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 void nfsd4_put_session(struct nfsd4_session *ses)
 101 {
 102         atomic_dec(&ses->se_ref);
 103 }
 104
 105 static bool is_session_dead(struct nfsd4_session *ses)
 106 {
 107         return ses->se_flags & NFS4_SESSION_DEAD;
 108 }
 109
 110 static __be32 mark_session_dead_locked(struct nfsd4_session *ses)
 111 {
 112         if (atomic_read(&ses->se_ref))
 113                 return nfserr_jukebox;
 114         ses->se_flags |= NFS4_SESSION_DEAD;
 115         return nfs_ok;
 116 }
 117
 118 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 119 {
 120         if (is_session_dead(ses))
 121                 return nfserr_badsession;
 122         atomic_inc(&ses->se_ref);
 123         return nfs_ok;
 124 }
 125
 126 void
 127 nfs4_unlock_state(void)
 128 {
 129         mutex_unlock(&client_mutex);
 130 }
 131
 132 static bool is_client_expired(struct nfs4_client *clp)
 133 {
 134         return clp->cl_time == 0;
 135 }
 136
 137 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
 138 {
 139         if (atomic_read(&clp->cl_refcount))
 140                 return nfserr_jukebox;
 141         clp->cl_time = 0;
 142         return nfs_ok;
 143 }
 144
 145 static __be32 mark_client_expired(struct nfs4_client *clp)
 146 {
 147         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 148         __be32 ret;
 149
 150         spin_lock(&nn->client_lock);
 151         ret = mark_client_expired_locked(clp);
 152         spin_unlock(&nn->client_lock);
 153         return ret;
 154 }
 155
 156 static __be32 get_client_locked(struct nfs4_client *clp)
 157 {
 158         if (is_client_expired(clp))
 159                 return nfserr_expired;
 160         atomic_inc(&clp->cl_refcount);
 161         return nfs_ok;
 162 }
 163
 164 /* must be called under the client_lock */
 165 static inline void
 166 renew_client_locked(struct nfs4_client *clp)
 167 {
 168         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 169
 170         if (is_client_expired(clp)) {
 171                 WARN_ON(1);
 172                 printk("%s: client (clientid %08x/%08x) already expired\n",
 173                         __func__,
 174                         clp->cl_clientid.cl_boot,
 175                         clp->cl_clientid.cl_id);
 176                 return;
 177         }
 178
 179         dprintk("renewing client (clientid %08x/%08x)\n",
 180                         clp->cl_clientid.cl_boot,
 181                         clp->cl_clientid.cl_id);
 182         list_move_tail(&clp->cl_lru, &nn->client_lru);
 183         clp->cl_time = get_seconds();
 184 }
 185
 186 static inline void
 187 renew_client(struct nfs4_client *clp)
 188 {
 189         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 190
 191         spin_lock(&nn->client_lock);
 192         renew_client_locked(clp);
 193         spin_unlock(&nn->client_lock);
 194 }
 195
 196 static void put_client_renew_locked(struct nfs4_client *clp)
 197 {
 198         if (!atomic_dec_and_test(&clp->cl_refcount))
 199                 return;
 200         if (!is_client_expired(clp))
 201                 renew_client_locked(clp);
 202 }
 203
 204 void put_client_renew(struct nfs4_client *clp)
 205 {
 206         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 207
 208         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 209                 return;
 210         if (!is_client_expired(clp))
 211                 renew_client_locked(clp);
 212         spin_unlock(&nn->client_lock);
 213 }
 214
 215
 216 static inline u32
 217 opaque_hashval(const void *ptr, int nbytes)
 218 {
 219         unsigned char *cptr = (unsigned char *) ptr;
 220
 221         u32 x = 0;
 222         while (nbytes--) {
 223                 x *= 37;
 224                 x += *cptr++;
 225         }
 226         return x;
 227 }
 228
 229 static void nfsd4_free_file(struct nfs4_file *f)
 230 {
 231         kmem_cache_free(file_slab, f);
 232 }
 233
 234 static inline void
 235 put_nfs4_file(struct nfs4_file *fi)
 236 {
 237         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 238                 hlist_del(&fi->fi_hash);
 239                 spin_unlock(&recall_lock);
 240                 iput(fi->fi_inode);
 241                 nfsd4_free_file(fi);
 242         }
 243 }
 244
 245 static inline void
 246 get_nfs4_file(struct nfs4_file *fi)
 247 {
 248         atomic_inc(&fi->fi_ref);
 249 }
 250
 251 static int num_delegations;
 252 unsigned long max_delegations;
 253
 254 /*
 255  * Open owner state (share locks)
 256  */
 257
 258 /* hash tables for lock and open owners */
 259 #define OWNER_HASH_BITS              8
 260 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 261 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 262
 263 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 264 {
 265         unsigned int ret;
 266
 267         ret = opaque_hashval(ownername->data, ownername->len);
 268         ret += clientid;
 269         return ret & OWNER_HASH_MASK;
 270 }
 271
 272 /* hash table for nfs4_file */
 273 #define FILE_HASH_BITS                   8
 274 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 275
 276 static unsigned int file_hashval(struct inode *ino)
 277 {
 278         /* XXX: why are we hashing on inode pointer, anyway? */
 279         return hash_ptr(ino, FILE_HASH_BITS);
 280 }
 281
 282 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 283
 284 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 285 {
 286         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 287         atomic_inc(&fp->fi_access[oflag]);
 288 }
 289
 290 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 291 {
 292         if (oflag == O_RDWR) {
 293                 __nfs4_file_get_access(fp, O_RDONLY);
 294                 __nfs4_file_get_access(fp, O_WRONLY);
 295         } else
 296                 __nfs4_file_get_access(fp, oflag);
 297 }
 298
 299 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 300 {
 301         if (fp->fi_fds[oflag]) {
 302                 fput(fp->fi_fds[oflag]);
 303                 fp->fi_fds[oflag] = NULL;
 304         }
 305 }
 306
 307 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 308 {
 309         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 310                 nfs4_file_put_fd(fp, oflag);
 311                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 312                         nfs4_file_put_fd(fp, O_RDWR);
 313         }
 314 }
 315
 316 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 317 {
 318         if (oflag == O_RDWR) {
 319                 __nfs4_file_put_access(fp, O_RDONLY);
 320                 __nfs4_file_put_access(fp, O_WRONLY);
 321         } else
 322                 __nfs4_file_put_access(fp, oflag);
 323 }
 324
 325 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct
 326 kmem_cache *slab)
 327 {
 328         struct idr *stateids = &cl->cl_stateids;
 329         struct nfs4_stid *stid;
 330         int new_id;
 331
 332         stid = kmem_cache_alloc(slab, GFP_KERNEL);
 333         if (!stid)
 334                 return NULL;
 335
 336         new_id = idr_alloc_cyclic(stateids, stid, 0, 0, GFP_KERNEL);
 337         if (new_id < 0)
 338                 goto out_free;
 339         stid->sc_client = cl;
 340         stid->sc_type = 0;
 341         stid->sc_stateid.si_opaque.so_id = new_id;
 342         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 343         /* Will be incremented before return to client: */
 344         stid->sc_stateid.si_generation = 0;
 345
 346         /*
 347          * It shouldn't be a problem to reuse an opaque stateid value.
 348          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 349          * example, a stray write retransmission could be accepted by
 350          * the server when it should have been rejected.  Therefore,
 351          * adopt a trick from the sctp code to attempt to maximize the
 352          * amount of time until an id is reused, by ensuring they always
 353          * "increase" (mod INT_MAX):
 354          */
 355         return stid;
 356 out_free:
 357         kmem_cache_free(slab, stid);
 358         return NULL;
 359 }
 360
 361 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 362 {
 363         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 364 }
 365
 366 static struct nfs4_delegation *
 367 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
 368 {
 369         struct nfs4_delegation *dp;
 370         struct nfs4_file *fp = stp->st_file;
 371
 372         dprintk("NFSD alloc_init_deleg\n");
 373         /*
 374          * Major work on the lease subsystem (for example, to support
 375          * calbacks on stat) will be required before we can support
 376          * write delegations properly.
 377          */
 378         if (type != NFS4_OPEN_DELEGATE_READ)
 379                 return NULL;
 380         if (fp->fi_had_conflict)
 381                 return NULL;
 382         if (num_delegations > max_delegations)
 383                 return NULL;
 384         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 385         if (dp == NULL)
 386                 return dp;
 387         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 388         /*
 389          * delegation seqid's are never incremented.  The 4.1 special
 390          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 391          * 0 anyway just for consistency and use 1:
 392          */
 393         dp->dl_stid.sc_stateid.si_generation = 1;
 394         num_delegations++;
 395         INIT_LIST_HEAD(&dp->dl_perfile);
 396         INIT_LIST_HEAD(&dp->dl_perclnt);
 397         INIT_LIST_HEAD(&dp->dl_recall_lru);
 398         get_nfs4_file(fp);
 399         dp->dl_file = fp;
 400         dp->dl_type = type;
 401         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 402         dp->dl_time = 0;
 403         atomic_set(&dp->dl_count, 1);
 404         nfsd4_init_callback(&dp->dl_recall);
 405         return dp;
 406 }
 407
 408 static void remove_stid(struct nfs4_stid *s)
 409 {
 410         struct idr *stateids = &s->sc_client->cl_stateids;
 411
 412         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 413 }
 414
 415 void
 416 nfs4_put_delegation(struct nfs4_delegation *dp)
 417 {
 418         if (atomic_dec_and_test(&dp->dl_count)) {
 419                 kmem_cache_free(deleg_slab, dp);
 420                 num_delegations--;
 421         }
 422 }
 423
 424 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 425 {
 426         if (atomic_dec_and_test(&fp->fi_delegees)) {
 427                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 428                 fp->fi_lease = NULL;
 429                 fput(fp->fi_deleg_file);
 430                 fp->fi_deleg_file = NULL;
 431         }
 432 }
 433
 434 static void unhash_stid(struct nfs4_stid *s)
 435 {
 436         s->sc_type = 0;
 437 }
 438
 439 /* Called under the state lock. */
 440 static void
 441 unhash_delegation(struct nfs4_delegation *dp)
 442 {
 443         list_del_init(&dp->dl_perclnt);
 444         spin_lock(&recall_lock);
 445         list_del_init(&dp->dl_perfile);
 446         list_del_init(&dp->dl_recall_lru);
 447         spin_unlock(&recall_lock);
 448         nfs4_put_deleg_lease(dp->dl_file);
 449         put_nfs4_file(dp->dl_file);
 450         dp->dl_file = NULL;
 451 }
 452
 453
 454
 455 static void destroy_revoked_delegation(struct nfs4_delegation *dp)
 456 {
 457         list_del_init(&dp->dl_recall_lru);
 458         remove_stid(&dp->dl_stid);
 459         nfs4_put_delegation(dp);
 460 }
 461
 462 static void destroy_delegation(struct nfs4_delegation *dp)
 463 {
 464         unhash_delegation(dp);
 465         remove_stid(&dp->dl_stid);
 466         nfs4_put_delegation(dp);
 467 }
 468
 469 static void revoke_delegation(struct nfs4_delegation *dp)
 470 {
 471         struct nfs4_client *clp = dp->dl_stid.sc_client;
 472
 473         if (clp->cl_minorversion == 0)
 474                 destroy_delegation(dp);
 475         else {
 476                 unhash_delegation(dp);
 477                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 478                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 479         }
 480 }
 481
 482 /*
 483  * SETCLIENTID state
 484  */
 485
 486 static unsigned int clientid_hashval(u32 id)
 487 {
 488         return id & CLIENT_HASH_MASK;
 489 }
 490
 491 static unsigned int clientstr_hashval(const char *name)
 492 {
 493         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 494 }
 495
 496 /*
 497  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 498  * st_{access,deny}_bmap field of the stateid, in order to track not
 499  * only what share bits are currently in force, but also what
 500  * combinations of share bits previous opens have used.  This allows us
 501  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 502  * return an error if the client attempt to downgrade to a combination
 503  * of share bits not explicable by closing some of its previous opens.
 504  *
 505  * XXX: This enforcement is actually incomplete, since we don't keep
 506  * track of access/deny bit combinations; so, e.g., we allow:
 507  *
 508  *      OPEN allow read, deny write
 509  *      OPEN allow both, deny none
 510  *      DOWNGRADE allow read, deny none
 511  *
 512  * which we should reject.
 513  */
 514 static unsigned int
 515 bmap_to_share_mode(unsigned long bmap) {
 516         int i;
 517         unsigned int access = 0;
 518
 519         for (i = 1; i < 4; i++) {
 520                 if (test_bit(i, &bmap))
 521                         access |= i;
 522         }
 523         return access;
 524 }
 525
 526 static bool
 527 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 528         unsigned int access, deny;
 529
 530         access = bmap_to_share_mode(stp->st_access_bmap);
 531         deny = bmap_to_share_mode(stp->st_deny_bmap);
 532         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 533                 return false;
 534         return true;
 535 }
 536
 537 /* set share access for a given stateid */
 538 static inline void
 539 set_access(u32 access, struct nfs4_ol_stateid *stp)
 540 {
 541         __set_bit(access, &stp->st_access_bmap);
 542 }
 543
 544 /* clear share access for a given stateid */
 545 static inline void
 546 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 547 {
 548         __clear_bit(access, &stp->st_access_bmap);
 549 }
 550
 551 /* test whether a given stateid has access */
 552 static inline bool
 553 test_access(u32 access, struct nfs4_ol_stateid *stp)
 554 {
 555         return test_bit(access, &stp->st_access_bmap);
 556 }
 557
 558 /* set share deny for a given stateid */
 559 static inline void
 560 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 561 {
 562         __set_bit(access, &stp->st_deny_bmap);
 563 }
 564
 565 /* clear share deny for a given stateid */
 566 static inline void
 567 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 568 {
 569         __clear_bit(access, &stp->st_deny_bmap);
 570 }
 571
 572 /* test whether a given stateid is denying specific access */
 573 static inline bool
 574 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 575 {
 576         return test_bit(access, &stp->st_deny_bmap);
 577 }
 578
 579 static int nfs4_access_to_omode(u32 access)
 580 {
 581         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 582         case NFS4_SHARE_ACCESS_READ:
 583                 return O_RDONLY;
 584         case NFS4_SHARE_ACCESS_WRITE:
 585                 return O_WRONLY;
 586         case NFS4_SHARE_ACCESS_BOTH:
 587                 return O_RDWR;
 588         }
 589         WARN_ON_ONCE(1);
 590         return O_RDONLY;
 591 }
 592
 593 /* release all access and file references for a given stateid */
 594 static void
 595 release_all_access(struct nfs4_ol_stateid *stp)
 596 {
 597         int i;
 598
 599         for (i = 1; i < 4; i++) {
 600                 if (test_access(i, stp))
 601                         nfs4_file_put_access(stp->st_file,
 602                                              nfs4_access_to_omode(i));
 603                 clear_access(i, stp);
 604         }
 605 }
 606
 607 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 608 {
 609         list_del(&stp->st_perfile);
 610         list_del(&stp->st_perstateowner);
 611 }
 612
 613 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 614 {
 615         release_all_access(stp);
 616         put_nfs4_file(stp->st_file);
 617         stp->st_file = NULL;
 618 }
 619
 620 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 621 {
 622         remove_stid(&stp->st_stid);
 623         kmem_cache_free(stateid_slab, stp);
 624 }
 625
 626 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 627 {
 628         struct file *file;
 629
 630         unhash_generic_stateid(stp);
 631         unhash_stid(&stp->st_stid);
 632         file = find_any_file(stp->st_file);
 633         if (file)
 634                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 635         close_generic_stateid(stp);
 636         free_generic_stateid(stp);
 637 }
 638
 639 static void unhash_lockowner(struct nfs4_lockowner *lo)
 640 {
 641         struct nfs4_ol_stateid *stp;
 642
 643         list_del(&lo->lo_owner.so_strhash);
 644         list_del(&lo->lo_perstateid);
 645         list_del(&lo->lo_owner_ino_hash);
 646         while (!list_empty(&lo->lo_owner.so_stateids)) {
 647                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 648                                 struct nfs4_ol_stateid, st_perstateowner);
 649                 release_lock_stateid(stp);
 650         }
 651 }
 652
 653 static void release_lockowner(struct nfs4_lockowner *lo)
 654 {
 655         unhash_lockowner(lo);
 656         nfs4_free_lockowner(lo);
 657 }
 658
 659 static void
 660 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 661 {
 662         struct nfs4_lockowner *lo;
 663
 664         while (!list_empty(&open_stp->st_lockowners)) {
 665                 lo = list_entry(open_stp->st_lockowners.next,
 666                                 struct nfs4_lockowner, lo_perstateid);
 667                 release_lockowner(lo);
 668         }
 669 }
 670
 671 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 672 {
 673         unhash_generic_stateid(stp);
 674         release_stateid_lockowners(stp);
 675         close_generic_stateid(stp);
 676 }
 677
 678 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 679 {
 680         unhash_open_stateid(stp);
 681         unhash_stid(&stp->st_stid);
 682         free_generic_stateid(stp);
 683 }
 684
 685 static void unhash_openowner(struct nfs4_openowner *oo)
 686 {
 687         struct nfs4_ol_stateid *stp;
 688
 689         list_del(&oo->oo_owner.so_strhash);
 690         list_del(&oo->oo_perclient);
 691         while (!list_empty(&oo->oo_owner.so_stateids)) {
 692                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 693                                 struct nfs4_ol_stateid, st_perstateowner);
 694                 release_open_stateid(stp);
 695         }
 696 }
 697
 698 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 699 {
 700         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 701
 702         if (s) {
 703                 unhash_stid(&s->st_stid);
 704                 free_generic_stateid(s);
 705                 oo->oo_last_closed_stid = NULL;
 706         }
 707 }
 708
 709 static void release_openowner(struct nfs4_openowner *oo)
 710 {
 711         unhash_openowner(oo);
 712         list_del(&oo->oo_close_lru);
 713         release_last_closed_stateid(oo);
 714         nfs4_free_openowner(oo);
 715 }
 716
 717 static inline int
 718 hash_sessionid(struct nfs4_sessionid *sessionid)
 719 {
 720         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 721
 722         return sid->sequence % SESSION_HASH_SIZE;
 723 }
 724
 725 #ifdef NFSD_DEBUG
 726 static inline void
 727 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 728 {
 729         u32 *ptr = (u32 *)(&sessionid->data[0]);
 730         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 731 }
 732 #else
 733 static inline void
 734 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 735 {
 736 }
 737 #endif
 738
 739 /*
 740  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
 741  * won't be used for replay.
