fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include "xdr4.h"
  45 #include "xdr4cb.h"
  46 #include "vfs.h"
  47 #include "current_stateid.h"
  48
  49 #include "netns.h"
  50
  51 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  52
  53 #define all_ones {{~0,~0},~0}
  54 static const stateid_t one_stateid = {
  55         .si_generation = ~0,
  56         .si_opaque = all_ones,
  57 };
  58 static const stateid_t zero_stateid = {
  59         /* all fields zero */
  60 };
  61 static const stateid_t currentstateid = {
  62         .si_generation = 1,
  63 };
  64
  65 static u64 current_sessionid = 1;
  66
  67 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  68 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  69 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  70
  71 /* forward declarations */
  72 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
  73
  74 /* Locking: */
  75
  76 /* Currently used for almost all code touching nfsv4 state: */
  77 static DEFINE_MUTEX(client_mutex);
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(recall_lock);
  85
  86 static struct kmem_cache *openowner_slab = NULL;
  87 static struct kmem_cache *lockowner_slab = NULL;
  88 static struct kmem_cache *file_slab = NULL;
  89 static struct kmem_cache *stateid_slab = NULL;
  90 static struct kmem_cache *deleg_slab = NULL;
  91
  92 void
  93 nfs4_lock_state(void)
  94 {
  95         mutex_lock(&client_mutex);
  96 }
  97
  98 static void free_session(struct nfsd4_session *);
  99
 100 static bool is_session_dead(struct nfsd4_session *ses)
 101 {
 102         return ses->se_flags & NFS4_SESSION_DEAD;
 103 }
 104
 105 void nfsd4_put_session(struct nfsd4_session *ses)
 106 {
 107         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 108                 free_session(ses);
 109 }
 110
 111 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 112 {
 113         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 114                 return nfserr_jukebox;
 115         ses->se_flags |= NFS4_SESSION_DEAD;
 116         return nfs_ok;
 117 }
 118
 119 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 120 {
 121         if (is_session_dead(ses))
 122                 return nfserr_badsession;
 123         atomic_inc(&ses->se_ref);
 124         return nfs_ok;
 125 }
 126
 127 void
 128 nfs4_unlock_state(void)
 129 {
 130         mutex_unlock(&client_mutex);
 131 }
 132
 133 static bool is_client_expired(struct nfs4_client *clp)
 134 {
 135         return clp->cl_time == 0;
 136 }
 137
 138 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
 139 {
 140         if (atomic_read(&clp->cl_refcount))
 141                 return nfserr_jukebox;
 142         clp->cl_time = 0;
 143         return nfs_ok;
 144 }
 145
 146 static __be32 mark_client_expired(struct nfs4_client *clp)
 147 {
 148         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 149         __be32 ret;
 150
 151         spin_lock(&nn->client_lock);
 152         ret = mark_client_expired_locked(clp);
 153         spin_unlock(&nn->client_lock);
 154         return ret;
 155 }
 156
 157 static __be32 get_client_locked(struct nfs4_client *clp)
 158 {
 159         if (is_client_expired(clp))
 160                 return nfserr_expired;
 161         atomic_inc(&clp->cl_refcount);
 162         return nfs_ok;
 163 }
 164
 165 /* must be called under the client_lock */
 166 static inline void
 167 renew_client_locked(struct nfs4_client *clp)
 168 {
 169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 170
 171         if (is_client_expired(clp)) {
 172                 WARN_ON(1);
 173                 printk("%s: client (clientid %08x/%08x) already expired\n",
 174                         __func__,
 175                         clp->cl_clientid.cl_boot,
 176                         clp->cl_clientid.cl_id);
 177                 return;
 178         }
 179
 180         dprintk("renewing client (clientid %08x/%08x)\n",
 181                         clp->cl_clientid.cl_boot,
 182                         clp->cl_clientid.cl_id);
 183         list_move_tail(&clp->cl_lru, &nn->client_lru);
 184         clp->cl_time = get_seconds();
 185 }
 186
 187 static inline void
 188 renew_client(struct nfs4_client *clp)
 189 {
 190         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 191
 192         spin_lock(&nn->client_lock);
 193         renew_client_locked(clp);
 194         spin_unlock(&nn->client_lock);
 195 }
 196
 197 static void put_client_renew_locked(struct nfs4_client *clp)
 198 {
 199         if (!atomic_dec_and_test(&clp->cl_refcount))
 200                 return;
 201         if (!is_client_expired(clp))
 202                 renew_client_locked(clp);
 203 }
 204
 205 void put_client_renew(struct nfs4_client *clp)
 206 {
 207         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 208
 209         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 210                 return;
 211         if (!is_client_expired(clp))
 212                 renew_client_locked(clp);
 213         spin_unlock(&nn->client_lock);
 214 }
 215
 216
 217 static inline u32
 218 opaque_hashval(const void *ptr, int nbytes)
 219 {
 220         unsigned char *cptr = (unsigned char *) ptr;
 221
 222         u32 x = 0;
 223         while (nbytes--) {
 224                 x *= 37;
 225                 x += *cptr++;
 226         }
 227         return x;
 228 }
 229
 230 static void nfsd4_free_file(struct nfs4_file *f)
 231 {
 232         kmem_cache_free(file_slab, f);
 233 }
 234
 235 static inline void
 236 put_nfs4_file(struct nfs4_file *fi)
 237 {
 238         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 239                 hlist_del(&fi->fi_hash);
 240                 spin_unlock(&recall_lock);
 241                 iput(fi->fi_inode);
 242                 nfsd4_free_file(fi);
 243         }
 244 }
 245
 246 static inline void
 247 get_nfs4_file(struct nfs4_file *fi)
 248 {
 249         atomic_inc(&fi->fi_ref);
 250 }
 251
 252 static int num_delegations;
 253 unsigned long max_delegations;
 254
 255 /*
 256  * Open owner state (share locks)
 257  */
 258
 259 /* hash tables for lock and open owners */
 260 #define OWNER_HASH_BITS              8
 261 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 262 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 263
 264 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 265 {
 266         unsigned int ret;
 267
 268         ret = opaque_hashval(ownername->data, ownername->len);
 269         ret += clientid;
 270         return ret & OWNER_HASH_MASK;
 271 }
 272
 273 /* hash table for nfs4_file */
 274 #define FILE_HASH_BITS                   8
 275 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 276
 277 static unsigned int file_hashval(struct inode *ino)
 278 {
 279         /* XXX: why are we hashing on inode pointer, anyway? */
 280         return hash_ptr(ino, FILE_HASH_BITS);
 281 }
 282
 283 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 284
 285 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 286 {
 287         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 288         atomic_inc(&fp->fi_access[oflag]);
 289 }
 290
 291 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 292 {
 293         if (oflag == O_RDWR) {
 294                 __nfs4_file_get_access(fp, O_RDONLY);
 295                 __nfs4_file_get_access(fp, O_WRONLY);
 296         } else
 297                 __nfs4_file_get_access(fp, oflag);
 298 }
 299
 300 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 301 {
 302         if (fp->fi_fds[oflag]) {
 303                 fput(fp->fi_fds[oflag]);
 304                 fp->fi_fds[oflag] = NULL;
 305         }
 306 }
 307
 308 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 309 {
 310         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 311                 nfs4_file_put_fd(fp, oflag);
 312                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 313                         nfs4_file_put_fd(fp, O_RDWR);
 314         }
 315 }
 316
 317 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 318 {
 319         if (oflag == O_RDWR) {
 320                 __nfs4_file_put_access(fp, O_RDONLY);
 321                 __nfs4_file_put_access(fp, O_WRONLY);
 322         } else
 323                 __nfs4_file_put_access(fp, oflag);
 324 }
 325
 326 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct
 327 kmem_cache *slab)
 328 {
 329         struct idr *stateids = &cl->cl_stateids;
 330         struct nfs4_stid *stid;
 331         int new_id;
 332
 333         stid = kmem_cache_alloc(slab, GFP_KERNEL);
 334         if (!stid)
 335                 return NULL;
 336
 337         new_id = idr_alloc_cyclic(stateids, stid, 0, 0, GFP_KERNEL);
 338         if (new_id < 0)
 339                 goto out_free;
 340         stid->sc_client = cl;
 341         stid->sc_type = 0;
 342         stid->sc_stateid.si_opaque.so_id = new_id;
 343         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 344         /* Will be incremented before return to client: */
 345         stid->sc_stateid.si_generation = 0;
 346
 347         /*
 348          * It shouldn't be a problem to reuse an opaque stateid value.
 349          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 350          * example, a stray write retransmission could be accepted by
 351          * the server when it should have been rejected.  Therefore,
 352          * adopt a trick from the sctp code to attempt to maximize the
 353          * amount of time until an id is reused, by ensuring they always
 354          * "increase" (mod INT_MAX):
 355          */
 356         return stid;
 357 out_free:
 358         kmem_cache_free(slab, stid);
 359         return NULL;
 360 }
 361
 362 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 363 {
 364         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 365 }
 366
 367 static struct nfs4_delegation *
 368 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh)
 369 {
 370         struct nfs4_delegation *dp;
 371
 372         dprintk("NFSD alloc_init_deleg\n");
 373         if (num_delegations > max_delegations)
 374                 return NULL;
 375         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 376         if (dp == NULL)
 377                 return dp;
 378         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 379         /*
 380          * delegation seqid's are never incremented.  The 4.1 special
 381          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 382          * 0 anyway just for consistency and use 1:
 383          */
 384         dp->dl_stid.sc_stateid.si_generation = 1;
 385         num_delegations++;
 386         INIT_LIST_HEAD(&dp->dl_perfile);
 387         INIT_LIST_HEAD(&dp->dl_perclnt);
 388         INIT_LIST_HEAD(&dp->dl_recall_lru);
 389         dp->dl_file = NULL;
 390         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 391         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 392         dp->dl_time = 0;
 393         atomic_set(&dp->dl_count, 1);
 394         nfsd4_init_callback(&dp->dl_recall);
 395         return dp;
 396 }
 397
 398 static void remove_stid(struct nfs4_stid *s)
 399 {
 400         struct idr *stateids = &s->sc_client->cl_stateids;
 401
 402         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 403 }
 404
 405 static void nfs4_free_stid(struct kmem_cache *slab, struct nfs4_stid *s)
 406 {
 407         kmem_cache_free(slab, s);
 408 }
 409
 410 void
 411 nfs4_put_delegation(struct nfs4_delegation *dp)
 412 {
 413         remove_stid(&dp->dl_stid);
 414         if (atomic_dec_and_test(&dp->dl_count)) {
 415                 nfs4_free_stid(deleg_slab, &dp->dl_stid);
 416                 num_delegations--;
 417         }
 418 }
 419
 420 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 421 {
 422         if (atomic_dec_and_test(&fp->fi_delegees)) {
 423                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 424                 fp->fi_lease = NULL;
 425                 fput(fp->fi_deleg_file);
 426                 fp->fi_deleg_file = NULL;
 427         }
 428 }
 429
 430 static void unhash_stid(struct nfs4_stid *s)
 431 {
 432         s->sc_type = 0;
 433 }
 434
 435 /* Called under the state lock. */
 436 static void
 437 unhash_delegation(struct nfs4_delegation *dp)
 438 {
 439         list_del_init(&dp->dl_perclnt);
 440         spin_lock(&recall_lock);
 441         list_del_init(&dp->dl_perfile);
 442         list_del_init(&dp->dl_recall_lru);
 443         spin_unlock(&recall_lock);
 444         nfs4_put_deleg_lease(dp->dl_file);
 445         put_nfs4_file(dp->dl_file);
 446         dp->dl_file = NULL;
 447 }
 448
 449
 450
 451 static void destroy_revoked_delegation(struct nfs4_delegation *dp)
 452 {
 453         list_del_init(&dp->dl_recall_lru);
 454         nfs4_put_delegation(dp);
 455 }
 456
 457 static void destroy_delegation(struct nfs4_delegation *dp)
 458 {
 459         unhash_delegation(dp);
 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         nfs4_free_stid(stateid_slab, &stp->st_stid);
 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         free_generic_stateid(stp);
 676 }
 677
 678 static void unhash_openowner(struct nfs4_openowner *oo)
 679 {
 680         struct nfs4_ol_stateid *stp;
 681
 682         list_del(&oo->oo_owner.so_strhash);
 683         list_del(&oo->oo_perclient);
 684         while (!list_empty(&oo->oo_owner.so_stateids)) {
 685                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 686                                 struct nfs4_ol_stateid, st_perstateowner);
 687                 release_open_stateid(stp);
 688         }
 689 }
 690
 691 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 692 {
 693         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 694
 695         if (s) {
 696                 free_generic_stateid(s);
 697                 oo->oo_last_closed_stid = NULL;
 698         }
 699 }
 700
 701 static void release_openowner(struct nfs4_openowner *oo)
 702 {
 703         unhash_openowner(oo);
 704         list_del(&oo->oo_close_lru);
 705         release_last_closed_stateid(oo);
 706         nfs4_free_openowner(oo);
 707 }
 708
 709 static inline int
 710 hash_sessionid(struct nfs4_sessionid *sessionid)
 711 {
 712         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 713
 714         return sid->sequence % SESSION_HASH_SIZE;
 715 }
 716
 717 #ifdef NFSD_DEBUG
 718 static inline void
 719 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 720 {
 721         u32 *ptr = (u32 *)(&sessionid->data[0]);
 722         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 723 }
 724 #else
 725 static inline void
 726 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 727 {
 728 }
 729 #endif
 730
 731 /*
 732  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
 733  * won't be used for replay.
 734  */
 735 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
 736 {
 737         struct nfs4_stateowner *so = cstate->replay_owner;
 738
 739         if (nfserr == nfserr_replay_me)
 740                 return;
 741
 742         if (!seqid_mutating_err(ntohl(nfserr))) {
 743                 cstate->replay_owner = NULL;
 744                 return;
 745         }
 746         if (!so)
 747                 return;
 748         if (so->so_is_open_owner)
 749                 release_last_closed_stateid(openowner(so));
 750         so->so_seqid++;
 751         return;
 752 }
 753
 754 static void
 755 gen_sessionid(struct nfsd4_session *ses)
 756 {
 757         struct nfs4_client *clp = ses->se_client;
 758         struct nfsd4_sessionid *sid;
 759
 760         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 761         sid->clientid = clp->cl_clientid;
 762         sid->sequence = current_sessionid++;
 763         sid->reserved = 0;
 764 }
 765
 766 /*
 767  * The protocol defines ca_maxresponssize_cached to include the size of
 768  * the rpc header, but all we need to cache is the data starting after
 769  * the end of the initial SEQUENCE operation--the rest we regenerate
 770  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 771  * value that is the number of bytes in our cache plus a few additional
 772  * bytes.  In order to stay on the safe side, and not promise more than
 773  * we can cache, those additional bytes must be the minimum possible: 24
 774  * bytes of rpc header (xid through accept state, with AUTH_NULL
 775  * verifier), 12 for the compound header (with zero-length tag), and 44
 776  * for the SEQUENCE op response:
 777  */
 778 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 779
 780 static void
 781 free_session_slots(struct nfsd4_session *ses)
 782 {
 783         int i;
 784
 785         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 786                 kfree(ses->se_slots[i]);
 787 }
 788
 789 /*
 790  * We don't actually need to cache the rpc and session headers, so we
 791  * can allocate a little less for each slot:
 792  */
 793 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
 794 {
 795         u32 size;
 796
 797         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
 798                 size = 0;
 799         else
 800                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 801         return size + sizeof(struct nfsd4_slot);
 802 }
 803
 804 /*
 805  * XXX: If we run out of reserved DRC memory we could (up to a point)
 806  * re-negotiate active sessions and reduce their slot usage to make
 807  * room for new connections. For now we just fail the create session.
