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/sunrpc/svcauth_gss.h>
  42 #include <linux/sunrpc/clnt.h>
  43 #include "xdr4.h"
  44 #include "vfs.h"
  45 #include "current_stateid.h"
  46
  47 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  48
  49 /* Globals */
  50 time_t nfsd4_lease = 90;     /* default lease time */
  51 time_t nfsd4_grace = 90;
  52 static time_t boot_time;
  53
  54 #define all_ones {{~0,~0},~0}
  55 static const stateid_t one_stateid = {
  56         .si_generation = ~0,
  57         .si_opaque = all_ones,
  58 };
  59 static const stateid_t zero_stateid = {
  60         /* all fields zero */
  61 };
  62 static const stateid_t currentstateid = {
  63         .si_generation = 1,
  64 };
  65
  66 static u64 current_sessionid = 1;
  67
  68 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  69 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  70 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  71
  72 /* forward declarations */
  73 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
  74
  75 /* Locking: */
  76
  77 /* Currently used for almost all code touching nfsv4 state: */
  78 static DEFINE_MUTEX(client_mutex);
  79
  80 /*
  81  * Currently used for the del_recall_lru and file hash table.  In an
  82  * effort to decrease the scope of the client_mutex, this spinlock may
  83  * eventually cover more:
  84  */
  85 static DEFINE_SPINLOCK(recall_lock);
  86
  87 static struct kmem_cache *openowner_slab = NULL;
  88 static struct kmem_cache *lockowner_slab = NULL;
  89 static struct kmem_cache *file_slab = NULL;
  90 static struct kmem_cache *stateid_slab = NULL;
  91 static struct kmem_cache *deleg_slab = NULL;
  92
  93 void
  94 nfs4_lock_state(void)
  95 {
  96         mutex_lock(&client_mutex);
  97 }
  98
  99 static void free_session(struct kref *);
 100
 101 /* Must be called under the client_lock */
 102 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 103 {
 104         kref_put(&ses->se_ref, free_session);
 105 }
 106
 107 static void nfsd4_get_session(struct nfsd4_session *ses)
 108 {
 109         kref_get(&ses->se_ref);
 110 }
 111
 112 void
 113 nfs4_unlock_state(void)
 114 {
 115         mutex_unlock(&client_mutex);
 116 }
 117
 118 static inline u32
 119 opaque_hashval(const void *ptr, int nbytes)
 120 {
 121         unsigned char *cptr = (unsigned char *) ptr;
 122
 123         u32 x = 0;
 124         while (nbytes--) {
 125                 x *= 37;
 126                 x += *cptr++;
 127         }
 128         return x;
 129 }
 130
 131 static struct list_head del_recall_lru;
 132
 133 static void nfsd4_free_file(struct nfs4_file *f)
 134 {
 135         kmem_cache_free(file_slab, f);
 136 }
 137
 138 static inline void
 139 put_nfs4_file(struct nfs4_file *fi)
 140 {
 141         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
 142                 list_del(&fi->fi_hash);
 143                 spin_unlock(&recall_lock);
 144                 iput(fi->fi_inode);
 145                 nfsd4_free_file(fi);
 146         }
 147 }
 148
 149 static inline void
 150 get_nfs4_file(struct nfs4_file *fi)
 151 {
 152         atomic_inc(&fi->fi_ref);
 153 }
 154
 155 static int num_delegations;
 156 unsigned int max_delegations;
 157
 158 /*
 159  * Open owner state (share locks)
 160  */
 161
 162 /* hash tables for lock and open owners */
 163 #define OWNER_HASH_BITS              8
 164 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 165 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 166
 167 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
 168 {
 169         unsigned int ret;
 170
 171         ret = opaque_hashval(ownername->data, ownername->len);
 172         ret += clientid;
 173         return ret & OWNER_HASH_MASK;
 174 }
 175
 176 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
 177
 178 /* hash table for nfs4_file */
 179 #define FILE_HASH_BITS                   8
 180 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 181
 182 static unsigned int file_hashval(struct inode *ino)
 183 {
 184         /* XXX: why are we hashing on inode pointer, anyway? */
 185         return hash_ptr(ino, FILE_HASH_BITS);
 186 }
 187
 188 static struct list_head file_hashtbl[FILE_HASH_SIZE];
 189
 190 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 191 {
 192         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
 193         atomic_inc(&fp->fi_access[oflag]);
 194 }
 195
 196 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
 197 {
 198         if (oflag == O_RDWR) {
 199                 __nfs4_file_get_access(fp, O_RDONLY);
 200                 __nfs4_file_get_access(fp, O_WRONLY);
 201         } else
 202                 __nfs4_file_get_access(fp, oflag);
 203 }
 204
 205 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
 206 {
 207         if (fp->fi_fds[oflag]) {
 208                 fput(fp->fi_fds[oflag]);
 209                 fp->fi_fds[oflag] = NULL;
 210         }
 211 }
 212
 213 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 214 {
 215         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
 216                 nfs4_file_put_fd(fp, oflag);
 217                 /*
 218                  * It's also safe to get rid of the RDWR open *if*
 219                  * we no longer have need of the other kind of access
 220                  * or if we already have the other kind of open:
 221                  */
 222                 if (fp->fi_fds[1-oflag]
 223                         || atomic_read(&fp->fi_access[1 - oflag]) == 0)
 224                         nfs4_file_put_fd(fp, O_RDWR);
 225         }
 226 }
 227
 228 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 229 {
 230         if (oflag == O_RDWR) {
 231                 __nfs4_file_put_access(fp, O_RDONLY);
 232                 __nfs4_file_put_access(fp, O_WRONLY);
 233         } else
 234                 __nfs4_file_put_access(fp, oflag);
 235 }
 236
 237 static inline int get_new_stid(struct nfs4_stid *stid)
 238 {
 239         static int min_stateid = 0;
 240         struct idr *stateids = &stid->sc_client->cl_stateids;
 241         int new_stid;
 242         int error;
 243
 244         error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
 245         /*
 246          * Note: the necessary preallocation was done in
 247          * nfs4_alloc_stateid().  The idr code caps the number of
 248          * preallocations that can exist at a time, but the state lock
 249          * prevents anyone from using ours before we get here:
 250          */
 251         BUG_ON(error);
 252         /*
 253          * It shouldn't be a problem to reuse an opaque stateid value.
 254          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 255          * example, a stray write retransmission could be accepted by
 256          * the server when it should have been rejected.  Therefore,
 257          * adopt a trick from the sctp code to attempt to maximize the
 258          * amount of time until an id is reused, by ensuring they always
 259          * "increase" (mod INT_MAX):
 260          */
 261
 262         min_stateid = new_stid+1;
 263         if (min_stateid == INT_MAX)
 264                 min_stateid = 0;
 265         return new_stid;
 266 }
 267
 268 static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
 269 {
 270         stateid_t *s = &stid->sc_stateid;
 271         int new_id;
 272
 273         stid->sc_type = type;
 274         stid->sc_client = cl;
 275         s->si_opaque.so_clid = cl->cl_clientid;
 276         new_id = get_new_stid(stid);
 277         s->si_opaque.so_id = (u32)new_id;
 278         /* Will be incremented before return to client: */
 279         s->si_generation = 0;
 280 }
 281
 282 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
 283 {
 284         struct idr *stateids = &cl->cl_stateids;
 285
 286         if (!idr_pre_get(stateids, GFP_KERNEL))
 287                 return NULL;
 288         /*
 289          * Note: if we fail here (or any time between now and the time
 290          * we actually get the new idr), we won't need to undo the idr
 291          * preallocation, since the idr code caps the number of
 292          * preallocated entries.
 293          */
 294         return kmem_cache_alloc(slab, GFP_KERNEL);
 295 }
 296
 297 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
 298 {
 299         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
 300 }
 301
 302 static struct nfs4_delegation *
 303 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
 304 {
 305         struct nfs4_delegation *dp;
 306         struct nfs4_file *fp = stp->st_file;
 307
 308         dprintk("NFSD alloc_init_deleg\n");
 309         /*
 310          * Major work on the lease subsystem (for example, to support
 311          * calbacks on stat) will be required before we can support
 312          * write delegations properly.
 313          */
 314         if (type != NFS4_OPEN_DELEGATE_READ)
 315                 return NULL;
 316         if (fp->fi_had_conflict)
 317                 return NULL;
 318         if (num_delegations > max_delegations)
 319                 return NULL;
 320         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
 321         if (dp == NULL)
 322                 return dp;
 323         init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
 324         /*
 325          * delegation seqid's are never incremented.  The 4.1 special
 326          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 327          * 0 anyway just for consistency and use 1:
 328          */
 329         dp->dl_stid.sc_stateid.si_generation = 1;
 330         num_delegations++;
 331         INIT_LIST_HEAD(&dp->dl_perfile);
 332         INIT_LIST_HEAD(&dp->dl_perclnt);
 333         INIT_LIST_HEAD(&dp->dl_recall_lru);
 334         get_nfs4_file(fp);
 335         dp->dl_file = fp;
 336         dp->dl_type = type;
 337         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
 338         dp->dl_time = 0;
 339         atomic_set(&dp->dl_count, 1);
 340         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
 341         return dp;
 342 }
 343
 344 void
 345 nfs4_put_delegation(struct nfs4_delegation *dp)
 346 {
 347         if (atomic_dec_and_test(&dp->dl_count)) {
 348                 dprintk("NFSD: freeing dp %p\n",dp);
 349                 put_nfs4_file(dp->dl_file);
 350                 kmem_cache_free(deleg_slab, dp);
 351                 num_delegations--;
 352         }
 353 }
 354
 355 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 356 {
 357         if (atomic_dec_and_test(&fp->fi_delegees)) {
 358                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
 359                 fp->fi_lease = NULL;
 360                 fput(fp->fi_deleg_file);
 361                 fp->fi_deleg_file = NULL;
 362         }
 363 }
 364
 365 static void unhash_stid(struct nfs4_stid *s)
 366 {
 367         struct idr *stateids = &s->sc_client->cl_stateids;
 368
 369         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
 370 }
 371
 372 /* Called under the state lock. */
 373 static void
 374 unhash_delegation(struct nfs4_delegation *dp)
 375 {
 376         unhash_stid(&dp->dl_stid);
 377         list_del_init(&dp->dl_perclnt);
 378         spin_lock(&recall_lock);
 379         list_del_init(&dp->dl_perfile);
 380         list_del_init(&dp->dl_recall_lru);
 381         spin_unlock(&recall_lock);
 382         nfs4_put_deleg_lease(dp->dl_file);
 383         nfs4_put_delegation(dp);
 384 }
 385
 386 /*
 387  * SETCLIENTID state
 388  */
 389
 390 /* client_lock protects the client lru list and session hash table */
 391 static DEFINE_SPINLOCK(client_lock);
 392
 393 /* Hash tables for nfs4_clientid state */
 394 #define CLIENT_HASH_BITS                 4
 395 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
 396 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
 397
 398 static unsigned int clientid_hashval(u32 id)
 399 {
 400         return id & CLIENT_HASH_MASK;
 401 }
 402
 403 static unsigned int clientstr_hashval(const char *name)
 404 {
 405         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 406 }
 407
 408 /*
 409  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
 410  * used in reboot/reset lease grace period processing
 411  *
 412  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
 413  * setclientid_confirmed info.
 414  *
 415  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
 416  * setclientid info.
 417  *
 418  * client_lru holds client queue ordered by nfs4_client.cl_time
 419  * for lease renewal.
 420  *
 421  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
 422  * for last close replay.
 423  */
 424 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
 425 static int reclaim_str_hashtbl_size = 0;
 426 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
 427 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
 428 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
 429 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
 430 static struct list_head client_lru;
 431 static struct list_head close_lru;
 432
 433 /*
 434  * We store the NONE, READ, WRITE, and BOTH bits separately in the
 435  * st_{access,deny}_bmap field of the stateid, in order to track not
 436  * only what share bits are currently in force, but also what
 437  * combinations of share bits previous opens have used.  This allows us
 438  * to enforce the recommendation of rfc 3530 14.2.19 that the server
 439  * return an error if the client attempt to downgrade to a combination
 440  * of share bits not explicable by closing some of its previous opens.
 441  *
 442  * XXX: This enforcement is actually incomplete, since we don't keep
 443  * track of access/deny bit combinations; so, e.g., we allow:
 444  *
 445  *      OPEN allow read, deny write
 446  *      OPEN allow both, deny none
 447  *      DOWNGRADE allow read, deny none
 448  *
 449  * which we should reject.
 450  */
 451 static unsigned int
 452 bmap_to_share_mode(unsigned long bmap) {
 453         int i;
 454         unsigned int access = 0;
 455
 456         for (i = 1; i < 4; i++) {
 457                 if (test_bit(i, &bmap))
 458                         access |= i;
 459         }
 460         return access;
 461 }
 462
 463 static bool
 464 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
 465         unsigned int access, deny;
 466
 467         access = bmap_to_share_mode(stp->st_access_bmap);
 468         deny = bmap_to_share_mode(stp->st_deny_bmap);
 469         if ((access & open->op_share_deny) || (deny & open->op_share_access))
 470                 return false;
 471         return true;
 472 }
 473
 474 /* set share access for a given stateid */
 475 static inline void
 476 set_access(u32 access, struct nfs4_ol_stateid *stp)
 477 {
 478         __set_bit(access, &stp->st_access_bmap);
 479 }
 480
 481 /* clear share access for a given stateid */
 482 static inline void
 483 clear_access(u32 access, struct nfs4_ol_stateid *stp)
 484 {
 485         __clear_bit(access, &stp->st_access_bmap);
 486 }
 487
 488 /* test whether a given stateid has access */
 489 static inline bool
 490 test_access(u32 access, struct nfs4_ol_stateid *stp)
 491 {
 492         return test_bit(access, &stp->st_access_bmap);
 493 }
 494
 495 /* set share deny for a given stateid */
 496 static inline void
 497 set_deny(u32 access, struct nfs4_ol_stateid *stp)
 498 {
 499         __set_bit(access, &stp->st_deny_bmap);
 500 }
 501
 502 /* clear share deny for a given stateid */
 503 static inline void
 504 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
 505 {
 506         __clear_bit(access, &stp->st_deny_bmap);
 507 }
 508
 509 /* test whether a given stateid is denying specific access */
 510 static inline bool
 511 test_deny(u32 access, struct nfs4_ol_stateid *stp)
 512 {
 513         return test_bit(access, &stp->st_deny_bmap);
 514 }
 515
 516 static int nfs4_access_to_omode(u32 access)
 517 {
 518         switch (access & NFS4_SHARE_ACCESS_BOTH) {
 519         case NFS4_SHARE_ACCESS_READ:
 520                 return O_RDONLY;
 521         case NFS4_SHARE_ACCESS_WRITE:
 522                 return O_WRONLY;
 523         case NFS4_SHARE_ACCESS_BOTH:
 524                 return O_RDWR;
 525         }
 526         BUG();
 527 }
 528
 529 /* release all access and file references for a given stateid */
 530 static void
 531 release_all_access(struct nfs4_ol_stateid *stp)
 532 {
 533         int i;
 534
 535         for (i = 1; i < 4; i++) {
 536                 if (test_access(i, stp))
 537                         nfs4_file_put_access(stp->st_file,
 538                                              nfs4_access_to_omode(i));
 539                 clear_access(i, stp);
 540         }
 541 }
 542
 543 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
 544 {
 545         list_del(&stp->st_perfile);
 546         list_del(&stp->st_perstateowner);
 547 }
 548
 549 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
 550 {
 551         release_all_access(stp);
 552         put_nfs4_file(stp->st_file);
 553         stp->st_file = NULL;
 554 }
 555
 556 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
 557 {
 558         kmem_cache_free(stateid_slab, stp);
 559 }
 560
 561 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
 562 {
 563         struct file *file;
 564
 565         unhash_generic_stateid(stp);
 566         unhash_stid(&stp->st_stid);
 567         file = find_any_file(stp->st_file);
 568         if (file)
 569                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
 570         close_generic_stateid(stp);
 571         free_generic_stateid(stp);
 572 }
 573
 574 static void unhash_lockowner(struct nfs4_lockowner *lo)
 575 {
 576         struct nfs4_ol_stateid *stp;
 577
 578         list_del(&lo->lo_owner.so_strhash);
 579         list_del(&lo->lo_perstateid);
 580         list_del(&lo->lo_owner_ino_hash);
 581         while (!list_empty(&lo->lo_owner.so_stateids)) {
 582                 stp = list_first_entry(&lo->lo_owner.so_stateids,
 583                                 struct nfs4_ol_stateid, st_perstateowner);
 584                 release_lock_stateid(stp);
 585         }
 586 }
 587
 588 static void release_lockowner(struct nfs4_lockowner *lo)
 589 {
 590         unhash_lockowner(lo);
 591         nfs4_free_lockowner(lo);
 592 }
 593
 594 static void
 595 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
 596 {
 597         struct nfs4_lockowner *lo;
 598
 599         while (!list_empty(&open_stp->st_lockowners)) {
 600                 lo = list_entry(open_stp->st_lockowners.next,
 601                                 struct nfs4_lockowner, lo_perstateid);
 602                 release_lockowner(lo);
 603         }
 604 }
 605
 606 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
 607 {
 608         unhash_generic_stateid(stp);
 609         release_stateid_lockowners(stp);
 610         close_generic_stateid(stp);
 611 }
 612
 613 static void release_open_stateid(struct nfs4_ol_stateid *stp)
 614 {
 615         unhash_open_stateid(stp);
 616         unhash_stid(&stp->st_stid);
 617         free_generic_stateid(stp);
 618 }
 619
 620 static void unhash_openowner(struct nfs4_openowner *oo)
 621 {
 622         struct nfs4_ol_stateid *stp;
 623
 624         list_del(&oo->oo_owner.so_strhash);
 625         list_del(&oo->oo_perclient);
 626         while (!list_empty(&oo->oo_owner.so_stateids)) {
 627                 stp = list_first_entry(&oo->oo_owner.so_stateids,
 628                                 struct nfs4_ol_stateid, st_perstateowner);
 629                 release_open_stateid(stp);
 630         }
 631 }
 632
 633 static void release_last_closed_stateid(struct nfs4_openowner *oo)
 634 {
 635         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
 636
 637         if (s) {
 638                 unhash_stid(&s->st_stid);
 639                 free_generic_stateid(s);
 640                 oo->oo_last_closed_stid = NULL;
 641         }
 642 }
 643
 644 static void release_openowner(struct nfs4_openowner *oo)
 645 {
 646         unhash_openowner(oo);
 647         list_del(&oo->oo_close_lru);
 648         release_last_closed_stateid(oo);
 649         nfs4_free_openowner(oo);
 650 }
 651
 652 #define SESSION_HASH_SIZE       512
 653 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
 654
 655 static inline int
 656 hash_sessionid(struct nfs4_sessionid *sessionid)
 657 {
 658         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
 659
 660         return sid->sequence % SESSION_HASH_SIZE;
 661 }
 662
 663 #ifdef NFSD_DEBUG
 664 static inline void
 665 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 666 {
 667         u32 *ptr = (u32 *)(&sessionid->data[0]);
 668         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
 669 }
 670 #else
 671 static inline void
 672 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
 673 {
 674 }
 675 #endif
 676
 677
 678 static void
 679 gen_sessionid(struct nfsd4_session *ses)
 680 {
 681         struct nfs4_client *clp = ses->se_client;
 682         struct nfsd4_sessionid *sid;
 683
 684         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
 685         sid->clientid = clp->cl_clientid;
 686         sid->sequence = current_sessionid++;
 687         sid->reserved = 0;
 688 }
 689
 690 /*
 691  * The protocol defines ca_maxresponssize_cached to include the size of
 692  * the rpc header, but all we need to cache is the data starting after
 693  * the end of the initial SEQUENCE operation--the rest we regenerate
 694  * each time.  Therefore we can advertise a ca_maxresponssize_cached
 695  * value that is the number of bytes in our cache plus a few additional
 696  * bytes.  In order to stay on the safe side, and not promise more than
 697  * we can cache, those additional bytes must be the minimum possible: 24
 698  * bytes of rpc header (xid through accept state, with AUTH_NULL
 699  * verifier), 12 for the compound header (with zero-length tag), and 44
 700  * for the SEQUENCE op response:
 701  */
 702 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
 703
 704 static void
 705 free_session_slots(struct nfsd4_session *ses)
 706 {
 707         int i;
 708
 709         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
 710                 kfree(ses->se_slots[i]);
 711 }
 712
 713 /*
 714  * We don't actually need to cache the rpc and session headers, so we
 715  * can allocate a little less for each slot:
 716  */
 717 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
 718 {
 719         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
 720 }
 721
 722 static int nfsd4_sanitize_slot_size(u32 size)
 723 {
 724         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
 725         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
 726
 727         return size;
 728 }
 729
 730 /*
 731  * XXX: If we run out of reserved DRC memory we could (up to a point)
 732  * re-negotiate active sessions and reduce their slot usage to make
 733  * room for new connections. For now we just fail the create session.