 742  */
 743 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
 744 {
 745         struct nfs4_stateowner *so = cstate->replay_owner;
 746
 747         if (nfserr == nfserr_replay_me)
 748                 return;
 749
 750         if (!seqid_mutating_err(ntohl(nfserr))) {
 751                 cstate->replay_owner = NULL;
 752                 return;
 753         }
 754         if (!so)
 755                 return;
 756         if (so->so_is_open_owner)
 757                 release_last_closed_stateid(openowner(so));
 758         so->so_seqid++;
 759         return;
 760 }
 761
 762 static void
 763 gen_sessionid(struct nfsd4_session *ses)
 764 {
 765         struct nfs4_client *clp = ses->se_client;
 766         struct nfsd4_sessionid *sid;
 767
 768         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 769         sid->clientid = clp->cl_clientid;
 770         sid->sequence = current_sessionid++;
 771         sid->reserved = 0;
 772 }
 773
 774 /*
 775  * The protocol defines ca_maxresponssize_cached to include the size of
 776  * the rpc header, but all we need to cache is the data starting after
 777  * the end of the initial SEQUENCE operation--the rest we regenerate
 778  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 779  * value that is the number of bytes in our cache plus a few additional
 780  * bytes.  In order to stay on the safe side, and not promise more than
 781  * we can cache, those additional bytes must be the minimum possible: 24
 782  * bytes of rpc header (xid through accept state, with AUTH_NULL
 783  * verifier), 12 for the compound header (with zero-length tag), and 44
 784  * for the SEQUENCE op response:
 785  */
 786 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 787
 788 static void
 789 free_session_slots(struct nfsd4_session *ses)
 790 {
 791         int i;
 792
 793         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 794                 kfree(ses->se_slots[i]);
 795 }
 796
 797 /*
 798  * We don't actually need to cache the rpc and session headers, so we
 799  * can allocate a little less for each slot:
 800  */
 801 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
 802 {
 803         u32 size;
 804
 805         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
 806                 size = 0;
 807         else
 808                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 809         return size + sizeof(struct nfsd4_slot);
 810 }
 811
 812 /*
 813  * XXX: If we run out of reserved DRC memory we could (up to a point)
 814  * re-negotiate active sessions and reduce their slot usage to make
 815  * room for new connections. For now we just fail the create session.
 816  */
 817 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
 818 {
 819         u32 slotsize = slot_bytes(ca);
 820         u32 num = ca->maxreqs;
 821         int avail;
 822
 823         spin_lock(&nfsd_drc_lock);
 824         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
 825                     nfsd_drc_max_mem - nfsd_drc_mem_used);
 826         num = min_t(int, num, avail / slotsize);
 827         nfsd_drc_mem_used += num * slotsize;
 828         spin_unlock(&nfsd_drc_lock);
 829
 830         return num;
 831 }
 832
 833 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
 834 {
 835         int slotsize = slot_bytes(ca);
 836
 837         spin_lock(&nfsd_drc_lock);
 838         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
 839         spin_unlock(&nfsd_drc_lock);
 840 }
 841
 842 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *attrs)
 843 {
 844         int numslots = attrs->maxreqs;
 845         int slotsize = slot_bytes(attrs);
 846         struct nfsd4_session *new;
 847         int mem, i;
 848
 849         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 850                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 851         mem = numslots * sizeof(struct nfsd4_slot *);
 852
 853         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 854         if (!new)
 855                 return NULL;
 856         /* allocate each struct nfsd4_slot and data cache in one piece */
 857         for (i = 0; i < numslots; i++) {
 858                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
 859                 if (!new->se_slots[i])
 860                         goto out_free;
 861         }
 862         return new;
 863 out_free:
 864         while (i--)
 865                 kfree(new->se_slots[i]);
 866         kfree(new);
 867         return NULL;
 868 }
 869
 870 static void free_conn(struct nfsd4_conn *c)
 871 {
 872         svc_xprt_put(c->cn_xprt);
 873         kfree(c);
 874 }
 875
 876 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 877 {
 878         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 879         struct nfs4_client *clp = c->cn_session->se_client;
 880
 881         spin_lock(&clp->cl_lock);
 882         if (!list_empty(&c->cn_persession)) {
 883                 list_del(&c->cn_persession);
 884                 free_conn(c);
 885         }
 886         nfsd4_probe_callback(clp);
 887         spin_unlock(&clp->cl_lock);
 888 }
 889
 890 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 891 {
 892         struct nfsd4_conn *conn;
 893
 894         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 895         if (!conn)
 896                 return NULL;
 897         svc_xprt_get(rqstp->rq_xprt);
 898         conn->cn_xprt = rqstp->rq_xprt;
 899         conn->cn_flags = flags;
 900         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 901         return conn;
 902 }
 903
 904 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 905 {
 906         conn->cn_session = ses;
 907         list_add(&conn->cn_persession, &ses->se_conns);
 908 }
 909
 910 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 911 {
 912         struct nfs4_client *clp = ses->se_client;
 913
 914         spin_lock(&clp->cl_lock);
 915         __nfsd4_hash_conn(conn, ses);
 916         spin_unlock(&clp->cl_lock);
 917 }
 918
 919 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 920 {
 921         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 922         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 923 }
 924
 925 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 926 {
 927         int ret;
 928
 929         nfsd4_hash_conn(conn, ses);
 930         ret = nfsd4_register_conn(conn);
 931         if (ret)
 932                 /* oops; xprt is already down: */
 933                 nfsd4_conn_lost(&conn->cn_xpt_user);
 934         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 935                 /* callback channel may be back up */
 936                 nfsd4_probe_callback(ses->se_client);
 937         }
 938 }
 939
 940 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 941 {
 942         u32 dir = NFS4_CDFC4_FORE;
 943
 944         if (cses->flags & SESSION4_BACK_CHAN)
 945                 dir |= NFS4_CDFC4_BACK;
 946         return alloc_conn(rqstp, dir);
 947 }
 948
 949 /* must be called under client_lock */
 950 static void nfsd4_del_conns(struct nfsd4_session *s)
 951 {
 952         struct nfs4_client *clp = s->se_client;
 953         struct nfsd4_conn *c;
 954
 955         spin_lock(&clp->cl_lock);
 956         while (!list_empty(&s->se_conns)) {
 957                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 958                 list_del_init(&c->cn_persession);
 959                 spin_unlock(&clp->cl_lock);
 960
 961                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 962                 free_conn(c);
 963
 964                 spin_lock(&clp->cl_lock);
 965         }
 966         spin_unlock(&clp->cl_lock);
 967 }
 968
 969 static void __free_session(struct nfsd4_session *ses)
 970 {
 971         free_session_slots(ses);
 972         kfree(ses);
 973 }
 974
 975 static void free_session(struct nfsd4_session *ses)
 976 {
 977         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
 978
 979         lockdep_assert_held(&nn->client_lock);
 980         nfsd4_del_conns(ses);
 981         nfsd4_put_drc_mem(&ses->se_fchannel);
 982         __free_session(ses);
 983 }
 984
 985 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 986 {
 987         int idx;
 988         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 989
 990         new->se_client = clp;
 991         gen_sessionid(new);
 992
 993         INIT_LIST_HEAD(&new->se_conns);
 994
 995         new->se_cb_seq_nr = 1;
 996         new->se_flags = cses->flags;
 997         new->se_cb_prog = cses->callback_prog;
 998         new->se_cb_sec = cses->cb_sec;
 999         atomic_set(&new->se_ref, 0);
1000         idx = hash_sessionid(&new->se_sessionid);
1001         spin_lock(&nn->client_lock);
1002         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1003         spin_lock(&clp->cl_lock);
1004         list_add(&new->se_perclnt, &clp->cl_sessions);
1005         spin_unlock(&clp->cl_lock);
1006         spin_unlock(&nn->client_lock);
1007         memcpy(&new->se_fchannel, &cses->fore_channel,
1008                         sizeof(struct nfsd4_channel_attrs));
1009         if (cses->flags & SESSION4_BACK_CHAN) {
1010                 struct sockaddr *sa = svc_addr(rqstp);
1011                 /*
1012                  * This is a little silly; with sessions there's no real
1013                  * use for the callback address.  Use the peer address
1014                  * as a reasonable default for now, but consider fixing
1015                  * the rpc client not to require an address in the
1016                  * future:
1017                  */
1018                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1019                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1020         }
1021 }
1022
1023 /* caller must hold client_lock */
1024 static struct nfsd4_session *
1025 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1026 {
1027         struct nfsd4_session *elem;
1028         int idx;
1029         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1030
1031         dump_sessionid(__func__, sessionid);
1032         idx = hash_sessionid(sessionid);
1033         /* Search in the appropriate list */
1034         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1035                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1036                             NFS4_MAX_SESSIONID_LEN)) {
1037                         return elem;
1038                 }
1039         }
1040
1041         dprintk("%s: session not found\n", __func__);
1042         return NULL;
1043 }
1044
1045 /* caller must hold client_lock */
1046 static void
1047 unhash_session(struct nfsd4_session *ses)
1048 {
1049         list_del(&ses->se_hash);
1050         spin_lock(&ses->se_client->cl_lock);
1051         list_del(&ses->se_perclnt);
1052         spin_unlock(&ses->se_client->cl_lock);
1053 }
1054
1055 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1056 static int
1057 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1058 {
1059         if (clid->cl_boot == nn->boot_time)
1060                 return 0;
1061         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1062                 clid->cl_boot, clid->cl_id, nn->boot_time);
1063         return 1;
1064 }
1065
1066 /*
1067  * XXX Should we use a slab cache ?
1068  * This type of memory management is somewhat inefficient, but we use it
1069  * anyway since SETCLIENTID is not a common operation.
1070  */
1071 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1072 {
1073         struct nfs4_client *clp;
1074
1075         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1076         if (clp == NULL)
1077                 return NULL;
1078         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1079         if (clp->cl_name.data == NULL) {
1080                 kfree(clp);
1081                 return NULL;
1082         }
1083         clp->cl_name.len = name.len;
1084         return clp;
1085 }
1086
1087 static inline void
1088 free_client(struct nfs4_client *clp)
1089 {
1090         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1091
1092         lockdep_assert_held(&nn->client_lock);
1093         while (!list_empty(&clp->cl_sessions)) {
1094                 struct nfsd4_session *ses;
1095                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1096                                 se_perclnt);
1097                 list_del(&ses->se_perclnt);
1098                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1099                 free_session(ses);
1100         }
1101         free_svc_cred(&clp->cl_cred);
1102         kfree(clp->cl_name.data);
1103         idr_destroy(&clp->cl_stateids);
1104         kfree(clp);
1105 }
1106
1107 /* must be called under the client_lock */
1108 static inline void
1109 unhash_client_locked(struct nfs4_client *clp)
1110 {
1111         struct nfsd4_session *ses;
1112
1113         list_del(&clp->cl_lru);
1114         spin_lock(&clp->cl_lock);
1115         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1116                 list_del_init(&ses->se_hash);
1117         spin_unlock(&clp->cl_lock);
1118 }
1119
1120 static void
1121 destroy_client(struct nfs4_client *clp)
1122 {
1123         struct nfs4_openowner *oo;
1124         struct nfs4_delegation *dp;
1125         struct list_head reaplist;
1126         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1127
1128         INIT_LIST_HEAD(&reaplist);
1129         spin_lock(&recall_lock);
1130         while (!list_empty(&clp->cl_delegations)) {
1131                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1132                 list_del_init(&dp->dl_perclnt);
1133                 list_move(&dp->dl_recall_lru, &reaplist);
1134         }
1135         spin_unlock(&recall_lock);
1136         while (!list_empty(&reaplist)) {
1137                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1138                 destroy_delegation(dp);
1139         }
1140         while (!list_empty(&clp->cl_openowners)) {
1141                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1142                 release_openowner(oo);
1143         }
1144         nfsd4_shutdown_callback(clp);
1145         if (clp->cl_cb_conn.cb_xprt)
1146                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1147         list_del(&clp->cl_idhash);
1148         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1149                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1150         else
1151                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1152         spin_lock(&nn->client_lock);
1153         unhash_client_locked(clp);
1154         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1155         free_client(clp);
1156         spin_unlock(&nn->client_lock);
1157 }
1158
1159 static void expire_client(struct nfs4_client *clp)
1160 {
1161         nfsd4_client_record_remove(clp);
1162         destroy_client(clp);
1163 }
1164
1165 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1166 {
1167         memcpy(target->cl_verifier.data, source->data,
1168                         sizeof(target->cl_verifier.data));
1169 }
1170
1171 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1172 {
1173         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1174         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1175 }
1176
1177 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1178 {
1179         if (source->cr_principal) {
1180                 target->cr_principal =
1181                                 kstrdup(source->cr_principal, GFP_KERNEL);
1182                 if (target->cr_principal == NULL)
1183                         return -ENOMEM;
1184         } else
1185                 target->cr_principal = NULL;
1186         target->cr_flavor = source->cr_flavor;
1187         target->cr_uid = source->cr_uid;
1188         target->cr_gid = source->cr_gid;
1189         target->cr_group_info = source->cr_group_info;
1190         get_group_info(target->cr_group_info);
1191         return 0;
1192 }
1193
1194 static long long
1195 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1196 {
1197         long long res;
1198
1199         res = o1->len - o2->len;
1200         if (res)
1201                 return res;
1202         return (long long)memcmp(o1->data, o2->data, o1->len);
1203 }
1204
1205 static int same_name(const char *n1, const char *n2)
1206 {
1207         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1208 }
1209
1210 static int
1211 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1212 {
1213         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1214 }
1215
1216 static int
1217 same_clid(clientid_t *cl1, clientid_t *cl2)
1218 {
1219         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1220 }
1221
1222 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1223 {
1224         int i;
1225
1226         if (g1->ngroups != g2->ngroups)
1227                 return false;
1228         for (i=0; i<g1->ngroups; i++)
1229                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1230                         return false;
1231         return true;
1232 }
1233
1234 /*
1235  * RFC 3530 language requires clid_inuse be returned when the
1236  * "principal" associated with a requests differs from that previously
1237  * used.  We use uid, gid's, and gss principal string as our best
1238  * approximation.  We also don't want to allow non-gss use of a client
1239  * established using gss: in theory cr_principal should catch that
1240  * change, but in practice cr_principal can be null even in the gss case
1241  * since gssd doesn't always pass down a principal string.
