Pileus Git - ~andy/linux/blob - fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include "xdr4.h"
  45 #include "xdr4cb.h"
  46 #include "vfs.h"
  47 #include "current_stateid.h"
  48
  49 #include "netns.h"
  50
  51 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  52
  53 #define all_ones {{~0,~0},~0}
  54 static const stateid_t one_stateid = {
  55         .si_generation = ~0,
  56         .si_opaque = all_ones,
  57 };
  58 static const stateid_t zero_stateid = {
  59         /* all fields zero */
  60 };
  61 static const stateid_t currentstateid = {
  62         .si_generation = 1,
  63 };
  64
  65 static u64 current_sessionid = 1;
  66
  67 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  68 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  69 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  70
  71 /* forward declarations */
  72 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
  73
  74 /* Locking: */
  75
  76 /* Currently used for almost all code touching nfsv4 state: */
  77 static DEFINE_MUTEX(client_mutex);
  78
  79 /*
  80  * Currently used for the del_recall_lru and file hash table.  In an
  81  * effort to decrease the scope of the client_mutex, this spinlock may
  82  * eventually cover more:
  83  */
  84 static DEFINE_SPINLOCK(recall_lock);
  85
  86 static struct kmem_cache *openowner_slab = NULL;
  87 static struct kmem_cache *lockowner_slab = NULL;
  88 static struct kmem_cache *file_slab = NULL;
  89 static struct kmem_cache *stateid_slab = NULL;
  90 static struct kmem_cache *deleg_slab = NULL;
  91
  92 void
  93 nfs4_lock_state(void)
  94 {
  95         mutex_lock(&client_mutex);
  96 }
  97
  98 static void free_session(struct nfsd4_session *);
  99
 100 void nfsd4_put_session(struct nfsd4_session *ses)
 101 {
 102         atomic_dec(&ses->se_ref);
 103 }
 104
 105 static bool is_session_dead(struct nfsd4_session *ses)
 106 {
 107         return ses->se_flags & NFS4_SESSION_DEAD;
 108 }
 109
 110 static __be32 mark_session_dead_locked(struct nfsd4_session *ses)
 111 {
 112         if (atomic_read(&ses->se_ref))
 113                 return nfserr_jukebox;
 114         ses->se_flags |= NFS4_SESSION_DEAD;
 115         return nfs_ok;
 116 }
 117
 118 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 119 {
 120         if (is_session_dead(ses))
 121                 return nfserr_badsession;
 122         atomic_inc(&ses->se_ref);
 123         return nfs_ok;
 124 }
 125
 126 void
 127 nfs4_unlock_state(void)
 128 {
 129         mutex_unlock(&client_mutex);
 130 }
 131
 132 static bool is_client_expired(struct nfs4_client *clp)
 133 {
 134         return clp->cl_time == 0;
 135 }
 136
 137 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
 138 {
 139         if (atomic_read(&clp->cl_refcount))
 140                 return nfserr_jukebox;
 141         clp->cl_time = 0;
 142         return nfs_ok;
 143 }
 144
 145 static __be32 mark_client_expired(struct nfs4_client *clp)
 146 {
 147         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 148         __be32 ret;
 149
 150         spin_lock(&nn->client_lock);
 151         ret = mark_client_expired_locked(clp);
 152         spin_unlock(&nn->client_lock);
 153         return ret;
 154 }
 155
 156 static __be32 get_client_locked(struct nfs4_client *clp)
 157 {
 158         if (is_client_expired(clp))
 159                 return nfserr_expired;
 160         atomic_inc(&clp->cl_refcount);
 161         return nfs_ok;
 162 }
 163
 164 /* must be called under the client_lock */
 165 static inline void
 166 renew_client_locked(struct nfs4_client *clp)
 167 {
 168         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 169
 170         if (is_client_expired(clp)) {
 171                 WARN_ON(1);
 172                 printk("%s: client (clientid %08x/%08x) already expired\n",
 173                         __func__,
 174                         clp->cl_clientid.cl_boot,
 175                         clp->cl_clientid.cl_id);
 176                 return;
 177         }
 178
 179         dprintk("renewing client (clientid %08x/%08x)\n",
 180                         clp->cl_clientid.cl_boot,
 181                         clp->cl_clientid.cl_id);
 182         list_move_tail(&clp->cl_lru, &nn->client_lru);
 183         clp->cl_time = get_seconds();
 184 }
 185
 186 static inline void
 187 renew_client(struct nfs4_client *clp)
 188 {
 189         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 190
 191         spin_lock(&nn->client_lock);
 192         renew_client_locked(clp);
 193         spin_unlock(&nn->client_lock);
 194 }
 195
 196 static void put_client_renew_locked(struct nfs4_client *clp)
 197 {
 198         if (!atomic_dec_and_test(&clp->cl_refcount))
 199                 return;
 200         if (!is_client_expired(clp))
 201                 renew_client_locked(clp);
 202 }
 203
 204 void put_client_renew(struct nfs4_client *clp)
 205 {
 206         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 207
 208         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 209                 return;
 210         if (!is_client_expired(clp))
 211                 renew_client_locked(clp);
 212         spin_unlock(&nn->client_lock);
 213 }
 214
 215
 216 static inline u32
 217 opaque_hashval(const void *ptr, int nbytes)
 218 {
 219         unsigned char *cptr = (unsigned char *) ptr;
 220
 221         u32 x = 0;
 222         while (nbytes--) {
 223                 x *= 37;
 224                 x += *cptr++;
 225         }
 226         return x;
 227 }
 228
 229 static void nfsd4_free_file(struct nfs4_file *f)
 230 {
 231         kmem_cache_free(file_slab, f);
 232 }
 233
 234 static inline void
 235 put_nfs4_file(struct nfs4_file *fi)
 236 {
 237         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 238                 hlist_del(&fi->fi_hash);
 239                 spin_unlock(&recall_lock);
 240                 iput(fi->fi_inode);
 241                 nfsd4_free_file(fi);
 242         }
 243 }
 244
 245 static inline void
 246 get_nfs4_file(struct nfs4_file *fi)
 247 {
 248         atomic_inc(&fi->fi_ref);
 249 }
 250
 251 static int num_delegations;
 252 unsigned long max_delegations;
 253
 254 /*
 255  * Open owner state (share locks)
 256  */
 257
 258 /* hash tables for lock and open owners */
 259 #define OWNER_HASH_BITS              8
 260 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 261 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 262
 263 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 264 {
 265         unsigned int ret;
 266
 267         ret = opaque_hashval(ownername->data, ownername->len);
 268         ret += clientid;
 269         return ret & OWNER_HASH_MASK;
 270 }
 271
 272 /* hash table for nfs4_file */
 273 #define FILE_HASH_BITS                   8
 274 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 275
 276 static unsigned int file_hashval(struct inode *ino)
 277 {
 278         /* XXX: why are we hashing on inode pointer, anyway? */
 279         return hash_ptr(ino, FILE_HASH_BITS);
 280 }
 281
 282 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 283
 284 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 285 {
 286         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 287         atomic_inc(&fp->fi_access[oflag]);
 288 }
 289
 290 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 291 {
 292         if (oflag == O_RDWR) {
 293                 __nfs4_file_get_access(fp, O_RDONLY);
 294                 __nfs4_file_get_access(fp, O_WRONLY);
 295         } else
 296                 __nfs4_file_get_access(fp, oflag);
 297 }
 298
 299 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 300 {
 301         if (fp->fi_fds[oflag]) {
 302                 fput(fp->fi_fds[oflag]);
 303                 fp->fi_fds[oflag] = NULL;
 304         }
 305 }
 306
 307 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 308 {
 309         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 310                 nfs4_file_put_fd(fp, oflag);
 311                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 312                         nfs4_file_put_fd(fp, O_RDWR);
 313         }
 314 }
 315
 316 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 317 {
 318         if (oflag == O_RDWR) {
 319                 __nfs4_file_put_access(fp, O_RDONLY);
 320                 __nfs4_file_put_access(fp, O_WRONLY);
 321         } else
 322                 __nfs4_file_put_access(fp, oflag);
 323 }
 324
 325 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct
 326 kmem_cache *slab)
 327 {
 328         struct idr *stateids = &cl->cl_stateids;
 329         struct nfs4_stid *stid;
 330         int new_id;
 331
 332         stid = kmem_cache_alloc(slab, GFP_KERNEL);
 333         if (!stid)
 334                 return NULL;
 335
 336         new_id = idr_alloc_cyclic(stateids, stid, 0, 0, GFP_KERNEL);
 337         if (new_id < 0)
 338                 goto out_free;
 339         stid->sc_client = cl;
 340         stid->sc_type = 0;
 341         stid->sc_stateid.si_opaque.so_id = new_id;
 342         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 343         /* Will be incremented before return to client: */
 344         stid->sc_stateid.si_generation = 0;
 345
 346         /*
 347          * It shouldn't be a problem to reuse an opaque stateid value.
 348          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 349          * example, a stray write retransmission could be accepted by
 350          * the server when it should have been rejected.  Therefore,
 351          * adopt a trick from the sctp code to attempt to maximize the
 352          * amount of time until an id is reused, by ensuring they always
 353          * "increase" (mod INT_MAX):
 354          */
 355         return stid;
 356 out_free:
 357         kmem_cache_free(slab, stid);
 358         return NULL;
 359 }
 360
 361 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 362 {
 363         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 364 }
 365
 366 static struct nfs4_delegation *
 367 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
 368 {
 369         struct nfs4_delegation *dp;
 370         struct nfs4_file *fp = stp->st_file;
 371
 372         dprintk("NFSD alloc_init_deleg\n");
 373         /*
 374          * Major work on the lease subsystem (for example, to support
 375          * calbacks on stat) will be required before we can support
 376          * write delegations properly.
 377          */
 378         if (type != NFS4_OPEN_DELEGATE_READ)
 379                 return NULL;
 380         if (fp->fi_had_conflict)
 381                 return NULL;
 382         if (num_delegations > max_delegations)
 383                 return NULL;
 384         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 385         if (dp == NULL)
 386                 return dp;
 387         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 388         /*
 389          * delegation seqid's are never incremented.  The 4.1 special
 390          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 391          * 0 anyway just for consistency and use 1:
 392          */
 393         dp->dl_stid.sc_stateid.si_generation = 1;
 394         num_delegations++;
 395         INIT_LIST_HEAD(&dp->dl_perfile);
 396         INIT_LIST_HEAD(&dp->dl_perclnt);
 397         INIT_LIST_HEAD(&dp->dl_recall_lru);
 398         get_nfs4_file(fp);
 399         dp->dl_file = fp;
 400         dp->dl_type = type;
 401         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 402         dp->dl_time = 0;
 403         atomic_set(&dp->dl_count, 1);
 404         nfsd4_init_callback(&dp->dl_recall);
 405         return dp;
 406 }
 407
 408 static void remove_stid(struct nfs4_stid *s)
 409 {
 410         struct idr *stateids = &s->sc_client->cl_stateids;
 411
 412         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 413 }
 414
 415 void
 416 nfs4_put_delegation(struct nfs4_delegation *dp)
 417 {
 418         if (atomic_dec_and_test(&dp->dl_count)) {
 419                 kmem_cache_free(deleg_slab, dp);
 420                 num_delegations--;
 421         }
 422 }
 423
 424 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 425 {
 426         if (atomic_dec_and_test(&fp->fi_delegees)) {
 427                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 428                 fp->fi_lease = NULL;
 429                 fput(fp->fi_deleg_file);
 430                 fp->fi_deleg_file = NULL;
 431         }
 432 }
 433
 434 static void unhash_stid(struct nfs4_stid *s)
 435 {
 436         s->sc_type = 0;
 437 }
 438
 439 /* Called under the state lock. */
 440 static void
 441 unhash_delegation(struct nfs4_delegation *dp)
 442 {
 443         list_del_init(&dp->dl_perclnt);
 444         spin_lock(&recall_lock);
 445         list_del_init(&dp->dl_perfile);
 446         list_del_init(&dp->dl_recall_lru);
 447         spin_unlock(&recall_lock);
 448         nfs4_put_deleg_lease(dp->dl_file);
 449         put_nfs4_file(dp->dl_file);
 450         dp->dl_file = NULL;
 451 }
 452
 453
 454
 455 static void destroy_revoked_delegation(struct nfs4_delegation *dp)
 456 {
 457         list_del_init(&dp->dl_recall_lru);
 458         remove_stid(&dp->dl_stid);
 459         nfs4_put_delegation(dp);
 460 }
 461
 462 static void destroy_delegation(struct nfs4_delegation *dp)
 463 {
 464         unhash_delegation(dp);
 465         remove_stid(&dp->dl_stid);
 466         nfs4_put_delegation(dp);
 467 }
 468
 469 static void revoke_delegation(struct nfs4_delegation *dp)
 470 {
 471         struct nfs4_client *clp = dp->dl_stid.sc_client;
 472
 473         if (clp->cl_minorversion == 0)
 474                 destroy_delegation(dp);
 475         else {
 476                 unhash_delegation(dp);
 477                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 478                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 479         }
 480 }
 481
 482 /*
 483  * SETCLIENTID state
 484  */
 485
 486 static unsigned int clientid_hashval(u32 id)
 487 {
 488         return id & CLIENT_HASH_MASK;
 489 }
 490
 491 static unsigned int clientstr_hashval(const char *name)
 492 {
 493         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 494 }
 495
 496 /*
 497  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 498  * st_{access,deny}_bmap field of the stateid, in order to track not
 499  * only what share bits are currently in force, but also what
 500  * combinations of share bits previous opens have used.  This allows us
 501  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 502  * return an error if the client attempt to downgrade to a combination
 503  * of share bits not explicable by closing some of its previous opens.
 504  *
 505  * XXX: This enforcement is actually incomplete, since we don't keep
 506  * track of access/deny bit combinations; so, e.g., we allow:
 507  *
 508  *      OPEN allow read, deny write
 509  *      OPEN allow both, deny none
 510  *      DOWNGRADE allow read, deny none
 511  *
 512  * which we should reject.
 513  */
 514 static unsigned int
 515 bmap_to_share_mode(unsigned long bmap) {
 516         int i;
 517         unsigned int access = 0;
 518
 519         for (i = 1; i < 4; i++) {
 520                 if (test_bit(i, &bmap))
 521                         access |= i;
 522         }
 523         return access;
 524 }
 525
 526 static bool
 527 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 528         unsigned int access, deny;
 529
 530         access = bmap_to_share_mode(stp->st_access_bmap);
 531         deny = bmap_to_share_mode(stp->st_deny_bmap);
 532         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 533                 return false;
 534         return true;
 535 }
 536
 537 /* set share access for a given stateid */
 538 static inline void
 539 set_access(u32 access, struct nfs4_ol_stateid *stp)
 540 {
 541         __set_bit(access, &stp->st_access_bmap);
 542 }
 543
 544 /* clear share access for a given stateid */
 545 static inline void
 546 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 547 {
 548         __clear_bit(access, &stp->st_access_bmap);
 549 }
 550
 551 /* test whether a given stateid has access */
 552 static inline bool
 553 test_access(u32 access, struct nfs4_ol_stateid *stp)
 554 {
 555         return test_bit(access, &stp->st_access_bmap);
 556 }
 557
 558 /* set share deny for a given stateid */
 559 static inline void
 560 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 561 {
 562         __set_bit(access, &stp->st_deny_bmap);
 563 }
 564
 565 /* clear share deny for a given stateid */
 566 static inline void
 567 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 568 {
 569         __clear_bit(access, &stp->st_deny_bmap);
 570 }
 571
 572 /* test whether a given stateid is denying specific access */
 573 static inline bool
 574 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 575 {
 576         return test_bit(access, &stp->st_deny_bmap);
 577 }
 578
 579 static int nfs4_access_to_omode(u32 access)
 580 {
 581         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 582         case NFS4_SHARE_ACCESS_READ:
 583                 return O_RDONLY;
 584         case NFS4_SHARE_ACCESS_WRITE:
 585                 return O_WRONLY;
 586         case NFS4_SHARE_ACCESS_BOTH:
 587                 return O_RDWR;
 588         }
 589         WARN_ON_ONCE(1);
 590         return O_RDONLY;
 591 }
 592
 593 /* release all access and file references for a given stateid */
 594 static void
 595 release_all_access(struct nfs4_ol_stateid *stp)
 596 {
 597         int i;
 598
 599         for (i = 1; i < 4; i++) {
 600                 if (test_access(i, stp))
 601                         nfs4_file_put_access(stp->st_file,
 602                                              nfs4_access_to_omode(i));
 603                 clear_access(i, stp);
 604         }
 605 }
 606
 607 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 608 {
 609         list_del(&stp->st_perfile);
 610         list_del(&stp->st_perstateowner);
 611 }
 612
 613 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 614 {
 615         release_all_access(stp);
 616         put_nfs4_file(stp->st_file);
 617         stp->st_file = NULL;
 618 }
 619
 620 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 621 {
 622         remove_stid(&stp->st_stid);
 623         kmem_cache_free(stateid_slab, stp);
 624 }
 625
 626 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 627 {
 628         struct file *file;
 629
 630         unhash_generic_stateid(stp);
 631         unhash_stid(&stp->st_stid);
 632         file = find_any_file(stp->st_file);
 633         if (file)
 634                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 635         close_generic_stateid(stp);
 636         free_generic_stateid(stp);
 637 }
 638
 639 static void unhash_lockowner(struct nfs4_lockowner *lo)
 640 {
 641         struct nfs4_ol_stateid *stp;
 642
 643         list_del(&lo->lo_owner.so_strhash);
 644         list_del(&lo->lo_perstateid);
 645         list_del(&lo->lo_owner_ino_hash);
 646         while (!list_empty(&lo->lo_owner.so_stateids)) {
 647                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 648                                 struct nfs4_ol_stateid, st_perstateowner);
 649                 release_lock_stateid(stp);
 650         }
 651 }
 652
 653 static void release_lockowner(struct nfs4_lockowner *lo)
 654 {
 655         unhash_lockowner(lo);
 656         nfs4_free_lockowner(lo);
 657 }
 658
 659 static void
 660 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 661 {
 662         struct nfs4_lockowner *lo;
 663
 664         while (!list_empty(&open_stp->st_lockowners)) {
 665                 lo = list_entry(open_stp->st_lockowners.next,
 666                                 struct nfs4_lockowner, lo_perstateid);
 667                 release_lockowner(lo);
 668         }
 669 }
 670
 671 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 672 {
 673         unhash_generic_stateid(stp);
 674         release_stateid_lockowners(stp);
 675         close_generic_stateid(stp);
 676 }
 677
 678 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 679 {
 680         unhash_open_stateid(stp);
 681         unhash_stid(&stp->st_stid);
 682         free_generic_stateid(stp);
 683 }
 684
 685 static void unhash_openowner(struct nfs4_openowner *oo)
 686 {
 687         struct nfs4_ol_stateid *stp;
 688
 689         list_del(&oo->oo_owner.so_strhash);
 690         list_del(&oo->oo_perclient);
 691         while (!list_empty(&oo->oo_owner.so_stateids)) {
 692                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 693                                 struct nfs4_ol_stateid, st_perstateowner);
 694                 release_open_stateid(stp);
 695         }
 696 }
 697
 698 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 699 {
 700         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 701
 702         if (s) {
 703                 unhash_stid(&s->st_stid);
 704                 free_generic_stateid(s);
 705                 oo->oo_last_closed_stid = NULL;
 706         }
 707 }
 708
 709 static void release_openowner(struct nfs4_openowner *oo)
 710 {
 711         unhash_openowner(oo);
 712         list_del(&oo->oo_close_lru);
 713         release_last_closed_stateid(oo);
 714         nfs4_free_openowner(oo);
 715 }
 716
 717 static inline int
 718 hash_sessionid(struct nfs4_sessionid *sessionid)
 719 {
 720         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 721
 722         return sid->sequence % SESSION_HASH_SIZE;
 723 }
 724
 725 #ifdef NFSD_DEBUG
 726 static inline void
 727 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 728 {
 729         u32 *ptr = (u32 *)(&sessionid->data[0]);
 730         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 731 }
 732 #else
 733 static inline void
 734 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 735 {
 736 }
 737 #endif
 738
 739 /*
 740  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
 741  * won't be used for replay.