 734  */
 735 static int nfsd4_get_drc_mem(int slotsize, u32 num)
 736 {
 737         int avail;
 738
 739         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
 740
 741         spin_lock(&nfsd_drc_lock);
 742         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
 743                         nfsd_drc_max_mem - nfsd_drc_mem_used);
 744         num = min_t(int, num, avail / slotsize);
 745         nfsd_drc_mem_used += num * slotsize;
 746         spin_unlock(&nfsd_drc_lock);
 747
 748         return num;
 749 }
 750
 751 static void nfsd4_put_drc_mem(int slotsize, int num)
 752 {
 753         spin_lock(&nfsd_drc_lock);
 754         nfsd_drc_mem_used -= slotsize * num;
 755         spin_unlock(&nfsd_drc_lock);
 756 }
 757
 758 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
 759 {
 760         struct nfsd4_session *new;
 761         int mem, i;
 762
 763         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
 764                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
 765         mem = numslots * sizeof(struct nfsd4_slot *);
 766
 767         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
 768         if (!new)
 769                 return NULL;
 770         /* allocate each struct nfsd4_slot and data cache in one piece */
 771         for (i = 0; i < numslots; i++) {
 772                 mem = sizeof(struct nfsd4_slot) + slotsize;
 773                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
 774                 if (!new->se_slots[i])
 775                         goto out_free;
 776         }
 777         return new;
 778 out_free:
 779         while (i--)
 780                 kfree(new->se_slots[i]);
 781         kfree(new);
 782         return NULL;
 783 }
 784
 785 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
 786 {
 787         u32 maxrpc = nfsd_serv->sv_max_mesg;
 788
 789         new->maxreqs = numslots;
 790         new->maxresp_cached = min_t(u32, req->maxresp_cached,
 791                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
 792         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
 793         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
 794         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
 795 }
 796
 797 static void free_conn(struct nfsd4_conn *c)
 798 {
 799         svc_xprt_put(c->cn_xprt);
 800         kfree(c);
 801 }
 802
 803 static void nfsd4_conn_lost(struct svc_xpt_user *u)
 804 {
 805         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
 806         struct nfs4_client *clp = c->cn_session->se_client;
 807
 808         spin_lock(&clp->cl_lock);
 809         if (!list_empty(&c->cn_persession)) {
 810                 list_del(&c->cn_persession);
 811                 free_conn(c);
 812         }
 813         spin_unlock(&clp->cl_lock);
 814         nfsd4_probe_callback(clp);
 815 }
 816
 817 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
 818 {
 819         struct nfsd4_conn *conn;
 820
 821         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
 822         if (!conn)
 823                 return NULL;
 824         svc_xprt_get(rqstp->rq_xprt);
 825         conn->cn_xprt = rqstp->rq_xprt;
 826         conn->cn_flags = flags;
 827         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
 828         return conn;
 829 }
 830
 831 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 832 {
 833         conn->cn_session = ses;
 834         list_add(&conn->cn_persession, &ses->se_conns);
 835 }
 836
 837 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
 838 {
 839         struct nfs4_client *clp = ses->se_client;
 840
 841         spin_lock(&clp->cl_lock);
 842         __nfsd4_hash_conn(conn, ses);
 843         spin_unlock(&clp->cl_lock);
 844 }
 845
 846 static int nfsd4_register_conn(struct nfsd4_conn *conn)
 847 {
 848         conn->cn_xpt_user.callback = nfsd4_conn_lost;
 849         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
 850 }
 851
 852 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
 853 {
 854         struct nfsd4_conn *conn;
 855         int ret;
 856
 857         conn = alloc_conn(rqstp, dir);
 858         if (!conn)
 859                 return nfserr_jukebox;
 860         nfsd4_hash_conn(conn, ses);
 861         ret = nfsd4_register_conn(conn);
 862         if (ret)
 863                 /* oops; xprt is already down: */
 864                 nfsd4_conn_lost(&conn->cn_xpt_user);
 865         return nfs_ok;
 866 }
 867
 868 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
 869 {
 870         u32 dir = NFS4_CDFC4_FORE;
 871
 872         if (ses->se_flags & SESSION4_BACK_CHAN)
 873                 dir |= NFS4_CDFC4_BACK;
 874
 875         return nfsd4_new_conn(rqstp, ses, dir);
 876 }
 877
 878 /* must be called under client_lock */
 879 static void nfsd4_del_conns(struct nfsd4_session *s)
 880 {
 881         struct nfs4_client *clp = s->se_client;
 882         struct nfsd4_conn *c;
 883
 884         spin_lock(&clp->cl_lock);
 885         while (!list_empty(&s->se_conns)) {
 886                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
 887                 list_del_init(&c->cn_persession);
 888                 spin_unlock(&clp->cl_lock);
 889
 890                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
 891                 free_conn(c);
 892
 893                 spin_lock(&clp->cl_lock);
 894         }
 895         spin_unlock(&clp->cl_lock);
 896 }
 897
 898 static void free_session(struct kref *kref)
 899 {
 900         struct nfsd4_session *ses;
 901         int mem;
 902
 903         BUG_ON(!spin_is_locked(&client_lock));
 904         ses = container_of(kref, struct nfsd4_session, se_ref);
 905         nfsd4_del_conns(ses);
 906         spin_lock(&nfsd_drc_lock);
 907         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
 908         nfsd_drc_mem_used -= mem;
 909         spin_unlock(&nfsd_drc_lock);
 910         free_session_slots(ses);
 911         kfree(ses);
 912 }
 913
 914 void nfsd4_put_session(struct nfsd4_session *ses)
 915 {
 916         spin_lock(&client_lock);
 917         nfsd4_put_session_locked(ses);
 918         spin_unlock(&client_lock);
 919 }
 920
 921 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
 922 {
 923         struct nfsd4_session *new;
 924         struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
 925         int numslots, slotsize;
 926         __be32 status;
 927         int idx;
 928
 929         /*
 930          * Note decreasing slot size below client's request may
 931          * make it difficult for client to function correctly, whereas
 932          * decreasing the number of slots will (just?) affect
 933          * performance.  When short on memory we therefore prefer to
 934          * decrease number of slots instead of their size.
 935          */
 936         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
 937         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
 938         if (numslots < 1)
 939                 return NULL;
 940
 941         new = alloc_session(slotsize, numslots);
 942         if (!new) {
 943                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
 944                 return NULL;
 945         }
 946         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
 947
 948         new->se_client = clp;
 949         gen_sessionid(new);
 950
 951         INIT_LIST_HEAD(&new->se_conns);
 952
 953         new->se_cb_seq_nr = 1;
 954         new->se_flags = cses->flags;
 955         new->se_cb_prog = cses->callback_prog;
 956         kref_init(&new->se_ref);
 957         idx = hash_sessionid(&new->se_sessionid);
 958         spin_lock(&client_lock);
 959         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
 960         spin_lock(&clp->cl_lock);
 961         list_add(&new->se_perclnt, &clp->cl_sessions);
 962         spin_unlock(&clp->cl_lock);
 963         spin_unlock(&client_lock);
 964
 965         status = nfsd4_new_conn_from_crses(rqstp, new);
 966         /* whoops: benny points out, status is ignored! (err, or bogus) */
 967         if (status) {
 968                 spin_lock(&client_lock);
 969                 free_session(&new->se_ref);
 970                 spin_unlock(&client_lock);
 971                 return NULL;
 972         }
 973         if (cses->flags & SESSION4_BACK_CHAN) {
 974                 struct sockaddr *sa = svc_addr(rqstp);
 975                 /*
 976                  * This is a little silly; with sessions there's no real
 977                  * use for the callback address.  Use the peer address
 978                  * as a reasonable default for now, but consider fixing
 979                  * the rpc client not to require an address in the
 980                  * future:
 981                  */
 982                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
 983                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
 984         }
 985         nfsd4_probe_callback(clp);
 986         return new;
 987 }
 988
 989 /* caller must hold client_lock */
 990 static struct nfsd4_session *
 991 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
 992 {
 993         struct nfsd4_session *elem;
 994         int idx;
 995
 996         dump_sessionid(__func__, sessionid);
 997         idx = hash_sessionid(sessionid);
 998         /* Search in the appropriate list */
 999         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
1000                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1001                             NFS4_MAX_SESSIONID_LEN)) {
1002                         return elem;
1003                 }
1004         }
1005
1006         dprintk("%s: session not found\n", __func__);
1007         return NULL;
1008 }
1009
1010 /* caller must hold client_lock */
1011 static void
1012 unhash_session(struct nfsd4_session *ses)
1013 {
1014         list_del(&ses->se_hash);
1015         spin_lock(&ses->se_client->cl_lock);
1016         list_del(&ses->se_perclnt);
1017         spin_unlock(&ses->se_client->cl_lock);
1018 }
1019
1020 /* must be called under the client_lock */
1021 static inline void
1022 renew_client_locked(struct nfs4_client *clp)
1023 {
1024         if (is_client_expired(clp)) {
1025                 WARN_ON(1);
1026                 printk("%s: client (clientid %08x/%08x) already expired\n",
1027                         __func__,
1028                         clp->cl_clientid.cl_boot,
1029                         clp->cl_clientid.cl_id);
1030                 return;
1031         }
1032
1033         dprintk("renewing client (clientid %08x/%08x)\n",
1034                         clp->cl_clientid.cl_boot,
1035                         clp->cl_clientid.cl_id);
1036         list_move_tail(&clp->cl_lru, &client_lru);
1037         clp->cl_time = get_seconds();
1038 }
1039
1040 static inline void
1041 renew_client(struct nfs4_client *clp)
1042 {
1043         spin_lock(&client_lock);
1044         renew_client_locked(clp);
1045         spin_unlock(&client_lock);
1046 }
1047
1048 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1049 static int
1050 STALE_CLIENTID(clientid_t *clid)
1051 {
1052         if (clid->cl_boot == boot_time)
1053                 return 0;
1054         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1055                 clid->cl_boot, clid->cl_id, boot_time);
1056         return 1;
1057 }
1058
1059 /*
1060  * XXX Should we use a slab cache ?
1061  * This type of memory management is somewhat inefficient, but we use it
1062  * anyway since SETCLIENTID is not a common operation.