 808  */
 809 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
 810 {
 811         u32 slotsize = slot_bytes(ca);
 812         u32 num = ca->maxreqs;
 813         int avail;
 814
 815         spin_lock(&nfsd_drc_lock);
 816         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
 817                     nfsd_drc_max_mem - nfsd_drc_mem_used);
 818         num = min_t(int, num, avail / slotsize);
 819         nfsd_drc_mem_used += num * slotsize;
 820         spin_unlock(&nfsd_drc_lock);
 821
 822         return num;
 823 }
 824
 825 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
 826 {
 827         int slotsize = slot_bytes(ca);
 828
 829         spin_lock(&nfsd_drc_lock);
 830         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
 831         spin_unlock(&nfsd_drc_lock);
 832 }
 833
 834 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *attrs)
 835 {
 836         int numslots = attrs->maxreqs;
 837         int slotsize = slot_bytes(attrs);
 838         struct nfsd4_session *new;
 839         int mem, i;
 840
 841         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 842                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 843         mem = numslots * sizeof(struct nfsd4_slot *);
 844
 845         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 846         if (!new)
 847                 return NULL;
 848         /* allocate each struct nfsd4_slot and data cache in one piece */
 849         for (i = 0; i < numslots; i++) {
 850                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
 851                 if (!new->se_slots[i])
 852                         goto out_free;
 853         }
 854         return new;
 855 out_free:
 856         while (i--)
 857                 kfree(new->se_slots[i]);
 858         kfree(new);
 859         return NULL;
 860 }
 861
 862 static void free_conn(struct nfsd4_conn *c)
 863 {
 864         svc_xprt_put(c->cn_xprt);
 865         kfree(c);
 866 }
 867
 868 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 869 {
 870         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 871         struct nfs4_client *clp = c->cn_session->se_client;
 872
 873         spin_lock(&clp->cl_lock);
 874         if (!list_empty(&c->cn_persession)) {
 875                 list_del(&c->cn_persession);
 876                 free_conn(c);
 877         }
 878         nfsd4_probe_callback(clp);
 879         spin_unlock(&clp->cl_lock);
 880 }
 881
 882 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 883 {
 884         struct nfsd4_conn *conn;
 885
 886         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 887         if (!conn)
 888                 return NULL;
 889         svc_xprt_get(rqstp->rq_xprt);
 890         conn->cn_xprt = rqstp->rq_xprt;
 891         conn->cn_flags = flags;
 892         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 893         return conn;
 894 }
 895
 896 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 897 {
 898         conn->cn_session = ses;
 899         list_add(&conn->cn_persession, &ses->se_conns);
 900 }
 901
 902 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 903 {
 904         struct nfs4_client *clp = ses->se_client;
 905
 906         spin_lock(&clp->cl_lock);
 907         __nfsd4_hash_conn(conn, ses);
 908         spin_unlock(&clp->cl_lock);
 909 }
 910
 911 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 912 {
 913         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 914         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 915 }
 916
 917 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 918 {
 919         int ret;
 920
 921         nfsd4_hash_conn(conn, ses);
 922         ret = nfsd4_register_conn(conn);
 923         if (ret)
 924                 /* oops; xprt is already down: */
 925                 nfsd4_conn_lost(&conn->cn_xpt_user);
 926         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 927                 /* callback channel may be back up */
 928                 nfsd4_probe_callback(ses->se_client);
 929         }
 930 }
 931
 932 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 933 {
 934         u32 dir = NFS4_CDFC4_FORE;
 935
 936         if (cses->flags & SESSION4_BACK_CHAN)
 937                 dir |= NFS4_CDFC4_BACK;
 938         return alloc_conn(rqstp, dir);
 939 }
 940
 941 /* must be called under client_lock */
 942 static void nfsd4_del_conns(struct nfsd4_session *s)
 943 {
 944         struct nfs4_client *clp = s->se_client;
 945         struct nfsd4_conn *c;
 946
 947         spin_lock(&clp->cl_lock);
 948         while (!list_empty(&s->se_conns)) {
 949                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 950                 list_del_init(&c->cn_persession);
 951                 spin_unlock(&clp->cl_lock);
 952
 953                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 954                 free_conn(c);
 955
 956                 spin_lock(&clp->cl_lock);
 957         }
 958         spin_unlock(&clp->cl_lock);
 959 }
 960
 961 static void __free_session(struct nfsd4_session *ses)
 962 {
 963         free_session_slots(ses);
 964         kfree(ses);
 965 }
 966
 967 static void free_session(struct nfsd4_session *ses)
 968 {
 969         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
 970
 971         lockdep_assert_held(&nn->client_lock);
 972         nfsd4_del_conns(ses);
 973         nfsd4_put_drc_mem(&ses->se_fchannel);
 974         __free_session(ses);
 975 }
 976
 977 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 978 {
 979         int idx;
 980         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 981
 982         new->se_client = clp;
 983         gen_sessionid(new);
 984
 985         INIT_LIST_HEAD(&new->se_conns);
 986
 987         new->se_cb_seq_nr = 1;
 988         new->se_flags = cses->flags;
 989         new->se_cb_prog = cses->callback_prog;
 990         new->se_cb_sec = cses->cb_sec;
 991         atomic_set(&new->se_ref, 0);
 992         idx = hash_sessionid(&new->se_sessionid);
 993         spin_lock(&nn->client_lock);
 994         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
 995         spin_lock(&clp->cl_lock);
 996         list_add(&new->se_perclnt, &clp->cl_sessions);
 997         spin_unlock(&clp->cl_lock);
 998         spin_unlock(&nn->client_lock);
 999         memcpy(&new->se_fchannel, &cses->fore_channel,
1000                         sizeof(struct nfsd4_channel_attrs));
1001         if (cses->flags & SESSION4_BACK_CHAN) {
1002                 struct sockaddr *sa = svc_addr(rqstp);
1003                 /*
1004                  * This is a little silly; with sessions there's no real
1005                  * use for the callback address.  Use the peer address
1006                  * as a reasonable default for now, but consider fixing
1007                  * the rpc client not to require an address in the
1008                  * future:
1009                  */
1010                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1011                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1012         }
1013 }
1014
1015 /* caller must hold client_lock */
1016 static struct nfsd4_session *
1017 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1018 {
1019         struct nfsd4_session *elem;
1020         int idx;
1021         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1022
1023         dump_sessionid(__func__, sessionid);
1024         idx = hash_sessionid(sessionid);
1025         /* Search in the appropriate list */
1026         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1027                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1028                             NFS4_MAX_SESSIONID_LEN)) {
1029                         return elem;
1030                 }
1031         }
1032
1033         dprintk("%s: session not found\n", __func__);
1034         return NULL;
1035 }
1036
1037 /* caller must hold client_lock */
1038 static void
1039 unhash_session(struct nfsd4_session *ses)
1040 {
1041         list_del(&ses->se_hash);
1042         spin_lock(&ses->se_client->cl_lock);
1043         list_del(&ses->se_perclnt);
1044         spin_unlock(&ses->se_client->cl_lock);
1045 }
1046
1047 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1048 static int
1049 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1050 {
1051         if (clid->cl_boot == nn->boot_time)
1052                 return 0;
1053         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1054                 clid->cl_boot, clid->cl_id, nn->boot_time);
1055         return 1;
1056 }
1057
1058 /*
1059  * XXX Should we use a slab cache ?
1060  * This type of memory management is somewhat inefficient, but we use it
1061  * anyway since SETCLIENTID is not a common operation.
1062  */
1063 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1064 {
1065         struct nfs4_client *clp;
1066
1067         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1068         if (clp == NULL)
1069                 return NULL;
1070         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1071         if (clp->cl_name.data == NULL) {
1072                 kfree(clp);
1073                 return NULL;
1074         }
1075         clp->cl_name.len = name.len;
1076         return clp;
1077 }
1078
1079 static inline void
1080 free_client(struct nfs4_client *clp)
1081 {
1082         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1083
1084         lockdep_assert_held(&nn->client_lock);
1085         while (!list_empty(&clp->cl_sessions)) {
1086                 struct nfsd4_session *ses;
1087                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1088                                 se_perclnt);
1089                 list_del(&ses->se_perclnt);
1090                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1091                 free_session(ses);
1092         }
1093         free_svc_cred(&clp->cl_cred);
1094         kfree(clp->cl_name.data);
1095         idr_destroy(&clp->cl_stateids);
1096         kfree(clp);
1097 }
1098
1099 /* must be called under the client_lock */
1100 static inline void
1101 unhash_client_locked(struct nfs4_client *clp)
1102 {
1103         struct nfsd4_session *ses;
1104
1105         list_del(&clp->cl_lru);
1106         spin_lock(&clp->cl_lock);
1107         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1108                 list_del_init(&ses->se_hash);
1109         spin_unlock(&clp->cl_lock);
1110 }
1111
1112 static void
1113 destroy_client(struct nfs4_client *clp)
1114 {
1115         struct nfs4_openowner *oo;
1116         struct nfs4_delegation *dp;
1117         struct list_head reaplist;
1118         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1119
1120         INIT_LIST_HEAD(&reaplist);
1121         spin_lock(&recall_lock);
1122         while (!list_empty(&clp->cl_delegations)) {
1123                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1124                 list_del_init(&dp->dl_perclnt);
1125                 list_move(&dp->dl_recall_lru, &reaplist);
1126         }
1127         spin_unlock(&recall_lock);
1128         while (!list_empty(&reaplist)) {
1129                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1130                 destroy_delegation(dp);
1131         }
1132         while (!list_empty(&clp->cl_openowners)) {
1133                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1134                 release_openowner(oo);
1135         }
1136         nfsd4_shutdown_callback(clp);
1137         if (clp->cl_cb_conn.cb_xprt)
1138                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1139         list_del(&clp->cl_idhash);
1140         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1141                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1142         else
1143                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1144         spin_lock(&nn->client_lock);
1145         unhash_client_locked(clp);
1146         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1147         free_client(clp);
1148         spin_unlock(&nn->client_lock);
1149 }
1150
1151 static void expire_client(struct nfs4_client *clp)
1152 {
1153         nfsd4_client_record_remove(clp);
1154         destroy_client(clp);
1155 }
1156
1157 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1158 {
1159         memcpy(target->cl_verifier.data, source->data,
1160                         sizeof(target->cl_verifier.data));
1161 }
1162
1163 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1164 {
1165         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1166         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1167 }
1168
1169 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1170 {
1171         if (source->cr_principal) {
1172                 target->cr_principal =
1173                                 kstrdup(source->cr_principal, GFP_KERNEL);
1174                 if (target->cr_principal == NULL)
1175                         return -ENOMEM;
1176         } else
1177                 target->cr_principal = NULL;
1178         target->cr_flavor = source->cr_flavor;
1179         target->cr_uid = source->cr_uid;
1180         target->cr_gid = source->cr_gid;
1181         target->cr_group_info = source->cr_group_info;
1182         get_group_info(target->cr_group_info);
1183         target->cr_gss_mech = source->cr_gss_mech;
1184         if (source->cr_gss_mech)
1185                 gss_mech_get(source->cr_gss_mech);
1186         return 0;
1187 }
1188
1189 static long long
1190 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1191 {
1192         long long res;
1193
1194         res = o1->len - o2->len;
1195         if (res)
1196                 return res;
1197         return (long long)memcmp(o1->data, o2->data, o1->len);
1198 }
1199
1200 static int same_name(const char *n1, const char *n2)
1201 {
1202         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1203 }
1204
1205 static int
1206 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1207 {
1208         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1209 }
1210
1211 static int
1212 same_clid(clientid_t *cl1, clientid_t *cl2)
1213 {
1214         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1215 }
1216
1217 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1218 {
1219         int i;
1220
1221         if (g1->ngroups != g2->ngroups)
1222                 return false;
1223         for (i=0; i<g1->ngroups; i++)
1224                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1225                         return false;
1226         return true;
1227 }
1228
1229 /*
1230  * RFC 3530 language requires clid_inuse be returned when the
1231  * "principal" associated with a requests differs from that previously
1232  * used.  We use uid, gid's, and gss principal string as our best
1233  * approximation.  We also don't want to allow non-gss use of a client
1234  * established using gss: in theory cr_principal should catch that
1235  * change, but in practice cr_principal can be null even in the gss case
1236  * since gssd doesn't always pass down a principal string.
1237  */
1238 static bool is_gss_cred(struct svc_cred *cr)
1239 {
1240         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1241         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1242 }
1243
1244
1245 static bool
1246 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1247 {
1248         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1249                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1250                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1251                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1252                 return false;
1253         if (cr1->cr_principal == cr2->cr_principal)
1254                 return true;
1255         if (!cr1->cr_principal || !cr2->cr_principal)
1256                 return false;
1257         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1258 }
1259
1260 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1261 {
1262         struct svc_cred *cr = &rqstp->rq_cred;
1263         u32 service;
1264
1265         if (!cr->cr_gss_mech)
1266                 return false;
1267         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1268         return service == RPC_GSS_SVC_INTEGRITY ||
1269                service == RPC_GSS_SVC_PRIVACY;
1270 }
1271
1272 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1273 {
1274         struct svc_cred *cr = &rqstp->rq_cred;
1275
1276         if (!cl->cl_mach_cred)
1277                 return true;
1278         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1279                 return false;
1280         if (!svc_rqst_integrity_protected(rqstp))
1281                 return false;
1282         if (!cr->cr_principal)
1283                 return false;
1284         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1285 }
1286
1287 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1288 {
1289         static u32 current_clientid = 1;
1290
1291         clp->cl_clientid.cl_boot = nn->boot_time;
1292         clp->cl_clientid.cl_id = current_clientid++;
1293 }
1294
1295 static void gen_confirm(struct nfs4_client *clp)
1296 {
1297         __be32 verf[2];
1298         static u32 i;
1299
1300         verf[0] = (__be32)get_seconds();
1301         verf[1] = (__be32)i++;
1302         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1303 }
1304
1305 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1306 {
1307         struct nfs4_stid *ret;
1308
1309         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1310         if (!ret || !ret->sc_type)
1311                 return NULL;
1312         return ret;
1313 }
1314
1315 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1316 {
1317         struct nfs4_stid *s;
1318
1319         s = find_stateid(cl, t);
1320         if (!s)
1321                 return NULL;
1322         if (typemask & s->sc_type)
1323                 return s;
1324         return NULL;
1325 }
1326
1327 static struct nfs4_client *create_client(struct xdr_netobj name,
1328                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1329 {
1330         struct nfs4_client *clp;
1331         struct sockaddr *sa = svc_addr(rqstp);
1332         int ret;
1333         struct net *net = SVC_NET(rqstp);
1334         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1335
1336         clp = alloc_client(name);
1337         if (clp == NULL)
1338                 return NULL;
1339
1340         INIT_LIST_HEAD(&clp->cl_sessions);
1341         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1342         if (ret) {
1343                 spin_lock(&nn->client_lock);
1344                 free_client(clp);
1345                 spin_unlock(&nn->client_lock);
1346                 return NULL;
1347         }
1348         idr_init(&clp->cl_stateids);
1349         atomic_set(&clp->cl_refcount, 0);
1350         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1351         INIT_LIST_HEAD(&clp->cl_idhash);
1352         INIT_LIST_HEAD(&clp->cl_openowners);
1353         INIT_LIST_HEAD(&clp->cl_delegations);
1354         INIT_LIST_HEAD(&clp->cl_lru);
1355         INIT_LIST_HEAD(&clp->cl_callbacks);
1356         INIT_LIST_HEAD(&clp->cl_revoked);
1357         spin_lock_init(&clp->cl_lock);
1358         nfsd4_init_callback(&clp->cl_cb_null);
1359         clp->cl_time = get_seconds();
1360         clear_bit(0, &clp->cl_cb_slot_busy);
1361         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1362         copy_verf(clp, verf);
1363         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1364         gen_confirm(clp);
1365         clp->cl_cb_session = NULL;
1366         clp->net = net;
1367         return clp;
1368 }
1369
1370 static void
1371 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1372 {
1373         struct rb_node **new = &(root->rb_node), *parent = NULL;
1374         struct nfs4_client *clp;
1375
1376         while (*new) {
1377                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1378                 parent = *new;
1379
1380                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1381                         new = &((*new)->rb_left);
1382                 else
1383                         new = &((*new)->rb_right);
1384         }
1385
1386         rb_link_node(&new_clp->cl_namenode, parent, new);
1387         rb_insert_color(&new_clp->cl_namenode, root);
1388 }
1389
1390 static struct nfs4_client *
1391 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1392 {
1393         long long cmp;
1394         struct rb_node *node = root->rb_node;
1395         struct nfs4_client *clp;
1396
1397         while (node) {
1398                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1399                 cmp = compare_blob(&clp->cl_name, name);
1400                 if (cmp > 0)
1401                         node = node->rb_left;
1402                 else if (cmp < 0)
1403                         node = node->rb_right;
1404                 else
1405                         return clp;
1406         }
1407         return NULL;
1408 }
1409
1410 static void
1411 add_to_unconfirmed(struct nfs4_client *clp)
1412 {
1413         unsigned int idhashval;
1414         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1415
1416         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1417         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1418         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1419         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1420         renew_client(clp);
1421 }
1422
1423 static void
1424 move_to_confirmed(struct nfs4_client *clp)
1425 {
1426         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1427         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1428
1429         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1430         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1431         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1432         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1433         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1434         renew_client(clp);
1435 }
1436
1437 static struct nfs4_client *
1438 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1439 {
1440         struct nfs4_client *clp;
1441         unsigned int idhashval = clientid_hashval(clid->cl_id);
1442
1443         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1444                 if (same_clid(&clp->cl_clientid, clid)) {
1445                         if ((bool)clp->cl_minorversion != sessions)
1446                                 return NULL;
1447                         renew_client(clp);
1448                         return clp;
1449                 }
1450         }
1451         return NULL;
1452 }
1453
1454 static struct nfs4_client *
1455 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1456 {
1457         struct list_head *tbl = nn->conf_id_hashtbl;
1458
1459         return find_client_in_id_table(tbl, clid, sessions);
1460 }
1461
1462 static struct nfs4_client *
1463 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1464 {
1465         struct list_head *tbl = nn->unconf_id_hashtbl;
1466
1467         return find_client_in_id_table(tbl, clid, sessions);
1468 }
1469
1470 static bool clp_used_exchangeid(struct nfs4_client *clp)
1471 {
1472         return clp->cl_exchange_flags != 0;
1473 }
1474
1475 static struct nfs4_client *
1476 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1477 {
1478         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1479 }
1480
1481 static struct nfs4_client *
1482 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1483 {
1484         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1485 }
1486
1487 static void
1488 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1489 {
1490         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1491         struct sockaddr *sa = svc_addr(rqstp);
1492         u32 scopeid = rpc_get_scope_id(sa);
1493         unsigned short expected_family;
1494
1495         /* Currently, we only support tcp and tcp6 for the callback channel */
1496         if (se->se_callback_netid_len == 3 &&
1497             !memcmp(se->se_callback_netid_val, "tcp", 3))
1498                 expected_family = AF_INET;
1499         else if (se->se_callback_netid_len == 4 &&
1500                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1501                 expected_family = AF_INET6;
1502         else
1503                 goto out_err;
1504
1505         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1506                                             se->se_callback_addr_len,
1507                                             (struct sockaddr *)&conn->cb_addr,
1508                                             sizeof(conn->cb_addr));
1509
1510         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1511                 goto out_err;
1512
1513         if (conn->cb_addr.ss_family == AF_INET6)
1514                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1515
1516         conn->cb_prog = se->se_callback_prog;
1517         conn->cb_ident = se->se_callback_ident;
1518         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1519         return;
1520 out_err:
1521         conn->cb_addr.ss_family = AF_UNSPEC;
1522         conn->cb_addrlen = 0;
1523         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1524                 "will not receive delegations\n",
1525                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1526
1527         return;
1528 }
1529
1530 /*
1531  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1532  */
1533 void
1534 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1535 {
1536         struct nfsd4_slot *slot = resp->cstate.slot;
1537         unsigned int base;
1538
1539         dprintk("--> %s slot %p\n", __func__, slot);
1540
1541         slot->sl_opcnt = resp->opcnt;
1542         slot->sl_status = resp->cstate.status;
1543
1544         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1545         if (nfsd4_not_cached(resp)) {
1546                 slot->sl_datalen = 0;
1547                 return;
1548         }
1549         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1550         base = (char *)resp->cstate.datap -
1551                                         (char *)resp->xbuf->head[0].iov_base;
1552         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1553                                     slot->sl_datalen))
1554                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1555         return;
1556 }
1557
1558 /*
1559  * Encode the replay sequence operation from the slot values.