1242  */
1243 static bool is_gss_cred(struct svc_cred *cr)
1244 {
1245         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1246         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1247 }
1248
1249
1250 static bool
1251 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1252 {
1253         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1254                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1255                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1256                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1257                 return false;
1258         if (cr1->cr_principal == cr2->cr_principal)
1259                 return true;
1260         if (!cr1->cr_principal || !cr2->cr_principal)
1261                 return false;
1262         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1263 }
1264
1265 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1266 {
1267         static u32 current_clientid = 1;
1268
1269         clp->cl_clientid.cl_boot = nn->boot_time;
1270         clp->cl_clientid.cl_id = current_clientid++;
1271 }
1272
1273 static void gen_confirm(struct nfs4_client *clp)
1274 {
1275         __be32 verf[2];
1276         static u32 i;
1277
1278         verf[0] = (__be32)get_seconds();
1279         verf[1] = (__be32)i++;
1280         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1281 }
1282
1283 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1284 {
1285         struct nfs4_stid *ret;
1286
1287         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1288         if (!ret || !ret->sc_type)
1289                 return NULL;
1290         return ret;
1291 }
1292
1293 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1294 {
1295         struct nfs4_stid *s;
1296
1297         s = find_stateid(cl, t);
1298         if (!s)
1299                 return NULL;
1300         if (typemask & s->sc_type)
1301                 return s;
1302         return NULL;
1303 }
1304
1305 static struct nfs4_client *create_client(struct xdr_netobj name,
1306                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1307 {
1308         struct nfs4_client *clp;
1309         struct sockaddr *sa = svc_addr(rqstp);
1310         int ret;
1311         struct net *net = SVC_NET(rqstp);
1312         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1313
1314         clp = alloc_client(name);
1315         if (clp == NULL)
1316                 return NULL;
1317
1318         INIT_LIST_HEAD(&clp->cl_sessions);
1319         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1320         if (ret) {
1321                 spin_lock(&nn->client_lock);
1322                 free_client(clp);
1323                 spin_unlock(&nn->client_lock);
1324                 return NULL;
1325         }
1326         idr_init(&clp->cl_stateids);
1327         atomic_set(&clp->cl_refcount, 0);
1328         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1329         INIT_LIST_HEAD(&clp->cl_idhash);
1330         INIT_LIST_HEAD(&clp->cl_openowners);
1331         INIT_LIST_HEAD(&clp->cl_delegations);
1332         INIT_LIST_HEAD(&clp->cl_lru);
1333         INIT_LIST_HEAD(&clp->cl_callbacks);
1334         INIT_LIST_HEAD(&clp->cl_revoked);
1335         spin_lock_init(&clp->cl_lock);
1336         nfsd4_init_callback(&clp->cl_cb_null);
1337         clp->cl_time = get_seconds();
1338         clear_bit(0, &clp->cl_cb_slot_busy);
1339         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1340         copy_verf(clp, verf);
1341         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1342         gen_confirm(clp);
1343         clp->cl_cb_session = NULL;
1344         clp->net = net;
1345         return clp;
1346 }
1347
1348 static void
1349 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1350 {
1351         struct rb_node **new = &(root->rb_node), *parent = NULL;
1352         struct nfs4_client *clp;
1353
1354         while (*new) {
1355                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1356                 parent = *new;
1357
1358                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1359                         new = &((*new)->rb_left);
1360                 else
1361                         new = &((*new)->rb_right);
1362         }
1363
1364         rb_link_node(&new_clp->cl_namenode, parent, new);
1365         rb_insert_color(&new_clp->cl_namenode, root);
1366 }
1367
1368 static struct nfs4_client *
1369 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1370 {
1371         long long cmp;
1372         struct rb_node *node = root->rb_node;
1373         struct nfs4_client *clp;
1374
1375         while (node) {
1376                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1377                 cmp = compare_blob(&clp->cl_name, name);
1378                 if (cmp > 0)
1379                         node = node->rb_left;
1380                 else if (cmp < 0)
1381                         node = node->rb_right;
1382                 else
1383                         return clp;
1384         }
1385         return NULL;
1386 }
1387
1388 static void
1389 add_to_unconfirmed(struct nfs4_client *clp)
1390 {
1391         unsigned int idhashval;
1392         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1393
1394         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1395         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1396         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1397         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1398         renew_client(clp);
1399 }
1400
1401 static void
1402 move_to_confirmed(struct nfs4_client *clp)
1403 {
1404         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1405         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1406
1407         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1408         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1409         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1410         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1411         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1412         renew_client(clp);
1413 }
1414
1415 static struct nfs4_client *
1416 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1417 {
1418         struct nfs4_client *clp;
1419         unsigned int idhashval = clientid_hashval(clid->cl_id);
1420
1421         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1422                 if (same_clid(&clp->cl_clientid, clid)) {
1423                         if ((bool)clp->cl_minorversion != sessions)
1424                                 return NULL;
1425                         renew_client(clp);
1426                         return clp;
1427                 }
1428         }
1429         return NULL;
1430 }
1431
1432 static struct nfs4_client *
1433 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1434 {
1435         struct list_head *tbl = nn->conf_id_hashtbl;
1436
1437         return find_client_in_id_table(tbl, clid, sessions);
1438 }
1439
1440 static struct nfs4_client *
1441 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1442 {
1443         struct list_head *tbl = nn->unconf_id_hashtbl;
1444
1445         return find_client_in_id_table(tbl, clid, sessions);
1446 }
1447
1448 static bool clp_used_exchangeid(struct nfs4_client *clp)
1449 {
1450         return clp->cl_exchange_flags != 0;
1451 }
1452
1453 static struct nfs4_client *
1454 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1455 {
1456         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1457 }
1458
1459 static struct nfs4_client *
1460 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1461 {
1462         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1463 }
1464
1465 static void
1466 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1467 {
1468         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1469         struct sockaddr *sa = svc_addr(rqstp);
1470         u32 scopeid = rpc_get_scope_id(sa);
1471         unsigned short expected_family;
1472
1473         /* Currently, we only support tcp and tcp6 for the callback channel */
1474         if (se->se_callback_netid_len == 3 &&
1475             !memcmp(se->se_callback_netid_val, "tcp", 3))
1476                 expected_family = AF_INET;
1477         else if (se->se_callback_netid_len == 4 &&
1478                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1479                 expected_family = AF_INET6;
1480         else
1481                 goto out_err;
1482
1483         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1484                                             se->se_callback_addr_len,
1485                                             (struct sockaddr *)&conn->cb_addr,
1486                                             sizeof(conn->cb_addr));
1487
1488         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1489                 goto out_err;
1490
1491         if (conn->cb_addr.ss_family == AF_INET6)
1492                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1493
1494         conn->cb_prog = se->se_callback_prog;
1495         conn->cb_ident = se->se_callback_ident;
1496         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1497         return;
1498 out_err:
1499         conn->cb_addr.ss_family = AF_UNSPEC;
1500         conn->cb_addrlen = 0;
1501         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1502                 "will not receive delegations\n",
1503                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1504
1505         return;
1506 }
1507
1508 /*
1509  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1510  */
1511 void
1512 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1513 {
1514         struct nfsd4_slot *slot = resp->cstate.slot;
1515         unsigned int base;
1516
1517         dprintk("--> %s slot %p\n", __func__, slot);
1518
1519         slot->sl_opcnt = resp->opcnt;
1520         slot->sl_status = resp->cstate.status;
1521
1522         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1523         if (nfsd4_not_cached(resp)) {
1524                 slot->sl_datalen = 0;
1525                 return;
1526         }
1527         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1528         base = (char *)resp->cstate.datap -
1529                                         (char *)resp->xbuf->head[0].iov_base;
1530         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1531                                     slot->sl_datalen))
1532                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1533         return;
1534 }
1535
1536 /*
1537  * Encode the replay sequence operation from the slot values.
1538  * If cachethis is FALSE encode the uncached rep error on the next
1539  * operation which sets resp->p and increments resp->opcnt for
1540  * nfs4svc_encode_compoundres.
1541  *
1542  */
1543 static __be32
1544 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1545                           struct nfsd4_compoundres *resp)
1546 {
1547         struct nfsd4_op *op;
1548         struct nfsd4_slot *slot = resp->cstate.slot;
1549
1550         /* Encode the replayed sequence operation */
1551         op = &args->ops[resp->opcnt - 1];
1552         nfsd4_encode_operation(resp, op);
1553
1554         /* Return nfserr_retry_uncached_rep in next operation. */
1555         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1556                 op = &args->ops[resp->opcnt++];
1557                 op->status = nfserr_retry_uncached_rep;
1558                 nfsd4_encode_operation(resp, op);
1559         }
1560         return op->status;
1561 }
1562
1563 /*
1564  * The sequence operation is not cached because we can use the slot and
1565  * session values.
1566  */
1567 __be32
1568 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1569                          struct nfsd4_sequence *seq)
1570 {
1571         struct nfsd4_slot *slot = resp->cstate.slot;
1572         __be32 status;
1573
1574         dprintk("--> %s slot %p\n", __func__, slot);
1575
1576         /* Either returns 0 or nfserr_retry_uncached */
1577         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1578         if (status == nfserr_retry_uncached_rep)
1579                 return status;
1580
1581         /* The sequence operation has been encoded, cstate->datap set. */
1582         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1583
1584         resp->opcnt = slot->sl_opcnt;
1585         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1586         status = slot->sl_status;
1587
1588         return status;
1589 }
1590
1591 /*
1592  * Set the exchange_id flags returned by the server.
1593  */
1594 static void
1595 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1596 {
1597         /* pNFS is not supported */
1598         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1599
1600         /* Referrals are supported, Migration is not. */
1601         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1602
1603         /* set the wire flags to return to client. */
1604         clid->flags = new->cl_exchange_flags;
1605 }
1606
1607 static bool client_has_state(struct nfs4_client *clp)
1608 {
1609         /*
1610          * Note clp->cl_openowners check isn't quite right: there's no
1611          * need to count owners without stateid's.
1612          *
1613          * Also note we should probably be using this in 4.0 case too.
1614          */
1615         return !list_empty(&clp->cl_openowners)
1616                 || !list_empty(&clp->cl_delegations)
1617                 || !list_empty(&clp->cl_sessions);
1618 }
1619
1620 __be32
1621 nfsd4_exchange_id(struct svc_rqst *rqstp,
1622                   struct nfsd4_compound_state *cstate,
1623                   struct nfsd4_exchange_id *exid)
1624 {
1625         struct nfs4_client *unconf, *conf, *new;
1626         __be32 status;
1627         char                    addr_str[INET6_ADDRSTRLEN];
1628         nfs4_verifier           verf = exid->verifier;
1629         struct sockaddr         *sa = svc_addr(rqstp);
1630         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1631         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1632
1633         rpc_ntop(sa, addr_str, sizeof(addr_str));
1634         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1635                 "ip_addr=%s flags %x, spa_how %d\n",
1636                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1637                 addr_str, exid->flags, exid->spa_how);
1638
1639         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1640                 return nfserr_inval;
1641
1642         /* Currently only support SP4_NONE */
1643         switch (exid->spa_how) {
1644         case SP4_NONE:
1645                 break;
1646         default:                                /* checked by xdr code */
1647                 WARN_ON_ONCE(1);
1648         case SP4_SSV:
1649                 return nfserr_encr_alg_unsupp;
1650         case SP4_MACH_CRED:
1651                 return nfserr_serverfault;      /* no excuse :-/ */
1652         }
1653
1654         /* Cases below refer to rfc 5661 section 18.35.4: */
1655         nfs4_lock_state();
1656         conf = find_confirmed_client_by_name(&exid->clname, nn);
1657         if (conf) {
1658                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1659                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1660
1661                 if (update) {
1662                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1663                                 status = nfserr_inval;
1664                                 goto out;
1665                         }
1666                         if (!creds_match) { /* case 9 */
1667                                 status = nfserr_perm;
1668                                 goto out;
1669                         }
1670                         if (!verfs_match) { /* case 8 */
1671                                 status = nfserr_not_same;
1672                                 goto out;
1673                         }
1674                         /* case 6 */
1675                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1676                         new = conf;
1677                         goto out_copy;
1678                 }
1679                 if (!creds_match) { /* case 3 */
1680                         if (client_has_state(conf)) {
1681                                 status = nfserr_clid_inuse;
1682                                 goto out;
1683                         }
1684                         expire_client(conf);
1685                         goto out_new;
1686                 }
1687                 if (verfs_match) { /* case 2 */
1688                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1689                         new = conf;
1690                         goto out_copy;
1691                 }
1692                 /* case 5, client reboot */
1693                 goto out_new;
1694         }
1695
1696         if (update) { /* case 7 */
1697                 status = nfserr_noent;
1698                 goto out;
1699         }
1700
1701         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1702         if (unconf) /* case 4, possible retry or client restart */
1703                 expire_client(unconf);
1704
1705         /* case 1 (normal case) */
1706 out_new:
1707         new = create_client(exid->clname, rqstp, &verf);
1708         if (new == NULL) {
1709                 status = nfserr_jukebox;
1710                 goto out;
1711         }
1712         new->cl_minorversion = 1;
1713
1714         gen_clid(new, nn);
1715         add_to_unconfirmed(new);
1716 out_copy:
1717         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1718         exid->clientid.cl_id = new->cl_clientid.cl_id;
1719
1720         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1721         nfsd4_set_ex_flags(new, exid);
1722
1723         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1724                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1725         status = nfs_ok;
1726
1727 out:
1728         nfs4_unlock_state();
1729         return status;
1730 }
1731
1732 static __be32
1733 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1734 {
1735         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1736                 slot_seqid);
1737
1738         /* The slot is in use, and no response has been sent. */
1739         if (slot_inuse) {
1740                 if (seqid == slot_seqid)
1741                         return nfserr_jukebox;
1742                 else
1743                         return nfserr_seq_misordered;
1744         }
1745         /* Note unsigned 32-bit arithmetic handles wraparound: */
1746         if (likely(seqid == slot_seqid + 1))
1747                 return nfs_ok;
1748         if (seqid == slot_seqid)
1749                 return nfserr_replay_cache;
1750         return nfserr_seq_misordered;
1751 }
1752
1753 /*
1754  * Cache the create session result into the create session single DRC
1755  * slot cache by saving the xdr structure. sl_seqid has been set.
1756  * Do this for solo or embedded create session operations.
1757  */
1758 static void
1759 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1760                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1761 {
1762         slot->sl_status = nfserr;
1763         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1764 }
1765
1766 static __be32
1767 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1768                             struct nfsd4_clid_slot *slot)
1769 {
1770         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1771         return slot->sl_status;
1772 }
1773
1774 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1775                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1776                         1 +     /* MIN tag is length with zero, only length */ \
1777                         3 +     /* version, opcount, opcode */ \
1778                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1779                                 /* seqid, slotID, slotID, cache */ \
1780                         4 ) * sizeof(__be32))
1781
1782 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1783                         2 +     /* verifier: AUTH_NULL, length 0 */\
1784                         1 +     /* status */ \
1785                         1 +     /* MIN tag is length with zero, only length */ \
1786                         3 +     /* opcount, opcode, opstatus*/ \
1787                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1788                                 /* seqid, slotID, slotID, slotID, status */ \
1789                         5 ) * sizeof(__be32))
1790
1791 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1792 {
1793         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
1794
1795         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
1796                 return nfserr_toosmall;
1797         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
1798                 return nfserr_toosmall;
1799         ca->headerpadsz = 0;
1800         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
1801         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
1802         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
1803         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
1804                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
1805         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
1806         /*
1807          * Note decreasing slot size below client's request may make it
1808          * difficult for client to function correctly, whereas
1809          * decreasing the number of slots will (just?) affect
1810          * performance.  When short on memory we therefore prefer to
1811          * decrease number of slots instead of their size.  Clients that
1812          * request larger slots than they need will get poor results:
1813          */
1814         ca->maxreqs = nfsd4_get_drc_mem(ca);
1815         if (!ca->maxreqs)
1816                 return nfserr_jukebox;
1817
1818         return nfs_ok;
1819 }
1820
1821 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
1822 {
1823         ca->headerpadsz = 0;
1824
1825         /*
1826          * These RPC_MAX_HEADER macros are overkill, especially since we
1827          * don't even do gss on the backchannel yet.  But this is still
1828          * less than 1k.  Tighten up this estimate in the unlikely event
1829          * it turns out to be a problem for some client:
1830          */
1831         if (ca->maxreq_sz < NFS4_enc_cb_recall_sz + RPC_MAX_HEADER_WITH_AUTH)
1832                 return nfserr_toosmall;
1833         if (ca->maxresp_sz < NFS4_dec_cb_recall_sz + RPC_MAX_REPHEADER_WITH_AUTH)
1834                 return nfserr_toosmall;
1835         ca->maxresp_cached = 0;
1836         if (ca->maxops < 2)
1837                 return nfserr_toosmall;
1838
1839         return nfs_ok;
1840 }
1841
1842 __be32
1843 nfsd4_create_session(struct svc_rqst *rqstp,
1844                      struct nfsd4_compound_state *cstate,
1845                      struct nfsd4_create_session *cr_ses)
1846 {
1847         struct sockaddr *sa = svc_addr(rqstp);
1848         struct nfs4_client *conf, *unconf;
1849         struct nfsd4_session *new;
1850         struct nfsd4_conn *conn;
1851         struct nfsd4_clid_slot *cs_slot = NULL;
1852         __be32 status = 0;
1853         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1854
1855         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1856                 return nfserr_inval;
1857         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
1858         if (status)
1859                 return status;
1860         status = check_backchannel_attrs(&cr_ses->back_channel);
1861         if (status)
1862                 return status;
1863         status = nfserr_jukebox;
1864         new = alloc_session(&cr_ses->fore_channel);
1865         if (!new)
1866                 goto out_release_drc_mem;
1867         conn = alloc_conn_from_crses(rqstp, cr_ses);
1868         if (!conn)
1869                 goto out_free_session;
1870
1871         nfs4_lock_state();
1872         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1873         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1874         WARN_ON_ONCE(conf && unconf);
1875
1876         if (conf) {
1877                 cs_slot = &conf->cl_cs_slot;
1878                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1879                 if (status == nfserr_replay_cache) {
1880                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1881                         goto out_free_conn;
1882                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1883                         status = nfserr_seq_misordered;
1884                         goto out_free_conn;
1885                 }
1886         } else if (unconf) {
1887                 struct nfs4_client *old;
1888                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1889                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1890                         status = nfserr_clid_inuse;
1891                         goto out_free_conn;
1892                 }
1893                 cs_slot = &unconf->cl_cs_slot;
1894                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1895                 if (status) {
1896                         /* an unconfirmed replay returns misordered */
1897                         status = nfserr_seq_misordered;
1898                         goto out_free_conn;
1899                 }
1900                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1901                 if (old) {
1902                         status = mark_client_expired(old);
1903                         if (status)
1904                                 goto out_free_conn;
1905                         expire_client(old);
1906                 }
1907                 move_to_confirmed(unconf);
1908                 conf = unconf;
1909         } else {
1910                 status = nfserr_stale_clientid;
1911                 goto out_free_conn;
1912         }
1913         status = nfs_ok;
1914         /*
1915          * We do not support RDMA or persistent sessions
1916          */
1917         cr_ses->flags &= ~SESSION4_PERSIST;
1918         cr_ses->flags &= ~SESSION4_RDMA;
1919
1920         init_session(rqstp, new, conf, cr_ses);
1921         nfsd4_init_conn(rqstp, conn, new);
1922
1923         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1924                NFS4_MAX_SESSIONID_LEN);
1925         cs_slot->sl_seqid++;
1926         cr_ses->seqid = cs_slot->sl_seqid;
1927
1928         /* cache solo and embedded create sessions under the state lock */
1929         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1930         nfs4_unlock_state();
1931         return status;
1932 out_free_conn:
1933         nfs4_unlock_state();
1934         free_conn(conn);
1935 out_free_session:
1936         __free_session(new);
1937 out_release_drc_mem:
1938         nfsd4_put_drc_mem(&cr_ses->fore_channel);
1939         return status;
1940 }
1941
1942 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1943 {
1944         switch (*dir) {
1945         case NFS4_CDFC4_FORE:
1946         case NFS4_CDFC4_BACK:
1947                 return nfs_ok;
1948         case NFS4_CDFC4_FORE_OR_BOTH:
1949         case NFS4_CDFC4_BACK_OR_BOTH:
1950                 *dir = NFS4_CDFC4_BOTH;
1951                 return nfs_ok;
1952         };
1953         return nfserr_inval;
1954 }
1955
1956 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1957 {
1958         struct nfsd4_session *session = cstate->session;
1959         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1960
1961         spin_lock(&nn->client_lock);
1962         session->se_cb_prog = bc->bc_cb_program;
1963         session->se_cb_sec = bc->bc_cb_sec;
1964         spin_unlock(&nn->client_lock);
1965
1966         nfsd4_probe_callback(session->se_client);
1967
1968         return nfs_ok;
1969 }
1970
1971 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1972                      struct nfsd4_compound_state *cstate,
1973                      struct nfsd4_bind_conn_to_session *bcts)
1974 {
1975         __be32 status;
1976         struct nfsd4_conn *conn;
1977         struct nfsd4_session *session;
1978         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1979
1980         if (!nfsd4_last_compound_op(rqstp))
1981                 return nfserr_not_only_op;
1982         nfs4_lock_state();
1983         spin_lock(&nn->client_lock);
1984         session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
1985         spin_unlock(&nn->client_lock);
1986         status = nfserr_badsession;
1987         if (!session)
1988                 goto out;
1989         status = nfsd4_map_bcts_dir(&bcts->dir);
1990         if (status)
1991                 goto out;
1992         conn = alloc_conn(rqstp, bcts->dir);
1993         status = nfserr_jukebox;
1994         if (!conn)
1995                 goto out;
1996         nfsd4_init_conn(rqstp, conn, session);
1997         status = nfs_ok;
1998 out:
1999         nfs4_unlock_state();
2000         return status;
2001 }
2002
2003 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2004 {
2005         if (!session)
2006                 return 0;
2007         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2008 }
2009
2010 __be32
2011 nfsd4_destroy_session(struct svc_rqst *r,
2012                       struct nfsd4_compound_state *cstate,
2013                       struct nfsd4_destroy_session *sessionid)
2014 {
2015         struct nfsd4_session *ses;
2016         __be32 status;
2017         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
2018
2019         nfs4_lock_state();
2020         status = nfserr_not_only_op;
2021         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2022                 if (!nfsd4_last_compound_op(r))
2023                         goto out;
2024         }
2025         dump_sessionid(__func__, &sessionid->sessionid);
2026         spin_lock(&nn->client_lock);
2027         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
2028         status = nfserr_badsession;
2029         if (!ses)
2030                 goto out_client_lock;
2031         status = mark_session_dead_locked(ses);
2032         if (status)
2033                 goto out_client_lock;
2034         unhash_session(ses);
2035         spin_unlock(&nn->client_lock);
2036
2037         nfsd4_probe_callback_sync(ses->se_client);
2038
2039         spin_lock(&nn->client_lock);
2040         free_session(ses);
2041         status = nfs_ok;
2042 out_client_lock:
2043         spin_unlock(&nn->client_lock);
2044 out:
2045         nfs4_unlock_state();
2046         return status;
2047 }
2048
2049 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2050 {
2051         struct nfsd4_conn *c;
2052
2053         list_for_each_entry(c, &s->se_conns, cn_persession) {
2054                 if (c->cn_xprt == xpt) {
2055                         return c;
2056                 }
2057         }
2058         return NULL;
2059 }
2060
2061 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2062 {
2063         struct nfs4_client *clp = ses->se_client;
2064         struct nfsd4_conn *c;
2065         int ret;
2066
2067         spin_lock(&clp->cl_lock);
2068         c = __nfsd4_find_conn(new->cn_xprt, ses);
2069         if (c) {
2070                 spin_unlock(&clp->cl_lock);
2071                 free_conn(new);
2072                 return;
2073         }
2074         __nfsd4_hash_conn(new, ses);
2075         spin_unlock(&clp->cl_lock);
2076         ret = nfsd4_register_conn(new);
2077         if (ret)
2078                 /* oops; xprt is already down: */
2079                 nfsd4_conn_lost(&new->cn_xpt_user);
2080         return;
2081 }
2082
2083 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2084 {
2085         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2086
2087         return args->opcnt > session->se_fchannel.maxops;
2088 }
2089
2090 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2091                                   struct nfsd4_session *session)
2092 {
2093         struct xdr_buf *xb = &rqstp->rq_arg;
2094
2095         return xb->len > session->se_fchannel.maxreq_sz;
2096 }
2097
2098 __be32
2099 nfsd4_sequence(struct svc_rqst *rqstp,
2100                struct nfsd4_compound_state *cstate,
2101                struct nfsd4_sequence *seq)
2102 {
2103         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2104         struct nfsd4_session *session;
2105         struct nfs4_client *clp;
2106         struct nfsd4_slot *slot;
2107         struct nfsd4_conn *conn;
2108         __be32 status;
2109         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2110
2111         if (resp->opcnt != 1)
2112                 return nfserr_sequence_pos;
2113
2114         /*
2115          * Will be either used or freed by nfsd4_sequence_check_conn
2116          * below.