1063  */
1064 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1065 {
1066         struct nfs4_client *clp;
1067
1068         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1069         if (clp == NULL)
1070                 return NULL;
1071         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1072         if (clp->cl_name.data == NULL) {
1073                 kfree(clp);
1074                 return NULL;
1075         }
1076         clp->cl_name.len = name.len;
1077         return clp;
1078 }
1079
1080 static inline void
1081 free_client(struct nfs4_client *clp)
1082 {
1083         BUG_ON(!spin_is_locked(&client_lock));
1084         while (!list_empty(&clp->cl_sessions)) {
1085                 struct nfsd4_session *ses;
1086                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1087                                 se_perclnt);
1088                 list_del(&ses->se_perclnt);
1089                 nfsd4_put_session_locked(ses);
1090         }
1091         free_svc_cred(&clp->cl_cred);
1092         kfree(clp->cl_name.data);
1093         kfree(clp);
1094 }
1095
1096 void
1097 release_session_client(struct nfsd4_session *session)
1098 {
1099         struct nfs4_client *clp = session->se_client;
1100
1101         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1102                 return;
1103         if (is_client_expired(clp)) {
1104                 free_client(clp);
1105                 session->se_client = NULL;
1106         } else
1107                 renew_client_locked(clp);
1108         spin_unlock(&client_lock);
1109 }
1110
1111 /* must be called under the client_lock */
1112 static inline void
1113 unhash_client_locked(struct nfs4_client *clp)
1114 {
1115         struct nfsd4_session *ses;
1116
1117         mark_client_expired(clp);
1118         list_del(&clp->cl_lru);
1119         spin_lock(&clp->cl_lock);
1120         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1121                 list_del_init(&ses->se_hash);
1122         spin_unlock(&clp->cl_lock);
1123 }
1124
1125 static void
1126 expire_client(struct nfs4_client *clp)
1127 {
1128         struct nfs4_openowner *oo;
1129         struct nfs4_delegation *dp;
1130         struct list_head reaplist;
1131
1132         INIT_LIST_HEAD(&reaplist);
1133         spin_lock(&recall_lock);
1134         while (!list_empty(&clp->cl_delegations)) {
1135                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1136                 list_del_init(&dp->dl_perclnt);
1137                 list_move(&dp->dl_recall_lru, &reaplist);
1138         }
1139         spin_unlock(&recall_lock);
1140         while (!list_empty(&reaplist)) {
1141                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1142                 unhash_delegation(dp);
1143         }
1144         while (!list_empty(&clp->cl_openowners)) {
1145                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1146                 release_openowner(oo);
1147         }
1148         nfsd4_shutdown_callback(clp);
1149         if (clp->cl_cb_conn.cb_xprt)
1150                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1151         list_del(&clp->cl_idhash);
1152         list_del(&clp->cl_strhash);
1153         spin_lock(&client_lock);
1154         unhash_client_locked(clp);
1155         if (atomic_read(&clp->cl_refcount) == 0)
1156                 free_client(clp);
1157         spin_unlock(&client_lock);
1158 }
1159
1160 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1161 {
1162         memcpy(target->cl_verifier.data, source->data,
1163                         sizeof(target->cl_verifier.data));
1164 }
1165
1166 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1167 {
1168         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1169         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1170 }
1171
1172 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1173 {
1174         if (source->cr_principal) {
1175                 target->cr_principal =
1176                                 kstrdup(source->cr_principal, GFP_KERNEL);
1177                 if (target->cr_principal == NULL)
1178                         return -ENOMEM;
1179         } else
1180                 target->cr_principal = NULL;
1181         target->cr_flavor = source->cr_flavor;
1182         target->cr_uid = source->cr_uid;
1183         target->cr_gid = source->cr_gid;
1184         target->cr_group_info = source->cr_group_info;
1185         get_group_info(target->cr_group_info);
1186         return 0;
1187 }
1188
1189 static int same_name(const char *n1, const char *n2)
1190 {
1191         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1192 }
1193
1194 static int
1195 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1196 {
1197         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1198 }
1199
1200 static int
1201 same_clid(clientid_t *cl1, clientid_t *cl2)
1202 {
1203         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1204 }
1205
1206 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1207 {
1208         int i;
1209
1210         if (g1->ngroups != g2->ngroups)
1211                 return false;
1212         for (i=0; i<g1->ngroups; i++)
1213                 if (GROUP_AT(g1, i) != GROUP_AT(g2, i))
1214                         return false;
1215         return true;
1216 }
1217
1218 static int
1219 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1220 {
1221         if ((cr1->cr_flavor != cr2->cr_flavor)
1222                 || (cr1->cr_uid != cr2->cr_uid)
1223                 || (cr1->cr_gid != cr2->cr_gid)
1224                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1225                 return false;
1226         if (cr1->cr_principal == cr2->cr_principal)
1227                 return true;
1228         if (!cr1->cr_principal || !cr2->cr_principal)
1229                 return false;
1230         return 0 == strcmp(cr1->cr_principal, cr1->cr_principal);
1231 }
1232
1233 static void gen_clid(struct nfs4_client *clp)
1234 {
1235         static u32 current_clientid = 1;
1236
1237         clp->cl_clientid.cl_boot = boot_time;
1238         clp->cl_clientid.cl_id = current_clientid++;
1239 }
1240
1241 static void gen_confirm(struct nfs4_client *clp)
1242 {
1243         __be32 verf[2];
1244         static u32 i;
1245
1246         verf[0] = (__be32)get_seconds();
1247         verf[1] = (__be32)i++;
1248         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1249 }
1250
1251 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1252 {
1253         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1254 }
1255
1256 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1257 {
1258         struct nfs4_stid *s;
1259
1260         s = find_stateid(cl, t);
1261         if (!s)
1262                 return NULL;
1263         if (typemask & s->sc_type)
1264                 return s;
1265         return NULL;
1266 }
1267
1268 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1269                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1270 {
1271         struct nfs4_client *clp;
1272         struct sockaddr *sa = svc_addr(rqstp);
1273         int ret;
1274
1275         clp = alloc_client(name);
1276         if (clp == NULL)
1277                 return NULL;
1278
1279         INIT_LIST_HEAD(&clp->cl_sessions);
1280         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1281         if (ret) {
1282                 spin_lock(&client_lock);
1283                 free_client(clp);
1284                 spin_unlock(&client_lock);
1285                 return NULL;
1286         }
1287         idr_init(&clp->cl_stateids);
1288         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1289         atomic_set(&clp->cl_refcount, 0);
1290         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1291         INIT_LIST_HEAD(&clp->cl_idhash);
1292         INIT_LIST_HEAD(&clp->cl_strhash);
1293         INIT_LIST_HEAD(&clp->cl_openowners);
1294         INIT_LIST_HEAD(&clp->cl_delegations);
1295         INIT_LIST_HEAD(&clp->cl_lru);
1296         INIT_LIST_HEAD(&clp->cl_callbacks);
1297         spin_lock_init(&clp->cl_lock);
1298         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1299         clp->cl_time = get_seconds();
1300         clear_bit(0, &clp->cl_cb_slot_busy);
1301         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1302         copy_verf(clp, verf);
1303         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1304         gen_confirm(clp);
1305         clp->cl_cb_session = NULL;
1306         return clp;
1307 }
1308
1309 static void
1310 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1311 {
1312         unsigned int idhashval;
1313
1314         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1315         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1316         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1317         renew_client(clp);
1318 }
1319
1320 static void
1321 move_to_confirmed(struct nfs4_client *clp)
1322 {
1323         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1324         unsigned int strhashval;
1325
1326         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1327         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1328         strhashval = clientstr_hashval(clp->cl_recdir);
1329         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1330         renew_client(clp);
1331 }
1332
1333 static struct nfs4_client *
1334 find_confirmed_client(clientid_t *clid)
1335 {
1336         struct nfs4_client *clp;
1337         unsigned int idhashval = clientid_hashval(clid->cl_id);
1338
1339         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1340                 if (same_clid(&clp->cl_clientid, clid)) {
1341                         renew_client(clp);
1342                         return clp;
1343                 }
1344         }
1345         return NULL;
1346 }
1347
1348 static struct nfs4_client *
1349 find_unconfirmed_client(clientid_t *clid)
1350 {
1351         struct nfs4_client *clp;
1352         unsigned int idhashval = clientid_hashval(clid->cl_id);
1353
1354         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1355                 if (same_clid(&clp->cl_clientid, clid))
1356                         return clp;
1357         }
1358         return NULL;
1359 }
1360
1361 static bool clp_used_exchangeid(struct nfs4_client *clp)
1362 {
1363         return clp->cl_exchange_flags != 0;
1364 }
1365
1366 static struct nfs4_client *
1367 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1368 {
1369         struct nfs4_client *clp;
1370
1371         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1372                 if (same_name(clp->cl_recdir, dname))
1373                         return clp;
1374         }
1375         return NULL;
1376 }
1377
1378 static struct nfs4_client *
1379 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1380 {
1381         struct nfs4_client *clp;
1382
1383         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1384                 if (same_name(clp->cl_recdir, dname))
1385                         return clp;
1386         }
1387         return NULL;
1388 }
1389
1390 static void
1391 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1392 {
1393         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1394         struct sockaddr *sa = svc_addr(rqstp);
1395         u32 scopeid = rpc_get_scope_id(sa);
1396         unsigned short expected_family;
1397
1398         /* Currently, we only support tcp and tcp6 for the callback channel */
1399         if (se->se_callback_netid_len == 3 &&
1400             !memcmp(se->se_callback_netid_val, "tcp", 3))
1401                 expected_family = AF_INET;
1402         else if (se->se_callback_netid_len == 4 &&
1403                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1404                 expected_family = AF_INET6;
1405         else
1406                 goto out_err;
1407
1408         conn->cb_addrlen = rpc_uaddr2sockaddr(&init_net, se->se_callback_addr_val,
1409                                             se->se_callback_addr_len,
1410                                             (struct sockaddr *)&conn->cb_addr,
1411                                             sizeof(conn->cb_addr));
1412
1413         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1414                 goto out_err;
1415
1416         if (conn->cb_addr.ss_family == AF_INET6)
1417                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1418
1419         conn->cb_prog = se->se_callback_prog;
1420         conn->cb_ident = se->se_callback_ident;
1421         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1422         return;
1423 out_err:
1424         conn->cb_addr.ss_family = AF_UNSPEC;
1425         conn->cb_addrlen = 0;
1426         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1427                 "will not receive delegations\n",
1428                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1429
1430         return;
1431 }
1432
1433 /*
1434  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1435  */
1436 void
1437 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1438 {
1439         struct nfsd4_slot *slot = resp->cstate.slot;
1440         unsigned int base;
1441
1442         dprintk("--> %s slot %p\n", __func__, slot);
1443
1444         slot->sl_opcnt = resp->opcnt;
1445         slot->sl_status = resp->cstate.status;
1446
1447         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1448         if (nfsd4_not_cached(resp)) {
1449                 slot->sl_datalen = 0;
1450                 return;
1451         }
1452         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1453         base = (char *)resp->cstate.datap -
1454                                         (char *)resp->xbuf->head[0].iov_base;
1455         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1456                                     slot->sl_datalen))
1457                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1458         return;
1459 }
1460
1461 /*
1462  * Encode the replay sequence operation from the slot values.
1463  * If cachethis is FALSE encode the uncached rep error on the next
1464  * operation which sets resp->p and increments resp->opcnt for
1465  * nfs4svc_encode_compoundres.
1466  *
1467  */
1468 static __be32
1469 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1470                           struct nfsd4_compoundres *resp)
1471 {
1472         struct nfsd4_op *op;
1473         struct nfsd4_slot *slot = resp->cstate.slot;
1474
1475         /* Encode the replayed sequence operation */
1476         op = &args->ops[resp->opcnt - 1];
1477         nfsd4_encode_operation(resp, op);
1478
1479         /* Return nfserr_retry_uncached_rep in next operation. */
1480         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1481                 op = &args->ops[resp->opcnt++];
1482                 op->status = nfserr_retry_uncached_rep;
1483                 nfsd4_encode_operation(resp, op);
1484         }
1485         return op->status;
1486 }
1487
1488 /*
1489  * The sequence operation is not cached because we can use the slot and
1490  * session values.
1491  */
1492 __be32
1493 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1494                          struct nfsd4_sequence *seq)
1495 {
1496         struct nfsd4_slot *slot = resp->cstate.slot;
1497         __be32 status;
1498
1499         dprintk("--> %s slot %p\n", __func__, slot);
1500
1501         /* Either returns 0 or nfserr_retry_uncached */
1502         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1503         if (status == nfserr_retry_uncached_rep)
1504                 return status;
1505
1506         /* The sequence operation has been encoded, cstate->datap set. */
1507         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1508
1509         resp->opcnt = slot->sl_opcnt;
1510         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1511         status = slot->sl_status;
1512
1513         return status;
1514 }
1515
1516 /*
1517  * Set the exchange_id flags returned by the server.
1518  */
1519 static void
1520 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1521 {
1522         /* pNFS is not supported */
1523         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1524
1525         /* Referrals are supported, Migration is not. */
1526         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1527
1528         /* set the wire flags to return to client. */
1529         clid->flags = new->cl_exchange_flags;
1530 }
1531
1532 static bool client_has_state(struct nfs4_client *clp)
1533 {
1534         /*
1535          * Note clp->cl_openowners check isn't quite right: there's no
1536          * need to count owners without stateid's.
1537          *
1538          * Also note we should probably be using this in 4.0 case too.
1539          */
1540         return list_empty(&clp->cl_openowners)
1541                 && list_empty(&clp->cl_delegations)
1542                 && list_empty(&clp->cl_sessions);
1543 }
1544
1545 __be32
1546 nfsd4_exchange_id(struct svc_rqst *rqstp,
1547                   struct nfsd4_compound_state *cstate,
1548                   struct nfsd4_exchange_id *exid)
1549 {
1550         struct nfs4_client *unconf, *conf, *new;
1551         __be32 status;
1552         unsigned int            strhashval;
1553         char                    dname[HEXDIR_LEN];
1554         char                    addr_str[INET6_ADDRSTRLEN];
1555         nfs4_verifier           verf = exid->verifier;
1556         struct sockaddr         *sa = svc_addr(rqstp);
1557         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1558
1559         rpc_ntop(sa, addr_str, sizeof(addr_str));
1560         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1561                 "ip_addr=%s flags %x, spa_how %d\n",
1562                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1563                 addr_str, exid->flags, exid->spa_how);
1564
1565         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1566                 return nfserr_inval;
1567
1568         /* Currently only support SP4_NONE */
1569         switch (exid->spa_how) {
1570         case SP4_NONE:
1571                 break;
1572         case SP4_SSV:
1573                 return nfserr_serverfault;
1574         default:
1575                 BUG();                          /* checked by xdr code */
1576         case SP4_MACH_CRED:
1577                 return nfserr_serverfault;      /* no excuse :-/ */
1578         }
1579
1580         status = nfs4_make_rec_clidname(dname, &exid->clname);
1581
1582         if (status)
1583                 return status;
1584
1585         strhashval = clientstr_hashval(dname);
1586
1587         /* Cases below refer to rfc 5661 section 18.35.4: */
1588         nfs4_lock_state();
1589         conf = find_confirmed_client_by_str(dname, strhashval);
1590         if (conf) {
1591                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1592                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1593
1594                 if (update) {
1595                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1596                                 status = nfserr_inval;
1597                                 goto out;
1598                         }
1599                         if (!creds_match) { /* case 9 */
1600                                 status = nfserr_perm;
1601                                 goto out;
1602                         }
1603                         if (!verfs_match) { /* case 8 */
1604                                 status = nfserr_not_same;
1605                                 goto out;
1606                         }
1607                         /* case 6 */
1608                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1609                         new = conf;
1610                         goto out_copy;
1611                 }
1612                 if (!creds_match) { /* case 3 */
1613                         if (client_has_state(conf)) {
1614                                 status = nfserr_clid_inuse;
1615                                 goto out;
1616                         }
1617                         goto expire_client;
1618                 }
1619                 if (verfs_match) { /* case 2 */
1620                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1621                         new = conf;
1622                         goto out_copy;
1623                 }
1624                 /* case 5, client reboot */
1625 expire_client:
1626                 expire_client(conf);
1627                 goto out_new;
1628         }
1629
1630         if (update) { /* case 7 */
1631                 status = nfserr_noent;
1632                 goto out;
1633         }
1634
1635         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1636         if (unconf) /* case 4, possible retry or client restart */
1637                 expire_client(unconf);
1638
1639         /* case 1 (normal case) */
1640 out_new:
1641         new = create_client(exid->clname, dname, rqstp, &verf);
1642         if (new == NULL) {
1643                 status = nfserr_jukebox;
1644                 goto out;
1645         }
1646
1647         gen_clid(new);
1648         add_to_unconfirmed(new, strhashval);
1649 out_copy:
1650         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1651         exid->clientid.cl_id = new->cl_clientid.cl_id;
1652
1653         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1654         nfsd4_set_ex_flags(new, exid);
1655
1656         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1657                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1658         status = nfs_ok;
1659
1660 out:
1661         nfs4_unlock_state();
1662         return status;
1663 }
1664
1665 static __be32
1666 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1667 {
1668         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1669                 slot_seqid);
1670
1671         /* The slot is in use, and no response has been sent. */
1672         if (slot_inuse) {
1673                 if (seqid == slot_seqid)
1674                         return nfserr_jukebox;
1675                 else
1676                         return nfserr_seq_misordered;
1677         }
1678         /* Note unsigned 32-bit arithmetic handles wraparound: */
1679         if (likely(seqid == slot_seqid + 1))
1680                 return nfs_ok;
1681         if (seqid == slot_seqid)
1682                 return nfserr_replay_cache;
1683         return nfserr_seq_misordered;
1684 }
1685
1686 /*
1687  * Cache the create session result into the create session single DRC
1688  * slot cache by saving the xdr structure. sl_seqid has been set.
1689  * Do this for solo or embedded create session operations.
1690  */
1691 static void
1692 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1693                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1694 {
1695         slot->sl_status = nfserr;
1696         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1697 }
1698
1699 static __be32
1700 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1701                             struct nfsd4_clid_slot *slot)
1702 {
1703         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1704         return slot->sl_status;
1705 }
1706
1707 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1708                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1709                         1 +     /* MIN tag is length with zero, only length */ \
1710                         3 +     /* version, opcount, opcode */ \
1711                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1712                                 /* seqid, slotID, slotID, cache */ \
1713                         4 ) * sizeof(__be32))
1714
1715 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1716                         2 +     /* verifier: AUTH_NULL, length 0 */\
1717                         1 +     /* status */ \
1718                         1 +     /* MIN tag is length with zero, only length */ \
1719                         3 +     /* opcount, opcode, opstatus*/ \
1720                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1721                                 /* seqid, slotID, slotID, slotID, status */ \
1722                         5 ) * sizeof(__be32))
1723
1724 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1725 {
1726         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1727                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1728 }
1729
1730 __be32
1731 nfsd4_create_session(struct svc_rqst *rqstp,
1732                      struct nfsd4_compound_state *cstate,
1733                      struct nfsd4_create_session *cr_ses)
1734 {
1735         struct sockaddr *sa = svc_addr(rqstp);
1736         struct nfs4_client *conf, *unconf;
1737         struct nfsd4_session *new;
1738         struct nfsd4_clid_slot *cs_slot = NULL;
1739         bool confirm_me = false;
1740         __be32 status = 0;
1741
1742         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1743                 return nfserr_inval;
1744
1745         nfs4_lock_state();
1746         unconf = find_unconfirmed_client(&cr_ses->clientid);
1747         conf = find_confirmed_client(&cr_ses->clientid);
1748
1749         if (conf) {
1750                 cs_slot = &conf->cl_cs_slot;
1751                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1752                 if (status == nfserr_replay_cache) {
1753                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1754                         goto out;
1755                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1756                         status = nfserr_seq_misordered;
1757                         goto out;
1758                 }
1759         } else if (unconf) {
1760                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1761                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1762                         status = nfserr_clid_inuse;
1763                         goto out;
1764                 }
1765                 cs_slot = &unconf->cl_cs_slot;
1766                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1767                 if (status) {
1768                         /* an unconfirmed replay returns misordered */
1769                         status = nfserr_seq_misordered;
1770                         goto out;
1771                 }
1772                 confirm_me = true;
1773                 conf = unconf;
1774         } else {
1775                 status = nfserr_stale_clientid;
1776                 goto out;
1777         }
1778
1779         /*
1780          * XXX: we should probably set this at creation time, and check
1781          * for consistent minorversion use throughout:
1782          */
1783         conf->cl_minorversion = 1;
1784         /*
1785          * We do not support RDMA or persistent sessions
1786          */
1787         cr_ses->flags &= ~SESSION4_PERSIST;
1788         cr_ses->flags &= ~SESSION4_RDMA;
1789
1790         status = nfserr_toosmall;
1791         if (check_forechannel_attrs(cr_ses->fore_channel))
1792                 goto out;
1793
1794         status = nfserr_jukebox;
1795         new = alloc_init_session(rqstp, conf, cr_ses);
1796         if (!new)
1797                 goto out;
1798         status = nfs_ok;
1799         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1800                NFS4_MAX_SESSIONID_LEN);
1801         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1802                 sizeof(struct nfsd4_channel_attrs));
1803         cs_slot->sl_seqid++;
1804         cr_ses->seqid = cs_slot->sl_seqid;
1805
1806         /* cache solo and embedded create sessions under the state lock */
1807         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1808         if (confirm_me)
1809                 move_to_confirmed(conf);
1810 out:
1811         nfs4_unlock_state();
1812         dprintk("%s returns %d\n", __func__, ntohl(status));
1813         return status;
1814 }
1815
1816 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1817 {
1818         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1819         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1820
1821         return argp->opcnt == resp->opcnt;
1822 }
1823
1824 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1825 {
1826         switch (*dir) {
1827         case NFS4_CDFC4_FORE:
1828         case NFS4_CDFC4_BACK:
1829                 return nfs_ok;
1830         case NFS4_CDFC4_FORE_OR_BOTH:
1831         case NFS4_CDFC4_BACK_OR_BOTH:
1832                 *dir = NFS4_CDFC4_BOTH;
1833                 return nfs_ok;
1834         };
1835         return nfserr_inval;
1836 }
1837
1838 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1839                      struct nfsd4_compound_state *cstate,
1840                      struct nfsd4_bind_conn_to_session *bcts)
1841 {
1842         __be32 status;
1843
1844         if (!nfsd4_last_compound_op(rqstp))
1845                 return nfserr_not_only_op;
1846         spin_lock(&client_lock);
1847         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1848         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1849          * client_lock iself: */
1850         if (cstate->session) {
1851                 nfsd4_get_session(cstate->session);
1852                 atomic_inc(&cstate->session->se_client->cl_refcount);
1853         }
1854         spin_unlock(&client_lock);
1855         if (!cstate->session)
1856                 return nfserr_badsession;
1857
1858         status = nfsd4_map_bcts_dir(&bcts->dir);
1859         if (!status)
1860                 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1861         return status;
1862 }
1863
1864 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1865 {
1866         if (!session)
1867                 return 0;
1868         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1869 }
1870
1871 __be32
1872 nfsd4_destroy_session(struct svc_rqst *r,
1873                       struct nfsd4_compound_state *cstate,
1874                       struct nfsd4_destroy_session *sessionid)
1875 {
1876         struct nfsd4_session *ses;
1877         __be32 status = nfserr_badsession;
1878
1879         /* Notes:
1880          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1881          * - Should we return nfserr_back_chan_busy if waiting for
1882          *   callbacks on to-be-destroyed session?