1560  * If cachethis is FALSE encode the uncached rep error on the next
1561  * operation which sets resp->p and increments resp->opcnt for
1562  * nfs4svc_encode_compoundres.
1563  *
1564  */
1565 static __be32
1566 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1567                           struct nfsd4_compoundres *resp)
1568 {
1569         struct nfsd4_op *op;
1570         struct nfsd4_slot *slot = resp->cstate.slot;
1571
1572         /* Encode the replayed sequence operation */
1573         op = &args->ops[resp->opcnt - 1];
1574         nfsd4_encode_operation(resp, op);
1575
1576         /* Return nfserr_retry_uncached_rep in next operation. */
1577         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1578                 op = &args->ops[resp->opcnt++];
1579                 op->status = nfserr_retry_uncached_rep;
1580                 nfsd4_encode_operation(resp, op);
1581         }
1582         return op->status;
1583 }
1584
1585 /*
1586  * The sequence operation is not cached because we can use the slot and
1587  * session values.
1588  */
1589 __be32
1590 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1591                          struct nfsd4_sequence *seq)
1592 {
1593         struct nfsd4_slot *slot = resp->cstate.slot;
1594         __be32 status;
1595
1596         dprintk("--> %s slot %p\n", __func__, slot);
1597
1598         /* Either returns 0 or nfserr_retry_uncached */
1599         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1600         if (status == nfserr_retry_uncached_rep)
1601                 return status;
1602
1603         /* The sequence operation has been encoded, cstate->datap set. */
1604         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1605
1606         resp->opcnt = slot->sl_opcnt;
1607         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1608         status = slot->sl_status;
1609
1610         return status;
1611 }
1612
1613 /*
1614  * Set the exchange_id flags returned by the server.
1615  */
1616 static void
1617 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1618 {
1619         /* pNFS is not supported */
1620         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1621
1622         /* Referrals are supported, Migration is not. */
1623         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1624
1625         /* set the wire flags to return to client. */
1626         clid->flags = new->cl_exchange_flags;
1627 }
1628
1629 static bool client_has_state(struct nfs4_client *clp)
1630 {
1631         /*
1632          * Note clp->cl_openowners check isn't quite right: there's no
1633          * need to count owners without stateid's.
1634          *
1635          * Also note we should probably be using this in 4.0 case too.
1636          */
1637         return !list_empty(&clp->cl_openowners)
1638                 || !list_empty(&clp->cl_delegations)
1639                 || !list_empty(&clp->cl_sessions);
1640 }
1641
1642 __be32
1643 nfsd4_exchange_id(struct svc_rqst *rqstp,
1644                   struct nfsd4_compound_state *cstate,
1645                   struct nfsd4_exchange_id *exid)
1646 {
1647         struct nfs4_client *unconf, *conf, *new;
1648         __be32 status;
1649         char                    addr_str[INET6_ADDRSTRLEN];
1650         nfs4_verifier           verf = exid->verifier;
1651         struct sockaddr         *sa = svc_addr(rqstp);
1652         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1653         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1654
1655         rpc_ntop(sa, addr_str, sizeof(addr_str));
1656         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1657                 "ip_addr=%s flags %x, spa_how %d\n",
1658                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1659                 addr_str, exid->flags, exid->spa_how);
1660
1661         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1662                 return nfserr_inval;
1663
1664         switch (exid->spa_how) {
1665         case SP4_MACH_CRED:
1666                 if (!svc_rqst_integrity_protected(rqstp))
1667                         return nfserr_inval;
1668         case SP4_NONE:
1669                 break;
1670         default:                                /* checked by xdr code */
1671                 WARN_ON_ONCE(1);
1672         case SP4_SSV:
1673                 return nfserr_encr_alg_unsupp;
1674         }
1675
1676         /* Cases below refer to rfc 5661 section 18.35.4: */
1677         nfs4_lock_state();
1678         conf = find_confirmed_client_by_name(&exid->clname, nn);
1679         if (conf) {
1680                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1681                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1682
1683                 if (update) {
1684                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1685                                 status = nfserr_inval;
1686                                 goto out;
1687                         }
1688                         if (!mach_creds_match(conf, rqstp)) {
1689                                 status = nfserr_wrong_cred;
1690                                 goto out;
1691                         }
1692                         if (!creds_match) { /* case 9 */
1693                                 status = nfserr_perm;
1694                                 goto out;
1695                         }
1696                         if (!verfs_match) { /* case 8 */
1697                                 status = nfserr_not_same;
1698                                 goto out;
1699                         }
1700                         /* case 6 */
1701                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1702                         new = conf;
1703                         goto out_copy;
1704                 }
1705                 if (!creds_match) { /* case 3 */
1706                         if (client_has_state(conf)) {
1707                                 status = nfserr_clid_inuse;
1708                                 goto out;
1709                         }
1710                         expire_client(conf);
1711                         goto out_new;
1712                 }
1713                 if (verfs_match) { /* case 2 */
1714                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1715                         new = conf;
1716                         goto out_copy;
1717                 }
1718                 /* case 5, client reboot */
1719                 goto out_new;
1720         }
1721
1722         if (update) { /* case 7 */
1723                 status = nfserr_noent;
1724                 goto out;
1725         }
1726
1727         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1728         if (unconf) /* case 4, possible retry or client restart */
1729                 expire_client(unconf);
1730
1731         /* case 1 (normal case) */
1732 out_new:
1733         new = create_client(exid->clname, rqstp, &verf);
1734         if (new == NULL) {
1735                 status = nfserr_jukebox;
1736                 goto out;
1737         }
1738         new->cl_minorversion = cstate->minorversion;
1739         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
1740
1741         gen_clid(new, nn);
1742         add_to_unconfirmed(new);
1743 out_copy:
1744         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1745         exid->clientid.cl_id = new->cl_clientid.cl_id;
1746
1747         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1748         nfsd4_set_ex_flags(new, exid);
1749
1750         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1751                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1752         status = nfs_ok;
1753
1754 out:
1755         nfs4_unlock_state();
1756         return status;
1757 }
1758
1759 static __be32
1760 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1761 {
1762         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1763                 slot_seqid);
1764
1765         /* The slot is in use, and no response has been sent. */
1766         if (slot_inuse) {
1767                 if (seqid == slot_seqid)
1768                         return nfserr_jukebox;
1769                 else
1770                         return nfserr_seq_misordered;
1771         }
1772         /* Note unsigned 32-bit arithmetic handles wraparound: */
1773         if (likely(seqid == slot_seqid + 1))
1774                 return nfs_ok;
1775         if (seqid == slot_seqid)
1776                 return nfserr_replay_cache;
1777         return nfserr_seq_misordered;
1778 }
1779
1780 /*
1781  * Cache the create session result into the create session single DRC
1782  * slot cache by saving the xdr structure. sl_seqid has been set.
1783  * Do this for solo or embedded create session operations.
1784  */
1785 static void
1786 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1787                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1788 {
1789         slot->sl_status = nfserr;
1790         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1791 }
1792
1793 static __be32
1794 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1795                             struct nfsd4_clid_slot *slot)
1796 {
1797         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1798         return slot->sl_status;
1799 }
1800
1801 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1802                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1803                         1 +     /* MIN tag is length with zero, only length */ \
1804                         3 +     /* version, opcount, opcode */ \
1805                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1806                                 /* seqid, slotID, slotID, cache */ \
1807                         4 ) * sizeof(__be32))
1808
1809 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1810                         2 +     /* verifier: AUTH_NULL, length 0 */\
1811                         1 +     /* status */ \
1812                         1 +     /* MIN tag is length with zero, only length */ \
1813                         3 +     /* opcount, opcode, opstatus*/ \
1814                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1815                                 /* seqid, slotID, slotID, slotID, status */ \
1816                         5 ) * sizeof(__be32))
1817
1818 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1819 {
1820         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
1821
1822         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
1823                 return nfserr_toosmall;
1824         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
1825                 return nfserr_toosmall;
1826         ca->headerpadsz = 0;
1827         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
1828         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
1829         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
1830         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
1831                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
1832         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
1833         /*
1834          * Note decreasing slot size below client's request may make it
1835          * difficult for client to function correctly, whereas
1836          * decreasing the number of slots will (just?) affect
1837          * performance.  When short on memory we therefore prefer to
1838          * decrease number of slots instead of their size.  Clients that
1839          * request larger slots than they need will get poor results:
1840          */
1841         ca->maxreqs = nfsd4_get_drc_mem(ca);
1842         if (!ca->maxreqs)
1843                 return nfserr_jukebox;
1844
1845         return nfs_ok;
1846 }
1847
1848 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
1849 {
1850         ca->headerpadsz = 0;
1851
1852         /*
1853          * These RPC_MAX_HEADER macros are overkill, especially since we
1854          * don't even do gss on the backchannel yet.  But this is still
1855          * less than 1k.  Tighten up this estimate in the unlikely event
1856          * it turns out to be a problem for some client:
1857          */
1858         if (ca->maxreq_sz < NFS4_enc_cb_recall_sz + RPC_MAX_HEADER_WITH_AUTH)
1859                 return nfserr_toosmall;
1860         if (ca->maxresp_sz < NFS4_dec_cb_recall_sz + RPC_MAX_REPHEADER_WITH_AUTH)
1861                 return nfserr_toosmall;
1862         ca->maxresp_cached = 0;
1863         if (ca->maxops < 2)
1864                 return nfserr_toosmall;
1865
1866         return nfs_ok;
1867 }
1868
1869 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
1870 {
1871         switch (cbs->flavor) {
1872         case RPC_AUTH_NULL:
1873         case RPC_AUTH_UNIX:
1874                 return nfs_ok;
1875         default:
1876                 /*
1877                  * GSS case: the spec doesn't allow us to return this
1878                  * error.  But it also doesn't allow us not to support
1879                  * GSS.
1880                  * I'd rather this fail hard than return some error the
1881                  * client might think it can already handle:
1882                  */
1883                 return nfserr_encr_alg_unsupp;
1884         }
1885 }
1886
1887 __be32
1888 nfsd4_create_session(struct svc_rqst *rqstp,
1889                      struct nfsd4_compound_state *cstate,
1890                      struct nfsd4_create_session *cr_ses)
1891 {
1892         struct sockaddr *sa = svc_addr(rqstp);
1893         struct nfs4_client *conf, *unconf;
1894         struct nfsd4_session *new;
1895         struct nfsd4_conn *conn;
1896         struct nfsd4_clid_slot *cs_slot = NULL;
1897         __be32 status = 0;
1898         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1899
1900         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1901                 return nfserr_inval;
1902         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
1903         if (status)
1904                 return status;
1905         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
1906         if (status)
1907                 return status;
1908         status = check_backchannel_attrs(&cr_ses->back_channel);
1909         if (status)
1910                 return status;
1911         status = nfserr_jukebox;
1912         new = alloc_session(&cr_ses->fore_channel);
1913         if (!new)
1914                 goto out_release_drc_mem;
1915         conn = alloc_conn_from_crses(rqstp, cr_ses);
1916         if (!conn)
1917                 goto out_free_session;
1918
1919         nfs4_lock_state();
1920         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1921         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1922         WARN_ON_ONCE(conf && unconf);
1923
1924         if (conf) {
1925                 status = nfserr_wrong_cred;
1926                 if (!mach_creds_match(conf, rqstp))
1927                         goto out_free_conn;
1928                 cs_slot = &conf->cl_cs_slot;
1929                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1930                 if (status == nfserr_replay_cache) {
1931                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1932                         goto out_free_conn;
1933                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1934                         status = nfserr_seq_misordered;
1935                         goto out_free_conn;
1936                 }
1937         } else if (unconf) {
1938                 struct nfs4_client *old;
1939                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1940                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1941                         status = nfserr_clid_inuse;
1942                         goto out_free_conn;
1943                 }
1944                 status = nfserr_wrong_cred;
1945                 if (!mach_creds_match(unconf, rqstp))
1946                         goto out_free_conn;
1947                 cs_slot = &unconf->cl_cs_slot;
1948                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1949                 if (status) {
1950                         /* an unconfirmed replay returns misordered */
1951                         status = nfserr_seq_misordered;
1952                         goto out_free_conn;
1953                 }
1954                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1955                 if (old) {
1956                         status = mark_client_expired(old);
1957                         if (status)
1958                                 goto out_free_conn;
1959                         expire_client(old);
1960                 }
1961                 move_to_confirmed(unconf);
1962                 conf = unconf;
1963         } else {
1964                 status = nfserr_stale_clientid;
1965                 goto out_free_conn;
1966         }
1967         status = nfs_ok;
1968         /*
1969          * We do not support RDMA or persistent sessions
1970          */
1971         cr_ses->flags &= ~SESSION4_PERSIST;
1972         cr_ses->flags &= ~SESSION4_RDMA;
1973
1974         init_session(rqstp, new, conf, cr_ses);
1975         nfsd4_init_conn(rqstp, conn, new);
1976
1977         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1978                NFS4_MAX_SESSIONID_LEN);
1979         cs_slot->sl_seqid++;
1980         cr_ses->seqid = cs_slot->sl_seqid;
1981
1982         /* cache solo and embedded create sessions under the state lock */
1983         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1984         nfs4_unlock_state();
1985         return status;
1986 out_free_conn:
1987         nfs4_unlock_state();
1988         free_conn(conn);
1989 out_free_session:
1990         __free_session(new);
1991 out_release_drc_mem:
1992         nfsd4_put_drc_mem(&cr_ses->fore_channel);
1993         return status;
1994 }
1995
1996 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1997 {
1998         switch (*dir) {
1999         case NFS4_CDFC4_FORE:
2000         case NFS4_CDFC4_BACK:
2001                 return nfs_ok;
2002         case NFS4_CDFC4_FORE_OR_BOTH:
2003         case NFS4_CDFC4_BACK_OR_BOTH:
2004                 *dir = NFS4_CDFC4_BOTH;
2005                 return nfs_ok;
2006         };
2007         return nfserr_inval;
2008 }
2009
2010 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2011 {
2012         struct nfsd4_session *session = cstate->session;
2013         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2014         __be32 status;
2015
2016         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2017         if (status)
2018                 return status;
2019         spin_lock(&nn->client_lock);
2020         session->se_cb_prog = bc->bc_cb_program;
2021         session->se_cb_sec = bc->bc_cb_sec;
2022         spin_unlock(&nn->client_lock);
2023
2024         nfsd4_probe_callback(session->se_client);
2025
2026         return nfs_ok;
2027 }
2028
2029 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2030                      struct nfsd4_compound_state *cstate,
2031                      struct nfsd4_bind_conn_to_session *bcts)
2032 {
2033         __be32 status;
2034         struct nfsd4_conn *conn;
2035         struct nfsd4_session *session;
2036         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2037
2038         if (!nfsd4_last_compound_op(rqstp))
2039                 return nfserr_not_only_op;
2040         nfs4_lock_state();
2041         spin_lock(&nn->client_lock);
2042         session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
2043         spin_unlock(&nn->client_lock);
2044         status = nfserr_badsession;
2045         if (!session)
2046                 goto out;
2047         status = nfserr_wrong_cred;
2048         if (!mach_creds_match(session->se_client, rqstp))
2049                 goto out;
2050         status = nfsd4_map_bcts_dir(&bcts->dir);
2051         if (status)
2052                 goto out;
2053         conn = alloc_conn(rqstp, bcts->dir);
2054         status = nfserr_jukebox;
2055         if (!conn)
2056                 goto out;
2057         nfsd4_init_conn(rqstp, conn, session);
2058         status = nfs_ok;
2059 out:
2060         nfs4_unlock_state();
2061         return status;
2062 }
2063
2064 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2065 {
2066         if (!session)
2067                 return 0;
2068         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2069 }
2070
2071 __be32
2072 nfsd4_destroy_session(struct svc_rqst *r,
2073                       struct nfsd4_compound_state *cstate,
2074                       struct nfsd4_destroy_session *sessionid)
2075 {
2076         struct nfsd4_session *ses;
2077         __be32 status;
2078         int ref_held_by_me = 0;
2079         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
2080
2081         nfs4_lock_state();
2082         status = nfserr_not_only_op;
2083         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2084                 if (!nfsd4_last_compound_op(r))
2085                         goto out;
2086                 ref_held_by_me++;
2087         }
2088         dump_sessionid(__func__, &sessionid->sessionid);
2089         spin_lock(&nn->client_lock);
2090         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
2091         status = nfserr_badsession;
2092         if (!ses)
2093                 goto out_client_lock;
2094         status = nfserr_wrong_cred;
2095         if (!mach_creds_match(ses->se_client, r))
2096                 goto out_client_lock;
2097         nfsd4_get_session_locked(ses);
2098         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2099         if (status)
2100                 goto out_put_session;
2101         unhash_session(ses);
2102         spin_unlock(&nn->client_lock);
2103
2104         nfsd4_probe_callback_sync(ses->se_client);
2105
2106         spin_lock(&nn->client_lock);
2107         status = nfs_ok;
2108 out_put_session:
2109         nfsd4_put_session(ses);
2110 out_client_lock:
2111         spin_unlock(&nn->client_lock);
2112 out:
2113         nfs4_unlock_state();
2114         return status;
2115 }
2116
2117 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2118 {
2119         struct nfsd4_conn *c;
2120
2121         list_for_each_entry(c, &s->se_conns, cn_persession) {
2122                 if (c->cn_xprt == xpt) {
2123                         return c;
2124                 }
2125         }
2126         return NULL;
2127 }
2128
2129 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2130 {
2131         struct nfs4_client *clp = ses->se_client;
2132         struct nfsd4_conn *c;
2133         __be32 status = nfs_ok;
2134         int ret;
2135
2136         spin_lock(&clp->cl_lock);
2137         c = __nfsd4_find_conn(new->cn_xprt, ses);
2138         if (c)
2139                 goto out_free;
2140         status = nfserr_conn_not_bound_to_session;
2141         if (clp->cl_mach_cred)
2142                 goto out_free;
2143         __nfsd4_hash_conn(new, ses);
2144         spin_unlock(&clp->cl_lock);
2145         ret = nfsd4_register_conn(new);
2146         if (ret)
2147                 /* oops; xprt is already down: */
2148                 nfsd4_conn_lost(&new->cn_xpt_user);
2149         return nfs_ok;
2150 out_free:
2151         spin_unlock(&clp->cl_lock);
2152         free_conn(new);
2153         return status;
2154 }
2155
2156 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2157 {
2158         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2159
2160         return args->opcnt > session->se_fchannel.maxops;
2161 }
2162
2163 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2164                                   struct nfsd4_session *session)
2165 {
2166         struct xdr_buf *xb = &rqstp->rq_arg;
2167
2168         return xb->len > session->se_fchannel.maxreq_sz;
2169 }
2170
2171 __be32
2172 nfsd4_sequence(struct svc_rqst *rqstp,
2173                struct nfsd4_compound_state *cstate,
2174                struct nfsd4_sequence *seq)
2175 {
2176         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2177         struct nfsd4_session *session;
2178         struct nfs4_client *clp;
2179         struct nfsd4_slot *slot;
2180         struct nfsd4_conn *conn;
2181         __be32 status;
2182         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2183
2184         if (resp->opcnt != 1)
2185                 return nfserr_sequence_pos;
2186
2187         /*
2188          * Will be either used or freed by nfsd4_sequence_check_conn
2189          * below.
