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         INIT_LIST_HEAD(&clp->cl_sessions);
1082         idr_init(&clp->cl_stateids);
1083         atomic_set(&clp->cl_refcount, 0);
1084         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1085         INIT_LIST_HEAD(&clp->cl_idhash);
1086         INIT_LIST_HEAD(&clp->cl_openowners);
1087         INIT_LIST_HEAD(&clp->cl_delegations);
1088         INIT_LIST_HEAD(&clp->cl_lru);
1089         INIT_LIST_HEAD(&clp->cl_callbacks);
1090         INIT_LIST_HEAD(&clp->cl_revoked);
1091         spin_lock_init(&clp->cl_lock);
1092         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1093         return clp;
1094 }
1095
1096 static inline void
1097 free_client(struct nfs4_client *clp)
1098 {
1099         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1100
1101         lockdep_assert_held(&nn->client_lock);
1102         while (!list_empty(&clp->cl_sessions)) {
1103                 struct nfsd4_session *ses;
1104                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1105                                 se_perclnt);
1106                 list_del(&ses->se_perclnt);
1107                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1108                 free_session(ses);
1109         }
1110         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1111         free_svc_cred(&clp->cl_cred);
1112         kfree(clp->cl_name.data);
1113         idr_destroy(&clp->cl_stateids);
1114         kfree(clp);
1115 }
1116
1117 /* must be called under the client_lock */
1118 static inline void
1119 unhash_client_locked(struct nfs4_client *clp)
1120 {
1121         struct nfsd4_session *ses;
1122
1123         list_del(&clp->cl_lru);
1124         spin_lock(&clp->cl_lock);
1125         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1126                 list_del_init(&ses->se_hash);
1127         spin_unlock(&clp->cl_lock);
1128 }
1129
1130 static void
1131 destroy_client(struct nfs4_client *clp)
1132 {
1133         struct nfs4_openowner *oo;
1134         struct nfs4_delegation *dp;
1135         struct list_head reaplist;
1136         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1137
1138         INIT_LIST_HEAD(&reaplist);
1139         spin_lock(&recall_lock);
1140         while (!list_empty(&clp->cl_delegations)) {
1141                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1142                 list_del_init(&dp->dl_perclnt);
1143                 list_move(&dp->dl_recall_lru, &reaplist);
1144         }
1145         spin_unlock(&recall_lock);
1146         while (!list_empty(&reaplist)) {
1147                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1148                 destroy_delegation(dp);
1149         }
1150         list_splice_init(&clp->cl_revoked, &reaplist);
1151         while (!list_empty(&reaplist)) {
1152                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1153                 destroy_revoked_delegation(dp);
1154         }
1155         while (!list_empty(&clp->cl_openowners)) {
1156                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1157                 release_openowner(oo);
1158         }
1159         nfsd4_shutdown_callback(clp);
1160         if (clp->cl_cb_conn.cb_xprt)
1161                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1162         list_del(&clp->cl_idhash);
1163         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1164                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1165         else
1166                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1167         spin_lock(&nn->client_lock);
1168         unhash_client_locked(clp);
1169         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1170         free_client(clp);
1171         spin_unlock(&nn->client_lock);
1172 }
1173
1174 static void expire_client(struct nfs4_client *clp)
1175 {
1176         nfsd4_client_record_remove(clp);
1177         destroy_client(clp);
1178 }
1179
1180 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1181 {
1182         memcpy(target->cl_verifier.data, source->data,
1183                         sizeof(target->cl_verifier.data));
1184 }
1185
1186 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1187 {
1188         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1189         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1190 }
1191
1192 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1193 {
1194         if (source->cr_principal) {
1195                 target->cr_principal =
1196                                 kstrdup(source->cr_principal, GFP_KERNEL);
1197                 if (target->cr_principal == NULL)
1198                         return -ENOMEM;
1199         } else
1200                 target->cr_principal = NULL;
1201         target->cr_flavor = source->cr_flavor;
1202         target->cr_uid = source->cr_uid;
1203         target->cr_gid = source->cr_gid;
1204         target->cr_group_info = source->cr_group_info;
1205         get_group_info(target->cr_group_info);
1206         target->cr_gss_mech = source->cr_gss_mech;
1207         if (source->cr_gss_mech)
1208                 gss_mech_get(source->cr_gss_mech);
1209         return 0;
1210 }
1211
1212 static long long
1213 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1214 {
1215         long long res;
1216
1217         res = o1->len - o2->len;
1218         if (res)
1219                 return res;
1220         return (long long)memcmp(o1->data, o2->data, o1->len);
1221 }
1222
1223 static int same_name(const char *n1, const char *n2)
1224 {
1225         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1226 }
1227
1228 static int
1229 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1230 {
1231         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1232 }
1233
1234 static int
1235 same_clid(clientid_t *cl1, clientid_t *cl2)
1236 {
1237         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1238 }
1239
1240 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1241 {
1242         int i;
1243
1244         if (g1->ngroups != g2->ngroups)
1245                 return false;
1246         for (i=0; i<g1->ngroups; i++)
1247                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1248                         return false;
1249         return true;
1250 }
1251
1252 /*
1253  * RFC 3530 language requires clid_inuse be returned when the
1254  * "principal" associated with a requests differs from that previously
1255  * used.  We use uid, gid's, and gss principal string as our best
1256  * approximation.  We also don't want to allow non-gss use of a client
1257  * established using gss: in theory cr_principal should catch that
1258  * change, but in practice cr_principal can be null even in the gss case
1259  * since gssd doesn't always pass down a principal string.
1260  */
1261 static bool is_gss_cred(struct svc_cred *cr)
1262 {
1263         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1264         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1265 }
1266
1267
1268 static bool
1269 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1270 {
1271         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1272                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1273                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1274                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1275                 return false;
1276         if (cr1->cr_principal == cr2->cr_principal)
1277                 return true;
1278         if (!cr1->cr_principal || !cr2->cr_principal)
1279                 return false;
1280         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1281 }
1282
1283 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1284 {
1285         struct svc_cred *cr = &rqstp->rq_cred;
1286         u32 service;
1287
1288         if (!cr->cr_gss_mech)
1289                 return false;
1290         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1291         return service == RPC_GSS_SVC_INTEGRITY ||
1292                service == RPC_GSS_SVC_PRIVACY;
1293 }
1294
1295 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1296 {
1297         struct svc_cred *cr = &rqstp->rq_cred;
1298
1299         if (!cl->cl_mach_cred)
1300                 return true;
1301         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1302                 return false;
1303         if (!svc_rqst_integrity_protected(rqstp))
1304                 return false;
1305         if (!cr->cr_principal)
1306                 return false;
1307         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1308 }
1309
1310 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1311 {
1312         static u32 current_clientid = 1;
1313
1314         clp->cl_clientid.cl_boot = nn->boot_time;
1315         clp->cl_clientid.cl_id = current_clientid++;
1316 }
1317
1318 static void gen_confirm(struct nfs4_client *clp)
1319 {
1320         __be32 verf[2];
1321         static u32 i;
1322
1323         verf[0] = (__be32)get_seconds();
1324         verf[1] = (__be32)i++;
1325         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1326 }
1327
1328 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1329 {
1330         struct nfs4_stid *ret;
1331
1332         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1333         if (!ret || !ret->sc_type)
1334                 return NULL;
1335         return ret;
1336 }
1337
1338 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1339 {
1340         struct nfs4_stid *s;
1341
1342         s = find_stateid(cl, t);
1343         if (!s)
1344                 return NULL;
1345         if (typemask & s->sc_type)
1346                 return s;
1347         return NULL;
1348 }
1349
1350 static struct nfs4_client *create_client(struct xdr_netobj name,
1351                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1352 {
1353         struct nfs4_client *clp;
1354         struct sockaddr *sa = svc_addr(rqstp);
1355         int ret;
1356         struct net *net = SVC_NET(rqstp);
1357         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1358
1359         clp = alloc_client(name);
1360         if (clp == NULL)
1361                 return NULL;
1362
1363         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1364         if (ret) {
1365                 spin_lock(&nn->client_lock);
1366                 free_client(clp);
1367                 spin_unlock(&nn->client_lock);
1368                 return NULL;
1369         }
1370         nfsd4_init_callback(&clp->cl_cb_null);
1371         clp->cl_time = get_seconds();
1372         clear_bit(0, &clp->cl_cb_slot_busy);
1373         copy_verf(clp, verf);
1374         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1375         gen_confirm(clp);
1376         clp->cl_cb_session = NULL;
1377         clp->net = net;
1378         return clp;
1379 }
1380
1381 static void
1382 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1383 {
1384         struct rb_node **new = &(root->rb_node), *parent = NULL;
1385         struct nfs4_client *clp;
1386
1387         while (*new) {
1388                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1389                 parent = *new;
1390
1391                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1392                         new = &((*new)->rb_left);
1393                 else
1394                         new = &((*new)->rb_right);
1395         }
1396
1397         rb_link_node(&new_clp->cl_namenode, parent, new);
1398         rb_insert_color(&new_clp->cl_namenode, root);
1399 }
1400
1401 static struct nfs4_client *
1402 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1403 {
1404         long long cmp;
1405         struct rb_node *node = root->rb_node;
1406         struct nfs4_client *clp;
1407
1408         while (node) {
1409                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1410                 cmp = compare_blob(&clp->cl_name, name);
1411                 if (cmp > 0)
1412                         node = node->rb_left;
1413                 else if (cmp < 0)
1414                         node = node->rb_right;
1415                 else
1416                         return clp;
1417         }
1418         return NULL;
1419 }
1420
1421 static void
1422 add_to_unconfirmed(struct nfs4_client *clp)
1423 {
1424         unsigned int idhashval;
1425         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1426
1427         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1428         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1429         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1430         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1431         renew_client(clp);
1432 }
1433
1434 static void
1435 move_to_confirmed(struct nfs4_client *clp)
1436 {
1437         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1438         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1439
1440         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1441         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1442         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1443         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1444         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1445         renew_client(clp);
1446 }
1447
1448 static struct nfs4_client *
1449 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1450 {
1451         struct nfs4_client *clp;
1452         unsigned int idhashval = clientid_hashval(clid->cl_id);
1453
1454         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1455                 if (same_clid(&clp->cl_clientid, clid)) {
1456                         if ((bool)clp->cl_minorversion != sessions)
1457                                 return NULL;
1458                         renew_client(clp);
1459                         return clp;
1460                 }
1461         }
1462         return NULL;
1463 }
1464
1465 static struct nfs4_client *
1466 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1467 {
1468         struct list_head *tbl = nn->conf_id_hashtbl;
1469
1470         return find_client_in_id_table(tbl, clid, sessions);
1471 }
1472
1473 static struct nfs4_client *
1474 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1475 {
1476         struct list_head *tbl = nn->unconf_id_hashtbl;
1477
1478         return find_client_in_id_table(tbl, clid, sessions);
1479 }
1480
1481 static bool clp_used_exchangeid(struct nfs4_client *clp)
1482 {
1483         return clp->cl_exchange_flags != 0;
1484 }
1485
1486 static struct nfs4_client *
1487 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1488 {
1489         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1490 }
1491
1492 static struct nfs4_client *
1493 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1494 {
1495         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1496 }
1497
1498 static void
1499 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1500 {
1501         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1502         struct sockaddr *sa = svc_addr(rqstp);
1503         u32 scopeid = rpc_get_scope_id(sa);
1504         unsigned short expected_family;
1505
1506         /* Currently, we only support tcp and tcp6 for the callback channel */
1507         if (se->se_callback_netid_len == 3 &&
1508             !memcmp(se->se_callback_netid_val, "tcp", 3))
1509                 expected_family = AF_INET;
1510         else if (se->se_callback_netid_len == 4 &&
1511                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1512                 expected_family = AF_INET6;
1513         else
1514                 goto out_err;
1515
1516         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1517                                             se->se_callback_addr_len,
1518                                             (struct sockaddr *)&conn->cb_addr,
1519                                             sizeof(conn->cb_addr));
1520
1521         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1522                 goto out_err;
1523
1524         if (conn->cb_addr.ss_family == AF_INET6)
1525                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1526
1527         conn->cb_prog = se->se_callback_prog;
1528         conn->cb_ident = se->se_callback_ident;
1529         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1530         return;
1531 out_err:
1532         conn->cb_addr.ss_family = AF_UNSPEC;
1533         conn->cb_addrlen = 0;
1534         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1535                 "will not receive delegations\n",
1536                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1537
1538         return;
1539 }
1540
1541 /*
1542  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1543  */
1544 void
1545 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1546 {
1547         struct nfsd4_slot *slot = resp->cstate.slot;
1548         unsigned int base;
1549
1550         dprintk("--> %s slot %p\n", __func__, slot);
1551
1552         slot->sl_opcnt = resp->opcnt;
1553         slot->sl_status = resp->cstate.status;
1554
1555         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1556         if (nfsd4_not_cached(resp)) {
1557                 slot->sl_datalen = 0;
1558                 return;
1559         }
1560         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1561         base = (char *)resp->cstate.datap -
1562                                         (char *)resp->xbuf->head[0].iov_base;
1563         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1564                                     slot->sl_datalen))
1565                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1566         return;
1567 }
1568
1569 /*
1570  * Encode the replay sequence operation from the slot values.
1571  * If cachethis is FALSE encode the uncached rep error on the next
1572  * operation which sets resp->p and increments resp->opcnt for
1573  * nfs4svc_encode_compoundres.
1574  *
1575  */
1576 static __be32
1577 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1578                           struct nfsd4_compoundres *resp)
1579 {
1580         struct nfsd4_op *op;
1581         struct nfsd4_slot *slot = resp->cstate.slot;
1582
1583         /* Encode the replayed sequence operation */
1584         op = &args->ops[resp->opcnt - 1];
1585         nfsd4_encode_operation(resp, op);
1586
1587         /* Return nfserr_retry_uncached_rep in next operation. */
1588         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1589                 op = &args->ops[resp->opcnt++];
1590                 op->status = nfserr_retry_uncached_rep;
1591                 nfsd4_encode_operation(resp, op);
1592         }
1593         return op->status;
1594 }
1595
1596 /*
1597  * The sequence operation is not cached because we can use the slot and
1598  * session values.
1599  */
1600 __be32
1601 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1602                          struct nfsd4_sequence *seq)
1603 {
1604         struct nfsd4_slot *slot = resp->cstate.slot;
1605         __be32 status;
1606
1607         dprintk("--> %s slot %p\n", __func__, slot);
1608
1609         /* Either returns 0 or nfserr_retry_uncached */
1610         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1611         if (status == nfserr_retry_uncached_rep)
1612                 return status;
1613
1614         /* The sequence operation has been encoded, cstate->datap set. */
1615         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1616
1617         resp->opcnt = slot->sl_opcnt;
1618         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1619         status = slot->sl_status;
1620
1621         return status;
1622 }
1623
1624 /*
1625  * Set the exchange_id flags returned by the server.