1883          * - Do we need to clear any callback info from previous session?
1884          */
1885
1886         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1887                 if (!nfsd4_last_compound_op(r))
1888                         return nfserr_not_only_op;
1889         }
1890         dump_sessionid(__func__, &sessionid->sessionid);
1891         spin_lock(&client_lock);
1892         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1893         if (!ses) {
1894                 spin_unlock(&client_lock);
1895                 goto out;
1896         }
1897
1898         unhash_session(ses);
1899         spin_unlock(&client_lock);
1900
1901         nfs4_lock_state();
1902         nfsd4_probe_callback_sync(ses->se_client);
1903         nfs4_unlock_state();
1904
1905         spin_lock(&client_lock);
1906         nfsd4_del_conns(ses);
1907         nfsd4_put_session_locked(ses);
1908         spin_unlock(&client_lock);
1909         status = nfs_ok;
1910 out:
1911         dprintk("%s returns %d\n", __func__, ntohl(status));
1912         return status;
1913 }
1914
1915 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1916 {
1917         struct nfsd4_conn *c;
1918
1919         list_for_each_entry(c, &s->se_conns, cn_persession) {
1920                 if (c->cn_xprt == xpt) {
1921                         return c;
1922                 }
1923         }
1924         return NULL;
1925 }
1926
1927 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1928 {
1929         struct nfs4_client *clp = ses->se_client;
1930         struct nfsd4_conn *c;
1931         int ret;
1932
1933         spin_lock(&clp->cl_lock);
1934         c = __nfsd4_find_conn(new->cn_xprt, ses);
1935         if (c) {
1936                 spin_unlock(&clp->cl_lock);
1937                 free_conn(new);
1938                 return;
1939         }
1940         __nfsd4_hash_conn(new, ses);
1941         spin_unlock(&clp->cl_lock);
1942         ret = nfsd4_register_conn(new);
1943         if (ret)
1944                 /* oops; xprt is already down: */
1945                 nfsd4_conn_lost(&new->cn_xpt_user);
1946         return;
1947 }
1948
1949 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1950 {
1951         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1952
1953         return args->opcnt > session->se_fchannel.maxops;
1954 }
1955
1956 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1957                                   struct nfsd4_session *session)
1958 {
1959         struct xdr_buf *xb = &rqstp->rq_arg;
1960
1961         return xb->len > session->se_fchannel.maxreq_sz;
1962 }
1963
1964 __be32
1965 nfsd4_sequence(struct svc_rqst *rqstp,
1966                struct nfsd4_compound_state *cstate,
1967                struct nfsd4_sequence *seq)
1968 {
1969         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1970         struct nfsd4_session *session;
1971         struct nfsd4_slot *slot;
1972         struct nfsd4_conn *conn;
1973         __be32 status;
1974
1975         if (resp->opcnt != 1)
1976                 return nfserr_sequence_pos;
1977
1978         /*
1979          * Will be either used or freed by nfsd4_sequence_check_conn
1980          * below.
1981          */
1982         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1983         if (!conn)
1984                 return nfserr_jukebox;
1985
1986         spin_lock(&client_lock);
1987         status = nfserr_badsession;
1988         session = find_in_sessionid_hashtbl(&seq->sessionid);
1989         if (!session)
1990                 goto out;
1991
1992         status = nfserr_too_many_ops;
1993         if (nfsd4_session_too_many_ops(rqstp, session))
1994                 goto out;
1995
1996         status = nfserr_req_too_big;
1997         if (nfsd4_request_too_big(rqstp, session))
1998                 goto out;
1999
2000         status = nfserr_badslot;
2001         if (seq->slotid >= session->se_fchannel.maxreqs)
2002                 goto out;
2003
2004         slot = session->se_slots[seq->slotid];
2005         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2006
2007         /* We do not negotiate the number of slots yet, so set the
2008          * maxslots to the session maxreqs which is used to encode
2009          * sr_highest_slotid and the sr_target_slot id to maxslots */
2010         seq->maxslots = session->se_fchannel.maxreqs;
2011
2012         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2013                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2014         if (status == nfserr_replay_cache) {
2015                 status = nfserr_seq_misordered;
2016                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2017                         goto out;
2018                 cstate->slot = slot;
2019                 cstate->session = session;
2020                 /* Return the cached reply status and set cstate->status
2021                  * for nfsd4_proc_compound processing */
2022                 status = nfsd4_replay_cache_entry(resp, seq);
2023                 cstate->status = nfserr_replay_cache;
2024                 goto out;
2025         }
2026         if (status)
2027                 goto out;
2028
2029         nfsd4_sequence_check_conn(conn, session);
2030         conn = NULL;
2031
2032         /* Success! bump slot seqid */
2033         slot->sl_seqid = seq->seqid;
2034         slot->sl_flags |= NFSD4_SLOT_INUSE;
2035         if (seq->cachethis)
2036                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2037         else
2038                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2039
2040         cstate->slot = slot;
2041         cstate->session = session;
2042
2043 out:
2044         /* Hold a session reference until done processing the compound. */
2045         if (cstate->session) {
2046                 struct nfs4_client *clp = session->se_client;
2047
2048                 nfsd4_get_session(cstate->session);
2049                 atomic_inc(&clp->cl_refcount);
2050                 switch (clp->cl_cb_state) {
2051                 case NFSD4_CB_DOWN:
2052                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2053                         break;
2054                 case NFSD4_CB_FAULT:
2055                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2056                         break;
2057                 default:
2058                         seq->status_flags = 0;
2059                 }
2060         }
2061         kfree(conn);
2062         spin_unlock(&client_lock);
2063         dprintk("%s: return %d\n", __func__, ntohl(status));
2064         return status;
2065 }
2066
2067 static inline bool has_resources(struct nfs4_client *clp)
2068 {
2069         return !list_empty(&clp->cl_openowners)
2070                 || !list_empty(&clp->cl_delegations)
2071                 || !list_empty(&clp->cl_sessions);
2072 }
2073
2074 __be32
2075 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2076 {
2077         struct nfs4_client *conf, *unconf, *clp;
2078         __be32 status = 0;
2079
2080         nfs4_lock_state();
2081         unconf = find_unconfirmed_client(&dc->clientid);
2082         conf = find_confirmed_client(&dc->clientid);
2083
2084         if (conf) {
2085                 clp = conf;
2086
2087                 if (!is_client_expired(conf) && has_resources(conf)) {
2088                         status = nfserr_clientid_busy;
2089                         goto out;
2090                 }
2091
2092                 /* rfc5661 18.50.3 */
2093                 if (cstate->session && conf == cstate->session->se_client) {
2094                         status = nfserr_clientid_busy;
2095                         goto out;
2096                 }
2097         } else if (unconf)
2098                 clp = unconf;
2099         else {
2100                 status = nfserr_stale_clientid;
2101                 goto out;
2102         }
2103
2104         expire_client(clp);
2105 out:
2106         nfs4_unlock_state();
2107         dprintk("%s return %d\n", __func__, ntohl(status));
2108         return status;
2109 }
2110
2111 __be32
2112 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2113 {
2114         __be32 status = 0;
2115
2116         if (rc->rca_one_fs) {
2117                 if (!cstate->current_fh.fh_dentry)
2118                         return nfserr_nofilehandle;
2119                 /*
2120                  * We don't take advantage of the rca_one_fs case.
2121                  * That's OK, it's optional, we can safely ignore it.
2122                  */
2123                  return nfs_ok;
2124         }
2125
2126         nfs4_lock_state();
2127         status = nfserr_complete_already;
2128         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2129                              &cstate->session->se_client->cl_flags))
2130                 goto out;
2131
2132         status = nfserr_stale_clientid;
2133         if (is_client_expired(cstate->session->se_client))
2134                 /*
2135                  * The following error isn't really legal.
2136                  * But we only get here if the client just explicitly
2137                  * destroyed the client.  Surely it no longer cares what
2138                  * error it gets back on an operation for the dead
2139                  * client.
2140                  */
2141                 goto out;
2142
2143         status = nfs_ok;
2144         nfsd4_client_record_create(cstate->session->se_client);
2145 out:
2146         nfs4_unlock_state();
2147         return status;
2148 }
2149
2150 __be32
2151 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2152                   struct nfsd4_setclientid *setclid)
2153 {
2154         struct xdr_netobj       clname = setclid->se_name;
2155         nfs4_verifier           clverifier = setclid->se_verf;
2156         unsigned int            strhashval;
2157         struct nfs4_client      *conf, *unconf, *new;
2158         __be32                  status;
2159         char                    dname[HEXDIR_LEN];
2160
2161         status = nfs4_make_rec_clidname(dname, &clname);
2162         if (status)
2163                 return status;
2164
2165         strhashval = clientstr_hashval(dname);
2166
2167         /* Cases below refer to rfc 3530 section 14.2.33: */
2168         nfs4_lock_state();
2169         conf = find_confirmed_client_by_str(dname, strhashval);
2170         if (conf) {
2171                 /* case 0: */
2172                 status = nfserr_clid_inuse;
2173                 if (clp_used_exchangeid(conf))
2174                         goto out;
2175                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2176                         char addr_str[INET6_ADDRSTRLEN];
2177                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2178                                  sizeof(addr_str));
2179                         dprintk("NFSD: setclientid: string in use by client "
2180                                 "at %s\n", addr_str);
2181                         goto out;
2182                 }
2183         }
2184         unconf = find_unconfirmed_client_by_str(dname, strhashval);
2185         if (unconf)
2186                 expire_client(unconf);
2187         status = nfserr_jukebox;
2188         new = create_client(clname, dname, rqstp, &clverifier);
2189         if (new == NULL)
2190                 goto out;
2191         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2192                 /* case 1: probable callback update */
2193                 copy_clid(new, conf);
2194         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2195                 gen_clid(new);
2196         /*
2197          * XXX: we should probably set this at creation time, and check
2198          * for consistent minorversion use throughout:
2199          */
2200         new->cl_minorversion = 0;
2201         gen_callback(new, setclid, rqstp);
2202         add_to_unconfirmed(new, strhashval);
2203         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2204         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2205         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2206         status = nfs_ok;
2207 out:
2208         nfs4_unlock_state();
2209         return status;
2210 }
2211
2212
2213 __be32
2214 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2215                          struct nfsd4_compound_state *cstate,
2216                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2217 {
2218         struct nfs4_client *conf, *unconf;
2219         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2220         clientid_t * clid = &setclientid_confirm->sc_clientid;
2221         __be32 status;
2222
2223         if (STALE_CLIENTID(clid))
2224                 return nfserr_stale_clientid;
2225         nfs4_lock_state();
2226
2227         conf = find_confirmed_client(clid);
2228         unconf = find_unconfirmed_client(clid);
2229         /*
2230          * We try hard to give out unique clientid's, so if we get an
2231          * attempt to confirm the same clientid with a different cred,
2232          * there's a bug somewhere.  Let's charitably assume it's our
2233          * bug.
2234          */
2235         status = nfserr_serverfault;
2236         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2237                 goto out;
2238         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2239                 goto out;
2240         /* cases below refer to rfc 3530 section 14.2.34: */
2241         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2242                 if (conf && !unconf) /* case 2: probable retransmit */
2243                         status = nfs_ok;
2244                 else /* case 4: client hasn't noticed we rebooted yet? */
2245                         status = nfserr_stale_clientid;
2246                 goto out;
2247         }
2248         status = nfs_ok;
2249         if (conf) { /* case 1: callback update */
2250                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2251                 nfsd4_probe_callback(conf);
2252                 expire_client(unconf);
2253         } else { /* case 3: normal case; new or rebooted client */
2254                 unsigned int hash = clientstr_hashval(unconf->cl_recdir);
2255
2256                 conf = find_confirmed_client_by_str(unconf->cl_recdir, hash);
2257                 if (conf) {
2258                         nfsd4_client_record_remove(conf);
2259                         expire_client(conf);
2260                 }
2261                 move_to_confirmed(unconf);
2262                 nfsd4_probe_callback(unconf);
2263         }
2264 out:
2265         nfs4_unlock_state();
2266         return status;
2267 }
2268
2269 static struct nfs4_file *nfsd4_alloc_file(void)
2270 {
2271         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2272 }
2273
2274 /* OPEN Share state helper functions */
2275 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2276 {
2277         unsigned int hashval = file_hashval(ino);
2278
2279         atomic_set(&fp->fi_ref, 1);
2280         INIT_LIST_HEAD(&fp->fi_hash);
2281         INIT_LIST_HEAD(&fp->fi_stateids);
2282         INIT_LIST_HEAD(&fp->fi_delegations);
2283         fp->fi_inode = igrab(ino);
2284         fp->fi_had_conflict = false;
2285         fp->fi_lease = NULL;
2286         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2287         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2288         spin_lock(&recall_lock);
2289         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2290         spin_unlock(&recall_lock);
2291 }
2292
2293 static void
2294 nfsd4_free_slab(struct kmem_cache **slab)
2295 {
2296         if (*slab == NULL)
2297                 return;
2298         kmem_cache_destroy(*slab);
2299         *slab = NULL;
2300 }
2301
2302 void
2303 nfsd4_free_slabs(void)
2304 {
2305         nfsd4_free_slab(&openowner_slab);
2306         nfsd4_free_slab(&lockowner_slab);
2307         nfsd4_free_slab(&file_slab);
2308         nfsd4_free_slab(&stateid_slab);
2309         nfsd4_free_slab(&deleg_slab);
2310 }
2311
2312 int
2313 nfsd4_init_slabs(void)
2314 {
2315         openowner_slab = kmem_cache_create("nfsd4_openowners",
2316                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2317         if (openowner_slab == NULL)
2318                 goto out_nomem;
2319         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2320                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2321         if (lockowner_slab == NULL)
2322                 goto out_nomem;
2323         file_slab = kmem_cache_create("nfsd4_files",
2324                         sizeof(struct nfs4_file), 0, 0, NULL);
2325         if (file_slab == NULL)
2326                 goto out_nomem;
2327         stateid_slab = kmem_cache_create("nfsd4_stateids",
2328                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2329         if (stateid_slab == NULL)
2330                 goto out_nomem;
2331         deleg_slab = kmem_cache_create("nfsd4_delegations",
2332                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2333         if (deleg_slab == NULL)
2334                 goto out_nomem;
2335         return 0;
2336 out_nomem:
2337         nfsd4_free_slabs();
2338         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2339         return -ENOMEM;
2340 }
2341
2342 void nfs4_free_openowner(struct nfs4_openowner *oo)
2343 {
2344         kfree(oo->oo_owner.so_owner.data);
2345         kmem_cache_free(openowner_slab, oo);
2346 }
2347
2348 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2349 {
2350         kfree(lo->lo_owner.so_owner.data);
2351         kmem_cache_free(lockowner_slab, lo);
2352 }
2353
2354 static void init_nfs4_replay(struct nfs4_replay *rp)
2355 {
2356         rp->rp_status = nfserr_serverfault;
2357         rp->rp_buflen = 0;
2358         rp->rp_buf = rp->rp_ibuf;
2359 }
2360
2361 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2362 {
2363         struct nfs4_stateowner *sop;
2364
2365         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2366         if (!sop)
2367                 return NULL;
2368
2369         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2370         if (!sop->so_owner.data) {
2371                 kmem_cache_free(slab, sop);
2372                 return NULL;
2373         }
2374         sop->so_owner.len = owner->len;
2375
2376         INIT_LIST_HEAD(&sop->so_stateids);
2377         sop->so_client = clp;
2378         init_nfs4_replay(&sop->so_replay);
2379         return sop;
2380 }
2381
2382 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2383 {
2384         list_add(&oo->oo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
2385         list_add(&oo->oo_perclient, &clp->cl_openowners);
2386 }
2387
2388 static struct nfs4_openowner *
2389 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2390         struct nfs4_openowner *oo;
2391
2392         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2393         if (!oo)
2394                 return NULL;
2395         oo->oo_owner.so_is_open_owner = 1;
2396         oo->oo_owner.so_seqid = open->op_seqid;
2397         oo->oo_flags = NFS4_OO_NEW;
2398         oo->oo_time = 0;
2399         oo->oo_last_closed_stid = NULL;
2400         INIT_LIST_HEAD(&oo->oo_close_lru);
2401         hash_openowner(oo, clp, strhashval);
2402         return oo;
2403 }
2404
2405 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2406         struct nfs4_openowner *oo = open->op_openowner;
2407         struct nfs4_client *clp = oo->oo_owner.so_client;
2408
2409         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2410         INIT_LIST_HEAD(&stp->st_lockowners);
2411         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2412         list_add(&stp->st_perfile, &fp->fi_stateids);
2413         stp->st_stateowner = &oo->oo_owner;
2414         get_nfs4_file(fp);
2415         stp->st_file = fp;
2416         stp->st_access_bmap = 0;
2417         stp->st_deny_bmap = 0;
2418         set_access(open->op_share_access, stp);
2419         set_deny(open->op_share_deny, stp);
2420         stp->st_openstp = NULL;
2421 }
2422
2423 static void
2424 move_to_close_lru(struct nfs4_openowner *oo)
2425 {
2426         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2427
2428         list_move_tail(&oo->oo_close_lru, &close_lru);
2429         oo->oo_time = get_seconds();
2430 }
2431
2432 static int
2433 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2434                                                         clientid_t *clid)
2435 {
2436         return (sop->so_owner.len == owner->len) &&
2437                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2438                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2439 }
2440
2441 static struct nfs4_openowner *
2442 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2443 {
2444         struct nfs4_stateowner *so;
2445         struct nfs4_openowner *oo;
2446
2447         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2448                 if (!so->so_is_open_owner)
2449                         continue;
2450                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2451                         oo = openowner(so);
2452                         renew_client(oo->oo_owner.so_client);
2453                         return oo;
2454                 }
2455         }
2456         return NULL;
2457 }
2458
2459 /* search file_hashtbl[] for file */
