values.
1570  * If cachethis is FALSE encode the uncached rep error on the next
1571  * operation which sets resp->p and increments resp->opcnt for
1572  * nfs4svc_encode_compoundres.
1573  *
1574  */
1575 static __be32
1576 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1577                           struct nfsd4_compoundres *resp)
1578 {
1579         struct nfsd4_op *op;
1580         struct nfsd4_slot *slot = resp->cstate.slot;
1581
1582         /* Encode the replayed sequence operation */
1583         op = &args->ops[resp->opcnt - 1];
1584         nfsd4_encode_operation(resp, op);
1585
1586         /* Return nfserr_retry_uncached_rep in next operation. */
1587         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1588                 op = &args->ops[resp->opcnt++];
1589                 op->status = nfserr_retry_uncached_rep;
1590                 nfsd4_encode_operation(resp, op);
1591         }
1592         return op->status;
1593 }
1594
1595 /*
1596  * The sequence operation is not cached because we can use the slot and
1597  * session values.
1598  */
1599 __be32
1600 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1601                          struct nfsd4_sequence *seq)
1602 {
1603         struct nfsd4_slot *slot = resp->cstate.slot;
1604         __be32 status;
1605
1606         dprintk("--> %s slot %p\n", __func__, slot);
1607
1608         /* Either returns 0 or nfserr_retry_uncached */
1609         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1610         if (status == nfserr_retry_uncached_rep)
1611                 return status;
1612
1613         /* The sequence operation has been encoded, cstate->datap set. */
1614         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1615
1616         resp->opcnt = slot->sl_opcnt;
1617         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1618         status = slot->sl_status;
1619
1620         return status;
1621 }
1622
1623 /*
1624  * Set the exchange_id flags returned by the server.
1625  */
1626 static void
1627 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1628 {
1629         /* pNFS is not supported */
1630         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1631
1632         /* Referrals are supported, Migration is not. */
1633         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1634
1635         /* set the wire flags to return to client. */
1636         clid->flags = new->cl_exchange_flags;
1637 }
1638
1639 static bool client_has_state(struct nfs4_client *clp)
1640 {
1641         /*
1642          * Note clp->cl_openowners check isn't quite right: there's no
1643          * need to count owners without stateid's.
1644          *
1645          * Also note we should probably be using this in 4.0 case too.
1646          */
1647         return !list_empty(&clp->cl_openowners)
1648                 || !list_empty(&clp->cl_delegations)
1649                 || !list_empty(&clp->cl_sessions);
1650 }
1651
1652 __be32
1653 nfsd4_exchange_id(struct svc_rqst *rqstp,
1654                   struct nfsd4_compound_state *cstate,
1655                   struct nfsd4_exchange_id *exid)
1656 {
1657         struct nfs4_client *unconf, *conf, *new;
1658         __be32 status;
1659         char                    addr_str[INET6_ADDRSTRLEN];
1660         nfs4_verifier           verf = exid->verifier;
1661         struct sockaddr         *sa = svc_addr(rqstp);
1662         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1663         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1664
1665         rpc_ntop(sa, addr_str, sizeof(addr_str));
1666         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1667                 "ip_addr=%s flags %x, spa_how %d\n",
1668                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1669                 addr_str, exid->flags, exid->spa_how);
1670
1671         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1672                 return nfserr_inval;
1673
1674         switch (exid->spa_how) {
1675         case SP4_MACH_CRED:
1676                 if (!svc_rqst_integrity_protected(rqstp))
1677                         return nfserr_inval;
1678         case SP4_NONE:
1679                 break;
1680         default:                                /* checked by xdr code */
1681                 WARN_ON_ONCE(1);
1682         case SP4_SSV:
1683                 return nfserr_encr_alg_unsupp;
1684         }
1685
1686         /* Cases below refer to rfc 5661 section 18.35.4: */
1687         nfs4_lock_state();
1688         conf = find_confirmed_client_by_name(&exid->clname, nn);
1689         if (conf) {
1690                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1691                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1692
1693                 if (update) {
1694                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1695                                 status = nfserr_inval;
1696                                 goto out;
1697                         }
1698                         if (!mach_creds_match(conf, rqstp)) {
1699                                 status = nfserr_wrong_cred;
1700                                 goto out;
1701                         }
1702                         if (!creds_match) { /* case 9 */
1703                                 status = nfserr_perm;
1704                                 goto out;
1705                         }
1706                         if (!verfs_match) { /* case 8 */
1707                                 status = nfserr_not_same;
1708                                 goto out;
1709                         }
1710                         /* case 6 */
1711                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1712                         new = conf;
1713                         goto out_copy;
1714                 }
1715                 if (!creds_match) { /* case 3 */
1716                         if (client_has_state(conf)) {
1717                                 status = nfserr_clid_inuse;
1718                                 goto out;
1719                         }
1720                         expire_client(conf);
1721                         goto out_new;
1722                 }
1723                 if (verfs_match) { /* case 2 */
1724                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1725                         new = conf;
1726                         goto out_copy;
1727                 }
1728                 /* case 5, client reboot */
1729                 goto out_new;
1730         }
1731
1732         if (update) { /* case 7 */
1733                 status = nfserr_noent;
1734                 goto out;
1735         }
1736
1737         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1738         if (unconf) /* case 4, possible retry or client restart */
1739                 expire_client(unconf);
1740
1741         /* case 1 (normal case) */
1742 out_new:
1743         new = create_client(exid->clname, rqstp, &verf);
1744         if (new == NULL) {
1745                 status = nfserr_jukebox;
1746                 goto out;
1747         }
1748         new->cl_minorversion = cstate->minorversion;
1749         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
1750
1751         gen_clid(new, nn);
1752         add_to_unconfirmed(new);
1753 out_copy:
1754         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1755         exid->clientid.cl_id = new->cl_clientid.cl_id;
1756
1757         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1758         nfsd4_set_ex_flags(new, exid);
1759
1760         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1761                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1762         status = nfs_ok;
1763
1764 out:
1765         nfs4_unlock_state();
1766         return status;
1767 }
1768
1769 static __be32
1770 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1771 {
1772         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1773                 slot_seqid);
1774
1775         /* The slot is in use, and no response has been sent. */
1776         if (slot_inuse) {
1777                 if (seqid == slot_seqid)
1778                         return nfserr_jukebox;
1779                 else
1780                         return nfserr_seq_misordered;
1781         }
1782         /* Note unsigned 32-bit arithmetic handles wraparound: */
1783         if (likely(seqid == slot_seqid + 1))
1784                 return nfs_ok;
1785         if (seqid == slot_seqid)
1786                 return nfserr_replay_cache;
1787         return nfserr_seq_misordered;
1788 }
1789
1790 /*
1791  * Cache the create session result into the create session single DRC
1792  * slot cache by saving the xdr structure. sl_seqid has been set.
1793  * Do this for solo or embedded create session operations.
1794  */
1795 static void
1796 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1797                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1798 {
1799         slot->sl_status = nfserr;
1800         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1801 }
1802
1803 static __be32
1804 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1805                             struct nfsd4_clid_slot *slot)
1806 {
1807         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1808         return slot->sl_status;
1809 }
1810
1811 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1812                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1813                         1 +     /* MIN tag is length with zero, only length */ \
1814                         3 +     /* version, opcount, opcode */ \
1815                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1816                                 /* seqid, slotID, slotID, cache */ \
1817                         4 ) * sizeof(__be32))
1818
1819 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1820                         2 +     /* verifier: AUTH_NULL, length 0 */\
1821                         1 +     /* status */ \
1822                         1 +     /* MIN tag is length with zero, only length */ \
1823                         3 +     /* opcount, opcode, opstatus*/ \
1824                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1825                                 /* seqid, slotID, slotID, slotID, status */ \
1826                         5 ) * sizeof(__be32))
1827
1828 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1829 {
1830         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
1831
1832         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
1833                 return nfserr_toosmall;
1834         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
1835                 return nfserr_toosmall;
1836         ca->headerpadsz = 0;
1837         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
1838         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
1839         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
1840         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
1841                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
1842         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
1843         /*
1844          * Note decreasing slot size below client's request may make it
1845          * difficult for client to function correctly, whereas
1846          * decreasing the number of slots will (just?) affect
1847          * performance.  When short on memory we therefore prefer to
1848          * decrease number of slots instead of their size.  Clients that
1849          * request larger slots than they need will get poor results:
1850          */
1851         ca->maxreqs = nfsd4_get_drc_mem(ca);
1852         if (!ca->maxreqs)
1853                 return nfserr_jukebox;
1854
1855         return nfs_ok;
1856 }
1857
1858 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
1859                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
1860 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
1861                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
1862
1863 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
1864 {
1865         ca->headerpadsz = 0;
1866
1867         /*
1868          * These RPC_MAX_HEADER macros are overkill, especially since we
1869          * don't even do gss on the backchannel yet.  But this is still
1870          * less than 1k.  Tighten up this estimate in the unlikely event
1871          * it turns out to be a problem for some client:
1872          */
1873         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
1874                 return nfserr_toosmall;
1875         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
1876                 return nfserr_toosmall;
1877         ca->maxresp_cached = 0;
1878         if (ca->maxops < 2)
1879                 return nfserr_toosmall;
1880
1881         return nfs_ok;
1882 }
1883
1884 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
1885 {
1886         switch (cbs->flavor) {
1887         case RPC_AUTH_NULL:
1888         case RPC_AUTH_UNIX:
1889                 return nfs_ok;
1890         default:
1891                 /*
1892                  * GSS case: the spec doesn't allow us to return this
1893                  * error.  But it also doesn't allow us not to support
1894                  * GSS.