1626  */
1627 static void
1628 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1629 {
1630         /* pNFS is not supported */
1631         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1632
1633         /* Referrals are supported, Migration is not. */
1634         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1635
1636         /* set the wire flags to return to client. */
1637         clid->flags = new->cl_exchange_flags;
1638 }
1639
1640 static bool client_has_state(struct nfs4_client *clp)
1641 {
1642         /*
1643          * Note clp->cl_openowners check isn't quite right: there's no
1644          * need to count owners without stateid's.
1645          *
1646          * Also note we should probably be using this in 4.0 case too.
1647          */
1648         return !list_empty(&clp->cl_openowners)
1649                 || !list_empty(&clp->cl_delegations)
1650                 || !list_empty(&clp->cl_sessions);
1651 }
1652
1653 __be32
1654 nfsd4_exchange_id(struct svc_rqst *rqstp,
1655                   struct nfsd4_compound_state *cstate,
1656                   struct nfsd4_exchange_id *exid)
1657 {
1658         struct nfs4_client *unconf, *conf, *new;
1659         __be32 status;
1660         char                    addr_str[INET6_ADDRSTRLEN];
1661         nfs4_verifier           verf = exid->verifier;
1662         struct sockaddr         *sa = svc_addr(rqstp);
1663         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1664         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1665
1666         rpc_ntop(sa, addr_str, sizeof(addr_str));
1667         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1668                 "ip_addr=%s flags %x, spa_how %d\n",
1669                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1670                 addr_str, exid->flags, exid->spa_how);
1671
1672         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1673                 return nfserr_inval;
1674
1675         switch (exid->spa_how) {
1676         case SP4_MACH_CRED:
1677                 if (!svc_rqst_integrity_protected(rqstp))
1678                         return nfserr_inval;
1679         case SP4_NONE:
1680                 break;
1681         default:                                /* checked by xdr code */
1682                 WARN_ON_ONCE(1);
1683         case SP4_SSV:
1684                 return nfserr_encr_alg_unsupp;
1685         }
1686
1687         /* Cases below refer to rfc 5661 section 18.35.4: */
1688         nfs4_lock_state();
1689         conf = find_confirmed_client_by_name(&exid->clname, nn);
1690         if (conf) {
1691                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1692                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1693
1694                 if (update) {
1695                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1696                                 status = nfserr_inval;
1697                                 goto out;
1698                         }
1699                         if (!mach_creds_match(conf, rqstp)) {
1700                                 status = nfserr_wrong_cred;
1701                                 goto out;
1702                         }
1703                         if (!creds_match) { /* case 9 */
1704                                 status = nfserr_perm;
1705                                 goto out;
1706                         }
1707                         if (!verfs_match) { /* case 8 */
1708                                 status = nfserr_not_same;
1709                                 goto out;
1710                         }
1711                         /* case 6 */
1712                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1713                         new = conf;
1714                         goto out_copy;
1715                 }
1716                 if (!creds_match) { /* case 3 */
1717                         if (client_has_state(conf)) {
1718                                 status = nfserr_clid_inuse;
1719                                 goto out;
1720                         }
1721                         expire_client(conf);
1722                         goto out_new;
1723                 }
1724                 if (verfs_match) { /* case 2 */
1725                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1726                         new = conf;
1727                         goto out_copy;
1728                 }
1729                 /* case 5, client reboot */
1730                 goto out_new;
1731         }
1732
1733         if (update) { /* case 7 */
1734                 status = nfserr_noent;
1735                 goto out;
1736         }
1737
1738         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1739         if (unconf) /* case 4, possible retry or client restart */
1740                 expire_client(unconf);
1741
1742         /* case 1 (normal case) */
1743 out_new:
1744         new = create_client(exid->clname, rqstp, &verf);
1745         if (new == NULL) {
1746                 status = nfserr_jukebox;
1747                 goto out;
1748         }
1749         new->cl_minorversion = cstate->minorversion;
1750         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
1751
1752         gen_clid(new, nn);
1753         add_to_unconfirmed(new);
1754 out_copy:
1755         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1756         exid->clientid.cl_id = new->cl_clientid.cl_id;
1757
1758         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1759         nfsd4_set_ex_flags(new, exid);
1760
1761         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1762                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1763         status = nfs_ok;
1764
1765 out:
1766         nfs4_unlock_state();
1767         return status;
1768 }
1769
1770 static __be32
1771 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1772 {
1773         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1774                 slot_seqid);
1775
1776         /* The slot is in use, and no response has been sent. */
1777         if (slot_inuse) {
1778                 if (seqid == slot_seqid)
1779                         return nfserr_jukebox;
1780                 else
1781                         return nfserr_seq_misordered;
1782         }
1783         /* Note unsigned 32-bit arithmetic handles wraparound: */
1784         if (likely(seqid == slot_seqid + 1))
1785                 return nfs_ok;
1786         if (seqid == slot_seqid)
1787                 return nfserr_replay_cache;
1788         return nfserr_seq_misordered;
1789 }
1790
1791 /*
1792  * Cache the create session result into the create session single DRC
1793  * slot cache by saving the xdr structure. sl_seqid has been set.
1794  * Do this for solo or embedded create session operations.
1795  */
1796 static void
1797 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1798                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1799 {
1800         slot->sl_status = nfserr;
1801         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1802 }
1803
1804 static __be32
1805 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1806                             struct nfsd4_clid_slot *slot)
1807 {
1808         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1809         return slot->sl_status;
1810 }
1811
1812 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1813                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1814                         1 +     /* MIN tag is length with zero, only length */ \
1815                         3 +     /* version, opcount, opcode */ \
1816                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1817                                 /* seqid, slotID, slotID, cache */ \
1818                         4 ) * sizeof(__be32))
1819
1820 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1821                         2 +     /* verifier: AUTH_NULL, length 0 */\
1822                         1 +     /* status */ \
1823                         1 +     /* MIN tag is length with zero, only length */ \
1824                         3 +     /* opcount, opcode, opstatus*/ \
1825                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1826                                 /* seqid, slotID, slotID, slotID, status */ \
1827                         5 ) * sizeof(__be32))
1828
1829 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1830 {
1831         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
1832
1833         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
1834                 return nfserr_toosmall;
1835         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
1836                 return nfserr_toosmall;
1837         ca->headerpadsz = 0;
1838         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
1839         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
1840         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
1841         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
1842                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
1843         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
1844         /*
1845          * Note decreasing slot size below client's request may make it
1846          * difficult for client to function correctly, whereas
1847          * decreasing the number of slots will (just?) affect
1848          * performance.  When short on memory we therefore prefer to
1849          * decrease number of slots instead of their size.  Clients that
1850          * request larger slots than they need will get poor results:
1851          */
1852         ca->maxreqs = nfsd4_get_drc_mem(ca);
1853         if (!ca->maxreqs)
1854                 return nfserr_jukebox;
1855
1856         return nfs_ok;
1857 }
1858
1859 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
1860                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
1861 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
1862                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
1863
1864 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
1865 {
1866         ca->headerpadsz = 0;
1867
1868         /*
1869          * These RPC_MAX_HEADER macros are overkill, especially since we
1870          * don't even do gss on the backchannel yet.  But this is still
1871          * less than 1k.  Tighten up this estimate in the unlikely event
1872          * it turns out to be a problem for some client:
1873          */
1874         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
1875                 return nfserr_toosmall;
1876         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
1877                 return nfserr_toosmall;
1878         ca->maxresp_cached = 0;
1879         if (ca->maxops < 2)
1880                 return nfserr_toosmall;
1881
1882         return nfs_ok;
1883 }
1884
1885 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
1886 {
1887         switch (cbs->flavor) {
1888         case RPC_AUTH_NULL:
1889         case RPC_AUTH_UNIX:
1890                 return nfs_ok;
1891         default:
1892                 /*
1893                  * GSS case: the spec doesn't allow us to return this
1894                  * error.  But it also doesn't allow us not to support
1895                  * GSS.
1896                  * I'd rather this fail hard than return some error the
1897                  * client might think it can already handle:
1898                  */
1899                 return nfserr_encr_alg_unsupp;
1900         }
1901 }
1902
1903 __be32
1904 nfsd4_create_session(struct svc_rqst *rqstp,
1905                      struct nfsd4_compound_state *cstate,
1906                      struct nfsd4_create_session *cr_ses)
1907 {
1908         struct sockaddr *sa = svc_addr(rqstp);
1909         struct nfs4_client *conf, *unconf;
1910         struct nfsd4_session *new;
1911         struct nfsd4_conn *conn;
1912         struct nfsd4_clid_slot *cs_slot = NULL;
1913         __be32 status = 0;
1914         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1915
1916         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1917                 return nfserr_inval;
1918         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
1919         if (status)
1920                 return status;
1921         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
1922         if (status)
1923                 return status;
1924         status = check_backchannel_attrs(&cr_ses->back_channel);
1925         if (status)
1926                 goto out_release_drc_mem;
1927         status = nfserr_jukebox;
1928         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
1929         if (!new)
1930                 goto out_release_drc_mem;
1931         conn = alloc_conn_from_crses(rqstp, cr_ses);
1932         if (!conn)
1933                 goto out_free_session;
1934
1935         nfs4_lock_state();
1936         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1937         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1938         WARN_ON_ONCE(conf && unconf);
1939
1940         if (conf) {
1941                 status = nfserr_wrong_cred;
1942                 if (!mach_creds_match(conf, rqstp))
1943                         goto out_free_conn;
1944                 cs_slot = &conf->cl_cs_slot;
1945                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1946                 if (status == nfserr_replay_cache) {
1947                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1948                         goto out_free_conn;
1949                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1950                         status = nfserr_seq_misordered;
1951                         goto out_free_conn;
1952                 }
1953         } else if (unconf) {
1954                 struct nfs4_client *old;
1955                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1956                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1957                         status = nfserr_clid_inuse;
1958                         goto out_free_conn;
1959                 }
1960                 status = nfserr_wrong_cred;
1961                 if (!mach_creds_match(unconf, rqstp))
1962                         goto out_free_conn;
1963                 cs_slot = &unconf->cl_cs_slot;
1964                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1965                 if (status) {
1966                         /* an unconfirmed replay returns misordered */
1967                         status = nfserr_seq_misordered;
1968                         goto out_free_conn;
1969                 }
1970                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1971                 if (old) {
1972                         status = mark_client_expired(old);
1973                         if (status)
1974                                 goto out_free_conn;
1975                         expire_client(old);
1976                 }
1977                 move_to_confirmed(unconf);
1978                 conf = unconf;
1979         } else {
1980                 status = nfserr_stale_clientid;
1981                 goto out_free_conn;
1982         }
1983         status = nfs_ok;
1984         /*
1985          * We do not support RDMA or persistent sessions
1986          */
1987         cr_ses->flags &= ~SESSION4_PERSIST;
1988         cr_ses->flags &= ~SESSION4_RDMA;
1989
1990         init_session(rqstp, new, conf, cr_ses);
1991         nfsd4_init_conn(rqstp, conn, new);
1992
1993         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1994                NFS4_MAX_SESSIONID_LEN);
1995         cs_slot->sl_seqid++;
1996         cr_ses->seqid = cs_slot->sl_seqid;
1997
1998         /* cache solo and embedded create sessions under the state lock */
1999         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2000         nfs4_unlock_state();
2001         return status;
2002 out_free_conn:
2003         nfs4_unlock_state();
2004         free_conn(conn);
2005 out_free_session:
2006         __free_session(new);
2007 out_release_drc_mem:
2008         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2009         return status;
2010 }
2011
2012 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2013 {
2014         switch (*dir) {
2015         case NFS4_CDFC4_FORE:
2016         case NFS4_CDFC4_BACK:
2017                 return nfs_ok;
2018         case NFS4_CDFC4_FORE_OR_BOTH:
2019         case NFS4_CDFC4_BACK_OR_BOTH:
2020                 *dir = NFS4_CDFC4_BOTH;
2021                 return nfs_ok;
2022         };
2023         return nfserr_inval;
2024 }
2025
2026 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2027 {
2028         struct nfsd4_session *session = cstate->session;
2029         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2030         __be32 status;
2031
2032         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2033         if (status)
2034                 return status;
2035         spin_lock(&nn->client_lock);
2036         session->se_cb_prog = bc->bc_cb_program;
2037         session->se_cb_sec = bc->bc_cb_sec;
2038         spin_unlock(&nn->client_lock);
2039
2040         nfsd4_probe_callback(session->se_client);
2041
2042         return nfs_ok;
2043 }
2044
2045 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2046                      struct nfsd4_compound_state *cstate,
2047                      struct nfsd4_bind_conn_to_session *bcts)
2048 {
2049         __be32 status;
2050         struct nfsd4_conn *conn;
2051         struct nfsd4_session *session;
2052         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2053
2054         if (!nfsd4_last_compound_op(rqstp))
2055                 return nfserr_not_only_op;
2056         nfs4_lock_state();
2057         spin_lock(&nn->client_lock);
2058         session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
2059         spin_unlock(&nn->client_lock);
2060         status = nfserr_badsession;
2061         if (!session)
2062                 goto out;
2063         status = nfserr_wrong_cred;
2064         if (!mach_creds_match(session->se_client, rqstp))
2065                 goto out;
2066         status = nfsd4_map_bcts_dir(&bcts->dir);
2067         if (status)
2068                 goto out;
2069         conn = alloc_conn(rqstp, bcts->dir);
2070         status = nfserr_jukebox;
2071         if (!conn)
2072                 goto out;
2073         nfsd4_init_conn(rqstp, conn, session);
2074         status = nfs_ok;
2075 out:
2076         nfs4_unlock_state();
2077         return status;
2078 }
2079
2080 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2081 {
2082         if (!session)
2083                 return 0;
2084         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2085 }
2086
2087 __be32
2088 nfsd4_destroy_session(struct svc_rqst *r,
2089                       struct nfsd4_compound_state *cstate,
2090                       struct nfsd4_destroy_session *sessionid)
2091 {
2092         struct nfsd4_session *ses;
2093         __be32 status;
2094         int ref_held_by_me = 0;
2095         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
2096
2097         nfs4_lock_state();
2098         status = nfserr_not_only_op;
2099         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2100                 if (!nfsd4_last_compound_op(r))
2101                         goto out;
2102                 ref_held_by_me++;
2103         }
2104         dump_sessionid(__func__, &sessionid->sessionid);
2105         spin_lock(&nn->client_lock);
2106         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
2107         status = nfserr_badsession;
2108         if (!ses)
2109                 goto out_client_lock;
2110         status = nfserr_wrong_cred;
2111         if (!mach_creds_match(ses->se_client, r))
2112                 goto out_client_lock;
2113         nfsd4_get_session_locked(ses);
2114         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2115         if (status)
2116                 goto out_put_session;
2117         unhash_session(ses);
2118         spin_unlock(&nn->client_lock);
2119
2120         nfsd4_probe_callback_sync(ses->se_client);
2121
2122         spin_lock(&nn->client_lock);
2123         status = nfs_ok;
2124 out_put_session:
2125         nfsd4_put_session(ses);
2126 out_client_lock:
2127         spin_unlock(&nn->client_lock);
2128 out:
2129         nfs4_unlock_state();
2130         return status;
2131 }
2132
2133 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2134 {
2135         struct nfsd4_conn *c;
2136
2137         list_for_each_entry(c, &s->se_conns, cn_persession) {
2138                 if (c->cn_xprt == xpt) {
2139                         return c;
2140                 }
2141         }
2142         return NULL;
2143 }
2144
2145 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2146 {
2147         struct nfs4_client *clp = ses->se_client;
2148         struct nfsd4_conn *c;
2149         __be32 status = nfs_ok;
2150         int ret;
2151
2152         spin_lock(&clp->cl_lock);
2153         c = __nfsd4_find_conn(new->cn_xprt, ses);
2154         if (c)
2155                 goto out_free;
2156         status = nfserr_conn_not_bound_to_session;
2157         if (clp->cl_mach_cred)
2158                 goto out_free;
2159         __nfsd4_hash_conn(new, ses);
2160         spin_unlock(&clp->cl_lock);
2161         ret = nfsd4_register_conn(new);
2162         if (ret)
2163                 /* oops; xprt is already down: */
2164                 nfsd4_conn_lost(&new->cn_xpt_user);
2165         return nfs_ok;
2166 out_free:
2167         spin_unlock(&clp->cl_lock);
2168         free_conn(new);
2169         return status;
2170 }
2171
2172 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2173 {
2174         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2175
2176         return args->opcnt > session->se_fchannel.maxops;
2177 }
2178
2179 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2180                                   struct nfsd4_session *session)
2181 {
2182         struct xdr_buf *xb = &rqstp->rq_arg;
2183
2184         return xb->len > session->se_fchannel.maxreq_sz;
2185 }
2186
2187 __be32
2188 nfsd4_sequence(struct svc_rqst *rqstp,
2189                struct nfsd4_compound_state *cstate,
2190                struct nfsd4_sequence *seq)
2191 {
2192         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2193         struct nfsd4_session *session;
2194         struct nfs4_client *clp;
2195         struct nfsd4_slot *slot;
2196         struct nfsd4_conn *conn;
2197         __be32 status;
2198         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2199
2200         if (resp->opcnt != 1)
2201                 return nfserr_sequence_pos;
2202
2203         /*
2204          * Will be either used or freed by nfsd4_sequence_check_conn
2205          * below.