 742  */
 743 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
 744 {
 745         struct nfs4_stateowner *so = cstate->replay_owner;
 746
 747         if (nfserr == nfserr_replay_me)
 748                 return;
 749
 750         if (!seqid_mutating_err(ntohl(nfserr))) {
 751                 cstate->replay_owner = NULL;
 752                 return;
 753         }
 754         if (!so)
 755                 return;
 756         if (so->so_is_open_owner)
 757                 release_last_closed_stateid(openowner(so));
 758         so->so_seqid++;
 759         return;
 760 }
 761
 762 static void
 763 gen_sessionid(struct nfsd4_session *ses)
 764 {
 765         struct nfs4_client *clp = ses->se_client;
 766         struct nfsd4_sessionid *sid;
 767
 768         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 769         sid->clientid = clp->cl_clientid;
 770         sid->sequence = current_sessionid++;
 771         sid->reserved = 0;
 772 }
 773
 774 /*
 775  * The protocol defines ca_maxresponssize_cached to include the size of
 776  * the rpc header, but all we need to cache is the data starting after
 777  * the end of the initial SEQUENCE operation--the rest we regenerate
 778  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 779  * value that is the number of bytes in our cache plus a few additional
 780  * bytes.  In order to stay on the safe side, and not promise more than
 781  * we can cache, those additional bytes must be the minimum possible: 24
 782  * bytes of rpc header (xid through accept state, with AUTH_NULL
 783  * verifier), 12 for the compound header (with zero-length tag), and 44
 784  * for the SEQUENCE op response:
 785  */
 786 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 787
 788 static void
 789 free_session_slots(struct nfsd4_session *ses)
 790 {
 791         int i;
 792
 793         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 794                 kfree(ses->se_slots[i]);
 795 }
 796
 797 /*
 798  * We don't actually need to cache the rpc and session headers, so we
 799  * can allocate a little less for each slot:
 800  */
 801 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
 802 {
 803         u32 size;
 804
 805         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
 806                 size = 0;
 807         else
 808                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 809         return size + sizeof(struct nfsd4_slot);
 810 }
 811
 812 /*
 813  * XXX: If we run out of reserved DRC memory we could (up to a point)
 814  * re-negotiate active sessions and reduce their slot usage to make
 815  * room for new connections. For now we just fail the create session.
 816  */
 817 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
 818 {
 819         u32 slotsize = slot_bytes(ca);
 820         u32 num = ca->maxreqs;
 821         int avail;
 822
 823         spin_lock(&nfsd_drc_lock);
 824         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
 825                     nfsd_drc_max_mem - nfsd_drc_mem_used);
 826         num = min_t(int, num, avail / slotsize);
 827         nfsd_drc_mem_used += num * slotsize;
 828         spin_unlock(&nfsd_drc_lock);
 829
 830         return num;
 831 }
 832
 833 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
 834 {
 835         int slotsize = slot_bytes(ca);
 836
 837         spin_lock(&nfsd_drc_lock);
 838         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
 839         spin_unlock(&nfsd_drc_lock);
 840 }
 841
 842 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *attrs)
 843 {
 844         int numslots = attrs->maxreqs;
 845         int slotsize = slot_bytes(attrs);
 846         struct nfsd4_session *new;
 847         int mem, i;
 848
 849         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 850                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 851         mem = numslots * sizeof(struct nfsd4_slot *);
 852
 853         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 854         if (!new)
 855                 return NULL;
 856         /* allocate each struct nfsd4_slot and data cache in one piece */
 857         for (i = 0; i < numslots; i++) {
 858                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
 859                 if (!new->se_slots[i])
 860                         goto out_free;
 861         }
 862         return new;
 863 out_free:
 864         while (i--)
 865                 kfree(new->se_slots[i]);
 866         kfree(new);
 867         return NULL;
 868 }
 869
 870 static void free_conn(struct nfsd4_conn *c)
 871 {
 872         svc_xprt_put(c->cn_xprt);
 873         kfree(c);
 874 }
 875
 876 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 877 {
 878         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 879         struct nfs4_client *clp = c->cn_session->se_client;
 880
 881         spin_lock(&clp->cl_lock);
 882         if (!list_empty(&c->cn_persession)) {
 883                 list_del(&c->cn_persession);
 884                 free_conn(c);
 885         }
 886         nfsd4_probe_callback(clp);
 887         spin_unlock(&clp->cl_lock);
 888 }
 889
 890 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 891 {
 892         struct nfsd4_conn *conn;
 893
 894         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 895         if (!conn)
 896                 return NULL;
 897         svc_xprt_get(rqstp->rq_xprt);
 898         conn->cn_xprt = rqstp->rq_xprt;
 899         conn->cn_flags = flags;
 900         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 901         return conn;
 902 }
 903
 904 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 905 {
 906         conn->cn_session = ses;
 907         list_add(&conn->cn_persession, &ses->se_conns);
 908 }
 909
 910 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 911 {
 912         struct nfs4_client *clp = ses->se_client;
 913
 914         spin_lock(&clp->cl_lock);
 915         __nfsd4_hash_conn(conn, ses);
 916         spin_unlock(&clp->cl_lock);
 917 }
 918
 919 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 920 {
 921         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 922         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 923 }
 924
 925 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
 926 {
 927         int ret;
 928
 929         nfsd4_hash_conn(conn, ses);
 930         ret = nfsd4_register_conn(conn);
 931         if (ret)
 932                 /* oops; xprt is already down: */
 933                 nfsd4_conn_lost(&conn->cn_xpt_user);
 934         if (conn->cn_flags & NFS4_CDFC4_BACK) {
 935                 /* callback channel may be back up */
 936                 nfsd4_probe_callback(ses->se_client);
 937         }
 938 }
 939
 940 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
 941 {
 942         u32 dir = NFS4_CDFC4_FORE;
 943
 944         if (cses->flags & SESSION4_BACK_CHAN)
 945                 dir |= NFS4_CDFC4_BACK;
 946         return alloc_conn(rqstp, dir);
 947 }
 948
 949 /* must be called under client_lock */
 950 static void nfsd4_del_conns(struct nfsd4_session *s)
 951 {
 952         struct nfs4_client *clp = s->se_client;
 953         struct nfsd4_conn *c;
 954
 955         spin_lock(&clp->cl_lock);
 956         while (!list_empty(&s->se_conns)) {
 957                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 958                 list_del_init(&c->cn_persession);
 959                 spin_unlock(&clp->cl_lock);
 960
 961                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 962                 free_conn(c);
 963
 964                 spin_lock(&clp->cl_lock);
 965         }
 966         spin_unlock(&clp->cl_lock);
 967 }
 968
 969 static void __free_session(struct nfsd4_session *ses)
 970 {
 971         free_session_slots(ses);
 972         kfree(ses);
 973 }
 974
 975 static void free_session(struct nfsd4_session *ses)
 976 {
 977         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
 978
 979         lockdep_assert_held(&nn->client_lock);
 980         nfsd4_del_conns(ses);
 981         nfsd4_put_drc_mem(&ses->se_fchannel);
 982         __free_session(ses);
 983 }
 984
 985 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 986 {
 987         int idx;
 988         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 989
 990         new->se_client = clp;
 991         gen_sessionid(new);
 992
 993         INIT_LIST_HEAD(&new->se_conns);
 994
 995         new->se_cb_seq_nr = 1;
 996         new->se_flags = cses->flags;
 997         new->se_cb_prog = cses->callback_prog;
 998         new->se_cb_sec = cses->cb_sec;
 999         atomic_set(&new->se_ref, 0);
1000         idx = hash_sessionid(&new->se_sessionid);
1001         spin_lock(&nn->client_lock);
1002         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1003         spin_lock(&clp->cl_lock);
1004         list_add(&new->se_perclnt, &clp->cl_sessions);
1005         spin_unlock(&clp->cl_lock);
1006         spin_unlock(&nn->client_lock);
1007         memcpy(&new->se_fchannel, &cses->fore_channel,
1008                         sizeof(struct nfsd4_channel_attrs));
1009         if (cses->flags & SESSION4_BACK_CHAN) {
1010                 struct sockaddr *sa = svc_addr(rqstp);
1011                 /*
1012                  * This is a little silly; with sessions there's no real
1013                  * use for the callback address.  Use the peer address
1014                  * as a reasonable default for now, but consider fixing
1015                  * the rpc client not to require an address in the
1016                  * future:
1017                  */
1018                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1019                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1020         }
1021 }
1022
1023 /* caller must hold client_lock */
1024 static struct nfsd4_session *
1025 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1026 {
1027         struct nfsd4_session *elem;
1028         int idx;
1029         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1030
1031         dump_sessionid(__func__, sessionid);
1032         idx = hash_sessionid(sessionid);
1033         /* Search in the appropriate list */
1034         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1035                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1036                             NFS4_MAX_SESSIONID_LEN)) {
1037                         return elem;
1038                 }
1039         }
1040
1041         dprintk("%s: session not found\n", __func__);
1042         return NULL;
1043 }
1044
1045 /* caller must hold client_lock */
1046 static void
1047 unhash_session(struct nfsd4_session *ses)
1048 {
1049         list_del(&ses->se_hash);
1050         spin_lock(&ses->se_client->cl_lock);
1051         list_del(&ses->se_perclnt);
1052         spin_unlock(&ses->se_client->cl_lock);
1053 }
1054
1055 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1056 static int
1057 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1058 {
1059         if (clid->cl_boot == nn->boot_time)
1060                 return 0;
1061         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1062                 clid->cl_boot, clid->cl_id, nn->boot_time);
1063         return 1;
1064 }
1065
1066 /*
1067  * XXX Should we use a slab cache ?
1068  * This type of memory management is somewhat inefficient, but we use it
1069  * anyway since SETCLIENTID is not a common operation.
1070  */
1071 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1072 {
1073         struct nfs4_client *clp;
1074
1075         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1076         if (clp == NULL)
1077                 return NULL;
1078         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1079         if (clp->cl_name.data == NULL) {
1080                 kfree(clp);
1081                 return NULL;
1082         }
1083         clp->cl_name.len = name.len;
1084         return clp;
1085 }
1086
1087 static inline void
1088 free_client(struct nfs4_client *clp)
1089 {
1090         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1091
1092         lockdep_assert_held(&nn->client_lock);
1093         while (!list_empty(&clp->cl_sessions)) {
1094                 struct nfsd4_session *ses;
1095                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1096                                 se_perclnt);
1097                 list_del(&ses->se_perclnt);
1098                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1099                 free_session(ses);
1100         }
1101         free_svc_cred(&clp->cl_cred);
1102         kfree(clp->cl_name.data);
1103         idr_destroy(&clp->cl_stateids);
1104         kfree(clp);
1105 }
1106
1107 /* must be called under the client_lock */
1108 static inline void
1109 unhash_client_locked(struct nfs4_client *clp)
1110 {
1111         struct nfsd4_session *ses;
1112
1113         list_del(&clp->cl_lru);
1114         spin_lock(&clp->cl_lock);
1115         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1116                 list_del_init(&ses->se_hash);
1117         spin_unlock(&clp->cl_lock);
1118 }
1119
1120 static void
1121 destroy_client(struct nfs4_client *clp)
1122 {
1123         struct nfs4_openowner *oo;
1124         struct nfs4_delegation *dp;
1125         struct list_head reaplist;
1126         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1127
1128         INIT_LIST_HEAD(&reaplist);
1129         spin_lock(&recall_lock);
1130         while (!list_empty(&clp->cl_delegations)) {
1131                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1132                 list_del_init(&dp->dl_perclnt);
1133                 list_move(&dp->dl_recall_lru, &reaplist);
1134         }
1135         spin_unlock(&recall_lock);
1136         while (!list_empty(&reaplist)) {
1137                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1138                 destroy_delegation(dp);
1139         }
1140         while (!list_empty(&clp->cl_openowners)) {
1141                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1142                 release_openowner(oo);
1143         }
1144         nfsd4_shutdown_callback(clp);
1145         if (clp->cl_cb_conn.cb_xprt)
1146                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1147         list_del(&clp->cl_idhash);
1148         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1149                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1150         else
1151                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1152         spin_lock(&nn->client_lock);
1153         unhash_client_locked(clp);
1154         WARN_ON_ONCE(atomic_read(&clp->cl_refcount));
1155         free_client(clp);
1156         spin_unlock(&nn->client_lock);
1157 }
1158
1159 static void expire_client(struct nfs4_client *clp)
1160 {
1161         nfsd4_client_record_remove(clp);
1162         destroy_client(clp);
1163 }
1164
1165 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1166 {
1167         memcpy(target->cl_verifier.data, source->data,
1168                         sizeof(target->cl_verifier.data));
1169 }
1170
1171 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1172 {
1173         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1174         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1175 }
1176
1177 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1178 {
1179         if (source->cr_principal) {
1180                 target->cr_principal =
1181                                 kstrdup(source->cr_principal, GFP_KERNEL);
1182                 if (target->cr_principal == NULL)
1183                         return -ENOMEM;
1184         } else
1185                 target->cr_principal = NULL;
1186         target->cr_flavor = source->cr_flavor;
1187         target->cr_uid = source->cr_uid;
1188         target->cr_gid = source->cr_gid;
1189         target->cr_group_info = source->cr_group_info;
1190         get_group_info(target->cr_group_info);
1191         target->cr_gss_mech = source->cr_gss_mech;
1192         if (source->cr_gss_mech)
1193                 gss_mech_get(source->cr_gss_mech);
1194         return 0;
1195 }
1196
1197 static long long
1198 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1199 {
1200         long long res;
1201
1202         res = o1->len - o2->len;
1203         if (res)
1204                 return res;
1205         return (long long)memcmp(o1->data, o2->data, o1->len);
1206 }
1207
1208 static int same_name(const char *n1, const char *n2)
1209 {
1210         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1211 }
1212
1213 static int
1214 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1215 {
1216         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1217 }
1218
1219 static int
1220 same_clid(clientid_t *cl1, clientid_t *cl2)
1221 {
1222         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1223 }
1224
1225 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1226 {
1227         int i;
1228
1229         if (g1->ngroups != g2->ngroups)
1230                 return false;
1231         for (i=0; i<g1->ngroups; i++)
1232                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1233                         return false;
1234         return true;
1235 }
1236
1237 /*
1238  * RFC 3530 language requires clid_inuse be returned when the
1239  * "principal" associated with a requests differs from that previously
1240  * used.  We use uid, gid's, and gss principal string as our best
1241  * approximation.  We also don't want to allow non-gss use of a client
1242  * established using gss: in theory cr_principal should catch that
1243  * change, but in practice cr_principal can be null even in the gss case
1244  * since gssd doesn't always pass down a principal string.