1895                  * I'd rather this fail hard than return some error the
1896                  * client might think it can already handle:
1897                  */
1898                 return nfserr_encr_alg_unsupp;
1899         }
1900 }
1901
1902 __be32
1903 nfsd4_create_session(struct svc_rqst *rqstp,
1904                      struct nfsd4_compound_state *cstate,
1905                      struct nfsd4_create_session *cr_ses)
1906 {
1907         struct sockaddr *sa = svc_addr(rqstp);
1908         struct nfs4_client *conf, *unconf;
1909         struct nfsd4_session *new;
1910         struct nfsd4_conn *conn;
1911         struct nfsd4_clid_slot *cs_slot = NULL;
1912         __be32 status = 0;
1913         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1914
1915         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1916                 return nfserr_inval;
1917         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
1918         if (status)
1919                 return status;
1920         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
1921         if (status)
1922                 return status;
1923         status = check_backchannel_attrs(&cr_ses->back_channel);
1924         if (status)
1925                 goto out_release_drc_mem;
1926         status = nfserr_jukebox;
1927         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
1928         if (!new)
1929                 goto out_release_drc_mem;
1930         conn = alloc_conn_from_crses(rqstp, cr_ses);
1931         if (!conn)
1932                 goto out_free_session;
1933
1934         nfs4_lock_state();
1935         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1936         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1937         WARN_ON_ONCE(conf && unconf);
1938
1939         if (conf) {
1940                 status = nfserr_wrong_cred;
1941                 if (!mach_creds_match(conf, rqstp))
1942                         goto out_free_conn;
1943                 cs_slot = &conf->cl_cs_slot;
1944                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1945                 if (status == nfserr_replay_cache) {
1946                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1947                         goto out_free_conn;
1948                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1949                         status = nfserr_seq_misordered;
1950                         goto out_free_conn;
1951                 }
1952         } else if (unconf) {
1953                 struct nfs4_client *old;
1954                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1955                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1956                         status = nfserr_clid_inuse;
1957                         goto out_free_conn;
1958                 }
1959                 status = nfserr_wrong_cred;
1960                 if (!mach_creds_match(unconf, rqstp))
1961                         goto out_free_conn;
1962                 cs_slot = &unconf->cl_cs_slot;
1963                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1964                 if (status) {
1965                         /* an unconfirmed replay returns misordered */
1966                         status = nfserr_seq_misordered;
1967                         goto out_free_conn;
1968                 }
1969                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1970                 if (old) {
1971                         status = mark_client_expired(old);
1972                         if (status)
1973                                 goto out_free_conn;
1974                         expire_client(old);
1975                 }
1976                 move_to_confirmed(unconf);
1977                 conf = unconf;
1978         } else {
1979                 status = nfserr_stale_clientid;
1980                 goto out_free_conn;
1981         }
1982         status = nfs_ok;
1983         /*
1984          * We do not support RDMA or persistent sessions
1985          */
1986         cr_ses->flags &= ~SESSION4_PERSIST;
1987         cr_ses->flags &= ~SESSION4_RDMA;
1988
1989         init_session(rqstp, new, conf, cr_ses);
1990         nfsd4_init_conn(rqstp, conn, new);
1991
1992         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1993                NFS4_MAX_SESSIONID_LEN);
1994         cs_slot->sl_seqid++;
1995         cr_ses->seqid = cs_slot->sl_seqid;
1996
1997         /* cache solo and embedded create sessions under the state lock */
1998         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1999         nfs4_unlock_state();
2000         return status;
2001 out_free_conn:
2002         nfs4_unlock_state();
2003         free_conn(conn);
2004 out_free_session:
2005         __free_session(new);
2006 out_release_drc_mem:
2007         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2008         return status;
2009 }
2010
2011 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2012 {
2013         switch (*dir) {
2014         case NFS4_CDFC4_FORE:
2015         case NFS4_CDFC4_BACK:
2016                 return nfs_ok;
2017         case NFS4_CDFC4_FORE_OR_BOTH:
2018         case NFS4_CDFC4_BACK_OR_BOTH:
2019                 *dir = NFS4_CDFC4_BOTH;
2020                 return nfs_ok;
2021         };
2022         return nfserr_inval;
2023 }
2024
2025 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2026 {
2027         struct nfsd4_session *session = cstate->session;
2028         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2029         __be32 status;
2030
2031         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2032         if (status)
2033                 return status;
2034         spin_lock(&nn->client_lock);
2035         session->se_cb_prog = bc->bc_cb_program;
2036         session->se_cb_sec = bc->bc_cb_sec;
2037         spin_unlock(&nn->client_lock);
2038
2039         nfsd4_probe_callback(session->se_client);
2040
2041         return nfs_ok;
2042 }
2043
2044 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2045                      struct nfsd4_compound_state *cstate,
2046                      struct nfsd4_bind_conn_to_session *bcts)
2047 {
2048         __be32 status;
2049         struct nfsd4_conn *conn;
2050         struct nfsd4_session *session;
2051         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2052
2053         if (!nfsd4_last_compound_op(rqstp))
2054                 return nfserr_not_only_op;
2055         nfs4_lock_state();
2056         spin_lock(&nn->client_lock);
2057         session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
2058         spin_unlock(&nn->client_lock);
2059         status = nfserr_badsession;
2060         if (!session)
2061                 goto out;
2062         status = nfserr_wrong_cred;
2063         if (!mach_creds_match(session->se_client, rqstp))
2064                 goto out;
2065         status = nfsd4_map_bcts_dir(&bcts->dir);
2066         if (status)
2067                 goto out;
2068         conn = alloc_conn(rqstp, bcts->dir);
2069         status = nfserr_jukebox;
2070         if (!conn)
2071                 goto out;
2072         nfsd4_init_conn(rqstp, conn, session);
2073         status = nfs_ok;
2074 out:
2075         nfs4_unlock_state();
2076         return status;
2077 }
2078
2079 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2080 {
2081         if (!session)
2082                 return 0;
2083         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2084 }
2085
2086 __be32
2087 nfsd4_destroy_session(struct svc_rqst *r,
2088                       struct nfsd4_compound_state *cstate,
2089                       struct nfsd4_destroy_session *sessionid)
2090 {
2091         struct nfsd4_session *ses;
2092         __be32 status;
2093         int ref_held_by_me = 0;
2094         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
2095
2096         nfs4_lock_state();
2097         status = nfserr_not_only_op;
2098         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2099                 if (!nfsd4_last_compound_op(r))
2100                         goto out;
2101                 ref_held_by_me++;
2102         }
2103         dump_sessionid(__func__, &sessionid->sessionid);
2104         spin_lock(&nn->client_lock);
2105         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
2106         status = nfserr_badsession;
2107         if (!ses)
2108                 goto out_client_lock;
2109         status = nfserr_wrong_cred;
2110         if (!mach_creds_match(ses->se_client, r))
2111                 goto out_client_lock;
2112         nfsd4_get_session_locked(ses);
2113         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2114         if (status)
2115                 goto out_put_session;
2116         unhash_session(ses);
2117         spin_unlock(&nn->client_lock);
2118
2119         nfsd4_probe_callback_sync(ses->se_client);
2120
2121         spin_lock(&nn->client_lock);
2122         status = nfs_ok;
2123 out_put_session:
2124         nfsd4_put_session(ses);
2125 out_client_lock:
2126         spin_unlock(&nn->client_lock);
2127 out:
2128         nfs4_unlock_state();
2129         return status;
2130 }
2131
2132 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2133 {
2134         struct nfsd4_conn *c;
2135
2136         list_for_each_entry(c, &s->se_conns, cn_persession) {
2137                 if (c->cn_xprt == xpt) {
2138                         return c;
2139                 }
2140         }
2141         return NULL;
2142 }
2143
2144 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2145 {
2146         struct nfs4_client *clp = ses->se_client;
2147         struct nfsd4_conn *c;
2148         __be32 status = nfs_ok;
2149         int ret;
2150
2151         spin_lock(&clp->cl_lock);
2152         c = __nfsd4_find_conn(new->cn_xprt, ses);
2153         if (c)
2154                 goto out_free;
2155         status = nfserr_conn_not_bound_to_session;
2156         if (clp->cl_mach_cred)
2157                 goto out_free;
2158         __nfsd4_hash_conn(new, ses);
2159         spin_unlock(&clp->cl_lock);
2160         ret = nfsd4_register_conn(new);
2161         if (ret)
2162                 /* oops; xprt is already down: */
2163                 nfsd4_conn_lost(&new->cn_xpt_user);
2164         return nfs_ok;
2165 out_free:
2166         spin_unlock(&clp->cl_lock);
2167         free_conn(new);
2168         return status;
2169 }
2170
2171 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2172 {
2173         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2174
2175         return args->opcnt > session->se_fchannel.maxops;
2176 }
2177
2178 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2179                                   struct nfsd4_session *session)
2180 {
2181         struct xdr_buf *xb = &rqstp->rq_arg;
2182
2183         return xb->len > session->se_fchannel.maxreq_sz;
2184 }
2185
2186 __be32
2187 nfsd4_sequence(struct svc_rqst *rqstp,
2188                struct nfsd4_compound_state *cstate,
2189                struct nfsd4_sequence *seq)
2190 {
2191         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2192         struct nfsd4_session *session;
2193         struct nfs4_client *clp;
2194         struct nfsd4_slot *slot;
2195         struct nfsd4_conn *conn;
2196         __be32 status;
2197         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2198
2199         if (resp->opcnt != 1)
2200                 return nfserr_sequence_pos;
2201
2202         /*
2203          * Will be either used or freed by nfsd4_sequence_check_conn
2204          * below.
2205          */
2206         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2207         if (!conn)
2208                 return nfserr_jukebox;
2209
2210         spin_lock(&nn->client_lock);
2211         status = nfserr_badsession;
2212         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2213         if (!session)
2214                 goto out_no_session;
2215         clp = session->se_client;
2216         status = get_client_locked(clp);
2217         if (status)
2218                 goto out_no_session;
2219         status = nfsd4_get_session_locked(session);
2220         if (status)
2221                 goto out_put_client;
2222
2223         status = nfserr_too_many_ops;
2224         if (nfsd4_session_too_many_ops(rqstp, session))
2225                 goto out_put_session;
2226
2227         status = nfserr_req_too_big;
2228         if (nfsd4_request_too_big(rqstp, session))
2229                 goto out_put_session;
2230
2231         status = nfserr_badslot;
2232         if (seq->slotid >= session->se_fchannel.maxreqs)
2233                 goto out_put_session;
2234
2235         slot = session->se_slots[seq->slotid];
2236         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2237
2238         /* We do not negotiate the number of slots yet, so set the
2239          * maxslots to the session maxreqs which is used to encode
2240          * sr_highest_slotid and the sr_target_slot id to maxslots */
2241         seq->maxslots = session->se_fchannel.maxreqs;
2242
2243         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2244                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2245         if (status == nfserr_replay_cache) {
2246                 status = nfserr_seq_misordered;
2247                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2248                         goto out_put_session;
2249                 cstate->slot = slot;
2250                 cstate->session = session;
2251                 /* Return the cached reply status and set cstate->status
2252                  * for nfsd4_proc_compound processing */
2253                 status = nfsd4_replay_cache_entry(resp, seq);
2254                 cstate->status = nfserr_replay_cache;
2255                 goto out;
2256         }
2257         if (status)
2258                 goto out_put_session;
2259
2260         status = nfsd4_sequence_check_conn(conn, session);
2261         conn = NULL;
2262         if (status)
2263                 goto out_put_session;
2264
2265         /* Success! bump slot seqid */
2266         slot->sl_seqid = seq->seqid;
2267         slot->sl_flags |= NFSD4_SLOT_INUSE;
2268         if (seq->cachethis)
2269                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2270         else
2271                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2272
2273         cstate->slot = slot;
2274         cstate->session = session;
2275
2276 out:
2277         switch (clp->cl_cb_state) {
2278         case NFSD4_CB_DOWN:
2279                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2280                 break;
2281         case NFSD4_CB_FAULT:
2282                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2283                 break;
2284         default:
2285                 seq->status_flags = 0;
2286         }
2287         if (!list_empty(&clp->cl_revoked))
2288                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2289 out_no_session:
2290         kfree(conn);
2291         spin_unlock(&nn->client_lock);
2292         return status;
2293 out_put_session:
2294         nfsd4_put_session(session);
2295 out_put_client:
2296         put_client_renew_locked(clp);
2297         goto out_no_session;
2298 }
2299
2300 __be32
2301 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2302 {
2303         struct nfs4_client *conf, *unconf, *clp;
2304         __be32 status = 0;
2305         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2306
2307         nfs4_lock_state();
2308         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2309         conf = find_confirmed_client(&dc->clientid, true, nn);
2310         WARN_ON_ONCE(conf && unconf);
2311
2312         if (conf) {
2313                 clp = conf;
2314
2315                 if (client_has_state(conf)) {
2316                         status = nfserr_clientid_busy;
2317                         goto out;
2318                 }
2319         } else if (unconf)
2320                 clp = unconf;
2321         else {
2322                 status = nfserr_stale_clientid;
2323                 goto out;
2324         }
2325         if (!mach_creds_match(clp, rqstp)) {
2326                 status = nfserr_wrong_cred;
2327                 goto out;
2328         }
2329         expire_client(clp);
2330 out:
2331         nfs4_unlock_state();
2332         return status;
2333 }
2334
2335 __be32
2336 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2337 {
2338         __be32 status = 0;
2339
2340         if (rc->rca_one_fs) {
2341                 if (!cstate->current_fh.fh_dentry)
2342                         return nfserr_nofilehandle;
2343                 /*
2344                  * We don't take advantage of the rca_one_fs case.
2345                  * That's OK, it's optional, we can safely ignore it.
2346                  */
2347                  return nfs_ok;
2348         }
2349
2350         nfs4_lock_state();
2351         status = nfserr_complete_already;
2352         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2353                              &cstate->session->se_client->cl_flags))
2354                 goto out;
2355
2356         status = nfserr_stale_clientid;
2357         if (is_client_expired(cstate->session->se_client))
2358                 /*
2359                  * The following error isn't really legal.
2360                  * But we only get here if the client just explicitly
2361                  * destroyed the client.  Surely it no longer cares what
2362                  * error it gets back on an operation for the dead
2363                  * client.
2364                  */
2365                 goto out;
2366
2367         status = nfs_ok;
2368         nfsd4_client_record_create(cstate->session->se_client);
2369 out:
2370         nfs4_unlock_state();
2371         return status;
2372 }
2373
2374 __be32
2375 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2376                   struct nfsd4_setclientid *setclid)
2377 {
2378         struct xdr_netobj       clname = setclid->se_name;
2379         nfs4_verifier           clverifier = setclid->se_verf;
2380         struct nfs4_client      *conf, *unconf, *new;
2381         __be32                  status;
2382         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2383
2384         /* Cases below refer to rfc 3530 section 14.2.33: */
2385         nfs4_lock_state();
2386         conf = find_confirmed_client_by_name(&clname, nn);
2387         if (conf) {
2388                 /* case 0: */
2389                 status = nfserr_clid_inuse;
2390                 if (clp_used_exchangeid(conf))
2391                         goto out;
2392                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2393                         char addr_str[INET6_ADDRSTRLEN];
2394                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2395                                  sizeof(addr_str));
2396                         dprintk("NFSD: setclientid: string in use by client "
2397                                 "at %s\n", addr_str);
2398                         goto out;
2399                 }
2400         }
2401         unconf = find_unconfirmed_client_by_name(&clname, nn);
2402         if (unconf)
2403                 expire_client(unconf);
2404         status = nfserr_jukebox;
2405         new = create_client(clname, rqstp, &clverifier);
2406         if (new == NULL)
2407                 goto out;
2408         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2409                 /* case 1: probable callback update */
2410                 copy_clid(new, conf);
2411         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2412                 gen_clid(new, nn);
2413         new->cl_minorversion = 0;
2414         gen_callback(new, setclid, rqstp);
2415         add_to_unconfirmed(new);
2416         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2417         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2418         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2419         status = nfs_ok;
2420 out:
2421         nfs4_unlock_state();
2422         return status;
2423 }
2424
2425
2426 __be32
2427 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2428                          struct nfsd4_compound_state *cstate,
2429                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2430 {
2431         struct nfs4_client *conf, *unconf;
2432         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2433         clientid_t * clid = &setclientid_confirm->sc_clientid;
2434         __be32 status;
2435         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2436
2437         if (STALE_CLIENTID(clid, nn))
2438                 return nfserr_stale_clientid;
2439         nfs4_lock_state();
2440
2441         conf = find_confirmed_client(clid, false, nn);
2442         unconf = find_unconfirmed_client(clid, false, nn);
2443         /*
2444          * We try hard to give out unique clientid's, so if we get an
2445          * attempt to confirm the same clientid with a different cred,
2446          * there's a bug somewhere.  Let's charitably assume it's our
2447          * bug.