2206          */
2207         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2208         if (!conn)
2209                 return nfserr_jukebox;
2210
2211         spin_lock(&nn->client_lock);
2212         status = nfserr_badsession;
2213         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2214         if (!session)
2215                 goto out_no_session;
2216         clp = session->se_client;
2217         status = get_client_locked(clp);
2218         if (status)
2219                 goto out_no_session;
2220         status = nfsd4_get_session_locked(session);
2221         if (status)
2222                 goto out_put_client;
2223
2224         status = nfserr_too_many_ops;
2225         if (nfsd4_session_too_many_ops(rqstp, session))
2226                 goto out_put_session;
2227
2228         status = nfserr_req_too_big;
2229         if (nfsd4_request_too_big(rqstp, session))
2230                 goto out_put_session;
2231
2232         status = nfserr_badslot;
2233         if (seq->slotid >= session->se_fchannel.maxreqs)
2234                 goto out_put_session;
2235
2236         slot = session->se_slots[seq->slotid];
2237         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2238
2239         /* We do not negotiate the number of slots yet, so set the
2240          * maxslots to the session maxreqs which is used to encode
2241          * sr_highest_slotid and the sr_target_slot id to maxslots */
2242         seq->maxslots = session->se_fchannel.maxreqs;
2243
2244         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2245                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2246         if (status == nfserr_replay_cache) {
2247                 status = nfserr_seq_misordered;
2248                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2249                         goto out_put_session;
2250                 cstate->slot = slot;
2251                 cstate->session = session;
2252                 /* Return the cached reply status and set cstate->status
2253                  * for nfsd4_proc_compound processing */
2254                 status = nfsd4_replay_cache_entry(resp, seq);
2255                 cstate->status = nfserr_replay_cache;
2256                 goto out;
2257         }
2258         if (status)
2259                 goto out_put_session;
2260
2261         status = nfsd4_sequence_check_conn(conn, session);
2262         conn = NULL;
2263         if (status)
2264                 goto out_put_session;
2265
2266         /* Success! bump slot seqid */
2267         slot->sl_seqid = seq->seqid;
2268         slot->sl_flags |= NFSD4_SLOT_INUSE;
2269         if (seq->cachethis)
2270                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2271         else
2272                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2273
2274         cstate->slot = slot;
2275         cstate->session = session;
2276
2277 out:
2278         switch (clp->cl_cb_state) {
2279         case NFSD4_CB_DOWN:
2280                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2281                 break;
2282         case NFSD4_CB_FAULT:
2283                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2284                 break;
2285         default:
2286                 seq->status_flags = 0;
2287         }
2288         if (!list_empty(&clp->cl_revoked))
2289                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2290 out_no_session:
2291         kfree(conn);
2292         spin_unlock(&nn->client_lock);
2293         return status;
2294 out_put_session:
2295         nfsd4_put_session(session);
2296 out_put_client:
2297         put_client_renew_locked(clp);
2298         goto out_no_session;
2299 }
2300
2301 __be32
2302 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2303 {
2304         struct nfs4_client *conf, *unconf, *clp;
2305         __be32 status = 0;
2306         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2307
2308         nfs4_lock_state();
2309         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2310         conf = find_confirmed_client(&dc->clientid, true, nn);
2311         WARN_ON_ONCE(conf && unconf);
2312
2313         if (conf) {
2314                 clp = conf;
2315
2316                 if (client_has_state(conf)) {
2317                         status = nfserr_clientid_busy;
2318                         goto out;
2319                 }
2320         } else if (unconf)
2321                 clp = unconf;
2322         else {
2323                 status = nfserr_stale_clientid;
2324                 goto out;
2325         }
2326         if (!mach_creds_match(clp, rqstp)) {
2327                 status = nfserr_wrong_cred;
2328                 goto out;
2329         }
2330         expire_client(clp);
2331 out:
2332         nfs4_unlock_state();
2333         return status;
2334 }
2335
2336 __be32
2337 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2338 {
2339         __be32 status = 0;
2340
2341         if (rc->rca_one_fs) {
2342                 if (!cstate->current_fh.fh_dentry)
2343                         return nfserr_nofilehandle;
2344                 /*
2345                  * We don't take advantage of the rca_one_fs case.
2346                  * That's OK, it's optional, we can safely ignore it.
2347                  */
2348                  return nfs_ok;
2349         }
2350
2351         nfs4_lock_state();
2352         status = nfserr_complete_already;
2353         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2354                              &cstate->session->se_client->cl_flags))
2355                 goto out;
2356
2357         status = nfserr_stale_clientid;
2358         if (is_client_expired(cstate->session->se_client))
2359                 /*
2360                  * The following error isn't really legal.
2361                  * But we only get here if the client just explicitly
2362                  * destroyed the client.  Surely it no longer cares what
2363                  * error it gets back on an operation for the dead
2364                  * client.
2365                  */
2366                 goto out;
2367
2368         status = nfs_ok;
2369         nfsd4_client_record_create(cstate->session->se_client);
2370 out:
2371         nfs4_unlock_state();
2372         return status;
2373 }
2374
2375 __be32
2376 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2377                   struct nfsd4_setclientid *setclid)
2378 {
2379         struct xdr_netobj       clname = setclid->se_name;
2380         nfs4_verifier           clverifier = setclid->se_verf;
2381         struct nfs4_client      *conf, *unconf, *new;
2382         __be32                  status;
2383         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2384
2385         /* Cases below refer to rfc 3530 section 14.2.33: */
2386         nfs4_lock_state();
2387         conf = find_confirmed_client_by_name(&clname, nn);
2388         if (conf) {
2389                 /* case 0: */
2390                 status = nfserr_clid_inuse;
2391                 if (clp_used_exchangeid(conf))
2392                         goto out;
2393                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2394                         char addr_str[INET6_ADDRSTRLEN];
2395                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2396                                  sizeof(addr_str));
2397                         dprintk("NFSD: setclientid: string in use by client "
2398                                 "at %s\n", addr_str);
2399                         goto out;
2400                 }
2401         }
2402         unconf = find_unconfirmed_client_by_name(&clname, nn);
2403         if (unconf)
2404                 expire_client(unconf);
2405         status = nfserr_jukebox;
2406         new = create_client(clname, rqstp, &clverifier);
2407         if (new == NULL)
2408                 goto out;
2409         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2410                 /* case 1: probable callback update */
2411                 copy_clid(new, conf);
2412         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2413                 gen_clid(new, nn);
2414         new->cl_minorversion = 0;
2415         gen_callback(new, setclid, rqstp);
2416         add_to_unconfirmed(new);
2417         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2418         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2419         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2420         status = nfs_ok;
2421 out:
2422         nfs4_unlock_state();
2423         return status;
2424 }
2425
2426
2427 __be32
2428 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2429                          struct nfsd4_compound_state *cstate,
2430                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2431 {
2432         struct nfs4_client *conf, *unconf;
2433         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2434         clientid_t * clid = &setclientid_confirm->sc_clientid;
2435         __be32 status;
2436         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2437
2438         if (STALE_CLIENTID(clid, nn))
2439                 return nfserr_stale_clientid;
2440         nfs4_lock_state();
2441
2442         conf = find_confirmed_client(clid, false, nn);
2443         unconf = find_unconfirmed_client(clid, false, nn);
2444         /*
2445          * We try hard to give out unique clientid's, so if we get an
2446          * attempt to confirm the same clientid with a different cred,
2447          * there's a bug somewhere.  Let's charitably assume it's our
2448          * bug.
2449          */
2450         status = nfserr_serverfault;
2451         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2452                 goto out;
2453         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2454                 goto out;
2455         /* cases below refer to rfc 3530 section 14.2.34: */
2456         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2457                 if (conf && !unconf) /* case 2: probable retransmit */
2458                         status = nfs_ok;
2459                 else /* case 4: client hasn't noticed we rebooted yet? */
2460                         status = nfserr_stale_clientid;
2461                 goto out;
2462         }
2463         status = nfs_ok;
2464         if (conf) { /* case 1: callback update */
2465                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2466                 nfsd4_probe_callback(conf);
2467                 expire_client(unconf);
2468         } else { /* case 3: normal case; new or rebooted client */
2469                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2470                 if (conf) {
2471                         status = mark_client_expired(conf);
2472                         if (status)
2473                                 goto out;
2474                         expire_client(conf);
2475                 }
2476                 move_to_confirmed(unconf);
2477                 nfsd4_probe_callback(unconf);
2478         }
2479 out:
2480         nfs4_unlock_state();
2481         return status;
2482 }
2483
2484 static struct nfs4_file *nfsd4_alloc_file(void)
2485 {
2486         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2487 }
2488
2489 /* OPEN Share state helper functions */
2490 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2491 {
2492         unsigned int hashval = file_hashval(ino);
2493
2494         atomic_set(&fp->fi_ref, 1);
2495         INIT_LIST_HEAD(&fp->fi_stateids);
2496         INIT_LIST_HEAD(&fp->fi_delegations);
2497         fp->fi_inode = igrab(ino);
2498         fp->fi_had_conflict = false;
2499         fp->fi_lease = NULL;
2500         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2501         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2502         spin_lock(&recall_lock);
2503         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2504         spin_unlock(&recall_lock);
2505 }
2506
2507 static void
2508 nfsd4_free_slab(struct kmem_cache **slab)
2509 {
2510         if (*slab == NULL)
2511                 return;
2512         kmem_cache_destroy(*slab);
2513         *slab = NULL;
2514 }
2515
2516 void
2517 nfsd4_free_slabs(void)
2518 {
2519         nfsd4_free_slab(&openowner_slab);
2520         nfsd4_free_slab(&lockowner_slab);
2521         nfsd4_free_slab(&file_slab);
2522         nfsd4_free_slab(&stateid_slab);
2523         nfsd4_free_slab(&deleg_slab);
2524 }
2525
2526 int
2527 nfsd4_init_slabs(void)
2528 {
2529         openowner_slab = kmem_cache_create("nfsd4_openowners",
2530                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2531         if (openowner_slab == NULL)
2532                 goto out_nomem;
2533         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2534                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2535         if (lockowner_slab == NULL)
2536                 goto out_nomem;
2537         file_slab = kmem_cache_create("nfsd4_files",
2538                         sizeof(struct nfs4_file), 0, 0, NULL);
2539         if (file_slab == NULL)
2540                 goto out_nomem;
2541         stateid_slab = kmem_cache_create("nfsd4_stateids",
2542                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2543         if (stateid_slab == NULL)
2544                 goto out_nomem;
2545         deleg_slab = kmem_cache_create("nfsd4_delegations",
2546                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2547         if (deleg_slab == NULL)
2548                 goto out_nomem;
2549         return 0;
2550 out_nomem:
2551         nfsd4_free_slabs();
2552         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2553         return -ENOMEM;
2554 }
2555
2556 void nfs4_free_openowner(struct nfs4_openowner *oo)
2557 {
2558         kfree(oo->oo_owner.so_owner.data);
2559         kmem_cache_free(openowner_slab, oo);
2560 }
2561
2562 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2563 {
2564         kfree(lo->lo_owner.so_owner.data);
2565         kmem_cache_free(lockowner_slab, lo);
2566 }
2567
2568 static void init_nfs4_replay(struct nfs4_replay *rp)
2569 {
2570         rp->rp_status = nfserr_serverfault;
2571         rp->rp_buflen = 0;
2572         rp->rp_buf = rp->rp_ibuf;
2573 }
2574
2575 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2576 {
2577         struct nfs4_stateowner *sop;
2578
2579         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2580         if (!sop)
2581                 return NULL;
2582
2583         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2584         if (!sop->so_owner.data) {
2585                 kmem_cache_free(slab, sop);
2586                 return NULL;
2587         }
2588         sop->so_owner.len = owner->len;
2589
2590         INIT_LIST_HEAD(&sop->so_stateids);
2591         sop->so_client = clp;
2592         init_nfs4_replay(&sop->so_replay);
2593         return sop;
2594 }
2595
2596 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2597 {
2598         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2599
2600         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2601         list_add(&oo->oo_perclient, &clp->cl_openowners);
2602 }
2603
2604 static struct nfs4_openowner *
2605 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2606         struct nfs4_openowner *oo;
2607
2608         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2609         if (!oo)
2610                 return NULL;
2611         oo->oo_owner.so_is_open_owner = 1;
2612         oo->oo_owner.so_seqid = open->op_seqid;
2613         oo->oo_flags = NFS4_OO_NEW;
2614         oo->oo_time = 0;
2615         oo->oo_last_closed_stid = NULL;
2616         INIT_LIST_HEAD(&oo->oo_close_lru);
2617         hash_openowner(oo, clp, strhashval);
2618         return oo;
2619 }
2620
2621 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2622         struct nfs4_openowner *oo = open->op_openowner;
2623
2624         stp->st_stid.sc_type = NFS4_OPEN_STID;
2625         INIT_LIST_HEAD(&stp->st_lockowners);
2626         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2627         list_add(&stp->st_perfile, &fp->fi_stateids);
2628         stp->st_stateowner = &oo->oo_owner;
2629         get_nfs4_file(fp);
2630         stp->st_file = fp;
2631         stp->st_access_bmap = 0;
2632         stp->st_deny_bmap = 0;
2633         set_access(open->op_share_access, stp);
2634         set_deny(open->op_share_deny, stp);
2635         stp->st_openstp = NULL;
2636 }
2637
2638 static void
2639 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2640 {
2641         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2642
2643         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2644
2645         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2646         oo->oo_time = get_seconds();
2647 }
2648
2649 static int
2650 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2651                                                         clientid_t *clid)
2652 {
2653         return (sop->so_owner.len == owner->len) &&
2654                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2655                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2656 }
2657
2658 static struct nfs4_openowner *
