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