1245  */
1246 static bool is_gss_cred(struct svc_cred *cr)
1247 {
1248         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1249         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1250 }
1251
1252
1253 static bool
1254 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1255 {
1256         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1257                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1258                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1259                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1260                 return false;
1261         if (cr1->cr_principal == cr2->cr_principal)
1262                 return true;
1263         if (!cr1->cr_principal || !cr2->cr_principal)
1264                 return false;
1265         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1266 }
1267
1268 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1269 {
1270         static u32 current_clientid = 1;
1271
1272         clp->cl_clientid.cl_boot = nn->boot_time;
1273         clp->cl_clientid.cl_id = current_clientid++;
1274 }
1275
1276 static void gen_confirm(struct nfs4_client *clp)
1277 {
1278         __be32 verf[2];
1279         static u32 i;
1280
1281         verf[0] = (__be32)get_seconds();
1282         verf[1] = (__be32)i++;
1283         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1284 }
1285
1286 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1287 {
1288         struct nfs4_stid *ret;
1289
1290         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1291         if (!ret || !ret->sc_type)
1292                 return NULL;
1293         return ret;
1294 }
1295
1296 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1297 {
1298         struct nfs4_stid *s;
1299
1300         s = find_stateid(cl, t);
1301         if (!s)
1302                 return NULL;
1303         if (typemask & s->sc_type)
1304                 return s;
1305         return NULL;
1306 }
1307
1308 static struct nfs4_client *create_client(struct xdr_netobj name,
1309                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1310 {
1311         struct nfs4_client *clp;
1312         struct sockaddr *sa = svc_addr(rqstp);
1313         int ret;
1314         struct net *net = SVC_NET(rqstp);
1315         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1316
1317         clp = alloc_client(name);
1318         if (clp == NULL)
1319                 return NULL;
1320
1321         INIT_LIST_HEAD(&clp->cl_sessions);
1322         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1323         if (ret) {
1324                 spin_lock(&nn->client_lock);
1325                 free_client(clp);
1326                 spin_unlock(&nn->client_lock);
1327                 return NULL;
1328         }
1329         idr_init(&clp->cl_stateids);
1330         atomic_set(&clp->cl_refcount, 0);
1331         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1332         INIT_LIST_HEAD(&clp->cl_idhash);
1333         INIT_LIST_HEAD(&clp->cl_openowners);
1334         INIT_LIST_HEAD(&clp->cl_delegations);
1335         INIT_LIST_HEAD(&clp->cl_lru);
1336         INIT_LIST_HEAD(&clp->cl_callbacks);
1337         INIT_LIST_HEAD(&clp->cl_revoked);
1338         spin_lock_init(&clp->cl_lock);
1339         nfsd4_init_callback(&clp->cl_cb_null);
1340         clp->cl_time = get_seconds();
1341         clear_bit(0, &clp->cl_cb_slot_busy);
1342         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1343         copy_verf(clp, verf);
1344         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1345         gen_confirm(clp);
1346         clp->cl_cb_session = NULL;
1347         clp->net = net;
1348         return clp;
1349 }
1350
1351 static void
1352 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1353 {
1354         struct rb_node **new = &(root->rb_node), *parent = NULL;
1355         struct nfs4_client *clp;
1356
1357         while (*new) {
1358                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1359                 parent = *new;
1360
1361                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1362                         new = &((*new)->rb_left);
1363                 else
1364                         new = &((*new)->rb_right);
1365         }
1366
1367         rb_link_node(&new_clp->cl_namenode, parent, new);
1368         rb_insert_color(&new_clp->cl_namenode, root);
1369 }
1370
1371 static struct nfs4_client *
1372 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1373 {
1374         long long cmp;
1375         struct rb_node *node = root->rb_node;
1376         struct nfs4_client *clp;
1377
1378         while (node) {
1379                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1380                 cmp = compare_blob(&clp->cl_name, name);
1381                 if (cmp > 0)
1382                         node = node->rb_left;
1383                 else if (cmp < 0)
1384                         node = node->rb_right;
1385                 else
1386                         return clp;
1387         }
1388         return NULL;
1389 }
1390
1391 static void
1392 add_to_unconfirmed(struct nfs4_client *clp)
1393 {
1394         unsigned int idhashval;
1395         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1396
1397         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1398         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1399         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1400         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1401         renew_client(clp);
1402 }
1403
1404 static void
1405 move_to_confirmed(struct nfs4_client *clp)
1406 {
1407         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1408         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1409
1410         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1411         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1412         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1413         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1414         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1415         renew_client(clp);
1416 }
1417
1418 static struct nfs4_client *
1419 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1420 {
1421         struct nfs4_client *clp;
1422         unsigned int idhashval = clientid_hashval(clid->cl_id);
1423
1424         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1425                 if (same_clid(&clp->cl_clientid, clid)) {
1426                         if ((bool)clp->cl_minorversion != sessions)
1427                                 return NULL;
1428                         renew_client(clp);
1429                         return clp;
1430                 }
1431         }
1432         return NULL;
1433 }
1434
1435 static struct nfs4_client *
1436 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1437 {
1438         struct list_head *tbl = nn->conf_id_hashtbl;
1439
1440         return find_client_in_id_table(tbl, clid, sessions);
1441 }
1442
1443 static struct nfs4_client *
1444 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1445 {
1446         struct list_head *tbl = nn->unconf_id_hashtbl;
1447
1448         return find_client_in_id_table(tbl, clid, sessions);
1449 }
1450
1451 static bool clp_used_exchangeid(struct nfs4_client *clp)
1452 {
1453         return clp->cl_exchange_flags != 0;
1454 }
1455
1456 static struct nfs4_client *
1457 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1458 {
1459         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1460 }
1461
1462 static struct nfs4_client *
1463 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1464 {
1465         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1466 }
1467
1468 static void
1469 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1470 {
1471         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1472         struct sockaddr *sa = svc_addr(rqstp);
1473         u32 scopeid = rpc_get_scope_id(sa);
1474         unsigned short expected_family;
1475
1476         /* Currently, we only support tcp and tcp6 for the callback channel */
1477         if (se->se_callback_netid_len == 3 &&
1478             !memcmp(se->se_callback_netid_val, "tcp", 3))
1479                 expected_family = AF_INET;
1480         else if (se->se_callback_netid_len == 4 &&
1481                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1482                 expected_family = AF_INET6;
1483         else
1484                 goto out_err;
1485
1486         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1487                                             se->se_callback_addr_len,
1488                                             (struct sockaddr *)&conn->cb_addr,
1489                                             sizeof(conn->cb_addr));
1490
1491         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1492                 goto out_err;
1493
1494         if (conn->cb_addr.ss_family == AF_INET6)
1495                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1496
1497         conn->cb_prog = se->se_callback_prog;
1498         conn->cb_ident = se->se_callback_ident;
1499         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1500         return;
1501 out_err:
1502         conn->cb_addr.ss_family = AF_UNSPEC;
1503         conn->cb_addrlen = 0;
1504         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1505                 "will not receive delegations\n",
1506                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1507
1508         return;
1509 }
1510
1511 /*
1512  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1513  */
1514 void
1515 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1516 {
1517         struct nfsd4_slot *slot = resp->cstate.slot;
1518         unsigned int base;
1519
1520         dprintk("--> %s slot %p\n", __func__, slot);
1521
1522         slot->sl_opcnt = resp->opcnt;
1523         slot->sl_status = resp->cstate.status;
1524
1525         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1526         if (nfsd4_not_cached(resp)) {
1527                 slot->sl_datalen = 0;
1528                 return;
1529         }
1530         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1531         base = (char *)resp->cstate.datap -
1532                                         (char *)resp->xbuf->head[0].iov_base;
1533         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1534                                     slot->sl_datalen))
1535                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1536         return;
1537 }
1538
1539 /*
1540  * Encode the replay sequence operation from the slot values.
1541  * If cachethis is FALSE encode the uncached rep error on the next
1542  * operation which sets resp->p and increments resp->opcnt for
1543  * nfs4svc_encode_compoundres.
1544  *
1545  */
1546 static __be32
1547 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1548                           struct nfsd4_compoundres *resp)
1549 {
1550         struct nfsd4_op *op;
1551         struct nfsd4_slot *slot = resp->cstate.slot;
1552
1553         /* Encode the replayed sequence operation */
1554         op = &args->ops[resp->opcnt - 1];
1555         nfsd4_encode_operation(resp, op);
1556
1557         /* Return nfserr_retry_uncached_rep in next operation. */
1558         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1559                 op = &args->ops[resp->opcnt++];
1560                 op->status = nfserr_retry_uncached_rep;
1561                 nfsd4_encode_operation(resp, op);
1562         }
1563         return op->status;
1564 }
1565
1566 /*
1567  * The sequence operation is not cached because we can use the slot and
1568  * session values.
1569  */
1570 __be32
1571 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1572                          struct nfsd4_sequence *seq)
1573 {
1574         struct nfsd4_slot *slot = resp->cstate.slot;
1575         __be32 status;
1576
1577         dprintk("--> %s slot %p\n", __func__, slot);
1578
1579         /* Either returns 0 or nfserr_retry_uncached */
1580         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1581         if (status == nfserr_retry_uncached_rep)
1582                 return status;
1583
1584         /* The sequence operation has been encoded, cstate->datap set. */
1585         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1586
1587         resp->opcnt = slot->sl_opcnt;
1588         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1589         status = slot->sl_status;
1590
1591         return status;
1592 }
1593
1594 /*
1595  * Set the exchange_id flags returned by the server.
1596  */
1597 static void
1598 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1599 {
1600         /* pNFS is not supported */
1601         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1602
1603         /* Referrals are supported, Migration is not. */
1604         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1605
1606         /* set the wire flags to return to client. */
1607         clid->flags = new->cl_exchange_flags;
1608 }
1609
1610 static bool client_has_state(struct nfs4_client *clp)
1611 {
1612         /*
1613          * Note clp->cl_openowners check isn't quite right: there's no
1614          * need to count owners without stateid's.
1615          *
1616          * Also note we should probably be using this in 4.0 case too.
1617          */
1618         return !list_empty(&clp->cl_openowners)
1619                 || !list_empty(&clp->cl_delegations)
1620                 || !list_empty(&clp->cl_sessions);
1621 }
1622
1623 __be32
1624 nfsd4_exchange_id(struct svc_rqst *rqstp,
1625                   struct nfsd4_compound_state *cstate,
1626                   struct nfsd4_exchange_id *exid)
1627 {
1628         struct nfs4_client *unconf, *conf, *new;
1629         __be32 status;
1630         char                    addr_str[INET6_ADDRSTRLEN];
1631         nfs4_verifier           verf = exid->verifier;
1632         struct sockaddr         *sa = svc_addr(rqstp);
1633         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1634         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1635
1636         rpc_ntop(sa, addr_str, sizeof(addr_str));
1637         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1638                 "ip_addr=%s flags %x, spa_how %d\n",
1639                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1640                 addr_str, exid->flags, exid->spa_how);
1641
1642         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1643                 return nfserr_inval;
1644
1645         /* Currently only support SP4_NONE */
1646         switch (exid->spa_how) {
1647         case SP4_NONE:
1648                 break;
1649         default:                                /* checked by xdr code */
1650                 WARN_ON_ONCE(1);
1651         case SP4_SSV:
1652                 return nfserr_encr_alg_unsupp;
1653         case SP4_MACH_CRED:
1654                 return nfserr_serverfault;      /* no excuse :-/ */
1655         }
1656
1657         /* Cases below refer to rfc 5661 section 18.35.4: */
1658         nfs4_lock_state();
1659         conf = find_confirmed_client_by_name(&exid->clname, nn);
1660         if (conf) {
1661                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1662                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1663
1664                 if (update) {
1665                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1666                                 status = nfserr_inval;
1667                                 goto out;
1668                         }
1669                         if (!creds_match) { /* case 9 */
1670                                 status = nfserr_perm;
1671                                 goto out;
1672                         }
1673                         if (!verfs_match) { /* case 8 */
1674                                 status = nfserr_not_same;
1675                                 goto out;
1676                         }
1677                         /* case 6 */
1678                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1679                         new = conf;
1680                         goto out_copy;
1681                 }
1682                 if (!creds_match) { /* case 3 */
1683                         if (client_has_state(conf)) {
1684                                 status = nfserr_clid_inuse;
1685                                 goto out;
1686                         }
1687                         expire_client(conf);
1688                         goto out_new;
1689                 }
1690                 if (verfs_match) { /* case 2 */
1691                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1692                         new = conf;
1693                         goto out_copy;
1694                 }
1695                 /* case 5, client reboot */
1696                 goto out_new;
1697         }
1698
1699         if (update) { /* case 7 */
1700                 status = nfserr_noent;
1701                 goto out;
1702         }
1703
1704         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1705         if (unconf) /* case 4, possible retry or client restart */
1706                 expire_client(unconf);
1707
1708         /* case 1 (normal case) */
1709 out_new:
1710         new = create_client(exid->clname, rqstp, &verf);
1711         if (new == NULL) {
1712                 status = nfserr_jukebox;
1713                 goto out;
1714         }
1715         new->cl_minorversion = cstate->minorversion;
1716
1717         gen_clid(new, nn);
1718         add_to_unconfirmed(new);
1719 out_copy:
1720         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1721         exid->clientid.cl_id = new->cl_clientid.cl_id;
1722
1723         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1724         nfsd4_set_ex_flags(new, exid);
1725
1726         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1727                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1728         status = nfs_ok;
1729
1730 out:
1731         nfs4_unlock_state();
1732         return status;
1733 }
1734
1735 static __be32
1736 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1737 {
1738         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1739                 slot_seqid);
1740
1741         /* The slot is in use, and no response has been sent. */
1742         if (slot_inuse) {
1743                 if (seqid == slot_seqid)
1744                         return nfserr_jukebox;
1745                 else
1746                         return nfserr_seq_misordered;
1747         }
1748         /* Note unsigned 32-bit arithmetic handles wraparound: */
1749         if (likely(seqid == slot_seqid + 1))
1750                 return nfs_ok;
1751         if (seqid == slot_seqid)
1752                 return nfserr_replay_cache;
1753         return nfserr_seq_misordered;
1754 }
1755
1756 /*
1757  * Cache the create session result into the create session single DRC
1758  * slot cache by saving the xdr structure. sl_seqid has been set.
1759  * Do this for solo or embedded create session operations.
1760  */
1761 static void
1762 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1763                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1764 {
1765         slot->sl_status = nfserr;
1766         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1767 }
1768
1769 static __be32
1770 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1771                             struct nfsd4_clid_slot *slot)
1772 {
1773         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1774         return slot->sl_status;
1775 }
1776
1777 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1778                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1779                         1 +     /* MIN tag is length with zero, only length */ \
1780                         3 +     /* version, opcount, opcode */ \
1781                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1782                                 /* seqid, slotID, slotID, cache */ \
1783                         4 ) * sizeof(__be32))
1784
1785 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1786                         2 +     /* verifier: AUTH_NULL, length 0 */\
1787                         1 +     /* status */ \
1788                         1 +     /* MIN tag is length with zero, only length */ \
1789                         3 +     /* opcount, opcode, opstatus*/ \
1790                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1791                                 /* seqid, slotID, slotID, slotID, status */ \
1792                         5 ) * sizeof(__be32))
1793
1794 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1795 {
1796         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
1797
1798         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
1799                 return nfserr_toosmall;
1800         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
1801                 return nfserr_toosmall;
1802         ca->headerpadsz = 0;
1803         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
1804         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
1805         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
1806         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
1807                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
1808         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
1809         /*
1810          * Note decreasing slot size below client's request may make it
1811          * difficult for client to function correctly, whereas
1812          * decreasing the number of slots will (just?) affect
1813          * performance.  When short on memory we therefore prefer to
1814          * decrease number of slots instead of their size.  Clients that
1815          * request larger slots than they need will get poor results:
1816          */
1817         ca->maxreqs = nfsd4_get_drc_mem(ca);
1818         if (!ca->maxreqs)
1819                 return nfserr_jukebox;
1820
1821         return nfs_ok;
1822 }
1823
1824 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
1825 {
1826         ca->headerpadsz = 0;
1827
1828         /*
1829          * These RPC_MAX_HEADER macros are overkill, especially since we
1830          * don't even do gss on the backchannel yet.  But this is still
1831          * less than 1k.  Tighten up this estimate in the unlikely event
1832          * it turns out to be a problem for some client:
1833          */
1834         if (ca->maxreq_sz < NFS4_enc_cb_recall_sz + RPC_MAX_HEADER_WITH_AUTH)
1835                 return nfserr_toosmall;
1836         if (ca->maxresp_sz < NFS4_dec_cb_recall_sz + RPC_MAX_REPHEADER_WITH_AUTH)
1837                 return nfserr_toosmall;
1838         ca->maxresp_cached = 0;
1839         if (ca->maxops < 2)
1840                 return nfserr_toosmall;
1841
1842         return nfs_ok;
1843 }
1844
1845 __be32
1846 nfsd4_create_session(struct svc_rqst *rqstp,
1847                      struct nfsd4_compound_state *cstate,
1848                      struct nfsd4_create_session *cr_ses)
1849 {
1850         struct sockaddr *sa = svc_addr(rqstp);
1851         struct nfs4_client *conf, *unconf;
1852         struct nfsd4_session *new;
1853         struct nfsd4_conn *conn;
1854         struct nfsd4_clid_slot *cs_slot = NULL;
1855         __be32 status = 0;
1856         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1857
1858         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1859                 return nfserr_inval;
1860         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
1861         if (status)
1862                 return status;
1863         status = check_backchannel_attrs(&cr_ses->back_channel);
1864         if (status)
1865                 return status;
1866         status = nfserr_jukebox;
1867         new = alloc_session(&cr_ses->fore_channel);
1868         if (!new)
1869                 goto out_release_drc_mem;
1870         conn = alloc_conn_from_crses(rqstp, cr_ses);
1871         if (!conn)
1872                 goto out_free_session;
1873
1874         nfs4_lock_state();
1875         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1876         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1877         WARN_ON_ONCE(conf && unconf);
1878
1879         if (conf) {
1880                 cs_slot = &conf->cl_cs_slot;
1881                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1882                 if (status == nfserr_replay_cache) {
1883                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1884                         goto out_free_conn;
1885                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1886                         status = nfserr_seq_misordered;
1887                         goto out_free_conn;
1888                 }
1889         } else if (unconf) {
1890                 struct nfs4_client *old;
1891                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1892                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1893                         status = nfserr_clid_inuse;
1894                         goto out_free_conn;
1895                 }
1896                 cs_slot = &unconf->cl_cs_slot;
1897                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1898                 if (status) {
1899                         /* an unconfirmed replay returns misordered */
1900                         status = nfserr_seq_misordered;
1901                         goto out_free_conn;
1902                 }
1903                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1904                 if (old) {
1905                         status = mark_client_expired(old);
1906                         if (status)
1907                                 goto out_free_conn;
1908                         expire_client(old);
1909                 }
1910                 move_to_confirmed(unconf);
1911                 conf = unconf;
1912         } else {
1913                 status = nfserr_stale_clientid;
1914                 goto out_free_conn;
1915         }
1916         status = nfs_ok;
1917         /*
1918          * We do not support RDMA or persistent sessions
1919          */
1920         cr_ses->flags &= ~SESSION4_PERSIST;
1921         cr_ses->flags &= ~SESSION4_RDMA;
1922
1923         init_session(rqstp, new, conf, cr_ses);
1924         nfsd4_init_conn(rqstp, conn, new);
1925
1926         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1927                NFS4_MAX_SESSIONID_LEN);
1928         cs_slot->sl_seqid++;
1929         cr_ses->seqid = cs_slot->sl_seqid;
1930
1931         /* cache solo and embedded create sessions under the state lock */
1932         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1933         nfs4_unlock_state();
1934         return status;
1935 out_free_conn:
1936         nfs4_unlock_state();
1937         free_conn(conn);
1938 out_free_session:
1939         __free_session(new);
1940 out_release_drc_mem:
1941         nfsd4_put_drc_mem(&cr_ses->fore_channel);
1942         return status;
1943 }
1944
1945 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1946 {
1947         switch (*dir) {
1948         case NFS4_CDFC4_FORE:
1949         case NFS4_CDFC4_BACK:
1950                 return nfs_ok;
1951         case NFS4_CDFC4_FORE_OR_BOTH:
1952         case NFS4_CDFC4_BACK_OR_BOTH:
1953                 *dir = NFS4_CDFC4_BOTH;
1954                 return nfs_ok;
1955         };
1956         return nfserr_inval;
1957 }
1958
1959 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1960 {
1961         struct nfsd4_session *session = cstate->session;
1962         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1963
1964         spin_lock(&nn->client_lock);
1965         session->se_cb_prog = bc->bc_cb_program;
1966         session->se_cb_sec = bc->bc_cb_sec;
1967         spin_unlock(&nn->client_lock);
1968
1969         nfsd4_probe_callback(session->se_client);
1970
1971         return nfs_ok;
1972 }
1973
1974 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1975                      struct nfsd4_compound_state *cstate,
1976                      struct nfsd4_bind_conn_to_session *bcts)
1977 {
1978         __be32 status;
1979         struct nfsd4_conn *conn;
1980         struct nfsd4_session *session;
1981         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1982
1983         if (!nfsd4_last_compound_op(rqstp))
1984                 return nfserr_not_only_op;
1985         nfs4_lock_state();
1986         spin_lock(&nn->client_lock);
1987         session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
1988         spin_unlock(&nn->client_lock);
1989         status = nfserr_badsession;
1990         if (!session)
1991                 goto out;
1992         status = nfsd4_map_bcts_dir(&bcts->dir);
1993         if (status)
1994                 goto out;
1995         conn = alloc_conn(rqstp, bcts->dir);
1996         status = nfserr_jukebox;
1997         if (!conn)
1998                 goto out;
1999         nfsd4_init_conn(rqstp, conn, session);
2000         status = nfs_ok;
2001 out:
2002         nfs4_unlock_state();
2003         return status;
2004 }
2005
2006 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2007 {
2008         if (!session)
2009                 return 0;
2010         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2011 }
2012
2013 __be32
2014 nfsd4_destroy_session(struct svc_rqst *r,
2015                       struct nfsd4_compound_state *cstate,
2016                       struct nfsd4_destroy_session *sessionid)
2017 {
2018         struct nfsd4_session *ses;
2019         __be32 status;
2020         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
2021
2022         nfs4_lock_state();
2023         status = nfserr_not_only_op;
2024         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2025                 if (!nfsd4_last_compound_op(r))
2026                         goto out;
2027         }
2028         dump_sessionid(__func__, &sessionid->sessionid);
2029         spin_lock(&nn->client_lock);
2030         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
2031         status = nfserr_badsession;
2032         if (!ses)
2033                 goto out_client_lock;
2034         status = mark_session_dead_locked(ses);
2035         if (status)
2036                 goto out_client_lock;
2037         unhash_session(ses);
2038         spin_unlock(&nn->client_lock);
2039
2040         nfsd4_probe_callback_sync(ses->se_client);
2041
2042         spin_lock(&nn->client_lock);
2043         free_session(ses);
2044         status = nfs_ok;
2045 out_client_lock:
2046         spin_unlock(&nn->client_lock);
2047 out:
2048         nfs4_unlock_state();
2049         return status;
2050 }
2051
2052 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2053 {
2054         struct nfsd4_conn *c;
2055
2056         list_for_each_entry(c, &s->se_conns, cn_persession) {
2057                 if (c->cn_xprt == xpt) {
2058                         return c;
2059                 }
2060         }
2061         return NULL;
2062 }
2063
2064 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2065 {
2066         struct nfs4_client *clp = ses->se_client;
2067         struct nfsd4_conn *c;
2068         int ret;
2069
2070         spin_lock(&clp->cl_lock);
2071         c = __nfsd4_find_conn(new->cn_xprt, ses);
2072         if (c) {
2073                 spin_unlock(&clp->cl_lock);
2074                 free_conn(new);
2075                 return;
2076         }
2077         __nfsd4_hash_conn(new, ses);
2078         spin_unlock(&clp->cl_lock);
2079         ret = nfsd4_register_conn(new);
2080         if (ret)
2081                 /* oops; xprt is already down: */
2082                 nfsd4_conn_lost(&new->cn_xpt_user);
2083         return;
2084 }
2085
2086 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2087 {
2088         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2089
2090         return args->opcnt > session->se_fchannel.maxops;
2091 }
2092
2093 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2094                                   struct nfsd4_session *session)
2095 {
2096         struct xdr_buf *xb = &rqstp->rq_arg;
2097
2098         return xb->len > session->se_fchannel.maxreq_sz;
2099 }
2100
2101 __be32
2102 nfsd4_sequence(struct svc_rqst *rqstp,
2103                struct nfsd4_compound_state *cstate,
2104                struct nfsd4_sequence *seq)
2105 {
2106         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2107         struct nfsd4_session *session;
2108         struct nfs4_client *clp;
2109         struct nfsd4_slot *slot;
2110         struct nfsd4_conn *conn;
2111         __be32 status;
2112         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2113
2114         if (resp->opcnt != 1)
2115                 return nfserr_sequence_pos;
2116
2117         /*
2118          * Will be either used or freed by nfsd4_sequence_check_conn
2119          * below.