2448          */
2449         status = nfserr_serverfault;
2450         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2451                 goto out;
2452         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2453                 goto out;
2454         /* cases below refer to rfc 3530 section 14.2.34: */
2455         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2456                 if (conf && !unconf) /* case 2: probable retransmit */
2457                         status = nfs_ok;
2458                 else /* case 4: client hasn't noticed we rebooted yet? */
2459                         status = nfserr_stale_clientid;
2460                 goto out;
2461         }
2462         status = nfs_ok;
2463         if (conf) { /* case 1: callback update */
2464                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2465                 nfsd4_probe_callback(conf);
2466                 expire_client(unconf);
2467         } else { /* case 3: normal case; new or rebooted client */
2468                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2469                 if (conf) {
2470                         status = mark_client_expired(conf);
2471                         if (status)
2472                                 goto out;
2473                         expire_client(conf);
2474                 }
2475                 move_to_confirmed(unconf);
2476                 nfsd4_probe_callback(unconf);
2477         }
2478 out:
2479         nfs4_unlock_state();
2480         return status;
2481 }
2482
2483 static struct nfs4_file *nfsd4_alloc_file(void)
2484 {
2485         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2486 }
2487
2488 /* OPEN Share state helper functions */
2489 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2490 {
2491         unsigned int hashval = file_hashval(ino);
2492
2493         atomic_set(&fp->fi_ref, 1);
2494         INIT_LIST_HEAD(&fp->fi_stateids);
2495         INIT_LIST_HEAD(&fp->fi_delegations);
2496         fp->fi_inode = igrab(ino);
2497         fp->fi_had_conflict = false;
2498         fp->fi_lease = NULL;
2499         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2500         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2501         spin_lock(&recall_lock);
2502         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2503         spin_unlock(&recall_lock);
2504 }
2505
2506 static void
2507 nfsd4_free_slab(struct kmem_cache **slab)
2508 {
2509         if (*slab == NULL)
2510                 return;
2511         kmem_cache_destroy(*slab);
2512         *slab = NULL;
2513 }
2514
2515 void
2516 nfsd4_free_slabs(void)
2517 {
2518         nfsd4_free_slab(&openowner_slab);
2519         nfsd4_free_slab(&lockowner_slab);
2520         nfsd4_free_slab(&file_slab);
2521         nfsd4_free_slab(&stateid_slab);
2522         nfsd4_free_slab(&deleg_slab);
2523 }
2524
2525 int
2526 nfsd4_init_slabs(void)
2527 {
2528         openowner_slab = kmem_cache_create("nfsd4_openowners",
2529                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2530         if (openowner_slab == NULL)
2531                 goto out_nomem;
2532         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2533                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2534         if (lockowner_slab == NULL)
2535                 goto out_nomem;
2536         file_slab = kmem_cache_create("nfsd4_files",
2537                         sizeof(struct nfs4_file), 0, 0, NULL);
2538         if (file_slab == NULL)
2539                 goto out_nomem;
2540         stateid_slab = kmem_cache_create("nfsd4_stateids",
2541                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2542         if (stateid_slab == NULL)
2543                 goto out_nomem;
2544         deleg_slab = kmem_cache_create("nfsd4_delegations",
2545                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2546         if (deleg_slab == NULL)
2547                 goto out_nomem;
2548         return 0;
2549 out_nomem:
2550         nfsd4_free_slabs();
2551         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2552         return -ENOMEM;
2553 }
2554
2555 void nfs4_free_openowner(struct nfs4_openowner *oo)
2556 {
2557         kfree(oo->oo_owner.so_owner.data);
2558         kmem_cache_free(openowner_slab, oo);
2559 }
2560
2561 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2562 {
2563         kfree(lo->lo_owner.so_owner.data);
2564         kmem_cache_free(lockowner_slab, lo);
2565 }
2566
2567 static void init_nfs4_replay(struct nfs4_replay *rp)
2568 {
2569         rp->rp_status = nfserr_serverfault;
2570         rp->rp_buflen = 0;
2571         rp->rp_buf = rp->rp_ibuf;
2572 }
2573
2574 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2575 {
2576         struct nfs4_stateowner *sop;
2577
2578         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2579         if (!sop)
2580                 return NULL;
2581
2582         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2583         if (!sop->so_owner.data) {
2584                 kmem_cache_free(slab, sop);
2585                 return NULL;
2586         }
2587         sop->so_owner.len = owner->len;
2588
2589         INIT_LIST_HEAD(&sop->so_stateids);
2590         sop->so_client = clp;
2591         init_nfs4_replay(&sop->so_replay);
2592         return sop;
2593 }
2594
2595 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2596 {
2597         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2598
2599         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2600         list_add(&oo->oo_perclient, &clp->cl_openowners);
2601 }
2602
2603 static struct nfs4_openowner *
2604 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2605         struct nfs4_openowner *oo;
2606
2607         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2608         if (!oo)
2609                 return NULL;
2610         oo->oo_owner.so_is_open_owner = 1;
2611         oo->oo_owner.so_seqid = open->op_seqid;
2612         oo->oo_flags = NFS4_OO_NEW;
2613         oo->oo_time = 0;
2614         oo->oo_last_closed_stid = NULL;
2615         INIT_LIST_HEAD(&oo->oo_close_lru);
2616         hash_openowner(oo, clp, strhashval);
2617         return oo;
2618 }
2619
2620 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2621         struct nfs4_openowner *oo = open->op_openowner;
2622
2623         stp->st_stid.sc_type = NFS4_OPEN_STID;
2624         INIT_LIST_HEAD(&stp->st_lockowners);
2625         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2626         list_add(&stp->st_perfile, &fp->fi_stateids);
2627         stp->st_stateowner = &oo->oo_owner;
2628         get_nfs4_file(fp);
2629         stp->st_file = fp;
2630         stp->st_access_bmap = 0;
2631         stp->st_deny_bmap = 0;
2632         set_access(open->op_share_access, stp);
2633         set_deny(open->op_share_deny, stp);
2634         stp->st_openstp = NULL;
2635 }
2636
2637 static void
2638 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2639 {
2640         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2641
2642         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2643
2644         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2645         oo->oo_time = get_seconds();
2646 }
2647
2648 static int
2649 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2650                                                         clientid_t *clid)
2651 {
2652         return (sop->so_owner.len == owner->len) &&
2653                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2654                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2655 }
2656
2657 static struct nfs4_openowner *
2658 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2659                         bool sessions, struct nfsd_net *nn)
2660 {
2661         struct nfs4_stateowner *so;
2662         struct nfs4_openowner *oo;
2663         struct nfs4_client *clp;
2664
2665         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2666                 if (!so->so_is_open_owner)
2667                         continue;
2668                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2669                         oo = openowner(so);
2670                         clp = oo->oo_owner.so_client;
2671                         if ((bool)clp->cl_minorversion != sessions)
2672                                 return NULL;
2673                         renew_client(oo->oo_owner.so_client);
2674                         return oo;
2675                 }
2676         }
2677         return NULL;
2678 }
2679
2680 /* search file_hashtbl[] for file */
2681 static struct nfs4_file *
2682 find_file(struct inode *ino)
2683 {
2684         unsigned int hashval = file_hashval(ino);
2685         struct nfs4_file *fp;
2686
2687         spin_lock(&recall_lock);
2688         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2689                 if (fp->fi_inode == ino) {
2690                         get_nfs4_file(fp);
2691                         spin_unlock(&recall_lock);
2692                         return fp;
2693                 }
2694         }
2695         spin_unlock(&recall_lock);
2696         return NULL;
2697 }
2698
2699 /*
2700  * Called to check deny when READ with all zero stateid or
2701  * WRITE with all zero or all one stateid
2702  */
2703 static __be32
2704 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2705 {
2706         struct inode *ino = current_fh->fh_dentry->d_inode;
2707         struct nfs4_file *fp;
2708         struct nfs4_ol_stateid *stp;
2709         __be32 ret;
2710
2711         fp = find_file(ino);
2712         if (!fp)
2713                 return nfs_ok;
2714         ret = nfserr_locked;
2715         /* Search for conflicting share reservations */
2716         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2717                 if (test_deny(deny_type, stp) ||
2718                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2719                         goto out;
2720         }
2721         ret = nfs_ok;
2722 out:
2723         put_nfs4_file(fp);
2724         return ret;
2725 }
2726
2727 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2728 {
2729         struct nfs4_client *clp = dp->dl_stid.sc_client;
2730         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2731
2732         /* We're assuming the state code never drops its reference
2733          * without first removing the lease.  Since we're in this lease
2734          * callback (and since the lease code is serialized by the kernel
2735          * lock) we know the server hasn't removed the lease yet, we know
2736          * it's safe to take a reference: */
2737         atomic_inc(&dp->dl_count);
2738
2739         list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
2740
2741         /* Only place dl_time is set; protected by i_lock: */
2742         dp->dl_time = get_seconds();
2743
2744         nfsd4_cb_recall(dp);
2745 }
2746
2747 /* Called from break_lease() with i_lock held. */
2748 static void nfsd_break_deleg_cb(struct file_lock *fl)
2749 {
2750         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2751         struct nfs4_delegation *dp;
2752
2753         if (!fp) {
2754                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2755                 return;
2756         }
2757         if (fp->fi_had_conflict) {
2758                 WARN(1, "duplicate break on %p\n", fp);
2759                 return;
2760         }
2761         /*
2762          * We don't want the locks code to timeout the lease for us;
2763          * we'll remove it ourself if a delegation isn't returned
2764          * in time:
2765          */
2766         fl->fl_break_time = 0;
2767
2768         spin_lock(&recall_lock);
2769         fp->fi_had_conflict = true;
2770         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2771                 nfsd_break_one_deleg(dp);
2772         spin_unlock(&recall_lock);
2773 }
2774
2775 static
2776 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2777 {
2778         if (arg & F_UNLCK)
2779                 return lease_modify(onlist, arg);
2780         else
2781                 return -EAGAIN;
2782 }
2783
2784 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2785         .lm_break = nfsd_break_deleg_cb,
2786         .lm_change = nfsd_change_deleg_cb,
2787 };
2788
2789 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2790 {
2791         if (nfsd4_has_session(cstate))
2792                 return nfs_ok;
2793         if (seqid == so->so_seqid - 1)
2794                 return nfserr_replay_me;
2795         if (seqid == so->so_seqid)
2796                 return nfs_ok;
2797         return nfserr_bad_seqid;
2798 }
2799
2800 __be32
2801 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2802                     struct nfsd4_open *open, struct nfsd_net *nn)
2803 {
2804         clientid_t *clientid = &open->op_clientid;
2805         struct nfs4_client *clp = NULL;
2806         unsigned int strhashval;
2807         struct nfs4_openowner *oo = NULL;
2808         __be32 status;
2809
2810         if (STALE_CLIENTID(&open->op_clientid, nn))
2811                 return nfserr_stale_clientid;
2812         /*
2813          * In case we need it later, after we've already created the
2814          * file and don't want to risk a further failure:
2815          */
2816         open->op_file = nfsd4_alloc_file();
2817         if (open->op_file == NULL)
2818                 return nfserr_jukebox;
2819
2820         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2821         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2822         open->op_openowner = oo;
2823         if (!oo) {
2824                 clp = find_confirmed_client(clientid, cstate->minorversion,
2825                                             nn);
2826                 if (clp == NULL)
2827                         return nfserr_expired;
2828                 goto new_owner;
2829         }
2830         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2831                 /* Replace unconfirmed owners without checking for replay. */
2832                 clp = oo->oo_owner.so_client;
2833                 release_openowner(oo);
2834                 open->op_openowner = NULL;
2835                 goto new_owner;
2836         }
2837         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2838         if (status)
2839                 return status;
2840         clp = oo->oo_owner.so_client;
2841         goto alloc_stateid;
2842 new_owner:
2843         oo = alloc_init_open_stateowner(strhashval, clp, open);
2844         if (oo == NULL)
2845                 return nfserr_jukebox;
2846         open->op_openowner = oo;
2847 alloc_stateid:
2848         open->op_stp = nfs4_alloc_stateid(clp);
2849         if (!open->op_stp)
2850                 return nfserr_jukebox;
2851         return nfs_ok;
2852 }
2853
2854 static inline __be32
2855 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2856 {
2857         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2858                 return nfserr_openmode;
2859         else
2860                 return nfs_ok;
2861 }
2862
2863 static int share_access_to_flags(u32 share_access)
2864 {
2865         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2866 }
2867
2868 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2869 {
2870         struct nfs4_stid *ret;
2871
2872         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2873         if (!ret)
2874                 return NULL;
2875         return delegstateid(ret);
2876 }
2877
2878 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2879 {
2880         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2881                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2882 }
2883
2884 static __be32
2885 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
2886                 struct nfs4_delegation **dp)
2887 {
2888         int flags;
2889         __be32 status = nfserr_bad_stateid;
2890
