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