2659 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2660                         bool sessions, struct nfsd_net *nn)
2661 {
2662         struct nfs4_stateowner *so;
2663         struct nfs4_openowner *oo;
2664         struct nfs4_client *clp;
2665
2666         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2667                 if (!so->so_is_open_owner)
2668                         continue;
2669                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2670                         oo = openowner(so);
2671                         clp = oo->oo_owner.so_client;
2672                         if ((bool)clp->cl_minorversion != sessions)
2673                                 return NULL;
2674                         renew_client(oo->oo_owner.so_client);
2675                         return oo;
2676                 }
2677         }
2678         return NULL;
2679 }
2680
2681 /* search file_hashtbl[] for file */
2682 static struct nfs4_file *
2683 find_file(struct inode *ino)
2684 {
2685         unsigned int hashval = file_hashval(ino);
2686         struct nfs4_file *fp;
2687
2688         spin_lock(&recall_lock);
2689         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2690                 if (fp->fi_inode == ino) {
2691                         get_nfs4_file(fp);
2692                         spin_unlock(&recall_lock);
2693                         return fp;
2694                 }
2695         }
2696         spin_unlock(&recall_lock);
2697         return NULL;
2698 }
2699
2700 /*
2701  * Called to check deny when READ with all zero stateid or
2702  * WRITE with all zero or all one stateid
2703  */
2704 static __be32
2705 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2706 {
2707         struct inode *ino = current_fh->fh_dentry->d_inode;
2708         struct nfs4_file *fp;
2709         struct nfs4_ol_stateid *stp;
2710         __be32 ret;
2711
2712         fp = find_file(ino);
2713         if (!fp)
2714                 return nfs_ok;
2715         ret = nfserr_locked;
2716         /* Search for conflicting share reservations */
2717         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2718                 if (test_deny(deny_type, stp) ||
2719                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2720                         goto out;
2721         }
2722         ret = nfs_ok;
2723 out:
2724         put_nfs4_file(fp);
2725         return ret;
2726 }
2727
2728 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2729 {
2730         struct nfs4_client *clp = dp->dl_stid.sc_client;
2731         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2732
2733         /* We're assuming the state code never drops its reference
2734          * without first removing the lease.  Since we're in this lease
2735          * callback (and since the lease code is serialized by the kernel
2736          * lock) we know the server hasn't removed the lease yet, we know
2737          * it's safe to take a reference: */
2738         atomic_inc(&dp->dl_count);
2739
2740         list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
2741
2742         /* Only place dl_time is set; protected by i_lock: */
2743         dp->dl_time = get_seconds();
2744
2745         nfsd4_cb_recall(dp);
2746 }
2747
2748 /* Called from break_lease() with i_lock held. */
2749 static void nfsd_break_deleg_cb(struct file_lock *fl)
2750 {
2751         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2752         struct nfs4_delegation *dp;
2753
2754         if (!fp) {
2755                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2756                 return;
2757         }
2758         if (fp->fi_had_conflict) {
2759                 WARN(1, "duplicate break on %p\n", fp);
2760                 return;
2761         }
2762         /*
2763          * We don't want the locks code to timeout the lease for us;
2764          * we'll remove it ourself if a delegation isn't returned
2765          * in time:
2766          */
2767         fl->fl_break_time = 0;
2768
2769         spin_lock(&recall_lock);
2770         fp->fi_had_conflict = true;
2771         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2772                 nfsd_break_one_deleg(dp);
2773         spin_unlock(&recall_lock);
2774 }
2775
2776 static
2777 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2778 {
2779         if (arg & F_UNLCK)
2780                 return lease_modify(onlist, arg);
2781         else
2782                 return -EAGAIN;
2783 }
2784
2785 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2786         .lm_break = nfsd_break_deleg_cb,
2787         .lm_change = nfsd_change_deleg_cb,
2788 };
2789
2790 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2791 {
2792         if (nfsd4_has_session(cstate))
2793                 return nfs_ok;
2794         if (seqid == so->so_seqid - 1)
2795                 return nfserr_replay_me;
2796         if (seqid == so->so_seqid)
2797                 return nfs_ok;
2798         return nfserr_bad_seqid;
2799 }
2800
2801 __be32
2802 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2803                     struct nfsd4_open *open, struct nfsd_net *nn)
2804 {
2805         clientid_t *clientid = &open->op_clientid;
2806         struct nfs4_client *clp = NULL;
2807         unsigned int strhashval;
2808         struct nfs4_openowner *oo = NULL;
2809         __be32 status;
2810
2811         if (STALE_CLIENTID(&open->op_clientid, nn))
2812                 return nfserr_stale_clientid;
2813         /*
2814          * In case we need it later, after we've already created the
2815          * file and don't want to risk a further failure:
2816          */
2817         open->op_file = nfsd4_alloc_file();
2818         if (open->op_file == NULL)
2819                 return nfserr_jukebox;
2820
2821         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2822         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2823         open->op_openowner = oo;
2824         if (!oo) {
2825                 clp = find_confirmed_client(clientid, cstate->minorversion,
2826                                             nn);
2827                 if (clp == NULL)
2828                         return nfserr_expired;
2829                 goto new_owner;
2830         }
2831         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2832                 /* Replace unconfirmed owners without checking for replay. */
2833                 clp = oo->oo_owner.so_client;
2834                 release_openowner(oo);
2835                 open->op_openowner = NULL;
2836                 goto new_owner;
2837         }
2838         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2839         if (status)
2840                 return status;
2841         clp = oo->oo_owner.so_client;
2842         goto alloc_stateid;
2843 new_owner:
2844         oo = alloc_init_open_stateowner(strhashval, clp, open);
2845         if (oo == NULL)
2846                 return nfserr_jukebox;
2847         open->op_openowner = oo;
2848 alloc_stateid:
2849         open->op_stp = nfs4_alloc_stateid(clp);
2850         if (!open->op_stp)
2851                 return nfserr_jukebox;
2852         return nfs_ok;
2853 }
2854
2855 static inline __be32
2856 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2857 {
2858         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2859                 return nfserr_openmode;
2860         else
2861                 return nfs_ok;
2862 }
2863
2864 static int share_access_to_flags(u32 share_access)
2865 {
2866         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2867 }
2868
2869 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2870 {
2871         struct nfs4_stid *ret;
2872
2873         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2874         if (!ret)
2875                 return NULL;
2876         return delegstateid(ret);
2877 }
2878
2879 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2880 {
2881         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2882                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2883 }
2884
2885 static __be32
2886 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
2887                 struct nfs4_delegation **dp)
2888 {
2889         int flags;
2890         __be32 status = nfserr_bad_stateid;
2891
2892         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2893         if (*dp == NULL)
2894                 goto out;
2895         flags = share_access_to_flags(open->op_share_access);
2896         status = nfs4_check_delegmode(*dp, flags);
2897         if (status)
2898                 *dp = NULL;
2899 out:
2900         if (!nfsd4_is_deleg_cur(open))
2901                 return nfs_ok;
2902         if (status)
2903                 return status;
2904         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2905         return nfs_ok;
2906 }
2907
2908 static __be32
2909 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2910 {
2911         struct nfs4_ol_stateid *local;
2912         struct nfs4_openowner *oo = open->op_openowner;
2913
2914         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2915                 /* ignore lock owners */
2916                 if (local->st_stateowner->so_is_open_owner == 0)
2917                         continue;
2918                 /* remember if we have seen this open owner */
2919                 if (local->st_stateowner == &oo->oo_owner)
2920                         *stpp = local;
2921                 /* check for conflicting share reservations */
2922                 if (!test_share(local, open))
2923                         return nfserr_share_denied;
2924         }
2925         return nfs_ok;
2926 }
2927
2928 static inline int nfs4_access_to_access(u32 nfs4_access)
2929 {
2930         int flags = 0;
2931
2932         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2933                 flags |= NFSD_MAY_READ;
2934         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2935                 flags |= NFSD_MAY_WRITE;
2936         return flags;
2937 }
2938
2939 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2940                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2941 {
2942         __be32 status;
2943         int oflag = nfs4_access_to_omode(open->op_share_access);
2944         int access = nfs4_access_to_access(open->op_share_access);
2945
2946         if (!fp->fi_fds[oflag]) {
2947                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2948                         &fp->fi_fds[oflag]);
2949                 if (status)
2950                         return status;
2951         }
2952         nfs4_file_get_access(fp, oflag);
2953
2954         return nfs_ok;
2955 }
2956
2957 static inline __be32
2958 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2959                 struct nfsd4_open *open)
2960 {
2961         struct iattr iattr = {
2962                 .ia_valid = ATTR_SIZE,
2963                 .ia_size = 0,
2964         };
2965         if (!open->op_truncate)
2966                 return 0;
2967         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2968                 return nfserr_inval;
2969         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2970 }
2971
2972 static __be32
2973 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)
2974 {
2975         u32 op_share_access = open->op_share_access;
2976         bool new_access;
2977         __be32 status;
2978
2979         new_access = !test_access(op_share_access, stp);
2980         if (new_access) {
2981                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2982                 if (status)
2983                         return status;
2984         }
2985         status = nfsd4_truncate(rqstp, cur_fh, open);
2986         if (status) {
2987                 if (new_access) {
2988                         int oflag = nfs4_access_to_omode(op_share_access);
2989                         nfs4_file_put_access(fp, oflag);
2990                 }
2991                 return status;
2992         }
2993         /* remember the open */
2994         set_access(op_share_access, stp);
2995         set_deny(open->op_share_deny, stp);
2996
2997         return nfs_ok;
2998 }
2999
3000
3001 static void
3002 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3003 {
3004         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3005 }
3006
3007 /* Should we give out recallable state?: */
3008 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3009 {
3010         if (clp->cl_cb_state == NFSD4_CB_UP)
3011                 return true;
3012         /*
3013          * In the sessions case, since we don't have to establish a
3014          * separate connection for callbacks, we assume it's OK
3015          * until we hear otherwise:
3016          */
3017         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3018 }
3019
3020 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
3021 {
3022         struct file_lock *fl;
3023
3024         fl = locks_alloc_lock();
3025         if (!fl)
3026                 return NULL;
3027         locks_init_lock(fl);
3028         fl->fl_lmops = &nfsd_lease_mng_ops;
3029         fl->fl_flags = FL_DELEG;
3030         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3031         fl->fl_end = OFFSET_MAX;
3032         fl->fl_owner = (fl_owner_t)(dp->dl_file);
3033         fl->fl_pid = current->tgid;
3034         return fl;
3035 }
3036
3037 static int nfs4_setlease(struct nfs4_delegation *dp)
3038 {
3039         struct nfs4_file *fp = dp->dl_file;
3040         struct file_lock *fl;
3041         int status;
3042
3043         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
3044         if (!fl)
3045                 return -ENOMEM;
3046         fl->fl_file = find_readable_file(fp);
3047         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
3048         if (status)
3049                 goto out_free;
3050         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3051         fp->fi_lease = fl;
3052         fp->fi_deleg_file = get_file(fl->fl_file);
3053         atomic_set(&fp->fi_delegees, 1);
3054         list_add(&dp->dl_perfile, &fp->fi_delegations);
3055         return 0;
3056 out_free:
3057         locks_free_lock(fl);
3058         return status;
3059 }
3060
3061 static int nfs4_set_delegation(struct nfs4_delegation *dp, struct nfs4_file *fp)
3062 {
3063         int status;
3064
3065         if (fp->fi_had_conflict)
3066                 return -EAGAIN;
3067         get_nfs4_file(fp);
3068         dp->dl_file = fp;
3069         if (!fp->fi_lease) {
3070                 status = nfs4_setlease(dp);
3071                 if (status)
3072                         goto out_free;
3073                 return 0;
3074         }
3075         spin_lock(&recall_lock);
3076         if (fp->fi_had_conflict) {
3077                 spin_unlock(&recall_lock);
3078                 status = -EAGAIN;
3079                 goto out_free;
3080         }
3081         atomic_inc(&fp->fi_delegees);
3082         list_add(&dp->dl_perfile, &fp->fi_delegations);
3083         spin_unlock(&recall_lock);
3084         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3085         return 0;
3086 out_free:
3087         put_nfs4_file(fp);
3088         dp->dl_file = fp;
3089         return status;
3090 }
3091
3092 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3093 {
3094         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3095         if (status == -EAGAIN)
3096                 open->op_why_no_deleg = WND4_CONTENTION;
3097         else {
3098                 open->op_why_no_deleg = WND4_RESOURCE;
3099                 switch (open->op_deleg_want) {
3100                 case NFS4_SHARE_WANT_READ_DELEG:
3101                 case NFS4_SHARE_WANT_WRITE_DELEG:
3102                 case NFS4_SHARE_WANT_ANY_DELEG:
3103                         break;
3104                 case NFS4_SHARE_WANT_CANCEL:
3105                         open->op_why_no_deleg = WND4_CANCELLED;
3106                         break;
3107                 case NFS4_SHARE_WANT_NO_DELEG:
3108                         WARN_ON_ONCE(1);
3109                 }
3110         }
3111 }
3112
3113 /*
3114  * Attempt to hand out a delegation.
3115  *
3116  * Note we don't support write delegations, and won't until the vfs has
3117  * proper support for them.
3118  */
3119 static void
3120 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
3121                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
3122 {
3123         struct nfs4_delegation *dp;
3124         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
3125         int cb_up;
3126         int status = 0;
3127
3128         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3129         open->op_recall = 0;
3130         switch (open->op_claim_type) {
3131                 case NFS4_OPEN_CLAIM_PREVIOUS:
3132                         if (!cb_up)
3133                                 open->op_recall = 1;
3134                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3135                                 goto out_no_deleg;
3136                         break;
3137                 case NFS4_OPEN_CLAIM_NULL:
3138                 case NFS4_OPEN_CLAIM_FH:
3139                         /*
3140                          * Let's not give out any delegations till everyone's
3141                          * had the chance to reclaim theirs....