2891         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2892         if (*dp == NULL)
2893                 goto out;
2894         flags = share_access_to_flags(open->op_share_access);
2895         status = nfs4_check_delegmode(*dp, flags);
2896         if (status)
2897                 *dp = NULL;
2898 out:
2899         if (!nfsd4_is_deleg_cur(open))
2900                 return nfs_ok;
2901         if (status)
2902                 return status;
2903         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2904         return nfs_ok;
2905 }
2906
2907 static __be32
2908 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2909 {
2910         struct nfs4_ol_stateid *local;
2911         struct nfs4_openowner *oo = open->op_openowner;
2912
2913         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2914                 /* ignore lock owners */
2915                 if (local->st_stateowner->so_is_open_owner == 0)
2916                         continue;
2917                 /* remember if we have seen this open owner */
2918                 if (local->st_stateowner == &oo->oo_owner)
2919                         *stpp = local;
2920                 /* check for conflicting share reservations */
2921                 if (!test_share(local, open))
2922                         return nfserr_share_denied;
2923         }
2924         return nfs_ok;
2925 }
2926
2927 static inline int nfs4_access_to_access(u32 nfs4_access)
2928 {
2929         int flags = 0;
2930
2931         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2932                 flags |= NFSD_MAY_READ;
2933         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2934                 flags |= NFSD_MAY_WRITE;
2935         return flags;
2936 }
2937
2938 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2939                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2940 {
2941         __be32 status;
2942         int oflag = nfs4_access_to_omode(open->op_share_access);
2943         int access = nfs4_access_to_access(open->op_share_access);
2944
2945         if (!fp->fi_fds[oflag]) {
2946                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2947                         &fp->fi_fds[oflag]);
2948                 if (status)
2949                         return status;
2950         }
2951         nfs4_file_get_access(fp, oflag);
2952
2953         return nfs_ok;
2954 }
2955
2956 static inline __be32
2957 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2958                 struct nfsd4_open *open)
2959 {
2960         struct iattr iattr = {
2961                 .ia_valid = ATTR_SIZE,
2962                 .ia_size = 0,
2963         };
2964         if (!open->op_truncate)
2965                 return 0;
2966         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2967                 return nfserr_inval;
2968         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2969 }
2970
2971 static __be32
2972 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)
2973 {
2974         u32 op_share_access = open->op_share_access;
2975         bool new_access;
2976         __be32 status;
2977
2978         new_access = !test_access(op_share_access, stp);
2979         if (new_access) {
2980                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2981                 if (status)
2982                         return status;
2983         }
2984         status = nfsd4_truncate(rqstp, cur_fh, open);
2985         if (status) {
2986                 if (new_access) {
2987                         int oflag = nfs4_access_to_omode(op_share_access);
2988                         nfs4_file_put_access(fp, oflag);
2989                 }
2990                 return status;
2991         }
2992         /* remember the open */
2993         set_access(op_share_access, stp);
2994         set_deny(open->op_share_deny, stp);
2995
2996         return nfs_ok;
2997 }
2998
2999
3000 static void
3001 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3002 {
3003         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3004 }
3005
3006 /* Should we give out recallable state?: */
3007 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3008 {
3009         if (clp->cl_cb_state == NFSD4_CB_UP)
3010                 return true;
3011         /*
3012          * In the sessions case, since we don't have to establish a
3013          * separate connection for callbacks, we assume it's OK
3014          * until we hear otherwise:
3015          */
3016         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3017 }
3018
3019 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
3020 {
3021         struct file_lock *fl;
3022
3023         fl = locks_alloc_lock();
3024         if (!fl)
3025                 return NULL;
3026         locks_init_lock(fl);
3027         fl->fl_lmops = &nfsd_lease_mng_ops;
3028         fl->fl_flags = FL_DELEG;
3029         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3030         fl->fl_end = OFFSET_MAX;
3031         fl->fl_owner = (fl_owner_t)(dp->dl_file);
3032         fl->fl_pid = current->tgid;
3033         return fl;
3034 }
3035
3036 static int nfs4_setlease(struct nfs4_delegation *dp)
3037 {
3038         struct nfs4_file *fp = dp->dl_file;
3039         struct file_lock *fl;
3040         int status;
3041
3042         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
3043         if (!fl)
3044                 return -ENOMEM;
3045         fl->fl_file = find_readable_file(fp);
3046         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
3047         if (status)
3048                 goto out_free;
3049         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3050         fp->fi_lease = fl;
3051         fp->fi_deleg_file = get_file(fl->fl_file);
3052         atomic_set(&fp->fi_delegees, 1);
3053         list_add(&dp->dl_perfile, &fp->fi_delegations);
3054         return 0;
3055 out_free:
3056         locks_free_lock(fl);
3057         return status;
3058 }
3059
3060 static int nfs4_set_delegation(struct nfs4_delegation *dp, struct nfs4_file *fp)
3061 {
3062         int status;
3063
3064         if (fp->fi_had_conflict)
3065                 return -EAGAIN;
3066         get_nfs4_file(fp);
3067         dp->dl_file = fp;
3068         if (!fp->fi_lease) {
3069                 status = nfs4_setlease(dp);
3070                 if (status)
3071                         goto out_free;
3072                 return 0;
3073         }
3074         spin_lock(&recall_lock);
3075         if (fp->fi_had_conflict) {
3076                 spin_unlock(&recall_lock);
3077                 status = -EAGAIN;
3078                 goto out_free;
3079         }
3080         atomic_inc(&fp->fi_delegees);
3081         list_add(&dp->dl_perfile, &fp->fi_delegations);
3082         spin_unlock(&recall_lock);
3083         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3084         return 0;
3085 out_free:
3086         put_nfs4_file(fp);
3087         dp->dl_file = fp;
3088         return status;
3089 }
3090
3091 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3092 {
3093         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3094         if (status == -EAGAIN)
3095                 open->op_why_no_deleg = WND4_CONTENTION;
3096         else {
3097                 open->op_why_no_deleg = WND4_RESOURCE;
3098                 switch (open->op_deleg_want) {
3099                 case NFS4_SHARE_WANT_READ_DELEG:
3100                 case NFS4_SHARE_WANT_WRITE_DELEG:
3101                 case NFS4_SHARE_WANT_ANY_DELEG:
3102                         break;
3103                 case NFS4_SHARE_WANT_CANCEL:
3104                         open->op_why_no_deleg = WND4_CANCELLED;
3105                         break;
3106                 case NFS4_SHARE_WANT_NO_DELEG:
3107                         WARN_ON_ONCE(1);
3108                 }
3109         }
3110 }
3111
3112 /*
3113  * Attempt to hand out a delegation.
3114  *
3115  * Note we don't support write delegations, and won't until the vfs has
3116  * proper support for them.
3117  */
3118 static void
3119 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
3120                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
3121 {
3122         struct nfs4_delegation *dp;
3123         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
3124         int cb_up;
3125         int status = 0;
3126
3127         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3128         open->op_recall = 0;
3129         switch (open->op_claim_type) {
3130                 case NFS4_OPEN_CLAIM_PREVIOUS:
3131                         if (!cb_up)
3132                                 open->op_recall = 1;
3133                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3134                                 goto out_no_deleg;
3135                         break;
3136                 case NFS4_OPEN_CLAIM_NULL:
3137                 case NFS4_OPEN_CLAIM_FH:
3138                         /*
3139                          * Let's not give out any delegations till everyone's
3140                          * had the chance to reclaim theirs....
3141                          */
3142                         if (locks_in_grace(net))
3143                                 goto out_no_deleg;
3144                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3145                                 goto out_no_deleg;
3146                         /*
3147                          * Also, if the file was opened for write or
3148                          * create, there's a good chance the client's
3149                          * about to write to it, resulting in an
3150                          * immediate recall (since we don't support
3151                          * write delegations):
3152                          */
3153                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3154                                 goto out_no_deleg;
3155                         if (open->op_create == NFS4_OPEN_CREATE)
3156                                 goto out_no_deleg;
3157                         break;
3158                 default:
3159                         goto out_no_deleg;
3160         }
3161         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh);
3162         if (dp == NULL)
3163                 goto out_no_deleg;
3164         status = nfs4_set_delegation(dp, stp->st_file);
3165         if (status)
3166                 goto out_free;
3167
3168         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3169
3170         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3171                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3172         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3173         return;
3174 out_free:
3175         remove_stid(&dp->dl_stid);
3176         nfs4_put_delegation(dp);
3177 out_no_deleg:
3178         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3179         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3180             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3181                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3182                 open->op_recall = 1;
3183         }
3184
3185         /* 4.1 client asking for a delegation? */
3186         if (open->op_deleg_want)
3187                 nfsd4_open_deleg_none_ext(open, status);
3188         return;
3189 }
3190
3191 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3192                                         struct nfs4_delegation *dp)
3193 {
3194         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3195             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3196                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3197                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3198         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3199                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3200                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3201                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3202         }
3203         /* Otherwise the client must be confused wanting a delegation
3204          * it already has, therefore we don't return
3205          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3206          */
3207 }
3208
3209 /*
3210  * called with nfs4_lock_state() held.
3211  */
3212 __be32
3213 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3214 {
3215         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3216         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3217         struct nfs4_file *fp = NULL;
3218         struct inode *ino = current_fh->fh_dentry->d_inode;
3219         struct nfs4_ol_stateid *stp = NULL;
3220         struct nfs4_delegation *dp = NULL;
3221         __be32 status;
3222
3223         /*
3224          * Lookup file; if found, lookup stateid and check open request,
3225          * and check for delegations in the process of being recalled.
3226          * If not found, create the nfs4_file struct
3227          */
3228         fp = find_file(ino);
3229         if (fp) {
3230                 if ((status = nfs4_check_open(fp, open, &stp)))
3231                         goto out;
3232                 status = nfs4_check_deleg(cl, open, &dp);
3233                 if (status)
3234                         goto out;
3235         } else {
3236                 status = nfserr_bad_stateid;
3237                 if (nfsd4_is_deleg_cur(open))
3238                         goto out;
3239                 status = nfserr_jukebox;
3240                 fp = open->op_file;
3241                 open->op_file = NULL;
3242                 nfsd4_init_file(fp, ino);
3243         }
3244
3245         /*
3246          * OPEN the file, or upgrade an existing OPEN.
3247          * If truncate fails, the OPEN fails.
3248          */
3249         if (stp) {
3250                 /* Stateid was found, this is an OPEN upgrade */
3251                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3252                 if (status)
3253                         goto out;
3254         } else {
3255                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3256                 if (status)
3257                         goto out;
3258                 status = nfsd4_truncate(rqstp, current_fh, open);
3259                 if (status)
3260                         goto out;
3261                 stp = open->op_stp;
3262                 open->op_stp = NULL;
3263                 init_open_stateid(stp, fp, open);
3264         }
3265         update_stateid(&stp->st_stid.sc_stateid);
3266         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3267
3268         if (nfsd4_has_session(&resp->cstate)) {
3269                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3270
3271                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3272                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3273                         open->op_why_no_deleg = WND4_NOT_WANTED;
3274                         goto nodeleg;
3275                 }
3276         }
3277
3278         /*
3279         * Attempt to hand out a delegation. No error return, because the
3280         * OPEN succeeds even if we fail.
3281         */
3282         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3283 nodeleg:
3284         status = nfs_ok;
3285
3286         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3287                 STATEID_VAL(&stp->st_stid.sc_stateid));
3288 out:
3289         /* 4.1 client trying to upgrade/downgrade delegation? */
3290         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3291             open->op_deleg_want)
3292                 nfsd4_deleg_xgrade_none_ext(open, dp);
3293
3294         if (fp)
3295                 put_nfs4_file(fp);
3296         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3297                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3298         /*
3299         * To finish the open response, we just need to set the rflags.