2190          */
2191         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2192         if (!conn)
2193                 return nfserr_jukebox;
2194
2195         spin_lock(&nn->client_lock);
2196         status = nfserr_badsession;
2197         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2198         if (!session)
2199                 goto out_no_session;
2200         clp = session->se_client;
2201         status = get_client_locked(clp);
2202         if (status)
2203                 goto out_no_session;
2204         status = nfsd4_get_session_locked(session);
2205         if (status)
2206                 goto out_put_client;
2207
2208         status = nfserr_too_many_ops;
2209         if (nfsd4_session_too_many_ops(rqstp, session))
2210                 goto out_put_session;
2211
2212         status = nfserr_req_too_big;
2213         if (nfsd4_request_too_big(rqstp, session))
2214                 goto out_put_session;
2215
2216         status = nfserr_badslot;
2217         if (seq->slotid >= session->se_fchannel.maxreqs)
2218                 goto out_put_session;
2219
2220         slot = session->se_slots[seq->slotid];
2221         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2222
2223         /* We do not negotiate the number of slots yet, so set the
2224          * maxslots to the session maxreqs which is used to encode
2225          * sr_highest_slotid and the sr_target_slot id to maxslots */
2226         seq->maxslots = session->se_fchannel.maxreqs;
2227
2228         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2229                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2230         if (status == nfserr_replay_cache) {
2231                 status = nfserr_seq_misordered;
2232                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2233                         goto out_put_session;
2234                 cstate->slot = slot;
2235                 cstate->session = session;
2236                 /* Return the cached reply status and set cstate->status
2237                  * for nfsd4_proc_compound processing */
2238                 status = nfsd4_replay_cache_entry(resp, seq);
2239                 cstate->status = nfserr_replay_cache;
2240                 goto out;
2241         }
2242         if (status)
2243                 goto out_put_session;
2244
2245         status = nfsd4_sequence_check_conn(conn, session);
2246         conn = NULL;
2247         if (status)
2248                 goto out_put_session;
2249
2250         /* Success! bump slot seqid */
2251         slot->sl_seqid = seq->seqid;
2252         slot->sl_flags |= NFSD4_SLOT_INUSE;
2253         if (seq->cachethis)
2254                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2255         else
2256                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2257
2258         cstate->slot = slot;
2259         cstate->session = session;
2260
2261 out:
2262         switch (clp->cl_cb_state) {
2263         case NFSD4_CB_DOWN:
2264                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2265                 break;
2266         case NFSD4_CB_FAULT:
2267                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2268                 break;
2269         default:
2270                 seq->status_flags = 0;
2271         }
2272         if (!list_empty(&clp->cl_revoked))
2273                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2274 out_no_session:
2275         kfree(conn);
2276         spin_unlock(&nn->client_lock);
2277         return status;
2278 out_put_session:
2279         nfsd4_put_session(session);
2280 out_put_client:
2281         put_client_renew_locked(clp);
2282         goto out_no_session;
2283 }
2284
2285 __be32
2286 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2287 {
2288         struct nfs4_client *conf, *unconf, *clp;
2289         __be32 status = 0;
2290         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2291
2292         nfs4_lock_state();
2293         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2294         conf = find_confirmed_client(&dc->clientid, true, nn);
2295         WARN_ON_ONCE(conf && unconf);
2296
2297         if (conf) {
2298                 clp = conf;
2299
2300                 if (client_has_state(conf)) {
2301                         status = nfserr_clientid_busy;
2302                         goto out;
2303                 }
2304         } else if (unconf)
2305                 clp = unconf;
2306         else {
2307                 status = nfserr_stale_clientid;
2308                 goto out;
2309         }
2310         if (!mach_creds_match(clp, rqstp)) {
2311                 status = nfserr_wrong_cred;
2312                 goto out;
2313         }
2314         expire_client(clp);
2315 out:
2316         nfs4_unlock_state();
2317         return status;
2318 }
2319
2320 __be32
2321 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2322 {
2323         __be32 status = 0;
2324
2325         if (rc->rca_one_fs) {
2326                 if (!cstate->current_fh.fh_dentry)
2327                         return nfserr_nofilehandle;
2328                 /*
2329                  * We don't take advantage of the rca_one_fs case.
2330                  * That's OK, it's optional, we can safely ignore it.
2331                  */
2332                  return nfs_ok;
2333         }
2334
2335         nfs4_lock_state();
2336         status = nfserr_complete_already;
2337         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2338                              &cstate->session->se_client->cl_flags))
2339                 goto out;
2340
2341         status = nfserr_stale_clientid;
2342         if (is_client_expired(cstate->session->se_client))
2343                 /*
2344                  * The following error isn't really legal.
2345                  * But we only get here if the client just explicitly
2346                  * destroyed the client.  Surely it no longer cares what
2347                  * error it gets back on an operation for the dead
2348                  * client.
2349                  */
2350                 goto out;
2351
2352         status = nfs_ok;
2353         nfsd4_client_record_create(cstate->session->se_client);
2354 out:
2355         nfs4_unlock_state();
2356         return status;
2357 }
2358
2359 __be32
2360 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2361                   struct nfsd4_setclientid *setclid)
2362 {
2363         struct xdr_netobj       clname = setclid->se_name;
2364         nfs4_verifier           clverifier = setclid->se_verf;
2365         struct nfs4_client      *conf, *unconf, *new;
2366         __be32                  status;
2367         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2368
2369         /* Cases below refer to rfc 3530 section 14.2.33: */
2370         nfs4_lock_state();
2371         conf = find_confirmed_client_by_name(&clname, nn);
2372         if (conf) {
2373                 /* case 0: */
2374                 status = nfserr_clid_inuse;
2375                 if (clp_used_exchangeid(conf))
2376                         goto out;
2377                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2378                         char addr_str[INET6_ADDRSTRLEN];
2379                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2380                                  sizeof(addr_str));
2381                         dprintk("NFSD: setclientid: string in use by client "
2382                                 "at %s\n", addr_str);
2383                         goto out;
2384                 }
2385         }
2386         unconf = find_unconfirmed_client_by_name(&clname, nn);
2387         if (unconf)
2388                 expire_client(unconf);
2389         status = nfserr_jukebox;
2390         new = create_client(clname, rqstp, &clverifier);
2391         if (new == NULL)
2392                 goto out;
2393         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2394                 /* case 1: probable callback update */
2395                 copy_clid(new, conf);
2396         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2397                 gen_clid(new, nn);
2398         new->cl_minorversion = 0;
2399         gen_callback(new, setclid, rqstp);
2400         add_to_unconfirmed(new);
2401         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2402         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2403         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2404         status = nfs_ok;
2405 out:
2406         nfs4_unlock_state();
2407         return status;
2408 }
2409
2410
2411 __be32
2412 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2413                          struct nfsd4_compound_state *cstate,
2414                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2415 {
2416         struct nfs4_client *conf, *unconf;
2417         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2418         clientid_t * clid = &setclientid_confirm->sc_clientid;
2419         __be32 status;
2420         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2421
2422         if (STALE_CLIENTID(clid, nn))
2423                 return nfserr_stale_clientid;
2424         nfs4_lock_state();
2425
2426         conf = find_confirmed_client(clid, false, nn);
2427         unconf = find_unconfirmed_client(clid, false, nn);
2428         /*
2429          * We try hard to give out unique clientid's, so if we get an
2430          * attempt to confirm the same clientid with a different cred,
2431          * there's a bug somewhere.  Let's charitably assume it's our
2432          * bug.
2433          */
2434         status = nfserr_serverfault;
2435         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2436                 goto out;
2437         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2438                 goto out;
2439         /* cases below refer to rfc 3530 section 14.2.34: */
2440         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2441                 if (conf && !unconf) /* case 2: probable retransmit */
2442                         status = nfs_ok;
2443                 else /* case 4: client hasn't noticed we rebooted yet? */
2444                         status = nfserr_stale_clientid;
2445                 goto out;
2446         }
2447         status = nfs_ok;
2448         if (conf) { /* case 1: callback update */
2449                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2450                 nfsd4_probe_callback(conf);
2451                 expire_client(unconf);
2452         } else { /* case 3: normal case; new or rebooted client */
2453                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2454                 if (conf) {
2455                         status = mark_client_expired(conf);
2456                         if (status)
2457                                 goto out;
2458                         expire_client(conf);
2459                 }
2460                 move_to_confirmed(unconf);
2461                 nfsd4_probe_callback(unconf);
2462         }
2463 out:
2464         nfs4_unlock_state();
2465         return status;
2466 }
2467
2468 static struct nfs4_file *nfsd4_alloc_file(void)
2469 {
2470         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2471 }
2472
2473 /* OPEN Share state helper functions */
2474 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2475 {
2476         unsigned int hashval = file_hashval(ino);
2477
2478         atomic_set(&fp->fi_ref, 1);
2479         INIT_LIST_HEAD(&fp->fi_stateids);
2480         INIT_LIST_HEAD(&fp->fi_delegations);
2481         fp->fi_inode = igrab(ino);
2482         fp->fi_had_conflict = false;
2483         fp->fi_lease = NULL;
2484         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2485         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2486         spin_lock(&recall_lock);
2487         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2488         spin_unlock(&recall_lock);
2489 }
2490
2491 static void
2492 nfsd4_free_slab(struct kmem_cache **slab)
2493 {
2494         if (*slab == NULL)
2495                 return;
2496         kmem_cache_destroy(*slab);
2497         *slab = NULL;
2498 }
2499
2500 void
2501 nfsd4_free_slabs(void)
2502 {
2503         nfsd4_free_slab(&openowner_slab);
2504         nfsd4_free_slab(&lockowner_slab);
2505         nfsd4_free_slab(&file_slab);
2506         nfsd4_free_slab(&stateid_slab);
2507         nfsd4_free_slab(&deleg_slab);
2508 }
2509
2510 int
2511 nfsd4_init_slabs(void)
2512 {
2513         openowner_slab = kmem_cache_create("nfsd4_openowners",
2514                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2515         if (openowner_slab == NULL)
2516                 goto out_nomem;
2517         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2518                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2519         if (lockowner_slab == NULL)
2520                 goto out_nomem;
2521         file_slab = kmem_cache_create("nfsd4_files",
2522                         sizeof(struct nfs4_file), 0, 0, NULL);
2523         if (file_slab == NULL)
2524                 goto out_nomem;
2525         stateid_slab = kmem_cache_create("nfsd4_stateids",
2526                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2527         if (stateid_slab == NULL)
2528                 goto out_nomem;
2529         deleg_slab = kmem_cache_create("nfsd4_delegations",
2530                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2531         if (deleg_slab == NULL)
2532                 goto out_nomem;
2533         return 0;
2534 out_nomem:
2535         nfsd4_free_slabs();
2536         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2537         return -ENOMEM;
2538 }
2539
2540 void nfs4_free_openowner(struct nfs4_openowner *oo)
2541 {
2542         kfree(oo->oo_owner.so_owner.data);
2543         kmem_cache_free(openowner_slab, oo);
2544 }
2545
2546 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2547 {
2548         kfree(lo->lo_owner.so_owner.data);
2549         kmem_cache_free(lockowner_slab, lo);
2550 }
2551
2552 static void init_nfs4_replay(struct nfs4_replay *rp)
2553 {
2554         rp->rp_status = nfserr_serverfault;
2555         rp->rp_buflen = 0;
2556         rp->rp_buf = rp->rp_ibuf;
2557 }
2558
2559 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2560 {
2561         struct nfs4_stateowner *sop;
2562
2563         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2564         if (!sop)
2565                 return NULL;
2566
2567         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2568         if (!sop->so_owner.data) {
2569                 kmem_cache_free(slab, sop);
2570                 return NULL;
2571         }
2572         sop->so_owner.len = owner->len;
2573
2574         INIT_LIST_HEAD(&sop->so_stateids);
2575         sop->so_client = clp;
2576         init_nfs4_replay(&sop->so_replay);
2577         return sop;
2578 }
2579
2580 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2581 {
2582         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2583
2584         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2585         list_add(&oo->oo_perclient, &clp->cl_openowners);
2586 }
2587
2588 static struct nfs4_openowner *
2589 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2590         struct nfs4_openowner *oo;
2591
2592         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2593         if (!oo)
2594                 return NULL;
2595         oo->oo_owner.so_is_open_owner = 1;
2596         oo->oo_owner.so_seqid = open->op_seqid;
2597         oo->oo_flags = NFS4_OO_NEW;
2598         oo->oo_time = 0;
2599         oo->oo_last_closed_stid = NULL;
2600         INIT_LIST_HEAD(&oo->oo_close_lru);
2601         hash_openowner(oo, clp, strhashval);
2602         return oo;
2603 }
2604
2605 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2606         struct nfs4_openowner *oo = open->op_openowner;
2607
2608         stp->st_stid.sc_type = NFS4_OPEN_STID;
2609         INIT_LIST_HEAD(&stp->st_lockowners);
2610         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2611         list_add(&stp->st_perfile, &fp->fi_stateids);
2612         stp->st_stateowner = &oo->oo_owner;
2613         get_nfs4_file(fp);
2614         stp->st_file = fp;
2615         stp->st_access_bmap = 0;
2616         stp->st_deny_bmap = 0;
2617         set_access(open->op_share_access, stp);
2618         set_deny(open->op_share_deny, stp);
2619         stp->st_openstp = NULL;
2620 }
2621
2622 static void
2623 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2624 {
2625         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2626
2627         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2628
2629         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2630         oo->oo_time = get_seconds();
2631 }
2632
2633 static int
2634 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2635                                                         clientid_t *clid)
2636 {
2637         return (sop->so_owner.len == owner->len) &&
2638                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2639                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2640 }
2641
2642 static struct nfs4_openowner *
2643 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2644                         bool sessions, struct nfsd_net *nn)
2645 {
2646         struct nfs4_stateowner *so;
2647         struct nfs4_openowner *oo;
2648         struct nfs4_client *clp;
2649
2650         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2651                 if (!so->so_is_open_owner)
2652                         continue;
2653                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2654                         oo = openowner(so);
2655                         clp = oo->oo_owner.so_client;
2656                         if ((bool)clp->cl_minorversion != sessions)
2657                                 return NULL;
2658                         renew_client(oo->oo_owner.so_client);
2659                         return oo;
2660                 }
2661         }
2662         return NULL;
2663 }
2664
2665 /* search file_hashtbl[] for file */
2666 static struct nfs4_file *
2667 find_file(struct inode *ino)
2668 {
2669         unsigned int hashval = file_hashval(ino);
2670         struct nfs4_file *fp;
2671
2672         spin_lock(&recall_lock);
2673         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2674                 if (fp->fi_inode == ino) {
2675                         get_nfs4_file(fp);
2676                         spin_unlock(&recall_lock);
2677                         return fp;
2678                 }
2679         }
2680         spin_unlock(&recall_lock);
2681         return NULL;
2682 }
2683
2684 /*
2685  * Called to check deny when READ with all zero stateid or
2686  * WRITE with all zero or all one stateid
2687  */
2688 static __be32
2689 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2690 {
2691         struct inode *ino = current_fh->fh_dentry->d_inode;
2692         struct nfs4_file *fp;
2693         struct nfs4_ol_stateid *stp;
2694         __be32 ret;
2695
2696         fp = find_file(ino);
2697         if (!fp)
2698                 return nfs_ok;
2699         ret = nfserr_locked;
2700         /* Search for conflicting share reservations */
2701         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2702                 if (test_deny(deny_type, stp) ||
2703                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2704                         goto out;
2705         }
2706         ret = nfs_ok;
2707 out:
2708         put_nfs4_file(fp);
2709         return ret;
2710 }
2711
2712 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2713 {
2714         struct nfs4_client *clp = dp->dl_stid.sc_client;
2715         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2716
2717         /* We're assuming the state code never drops its reference
2718          * without first removing the lease.  Since we're in this lease
2719          * callback (and since the lease code is serialized by the kernel
2720          * lock) we know the server hasn't removed the lease yet, we know
2721          * it's safe to take a reference: */
2722         atomic_inc(&dp->dl_count);
2723
2724         list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
2725
2726         /* Only place dl_time is set; protected by i_lock: */
2727         dp->dl_time = get_seconds();
2728
2729         nfsd4_cb_recall(dp);
2730 }
2731
2732 /* Called from break_lease() with i_lock held. */
2733 static void nfsd_break_deleg_cb(struct file_lock *fl)
2734 {
2735         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2736         struct nfs4_delegation *dp;