2117          */
2118         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2119         if (!conn)
2120                 return nfserr_jukebox;
2121
2122         spin_lock(&nn->client_lock);
2123         status = nfserr_badsession;
2124         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2125         if (!session)
2126                 goto out_no_session;
2127         clp = session->se_client;
2128         status = get_client_locked(clp);
2129         if (status)
2130                 goto out_no_session;
2131         status = nfsd4_get_session_locked(session);
2132         if (status)
2133                 goto out_put_client;
2134
2135         status = nfserr_too_many_ops;
2136         if (nfsd4_session_too_many_ops(rqstp, session))
2137                 goto out_put_session;
2138
2139         status = nfserr_req_too_big;
2140         if (nfsd4_request_too_big(rqstp, session))
2141                 goto out_put_session;
2142
2143         status = nfserr_badslot;
2144         if (seq->slotid >= session->se_fchannel.maxreqs)
2145                 goto out_put_session;
2146
2147         slot = session->se_slots[seq->slotid];
2148         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2149
2150         /* We do not negotiate the number of slots yet, so set the
2151          * maxslots to the session maxreqs which is used to encode
2152          * sr_highest_slotid and the sr_target_slot id to maxslots */
2153         seq->maxslots = session->se_fchannel.maxreqs;
2154
2155         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2156                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2157         if (status == nfserr_replay_cache) {
2158                 status = nfserr_seq_misordered;
2159                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2160                         goto out_put_session;
2161                 cstate->slot = slot;
2162                 cstate->session = session;
2163                 /* Return the cached reply status and set cstate->status
2164                  * for nfsd4_proc_compound processing */
2165                 status = nfsd4_replay_cache_entry(resp, seq);
2166                 cstate->status = nfserr_replay_cache;
2167                 goto out;
2168         }
2169         if (status)
2170                 goto out_put_session;
2171
2172         nfsd4_sequence_check_conn(conn, session);
2173         conn = NULL;
2174
2175         /* Success! bump slot seqid */
2176         slot->sl_seqid = seq->seqid;
2177         slot->sl_flags |= NFSD4_SLOT_INUSE;
2178         if (seq->cachethis)
2179                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2180         else
2181                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2182
2183         cstate->slot = slot;
2184         cstate->session = session;
2185
2186 out:
2187         switch (clp->cl_cb_state) {
2188         case NFSD4_CB_DOWN:
2189                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2190                 break;
2191         case NFSD4_CB_FAULT:
2192                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2193                 break;
2194         default:
2195                 seq->status_flags = 0;
2196         }
2197         if (!list_empty(&clp->cl_revoked))
2198                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2199 out_no_session:
2200         kfree(conn);
2201         spin_unlock(&nn->client_lock);
2202         return status;
2203 out_put_session:
2204         nfsd4_put_session(session);
2205 out_put_client:
2206         put_client_renew_locked(clp);
2207         goto out_no_session;
2208 }
2209
2210 __be32
2211 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2212 {
2213         struct nfs4_client *conf, *unconf, *clp;
2214         __be32 status = 0;
2215         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2216
2217         nfs4_lock_state();
2218         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2219         conf = find_confirmed_client(&dc->clientid, true, nn);
2220         WARN_ON_ONCE(conf && unconf);
2221
2222         if (conf) {
2223                 clp = conf;
2224
2225                 if (client_has_state(conf)) {
2226                         status = nfserr_clientid_busy;
2227                         goto out;
2228                 }
2229         } else if (unconf)
2230                 clp = unconf;
2231         else {
2232                 status = nfserr_stale_clientid;
2233                 goto out;
2234         }
2235
2236         expire_client(clp);
2237 out:
2238         nfs4_unlock_state();
2239         return status;
2240 }
2241
2242 __be32
2243 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2244 {
2245         __be32 status = 0;
2246
2247         if (rc->rca_one_fs) {
2248                 if (!cstate->current_fh.fh_dentry)
2249                         return nfserr_nofilehandle;
2250                 /*
2251                  * We don't take advantage of the rca_one_fs case.
2252                  * That's OK, it's optional, we can safely ignore it.
2253                  */
2254                  return nfs_ok;
2255         }
2256
2257         nfs4_lock_state();
2258         status = nfserr_complete_already;
2259         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2260                              &cstate->session->se_client->cl_flags))
2261                 goto out;
2262
2263         status = nfserr_stale_clientid;
2264         if (is_client_expired(cstate->session->se_client))
2265                 /*
2266                  * The following error isn't really legal.
2267                  * But we only get here if the client just explicitly
2268                  * destroyed the client.  Surely it no longer cares what
2269                  * error it gets back on an operation for the dead
2270                  * client.
2271                  */
2272                 goto out;
2273
2274         status = nfs_ok;
2275         nfsd4_client_record_create(cstate->session->se_client);
2276 out:
2277         nfs4_unlock_state();
2278         return status;
2279 }
2280
2281 __be32
2282 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2283                   struct nfsd4_setclientid *setclid)
2284 {
2285         struct xdr_netobj       clname = setclid->se_name;
2286         nfs4_verifier           clverifier = setclid->se_verf;
2287         struct nfs4_client      *conf, *unconf, *new;
2288         __be32                  status;
2289         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2290
2291         /* Cases below refer to rfc 3530 section 14.2.33: */
2292         nfs4_lock_state();
2293         conf = find_confirmed_client_by_name(&clname, nn);
2294         if (conf) {
2295                 /* case 0: */
2296                 status = nfserr_clid_inuse;
2297                 if (clp_used_exchangeid(conf))
2298                         goto out;
2299                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2300                         char addr_str[INET6_ADDRSTRLEN];
2301                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2302                                  sizeof(addr_str));
2303                         dprintk("NFSD: setclientid: string in use by client "
2304                                 "at %s\n", addr_str);
2305                         goto out;
2306                 }
2307         }
2308         unconf = find_unconfirmed_client_by_name(&clname, nn);
2309         if (unconf)
2310                 expire_client(unconf);
2311         status = nfserr_jukebox;
2312         new = create_client(clname, rqstp, &clverifier);
2313         if (new == NULL)
2314                 goto out;
2315         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2316                 /* case 1: probable callback update */
2317                 copy_clid(new, conf);
2318         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2319                 gen_clid(new, nn);
2320         new->cl_minorversion = 0;
2321         gen_callback(new, setclid, rqstp);
2322         add_to_unconfirmed(new);
2323         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2324         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2325         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2326         status = nfs_ok;
2327 out:
2328         nfs4_unlock_state();
2329         return status;
2330 }
2331
2332
2333 __be32
2334 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2335                          struct nfsd4_compound_state *cstate,
2336                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2337 {
2338         struct nfs4_client *conf, *unconf;
2339         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2340         clientid_t * clid = &setclientid_confirm->sc_clientid;
2341         __be32 status;
2342         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2343
2344         if (STALE_CLIENTID(clid, nn))
2345                 return nfserr_stale_clientid;
2346         nfs4_lock_state();
2347
2348         conf = find_confirmed_client(clid, false, nn);
2349         unconf = find_unconfirmed_client(clid, false, nn);
2350         /*
2351          * We try hard to give out unique clientid's, so if we get an
2352          * attempt to confirm the same clientid with a different cred,
2353          * there's a bug somewhere.  Let's charitably assume it's our
2354          * bug.
2355          */
2356         status = nfserr_serverfault;
2357         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2358                 goto out;
2359         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2360                 goto out;
2361         /* cases below refer to rfc 3530 section 14.2.34: */
2362         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2363                 if (conf && !unconf) /* case 2: probable retransmit */
2364                         status = nfs_ok;
2365                 else /* case 4: client hasn't noticed we rebooted yet? */
2366                         status = nfserr_stale_clientid;
2367                 goto out;
2368         }
2369         status = nfs_ok;
2370         if (conf) { /* case 1: callback update */
2371                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2372                 nfsd4_probe_callback(conf);
2373                 expire_client(unconf);
2374         } else { /* case 3: normal case; new or rebooted client */
2375                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2376                 if (conf) {
2377                         status = mark_client_expired(conf);
2378                         if (status)
2379                                 goto out;
2380                         expire_client(conf);
2381                 }
2382                 move_to_confirmed(unconf);
2383                 nfsd4_probe_callback(unconf);
2384         }
2385 out:
2386         nfs4_unlock_state();
2387         return status;
2388 }
2389
2390 static struct nfs4_file *nfsd4_alloc_file(void)
2391 {
2392         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2393 }
2394
2395 /* OPEN Share state helper functions */
2396 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2397 {
2398         unsigned int hashval = file_hashval(ino);
2399
2400         atomic_set(&fp->fi_ref, 1);
2401         INIT_LIST_HEAD(&fp->fi_stateids);
2402         INIT_LIST_HEAD(&fp->fi_delegations);
2403         fp->fi_inode = igrab(ino);
2404         fp->fi_had_conflict = false;
2405         fp->fi_lease = NULL;
2406         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2407         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2408         spin_lock(&recall_lock);
2409         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2410         spin_unlock(&recall_lock);
2411 }
2412
2413 static void
2414 nfsd4_free_slab(struct kmem_cache **slab)
2415 {
2416         if (*slab == NULL)
2417                 return;
2418         kmem_cache_destroy(*slab);
2419         *slab = NULL;
2420 }
2421
2422 void
2423 nfsd4_free_slabs(void)
2424 {
2425         nfsd4_free_slab(&openowner_slab);
2426         nfsd4_free_slab(&lockowner_slab);
2427         nfsd4_free_slab(&file_slab);
2428         nfsd4_free_slab(&stateid_slab);
2429         nfsd4_free_slab(&deleg_slab);
2430 }
2431
2432 int
2433 nfsd4_init_slabs(void)
2434 {
2435         openowner_slab = kmem_cache_create("nfsd4_openowners",
2436                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2437         if (openowner_slab == NULL)
2438                 goto out_nomem;
2439         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2440                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2441         if (lockowner_slab == NULL)
2442                 goto out_nomem;
2443         file_slab = kmem_cache_create("nfsd4_files",
2444                         sizeof(struct nfs4_file), 0, 0, NULL);
2445         if (file_slab == NULL)
2446                 goto out_nomem;
2447         stateid_slab = kmem_cache_create("nfsd4_stateids",
2448                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2449         if (stateid_slab == NULL)
2450                 goto out_nomem;
2451         deleg_slab = kmem_cache_create("nfsd4_delegations",
2452                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2453         if (deleg_slab == NULL)
2454                 goto out_nomem;
2455         return 0;
2456 out_nomem:
2457         nfsd4_free_slabs();
2458         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2459         return -ENOMEM;
2460 }
2461
2462 void nfs4_free_openowner(struct nfs4_openowner *oo)
2463 {
2464         kfree(oo->oo_owner.so_owner.data);
2465         kmem_cache_free(openowner_slab, oo);
2466 }
2467
2468 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2469 {
2470         kfree(lo->lo_owner.so_owner.data);
2471         kmem_cache_free(lockowner_slab, lo);
2472 }
2473
2474 static void init_nfs4_replay(struct nfs4_replay *rp)
2475 {
2476         rp->rp_status = nfserr_serverfault;
2477         rp->rp_buflen = 0;
2478         rp->rp_buf = rp->rp_ibuf;
2479 }
2480
2481 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2482 {
2483         struct nfs4_stateowner *sop;
2484
2485         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2486         if (!sop)
2487                 return NULL;
2488
2489         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2490         if (!sop->so_owner.data) {
2491                 kmem_cache_free(slab, sop);
2492                 return NULL;
2493         }
2494         sop->so_owner.len = owner->len;
2495
2496         INIT_LIST_HEAD(&sop->so_stateids);
2497         sop->so_client = clp;
2498         init_nfs4_replay(&sop->so_replay);
2499         return sop;
2500 }
2501
2502 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2503 {
2504         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2505
2506         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2507         list_add(&oo->oo_perclient, &clp->cl_openowners);
2508 }
2509
2510 static struct nfs4_openowner *
2511 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2512         struct nfs4_openowner *oo;
2513
2514         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2515         if (!oo)
2516                 return NULL;
2517         oo->oo_owner.so_is_open_owner = 1;
2518         oo->oo_owner.so_seqid = open->op_seqid;
2519         oo->oo_flags = NFS4_OO_NEW;
2520         oo->oo_time = 0;
2521         oo->oo_last_closed_stid = NULL;
2522         INIT_LIST_HEAD(&oo->oo_close_lru);
2523         hash_openowner(oo, clp, strhashval);
2524         return oo;
2525 }
2526
2527 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2528         struct nfs4_openowner *oo = open->op_openowner;
2529
2530         stp->st_stid.sc_type = NFS4_OPEN_STID;
2531         INIT_LIST_HEAD(&stp->st_lockowners);
2532         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2533         list_add(&stp->st_perfile, &fp->fi_stateids);
2534         stp->st_stateowner = &oo->oo_owner;
2535         get_nfs4_file(fp);
2536         stp->st_file = fp;
2537         stp->st_access_bmap = 0;
2538         stp->st_deny_bmap = 0;
2539         set_access(open->op_share_access, stp);
2540         set_deny(open->op_share_deny, stp);
2541         stp->st_openstp = NULL;
2542 }
2543
2544 static void
2545 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2546 {
2547         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2548
2549         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2550
2551         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2552         oo->oo_time = get_seconds();
2553 }
2554
2555 static int
2556 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2557                                                         clientid_t *clid)
2558 {
2559         return (sop->so_owner.len == owner->len) &&
2560                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2561                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2562 }
2563
2564 static struct nfs4_openowner *
2565 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2566                         bool sessions, struct nfsd_net *nn)
2567 {
2568         struct nfs4_stateowner *so;
2569         struct nfs4_openowner *oo;
2570         struct nfs4_client *clp;
2571
2572         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2573                 if (!so->so_is_open_owner)
2574                         continue;
2575                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2576                         oo = openowner(so);
2577                         clp = oo->oo_owner.so_client;
2578                         if ((bool)clp->cl_minorversion != sessions)
2579                                 return NULL;
2580                         renew_client(oo->oo_owner.so_client);
2581                         return oo;
2582                 }
2583         }
2584         return NULL;
2585 }
2586
2587 /* search file_hashtbl[] for file */
2588 static struct nfs4_file *
2589 find_file(struct inode *ino)
2590 {
2591         unsigned int hashval = file_hashval(ino);
2592         struct nfs4_file *fp;
2593
2594         spin_lock(&recall_lock);
2595         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2596                 if (fp->fi_inode == ino) {
2597                         get_nfs4_file(fp);
2598                         spin_unlock(&recall_lock);
2599                         return fp;
2600                 }
2601         }
2602         spin_unlock(&recall_lock);
2603         return NULL;
2604 }
2605
2606 /*
2607  * Called to check deny when READ with all zero stateid or
2608  * WRITE with all zero or all one stateid
2609  */
2610 static __be32
2611 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2612 {
2613         struct inode *ino = current_fh->fh_dentry->d_inode;
2614         struct nfs4_file *fp;
2615         struct nfs4_ol_stateid *stp;
2616         __be32 ret;
2617
2618         fp = find_file(ino);
2619         if (!fp)
2620                 return nfs_ok;
2621         ret = nfserr_locked;
2622         /* Search for conflicting share reservations */
2623         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2624                 if (test_deny(deny_type, stp) ||
2625                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2626                         goto out;
2627         }
2628         ret = nfs_ok;
2629 out:
2630         put_nfs4_file(fp);
2631         return ret;
2632 }
2633
2634 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2635 {
2636         struct nfs4_client *clp = dp->dl_stid.sc_client;
2637         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2638
2639         /* We're assuming the state code never drops its reference
2640          * without first removing the lease.  Since we're in this lease
2641          * callback (and since the lease code is serialized by the kernel
2642          * lock) we know the server hasn't removed the lease yet, we know
2643          * it's safe to take a reference: */
2644         atomic_inc(&dp->dl_count);
2645
2646         list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
2647
2648         /* only place dl_time is set. protected by lock_flocks*/