2460 static struct nfs4_file *
2461 find_file(struct inode *ino)
2462 {
2463         unsigned int hashval = file_hashval(ino);
2464         struct nfs4_file *fp;
2465
2466         spin_lock(&recall_lock);
2467         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2468                 if (fp->fi_inode == ino) {
2469                         get_nfs4_file(fp);
2470                         spin_unlock(&recall_lock);
2471                         return fp;
2472                 }
2473         }
2474         spin_unlock(&recall_lock);
2475         return NULL;
2476 }
2477
2478 /*
2479  * Called to check deny when READ with all zero stateid or
2480  * WRITE with all zero or all one stateid
2481  */
2482 static __be32
2483 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2484 {
2485         struct inode *ino = current_fh->fh_dentry->d_inode;
2486         struct nfs4_file *fp;
2487         struct nfs4_ol_stateid *stp;
2488         __be32 ret;
2489
2490         dprintk("NFSD: nfs4_share_conflict\n");
2491
2492         fp = find_file(ino);
2493         if (!fp)
2494                 return nfs_ok;
2495         ret = nfserr_locked;
2496         /* Search for conflicting share reservations */
2497         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2498                 if (test_deny(deny_type, stp) ||
2499                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2500                         goto out;
2501         }
2502         ret = nfs_ok;
2503 out:
2504         put_nfs4_file(fp);
2505         return ret;
2506 }
2507
2508 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2509 {
2510         /* We're assuming the state code never drops its reference
2511          * without first removing the lease.  Since we're in this lease
2512          * callback (and since the lease code is serialized by the kernel
2513          * lock) we know the server hasn't removed the lease yet, we know
2514          * it's safe to take a reference: */
2515         atomic_inc(&dp->dl_count);
2516
2517         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2518
2519         /* only place dl_time is set. protected by lock_flocks*/
2520         dp->dl_time = get_seconds();
2521
2522         nfsd4_cb_recall(dp);
2523 }
2524
2525 /* Called from break_lease() with lock_flocks() held. */
2526 static void nfsd_break_deleg_cb(struct file_lock *fl)
2527 {
2528         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2529         struct nfs4_delegation *dp;
2530
2531         BUG_ON(!fp);
2532         /* We assume break_lease is only called once per lease: */
2533         BUG_ON(fp->fi_had_conflict);
2534         /*
2535          * We don't want the locks code to timeout the lease for us;
2536          * we'll remove it ourself if a delegation isn't returned
2537          * in time:
2538          */
2539         fl->fl_break_time = 0;
2540
2541         spin_lock(&recall_lock);
2542         fp->fi_had_conflict = true;
2543         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2544                 nfsd_break_one_deleg(dp);
2545         spin_unlock(&recall_lock);
2546 }
2547
2548 static
2549 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2550 {
2551         if (arg & F_UNLCK)
2552                 return lease_modify(onlist, arg);
2553         else
2554                 return -EAGAIN;
2555 }
2556
2557 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2558         .lm_break = nfsd_break_deleg_cb,
2559         .lm_change = nfsd_change_deleg_cb,
2560 };
2561
2562 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2563 {
2564         if (nfsd4_has_session(cstate))
2565                 return nfs_ok;
2566         if (seqid == so->so_seqid - 1)
2567                 return nfserr_replay_me;
2568         if (seqid == so->so_seqid)
2569                 return nfs_ok;
2570         return nfserr_bad_seqid;
2571 }
2572
2573 __be32
2574 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2575                     struct nfsd4_open *open)
2576 {
2577         clientid_t *clientid = &open->op_clientid;
2578         struct nfs4_client *clp = NULL;
2579         unsigned int strhashval;
2580         struct nfs4_openowner *oo = NULL;
2581         __be32 status;
2582
2583         if (STALE_CLIENTID(&open->op_clientid))
2584                 return nfserr_stale_clientid;
2585         /*
2586          * In case we need it later, after we've already created the
2587          * file and don't want to risk a further failure:
2588          */
2589         open->op_file = nfsd4_alloc_file();
2590         if (open->op_file == NULL)
2591                 return nfserr_jukebox;
2592
2593         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2594         oo = find_openstateowner_str(strhashval, open);
2595         open->op_openowner = oo;
2596         if (!oo) {
2597                 clp = find_confirmed_client(clientid);
2598                 if (clp == NULL)
2599                         return nfserr_expired;
2600                 goto new_owner;
2601         }
2602         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2603                 /* Replace unconfirmed owners without checking for replay. */
2604                 clp = oo->oo_owner.so_client;
2605                 release_openowner(oo);
2606                 open->op_openowner = NULL;
2607                 goto new_owner;
2608         }
2609         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2610         if (status)
2611                 return status;
2612         clp = oo->oo_owner.so_client;
2613         goto alloc_stateid;
2614 new_owner:
2615         oo = alloc_init_open_stateowner(strhashval, clp, open);
2616         if (oo == NULL)
2617                 return nfserr_jukebox;
2618         open->op_openowner = oo;
2619 alloc_stateid:
2620         open->op_stp = nfs4_alloc_stateid(clp);
2621         if (!open->op_stp)
2622                 return nfserr_jukebox;
2623         return nfs_ok;
2624 }
2625
2626 static inline __be32
2627 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2628 {
2629         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2630                 return nfserr_openmode;
2631         else
2632                 return nfs_ok;
2633 }
2634
2635 static int share_access_to_flags(u32 share_access)
2636 {
2637         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2638 }
2639
2640 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2641 {
2642         struct nfs4_stid *ret;
2643
2644         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2645         if (!ret)
2646                 return NULL;
2647         return delegstateid(ret);
2648 }
2649
2650 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2651 {
2652         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2653                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2654 }
2655
2656 static __be32
2657 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2658                 struct nfs4_delegation **dp)
2659 {
2660         int flags;
2661         __be32 status = nfserr_bad_stateid;
2662
2663         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2664         if (*dp == NULL)
2665                 goto out;
2666         flags = share_access_to_flags(open->op_share_access);
2667         status = nfs4_check_delegmode(*dp, flags);
2668         if (status)
2669                 *dp = NULL;
2670 out:
2671         if (!nfsd4_is_deleg_cur(open))
2672                 return nfs_ok;
2673         if (status)
2674                 return status;
2675         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2676         return nfs_ok;
2677 }
2678
2679 static __be32
2680 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2681 {
2682         struct nfs4_ol_stateid *local;
2683         struct nfs4_openowner *oo = open->op_openowner;
2684
2685         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2686                 /* ignore lock owners */
2687                 if (local->st_stateowner->so_is_open_owner == 0)
2688                         continue;
2689                 /* remember if we have seen this open owner */
2690                 if (local->st_stateowner == &oo->oo_owner)
2691                         *stpp = local;
2692                 /* check for conflicting share reservations */
2693                 if (!test_share(local, open))
2694                         return nfserr_share_denied;
2695         }
2696         return nfs_ok;
2697 }
2698
2699 static void nfs4_free_stateid(struct nfs4_ol_stateid *s)
2700 {
2701         kmem_cache_free(stateid_slab, s);
2702 }
2703
2704 static inline int nfs4_access_to_access(u32 nfs4_access)
2705 {
2706         int flags = 0;
2707
2708         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2709                 flags |= NFSD_MAY_READ;
2710         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2711                 flags |= NFSD_MAY_WRITE;
2712         return flags;
2713 }
2714
2715 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2716                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2717 {
2718         __be32 status;
2719         int oflag = nfs4_access_to_omode(open->op_share_access);
2720         int access = nfs4_access_to_access(open->op_share_access);
2721
2722         if (!fp->fi_fds[oflag]) {
2723                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2724                         &fp->fi_fds[oflag]);
2725                 if (status)
2726                         return status;
2727         }
2728         nfs4_file_get_access(fp, oflag);
2729
2730         return nfs_ok;
2731 }
2732
2733 static inline __be32
2734 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2735                 struct nfsd4_open *open)
2736 {
2737         struct iattr iattr = {
2738                 .ia_valid = ATTR_SIZE,
2739                 .ia_size = 0,
2740         };
2741         if (!open->op_truncate)
2742                 return 0;
2743         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2744                 return nfserr_inval;
2745         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2746 }
2747
2748 static __be32
2749 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)
2750 {
2751         u32 op_share_access = open->op_share_access;
2752         bool new_access;
2753         __be32 status;
2754
2755         new_access = !test_access(op_share_access, stp);
2756         if (new_access) {
2757                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2758                 if (status)
2759                         return status;
2760         }
2761         status = nfsd4_truncate(rqstp, cur_fh, open);
2762         if (status) {
2763                 if (new_access) {
2764                         int oflag = nfs4_access_to_omode(op_share_access);
2765                         nfs4_file_put_access(fp, oflag);
2766                 }
2767                 return status;
2768         }
2769         /* remember the open */
2770         set_access(op_share_access, stp);
2771         set_deny(open->op_share_deny, stp);
2772
2773         return nfs_ok;
2774 }
2775
2776
2777 static void
2778 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2779 {
2780         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2781 }
2782
2783 /* Should we give out recallable state?: */
2784 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2785 {
2786         if (clp->cl_cb_state == NFSD4_CB_UP)
2787                 return true;
2788         /*
2789          * In the sessions case, since we don't have to establish a
2790          * separate connection for callbacks, we assume it's OK
2791          * until we hear otherwise:
2792          */
2793         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2794 }
2795
2796 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2797 {
2798         struct file_lock *fl;
2799
2800         fl = locks_alloc_lock();
2801         if (!fl)
2802                 return NULL;
2803         locks_init_lock(fl);
2804         fl->fl_lmops = &nfsd_lease_mng_ops;
2805         fl->fl_flags = FL_LEASE;
2806         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2807         fl->fl_end = OFFSET_MAX;
2808         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2809         fl->fl_pid = current->tgid;
2810         return fl;
2811 }
2812
2813 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2814 {
2815         struct nfs4_file *fp = dp->dl_file;
2816         struct file_lock *fl;
2817         int status;
2818
2819         fl = nfs4_alloc_init_lease(dp, flag);
2820         if (!fl)
2821                 return -ENOMEM;
2822         fl->fl_file = find_readable_file(fp);
2823         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2824         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2825         if (status) {
2826                 list_del_init(&dp->dl_perclnt);
2827                 locks_free_lock(fl);
2828                 return -ENOMEM;
2829         }
2830         fp->fi_lease = fl;
2831         fp->fi_deleg_file = fl->fl_file;
2832         get_file(fp->fi_deleg_file);
2833         atomic_set(&fp->fi_delegees, 1);
2834         list_add(&dp->dl_perfile, &fp->fi_delegations);
2835         return 0;
2836 }
2837
2838 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2839 {
2840         struct nfs4_file *fp = dp->dl_file;
2841
2842         if (!fp->fi_lease)
2843                 return nfs4_setlease(dp, flag);
2844         spin_lock(&recall_lock);
2845         if (fp->fi_had_conflict) {
2846                 spin_unlock(&recall_lock);
2847                 return -EAGAIN;
2848         }
2849         atomic_inc(&fp->fi_delegees);
2850         list_add(&dp->dl_perfile, &fp->fi_delegations);
2851         spin_unlock(&recall_lock);
2852         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2853         return 0;
2854 }
2855
2856 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2857 {
2858         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2859         if (status == -EAGAIN)
2860                 open->op_why_no_deleg = WND4_CONTENTION;
2861         else {
2862                 open->op_why_no_deleg = WND4_RESOURCE;
2863                 switch (open->op_deleg_want) {
2864                 case NFS4_SHARE_WANT_READ_DELEG:
2865                 case NFS4_SHARE_WANT_WRITE_DELEG:
2866                 case NFS4_SHARE_WANT_ANY_DELEG:
2867                         break;
2868                 case NFS4_SHARE_WANT_CANCEL:
2869                         open->op_why_no_deleg = WND4_CANCELLED;
2870                         break;
2871                 case NFS4_SHARE_WANT_NO_DELEG:
2872                         BUG();  /* not supposed to get here */
2873                 }
2874         }
2875 }
2876
2877 /*
2878  * Attempt to hand out a delegation.
2879  */
2880 static void
2881 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2882 {
2883         struct nfs4_delegation *dp;
2884         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2885         int cb_up;
2886         int status = 0, flag = 0;
2887
2888         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2889         flag = NFS4_OPEN_DELEGATE_NONE;
2890         open->op_recall = 0;
2891         switch (open->op_claim_type) {
2892                 case NFS4_OPEN_CLAIM_PREVIOUS:
2893                         if (!cb_up)
2894                                 open->op_recall = 1;
2895                         flag = open->op_delegate_type;
2896                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2897                                 goto out;
2898                         break;
2899                 case NFS4_OPEN_CLAIM_NULL:
2900                         /* Let's not give out any delegations till everyone's
2901                          * had the chance to reclaim theirs.... */
2902                         if (locks_in_grace())
2903                                 goto out;
2904                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2905                                 goto out;
2906                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2907                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2908                         else
2909                                 flag = NFS4_OPEN_DELEGATE_READ;
2910                         break;
2911                 default:
2912                         goto out;
2913         }
2914
2915         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2916         if (dp == NULL)
2917                 goto out_no_deleg;
2918         status = nfs4_set_delegation(dp, flag);
2919         if (status)
2920                 goto out_free;
2921
2922         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2923
2924         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2925                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2926 out:
2927         open->op_delegate_type = flag;
2928         if (flag == NFS4_OPEN_DELEGATE_NONE) {
2929                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2930                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2931                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2932
2933                 /* 4.1 client asking for a delegation? */
2934                 if (open->op_deleg_want)
2935                         nfsd4_open_deleg_none_ext(open, status);
2936         }
2937         return;
2938 out_free:
2939         nfs4_put_delegation(dp);
2940 out_no_deleg:
2941         flag = NFS4_OPEN_DELEGATE_NONE;
2942         goto out;
2943 }
2944
2945 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
2946                                         struct nfs4_delegation *dp)
2947 {
2948         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
2949             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2950                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2951                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
2952         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
2953                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2954                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2955                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
2956         }
2957         /* Otherwise the client must be confused wanting a delegation
2958          * it already has, therefore we don't return
2959          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
2960          */
2961 }
2962
2963 /*
2964  * called with nfs4_lock_state() held.
2965  */
2966 __be32
2967 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2968 {
2969         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2970         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
2971         struct nfs4_file *fp = NULL;
2972         struct inode *ino = current_fh->fh_dentry->d_inode;
2973         struct nfs4_ol_stateid *stp = NULL;
2974         struct nfs4_delegation *dp = NULL;
2975         __be32 status;
2976
2977         /*
2978          * Lookup file; if found, lookup stateid and check open request,
2979          * and check for delegations in the process of being recalled.
2980          * If not found, create the nfs4_file struct
2981          */
2982         fp = find_file(ino);
2983         if (fp) {
2984                 if ((status = nfs4_check_open(fp, open, &stp)))
2985                         goto out;
2986                 status = nfs4_check_deleg(cl, fp, open, &dp);
2987                 if (status)
2988                         goto out;
2989         } else {
2990                 status = nfserr_bad_stateid;
2991                 if (nfsd4_is_deleg_cur(open))
2992                         goto out;
2993                 status = nfserr_jukebox;
2994                 fp = open->op_file;
2995                 open->op_file = NULL;
2996                 nfsd4_init_file(fp, ino);
2997         }
2998
2999         /*
3000          * OPEN the file, or upgrade an existing OPEN.
3001          * If truncate fails, the OPEN fails.
3002          */
3003         if (stp) {
3004                 /* Stateid was found, this is an OPEN upgrade */
3005                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3006                 if (status)
3007                         goto out;
3008         } else {
3009                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3010                 if (status)
3011                         goto out;
3012                 stp = open->op_stp;
3013                 open->op_stp = NULL;
3014                 init_open_stateid(stp, fp, open);
3015                 status = nfsd4_truncate(rqstp, current_fh, open);
3016                 if (status) {
3017                         release_open_stateid(stp);
3018                         goto out;
3019                 }
3020         }
3021         update_stateid(&stp->st_stid.sc_stateid);
3022         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3023
3024         if (nfsd4_has_session(&resp->cstate)) {
3025                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3026
3027                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3028                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3029                         open->op_why_no_deleg = WND4_NOT_WANTED;
3030                         goto nodeleg;
3031                 }
3032         }
3033
3034         /*
3035         * Attempt to hand out a delegation. No error return, because the
3036         * OPEN succeeds even if we fail.