2737
2738         if (!fp) {
2739                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2740                 return;
2741         }
2742         if (fp->fi_had_conflict) {
2743                 WARN(1, "duplicate break on %p\n", fp);
2744                 return;
2745         }
2746         /*
2747          * We don't want the locks code to timeout the lease for us;
2748          * we'll remove it ourself if a delegation isn't returned
2749          * in time:
2750          */
2751         fl->fl_break_time = 0;
2752
2753         spin_lock(&recall_lock);
2754         fp->fi_had_conflict = true;
2755         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2756                 nfsd_break_one_deleg(dp);
2757         spin_unlock(&recall_lock);
2758 }
2759
2760 static
2761 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2762 {
2763         if (arg & F_UNLCK)
2764                 return lease_modify(onlist, arg);
2765         else
2766                 return -EAGAIN;
2767 }
2768
2769 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2770         .lm_break = nfsd_break_deleg_cb,
2771         .lm_change = nfsd_change_deleg_cb,
2772 };
2773
2774 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2775 {
2776         if (nfsd4_has_session(cstate))
2777                 return nfs_ok;
2778         if (seqid == so->so_seqid - 1)
2779                 return nfserr_replay_me;
2780         if (seqid == so->so_seqid)
2781                 return nfs_ok;
2782         return nfserr_bad_seqid;
2783 }
2784
2785 __be32
2786 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2787                     struct nfsd4_open *open, struct nfsd_net *nn)
2788 {
2789         clientid_t *clientid = &open->op_clientid;
2790         struct nfs4_client *clp = NULL;
2791         unsigned int strhashval;
2792         struct nfs4_openowner *oo = NULL;
2793         __be32 status;
2794
2795         if (STALE_CLIENTID(&open->op_clientid, nn))
2796                 return nfserr_stale_clientid;
2797         /*
2798          * In case we need it later, after we've already created the
2799          * file and don't want to risk a further failure:
2800          */
2801         open->op_file = nfsd4_alloc_file();
2802         if (open->op_file == NULL)
2803                 return nfserr_jukebox;
2804
2805         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2806         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2807         open->op_openowner = oo;
2808         if (!oo) {
2809                 clp = find_confirmed_client(clientid, cstate->minorversion,
2810                                             nn);
2811                 if (clp == NULL)
2812                         return nfserr_expired;
2813                 goto new_owner;
2814         }
2815         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2816                 /* Replace unconfirmed owners without checking for replay. */
2817                 clp = oo->oo_owner.so_client;
2818                 release_openowner(oo);
2819                 open->op_openowner = NULL;
2820                 goto new_owner;
2821         }
2822         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2823         if (status)
2824                 return status;
2825         clp = oo->oo_owner.so_client;
2826         goto alloc_stateid;
2827 new_owner:
2828         oo = alloc_init_open_stateowner(strhashval, clp, open);
2829         if (oo == NULL)
2830                 return nfserr_jukebox;
2831         open->op_openowner = oo;
2832 alloc_stateid:
2833         open->op_stp = nfs4_alloc_stateid(clp);
2834         if (!open->op_stp)
2835                 return nfserr_jukebox;
2836         return nfs_ok;
2837 }
2838
2839 static inline __be32
2840 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2841 {
2842         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2843                 return nfserr_openmode;
2844         else
2845                 return nfs_ok;
2846 }
2847
2848 static int share_access_to_flags(u32 share_access)
2849 {
2850         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2851 }
2852
2853 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2854 {
2855         struct nfs4_stid *ret;
2856
2857         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2858         if (!ret)
2859                 return NULL;
2860         return delegstateid(ret);
2861 }
2862
2863 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2864 {
2865         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2866                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2867 }
2868
2869 static __be32
2870 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
2871                 struct nfs4_delegation **dp)
2872 {
2873         int flags;
2874         __be32 status = nfserr_bad_stateid;
2875
2876         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2877         if (*dp == NULL)
2878                 goto out;
2879         flags = share_access_to_flags(open->op_share_access);
2880         status = nfs4_check_delegmode(*dp, flags);
2881         if (status)
2882                 *dp = NULL;
2883 out:
2884         if (!nfsd4_is_deleg_cur(open))
2885                 return nfs_ok;
2886         if (status)
2887                 return status;
2888         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2889         return nfs_ok;
2890 }
2891
2892 static __be32
2893 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2894 {
2895         struct nfs4_ol_stateid *local;
2896         struct nfs4_openowner *oo = open->op_openowner;
2897
2898         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2899                 /* ignore lock owners */
2900                 if (local->st_stateowner->so_is_open_owner == 0)
2901                         continue;
2902                 /* remember if we have seen this open owner */
2903                 if (local->st_stateowner == &oo->oo_owner)
2904                         *stpp = local;
2905                 /* check for conflicting share reservations */
2906                 if (!test_share(local, open))
2907                         return nfserr_share_denied;
2908         }
2909         return nfs_ok;
2910 }
2911
2912 static inline int nfs4_access_to_access(u32 nfs4_access)
2913 {
2914         int flags = 0;
2915
2916         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2917                 flags |= NFSD_MAY_READ;
2918         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2919                 flags |= NFSD_MAY_WRITE;
2920         return flags;
2921 }
2922
2923 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2924                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2925 {
2926         __be32 status;
2927         int oflag = nfs4_access_to_omode(open->op_share_access);
2928         int access = nfs4_access_to_access(open->op_share_access);
2929
2930         if (!fp->fi_fds[oflag]) {
2931                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2932                         &fp->fi_fds[oflag]);
2933                 if (status)
2934                         return status;
2935         }
2936         nfs4_file_get_access(fp, oflag);
2937
2938         return nfs_ok;
2939 }
2940
2941 static inline __be32
2942 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2943                 struct nfsd4_open *open)
2944 {
2945         struct iattr iattr = {
2946                 .ia_valid = ATTR_SIZE,
2947                 .ia_size = 0,
2948         };
2949         if (!open->op_truncate)
2950                 return 0;
2951         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2952                 return nfserr_inval;
2953         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2954 }
2955
2956 static __be32
2957 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)
2958 {
2959         u32 op_share_access = open->op_share_access;
2960         bool new_access;
2961         __be32 status;
2962
2963         new_access = !test_access(op_share_access, stp);
2964         if (new_access) {
2965                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2966                 if (status)
2967                         return status;
2968         }
2969         status = nfsd4_truncate(rqstp, cur_fh, open);
2970         if (status) {
2971                 if (new_access) {
2972                         int oflag = nfs4_access_to_omode(op_share_access);
2973                         nfs4_file_put_access(fp, oflag);
2974                 }
2975                 return status;
2976         }
2977         /* remember the open */
2978         set_access(op_share_access, stp);
2979         set_deny(open->op_share_deny, stp);
2980
2981         return nfs_ok;
2982 }
2983
2984
2985 static void
2986 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2987 {
2988         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2989 }
2990
2991 /* Should we give out recallable state?: */
2992 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2993 {
2994         if (clp->cl_cb_state == NFSD4_CB_UP)
2995                 return true;
2996         /*
2997          * In the sessions case, since we don't have to establish a
2998          * separate connection for callbacks, we assume it's OK
2999          * until we hear otherwise:
3000          */
3001         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3002 }
3003
3004 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
3005 {
3006         struct file_lock *fl;
3007
3008         fl = locks_alloc_lock();
3009         if (!fl)
3010                 return NULL;
3011         locks_init_lock(fl);
3012         fl->fl_lmops = &nfsd_lease_mng_ops;
3013         fl->fl_flags = FL_LEASE;
3014         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3015         fl->fl_end = OFFSET_MAX;
3016         fl->fl_owner = (fl_owner_t)(dp->dl_file);
3017         fl->fl_pid = current->tgid;
3018         return fl;
3019 }
3020
3021 static int nfs4_setlease(struct nfs4_delegation *dp)
3022 {
3023         struct nfs4_file *fp = dp->dl_file;
3024         struct file_lock *fl;
3025         int status;
3026
3027         fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
3028         if (!fl)
3029                 return -ENOMEM;
3030         fl->fl_file = find_readable_file(fp);
3031         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3032         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
3033         if (status) {
3034                 list_del_init(&dp->dl_perclnt);
3035                 locks_free_lock(fl);
3036                 return status;
3037         }
3038         fp->fi_lease = fl;
3039         fp->fi_deleg_file = get_file(fl->fl_file);
3040         atomic_set(&fp->fi_delegees, 1);
3041         list_add(&dp->dl_perfile, &fp->fi_delegations);
3042         return 0;
3043 }
3044
3045 static int nfs4_set_delegation(struct nfs4_delegation *dp, struct nfs4_file *fp)
3046 {
3047         int status;
3048
3049         if (fp->fi_had_conflict)
3050                 return -EAGAIN;
3051         get_nfs4_file(fp);
3052         dp->dl_file = fp;
3053         if (!fp->fi_lease) {
3054                 status = nfs4_setlease(dp);
3055                 if (status)
3056                         goto out_free;
3057                 return 0;
3058         }
3059         spin_lock(&recall_lock);
3060         if (fp->fi_had_conflict) {
3061                 spin_unlock(&recall_lock);
3062                 status = -EAGAIN;
3063                 goto out_free;
3064         }
3065         atomic_inc(&fp->fi_delegees);
3066         list_add(&dp->dl_perfile, &fp->fi_delegations);
3067         spin_unlock(&recall_lock);
3068         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
3069         return 0;
3070 out_free:
3071         put_nfs4_file(fp);
3072         dp->dl_file = fp;
3073         return status;
3074 }
3075
3076 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3077 {
3078         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3079         if (status == -EAGAIN)
3080                 open->op_why_no_deleg = WND4_CONTENTION;
3081         else {
3082                 open->op_why_no_deleg = WND4_RESOURCE;
3083                 switch (open->op_deleg_want) {
3084                 case NFS4_SHARE_WANT_READ_DELEG:
3085                 case NFS4_SHARE_WANT_WRITE_DELEG:
3086                 case NFS4_SHARE_WANT_ANY_DELEG:
3087                         break;
3088                 case NFS4_SHARE_WANT_CANCEL:
3089                         open->op_why_no_deleg = WND4_CANCELLED;
3090                         break;
3091                 case NFS4_SHARE_WANT_NO_DELEG:
3092                         WARN_ON_ONCE(1);
3093                 }
3094         }
3095 }
3096
3097 /*
3098  * Attempt to hand out a delegation.
3099  *
3100  * Note we don't support write delegations, and won't until the vfs has
3101  * proper support for them.
3102  */
3103 static void
3104 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
3105                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
3106 {
3107         struct nfs4_delegation *dp;
3108         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
3109         int cb_up;
3110         int status = 0;
3111
3112         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3113         open->op_recall = 0;
3114         switch (open->op_claim_type) {
3115                 case NFS4_OPEN_CLAIM_PREVIOUS:
3116                         if (!cb_up)
3117                                 open->op_recall = 1;
3118                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3119                                 goto out_no_deleg;
3120                         break;
3121                 case NFS4_OPEN_CLAIM_NULL:
3122                         /*
3123                          * Let's not give out any delegations till everyone's
3124                          * had the chance to reclaim theirs....
3125                          */
3126                         if (locks_in_grace(net))
3127                                 goto out_no_deleg;
3128                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3129                                 goto out_no_deleg;
3130                         /*
3131                          * Also, if the file was opened for write or
3132                          * create, there's a good chance the client's
3133                          * about to write to it, resulting in an
3134                          * immediate recall (since we don't support
3135                          * write delegations):
3136                          */
3137                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3138                                 goto out_no_deleg;
3139                         if (open->op_create == NFS4_OPEN_CREATE)
3140                                 goto out_no_deleg;
3141                         break;
3142                 default:
3143                         goto out_no_deleg;
3144         }
3145         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh);
3146         if (dp == NULL)
3147                 goto out_no_deleg;
3148         status = nfs4_set_delegation(dp, stp->st_file);
3149         if (status)
3150                 goto out_free;
3151
3152         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3153
3154         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3155                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3156         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3157         return;
3158 out_free:
3159         nfs4_put_delegation(dp);
3160 out_no_deleg:
3161         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3162         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3163             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3164                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3165                 open->op_recall = 1;
3166         }
3167
3168         /* 4.1 client asking for a delegation? */
3169         if (open->op_deleg_want)
3170                 nfsd4_open_deleg_none_ext(open, status);
3171         return;
3172 }
3173
3174 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3175                                         struct nfs4_delegation *dp)
3176 {
3177         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3178             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3179                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3180                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3181         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3182                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3183                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3184                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3185         }
3186         /* Otherwise the client must be confused wanting a delegation
3187          * it already has, therefore we don't return
3188          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3189          */
3190 }
3191
3192 /*
3193  * called with nfs4_lock_state() held.
3194  */
3195 __be32
3196 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3197 {
3198         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3199         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3200         struct nfs4_file *fp = NULL;
3201         struct inode *ino = current_fh->fh_dentry->d_inode;
3202         struct nfs4_ol_stateid *stp = NULL;
3203         struct nfs4_delegation *dp = NULL;
3204         __be32 status;
3205
3206         /*
3207          * Lookup file; if found, lookup stateid and check open request,
3208          * and check for delegations in the process of being recalled.
3209          * If not found, create the nfs4_file struct
3210          */
3211         fp = find_file(ino);
3212         if (fp) {
3213                 if ((status = nfs4_check_open(fp, open, &stp)))
3214                         goto out;
3215                 status = nfs4_check_deleg(cl, open, &dp);
3216                 if (status)
3217                         goto out;
3218         } else {
3219                 status = nfserr_bad_stateid;
3220                 if (nfsd4_is_deleg_cur(open))
3221                         goto out;
3222                 status = nfserr_jukebox;
3223                 fp = open->op_file;
3224                 open->op_file = NULL;
3225                 nfsd4_init_file(fp, ino);
3226         }
3227
3228         /*
3229          * OPEN the file, or upgrade an existing OPEN.
3230          * If truncate fails, the OPEN fails.
3231          */
3232         if (stp) {
3233                 /* Stateid was found, this is an OPEN upgrade */
3234                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3235                 if (status)
3236                         goto out;
3237         } else {
3238                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3239                 if (status)
3240                         goto out;
3241                 status = nfsd4_truncate(rqstp, current_fh, open);
3242                 if (status)
3243                         goto out;
3244                 stp = open->op_stp;
3245                 open->op_stp = NULL;
3246                 init_open_stateid(stp, fp, open);
3247         }
3248         update_stateid(&stp->st_stid.sc_stateid);
3249         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3250
3251         if (nfsd4_has_session(&resp->cstate)) {
3252                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3253
3254                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3255                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3256                         open->op_why_no_deleg = WND4_NOT_WANTED;
3257                         goto nodeleg;
3258                 }
3259         }
3260
3261         /*
3262         * Attempt to hand out a delegation. No error return, because the
3263         * OPEN succeeds even if we fail.
3264         */
3265         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3266 nodeleg:
3267         status = nfs_ok;
3268
3269         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3270                 STATEID_VAL(&stp->st_stid.sc_stateid));
3271 out:
3272         /* 4.1 client trying to upgrade/downgrade delegation? */
3273         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3274             open->op_deleg_want)
3275                 nfsd4_deleg_xgrade_none_ext(open, dp);
3276
3277         if (fp)
3278                 put_nfs4_file(fp);
3279         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3280                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3281         /*
3282         * To finish the open response, we just need to set the rflags.