3142                          */
3143                         if (locks_in_grace(net))
3144                                 goto out_no_deleg;
3145                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3146                                 goto out_no_deleg;
3147                         /*
3148                          * Also, if the file was opened for write or
3149                          * create, there's a good chance the client's
3150                          * about to write to it, resulting in an
3151                          * immediate recall (since we don't support
3152                          * write delegations):
3153                          */
3154                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3155                                 goto out_no_deleg;
3156                         if (open->op_create == NFS4_OPEN_CREATE)
3157                                 goto out_no_deleg;
3158                         break;
3159                 default:
3160                         goto out_no_deleg;
3161         }
3162         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh);
3163         if (dp == NULL)
3164                 goto out_no_deleg;
3165         status = nfs4_set_delegation(dp, stp->st_file);
3166         if (status)
3167                 goto out_free;
3168
3169         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3170
3171         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3172                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3173         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3174         return;
3175 out_free:
3176         remove_stid(&dp->dl_stid);
3177         nfs4_put_delegation(dp);
3178 out_no_deleg:
3179         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3180         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3181             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3182                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3183                 open->op_recall = 1;
3184         }
3185
3186         /* 4.1 client asking for a delegation? */
3187         if (open->op_deleg_want)
3188                 nfsd4_open_deleg_none_ext(open, status);
3189         return;
3190 }
3191
3192 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3193                                         struct nfs4_delegation *dp)
3194 {
3195         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3196             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3197                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3198                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3199         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3200                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3201                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3202                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3203         }
3204         /* Otherwise the client must be confused wanting a delegation
3205          * it already has, therefore we don't return
3206          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3207          */
3208 }
3209
3210 /*
3211  * called with nfs4_lock_state() held.
3212  */
3213 __be32
3214 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3215 {
3216         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3217         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3218         struct nfs4_file *fp = NULL;
3219         struct inode *ino = current_fh->fh_dentry->d_inode;
3220         struct nfs4_ol_stateid *stp = NULL;
3221         struct nfs4_delegation *dp = NULL;
3222         __be32 status;
3223
3224         /*
3225          * Lookup file; if found, lookup stateid and check open request,
3226          * and check for delegations in the process of being recalled.
3227          * If not found, create the nfs4_file struct
3228          */
3229         fp = find_file(ino);
3230         if (fp) {
3231                 if ((status = nfs4_check_open(fp, open, &stp)))
3232                         goto out;
3233                 status = nfs4_check_deleg(cl, open, &dp);
3234                 if (status)
3235                         goto out;
3236         } else {
3237                 status = nfserr_bad_stateid;
3238                 if (nfsd4_is_deleg_cur(open))
3239                         goto out;
3240                 status = nfserr_jukebox;
3241                 fp = open->op_file;
3242                 open->op_file = NULL;
3243                 nfsd4_init_file(fp, ino);
3244         }
3245
3246         /*
3247          * OPEN the file, or upgrade an existing OPEN.
3248          * If truncate fails, the OPEN fails.
3249          */
3250         if (stp) {
3251                 /* Stateid was found, this is an OPEN upgrade */
3252                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3253                 if (status)
3254                         goto out;
3255         } else {
3256                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3257                 if (status)
3258                         goto out;
3259                 status = nfsd4_truncate(rqstp, current_fh, open);
3260                 if (status)
3261                         goto out;
3262                 stp = open->op_stp;
3263                 open->op_stp = NULL;
3264                 init_open_stateid(stp, fp, open);
3265         }
3266         update_stateid(&stp->st_stid.sc_stateid);
3267         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3268
3269         if (nfsd4_has_session(&resp->cstate)) {
3270                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3271
3272                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3273                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3274                         open->op_why_no_deleg = WND4_NOT_WANTED;
3275                         goto nodeleg;
3276                 }
3277         }
3278
3279         /*
3280         * Attempt to hand out a delegation. No error return, because the
3281         * OPEN succeeds even if we fail.
3282         */
3283         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3284 nodeleg:
3285         status = nfs_ok;
3286
3287         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3288                 STATEID_VAL(&stp->st_stid.sc_stateid));
3289 out:
3290         /* 4.1 client trying to upgrade/downgrade delegation? */
3291         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3292             open->op_deleg_want)
3293                 nfsd4_deleg_xgrade_none_ext(open, dp);
3294
3295         if (fp)
3296                 put_nfs4_file(fp);
3297         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3298                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3299         /*
3300         * To finish the open response, we just need to set the rflags.
3301         */
3302         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3303         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3304             !nfsd4_has_session(&resp->cstate))
3305                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3306
3307         return status;
3308 }
3309
3310 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3311 {
3312         if (open->op_openowner) {
3313                 struct nfs4_openowner *oo = open->op_openowner;
3314
3315                 if (!list_empty(&oo->oo_owner.so_stateids))
3316                         list_del_init(&oo->oo_close_lru);
3317                 if (oo->oo_flags & NFS4_OO_NEW) {
3318                         if (status) {
3319                                 release_openowner(oo);
3320                                 open->op_openowner = NULL;
3321                         } else
3322                                 oo->oo_flags &= ~NFS4_OO_NEW;
3323                 }
3324         }
3325         if (open->op_file)
3326                 nfsd4_free_file(open->op_file);
3327         if (open->op_stp)
3328                 free_generic_stateid(open->op_stp);
3329 }
3330
3331 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3332 {
3333         struct nfs4_client *found;
3334
3335         if (STALE_CLIENTID(clid, nn))
3336                 return nfserr_stale_clientid;
3337         found = find_confirmed_client(clid, session, nn);
3338         if (clp)
3339                 *clp = found;
3340         return found ? nfs_ok : nfserr_expired;
3341 }
3342
3343 __be32
3344 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3345             clientid_t *clid)
3346 {
3347         struct nfs4_client *clp;
3348         __be32 status;
3349         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3350
3351         nfs4_lock_state();
3352         dprintk("process_renew(%08x/%08x): starting\n",
3353                         clid->cl_boot, clid->cl_id);
3354         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3355         if (status)
3356                 goto out;
3357         status = nfserr_cb_path_down;
3358         if (!list_empty(&clp->cl_delegations)
3359                         && clp->cl_cb_state != NFSD4_CB_UP)
3360                 goto out;
3361         status = nfs_ok;
3362 out:
3363         nfs4_unlock_state();
3364         return status;
3365 }
3366
3367 static void
3368 nfsd4_end_grace(struct nfsd_net *nn)
3369 {
3370         /* do nothing if grace period already ended */
3371         if (nn->grace_ended)
3372                 return;
3373
3374         dprintk("NFSD: end of grace period\n");
3375         nn->grace_ended = true;
3376         nfsd4_record_grace_done(nn, nn->boot_time);
3377         locks_end_grace(&nn->nfsd4_manager);
3378         /*
3379          * Now that every NFSv4 client has had the chance to recover and
3380          * to see the (possibly new, possibly shorter) lease time, we
3381          * can safely set the next grace time to the current lease time:
3382          */
3383         nn->nfsd4_grace = nn->nfsd4_lease;
3384 }
3385
3386 static time_t
3387 nfs4_laundromat(struct nfsd_net *nn)
3388 {
3389         struct nfs4_client *clp;
3390         struct nfs4_openowner *oo;
3391         struct nfs4_delegation *dp;
3392         struct list_head *pos, *next, reaplist;
3393         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3394         time_t t, clientid_val = nn->nfsd4_lease;
3395         time_t u, test_val = nn->nfsd4_lease;
3396
3397         nfs4_lock_state();
3398
3399         dprintk("NFSD: laundromat service - starting\n");
3400         nfsd4_end_grace(nn);
3401         INIT_LIST_HEAD(&reaplist);
3402         spin_lock(&nn->client_lock);
3403         list_for_each_safe(pos, next, &nn->client_lru) {
3404                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3405                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3406                         t = clp->cl_time - cutoff;
3407                         if (clientid_val > t)
3408                                 clientid_val = t;
3409                         break;
3410                 }
3411                 if (mark_client_expired_locked(clp)) {
3412                         dprintk("NFSD: client in use (clientid %08x)\n",
3413                                 clp->cl_clientid.cl_id);
3414                         continue;
3415                 }
3416                 list_move(&clp->cl_lru, &reaplist);
3417         }
3418         spin_unlock(&nn->client_lock);
3419         list_for_each_safe(pos, next, &reaplist) {
3420                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3421                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3422                         clp->cl_clientid.cl_id);
3423                 expire_client(clp);
3424         }
3425         spin_lock(&recall_lock);
3426         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3427                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3428                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3429                         continue;
3430                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3431                         u = dp->dl_time - cutoff;
3432                         if (test_val > u)
3433                                 test_val = u;
3434                         break;
3435                 }
3436                 list_move(&dp->dl_recall_lru, &reaplist);
3437         }
3438         spin_unlock(&recall_lock);
3439         list_for_each_safe(pos, next, &reaplist) {
3440                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3441                 revoke_delegation(dp);
3442         }
3443         test_val = nn->nfsd4_lease;
3444         list_for_each_safe(pos, next, &nn->close_lru) {
3445                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3446                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3447                         u = oo->oo_time - cutoff;
3448                         if (test_val > u)
3449                                 test_val = u;
3450                         break;
3451                 }
3452                 release_openowner(oo);
3453         }
3454         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3455                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3456         nfs4_unlock_state();
3457         return clientid_val;
3458 }
3459
3460 static struct workqueue_struct *laundry_wq;
3461 static void laundromat_main(struct work_struct *);
3462
3463 static void
3464 laundromat_main(struct work_struct *laundry)
3465 {
3466         time_t t;
3467         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3468                                                   work);
3469         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3470                                            laundromat_work);
3471
3472         t = nfs4_laundromat(nn);
3473         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3474         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3475 }
3476
3477 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3478 {
3479         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3480                 return nfserr_bad_stateid;
3481         return nfs_ok;
3482 }
3483
3484 static inline int
3485 access_permit_read(struct nfs4_ol_stateid *stp)
3486 {
3487         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3488                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3489                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3490 }
3491
3492 static inline int
3493 access_permit_write(struct nfs4_ol_stateid *stp)
3494 {
3495         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3496                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3497 }
3498
3499 static
3500 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3501 {
3502         __be32 status = nfserr_openmode;
3503
3504         /* For lock stateid's, we test the parent open, not the lock: */
3505         if (stp->st_openstp)
3506                 stp = stp->st_openstp;
3507         if ((flags & WR_STATE) && !access_permit_write(stp))
3508                 goto out;
3509         if ((flags & RD_STATE) && !access_permit_read(stp))
3510                 goto out;
3511         status = nfs_ok;
3512 out:
3513         return status;
3514 }
3515
3516 static inline __be32
3517 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3518 {
3519         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3520                 return nfs_ok;
3521         else if (locks_in_grace(net)) {
3522                 /* Answer in remaining cases depends on existence of
3523                  * conflicting state; so we must wait out the grace period. */
3524                 return nfserr_grace;
3525         } else if (flags & WR_STATE)
3526                 return nfs4_share_conflict(current_fh,
3527                                 NFS4_SHARE_DENY_WRITE);
3528         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3529                 return nfs4_share_conflict(current_fh,
3530                                 NFS4_SHARE_DENY_READ);
3531 }
3532
3533 /*
3534  * Allow READ/WRITE during grace period on recovered state only for files
3535  * that are not able to provide mandatory locking.
3536  */
3537 static inline int
3538 grace_disallows_io(struct net *net, struct inode *inode)
3539 {
3540         return locks_in_grace(net) && mandatory_lock(inode);
3541 }
3542
3543 /* Returns true iff a is later than b: */
3544 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3545 {
3546         return (s32)(a->si_generation - b->si_generation) > 0;
3547 }
3548
3549 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3550 {
3551         /*
3552          * When sessions are used the stateid generation number is ignored
3553          * when it is zero.