3037         */
3038         nfs4_open_delegation(current_fh, open, stp);
3039 nodeleg:
3040         status = nfs_ok;
3041
3042         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3043                 STATEID_VAL(&stp->st_stid.sc_stateid));
3044 out:
3045         /* 4.1 client trying to upgrade/downgrade delegation? */
3046         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3047             open->op_deleg_want)
3048                 nfsd4_deleg_xgrade_none_ext(open, dp);
3049
3050         if (fp)
3051                 put_nfs4_file(fp);
3052         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3053                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3054         /*
3055         * To finish the open response, we just need to set the rflags.
3056         */
3057         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3058         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3059             !nfsd4_has_session(&resp->cstate))
3060                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3061
3062         return status;
3063 }
3064
3065 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3066 {
3067         if (open->op_openowner) {
3068                 struct nfs4_openowner *oo = open->op_openowner;
3069
3070                 if (!list_empty(&oo->oo_owner.so_stateids))
3071                         list_del_init(&oo->oo_close_lru);
3072                 if (oo->oo_flags & NFS4_OO_NEW) {
3073                         if (status) {
3074                                 release_openowner(oo);
3075                                 open->op_openowner = NULL;
3076                         } else
3077                                 oo->oo_flags &= ~NFS4_OO_NEW;
3078                 }
3079         }
3080         if (open->op_file)
3081                 nfsd4_free_file(open->op_file);
3082         if (open->op_stp)
3083                 nfs4_free_stateid(open->op_stp);
3084 }
3085
3086 __be32
3087 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3088             clientid_t *clid)
3089 {
3090         struct nfs4_client *clp;
3091         __be32 status;
3092
3093         nfs4_lock_state();
3094         dprintk("process_renew(%08x/%08x): starting\n",
3095                         clid->cl_boot, clid->cl_id);
3096         status = nfserr_stale_clientid;
3097         if (STALE_CLIENTID(clid))
3098                 goto out;
3099         clp = find_confirmed_client(clid);
3100         status = nfserr_expired;
3101         if (clp == NULL) {
3102                 /* We assume the client took too long to RENEW. */
3103                 dprintk("nfsd4_renew: clientid not found!\n");
3104                 goto out;
3105         }
3106         status = nfserr_cb_path_down;
3107         if (!list_empty(&clp->cl_delegations)
3108                         && clp->cl_cb_state != NFSD4_CB_UP)
3109                 goto out;
3110         status = nfs_ok;
3111 out:
3112         nfs4_unlock_state();
3113         return status;
3114 }
3115
3116 static struct lock_manager nfsd4_manager = {
3117 };
3118
3119 static bool grace_ended;
3120
3121 static void
3122 nfsd4_end_grace(void)
3123 {
3124         /* do nothing if grace period already ended */
3125         if (grace_ended)
3126                 return;
3127
3128         dprintk("NFSD: end of grace period\n");
3129         grace_ended = true;
3130         nfsd4_record_grace_done(&init_net, boot_time);
3131         locks_end_grace(&nfsd4_manager);
3132         /*
3133          * Now that every NFSv4 client has had the chance to recover and
3134          * to see the (possibly new, possibly shorter) lease time, we
3135          * can safely set the next grace time to the current lease time:
3136          */
3137         nfsd4_grace = nfsd4_lease;
3138 }
3139
3140 static time_t
3141 nfs4_laundromat(void)
3142 {
3143         struct nfs4_client *clp;
3144         struct nfs4_openowner *oo;
3145         struct nfs4_delegation *dp;
3146         struct list_head *pos, *next, reaplist;
3147         time_t cutoff = get_seconds() - nfsd4_lease;
3148         time_t t, clientid_val = nfsd4_lease;
3149         time_t u, test_val = nfsd4_lease;
3150
3151         nfs4_lock_state();
3152
3153         dprintk("NFSD: laundromat service - starting\n");
3154         nfsd4_end_grace();
3155         INIT_LIST_HEAD(&reaplist);
3156         spin_lock(&client_lock);
3157         list_for_each_safe(pos, next, &client_lru) {
3158                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3159                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3160                         t = clp->cl_time - cutoff;
3161                         if (clientid_val > t)
3162                                 clientid_val = t;
3163                         break;
3164                 }
3165                 if (atomic_read(&clp->cl_refcount)) {
3166                         dprintk("NFSD: client in use (clientid %08x)\n",
3167                                 clp->cl_clientid.cl_id);
3168                         continue;
3169                 }
3170                 unhash_client_locked(clp);
3171                 list_add(&clp->cl_lru, &reaplist);
3172         }
3173         spin_unlock(&client_lock);
3174         list_for_each_safe(pos, next, &reaplist) {
3175                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3176                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3177                         clp->cl_clientid.cl_id);
3178                 nfsd4_client_record_remove(clp);
3179                 expire_client(clp);
3180         }
3181         spin_lock(&recall_lock);
3182         list_for_each_safe(pos, next, &del_recall_lru) {
3183                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3184                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3185                         u = dp->dl_time - cutoff;
3186                         if (test_val > u)
3187                                 test_val = u;
3188                         break;
3189                 }
3190                 list_move(&dp->dl_recall_lru, &reaplist);
3191         }
3192         spin_unlock(&recall_lock);
3193         list_for_each_safe(pos, next, &reaplist) {
3194                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3195                 unhash_delegation(dp);
3196         }
3197         test_val = nfsd4_lease;
3198         list_for_each_safe(pos, next, &close_lru) {
3199                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3200                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3201                         u = oo->oo_time - cutoff;
3202                         if (test_val > u)
3203                                 test_val = u;
3204                         break;
3205                 }
3206                 release_openowner(oo);
3207         }
3208         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3209                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3210         nfs4_unlock_state();
3211         return clientid_val;
3212 }
3213
3214 static struct workqueue_struct *laundry_wq;
3215 static void laundromat_main(struct work_struct *);
3216 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3217
3218 static void
3219 laundromat_main(struct work_struct *not_used)
3220 {
3221         time_t t;
3222
3223         t = nfs4_laundromat();
3224         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3225         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3226 }
3227
3228 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3229 {
3230         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3231                 return nfserr_bad_stateid;
3232         return nfs_ok;
3233 }
3234
3235 static int
3236 STALE_STATEID(stateid_t *stateid)
3237 {
3238         if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3239                 return 0;
3240         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3241                 STATEID_VAL(stateid));
3242         return 1;
3243 }
3244
3245 static inline int
3246 access_permit_read(struct nfs4_ol_stateid *stp)
3247 {
3248         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3249                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3250                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3251 }
3252
3253 static inline int
3254 access_permit_write(struct nfs4_ol_stateid *stp)
3255 {
3256         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3257                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3258 }
3259
3260 static
3261 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3262 {
3263         __be32 status = nfserr_openmode;
3264
3265         /* For lock stateid's, we test the parent open, not the lock: */
3266         if (stp->st_openstp)
3267                 stp = stp->st_openstp;
3268         if ((flags & WR_STATE) && !access_permit_write(stp))
3269                 goto out;
3270         if ((flags & RD_STATE) && !access_permit_read(stp))
3271                 goto out;
3272         status = nfs_ok;
3273 out:
3274         return status;
3275 }
3276
3277 static inline __be32
3278 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3279 {
3280         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3281                 return nfs_ok;
3282         else if (locks_in_grace()) {
3283                 /* Answer in remaining cases depends on existence of
3284                  * conflicting state; so we must wait out the grace period. */
3285                 return nfserr_grace;
3286         } else if (flags & WR_STATE)
3287                 return nfs4_share_conflict(current_fh,
3288                                 NFS4_SHARE_DENY_WRITE);
3289         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3290                 return nfs4_share_conflict(current_fh,
3291                                 NFS4_SHARE_DENY_READ);
3292 }
3293
3294 /*
3295  * Allow READ/WRITE during grace period on recovered state only for files
3296  * that are not able to provide mandatory locking.
3297  */
3298 static inline int
3299 grace_disallows_io(struct inode *inode)
3300 {
3301         return locks_in_grace() && mandatory_lock(inode);
3302 }
3303
3304 /* Returns true iff a is later than b: */
3305 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3306 {
3307         return (s32)a->si_generation - (s32)b->si_generation > 0;
3308 }
3309
3310 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3311 {
3312         /*
3313          * When sessions are used the stateid generation number is ignored
3314          * when it is zero.
3315          */
3316         if (has_session && in->si_generation == 0)
3317                 return nfs_ok;
3318
3319         if (in->si_generation == ref->si_generation)
3320                 return nfs_ok;
3321
3322         /* If the client sends us a stateid from the future, it's buggy: */
3323         if (stateid_generation_after(in, ref))
3324                 return nfserr_bad_stateid;
3325         /*
3326          * However, we could see a stateid from the past, even from a
3327          * non-buggy client.  For example, if the client sends a lock
3328          * while some IO is outstanding, the lock may bump si_generation
3329          * while the IO is still in flight.  The client could avoid that
3330          * situation by waiting for responses on all the IO requests,
3331          * but better performance may result in retrying IO that
3332          * receives an old_stateid error if requests are rarely
3333          * reordered in flight:
3334          */
3335         return nfserr_old_stateid;
3336 }
3337
3338 __be32 nfs4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3339 {
3340         struct nfs4_stid *s;
3341         struct nfs4_ol_stateid *ols;
3342         __be32 status;
3343
3344         if (STALE_STATEID(stateid))
3345                 return nfserr_stale_stateid;
3346
3347         s = find_stateid(cl, stateid);
3348         if (!s)
3349                  return nfserr_stale_stateid;
3350         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3351         if (status)
3352                 return status;
3353         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3354                 return nfs_ok;
3355         ols = openlockstateid(s);
3356         if (ols->st_stateowner->so_is_open_owner
3357             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3358                 return nfserr_bad_stateid;
3359         return nfs_ok;
3360 }
3361
3362 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3363 {
3364         struct nfs4_client *cl;
3365
3366         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3367                 return nfserr_bad_stateid;
3368         if (STALE_STATEID(stateid))
3369                 return nfserr_stale_stateid;
3370         cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3371         if (!cl)
3372                 return nfserr_expired;
3373         *s = find_stateid_by_type(cl, stateid, typemask);
3374         if (!*s)
3375                 return nfserr_bad_stateid;
3376         return nfs_ok;
3377
3378 }
3379
3380 /*
3381 * Checks for stateid operations
3382 */
3383 __be32
3384 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3385                            stateid_t *stateid, int flags, struct file **filpp)
3386 {
3387         struct nfs4_stid *s;
3388         struct nfs4_ol_stateid *stp = NULL;
3389         struct nfs4_delegation *dp = NULL;
3390         struct svc_fh *current_fh = &cstate->current_fh;
3391         struct inode *ino = current_fh->fh_dentry->d_inode;
3392         __be32 status;
3393
3394         if (filpp)
3395                 *filpp = NULL;
3396
3397         if (grace_disallows_io(ino))
3398                 return nfserr_grace;
3399
3400         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3401                 return check_special_stateids(current_fh, stateid, flags);
3402
3403         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s);
3404         if (status)
3405                 return status;
3406         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3407         if (status)
3408                 goto out;
3409         switch (s->sc_type) {
3410         case NFS4_DELEG_STID:
3411                 dp = delegstateid(s);
3412                 status = nfs4_check_delegmode(dp, flags);
3413                 if (status)
3414                         goto out;
3415                 if (filpp) {
3416                         *filpp = dp->dl_file->fi_deleg_file;
3417                         BUG_ON(!*filpp);
3418                 }
3419                 break;
3420         case NFS4_OPEN_STID:
3421         case NFS4_LOCK_STID:
3422                 stp = openlockstateid(s);
3423                 status = nfs4_check_fh(current_fh, stp);
3424                 if (status)
3425                         goto out;
3426                 if (stp->st_stateowner->so_is_open_owner
3427                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3428                         goto out;
3429                 status = nfs4_check_openmode(stp, flags);
3430                 if (status)
3431                         goto out;
3432                 if (filpp) {
3433                         if (flags & RD_STATE)
3434                                 *filpp = find_readable_file(stp->st_file);
3435                         else
3436                                 *filpp = find_writeable_file(stp->st_file);
3437                 }
3438                 break;
3439         default:
3440                 return nfserr_bad_stateid;
3441         }
3442         status = nfs_ok;
3443 out:
3444         return status;
3445 }
3446
3447 static __be32
3448 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3449 {
3450         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3451                 return nfserr_locks_held;
3452         release_lock_stateid(stp);
3453         return nfs_ok;
3454 }
3455
3456 /*
3457  * Test if the stateid is valid
3458  */
3459 __be32
3460 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3461                    struct nfsd4_test_stateid *test_stateid)
3462 {
3463         struct nfsd4_test_stateid_id *stateid;
3464         struct nfs4_client *cl = cstate->session->se_client;
3465
3466         nfs4_lock_state();
3467         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3468                 stateid->ts_id_status = nfs4_validate_stateid(cl, &stateid->ts_id_stateid);
3469         nfs4_unlock_state();
3470
3471         return nfs_ok;
3472 }
3473
3474 __be32
3475 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3476                    struct nfsd4_free_stateid *free_stateid)
3477 {
3478         stateid_t *stateid = &free_stateid->fr_stateid;
3479         struct nfs4_stid *s;
3480         struct nfs4_client *cl = cstate->session->se_client;
3481         __be32 ret = nfserr_bad_stateid;
3482
3483         nfs4_lock_state();
3484         s = find_stateid(cl, stateid);
3485         if (!s)
3486                 goto out;
3487         switch (s->sc_type) {
3488         case NFS4_DELEG_STID:
3489                 ret = nfserr_locks_held;
3490                 goto out;
3491         case NFS4_OPEN_STID:
3492         case NFS4_LOCK_STID:
3493                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3494                 if (ret)
3495                         goto out;
3496                 if (s->sc_type == NFS4_LOCK_STID)
3497                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3498                 else
3499                         ret = nfserr_locks_held;
3500                 break;
3501         default:
3502                 ret = nfserr_bad_stateid;
3503         }
3504 out:
3505         nfs4_unlock_state();
3506         return ret;
3507 }
3508
3509 static inline int
3510 setlkflg (int type)
3511 {
3512         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3513                 RD_STATE : WR_STATE;
3514 }
3515
3516 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3517 {
3518         struct svc_fh *current_fh = &cstate->current_fh;
3519         struct nfs4_stateowner *sop = stp->st_stateowner;
3520         __be32 status;
3521
3522         status = nfsd4_check_seqid(cstate, sop, seqid);
3523         if (status)
3524                 return status;
3525         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3526                 /*
3527                  * "Closed" stateid's exist *only* to return
3528                  * nfserr_replay_me from the previous step.
3529                  */
3530                 return nfserr_bad_stateid;
3531         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3532         if (status)
3533                 return status;
3534         return nfs4_check_fh(current_fh, stp);
3535 }
3536
3537 /*
3538  * Checks for sequence id mutating operations.