2649         dp->dl_time = get_seconds();
2650
2651         nfsd4_cb_recall(dp);
2652 }
2653
2654 /* Called from break_lease() with lock_flocks() held. */
2655 static void nfsd_break_deleg_cb(struct file_lock *fl)
2656 {
2657         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2658         struct nfs4_delegation *dp;
2659
2660         if (!fp) {
2661                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2662                 return;
2663         }
2664         if (fp->fi_had_conflict) {
2665                 WARN(1, "duplicate break on %p\n", fp);
2666                 return;
2667         }
2668         /*
2669          * We don't want the locks code to timeout the lease for us;
2670          * we'll remove it ourself if a delegation isn't returned
2671          * in time:
2672          */
2673         fl->fl_break_time = 0;
2674
2675         spin_lock(&recall_lock);
2676         fp->fi_had_conflict = true;
2677         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2678                 nfsd_break_one_deleg(dp);
2679         spin_unlock(&recall_lock);
2680 }
2681
2682 static
2683 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2684 {
2685         if (arg & F_UNLCK)
2686                 return lease_modify(onlist, arg);
2687         else
2688                 return -EAGAIN;
2689 }
2690
2691 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2692         .lm_break = nfsd_break_deleg_cb,
2693         .lm_change = nfsd_change_deleg_cb,
2694 };
2695
2696 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2697 {
2698         if (nfsd4_has_session(cstate))
2699                 return nfs_ok;
2700         if (seqid == so->so_seqid - 1)
2701                 return nfserr_replay_me;
2702         if (seqid == so->so_seqid)
2703                 return nfs_ok;
2704         return nfserr_bad_seqid;
2705 }
2706
2707 __be32
2708 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2709                     struct nfsd4_open *open, struct nfsd_net *nn)
2710 {
2711         clientid_t *clientid = &open->op_clientid;
2712         struct nfs4_client *clp = NULL;
2713         unsigned int strhashval;
2714         struct nfs4_openowner *oo = NULL;
2715         __be32 status;
2716
2717         if (STALE_CLIENTID(&open->op_clientid, nn))
2718                 return nfserr_stale_clientid;
2719         /*
2720          * In case we need it later, after we've already created the
2721          * file and don't want to risk a further failure:
2722          */
2723         open->op_file = nfsd4_alloc_file();
2724         if (open->op_file == NULL)
2725                 return nfserr_jukebox;
2726
2727         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2728         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2729         open->op_openowner = oo;
2730         if (!oo) {
2731                 clp = find_confirmed_client(clientid, cstate->minorversion,
2732                                             nn);
2733                 if (clp == NULL)
2734                         return nfserr_expired;
2735                 goto new_owner;
2736         }
2737         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2738                 /* Replace unconfirmed owners without checking for replay. */
2739                 clp = oo->oo_owner.so_client;
2740                 release_openowner(oo);
2741                 open->op_openowner = NULL;
2742                 goto new_owner;
2743         }
2744         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2745         if (status)
2746                 return status;
2747         clp = oo->oo_owner.so_client;
2748         goto alloc_stateid;
2749 new_owner:
2750         oo = alloc_init_open_stateowner(strhashval, clp, open);
2751         if (oo == NULL)
2752                 return nfserr_jukebox;
2753         open->op_openowner = oo;
2754 alloc_stateid:
2755         open->op_stp = nfs4_alloc_stateid(clp);
2756         if (!open->op_stp)
2757                 return nfserr_jukebox;
2758         return nfs_ok;
2759 }
2760
2761 static inline __be32
2762 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2763 {
2764         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2765                 return nfserr_openmode;
2766         else
2767                 return nfs_ok;
2768 }
2769
2770 static int share_access_to_flags(u32 share_access)
2771 {
2772         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2773 }
2774
2775 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2776 {
2777         struct nfs4_stid *ret;
2778
2779         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2780         if (!ret)
2781                 return NULL;
2782         return delegstateid(ret);
2783 }
2784
2785 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2786 {
2787         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2788                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2789 }
2790
2791 static __be32
2792 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
2793                 struct nfs4_delegation **dp)
2794 {
2795         int flags;
2796         __be32 status = nfserr_bad_stateid;
2797
2798         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2799         if (*dp == NULL)
2800                 goto out;
2801         flags = share_access_to_flags(open->op_share_access);
2802         status = nfs4_check_delegmode(*dp, flags);
2803         if (status)
2804                 *dp = NULL;
2805 out:
2806         if (!nfsd4_is_deleg_cur(open))
2807                 return nfs_ok;
2808         if (status)
2809                 return status;
2810         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2811         return nfs_ok;
2812 }
2813
2814 static __be32
2815 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2816 {
2817         struct nfs4_ol_stateid *local;
2818         struct nfs4_openowner *oo = open->op_openowner;
2819
2820         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2821                 /* ignore lock owners */
2822                 if (local->st_stateowner->so_is_open_owner == 0)
2823                         continue;
2824                 /* remember if we have seen this open owner */
2825                 if (local->st_stateowner == &oo->oo_owner)
2826                         *stpp = local;
2827                 /* check for conflicting share reservations */
2828                 if (!test_share(local, open))
2829                         return nfserr_share_denied;
2830         }
2831         return nfs_ok;
2832 }
2833
2834 static inline int nfs4_access_to_access(u32 nfs4_access)
2835 {
2836         int flags = 0;
2837
2838         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2839                 flags |= NFSD_MAY_READ;
2840         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2841                 flags |= NFSD_MAY_WRITE;
2842         return flags;
2843 }
2844
2845 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2846                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2847 {
2848         __be32 status;
2849         int oflag = nfs4_access_to_omode(open->op_share_access);
2850         int access = nfs4_access_to_access(open->op_share_access);
2851
2852         if (!fp->fi_fds[oflag]) {
2853                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2854                         &fp->fi_fds[oflag]);
2855                 if (status)
2856                         return status;
2857         }
2858         nfs4_file_get_access(fp, oflag);
2859
2860         return nfs_ok;
2861 }
2862
2863 static inline __be32
2864 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2865                 struct nfsd4_open *open)
2866 {
2867         struct iattr iattr = {
2868                 .ia_valid = ATTR_SIZE,
2869                 .ia_size = 0,
2870         };
2871         if (!open->op_truncate)
2872                 return 0;
2873         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2874                 return nfserr_inval;
2875         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2876 }
2877
2878 static __be32
2879 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)
2880 {
2881         u32 op_share_access = open->op_share_access;
2882         bool new_access;
2883         __be32 status;
2884
2885         new_access = !test_access(op_share_access, stp);
2886         if (new_access) {
2887                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2888                 if (status)
2889                         return status;
2890         }
2891         status = nfsd4_truncate(rqstp, cur_fh, open);
2892         if (status) {
2893                 if (new_access) {
2894                         int oflag = nfs4_access_to_omode(op_share_access);
2895                         nfs4_file_put_access(fp, oflag);
2896                 }
2897                 return status;
2898         }
2899         /* remember the open */
2900         set_access(op_share_access, stp);
2901         set_deny(open->op_share_deny, stp);
2902
2903         return nfs_ok;
2904 }
2905
2906
2907 static void
2908 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2909 {
2910         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2911 }
2912
2913 /* Should we give out recallable state?: */
2914 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2915 {
2916         if (clp->cl_cb_state == NFSD4_CB_UP)
2917                 return true;
2918         /*
2919          * In the sessions case, since we don't have to establish a
2920          * separate connection for callbacks, we assume it's OK
2921          * until we hear otherwise:
2922          */
2923         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2924 }
2925
2926 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2927 {
2928         struct file_lock *fl;
2929
2930         fl = locks_alloc_lock();
2931         if (!fl)
2932                 return NULL;
2933         locks_init_lock(fl);
2934         fl->fl_lmops = &nfsd_lease_mng_ops;
2935         fl->fl_flags = FL_LEASE;
2936         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2937         fl->fl_end = OFFSET_MAX;
2938         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2939         fl->fl_pid = current->tgid;
2940         return fl;
2941 }
2942
2943 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2944 {
2945         struct nfs4_file *fp = dp->dl_file;
2946         struct file_lock *fl;
2947         int status;
2948
2949         fl = nfs4_alloc_init_lease(dp, flag);
2950         if (!fl)
2951                 return -ENOMEM;
2952         fl->fl_file = find_readable_file(fp);
2953         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2954         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2955         if (status) {
2956                 list_del_init(&dp->dl_perclnt);
2957                 locks_free_lock(fl);
2958                 return -ENOMEM;
2959         }
2960         fp->fi_lease = fl;
2961         fp->fi_deleg_file = get_file(fl->fl_file);
2962         atomic_set(&fp->fi_delegees, 1);
2963         list_add(&dp->dl_perfile, &fp->fi_delegations);
2964         return 0;
2965 }
2966
2967 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2968 {
2969         struct nfs4_file *fp = dp->dl_file;
2970
2971         if (!fp->fi_lease)
2972                 return nfs4_setlease(dp, flag);
2973         spin_lock(&recall_lock);
2974         if (fp->fi_had_conflict) {
2975                 spin_unlock(&recall_lock);
2976                 return -EAGAIN;
2977         }
2978         atomic_inc(&fp->fi_delegees);
2979         list_add(&dp->dl_perfile, &fp->fi_delegations);
2980         spin_unlock(&recall_lock);
2981         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2982         return 0;
2983 }
2984
2985 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2986 {
2987         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2988         if (status == -EAGAIN)
2989                 open->op_why_no_deleg = WND4_CONTENTION;
2990         else {
2991                 open->op_why_no_deleg = WND4_RESOURCE;
2992                 switch (open->op_deleg_want) {
2993                 case NFS4_SHARE_WANT_READ_DELEG:
2994                 case NFS4_SHARE_WANT_WRITE_DELEG:
2995                 case NFS4_SHARE_WANT_ANY_DELEG:
2996                         break;
2997                 case NFS4_SHARE_WANT_CANCEL:
2998                         open->op_why_no_deleg = WND4_CANCELLED;
2999                         break;
3000                 case NFS4_SHARE_WANT_NO_DELEG:
3001                         WARN_ON_ONCE(1);
3002                 }
3003         }
3004 }
3005
3006 /*
3007  * Attempt to hand out a delegation.
3008  */
3009 static void
3010 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
3011                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
3012 {
3013         struct nfs4_delegation *dp;
3014         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
3015         int cb_up;
3016         int status = 0, flag = 0;
3017
3018         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3019         flag = NFS4_OPEN_DELEGATE_NONE;
3020         open->op_recall = 0;
3021         switch (open->op_claim_type) {
3022                 case NFS4_OPEN_CLAIM_PREVIOUS:
3023                         if (!cb_up)
3024                                 open->op_recall = 1;
3025                         flag = open->op_delegate_type;
3026                         if (flag == NFS4_OPEN_DELEGATE_NONE)
3027                                 goto out;
3028                         break;
3029                 case NFS4_OPEN_CLAIM_NULL:
3030                         /* Let's not give out any delegations till everyone's
3031                          * had the chance to reclaim theirs.... */
3032                         if (locks_in_grace(net))
3033                                 goto out;
3034                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3035                                 goto out;
3036                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3037                                 flag = NFS4_OPEN_DELEGATE_WRITE;
3038                         else
3039                                 flag = NFS4_OPEN_DELEGATE_READ;
3040                         break;
3041                 default:
3042                         goto out;
3043         }
3044
3045         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
3046         if (dp == NULL)
3047                 goto out_no_deleg;
3048         status = nfs4_set_delegation(dp, flag);
3049         if (status)
3050                 goto out_free;
3051
3052         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3053
3054         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3055                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3056 out:
3057         open->op_delegate_type = flag;
3058         if (flag == NFS4_OPEN_DELEGATE_NONE) {
3059                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3060                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
3061                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3062
3063                 /* 4.1 client asking for a delegation? */
3064                 if (open->op_deleg_want)
3065                         nfsd4_open_deleg_none_ext(open, status);
3066         }
3067         return;
3068 out_free:
3069         unhash_stid(&dp->dl_stid);
3070         nfs4_put_delegation(dp);
3071 out_no_deleg:
3072         flag = NFS4_OPEN_DELEGATE_NONE;
3073         goto out;
3074 }
3075
3076 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3077                                         struct nfs4_delegation *dp)
3078 {
3079         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3080             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3081                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3082                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3083         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3084                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3085                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3086                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3087         }
3088         /* Otherwise the client must be confused wanting a delegation
3089          * it already has, therefore we don't return
3090          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3091          */
3092 }
3093
3094 /*
3095  * called with nfs4_lock_state() held.
3096  */
3097 __be32
3098 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3099 {
3100         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3101         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3102         struct nfs4_file *fp = NULL;
3103         struct inode *ino = current_fh->fh_dentry->d_inode;
3104         struct nfs4_ol_stateid *stp = NULL;
3105         struct nfs4_delegation *dp = NULL;
3106         __be32 status;
3107
3108         /*
3109          * Lookup file; if found, lookup stateid and check open request,
3110          * and check for delegations in the process of being recalled.
3111          * If not found, create the nfs4_file struct
3112          */
3113         fp = find_file(ino);
3114         if (fp) {
3115                 if ((status = nfs4_check_open(fp, open, &stp)))
3116                         goto out;
3117                 status = nfs4_check_deleg(cl, open, &dp);
3118                 if (status)
3119                         goto out;
3120         } else {
3121                 status = nfserr_bad_stateid;
3122                 if (nfsd4_is_deleg_cur(open))
3123                         goto out;
3124                 status = nfserr_jukebox;
3125                 fp = open->op_file;
3126                 open->op_file = NULL;
3127                 nfsd4_init_file(fp, ino);
3128         }
3129
3130         /*
3131          * OPEN the file, or upgrade an existing OPEN.
3132          * If truncate fails, the OPEN fails.
3133          */
3134         if (stp) {
3135                 /* Stateid was found, this is an OPEN upgrade */
3136                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3137                 if (status)
3138                         goto out;
3139         } else {
3140                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3141                 if (status)
3142                         goto out;
3143                 status = nfsd4_truncate(rqstp, current_fh, open);
3144                 if (status)
3145                         goto out;
3146                 stp = open->op_stp;
3147                 open->op_stp = NULL;
3148                 init_open_stateid(stp, fp, open);
3149         }
3150         update_stateid(&stp->st_stid.sc_stateid);
3151         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3152
3153         if (nfsd4_has_session(&resp->cstate)) {
3154                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3155
3156                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3157                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3158                         open->op_why_no_deleg = WND4_NOT_WANTED;
3159                         goto nodeleg;
3160                 }
3161         }
3162
3163         /*
3164         * Attempt to hand out a delegation. No error return, because the
3165         * OPEN succeeds even if we fail.
3166         */
3167         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3168 nodeleg:
3169         status = nfs_ok;
3170
3171         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3172                 STATEID_VAL(&stp->st_stid.sc_stateid));
3173 out:
3174         /* 4.1 client trying to upgrade/downgrade delegation? */
3175         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3176             open->op_deleg_want)
3177                 nfsd4_deleg_xgrade_none_ext(open, dp);
3178
3179         if (fp)
3180                 put_nfs4_file(fp);
3181         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3182                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3183         /*
3184         * To finish the open response, we just need to set the rflags.