3283         */
3284         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3285         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3286             !nfsd4_has_session(&resp->cstate))
3287                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3288
3289         return status;
3290 }
3291
3292 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3293 {
3294         if (open->op_openowner) {
3295                 struct nfs4_openowner *oo = open->op_openowner;
3296
3297                 if (!list_empty(&oo->oo_owner.so_stateids))
3298                         list_del_init(&oo->oo_close_lru);
3299                 if (oo->oo_flags & NFS4_OO_NEW) {
3300                         if (status) {
3301                                 release_openowner(oo);
3302                                 open->op_openowner = NULL;
3303                         } else
3304                                 oo->oo_flags &= ~NFS4_OO_NEW;
3305                 }
3306         }
3307         if (open->op_file)
3308                 nfsd4_free_file(open->op_file);
3309         if (open->op_stp)
3310                 free_generic_stateid(open->op_stp);
3311 }
3312
3313 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3314 {
3315         struct nfs4_client *found;
3316
3317         if (STALE_CLIENTID(clid, nn))
3318                 return nfserr_stale_clientid;
3319         found = find_confirmed_client(clid, session, nn);
3320         if (clp)
3321                 *clp = found;
3322         return found ? nfs_ok : nfserr_expired;
3323 }
3324
3325 __be32
3326 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3327             clientid_t *clid)
3328 {
3329         struct nfs4_client *clp;
3330         __be32 status;
3331         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3332
3333         nfs4_lock_state();
3334         dprintk("process_renew(%08x/%08x): starting\n",
3335                         clid->cl_boot, clid->cl_id);
3336         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3337         if (status)
3338                 goto out;
3339         status = nfserr_cb_path_down;
3340         if (!list_empty(&clp->cl_delegations)
3341                         && clp->cl_cb_state != NFSD4_CB_UP)
3342                 goto out;
3343         status = nfs_ok;
3344 out:
3345         nfs4_unlock_state();
3346         return status;
3347 }
3348
3349 static void
3350 nfsd4_end_grace(struct nfsd_net *nn)
3351 {
3352         /* do nothing if grace period already ended */
3353         if (nn->grace_ended)
3354                 return;
3355
3356         dprintk("NFSD: end of grace period\n");
3357         nn->grace_ended = true;
3358         nfsd4_record_grace_done(nn, nn->boot_time);
3359         locks_end_grace(&nn->nfsd4_manager);
3360         /*
3361          * Now that every NFSv4 client has had the chance to recover and
3362          * to see the (possibly new, possibly shorter) lease time, we
3363          * can safely set the next grace time to the current lease time:
3364          */
3365         nn->nfsd4_grace = nn->nfsd4_lease;
3366 }
3367
3368 static time_t
3369 nfs4_laundromat(struct nfsd_net *nn)
3370 {
3371         struct nfs4_client *clp;
3372         struct nfs4_openowner *oo;
3373         struct nfs4_delegation *dp;
3374         struct list_head *pos, *next, reaplist;
3375         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3376         time_t t, clientid_val = nn->nfsd4_lease;
3377         time_t u, test_val = nn->nfsd4_lease;
3378
3379         nfs4_lock_state();
3380
3381         dprintk("NFSD: laundromat service - starting\n");
3382         nfsd4_end_grace(nn);
3383         INIT_LIST_HEAD(&reaplist);
3384         spin_lock(&nn->client_lock);
3385         list_for_each_safe(pos, next, &nn->client_lru) {
3386                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3387                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3388                         t = clp->cl_time - cutoff;
3389                         if (clientid_val > t)
3390                                 clientid_val = t;
3391                         break;
3392                 }
3393                 if (mark_client_expired_locked(clp)) {
3394                         dprintk("NFSD: client in use (clientid %08x)\n",
3395                                 clp->cl_clientid.cl_id);
3396                         continue;
3397                 }
3398                 list_move(&clp->cl_lru, &reaplist);
3399         }
3400         spin_unlock(&nn->client_lock);
3401         list_for_each_safe(pos, next, &reaplist) {
3402                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3403                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3404                         clp->cl_clientid.cl_id);
3405                 expire_client(clp);
3406         }
3407         spin_lock(&recall_lock);
3408         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3409                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3410                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3411                         continue;
3412                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3413                         u = dp->dl_time - cutoff;
3414                         if (test_val > u)
3415                                 test_val = u;
3416                         break;
3417                 }
3418                 list_move(&dp->dl_recall_lru, &reaplist);
3419         }
3420         spin_unlock(&recall_lock);
3421         list_for_each_safe(pos, next, &reaplist) {
3422                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3423                 revoke_delegation(dp);
3424         }
3425         test_val = nn->nfsd4_lease;
3426         list_for_each_safe(pos, next, &nn->close_lru) {
3427                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3428                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3429                         u = oo->oo_time - cutoff;
3430                         if (test_val > u)
3431                                 test_val = u;
3432                         break;
3433                 }
3434                 release_openowner(oo);
3435         }
3436         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3437                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3438         nfs4_unlock_state();
3439         return clientid_val;
3440 }
3441
3442 static struct workqueue_struct *laundry_wq;
3443 static void laundromat_main(struct work_struct *);
3444
3445 static void
3446 laundromat_main(struct work_struct *laundry)
3447 {
3448         time_t t;
3449         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3450                                                   work);
3451         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3452                                            laundromat_work);
3453
3454         t = nfs4_laundromat(nn);
3455         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3456         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3457 }
3458
3459 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3460 {
3461         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3462                 return nfserr_bad_stateid;
3463         return nfs_ok;
3464 }
3465
3466 static inline int
3467 access_permit_read(struct nfs4_ol_stateid *stp)
3468 {
3469         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3470                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3471                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3472 }
3473
3474 static inline int
3475 access_permit_write(struct nfs4_ol_stateid *stp)
3476 {
3477         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3478                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3479 }
3480
3481 static
3482 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3483 {
3484         __be32 status = nfserr_openmode;
3485
3486         /* For lock stateid's, we test the parent open, not the lock: */
3487         if (stp->st_openstp)
3488                 stp = stp->st_openstp;
3489         if ((flags & WR_STATE) && !access_permit_write(stp))
3490                 goto out;
3491         if ((flags & RD_STATE) && !access_permit_read(stp))
3492                 goto out;
3493         status = nfs_ok;
3494 out:
3495         return status;
3496 }
3497
3498 static inline __be32
3499 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3500 {
3501         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3502                 return nfs_ok;
3503         else if (locks_in_grace(net)) {
3504                 /* Answer in remaining cases depends on existence of
3505                  * conflicting state; so we must wait out the grace period. */
3506                 return nfserr_grace;
3507         } else if (flags & WR_STATE)
3508                 return nfs4_share_conflict(current_fh,
3509                                 NFS4_SHARE_DENY_WRITE);
3510         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3511                 return nfs4_share_conflict(current_fh,
3512                                 NFS4_SHARE_DENY_READ);
3513 }
3514
3515 /*
3516  * Allow READ/WRITE during grace period on recovered state only for files
3517  * that are not able to provide mandatory locking.
3518  */
3519 static inline int
3520 grace_disallows_io(struct net *net, struct inode *inode)
3521 {
3522         return locks_in_grace(net) && mandatory_lock(inode);
3523 }
3524
3525 /* Returns true iff a is later than b: */
3526 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3527 {
3528         return (s32)(a->si_generation - b->si_generation) > 0;
3529 }
3530
3531 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3532 {
3533         /*
3534          * When sessions are used the stateid generation number is ignored
3535          * when it is zero.
3536          */
3537         if (has_session && in->si_generation == 0)
3538                 return nfs_ok;
3539
3540         if (in->si_generation == ref->si_generation)
3541                 return nfs_ok;
3542
3543         /* If the client sends us a stateid from the future, it's buggy: */
3544         if (stateid_generation_after(in, ref))
3545                 return nfserr_bad_stateid;
3546         /*
3547          * However, we could see a stateid from the past, even from a
3548          * non-buggy client.  For example, if the client sends a lock
3549          * while some IO is outstanding, the lock may bump si_generation
3550          * while the IO is still in flight.  The client could avoid that
3551          * situation by waiting for responses on all the IO requests,
3552          * but better performance may result in retrying IO that
3553          * receives an old_stateid error if requests are rarely
3554          * reordered in flight:
3555          */
3556         return nfserr_old_stateid;
3557 }
3558
3559 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3560 {
3561         struct nfs4_stid *s;
3562         struct nfs4_ol_stateid *ols;
3563         __be32 status;
3564
3565         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3566                 return nfserr_bad_stateid;
3567         /* Client debugging aid. */
3568         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3569                 char addr_str[INET6_ADDRSTRLEN];
3570                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3571                                  sizeof(addr_str));
3572                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3573                                         "with incorrect client ID\n", addr_str);
3574                 return nfserr_bad_stateid;
3575         }
3576         s = find_stateid(cl, stateid);
3577         if (!s)
3578                 return nfserr_bad_stateid;
3579         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3580         if (status)
3581                 return status;
3582         switch (s->sc_type) {
3583         case NFS4_DELEG_STID:
3584                 return nfs_ok;
3585         case NFS4_REVOKED_DELEG_STID:
3586                 return nfserr_deleg_revoked;
3587         case NFS4_OPEN_STID:
3588         case NFS4_LOCK_STID:
3589                 ols = openlockstateid(s);
3590                 if (ols->st_stateowner->so_is_open_owner
3591                                 && !(openowner(ols->st_stateowner)->oo_flags
3592                                                 & NFS4_OO_CONFIRMED))
3593                         return nfserr_bad_stateid;
3594                 return nfs_ok;
3595         default:
3596                 printk("unknown stateid type %x\n", s->sc_type);
3597         case NFS4_CLOSED_STID:
3598                 return nfserr_bad_stateid;
3599         }
3600 }
3601
3602 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3603                                    struct nfs4_stid **s, bool sessions,
3604                                    struct nfsd_net *nn)
3605 {
3606         struct nfs4_client *cl;
3607         __be32 status;
3608
3609         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3610                 return nfserr_bad_stateid;
3611         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3612                                                         nn, &cl);
3613         if (status == nfserr_stale_clientid)
3614                 return nfserr_stale_stateid;
3615         if (status)
3616                 return status;
3617         *s = find_stateid_by_type(cl, stateid, typemask);
3618         if (!*s)
3619                 return nfserr_bad_stateid;
3620         return nfs_ok;
3621 }
3622
3623 /*
3624 * Checks for stateid operations
3625 */
3626 __be32
3627 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3628                            stateid_t *stateid, int flags, struct file **filpp)
3629 {
3630         struct nfs4_stid *s;
3631         struct nfs4_ol_stateid *stp = NULL;
3632         struct nfs4_delegation *dp = NULL;
3633         struct svc_fh *current_fh = &cstate->current_fh;
3634         struct inode *ino = current_fh->fh_dentry->d_inode;
3635         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3636         __be32 status;
3637
3638         if (filpp)
3639                 *filpp = NULL;
3640
3641         if (grace_disallows_io(net, ino))
3642                 return nfserr_grace;
3643
3644         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3645                 return check_special_stateids(net, current_fh, stateid, flags);
3646
3647         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3648                                       &s, cstate->minorversion, nn);
3649         if (status)
3650                 return status;
3651         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3652         if (status)
3653                 goto out;
3654         switch (s->sc_type) {
3655         case NFS4_DELEG_STID:
3656                 dp = delegstateid(s);
3657                 status = nfs4_check_delegmode(dp, flags);
3658                 if (status)
3659                         goto out;
3660                 if (filpp) {
3661                         *filpp = dp->dl_file->fi_deleg_file;
3662                         if (!*filpp) {
3663                                 WARN_ON_ONCE(1);
3664                                 status = nfserr_serverfault;
3665                                 goto out;
3666                         }
3667                 }
3668                 break;
3669         case NFS4_OPEN_STID:
3670         case NFS4_LOCK_STID:
3671                 stp = openlockstateid(s);
3672                 status = nfs4_check_fh(current_fh, stp);
3673                 if (status)
3674                         goto out;
3675                 if (stp->st_stateowner->so_is_open_owner
3676                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3677                         goto out;
3678                 status = nfs4_check_openmode(stp, flags);
3679                 if (status)
3680                         goto out;
3681                 if (filpp) {
3682                         if (flags & RD_STATE)
3683                                 *filpp = find_readable_file(stp->st_file);
3684                         else
3685                                 *filpp = find_writeable_file(stp->st_file);
3686                 }
3687                 break;
3688         default:
3689                 return nfserr_bad_stateid;
3690         }
3691         status = nfs_ok;
3692 out:
3693         return status;
3694 }
3695
3696 static __be32
3697 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3698 {
3699         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3700                 return nfserr_locks_held;
3701         release_lock_stateid(stp);
3702         return nfs_ok;
3703 }
3704
3705 /*
3706  * Test if the stateid is valid
3707  */
3708 __be32
3709 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3710                    struct nfsd4_test_stateid *test_stateid)
3711 {
3712         struct nfsd4_test_stateid_id *stateid;
3713         struct nfs4_client *cl = cstate->session->se_client;
3714
3715         nfs4_lock_state();
3716         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3717                 stateid->ts_id_status =
3718                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3719         nfs4_unlock_state();
3720
3721         return nfs_ok;
3722 }
3723
3724 __be32
3725 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3726                    struct nfsd4_free_stateid *free_stateid)
3727 {
3728         stateid_t *stateid = &free_stateid->fr_stateid;
3729         struct nfs4_stid *s;
3730         struct nfs4_delegation *dp;
3731         struct nfs4_client *cl = cstate->session->se_client;
3732         __be32 ret = nfserr_bad_stateid;
3733
3734         nfs4_lock_state();
3735         s = find_stateid(cl, stateid);
3736         if (!s)
3737                 goto out;
3738         switch (s->sc_type) {
3739         case NFS4_DELEG_STID:
3740                 ret = nfserr_locks_held;
3741                 goto out;
3742         case NFS4_OPEN_STID:
3743         case NFS4_LOCK_STID:
3744                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3745                 if (ret)
3746                         goto out;
3747                 if (s->sc_type == NFS4_LOCK_STID)
3748                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3749                 else
3750                         ret = nfserr_locks_held;
3751                 break;
3752         case NFS4_REVOKED_DELEG_STID:
3753                 dp = delegstateid(s);
3754                 destroy_revoked_delegation(dp);
3755                 ret = nfs_ok;
3756                 break;
3757         default:
3758                 ret = nfserr_bad_stateid;
3759         }
3760 out:
3761         nfs4_unlock_state();
3762         return ret;
3763 }
3764
3765 static inline int
3766 setlkflg (int type)
3767 {
3768         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3769                 RD_STATE : WR_STATE;
3770 }
3771
3772 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3773 {
3774         struct svc_fh *current_fh = &cstate->current_fh;
3775         struct nfs4_stateowner *sop = stp->st_stateowner;
3776         __be32 status;
3777
3778         status = nfsd4_check_seqid(cstate, sop, seqid);
3779         if (status)
3780                 return status;
3781         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
3782                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
3783                 /*
3784                  * "Closed" stateid's exist *only* to return
3785                  * nfserr_replay_me from the previous step, and
3786                  * revoked delegations are kept only for free_stateid.
3787                  */
3788                 return nfserr_bad_stateid;
3789         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3790         if (status)
3791                 return status;
3792         return nfs4_check_fh(current_fh, stp);
3793 }
3794
3795 /*
3796  * Checks for sequence id mutating operations.
3797  */
3798 static __be32
3799 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3800                          stateid_t *stateid, char typemask,
3801                          struct nfs4_ol_stateid **stpp,
3802                          struct nfsd_net *nn)
3803 {
3804         __be32 status;
3805         struct nfs4_stid *s;
3806
3807         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3808                 seqid, STATEID_VAL(stateid));
3809
3810         *stpp = NULL;
3811         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3812                                       cstate->minorversion, nn);
3813         if (status)
3814                 return status;
3815         *stpp = openlockstateid(s);
3816         if (!nfsd4_has_session(cstate))
3817                 cstate->replay_owner = (*stpp)->st_stateowner;
3818
3819         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3820 }
3821
3822 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3823                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3824 {
3825         __be32 status;
3826         struct nfs4_openowner *oo;
3827
3828         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3829                                                 NFS4_OPEN_STID, stpp, nn);
3830         if (status)
3831                 return status;
3832         oo = openowner((*stpp)->st_stateowner);
3833         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3834                 return nfserr_bad_stateid;
3835         return nfs_ok;
3836 }
3837
3838 __be32
3839 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3840                    struct nfsd4_open_confirm *oc)
3841 {
3842         __be32 status;
3843         struct nfs4_openowner *oo;
3844         struct nfs4_ol_stateid *stp;
3845         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3846
3847         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
3848                         cstate->current_fh.fh_dentry);
3849
3850         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3851         if (status)
3852                 return status;
3853
3854         nfs4_lock_state();
3855
3856         status = nfs4_preprocess_seqid_op(cstate,
3857                                         oc->oc_seqid, &oc->oc_req_stateid,
3858                                         NFS4_OPEN_STID, &stp, nn);
3859         if (status)
3860                 goto out;
3861         oo = openowner(stp->st_stateowner);
3862         status = nfserr_bad_stateid;
3863         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3864                 goto out;
3865         oo->oo_flags |= NFS4_OO_CONFIRMED;
3866         update_stateid(&stp->st_stid.sc_stateid);
3867         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3868         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3869                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3870
3871         nfsd4_client_record_create(oo->oo_owner.so_client);
3872         status = nfs_ok;
3873 out:
3874         nfsd4_bump_seqid(cstate, status);
3875         if (!cstate->replay_owner)
3876                 nfs4_unlock_state();
3877         return status;
3878 }
3879
3880 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3881 {
3882         if (!test_access(access, stp))
3883                 return;
3884         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3885         clear_access(access, stp);
3886 }
3887
3888 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3889 {
3890         switch (to_access) {
3891         case NFS4_SHARE_ACCESS_READ:
3892                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3893                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3894                 break;
3895         case NFS4_SHARE_ACCESS_WRITE:
3896                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3897                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3898                 break;
3899         case NFS4_SHARE_ACCESS_BOTH:
3900                 break;
3901         default:
3902                 WARN_ON_ONCE(1);
3903         }
3904 }
3905
3906 static void
3907 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3908 {
3909         int i;
3910         for (i = 0; i < 4; i++) {
3911                 if ((i & deny) != i)
3912                         clear_deny(i, stp);
3913         }
3914 }
3915
3916 __be32
3917 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3918                      struct nfsd4_compound_state *cstate,
3919                      struct nfsd4_open_downgrade *od)
3920 {
3921         __be32 status;
3922         struct nfs4_ol_stateid *stp;
3923         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3924
3925         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
3926                         cstate->current_fh.fh_dentry);
3927
3928         /* We don't yet support WANT bits: */
3929         if (od->od_deleg_want)
3930                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3931                         od->od_deleg_want);
3932
3933         nfs4_lock_state();
3934         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3935                                         &od->od_stateid, &stp, nn);
3936         if (status)
3937                 goto out;
3938         status = nfserr_inval;
3939         if (!test_access(od->od_share_access, stp)) {
3940                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3941                         stp->st_access_bmap, od->od_share_access);
3942                 goto out;
3943         }
3944         if (!test_deny(od->od_share_deny, stp)) {
3945                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3946                         stp->st_deny_bmap, od->od_share_deny);
3947                 goto out;
3948         }
3949         nfs4_stateid_downgrade(stp, od->od_share_access);
3950
3951         reset_union_bmap_deny(od->od_share_deny, stp);
3952
3953         update_stateid(&stp->st_stid.sc_stateid);
3954         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3955         status = nfs_ok;
3956 out:
3957         nfsd4_bump_seqid(cstate, status);
3958         if (!cstate->replay_owner)
3959                 nfs4_unlock_state();
3960         return status;
3961 }
3962
3963 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3964 {
3965         unhash_open_stateid(s);
3966         s->st_stid.sc_type = NFS4_CLOSED_STID;
3967 }
3968
3969 /*
3970  * nfs4_unlock_state() called after encode
3971  */
3972 __be32
3973 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3974             struct nfsd4_close *close)
3975 {
3976         __be32 status;
3977         struct nfs4_openowner *oo;
3978         struct nfs4_ol_stateid *stp;
3979         struct net *net = SVC_NET(rqstp);
3980         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3981
3982         dprintk("NFSD: nfsd4_close on file %pd\n",
3983                         cstate->current_fh.fh_dentry);
3984
3985         nfs4_lock_state();
3986         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3987                                         &close->cl_stateid,
3988                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3989                                         &stp, nn);
3990         nfsd4_bump_seqid(cstate, status);
3991         if (status)
3992                 goto out;
3993         oo = openowner(stp->st_stateowner);
3994         update_stateid(&stp->st_stid.sc_stateid);
3995         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3996
3997         nfsd4_close_open_stateid(stp);
3998
3999         if (cstate->minorversion)
4000                 free_generic_stateid(stp);
4001         else
4002                 oo->oo_last_closed_stid = stp;
4003
4004         if (list_empty(&oo->oo_owner.so_stateids)) {
4005                 if (cstate->minorversion)
4006                         release_openowner(oo);
4007                 else {
4008                         /*
4009                          * In the 4.0 case we need to keep the owners around a
4010                          * little while to handle CLOSE replay.