3554          */
3555         if (has_session && in->si_generation == 0)
3556                 return nfs_ok;
3557
3558         if (in->si_generation == ref->si_generation)
3559                 return nfs_ok;
3560
3561         /* If the client sends us a stateid from the future, it's buggy: */
3562         if (stateid_generation_after(in, ref))
3563                 return nfserr_bad_stateid;
3564         /*
3565          * However, we could see a stateid from the past, even from a
3566          * non-buggy client.  For example, if the client sends a lock
3567          * while some IO is outstanding, the lock may bump si_generation
3568          * while the IO is still in flight.  The client could avoid that
3569          * situation by waiting for responses on all the IO requests,
3570          * but better performance may result in retrying IO that
3571          * receives an old_stateid error if requests are rarely
3572          * reordered in flight:
3573          */
3574         return nfserr_old_stateid;
3575 }
3576
3577 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3578 {
3579         struct nfs4_stid *s;
3580         struct nfs4_ol_stateid *ols;
3581         __be32 status;
3582
3583         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3584                 return nfserr_bad_stateid;
3585         /* Client debugging aid. */
3586         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3587                 char addr_str[INET6_ADDRSTRLEN];
3588                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3589                                  sizeof(addr_str));
3590                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3591                                         "with incorrect client ID\n", addr_str);
3592                 return nfserr_bad_stateid;
3593         }
3594         s = find_stateid(cl, stateid);
3595         if (!s)
3596                 return nfserr_bad_stateid;
3597         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3598         if (status)
3599                 return status;
3600         switch (s->sc_type) {
3601         case NFS4_DELEG_STID:
3602                 return nfs_ok;
3603         case NFS4_REVOKED_DELEG_STID:
3604                 return nfserr_deleg_revoked;
3605         case NFS4_OPEN_STID:
3606         case NFS4_LOCK_STID:
3607                 ols = openlockstateid(s);
3608                 if (ols->st_stateowner->so_is_open_owner
3609                                 && !(openowner(ols->st_stateowner)->oo_flags
3610                                                 & NFS4_OO_CONFIRMED))
3611                         return nfserr_bad_stateid;
3612                 return nfs_ok;
3613         default:
3614                 printk("unknown stateid type %x\n", s->sc_type);
3615         case NFS4_CLOSED_STID:
3616                 return nfserr_bad_stateid;
3617         }
3618 }
3619
3620 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3621                                    struct nfs4_stid **s, bool sessions,
3622                                    struct nfsd_net *nn)
3623 {
3624         struct nfs4_client *cl;
3625         __be32 status;
3626
3627         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3628                 return nfserr_bad_stateid;
3629         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3630                                                         nn, &cl);
3631         if (status == nfserr_stale_clientid)
3632                 return nfserr_stale_stateid;
3633         if (status)
3634                 return status;
3635         *s = find_stateid_by_type(cl, stateid, typemask);
3636         if (!*s)
3637                 return nfserr_bad_stateid;
3638         return nfs_ok;
3639 }
3640
3641 /*
3642 * Checks for stateid operations
3643 */
3644 __be32
3645 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3646                            stateid_t *stateid, int flags, struct file **filpp)
3647 {
3648         struct nfs4_stid *s;
3649         struct nfs4_ol_stateid *stp = NULL;
3650         struct nfs4_delegation *dp = NULL;
3651         struct svc_fh *current_fh = &cstate->current_fh;
3652         struct inode *ino = current_fh->fh_dentry->d_inode;
3653         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3654         __be32 status;
3655
3656         if (filpp)
3657                 *filpp = NULL;
3658
3659         if (grace_disallows_io(net, ino))
3660                 return nfserr_grace;
3661
3662         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3663                 return check_special_stateids(net, current_fh, stateid, flags);
3664
3665         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3666                                       &s, cstate->minorversion, nn);
3667         if (status)
3668                 return status;
3669         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3670         if (status)
3671                 goto out;
3672         switch (s->sc_type) {
3673         case NFS4_DELEG_STID:
3674                 dp = delegstateid(s);
3675                 status = nfs4_check_delegmode(dp, flags);
3676                 if (status)
3677                         goto out;
3678                 if (filpp) {
3679                         *filpp = dp->dl_file->fi_deleg_file;
3680                         if (!*filpp) {
3681                                 WARN_ON_ONCE(1);
3682                                 status = nfserr_serverfault;
3683                                 goto out;
3684                         }
3685                 }
3686                 break;
3687         case NFS4_OPEN_STID:
3688         case NFS4_LOCK_STID:
3689                 stp = openlockstateid(s);
3690                 status = nfs4_check_fh(current_fh, stp);
3691                 if (status)
3692                         goto out;
3693                 if (stp->st_stateowner->so_is_open_owner
3694                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3695                         goto out;
3696                 status = nfs4_check_openmode(stp, flags);
3697                 if (status)
3698                         goto out;
3699                 if (filpp) {
3700                         if (flags & RD_STATE)
3701                                 *filpp = find_readable_file(stp->st_file);
3702                         else
3703                                 *filpp = find_writeable_file(stp->st_file);
3704                 }
3705                 break;
3706         default:
3707                 return nfserr_bad_stateid;
3708         }
3709         status = nfs_ok;
3710 out:
3711         return status;
3712 }
3713
3714 static __be32
3715 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3716 {
3717         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
3718
3719         if (check_for_locks(stp->st_file, lo))
3720                 return nfserr_locks_held;
3721         /*
3722          * Currently there's a 1-1 lock stateid<->lockowner
3723          * correspondance, and we have to delete the lockowner when we
3724          * delete the lock stateid:
3725          */
3726         release_lockowner(lo);
3727         return nfs_ok;
3728 }
3729
3730 /*
3731  * Test if the stateid is valid
3732  */
3733 __be32
3734 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3735                    struct nfsd4_test_stateid *test_stateid)
3736 {
3737         struct nfsd4_test_stateid_id *stateid;
3738         struct nfs4_client *cl = cstate->session->se_client;
3739
3740         nfs4_lock_state();
3741         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3742                 stateid->ts_id_status =
3743                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3744         nfs4_unlock_state();
3745
3746         return nfs_ok;
3747 }
3748
3749 __be32
3750 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3751                    struct nfsd4_free_stateid *free_stateid)
3752 {
3753         stateid_t *stateid = &free_stateid->fr_stateid;
3754         struct nfs4_stid *s;
3755         struct nfs4_delegation *dp;
3756         struct nfs4_client *cl = cstate->session->se_client;
3757         __be32 ret = nfserr_bad_stateid;
3758
3759         nfs4_lock_state();
3760         s = find_stateid(cl, stateid);
3761         if (!s)
3762                 goto out;
3763         switch (s->sc_type) {
3764         case NFS4_DELEG_STID:
3765                 ret = nfserr_locks_held;
3766                 goto out;
3767         case NFS4_OPEN_STID:
3768         case NFS4_LOCK_STID:
3769                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3770                 if (ret)
3771                         goto out;
3772                 if (s->sc_type == NFS4_LOCK_STID)
3773                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3774                 else
3775                         ret = nfserr_locks_held;
3776                 break;
3777         case NFS4_REVOKED_DELEG_STID:
3778                 dp = delegstateid(s);
3779                 destroy_revoked_delegation(dp);
3780                 ret = nfs_ok;
3781                 break;
3782         default:
3783                 ret = nfserr_bad_stateid;
3784         }
3785 out:
3786         nfs4_unlock_state();
3787         return ret;
3788 }
3789
3790 static inline int
3791 setlkflg (int type)
3792 {
3793         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3794                 RD_STATE : WR_STATE;
3795 }
3796
3797 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3798 {
3799         struct svc_fh *current_fh = &cstate->current_fh;
3800         struct nfs4_stateowner *sop = stp->st_stateowner;
3801         __be32 status;
3802
3803         status = nfsd4_check_seqid(cstate, sop, seqid);
3804         if (status)
3805                 return status;
3806         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
3807                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
3808                 /*
3809                  * "Closed" stateid's exist *only* to return
3810                  * nfserr_replay_me from the previous step, and
3811                  * revoked delegations are kept only for free_stateid.
3812                  */
3813                 return nfserr_bad_stateid;
3814         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3815         if (status)
3816                 return status;
3817         return nfs4_check_fh(current_fh, stp);
3818 }
3819
3820 /*
3821  * Checks for sequence id mutating operations.
3822  */
3823 static __be32
3824 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3825                          stateid_t *stateid, char typemask,
3826                          struct nfs4_ol_stateid **stpp,
3827                          struct nfsd_net *nn)
3828 {
3829         __be32 status;
3830         struct nfs4_stid *s;
3831
3832         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3833                 seqid, STATEID_VAL(stateid));
3834
3835         *stpp = NULL;
3836         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3837                                       cstate->minorversion, nn);
3838         if (status)
3839                 return status;
3840         *stpp = openlockstateid(s);
3841         if (!nfsd4_has_session(cstate))
3842                 cstate->replay_owner = (*stpp)->st_stateowner;
3843
3844         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3845 }
3846
3847 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3848                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3849 {
3850         __be32 status;
3851         struct nfs4_openowner *oo;
3852
3853         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3854                                                 NFS4_OPEN_STID, stpp, nn);
3855         if (status)
3856                 return status;
3857         oo = openowner((*stpp)->st_stateowner);
3858         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3859                 return nfserr_bad_stateid;
3860         return nfs_ok;
3861 }
3862
3863 __be32
3864 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3865                    struct nfsd4_open_confirm *oc)
3866 {
3867         __be32 status;
3868         struct nfs4_openowner *oo;
3869         struct nfs4_ol_stateid *stp;
3870         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3871
3872         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
3873                         cstate->current_fh.fh_dentry);
3874
3875         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3876         if (status)
3877                 return status;
3878
3879         nfs4_lock_state();
3880
3881         status = nfs4_preprocess_seqid_op(cstate,
3882                                         oc->oc_seqid, &oc->oc_req_stateid,
3883                                         NFS4_OPEN_STID, &stp, nn);
3884         if (status)
3885                 goto out;
3886         oo = openowner(stp->st_stateowner);
3887         status = nfserr_bad_stateid;
3888         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3889                 goto out;
3890         oo->oo_flags |= NFS4_OO_CONFIRMED;
3891         update_stateid(&stp->st_stid.sc_stateid);
3892         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3893         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3894                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3895
3896         nfsd4_client_record_create(oo->oo_owner.so_client);
3897         status = nfs_ok;
3898 out:
3899         nfsd4_bump_seqid(cstate, status);
3900         if (!cstate->replay_owner)
3901                 nfs4_unlock_state();
3902         return status;
3903 }
3904
3905 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3906 {
3907         if (!test_access(access, stp))
3908                 return;
3909         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3910         clear_access(access, stp);
3911 }
3912
3913 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3914 {
3915         switch (to_access) {
3916         case NFS4_SHARE_ACCESS_READ:
3917                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3918                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3919                 break;
3920         case NFS4_SHARE_ACCESS_WRITE:
3921                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3922                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3923                 break;
3924         case NFS4_SHARE_ACCESS_BOTH:
3925                 break;
3926         default:
3927                 WARN_ON_ONCE(1);
3928         }
3929 }
3930
3931 static void
3932 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3933 {
3934         int i;
3935         for (i = 0; i < 4; i++) {
3936                 if ((i & deny) != i)
3937                         clear_deny(i, stp);
3938         }
3939 }
3940
3941 __be32
3942 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3943                      struct nfsd4_compound_state *cstate,
3944                      struct nfsd4_open_downgrade *od)
3945 {
3946         __be32 status;
3947         struct nfs4_ol_stateid *stp;
3948         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3949
3950         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
3951                         cstate->current_fh.fh_dentry);
3952
3953         /* We don't yet support WANT bits: */
3954         if (od->od_deleg_want)
3955                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3956                         od->od_deleg_want);
3957
3958         nfs4_lock_state();
3959         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3960                                         &od->od_stateid, &stp, nn);
3961         if (status)
3962                 goto out;
3963         status = nfserr_inval;
3964         if (!test_access(od->od_share_access, stp)) {
3965                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3966                         stp->st_access_bmap, od->od_share_access);
3967                 goto out;
3968         }
3969         if (!test_deny(od->od_share_deny, stp)) {
3970                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3971                         stp->st_deny_bmap, od->od_share_deny);
3972                 goto out;
3973         }
3974         nfs4_stateid_downgrade(stp, od->od_share_access);
3975
3976         reset_union_bmap_deny(od->od_share_deny, stp);
3977
3978         update_stateid(&stp->st_stid.sc_stateid);
3979         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3980         status = nfs_ok;
3981 out:
3982         nfsd4_bump_seqid(cstate, status);
3983         if (!cstate->replay_owner)
3984                 nfs4_unlock_state();
3985         return status;
3986 }
3987
3988 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3989 {
3990         unhash_open_stateid(s);
3991         s->st_stid.sc_type = NFS4_CLOSED_STID;
3992 }
3993
3994 /*
3995  * nfs4_unlock_state() called after encode
3996  */
3997 __be32
3998 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3999             struct nfsd4_close *close)
4000 {
4001         __be32 status;
4002         struct nfs4_openowner *oo;
4003         struct nfs4_ol_stateid *stp;
4004         struct net *net = SVC_NET(rqstp);
4005         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4006
4007         dprintk("NFSD: nfsd4_close on file %pd\n",
4008                         cstate->current_fh.fh_dentry);
4009
4010         nfs4_lock_state();
4011         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4012                                         &close->cl_stateid,
4013                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
4014                                         &stp, nn);
4015         nfsd4_bump_seqid(cstate, status);
4016         if (status)
4017                 goto out;
4018         oo = openowner(stp->st_stateowner);
4019         update_stateid(&stp->st_stid.sc_stateid);
4020         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4021
4022         nfsd4_close_open_stateid(stp);
4023
4024         if (cstate->minorversion)
4025                 free_generic_stateid(stp);
4026         else
4027                 oo->oo_last_closed_stid = stp;
4028
4029         if (list_empty(&oo->oo_owner.so_stateids)) {
4030                 if (cstate->minorversion)
4031                         release_openowner(oo);
4032                 else {
4033                         /*
4034                          * In the 4.0 case we need to keep the owners around a
4035                          * little while to handle CLOSE replay.
4036                          */
4037                         move_to_close_lru(oo, SVC_NET(rqstp));
4038                 }
4039         }
4040 out:
4041         if (!cstate->replay_owner)
4042                 nfs4_unlock_state();
4043         return status;
4044 }
4045
4046 __be32
4047 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4048                   struct nfsd4_delegreturn *dr)
4049 {
4050         struct nfs4_delegation *dp;
4051         stateid_t *stateid = &dr->dr_stateid;
4052         struct nfs4_stid *s;
4053         __be32 status;
4054         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4055
4056         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4057                 return status;
4058
4059         nfs4_lock_state();
4060         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
4061                                       cstate->minorversion, nn);
4062         if (status)
4063                 goto out;
4064         dp = delegstateid(s);
4065         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4066         if (status)
4067                 goto out;
4068
4069         destroy_delegation(dp);
4070 out:
4071         nfs4_unlock_state();
4072
4073         return status;
4074 }
4075
4076
4077 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4078
4079 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
4080
4081 static inline u64
4082 end_offset(u64 start, u64 len)
4083 {
4084         u64 end;
4085
4086         end = start + len;
4087         return end >= start ? end: NFS4_MAX_UINT64;
4088 }
4089
4090 /* last octet in a range */
4091 static inline u64
4092 last_byte_offset(u64 start, u64 len)
4093 {
4094         u64 end;
4095
4096         WARN_ON_ONCE(!len);
4097         end = start + len;
4098         return end > start ? end - 1: NFS4_MAX_UINT64;
4099 }
4100
4101 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
4102 {
4103         return (file_hashval(inode) + cl_id
4104                         + opaque_hashval(ownername->data, ownername->len))
4105                 & LOCKOWNER_INO_HASH_MASK;
4106 }
4107
4108 /*
4109  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4110  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4111  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4112  * locking, this prevents us from being completely protocol-compliant.  The
4113  * real solution to this problem is to start using unsigned file offsets in
4114  * the VFS, but this is a very deep change!
4115  */
4116 static inline void
4117 nfs4_transform_lock_offset(struct file_lock *lock)
4118 {
4119         if (lock->fl_start < 0)
4120                 lock->fl_start = OFFSET_MAX;
4121         if (lock->fl_end < 0)
4122                 lock->fl_end = OFFSET_MAX;
4123 }
4124
4125 /* Hack!: For now, we're defining this just so we can use a pointer to it
4126  * as a unique cookie to identify our (NFSv4's) posix locks. */
4127 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4128 };
4129
4130 static inline void
4131 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4132 {
4133         struct nfs4_lockowner *lo;
4134
4135         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4136                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4137                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4138                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4139                 if (!deny->ld_owner.data)
4140                         /* We just don't care that much */
4141                         goto nevermind;
4142                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4143                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4144         } else {
4145 nevermind:
4146                 deny->ld_owner.len = 0;
4147                 deny->ld_owner.data = NULL;
4148                 deny->ld_clientid.cl_boot = 0;
4149                 deny->ld_clientid.cl_id = 0;
4150         }
4151         deny->ld_start = fl->fl_start;
4152         deny->ld_length = NFS4_MAX_UINT64;
4153         if (fl->fl_end != NFS4_MAX_UINT64)
4154                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4155         deny->ld_type = NFS4_READ_LT;
4156         if (fl->fl_type != F_RDLCK)
4157                 deny->ld_type = NFS4_WRITE_LT;
4158 }
4159
4160 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
4161 {
4162         struct nfs4_ol_stateid *lst;
4163
4164         if (!same_owner_str(&lo->lo_owner, owner, clid))
4165                 return false;
4166         if (list_empty(&lo->lo_owner.so_stateids)) {
4167                 WARN_ON_ONCE(1);
4168                 return false;
4169         }
4170         lst = list_first_entry(&lo->lo_owner.so_stateids,
4171                                struct nfs4_ol_stateid, st_perstateowner);
4172         return lst->st_file->fi_inode == inode;
4173 }
4174
4175 static struct nfs4_lockowner *
4176 find_lockowner_str(struct inode *inode, clientid_t *clid,
4177                    struct xdr_netobj *owner, struct nfsd_net *nn)
4178 {
4179         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4180         struct nfs4_lockowner *lo;
4181
4182         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4183                 if (same_lockowner_ino(lo, inode, clid, owner))
4184                         return lo;
4185         }
4186         return NULL;
4187 }
4188
4189 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4190 {
4191         struct inode *inode = open_stp->st_file->fi_inode;
4192         unsigned int inohash = lockowner_ino_hashval(inode,
4193                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4194         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4195
4196         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4197         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4198         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4199 }
4200
4201 /*
4202  * Alloc a lock owner structure.