3300         */
3301         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3302         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3303             !nfsd4_has_session(&resp->cstate))
3304                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3305
3306         return status;
3307 }
3308
3309 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3310 {
3311         if (open->op_openowner) {
3312                 struct nfs4_openowner *oo = open->op_openowner;
3313
3314                 if (!list_empty(&oo->oo_owner.so_stateids))
3315                         list_del_init(&oo->oo_close_lru);
3316                 if (oo->oo_flags & NFS4_OO_NEW) {
3317                         if (status) {
3318                                 release_openowner(oo);
3319                                 open->op_openowner = NULL;
3320                         } else
3321                                 oo->oo_flags &= ~NFS4_OO_NEW;
3322                 }
3323         }
3324         if (open->op_file)
3325                 nfsd4_free_file(open->op_file);
3326         if (open->op_stp)
3327                 free_generic_stateid(open->op_stp);
3328 }
3329
3330 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3331 {
3332         struct nfs4_client *found;
3333
3334         if (STALE_CLIENTID(clid, nn))
3335                 return nfserr_stale_clientid;
3336         found = find_confirmed_client(clid, session, nn);
3337         if (clp)
3338                 *clp = found;
3339         return found ? nfs_ok : nfserr_expired;
3340 }
3341
3342 __be32
3343 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3344             clientid_t *clid)
3345 {
3346         struct nfs4_client *clp;
3347         __be32 status;
3348         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3349
3350         nfs4_lock_state();
3351         dprintk("process_renew(%08x/%08x): starting\n",
3352                         clid->cl_boot, clid->cl_id);
3353         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3354         if (status)
3355                 goto out;
3356         status = nfserr_cb_path_down;
3357         if (!list_empty(&clp->cl_delegations)
3358                         && clp->cl_cb_state != NFSD4_CB_UP)
3359                 goto out;
3360         status = nfs_ok;
3361 out:
3362         nfs4_unlock_state();
3363         return status;
3364 }
3365
3366 static void
3367 nfsd4_end_grace(struct nfsd_net *nn)
3368 {
3369         /* do nothing if grace period already ended */
3370         if (nn->grace_ended)
3371                 return;
3372
3373         dprintk("NFSD: end of grace period\n");
3374         nn->grace_ended = true;
3375         nfsd4_record_grace_done(nn, nn->boot_time);
3376         locks_end_grace(&nn->nfsd4_manager);
3377         /*
3378          * Now that every NFSv4 client has had the chance to recover and
3379          * to see the (possibly new, possibly shorter) lease time, we
3380          * can safely set the next grace time to the current lease time:
3381          */
3382         nn->nfsd4_grace = nn->nfsd4_lease;
3383 }
3384
3385 static time_t
3386 nfs4_laundromat(struct nfsd_net *nn)
3387 {
3388         struct nfs4_client *clp;
3389         struct nfs4_openowner *oo;
3390         struct nfs4_delegation *dp;
3391         struct list_head *pos, *next, reaplist;
3392         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3393         time_t t, clientid_val = nn->nfsd4_lease;
3394         time_t u, test_val = nn->nfsd4_lease;
3395
3396         nfs4_lock_state();
3397
3398         dprintk("NFSD: laundromat service - starting\n");
3399         nfsd4_end_grace(nn);
3400         INIT_LIST_HEAD(&reaplist);
3401         spin_lock(&nn->client_lock);
3402         list_for_each_safe(pos, next, &nn->client_lru) {
3403                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3404                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3405                         t = clp->cl_time - cutoff;
3406                         if (clientid_val > t)
3407                                 clientid_val = t;
3408                         break;
3409                 }
3410                 if (mark_client_expired_locked(clp)) {
3411                         dprintk("NFSD: client in use (clientid %08x)\n",
3412                                 clp->cl_clientid.cl_id);
3413                         continue;
3414                 }
3415                 list_move(&clp->cl_lru, &reaplist);
3416         }
3417         spin_unlock(&nn->client_lock);
3418         list_for_each_safe(pos, next, &reaplist) {
3419                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3420                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3421                         clp->cl_clientid.cl_id);
3422                 expire_client(clp);
3423         }
3424         spin_lock(&recall_lock);
3425         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3426                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3427                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3428                         continue;
3429                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3430                         u = dp->dl_time - cutoff;
3431                         if (test_val > u)
3432                                 test_val = u;
3433                         break;
3434                 }
3435                 list_move(&dp->dl_recall_lru, &reaplist);
3436         }
3437         spin_unlock(&recall_lock);
3438         list_for_each_safe(pos, next, &reaplist) {
3439                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3440                 revoke_delegation(dp);
3441         }
3442         test_val = nn->nfsd4_lease;
3443         list_for_each_safe(pos, next, &nn->close_lru) {
3444                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3445                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3446                         u = oo->oo_time - cutoff;
3447                         if (test_val > u)
3448                                 test_val = u;
3449                         break;
3450                 }
3451                 release_openowner(oo);
3452         }
3453         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3454                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3455         nfs4_unlock_state();
3456         return clientid_val;
3457 }
3458
3459 static struct workqueue_struct *laundry_wq;
3460 static void laundromat_main(struct work_struct *);
3461
3462 static void
3463 laundromat_main(struct work_struct *laundry)
3464 {
3465         time_t t;
3466         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3467                                                   work);
3468         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3469                                            laundromat_work);
3470
3471         t = nfs4_laundromat(nn);
3472         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3473         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3474 }
3475
3476 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3477 {
3478         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3479                 return nfserr_bad_stateid;
3480         return nfs_ok;
3481 }
3482
3483 static inline int
3484 access_permit_read(struct nfs4_ol_stateid *stp)
3485 {
3486         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3487                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3488                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3489 }
3490
3491 static inline int
3492 access_permit_write(struct nfs4_ol_stateid *stp)
3493 {
3494         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3495                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3496 }
3497
3498 static
3499 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3500 {
3501         __be32 status = nfserr_openmode;
3502
3503         /* For lock stateid's, we test the parent open, not the lock: */
3504         if (stp->st_openstp)
3505                 stp = stp->st_openstp;
3506         if ((flags & WR_STATE) && !access_permit_write(stp))
3507                 goto out;
3508         if ((flags & RD_STATE) && !access_permit_read(stp))
3509                 goto out;
3510         status = nfs_ok;
3511 out:
3512         return status;
3513 }
3514
3515 static inline __be32
3516 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3517 {
3518         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3519                 return nfs_ok;
3520         else if (locks_in_grace(net)) {
3521                 /* Answer in remaining cases depends on existence of
3522                  * conflicting state; so we must wait out the grace period. */
3523                 return nfserr_grace;
3524         } else if (flags & WR_STATE)
3525                 return nfs4_share_conflict(current_fh,
3526                                 NFS4_SHARE_DENY_WRITE);
3527         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3528                 return nfs4_share_conflict(current_fh,
3529                                 NFS4_SHARE_DENY_READ);
3530 }
3531
3532 /*
3533  * Allow READ/WRITE during grace period on recovered state only for files
3534  * that are not able to provide mandatory locking.
3535  */
3536 static inline int
3537 grace_disallows_io(struct net *net, struct inode *inode)
3538 {
3539         return locks_in_grace(net) && mandatory_lock(inode);
3540 }
3541
3542 /* Returns true iff a is later than b: */
3543 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3544 {
3545         return (s32)(a->si_generation - b->si_generation) > 0;
3546 }
3547
3548 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3549 {
3550         /*
3551          * When sessions are used the stateid generation number is ignored
3552          * when it is zero.
3553          */
3554         if (has_session && in->si_generation == 0)
3555                 return nfs_ok;
3556
3557         if (in->si_generation == ref->si_generation)
3558                 return nfs_ok;
3559
3560         /* If the client sends us a stateid from the future, it's buggy: */
3561         if (stateid_generation_after(in, ref))
3562                 return nfserr_bad_stateid;
3563         /*
3564          * However, we could see a stateid from the past, even from a
3565          * non-buggy client.  For example, if the client sends a lock
3566          * while some IO is outstanding, the lock may bump si_generation
3567          * while the IO is still in flight.  The client could avoid that
3568          * situation by waiting for responses on all the IO requests,
3569          * but better performance may result in retrying IO that
3570          * receives an old_stateid error if requests are rarely
3571          * reordered in flight:
3572          */
3573         return nfserr_old_stateid;
3574 }
3575
3576 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3577 {
3578         struct nfs4_stid *s;
3579         struct nfs4_ol_stateid *ols;
3580         __be32 status;
3581
3582         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3583                 return nfserr_bad_stateid;
3584         /* Client debugging aid. */
3585         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3586                 char addr_str[INET6_ADDRSTRLEN];
3587                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3588                                  sizeof(addr_str));
3589                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3590                                         "with incorrect client ID\n", addr_str);
3591                 return nfserr_bad_stateid;
3592         }
3593         s = find_stateid(cl, stateid);
3594         if (!s)
3595                 return nfserr_bad_stateid;
3596         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3597         if (status)
3598                 return status;
3599         switch (s->sc_type) {
3600         case NFS4_DELEG_STID:
3601                 return nfs_ok;
3602         case NFS4_REVOKED_DELEG_STID:
3603                 return nfserr_deleg_revoked;
3604         case NFS4_OPEN_STID:
3605         case NFS4_LOCK_STID:
3606                 ols = openlockstateid(s);
3607                 if (ols->st_stateowner->so_is_open_owner
3608                                 && !(openowner(ols->st_stateowner)->oo_flags
3609                                                 & NFS4_OO_CONFIRMED))
3610                         return nfserr_bad_stateid;
3611                 return nfs_ok;
3612         default:
3613                 printk("unknown stateid type %x\n", s->sc_type);
3614         case NFS4_CLOSED_STID:
3615                 return nfserr_bad_stateid;
3616         }
3617 }
3618
3619 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3620                                    struct nfs4_stid **s, bool sessions,
3621                                    struct nfsd_net *nn)
3622 {
3623         struct nfs4_client *cl;
3624         __be32 status;
3625
3626         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3627                 return nfserr_bad_stateid;
3628         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3629                                                         nn, &cl);
3630         if (status == nfserr_stale_clientid)
3631                 return nfserr_stale_stateid;
3632         if (status)
3633                 return status;
3634         *s = find_stateid_by_type(cl, stateid, typemask);
3635         if (!*s)
3636                 return nfserr_bad_stateid;
3637         return nfs_ok;
3638 }
3639
3640 /*
3641 * Checks for stateid operations
3642 */
3643 __be32
3644 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3645                            stateid_t *stateid, int flags, struct file **filpp)
3646 {
3647         struct nfs4_stid *s;
3648         struct nfs4_ol_stateid *stp = NULL;
3649         struct nfs4_delegation *dp = NULL;
3650         struct svc_fh *current_fh = &cstate->current_fh;
3651         struct inode *ino = current_fh->fh_dentry->d_inode;
3652         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3653         __be32 status;
3654
3655         if (filpp)
3656                 *filpp = NULL;
3657
3658         if (grace_disallows_io(net, ino))
3659                 return nfserr_grace;
3660
3661         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3662                 return check_special_stateids(net, current_fh, stateid, flags);
3663
3664         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3665                                       &s, cstate->minorversion, nn);
3666         if (status)
3667                 return status;
3668         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3669         if (status)
3670                 goto out;
3671         switch (s->sc_type) {
3672         case NFS4_DELEG_STID:
3673                 dp = delegstateid(s);
3674                 status = nfs4_check_delegmode(dp, flags);
3675                 if (status)
3676                         goto out;
3677                 if (filpp) {
3678                         *filpp = dp->dl_file->fi_deleg_file;
3679                         if (!*filpp) {
3680                                 WARN_ON_ONCE(1);
3681                                 status = nfserr_serverfault;
3682                                 goto out;
3683                         }
3684                 }
3685                 break;
3686         case NFS4_OPEN_STID:
3687         case NFS4_LOCK_STID:
3688                 stp = openlockstateid(s);
3689                 status = nfs4_check_fh(current_fh, stp);
3690                 if (status)
3691                         goto out;
3692                 if (stp->st_stateowner->so_is_open_owner
3693                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3694                         goto out;
3695                 status = nfs4_check_openmode(stp, flags);
3696                 if (status)
3697                         goto out;
3698                 if (filpp) {
3699                         if (flags & RD_STATE)
3700                                 *filpp = find_readable_file(stp->st_file);
3701                         else
3702                                 *filpp = find_writeable_file(stp->st_file);
3703                 }
3704                 break;
3705         default:
3706                 return nfserr_bad_stateid;
3707         }
3708         status = nfs_ok;
3709 out:
3710         return status;
3711 }
3712
3713 static __be32
3714 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3715 {
3716         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3717                 return nfserr_locks_held;
3718         release_lock_stateid(stp);
3719         return nfs_ok;
3720 }
3721
3722 /*
3723  * Test if the stateid is valid
3724  */
3725 __be32
3726 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3727                    struct nfsd4_test_stateid *test_stateid)
3728 {
3729         struct nfsd4_test_stateid_id *stateid;
3730         struct nfs4_client *cl = cstate->session->se_client;
3731
3732         nfs4_lock_state();
3733         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3734                 stateid->ts_id_status =
3735                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3736         nfs4_unlock_state();
3737
3738         return nfs_ok;
3739 }
3740
3741 __be32
3742 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3743                    struct nfsd4_free_stateid *free_stateid)
3744 {
3745         stateid_t *stateid = &free_stateid->fr_stateid;
3746         struct nfs4_stid *s;
3747         struct nfs4_delegation *dp;
3748         struct nfs4_client *cl = cstate->session->se_client;
3749         __be32 ret = nfserr_bad_stateid;
3750
3751         nfs4_lock_state();
3752         s = find_stateid(cl, stateid);
3753         if (!s)
3754                 goto out;
3755         switch (s->sc_type) {
3756         case NFS4_DELEG_STID:
3757                 ret = nfserr_locks_held;
3758                 goto out;
3759         case NFS4_OPEN_STID:
3760         case NFS4_LOCK_STID:
3761                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3762                 if (ret)
3763                         goto out;
3764                 if (s->sc_type == NFS4_LOCK_STID)
3765                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3766                 else
3767                         ret = nfserr_locks_held;
3768                 break;
3769         case NFS4_REVOKED_DELEG_STID:
3770                 dp = delegstateid(s);
3771                 destroy_revoked_delegation(dp);
3772                 ret = nfs_ok;
3773                 break;
3774         default:
3775                 ret = nfserr_bad_stateid;
3776         }
3777 out:
3778         nfs4_unlock_state();
3779         return ret;
3780 }
3781
3782 static inline int
3783 setlkflg (int type)
3784 {
3785         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3786                 RD_STATE : WR_STATE;
3787 }
3788
3789 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3790 {
3791         struct svc_fh *current_fh = &cstate->current_fh;
3792         struct nfs4_stateowner *sop = stp->st_stateowner;
3793         __be32 status;
3794
3795         status = nfsd4_check_seqid(cstate, sop, seqid);
3796         if (status)
3797                 return status;
3798         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
3799                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
3800                 /*
3801                  * "Closed" stateid's exist *only* to return
3802                  * nfserr_replay_me from the previous step, and
3803                  * revoked delegations are kept only for free_stateid.