4011                          */
4012                         move_to_close_lru(oo, SVC_NET(rqstp));
4013                 }
4014         }
4015 out:
4016         if (!cstate->replay_owner)
4017                 nfs4_unlock_state();
4018         return status;
4019 }
4020
4021 __be32
4022 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4023                   struct nfsd4_delegreturn *dr)
4024 {
4025         struct nfs4_delegation *dp;
4026         stateid_t *stateid = &dr->dr_stateid;
4027         struct nfs4_stid *s;
4028         __be32 status;
4029         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4030
4031         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4032                 return status;
4033
4034         nfs4_lock_state();
4035         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
4036                                       cstate->minorversion, nn);
4037         if (status)
4038                 goto out;
4039         dp = delegstateid(s);
4040         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4041         if (status)
4042                 goto out;
4043
4044         destroy_delegation(dp);
4045 out:
4046         nfs4_unlock_state();
4047
4048         return status;
4049 }
4050
4051
4052 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
4053
4054 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
4055
4056 static inline u64
4057 end_offset(u64 start, u64 len)
4058 {
4059         u64 end;
4060
4061         end = start + len;
4062         return end >= start ? end: NFS4_MAX_UINT64;
4063 }
4064
4065 /* last octet in a range */
4066 static inline u64
4067 last_byte_offset(u64 start, u64 len)
4068 {
4069         u64 end;
4070
4071         WARN_ON_ONCE(!len);
4072         end = start + len;
4073         return end > start ? end - 1: NFS4_MAX_UINT64;
4074 }
4075
4076 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
4077 {
4078         return (file_hashval(inode) + cl_id
4079                         + opaque_hashval(ownername->data, ownername->len))
4080                 & LOCKOWNER_INO_HASH_MASK;
4081 }
4082
4083 /*
4084  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4085  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4086  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
4087  * locking, this prevents us from being completely protocol-compliant.  The
4088  * real solution to this problem is to start using unsigned file offsets in
4089  * the VFS, but this is a very deep change!
4090  */
4091 static inline void
4092 nfs4_transform_lock_offset(struct file_lock *lock)
4093 {
4094         if (lock->fl_start < 0)
4095                 lock->fl_start = OFFSET_MAX;
4096         if (lock->fl_end < 0)
4097                 lock->fl_end = OFFSET_MAX;
4098 }
4099
4100 /* Hack!: For now, we're defining this just so we can use a pointer to it
4101  * as a unique cookie to identify our (NFSv4's) posix locks. */
4102 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4103 };
4104
4105 static inline void
4106 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4107 {
4108         struct nfs4_lockowner *lo;
4109
4110         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4111                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4112                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4113                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4114                 if (!deny->ld_owner.data)
4115                         /* We just don't care that much */
4116                         goto nevermind;
4117                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4118                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4119         } else {
4120 nevermind:
4121                 deny->ld_owner.len = 0;
4122                 deny->ld_owner.data = NULL;
4123                 deny->ld_clientid.cl_boot = 0;
4124                 deny->ld_clientid.cl_id = 0;
4125         }
4126         deny->ld_start = fl->fl_start;
4127         deny->ld_length = NFS4_MAX_UINT64;
4128         if (fl->fl_end != NFS4_MAX_UINT64)
4129                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4130         deny->ld_type = NFS4_READ_LT;
4131         if (fl->fl_type != F_RDLCK)
4132                 deny->ld_type = NFS4_WRITE_LT;
4133 }
4134
4135 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
4136 {
4137         struct nfs4_ol_stateid *lst;
4138
4139         if (!same_owner_str(&lo->lo_owner, owner, clid))
4140                 return false;
4141         lst = list_first_entry(&lo->lo_owner.so_stateids,
4142                                struct nfs4_ol_stateid, st_perstateowner);
4143         return lst->st_file->fi_inode == inode;
4144 }
4145
4146 static struct nfs4_lockowner *
4147 find_lockowner_str(struct inode *inode, clientid_t *clid,
4148                    struct xdr_netobj *owner, struct nfsd_net *nn)
4149 {
4150         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4151         struct nfs4_lockowner *lo;
4152
4153         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4154                 if (same_lockowner_ino(lo, inode, clid, owner))
4155                         return lo;
4156         }
4157         return NULL;
4158 }
4159
4160 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4161 {
4162         struct inode *inode = open_stp->st_file->fi_inode;
4163         unsigned int inohash = lockowner_ino_hashval(inode,
4164                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4165         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4166
4167         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4168         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4169         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4170 }
4171
4172 /*
4173  * Alloc a lock owner structure.
4174  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4175  * occurred.
4176  *
4177  * strhashval = ownerstr_hashval
4178  */
4179
4180 static struct nfs4_lockowner *
4181 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4182         struct nfs4_lockowner *lo;
4183
4184         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4185         if (!lo)
4186                 return NULL;
4187         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4188         lo->lo_owner.so_is_open_owner = 0;
4189         /* It is the openowner seqid that will be incremented in encode in the
4190          * case of new lockowners; so increment the lock seqid manually: */
4191         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4192         hash_lockowner(lo, strhashval, clp, open_stp);
4193         return lo;
4194 }
4195
4196 static struct nfs4_ol_stateid *
4197 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4198 {
4199         struct nfs4_ol_stateid *stp;
4200         struct nfs4_client *clp = lo->lo_owner.so_client;
4201
4202         stp = nfs4_alloc_stateid(clp);
4203         if (stp == NULL)
4204                 return NULL;
4205         stp->st_stid.sc_type = NFS4_LOCK_STID;
4206         list_add(&stp->st_perfile, &fp->fi_stateids);
4207         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4208         stp->st_stateowner = &lo->lo_owner;
4209         get_nfs4_file(fp);
4210         stp->st_file = fp;
4211         stp->st_access_bmap = 0;
4212         stp->st_deny_bmap = open_stp->st_deny_bmap;
4213         stp->st_openstp = open_stp;
4214         return stp;
4215 }
4216
4217 static int
4218 check_lock_length(u64 offset, u64 length)
4219 {
4220         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4221              LOFF_OVERFLOW(offset, length)));
4222 }
4223
4224 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4225 {
4226         struct nfs4_file *fp = lock_stp->st_file;
4227         int oflag = nfs4_access_to_omode(access);
4228
4229         if (test_access(access, lock_stp))
4230                 return;
4231         nfs4_file_get_access(fp, oflag);
4232         set_access(access, lock_stp);
4233 }
4234
4235 static __be32 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate, struct nfs4_ol_stateid *ost, struct nfsd4_lock *lock, struct nfs4_ol_stateid **lst, bool *new)
4236 {
4237         struct nfs4_file *fi = ost->st_file;
4238         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4239         struct nfs4_client *cl = oo->oo_owner.so_client;
4240         struct nfs4_lockowner *lo;
4241         unsigned int strhashval;
4242         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4243
4244         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4245                                 &lock->v.new.owner, nn);
4246         if (lo) {
4247                 if (!cstate->minorversion)
4248                         return nfserr_bad_seqid;
4249                 /* XXX: a lockowner always has exactly one stateid: */
4250                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4251                                 struct nfs4_ol_stateid, st_perstateowner);
4252                 return nfs_ok;
4253         }
4254         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4255                         &lock->v.new.owner);
4256         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4257         if (lo == NULL)
4258                 return nfserr_jukebox;
4259         *lst = alloc_init_lock_stateid(lo, fi, ost);
4260         if (*lst == NULL) {
4261                 release_lockowner(lo);
4262                 return nfserr_jukebox;
4263         }
4264         *new = true;
4265         return nfs_ok;
4266 }
4267
4268 /*
4269  *  LOCK operation
4270  */
4271 __be32
4272 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4273            struct nfsd4_lock *lock)
4274 {
4275         struct nfs4_openowner *open_sop = NULL;
4276         struct nfs4_lockowner *lock_sop = NULL;
4277         struct nfs4_ol_stateid *lock_stp;
4278         struct file *filp = NULL;
4279         struct file_lock *file_lock = NULL;
4280         struct file_lock *conflock = NULL;
4281         __be32 status = 0;
4282         bool new_state = false;
4283         int lkflg;
4284         int err;
4285         struct net *net = SVC_NET(rqstp);
4286         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4287
4288         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4289                 (long long) lock->lk_offset,
4290                 (long long) lock->lk_length);
4291
4292         if (check_lock_length(lock->lk_offset, lock->lk_length))
4293                  return nfserr_inval;
4294
4295         if ((status = fh_verify(rqstp, &cstate->current_fh,
4296                                 S_IFREG, NFSD_MAY_LOCK))) {
4297                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4298                 return status;
4299         }
4300
4301         nfs4_lock_state();
4302
4303         if (lock->lk_is_new) {
4304                 struct nfs4_ol_stateid *open_stp = NULL;
4305
4306                 if (nfsd4_has_session(cstate))
4307                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4308                         memcpy(&lock->v.new.clientid,
4309                                 &cstate->session->se_client->cl_clientid,
4310                                 sizeof(clientid_t));
4311
4312                 status = nfserr_stale_clientid;
4313                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4314                         goto out;
4315
4316                 /* validate and update open stateid and open seqid */
4317                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4318                                         lock->lk_new_open_seqid,
4319                                         &lock->lk_new_open_stateid,
4320                                         &open_stp, nn);
4321                 if (status)
4322                         goto out;
4323                 open_sop = openowner(open_stp->st_stateowner);
4324                 status = nfserr_bad_stateid;
4325                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4326                                                 &lock->v.new.clientid))
4327                         goto out;
4328                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4329                                                         &lock_stp, &new_state);
4330         } else
4331                 status = nfs4_preprocess_seqid_op(cstate,
4332                                        lock->lk_old_lock_seqid,
4333                                        &lock->lk_old_lock_stateid,
4334                                        NFS4_LOCK_STID, &lock_stp, nn);
4335         if (status)
4336                 goto out;
4337         lock_sop = lockowner(lock_stp->st_stateowner);
4338
4339         lkflg = setlkflg(lock->lk_type);
4340         status = nfs4_check_openmode(lock_stp, lkflg);
4341         if (status)
4342                 goto out;
4343
4344         status = nfserr_grace;
4345         if (locks_in_grace(net) && !lock->lk_reclaim)
4346                 goto out;
4347         status = nfserr_no_grace;
4348         if (!locks_in_grace(net) && lock->lk_reclaim)
4349                 goto out;
4350
4351         file_lock = locks_alloc_lock();
4352         if (!file_lock) {
4353                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4354                 status = nfserr_jukebox;
4355                 goto out;
4356         }
4357
4358         locks_init_lock(file_lock);
4359         switch (lock->lk_type) {
4360                 case NFS4_READ_LT:
4361                 case NFS4_READW_LT:
4362                         filp = find_readable_file(lock_stp->st_file);
4363                         if (filp)
4364                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4365                         file_lock->fl_type = F_RDLCK;
4366                         break;
4367                 case NFS4_WRITE_LT:
4368                 case NFS4_WRITEW_LT:
4369                         filp = find_writeable_file(lock_stp->st_file);
4370                         if (filp)
4371                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4372                         file_lock->fl_type = F_WRLCK;
4373                         break;
4374                 default:
4375                         status = nfserr_inval;
4376                 goto out;
4377         }
4378         if (!filp) {
4379                 status = nfserr_openmode;
4380                 goto out;
4381         }
4382         file_lock->fl_owner = (fl_owner_t)lock_sop;
4383         file_lock->fl_pid = current->tgid;
4384         file_lock->fl_file = filp;
4385         file_lock->fl_flags = FL_POSIX;
4386         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4387         file_lock->fl_start = lock->lk_offset;
4388         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4389         nfs4_transform_lock_offset(file_lock);
4390
4391         conflock = locks_alloc_lock();
4392         if (!conflock) {
4393                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4394                 status = nfserr_jukebox;
4395                 goto out;
4396         }
4397
4398         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4399         switch (-err) {
4400         case 0: /* success! */
4401                 update_stateid(&lock_stp->st_stid.sc_stateid);
4402                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4403                                 sizeof(stateid_t));
4404                 status = 0;
4405                 break;
4406         case (EAGAIN):          /* conflock holds conflicting lock */
4407                 status = nfserr_denied;
4408                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4409                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4410                 break;
4411         case (EDEADLK):
4412                 status = nfserr_deadlock;
4413                 break;
4414         default:
4415                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4416                 status = nfserrno(err);
4417                 break;
4418         }
4419 out:
4420         if (status && new_state)
4421                 release_lockowner(lock_sop);
4422         nfsd4_bump_seqid(cstate, status);
4423         if (!cstate->replay_owner)
4424                 nfs4_unlock_state();
4425         if (file_lock)
4426                 locks_free_lock(file_lock);
4427         if (conflock)
4428                 locks_free_lock(conflock);
4429         return status;
4430 }
4431
4432 /*
4433  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4434  * so we do a temporary open here just to get an open file to pass to
4435  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4436  * inode operation.)