2120          */
2121         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2122         if (!conn)
2123                 return nfserr_jukebox;
2124
2125         spin_lock(&nn->client_lock);
2126         status = nfserr_badsession;
2127         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2128         if (!session)
2129                 goto out_no_session;
2130         clp = session->se_client;
2131         status = get_client_locked(clp);
2132         if (status)
2133                 goto out_no_session;
2134         status = nfsd4_get_session_locked(session);
2135         if (status)
2136                 goto out_put_client;
2137
2138         status = nfserr_too_many_ops;
2139         if (nfsd4_session_too_many_ops(rqstp, session))
2140                 goto out_put_session;
2141
2142         status = nfserr_req_too_big;
2143         if (nfsd4_request_too_big(rqstp, session))
2144                 goto out_put_session;
2145
2146         status = nfserr_badslot;
2147         if (seq->slotid >= session->se_fchannel.maxreqs)
2148                 goto out_put_session;
2149
2150         slot = session->se_slots[seq->slotid];
2151         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2152
2153         /* We do not negotiate the number of slots yet, so set the
2154          * maxslots to the session maxreqs which is used to encode
2155          * sr_highest_slotid and the sr_target_slot id to maxslots */
2156         seq->maxslots = session->se_fchannel.maxreqs;
2157
2158         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2159                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2160         if (status == nfserr_replay_cache) {
2161                 status = nfserr_seq_misordered;
2162                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2163                         goto out_put_session;
2164                 cstate->slot = slot;
2165                 cstate->session = session;
2166                 /* Return the cached reply status and set cstate->status
2167                  * for nfsd4_proc_compound processing */
2168                 status = nfsd4_replay_cache_entry(resp, seq);
2169                 cstate->status = nfserr_replay_cache;
2170                 goto out;
2171         }
2172         if (status)
2173                 goto out_put_session;
2174
2175         nfsd4_sequence_check_conn(conn, session);
2176         conn = NULL;
2177
2178         /* Success! bump slot seqid */
2179         slot->sl_seqid = seq->seqid;
2180         slot->sl_flags |= NFSD4_SLOT_INUSE;
2181         if (seq->cachethis)
2182                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2183         else
2184                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2185
2186         cstate->slot = slot;
2187         cstate->session = session;
2188
2189 out:
2190         switch (clp->cl_cb_state) {
2191         case NFSD4_CB_DOWN:
2192                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2193                 break;
2194         case NFSD4_CB_FAULT:
2195                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2196                 break;
2197         default:
2198                 seq->status_flags = 0;
2199         }
2200         if (!list_empty(&clp->cl_revoked))
2201                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2202 out_no_session:
2203         kfree(conn);
2204         spin_unlock(&nn->client_lock);
2205         return status;
2206 out_put_session:
2207         nfsd4_put_session(session);
2208 out_put_client:
2209         put_client_renew_locked(clp);
2210         goto out_no_session;
2211 }
2212
2213 __be32
2214 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2215 {
2216         struct nfs4_client *conf, *unconf, *clp;
2217         __be32 status = 0;
2218         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2219
2220         nfs4_lock_state();
2221         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2222         conf = find_confirmed_client(&dc->clientid, true, nn);
2223         WARN_ON_ONCE(conf && unconf);
2224
2225         if (conf) {
2226                 clp = conf;
2227
2228                 if (client_has_state(conf)) {
2229                         status = nfserr_clientid_busy;
2230                         goto out;
2231                 }
2232         } else if (unconf)
2233                 clp = unconf;
2234         else {
2235                 status = nfserr_stale_clientid;
2236                 goto out;
2237         }
2238
2239         expire_client(clp);
2240 out:
2241         nfs4_unlock_state();
2242         return status;
2243 }
2244
2245 __be32
2246 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2247 {
2248         __be32 status = 0;
2249
2250         if (rc->rca_one_fs) {
2251                 if (!cstate->current_fh.fh_dentry)
2252                         return nfserr_nofilehandle;
2253                 /*
2254                  * We don't take advantage of the rca_one_fs case.
2255                  * That's OK, it's optional, we can safely ignore it.
2256                  */
2257                  return nfs_ok;
2258         }
2259
2260         nfs4_lock_state();
2261         status = nfserr_complete_already;
2262         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2263                              &cstate->session->se_client->cl_flags))
2264                 goto out;
2265
2266         status = nfserr_stale_clientid;
2267         if (is_client_expired(cstate->session->se_client))
2268                 /*
2269                  * The following error isn't really legal.
2270                  * But we only get here if the client just explicitly
2271                  * destroyed the client.  Surely it no longer cares what
2272                  * error it gets back on an operation for the dead
2273                  * client.
2274                  */
2275                 goto out;
2276
2277         status = nfs_ok;
2278         nfsd4_client_record_create(cstate->session->se_client);
2279 out:
2280         nfs4_unlock_state();
2281         return status;
2282 }
2283
2284 __be32
2285 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2286                   struct nfsd4_setclientid *setclid)
2287 {
2288         struct xdr_netobj       clname = setclid->se_name;
2289         nfs4_verifier           clverifier = setclid->se_verf;
2290         struct nfs4_client      *conf, *unconf, *new;
2291         __be32                  status;
2292         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2293
2294         /* Cases below refer to rfc 3530 section 14.2.33: */
2295         nfs4_lock_state();
2296         conf = find_confirmed_client_by_name(&clname, nn);
2297         if (conf) {
2298                 /* case 0: */
2299                 status = nfserr_clid_inuse;
2300                 if (clp_used_exchangeid(conf))
2301                         goto out;
2302                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2303                         char addr_str[INET6_ADDRSTRLEN];
2304                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2305                                  sizeof(addr_str));
2306                         dprintk("NFSD: setclientid: string in use by client "
2307                                 "at %s\n", addr_str);
2308                         goto out;
2309                 }
2310         }
2311         unconf = find_unconfirmed_client_by_name(&clname, nn);
2312         if (unconf)
2313                 expire_client(unconf);
2314         status = nfserr_jukebox;
2315         new = create_client(clname, rqstp, &clverifier);
2316         if (new == NULL)
2317                 goto out;
2318         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2319                 /* case 1: probable callback update */
2320                 copy_clid(new, conf);
2321         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2322                 gen_clid(new, nn);
2323         new->cl_minorversion = 0;
2324         gen_callback(new, setclid, rqstp);
2325         add_to_unconfirmed(new);
2326         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2327         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2328         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2329         status = nfs_ok;
2330 out:
2331         nfs4_unlock_state();
2332         return status;
2333 }
2334
2335
2336 __be32
2337 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2338                          struct nfsd4_compound_state *cstate,
2339                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2340 {
2341         struct nfs4_client *conf, *unconf;
2342         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2343         clientid_t * clid = &setclientid_confirm->sc_clientid;
2344         __be32 status;
2345         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2346
2347         if (STALE_CLIENTID(clid, nn))
2348                 return nfserr_stale_clientid;
2349         nfs4_lock_state();
2350
2351         conf = find_confirmed_client(clid, false, nn);
2352         unconf = find_unconfirmed_client(clid, false, nn);
2353         /*
2354          * We try hard to give out unique clientid's, so if we get an
2355          * attempt to confirm the same clientid with a different cred,
2356          * there's a bug somewhere.  Let's charitably assume it's our
2357          * bug.
2358          */
2359         status = nfserr_serverfault;
2360         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2361                 goto out;
2362         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2363                 goto out;
2364         /* cases below refer to rfc 3530 section 14.2.34: */
2365         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2366                 if (conf && !unconf) /* case 2: probable retransmit */
2367                         status = nfs_ok;
2368                 else /* case 4: client hasn't noticed we rebooted yet? */
2369                         status = nfserr_stale_clientid;
2370                 goto out;
2371         }
2372         status = nfs_ok;
2373         if (conf) { /* case 1: callback update */
2374                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2375                 nfsd4_probe_callback(conf);
2376                 expire_client(unconf);
2377         } else { /* case 3: normal case; new or rebooted client */
2378                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2379                 if (conf) {
2380                         status = mark_client_expired(conf);
2381                         if (status)
2382                                 goto out;
2383                         expire_client(conf);
2384                 }
2385                 move_to_confirmed(unconf);
2386                 nfsd4_probe_callback(unconf);
2387         }
2388 out:
2389         nfs4_unlock_state();
2390         return status;
2391 }
2392
2393 static struct nfs4_file *nfsd4_alloc_file(void)
2394 {
2395         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2396 }
2397
2398 /* OPEN Share state helper functions */
2399 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2400 {
2401         unsigned int hashval = file_hashval(ino);
2402
2403         atomic_set(&fp->fi_ref, 1);
2404         INIT_LIST_HEAD(&fp->fi_stateids);
2405         INIT_LIST_HEAD(&fp->fi_delegations);
2406         fp->fi_inode = igrab(ino);
2407         fp->fi_had_conflict = false;
2408         fp->fi_lease = NULL;
2409         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2410         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2411         spin_lock(&recall_lock);
2412         hlist_add_head(&fp->fi_hash, &file_hashtbl[hashval]);
2413         spin_unlock(&recall_lock);
2414 }
2415
2416 static void
2417 nfsd4_free_slab(struct kmem_cache **slab)
2418 {
2419         if (*slab == NULL)
2420                 return;
2421         kmem_cache_destroy(*slab);
2422         *slab = NULL;
2423 }
2424
2425 void
2426 nfsd4_free_slabs(void)
2427 {
2428         nfsd4_free_slab(&openowner_slab);
2429         nfsd4_free_slab(&lockowner_slab);
2430         nfsd4_free_slab(&file_slab);
2431         nfsd4_free_slab(&stateid_slab);
2432         nfsd4_free_slab(&deleg_slab);
2433 }
2434
2435 int
2436 nfsd4_init_slabs(void)
2437 {
2438         openowner_slab = kmem_cache_create("nfsd4_openowners",
2439                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2440         if (openowner_slab == NULL)
2441                 goto out_nomem;
2442         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2443                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2444         if (lockowner_slab == NULL)
2445                 goto out_nomem;
2446         file_slab = kmem_cache_create("nfsd4_files",
2447                         sizeof(struct nfs4_file), 0, 0, NULL);
2448         if (file_slab == NULL)
2449                 goto out_nomem;
2450         stateid_slab = kmem_cache_create("nfsd4_stateids",
2451                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2452         if (stateid_slab == NULL)
2453                 goto out_nomem;
2454         deleg_slab = kmem_cache_create("nfsd4_delegations",
2455                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2456         if (deleg_slab == NULL)
2457                 goto out_nomem;
2458         return 0;
2459 out_nomem:
2460         nfsd4_free_slabs();
2461         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2462         return -ENOMEM;
2463 }
2464
2465 void nfs4_free_openowner(struct nfs4_openowner *oo)
2466 {
2467         kfree(oo->oo_owner.so_owner.data);
2468         kmem_cache_free(openowner_slab, oo);
2469 }
2470
2471 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2472 {
2473         kfree(lo->lo_owner.so_owner.data);
2474         kmem_cache_free(lockowner_slab, lo);
2475 }
2476
2477 static void init_nfs4_replay(struct nfs4_replay *rp)
2478 {
2479         rp->rp_status = nfserr_serverfault;
2480         rp->rp_buflen = 0;
2481         rp->rp_buf = rp->rp_ibuf;
2482 }
2483
2484 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2485 {
2486         struct nfs4_stateowner *sop;
2487
2488         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2489         if (!sop)
2490                 return NULL;
2491
2492         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2493         if (!sop->so_owner.data) {
2494                 kmem_cache_free(slab, sop);
2495                 return NULL;
2496         }
2497         sop->so_owner.len = owner->len;
2498
2499         INIT_LIST_HEAD(&sop->so_stateids);
2500         sop->so_client = clp;
2501         init_nfs4_replay(&sop->so_replay);
2502         return sop;
2503 }
2504
2505 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2506 {
2507         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2508
2509         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2510         list_add(&oo->oo_perclient, &clp->cl_openowners);
2511 }
2512
2513 static struct nfs4_openowner *
2514 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2515         struct nfs4_openowner *oo;
2516
2517         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2518         if (!oo)
2519                 return NULL;
2520         oo->oo_owner.so_is_open_owner = 1;
2521         oo->oo_owner.so_seqid = open->op_seqid;
2522         oo->oo_flags = NFS4_OO_NEW;
2523         oo->oo_time = 0;
2524         oo->oo_last_closed_stid = NULL;
2525         INIT_LIST_HEAD(&oo->oo_close_lru);
2526         hash_openowner(oo, clp, strhashval);
2527         return oo;
2528 }
2529
2530 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2531         struct nfs4_openowner *oo = open->op_openowner;
2532
2533         stp->st_stid.sc_type = NFS4_OPEN_STID;
2534         INIT_LIST_HEAD(&stp->st_lockowners);
2535         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2536         list_add(&stp->st_perfile, &fp->fi_stateids);
2537         stp->st_stateowner = &oo->oo_owner;
2538         get_nfs4_file(fp);
2539         stp->st_file = fp;
2540         stp->st_access_bmap = 0;
2541         stp->st_deny_bmap = 0;
2542         set_access(open->op_share_access, stp);
2543         set_deny(open->op_share_deny, stp);
2544         stp->st_openstp = NULL;
2545 }
2546
2547 static void
2548 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2549 {
2550         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2551
2552         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2553
2554         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2555         oo->oo_time = get_seconds();
2556 }
2557
2558 static int
2559 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2560                                                         clientid_t *clid)
2561 {
2562         return (sop->so_owner.len == owner->len) &&
2563                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2564                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2565 }
2566
2567 static struct nfs4_openowner *
2568 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2569                         bool sessions, struct nfsd_net *nn)
2570 {
2571         struct nfs4_stateowner *so;
2572         struct nfs4_openowner *oo;
2573         struct nfs4_client *clp;
2574
2575         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2576                 if (!so->so_is_open_owner)
2577                         continue;
2578                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2579                         oo = openowner(so);
2580                         clp = oo->oo_owner.so_client;
2581                         if ((bool)clp->cl_minorversion != sessions)
2582                                 return NULL;
2583                         renew_client(oo->oo_owner.so_client);
2584                         return oo;
2585                 }
2586         }
2587         return NULL;
2588 }
2589
2590 /* search file_hashtbl[] for file */
2591 static struct nfs4_file *
2592 find_file(struct inode *ino)
2593 {
2594         unsigned int hashval = file_hashval(ino);
2595         struct nfs4_file *fp;
2596
2597         spin_lock(&recall_lock);
2598         hlist_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2599                 if (fp->fi_inode == ino) {
2600                         get_nfs4_file(fp);
2601                         spin_unlock(&recall_lock);
2602                         return fp;
2603                 }
2604         }
2605         spin_unlock(&recall_lock);
2606         return NULL;
2607 }
2608
2609 /*
2610  * Called to check deny when READ with all zero stateid or
2611  * WRITE with all zero or all one stateid
2612  */
2613 static __be32
2614 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2615 {
2616         struct inode *ino = current_fh->fh_dentry->d_inode;
2617         struct nfs4_file *fp;
2618         struct nfs4_ol_stateid *stp;
2619         __be32 ret;
2620
2621         fp = find_file(ino);
2622         if (!fp)
2623                 return nfs_ok;
2624         ret = nfserr_locked;
2625         /* Search for conflicting share reservations */
2626         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2627                 if (test_deny(deny_type, stp) ||
2628                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2629                         goto out;
2630         }
2631         ret = nfs_ok;
2632 out:
2633         put_nfs4_file(fp);
2634         return ret;
2635 }
2636
2637 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2638 {
2639         struct nfs4_client *clp = dp->dl_stid.sc_client;
2640         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2641
2642         /* We're assuming the state code never drops its reference
2643          * without first removing the lease.  Since we're in this lease
2644          * callback (and since the lease code is serialized by the kernel
2645          * lock) we know the server hasn't removed the lease yet, we know
2646          * it's safe to take a reference: */
2647         atomic_inc(&dp->dl_count);
2648
2649         list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
2650
2651         /* only place dl_time is set. protected by lock_flocks*/