3804                  */
3805                 return nfserr_bad_stateid;
3806         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3807         if (status)
3808                 return status;
3809         return nfs4_check_fh(current_fh, stp);
3810 }
3811
3812 /*
3813  * Checks for sequence id mutating operations.
3814  */
3815 static __be32
3816 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3817                          stateid_t *stateid, char typemask,
3818                          struct nfs4_ol_stateid **stpp,
3819                          struct nfsd_net *nn)
3820 {
3821         __be32 status;
3822         struct nfs4_stid *s;
3823
3824         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3825                 seqid, STATEID_VAL(stateid));
3826
3827         *stpp = NULL;
3828         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3829                                       cstate->minorversion, nn);
3830         if (status)
3831                 return status;
3832         *stpp = openlockstateid(s);
3833         if (!nfsd4_has_session(cstate))
3834                 cstate->replay_owner = (*stpp)->st_stateowner;
3835
3836         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3837 }
3838
3839 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3840                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3841 {
3842         __be32 status;
3843         struct nfs4_openowner *oo;
3844
3845         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3846                                                 NFS4_OPEN_STID, stpp, nn);
3847         if (status)
3848                 return status;
3849         oo = openowner((*stpp)->st_stateowner);
3850         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3851                 return nfserr_bad_stateid;
3852         return nfs_ok;
3853 }
3854
3855 __be32
3856 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3857                    struct nfsd4_open_confirm *oc)
3858 {
3859         __be32 status;
3860         struct nfs4_openowner *oo;
3861         struct nfs4_ol_stateid *stp;
3862         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3863
3864         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
3865                         cstate->current_fh.fh_dentry);
3866
3867         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3868         if (status)
3869                 return status;
3870
3871         nfs4_lock_state();
3872
3873         status = nfs4_preprocess_seqid_op(cstate,
3874                                         oc->oc_seqid, &oc->oc_req_stateid,
3875                                         NFS4_OPEN_STID, &stp, nn);
3876         if (status)
3877                 goto out;
3878         oo = openowner(stp->st_stateowner);
3879         status = nfserr_bad_stateid;
3880         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3881                 goto out;
3882         oo->oo_flags |= NFS4_OO_CONFIRMED;
3883         update_stateid(&stp->st_stid.sc_stateid);
3884         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3885         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3886                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3887
3888         nfsd4_client_record_create(oo->oo_owner.so_client);
3889         status = nfs_ok;
3890 out:
3891         nfsd4_bump_seqid(cstate, status);
3892         if (!cstate->replay_owner)
3893                 nfs4_unlock_state();
3894         return status;
3895 }
3896
3897 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3898 {
3899         if (!test_access(access, stp))
3900                 return;
3901         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3902         clear_access(access, stp);
3903 }
3904
3905 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3906 {
3907         switch (to_access) {
3908         case NFS4_SHARE_ACCESS_READ:
3909                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3910                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3911                 break;
3912         case NFS4_SHARE_ACCESS_WRITE:
3913                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3914                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3915                 break;
3916         case NFS4_SHARE_ACCESS_BOTH:
3917                 break;
3918         default:
3919                 WARN_ON_ONCE(1);
3920         }
3921 }
3922
3923 static void
3924 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3925 {
3926         int i;
3927         for (i = 0; i < 4; i++) {
3928                 if ((i & deny) != i)
3929                         clear_deny(i, stp);
3930         }
3931 }
3932
3933 __be32
3934 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3935                      struct nfsd4_compound_state *cstate,
3936                      struct nfsd4_open_downgrade *od)
3937 {
3938         __be32 status;
3939         struct nfs4_ol_stateid *stp;
3940         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3941
3942         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
3943                         cstate->current_fh.fh_dentry);
3944
3945         /* We don't yet support WANT bits: */
3946         if (od->od_deleg_want)
3947                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3948                         od->od_deleg_want);
3949
3950         nfs4_lock_state();
3951         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3952                                         &od->od_stateid, &stp, nn);
3953         if (status)
3954                 goto out;
3955         status = nfserr_inval;
3956         if (!test_access(od->od_share_access, stp)) {
3957                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3958                         stp->st_access_bmap, od->od_share_access);
3959                 goto out;
3960         }
3961         if (!test_deny(od->od_share_deny, stp)) {
3962                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3963                         stp->st_deny_bmap, od->od_share_deny);
3964                 goto out;
3965         }
3966         nfs4_stateid_downgrade(stp, od->od_share_access);
3967
3968         reset_union_bmap_deny(od->od_share_deny, stp);
3969
3970         update_stateid(&stp->st_stid.sc_stateid);
3971         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3972         status = nfs_ok;
3973 out:
3974         nfsd4_bump_seqid(cstate, status);
3975         if (!cstate->replay_owner)
3976                 nfs4_unlock_state();
3977         return status;
3978 }
3979
3980 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3981 {
3982         unhash_open_stateid(s);
3983         s->st_stid.sc_type = NFS4_CLOSED_STID;
3984 }
3985
3986 /*
3987  * nfs4_unlock_state() called after encode
3988  */
3989 __be32
3990 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3991             struct nfsd4_close *close)
3992 {
3993         __be32 status;
3994         struct nfs4_openowner *oo;
3995         struct nfs4_ol_stateid *stp;
3996         struct net *net = SVC_NET(rqstp);
3997         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3998
3999         dprintk("NFSD: nfsd4_close on file %pd\n",
4000                         cstate->current_fh.fh_dentry);
4001
4002         nfs4_lock_state();
4003         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4004                                         &close->cl_stateid,
4005                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4006                                         &stp, nn);
4007         nfsd4_bump_seqid(cstate, status);
4008         if (status)
4009                 goto out;
4010         oo = openowner(stp->st_stateowner);
4011         update_stateid(&stp->st_stid.sc_stateid);
4012         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4013
4014         nfsd4_close_open_stateid(stp);
4015
4016         if (cstate->minorversion)
4017                 free_generic_stateid(stp);
4018         else
4019                 oo->oo_last_closed_stid = stp;
4020
4021         if (list_empty(&oo->oo_owner.so_stateids)) {
4022                 if (cstate->minorversion)
4023                         release_openowner(oo);
4024                 else {
4025                         /*
4026                          * In the 4.0 case we need to keep the owners around a
4027                          * little while to handle CLOSE replay.
4028                          */
4029                         move_to_close_lru(oo, SVC_NET(rqstp));
4030                 }
4031         }
4032 out:
4033         if (!cstate->replay_owner)
4034                 nfs4_unlock_state();
4035         return status;
4036 }
4037
4038 __be32
4039 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4040                   struct nfsd4_delegreturn *dr)
4041 {
4042         struct nfs4_delegation *dp;
4043         stateid_t *stateid = &dr->dr_stateid;
4044         struct nfs4_stid *s;
4045         __be32 status;
4046         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4047
4048         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4049                 return status;
4050
4051         nfs4_lock_state();
4052         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
4053                                       cstate->minorversion, nn);
4054         if (status)
4055                 goto out;
4056         dp = delegstateid(s);
4057         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4058         if (status)
4059                 goto out;
4060
4061         destroy_delegation(dp);
4062 out:
4063         nfs4_unlock_state();
4064
4065         return status;
4066 }
4067
4068
4069 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4070
4071 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
4072
4073 static inline u64
4074 end_offset(u64 start, u64 len)
4075 {
4076         u64 end;
4077
4078         end = start + len;
4079         return end >= start ? end: NFS4_MAX_UINT64;
4080 }
4081
4082 /* last octet in a range */
4083 static inline u64
4084 last_byte_offset(u64 start, u64 len)
4085 {
4086         u64 end;
4087
4088         WARN_ON_ONCE(!len);
4089         end = start + len;
4090         return end > start ? end - 1: NFS4_MAX_UINT64;
4091 }
4092
4093 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
4094 {
4095         return (file_hashval(inode) + cl_id
4096                         + opaque_hashval(ownername->data, ownername->len))
4097                 & LOCKOWNER_INO_HASH_MASK;
4098 }
4099
4100 /*
4101  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4102  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4103  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4104  * locking, this prevents us from being completely protocol-compliant.  The
4105  * real solution to this problem is to start using unsigned file offsets in
4106  * the VFS, but this is a very deep change!
4107  */
4108 static inline void
4109 nfs4_transform_lock_offset(struct file_lock *lock)
4110 {
4111         if (lock->fl_start < 0)
4112                 lock->fl_start = OFFSET_MAX;
4113         if (lock->fl_end < 0)
4114                 lock->fl_end = OFFSET_MAX;
4115 }
4116
4117 /* Hack!: For now, we're defining this just so we can use a pointer to it
4118  * as a unique cookie to identify our (NFSv4's) posix locks. */
4119 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4120 };
4121
4122 static inline void
4123 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4124 {
4125         struct nfs4_lockowner *lo;
4126
4127         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4128                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4129                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4130                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4131                 if (!deny->ld_owner.data)
4132                         /* We just don't care that much */
4133                         goto nevermind;
4134                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4135                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4136         } else {
4137 nevermind:
4138                 deny->ld_owner.len = 0;
4139                 deny->ld_owner.data = NULL;
4140                 deny->ld_clientid.cl_boot = 0;
4141                 deny->ld_clientid.cl_id = 0;
4142         }
4143         deny->ld_start = fl->fl_start;
4144         deny->ld_length = NFS4_MAX_UINT64;
4145         if (fl->fl_end != NFS4_MAX_UINT64)
4146                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4147         deny->ld_type = NFS4_READ_LT;
4148         if (fl->fl_type != F_RDLCK)
4149                 deny->ld_type = NFS4_WRITE_LT;
4150 }
4151
4152 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
4153 {
4154         struct nfs4_ol_stateid *lst;
4155
4156         if (!same_owner_str(&lo->lo_owner, owner, clid))
4157                 return false;
4158         lst = list_first_entry(&lo->lo_owner.so_stateids,
4159                                struct nfs4_ol_stateid, st_perstateowner);
4160         return lst->st_file->fi_inode == inode;
4161 }
4162
4163 static struct nfs4_lockowner *
4164 find_lockowner_str(struct inode *inode, clientid_t *clid,
4165                    struct xdr_netobj *owner, struct nfsd_net *nn)
4166 {
4167         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4168         struct nfs4_lockowner *lo;
4169
4170         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4171                 if (same_lockowner_ino(lo, inode, clid, owner))
4172                         return lo;
4173         }
4174         return NULL;
4175 }
4176
4177 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4178 {
4179         struct inode *inode = open_stp->st_file->fi_inode;
4180         unsigned int inohash = lockowner_ino_hashval(inode,
4181                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4182         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4183
4184         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4185         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4186         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4187 }
4188
4189 /*
4190  * Alloc a lock owner structure.
4191  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4192  * occurred.