4203  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4204  * occurred.
4205  *
4206  * strhashval = ownerstr_hashval
4207  */
4208
4209 static struct nfs4_lockowner *
4210 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4211         struct nfs4_lockowner *lo;
4212
4213         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4214         if (!lo)
4215                 return NULL;
4216         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4217         lo->lo_owner.so_is_open_owner = 0;
4218         /* It is the openowner seqid that will be incremented in encode in the
4219          * case of new lockowners; so increment the lock seqid manually: */
4220         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4221         hash_lockowner(lo, strhashval, clp, open_stp);
4222         return lo;
4223 }
4224
4225 static struct nfs4_ol_stateid *
4226 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4227 {
4228         struct nfs4_ol_stateid *stp;
4229         struct nfs4_client *clp = lo->lo_owner.so_client;
4230
4231         stp = nfs4_alloc_stateid(clp);
4232         if (stp == NULL)
4233                 return NULL;
4234         stp->st_stid.sc_type = NFS4_LOCK_STID;
4235         list_add(&stp->st_perfile, &fp->fi_stateids);
4236         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4237         stp->st_stateowner = &lo->lo_owner;
4238         get_nfs4_file(fp);
4239         stp->st_file = fp;
4240         stp->st_access_bmap = 0;
4241         stp->st_deny_bmap = open_stp->st_deny_bmap;
4242         stp->st_openstp = open_stp;
4243         return stp;
4244 }
4245
4246 static int
4247 check_lock_length(u64 offset, u64 length)
4248 {
4249         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4250              LOFF_OVERFLOW(offset, length)));
4251 }
4252
4253 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4254 {
4255         struct nfs4_file *fp = lock_stp->st_file;
4256         int oflag = nfs4_access_to_omode(access);
4257
4258         if (test_access(access, lock_stp))
4259                 return;
4260         nfs4_file_get_access(fp, oflag);
4261         set_access(access, lock_stp);
4262 }
4263
4264 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)
4265 {
4266         struct nfs4_file *fi = ost->st_file;
4267         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4268         struct nfs4_client *cl = oo->oo_owner.so_client;
4269         struct nfs4_lockowner *lo;
4270         unsigned int strhashval;
4271         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4272
4273         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4274                                 &lock->v.new.owner, nn);
4275         if (lo) {
4276                 if (!cstate->minorversion)
4277                         return nfserr_bad_seqid;
4278                 /* XXX: a lockowner always has exactly one stateid: */
4279                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4280                                 struct nfs4_ol_stateid, st_perstateowner);
4281                 return nfs_ok;
4282         }
4283         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4284                         &lock->v.new.owner);
4285         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4286         if (lo == NULL)
4287                 return nfserr_jukebox;
4288         *lst = alloc_init_lock_stateid(lo, fi, ost);
4289         if (*lst == NULL) {
4290                 release_lockowner(lo);
4291                 return nfserr_jukebox;
4292         }
4293         *new = true;
4294         return nfs_ok;
4295 }
4296
4297 /*
4298  *  LOCK operation
4299  */
4300 __be32
4301 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4302            struct nfsd4_lock *lock)
4303 {
4304         struct nfs4_openowner *open_sop = NULL;
4305         struct nfs4_lockowner *lock_sop = NULL;
4306         struct nfs4_ol_stateid *lock_stp;
4307         struct file *filp = NULL;
4308         struct file_lock *file_lock = NULL;
4309         struct file_lock *conflock = NULL;
4310         __be32 status = 0;
4311         bool new_state = false;
4312         int lkflg;
4313         int err;
4314         struct net *net = SVC_NET(rqstp);
4315         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4316
4317         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4318                 (long long) lock->lk_offset,
4319                 (long long) lock->lk_length);
4320
4321         if (check_lock_length(lock->lk_offset, lock->lk_length))
4322                  return nfserr_inval;
4323
4324         if ((status = fh_verify(rqstp, &cstate->current_fh,
4325                                 S_IFREG, NFSD_MAY_LOCK))) {
4326                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4327                 return status;
4328         }
4329
4330         nfs4_lock_state();
4331
4332         if (lock->lk_is_new) {
4333                 struct nfs4_ol_stateid *open_stp = NULL;
4334
4335                 if (nfsd4_has_session(cstate))
4336                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4337                         memcpy(&lock->v.new.clientid,
4338                                 &cstate->session->se_client->cl_clientid,
4339                                 sizeof(clientid_t));
4340
4341                 status = nfserr_stale_clientid;
4342                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4343                         goto out;
4344
4345                 /* validate and update open stateid and open seqid */
4346                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4347                                         lock->lk_new_open_seqid,
4348                                         &lock->lk_new_open_stateid,
4349                                         &open_stp, nn);
4350                 if (status)
4351                         goto out;
4352                 open_sop = openowner(open_stp->st_stateowner);
4353                 status = nfserr_bad_stateid;
4354                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4355                                                 &lock->v.new.clientid))
4356                         goto out;
4357                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4358                                                         &lock_stp, &new_state);
4359         } else
4360                 status = nfs4_preprocess_seqid_op(cstate,
4361                                        lock->lk_old_lock_seqid,
4362                                        &lock->lk_old_lock_stateid,
4363                                        NFS4_LOCK_STID, &lock_stp, nn);
4364         if (status)
4365                 goto out;
4366         lock_sop = lockowner(lock_stp->st_stateowner);
4367
4368         lkflg = setlkflg(lock->lk_type);
4369         status = nfs4_check_openmode(lock_stp, lkflg);
4370         if (status)
4371                 goto out;
4372
4373         status = nfserr_grace;
4374         if (locks_in_grace(net) && !lock->lk_reclaim)
4375                 goto out;
4376         status = nfserr_no_grace;
4377         if (!locks_in_grace(net) && lock->lk_reclaim)
4378                 goto out;
4379
4380         file_lock = locks_alloc_lock();
4381         if (!file_lock) {
4382                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4383                 status = nfserr_jukebox;
4384                 goto out;
4385         }
4386
4387         locks_init_lock(file_lock);
4388         switch (lock->lk_type) {
4389                 case NFS4_READ_LT:
4390                 case NFS4_READW_LT:
4391                         filp = find_readable_file(lock_stp->st_file);
4392                         if (filp)
4393                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4394                         file_lock->fl_type = F_RDLCK;
4395                         break;
4396                 case NFS4_WRITE_LT:
4397                 case NFS4_WRITEW_LT:
4398                         filp = find_writeable_file(lock_stp->st_file);
4399                         if (filp)
4400                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4401                         file_lock->fl_type = F_WRLCK;
4402                         break;
4403                 default:
4404                         status = nfserr_inval;
4405                 goto out;
4406         }
4407         if (!filp) {
4408                 status = nfserr_openmode;
4409                 goto out;
4410         }
4411         file_lock->fl_owner = (fl_owner_t)lock_sop;
4412         file_lock->fl_pid = current->tgid;
4413         file_lock->fl_file = filp;
4414         file_lock->fl_flags = FL_POSIX;
4415         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4416         file_lock->fl_start = lock->lk_offset;
4417         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4418         nfs4_transform_lock_offset(file_lock);
4419
4420         conflock = locks_alloc_lock();
4421         if (!conflock) {
4422                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4423                 status = nfserr_jukebox;
4424                 goto out;
4425         }
4426
4427         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4428         switch (-err) {
4429         case 0: /* success! */
4430                 update_stateid(&lock_stp->st_stid.sc_stateid);
4431                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4432                                 sizeof(stateid_t));
4433                 status = 0;
4434                 break;
4435         case (EAGAIN):          /* conflock holds conflicting lock */
4436                 status = nfserr_denied;
4437                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4438                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4439                 break;
4440         case (EDEADLK):
4441                 status = nfserr_deadlock;
4442                 break;
4443         default:
4444                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4445                 status = nfserrno(err);
4446                 break;
4447         }
4448 out:
4449         if (status && new_state)
4450                 release_lockowner(lock_sop);
4451         nfsd4_bump_seqid(cstate, status);
4452         if (!cstate->replay_owner)
4453                 nfs4_unlock_state();
4454         if (file_lock)
4455                 locks_free_lock(file_lock);
4456         if (conflock)
4457                 locks_free_lock(conflock);
4458         return status;
4459 }
4460
4461 /*
4462  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4463  * so we do a temporary open here just to get an open file to pass to
4464  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4465  * inode operation.)
4466  */
4467 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4468 {
4469         struct file *file;
4470         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4471         if (!err) {
4472                 err = nfserrno(vfs_test_lock(file, lock));
4473                 nfsd_close(file);
4474         }
4475         return err;
4476 }
4477
4478 /*
4479  * LOCKT operation
4480  */
4481 __be32
4482 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4483             struct nfsd4_lockt *lockt)
4484 {
4485         struct inode *inode;
4486         struct file_lock *file_lock = NULL;
4487         struct nfs4_lockowner *lo;
4488         __be32 status;
4489         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4490
4491         if (locks_in_grace(SVC_NET(rqstp)))
4492                 return nfserr_grace;
4493
4494         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4495                  return nfserr_inval;
4496
4497         nfs4_lock_state();
4498
4499         if (!nfsd4_has_session(cstate)) {
4500                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4501                 if (status)
4502                         goto out;
4503         }
4504
4505         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4506                 goto out;
4507
4508         inode = cstate->current_fh.fh_dentry->d_inode;
4509         file_lock = locks_alloc_lock();
4510         if (!file_lock) {
4511                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4512                 status = nfserr_jukebox;
4513                 goto out;
4514         }
4515         locks_init_lock(file_lock);
4516         switch (lockt->lt_type) {
4517                 case NFS4_READ_LT:
4518                 case NFS4_READW_LT:
4519                         file_lock->fl_type = F_RDLCK;
4520                 break;
4521                 case NFS4_WRITE_LT:
4522                 case NFS4_WRITEW_LT:
4523                         file_lock->fl_type = F_WRLCK;
4524                 break;
4525                 default:
4526                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4527                         status = nfserr_inval;
4528                 goto out;
4529         }
4530
4531         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4532         if (lo)
4533                 file_lock->fl_owner = (fl_owner_t)lo;
4534         file_lock->fl_pid = current->tgid;
4535         file_lock->fl_flags = FL_POSIX;
4536
4537         file_lock->fl_start = lockt->lt_offset;
4538         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4539
4540         nfs4_transform_lock_offset(file_lock);
4541
4542         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4543         if (status)
4544                 goto out;
4545
4546         if (file_lock->fl_type != F_UNLCK) {
4547                 status = nfserr_denied;
4548                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4549         }
4550 out:
4551         nfs4_unlock_state();
4552         if (file_lock)
4553                 locks_free_lock(file_lock);
4554         return status;
4555 }
4556
4557 __be32
4558 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4559             struct nfsd4_locku *locku)
4560 {
4561         struct nfs4_ol_stateid *stp;
4562         struct file *filp = NULL;
4563         struct file_lock *file_lock = NULL;
4564         __be32 status;
4565         int err;
4566         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4567
4568         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4569                 (long long) locku->lu_offset,
4570                 (long long) locku->lu_length);
4571
4572         if (check_lock_length(locku->lu_offset, locku->lu_length))
4573                  return nfserr_inval;
4574
4575         nfs4_lock_state();
4576
4577         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4578                                         &locku->lu_stateid, NFS4_LOCK_STID,
4579                                         &stp, nn);
4580         if (status)
4581                 goto out;
4582         filp = find_any_file(stp->st_file);
4583         if (!filp) {
4584                 status = nfserr_lock_range;
4585                 goto out;
4586         }
4587         file_lock = locks_alloc_lock();
4588         if (!file_lock) {
4589                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4590                 status = nfserr_jukebox;
4591                 goto out;
4592         }
4593         locks_init_lock(file_lock);
4594         file_lock->fl_type = F_UNLCK;
4595         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4596         file_lock->fl_pid = current->tgid;
4597         file_lock->fl_file = filp;
4598         file_lock->fl_flags = FL_POSIX;
4599         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4600         file_lock->fl_start = locku->lu_offset;
4601
4602         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4603                                                 locku->lu_length);
4604         nfs4_transform_lock_offset(file_lock);
4605
4606         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4607         if (err) {
4608                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4609                 goto out_nfserr;
4610         }
4611         update_stateid(&stp->st_stid.sc_stateid);
4612         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4613
4614 out:
4615         nfsd4_bump_seqid(cstate, status);
4616         if (!cstate->replay_owner)
4617                 nfs4_unlock_state();
4618         if (file_lock)
4619                 locks_free_lock(file_lock);
4620         return status;
4621
4622 out_nfserr:
4623         status = nfserrno(err);
4624         goto out;
4625 }
4626
4627 /*
4628  * returns
4629  *      1: locks held by lockowner
4630  *      0: no locks held by lockowner
4631  */
4632 static int
4633 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4634 {
4635         struct file_lock **flpp;
4636         struct inode *inode = filp->fi_inode;
4637         int status = 0;
4638
4639         spin_lock(&inode->i_lock);
4640         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4641                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4642                         status = 1;
4643                         goto out;
4644                 }
4645         }
4646 out:
4647         spin_unlock(&inode->i_lock);
4648         return status;
4649 }
4650
4651 __be32
4652 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4653                         struct nfsd4_compound_state *cstate,
4654                         struct nfsd4_release_lockowner *rlockowner)
4655 {
4656         clientid_t *clid = &rlockowner->rl_clientid;
4657         struct nfs4_stateowner *sop;
4658         struct nfs4_lockowner *lo;
4659         struct nfs4_ol_stateid *stp;
4660         struct xdr_netobj *owner = &rlockowner->rl_owner;
4661         struct list_head matches;
4662         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4663         __be32 status;
4664         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4665
4666         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4667                 clid->cl_boot, clid->cl_id);
4668
4669         nfs4_lock_state();
4670
4671         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4672         if (status)
4673                 goto out;
4674
4675         status = nfserr_locks_held;
4676         INIT_LIST_HEAD(&matches);
4677
4678         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4679                 if (sop->so_is_open_owner)
4680                         continue;
4681                 if (!same_owner_str(sop, owner, clid))
4682                         continue;
4683                 list_for_each_entry(stp, &sop->so_stateids,
4684                                 st_perstateowner) {
4685                         lo = lockowner(sop);
4686                         if (check_for_locks(stp->st_file, lo))
4687                                 goto out;
4688                         list_add(&lo->lo_list, &matches);
4689                 }
4690         }
4691         /* Clients probably won't expect us to return with some (but not all)
4692          * of the lockowner state released; so don't release any until all
4693          * have been checked. */
4694         status = nfs_ok;
4695         while (!list_empty(&matches)) {
4696                 lo = list_entry(matches.next, struct nfs4_lockowner,
4697                                                                 lo_list);
4698                 /* unhash_stateowner deletes so_perclient only
4699                  * for openowners. */
4700                 list_del(&lo->lo_list);
4701                 release_lockowner(lo);
4702         }
4703 out:
4704         nfs4_unlock_state();
4705         return status;
4706 }
4707
4708 static inline struct nfs4_client_reclaim *
4709 alloc_reclaim(void)
4710 {
4711         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4712 }
4713
4714 bool
4715 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4716 {
4717         struct nfs4_client_reclaim *crp;
4718
4719         crp = nfsd4_find_reclaim_client(name, nn);
4720         return (crp && crp->cr_clp);
4721 }
4722
4723 /*
4724  * failure => all reset bets are off, nfserr_no_grace...