3185         */
3186         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3187         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3188             !nfsd4_has_session(&resp->cstate))
3189                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3190
3191         return status;
3192 }
3193
3194 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3195 {
3196         if (open->op_openowner) {
3197                 struct nfs4_openowner *oo = open->op_openowner;
3198
3199                 if (!list_empty(&oo->oo_owner.so_stateids))
3200                         list_del_init(&oo->oo_close_lru);
3201                 if (oo->oo_flags & NFS4_OO_NEW) {
3202                         if (status) {
3203                                 release_openowner(oo);
3204                                 open->op_openowner = NULL;
3205                         } else
3206                                 oo->oo_flags &= ~NFS4_OO_NEW;
3207                 }
3208         }
3209         if (open->op_file)
3210                 nfsd4_free_file(open->op_file);
3211         if (open->op_stp)
3212                 free_generic_stateid(open->op_stp);
3213 }
3214
3215 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3216 {
3217         struct nfs4_client *found;
3218
3219         if (STALE_CLIENTID(clid, nn))
3220                 return nfserr_stale_clientid;
3221         found = find_confirmed_client(clid, session, nn);
3222         if (clp)
3223                 *clp = found;
3224         return found ? nfs_ok : nfserr_expired;
3225 }
3226
3227 __be32
3228 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3229             clientid_t *clid)
3230 {
3231         struct nfs4_client *clp;
3232         __be32 status;
3233         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3234
3235         nfs4_lock_state();
3236         dprintk("process_renew(%08x/%08x): starting\n",
3237                         clid->cl_boot, clid->cl_id);
3238         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3239         if (status)
3240                 goto out;
3241         status = nfserr_cb_path_down;
3242         if (!list_empty(&clp->cl_delegations)
3243                         && clp->cl_cb_state != NFSD4_CB_UP)
3244                 goto out;
3245         status = nfs_ok;
3246 out:
3247         nfs4_unlock_state();
3248         return status;
3249 }
3250
3251 static void
3252 nfsd4_end_grace(struct nfsd_net *nn)
3253 {
3254         /* do nothing if grace period already ended */
3255         if (nn->grace_ended)
3256                 return;
3257
3258         dprintk("NFSD: end of grace period\n");
3259         nn->grace_ended = true;
3260         nfsd4_record_grace_done(nn, nn->boot_time);
3261         locks_end_grace(&nn->nfsd4_manager);
3262         /*
3263          * Now that every NFSv4 client has had the chance to recover and
3264          * to see the (possibly new, possibly shorter) lease time, we
3265          * can safely set the next grace time to the current lease time:
3266          */
3267         nn->nfsd4_grace = nn->nfsd4_lease;
3268 }
3269
3270 static time_t
3271 nfs4_laundromat(struct nfsd_net *nn)
3272 {
3273         struct nfs4_client *clp;
3274         struct nfs4_openowner *oo;
3275         struct nfs4_delegation *dp;
3276         struct list_head *pos, *next, reaplist;
3277         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3278         time_t t, clientid_val = nn->nfsd4_lease;
3279         time_t u, test_val = nn->nfsd4_lease;
3280
3281         nfs4_lock_state();
3282
3283         dprintk("NFSD: laundromat service - starting\n");
3284         nfsd4_end_grace(nn);
3285         INIT_LIST_HEAD(&reaplist);
3286         spin_lock(&nn->client_lock);
3287         list_for_each_safe(pos, next, &nn->client_lru) {
3288                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3289                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3290                         t = clp->cl_time - cutoff;
3291                         if (clientid_val > t)
3292                                 clientid_val = t;
3293                         break;
3294                 }
3295                 if (mark_client_expired_locked(clp)) {
3296                         dprintk("NFSD: client in use (clientid %08x)\n",
3297                                 clp->cl_clientid.cl_id);
3298                         continue;
3299                 }
3300                 list_move(&clp->cl_lru, &reaplist);
3301         }
3302         spin_unlock(&nn->client_lock);
3303         list_for_each_safe(pos, next, &reaplist) {
3304                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3305                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3306                         clp->cl_clientid.cl_id);
3307                 expire_client(clp);
3308         }
3309         spin_lock(&recall_lock);
3310         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3311                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3312                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3313                         continue;
3314                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3315                         u = dp->dl_time - cutoff;
3316                         if (test_val > u)
3317                                 test_val = u;
3318                         break;
3319                 }
3320                 list_move(&dp->dl_recall_lru, &reaplist);
3321         }
3322         spin_unlock(&recall_lock);
3323         list_for_each_safe(pos, next, &reaplist) {
3324                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3325                 revoke_delegation(dp);
3326         }
3327         test_val = nn->nfsd4_lease;
3328         list_for_each_safe(pos, next, &nn->close_lru) {
3329                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3330                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3331                         u = oo->oo_time - cutoff;
3332                         if (test_val > u)
3333                                 test_val = u;
3334                         break;
3335                 }
3336                 release_openowner(oo);
3337         }
3338         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3339                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3340         nfs4_unlock_state();
3341         return clientid_val;
3342 }
3343
3344 static struct workqueue_struct *laundry_wq;
3345 static void laundromat_main(struct work_struct *);
3346
3347 static void
3348 laundromat_main(struct work_struct *laundry)
3349 {
3350         time_t t;
3351         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3352                                                   work);
3353         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3354                                            laundromat_work);
3355
3356         t = nfs4_laundromat(nn);
3357         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3358         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3359 }
3360
3361 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3362 {
3363         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3364                 return nfserr_bad_stateid;
3365         return nfs_ok;
3366 }
3367
3368 static inline int
3369 access_permit_read(struct nfs4_ol_stateid *stp)
3370 {
3371         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3372                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3373                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3374 }
3375
3376 static inline int
3377 access_permit_write(struct nfs4_ol_stateid *stp)
3378 {
3379         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3380                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3381 }
3382
3383 static
3384 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3385 {
3386         __be32 status = nfserr_openmode;
3387
3388         /* For lock stateid's, we test the parent open, not the lock: */
3389         if (stp->st_openstp)
3390                 stp = stp->st_openstp;
3391         if ((flags & WR_STATE) && !access_permit_write(stp))
3392                 goto out;
3393         if ((flags & RD_STATE) && !access_permit_read(stp))
3394                 goto out;
3395         status = nfs_ok;
3396 out:
3397         return status;
3398 }
3399
3400 static inline __be32
3401 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3402 {
3403         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3404                 return nfs_ok;
3405         else if (locks_in_grace(net)) {
3406                 /* Answer in remaining cases depends on existence of
3407                  * conflicting state; so we must wait out the grace period. */
3408                 return nfserr_grace;
3409         } else if (flags & WR_STATE)
3410                 return nfs4_share_conflict(current_fh,
3411                                 NFS4_SHARE_DENY_WRITE);
3412         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3413                 return nfs4_share_conflict(current_fh,
3414                                 NFS4_SHARE_DENY_READ);
3415 }
3416
3417 /*
3418  * Allow READ/WRITE during grace period on recovered state only for files
3419  * that are not able to provide mandatory locking.
3420  */
3421 static inline int
3422 grace_disallows_io(struct net *net, struct inode *inode)
3423 {
3424         return locks_in_grace(net) && mandatory_lock(inode);
3425 }
3426
3427 /* Returns true iff a is later than b: */
3428 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3429 {
3430         return (s32)a->si_generation - (s32)b->si_generation > 0;
3431 }
3432
3433 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3434 {
3435         /*
3436          * When sessions are used the stateid generation number is ignored
3437          * when it is zero.
3438          */
3439         if (has_session && in->si_generation == 0)
3440                 return nfs_ok;
3441
3442         if (in->si_generation == ref->si_generation)
3443                 return nfs_ok;
3444
3445         /* If the client sends us a stateid from the future, it's buggy: */
3446         if (stateid_generation_after(in, ref))
3447                 return nfserr_bad_stateid;
3448         /*
3449          * However, we could see a stateid from the past, even from a
3450          * non-buggy client.  For example, if the client sends a lock
3451          * while some IO is outstanding, the lock may bump si_generation
3452          * while the IO is still in flight.  The client could avoid that
3453          * situation by waiting for responses on all the IO requests,
3454          * but better performance may result in retrying IO that
3455          * receives an old_stateid error if requests are rarely
3456          * reordered in flight:
3457          */
3458         return nfserr_old_stateid;
3459 }
3460
3461 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3462 {
3463         struct nfs4_stid *s;
3464         struct nfs4_ol_stateid *ols;
3465         __be32 status;
3466
3467         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3468                 return nfserr_bad_stateid;
3469         /* Client debugging aid. */
3470         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3471                 char addr_str[INET6_ADDRSTRLEN];
3472                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3473                                  sizeof(addr_str));
3474                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3475                                         "with incorrect client ID\n", addr_str);
3476                 return nfserr_bad_stateid;
3477         }
3478         s = find_stateid(cl, stateid);
3479         if (!s)
3480                 return nfserr_bad_stateid;
3481         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3482         if (status)
3483                 return status;
3484         switch (s->sc_type) {
3485         case NFS4_DELEG_STID:
3486                 return nfs_ok;
3487         case NFS4_REVOKED_DELEG_STID:
3488                 return nfserr_deleg_revoked;
3489         case NFS4_OPEN_STID:
3490         case NFS4_LOCK_STID:
3491                 ols = openlockstateid(s);
3492                 if (ols->st_stateowner->so_is_open_owner
3493                                 && !(openowner(ols->st_stateowner)->oo_flags
3494                                                 & NFS4_OO_CONFIRMED))
3495                         return nfserr_bad_stateid;
3496                 return nfs_ok;
3497         default:
3498                 printk("unknown stateid type %x\n", s->sc_type);
3499         case NFS4_CLOSED_STID:
3500                 return nfserr_bad_stateid;
3501         }
3502 }
3503
3504 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3505                                    struct nfs4_stid **s, bool sessions,
3506                                    struct nfsd_net *nn)
3507 {
3508         struct nfs4_client *cl;
3509         __be32 status;
3510
3511         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3512                 return nfserr_bad_stateid;
3513         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3514                                                         nn, &cl);
3515         if (status == nfserr_stale_clientid)
3516                 return nfserr_stale_stateid;
3517         if (status)
3518                 return status;
3519         *s = find_stateid_by_type(cl, stateid, typemask);
3520         if (!*s)
3521                 return nfserr_bad_stateid;
3522         return nfs_ok;
3523 }
3524
3525 /*
3526 * Checks for stateid operations
3527 */
3528 __be32
3529 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3530                            stateid_t *stateid, int flags, struct file **filpp)
3531 {
3532         struct nfs4_stid *s;
3533         struct nfs4_ol_stateid *stp = NULL;
3534         struct nfs4_delegation *dp = NULL;
3535         struct svc_fh *current_fh = &cstate->current_fh;
3536         struct inode *ino = current_fh->fh_dentry->d_inode;
3537         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3538         __be32 status;
3539
3540         if (filpp)
3541                 *filpp = NULL;
3542
3543         if (grace_disallows_io(net, ino))
3544                 return nfserr_grace;
3545
3546         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3547                 return check_special_stateids(net, current_fh, stateid, flags);
3548
3549         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3550                                       &s, cstate->minorversion, nn);
3551         if (status)
3552                 return status;
3553         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3554         if (status)
3555                 goto out;
3556         switch (s->sc_type) {
3557         case NFS4_DELEG_STID:
3558                 dp = delegstateid(s);
3559                 status = nfs4_check_delegmode(dp, flags);
3560                 if (status)
3561                         goto out;
3562                 if (filpp) {
3563                         *filpp = dp->dl_file->fi_deleg_file;
3564                         if (!*filpp) {
3565                                 WARN_ON_ONCE(1);
3566                                 status = nfserr_serverfault;
3567                                 goto out;
3568                         }
3569                 }
3570                 break;
3571         case NFS4_OPEN_STID:
3572         case NFS4_LOCK_STID:
3573                 stp = openlockstateid(s);
3574                 status = nfs4_check_fh(current_fh, stp);
3575                 if (status)
3576                         goto out;
3577                 if (stp->st_stateowner->so_is_open_owner
3578                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3579                         goto out;
3580                 status = nfs4_check_openmode(stp, flags);
3581                 if (status)
3582                         goto out;
3583                 if (filpp) {
3584                         if (flags & RD_STATE)
3585                                 *filpp = find_readable_file(stp->st_file);
3586                         else
3587                                 *filpp = find_writeable_file(stp->st_file);
3588                 }
3589                 break;
3590         default:
3591                 return nfserr_bad_stateid;
3592         }
3593         status = nfs_ok;
3594 out:
3595         return status;
3596 }
3597
3598 static __be32
3599 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3600 {
3601         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3602                 return nfserr_locks_held;
3603         release_lock_stateid(stp);
3604         return nfs_ok;
3605 }
3606
3607 /*
3608  * Test if the stateid is valid
3609  */
3610 __be32
3611 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3612                    struct nfsd4_test_stateid *test_stateid)
3613 {
3614         struct nfsd4_test_stateid_id *stateid;
3615         struct nfs4_client *cl = cstate->session->se_client;
3616
3617         nfs4_lock_state();
3618         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3619                 stateid->ts_id_status =
3620                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3621         nfs4_unlock_state();
3622
3623         return nfs_ok;
3624 }
3625
3626 __be32
3627 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3628                    struct nfsd4_free_stateid *free_stateid)
3629 {
3630         stateid_t *stateid = &free_stateid->fr_stateid;
3631         struct nfs4_stid *s;
3632         struct nfs4_delegation *dp;
3633         struct nfs4_client *cl = cstate->session->se_client;
3634         __be32 ret = nfserr_bad_stateid;
3635
3636         nfs4_lock_state();
3637         s = find_stateid(cl, stateid);
3638         if (!s)
3639                 goto out;
3640         switch (s->sc_type) {
3641         case NFS4_DELEG_STID:
3642                 ret = nfserr_locks_held;
3643                 goto out;
3644         case NFS4_OPEN_STID:
3645         case NFS4_LOCK_STID:
3646                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3647                 if (ret)
3648                         goto out;
3649                 if (s->sc_type == NFS4_LOCK_STID)
3650                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3651                 else
3652                         ret = nfserr_locks_held;
3653                 break;
3654         case NFS4_REVOKED_DELEG_STID:
3655                 dp = delegstateid(s);
3656                 destroy_revoked_delegation(dp);
3657                 ret = nfs_ok;
3658                 break;
3659         default:
3660                 ret = nfserr_bad_stateid;
3661         }
3662 out:
3663         nfs4_unlock_state();
3664         return ret;
3665 }
3666
3667 static inline int
3668 setlkflg (int type)
3669 {
3670         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3671                 RD_STATE : WR_STATE;
3672 }
3673
3674 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3675 {
3676         struct svc_fh *current_fh = &cstate->current_fh;
3677         struct nfs4_stateowner *sop = stp->st_stateowner;
3678         __be32 status;
3679
3680         status = nfsd4_check_seqid(cstate, sop, seqid);
3681         if (status)
3682                 return status;
3683         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
3684                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
3685                 /*
3686                  * "Closed" stateid's exist *only* to return
3687                  * nfserr_replay_me from the previous step, and
3688                  * revoked delegations are kept only for free_stateid.
3689                  */
3690                 return nfserr_bad_stateid;
3691         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3692         if (status)
3693                 return status;
3694         return nfs4_check_fh(current_fh, stp);
3695 }
3696
3697 /*
3698  * Checks for sequence id mutating operations.
3699  */
3700 static __be32
3701 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3702                          stateid_t *stateid, char typemask,
3703                          struct nfs4_ol_stateid **stpp,
3704                          struct nfsd_net *nn)
3705 {
3706         __be32 status;
3707         struct nfs4_stid *s;
3708
3709         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3710                 seqid, STATEID_VAL(stateid));
3711
3712         *stpp = NULL;
3713         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3714                                       cstate->minorversion, nn);
3715         if (status)
3716                 return status;
3717         *stpp = openlockstateid(s);
3718         if (!nfsd4_has_session(cstate))
3719                 cstate->replay_owner = (*stpp)->st_stateowner;
3720
3721         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3722 }
3723
3724 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3725                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3726 {
3727         __be32 status;
3728         struct nfs4_openowner *oo;
3729
3730         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3731                                                 NFS4_OPEN_STID, stpp, nn);
3732         if (status)
3733                 return status;
3734         oo = openowner((*stpp)->st_stateowner);
3735         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3736                 return nfserr_bad_stateid;
3737         return nfs_ok;
3738 }
3739
3740 __be32
3741 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3742                    struct nfsd4_open_confirm *oc)
3743 {
3744         __be32 status;
3745         struct nfs4_openowner *oo;
3746         struct nfs4_ol_stateid *stp;
3747         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3748
3749         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3750                         (int)cstate->current_fh.fh_dentry->d_name.len,
3751                         cstate->current_fh.fh_dentry->d_name.name);
3752
3753         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3754         if (status)
3755                 return status;
3756
3757         nfs4_lock_state();
3758
3759         status = nfs4_preprocess_seqid_op(cstate,
3760                                         oc->oc_seqid, &oc->oc_req_stateid,
3761                                         NFS4_OPEN_STID, &stp, nn);
3762         if (status)
3763                 goto out;
3764         oo = openowner(stp->st_stateowner);
3765         status = nfserr_bad_stateid;
3766         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3767                 goto out;
3768         oo->oo_flags |= NFS4_OO_CONFIRMED;
3769         update_stateid(&stp->st_stid.sc_stateid);
3770         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3771         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3772                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3773
3774         nfsd4_client_record_create(oo->oo_owner.so_client);
3775         status = nfs_ok;
3776 out:
3777         nfsd4_bump_seqid(cstate, status);
3778         if (!cstate->replay_owner)
3779                 nfs4_unlock_state();
3780         return status;
3781 }
3782
3783 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3784 {
3785         if (!test_access(access, stp))
3786                 return;
3787         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3788         clear_access(access, stp);
3789 }
3790
3791 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3792 {
3793         switch (to_access) {
3794         case NFS4_SHARE_ACCESS_READ:
3795                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3796                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3797                 break;
3798         case NFS4_SHARE_ACCESS_WRITE:
3799                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3800                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3801                 break;
3802         case NFS4_SHARE_ACCESS_BOTH:
3803                 break;
3804         default:
3805                 WARN_ON_ONCE(1);
3806         }
3807 }
3808
3809 static void
3810 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3811 {
3812         int i;
3813         for (i = 0; i < 4; i++) {
3814                 if ((i & deny) != i)
3815                         clear_deny(i, stp);
3816         }
3817 }
3818
3819 __be32
3820 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3821                      struct nfsd4_compound_state *cstate,
3822                      struct nfsd4_open_downgrade *od)
3823 {
3824         __be32 status;
3825         struct nfs4_ol_stateid *stp;
3826         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3827
3828         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3829                         (int)cstate->current_fh.fh_dentry->d_name.len,
3830                         cstate->current_fh.fh_dentry->d_name.name);
3831
3832         /* We don't yet support WANT bits: */
3833         if (od->od_deleg_want)
3834                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3835                         od->od_deleg_want);
3836
3837         nfs4_lock_state();
3838         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3839                                         &od->od_stateid, &stp, nn);
3840         if (status)
3841                 goto out;
3842         status = nfserr_inval;
3843         if (!test_access(od->od_share_access, stp)) {
3844                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3845                         stp->st_access_bmap, od->od_share_access);
3846                 goto out;
3847         }
3848         if (!test_deny(od->od_share_deny, stp)) {
3849                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3850                         stp->st_deny_bmap, od->od_share_deny);
3851                 goto out;
3852         }
3853         nfs4_stateid_downgrade(stp, od->od_share_access);
3854
3855         reset_union_bmap_deny(od->od_share_deny, stp);
3856
3857         update_stateid(&stp->st_stid.sc_stateid);
3858         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3859         status = nfs_ok;
3860 out:
3861         nfsd4_bump_seqid(cstate, status);
3862         if (!cstate->replay_owner)
3863                 nfs4_unlock_state();
3864         return status;
3865 }
3866
3867 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3868 {
3869         unhash_open_stateid(s);
3870         s->st_stid.sc_type = NFS4_CLOSED_STID;
3871 }
3872
3873 /*
3874  * nfs4_unlock_state() called after encode
3875  */
3876 __be32
3877 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3878             struct nfsd4_close *close)
3879 {
3880         __be32 status;
3881         struct nfs4_openowner *oo;
3882         struct nfs4_ol_stateid *stp;
3883         struct net *net = SVC_NET(rqstp);
3884         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3885
3886         dprintk("NFSD: nfsd4_close on file %.*s\n",
3887                         (int)cstate->current_fh.fh_dentry->d_name.len,
3888                         cstate->current_fh.fh_dentry->d_name.name);
3889
3890         nfs4_lock_state();
3891         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3892                                         &close->cl_stateid,
3893                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3894                                         &stp, nn);
3895         nfsd4_bump_seqid(cstate, status);
3896         if (status)
3897                 goto out;
3898         oo = openowner(stp->st_stateowner);
3899         update_stateid(&stp->st_stid.sc_stateid);
3900         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3901
3902         nfsd4_close_open_stateid(stp);
3903
3904         if (cstate->minorversion) {
3905                 unhash_stid(&stp->st_stid);
3906                 free_generic_stateid(stp);
3907         } else
3908                 oo->oo_last_closed_stid = stp;
3909
3910         if (list_empty(&oo->oo_owner.so_stateids)) {
3911                 if (cstate->minorversion)
3912                         release_openowner(oo);
3913                 else {
3914                         /*
3915                          * In the 4.0 case we need to keep the owners around a
3916                          * little while to handle CLOSE replay.