4193  *
4194  * strhashval = ownerstr_hashval
4195  */
4196
4197 static struct nfs4_lockowner *
4198 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4199         struct nfs4_lockowner *lo;
4200
4201         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4202         if (!lo)
4203                 return NULL;
4204         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4205         lo->lo_owner.so_is_open_owner = 0;
4206         /* It is the openowner seqid that will be incremented in encode in the
4207          * case of new lockowners; so increment the lock seqid manually: */
4208         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4209         hash_lockowner(lo, strhashval, clp, open_stp);
4210         return lo;
4211 }
4212
4213 static struct nfs4_ol_stateid *
4214 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4215 {
4216         struct nfs4_ol_stateid *stp;
4217         struct nfs4_client *clp = lo->lo_owner.so_client;
4218
4219         stp = nfs4_alloc_stateid(clp);
4220         if (stp == NULL)
4221                 return NULL;
4222         stp->st_stid.sc_type = NFS4_LOCK_STID;
4223         list_add(&stp->st_perfile, &fp->fi_stateids);
4224         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4225         stp->st_stateowner = &lo->lo_owner;
4226         get_nfs4_file(fp);
4227         stp->st_file = fp;
4228         stp->st_access_bmap = 0;
4229         stp->st_deny_bmap = open_stp->st_deny_bmap;
4230         stp->st_openstp = open_stp;
4231         return stp;
4232 }
4233
4234 static int
4235 check_lock_length(u64 offset, u64 length)
4236 {
4237         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4238              LOFF_OVERFLOW(offset, length)));
4239 }
4240
4241 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4242 {
4243         struct nfs4_file *fp = lock_stp->st_file;
4244         int oflag = nfs4_access_to_omode(access);
4245
4246         if (test_access(access, lock_stp))
4247                 return;
4248         nfs4_file_get_access(fp, oflag);
4249         set_access(access, lock_stp);
4250 }
4251
4252 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)
4253 {
4254         struct nfs4_file *fi = ost->st_file;
4255         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4256         struct nfs4_client *cl = oo->oo_owner.so_client;
4257         struct nfs4_lockowner *lo;
4258         unsigned int strhashval;
4259         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4260
4261         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4262                                 &lock->v.new.owner, nn);
4263         if (lo) {
4264                 if (!cstate->minorversion)
4265                         return nfserr_bad_seqid;
4266                 /* XXX: a lockowner always has exactly one stateid: */
4267                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4268                                 struct nfs4_ol_stateid, st_perstateowner);
4269                 return nfs_ok;
4270         }
4271         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4272                         &lock->v.new.owner);
4273         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4274         if (lo == NULL)
4275                 return nfserr_jukebox;
4276         *lst = alloc_init_lock_stateid(lo, fi, ost);
4277         if (*lst == NULL) {
4278                 release_lockowner(lo);
4279                 return nfserr_jukebox;
4280         }
4281         *new = true;
4282         return nfs_ok;
4283 }
4284
4285 /*
4286  *  LOCK operation
4287  */
4288 __be32
4289 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4290            struct nfsd4_lock *lock)
4291 {
4292         struct nfs4_openowner *open_sop = NULL;
4293         struct nfs4_lockowner *lock_sop = NULL;
4294         struct nfs4_ol_stateid *lock_stp;
4295         struct file *filp = NULL;
4296         struct file_lock *file_lock = NULL;
4297         struct file_lock *conflock = NULL;
4298         __be32 status = 0;
4299         bool new_state = false;
4300         int lkflg;
4301         int err;
4302         struct net *net = SVC_NET(rqstp);
4303         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4304
4305         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4306                 (long long) lock->lk_offset,
4307                 (long long) lock->lk_length);
4308
4309         if (check_lock_length(lock->lk_offset, lock->lk_length))
4310                  return nfserr_inval;
4311
4312         if ((status = fh_verify(rqstp, &cstate->current_fh,
4313                                 S_IFREG, NFSD_MAY_LOCK))) {
4314                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4315                 return status;
4316         }
4317
4318         nfs4_lock_state();
4319
4320         if (lock->lk_is_new) {
4321                 struct nfs4_ol_stateid *open_stp = NULL;
4322
4323                 if (nfsd4_has_session(cstate))
4324                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4325                         memcpy(&lock->v.new.clientid,
4326                                 &cstate->session->se_client->cl_clientid,
4327                                 sizeof(clientid_t));
4328
4329                 status = nfserr_stale_clientid;
4330                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4331                         goto out;
4332
4333                 /* validate and update open stateid and open seqid */
4334                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4335                                         lock->lk_new_open_seqid,
4336                                         &lock->lk_new_open_stateid,
4337                                         &open_stp, nn);
4338                 if (status)
4339                         goto out;
4340                 open_sop = openowner(open_stp->st_stateowner);
4341                 status = nfserr_bad_stateid;
4342                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4343                                                 &lock->v.new.clientid))
4344                         goto out;
4345                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4346                                                         &lock_stp, &new_state);
4347         } else
4348                 status = nfs4_preprocess_seqid_op(cstate,
4349                                        lock->lk_old_lock_seqid,
4350                                        &lock->lk_old_lock_stateid,
4351                                        NFS4_LOCK_STID, &lock_stp, nn);
4352         if (status)
4353                 goto out;
4354         lock_sop = lockowner(lock_stp->st_stateowner);
4355
4356         lkflg = setlkflg(lock->lk_type);
4357         status = nfs4_check_openmode(lock_stp, lkflg);
4358         if (status)
4359                 goto out;
4360
4361         status = nfserr_grace;
4362         if (locks_in_grace(net) && !lock->lk_reclaim)
4363                 goto out;
4364         status = nfserr_no_grace;
4365         if (!locks_in_grace(net) && lock->lk_reclaim)
4366                 goto out;
4367
4368         file_lock = locks_alloc_lock();
4369         if (!file_lock) {
4370                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4371                 status = nfserr_jukebox;
4372                 goto out;
4373         }
4374
4375         locks_init_lock(file_lock);
4376         switch (lock->lk_type) {
4377                 case NFS4_READ_LT:
4378                 case NFS4_READW_LT:
4379                         filp = find_readable_file(lock_stp->st_file);
4380                         if (filp)
4381                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4382                         file_lock->fl_type = F_RDLCK;
4383                         break;
4384                 case NFS4_WRITE_LT:
4385                 case NFS4_WRITEW_LT:
4386                         filp = find_writeable_file(lock_stp->st_file);
4387                         if (filp)
4388                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4389                         file_lock->fl_type = F_WRLCK;
4390                         break;
4391                 default:
4392                         status = nfserr_inval;
4393                 goto out;
4394         }
4395         if (!filp) {
4396                 status = nfserr_openmode;
4397                 goto out;
4398         }
4399         file_lock->fl_owner = (fl_owner_t)lock_sop;
4400         file_lock->fl_pid = current->tgid;
4401         file_lock->fl_file = filp;
4402         file_lock->fl_flags = FL_POSIX;
4403         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4404         file_lock->fl_start = lock->lk_offset;
4405         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4406         nfs4_transform_lock_offset(file_lock);
4407
4408         conflock = locks_alloc_lock();
4409         if (!conflock) {
4410                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4411                 status = nfserr_jukebox;
4412                 goto out;
4413         }
4414
4415         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4416         switch (-err) {
4417         case 0: /* success! */
4418                 update_stateid(&lock_stp->st_stid.sc_stateid);
4419                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4420                                 sizeof(stateid_t));
4421                 status = 0;
4422                 break;
4423         case (EAGAIN):          /* conflock holds conflicting lock */
4424                 status = nfserr_denied;
4425                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4426                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4427                 break;
4428         case (EDEADLK):
4429                 status = nfserr_deadlock;
4430                 break;
4431         default:
4432                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4433                 status = nfserrno(err);
4434                 break;
4435         }
4436 out:
4437         if (status && new_state)
4438                 release_lockowner(lock_sop);
4439         nfsd4_bump_seqid(cstate, status);
4440         if (!cstate->replay_owner)
4441                 nfs4_unlock_state();
4442         if (file_lock)
4443                 locks_free_lock(file_lock);
4444         if (conflock)
4445                 locks_free_lock(conflock);
4446         return status;
4447 }
4448
4449 /*
4450  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4451  * so we do a temporary open here just to get an open file to pass to
4452  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4453  * inode operation.)
4454  */
4455 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4456 {
4457         struct file *file;
4458         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4459         if (!err) {
4460                 err = nfserrno(vfs_test_lock(file, lock));
4461                 nfsd_close(file);
4462         }
4463         return err;
4464 }
4465
4466 /*
4467  * LOCKT operation
4468  */
4469 __be32
4470 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4471             struct nfsd4_lockt *lockt)
4472 {
4473         struct inode *inode;
4474         struct file_lock *file_lock = NULL;
4475         struct nfs4_lockowner *lo;
4476         __be32 status;
4477         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4478
4479         if (locks_in_grace(SVC_NET(rqstp)))
4480                 return nfserr_grace;
4481
4482         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4483                  return nfserr_inval;
4484
4485         nfs4_lock_state();
4486
4487         if (!nfsd4_has_session(cstate)) {
4488                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4489                 if (status)
4490                         goto out;
4491         }
4492
4493         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4494                 goto out;
4495
4496         inode = cstate->current_fh.fh_dentry->d_inode;
4497         file_lock = locks_alloc_lock();
4498         if (!file_lock) {
4499                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4500                 status = nfserr_jukebox;
4501                 goto out;
4502         }
4503         locks_init_lock(file_lock);
4504         switch (lockt->lt_type) {
4505                 case NFS4_READ_LT:
4506                 case NFS4_READW_LT:
4507                         file_lock->fl_type = F_RDLCK;
4508                 break;
4509                 case NFS4_WRITE_LT:
4510                 case NFS4_WRITEW_LT:
4511                         file_lock->fl_type = F_WRLCK;
4512                 break;
4513                 default:
4514                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4515                         status = nfserr_inval;
4516                 goto out;
4517         }
4518
4519         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4520         if (lo)
4521                 file_lock->fl_owner = (fl_owner_t)lo;
4522         file_lock->fl_pid = current->tgid;
4523         file_lock->fl_flags = FL_POSIX;
4524
4525         file_lock->fl_start = lockt->lt_offset;
4526         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4527
4528         nfs4_transform_lock_offset(file_lock);
4529
4530         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4531         if (status)
4532                 goto out;
4533
4534         if (file_lock->fl_type != F_UNLCK) {
4535                 status = nfserr_denied;
4536                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4537         }
4538 out:
4539         nfs4_unlock_state();
4540         if (file_lock)
4541                 locks_free_lock(file_lock);
4542         return status;
4543 }
4544
4545 __be32
4546 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4547             struct nfsd4_locku *locku)
4548 {
4549         struct nfs4_ol_stateid *stp;
4550         struct file *filp = NULL;
4551         struct file_lock *file_lock = NULL;
4552         __be32 status;
4553         int err;
4554         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4555
4556         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4557                 (long long) locku->lu_offset,
4558                 (long long) locku->lu_length);
4559
4560         if (check_lock_length(locku->lu_offset, locku->lu_length))
4561                  return nfserr_inval;
4562
4563         nfs4_lock_state();
4564
4565         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4566                                         &locku->lu_stateid, NFS4_LOCK_STID,
4567                                         &stp, nn);
4568         if (status)
4569                 goto out;
4570         filp = find_any_file(stp->st_file);
4571         if (!filp) {
4572                 status = nfserr_lock_range;
4573                 goto out;
4574         }
4575         file_lock = locks_alloc_lock();
4576         if (!file_lock) {
4577                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4578                 status = nfserr_jukebox;
4579                 goto out;
4580         }
4581         locks_init_lock(file_lock);
4582         file_lock->fl_type = F_UNLCK;
4583         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4584         file_lock->fl_pid = current->tgid;
4585         file_lock->fl_file = filp;
4586         file_lock->fl_flags = FL_POSIX;
4587         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4588         file_lock->fl_start = locku->lu_offset;
4589
4590         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4591                                                 locku->lu_length);
4592         nfs4_transform_lock_offset(file_lock);
4593
4594         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4595         if (err) {
4596                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4597                 goto out_nfserr;
4598         }
4599         update_stateid(&stp->st_stid.sc_stateid);
4600         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4601
4602 out:
4603         nfsd4_bump_seqid(cstate, status);
4604         if (!cstate->replay_owner)
4605                 nfs4_unlock_state();
4606         if (file_lock)
4607                 locks_free_lock(file_lock);
4608         return status;
4609
4610 out_nfserr:
4611         status = nfserrno(err);
4612         goto out;
4613 }
4614
4615 /*
4616  * returns
4617  *      1: locks held by lockowner
4618  *      0: no locks held by lockowner
4619  */
4620 static int
4621 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4622 {
4623         struct file_lock **flpp;
4624         struct inode *inode = filp->fi_inode;
4625         int status = 0;
4626
4627         spin_lock(&inode->i_lock);
4628         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4629                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4630                         status = 1;
4631                         goto out;
4632                 }
4633         }
4634 out:
4635         spin_unlock(&inode->i_lock);
4636         return status;
4637 }
4638
4639 __be32
4640 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4641                         struct nfsd4_compound_state *cstate,
4642                         struct nfsd4_release_lockowner *rlockowner)
4643 {
4644         clientid_t *clid = &rlockowner->rl_clientid;
4645         struct nfs4_stateowner *sop;
4646         struct nfs4_lockowner *lo;
4647         struct nfs4_ol_stateid *stp;
4648         struct xdr_netobj *owner = &rlockowner->rl_owner;
4649         struct list_head matches;
4650         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4651         __be32 status;
4652         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4653
4654         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4655                 clid->cl_boot, clid->cl_id);
4656
4657         nfs4_lock_state();
4658
4659         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4660         if (status)
4661                 goto out;
4662
4663         status = nfserr_locks_held;
4664         INIT_LIST_HEAD(&matches);
4665
4666         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4667                 if (sop->so_is_open_owner)
4668                         continue;
4669                 if (!same_owner_str(sop, owner, clid))
4670                         continue;
4671                 list_for_each_entry(stp, &sop->so_stateids,
4672                                 st_perstateowner) {
4673                         lo = lockowner(sop);
4674                         if (check_for_locks(stp->st_file, lo))
4675                                 goto out;
4676                         list_add(&lo->lo_list, &matches);
4677                 }
4678         }
4679         /* Clients probably won't expect us to return with some (but not all)
4680          * of the lockowner state released; so don't release any until all
4681          * have been checked. */
4682         status = nfs_ok;
4683         while (!list_empty(&matches)) {
4684                 lo = list_entry(matches.next, struct nfs4_lockowner,
4685                                                                 lo_list);
4686                 /* unhash_stateowner deletes so_perclient only
4687                  * for openowners. */
4688                 list_del(&lo->lo_list);
4689                 release_lockowner(lo);
4690         }
4691 out:
4692         nfs4_unlock_state();
4693         return status;
4694 }
4695
4696 static inline struct nfs4_client_reclaim *
4697 alloc_reclaim(void)
4698 {
4699         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4700 }
4701
4702 bool
4703 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4704 {
4705         struct nfs4_client_reclaim *crp;
4706
4707         crp = nfsd4_find_reclaim_client(name, nn);
4708         return (crp && crp->cr_clp);
4709 }
4710
4711 /*
4712  * failure => all reset bets are off, nfserr_no_grace...