2652         dp->dl_time = get_seconds();
2653
2654         nfsd4_cb_recall(dp);
2655 }
2656
2657 /* Called from break_lease() with lock_flocks() held. */
2658 static void nfsd_break_deleg_cb(struct file_lock *fl)
2659 {
2660         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2661         struct nfs4_delegation *dp;
2662
2663         if (!fp) {
2664                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2665                 return;
2666         }
2667         if (fp->fi_had_conflict) {
2668                 WARN(1, "duplicate break on %p\n", fp);
2669                 return;
2670         }
2671         /*
2672          * We don't want the locks code to timeout the lease for us;
2673          * we'll remove it ourself if a delegation isn't returned
2674          * in time:
2675          */
2676         fl->fl_break_time = 0;
2677
2678         spin_lock(&recall_lock);
2679         fp->fi_had_conflict = true;
2680         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2681                 nfsd_break_one_deleg(dp);
2682         spin_unlock(&recall_lock);
2683 }
2684
2685 static
2686 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2687 {
2688         if (arg & F_UNLCK)
2689                 return lease_modify(onlist, arg);
2690         else
2691                 return -EAGAIN;
2692 }
2693
2694 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2695         .lm_break = nfsd_break_deleg_cb,
2696         .lm_change = nfsd_change_deleg_cb,
2697 };
2698
2699 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2700 {
2701         if (nfsd4_has_session(cstate))
2702                 return nfs_ok;
2703         if (seqid == so->so_seqid - 1)
2704                 return nfserr_replay_me;
2705         if (seqid == so->so_seqid)
2706                 return nfs_ok;
2707         return nfserr_bad_seqid;
2708 }
2709
2710 __be32
2711 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2712                     struct nfsd4_open *open, struct nfsd_net *nn)
2713 {
2714         clientid_t *clientid = &open->op_clientid;
2715         struct nfs4_client *clp = NULL;
2716         unsigned int strhashval;
2717         struct nfs4_openowner *oo = NULL;
2718         __be32 status;
2719
2720         if (STALE_CLIENTID(&open->op_clientid, nn))
2721                 return nfserr_stale_clientid;
2722         /*
2723          * In case we need it later, after we've already created the
2724          * file and don't want to risk a further failure:
2725          */
2726         open->op_file = nfsd4_alloc_file();
2727         if (open->op_file == NULL)
2728                 return nfserr_jukebox;
2729
2730         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2731         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2732         open->op_openowner = oo;
2733         if (!oo) {
2734                 clp = find_confirmed_client(clientid, cstate->minorversion,
2735                                             nn);
2736                 if (clp == NULL)
2737                         return nfserr_expired;
2738                 goto new_owner;
2739         }
2740         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2741                 /* Replace unconfirmed owners without checking for replay. */
2742                 clp = oo->oo_owner.so_client;
2743                 release_openowner(oo);
2744                 open->op_openowner = NULL;
2745                 goto new_owner;
2746         }
2747         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2748         if (status)
2749                 return status;
2750         clp = oo->oo_owner.so_client;
2751         goto alloc_stateid;
2752 new_owner:
2753         oo = alloc_init_open_stateowner(strhashval, clp, open);
2754         if (oo == NULL)
2755                 return nfserr_jukebox;
2756         open->op_openowner = oo;
2757 alloc_stateid:
2758         open->op_stp = nfs4_alloc_stateid(clp);
2759         if (!open->op_stp)
2760                 return nfserr_jukebox;
2761         return nfs_ok;
2762 }
2763
2764 static inline __be32
2765 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2766 {
2767         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2768                 return nfserr_openmode;
2769         else
2770                 return nfs_ok;
2771 }
2772
2773 static int share_access_to_flags(u32 share_access)
2774 {
2775         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2776 }
2777
2778 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2779 {
2780         struct nfs4_stid *ret;
2781
2782         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2783         if (!ret)
2784                 return NULL;
2785         return delegstateid(ret);
2786 }
2787
2788 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2789 {
2790         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2791                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2792 }
2793
2794 static __be32
2795 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
2796                 struct nfs4_delegation **dp)
2797 {
2798         int flags;
2799         __be32 status = nfserr_bad_stateid;
2800
2801         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2802         if (*dp == NULL)
2803                 goto out;
2804         flags = share_access_to_flags(open->op_share_access);
2805         status = nfs4_check_delegmode(*dp, flags);
2806         if (status)
2807                 *dp = NULL;
2808 out:
2809         if (!nfsd4_is_deleg_cur(open))
2810                 return nfs_ok;
2811         if (status)
2812                 return status;
2813         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2814         return nfs_ok;
2815 }
2816
2817 static __be32
2818 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2819 {
2820         struct nfs4_ol_stateid *local;
2821         struct nfs4_openowner *oo = open->op_openowner;
2822
2823         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2824                 /* ignore lock owners */
2825                 if (local->st_stateowner->so_is_open_owner == 0)
2826                         continue;
2827                 /* remember if we have seen this open owner */
2828                 if (local->st_stateowner == &oo->oo_owner)
2829                         *stpp = local;
2830                 /* check for conflicting share reservations */
2831                 if (!test_share(local, open))
2832                         return nfserr_share_denied;
2833         }
2834         return nfs_ok;
2835 }
2836
2837 static inline int nfs4_access_to_access(u32 nfs4_access)
2838 {
2839         int flags = 0;
2840
2841         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2842                 flags |= NFSD_MAY_READ;
2843         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2844                 flags |= NFSD_MAY_WRITE;
2845         return flags;
2846 }
2847
2848 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2849                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2850 {
2851         __be32 status;
2852         int oflag = nfs4_access_to_omode(open->op_share_access);
2853         int access = nfs4_access_to_access(open->op_share_access);
2854
2855         if (!fp->fi_fds[oflag]) {
2856                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2857                         &fp->fi_fds[oflag]);
2858                 if (status)
2859                         return status;
2860         }
2861         nfs4_file_get_access(fp, oflag);
2862
2863         return nfs_ok;
2864 }
2865
2866 static inline __be32
2867 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2868                 struct nfsd4_open *open)
2869 {
2870         struct iattr iattr = {
2871                 .ia_valid = ATTR_SIZE,
2872                 .ia_size = 0,
2873         };
2874         if (!open->op_truncate)
2875                 return 0;
2876         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2877                 return nfserr_inval;
2878         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2879 }
2880
2881 static __be32
2882 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)
2883 {
2884         u32 op_share_access = open->op_share_access;
2885         bool new_access;
2886         __be32 status;
2887
2888         new_access = !test_access(op_share_access, stp);
2889         if (new_access) {
2890                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2891                 if (status)
2892                         return status;
2893         }
2894         status = nfsd4_truncate(rqstp, cur_fh, open);
2895         if (status) {
2896                 if (new_access) {
2897                         int oflag = nfs4_access_to_omode(op_share_access);
2898                         nfs4_file_put_access(fp, oflag);
2899                 }
2900                 return status;
2901         }
2902         /* remember the open */
2903         set_access(op_share_access, stp);
2904         set_deny(open->op_share_deny, stp);
2905
2906         return nfs_ok;
2907 }
2908
2909
2910 static void
2911 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2912 {
2913         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2914 }
2915
2916 /* Should we give out recallable state?: */
2917 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2918 {
2919         if (clp->cl_cb_state == NFSD4_CB_UP)
2920                 return true;
2921         /*
2922          * In the sessions case, since we don't have to establish a
2923          * separate connection for callbacks, we assume it's OK
2924          * until we hear otherwise:
2925          */
2926         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2927 }
2928
2929 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2930 {
2931         struct file_lock *fl;
2932
2933         fl = locks_alloc_lock();
2934         if (!fl)
2935                 return NULL;
2936         locks_init_lock(fl);
2937         fl->fl_lmops = &nfsd_lease_mng_ops;
2938         fl->fl_flags = FL_LEASE;
2939         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2940         fl->fl_end = OFFSET_MAX;
2941         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2942         fl->fl_pid = current->tgid;
2943         return fl;
2944 }
2945
2946 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2947 {
2948         struct nfs4_file *fp = dp->dl_file;
2949         struct file_lock *fl;
2950         int status;
2951
2952         fl = nfs4_alloc_init_lease(dp, flag);
2953         if (!fl)
2954                 return -ENOMEM;
2955         fl->fl_file = find_readable_file(fp);
2956         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2957         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2958         if (status) {
2959                 list_del_init(&dp->dl_perclnt);
2960                 locks_free_lock(fl);
2961                 return -ENOMEM;
2962         }
2963         fp->fi_lease = fl;
2964         fp->fi_deleg_file = get_file(fl->fl_file);
2965         atomic_set(&fp->fi_delegees, 1);
2966         list_add(&dp->dl_perfile, &fp->fi_delegations);
2967         return 0;
2968 }
2969
2970 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2971 {
2972         struct nfs4_file *fp = dp->dl_file;
2973
2974         if (!fp->fi_lease)
2975                 return nfs4_setlease(dp, flag);
2976         spin_lock(&recall_lock);
2977         if (fp->fi_had_conflict) {
2978                 spin_unlock(&recall_lock);
2979                 return -EAGAIN;
2980         }
2981         atomic_inc(&fp->fi_delegees);
2982         list_add(&dp->dl_perfile, &fp->fi_delegations);
2983         spin_unlock(&recall_lock);
2984         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2985         return 0;
2986 }
2987
2988 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2989 {
2990         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2991         if (status == -EAGAIN)
2992                 open->op_why_no_deleg = WND4_CONTENTION;
2993         else {
2994                 open->op_why_no_deleg = WND4_RESOURCE;
2995                 switch (open->op_deleg_want) {
2996                 case NFS4_SHARE_WANT_READ_DELEG:
2997                 case NFS4_SHARE_WANT_WRITE_DELEG:
2998                 case NFS4_SHARE_WANT_ANY_DELEG:
2999                         break;
3000                 case NFS4_SHARE_WANT_CANCEL:
3001                         open->op_why_no_deleg = WND4_CANCELLED;
3002                         break;
3003                 case NFS4_SHARE_WANT_NO_DELEG:
3004                         WARN_ON_ONCE(1);
3005                 }
3006         }
3007 }
3008
3009 /*
3010  * Attempt to hand out a delegation.
3011  */
3012 static void
3013 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
3014                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
3015 {
3016         struct nfs4_delegation *dp;
3017         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
3018         int cb_up;
3019         int status = 0, flag = 0;
3020
3021         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3022         flag = NFS4_OPEN_DELEGATE_NONE;
3023         open->op_recall = 0;
3024         switch (open->op_claim_type) {
3025                 case NFS4_OPEN_CLAIM_PREVIOUS:
3026                         if (!cb_up)
3027                                 open->op_recall = 1;
3028                         flag = open->op_delegate_type;
3029                         if (flag == NFS4_OPEN_DELEGATE_NONE)
3030                                 goto out;
3031                         break;
3032                 case NFS4_OPEN_CLAIM_NULL:
3033                         /* Let's not give out any delegations till everyone's
3034                          * had the chance to reclaim theirs.... */
3035                         if (locks_in_grace(net))
3036                                 goto out;
3037                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3038                                 goto out;
3039                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3040                                 flag = NFS4_OPEN_DELEGATE_WRITE;
3041                         else
3042                                 flag = NFS4_OPEN_DELEGATE_READ;
3043                         break;
3044                 default:
3045                         goto out;
3046         }
3047
3048         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
3049         if (dp == NULL)
3050                 goto out_no_deleg;
3051         status = nfs4_set_delegation(dp, flag);
3052         if (status)
3053                 goto out_free;
3054
3055         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3056
3057         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3058                 STATEID_VAL(&dp->dl_stid.sc_stateid));
3059 out:
3060         open->op_delegate_type = flag;
3061         if (flag == NFS4_OPEN_DELEGATE_NONE) {
3062                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3063                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
3064                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3065
3066                 /* 4.1 client asking for a delegation? */
3067                 if (open->op_deleg_want)
3068                         nfsd4_open_deleg_none_ext(open, status);
3069         }
3070         return;
3071 out_free:
3072         unhash_stid(&dp->dl_stid);
3073         nfs4_put_delegation(dp);
3074 out_no_deleg:
3075         flag = NFS4_OPEN_DELEGATE_NONE;
3076         goto out;
3077 }
3078
3079 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3080                                         struct nfs4_delegation *dp)
3081 {
3082         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3083             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3084                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3085                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3086         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3087                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3088                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3089                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3090         }
3091         /* Otherwise the client must be confused wanting a delegation
3092          * it already has, therefore we don't return
3093          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3094          */
3095 }
3096
3097 /*
3098  * called with nfs4_lock_state() held.
3099  */
3100 __be32
3101 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3102 {
3103         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3104         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3105         struct nfs4_file *fp = NULL;
3106         struct inode *ino = current_fh->fh_dentry->d_inode;
3107         struct nfs4_ol_stateid *stp = NULL;
3108         struct nfs4_delegation *dp = NULL;
3109         __be32 status;
3110
3111         /*
3112          * Lookup file; if found, lookup stateid and check open request,
3113          * and check for delegations in the process of being recalled.
3114          * If not found, create the nfs4_file struct
3115          */
3116         fp = find_file(ino);
3117         if (fp) {
3118                 if ((status = nfs4_check_open(fp, open, &stp)))
3119                         goto out;
3120                 status = nfs4_check_deleg(cl, open, &dp);
3121                 if (status)
3122                         goto out;
3123         } else {
3124                 status = nfserr_bad_stateid;
3125                 if (nfsd4_is_deleg_cur(open))
3126                         goto out;
3127                 status = nfserr_jukebox;
3128                 fp = open->op_file;
3129                 open->op_file = NULL;
3130                 nfsd4_init_file(fp, ino);
3131         }
3132
3133         /*
3134          * OPEN the file, or upgrade an existing OPEN.
3135          * If truncate fails, the OPEN fails.
3136          */
3137         if (stp) {
3138                 /* Stateid was found, this is an OPEN upgrade */
3139                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3140                 if (status)
3141                         goto out;
3142         } else {
3143                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3144                 if (status)
3145                         goto out;
3146                 status = nfsd4_truncate(rqstp, current_fh, open);
3147                 if (status)
3148                         goto out;
3149                 stp = open->op_stp;
3150                 open->op_stp = NULL;
3151                 init_open_stateid(stp, fp, open);
3152         }
3153         update_stateid(&stp->st_stid.sc_stateid);
3154         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3155
3156         if (nfsd4_has_session(&resp->cstate)) {
3157                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3158
3159                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3160                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3161                         open->op_why_no_deleg = WND4_NOT_WANTED;
3162                         goto nodeleg;
3163                 }
3164         }
3165
3166         /*
3167         * Attempt to hand out a delegation. No error return, because the
3168         * OPEN succeeds even if we fail.
3169         */
3170         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3171 nodeleg:
3172         status = nfs_ok;
3173
3174         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3175                 STATEID_VAL(&stp->st_stid.sc_stateid));
3176 out:
3177         /* 4.1 client trying to upgrade/downgrade delegation? */
3178         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3179             open->op_deleg_want)
3180                 nfsd4_deleg_xgrade_none_ext(open, dp);
3181
3182         if (fp)
3183                 put_nfs4_file(fp);
3184         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3185                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3186         /*
3187         * To finish the open response, we just need to set the rflags.