3539  */
3540 static __be32
3541 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3542                          stateid_t *stateid, char typemask,
3543                          struct nfs4_ol_stateid **stpp)
3544 {
3545         __be32 status;
3546         struct nfs4_stid *s;
3547
3548         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3549                 seqid, STATEID_VAL(stateid));
3550
3551         *stpp = NULL;
3552         status = nfsd4_lookup_stateid(stateid, typemask, &s);
3553         if (status)
3554                 return status;
3555         *stpp = openlockstateid(s);
3556         cstate->replay_owner = (*stpp)->st_stateowner;
3557
3558         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3559 }
3560
3561 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3562 {
3563         __be32 status;
3564         struct nfs4_openowner *oo;
3565
3566         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3567                                                 NFS4_OPEN_STID, stpp);
3568         if (status)
3569                 return status;
3570         oo = openowner((*stpp)->st_stateowner);
3571         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3572                 return nfserr_bad_stateid;
3573         return nfs_ok;
3574 }
3575
3576 __be32
3577 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3578                    struct nfsd4_open_confirm *oc)
3579 {
3580         __be32 status;
3581         struct nfs4_openowner *oo;
3582         struct nfs4_ol_stateid *stp;
3583
3584         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3585                         (int)cstate->current_fh.fh_dentry->d_name.len,
3586                         cstate->current_fh.fh_dentry->d_name.name);
3587
3588         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3589         if (status)
3590                 return status;
3591
3592         nfs4_lock_state();
3593
3594         status = nfs4_preprocess_seqid_op(cstate,
3595                                         oc->oc_seqid, &oc->oc_req_stateid,
3596                                         NFS4_OPEN_STID, &stp);
3597         if (status)
3598                 goto out;
3599         oo = openowner(stp->st_stateowner);
3600         status = nfserr_bad_stateid;
3601         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3602                 goto out;
3603         oo->oo_flags |= NFS4_OO_CONFIRMED;
3604         update_stateid(&stp->st_stid.sc_stateid);
3605         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3606         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3607                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3608
3609         nfsd4_client_record_create(oo->oo_owner.so_client);
3610         status = nfs_ok;
3611 out:
3612         if (!cstate->replay_owner)
3613                 nfs4_unlock_state();
3614         return status;
3615 }
3616
3617 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3618 {
3619         if (!test_access(access, stp))
3620                 return;
3621         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3622         clear_access(access, stp);
3623 }
3624
3625 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3626 {
3627         switch (to_access) {
3628         case NFS4_SHARE_ACCESS_READ:
3629                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3630                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3631                 break;
3632         case NFS4_SHARE_ACCESS_WRITE:
3633                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3634                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3635                 break;
3636         case NFS4_SHARE_ACCESS_BOTH:
3637                 break;
3638         default:
3639                 BUG();
3640         }
3641 }
3642
3643 static void
3644 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3645 {
3646         int i;
3647         for (i = 0; i < 4; i++) {
3648                 if ((i & deny) != i)
3649                         clear_deny(i, stp);
3650         }
3651 }
3652
3653 __be32
3654 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3655                      struct nfsd4_compound_state *cstate,
3656                      struct nfsd4_open_downgrade *od)
3657 {
3658         __be32 status;
3659         struct nfs4_ol_stateid *stp;
3660
3661         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3662                         (int)cstate->current_fh.fh_dentry->d_name.len,
3663                         cstate->current_fh.fh_dentry->d_name.name);
3664
3665         /* We don't yet support WANT bits: */
3666         if (od->od_deleg_want)
3667                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3668                         od->od_deleg_want);
3669
3670         nfs4_lock_state();
3671         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3672                                         &od->od_stateid, &stp);
3673         if (status)
3674                 goto out;
3675         status = nfserr_inval;
3676         if (!test_access(od->od_share_access, stp)) {
3677                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3678                         stp->st_access_bmap, od->od_share_access);
3679                 goto out;
3680         }
3681         if (!test_deny(od->od_share_deny, stp)) {
3682                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3683                         stp->st_deny_bmap, od->od_share_deny);
3684                 goto out;
3685         }
3686         nfs4_stateid_downgrade(stp, od->od_share_access);
3687
3688         reset_union_bmap_deny(od->od_share_deny, stp);
3689
3690         update_stateid(&stp->st_stid.sc_stateid);
3691         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3692         status = nfs_ok;
3693 out:
3694         if (!cstate->replay_owner)
3695                 nfs4_unlock_state();
3696         return status;
3697 }
3698
3699 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3700 {
3701         struct nfs4_openowner *oo;
3702         struct nfs4_ol_stateid *s;
3703
3704         if (!so->so_is_open_owner)
3705                 return;
3706         oo = openowner(so);
3707         s = oo->oo_last_closed_stid;
3708         if (!s)
3709                 return;
3710         if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3711                 /* Release the last_closed_stid on the next seqid bump: */
3712                 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3713                 return;
3714         }
3715         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3716         release_last_closed_stateid(oo);
3717 }
3718
3719 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3720 {
3721         unhash_open_stateid(s);
3722         s->st_stid.sc_type = NFS4_CLOSED_STID;
3723 }
3724
3725 /*
3726  * nfs4_unlock_state() called after encode
3727  */
3728 __be32
3729 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3730             struct nfsd4_close *close)
3731 {
3732         __be32 status;
3733         struct nfs4_openowner *oo;
3734         struct nfs4_ol_stateid *stp;
3735
3736         dprintk("NFSD: nfsd4_close on file %.*s\n",
3737                         (int)cstate->current_fh.fh_dentry->d_name.len,
3738                         cstate->current_fh.fh_dentry->d_name.name);
3739
3740         nfs4_lock_state();
3741         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3742                                         &close->cl_stateid,
3743                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3744                                         &stp);
3745         if (status)
3746                 goto out;
3747         oo = openowner(stp->st_stateowner);
3748         status = nfs_ok;
3749         update_stateid(&stp->st_stid.sc_stateid);
3750         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3751
3752         nfsd4_close_open_stateid(stp);
3753         oo->oo_last_closed_stid = stp;
3754
3755         /* place unused nfs4_stateowners on so_close_lru list to be
3756          * released by the laundromat service after the lease period
3757          * to enable us to handle CLOSE replay
3758          */
3759         if (list_empty(&oo->oo_owner.so_stateids))
3760                 move_to_close_lru(oo);
3761 out:
3762         if (!cstate->replay_owner)
3763                 nfs4_unlock_state();
3764         return status;
3765 }
3766
3767 __be32
3768 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3769                   struct nfsd4_delegreturn *dr)
3770 {
3771         struct nfs4_delegation *dp;
3772         stateid_t *stateid = &dr->dr_stateid;
3773         struct nfs4_stid *s;
3774         struct inode *inode;
3775         __be32 status;
3776
3777         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3778                 return status;
3779         inode = cstate->current_fh.fh_dentry->d_inode;
3780
3781         nfs4_lock_state();
3782         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s);
3783         if (status)
3784                 goto out;
3785         dp = delegstateid(s);
3786         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3787         if (status)
3788                 goto out;
3789
3790         unhash_delegation(dp);
3791 out:
3792         nfs4_unlock_state();
3793
3794         return status;
3795 }
3796
3797
3798 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3799
3800 #define LOCKOWNER_INO_HASH_BITS 8
3801 #define LOCKOWNER_INO_HASH_SIZE (1 << LOCKOWNER_INO_HASH_BITS)
3802 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3803
3804 static inline u64
3805 end_offset(u64 start, u64 len)
3806 {
3807         u64 end;
3808
3809         end = start + len;
3810         return end >= start ? end: NFS4_MAX_UINT64;
3811 }
3812
3813 /* last octet in a range */
3814 static inline u64
3815 last_byte_offset(u64 start, u64 len)
3816 {
3817         u64 end;
3818
3819         BUG_ON(!len);
3820         end = start + len;
3821         return end > start ? end - 1: NFS4_MAX_UINT64;
3822 }
3823
3824 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3825 {
3826         return (file_hashval(inode) + cl_id
3827                         + opaque_hashval(ownername->data, ownername->len))
3828                 & LOCKOWNER_INO_HASH_MASK;
3829 }
3830
3831 static struct list_head lockowner_ino_hashtbl[LOCKOWNER_INO_HASH_SIZE];
3832
3833 /*
3834  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3835  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3836  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3837  * locking, this prevents us from being completely protocol-compliant.  The
3838  * real solution to this problem is to start using unsigned file offsets in
3839  * the VFS, but this is a very deep change!
3840  */
3841 static inline void
3842 nfs4_transform_lock_offset(struct file_lock *lock)
3843 {
3844         if (lock->fl_start < 0)
3845                 lock->fl_start = OFFSET_MAX;
3846         if (lock->fl_end < 0)
3847                 lock->fl_end = OFFSET_MAX;
3848 }
3849
3850 /* Hack!: For now, we're defining this just so we can use a pointer to it
3851  * as a unique cookie to identify our (NFSv4's) posix locks. */
3852 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3853 };
3854
3855 static inline void
3856 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3857 {
3858         struct nfs4_lockowner *lo;
3859
3860         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3861                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3862                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3863                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3864                 if (!deny->ld_owner.data)
3865                         /* We just don't care that much */
3866                         goto nevermind;
3867                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3868                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3869         } else {
3870 nevermind:
3871                 deny->ld_owner.len = 0;
3872                 deny->ld_owner.data = NULL;
3873                 deny->ld_clientid.cl_boot = 0;
3874                 deny->ld_clientid.cl_id = 0;
3875         }
3876         deny->ld_start = fl->fl_start;
3877         deny->ld_length = NFS4_MAX_UINT64;
3878         if (fl->fl_end != NFS4_MAX_UINT64)
3879                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3880         deny->ld_type = NFS4_READ_LT;
3881         if (fl->fl_type != F_RDLCK)
3882                 deny->ld_type = NFS4_WRITE_LT;
3883 }
3884
3885 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3886 {
3887         struct nfs4_ol_stateid *lst;
3888
3889         if (!same_owner_str(&lo->lo_owner, owner, clid))
3890                 return false;
3891         lst = list_first_entry(&lo->lo_owner.so_stateids,
3892                                struct nfs4_ol_stateid, st_perstateowner);
3893         return lst->st_file->fi_inode == inode;
3894 }
3895
3896 static struct nfs4_lockowner *
3897 find_lockowner_str(struct inode *inode, clientid_t *clid,
3898                 struct xdr_netobj *owner)
3899 {
3900         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3901         struct nfs4_lockowner *lo;
3902
3903         list_for_each_entry(lo, &lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3904                 if (same_lockowner_ino(lo, inode, clid, owner))
3905                         return lo;
3906         }
3907         return NULL;
3908 }
3909
3910 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3911 {
3912         struct inode *inode = open_stp->st_file->fi_inode;
3913         unsigned int inohash = lockowner_ino_hashval(inode,
3914                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
3915
3916         list_add(&lo->lo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
3917         list_add(&lo->lo_owner_ino_hash, &lockowner_ino_hashtbl[inohash]);
3918         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3919 }
3920
3921 /*
3922  * Alloc a lock owner structure.
3923  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3924  * occurred.
3925  *
3926  * strhashval = ownerstr_hashval
3927  */
3928
3929 static struct nfs4_lockowner *
3930 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
3931         struct nfs4_lockowner *lo;
3932
3933         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
3934         if (!lo)
3935                 return NULL;
3936         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
3937         lo->lo_owner.so_is_open_owner = 0;
3938         /* It is the openowner seqid that will be incremented in encode in the
3939          * case of new lockowners; so increment the lock seqid manually: */
3940         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
3941         hash_lockowner(lo, strhashval, clp, open_stp);
3942         return lo;
3943 }
3944
3945 static struct nfs4_ol_stateid *
3946 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
3947 {
3948         struct nfs4_ol_stateid *stp;
3949         struct nfs4_client *clp = lo->lo_owner.so_client;
3950
3951         stp = nfs4_alloc_stateid(clp);
3952         if (stp == NULL)
3953                 return NULL;
3954         init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
3955         list_add(&stp->st_perfile, &fp->fi_stateids);
3956         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
3957         stp->st_stateowner = &lo->lo_owner;
3958         get_nfs4_file(fp);
3959         stp->st_file = fp;
3960         stp->st_access_bmap = 0;
3961         stp->st_deny_bmap = open_stp->st_deny_bmap;
3962         stp->st_openstp = open_stp;
3963         return stp;
3964 }
3965
3966 static int
3967 check_lock_length(u64 offset, u64 length)
3968 {
3969         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3970              LOFF_OVERFLOW(offset, length)));
3971 }
3972
3973 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
3974 {
3975         struct nfs4_file *fp = lock_stp->st_file;
3976         int oflag = nfs4_access_to_omode(access);
3977
3978         if (test_access(access, lock_stp))
3979                 return;
3980         nfs4_file_get_access(fp, oflag);
3981         set_access(access, lock_stp);
3982 }
3983
3984 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)
3985 {
3986         struct nfs4_file *fi = ost->st_file;
3987         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
3988         struct nfs4_client *cl = oo->oo_owner.so_client;
3989         struct nfs4_lockowner *lo;
3990         unsigned int strhashval;
3991
3992         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid, &lock->v.new.owner);
3993         if (lo) {
3994                 if (!cstate->minorversion)
3995                         return nfserr_bad_seqid;
3996                 /* XXX: a lockowner always has exactly one stateid: */
3997                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
3998                                 struct nfs4_ol_stateid, st_perstateowner);
3999                 return nfs_ok;
4000         }
4001         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4002                         &lock->v.new.owner);
4003         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4004         if (lo == NULL)
4005                 return nfserr_jukebox;
4006         *lst = alloc_init_lock_stateid(lo, fi, ost);
4007         if (*lst == NULL) {
4008                 release_lockowner(lo);
4009                 return nfserr_jukebox;
4010         }
4011         *new = true;
4012         return nfs_ok;
4013 }
4014
4015 /*
4016  *  LOCK operation
4017  */
4018 __be32
4019 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4020            struct nfsd4_lock *lock)
4021 {
4022         struct nfs4_openowner *open_sop = NULL;
4023         struct nfs4_lockowner *lock_sop = NULL;
4024         struct nfs4_ol_stateid *lock_stp;
4025         struct file *filp = NULL;
4026         struct file_lock file_lock;
4027         struct file_lock conflock;
4028         __be32 status = 0;
4029         bool new_state = false;
4030         int lkflg;
4031         int err;
4032
4033         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4034                 (long long) lock->lk_offset,
4035                 (long long) lock->lk_length);
4036
4037         if (check_lock_length(lock->lk_offset, lock->lk_length))
4038                  return nfserr_inval;
4039
4040         if ((status = fh_verify(rqstp, &cstate->current_fh,
4041                                 S_IFREG, NFSD_MAY_LOCK))) {
4042                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4043                 return status;
4044         }
4045
4046         nfs4_lock_state();
4047
4048         if (lock->lk_is_new) {
4049                 /*
4050                  * Client indicates that this is a new lockowner.
4051                  * Use open owner and open stateid to create lock owner and
4052                  * lock stateid.
4053                  */
4054                 struct nfs4_ol_stateid *open_stp = NULL;
4055
4056                 if (nfsd4_has_session(cstate))
4057                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4058                         memcpy(&lock->v.new.clientid,
4059                                 &cstate->session->se_client->cl_clientid,
4060                                 sizeof(clientid_t));
4061
4062                 status = nfserr_stale_clientid;
4063                 if (STALE_CLIENTID(&lock->lk_new_clientid))
4064                         goto out;
4065
4066                 /* validate and update open stateid and open seqid */
4067                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4068                                         lock->lk_new_open_seqid,
4069                                         &lock->lk_new_open_stateid,
4070                                         &open_stp);
4071                 if (status)
4072                         goto out;
4073                 open_sop = openowner(open_stp->st_stateowner);
4074                 status = nfserr_bad_stateid;
4075                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4076                                                 &lock->v.new.clientid))
4077                         goto out;
4078                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4079                                                         &lock_stp, &new_state);
4080                 if (status)
4081                         goto out;
4082         } else {
4083                 /* lock (lock owner + lock stateid) already exists */
4084                 status = nfs4_preprocess_seqid_op(cstate,
4085                                        lock->lk_old_lock_seqid,
4086                                        &lock->lk_old_lock_stateid,
4087                                        NFS4_LOCK_STID, &lock_stp);
4088                 if (status)
4089                         goto out;
4090         }
4091         lock_sop = lockowner(lock_stp->st_stateowner);
4092
4093         lkflg = setlkflg(lock->lk_type);
4094         status = nfs4_check_openmode(lock_stp, lkflg);
4095         if (status)
4096                 goto out;
4097
4098         status = nfserr_grace;
4099         if (locks_in_grace() && !lock->lk_reclaim)
4100                 goto out;
4101         status = nfserr_no_grace;
4102         if (!locks_in_grace() && lock->lk_reclaim)
4103                 goto out;
4104
4105         locks_init_lock(&file_lock);
4106         switch (lock->lk_type) {
4107                 case NFS4_READ_LT:
4108                 case NFS4_READW_LT:
4109                         filp = find_readable_file(lock_stp->st_file);
4110                         if (filp)
4111                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4112                         file_lock.fl_type = F_RDLCK;
4113                         break;
4114                 case NFS4_WRITE_LT:
4115                 case NFS4_WRITEW_LT:
4116                         filp = find_writeable_file(lock_stp->st_file);
4117                         if (filp)
4118                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4119                         file_lock.fl_type = F_WRLCK;
4120                         break;
4121                 default:
4122                         status = nfserr_inval;
4123                 goto out;
4124         }
4125         if (!filp) {
4126                 status = nfserr_openmode;
4127                 goto out;
4128         }
4129         file_lock.fl_owner = (fl_owner_t)lock_sop;
4130         file_lock.fl_pid = current->tgid;
4131         file_lock.fl_file = filp;
4132         file_lock.fl_flags = FL_POSIX;
4133         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4134
4135         file_lock.fl_start = lock->lk_offset;
4136         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4137         nfs4_transform_lock_offset(&file_lock);
4138
4139         /*
4140         * Try to lock the file in the VFS.
4141         * Note: locks.c uses the BKL to protect the inode's lock list.
4142         */
4143
4144         err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4145         switch (-err) {
4146         case 0: /* success! */
4147                 update_stateid(&lock_stp->st_stid.sc_stateid);
4148                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4149                                 sizeof(stateid_t));
4150                 status = 0;
4151                 break;
4152         case (EAGAIN):          /* conflock holds conflicting lock */
4153                 status = nfserr_denied;
4154                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4155                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4156                 break;
4157         case (EDEADLK):
4158                 status = nfserr_deadlock;
4159                 break;
4160         default:
4161                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4162                 status = nfserrno(err);
4163                 break;
4164         }
4165 out:
4166         if (status && new_state)
4167                 release_lockowner(lock_sop);
4168         if (!cstate->replay_owner)
4169                 nfs4_unlock_state();
4170         return status;
4171 }
4172
4173 /*
4174  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4175  * so we do a temporary open here just to get an open file to pass to
4176  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4177  * inode operation.)