4725  */
4726 struct nfs4_client_reclaim *
4727 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4728 {
4729         unsigned int strhashval;
4730         struct nfs4_client_reclaim *crp;
4731
4732         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4733         crp = alloc_reclaim();
4734         if (crp) {
4735                 strhashval = clientstr_hashval(name);
4736                 INIT_LIST_HEAD(&crp->cr_strhash);
4737                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4738                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4739                 crp->cr_clp = NULL;
4740                 nn->reclaim_str_hashtbl_size++;
4741         }
4742         return crp;
4743 }
4744
4745 void
4746 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4747 {
4748         list_del(&crp->cr_strhash);
4749         kfree(crp);
4750         nn->reclaim_str_hashtbl_size--;
4751 }
4752
4753 void
4754 nfs4_release_reclaim(struct nfsd_net *nn)
4755 {
4756         struct nfs4_client_reclaim *crp = NULL;
4757         int i;
4758
4759         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4760                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4761                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4762                                         struct nfs4_client_reclaim, cr_strhash);
4763                         nfs4_remove_reclaim_record(crp, nn);
4764                 }
4765         }
4766         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4767 }
4768
4769 /*
4770  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4771 struct nfs4_client_reclaim *
4772 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4773 {
4774         unsigned int strhashval;
4775         struct nfs4_client_reclaim *crp = NULL;
4776
4777         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4778
4779         strhashval = clientstr_hashval(recdir);
4780         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4781                 if (same_name(crp->cr_recdir, recdir)) {
4782                         return crp;
4783                 }
4784         }
4785         return NULL;
4786 }
4787
4788 /*
4789 * Called from OPEN. Look for clientid in reclaim list.
4790 */
4791 __be32
4792 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4793 {
4794         struct nfs4_client *clp;
4795
4796         /* find clientid in conf_id_hashtbl */
4797         clp = find_confirmed_client(clid, sessions, nn);
4798         if (clp == NULL)
4799                 return nfserr_reclaim_bad;
4800
4801         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4802 }
4803
4804 #ifdef CONFIG_NFSD_FAULT_INJECTION
4805
4806 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4807 {
4808         if (mark_client_expired(clp))
4809                 return 0;
4810         expire_client(clp);
4811         return 1;
4812 }
4813
4814 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4815 {
4816         char buf[INET6_ADDRSTRLEN];
4817         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4818         printk(KERN_INFO "NFS Client: %s\n", buf);
4819         return 1;
4820 }
4821
4822 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4823                              const char *type)
4824 {
4825         char buf[INET6_ADDRSTRLEN];
4826         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4827         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4828 }
4829
4830 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4831 {
4832         struct nfs4_openowner *oop;
4833         struct nfs4_lockowner *lop, *lo_next;
4834         struct nfs4_ol_stateid *stp, *st_next;
4835         u64 count = 0;
4836
4837         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4838                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4839                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4840                                 if (func)
4841                                         func(lop);
4842                                 if (++count == max)
4843                                         return count;
4844                         }
4845                 }
4846         }
4847
4848         return count;
4849 }
4850
4851 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4852 {
4853         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4854 }
4855
4856 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4857 {
4858         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4859         nfsd_print_count(clp, count, "locked files");
4860         return count;
4861 }
4862
4863 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4864 {
4865         struct nfs4_openowner *oop, *next;
4866         u64 count = 0;
4867
4868         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4869                 if (func)
4870                         func(oop);
4871                 if (++count == max)
4872                         break;
4873         }
4874
4875         return count;
4876 }
4877
4878 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4879 {
4880         return nfsd_foreach_client_open(clp, max, release_openowner);
4881 }
4882
4883 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4884 {
4885         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4886         nfsd_print_count(clp, count, "open files");
4887         return count;
4888 }
4889
4890 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4891                                      struct list_head *victims)
4892 {
4893         struct nfs4_delegation *dp, *next;
4894         u64 count = 0;
4895
4896         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4897                 if (victims)
4898                         list_move(&dp->dl_recall_lru, victims);
4899                 if (++count == max)
4900                         break;
4901         }
4902         return count;
4903 }
4904
4905 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4906 {
4907         struct nfs4_delegation *dp, *next;
4908         LIST_HEAD(victims);
4909         u64 count;
4910
4911         spin_lock(&recall_lock);
4912         count = nfsd_find_all_delegations(clp, max, &victims);
4913         spin_unlock(&recall_lock);
4914
4915         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4916                 revoke_delegation(dp);
4917
4918         return count;
4919 }
4920
4921 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4922 {
4923         struct nfs4_delegation *dp, *next;
4924         LIST_HEAD(victims);
4925         u64 count;
4926
4927         spin_lock(&recall_lock);
4928         count = nfsd_find_all_delegations(clp, max, &victims);
4929         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4930                 nfsd_break_one_deleg(dp);
4931         spin_unlock(&recall_lock);
4932
4933         return count;
4934 }
4935
4936 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4937 {
4938         u64 count = 0;
4939
4940         spin_lock(&recall_lock);
4941         count = nfsd_find_all_delegations(clp, max, NULL);
4942         spin_unlock(&recall_lock);
4943
4944         nfsd_print_count(clp, count, "delegations");
4945         return count;
4946 }
4947
4948 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4949 {
4950         struct nfs4_client *clp, *next;
4951         u64 count = 0;
4952         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4953
4954         if (!nfsd_netns_ready(nn))
4955                 return 0;
4956
4957         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4958                 count += func(clp, max - count);
4959                 if ((max != 0) && (count >= max))
4960                         break;
4961         }
4962
4963         return count;
4964 }
4965
4966 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4967 {
4968         struct nfs4_client *clp;
4969         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4970
4971         if (!nfsd_netns_ready(nn))
4972                 return NULL;
4973
4974         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4975                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4976                         return clp;
4977         }
4978         return NULL;
4979 }
4980
4981 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4982
4983 /* initialization to perform at module load time: */
4984
4985 void
4986 nfs4_state_init(void)
4987 {
4988 }
4989
4990 /*
4991  * Since the lifetime of a delegation isn't limited to that of an open, a
4992  * client may quite reasonably hang on to a delegation as long as it has
4993  * the inode cached.  This becomes an obvious problem the first time a
4994  * client's inode cache approaches the size of the server's total memory.
4995  *
4996  * For now we avoid this problem by imposing a hard limit on the number
4997  * of delegations, which varies according to the server's memory size.
4998  */
4999 static void
5000 set_max_delegations(void)
5001 {
5002         /*
5003          * Allow at most 4 delegations per megabyte of RAM.  Quick
5004          * estimates suggest that in the worst case (where every delegation
5005          * is for a different inode), a delegation could take about 1.5K,
5006          * giving a worst case usage of about 6% of memory.
5007          */
5008         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
5009 }
5010
5011 static int nfs4_state_create_net(struct net *net)
5012 {
5013         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5014         int i;
5015
5016         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
5017                         CLIENT_HASH_SIZE, GFP_KERNEL);
5018         if (!nn->conf_id_hashtbl)
5019                 goto err;
5020         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
5021                         CLIENT_HASH_SIZE, GFP_KERNEL);
5022         if (!nn->unconf_id_hashtbl)
5023                 goto err_unconf_id;
5024         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
5025                         OWNER_HASH_SIZE, GFP_KERNEL);
5026         if (!nn->ownerstr_hashtbl)
5027                 goto err_ownerstr;
5028         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
5029                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
5030         if (!nn->lockowner_ino_hashtbl)
5031                 goto err_lockowner_ino;
5032         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
5033                         SESSION_HASH_SIZE, GFP_KERNEL);
5034         if (!nn->sessionid_hashtbl)
5035                 goto err_sessionid;
5036
5037         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5038                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
5039                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
5040         }
5041         for (i = 0; i < OWNER_HASH_SIZE; i++)
5042                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
5043         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
5044                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
5045         for (i = 0; i < SESSION_HASH_SIZE; i++)
5046                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
5047         nn->conf_name_tree = RB_ROOT;
5048         nn->unconf_name_tree = RB_ROOT;
5049         INIT_LIST_HEAD(&nn->client_lru);
5050         INIT_LIST_HEAD(&nn->close_lru);
5051         INIT_LIST_HEAD(&nn->del_recall_lru);
5052         spin_lock_init(&nn->client_lock);
5053
5054         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
5055         get_net(net);
5056
5057         return 0;
5058
5059 err_sessionid:
5060         kfree(nn->lockowner_ino_hashtbl);
5061 err_lockowner_ino:
5062         kfree(nn->ownerstr_hashtbl);
5063 err_ownerstr:
5064         kfree(nn->unconf_id_hashtbl);
5065 err_unconf_id:
5066         kfree(nn->conf_id_hashtbl);
5067 err:
5068         return -ENOMEM;
5069 }
5070
5071 static void
5072 nfs4_state_destroy_net(struct net *net)
5073 {
5074         int i;
5075         struct nfs4_client *clp = NULL;
5076         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5077
5078         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5079                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
5080                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
5081                         destroy_client(clp);
5082                 }
5083         }
5084
5085         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5086                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
5087                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
5088                         destroy_client(clp);
5089                 }
5090         }
5091
5092         kfree(nn->sessionid_hashtbl);
5093         kfree(nn->lockowner_ino_hashtbl);
5094         kfree(nn->ownerstr_hashtbl);
5095         kfree(nn->unconf_id_hashtbl);
5096         kfree(nn->conf_id_hashtbl);
5097         put_net(net);
5098 }
5099
5100 int
5101 nfs4_state_start_net(struct net *net)
5102 {
5103         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5104         int ret;
5105
5106         ret = nfs4_state_create_net(net);
5107         if (ret)
5108                 return ret;
5109         nfsd4_client_tracking_init(net);
5110         nn->boot_time = get_seconds();
5111         locks_start_grace(net, &nn->nfsd4_manager);
5112         nn->grace_ended = false;
5113         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
5114                nn->nfsd4_grace, net);
5115         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
5116         return 0;
5117 }
5118
5119 /* initialization to perform when the nfsd service is started: */
5120
5121 int
5122 nfs4_state_start(void)
5123 {
5124         int ret;
5125
5126         ret = set_callback_cred();
5127         if (ret)
5128                 return -ENOMEM;
5129         laundry_wq = create_singlethread_workqueue("nfsd4");
5130         if (laundry_wq == NULL) {
5131                 ret = -ENOMEM;
5132                 goto out_recovery;
5133         }
5134         ret = nfsd4_create_callback_queue();
5135         if (ret)
5136                 goto out_free_laundry;
5137
5138         set_max_delegations();
5139
5140         return 0;
5141
5142 out_free_laundry:
5143         destroy_workqueue(laundry_wq);
5144 out_recovery:
5145         return ret;
5146 }
5147
5148 void
5149 nfs4_state_shutdown_net(struct net *net)
5150 {
5151         struct nfs4_delegation *dp = NULL;
5152         struct list_head *pos, *next, reaplist;
5153         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5154
5155         cancel_delayed_work_sync(&nn->laundromat_work);
5156         locks_end_grace(&nn->nfsd4_manager);
5157
5158         nfs4_lock_state();
5159         INIT_LIST_HEAD(&reaplist);
5160         spin_lock(&recall_lock);
5161         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5162                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5163                 list_move(&dp->dl_recall_lru, &reaplist);
5164         }
5165         spin_unlock(&recall_lock);
5166         list_for_each_safe(pos, next, &reaplist) {
5167                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5168                 destroy_delegation(dp);
5169         }
5170
5171         nfsd4_client_tracking_exit(net);
5172         nfs4_state_destroy_net(net);
5173         nfs4_unlock_state();
5174 }
5175
5176 void
5177 nfs4_state_shutdown(void)
5178 {
5179         destroy_workqueue(laundry_wq);
5180         nfsd4_destroy_callback_queue();
5181 }
5182
5183 static void
5184 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5185 {
5186         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5187                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5188 }
5189
5190 static void
5191 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5192 {
5193         if (cstate->minorversion) {
5194                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5195                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5196         }
5197 }
5198
5199 void
5200 clear_current_stateid(struct nfsd4_compound_state *cstate)
5201 {
5202         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5203 }
5204
5205 /*
5206  * functions to set current state id
5207  */
5208 void
5209 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5210 {
5211         put_stateid(cstate, &odp->od_stateid);
5212 }
5213
5214 void
5215 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5216 {
5217         put_stateid(cstate, &open->op_stateid);
5218 }
5219
5220 void
5221 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5222 {
5223         put_stateid(cstate, &close->cl_stateid);
5224 }
5225
5226 void
5227 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5228 {
5229         put_stateid(cstate, &lock->lk_resp_stateid);
5230 }
5231
5232 /*
5233  * functions to consume current state id
5234  */
5235
5236 void
5237 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5238 {
5239         get_stateid(cstate, &odp->od_stateid);
5240 }
5241
5242 void
5243 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5244 {
5245         get_stateid(cstate, &drp->dr_stateid);
5246 }
5247
5248 void
5249 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5250 {
5251         get_stateid(cstate, &fsp->fr_stateid);
5252 }
5253
5254 void
5255 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5256 {
5257         get_stateid(cstate, &setattr->sa_stateid);
5258 }
5259
5260 void
5261 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5262 {
5263         get_stateid(cstate, &close->cl_stateid);
5264 }
5265
5266 void
5267 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5268 {
5269         get_stateid(cstate, &locku->lu_stateid);
5270 }
5271
5272 void
5273 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5274 {
5275         get_stateid(cstate, &read->rd_stateid);
5276 }
5277
5278 void
5279 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5280 {
5281         get_stateid(cstate, &write->wr_stateid);
5282 }