3188         */
3189         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3190         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3191             !nfsd4_has_session(&resp->cstate))
3192                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3193
3194         return status;
3195 }
3196
3197 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3198 {
3199         if (open->op_openowner) {
3200                 struct nfs4_openowner *oo = open->op_openowner;
3201
3202                 if (!list_empty(&oo->oo_owner.so_stateids))
3203                         list_del_init(&oo->oo_close_lru);
3204                 if (oo->oo_flags & NFS4_OO_NEW) {
3205                         if (status) {
3206                                 release_openowner(oo);
3207                                 open->op_openowner = NULL;
3208                         } else
3209                                 oo->oo_flags &= ~NFS4_OO_NEW;
3210                 }
3211         }
3212         if (open->op_file)
3213                 nfsd4_free_file(open->op_file);
3214         if (open->op_stp)
3215                 free_generic_stateid(open->op_stp);
3216 }
3217
3218 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3219 {
3220         struct nfs4_client *found;
3221
3222         if (STALE_CLIENTID(clid, nn))
3223                 return nfserr_stale_clientid;
3224         found = find_confirmed_client(clid, session, nn);
3225         if (clp)
3226                 *clp = found;
3227         return found ? nfs_ok : nfserr_expired;
3228 }
3229
3230 __be32
3231 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3232             clientid_t *clid)
3233 {
3234         struct nfs4_client *clp;
3235         __be32 status;
3236         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3237
3238         nfs4_lock_state();
3239         dprintk("process_renew(%08x/%08x): starting\n",
3240                         clid->cl_boot, clid->cl_id);
3241         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3242         if (status)
3243                 goto out;
3244         status = nfserr_cb_path_down;
3245         if (!list_empty(&clp->cl_delegations)
3246                         && clp->cl_cb_state != NFSD4_CB_UP)
3247                 goto out;
3248         status = nfs_ok;
3249 out:
3250         nfs4_unlock_state();
3251         return status;
3252 }
3253
3254 static void
3255 nfsd4_end_grace(struct nfsd_net *nn)
3256 {
3257         /* do nothing if grace period already ended */
3258         if (nn->grace_ended)
3259                 return;
3260
3261         dprintk("NFSD: end of grace period\n");
3262         nn->grace_ended = true;
3263         nfsd4_record_grace_done(nn, nn->boot_time);
3264         locks_end_grace(&nn->nfsd4_manager);
3265         /*
3266          * Now that every NFSv4 client has had the chance to recover and
3267          * to see the (possibly new, possibly shorter) lease time, we
3268          * can safely set the next grace time to the current lease time:
3269          */
3270         nn->nfsd4_grace = nn->nfsd4_lease;
3271 }
3272
3273 static time_t
3274 nfs4_laundromat(struct nfsd_net *nn)
3275 {
3276         struct nfs4_client *clp;
3277         struct nfs4_openowner *oo;
3278         struct nfs4_delegation *dp;
3279         struct list_head *pos, *next, reaplist;
3280         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3281         time_t t, clientid_val = nn->nfsd4_lease;
3282         time_t u, test_val = nn->nfsd4_lease;
3283
3284         nfs4_lock_state();
3285
3286         dprintk("NFSD: laundromat service - starting\n");
3287         nfsd4_end_grace(nn);
3288         INIT_LIST_HEAD(&reaplist);
3289         spin_lock(&nn->client_lock);
3290         list_for_each_safe(pos, next, &nn->client_lru) {
3291                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3292                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3293                         t = clp->cl_time - cutoff;
3294                         if (clientid_val > t)
3295                                 clientid_val = t;
3296                         break;
3297                 }
3298                 if (mark_client_expired_locked(clp)) {
3299                         dprintk("NFSD: client in use (clientid %08x)\n",
3300                                 clp->cl_clientid.cl_id);
3301                         continue;
3302                 }
3303                 list_move(&clp->cl_lru, &reaplist);
3304         }
3305         spin_unlock(&nn->client_lock);
3306         list_for_each_safe(pos, next, &reaplist) {
3307                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3308                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3309                         clp->cl_clientid.cl_id);
3310                 expire_client(clp);
3311         }
3312         spin_lock(&recall_lock);
3313         list_for_each_safe(pos, next, &nn->del_recall_lru) {
3314                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3315                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3316                         continue;
3317                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3318                         u = dp->dl_time - cutoff;
3319                         if (test_val > u)
3320                                 test_val = u;
3321                         break;
3322                 }
3323                 list_move(&dp->dl_recall_lru, &reaplist);
3324         }
3325         spin_unlock(&recall_lock);
3326         list_for_each_safe(pos, next, &reaplist) {
3327                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3328                 revoke_delegation(dp);
3329         }
3330         test_val = nn->nfsd4_lease;
3331         list_for_each_safe(pos, next, &nn->close_lru) {
3332                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3333                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3334                         u = oo->oo_time - cutoff;
3335                         if (test_val > u)
3336                                 test_val = u;
3337                         break;
3338                 }
3339                 release_openowner(oo);
3340         }
3341         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3342                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3343         nfs4_unlock_state();
3344         return clientid_val;
3345 }
3346
3347 static struct workqueue_struct *laundry_wq;
3348 static void laundromat_main(struct work_struct *);
3349
3350 static void
3351 laundromat_main(struct work_struct *laundry)
3352 {
3353         time_t t;
3354         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3355                                                   work);
3356         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3357                                            laundromat_work);
3358
3359         t = nfs4_laundromat(nn);
3360         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3361         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3362 }
3363
3364 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3365 {
3366         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3367                 return nfserr_bad_stateid;
3368         return nfs_ok;
3369 }
3370
3371 static inline int
3372 access_permit_read(struct nfs4_ol_stateid *stp)
3373 {
3374         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3375                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3376                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3377 }
3378
3379 static inline int
3380 access_permit_write(struct nfs4_ol_stateid *stp)
3381 {
3382         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3383                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3384 }
3385
3386 static
3387 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3388 {
3389         __be32 status = nfserr_openmode;
3390
3391         /* For lock stateid's, we test the parent open, not the lock: */
3392         if (stp->st_openstp)
3393                 stp = stp->st_openstp;
3394         if ((flags & WR_STATE) && !access_permit_write(stp))
3395                 goto out;
3396         if ((flags & RD_STATE) && !access_permit_read(stp))
3397                 goto out;
3398         status = nfs_ok;
3399 out:
3400         return status;
3401 }
3402
3403 static inline __be32
3404 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3405 {
3406         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3407                 return nfs_ok;
3408         else if (locks_in_grace(net)) {
3409                 /* Answer in remaining cases depends on existence of
3410                  * conflicting state; so we must wait out the grace period. */
3411                 return nfserr_grace;
3412         } else if (flags & WR_STATE)
3413                 return nfs4_share_conflict(current_fh,
3414                                 NFS4_SHARE_DENY_WRITE);
3415         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3416                 return nfs4_share_conflict(current_fh,
3417                                 NFS4_SHARE_DENY_READ);
3418 }
3419
3420 /*
3421  * Allow READ/WRITE during grace period on recovered state only for files
3422  * that are not able to provide mandatory locking.
3423  */
3424 static inline int
3425 grace_disallows_io(struct net *net, struct inode *inode)
3426 {
3427         return locks_in_grace(net) && mandatory_lock(inode);
3428 }
3429
3430 /* Returns true iff a is later than b: */
3431 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3432 {
3433         return (s32)(a->si_generation - b->si_generation) > 0;
3434 }
3435
3436 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3437 {
3438         /*
3439          * When sessions are used the stateid generation number is ignored
3440          * when it is zero.
3441          */
3442         if (has_session && in->si_generation == 0)
3443                 return nfs_ok;
3444
3445         if (in->si_generation == ref->si_generation)
3446                 return nfs_ok;
3447
3448         /* If the client sends us a stateid from the future, it's buggy: */
3449         if (stateid_generation_after(in, ref))
3450                 return nfserr_bad_stateid;
3451         /*
3452          * However, we could see a stateid from the past, even from a
3453          * non-buggy client.  For example, if the client sends a lock
3454          * while some IO is outstanding, the lock may bump si_generation
3455          * while the IO is still in flight.  The client could avoid that
3456          * situation by waiting for responses on all the IO requests,
3457          * but better performance may result in retrying IO that
3458          * receives an old_stateid error if requests are rarely
3459          * reordered in flight:
3460          */
3461         return nfserr_old_stateid;
3462 }
3463
3464 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3465 {
3466         struct nfs4_stid *s;
3467         struct nfs4_ol_stateid *ols;
3468         __be32 status;
3469
3470         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3471                 return nfserr_bad_stateid;
3472         /* Client debugging aid. */
3473         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3474                 char addr_str[INET6_ADDRSTRLEN];
3475                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3476                                  sizeof(addr_str));
3477                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3478                                         "with incorrect client ID\n", addr_str);
3479                 return nfserr_bad_stateid;
3480         }
3481         s = find_stateid(cl, stateid);
3482         if (!s)
3483                 return nfserr_bad_stateid;
3484         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3485         if (status)
3486                 return status;
3487         switch (s->sc_type) {
3488         case NFS4_DELEG_STID:
3489                 return nfs_ok;
3490         case NFS4_REVOKED_DELEG_STID:
3491                 return nfserr_deleg_revoked;
3492         case NFS4_OPEN_STID:
3493         case NFS4_LOCK_STID:
3494                 ols = openlockstateid(s);
3495                 if (ols->st_stateowner->so_is_open_owner
3496                                 && !(openowner(ols->st_stateowner)->oo_flags
3497                                                 & NFS4_OO_CONFIRMED))
3498                         return nfserr_bad_stateid;
3499                 return nfs_ok;
3500         default:
3501                 printk("unknown stateid type %x\n", s->sc_type);
3502         case NFS4_CLOSED_STID:
3503                 return nfserr_bad_stateid;
3504         }
3505 }
3506
3507 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3508                                    struct nfs4_stid **s, bool sessions,
3509                                    struct nfsd_net *nn)
3510 {
3511         struct nfs4_client *cl;
3512         __be32 status;
3513
3514         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3515                 return nfserr_bad_stateid;
3516         status = lookup_clientid(&stateid->si_opaque.so_clid, sessions,
3517                                                         nn, &cl);
3518         if (status == nfserr_stale_clientid)
3519                 return nfserr_stale_stateid;
3520         if (status)
3521                 return status;
3522         *s = find_stateid_by_type(cl, stateid, typemask);
3523         if (!*s)
3524                 return nfserr_bad_stateid;
3525         return nfs_ok;
3526 }
3527
3528 /*
3529 * Checks for stateid operations
3530 */
3531 __be32
3532 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3533                            stateid_t *stateid, int flags, struct file **filpp)
3534 {
3535         struct nfs4_stid *s;
3536         struct nfs4_ol_stateid *stp = NULL;
3537         struct nfs4_delegation *dp = NULL;
3538         struct svc_fh *current_fh = &cstate->current_fh;
3539         struct inode *ino = current_fh->fh_dentry->d_inode;
3540         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3541         __be32 status;
3542
3543         if (filpp)
3544                 *filpp = NULL;
3545
3546         if (grace_disallows_io(net, ino))
3547                 return nfserr_grace;
3548
3549         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3550                 return check_special_stateids(net, current_fh, stateid, flags);
3551
3552         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3553                                       &s, cstate->minorversion, nn);
3554         if (status)
3555                 return status;
3556         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3557         if (status)
3558                 goto out;
3559         switch (s->sc_type) {
3560         case NFS4_DELEG_STID:
3561                 dp = delegstateid(s);
3562                 status = nfs4_check_delegmode(dp, flags);
3563                 if (status)
3564                         goto out;
3565                 if (filpp) {
3566                         *filpp = dp->dl_file->fi_deleg_file;
3567                         if (!*filpp) {
3568                                 WARN_ON_ONCE(1);
3569                                 status = nfserr_serverfault;
3570                                 goto out;
3571                         }
3572                 }
3573                 break;
3574         case NFS4_OPEN_STID:
3575         case NFS4_LOCK_STID:
3576                 stp = openlockstateid(s);
3577                 status = nfs4_check_fh(current_fh, stp);
3578                 if (status)
3579                         goto out;
3580                 if (stp->st_stateowner->so_is_open_owner
3581                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3582                         goto out;
3583                 status = nfs4_check_openmode(stp, flags);
3584                 if (status)
3585                         goto out;
3586                 if (filpp) {
3587                         if (flags & RD_STATE)
3588                                 *filpp = find_readable_file(stp->st_file);
3589                         else
3590                                 *filpp = find_writeable_file(stp->st_file);
3591                 }
3592                 break;
3593         default:
3594                 return nfserr_bad_stateid;
3595         }
3596         status = nfs_ok;
3597 out:
3598         return status;
3599 }
3600
3601 static __be32
3602 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3603 {
3604         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3605                 return nfserr_locks_held;
3606         release_lock_stateid(stp);
3607         return nfs_ok;
3608 }
3609
3610 /*
3611  * Test if the stateid is valid
3612  */
3613 __be32
3614 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3615                    struct nfsd4_test_stateid *test_stateid)
3616 {
3617         struct nfsd4_test_stateid_id *stateid;
3618         struct nfs4_client *cl = cstate->session->se_client;
3619
3620         nfs4_lock_state();
3621         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3622                 stateid->ts_id_status =
3623                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3624         nfs4_unlock_state();
3625
3626         return nfs_ok;
3627 }
3628
3629 __be32
3630 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3631                    struct nfsd4_free_stateid *free_stateid)
3632 {
3633         stateid_t *stateid = &free_stateid->fr_stateid;
3634         struct nfs4_stid *s;
3635         struct nfs4_delegation *dp;
3636         struct nfs4_client *cl = cstate->session->se_client;
3637         __be32 ret = nfserr_bad_stateid;
3638
3639         nfs4_lock_state();
3640         s = find_stateid(cl, stateid);
3641         if (!s)
3642                 goto out;
3643         switch (s->sc_type) {
3644         case NFS4_DELEG_STID:
3645                 ret = nfserr_locks_held;
3646                 goto out;
3647         case NFS4_OPEN_STID:
3648         case NFS4_LOCK_STID:
3649                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3650                 if (ret)
3651                         goto out;
3652                 if (s->sc_type == NFS4_LOCK_STID)
3653                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3654                 else
3655                         ret = nfserr_locks_held;
3656                 break;
3657         case NFS4_REVOKED_DELEG_STID:
3658                 dp = delegstateid(s);
3659                 destroy_revoked_delegation(dp);
3660                 ret = nfs_ok;
3661                 break;
3662         default:
3663                 ret = nfserr_bad_stateid;
3664         }
3665 out:
3666         nfs4_unlock_state();
3667         return ret;
3668 }
3669
3670 static inline int
3671 setlkflg (int type)
3672 {
3673         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3674                 RD_STATE : WR_STATE;
3675 }
3676
3677 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3678 {
3679         struct svc_fh *current_fh = &cstate->current_fh;
3680         struct nfs4_stateowner *sop = stp->st_stateowner;
3681         __be32 status;
3682
3683         status = nfsd4_check_seqid(cstate, sop, seqid);
3684         if (status)
3685                 return status;
3686         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
3687                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
3688                 /*
3689                  * "Closed" stateid's exist *only* to return
3690                  * nfserr_replay_me from the previous step, and
3691                  * revoked delegations are kept only for free_stateid.
3692                  */
3693                 return nfserr_bad_stateid;
3694         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3695         if (status)
3696                 return status;
3697         return nfs4_check_fh(current_fh, stp);
3698 }
3699
3700 /*
3701  * Checks for sequence id mutating operations.
3702  */
3703 static __be32
3704 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3705                          stateid_t *stateid, char typemask,
3706                          struct nfs4_ol_stateid **stpp,
3707                          struct nfsd_net *nn)
3708 {
3709         __be32 status;
3710         struct nfs4_stid *s;
3711
3712         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3713                 seqid, STATEID_VAL(stateid));
3714
3715         *stpp = NULL;
3716         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3717                                       cstate->minorversion, nn);
3718         if (status)
3719                 return status;
3720         *stpp = openlockstateid(s);
3721         if (!nfsd4_has_session(cstate))
3722                 cstate->replay_owner = (*stpp)->st_stateowner;
3723
3724         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3725 }
3726
3727 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3728                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3729 {
3730         __be32 status;
3731         struct nfs4_openowner *oo;
3732
3733         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3734                                                 NFS4_OPEN_STID, stpp, nn);
3735         if (status)
3736                 return status;
3737         oo = openowner((*stpp)->st_stateowner);
3738         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3739                 return nfserr_bad_stateid;
3740         return nfs_ok;
3741 }
3742
3743 __be32
3744 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3745                    struct nfsd4_open_confirm *oc)
3746 {
3747         __be32 status;
3748         struct nfs4_openowner *oo;
3749         struct nfs4_ol_stateid *stp;
3750         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3751
3752         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3753                         (int)cstate->current_fh.fh_dentry->d_name.len,
3754                         cstate->current_fh.fh_dentry->d_name.name);
3755
3756         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3757         if (status)
3758                 return status;
3759
3760         nfs4_lock_state();
3761
3762         status = nfs4_preprocess_seqid_op(cstate,
3763                                         oc->oc_seqid, &oc->oc_req_stateid,
3764                                         NFS4_OPEN_STID, &stp, nn);
3765         if (status)
3766                 goto out;
3767         oo = openowner(stp->st_stateowner);
3768         status = nfserr_bad_stateid;
3769         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3770                 goto out;
3771         oo->oo_flags |= NFS4_OO_CONFIRMED;
3772         update_stateid(&stp->st_stid.sc_stateid);
3773         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3774         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3775                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3776
3777         nfsd4_client_record_create(oo->oo_owner.so_client);
3778         status = nfs_ok;
3779 out:
3780         nfsd4_bump_seqid(cstate, status);
3781         if (!cstate->replay_owner)
3782                 nfs4_unlock_state();
3783         return status;
3784 }
3785
3786 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3787 {
3788         if (!test_access(access, stp))
3789                 return;
3790         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3791         clear_access(access, stp);
3792 }
3793
3794 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3795 {
3796         switch (to_access) {
3797         case NFS4_SHARE_ACCESS_READ:
3798                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3799                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3800                 break;
3801         case NFS4_SHARE_ACCESS_WRITE:
3802                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3803                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3804                 break;
3805         case NFS4_SHARE_ACCESS_BOTH:
3806                 break;
3807         default:
3808                 WARN_ON_ONCE(1);
3809         }
3810 }
3811
3812 static void
3813 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3814 {
3815         int i;
3816         for (i = 0; i < 4; i++) {
3817                 if ((i & deny) != i)
3818                         clear_deny(i, stp);
3819         }
3820 }
3821
3822 __be32
3823 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3824                      struct nfsd4_compound_state *cstate,
3825                      struct nfsd4_open_downgrade *od)
3826 {
3827         __be32 status;
3828         struct nfs4_ol_stateid *stp;
3829         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3830
3831         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3832                         (int)cstate->current_fh.fh_dentry->d_name.len,
3833                         cstate->current_fh.fh_dentry->d_name.name);
3834
3835         /* We don't yet support WANT bits: */
3836         if (od->od_deleg_want)
3837                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3838                         od->od_deleg_want);
3839
3840         nfs4_lock_state();
3841         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3842                                         &od->od_stateid, &stp, nn);
3843         if (status)
3844                 goto out;
3845         status = nfserr_inval;
3846         if (!test_access(od->od_share_access, stp)) {
3847                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3848                         stp->st_access_bmap, od->od_share_access);
3849                 goto out;
3850         }
3851         if (!test_deny(od->od_share_deny, stp)) {
3852                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3853                         stp->st_deny_bmap, od->od_share_deny);
3854                 goto out;
3855         }
3856         nfs4_stateid_downgrade(stp, od->od_share_access);
3857
3858         reset_union_bmap_deny(od->od_share_deny, stp);
3859
3860         update_stateid(&stp->st_stid.sc_stateid);
3861         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3862         status = nfs_ok;
3863 out:
3864         nfsd4_bump_seqid(cstate, status);
3865         if (!cstate->replay_owner)
3866                 nfs4_unlock_state();
3867         return status;
3868 }
3869
3870 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3871 {
3872         unhash_open_stateid(s);
3873         s->st_stid.sc_type = NFS4_CLOSED_STID;
3874 }
3875
3876 /*
3877  * nfs4_unlock_state() called after encode
3878  */
3879 __be32
3880 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3881             struct nfsd4_close *close)
3882 {
3883         __be32 status;
3884         struct nfs4_openowner *oo;
3885         struct nfs4_ol_stateid *stp;
3886         struct net *net = SVC_NET(rqstp);
3887         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3888
3889         dprintk("NFSD: nfsd4_close on file %.*s\n",
3890                         (int)cstate->current_fh.fh_dentry->d_name.len,
3891                         cstate->current_fh.fh_dentry->d_name.name);
3892
3893         nfs4_lock_state();
3894         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3895                                         &close->cl_stateid,
3896                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3897                                         &stp, nn);
3898         nfsd4_bump_seqid(cstate, status);
3899         if (status)
3900                 goto out;
3901         oo = openowner(stp->st_stateowner);
3902         update_stateid(&stp->st_stid.sc_stateid);
3903         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3904
3905         nfsd4_close_open_stateid(stp);
3906
3907         if (cstate->minorversion) {
3908                 unhash_stid(&stp->st_stid);
3909                 free_generic_stateid(stp);
3910         } else
3911                 oo->oo_last_closed_stid = stp;
3912
3913         if (list_empty(&oo->oo_owner.so_stateids)) {
3914                 if (cstate->minorversion)
3915                         release_openowner(oo);
3916                 else {
3917                         /*
3918                          * In the 4.0 case we need to keep the owners around a
3919                          * little while to handle CLOSE replay.