4713  */
4714 struct nfs4_client_reclaim *
4715 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4716 {
4717         unsigned int strhashval;
4718         struct nfs4_client_reclaim *crp;
4719
4720         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4721         crp = alloc_reclaim();
4722         if (crp) {
4723                 strhashval = clientstr_hashval(name);
4724                 INIT_LIST_HEAD(&crp->cr_strhash);
4725                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4726                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4727                 crp->cr_clp = NULL;
4728                 nn->reclaim_str_hashtbl_size++;
4729         }
4730         return crp;
4731 }
4732
4733 void
4734 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4735 {
4736         list_del(&crp->cr_strhash);
4737         kfree(crp);
4738         nn->reclaim_str_hashtbl_size--;
4739 }
4740
4741 void
4742 nfs4_release_reclaim(struct nfsd_net *nn)
4743 {
4744         struct nfs4_client_reclaim *crp = NULL;
4745         int i;
4746
4747         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4748                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4749                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4750                                         struct nfs4_client_reclaim, cr_strhash);
4751                         nfs4_remove_reclaim_record(crp, nn);
4752                 }
4753         }
4754         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4755 }
4756
4757 /*
4758  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4759 struct nfs4_client_reclaim *
4760 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4761 {
4762         unsigned int strhashval;
4763         struct nfs4_client_reclaim *crp = NULL;
4764
4765         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4766
4767         strhashval = clientstr_hashval(recdir);
4768         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4769                 if (same_name(crp->cr_recdir, recdir)) {
4770                         return crp;
4771                 }
4772         }
4773         return NULL;
4774 }
4775
4776 /*
4777 * Called from OPEN. Look for clientid in reclaim list.
4778 */
4779 __be32
4780 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4781 {
4782         struct nfs4_client *clp;
4783
4784         /* find clientid in conf_id_hashtbl */
4785         clp = find_confirmed_client(clid, sessions, nn);
4786         if (clp == NULL)
4787                 return nfserr_reclaim_bad;
4788
4789         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4790 }
4791
4792 #ifdef CONFIG_NFSD_FAULT_INJECTION
4793
4794 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4795 {
4796         if (mark_client_expired(clp))
4797                 return 0;
4798         expire_client(clp);
4799         return 1;
4800 }
4801
4802 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4803 {
4804         char buf[INET6_ADDRSTRLEN];
4805         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4806         printk(KERN_INFO "NFS Client: %s\n", buf);
4807         return 1;
4808 }
4809
4810 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4811                              const char *type)
4812 {
4813         char buf[INET6_ADDRSTRLEN];
4814         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4815         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4816 }
4817
4818 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4819 {
4820         struct nfs4_openowner *oop;
4821         struct nfs4_lockowner *lop, *lo_next;
4822         struct nfs4_ol_stateid *stp, *st_next;
4823         u64 count = 0;
4824
4825         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4826                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4827                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4828                                 if (func)
4829                                         func(lop);
4830                                 if (++count == max)
4831                                         return count;
4832                         }
4833                 }
4834         }
4835
4836         return count;
4837 }
4838
4839 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4840 {
4841         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4842 }
4843
4844 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4845 {
4846         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4847         nfsd_print_count(clp, count, "locked files");
4848         return count;
4849 }
4850
4851 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4852 {
4853         struct nfs4_openowner *oop, *next;
4854         u64 count = 0;
4855
4856         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4857                 if (func)
4858                         func(oop);
4859                 if (++count == max)
4860                         break;
4861         }
4862
4863         return count;
4864 }
4865
4866 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4867 {
4868         return nfsd_foreach_client_open(clp, max, release_openowner);
4869 }
4870
4871 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4872 {
4873         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4874         nfsd_print_count(clp, count, "open files");
4875         return count;
4876 }
4877
4878 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4879                                      struct list_head *victims)
4880 {
4881         struct nfs4_delegation *dp, *next;
4882         u64 count = 0;
4883
4884         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4885                 if (victims)
4886                         list_move(&dp->dl_recall_lru, victims);
4887                 if (++count == max)
4888                         break;
4889         }
4890         return count;
4891 }
4892
4893 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4894 {
4895         struct nfs4_delegation *dp, *next;
4896         LIST_HEAD(victims);
4897         u64 count;
4898
4899         spin_lock(&recall_lock);
4900         count = nfsd_find_all_delegations(clp, max, &victims);
4901         spin_unlock(&recall_lock);
4902
4903         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4904                 revoke_delegation(dp);
4905
4906         return count;
4907 }
4908
4909 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4910 {
4911         struct nfs4_delegation *dp, *next;
4912         LIST_HEAD(victims);
4913         u64 count;
4914
4915         spin_lock(&recall_lock);
4916         count = nfsd_find_all_delegations(clp, max, &victims);
4917         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4918                 nfsd_break_one_deleg(dp);
4919         spin_unlock(&recall_lock);
4920
4921         return count;
4922 }
4923
4924 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4925 {
4926         u64 count = 0;
4927
4928         spin_lock(&recall_lock);
4929         count = nfsd_find_all_delegations(clp, max, NULL);
4930         spin_unlock(&recall_lock);
4931
4932         nfsd_print_count(clp, count, "delegations");
4933         return count;
4934 }
4935
4936 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4937 {
4938         struct nfs4_client *clp, *next;
4939         u64 count = 0;
4940         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4941
4942         if (!nfsd_netns_ready(nn))
4943                 return 0;
4944
4945         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4946                 count += func(clp, max - count);
4947                 if ((max != 0) && (count >= max))
4948                         break;
4949         }
4950
4951         return count;
4952 }
4953
4954 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4955 {
4956         struct nfs4_client *clp;
4957         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4958
4959         if (!nfsd_netns_ready(nn))
4960                 return NULL;
4961
4962         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4963                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4964                         return clp;
4965         }
4966         return NULL;
4967 }
4968
4969 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4970
4971 /* initialization to perform at module load time: */
4972
4973 void
4974 nfs4_state_init(void)
4975 {
4976 }
4977
4978 /*
4979  * Since the lifetime of a delegation isn't limited to that of an open, a
4980  * client may quite reasonably hang on to a delegation as long as it has
4981  * the inode cached.  This becomes an obvious problem the first time a
4982  * client's inode cache approaches the size of the server's total memory.
4983  *
4984  * For now we avoid this problem by imposing a hard limit on the number
4985  * of delegations, which varies according to the server's memory size.
4986  */
4987 static void
4988 set_max_delegations(void)
4989 {
4990         /*
4991          * Allow at most 4 delegations per megabyte of RAM.  Quick
4992          * estimates suggest that in the worst case (where every delegation
4993          * is for a different inode), a delegation could take about 1.5K,
4994          * giving a worst case usage of about 6% of memory.
4995          */
4996         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4997 }
4998
4999 static int nfs4_state_create_net(struct net *net)
5000 {
5001         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5002         int i;
5003
5004         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
5005                         CLIENT_HASH_SIZE, GFP_KERNEL);
5006         if (!nn->conf_id_hashtbl)
5007                 goto err;
5008         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
5009                         CLIENT_HASH_SIZE, GFP_KERNEL);
5010         if (!nn->unconf_id_hashtbl)
5011                 goto err_unconf_id;
5012         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
5013                         OWNER_HASH_SIZE, GFP_KERNEL);
5014         if (!nn->ownerstr_hashtbl)
5015                 goto err_ownerstr;
5016         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
5017                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
5018         if (!nn->lockowner_ino_hashtbl)
5019                 goto err_lockowner_ino;
5020         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
5021                         SESSION_HASH_SIZE, GFP_KERNEL);
5022         if (!nn->sessionid_hashtbl)
5023                 goto err_sessionid;
5024
5025         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5026                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
5027                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
5028         }
5029         for (i = 0; i < OWNER_HASH_SIZE; i++)
5030                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
5031         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
5032                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
5033         for (i = 0; i < SESSION_HASH_SIZE; i++)
5034                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
5035         nn->conf_name_tree = RB_ROOT;
5036         nn->unconf_name_tree = RB_ROOT;
5037         INIT_LIST_HEAD(&nn->client_lru);
5038         INIT_LIST_HEAD(&nn->close_lru);
5039         INIT_LIST_HEAD(&nn->del_recall_lru);
5040         spin_lock_init(&nn->client_lock);
5041
5042         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
5043         get_net(net);
5044
5045         return 0;
5046
5047 err_sessionid:
5048         kfree(nn->lockowner_ino_hashtbl);
5049 err_lockowner_ino:
5050         kfree(nn->ownerstr_hashtbl);
5051 err_ownerstr:
5052         kfree(nn->unconf_id_hashtbl);
5053 err_unconf_id:
5054         kfree(nn->conf_id_hashtbl);
5055 err:
5056         return -ENOMEM;
5057 }
5058
5059 static void
5060 nfs4_state_destroy_net(struct net *net)
5061 {
5062         int i;
5063         struct nfs4_client *clp = NULL;
5064         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5065         struct rb_node *node, *tmp;
5066
5067         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5068                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
5069                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
5070                         destroy_client(clp);
5071                 }
5072         }
5073
5074         node = rb_first(&nn->unconf_name_tree);
5075         while (node != NULL) {
5076                 tmp = node;
5077                 node = rb_next(tmp);
5078                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
5079                 rb_erase(tmp, &nn->unconf_name_tree);
5080                 destroy_client(clp);
5081         }
5082
5083         kfree(nn->sessionid_hashtbl);
5084         kfree(nn->lockowner_ino_hashtbl);
5085         kfree(nn->ownerstr_hashtbl);
5086         kfree(nn->unconf_id_hashtbl);
5087         kfree(nn->conf_id_hashtbl);
5088         put_net(net);
5089 }
5090
5091 int
5092 nfs4_state_start_net(struct net *net)
5093 {
5094         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5095         int ret;
5096
5097         ret = nfs4_state_create_net(net);
5098         if (ret)
5099                 return ret;
5100         nfsd4_client_tracking_init(net);
5101         nn->boot_time = get_seconds();
5102         locks_start_grace(net, &nn->nfsd4_manager);
5103         nn->grace_ended = false;
5104         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
5105                nn->nfsd4_grace, net);
5106         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
5107         return 0;
5108 }
5109
5110 /* initialization to perform when the nfsd service is started: */
5111
5112 int
5113 nfs4_state_start(void)
5114 {
5115         int ret;
5116
5117         ret = set_callback_cred();
5118         if (ret)
5119                 return -ENOMEM;
5120         laundry_wq = create_singlethread_workqueue("nfsd4");
5121         if (laundry_wq == NULL) {
5122                 ret = -ENOMEM;
5123                 goto out_recovery;
5124         }
5125         ret = nfsd4_create_callback_queue();
5126         if (ret)
5127                 goto out_free_laundry;
5128
5129         set_max_delegations();
5130
5131         return 0;
5132
5133 out_free_laundry:
5134         destroy_workqueue(laundry_wq);
5135 out_recovery:
5136         return ret;
5137 }
5138
5139 void
5140 nfs4_state_shutdown_net(struct net *net)
5141 {
5142         struct nfs4_delegation *dp = NULL;
5143         struct list_head *pos, *next, reaplist;
5144         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5145
5146         cancel_delayed_work_sync(&nn->laundromat_work);
5147         locks_end_grace(&nn->nfsd4_manager);
5148
5149         nfs4_lock_state();
5150         INIT_LIST_HEAD(&reaplist);
5151         spin_lock(&recall_lock);
5152         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5153                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5154                 list_move(&dp->dl_recall_lru, &reaplist);
5155         }
5156         spin_unlock(&recall_lock);
5157         list_for_each_safe(pos, next, &reaplist) {
5158                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5159                 destroy_delegation(dp);
5160         }
5161
5162         nfsd4_client_tracking_exit(net);
5163         nfs4_state_destroy_net(net);
5164         nfs4_unlock_state();
5165 }
5166
5167 void
5168 nfs4_state_shutdown(void)
5169 {
5170         destroy_workqueue(laundry_wq);
5171         nfsd4_destroy_callback_queue();
5172 }
5173
5174 static void
5175 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5176 {
5177         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5178                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5179 }
5180
5181 static void
5182 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5183 {
5184         if (cstate->minorversion) {
5185                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5186                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5187         }
5188 }
5189
5190 void
5191 clear_current_stateid(struct nfsd4_compound_state *cstate)
5192 {
5193         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5194 }
5195
5196 /*
5197  * functions to set current state id
5198  */
5199 void
5200 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5201 {
5202         put_stateid(cstate, &odp->od_stateid);
5203 }
5204
5205 void
5206 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5207 {
5208         put_stateid(cstate, &open->op_stateid);
5209 }
5210
5211 void
5212 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5213 {
5214         put_stateid(cstate, &close->cl_stateid);
5215 }
5216
5217 void
5218 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5219 {
5220         put_stateid(cstate, &lock->lk_resp_stateid);
5221 }
5222
5223 /*
5224  * functions to consume current state id
5225  */
5226
5227 void
5228 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5229 {
5230         get_stateid(cstate, &odp->od_stateid);
5231 }
5232
5233 void
5234 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5235 {
5236         get_stateid(cstate, &drp->dr_stateid);
5237 }
5238
5239 void
5240 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5241 {
5242         get_stateid(cstate, &fsp->fr_stateid);
5243 }
5244
5245 void
5246 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5247 {
5248         get_stateid(cstate, &setattr->sa_stateid);
5249 }
5250
5251 void
5252 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5253 {
5254         get_stateid(cstate, &close->cl_stateid);
5255 }
5256
5257 void
5258 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5259 {
5260         get_stateid(cstate, &locku->lu_stateid);
5261 }
5262
5263 void
5264 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5265 {
5266         get_stateid(cstate, &read->rd_stateid);
5267 }
5268
5269 void
5270 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5271 {
5272         get_stateid(cstate, &write->wr_stateid);
5273 }