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