3920                          */
3921                         move_to_close_lru(oo, SVC_NET(rqstp));
3922                 }
3923         }
3924 out:
3925         if (!cstate->replay_owner)
3926                 nfs4_unlock_state();
3927         return status;
3928 }
3929
3930 __be32
3931 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3932                   struct nfsd4_delegreturn *dr)
3933 {
3934         struct nfs4_delegation *dp;
3935         stateid_t *stateid = &dr->dr_stateid;
3936         struct nfs4_stid *s;
3937         __be32 status;
3938         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3939
3940         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3941                 return status;
3942
3943         nfs4_lock_state();
3944         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
3945                                       cstate->minorversion, nn);
3946         if (status)
3947                 goto out;
3948         dp = delegstateid(s);
3949         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3950         if (status)
3951                 goto out;
3952
3953         destroy_delegation(dp);
3954 out:
3955         nfs4_unlock_state();
3956
3957         return status;
3958 }
3959
3960
3961 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3962
3963 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3964
3965 static inline u64
3966 end_offset(u64 start, u64 len)
3967 {
3968         u64 end;
3969
3970         end = start + len;
3971         return end >= start ? end: NFS4_MAX_UINT64;
3972 }
3973
3974 /* last octet in a range */
3975 static inline u64
3976 last_byte_offset(u64 start, u64 len)
3977 {
3978         u64 end;
3979
3980         WARN_ON_ONCE(!len);
3981         end = start + len;
3982         return end > start ? end - 1: NFS4_MAX_UINT64;
3983 }
3984
3985 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3986 {
3987         return (file_hashval(inode) + cl_id
3988                         + opaque_hashval(ownername->data, ownername->len))
3989                 & LOCKOWNER_INO_HASH_MASK;
3990 }
3991
3992 /*
3993  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3994  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3995  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3996  * locking, this prevents us from being completely protocol-compliant.  The
3997  * real solution to this problem is to start using unsigned file offsets in
3998  * the VFS, but this is a very deep change!
3999  */
4000 static inline void
4001 nfs4_transform_lock_offset(struct file_lock *lock)
4002 {
4003         if (lock->fl_start < 0)
4004                 lock->fl_start = OFFSET_MAX;
4005         if (lock->fl_end < 0)
4006                 lock->fl_end = OFFSET_MAX;
4007 }
4008
4009 /* Hack!: For now, we're defining this just so we can use a pointer to it
4010  * as a unique cookie to identify our (NFSv4's) posix locks. */
4011 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
4012 };
4013
4014 static inline void
4015 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4016 {
4017         struct nfs4_lockowner *lo;
4018
4019         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4020                 lo = (struct nfs4_lockowner *) fl->fl_owner;
4021                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4022                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
4023                 if (!deny->ld_owner.data)
4024                         /* We just don't care that much */
4025                         goto nevermind;
4026                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4027                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4028         } else {
4029 nevermind:
4030                 deny->ld_owner.len = 0;
4031                 deny->ld_owner.data = NULL;
4032                 deny->ld_clientid.cl_boot = 0;
4033                 deny->ld_clientid.cl_id = 0;
4034         }
4035         deny->ld_start = fl->fl_start;
4036         deny->ld_length = NFS4_MAX_UINT64;
4037         if (fl->fl_end != NFS4_MAX_UINT64)
4038                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4039         deny->ld_type = NFS4_READ_LT;
4040         if (fl->fl_type != F_RDLCK)
4041                 deny->ld_type = NFS4_WRITE_LT;
4042 }
4043
4044 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
4045 {
4046         struct nfs4_ol_stateid *lst;
4047
4048         if (!same_owner_str(&lo->lo_owner, owner, clid))
4049                 return false;
4050         lst = list_first_entry(&lo->lo_owner.so_stateids,
4051                                struct nfs4_ol_stateid, st_perstateowner);
4052         return lst->st_file->fi_inode == inode;
4053 }
4054
4055 static struct nfs4_lockowner *
4056 find_lockowner_str(struct inode *inode, clientid_t *clid,
4057                    struct xdr_netobj *owner, struct nfsd_net *nn)
4058 {
4059         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4060         struct nfs4_lockowner *lo;
4061
4062         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4063                 if (same_lockowner_ino(lo, inode, clid, owner))
4064                         return lo;
4065         }
4066         return NULL;
4067 }
4068
4069 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4070 {
4071         struct inode *inode = open_stp->st_file->fi_inode;
4072         unsigned int inohash = lockowner_ino_hashval(inode,
4073                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4074         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4075
4076         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4077         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4078         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4079 }
4080
4081 /*
4082  * Alloc a lock owner structure.
4083  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4084  * occurred.
4085  *
4086  * strhashval = ownerstr_hashval
4087  */
4088
4089 static struct nfs4_lockowner *
4090 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4091         struct nfs4_lockowner *lo;
4092
4093         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4094         if (!lo)
4095                 return NULL;
4096         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4097         lo->lo_owner.so_is_open_owner = 0;
4098         /* It is the openowner seqid that will be incremented in encode in the
4099          * case of new lockowners; so increment the lock seqid manually: */
4100         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4101         hash_lockowner(lo, strhashval, clp, open_stp);
4102         return lo;
4103 }
4104
4105 static struct nfs4_ol_stateid *
4106 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4107 {
4108         struct nfs4_ol_stateid *stp;
4109         struct nfs4_client *clp = lo->lo_owner.so_client;
4110
4111         stp = nfs4_alloc_stateid(clp);
4112         if (stp == NULL)
4113                 return NULL;
4114         stp->st_stid.sc_type = NFS4_LOCK_STID;
4115         list_add(&stp->st_perfile, &fp->fi_stateids);
4116         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4117         stp->st_stateowner = &lo->lo_owner;
4118         get_nfs4_file(fp);
4119         stp->st_file = fp;
4120         stp->st_access_bmap = 0;
4121         stp->st_deny_bmap = open_stp->st_deny_bmap;
4122         stp->st_openstp = open_stp;
4123         return stp;
4124 }
4125
4126 static int
4127 check_lock_length(u64 offset, u64 length)
4128 {
4129         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4130              LOFF_OVERFLOW(offset, length)));
4131 }
4132
4133 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4134 {
4135         struct nfs4_file *fp = lock_stp->st_file;
4136         int oflag = nfs4_access_to_omode(access);
4137
4138         if (test_access(access, lock_stp))
4139                 return;
4140         nfs4_file_get_access(fp, oflag);
4141         set_access(access, lock_stp);
4142 }
4143
4144 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)
4145 {
4146         struct nfs4_file *fi = ost->st_file;
4147         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4148         struct nfs4_client *cl = oo->oo_owner.so_client;
4149         struct nfs4_lockowner *lo;
4150         unsigned int strhashval;
4151         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4152
4153         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4154                                 &lock->v.new.owner, nn);
4155         if (lo) {
4156                 if (!cstate->minorversion)
4157                         return nfserr_bad_seqid;
4158                 /* XXX: a lockowner always has exactly one stateid: */
4159                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4160                                 struct nfs4_ol_stateid, st_perstateowner);
4161                 return nfs_ok;
4162         }
4163         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4164                         &lock->v.new.owner);
4165         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4166         if (lo == NULL)
4167                 return nfserr_jukebox;
4168         *lst = alloc_init_lock_stateid(lo, fi, ost);
4169         if (*lst == NULL) {
4170                 release_lockowner(lo);
4171                 return nfserr_jukebox;
4172         }
4173         *new = true;
4174         return nfs_ok;
4175 }
4176
4177 /*
4178  *  LOCK operation
4179  */
4180 __be32
4181 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4182            struct nfsd4_lock *lock)
4183 {
4184         struct nfs4_openowner *open_sop = NULL;
4185         struct nfs4_lockowner *lock_sop = NULL;
4186         struct nfs4_ol_stateid *lock_stp;
4187         struct file *filp = NULL;
4188         struct file_lock *file_lock = NULL;
4189         struct file_lock *conflock = NULL;
4190         __be32 status = 0;
4191         bool new_state = false;
4192         int lkflg;
4193         int err;
4194         struct net *net = SVC_NET(rqstp);
4195         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4196
4197         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4198                 (long long) lock->lk_offset,
4199                 (long long) lock->lk_length);
4200
4201         if (check_lock_length(lock->lk_offset, lock->lk_length))
4202                  return nfserr_inval;
4203
4204         if ((status = fh_verify(rqstp, &cstate->current_fh,
4205                                 S_IFREG, NFSD_MAY_LOCK))) {
4206                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4207                 return status;
4208         }
4209
4210         nfs4_lock_state();
4211
4212         if (lock->lk_is_new) {
4213                 struct nfs4_ol_stateid *open_stp = NULL;
4214
4215                 if (nfsd4_has_session(cstate))
4216                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4217                         memcpy(&lock->v.new.clientid,
4218                                 &cstate->session->se_client->cl_clientid,
4219                                 sizeof(clientid_t));
4220
4221                 status = nfserr_stale_clientid;
4222                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4223                         goto out;
4224
4225                 /* validate and update open stateid and open seqid */
4226                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4227                                         lock->lk_new_open_seqid,
4228                                         &lock->lk_new_open_stateid,
4229                                         &open_stp, nn);
4230                 if (status)
4231                         goto out;
4232                 open_sop = openowner(open_stp->st_stateowner);
4233                 status = nfserr_bad_stateid;
4234                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4235                                                 &lock->v.new.clientid))
4236                         goto out;
4237                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4238                                                         &lock_stp, &new_state);
4239         } else
4240                 status = nfs4_preprocess_seqid_op(cstate,
4241                                        lock->lk_old_lock_seqid,
4242                                        &lock->lk_old_lock_stateid,
4243                                        NFS4_LOCK_STID, &lock_stp, nn);
4244         if (status)
4245                 goto out;
4246         lock_sop = lockowner(lock_stp->st_stateowner);
4247
4248         lkflg = setlkflg(lock->lk_type);
4249         status = nfs4_check_openmode(lock_stp, lkflg);
4250         if (status)
4251                 goto out;
4252
4253         status = nfserr_grace;
4254         if (locks_in_grace(net) && !lock->lk_reclaim)
4255                 goto out;
4256         status = nfserr_no_grace;
4257         if (!locks_in_grace(net) && lock->lk_reclaim)
4258                 goto out;
4259
4260         file_lock = locks_alloc_lock();
4261         if (!file_lock) {
4262                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4263                 status = nfserr_jukebox;
4264                 goto out;
4265         }
4266
4267         locks_init_lock(file_lock);
4268         switch (lock->lk_type) {
4269                 case NFS4_READ_LT:
4270                 case NFS4_READW_LT:
4271                         filp = find_readable_file(lock_stp->st_file);
4272                         if (filp)
4273                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4274                         file_lock->fl_type = F_RDLCK;
4275                         break;
4276                 case NFS4_WRITE_LT:
4277                 case NFS4_WRITEW_LT:
4278                         filp = find_writeable_file(lock_stp->st_file);
4279                         if (filp)
4280                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4281                         file_lock->fl_type = F_WRLCK;
4282                         break;
4283                 default:
4284                         status = nfserr_inval;
4285                 goto out;
4286         }
4287         if (!filp) {
4288                 status = nfserr_openmode;
4289                 goto out;
4290         }
4291         file_lock->fl_owner = (fl_owner_t)lock_sop;
4292         file_lock->fl_pid = current->tgid;
4293         file_lock->fl_file = filp;
4294         file_lock->fl_flags = FL_POSIX;
4295         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4296         file_lock->fl_start = lock->lk_offset;
4297         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4298         nfs4_transform_lock_offset(file_lock);
4299
4300         conflock = locks_alloc_lock();
4301         if (!conflock) {
4302                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4303                 status = nfserr_jukebox;
4304                 goto out;
4305         }
4306
4307         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4308         switch (-err) {
4309         case 0: /* success! */
4310                 update_stateid(&lock_stp->st_stid.sc_stateid);
4311                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4312                                 sizeof(stateid_t));
4313                 status = 0;
4314                 break;
4315         case (EAGAIN):          /* conflock holds conflicting lock */
4316                 status = nfserr_denied;
4317                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4318                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4319                 break;
4320         case (EDEADLK):
4321                 status = nfserr_deadlock;
4322                 break;
4323         default:
4324                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4325                 status = nfserrno(err);
4326                 break;
4327         }
4328 out:
4329         if (status && new_state)
4330                 release_lockowner(lock_sop);
4331         nfsd4_bump_seqid(cstate, status);
4332         if (!cstate->replay_owner)
4333                 nfs4_unlock_state();
4334         if (file_lock)
4335                 locks_free_lock(file_lock);
4336         if (conflock)
4337                 locks_free_lock(conflock);
4338         return status;
4339 }
4340
4341 /*
4342  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4343  * so we do a temporary open here just to get an open file to pass to
4344  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4345  * inode operation.)