4178  */
4179 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4180 {
4181         struct file *file;
4182         int err;
4183
4184         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4185         if (err)
4186                 return err;
4187         err = vfs_test_lock(file, lock);
4188         nfsd_close(file);
4189         return err;
4190 }
4191
4192 /*
4193  * LOCKT operation
4194  */
4195 __be32
4196 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4197             struct nfsd4_lockt *lockt)
4198 {
4199         struct inode *inode;
4200         struct file_lock file_lock;
4201         struct nfs4_lockowner *lo;
4202         int error;
4203         __be32 status;
4204
4205         if (locks_in_grace())
4206                 return nfserr_grace;
4207
4208         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4209                  return nfserr_inval;
4210
4211         nfs4_lock_state();
4212
4213         status = nfserr_stale_clientid;
4214         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4215                 goto out;
4216
4217         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4218                 goto out;
4219
4220         inode = cstate->current_fh.fh_dentry->d_inode;
4221         locks_init_lock(&file_lock);
4222         switch (lockt->lt_type) {
4223                 case NFS4_READ_LT:
4224                 case NFS4_READW_LT:
4225                         file_lock.fl_type = F_RDLCK;
4226                 break;
4227                 case NFS4_WRITE_LT:
4228                 case NFS4_WRITEW_LT:
4229                         file_lock.fl_type = F_WRLCK;
4230                 break;
4231                 default:
4232                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4233                         status = nfserr_inval;
4234                 goto out;
4235         }
4236
4237         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4238         if (lo)
4239                 file_lock.fl_owner = (fl_owner_t)lo;
4240         file_lock.fl_pid = current->tgid;
4241         file_lock.fl_flags = FL_POSIX;
4242
4243         file_lock.fl_start = lockt->lt_offset;
4244         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4245
4246         nfs4_transform_lock_offset(&file_lock);
4247
4248         status = nfs_ok;
4249         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4250         if (error) {
4251                 status = nfserrno(error);
4252                 goto out;
4253         }
4254         if (file_lock.fl_type != F_UNLCK) {
4255                 status = nfserr_denied;
4256                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4257         }
4258 out:
4259         nfs4_unlock_state();
4260         return status;
4261 }
4262
4263 __be32
4264 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4265             struct nfsd4_locku *locku)
4266 {
4267         struct nfs4_ol_stateid *stp;
4268         struct file *filp = NULL;
4269         struct file_lock file_lock;
4270         __be32 status;
4271         int err;
4272
4273         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4274                 (long long) locku->lu_offset,
4275                 (long long) locku->lu_length);
4276
4277         if (check_lock_length(locku->lu_offset, locku->lu_length))
4278                  return nfserr_inval;
4279
4280         nfs4_lock_state();
4281
4282         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4283                                         &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4284         if (status)
4285                 goto out;
4286         filp = find_any_file(stp->st_file);
4287         if (!filp) {
4288                 status = nfserr_lock_range;
4289                 goto out;
4290         }
4291         BUG_ON(!filp);
4292         locks_init_lock(&file_lock);
4293         file_lock.fl_type = F_UNLCK;
4294         file_lock.fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4295         file_lock.fl_pid = current->tgid;
4296         file_lock.fl_file = filp;
4297         file_lock.fl_flags = FL_POSIX;
4298         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4299         file_lock.fl_start = locku->lu_offset;
4300
4301         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4302         nfs4_transform_lock_offset(&file_lock);
4303
4304         /*
4305         *  Try to unlock the file in the VFS.
4306         */
4307         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4308         if (err) {
4309                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4310                 goto out_nfserr;
4311         }
4312         /*
4313         * OK, unlock succeeded; the only thing left to do is update the stateid.
4314         */
4315         update_stateid(&stp->st_stid.sc_stateid);
4316         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4317
4318 out:
4319         if (!cstate->replay_owner)
4320                 nfs4_unlock_state();
4321         return status;
4322
4323 out_nfserr:
4324         status = nfserrno(err);
4325         goto out;
4326 }
4327
4328 /*
4329  * returns
4330  *      1: locks held by lockowner
4331  *      0: no locks held by lockowner
4332  */
4333 static int
4334 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4335 {
4336         struct file_lock **flpp;
4337         struct inode *inode = filp->fi_inode;
4338         int status = 0;
4339
4340         lock_flocks();
4341         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4342                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4343                         status = 1;
4344                         goto out;
4345                 }
4346         }
4347 out:
4348         unlock_flocks();
4349         return status;
4350 }
4351
4352 __be32
4353 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4354                         struct nfsd4_compound_state *cstate,
4355                         struct nfsd4_release_lockowner *rlockowner)
4356 {
4357         clientid_t *clid = &rlockowner->rl_clientid;
4358         struct nfs4_stateowner *sop;
4359         struct nfs4_lockowner *lo;
4360         struct nfs4_ol_stateid *stp;
4361         struct xdr_netobj *owner = &rlockowner->rl_owner;
4362         struct list_head matches;
4363         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4364         __be32 status;
4365
4366         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4367                 clid->cl_boot, clid->cl_id);
4368
4369         /* XXX check for lease expiration */
4370
4371         status = nfserr_stale_clientid;
4372         if (STALE_CLIENTID(clid))
4373                 return status;
4374
4375         nfs4_lock_state();
4376
4377         status = nfserr_locks_held;
4378         INIT_LIST_HEAD(&matches);
4379
4380         list_for_each_entry(sop, &ownerstr_hashtbl[hashval], so_strhash) {
4381                 if (sop->so_is_open_owner)
4382                         continue;
4383                 if (!same_owner_str(sop, owner, clid))
4384                         continue;
4385                 list_for_each_entry(stp, &sop->so_stateids,
4386                                 st_perstateowner) {
4387                         lo = lockowner(sop);
4388                         if (check_for_locks(stp->st_file, lo))
4389                                 goto out;
4390                         list_add(&lo->lo_list, &matches);
4391                 }
4392         }
4393         /* Clients probably won't expect us to return with some (but not all)
4394          * of the lockowner state released; so don't release any until all
4395          * have been checked. */
4396         status = nfs_ok;
4397         while (!list_empty(&matches)) {
4398                 lo = list_entry(matches.next, struct nfs4_lockowner,
4399                                                                 lo_list);
4400                 /* unhash_stateowner deletes so_perclient only
4401                  * for openowners. */
4402                 list_del(&lo->lo_list);
4403                 release_lockowner(lo);
4404         }
4405 out:
4406         nfs4_unlock_state();
4407         return status;
4408 }
4409
4410 static inline struct nfs4_client_reclaim *
4411 alloc_reclaim(void)
4412 {
4413         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4414 }
4415
4416 int
4417 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4418 {
4419         unsigned int strhashval = clientstr_hashval(name);
4420         struct nfs4_client *clp;
4421
4422         clp = find_confirmed_client_by_str(name, strhashval);
4423         if (!clp)
4424                 return 0;
4425         return test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
4426 }
4427
4428 /*
4429  * failure => all reset bets are off, nfserr_no_grace...
4430  */
4431 int
4432 nfs4_client_to_reclaim(const char *name)
4433 {
4434         unsigned int strhashval;
4435         struct nfs4_client_reclaim *crp = NULL;
4436
4437         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4438         crp = alloc_reclaim();
4439         if (!crp)
4440                 return 0;
4441         strhashval = clientstr_hashval(name);
4442         INIT_LIST_HEAD(&crp->cr_strhash);
4443         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4444         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4445         reclaim_str_hashtbl_size++;
4446         return 1;
4447 }
4448
4449 void
4450 nfs4_release_reclaim(void)
4451 {
4452         struct nfs4_client_reclaim *crp = NULL;
4453         int i;
4454
4455         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4456                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4457                         crp = list_entry(reclaim_str_hashtbl[i].next,
4458                                         struct nfs4_client_reclaim, cr_strhash);
4459                         list_del(&crp->cr_strhash);
4460                         kfree(crp);
4461                         reclaim_str_hashtbl_size--;
4462                 }
4463         }
4464         BUG_ON(reclaim_str_hashtbl_size);
4465 }
4466
4467 /*
4468  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4469 struct nfs4_client_reclaim *
4470 nfsd4_find_reclaim_client(struct nfs4_client *clp)
4471 {
4472         unsigned int strhashval;
4473         struct nfs4_client_reclaim *crp = NULL;
4474
4475         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4476                             clp->cl_name.len, clp->cl_name.data,
4477                             clp->cl_recdir);
4478
4479         /* find clp->cl_name in reclaim_str_hashtbl */
4480         strhashval = clientstr_hashval(clp->cl_recdir);
4481         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4482                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4483                         return crp;
4484                 }
4485         }
4486         return NULL;
4487 }
4488
4489 /*
4490 * Called from OPEN. Look for clientid in reclaim list.
4491 */
4492 __be32
4493 nfs4_check_open_reclaim(clientid_t *clid)
4494 {
4495         struct nfs4_client *clp;
4496
4497         /* find clientid in conf_id_hashtbl */
4498         clp = find_confirmed_client(clid);
4499         if (clp == NULL)
4500                 return nfserr_reclaim_bad;
4501
4502         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4503 }
4504
4505 #ifdef CONFIG_NFSD_FAULT_INJECTION
4506
4507 void nfsd_forget_clients(u64 num)
4508 {
4509         struct nfs4_client *clp, *next;
4510         int count = 0;
4511
4512         nfs4_lock_state();
4513         list_for_each_entry_safe(clp, next, &client_lru, cl_lru) {
4514                 nfsd4_client_record_remove(clp);
4515                 expire_client(clp);
4516                 if (++count == num)
4517                         break;
4518         }
4519         nfs4_unlock_state();
4520
4521         printk(KERN_INFO "NFSD: Forgot %d clients", count);
4522 }
4523
4524 static void release_lockowner_sop(struct nfs4_stateowner *sop)
4525 {
4526         release_lockowner(lockowner(sop));
4527 }
4528
4529 static void release_openowner_sop(struct nfs4_stateowner *sop)
4530 {
4531         release_openowner(openowner(sop));
4532 }
4533
4534 static int nfsd_release_n_owners(u64 num, bool is_open_owner,
4535                                 void (*release_sop)(struct nfs4_stateowner *))
4536 {
4537         int i, count = 0;
4538         struct nfs4_stateowner *sop, *next;
4539
4540         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4541                 list_for_each_entry_safe(sop, next, &ownerstr_hashtbl[i], so_strhash) {
4542                         if (sop->so_is_open_owner != is_open_owner)
4543                                 continue;
4544                         release_sop(sop);
4545                         if (++count == num)
4546                                 return count;
4547                 }
4548         }
4549         return count;
4550 }
4551
4552 void nfsd_forget_locks(u64 num)
4553 {
4554         int count;
4555
4556         nfs4_lock_state();
4557         count = nfsd_release_n_owners(num, false, release_lockowner_sop);
4558         nfs4_unlock_state();
4559
4560         printk(KERN_INFO "NFSD: Forgot %d locks", count);
4561 }
4562
4563 void nfsd_forget_openowners(u64 num)
4564 {
4565         int count;
4566
4567         nfs4_lock_state();
4568         count = nfsd_release_n_owners(num, true, release_openowner_sop);
4569         nfs4_unlock_state();
4570
4571         printk(KERN_INFO "NFSD: Forgot %d open owners", count);
4572 }
4573
4574 int nfsd_process_n_delegations(u64 num, void (*deleg_func)(struct nfs4_delegation *))
4575 {
4576         int i, count = 0;
4577         struct nfs4_file *fp, *fnext;
4578         struct nfs4_delegation *dp, *dnext;
4579
4580         for (i = 0; i < FILE_HASH_SIZE; i++) {
4581                 list_for_each_entry_safe(fp, fnext, &file_hashtbl[i], fi_hash) {
4582                         list_for_each_entry_safe(dp, dnext, &fp->fi_delegations, dl_perfile) {
4583                                 deleg_func(dp);
4584                                 if (++count == num)
4585                                         return count;
4586                         }
4587                 }
4588         }
4589
4590         return count;
4591 }
4592
4593 void nfsd_forget_delegations(u64 num)
4594 {
4595         unsigned int count;
4596
4597         nfs4_lock_state();
4598         count = nfsd_process_n_delegations(num, unhash_delegation);
4599         nfs4_unlock_state();
4600
4601         printk(KERN_INFO "NFSD: Forgot %d delegations", count);
4602 }
4603
4604 void nfsd_recall_delegations(u64 num)
4605 {
4606         unsigned int count;
4607
4608         nfs4_lock_state();
4609         spin_lock(&recall_lock);
4610         count = nfsd_process_n_delegations(num, nfsd_break_one_deleg);
4611         spin_unlock(&recall_lock);
4612         nfs4_unlock_state();
4613
4614         printk(KERN_INFO "NFSD: Recalled %d delegations", count);
4615 }
4616
4617 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4618
4619 /* initialization to perform at module load time: */
4620
4621 void
4622 nfs4_state_init(void)
4623 {
4624         int i;
4625
4626         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4627                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4628                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4629                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4630                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4631                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4632         }
4633         for (i = 0; i < SESSION_HASH_SIZE; i++)
4634                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4635         for (i = 0; i < FILE_HASH_SIZE; i++) {
4636                 INIT_LIST_HEAD(&file_hashtbl[i]);
4637         }
4638         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4639                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4640         }
4641         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4642                 INIT_LIST_HEAD(&lockowner_ino_hashtbl[i]);
4643         INIT_LIST_HEAD(&close_lru);
4644         INIT_LIST_HEAD(&client_lru);
4645         INIT_LIST_HEAD(&del_recall_lru);
4646         reclaim_str_hashtbl_size = 0;
4647 }
4648
4649 /*
4650  * Since the lifetime of a delegation isn't limited to that of an open, a
4651  * client may quite reasonably hang on to a delegation as long as it has
4652  * the inode cached.  This becomes an obvious problem the first time a
4653  * client's inode cache approaches the size of the server's total memory.
4654  *
4655  * For now we avoid this problem by imposing a hard limit on the number
4656  * of delegations, which varies according to the server's memory size.
4657  */
4658 static void
4659 set_max_delegations(void)
4660 {
4661         /*
4662          * Allow at most 4 delegations per megabyte of RAM.  Quick
4663          * estimates suggest that in the worst case (where every delegation
4664          * is for a different inode), a delegation could take about 1.5K,
4665          * giving a worst case usage of about 6% of memory.
4666          */
4667         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4668 }
4669
4670 /* initialization to perform when the nfsd service is started: */
4671
4672 int
4673 nfs4_state_start(void)
4674 {
4675         int ret;
4676
4677         /*
4678          * FIXME: For now, we hang most of the pernet global stuff off of
4679          * init_net until nfsd is fully containerized. Eventually, we'll
4680          * need to pass a net pointer into this function, take a reference
4681          * to that instead and then do most of the rest of this on a per-net
4682          * basis.
4683          */
4684         get_net(&init_net);
4685         nfsd4_client_tracking_init(&init_net);
4686         boot_time = get_seconds();
4687         locks_start_grace(&nfsd4_manager);
4688         grace_ended = false;
4689         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4690                nfsd4_grace);
4691         ret = set_callback_cred();
4692         if (ret) {
4693                 ret = -ENOMEM;
4694                 goto out_recovery;
4695         }
4696         laundry_wq = create_singlethread_workqueue("nfsd4");
4697         if (laundry_wq == NULL) {
4698                 ret = -ENOMEM;
4699                 goto out_recovery;
4700         }
4701         ret = nfsd4_create_callback_queue();
4702         if (ret)
4703                 goto out_free_laundry;
4704         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4705         set_max_delegations();
4706         return 0;
4707 out_free_laundry:
4708         destroy_workqueue(laundry_wq);
4709 out_recovery:
4710         nfsd4_client_tracking_exit(&init_net);
4711         put_net(&init_net);
4712         return ret;
4713 }
4714
4715 static void
4716 __nfs4_state_shutdown(void)
4717 {
4718         int i;
4719         struct nfs4_client *clp = NULL;
4720         struct nfs4_delegation *dp = NULL;
4721         struct list_head *pos, *next, reaplist;
4722
4723         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4724                 while (!list_empty(&conf_id_hashtbl[i])) {
4725                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4726                         expire_client(clp);
4727                 }
4728                 while (!list_empty(&unconf_str_hashtbl[i])) {
4729                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4730                         expire_client(clp);
4731                 }
4732         }
4733         INIT_LIST_HEAD(&reaplist);
4734         spin_lock(&recall_lock);
4735         list_for_each_safe(pos, next, &del_recall_lru) {
4736                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4737                 list_move(&dp->dl_recall_lru, &reaplist);
4738         }
4739         spin_unlock(&recall_lock);
4740         list_for_each_safe(pos, next, &reaplist) {
4741                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4742                 unhash_delegation(dp);
4743         }
4744
4745         nfsd4_client_tracking_exit(&init_net);
4746         put_net(&init_net);
4747 }
4748
4749 void
4750 nfs4_state_shutdown(void)
4751 {
4752         cancel_delayed_work_sync(&laundromat_work);
4753         destroy_workqueue(laundry_wq);
4754         locks_end_grace(&nfsd4_manager);
4755         nfs4_lock_state();
4756         __nfs4_state_shutdown();
4757         nfs4_unlock_state();
4758         nfsd4_destroy_callback_queue();
4759 }
4760
4761 static void
4762 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4763 {
4764         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4765                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4766 }
4767
4768 static void
4769 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4770 {
4771         if (cstate->minorversion) {
4772                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4773                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4774         }
4775 }
4776
4777 void
4778 clear_current_stateid(struct nfsd4_compound_state *cstate)
4779 {
4780         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4781 }
4782
4783 /*
4784  * functions to set current state id
4785  */
4786 void
4787 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4788 {
4789         put_stateid(cstate, &odp->od_stateid);
4790 }
4791
4792 void
4793 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
4794 {
4795         put_stateid(cstate, &open->op_stateid);
4796 }
4797
4798 void
4799 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4800 {
4801         put_stateid(cstate, &close->cl_stateid);
4802 }
4803
4804 void
4805 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
4806 {
4807         put_stateid(cstate, &lock->lk_resp_stateid);
4808 }
4809
4810 /*
4811  * functions to consume current state id
4812  */
4813
4814 void
4815 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4816 {
4817         get_stateid(cstate, &odp->od_stateid);
4818 }
4819
4820 void
4821 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
4822 {
4823         get_stateid(cstate, &drp->dr_stateid);
4824 }
4825
4826 void
4827 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
4828 {
4829         get_stateid(cstate, &fsp->fr_stateid);
4830 }
4831
4832 void
4833 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
4834 {
4835         get_stateid(cstate, &setattr->sa_stateid);
4836 }
4837
4838 void
4839 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4840 {
4841         get_stateid(cstate, &close->cl_stateid);
4842 }
4843
4844 void
4845 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
4846 {
4847         get_stateid(cstate, &locku->lu_stateid);
4848 }
4849
4850 void
4851 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
4852 {
4853         get_stateid(cstate, &read->rd_stateid);
4854 }
4855
4856 void
4857 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
4858 {
4859         get_stateid(cstate, &write->wr_stateid);
4860 }