3917                          */
3918                         move_to_close_lru(oo, SVC_NET(rqstp));
3919                 }
3920         }
3921 out:
3922         if (!cstate->replay_owner)
3923                 nfs4_unlock_state();
3924         return status;
3925 }
3926
3927 __be32
3928 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3929                   struct nfsd4_delegreturn *dr)
3930 {
3931         struct nfs4_delegation *dp;
3932         stateid_t *stateid = &dr->dr_stateid;
3933         struct nfs4_stid *s;
3934         __be32 status;
3935         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3936
3937         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3938                 return status;
3939
3940         nfs4_lock_state();
3941         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
3942                                       cstate->minorversion, nn);
3943         if (status)
3944                 goto out;
3945         dp = delegstateid(s);
3946         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3947         if (status)
3948                 goto out;
3949
3950         destroy_delegation(dp);
3951 out:
3952         nfs4_unlock_state();
3953
3954         return status;
3955 }
3956
3957
3958 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3959
3960 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3961
3962 static inline u64
3963 end_offset(u64 start, u64 len)
3964 {
3965         u64 end;
3966
3967         end = start + len;
3968         return end >= start ? end: NFS4_MAX_UINT64;
3969 }
3970
3971 /* last octet in a range */
3972 static inline u64
3973 last_byte_offset(u64 start, u64 len)
3974 {
3975         u64 end;
3976
3977         WARN_ON_ONCE(!len);
3978         end = start + len;
3979         return end > start ? end - 1: NFS4_MAX_UINT64;
3980 }
3981
3982 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3983 {
3984         return (file_hashval(inode) + cl_id
3985                         + opaque_hashval(ownername->data, ownername->len))
3986                 & LOCKOWNER_INO_HASH_MASK;
3987 }
3988
3989 /*
3990  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3991  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3992  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3993  * locking, this prevents us from being completely protocol-compliant.  The
3994  * real solution to this problem is to start using unsigned file offsets in
3995  * the VFS, but this is a very deep change!
3996  */
3997 static inline void
3998 nfs4_transform_lock_offset(struct file_lock *lock)
3999 {
4000         if (lock->fl_start < 0)
4001                 lock->fl_start = OFFSET_MAX;
4002         if (lock->fl_end < 0)
4003                 lock->fl_end = OFFSET_MAX;
4004 }
4005
4006 /* Hack!: For now, we're defining this just so we can use a pointer to it
4007  * as a unique cookie to identify our (NFSv4's) posix locks. */
4008 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4009 };
4010
4011 static inline void
4012 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4013 {
4014         struct nfs4_lockowner *lo;
4015
4016         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4017                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4018                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4019                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4020                 if (!deny->ld_owner.data)
4021                         /* We just don't care that much */
4022                         goto nevermind;
4023                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4024                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4025         } else {
4026 nevermind:
4027                 deny->ld_owner.len = 0;
4028                 deny->ld_owner.data = NULL;
4029                 deny->ld_clientid.cl_boot = 0;
4030                 deny->ld_clientid.cl_id = 0;
4031         }
4032         deny->ld_start = fl->fl_start;
4033         deny->ld_length = NFS4_MAX_UINT64;
4034         if (fl->fl_end != NFS4_MAX_UINT64)
4035                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4036         deny->ld_type = NFS4_READ_LT;
4037         if (fl->fl_type != F_RDLCK)
4038                 deny->ld_type = NFS4_WRITE_LT;
4039 }
4040
4041 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
4042 {
4043         struct nfs4_ol_stateid *lst;
4044
4045         if (!same_owner_str(&lo->lo_owner, owner, clid))
4046                 return false;
4047         lst = list_first_entry(&lo->lo_owner.so_stateids,
4048                                struct nfs4_ol_stateid, st_perstateowner);
4049         return lst->st_file->fi_inode == inode;
4050 }
4051
4052 static struct nfs4_lockowner *
4053 find_lockowner_str(struct inode *inode, clientid_t *clid,
4054                    struct xdr_netobj *owner, struct nfsd_net *nn)
4055 {
4056         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4057         struct nfs4_lockowner *lo;
4058
4059         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4060                 if (same_lockowner_ino(lo, inode, clid, owner))
4061                         return lo;
4062         }
4063         return NULL;
4064 }
4065
4066 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4067 {
4068         struct inode *inode = open_stp->st_file->fi_inode;
4069         unsigned int inohash = lockowner_ino_hashval(inode,
4070                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4071         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4072
4073         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4074         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4075         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4076 }
4077
4078 /*
4079  * Alloc a lock owner structure.
4080  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4081  * occurred.
4082  *
4083  * strhashval = ownerstr_hashval
4084  */
4085
4086 static struct nfs4_lockowner *
4087 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4088         struct nfs4_lockowner *lo;
4089
4090         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4091         if (!lo)
4092                 return NULL;
4093         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4094         lo->lo_owner.so_is_open_owner = 0;
4095         /* It is the openowner seqid that will be incremented in encode in the
4096          * case of new lockowners; so increment the lock seqid manually: */
4097         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4098         hash_lockowner(lo, strhashval, clp, open_stp);
4099         return lo;
4100 }
4101
4102 static struct nfs4_ol_stateid *
4103 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4104 {
4105         struct nfs4_ol_stateid *stp;
4106         struct nfs4_client *clp = lo->lo_owner.so_client;
4107
4108         stp = nfs4_alloc_stateid(clp);
4109         if (stp == NULL)
4110                 return NULL;
4111         stp->st_stid.sc_type = NFS4_LOCK_STID;
4112         list_add(&stp->st_perfile, &fp->fi_stateids);
4113         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4114         stp->st_stateowner = &lo->lo_owner;
4115         get_nfs4_file(fp);
4116         stp->st_file = fp;
4117         stp->st_access_bmap = 0;
4118         stp->st_deny_bmap = open_stp->st_deny_bmap;
4119         stp->st_openstp = open_stp;
4120         return stp;
4121 }
4122
4123 static int
4124 check_lock_length(u64 offset, u64 length)
4125 {
4126         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4127              LOFF_OVERFLOW(offset, length)));
4128 }
4129
4130 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4131 {
4132         struct nfs4_file *fp = lock_stp->st_file;
4133         int oflag = nfs4_access_to_omode(access);
4134
4135         if (test_access(access, lock_stp))
4136                 return;
4137         nfs4_file_get_access(fp, oflag);
4138         set_access(access, lock_stp);
4139 }
4140
4141 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)
4142 {
4143         struct nfs4_file *fi = ost->st_file;
4144         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4145         struct nfs4_client *cl = oo->oo_owner.so_client;
4146         struct nfs4_lockowner *lo;
4147         unsigned int strhashval;
4148         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4149
4150         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4151                                 &lock->v.new.owner, nn);
4152         if (lo) {
4153                 if (!cstate->minorversion)
4154                         return nfserr_bad_seqid;
4155                 /* XXX: a lockowner always has exactly one stateid: */
4156                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4157                                 struct nfs4_ol_stateid, st_perstateowner);
4158                 return nfs_ok;
4159         }
4160         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4161                         &lock->v.new.owner);
4162         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4163         if (lo == NULL)
4164                 return nfserr_jukebox;
4165         *lst = alloc_init_lock_stateid(lo, fi, ost);
4166         if (*lst == NULL) {
4167                 release_lockowner(lo);
4168                 return nfserr_jukebox;
4169         }
4170         *new = true;
4171         return nfs_ok;
4172 }
4173
4174 /*
4175  *  LOCK operation
4176  */
4177 __be32
4178 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4179            struct nfsd4_lock *lock)
4180 {
4181         struct nfs4_openowner *open_sop = NULL;
4182         struct nfs4_lockowner *lock_sop = NULL;
4183         struct nfs4_ol_stateid *lock_stp;
4184         struct file *filp = NULL;
4185         struct file_lock *file_lock = NULL;
4186         struct file_lock *conflock = NULL;
4187         __be32 status = 0;
4188         bool new_state = false;
4189         int lkflg;
4190         int err;
4191         struct net *net = SVC_NET(rqstp);
4192         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4193
4194         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4195                 (long long) lock->lk_offset,
4196                 (long long) lock->lk_length);
4197
4198         if (check_lock_length(lock->lk_offset, lock->lk_length))
4199                  return nfserr_inval;
4200
4201         if ((status = fh_verify(rqstp, &cstate->current_fh,
4202                                 S_IFREG, NFSD_MAY_LOCK))) {
4203                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4204                 return status;
4205         }
4206
4207         nfs4_lock_state();
4208
4209         if (lock->lk_is_new) {
4210                 struct nfs4_ol_stateid *open_stp = NULL;
4211
4212                 if (nfsd4_has_session(cstate))
4213                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4214                         memcpy(&lock->v.new.clientid,
4215                                 &cstate->session->se_client->cl_clientid,
4216                                 sizeof(clientid_t));
4217
4218                 status = nfserr_stale_clientid;
4219                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4220                         goto out;
4221
4222                 /* validate and update open stateid and open seqid */
4223                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4224                                         lock->lk_new_open_seqid,
4225                                         &lock->lk_new_open_stateid,
4226                                         &open_stp, nn);
4227                 if (status)
4228                         goto out;
4229                 open_sop = openowner(open_stp->st_stateowner);
4230                 status = nfserr_bad_stateid;
4231                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4232                                                 &lock->v.new.clientid))
4233                         goto out;
4234                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4235                                                         &lock_stp, &new_state);
4236         } else
4237                 status = nfs4_preprocess_seqid_op(cstate,
4238                                        lock->lk_old_lock_seqid,
4239                                        &lock->lk_old_lock_stateid,
4240                                        NFS4_LOCK_STID, &lock_stp, nn);
4241         if (status)
4242                 goto out;
4243         lock_sop = lockowner(lock_stp->st_stateowner);
4244
4245         lkflg = setlkflg(lock->lk_type);
4246         status = nfs4_check_openmode(lock_stp, lkflg);
4247         if (status)
4248                 goto out;
4249
4250         status = nfserr_grace;
4251         if (locks_in_grace(net) && !lock->lk_reclaim)
4252                 goto out;
4253         status = nfserr_no_grace;
4254         if (!locks_in_grace(net) && lock->lk_reclaim)
4255                 goto out;
4256
4257         file_lock = locks_alloc_lock();
4258         if (!file_lock) {
4259                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4260                 status = nfserr_jukebox;
4261                 goto out;
4262         }
4263
4264         locks_init_lock(file_lock);
4265         switch (lock->lk_type) {
4266                 case NFS4_READ_LT:
4267                 case NFS4_READW_LT:
4268                         filp = find_readable_file(lock_stp->st_file);
4269                         if (filp)
4270                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4271                         file_lock->fl_type = F_RDLCK;
4272                         break;
4273                 case NFS4_WRITE_LT:
4274                 case NFS4_WRITEW_LT:
4275                         filp = find_writeable_file(lock_stp->st_file);
4276                         if (filp)
4277                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4278                         file_lock->fl_type = F_WRLCK;
4279                         break;
4280                 default:
4281                         status = nfserr_inval;
4282                 goto out;
4283         }
4284         if (!filp) {
4285                 status = nfserr_openmode;
4286                 goto out;
4287         }
4288         file_lock->fl_owner = (fl_owner_t)lock_sop;
4289         file_lock->fl_pid = current->tgid;
4290         file_lock->fl_file = filp;
4291         file_lock->fl_flags = FL_POSIX;
4292         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4293         file_lock->fl_start = lock->lk_offset;
4294         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4295         nfs4_transform_lock_offset(file_lock);
4296
4297         conflock = locks_alloc_lock();
4298         if (!conflock) {
4299                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4300                 status = nfserr_jukebox;
4301                 goto out;
4302         }
4303
4304         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4305         switch (-err) {
4306         case 0: /* success! */
4307                 update_stateid(&lock_stp->st_stid.sc_stateid);
4308                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4309                                 sizeof(stateid_t));
4310                 status = 0;
4311                 break;
4312         case (EAGAIN):          /* conflock holds conflicting lock */
4313                 status = nfserr_denied;
4314                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4315                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4316                 break;
4317         case (EDEADLK):
4318                 status = nfserr_deadlock;
4319                 break;
4320         default:
4321                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4322                 status = nfserrno(err);
4323                 break;
4324         }
4325 out:
4326         if (status && new_state)
4327                 release_lockowner(lock_sop);
4328         nfsd4_bump_seqid(cstate, status);
4329         if (!cstate->replay_owner)
4330                 nfs4_unlock_state();
4331         if (file_lock)
4332                 locks_free_lock(file_lock);
4333         if (conflock)
4334                 locks_free_lock(conflock);
4335         return status;
4336 }
4337
4338 /*
4339  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4340  * so we do a temporary open here just to get an open file to pass to
4341  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4342  * inode operation.)