4346  */
4347 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4348 {
4349         struct file *file;
4350         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4351         if (!err) {
4352                 err = nfserrno(vfs_test_lock(file, lock));
4353                 nfsd_close(file);
4354         }
4355         return err;
4356 }
4357
4358 /*
4359  * LOCKT operation
4360  */
4361 __be32
4362 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4363             struct nfsd4_lockt *lockt)
4364 {
4365         struct inode *inode;
4366         struct file_lock *file_lock = NULL;
4367         struct nfs4_lockowner *lo;
4368         __be32 status;
4369         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4370
4371         if (locks_in_grace(SVC_NET(rqstp)))
4372                 return nfserr_grace;
4373
4374         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4375                  return nfserr_inval;
4376
4377         nfs4_lock_state();
4378
4379         if (!nfsd4_has_session(cstate)) {
4380                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4381                 if (status)
4382                         goto out;
4383         }
4384
4385         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4386                 goto out;
4387
4388         inode = cstate->current_fh.fh_dentry->d_inode;
4389         file_lock = locks_alloc_lock();
4390         if (!file_lock) {
4391                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4392                 status = nfserr_jukebox;
4393                 goto out;
4394         }
4395         locks_init_lock(file_lock);
4396         switch (lockt->lt_type) {
4397                 case NFS4_READ_LT:
4398                 case NFS4_READW_LT:
4399                         file_lock->fl_type = F_RDLCK;
4400                 break;
4401                 case NFS4_WRITE_LT:
4402                 case NFS4_WRITEW_LT:
4403                         file_lock->fl_type = F_WRLCK;
4404                 break;
4405                 default:
4406                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4407                         status = nfserr_inval;
4408                 goto out;
4409         }
4410
4411         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4412         if (lo)
4413                 file_lock->fl_owner = (fl_owner_t)lo;
4414         file_lock->fl_pid = current->tgid;
4415         file_lock->fl_flags = FL_POSIX;
4416
4417         file_lock->fl_start = lockt->lt_offset;
4418         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4419
4420         nfs4_transform_lock_offset(file_lock);
4421
4422         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4423         if (status)
4424                 goto out;
4425
4426         if (file_lock->fl_type != F_UNLCK) {
4427                 status = nfserr_denied;
4428                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4429         }
4430 out:
4431         nfs4_unlock_state();
4432         if (file_lock)
4433                 locks_free_lock(file_lock);
4434         return status;
4435 }
4436
4437 __be32
4438 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4439             struct nfsd4_locku *locku)
4440 {
4441         struct nfs4_lockowner *lo;
4442         struct nfs4_ol_stateid *stp;
4443         struct file *filp = NULL;
4444         struct file_lock *file_lock = NULL;
4445         __be32 status;
4446         int err;
4447         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4448
4449         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4450                 (long long) locku->lu_offset,
4451                 (long long) locku->lu_length);
4452
4453         if (check_lock_length(locku->lu_offset, locku->lu_length))
4454                  return nfserr_inval;
4455
4456         nfs4_lock_state();
4457
4458         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4459                                         &locku->lu_stateid, NFS4_LOCK_STID,
4460                                         &stp, nn);
4461         if (status)
4462                 goto out;
4463         filp = find_any_file(stp->st_file);
4464         if (!filp) {
4465                 status = nfserr_lock_range;
4466                 goto out;
4467         }
4468         file_lock = locks_alloc_lock();
4469         if (!file_lock) {
4470                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4471                 status = nfserr_jukebox;
4472                 goto out;
4473         }
4474         lo = lockowner(stp->st_stateowner);
4475         locks_init_lock(file_lock);
4476         file_lock->fl_type = F_UNLCK;
4477         file_lock->fl_owner = (fl_owner_t)lo;
4478         file_lock->fl_pid = current->tgid;
4479         file_lock->fl_file = filp;
4480         file_lock->fl_flags = FL_POSIX;
4481         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4482         file_lock->fl_start = locku->lu_offset;
4483
4484         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4485                                                 locku->lu_length);
4486         nfs4_transform_lock_offset(file_lock);
4487
4488         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4489         if (err) {
4490                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4491                 goto out_nfserr;
4492         }
4493         update_stateid(&stp->st_stid.sc_stateid);
4494         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4495
4496         if (nfsd4_has_session(cstate) && !check_for_locks(stp->st_file, lo)) {
4497                 WARN_ON_ONCE(cstate->replay_owner);
4498                 release_lockowner(lo);
4499         }
4500
4501 out:
4502         nfsd4_bump_seqid(cstate, status);
4503         if (!cstate->replay_owner)
4504                 nfs4_unlock_state();
4505         if (file_lock)
4506                 locks_free_lock(file_lock);
4507         return status;
4508
4509 out_nfserr:
4510         status = nfserrno(err);
4511         goto out;
4512 }
4513
4514 /*
4515  * returns
4516  *      1: locks held by lockowner
4517  *      0: no locks held by lockowner
4518  */
4519 static int
4520 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4521 {
4522         struct file_lock **flpp;
4523         struct inode *inode = filp->fi_inode;
4524         int status = 0;
4525
4526         lock_flocks();
4527         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4528                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4529                         status = 1;
4530                         goto out;
4531                 }
4532         }
4533 out:
4534         unlock_flocks();
4535         return status;
4536 }
4537
4538 __be32
4539 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4540                         struct nfsd4_compound_state *cstate,
4541                         struct nfsd4_release_lockowner *rlockowner)
4542 {
4543         clientid_t *clid = &rlockowner->rl_clientid;
4544         struct nfs4_stateowner *sop;
4545         struct nfs4_lockowner *lo;
4546         struct nfs4_ol_stateid *stp;
4547         struct xdr_netobj *owner = &rlockowner->rl_owner;
4548         struct list_head matches;
4549         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4550         __be32 status;
4551         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4552
4553         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4554                 clid->cl_boot, clid->cl_id);
4555
4556         nfs4_lock_state();
4557
4558         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4559         if (status)
4560                 goto out;
4561
4562         status = nfserr_locks_held;
4563         INIT_LIST_HEAD(&matches);
4564
4565         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4566                 if (sop->so_is_open_owner)
4567                         continue;
4568                 if (!same_owner_str(sop, owner, clid))
4569                         continue;
4570                 list_for_each_entry(stp, &sop->so_stateids,
4571                                 st_perstateowner) {
4572                         lo = lockowner(sop);
4573                         if (check_for_locks(stp->st_file, lo))
4574                                 goto out;
4575                         list_add(&lo->lo_list, &matches);
4576                 }
4577         }
4578         /* Clients probably won't expect us to return with some (but not all)
4579          * of the lockowner state released; so don't release any until all
4580          * have been checked. */
4581         status = nfs_ok;
4582         while (!list_empty(&matches)) {
4583                 lo = list_entry(matches.next, struct nfs4_lockowner,
4584                                                                 lo_list);
4585                 /* unhash_stateowner deletes so_perclient only
4586                  * for openowners. */
4587                 list_del(&lo->lo_list);
4588                 release_lockowner(lo);
4589         }
4590 out:
4591         nfs4_unlock_state();
4592         return status;
4593 }
4594
4595 static inline struct nfs4_client_reclaim *
4596 alloc_reclaim(void)
4597 {
4598         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4599 }
4600
4601 bool
4602 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4603 {
4604         struct nfs4_client_reclaim *crp;
4605
4606         crp = nfsd4_find_reclaim_client(name, nn);
4607         return (crp && crp->cr_clp);
4608 }
4609
4610 /*
4611  * failure => all reset bets are off, nfserr_no_grace...
4612  */
4613 struct nfs4_client_reclaim *
4614 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4615 {
4616         unsigned int strhashval;
4617         struct nfs4_client_reclaim *crp;
4618
4619         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4620         crp = alloc_reclaim();
4621         if (crp) {
4622                 strhashval = clientstr_hashval(name);
4623                 INIT_LIST_HEAD(&crp->cr_strhash);
4624                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4625                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4626                 crp->cr_clp = NULL;
4627                 nn->reclaim_str_hashtbl_size++;
4628         }
4629         return crp;
4630 }
4631
4632 void
4633 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4634 {
4635         list_del(&crp->cr_strhash);
4636         kfree(crp);
4637         nn->reclaim_str_hashtbl_size--;
4638 }
4639
4640 void
4641 nfs4_release_reclaim(struct nfsd_net *nn)
4642 {
4643         struct nfs4_client_reclaim *crp = NULL;
4644         int i;
4645
4646         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4647                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4648                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4649                                         struct nfs4_client_reclaim, cr_strhash);
4650                         nfs4_remove_reclaim_record(crp, nn);
4651                 }
4652         }
4653         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4654 }
4655
4656 /*
4657  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4658 struct nfs4_client_reclaim *
4659 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4660 {
4661         unsigned int strhashval;
4662         struct nfs4_client_reclaim *crp = NULL;
4663
4664         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4665
4666         strhashval = clientstr_hashval(recdir);
4667         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4668                 if (same_name(crp->cr_recdir, recdir)) {
4669                         return crp;
4670                 }
4671         }
4672         return NULL;
4673 }
4674
4675 /*
4676 * Called from OPEN. Look for clientid in reclaim list.
4677 */
4678 __be32
4679 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4680 {
4681         struct nfs4_client *clp;
4682
4683         /* find clientid in conf_id_hashtbl */
4684         clp = find_confirmed_client(clid, sessions, nn);
4685         if (clp == NULL)
4686                 return nfserr_reclaim_bad;
4687
4688         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4689 }
4690
4691 #ifdef CONFIG_NFSD_FAULT_INJECTION
4692
4693 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4694 {
4695         if (mark_client_expired(clp))
4696                 return 0;
4697         expire_client(clp);
4698         return 1;
4699 }
4700
4701 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4702 {
4703         char buf[INET6_ADDRSTRLEN];
4704         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4705         printk(KERN_INFO "NFS Client: %s\n", buf);
4706         return 1;
4707 }
4708
4709 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4710                              const char *type)
4711 {
4712         char buf[INET6_ADDRSTRLEN];
4713         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4714         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4715 }
4716
4717 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4718 {
4719         struct nfs4_openowner *oop;
4720         struct nfs4_lockowner *lop, *lo_next;
4721         struct nfs4_ol_stateid *stp, *st_next;
4722         u64 count = 0;
4723
4724         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4725                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4726                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4727                                 if (func)
4728                                         func(lop);
4729                                 if (++count == max)
4730                                         return count;
4731                         }
4732                 }
4733         }
4734
4735         return count;
4736 }
4737
4738 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4739 {
4740         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4741 }
4742
4743 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4744 {
4745         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4746         nfsd_print_count(clp, count, "locked files");
4747         return count;
4748 }
4749
4750 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4751 {
4752         struct nfs4_openowner *oop, *next;
4753         u64 count = 0;
4754
4755         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4756                 if (func)
4757                         func(oop);
4758                 if (++count == max)
4759                         break;
4760         }
4761
4762         return count;
4763 }
4764
4765 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4766 {
4767         return nfsd_foreach_client_open(clp, max, release_openowner);
4768 }
4769
4770 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4771 {
4772         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4773         nfsd_print_count(clp, count, "open files");
4774         return count;
4775 }
4776
4777 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4778                                      struct list_head *victims)
4779 {
4780         struct nfs4_delegation *dp, *next;
4781         u64 count = 0;
4782
4783         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4784                 if (victims)
4785                         list_move(&dp->dl_recall_lru, victims);
4786                 if (++count == max)
4787                         break;
4788         }
4789         return count;
4790 }
4791
4792 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4793 {
4794         struct nfs4_delegation *dp, *next;
4795         LIST_HEAD(victims);
4796         u64 count;
4797
4798         spin_lock(&recall_lock);
4799         count = nfsd_find_all_delegations(clp, max, &victims);
4800         spin_unlock(&recall_lock);
4801
4802         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4803                 revoke_delegation(dp);
4804
4805         return count;
4806 }
4807
4808 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4809 {
4810         struct nfs4_delegation *dp, *next;
4811         LIST_HEAD(victims);
4812         u64 count;
4813
4814         spin_lock(&recall_lock);
4815         count = nfsd_find_all_delegations(clp, max, &victims);
4816         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4817                 nfsd_break_one_deleg(dp);
4818         spin_unlock(&recall_lock);
4819
4820         return count;
4821 }
4822
4823 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4824 {
4825         u64 count = 0;
4826
4827         spin_lock(&recall_lock);
4828         count = nfsd_find_all_delegations(clp, max, NULL);
4829         spin_unlock(&recall_lock);
4830
4831         nfsd_print_count(clp, count, "delegations");
4832         return count;
4833 }
4834
4835 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4836 {
4837         struct nfs4_client *clp, *next;
4838         u64 count = 0;
4839         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4840
4841         if (!nfsd_netns_ready(nn))
4842                 return 0;
4843
4844         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4845                 count += func(clp, max - count);
4846                 if ((max != 0) && (count >= max))
4847                         break;
4848         }
4849
4850         return count;
4851 }
4852
4853 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4854 {
4855         struct nfs4_client *clp;
4856         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4857
4858         if (!nfsd_netns_ready(nn))
4859                 return NULL;
4860
4861         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4862                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4863                         return clp;
4864         }
4865         return NULL;
4866 }
4867
4868 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4869
4870 /* initialization to perform at module load time: */
4871
4872 void
4873 nfs4_state_init(void)
4874 {
4875 }
4876
4877 /*
4878  * Since the lifetime of a delegation isn't limited to that of an open, a
4879  * client may quite reasonably hang on to a delegation as long as it has
4880  * the inode cached.  This becomes an obvious problem the first time a
4881  * client's inode cache approaches the size of the server's total memory.
4882  *
4883  * For now we avoid this problem by imposing a hard limit on the number
4884  * of delegations, which varies according to the server's memory size.
4885  */
4886 static void
4887 set_max_delegations(void)
4888 {
4889         /*
4890          * Allow at most 4 delegations per megabyte of RAM.  Quick
4891          * estimates suggest that in the worst case (where every delegation
4892          * is for a different inode), a delegation could take about 1.5K,
4893          * giving a worst case usage of about 6% of memory.
4894          */
4895         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4896 }
4897
4898 static int nfs4_state_create_net(struct net *net)
4899 {
4900         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4901         int i;
4902
4903         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4904                         CLIENT_HASH_SIZE, GFP_KERNEL);
4905         if (!nn->conf_id_hashtbl)
4906                 goto err;
4907         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4908                         CLIENT_HASH_SIZE, GFP_KERNEL);
4909         if (!nn->unconf_id_hashtbl)
4910                 goto err_unconf_id;
4911         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4912                         OWNER_HASH_SIZE, GFP_KERNEL);
4913         if (!nn->ownerstr_hashtbl)
4914                 goto err_ownerstr;
4915         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
4916                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
4917         if (!nn->lockowner_ino_hashtbl)
4918                 goto err_lockowner_ino;
4919         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
4920                         SESSION_HASH_SIZE, GFP_KERNEL);
4921         if (!nn->sessionid_hashtbl)
4922                 goto err_sessionid;
4923
4924         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4925                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
4926                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
4927         }
4928         for (i = 0; i < OWNER_HASH_SIZE; i++)
4929                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
4930         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4931                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
4932         for (i = 0; i < SESSION_HASH_SIZE; i++)
4933                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
4934         nn->conf_name_tree = RB_ROOT;
4935         nn->unconf_name_tree = RB_ROOT;
4936         INIT_LIST_HEAD(&nn->client_lru);
4937         INIT_LIST_HEAD(&nn->close_lru);
4938         INIT_LIST_HEAD(&nn->del_recall_lru);
4939         spin_lock_init(&nn->client_lock);
4940
4941         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
4942         get_net(net);
4943
4944         return 0;
4945
4946 err_sessionid:
4947         kfree(nn->lockowner_ino_hashtbl);
4948 err_lockowner_ino:
4949         kfree(nn->ownerstr_hashtbl);
4950 err_ownerstr:
4951         kfree(nn->unconf_id_hashtbl);
4952 err_unconf_id:
4953         kfree(nn->conf_id_hashtbl);
4954 err:
4955         return -ENOMEM;
4956 }
4957
4958 static void
4959 nfs4_state_destroy_net(struct net *net)
4960 {
4961         int i;
4962         struct nfs4_client *clp = NULL;
4963         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4964         struct rb_node *node, *tmp;
4965
4966         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4967                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
4968                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4969                         destroy_client(clp);
4970                 }
4971         }
4972
4973         node = rb_first(&nn->unconf_name_tree);
4974         while (node != NULL) {
4975                 tmp = node;
4976                 node = rb_next(tmp);
4977                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
4978                 rb_erase(tmp, &nn->unconf_name_tree);
4979                 destroy_client(clp);
4980         }
4981
4982         kfree(nn->sessionid_hashtbl);
4983         kfree(nn->lockowner_ino_hashtbl);
4984         kfree(nn->ownerstr_hashtbl);
4985         kfree(nn->unconf_id_hashtbl);
4986         kfree(nn->conf_id_hashtbl);
4987         put_net(net);
4988 }
4989
4990 int
4991 nfs4_state_start_net(struct net *net)
4992 {
4993         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4994         int ret;
4995
4996         ret = nfs4_state_create_net(net);
4997         if (ret)
4998                 return ret;
4999         nfsd4_client_tracking_init(net);
5000         nn->boot_time = get_seconds();
5001         locks_start_grace(net, &nn->nfsd4_manager);
5002         nn->grace_ended = false;
5003         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
5004                nn->nfsd4_grace, net);
5005         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
5006         return 0;
5007 }
5008
5009 /* initialization to perform when the nfsd service is started: */
5010
5011 int
5012 nfs4_state_start(void)
5013 {
5014         int ret;
5015
5016         ret = set_callback_cred();
5017         if (ret)
5018                 return -ENOMEM;
5019         laundry_wq = create_singlethread_workqueue("nfsd4");
5020         if (laundry_wq == NULL) {
5021                 ret = -ENOMEM;
5022                 goto out_recovery;
5023         }
5024         ret = nfsd4_create_callback_queue();
5025         if (ret)
5026                 goto out_free_laundry;
5027
5028         set_max_delegations();
5029
5030         return 0;
5031
5032 out_free_laundry:
5033         destroy_workqueue(laundry_wq);
5034 out_recovery:
5035         return ret;
5036 }
5037
5038 /* should be called with the state lock held */
5039 void
5040 nfs4_state_shutdown_net(struct net *net)
5041 {
5042         struct nfs4_delegation *dp = NULL;
5043         struct list_head *pos, *next, reaplist;
5044         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5045
5046         cancel_delayed_work_sync(&nn->laundromat_work);
5047         locks_end_grace(&nn->nfsd4_manager);
5048
5049         INIT_LIST_HEAD(&reaplist);
5050         spin_lock(&recall_lock);
5051         list_for_each_safe(pos, next, &nn->del_recall_lru) {
5052                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5053                 list_move(&dp->dl_recall_lru, &reaplist);
5054         }
5055         spin_unlock(&recall_lock);
5056         list_for_each_safe(pos, next, &reaplist) {
5057                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5058                 destroy_delegation(dp);
5059         }
5060
5061         nfsd4_client_tracking_exit(net);
5062         nfs4_state_destroy_net(net);
5063 }
5064
5065 void
5066 nfs4_state_shutdown(void)
5067 {
5068         destroy_workqueue(laundry_wq);
5069         nfsd4_destroy_callback_queue();
5070 }
5071
5072 static void
5073 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5074 {
5075         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5076                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5077 }
5078
5079 static void
5080 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5081 {
5082         if (cstate->minorversion) {
5083                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5084                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5085         }
5086 }
5087
5088 void
5089 clear_current_stateid(struct nfsd4_compound_state *cstate)
5090 {
5091         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5092 }
5093
5094 /*
5095  * functions to set current state id
5096  */
5097 void
5098 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5099 {
5100         put_stateid(cstate, &odp->od_stateid);
5101 }
5102
5103 void
5104 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5105 {
5106         put_stateid(cstate, &open->op_stateid);
5107 }
5108
5109 void
5110 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5111 {
5112         put_stateid(cstate, &close->cl_stateid);
5113 }
5114
5115 void
5116 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5117 {
5118         put_stateid(cstate, &lock->lk_resp_stateid);
5119 }
5120
5121 /*
5122  * functions to consume current state id
5123  */
5124
5125 void
5126 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5127 {
5128         get_stateid(cstate, &odp->od_stateid);
5129 }
5130
5131 void
5132 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5133 {
5134         get_stateid(cstate, &drp->dr_stateid);
5135 }
5136
5137 void
5138 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5139 {
5140         get_stateid(cstate, &fsp->fr_stateid);
5141 }
5142
5143 void
5144 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5145 {
5146         get_stateid(cstate, &setattr->sa_stateid);
5147 }
5148
5149 void
5150 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5151 {
5152         get_stateid(cstate, &close->cl_stateid);
5153 }
5154
5155 void
5156 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5157 {
5158         get_stateid(cstate, &locku->lu_stateid);
5159 }
5160
5161 void
5162 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5163 {
5164         get_stateid(cstate, &read->rd_stateid);
5165 }
5166
5167 void
5168 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5169 {
5170         get_stateid(cstate, &write->wr_stateid);
5171 }