4437  */
4438 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4439 {
4440         struct file *file;
4441         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4442         if (!err) {
4443                 err = nfserrno(vfs_test_lock(file, lock));
4444                 nfsd_close(file);
4445         }
4446         return err;
4447 }
4448
4449 /*
4450  * LOCKT operation
4451  */
4452 __be32
4453 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4454             struct nfsd4_lockt *lockt)
4455 {
4456         struct inode *inode;
4457         struct file_lock *file_lock = NULL;
4458         struct nfs4_lockowner *lo;
4459         __be32 status;
4460         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4461
4462         if (locks_in_grace(SVC_NET(rqstp)))
4463                 return nfserr_grace;
4464
4465         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4466                  return nfserr_inval;
4467
4468         nfs4_lock_state();
4469
4470         if (!nfsd4_has_session(cstate)) {
4471                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4472                 if (status)
4473                         goto out;
4474         }
4475
4476         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4477                 goto out;
4478
4479         inode = cstate->current_fh.fh_dentry->d_inode;
4480         file_lock = locks_alloc_lock();
4481         if (!file_lock) {
4482                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4483                 status = nfserr_jukebox;
4484                 goto out;
4485         }
4486         locks_init_lock(file_lock);
4487         switch (lockt->lt_type) {
4488                 case NFS4_READ_LT:
4489                 case NFS4_READW_LT:
4490                         file_lock->fl_type = F_RDLCK;
4491                 break;
4492                 case NFS4_WRITE_LT:
4493                 case NFS4_WRITEW_LT:
4494                         file_lock->fl_type = F_WRLCK;
4495                 break;
4496                 default:
4497                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4498                         status = nfserr_inval;
4499                 goto out;
4500         }
4501
4502         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4503         if (lo)
4504                 file_lock->fl_owner = (fl_owner_t)lo;
4505         file_lock->fl_pid = current->tgid;
4506         file_lock->fl_flags = FL_POSIX;
4507
4508         file_lock->fl_start = lockt->lt_offset;
4509         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4510
4511         nfs4_transform_lock_offset(file_lock);
4512
4513         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4514         if (status)
4515                 goto out;
4516
4517         if (file_lock->fl_type != F_UNLCK) {
4518                 status = nfserr_denied;
4519                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4520         }
4521 out:
4522         nfs4_unlock_state();
4523         if (file_lock)
4524                 locks_free_lock(file_lock);
4525         return status;
4526 }
4527
4528 __be32
4529 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4530             struct nfsd4_locku *locku)
4531 {
4532         struct nfs4_ol_stateid *stp;
4533         struct file *filp = NULL;
4534         struct file_lock *file_lock = NULL;
4535         __be32 status;
4536         int err;
4537         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4538
4539         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4540                 (long long) locku->lu_offset,
4541                 (long long) locku->lu_length);
4542
4543         if (check_lock_length(locku->lu_offset, locku->lu_length))
4544                  return nfserr_inval;
4545
4546         nfs4_lock_state();
4547
4548         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4549                                         &locku->lu_stateid, NFS4_LOCK_STID,
4550                                         &stp, nn);
4551         if (status)
4552                 goto out;
4553         filp = find_any_file(stp->st_file);
4554         if (!filp) {
4555                 status = nfserr_lock_range;
4556                 goto out;
4557         }
4558         file_lock = locks_alloc_lock();
4559         if (!file_lock) {
4560                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4561                 status = nfserr_jukebox;
4562                 goto out;
4563         }
4564         locks_init_lock(file_lock);
4565         file_lock->fl_type = F_UNLCK;
4566         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4567         file_lock->fl_pid = current->tgid;
4568         file_lock->fl_file = filp;
4569         file_lock->fl_flags = FL_POSIX;
4570         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4571         file_lock->fl_start = locku->lu_offset;
4572
4573         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4574                                                 locku->lu_length);
4575         nfs4_transform_lock_offset(file_lock);
4576
4577         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4578         if (err) {
4579                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4580                 goto out_nfserr;
4581         }
4582         update_stateid(&stp->st_stid.sc_stateid);
4583         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4584
4585 out:
4586         nfsd4_bump_seqid(cstate, status);
4587         if (!cstate->replay_owner)
4588                 nfs4_unlock_state();
4589         if (file_lock)
4590                 locks_free_lock(file_lock);
4591         return status;
4592
4593 out_nfserr:
4594         status = nfserrno(err);
4595         goto out;
4596 }
4597
4598 /*
4599  * returns
4600  *      1: locks held by lockowner
4601  *      0: no locks held by lockowner
4602  */
4603 static int
4604 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4605 {
4606         struct file_lock **flpp;
4607         struct inode *inode = filp->fi_inode;
4608         int status = 0;
4609
4610         spin_lock(&inode->i_lock);
4611         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4612                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4613                         status = 1;
4614                         goto out;
4615                 }
4616         }
4617 out:
4618         spin_unlock(&inode->i_lock);
4619         return status;
4620 }
4621
4622 __be32
4623 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4624                         struct nfsd4_compound_state *cstate,
4625                         struct nfsd4_release_lockowner *rlockowner)
4626 {
4627         clientid_t *clid = &rlockowner->rl_clientid;
4628         struct nfs4_stateowner *sop;
4629         struct nfs4_lockowner *lo;
4630         struct nfs4_ol_stateid *stp;
4631         struct xdr_netobj *owner = &rlockowner->rl_owner;
4632         struct list_head matches;
4633         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4634         __be32 status;
4635         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4636
4637         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4638                 clid->cl_boot, clid->cl_id);
4639
4640         nfs4_lock_state();
4641
4642         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4643         if (status)
4644                 goto out;
4645
4646         status = nfserr_locks_held;
4647         INIT_LIST_HEAD(&matches);
4648
4649         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4650                 if (sop->so_is_open_owner)
4651                         continue;
4652                 if (!same_owner_str(sop, owner, clid))
4653                         continue;
4654                 list_for_each_entry(stp, &sop->so_stateids,
4655                                 st_perstateowner) {
4656                         lo = lockowner(sop);
4657                         if (check_for_locks(stp->st_file, lo))
4658                                 goto out;
4659                         list_add(&lo->lo_list, &matches);
4660                 }
4661         }
4662         /* Clients probably won't expect us to return with some (but not all)
4663          * of the lockowner state released; so don't release any until all
4664          * have been checked. */
4665         status = nfs_ok;
4666         while (!list_empty(&matches)) {
4667                 lo = list_entry(matches.next, struct nfs4_lockowner,
4668                                                                 lo_list);
4669                 /* unhash_stateowner deletes so_perclient only
4670                  * for openowners. */
4671                 list_del(&lo->lo_list);
4672                 release_lockowner(lo);
4673         }
4674 out:
4675         nfs4_unlock_state();
4676         return status;
4677 }
4678
4679 static inline struct nfs4_client_reclaim *
4680 alloc_reclaim(void)
4681 {
4682         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4683 }
4684
4685 bool
4686 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4687 {
4688         struct nfs4_client_reclaim *crp;
4689
4690         crp = nfsd4_find_reclaim_client(name, nn);
4691         return (crp && crp->cr_clp);
4692 }
4693
4694 /*
4695  * failure => all reset bets are off, nfserr_no_grace...
4696  */
4697 struct nfs4_client_reclaim *
4698 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4699 {
4700         unsigned int strhashval;
4701         struct nfs4_client_reclaim *crp;
4702
4703         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4704         crp = alloc_reclaim();
4705         if (crp) {
4706                 strhashval = clientstr_hashval(name);
4707                 INIT_LIST_HEAD(&crp->cr_strhash);
4708                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4709                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4710                 crp->cr_clp = NULL;
4711                 nn->reclaim_str_hashtbl_size++;
4712         }
4713         return crp;
4714 }
4715
4716 void
4717 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4718 {
4719         list_del(&crp->cr_strhash);
4720         kfree(crp);
4721         nn->reclaim_str_hashtbl_size--;
4722 }
4723
4724 void
4725 nfs4_release_reclaim(struct nfsd_net *nn)
4726 {
4727         struct nfs4_client_reclaim *crp = NULL;
4728         int i;
4729
4730         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4731                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4732                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4733                                         struct nfs4_client_reclaim, cr_strhash);
4734                         nfs4_remove_reclaim_record(crp, nn);
4735                 }
4736         }
4737         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4738 }
4739
4740 /*
4741  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4742 struct nfs4_client_reclaim *
4743 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4744 {
4745         unsigned int strhashval;
4746         struct nfs4_client_reclaim *crp = NULL;
4747
4748         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4749
4750         strhashval = clientstr_hashval(recdir);
4751         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4752                 if (same_name(crp->cr_recdir, recdir)) {
4753                         return crp;
4754                 }
4755         }
4756         return NULL;
4757 }
4758
4759 /*
4760 * Called from OPEN. Look for clientid in reclaim list.
4761 */
4762 __be32
4763 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4764 {
4765         struct nfs4_client *clp;
4766
4767         /* find clientid in conf_id_hashtbl */
4768         clp = find_confirmed_client(clid, sessions, nn);
4769         if (clp == NULL)
4770                 return nfserr_reclaim_bad;
4771
4772         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4773 }
4774
4775 #ifdef CONFIG_NFSD_FAULT_INJECTION
4776
4777 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4778 {
4779         if (mark_client_expired(clp))
4780                 return 0;
4781         expire_client(clp);
4782         return 1;
4783 }
4784
4785 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4786 {
4787         char buf[INET6_ADDRSTRLEN];
4788         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4789         printk(KERN_INFO "NFS Client: %s\n", buf);
4790         return 1;
4791 }
4792
4793 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4794                              const char *type)
4795 {
4796         char buf[INET6_ADDRSTRLEN];
4797         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4798         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4799 }
4800
4801 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4802 {
4803         struct nfs4_openowner *oop;
4804         struct nfs4_lockowner *lop, *lo_next;
4805         struct nfs4_ol_stateid *stp, *st_next;
4806         u64 count = 0;
4807
4808         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4809                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4810                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4811                                 if (func)
4812                                         func(lop);
4813                                 if (++count == max)
4814                                         return count;
4815                         }
4816                 }
4817         }
4818
4819         return count;
4820 }
4821
4822 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4823 {
4824         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4825 }
4826
4827 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4828 {
4829         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4830         nfsd_print_count(clp, count, "locked files");
4831         return count;
4832 }
4833
4834 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4835 {
4836         struct nfs4_openowner *oop, *next;
4837         u64 count = 0;
4838
4839         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4840                 if (func)
4841                         func(oop);
4842                 if (++count == max)
4843                         break;
4844         }
4845
4846         return count;
4847 }
4848
4849 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4850 {
4851         return nfsd_foreach_client_open(clp, max, release_openowner);
4852 }
4853
4854 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4855 {
4856         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4857         nfsd_print_count(clp, count, "open files");
4858         return count;
4859 }
4860
4861 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4862                                      struct list_head *victims)
4863 {
4864         struct nfs4_delegation *dp, *next;
4865         u64 count = 0;
4866
4867         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4868                 if (victims)
4869                         list_move(&dp->dl_recall_lru, victims);
4870                 if (++count == max)
4871                         break;
4872         }
4873         return count;
4874 }
4875
4876 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4877 {
4878         struct nfs4_delegation *dp, *next;
4879         LIST_HEAD(victims);
4880         u64 count;
4881
4882         spin_lock(&recall_lock);
4883         count = nfsd_find_all_delegations(clp, max, &victims);
4884         spin_unlock(&recall_lock);
4885
4886         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4887                 revoke_delegation(dp);
4888
4889         return count;
4890 }
4891
4892 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4893 {
4894         struct nfs4_delegation *dp, *next;
4895         LIST_HEAD(victims);
4896         u64 count;
4897
4898         spin_lock(&recall_lock);
4899         count = nfsd_find_all_delegations(clp, max, &victims);
4900         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4901                 nfsd_break_one_deleg(dp);
4902         spin_unlock(&recall_lock);
4903
4904         return count;
4905 }
4906
4907 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4908 {
4909         u64 count = 0;
4910
4911         spin_lock(&recall_lock);
4912         count = nfsd_find_all_delegations(clp, max, NULL);
4913         spin_unlock(&recall_lock);
4914
4915         nfsd_print_count(clp, count, "delegations");
4916         return count;
4917 }
4918
4919 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4920 {
4921         struct nfs4_client *clp, *next;
4922         u64 count = 0;
4923         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4924
4925         if (!nfsd_netns_ready(nn))
4926                 return 0;
4927
4928         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4929                 count += func(clp, max - count);
4930                 if ((max != 0) && (count >= max))
4931                         break;
4932         }
4933
4934         return count;
4935 }
4936
4937 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4938 {
4939         struct nfs4_client *clp;
4940         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4941
4942         if (!nfsd_netns_ready(nn))
4943                 return NULL;
4944
4945         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4946                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4947                         return clp;
4948         }
4949         return NULL;
4950 }
4951
4952 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4953
4954 /* initialization to perform at module load time: */
4955
4956 void
4957 nfs4_state_init(void)
4958 {
4959 }
4960
4961 /*
4962  * Since the lifetime of a delegation isn't limited to that of an open, a
4963  * client may quite reasonably hang on to a delegation as long as it has
4964  * the inode cached.  This becomes an obvious problem the first time a
4965  * client's inode cache approaches the size of the server's total memory.
4966  *
4967  * For now we avoid this problem by imposing a hard limit on the number
4968  * of delegations, which varies according to the server's memory size.
4969  */
4970 static void
4971 set_max_delegations(void)
4972 {
4973         /*
4974          * Allow at most 4 delegations per megabyte of RAM.  Quick
4975          * estimates suggest that in the worst case (where every delegation
4976          * is for a different inode), a delegation could take about 1.5K,
4977          * giving a worst case usage of about 6% of memory.
4978          */
4979         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4980 }
4981
4982 static int nfs4_state_create_net(struct net *net)
4983 {
4984         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4985         int i;
4986
4987         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4988                         CLIENT_HASH_SIZE, GFP_KERNEL);
4989         if (!nn->conf_id_hashtbl)
4990                 goto err;
4991         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4992                         CLIENT_HASH_SIZE, GFP_KERNEL);
4993         if (!nn->unconf_id_hashtbl)
4994                 goto err_unconf_id;
4995         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4996                         OWNER_HASH_SIZE, GFP_KERNEL);
4997         if (!nn->ownerstr_hashtbl)
4998                 goto err_ownerstr;
4999         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
5000                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
5001         if (!nn->lockowner_ino_hashtbl)
5002                 goto err_lockowner_ino;
5003         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
5004                         SESSION_HASH_SIZE, GFP_KERNEL);
5005         if (!nn->sessionid_hashtbl)
5006                 goto err_sessionid;
5007
5008         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5009                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
5010                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
5011         }
5012         for (i = 0; i < OWNER_HASH_SIZE; i++)
5013                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
5014         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
5015                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
5016         for (i = 0; i < SESSION_HASH_SIZE; i++)
5017                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
5018         nn->conf_name_tree = RB_ROOT;
5019         nn->unconf_name_tree = RB_ROOT;
5020         INIT_LIST_HEAD(&nn->client_lru);
5021         INIT_LIST_HEAD(&nn->close_lru);
5022         INIT_LIST_HEAD(&nn->del_recall_lru);
5023         spin_lock_init(&nn->client_lock);
5024
5025         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
5026         get_net(net);
5027
5028         return 0;
5029
5030 err_sessionid:
5031         kfree(nn->lockowner_ino_hashtbl);
5032 err_lockowner_ino:
5033         kfree(nn->ownerstr_hashtbl);
5034 err_ownerstr:
5035         kfree(nn->unconf_id_hashtbl);
5036 err_unconf_id:
5037         kfree(nn->conf_id_hashtbl);
5038 err:
5039         return -ENOMEM;
5040 }
5041
5042 static void
5043 nfs4_state_destroy_net(struct net *net)
5044 {
5045         int i;
5046         struct nfs4_client *clp = NULL;
5047         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5048         struct rb_node *node, *tmp;
5049
5050         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5051                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
5052                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
5053                         destroy_client(clp);
5054                 }
5055         }
5056
5057         node = rb_first(&nn->unconf_name_tree);
5058         while (node != NULL) {
5059                 tmp = node;
5060                 node = rb_next(tmp);
5061                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
5062                 rb_erase(tmp, &nn->unconf_name_tree);
5063                 destroy_client(clp);
5064         }
5065
5066         kfree(nn->sessionid_hashtbl);
5067         kfree(nn->lockowner_ino_hashtbl);
5068         kfree(nn->ownerstr_hashtbl);
5069         kfree(nn->unconf_id_hashtbl);
5070         kfree(nn->conf_id_hashtbl);
5071         put_net(net);
5072 }
5073
5074 int
5075 nfs4_state_start_net(struct net *net)
5076 {
5077         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5078         int ret;
5079
5080         ret = nfs4_state_create_net(net);
5081         if (ret)
5082                 return ret;
5083         nfsd4_client_tracking_init(net);
5084         nn->boot_time = get_seconds();
5085         locks_start_grace(net, &nn->nfsd4_manager);
5086         nn->grace_ended = false;
5087         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
5088                nn->nfsd4_grace, net);
5089         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
5090         return 0;
5091 }
5092
5093 /* initialization to perform when the nfsd service is started: */
5094
5095 int
5096 nfs4_state_start(void)
5097 {
5098         int ret;
5099
5100         ret = set_callback_cred();
5101         if (ret)
5102                 return -ENOMEM;
5103         laundry_wq = create_singlethread_workqueue("nfsd4");
5104         if (laundry_wq == NULL) {
5105                 ret = -ENOMEM;
5106                 goto out_recovery;
5107         }
5108         ret = nfsd4_create_callback_queue();
5109         if (ret)
5110                 goto out_free_laundry;
5111
5112         set_max_delegations();
5113
5114         return 0;
5115
5116 out_free_laundry:
5117         destroy_workqueue(laundry_wq);
5118 out_recovery:
5119         return ret;
5120 }
5121
5122 /* should be called with the state lock held */
5123 void
5124 nfs4_state_shutdown_net(struct net *net)
5125 {
5126         struct nfs4_delegation *dp = NULL;
5127         struct list_head *pos, *next, reaplist;
5128         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5129
5130         cancel_delayed_work_sync(&nn->laundromat_work);
5131         locks_end_grace(&nn->nfsd4_manager);
5132
5133         INIT_LIST_HEAD(&reaplist);
5134         spin_lock(&recall_lock);
5135         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5136                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5137                 list_move(&dp->dl_recall_lru, &reaplist);
5138         }
5139         spin_unlock(&recall_lock);
5140         list_for_each_safe(pos, next, &reaplist) {
5141                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5142                 destroy_delegation(dp);
5143         }
5144
5145         nfsd4_client_tracking_exit(net);
5146         nfs4_state_destroy_net(net);
5147 }
5148
5149 void
5150 nfs4_state_shutdown(void)
5151 {
5152         destroy_workqueue(laundry_wq);
5153         nfsd4_destroy_callback_queue();
5154 }
5155
5156 static void
5157 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5158 {
5159         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5160                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5161 }
5162
5163 static void
5164 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5165 {
5166         if (cstate->minorversion) {
5167                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5168                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5169         }
5170 }
5171
5172 void
5173 clear_current_stateid(struct nfsd4_compound_state *cstate)
5174 {
5175         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5176 }
5177
5178 /*
5179  * functions to set current state id
5180  */
5181 void
5182 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5183 {
5184         put_stateid(cstate, &odp->od_stateid);
5185 }
5186
5187 void
5188 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5189 {
5190         put_stateid(cstate, &open->op_stateid);
5191 }
5192
5193 void
5194 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5195 {
5196         put_stateid(cstate, &close->cl_stateid);
5197 }
5198
5199 void
5200 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5201 {
5202         put_stateid(cstate, &lock->lk_resp_stateid);
5203 }
5204
5205 /*
5206  * functions to consume current state id
5207  */
5208
5209 void
5210 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5211 {
5212         get_stateid(cstate, &odp->od_stateid);
5213 }
5214
5215 void
5216 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5217 {
5218         get_stateid(cstate, &drp->dr_stateid);
5219 }
5220
5221 void
5222 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5223 {
5224         get_stateid(cstate, &fsp->fr_stateid);
5225 }
5226
5227 void
5228 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5229 {
5230         get_stateid(cstate, &setattr->sa_stateid);
5231 }
5232
5233 void
5234 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5235 {
5236         get_stateid(cstate, &close->cl_stateid);
5237 }
5238
5239 void
5240 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5241 {
5242         get_stateid(cstate, &locku->lu_stateid);
5243 }
5244
5245 void
5246 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5247 {
5248         get_stateid(cstate, &read->rd_stateid);
5249 }
5250
5251 void
5252 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5253 {
5254         get_stateid(cstate, &write->wr_stateid);
5255 }