4343  */
4344 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4345 {
4346         struct file *file;
4347         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4348         if (!err) {
4349                 err = nfserrno(vfs_test_lock(file, lock));
4350                 nfsd_close(file);
4351         }
4352         return err;
4353 }
4354
4355 /*
4356  * LOCKT operation
4357  */
4358 __be32
4359 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4360             struct nfsd4_lockt *lockt)
4361 {
4362         struct inode *inode;
4363         struct file_lock *file_lock = NULL;
4364         struct nfs4_lockowner *lo;
4365         __be32 status;
4366         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4367
4368         if (locks_in_grace(SVC_NET(rqstp)))
4369                 return nfserr_grace;
4370
4371         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4372                  return nfserr_inval;
4373
4374         nfs4_lock_state();
4375
4376         if (!nfsd4_has_session(cstate)) {
4377                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4378                 if (status)
4379                         goto out;
4380         }
4381
4382         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4383                 goto out;
4384
4385         inode = cstate->current_fh.fh_dentry->d_inode;
4386         file_lock = locks_alloc_lock();
4387         if (!file_lock) {
4388                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4389                 status = nfserr_jukebox;
4390                 goto out;
4391         }
4392         locks_init_lock(file_lock);
4393         switch (lockt->lt_type) {
4394                 case NFS4_READ_LT:
4395                 case NFS4_READW_LT:
4396                         file_lock->fl_type = F_RDLCK;
4397                 break;
4398                 case NFS4_WRITE_LT:
4399                 case NFS4_WRITEW_LT:
4400                         file_lock->fl_type = F_WRLCK;
4401                 break;
4402                 default:
4403                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4404                         status = nfserr_inval;
4405                 goto out;
4406         }
4407
4408         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4409         if (lo)
4410                 file_lock->fl_owner = (fl_owner_t)lo;
4411         file_lock->fl_pid = current->tgid;
4412         file_lock->fl_flags = FL_POSIX;
4413
4414         file_lock->fl_start = lockt->lt_offset;
4415         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4416
4417         nfs4_transform_lock_offset(file_lock);
4418
4419         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4420         if (status)
4421                 goto out;
4422
4423         if (file_lock->fl_type != F_UNLCK) {
4424                 status = nfserr_denied;
4425                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4426         }
4427 out:
4428         nfs4_unlock_state();
4429         if (file_lock)
4430                 locks_free_lock(file_lock);
4431         return status;
4432 }
4433
4434 __be32
4435 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4436             struct nfsd4_locku *locku)
4437 {
4438         struct nfs4_lockowner *lo;
4439         struct nfs4_ol_stateid *stp;
4440         struct file *filp = NULL;
4441         struct file_lock *file_lock = NULL;
4442         __be32 status;
4443         int err;
4444         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4445
4446         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4447                 (long long) locku->lu_offset,
4448                 (long long) locku->lu_length);
4449
4450         if (check_lock_length(locku->lu_offset, locku->lu_length))
4451                  return nfserr_inval;
4452
4453         nfs4_lock_state();
4454
4455         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4456                                         &locku->lu_stateid, NFS4_LOCK_STID,
4457                                         &stp, nn);
4458         if (status)
4459                 goto out;
4460         filp = find_any_file(stp->st_file);
4461         if (!filp) {
4462                 status = nfserr_lock_range;
4463                 goto out;
4464         }
4465         file_lock = locks_alloc_lock();
4466         if (!file_lock) {
4467                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4468                 status = nfserr_jukebox;
4469                 goto out;
4470         }
4471         lo = lockowner(stp->st_stateowner);
4472         locks_init_lock(file_lock);
4473         file_lock->fl_type = F_UNLCK;
4474         file_lock->fl_owner = (fl_owner_t)lo;
4475         file_lock->fl_pid = current->tgid;
4476         file_lock->fl_file = filp;
4477         file_lock->fl_flags = FL_POSIX;
4478         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4479         file_lock->fl_start = locku->lu_offset;
4480
4481         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4482                                                 locku->lu_length);
4483         nfs4_transform_lock_offset(file_lock);
4484
4485         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4486         if (err) {
4487                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4488                 goto out_nfserr;
4489         }
4490         update_stateid(&stp->st_stid.sc_stateid);
4491         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4492
4493         if (nfsd4_has_session(cstate) && !check_for_locks(stp->st_file, lo)) {
4494                 WARN_ON_ONCE(cstate->replay_owner);
4495                 release_lockowner(lo);
4496         }
4497
4498 out:
4499         nfsd4_bump_seqid(cstate, status);
4500         if (!cstate->replay_owner)
4501                 nfs4_unlock_state();
4502         if (file_lock)
4503                 locks_free_lock(file_lock);
4504         return status;
4505
4506 out_nfserr:
4507         status = nfserrno(err);
4508         goto out;
4509 }
4510
4511 /*
4512  * returns
4513  *      1: locks held by lockowner
4514  *      0: no locks held by lockowner
4515  */
4516 static int
4517 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4518 {
4519         struct file_lock **flpp;
4520         struct inode *inode = filp->fi_inode;
4521         int status = 0;
4522
4523         lock_flocks();
4524         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4525                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4526                         status = 1;
4527                         goto out;
4528                 }
4529         }
4530 out:
4531         unlock_flocks();
4532         return status;
4533 }
4534
4535 __be32
4536 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4537                         struct nfsd4_compound_state *cstate,
4538                         struct nfsd4_release_lockowner *rlockowner)
4539 {
4540         clientid_t *clid = &rlockowner->rl_clientid;
4541         struct nfs4_stateowner *sop;
4542         struct nfs4_lockowner *lo;
4543         struct nfs4_ol_stateid *stp;
4544         struct xdr_netobj *owner = &rlockowner->rl_owner;
4545         struct list_head matches;
4546         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4547         __be32 status;
4548         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4549
4550         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4551                 clid->cl_boot, clid->cl_id);
4552
4553         nfs4_lock_state();
4554
4555         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4556         if (status)
4557                 goto out;
4558
4559         status = nfserr_locks_held;
4560         INIT_LIST_HEAD(&matches);
4561
4562         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4563                 if (sop->so_is_open_owner)
4564                         continue;
4565                 if (!same_owner_str(sop, owner, clid))
4566                         continue;
4567                 list_for_each_entry(stp, &sop->so_stateids,
4568                                 st_perstateowner) {
4569                         lo = lockowner(sop);
4570                         if (check_for_locks(stp->st_file, lo))
4571                                 goto out;
4572                         list_add(&lo->lo_list, &matches);
4573                 }
4574         }
4575         /* Clients probably won't expect us to return with some (but not all)
4576          * of the lockowner state released; so don't release any until all
4577          * have been checked. */
4578         status = nfs_ok;
4579         while (!list_empty(&matches)) {
4580                 lo = list_entry(matches.next, struct nfs4_lockowner,
4581                                                                 lo_list);
4582                 /* unhash_stateowner deletes so_perclient only
4583                  * for openowners. */
4584                 list_del(&lo->lo_list);
4585                 release_lockowner(lo);
4586         }
4587 out:
4588         nfs4_unlock_state();
4589         return status;
4590 }
4591
4592 static inline struct nfs4_client_reclaim *
4593 alloc_reclaim(void)
4594 {
4595         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4596 }
4597
4598 bool
4599 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4600 {
4601         struct nfs4_client_reclaim *crp;
4602
4603         crp = nfsd4_find_reclaim_client(name, nn);
4604         return (crp && crp->cr_clp);
4605 }
4606
4607 /*
4608  * failure => all reset bets are off, nfserr_no_grace...
4609  */
4610 struct nfs4_client_reclaim *
4611 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4612 {
4613         unsigned int strhashval;
4614         struct nfs4_client_reclaim *crp;
4615
4616         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4617         crp = alloc_reclaim();
4618         if (crp) {
4619                 strhashval = clientstr_hashval(name);
4620                 INIT_LIST_HEAD(&crp->cr_strhash);
4621                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4622                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4623                 crp->cr_clp = NULL;
4624                 nn->reclaim_str_hashtbl_size++;
4625         }
4626         return crp;
4627 }
4628
4629 void
4630 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4631 {
4632         list_del(&crp->cr_strhash);
4633         kfree(crp);
4634         nn->reclaim_str_hashtbl_size--;
4635 }
4636
4637 void
4638 nfs4_release_reclaim(struct nfsd_net *nn)
4639 {
4640         struct nfs4_client_reclaim *crp = NULL;
4641         int i;
4642
4643         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4644                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4645                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4646                                         struct nfs4_client_reclaim, cr_strhash);
4647                         nfs4_remove_reclaim_record(crp, nn);
4648                 }
4649         }
4650         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4651 }
4652
4653 /*
4654  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4655 struct nfs4_client_reclaim *
4656 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4657 {
4658         unsigned int strhashval;
4659         struct nfs4_client_reclaim *crp = NULL;
4660
4661         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4662
4663         strhashval = clientstr_hashval(recdir);
4664         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4665                 if (same_name(crp->cr_recdir, recdir)) {
4666                         return crp;
4667                 }
4668         }
4669         return NULL;
4670 }
4671
4672 /*
4673 * Called from OPEN. Look for clientid in reclaim list.
4674 */
4675 __be32
4676 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4677 {
4678         struct nfs4_client *clp;
4679
4680         /* find clientid in conf_id_hashtbl */
4681         clp = find_confirmed_client(clid, sessions, nn);
4682         if (clp == NULL)
4683                 return nfserr_reclaim_bad;
4684
4685         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4686 }
4687
4688 #ifdef CONFIG_NFSD_FAULT_INJECTION
4689
4690 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4691 {
4692         if (mark_client_expired(clp))
4693                 return 0;
4694         expire_client(clp);
4695         return 1;
4696 }
4697
4698 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4699 {
4700         char buf[INET6_ADDRSTRLEN];
4701         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4702         printk(KERN_INFO "NFS Client: %s\n", buf);
4703         return 1;
4704 }
4705
4706 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4707                              const char *type)
4708 {
4709         char buf[INET6_ADDRSTRLEN];
4710         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4711         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4712 }
4713
4714 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4715 {
4716         struct nfs4_openowner *oop;
4717         struct nfs4_lockowner *lop, *lo_next;
4718         struct nfs4_ol_stateid *stp, *st_next;
4719         u64 count = 0;
4720
4721         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4722                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4723                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4724                                 if (func)
4725                                         func(lop);
4726                                 if (++count == max)
4727                                         return count;
4728                         }
4729                 }
4730         }
4731
4732         return count;
4733 }
4734
4735 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4736 {
4737         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4738 }
4739
4740 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4741 {
4742         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4743         nfsd_print_count(clp, count, "locked files");
4744         return count;
4745 }
4746
4747 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4748 {
4749         struct nfs4_openowner *oop, *next;
4750         u64 count = 0;
4751
4752         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4753                 if (func)
4754                         func(oop);
4755                 if (++count == max)
4756                         break;
4757         }
4758
4759         return count;
4760 }
4761
4762 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4763 {
4764         return nfsd_foreach_client_open(clp, max, release_openowner);
4765 }
4766
4767 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4768 {
4769         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4770         nfsd_print_count(clp, count, "open files");
4771         return count;
4772 }
4773
4774 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4775                                      struct list_head *victims)
4776 {
4777         struct nfs4_delegation *dp, *next;
4778         u64 count = 0;
4779
4780         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4781                 if (victims)
4782                         list_move(&dp->dl_recall_lru, victims);
4783                 if (++count == max)
4784                         break;
4785         }
4786         return count;
4787 }
4788
4789 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4790 {
4791         struct nfs4_delegation *dp, *next;
4792         LIST_HEAD(victims);
4793         u64 count;
4794
4795         spin_lock(&recall_lock);
4796         count = nfsd_find_all_delegations(clp, max, &victims);
4797         spin_unlock(&recall_lock);
4798
4799         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4800                 revoke_delegation(dp);
4801
4802         return count;
4803 }
4804
4805 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4806 {
4807         struct nfs4_delegation *dp, *next;
4808         LIST_HEAD(victims);
4809         u64 count;
4810
4811         spin_lock(&recall_lock);
4812         count = nfsd_find_all_delegations(clp, max, &victims);
4813         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4814                 nfsd_break_one_deleg(dp);
4815         spin_unlock(&recall_lock);
4816
4817         return count;
4818 }
4819
4820 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4821 {
4822         u64 count = 0;
4823
4824         spin_lock(&recall_lock);
4825         count = nfsd_find_all_delegations(clp, max, NULL);
4826         spin_unlock(&recall_lock);
4827
4828         nfsd_print_count(clp, count, "delegations");
4829         return count;
4830 }
4831
4832 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4833 {
4834         struct nfs4_client *clp, *next;
4835         u64 count = 0;
4836         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4837
4838         if (!nfsd_netns_ready(nn))
4839                 return 0;
4840
4841         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4842                 count += func(clp, max - count);
4843                 if ((max != 0) && (count >= max))
4844                         break;
4845         }
4846
4847         return count;
4848 }
4849
4850 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4851 {
4852         struct nfs4_client *clp;
4853         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4854
4855         if (!nfsd_netns_ready(nn))
4856                 return NULL;
4857
4858         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4859                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4860                         return clp;
4861         }
4862         return NULL;
4863 }
4864
4865 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4866
4867 /* initialization to perform at module load time: */
4868
4869 void
4870 nfs4_state_init(void)
4871 {
4872 }
4873
4874 /*
4875  * Since the lifetime of a delegation isn't limited to that of an open, a
4876  * client may quite reasonably hang on to a delegation as long as it has
4877  * the inode cached.  This becomes an obvious problem the first time a
4878  * client's inode cache approaches the size of the server's total memory.
4879  *
4880  * For now we avoid this problem by imposing a hard limit on the number
4881  * of delegations, which varies according to the server's memory size.
4882  */
4883 static void
4884 set_max_delegations(void)
4885 {
4886         /*
4887          * Allow at most 4 delegations per megabyte of RAM.  Quick
4888          * estimates suggest that in the worst case (where every delegation
4889          * is for a different inode), a delegation could take about 1.5K,
4890          * giving a worst case usage of about 6% of memory.
4891          */
4892         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4893 }
4894
4895 static int nfs4_state_create_net(struct net *net)
4896 {
4897         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4898         int i;
4899
4900         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4901                         CLIENT_HASH_SIZE, GFP_KERNEL);
4902         if (!nn->conf_id_hashtbl)
4903                 goto err;
4904         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4905                         CLIENT_HASH_SIZE, GFP_KERNEL);
4906         if (!nn->unconf_id_hashtbl)
4907                 goto err_unconf_id;
4908         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4909                         OWNER_HASH_SIZE, GFP_KERNEL);
4910         if (!nn->ownerstr_hashtbl)
4911                 goto err_ownerstr;
4912         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
4913                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
4914         if (!nn->lockowner_ino_hashtbl)
4915                 goto err_lockowner_ino;
4916         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
4917                         SESSION_HASH_SIZE, GFP_KERNEL);
4918         if (!nn->sessionid_hashtbl)
4919                 goto err_sessionid;
4920
4921         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4922                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
4923                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
4924         }
4925         for (i = 0; i < OWNER_HASH_SIZE; i++)
4926                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
4927         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4928                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
4929         for (i = 0; i < SESSION_HASH_SIZE; i++)
4930                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
4931         nn->conf_name_tree = RB_ROOT;
4932         nn->unconf_name_tree = RB_ROOT;
4933         INIT_LIST_HEAD(&nn->client_lru);
4934         INIT_LIST_HEAD(&nn->close_lru);
4935         INIT_LIST_HEAD(&nn->del_recall_lru);
4936         spin_lock_init(&nn->client_lock);
4937
4938         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
4939         get_net(net);
4940
4941         return 0;
4942
4943 err_sessionid:
4944         kfree(nn->lockowner_ino_hashtbl);
4945 err_lockowner_ino:
4946         kfree(nn->ownerstr_hashtbl);
4947 err_ownerstr:
4948         kfree(nn->unconf_id_hashtbl);
4949 err_unconf_id:
4950         kfree(nn->conf_id_hashtbl);
4951 err:
4952         return -ENOMEM;
4953 }
4954
4955 static void
4956 nfs4_state_destroy_net(struct net *net)
4957 {
4958         int i;
4959         struct nfs4_client *clp = NULL;
4960         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4961         struct rb_node *node, *tmp;
4962
4963         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4964                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
4965                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4966                         destroy_client(clp);
4967                 }
4968         }
4969
4970         node = rb_first(&nn->unconf_name_tree);
4971         while (node != NULL) {
4972                 tmp = node;
4973                 node = rb_next(tmp);
4974                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
4975                 rb_erase(tmp, &nn->unconf_name_tree);
4976                 destroy_client(clp);
4977         }
4978
4979         kfree(nn->sessionid_hashtbl);
4980         kfree(nn->lockowner_ino_hashtbl);
4981         kfree(nn->ownerstr_hashtbl);
4982         kfree(nn->unconf_id_hashtbl);
4983         kfree(nn->conf_id_hashtbl);
4984         put_net(net);
4985 }
4986
4987 int
4988 nfs4_state_start_net(struct net *net)
4989 {
4990         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4991         int ret;
4992
4993         ret = nfs4_state_create_net(net);
4994         if (ret)
4995                 return ret;
4996         nfsd4_client_tracking_init(net);
4997         nn->boot_time = get_seconds();
4998         locks_start_grace(net, &nn->nfsd4_manager);
4999         nn->grace_ended = false;
5000         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
5001                nn->nfsd4_grace, net);
5002         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
5003         return 0;
5004 }
5005
5006 /* initialization to perform when the nfsd service is started: */
5007
5008 int
5009 nfs4_state_start(void)
5010 {
5011         int ret;
5012
5013         ret = set_callback_cred();
5014         if (ret)
5015                 return -ENOMEM;
5016         laundry_wq = create_singlethread_workqueue("nfsd4");
5017         if (laundry_wq == NULL) {
5018                 ret = -ENOMEM;
5019                 goto out_recovery;
5020         }
5021         ret = nfsd4_create_callback_queue();
5022         if (ret)
5023                 goto out_free_laundry;
5024
5025         set_max_delegations();
5026
5027         return 0;
5028
5029 out_free_laundry:
5030         destroy_workqueue(laundry_wq);
5031 out_recovery:
5032         return ret;
5033 }
5034
5035 /* should be called with the state lock held */
5036 void
5037 nfs4_state_shutdown_net(struct net *net)
5038 {
5039         struct nfs4_delegation *dp = NULL;
5040         struct list_head *pos, *next, reaplist;
5041         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5042
5043         cancel_delayed_work_sync(&nn->laundromat_work);
5044         locks_end_grace(&nn->nfsd4_manager);
5045
5046         INIT_LIST_HEAD(&reaplist);
5047         spin_lock(&recall_lock);
5048         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5049                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5050                 list_move(&dp->dl_recall_lru, &reaplist);
5051         }
5052         spin_unlock(&recall_lock);
5053         list_for_each_safe(pos, next, &reaplist) {
5054                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5055                 destroy_delegation(dp);
5056         }
5057
5058         nfsd4_client_tracking_exit(net);
5059         nfs4_state_destroy_net(net);
5060 }
5061
5062 void
5063 nfs4_state_shutdown(void)
5064 {
5065         destroy_workqueue(laundry_wq);
5066         nfsd4_destroy_callback_queue();
5067 }
5068
5069 static void
5070 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5071 {
5072         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5073                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5074 }
5075
5076 static void
5077 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5078 {
5079         if (cstate->minorversion) {
5080                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5081                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5082         }
5083 }
5084
5085 void
5086 clear_current_stateid(struct nfsd4_compound_state *cstate)
5087 {
5088         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5089 }
5090
5091 /*
5092  * functions to set current state id
5093  */
5094 void
5095 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5096 {
5097         put_stateid(cstate, &odp->od_stateid);
5098 }
5099
5100 void
5101 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5102 {
5103         put_stateid(cstate, &open->op_stateid);
5104 }
5105
5106 void
5107 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5108 {
5109         put_stateid(cstate, &close->cl_stateid);
5110 }
5111
5112 void
5113 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5114 {
5115         put_stateid(cstate, &lock->lk_resp_stateid);
5116 }
5117
5118 /*
5119  * functions to consume current state id
5120  */
5121
5122 void
5123 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5124 {
5125         get_stateid(cstate, &odp->od_stateid);
5126 }
5127
5128 void
5129 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5130 {
5131         get_stateid(cstate, &drp->dr_stateid);
5132 }
5133
5134 void
5135 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5136 {
5137         get_stateid(cstate, &fsp->fr_stateid);
5138 }
5139
5140 void
5141 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5142 {
5143         get_stateid(cstate, &setattr->sa_stateid);
5144 }
5145
5146 void
5147 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5148 {
5149         get_stateid(cstate, &close->cl_stateid);
5150 }
5151
5152 void
5153 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5154 {
5155         get_stateid(cstate, &locku->lu_stateid);
5156 }
5157
5158 void
5159 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5160 {
5161         get_stateid(cstate, &read->rd_stateid);
5162 }
5163
5164 void
5165 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5166 {
5167         get_stateid(cstate, &write->wr_stateid);
5168 }